Findings of the Association for Computational Linguistics: ACL 2026

WikiVideo: Article Generation from Multiple Videos (2026.findings-acl)

Challenge:	Existing methods for extracting present and past personal emotion expressions from text-based crisis conversations are lacking in clinically relevant areas.
Approach:	They propose a method for extracting present and past personal emotion expressions from text-based crisis conversations and train a transformer-based model that captures contextual distinctions between true personal emotion and other mentions.
Outcome:	The proposed method outperforms a regex and a model trained on real conversation data and achieves an F1 score of 0.856.

Copied to clipboard

Alexander Martin, Reno Kriz, William Gantt Walden, Kate Sanders, Hannah Recknor, Eugene Yang, Francis Ferraro, Benjamin Van Durme

Large Language Models Are Effective Human Annotation Assistants, But Not Good Independent Annotators (2026.findings-acl)

Challenge:	Existing methods for retrieval-augmented generation focus on text rather than video.
Approach:	They propose a benchmark to generate Wikipedia-style articles from multiple videos . they propose 'collaborative article generation' that leverages an r1-style reasoning model and a VideoLLM to draw higher-level inferences about the target event than is possible with VideoLLms alone.
Outcome:	The proposed method outperforms existing methods in oracle retrieval and RAG settings while suggesting promising avenues for future work.

Copied to clipboard

Feng Gu, Zongxia Li, Carlos R. Colon, Benjamin Evans, Ishani Mondal, Jordan Lee Boyd-Graber

Accelerating LLM Fine-Tuning via Embedding Knowledge Transfer (2026.findings-acl)

Challenge:	State-of-the-art NLP models are expensive and inefficient for event annotation.
Approach:	They propose to integrate LLMs into a holistic workflow that summarizes news with event coreference resolution and argument extraction in three modes: AI-only, AI assistance, and human only.
Outcome:	The proposed workflow integrates LLMs to alleviate human labor in a holistic pipeline.

Copied to clipboard

Meishu Peng, Ziyue Zhang, Yi Zhang, Pengyang Wang, Zixuan Yuan, Denghui Zhang

Understanding Secret Leakage Risks in Code LLMs: A Tokenization Perspective (2026.findings-acl)

Challenge:	Existing studies on parameter-efficient fine-tuning (PEFT) have produced many state-of-the-art results by adapting LLMs to new tasks, but it requires substantial training data and time to enhance model performance.
Approach:	They propose a parameter-efficient fine-tuning framework which efficiently transfers knowledge from a small expert model to a target large model via embedding layers.
Outcome:	The proposed framework accelerates domain-specific fine-tuning, improves model performance and remains robust across diverse model families and PEFT methods.

Copied to clipboard

Meifang Chen, Zhe Yang, Huang Nianchen, Yizhan Huang, Yichen LI, Zihan Li, Michael R. Lyu

MALLM-GAN: Multi-Agent Large Language Model as Generative Adversarial Network for Synthesizing Tabular Data (2026.findings-acl)

Challenge:	Code Large Language Models (CLLMs) are reshaping how software is built, maintained, and evolved.
Approach:	They propose to use BPE tokenization to inadvertently leak code secrets . they propose to mitigate the gibberish bias by using a newer tokenizer .
Outcome:	The proposed model is based on a novel method that can be used to detect and mitigate gibberish bias in CLLMs.

Copied to clipboard

Yaobin Ling, Xiaoqian Jiang, Yejin Kim

EMCompress: Video-LLMs with Endomorphic Multimodal Compression (2026.findings-acl)

Challenge:	Existing models for tabular data generation require large amounts of data to train effectively.
Approach:	They propose a framework to generate tabular data powered by large language models that emulates a Generative Adversarial Network.
Outcome:	The proposed framework outperforms state-of-the-art models while keeping privacy of real data.

Copied to clipboard

Zheyu Fan, Jiateng Liu, Yuji Zhang, Zihan Wang, Yi R. Fung, Manling Li, Heng Ji

Learning from Contrastive Prompts: An Automated Prompt Optimization Framework (2026.findings-acl)

Challenge:	Static, sparse frame sampling either dilutes evidence across task-irrelevant segments at significant cost or misses fine-grained temporal semantics altogether.
Approach:	They propose a novel task that compresses multimodal input while preserving answer invariance across reasonable downstream models.
Outcome:	The proposed task surpasses prior methods by 0.40 F-1 with competitive query rewriting.

Copied to clipboard

Mingqi Li, Karan Aggarwal, Yong Xie, Aitzaz Ahmad, Stephen Lau

Private Seeds, Public LLMs: Realistic and Privacy-Preserving Synthetic Data Generation (2026.findings-acl)

Challenge:	Existing prompt optimization methods often underperform due to learning exclusively from incorrect samples.
Approach:	They propose a framework that leverages contrastive prompts to distinguish between high- and low-performing cases.
Outcome:	The proposed framework can generalize across open and proprietary models and NLU benchmarks.

Copied to clipboard

Qian Ma, Sarah Rajtmajer

ToolScope: An Agentic Framework for Vision-Guided and Long-Horizon Tool Use (2026.findings-acl)

Challenge:	Large language models (LLMs) are a powerful tool for creating synthetic replicas of private text.
Approach:	They propose a method for creating privacy-preserving synthetic data using private seeds and a formal differential privacy mechanism.
Outcome:	The proposed method achieves high fidelity to private data while providing strong privacy protection.

Copied to clipboard

Mengjie Deng, Guanting Dong, Zhicheng Dou

VChain: Chain-of-Visual-Thought for Reasoning in Video Generation (2026.findings-acl)

Challenge:	Recent advances in large language models have demonstrated remarkable problem-solving capabilities . however, enabling multimodal large language model to flexibly and efficiently utilize external tools remains a challenge .
Approach:	They introduce an agentic framework to unify global planning with local multimodal perception . they evaluate ToolScope on four VQA benchmarks across diverse domains .
Outcome:	The proposed framework unifies global planning with local multimodal perception . it adopts a specialized Perceive tool to mitigate visual context degradation in long-horizon VQA task.

Copied to clipboard

Ziqi Huang, Ning Yu, Gordon Chen, Haonan Qiu, Paul Debevec, Ziwei Liu

Diversity Collapse in Multi-Agent LLM Systems: Structural Coupling and Collective Failure in Open-Ended Idea Generation (2026.findings-acl)

Challenge:	Recent video generation models struggle to synthesize complex dynamics with a coherent chain of consequences.
Approach:	They propose a framework that injects visual reasoning signals from multimodal models into video generation.
Outcome:	a new framework that leverages multimodal models to generate sparse keyframes significantly improves quality of generated videos.

Copied to clipboard

Nuo Chen, Yicheng Tong, Yuzhe Yang, Yufei He, Xueyi Zhang, Zou Qingyun, Qian Wang, Bingsheng He

Challenge:	Multi-agent systems (MAS) are increasingly used for open-ended idea generation . when and why collective interaction expands the solution space remains unclear .
Approach:	They propose to study diversity in multi-agent systems across three bottom-up levels: model intelligence, agent cognition, and system dynamics.
Outcome:	The proposed model yields diminishing diversity despite higher quality . the proposed model fails to expand diversity and causes it to collapse .

Justice in Judgment: Unveiling (Hidden) Bias in LLM-assisted Peer Reviews (2026.findings-acl)

Copied to clipboard

Sai Suresh Macharla Vasu, Ivaxi Sheth, Hui-Po Wang, Ruta Binkyte, Mario Fritz

A Study of LLMs’ Preferences for Libraries and Programming Languages (2026.findings-acl)

Challenge:	Existing studies show that large language models carry implicit biases across race, gender, and religion . prior studies documented such biase based on text generation and classification tasks .
Approach:	They investigate bias in large language models by controlling metadata on author metadata . authors found affiliation bias favoring authors from highly ranked institutions .
Outcome:	The proposed model favors authors from highly ranked institutions, the authors show . the model also favors author affiliations from highly-ranked institutions .

Copied to clipboard

Lukas Twist, Jie M. Zhang, Mark Harman, Don Syme, Joost Noppen, Helen Yannakoudakis, Detlef Nauck

Unveiling Inherent Visual Grounding in Multimodal LLMs for Text-Rich Images (2026.findings-acl)

Challenge:	Existing evaluations focus on functional correctness or syntactic validity, overlooking how LLMs make critical design choices such as which library or programming language to use.
Approach:	They conduct the first systematic study of LLMs’ preferences for libraries and programming languages when generating code, covering eight different LLM.
Outcome:	The proposed benchmarks show that LLMs prioritize familiarity and popularity over suitability and task-specific optimality.

Copied to clipboard

Shijie Zhou, Jihyung Kil, Ming Li, Jiuxiang Gu, Curtis Wigington, Rajiv Jain, Changyou Chen, Ruiyi Zhang

PUMA: Projected Universal Multilingual ASR for Low-Resource Settings. Application to Diverse African Languages (2026.findings-acl)

Challenge:	Existing multimodal large language model (MLLM) approaches struggle to align query tokens with visual–text patches, heavily relying on lengthy OCR inputs.
Approach:	They propose an OCR-free approach that leverages the MLLM's inherent multi-head attention for multi-patch grounding.
Outcome:	Empirical results show that the proposed approach outperforms existing approaches on challenging document grounding benchmarks.

Copied to clipboard

Ilyes Oukid, Bilal Faye, Hanane Azzag, Mustapha Lebbah, Said Yacine Boulahia

MGPO: Thinking with Images via Multi-Turn Grounding-Based Reinforcement Learning (2026.findings-acl)

Challenge:	Existing multilingual ASR models fail to generalize to low-resource languages while remaining costly to scale.
Approach:	They propose a multilingual ASR model that integrates a learnable language token with acoustic representations to enable language-aware processing.
Outcome:	The proposed model improves low-resource performance with reduced model complexity on African languages.

Copied to clipboard

Xinyu Huang, Yuhao Dong, Weiwei Tian, Bo Li, Rui Feng, Ziwei Liu

How Grounded is Wikipedia? A Study on Structured Evidential Support and Retrieval (2026.findings-acl)

Challenge:	State-of-the-art large multimodal models face challenges when processing high-resolution images, as these inputs are converted into enormous visual tokens, many of which are irrelevant to the downstream task.
Approach:	They propose a multi-turn grounding-based policy optimization framework that enables LMMs to iteratively focus on key visual regions by automatically cropping sub-images based on model-predicted grounding coordinates within a multiple-turn conversation framework.
Outcome:	The proposed framework improves on Qwen2.5-VL-7B with 21K samples and surpasses OpenAI’s o1 and GPT-4o models on the out-of-distribution (OOD) V* Bench.

Copied to clipboard

William Gantt Walden, Kathryn Ricci, Miriam Wanner, Zhengping Jiang, Chandler May, Rongkun Zhou, Benjamin Van Durme

Jailbreaking Attacks vs. Content Safety Filters: How Far Are We in the LLM Safety Arms Race? (2026.findings-acl)

Challenge:	22% of claims in Wikipedia lead sections are unsupported by the article body . 30% of annotated claims in the article body are unbacked by their (publicly accessible) sources .
Approach:	They analyze Wikipedia's claim support annotations using a large-scale dataset . they find that 22% of Wikipedia claims are unsupported by the article body .
Outcome:	The proposed dataset analyzes claims support annotations on biographical Wikipedia articles.

Copied to clipboard

Yuan Xin, Dingfan Chen, Linyi Yang, Michael Backes, Xiao Zhang

BrowseConf: Confidence-Guided Test-Time Scaling for Web Agents (2026.findings-acl)

Challenge:	Existing studies have focused on the models, neglecting the full deployment pipeline . previous studies have underestimated the practical success of these attacks .
Approach:	They evaluate the effectiveness of jailbreak attacks targeting LLM safety alignment . they highlight critical gaps and call for further refinement of detection accuracy and usability .
Outcome:	The proposed attacks can detect at least one safety filter across the entire deployment pipeline.

Copied to clipboard

Litu Ou, Kuan Li, Huifeng Yin, Liwen Zhang, Zhongwang Zhang, Xixi Wu, Rui Ye, Zile Qiao, Yong Jiang, Pengjun Xie, Fei Huang, Jingren Zhou

Challenge:	Existing work on confidence in LLMs is limited.
Approach:	They propose to use confidence scores to determine model answer quality and encourage model to try again until it reaches satisfactory confidence level.
Outcome:	The proposed methods significantly reduce token consumption while demonstrating competitive performance compared to baseline fixed budget methods.

HyperEdit: Unlocking Instruction-based Text Editing in LLMs via Hypernetworks (2026.findings-acl)

Copied to clipboard

Yiming Zeng, Jinghan Cao, Zexin Li, Wanhao Yu, Zhankai Ye, Dawei Xiang, Ting Hua, Xin Liu, Shangqian Gao, Tingting Yu

Strategy-Induct: Task-Level Strategy Induction for Instruction Generation (2026.findings-acl)

Challenge:	Existing approaches treat instruction-based text editing as a generic text generation problem. Existing methods either over-edit or fail to apply modifications consistently.
Approach:	They propose a framework that processes each editing request to best align with it.
Outcome:	The proposed framework achieves 9% improvement over the state-of-the-art model.

Copied to clipboard

Po-Chun Chen, Hen-Hsen Huang, Hsin-Hsi Chen

Idiom Understanding as a Tool to Measure the Dialect Gap (2026.findings-acl)

Challenge:	Existing methods for task-level instruction generation rely on input-output pairs . obtaining labeled answers can be difficult or costly, limiting generalization across architectures.
Approach:	They propose a framework that derives task-level instructions solely from a small set of example questions without requiring labeled answers.
Outcome:	The proposed framework outperforms state-of-the-art methods in question-only settings.

Copied to clipboard

David Beauchemin, Yan Tremblay, Mohamed Amine Youssef, Richard Khoury

Challenge:	idiom understanding and dialect understanding are well-established benchmarks in natural language processing . a language model trained in one of these dialects will have trouble making sense of the idiomatics from the other two .
Approach:	They propose to combine idiom understanding and dialect understanding to test regional idiomatics . they propose to use regional ids as benchmarks for other natural language processing languages .
Outcome:	The proposed benchmarks are based on idiomatic and dialect understanding datasets for french and francais . the results show prestige-language proficiency does not guarantee regional dialect understanding .

Logic Jailbreak: Efficiently Unlocking LLM Safety Restrictions Through Formal Logical Expression (2026.findings-acl)

Copied to clipboard

Jingyu Peng, Maolin Wang, Nan Wang, Jiatong Li, Yuchen Li, Yuyang Ye, Wanyu Wang, Pengyue Jia, Kai Zhang, Xiangyu Zhao

Two Heads Are Better Than One: Audio-Visual Speech Error Correction with Dual Hypotheses (2026.findings-acl)

Challenge:	Despite advances in aligning LLMs with human values, current safety mechanisms remain vulnerable to jailbreak attacks.
Approach:	They propose a black-box jailbreak method that uses logical expression translation to bypass LLM safety mechanisms.
Outcome:	The proposed method exploits the distributional gap between alignment data and logic-expressed inputs while preserving the underlying semantic intent and readability while evading safety constraints.

Copied to clipboard

Sungnyun Kim, Kangwook Jang, Sungwoo Cho, Joon Son Chung, Hoi-Rin Kim, Se-Young Yun

Towards Hierarchical Multi-Step Reward Models for Enhanced Reasoning in Large Language Models (2026.findings-acl)

Challenge:	Recent advances have introduced GER frameworks that utilize LLMs to refine ASR outputs.
Approach:	They propose a framework that allows a large language model to compose independent N-best hypotheses from separate automatic speech recognition (ASR) and visual speech recognition models.
Outcome:	The proposed framework achieves 57.7% error rate gain over standard ASR baseline, compared to single-stream approaches that achieve only 10% gain.

Copied to clipboard

Teng Wang, Jiang Zhangyi, Zhenqi He, Hailei Gong, Shenyang Tong, Wenhan Yang, Zeyu Li, Yanan Zheng, Zifan He, Zewen Ye, Shengjie Ma, Jianping Zhang

SciVQR: A Multidisciplinary Multimodal Benchmark for Advanced Scientific Reasoning Evaluation (2026.findings-acl)

Challenge:	Existing Process Reward Models (PRMs) are vulnerable to reward hacking and require expensive, large-scale annotation of reasoning steps.
Approach:	They propose a reward model approach which evaluates both individual and consecutive reasoning steps from fine-grained and coarse-grounded level.
Outcome:	Empirical results show that the proposed model performs better than existing PRMs and is more robust than existing models.

Copied to clipboard

Longteng Guo, Xuanxu Lin, Dongze Hao, Tongtian Yue, Pengkang Huo, Jiatong Ma, Yuchen Liu, Jing Liu

UrBLiMP: A Benchmark for Evaluating the Linguistic Competence of Large Language Models in Urdu (2026.findings-acl)

Challenge:	Existing benchmarks for multimodal large language models fail to capture complexity and traceability of reasoning processes . SciVQR includes domain-specific visuals and challenges models to combine visual comprehension with reasoning.
Approach:	They propose a multimodal benchmark for scientific reasoning covering 54 subfields . SciVQR includes domain-specific visuals and challenges models to combine visual comprehension with reasoning .
Outcome:	SciVQR evaluates 54 subfields in mathematics, physics, chemistry, geography, astronomy, and biology . the results highlight the need for improved multi-step reasoning and integration of interdisciplinary knowledge .

Copied to clipboard

Farah Adeeba, Brian Dillon, Hassan Sajjad, Rajesh Bhatt

ANCHOR: LLM-driven Subject Conditioning for Text-to-Image Synthesis (2026.findings-acl)

Challenge:	Evaluating how large language models capture grammatical structure of low-resource languages remains underexplored.
Approach:	They evaluate a set of 5,696 minimal pairs that contrast grammatical acceptability across ten core syntactic and morpho-syntactical phenomena in Urdu.
Outcome:	The proposed framework compares multilingual models with the proprietary model . the proposed framework achieves the highest average accuracy on regular phenomena .

Copied to clipboard

Aashish Anantha Ramakrishnan, Sharon X Huang, Dongwon Lee

Personalized Benchmarking: Evaluating LLMs by Individual Preferences (2026.findings-acl)

Challenge:	Text-to-image (T2I) models are trained on literal, object-centric prompts designed to reflect the visible contents of an image.
Approach:	They propose a method to extract key subjects and enhance their representation at embedding-level using Large Language Models.
Outcome:	The proposed model significantly improves image-caption consistency and human preference alignment.

Copied to clipboard

Cristina Garbacea, Heran Wang, Chenhao Tan

Challenge:	Current benchmarks average preferences across all users to compute aggregate ratings . this overlooks individual user preferences when establishing model rankings .
Approach:	They compute personalized model rankings using ELO ratings and Bradley-Terry coefficients . they find users exhibit substantial heterogeneity in topical interests and communication styles .
Outcome:	The results show that individual rankings of LLM models diverge dramatically from aggregate rankings . a compact combination of topic and style features provides a useful feature space .

Op-Fed: Opinion, Stance, and Monetary Policy Annotations on FOMC Transcripts Using Active Learning (2026.findings-acl)

Copied to clipboard

Alisa Kanganis, Katherine A. Keith

C3D: Enhancing LLM Reasoning via Confidence-Guided Contrastive Decoding (2026.findings-acl)

Challenge:	Op-Fed dataset captures monetary policy stance, but there are still questions about FOMC decision making .
Approach:	They use a dataset of 1044 human-annotated sentences and their contexts from FOMC transcripts to capture monetary policy stance.
Outcome:	The proposed dataset captures whether an individual FOMC member expresses support for tightening or loosening policy.

Copied to clipboard

Yufeng Zhang, Xuepeng Wang, Lingxiang Wu, Jinqiao Wang

Do LLMs Really Know What They Don’t Know? Internal States Mainly Reflect Knowledge Recall Rather Than Truthfulness (2026.findings-acl)

Challenge:	Large language models are prone to distraction by contextual information during reasoning tasks.
Approach:	They propose a decoding method that uses predicted logits to estimate the model's confidence.
Outcome:	The proposed method reveals how the model dynamically activates and adjusts its consideration of each premise as reasoning progresses.

Copied to clipboard

Chi Seng Cheang, Hou Pong Chan, Wenxuan Zhang, Yang Deng

Enhancing Hallucination Detection via Future Context (2026.findings-acl)

Challenge:	Recent work suggests that large language models (LLMs) produce hallucinated and factually correct outputs.
Approach:	They propose a taxonomy categorizing hallucinations into Unassociated Hallucination (UH) and Associated Hallucinian (AH) they propose to use internal signals to distinguish hallucinos from factual errors .
Outcome:	The proposed taxonomy categorizes hallucinations into Unassociated Hallucination (UH) and Associated Hallucinications (AHs) based on the proposed taxonomic, the authors show that hidden states reflect whether the model is recalling parametric knowledge rather than the truthfulness of the output itself.

Copied to clipboard

Joosung Lee, Cheonbok Park, Hwiyeol Jo, Jeonghoon Kim, Joonsuk Park, Kang Min Yoo

MTO: A Multi-turn Conversational Text-to-OverpassQL Dataset for Enhanced OpenStreetMap Query Generation (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are widely used to generate plausible text on online platforms, without revealing the generation process.
Approach:	They propose a framework for detection of hallucinations in black-box generators by analyzing future contexts.
Outcome:	The proposed framework improves on existing methods and demonstrates that it is feasible to integrate it with other models.

Copied to clipboard

Haodi Zhang, Xinrui Zhu, Mingze Kong, Zhidan Liu, Tao Fan, Kaishun Wu, Yuanfeng Song

A Dual-Phase Self-Evolution Framework for Large Language Models (2026.findings-acl)

Challenge:	a framework for constructing multi-turn Text-to-OverpassQL dialogue datasets is proposed . a dataset of over 7,800 dialogues contains more than 20,000 individual utterances .
Approach:	They propose a framework for constructing multi-turn Text-to-OverpassQL dialogue datasets . they convert Overpass queries into syntax trees using a custom parser based on OverpassQl .
Outcome:	The proposed dataset includes over 7,800 dialogues, each containing 2 to 4 user utterances . it is the first multi-turn Text-to-OverpassQL dataset built upon the OverpassNL corpus .

Copied to clipboard

Haoran Sun, Zekun Zhang, Shaoning Zeng

Preference-Aware Memory Update for Long-Term LLM Agents (2026.findings-acl)

Challenge:	Existing strategies to optimize LLMs through pretraining fail to enhance domain cognition.
Approach:	They propose a dual-phase self-evolution framework that integrates user preference adaptation and domain-specific competence to optimize LLMs.
Outcome:	The proposed framework outperforms Supervised Fine-Tuning, Preference Optimization, and Memory-Augmented baselines on general NLP benchmarks and long-term dialogue tasks.

Copied to clipboard

Haoran Sun, Zekun Zhang, Shaoning Zeng

Understanding In-Context Learning Beyond Transformers: An Investigation of State Space and Hybrid Architectures (2026.findings-acl)

Challenge:	Existing methods for integrating long-term memory do not provide dynamic and personalized memory refinement.
Approach:	They propose a long-term memory update mechanism that enables dynamic and personalized memory refinement.
Outcome:	The proposed mechanism improves the performance of LLM-based agents in five tasks.

Copied to clipboard

Shenran Wang, Timothy Tin-Long Tse, Jian Zhu

C2DLM: Causal Concept-Guided Diffusion Large Language Models (2026.findings-acl)

Challenge:	In-context learning is an emergent ability from pretrained Large Language Models (LLMs).
Approach:	They perform in-depth evaluations of in-context learning on transformers and hybrid large language models using behavioral probing and intervention-based methods.
Outcome:	The proposed model performs well on state-of-the-art transformer, state-space, and hybrid large language models.

Copied to clipboard

Kairong Han, Nuanqiao Shan, Ziyu Zhao, Zijing Hu, Xinpeng Dong, Ye Jun Jian, Lujia Pan, Fei Wu, Kun Kuang

CMTD: Cognitive Modeling with Traits and Distortions for Multimodal Emotion Recognition in Conversations (2026.findings-acl)

Challenge:	Autoregressive (AR) and diffusion language models (DLMs) suffer from insufficient reasoning capabilities.
Approach:	They propose a fully connected Diffusion Language Model that uses a concept-level causal graph to guide attention to learn causal relationships between concepts.
Outcome:	The proposed model achieves a 12% improvement and 3.2 training speedup on the COT-OrderPerturb task, along with an average gain of 1.31% across six downstream reasoning tasks.

Copied to clipboard

Minh-Tien Nguyen, Huu-Loi Le, Manh-Cuong Phan, Hajime Hotta

Bridging the Culture Gap: A Framework for LLM-Driven Socio-Cultural Localization of Math Word Problems in Low-Resource Languages (2026.findings-acl)

Challenge:	Experimental results show that traits temper negativity bias from distortions, and cognitive modeling with psychological, visual, and acoustic information can improve the performance of MERC.
Approach:	They propose a framework for multimodal emotion recognition in conversations that takes advantage of stable personality traits, dynamic cognitive distortions, visual and acoustic features of interlocutors to enhance the emotional intelligence of LLMs.
Outcome:	Experimental results show that traits temper negativity bias from distortions, and cognitive modeling with psychological, visual, and acoustic information can improve the performance of MERC.

Copied to clipboard

Israel Abebe Azime, Tadesse Destaw Belay, Dietrich Klakow, Philipp Slusallek, Anshuman Chhabra

Stop Guessing When to Stop Testing: Efficient Model Evaluation with Just Enough Data (2026.findings-acl)

Challenge:	Existing multilingual benchmarks that use translations retain English-centric entities.
Approach:	They propose a framework that culturally localizes translated datasets into variants enriched with local entities.
Outcome:	The proposed framework mitigates English-centric entity bias and improves model robustness when native entities are introduced across languages.

Copied to clipboard

Ofir Arviv, Kristjan Greenewald, Yotam Perlitz, Hadar Mulian, Michal Shmueli-Scheuer, Leshem Choshen

AnyGraph: Graph Foundation Model in the Wild (2026.findings-acl)

Challenge:	Current evaluation practices, typically employing fixed-size benchmarks, are inherently wasteful, continuing to the predetermined sample size even when the CI reaches 2.5, saving 80% of the evaluation cost.
Approach:	They propose an adaptive evaluation framework that combines sequential testing with stopping criteria tailored to common evaluation needs such as diminishing returns detection and minimum detectable effect size.
Outcome:	The proposed framework reduces computational cost and reliability while maintaining statistical significance.

Copied to clipboard

Lianghao Xia, Chao Huang

DOSE: Data Selection for Multi-Modal LLMs via Off-the-Shelf Models (2026.findings-acl)

Challenge:	Existing graph learning models struggle to extract generalizable insights from heterogeneous graph data, requiring extensive fine-tuning and limiting versatility across domains.
Approach:	They propose a graph foundation model that can handle key challenges such as Structure Heterogenity and Feature Heterogenicity.
Outcome:	The proposed model can handle key challenges such as structure heterogeneity, Feature heterogenity and fast adaptation across domains.

Copied to clipboard

Biao Wu, Yiwu Zhong, Meng Fang, Ling Chen

MergeIT: From Selection to Merging for Efficient Instruction Tuning (2026.findings-acl)

Challenge:	Existing data filtering methods are expensive because they are trained on the same data they are meant to screen.
Approach:	They propose to use off-the-shelf pretrained models that have never seen the target data to select training samples for larger and stronger multimodal models without task-specific training.
Outcome:	The proposed method can achieve comparable or even better results than those trained on the full dataset in standard VQA and math benchmarks.

Copied to clipboard

Hongyi Cai, Yuqian Fu, Hongming Fu, Bo Zhao

The Illusion of Insight in Reasoning Models (2026.findings-acl)

Challenge:	Existing methods for instruction tuning rely on LLMs to score instruction quality . existing methods rely only on Llms to rank instruction quality, but this approach is expensive and time-consuming .
Approach:	They propose a novel LLM-based Merging strategy for better Instruction Tuning that shifts the focus from selection to synthesis.
Outcome:	The proposed method reduces time and computational cost while preserving diversity and reducing redundancy.

Copied to clipboard

Liv G. d’Aliberti, Manoel Horta Ribeiro

Adaptive Layer Selection for Layer-Wise Token Pruning in LLM Inference (2026.findings-acl)

Challenge:	Existing evidence suggests that reasoning models exhibit "Aha!" moments . however, it remains unclear whether such intrinsic shifts in reasoning strategy improve performance .
Approach:	They analyze reasoning traces, training checks, three reasoning domains to detect mid-reasoning shifts . they find that reasoning shifts are rare, do not become more frequent with training .
Outcome:	The proposed method detects mid-reasoning shifts and instrument training runs.

Copied to clipboard

Rei Taniguchi, Yuyang Dong, Makoto Onizuka, Chuan Xiao

Challenge:	Large language models (LLMs) have demonstrated remarkable capabilities in processing long contexts.
Approach:	They propose a training-free method that adaptively chooses the selection layer for KV cache reduction . they exploit the variance of token ranks ordered by attention score to optimize decoding .
Outcome:	The proposed method outperforms state-of-the-art token pruning methods on InfiniteBench, RULER, and NIAH benchmarks.

Aligning What LLMs Do and Say: Towards Self-Consistent Explanations (2026.findings-acl)

Copied to clipboard

Sahar Admoni, Ofra Amir, Assaf Hallak, Yftah Ziser

TARo: Token-level Adaptive Routing for LLM Test-time Alignment (2026.findings-acl)

Challenge:	Large language models (LLMs) are often prompted to produce natural language explanations, but the features driving the answer are often different from those emphasized in their explanations.
Approach:	They propose a large-scale benchmark linking model decisions with diverse explanations and attribution vectors across datasets, methods, and model families to address this gap.
Outcome:	The proposed model generates an answer where the word NLP in the prompt has high feature importance.

Copied to clipboard

Arushi Rai, Qiang Zhang, Hanqing Zeng, Yunkai Zhang, Dipesh Tamboli, Xiangjun Fan, Zhuokai Zhao, Lizhu Zhang

ProRank: Prompt Warmup via Reinforcement Learning for Small Language Models Reranking (2026.findings-acl)

Challenge:	Large language models (LLMs) exhibit strong reasoning capabilities but typically require expensive post-training to reach high performance.
Approach:	They propose to use token-level Adaptive Routing to steer frozen LLMs toward structured reasoning entirely at inference time.
Outcome:	Extensive experiments show that TARo significantly improves reasoning performance by up to +22.4% over base model and +8.4% .

Copied to clipboard

Xianming LI, Aamir Shakir, Rui Huang, Julius Lipp, Benjamin Clavié, Jing Li

ChemAmp: Amplified Chemistry Tools via Composable Agents (2026.findings-acl)

Challenge:	Recent Large Language Models (LLMs) have demonstrated remarkable performance in document reranking tasks.
Approach:	They propose a two-stage training approach for document reranking using reinforcement learning and fine-grained score learning.
Outcome:	The proposed approach outperforms open-source and proprietary reranking models on BEIR benchmark.

Copied to clipboard

Zhucong Li, Powei Chang, Jin Xiao, Zhijian Zhou, Qianyu He, Jiaqing Liang, Fenglei Cao, Xu Yinghui, Yuan Qi

Challenge:	LLM-based agents are powerful tools for automating complex scientific workflows, especially in chemistry, but their single-task performance is limited by tool constraints.
Approach:	They propose a framework that optimizes the collective capabilities of specialized tools by dynamic coordination within individual tasks.
Outcome:	The proposed framework outperforms chemistry-specialized models, generalist LLMs, and agent systems with tool orchestration.

CodeContests-O: Powering LLMs via Feedback-Driven Iterative Test Case Generation (2026.findings-acl)

Copied to clipboard

Jianfeng Cai, Jinhua Zhu, Ruopei Sun, Kangwen Zhao, Dongyun Xue, Mingxiao Feng, Wengang Zhou, Houqiang Li

Hybrid-Vector Retrieval for Visually Rich Documents: Combining Single-Vector Efficiency and Multi-Vector Accuracy (2026.findings-acl)

Challenge:	Existing approaches to synthesize test cases using Large Language Models (LLMs) rely on the model’s intrinsic generation capabilities without external feedback, resulting in insufficiently diverse cases.
Approach:	They propose a feedback-driven iterative framework that leverages Large Language Models to generate initial test cases, execute them against known correct and incorrect solutions, and utilizes the failed results as feedback to guide the LLM in refining the test cases toward high fidelity and discriminability.
Outcome:	The proposed method outperforms the existing codecontests and codecontests+ models by 4.30% and 8.78%.

Copied to clipboard

Juyeon Kim, Geon Lee, Dongwon Choi, Taeuk Kim, Kijung Shin

Multi-Scale Spectral Selection and Entropy-Guided Uncertainty Fusion for Multimodal Rumor Detection (2026.findings-acl)

Challenge:	Existing approaches to document retrieval are coarse and efficient, but expensive.
Approach:	a plug-and-play two-stage hybrid-vector framework is proposed to retrieve visually rich documents . HEAVEN efficiently retrieves candidate pages using a single-vektor method over VS-Pages . it also reranks candidates with a multi-vecctor method while filtering query tokens by linguistic importance .
Outcome:	HEAVEN achieves 99.87% of the Recall@1 performance of multi-vector models on average . it reduces per-query computation by 99.8%, achieving efficiency and accuracy .

Copied to clipboard

Zongliang Han, Wenyu Guo, Guoqing Jin, Yang Liu, Fan Li, Dong Yu, Yan Song, null Zhangfengzhen

Covariance Matrix-Driven Image Channel Allocation for Multimodal Fake News Detection (2026.findings-acl)

Challenge:	Existing methods for multimodal content detection fail to capture cross-modal semantic inconsistencies and ignore inherent noise in multimodal features.
Approach:	They propose a multimodal rumor detection method based on a frequency domain spectral selection method and entropy-guided uncertainty fusion method to capture cross-modal semantic inconsistencies.
Outcome:	The proposed method outperforms state-of-the-art methods in multimodal rumor detection . it shows stronger detection capability and robustness on multiple datasets .

Copied to clipboard

Zongliang Han, Wenyu Guo, Guoqing Jin, Yang Liu, Yan Song, Dong Yu, Wang Min

Mechanistic Insights into Deferred Semantic Drift in LLMs (2026.findings-acl)

Challenge:	Existing methods focus on designing efficient multimodal fusion frameworks to bridge the semantic gap between images and texts.
Approach:	They propose a covariance matrix-driven image channel allocation method that expands the number of original channel maps and assigns importance scores to the expanded channel maps.
Outcome:	The proposed method achieves state-of-the-art on three public multimodal fake news detection benchmark datasets.

Copied to clipboard

Jingjie Zeng, Huayang Li, Liang Yang, Shaowu Zhang, Yuanyuan Sun, Hongfei Lin

Challenge:	Large Language Models (LLMs) face a fundamental challenge with delayed disambiguation: how can a model update the meaning of an early, ambiguous token when clarifying context only appears later in the sequence?
Approach:	They propose a method to defer semantic re-evaluation to subsequent tokens in a process they call "Deferred Semantic Drift" they demonstrate this mechanism in metaphor comprehension and provide causal validation by steering model outputs towards literal or metaphorical meanings via targeted activation interventions.
Outcome:	The proposed model can update the meaning of an ambiguous word when clarifying context arrives only after it has been processed.

JointCQ: Improving Factual Hallucination Detection with Joint Claim and Query Generation (2026.findings-acl)

Copied to clipboard

Fan Xu, Huixuan Zhang, Zhenliang Zhang, Jiahao Wang, Xiaojun Wan

Pseudo2Real: Task Arithmetic for Pseudo-Label Correction in Automatic Speech Recognition (2026.findings-acl)

Challenge:	Existing methods for detecting factual hallucinations in generated content exhibit limitations in the first two stages of the halluciation detection pipeline.
Approach:	They propose a joint claim-and-query generation framework that can detect factual hallucinations in generated content.
Outcome:	The proposed method outperforms existing methods on open-domain QA hallucination detection benchmarks.

Copied to clipboard

Yi-Cheng Lin, Yu-Hsuan Li Liang, Hsuan Su, Tzu-Quan Lin, Shang-Tse Chen, Yun-Nung Chen, Hung-yi Lee

SGA-MCTS: Decoupling Planning from Execution via Training-Free Atomic Experience Retrieval (2026.findings-acl)

Challenge:	ASR models can be used to correct accent-specific errors without ground truth . pseudo-labels inherit the teacher model's systematic biases, authors say .
Approach:	They propose a parameter-space correction technique that captures pseudo-label biases . they propose achieving up to 35% relative WER reduction on a pseudo-labeled target model .
Outcome:	The proposed model achieves 35% relative WER reduction on ten African accents with the Whisper tiny model.

Copied to clipboard

Xin Xie, Dongyun Xue, Wuguannan Yao, Mingxiao Feng, Wengang Zhou, Xiang Qi, Houqiang Li, Peng Zhang

Robustness via Referencing: Defending against Prompt Injection Attacks by Referencing the Executed Instruction (2026.findings-acl)

Challenge:	a new framework casts LLM planning as non-parametric retrieval, but high latency of inference-time search and supervised fine-tuning are limitations.
Approach:	They propose a framework that casts LLM planning as non-parametric retrieval . they leverage Monte Carlo Tree Search to explore the solution space .
Outcome:	Empirical results show that SGA-MCTS can match the performance of SOTA systems without task-specific fine-tuning.

Copied to clipboard

Yulin Chen, Haoran Li, Yuan Sui, Yue Liu, Yufei He, Xiaoling Bai, Chi Fei, Li Yabo, Haozhe Ma, Yangqiu Song, Bryan Hooi

Differentially Private Synthetic Text Generation for Retrieval-Augmented Generation (RAG) (2026.findings-acl)

Challenge:	Prompt injection attacks manipulate large language models (LLMs) by misleading them to deviate from the original input instructions and execute maliciously injected instructions.
Approach:	They propose a prompt injection defense method that suppresses the model's instruction-following tendencies rather than suppressing them.
Outcome:	The proposed method outperforms prompt-engineering-based approaches and fine-tuning methods and reduces the ASR to nearly 0% in some scenarios.

Copied to clipboard

Junki Mori, Kazuya Kakizaki, Taiki Miyagawa, Jun Sakuma

When Slower Isn’t Truer: Inverse Scaling Law of Truthfulness in Multimodal Reasoning (2026.findings-acl)

Challenge:	Existing private RAG methods rely on query-time differential privacy (DP) Existing studies have identified significant privacy risks when their databases contain sensitive information.
Approach:	They propose a framework that generates differentially private RAG databases using LLMs . Unlike prior methods, the synthetic text can be reused once created .
Outcome:	Experiments show that DP-SynRAG achieves superior performance to state-of-the-art RAG systems while maintaining a fixed privacy budget.

Copied to clipboard

Sitong Fang, Wenjing Cao, Jiahao Li, Xuyao Wang, Chi-Min Chan, Sirui Han, Juntao Dai, Yike Guo, Yaodong Yang, Jiaming Ji

StressTest: Can YOUR Speech LM Handle the Stress? (2026.findings-acl)

Challenge:	a study of slow reasoning models for multimodal reasoning finds that they are more prone to fabricating plausible yet false details when confronted with incomplete or misleading visual inputs.
Approach:	They conduct the first systematic study of the inverse scaling law in slow-thinking paradigms for multimodal reasoning.
Outcome:	The findings suggest that slower reasoning models are more prone to fabricating false details . the study analyzed 5,000-sample hierarchical prompt dataset by 50 participants .

Copied to clipboard

Iddo Yosha, Gallil Maimon, Yossi Adi

EVADE: LLM-Based Explanation Generation and Validation for Error Detection in NLI (2026.findings-acl)

Challenge:	Recent speech-aware language models (SLMs) have enabled direct audio processing, allowing models to access the full expressive range of spoken language.
Approach:	They propose a data generation pipeline that simulates change of meaning implied by stress variation and propose 'stresstest' to evaluate models' ability to distinguish between meanings of speech based on stress pattern.
Outcome:	The proposed model outperforms existing models on sentence stress reasoning and detection.

Copied to clipboard

Longfei Zuo, Barbara Plank, Siyao Peng

AgenticRAGTracer: A Hop-Aware Benchmark for Diagnosing Multi-Step Retrieval Reasoning in Agentic RAG (2026.findings-acl)

Challenge:	Human label variation (HLV) arises when multiple labels are valid for the same instance.
Approach:	They propose a framework for generating and validating explanations to detect errors using large language models (LLMs) EVADE framework provides broader explanation coverage and requires less human intervention .
Outcome:	The proposed framework provides broader explanation coverage, requires less human intervention, and delivers better downstream performance in predicting label distributions.

Copied to clipboard

Qijie You, Wenkai Yu, Hao Liang, Zhen Hao Wong, Wentao Zhang

Attention to Non-Adopters (2026.findings-acl)

Challenge:	Existing benchmarks provide only final questions and answers, while lacking intermediate hop-level questions that gradually connect atomic questions to the final multi-hop query.
Approach:	They propose to build a multi-hop reasoning model that is primarily constructed automatically by large language models and designed to support step-by-step validation.
Outcome:	The proposed benchmark spans multiple domains, contains 1,305 data points, and has no overlap with existing mainstream benchmarks.

Copied to clipboard

Kaitlyn Zhou, Kristina Gligorić, Myra Cheng, Michelle S. Lam, Vyoma Raman, Boluwatife Aminu, Caeley Woo, Michael Brockman, Hannah Cha, Dan Jurafsky

Ready to Translate, Not to Represent? Bias and Performance Gaps in Multilingual LLMs Across Language Families and Domains (2026.findings-acl)

Challenge:	incorporating non-adopter perspectives is essential for developing useful and capable LLMs, argues a new study.
Approach:	They argue that incorporating non-adopter perspectives is essential for developing broadly useful and capable LLMs.
Outcome:	The proposed method will risk missing tasks prioritized by non-adopters, the authors argue . they show that non-dots diverge from those of current users, and non-no-acopter needs point towards novel reasoning tasks.

Copied to clipboard

Md. Faiyaz Abdullah Sayeedi, Subhey Sadi Rahman, Md. Mahbub Alam, Md. Adnanul Islam, Jannatul Ferdous Deepti, Tasnim Mohiuddin, Md Mofijul Islam, Swakkhar Shatabda

Music Audio-Visual Question Answering Requires Specialized Multimodal Designs (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have redefined Machine Translation, enabling context-aware and fluent translations across hundreds of languages and textual domains.
Approach:	They propose a framework and dataset to evaluate the translation quality and fairness of open-source LLMs.
Outcome:	The proposed framework and dataset evaluates translation quality and fairness of open-source LLMs.

Copied to clipboard

Wenhao You, Xingjian Diao, Wenjun Huang, Chunhui Zhang, Keyi Kong, Weiyi Wu, Chiyu Ma, Zhongyu Ouyang, Tingxuan Wu, Ming Cheng, Soroush Vosoughi, Jiang Gui

TemporalVLM: Video LLMs for Temporal Reasoning in Long Videos (2026.findings-acl)

Challenge:	Music audio-visual question answering presents unique challenges with dense audio-visual content, intricate temporal dynamics, and the need for domain-specific knowledge.
Approach:	They analyze Music AVQA datasets and analyze their results to identify key design patterns . they propose concrete future directions for incorporating musical priors .
Outcome:	The proposed architectures are critical for success in Music AVQA, the authors argue . they suggest concrete future directions for incorporating musical priors .

Copied to clipboard

Fawad Javed Fateh, Umer Ahmed, Hamza Khan, Zeeshan Zia, Quoc-Huy Tran

Mitigating Judgment Preference Bias in Large Language Models through Group-Based Polling (2026.findings-acl)

Challenge:	Several video understanding applications require the ability of temporal reasoning.
Approach:	They propose a video large language model for temporal reasoning and fine-grained understanding in long videos.
Outcome:	The proposed model outperforms existing methods in time and motion studies and temporal action segmentation evaluations.

Copied to clipboard

Shuliang Liu, Zhipeng Xu, Zhenghao Liu, Yukun Yan, Minghe Yu, Yu Gu, Chong Chen, Huiyuan Xie, Ge Yu

Enhancing Long-Chain Reasoning Distillation through Error-Aware Self-Reflection (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are used as automatic evaluators to provide accurate and reliable assessments.
Approach:	They propose a framework that integrates LLM-based judgment models into a multi-agent system and simulates the interactive client-server polling mechanism.
Outcome:	The proposed framework outperforms supervised models trained on annotated judgment data while requiring no human-labeled annotations.

Copied to clipboard

Zhuoyang Wu, Xinze Li, Zhenghao Liu, Yukun Yan, Zhiyuan Liu, Minghe Yu, Cheng Yang, Yu Gu, Ge Yu, Maosong Sun

New Compendium of a Myriad of Plants: A New Dataset Describing Ancient Chinese Plants (2026.findings-acl)

Challenge:	Existing studies treat SLMs as student models and use long-form Chains-of-Thought (CoTs) as supervision signals for Supervised Fine-Tuning (SFT). Existing research focuses on distilling reasoning ability from LLMs to enhance the mathematical reasoning performance of small-scale models.
Approach:	They propose a framework that refines teacher CoTs through an error-aware reflection process to enable the student model to construct more tailored teacher Cots.
Outcome:	Experiments on multiple mathematical reasoning benchmarks show that ORION improves performance by more than 2% over all baselines.

Copied to clipboard

Xiaobin Shen, Zhongqing Wang, Shichen Li, Chu-Ren Huang, Guodong Zhou

Generative Interfaces for Language Models (2026.findings-acl)

Challenge:	Existing approaches to digitize ancient Chinese texts and extract information from them are shallow and inconsistent with modern realities.
Approach:	They propose to expand ancient Chinese datasets using large language model . they focus on Great Compendium of Myriad Flowers, an ancient plants dataset .
Outcome:	The proposed model can extract plant-related information from classical Chinese poetry and prose.

Copied to clipboard

Jiaqi Chen, Yanzhe Zhang, Yutong Zhang, Yijia Shao, Diyi Yang

EvoEdit: Evolving Null-space Alignment for Robust and Efficient Knowledge Editing (2026.findings-acl)

Challenge:	Large language models are increasingly seen as assistants, copilots, and consultants . however, their linear request-response format often makes interactions inefficient in multi-turn tasks .
Approach:	They propose a paradigm in which large language models respond to user queries by generating user interfaces that enable more adaptive and interactive engagement.
Outcome:	The proposed paradigm outperforms traditional chat-based interfaces in many tasks and interaction patterns.

Copied to clipboard

Sicheng Lyu, Yu Gu, Xinyu Wang, Jerry Huang, Sitao Luan, Yufei Cui, Xiao-Wen Chang, Peng Lu

Efficient Reasoning for LLMs through Speculative Chain-of-Thought (2026.findings-acl)

Challenge:	Existing approaches to modifying large language models require continual updates to rectify outdated or erroneous knowledge.
Approach:	They propose a model editing strategy that mitigates catastrophic interference through sequential null-space alignment.
Outcome:	EvoEdit achieves better or comparable performance than prior state-of-the-art techniques with up to 3.53 speedup.

Copied to clipboard

Jikai Wang, Juntao Li, Jianye Hou, Yan Bowen, Lijun Wu, Min Zhang

From Coarse to Fine: Self-Adaptive Hierarchical Planning for LLM Agents (2026.findings-acl)

Challenge:	Existing methods for efficient reasoning focus on reducing the number of model parameters or shortening the chain-of-thought length.
Approach:	They propose a speculative chain-of-thought (SCoT) method to reduce reasoning latency by accelerating average reasoning speed through large and small model collaboration.
Outcome:	The proposed method reduces reasoning latency by 48%66% and 21%49% on GSM8K, MATH, GaoKao, CollegeMath and Olympiad datasets.

Copied to clipboard

Haoran Tan, Zeyu Zhang, Chen Ma, Tianze Liu, Quanyu Dai, Xu Chen

Challenge:	Existing plans for large language model-based agents are limited by their granularity and lack flexibility.
Approach:	They propose a self-adaptive hierarchical planning mechanism that mimics human planning strategies and generates self-adapted hierarchic plans tailored to the varying difficulty levels of different tasks.
Outcome:	The proposed method significantly improves task execution success rates while mitigating overthinking at the planning level, providing a flexible and efficient solution for multi-step complex decision-making tasks.

Data-Centric Perspectives on Agentic Retrieval-Augmented Generation: A Survey (2026.findings-acl)

Copied to clipboard

Jingwen Deng, Jihao Huang, Zhen Hao Wong, Hao Liang, Quanqing Xu, Bin Cui, Wentao Zhang

WISCA: A Lightweight Model Transition Method to Improve LLM Training via Weight Scaling (2026.findings-acl)

Challenge:	Large Language Models (LLMs) excel at natural language understanding and generation, yet rely on static pre-training data.
Approach:	They propose to augment Large Language Models with external retrieval to ground model outputs . traditional RAG is constrained by a fixed retrieve-then-generate routine . authors aim to guide creation of high-quality datasets for next generation of adaptive LLM agents .
Outcome:	The proposed model can decompose tasks, issue exploratory queries, and refine evidence through iterative retrieval.

Copied to clipboard

Jiacheng Li, Jianchao Tan, Zhidong Yang, Pingwei Sun, Feiye Huo, Jiayu Qin, Xiangyu Zhang, Maoxin He, Guangming Tan, Weile Jia, Xunliang Cai, Tong Zhao

Disentangling Codemixing in Chats: The NUS ABC Codemixed Corpus (2026.findings-acl)

Challenge:	Recent advances in training optimization for Transformer-based large language models lack systematic optimization of weight patterns during training.
Approach:	They propose a Weight Scaling method that rescales weights while preserving model outputs to improve model training efficiency and model quality.
Outcome:	The proposed method significantly improves convergence quality and loss reduction in LLMs with Grouped Query Attention architectures and LoRA fine-tuning tasks.

Copied to clipboard

Svetlana Churina, Akshat Gupta, Nur Insyirah Binte Imam Mujtahid, Kokil Jaidka

Challenge:	Existing studies on code-mixing have not been able to model human interactions in context.
Approach:	They propose to use a general-purpose code-mixing corpus to model human interactions and relationships in context while maintaining ethical standards.
Outcome:	The proposed corpus includes over 355,641 messages spanning various code-mixing patterns, with a primary focus on English, Mandarin, and other languages.

When Thoughts Meet Facts: Reusable Reasoning for Long-Context LMs (2026.findings-acl)

Copied to clipboard

Soyeong Jeong, Taehee Jung, Sung Ju Hwang, Joo-Kyung Kim, Dongyeop Kang

Infinity-Parser: Layout-Aware Reinforcement Learning with High-quality Document Parsing Dataset (2026.findings-acl)

Challenge:	Recent Long-Context Language Models (LCLMs) do not capture how evidence should be connected . a new framework that integrates thought templates into LCLM frameworks is proving useful .
Approach:	They propose a framework that iteratively refines reusable reasoning patterns derived from prior problem solving to improve their templates.
Outcome:	The proposed framework outperforms baselines on knowledge-intensive multi-hop reasoning benchmarks and practical scenarios without retrieval.

Copied to clipboard

Baode Wang, Biao Wu, Weizhen Li, Meng Fang, Zuming Huang, Jun Huang, Yanjie Liang, Haozhe Wang, Ling Chen, Wei Chu, Yuan Qi

Lunar-Bench: Towards Evaluating Task-Oriented Reasoning of LLMs in Lunar Exploration Scenarios (2026.findings-acl)

Challenge:	Existing supervised fine-tuning methods struggle to generalize across document types, leading to poor performance.
Approach:	They propose layoutRL, a reinforcement learning framework that optimizes layout understanding through composite rewards integrating normalized edit distance, paragraph count accuracy, and reading order preservation.
Outcome:	The proposed model outperforms specialized document parsing systems and general-purpose vision-language models on a broad range of document types, languages, and structural complexities.

Copied to clipboard

Xin-Yu Xiao, Ye Tian, Erwei Yin, Zhixian He, Shiqi Wang, Yalei Liu, Qianchen Xia

BV-Blend: Uncertainty-Weighted Historical Baselines for Stable Critic-Free RL with Verifiable Rewards (2026.findings-acl)

Challenge:	Existing benchmarks focus on static, context-independent reasoning tasks and fail to capture constraints and dependencies of lunar missions.
Approach:	They propose a benchmark to assess the task-oriented reasoning and decision-making performance of large language models through 3,000 tasks derived from mission procedures and documentation.
Outcome:	The proposed model achieves 47.8% accuracy compared with 65.1% for human experts on 36 representative missions.

Copied to clipboard

Yupeng Chang, Yuan Wu, Yi Chang

Rethinking Research on Stereotypes: An Analysis through Social Psychological and Computational Perspectives (2026.findings-acl)

Challenge:	Critic-free reinforcement learning with verifiable rewards (RLVR) is a practical paradigm for aligning Large Language Models.
Approach:	They propose a framework that stabilizes advantage estimation by combining prompt-local on-policy statistics with semantic-cluster-conditioned historical moments.
Outcome:	Experiments show that RLVR improves training stability and performance compared to critic-based methods . compared with other approaches, RL VR improves in cold-start regimes with binary verifiers .

Copied to clipboard

Kaustubh Shivshankar Shejole, Pushpak Bhattacharyya

Do LLMs Catch Their Own Mistakes? A Comprehensive Benchmark for Reflective Tool Use LLMs (2026.findings-acl)

Challenge:	Existing research on stereotypical biases ignores literature on them and results in resource wastage.
Approach:	They argue that stereotypes are social constructs shaping human perception and behavior that can produce harmful outcomes under specific conditions.
Outcome:	The proposed models can inherit and amplify stereotypes under certain conditions.

Copied to clipboard

Zheyuan Liu, Liqiang Xiao, Yang Li, Hyokun Yun, Lihong Li, Chao Zhang, Meng Jiang

Understanding Generalization in Role-Playing Models via Information Theory (2026.findings-acl)

Challenge:	Existing benchmarks primarily evaluate planning and execution success, overlooking the self-reflective dimension of tool use.
Approach:	They propose a benchmark to assess LLMs’ self-reflective reasoning in tool-augmented multi-turn dialogues.
Outcome:	The proposed benchmark covers 10 domains with 88 distinct APIs and 968 annotated dialogues, systematically injecting diverse error types arising from both user and assistant behavior.

Copied to clipboard

Yongqi Li, Hao Lang, Fei Huang, Tieyun Qian, Yongbin Li

A Multilingual Dataset and Empirical Validation for the Mutual Reinforcement Effect in Information Extraction (2026.findings-acl)

Challenge:	Existing methods to characterize RPM generalization are inadequate . existing methods do not provide a fine-grained diagnosis of distribution shifts .
Approach:	They propose a reasoning-based effective mutual information difference (R-EMID) to measure RPM performance degradation in an interpretable way.
Outcome:	The proposed model predicts the worst-case generalization performance of RPMs and reveals how shifts contribute to degradation.

Copied to clipboard

Chengguang Gan, Sunbowen Lee, Qingyu Yin, Yunhao Liang, Xinyang He, Hanjun Wei, Younghun Lim, Shijian Wang, Hexiang Huang, QingHao Zhang, Shiwen Ni, Tatsunori Mori

TRACE: An Experiential Framework for Coherent Multi-hop Knowledge Graph Question Answering (2026.findings-acl)

Challenge:	Existing work on the Mutual Reinforcement Effect in information extraction has not been empirically validated . 76 percent of the 21 sub-datasets exhibit the Mutual Reforcement effect across languages .
Approach:	They propose a multilingual MRE mix dataset that integrates 21 sub-datasets covering English, Japanese, and Chinese.
Outcome:	The proposed framework reduces manual annotation effort while preserving structural requirements of MRE tasks.

Copied to clipboard

Yingxu Wang, Jiaxin Huang, Mengzhu Wang, Nan Yin

Can LLMs See Without Pixels? Benchmarking Spatial Intelligence from Textual Descriptions (2026.findings-acl)

Challenge:	Existing methods for multihop Knowledge Graph Question Answering (KGQA) treat each reasoning step independently and fail to leverage experience from prior explorations, leading to fragmented reasoning and redundant exploration.
Approach:	They propose a framework that unifies LLM-driven contextual reasoning with exploration prior integration to enhance coherence and robustness of multihop KGQA.
Outcome:	Extensive experiments on multiple KGQA benchmarks show that TRACE outperforms state-of-the-art methods.

Copied to clipboard

Zhongbin Guo, Zhen Yang, Yushan Li, Xinyue Zhang, Wenyu Gao, Jiacheng Wang, Chengzhi Li, Xiangrui Liu, Ping Jian

Lil: Less is Less When Applying Post-Training Sparse-Attention Algorithms in Long-Decode Stage (2026.findings-acl)

Challenge:	Existing advances in Spatial Intelligence rely on vision-Language Models . however, a critical question remains: does spatial understanding originate from visual encoders?
Approach:	They propose to evaluate the SI performance of Large Language Models without pixel-level input.
Outcome:	The proposed benchmark challenges large language models to perform symbolic reasoning rather than visual pattern matching.

Copied to clipboard

Junhao Hu, Fangze Li, Mingtao Xu, Feifan Meng, Shiju Zhao, Tiancheng Hu, Ting Peng, Anmin Liu, Wenrui Huang, Chenxu Liu, Ziyue Hua, Tao Xie

Challenge:	Prior work typically decomposes inference into prefill and decode stages, with the decode stage dominating total latency.
Approach:	They propose an algorithm that detects threshold where information loss exceeds information gain during sparse decoding to reduce token consumption by up to 90% and a marginal accuracy degradation of less than 2%.
Outcome:	The proposed algorithm reduces token consumption by 90% with a marginal accuracy degradation of less than 2% across reasoning-intensive benchmarks.

Static Models, Dynamic World: A Unified Perspective on Temporal Perception in Large Language Models (2026.findings-acl)

Copied to clipboard

Chenhao Li, Dandan Song, Changzhi Zhou, Jun Yang, Yuhang Tian, Huipeng Ma, Guangyuan Feng, Luan Zhang, Xudong Li, Ke Duan

Decomposed Trust: Privacy, Adversarial Robustness, Ethics, and Fairness in Low-Rank LLMs (2026.findings-acl)

Challenge:	Large language models are trained on static corpora but deployed in a dynamic world . a foundational tension remains between time and the ability to understand it .
Approach:	They formalize temporal queries in an information-theoretic framework based on parametric reachability of temporal premises and answers.
Outcome:	The proposed framework formalizes temporal queries in an information-theoretic framework based on parametric reachability of temporal premises and answers . the framework induces four temporal information regimes corresponding to internal reasoning, answer recency, premise anchoring, and genuine world indeterminacy .

Copied to clipboard

Daniel Agyei Asante, Md Mokarram Chowdhury, Yang Li

Beyond Token Length: Step Pruner for Efficient and Accurate Reasoning in Large Language Models (2026.findings-acl)

Challenge:	Large language models (LLMs) have driven major advances across domains, yet their massive size hinders deployment in resource-constrained settings.
Approach:	They propose to compress large language models to reduce computation and memory consumption while maintaining accuracy.
Outcome:	The proposed algorithms preserve training data privacy but weaken the protection of personally identifiable information during conversations.

Copied to clipboard

Canhui Wu, Qiong Cao, Chang Li, Zhenfang Wang, Chao Xue, Yuwei Fan, Wei Xi, Xiaodong He

Exploring Reasoning Reward Model for Agents (2026.findings-acl)

Challenge:	Existing reinforcement learning methods for large reasoning models suffer from excessive verbosity, known as "overthinking." Existing models penalize generated tokens to promote conciseness, but these methods encounter two challenges: they may develop hacking behavior in later stages of training by discarding reasoning steps.
Approach:	They propose a framework that steers large reasoning models toward more efficient reasoning . they prioritize correctness while imposing penalties for redundant steps .
Outcome:	The proposed framework reduces token usage by 69.7% on AIME24.

Copied to clipboard

Kaixuan Fan, Kaituo Feng, Manyuan Zhang, Tianshuo Peng, Zhixun Li, Yilei Jiang, Shuang Chen, Xiangyu Yue

MTMCS-Bench: Evaluating Contextual Safety of Multimodal Large Language Models in Multi-Turn Dialogues (2026.findings-acl)

Challenge:	Existing methods for agentic reinforcement learning rely on sparse outcome-based reward for training, leading to suboptimal results.
Approach:	They propose an agent-based reward model that produces structured feedback for agentic trajectories, including an explicit reasoning trace and a focused critique.
Outcome:	The proposed model produces structured feedback for agentic trajectories including an explicit reasoning trace, a focused critique, and an overall score that evaluates process performance.

Copied to clipboard

Zheyuan Liu, Dongwhi Kim, Yixin Wan, Xiangchi Yuan, Zhaoxuan Tan, Fengran Mo, Meng Jiang

REAP: Towards Effective Training-Free Chemical Reasoning with Explicit Atomic Priors (2026.findings-acl)

Challenge:	Existing contextual safety benchmarks are mostly single-turn and miss how malicious intent can emerge gradually or how the same scene can support both benign and exploitative goals.
Approach:	They propose a benchmark that evaluates contextual safety in multimodal large language models . they observe persistent trade-offs between contextual safety and utility .
Outcome:	The proposed model combines multi-turn and multi-switch scenarios to evaluate safety in multimodal large language models.

Copied to clipboard

Mingxu Zhang, Dazhong Shen, Qi Zhang, Ying Sun

Towards Self-Improving Error Diagnosis in Multi-Agent Systems (2026.findings-acl)

Challenge:	Current approaches to instill explicit priors into LLMs often suffer from an information bottleneck .
Approach:	They propose a training-free framework that equips LLMs with an external knowledge base, enabling them to reason over retrieved chemical priors dynamically.
Outcome:	Experiments show that REAP outperforms current reasoning methods and rivals state-of-the-art training-based models.

Copied to clipboard

Jiazheng Li, Emine Yilmaz, Bei Chen, Thu Le

MulDimIF: A Multi-Dimensional Constraint Framework for Evaluating and Improving Instruction Following in Large Language Models (2026.findings-acl)

Challenge:	Existing diagnostic approaches rely on expensive expert annotations and ”LLM-as-a-judge” paradigms.
Approach:	They propose a framework for semantic failure attribution that identifies responsible agents and the originating error step.
Outcome:	The proposed framework outperforms baselines in step-level localization and validation.

Copied to clipboard

Junjie Ye, Caishuang Huang, Zhuohan Chen, Wenjie Fu, Chenyuan Yang, Leyi Yang, Yilong Wu, Peng Wang, Meng Zhou, Xiaolong Yang, Tao Gui, Qi Zhang, Zhongchao Shi, Jianping Fan, Xuanjing Huang

Text2Mem: A Unified Memory Operation Language for Memory Operating System (2026.findings-acl)

Challenge:	Existing research has focused on constraint categories, offering little guidance for improving instruction following abilities.
Approach:	They propose a multi-dimensional constraint framework that allows for instruction following . they construct 9,106 code-verifiable samples and evaluate 18 LLMs .
Outcome:	The proposed framework improves instruction following performance without compromising general performance.

Copied to clipboard

Leo Wang, Lihai Yang, Boyu Chen, Kerun Xu, Gongyi Zou, Bo Tang, Feiyu Xiong, Siheng Chen, Zhiyu li

Challenge:	Existing memory frameworks lack a formal, executable specification for memory control.
Approach:	They propose a unified memory operation language that standardizes translation of natural-language instructions into reliable execution.
Outcome:	The proposed language standardizes translation of natural-language instructions into reliable execution.

AMO-Bench: Large Language Models Still Struggle in High School Math Competitions (2026.findings-acl)

Copied to clipboard

Junlin Liu, Shengnan An, Shuang Zhou, Dan Ma, Yehao Lin, Xinxuan Lv, Xuanlin Wang, Xiaoyu Li, Ziwen Wang, Xuezhi Cao, Xunliang Cai

Solver-Independent Automated Problem Formulation via LLMs for High-Cost Simulation-Driven Design (2026.findings-acl)

Challenge:	Existing benchmarks for mathematical reasoning are becoming less effective due to performance saturation.
Approach:	They propose to use a mathematical reasoning benchmark with Olympiad difficulty to evaluate top-tier LLMs.
Outcome:	The proposed benchmarks are cross-validated by experts to meet IMO difficulty standards and entirely original problems to prevent performance leakages from data memorization.

Copied to clipboard

Yuchen Li, Handing Wang, Bing Xue, Mengjie Zhang, Yaochu Jin

When Facts Change: Temporal Knowledge Conflict Resolution in LLMs (2026.findings-acl)

Challenge:	Existing approaches to translating ambiguous design requirements into a mathematical optimization formulation are expensive and time-consuming.
Approach:	They propose a solver-independent framework that converts engineers’ natural language requirements into executable optimization models.
Outcome:	The proposed framework outperforms existing methods in the accuracy of requirement formalization and quality of resulting radiation efficiency curves on antenna design.

Copied to clipboard

Jonas Wallat, Wolfgang Nejdl, Sandipan Sikdar

SCRIPT: A Subcharacter Compositional Representation Injection Module for Korean Pre-Trained Language Models (2026.findings-acl)

Challenge:	Large language models are increasingly used in retrieval-augmented generation systems to reconcile knowledge conflicts between parametric memory and contextual inputs.
Approach:	They propose to use mutability to resolve temporal misalignment in large language models to compare stable and recently updated facts from Wikidata to determine if mutable models can serve as a mediating signal in this process.
Outcome:	The proposed model can produce reasoning for facts that actually changed but rarely for stable ones, whereas smaller models rarely detect conflict, while larger models detect it but fail to act on mutability judgments.

Copied to clipboard

SungHo Kim, Juhyeong Park, Eda Atalay, SangKeun Lee

PibE-MPP: A Play-it-by-Ear Masking Performance Plug-in for LLMs (2026.findings-acl)

Challenge:	Korean uses a featural writing system in which each character is composed of subcharacter units known as Jamo.
Approach:	They propose a model-agnostic module that injects subcharacter compositional knowledge into Korean language models.
Outcome:	a new module improves embeddings of Korean subwords with structural granularity . the module improve grammatical regularities and semantic cohesive variations .

Copied to clipboard

Mengwei Wang, Simin Niu, Xun Liang, Yuefeng Ma, Sensen Zhang, Jiawei Yang, Shichao Song, Hanyu Wang, Huayi Lai

RoleCDE: Benchmarking and Mitigating Role–Alignment Trade-offs in Role-Playing Agents (2026.findings-acl)

Challenge:	Random masking is a widely adopted classic baseline in large language models (LLMs).
Approach:	They propose a play-it-by-ear masking performance plug-in which enables LLMs to adaptively select masking target combinations for each task.
Outcome:	The proposed performance plug-in retains the advantages and mitigates the drawbacks of random masking in large language models.

Copied to clipboard

Huayi Lai, Shichao Song, Simin Niu, Hanyu Wang, Jiawei Yang, Zhouxing Wang, Zhiqiang Yin, Xun Liang

PolicyLLM: Towards Excellent Comprehension of Public Policy for Large Language Models (2026.findings-acl)

Challenge:	Existing benchmarks for role-playing agents only evaluate surface-level fidelity and provide limited insight into decision making under role–alignment value conflicts.
Approach:	They propose a benchmark to evaluate RPAs under role–alignment value conflicts . they use 8k diverse role profiles and 240k dilemma instances to evaluate role-aware decision making .
Outcome:	The proposed benchmark covers 8k diverse role profiles and scenarios and nearly 240k dilemma instances across three difficulty levels and eight role categories.

Copied to clipboard

Han Bao, Penghao Zhang, Yue Huang, Zhengqing Yuan, Yanchi Ru, SU Rui, Yujun Zhou, Xiangqi Wang, Kehan Guo, Nitesh V Chawla, Yanfang Ye, Xiangliang Zhang

BOOKAGENT: Orchestrating Safety-Aware Visual Narratives via Multi-Agent Cognitive Calibration (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are increasingly integrated into real-world decision-making, but their ability to comprehend and reason about policy-related content remains underexplored.
Approach:	They propose a bilingual benchmark evaluating policy comprehension comprising 21K cases across a broad spectrum of policy areas.
Outcome:	The proposed model shows stronger performance on application-oriented policy tasks than on memorization or conceptual understanding, and yields the highest accuracy on structured reasoning tasks.

Copied to clipboard

Bo Gao, Chang Liu, Yuyang Miao, SiYuan Ma, Ser-Nam Lim

Feedback-Driven Tool-Use Improvements in Large Language Models via Automated Build Environments (2026.findings-acl)

Challenge:	Existing work on illustrated storybooks decomposes this task into separate stages, limiting multi-modal grounding . et al. proposes a safety-aware multi-agent collaboration framework for high-quality, safety-conscious visual narratives .
Approach:	They propose a safety-aware multi-agent collaboration framework for illustrated storybooks . the framework jointly plans, scripts, illustrates, and globally corrects inconsistencies .
Outcome:	a novel framework outperforms existing methods in safety and coherence, and improves visual consistency . the framework is available on github at https://github.com/bogao-code/BookAgent/main .

Copied to clipboard

Junjie Ye, Changhao Jiang, Zhengyin Du, Yufei Xu, Xuesong Yao, Zhiheng Xi, Xiaoran Fan, Qi Zhang, Tao Gui, Xuanjing Huang, Jiecao Chen

Do Domain-specific Experts exist in MoE-based LLMs? (2026.findings-acl)

Challenge:	Currently, there are no efficient reinforcement learning (RL) frameworks specifically designed for tool use.
Approach:	They propose an automated environment construction pipeline that incorporates scenario decomposition, document generation, function integration, complexity scaling, and localized deployment to enable high-quality training environments without external tools.
Outcome:	The proposed framework significantly improves the models’ tool-use performance without degrading their general capabilities.

Copied to clipboard

Giang Do, Hung Le, Truyen Tran

AutoUE: Automated Generation of 3D Games in Unreal Engine via Multi-Agent Systems (2026.findings-acl)

Challenge:	Existing studies on domain-specific experts in Large Language Models (LLMs) are still lacking.
Approach:	They propose a training-free framework that introduces zero additional inference cost and outperforms well-trained MoE-based LLMs.
Outcome:	The proposed framework outperforms well-trained MoE-based LLMs and strong baselines across target and non-target domains.

Copied to clipboard

Lei Yin, Wentao Cheng, Zhida Qin, Tianyu Huang, Yidong Li, Gangyi Ding

Comprehensive Benchmarking of Long-Form Speech Generation in Diverse Scenarios (2026.findings-acl)

Challenge:	Recent advances in Large Language Models (LLMs) and generative models have motivated studies on automated game generation from natural language descriptions.
Approach:	They propose a novel multi-agent system, AutoUE, which coordinates multiple agents to end-to-end generate 3D games, covering model retrieval, scene generation, gameplay and interaction code synthesis, and automated game testing for evaluation.
Outcome:	The proposed system covers model retrieval, scene generation, gameplay and interaction code synthesis, and automated game testing for evaluation.

Copied to clipboard

Changhao Pan, Rui Yang, Han Wang, Zhuan Zhou, Xuming He, Wenxiang Guo, Ziyue Jiang, Ruiqi Li, Yu Zhang, Chenyuhao Wen, Ke Lei, Xiang Yin, Jingyu Lu, Zhiyuan Zhu, Zhou Zhao

CAST: Achieving Stable LLM-based Text Analysis for Data Analytics (2026.findings-acl)

Challenge:	Existing evaluation benchmarks for long-form speech are limited to limited domains, creating a significant gap with the diverse downstream applications.
Approach:	They propose a benchmark that decomposes "long-form speech quality" into specific, disentangled dimensions.
Outcome:	The proposed benchmark decomposes “long-form speech quality” into specific, disentangled dimensions.

Copied to clipboard

Jinxiang Xie, Zihao Li, Wei He, Rui Ding, Shi Han, Dongmei Zhang

WavAlign: Enhancing Intelligence and Expressiveness in Spoken Dialogue Models via Adaptive Hybrid Post-Training (2026.findings-acl)

Challenge:	Text analysis of tabular data relies on two core operations: summarization for corpus-level theme extraction and tagging for row-level labeling.
Approach:	They propose a framework that enhances output stability by constraining the model’s latent reasoning trajectory.
Outcome:	The proposed framework improves stability by constraining the model's latent reasoning trajectory.

Copied to clipboard

Yifu Chen, Shengpeng Ji, Qian Chen, Tianle Liang, Yangzhuo Li, Ziqing Wang, Wen Wang, Jingyu Lu, Haoxiao Wang, Xueyi Pu, Fan Zhuo, Zhou Zhao

OmniCompliance-100K: A Multi-Domain, Rule-Grounded, Real-World Safety Compliance Dataset (2026.findings-acl)

Challenge:	End-to-end spoken dialogue models have higher potential ceiling in expressiveness and perceptual ability than cascaded systems.
Approach:	They propose a modality-aware adaptive post-training recipe that constrains preference updates to the semantic channel and improves acoustic behavior via explicit anchoring.
Outcome:	The proposed model improves speech quality and expressiveness across spoken dialogue benchmarks and architectures.

Copied to clipboard

Wenbin Hu, Huihao Jing, Haochen Shi, Changxuan Fan, Haoran Li, Yangqiu Song

AutoMonitor-Bench: Evaluating the Reliability of LLM-Based Misbehavior Monitor (2026.findings-acl)

Challenge:	Existing LLM safety datasets rely on ad-hoc taxonomies and lack rule-grounded, real-world cases.
Approach:	They construct a rule-grounded, real-world case dataset OmniCompliance-100K from a compliance perspective using a powerful web-searching agent.
Outcome:	The proposed dataset spans 74 regulations and policies across a wide range of domains including security and privacy regulations, content safety and user data privacy policies, financial security requirements, medical device risk management standards, educational integrity guidelines, and protections of fundamental human rights.

Copied to clipboard

Shu Yang, Jingyu Hu, Tong Li, Hanqi Yan, Wenxuan Wang, Di Wang

Challenge:	AutoMonitor-Bench evaluates the reliability of LLM-based misbehavior monitors across diverse tasks and failure modes.
Approach:	They introduce AutoMonitor-Bench, a benchmark designed to evaluate misbehavior monitors across diverse tasks and failure modes.
Outcome:	The new benchmark evaluates the reliability of LLM-based misbehavior monitors across tasks and failure modes.

Adaptive and Representative Multi-Interest Modeling for Recommendation with Large Language Model (2026.findings-acl)

Copied to clipboard

Ziyan Wang, Yingpeng Du, Tianjun Wei, Haoyan Chua, Jieyi Bi, Jie Zhang, Zhu Sun

MathMixup: Boosting LLM Mathematical Reasoning with Difficulty-Controllable Data Synthesis and Curriculum Learning (2026.findings-acl)

Challenge:	Existing methods for multi-interest analysis of users rely on heuristic assumptions . however, the granularity of raw generation of LLMs is agnostic, leading to overly fine or coarse interest grouping.
Approach:	They propose an LLM-driven adaptive and representative multi-interest modeling framework that exploits the agnostic granularity of LLMs for multi-interest analysis.
Outcome:	The proposed model outperforms baselines on real-world datasets.

Copied to clipboard

Xuchen Li, Jing Chen, Xuzhao Li, Hao Liang, Xiaohuan Zhou, Taifeng Wang, Wentao Zhang

The Model Agreed, But Didn’t Learn: Diagnosing Surface Compliance in Large Language Models (2026.findings-acl)

Challenge:	Existing data synthesis methods suffer from limited diversity and lack precise control over problem difficulty, making them insufficient for efficient training paradigms such as curriculum learning.
Approach:	They propose a data synthesis paradigm that generates high-quality, difficulty-controllable mathematical reasoning problems through hybrid and decomposed strategies.
Outcome:	The proposed paradigm outperforms existing methods and improves mathematical reasoning abilities.

Copied to clipboard

Xiaojie Gu, Ziying Huang, Weicong Hong, Jian Xie, Renze Lou, Kai Zhang

AWARE: Agentic Knowledge Warehousing for Contextual Intelligence (2026.findings-acl)

Challenge:	Large Language Models internalize vast world knowledge as parametric memory, yet inherit the staleness and errors of their source corpora.
Approach:	They propose a framework that subjects models to discriminative self-assessment under diverse contextual pressures to scrutinize subtle behavioral nuances induced by memory modifications.
Outcome:	The proposed framework achieves high benchmarks without overwriting internal beliefs, while recursive modifications accumulate representational residues, triggering cognitive instability and permanently diminishing the reversibility of the model’s memory state.

Copied to clipboard

Hongjin Qian, Siqi Bao, Zhao Cao, Zheng Liu

Alignment Tuning for Large Language Models: A Data-Centric Lens on Alignment Data Pipelines (2026.findings-acl)

Challenge:	Large language models excel in information seeking tasks, but their knowledge is limited in coverage and timeliness.
Approach:	They propose an agentic knowledge warehousing framework that transforms unstructured data into minimal, task-conditioned knowledge representations consumable by LLMs.
Outcome:	Experiments on GAIA, WebWalker, and BrowseComp-Plus show improvements over baselines.

Copied to clipboard

Hwanjun Song

Purging the Gray Zone: Latent-Geometric Denoising for Precise Knowledge Boundary Awareness (2026.findings-acl)

Challenge:	a large number of alignment tuning literature is organized around optimization objectives, but the construction of alignment data is often treated implicitly.
Approach:	They propose to decompose alignment data construction into three interacting stages . they identify recurring design trade-offs and failure modes observed across prior alignment methods .
Outcome:	The proposed framework identifies recurring design trade-offs and failure modes observed across prior alignment methods and outlines open challenges for alignment data pipelines including prompt-level alignment, agentic settings, and alignment under evolving objectives.

Copied to clipboard

Hao An, Yibin Lou, Jiayi Guo, Yang Xu

Temporal Token Matters: Investigating and Interpreting the Consistency of Temporal Ordering in Large Language Models (2026.findings-acl)

Challenge:	Existing abstention fine-tuning methods cause models to suffer from label noise near the decision boundaries.
Approach:	They propose a latent space representation perspective for abstention fine-tuning . they propose 'geometric denoising' framework that constructs a truth hyperplane .
Outcome:	The proposed framework significantly enhances model truthfulness and demonstrates strong generalization in out-of-distribution scenarios.

Copied to clipboard

Zhen Yang, Xinyue Zhang, Ping Jian, Chengzhi Li, Zhongbin Guo, Jiaping Feng, Wenpeng Lu

Towards LLM Agents for Earth Observation (2026.findings-acl)

Challenge:	Large Language Models (LLMs) exhibit notable deficiencies in temporal reasoning . phrasing changes can lead LLMs to produce inconsistent outputs .
Approach:	They investigate the mechanistic interpretability of temporal ordering within event temporal reasoning . they identify a sparse subset of attention heads that are causally responsible for reasoning outcomes .
Outcome:	The proposed model outperforms other models in a variety of tasks and is validated by intervention-based experiments.

Copied to clipboard

Chia Hsiang Kao, Wenting Zhao, Cheryl Lam, Aarush Umap, Shreelekha Revankar, Samuel Speas, Snehal Bhagat, Rajeev Datta, Cheng Perng Phoo, Utkarsh Mall, Carl Vondrick, Kavita Bala, Bharath Hariharan

Challenge:	specialized automated systems for specific earth observation tasks lack flexibility for general-purpose, customized queries.
Approach:	They propose a coding benchmark of 408 yes/no questions from NASA Earth Observatory articles . they analyze the impact of using JavaScript API versus Python and the effect of providing documentation .
Outcome:	The proposed frameworks reduce errors by 60%, but are only marginally above random chance.

Jamendo-MT-QA: A Benchmark for Multi-Track Comparative Music Question Answering (2026.findings-acl)

Copied to clipboard

Junyoung Koh, Jaeyun Lee, Soo Yong Kim, Gyu Hyeong Choi, Jung In Koh, Jordan Phillips, Yeonjin Lee, Min Song

Chinese Live-Streaming E-Commerce Morph Resolution: Datasets and Methods (2026.findings-acl)

Challenge:	Existing benchmarks for music question answering do not systematically evaluate reasoning across tracks.
Approach:	They propose a dataset and benchmark for multi-track comparative question answering . they construct 36,519 comparative QA items over 12,173 track pairs .
Outcome:	The proposed dataset and benchmark for multi-track comparative question answering is based on the Jamendo-QA dataset.

Copied to clipboard

Jipeng Qiang, Jiahao Zhu, Yi Zhu, Chaowei Zhang

MemoBrain: Executive Memory as an Agentic Brain for Reasoning (2026.findings-acl)

Challenge:	Live-stream E-commerce faces significant challenges from morphs, deliberate linguistic variants used to evade real-time voice filters and amplify product claims illegally.
Approach:	They propose a framework that resolves morphs and generates structured explanations . they propose morph-aware dual-output refinement framework that detects inconsistencies .
Outcome:	The proposed framework improves morph resolution accuracy and interpretability.

Copied to clipboard

Hongjin Qian, Zhao Cao, Zheng Liu

Exploring the Capability Boundaries of LLMs in Mastering of Chinese Chouxiang Language (2026.findings-acl)

Challenge:	Large language models (LLMs) are inherently long-horizon, causing reasoning traces and tool artifacts to accumulate and strain the working context of large language models.
Approach:	They propose a model that constructs a dependency-aware memory over reasoning steps and captures salient intermediate states and their logical relations.
Outcome:	The proposed model prunes invalid steps, folds completed sub-trajectories, and preserves a compact, high-salience reasoning backbone under a fixed context budget.

Copied to clipboard

Dianqing Lin, Tian Lan, Jiali Zhu, Jiang Li, Wei Chen, Xu Liu, null Aruukhan, Xiangdong Su, Hongxu Hou, Guanglai Gao

RECOR: Reasoning-focused Multi-turn Conversational Retrieval Benchmark (2026.findings-acl)

Challenge:	Current state-of-the-art LLMs exhibit clear limitations on multiple tasks, while performing well on tasks that involve contextual semantic understanding.
Approach:	They propose a mouse-based benchmark to evaluate LLMs' performance on NLP tasks involving Chouxiang Language.
Outcome:	The proposed benchmark evaluates the performance of LLMs on six NLP tasks involving Chouxiang Language.

Copied to clipboard

Mohammed Ali, Abdelrahman Abdallah, Amit Agarwal, Hitesh Laxmichand Patel, Adam Jatowt

Data Selection for Multi-turn Dialogue Instruction Tuning (2026.findings-acl)

Challenge:	Existing benchmarks treat multi-turn conversation and reasoning-intensive retrieval separately, yet real-world information seeking requires both.
Approach:	They propose a framework that transforms complex queries into fact-grounded multi-turn dialogues through multi-level validation.
Outcome:	The proposed framework outperforms existing systems in a number of domains and can be used to improve multi-turn conversation retrieval.

Copied to clipboard

Bo Li, Shikun Zhang, Wei Ye

AgentGC: Evolutionary Learning-based Lossless Compression for Genomics Data with LLM-driven Multiple Agent (2026.findings-acl)

Challenge:	Instruction-tuned language models often use noisy multi-turn dialogue datasets with topic drift, repetitive chitchat, and mismatched answer formats across turns.
Approach:	They propose a dialogue-level framework that scores whole conversations rather than isolated turns.
Outcome:	The proposed framework outperforms strong single-turn selectors, dialogue-level LLM scorers and heuristic baselines on three multi-turn benchmarks and an in-domain Banking test set.

Copied to clipboard

Sun Hui, Ding Yanfeng, Huidong Ma, Chang Xu, Keyan Jin, Lizheng Zu, Cheng Zhong, Xiaoguang Liu, Gang Wang, Wentong Cai

Cognitive Modeling

Challenge:	Lossless compression has made significant advancements in Genomics Data storage, sharing and management.
Approach:	They propose a novel agent-based GD Compressor with 3 layers with a multi-agent named Leader and Worker.
Outcome:	The proposed method improves on existing methods with low-level modeling and limited adaptability and user-unfriendly interface.

Think Outside the Policy: In-Context Steered Policy Optimization (2026.findings-acl)

Copied to clipboard

Hsiu-Yuan Huang, Chenming Tang, Weijie Liu, Clive Bai, Saiyong Yang, Yunfang Wu

Where CoT Reasoning Commits: Entropy Traces Identify Interpretable Attention Heads (2026.findings-acl)

Challenge:	Existing Reinforcement Learning from Verifiable Rewards (RLVR) methods exhibit limited exploration due to reliance on on-policy rollouts which are limited to the current policy’s distribution, resulting in narrow trajectory diversity.
Approach:	They propose a framework that leverages the in-context learning capability of Large Reasoning Models to provide expert guidance using existing datasets.
Outcome:	The proposed framework improves RLVR performance and training stability on mathematical reasoning benchmarks.

Copied to clipboard

Tianhe Zhang, Yonghong Deng, Ping Jian, Zhen Yang, Boyang Wang, Xinyue Zhang

From Coarse to Fine: Benchmarking and Reward Modeling for Writing-Centric Generation Tasks (2026.findings-acl)

Challenge:	a growing body of work suggests a disconnect between the generated rationale and the model's actual choice.
Approach:	They propose a mechanism-aware framework that interprets the evolving "choice state" of attention heads during CoT generation . they identify a set of intervention targets and perform Selective Head Fine-Tuning .
Outcome:	The proposed framework interprets the "choice state" of attention heads during CoT generation . it detects two functional behaviors: Steadfast Heads and Wavering Heads .

Copied to clipboard

Qingyu Ren, Tianjun Pan, Xingzhou Chen, Xuhong Wang

DentalGPT: Incentivizing Multimodal Reasoning in Dentistry (2026.findings-acl)

Challenge:	Existing evaluation benchmarks for writing reward models are coarse-grained.
Approach:	They propose a benchmark and a fine-grained training framework to evaluate writing reward models.
Outcome:	The proposed model improves on various writing benchmarks and exhibits strong generalization.

Copied to clipboard

Zhenyang Cai, Jiaming Zhang, Junjie Zhao, Ziyi Zeng, Yanchao Li, Liang Jingyi, Junying Chen, Yunjin Yang, Jiajun You, Shuzhi Deng, null Xieruiqiii, Yuanting Chen, Xiangyi Feng, Jianquan Li, Liangyi Chen, Junwen Wang, Shan Jiang, Benyou Wang

SARA: Unlocking Multilingual Knowledge in Mixture-of-Experts via Semantically Anchored Routing Alignment (2026.findings-acl)

Challenge:	Current multimodal large language models (MLLMs) show limited understanding of dental images.
Approach:	They propose a dental-specialized multimodal large language model trained via staged multimodal alignment and reinforcement learning.
Outcome:	The proposed model outperforms state-of-the-art models on disease classification and dental VQA tasks.

Copied to clipboard

Tianyu Dong, Yangyang Liu, Jiang Zhou, Xinwei Wu, Xiaohu Zhao, Hao Wang, Heng Liu, Linlong Xu, Longyue Wang, Weihua Luo, Shaolin Zhu, Deyi Xiong

Eliciting Medical Reasoning with Knowledge-enhanced Data Synthesis: A Semi-Supervised Reinforcement Learning Approach (2026.findings-acl)

Challenge:	Low-resource language tokens are often routed to different experts than those activated by high-resourced inputs, which hinders their efficacy in multilingual contexts.
Approach:	They propose a framework to transfer specialized capabilities from high-resource languages as anchors to low-resourced languages by using a symmetric Jensen-Shannon constraint.
Outcome:	The proposed framework outperforms standard instruction tuning on 5 low-resource languages and 3 benchmarks.

Copied to clipboard

Haolin Li, Shuyang Jiang, Ruipeng Zhang, Jiangchao Yao, Ya Zhang, Yanfeng Wang

AlphaQT-Bench: Diagnosing the Gap between Financial Code Generation and Quantitative Reasoning in LLMs (2026.findings-acl)

Challenge:	Existing methods to enhance medical reasoning lack high-quality data.
Approach:	They propose a medical knowledge-enhanced data Synthesis and Semi-supervised Reinforcement learning framework that uses rare disease knowledge to synthesize distribution-controllable reasoning questions.
Outcome:	The proposed method outperforms existing methods across ten medical benchmarks and achieves up to 5.93% gain on rare diseases tasks.

Copied to clipboard

Sichun Luo, Yi Huang, Shichang Meng, Fengyuan Liu, Mukai Li, Qinghua Yao, Zefa Hu, Junlan Feng, Qi Liu

Fine-tuning with Hierarchical Prompting for Robust Propaganda Classification Across Annotation Schemas (2026.findings-acl)

Challenge:	Existing benchmarks rely on outcome-driven metrics such as profitability and look-ahead bias.
Approach:	They propose a diagnostic benchmark for instruction-grounded financial code generation under strict semantic and temporal constraints.
Outcome:	The proposed benchmarks show that the models fail under causal, structural, or functional constraints.

Copied to clipboard

Lukas Stähelin, Veronika Solopova, Max Upravitelev, David Kaplan, Premtim Sahitaj, Ariana Sahitaj, Charlott Jakob, Sebastian Möller, Vera Schmitt

ComfyFlow: Benchmarking LLMs for AIGC Workflow Generation (2026.findings-acl)

Challenge:	Propaganda detection in social media is challenging due to noisy, short texts and low annotation agreements.
Approach:	They propose a new intent-focused taxonomy of propaganda techniques and compare it against an established, higher-agreement schema.
Outcome:	The proposed taxonomy outperforms existing models and reveals methodological differences hidden in base models.

Copied to clipboard

Zhenran Xu, Yiyu Wang, Yunxin li, Muyang Ye, null Yangxue, Kai Chen, Longyue Wang, Weihua Luo, Baotian Hu, Min Zhang

SEA-Guard: Culturally Grounded Multilingual Safeguard for Southeast Asia (2026.findings-acl)

Challenge:	Large language models (LLMs) have shown promising advances in tackling human-level tasks, but generating workflows for collaborative AI systems remains a critical and challenging step.
Approach:	They propose a benchmark to evaluate LLMs’ ability to generate executable and instruction-following AIGC workflows in ComfyUI.
Outcome:	The proposed benchmarks show that LLMs can generate executable and instruction-following AIGC workflows in ComfyUI.

Copied to clipboard

Panuthep Tasawong, Jian Gang Ngui, Alham Fikri Aji, Trevor Cohn, Peerat Limkonchotiwat

Prompt Injection in Automated Résumé Screening with Large Language Models: Single and Multi-Injection Settings (2026.findings-acl)

Challenge:	Existing safeguard models rely on translation of English datasets, missing regional and cultural nuances.
Approach:	They propose a framework to generate culturally grounded safety datasets for Southeast Asia . SEA-Guard family is the first multilingual safeguard model grounded in SEA cultural contexts .
Outcome:	The proposed model outperforms existing safeguard models in detecting regionally sensitive content while maintaining strong general safety performance.

Copied to clipboard

Preet Baxi, Jiannan Xu, Jane Yi Jiang, Stefanus Jasin

FuseSearch: Learning Adaptive Parallel Execution for Efficient Code Localization (2026.findings-acl)

Challenge:	Large language models (LLMs) are increasingly used to screen and rank job applicants . a recent study shows that prompt injection can improve ranking when résumé quality is homogeneous .
Approach:	They study whether prompt injection distorts quality-based candidate selection . they find that it improves rankings when résumé quality is homogeneous .
Outcome:	The proposed model improves rankings when résumé quality is homogeneous and few candidates inject.

Copied to clipboard

Ke Xu, Siyang Xiao, Ming Liang, Yichen Yu, Zhixiang Wang, Jingxuan Xu, Dajun Chen, Wei Jiang, Yong Li

Revisiting Self-Play Preference Optimization: On the Role of Prompt Difficulty (2026.findings-acl)

Challenge:	Existing parallel code localization agents suffer from a 34.9% redundant tool invocation rate . specialized localization agent that operate as dedicated search components is needed to achieve high localization accuracy.
Approach:	They propose a parallel code localization system that reframes parallel code execution as a quality–efficiency co-optimization problem.
Outcome:	The proposed method matches SOTA performance while being 93.6% faster.

Copied to clipboard

Yao Xiao, Jung-jae Kim, Roy Ka-Wei Lee, Lidong Bing

Multi-Agent Comedy Club: Investigating Community Discussion Effects on LLM Humor Generation (2026.findings-acl)

Challenge:	incorporating difficult prompts into training fails to enhance overall performance, e.g., as prompt difficulty decreases.
Approach:	They investigate how prompts of varying difficulty influence self-play preference optimization . they use the reward of sampled responses of a prompt as a proxy for its difficulty .
Outcome:	The proposed model improves on difficult prompts and easy prompts, but fails to train on difficult ones and learns from failures.

Copied to clipboard

Shiwei Hong, Lingyao Li, Ethan Z. Rong, Chenxinran Shen, Zhicong Lu

ComfyUI-R1: Exploring Reasoning Models for Workflow Generation (2026.findings-acl)

Challenge:	Existing studies on the use of multi-turn interaction and feedback for LLM writing focus on prompts and localized feedback.
Approach:	They build a controlled multi-agent sandbox that instantiates a small standup comedy community and allows it to manipu-late whether public reception is generated, logged, and fed back into later rounds.
Outcome:	The proposed model improves craft/clarity and social response with occasional increases in aggressive humor.

Copied to clipboard

Zhenran Xu, Yiyu Wang, null Yangxue, Longyue Wang, Weihua Luo, Kaifu Zhang, Baotian Hu, Min Zhang

Hard to Be Heard: Phoneme-Level ASR Analysis of Phonologically Complex, Low-Resource Endangered Languages (2026.findings-acl)

Challenge:	ComfyUI-R1 is the first large reasoning model for automated workflow generation.
Approach:	They propose a large reasoning model for automated workflow generation that builds on curated knowledge bases and a two-stage framework to fine-tune models for cold start and reinforcement learning for incentivizing reasoning capability.
Outcome:	The proposed model achieves 97% format validity rate, high pass rate, node-level and graph-level F1 scores, surpassing prior state-of-the-art methods that employ leading closed-source models such as GPT-4o and Claude series.

Copied to clipboard

V.S.D.S.Mahesh Akavarapu, Michael Daniel, Gerhard Jäger

Beyond Surface-Level Pattern Trap: LLM Agents for Faster and Smarter Cross-Architecture Code Migration (2026.findings-acl)

Challenge:	a phoneme-level analysis of automatic speech recognition (ASR) is performed on two low-resource, typologically complex East Caucasian languages.
Approach:	They propose a phoneme-level analysis of automatic speech recognition for two East Caucasian languages, Archi and Rutul.
Outcome:	The proposed model improves on existing models and improves in low-resource settings.

Copied to clipboard

Weijia Li, KE Gao, Pengfei Chen, Jiajie Li, Xinyu Wang, Yiran Le, Yize Wu, Ling Li

PairCoder: Pair Programming-Inspired Two-Agent Collaboration for Code Generation (2026.findings-acl)

Challenge:	cross-architecture code migration is a resource-intensive and errorprone task.
Approach:	a framework for cross-architecture code migration is proposed to decouple implementation details through functional mining and code refactoring.
Outcome:	a new framework improves performance and correctness over state-of-the-art frameworks on OpenCV migration tasks.

Copied to clipboard

Junhao Chen, Xiang Li, Yibin Xu, Yuehan Cui, Fangsheng Weng, Hao Zhao, Fei Ma, Qi Tian

OptiVerse: A Comprehensive Benchmark towards Optimization Problem Solving (2026.findings-acl)

Challenge:	Existing multi agent frameworks for large language models are brittle on code generation tasks.
Approach:	They propose a framework that brings pair programming to autonomous LLM collaboration.
Outcome:	Using PairCoder, large language models achieve better results on code generation tasks and reduce token usage by 40% to 70% on eight representative backbones.

Copied to clipboard

Xinyu Zhang, Boxuan Zhang, Yuchen Wan, Lingling Zhang, YiXing Yao, Bifan Wei, Yaqiang Wu, Jun Liu

DiagnosisArena: Benchmarking Diagnostic Reasoning for Large Language Models (2026.findings-acl)

Challenge:	Existing benchmarks focus on Mathematical Programming and Combinatorial Optimization, hindering comprehensive evaluation.
Approach:	They propose a benchmarking tool that compares 1,000 curated optimization problems across three difficulty levels.
Outcome:	The proposed model improves performance on hard problems while maintaining 27% accuracy.

Copied to clipboard

Yakun Zhu, Zhongzhen Huang, Linjie Mu, Yutong Huang, Wei Nie, Jiaji Liu, Shaoting Zhang, Pengfei Liu, Xiaofan Zhang

AMA: Adaptive Memory via Multi-Agent Collaboration (2026.findings-acl)

Challenge:	Existing medical benchmarks for diagnostic reasoning are limited in their ability to perform complex tasks.
Approach:	They propose to benchmark diagnostic capabilities of large language models to assess their accuracy and generalization bottlenecks.
Outcome:	The proposed model achieves 45.82%, 31.09%, and 17.79% accuracy, compared to current models, o3-mini, e1 and DeepSeek-R1 .

Copied to clipboard

Weiquan Huang, Zixuan Wang, Hehai Lin, Sudong Wang, Bo Xu, Qian Li, Beier Zhu, Linyi Yang, Chengwei Qin

WESR: A Benchmark and Strong Baseline for Word-level Event-Speech Recognition (2026.findings-acl)

Challenge:	Existing approaches to longterm memory rely on rigid retrieval granularity, accumulation-heavy maintenance strategies, and coarse-grained update mechanisms.
Approach:	They propose a framework that leverages coordinated agents to manage memory across multiple granularities.
Outcome:	The proposed framework outperforms state-of-the-art benchmarks while reducing token consumption by approximately 80%.

Copied to clipboard

Chenchen Yang, Kexin Huang, Liwei Fan, Qian Tu, Botian Jiang, Dong Zhang, Linqi Yin, Shimin Li, Zhaoye Fei, Qinyuan Cheng, Xipeng Qiu

Applicability Condition Extraction for Therapeutic Drug-Disease Relations (2026.findings-acl)

Challenge:	aaron carroll: the precise localization of non-verbal vocal events remains a critical yet under-explored challenge. carroll says current methods suffer from insufficient task definitions with limited category coverage. carrol: knowing exactly where an event occurred is not enough; knowing exactly what it happened is.
Approach:	They propose a taxonomy of 21 vocal events with a new categorization into discrete versus continuous types.
Outcome:	The proposed model disentangles ASR errors from event detection while maintaining ASR quality.

Copied to clipboard

Guanting Luo, Noriki Nishida, Yuji Matsumoto, Yuki Arase

ACR: Adaptive Context Refactoring via Context Refactoring Operators for Multi-Turn Dialogue (2026.findings-acl)

Challenge:	Existing methods for identifying conditions under which a drug can be effective are limited . et al., j. n. d., al. c., and dr. m. s., 2005, are not able to identify context-specific conditions for therapeutic drug–disease relations.
Approach:	They propose to annotate triples of drugs, diseases, and applicability conditions from biomedical literature.
Outcome:	The proposed method outperforms baselines across evaluation settings.

Copied to clipboard

Jiawei Shen, Jia Zhu, Hanghui Guo, Weijie Shi, Yue Cui, Qingyu Niu, Guoqing Ma, Jingjiang Liu, Yidan Liang, Yilin Wang, Shimin Di, Jiajie Xu

ALGOGEN: Tool-Generated Verifiable Traces for Reliable Algorithm Visualization (2026.findings-acl)

Challenge:	Existing approaches to multi-turn dialogues lack contextual consistency and dependencies, and models struggle to maintain factual faithfulness as interaction turns increase.
Approach:	They propose an adaptive context refactoring framework that monitors and reshapes the interaction history to mitigate contextual inertia and state drift.
Outcome:	The proposed model outperforms baselines while reducing token consumption.

Copied to clipboard

null Liaokunpeng, Yuexiao Ma, Yisheng Lin, Hualin Zeng, Xiawu Zheng, Rongrong Ji

PivotAttack: Rethinking the Search Trajectory in Hard-Label Text Attacks via Pivot Words (2026.findings-acl)

Challenge:	Recent LLM-based systems require simulation of algorithm flow and video rendering constraints.
Approach:	They propose a paradigm that decouples algorithm execution from rendering.
Outcome:	The proposed paradigm reduces execution success rates, element overlap, and inter-frame inconsistencies.

Copied to clipboard

Yuzhi Liang, Shiliang Xiao, Jingsong Wei, Qiliang Lin, Xia Li

MiMIC: Mitigating Visual Modality Collapse in Universal Multimodal Retrieval While Avoiding Semantic Misalignment (2026.findings-acl)

Challenge:	Existing hard-label text attacks rely on inefficient "outside-in" strategies that traverse vast search spaces.
Approach:	They propose a query-efficient "inside-out" framework that perturbs Pivot Sets to induce label flips.
Outcome:	The proposed framework outperforms state-of-the-art methods in both Attack Success Rate and query efficiency.

Copied to clipboard

Li Juan, Chuanghao Ding, Xujie Zhang, Cam-Tu Nguyen

Towards Proactive Personalization through Profile Customization for Individual Users in Dialogues (2026.findings-acl)

Challenge:	Existing UMR methods can be broadly divided into two categories: early-fusion approaches, such as Marvel, which projects visual features into the language model space for integrating with text modality, and late-fusion methods, such UniVL-DR, which encode visual and textual inputs using separate encoders and obtain fused embeddings through addition.
Approach:	They propose to map different modalities into a shared embedding space for multi-modal retrieval.
Outcome:	Experiments on the WebQA+ and EVQA+ datasets show that MiMIC outperforms both early- and late-fusion approaches.

Copied to clipboard

Xiaotian Zhang, Yuan Wang, Ruizhe Chen, Zeya Wang, Runchen Hou, Zuozhu Liu

Beyond Stochastic Exploration: What Makes Training Data Valuable for Agentic Search (2026.findings-acl)

Challenge:	Existing alignment methods focus on universal human values or static, single-turn preferences, thereby failing to address the critical needs of long-term personalization and the initial user cold-start problem.
Approach:	They propose a user-centric lifelong agent that continuously infers and adapts to user preferences.
Outcome:	The proposed agent achieves superior performance over strong prompt-based and policy optimization baselines, not only in idealized but also in noisy conversational contexts.

Copied to clipboard

Chuzhan Hao, Wenfeng Feng, Guochao Jiang, Guofeng Quan, Guohua Liu, Yuewei Zhang

Can LLMs Hear the Dogwhistle? (2026.findings-acl)

Challenge:	Existing RL-based search agents rely on stochastic exploration, leading to inefficient reasoning trajectories and unstable training.
Approach:	They propose a framework to enhance the performance and training stability of search agents by transforming raw reasoning trajectories into hierarchical experience knowledge.
Outcome:	The proposed framework exhibits strong cross-task and cross-algorithm generalizations on multiple complex agentic search and mathematical reasoning benchmarks.

Copied to clipboard

Yifan Liu, Yi Lin, Xinwei Guo, Ziwei Wang, Jiaxin Zhang, Guanhua Chen, Haiyan Wu, Xiangyu Zhao, Xin Yao, Xuetao Wei

HiPrune: Hierarchical Attention for Efficient Token Pruning in Vision-Language Models (2026.findings-acl)

Challenge:	Existing safety benchmarks focus on explicitly harmful content, but ignore context-dependent expressions such as dogwhistles.
Approach:	They propose a benchmark for evaluating LLM safety under dogwhistle-driven prompts . their findings expose a blind spot in current safety evaluation practices .
Outcome:	The proposed benchmark compared safety performance with toxic terms using dogwhistle-driven prompts.

Copied to clipboard

Jizhihui Liu, Guangdao Zhu, Feiyi Du, Niu Lian, Jun Li, Bin Chen, Weili Guan, Yaowei Wang

MIThinker: A Plug-and-Play Policy-Optimized Thinker For Motivational Interviewing Counseling (2026.findings-acl)

Challenge:	Existing methods for visual token pruning lack insight into the intrinsic property of the vision encoder . et al., 2017: 99.3% of task accuracy with only 1/3 of the tokens.
Approach:	They propose a model-agnostic token pruning method that trains without training . they propose 'HiPrune' method which prunes visual tokens according to their attention .
Outcome:	The proposed method achieves 99.3% of task accuracy with only 1/3 of the tokens . it reduces inference FLOPs by 58.7% and maintains 99.99% accuracy with 2/9 tokens.

Copied to clipboard

Yizhe Yang, Palakorn Achananuparp, Heyan Huang, Jing Jiang, Ee-Peng Lim

The Art of (Mis)alignment: How Fine-Tuning Methods Effectively Misalign and Realign LLMs in Post-Training (2026.findings-acl)

Challenge:	Existing reasoning large language models (LLMs) generate responses without explicitly aligning thoughts with counseling techniques, limiting their effectiveness.
Approach:	They propose a lightweight thinking model that generates therapeutic thoughts to guide MI counseling agents in strategy selection and response generation.
Outcome:	The proposed model achieves theory-of-mind assessment comparable to state-of the-art systems with an order of magnitude less computation.

Copied to clipboard

Rui Zhang, Hongwei Li, Yun Shen, Xinyue Shen, Wenbo Jiang, Guowen Xu, Yang Liu, Michael Backes, Yang Zhang

ADaPT: Token-Level Decoupling for Efficient Large Reasoning Models (2026.findings-acl)

Challenge:	Misaligned large language models can magnify harm by exploiting them to undermine safety . et al., 2022b; Bai e.t., 2023): misalignment, realignment and model-specific resistance are important .
Approach:	They evaluate four methods to identify a mechanism asymmetry between attack and defense . they find that ORPO is most effective for misalignment, but DPO excels in realignment .
Outcome:	The proposed methods show a mechanism asymmetry between attack and defense . the proposed methods excel in realignment, but at the expense of model utility .

Copied to clipboard

Tingyun li, Zishang Jiang, Jinyi Han, Xinyi Wang, Sihang Jiang, Han Xia, Zhaoqian Dai, Ma Shuguang, Fei Yu, Jiaqing Liang, Yanghua Xiao

Talk to Your Slides: High-Efficiency Slide Editing via Language-Driven Structured Data Manipulation (2026.findings-acl)

Challenge:	Existing efficiency-oriented methods attempt to shorten or mix reasoning strategies, yet often degrade reasoning capability.
Approach:	They propose a token-level dual-process framework that explicitly decouples efficiency and correctness signals during training.
Outcome:	The proposed framework reduces inference cost while maintaining strong reasoning ability across multiple benchmarks.

Copied to clipboard

Kyudan Jung, Hojun Cho, Jooyeol Yun, Soyoung Yang, Jaehyeok Jang, Jaegul Choo

Verifiable Parameterization of Bayesian Networks from Scientific Literature: Unlocking Unstructured Empirical Evidence (2026.findings-acl)

Challenge:	Talk-to-Your-Slides is a high-efficiency slide editing agent that uses language-driven structured data manipulation instead of the image modality.
Approach:	They propose a language-driven slide editing agent that uses language-based structured data manipulation instead of image modality.
Outcome:	The proposed system achieves faster processing and better instruction fidelity than GUI-based agents.

Copied to clipboard

Jonas Gottal, Florian Matthes

PRIME: Ultra-Low-Rank Principal–Residual Model Merging (2026.findings-acl)

Challenge:	Current methods to learn conditional probabilities from raw tabular data are limited due to privacy concerns or general lack of access to data.
Approach:	They propose to reconstruct local conditional probability tables solely from statistical summaries to parameterize Bayesian Networks.
Outcome:	The proposed methods outperform baseline methods while ensuring transparency and verifiability.

Copied to clipboard

Seung-Ho Lee, Kyungsu Lee, Bazarvaani Zuchi, Jeongmin Ahn, Insuk Seo, Donghyeon Jeon, Inho Kang, Seung-Hoon Na

Don’t Tell the Answer, Truly Guide the Reasoning During RL Rollouts (2026.findings-acl)

Challenge:	Existing methods for model merging have been limited by task-specific performance and task-related tasks.
Approach:	They propose an ultra-low-rank principal-residual model merging framework that decomposes task vector merging into two stages.
Outcome:	Experiments on eight natural language processing tasks show that PRIME outperforms existing models while preserving the task-specific capabilities of the original models.

Copied to clipboard

Xinyi Wang, Jinyi Han, Zishang Jiang, Tingyun li, Jiaqing Liang, Sihang Jiang, Zhaoqian Dai, Ma Shuguang, Fei Yu, Yanghua Xiao

Mask Tokens as Prophet: Fine-Grained Cache Eviction for Efficient dLLM Inference (2026.findings-acl)

Challenge:	Existing methods such as GRPO often break down when task difficulty exceeds the model’s capacity, resulting in sparse rewards and inefficient training.
Approach:	They propose to measure the compatibility between external guidance and a model's intrinsic policy by introducing an adaptive framework to enhance reasoning performance while explicitly preserving high Affinity.
Outcome:	The proposed framework outperforms baseline models while maintaining high Affinity.

Copied to clipboard

Jianuo Huang, Yaojie Zhang, Yicun Yang, Benhao Huang, Linfeng Zhang

SPaCe: Unlocking Sample-Efficient Large Language Models Training With Self-Pace Curriculum Learning (2026.findings-acl)

Challenge:	Existing cache eviction strategies for autoregressive language models fail to account for the role of mask tokens and specific characteristics in dLLMs.
Approach:	They propose a training-free cache eviction framework tailored to dLLMs that denies a fully masked sequence and allows parallel decoding at the expense of memory and computation.
Outcome:	The proposed framework reduces the cost of memory and cache eviction and improves efficiency by reducing allocation in intermediate layers and concentrating resources on prompt-preferring heads.

Copied to clipboard

Van Dai Do, Manh Nguyen, Svetha Venkatesh, Hung Le

Crossroads of Optimization under Uncertainty: How to Choose the Optimal Model (2026.findings-acl)

Challenge:	Existing training pipelines sample training examples uniformly across steps or epochs, ignoring differences in difficulty, redundancy, and learning value, which slows learning and wastes computation.
Approach:	They propose a self-paced learning framework that enables efficient learning based on the capability of the model being trained through optimizing which data to use and when.
Outcome:	The proposed framework achieves comparable or better accuracy than state-of-the-art baselines while using up to (100 times) fewer samples.

Copied to clipboard

Chengxi She, Zhiqiang Chen, Xingyu Lu, Caihua Chen, Piaoyang Zhao, Xuedong Wang

Challenge:	Existing approaches to Optimization under Uncertainty (OuU) have inherent limitations and advantages.
Approach:	They propose a framework that automates the modeling and solving of six types of uncertainty models and generates mapping pairs to explore the potential relationship between optimization problems and optimal models.
Outcome:	The proposed framework achieves superior performance even on specific model types, with correlation analysis showing that data scale and specific scenario significantly influence model selection.

Know Thy Enemy: Securing LLMs Against Prompt Injection via Diverse Data Synthesis and Instruction-Level Chain-of-Thought Learning (2026.findings-acl)

Copied to clipboard

Zhiyuan Chang, Mingyang Li, Yuekai Huang, Ziyou Jiang, Xiaojun Jia, Qian Xiong, Junjie Wang, Zhaoyang Li, Qing Wang

ModularMoE: Fast LLM Customization with Parameter-Sharing Mixture-of-Experts for Low-Resource Settings (2026.findings-acl)

Challenge:	Large language model (LLM)-integrated applications face security vulnerabilities from prompt injection (PI) attacks.
Approach:	They propose a model enhancement method that synthesizes diverse training data and employs instruction-level chain-of-thought fine-tuning to enable LLMs to effectively identify and reject malicious instructions regardless of their source or position in the context.
Outcome:	The proposed method outperforms baselines in three critical dimensions while maintaining utility performance without degradation.

Copied to clipboard

Jiaxing Liu, Qi Qi, Haifeng Sun, Dunjun Li, Zirui Zhuang, Bo He, Xiang Yang, Cong Liu, Jianxin Liao, Jingyu Wang

UCGRec: User-Centric Graph Learning for LLM-based Sequential Recommendation (2026.findings-acl)

Challenge:	Large Language Models impose significant computational and storage burdens on personal devices . existing customization approaches incur excessive computational costs or lead to suboptimal performance .
Approach:	They propose a training framework that converts pre-trained LLMs into parameter-sharing MoE models for lightweight deployment.
Outcome:	The proposed training framework outperforms state-of-the-art training frameworks at the same sparsity level while delivering up to 2.71 inference speedup.

Copied to clipboard

HanBeul Kim, CheolWon Na, Suyoung Bae, YunSeok Choi, Jee-Hyong Lee

HeteroRAG: A Heterogeneous Retrieval-Augmented Generation Framework for Medical Vision Language Tasks (2026.findings-acl)

Challenge:	Existing methods for sequential recommendation rely primarily on item descriptions or utilize user preferences independently.
Approach:	They propose a method that integrates diverse user-relevant preference signals into a unified user-centric graph and injects the graph-based knowledge into the LLM through end-to-end training with graph neural networks.
Outcome:	The proposed method outperforms conventional and state-of-the-art methods on four widely used sequential real-world recommendation datasets.

Copied to clipboard

Zhe Chen, Yusheng Liao, Zhiyuan Zhu, Haolin Li, Hongcheng Liu, Yanfeng Wang, Yu Wang

Subgraph-Guided Executable Logical Form Generation for Knowledge Base Question Answering (2026.findings-acl)

Challenge:	Medical large vision-language models suffer from factual inaccuracies and unreliable outputs.
Approach:	They propose a framework that enhances Med-LVLMs through heterogeneous knowledge sources.
Outcome:	The proposed framework improves Med-LVLMs through heterogeneous knowledge sources.

Copied to clipboard

Yuhang Tian, Dandan Song, Zhijing Wu, Changzhi Zhou, Jun Yang, Huipeng Ma, Chenhao Li, Luan Zhang, Yading Li, Xudong Li, Shenxi Liu, Jing Jiang

Towards Reliable Paper Contributions Annotation in the ACL Rolling Review (2026.findings-acl)

Challenge:	Existing retrieval-augmented approaches focus on ignoring the structural information of the Knowledge Base (KB) and the question.
Approach:	They propose a structure-aware subgraph retrieval stage that ranks candidate subgraphs by aligning them with the question’s structure, along with semantic relevance.
Outcome:	Experiments on GrailQA, WebQSP, and GraphQuestions show that the proposed framework achieves state-of-the-art performance.

Copied to clipboard

Julien Aubert-Béduchaud, Florian Boudin, Akiko Aizawa, Beatrice Daille, Richard Dufour

Memory Dial: A Training Framework for Controllable Memorization in Language Models (2026.findings-acl)

Challenge:	Identifying the types of contributions an article makes can help readers grasp its significance.
Approach:	They propose to use a typology to categorize articles by their contributions to improve review quality and fairness.
Outcome:	The ACL Rolling Review (ARR) introduced a typology requiring authors to specify their contributions to improve review quality and fairness.

Copied to clipboard

Xiangbo Zhang, Ali Emami

SEAD: Self-Evolving Agent for Multi-Turn Service Dialogue (2026.findings-acl)

Challenge:	Existing approaches to memorization detection are post-hoc . large language models can reproduce training data verbatim, complicating accuracy estimates .
Approach:	They propose a training framework that makes memorization an explicit variable.
Outcome:	The proposed framework produces models identical in architecture, data, and optimization, but varying in memorization pressure.

Copied to clipboard

Yuqin Dai, Ning Gao, Wei Zhang, Jie Wang, null Luozichen, Ruiyuan Wu, Jinpeng Wang, Chaozheng Wang

DataSciBench: An LLM Agent Benchmark for Data Science (2026.findings-acl)

Challenge:	Large Language Models have demonstrated remarkable capabilities in open-domain dialogues, but their performance in service dialogues remains suboptimal.
Approach:	They propose a framework that enables agents to learn effective strategies without large-scale human annotations.
Outcome:	The proposed framework decouples user modeling into two components that provide adaptive training scenarios rather than acting as an unfair adversary.

Copied to clipboard

Dan Zhang, Sining Zhoubian, Min Cai, Fengzu Li, Lekang Yang, Wei Wang, Tianjiao Dong, Ziniu Hu, Jie Tang, Yisong Yue

Challenge:	Existing benchmarks focus on single task, simple evaluation metrics, and readily available ground truth (GT) DataSciBench is built on curated, natural, and challenging prompts with complex evaluation criteria and uncertain GT.
Approach:	They propose a benchmark for evaluating Large Language Models in data science that integrates LLM-based self-consistency and human verification to ensure accuracy.
Outcome:	The proposed framework outperforms open-source models in all metrics and offers rigorous insights into LLM strengths and weaknesses.

PVPO: Pre-Estimated Value-Based Policy Optimization for Agentic Reasoning (2026.findings-acl)

Copied to clipboard

Wenfeng Feng, Penghong Zhao, Guochao Jiang, Chuzhan Hao, Guohua Liu, Yuewei Zhang

Momoka-RAG: MCTS-Organized Mapping of Knowledge Associations for Long-Document Retrieval Augmented Generation (2026.findings-acl)

Challenge:	extending grouping-based methods to agentic reasoning presents unique challenges . frequent environment interactions and tool invocations render intra-group advantage estimation unstable .
Approach:	They propose a grouping-based method that uses a single round of rollouts to stabilize advantage estimation.
Outcome:	a new RL framework outperforms grouping-based methods in retrieval tasks and advanced mathematical reasoning benchmarks.

Copied to clipboard

Wenyu Tao, Xiaofen Xing, Zeliang Li, Xiangmin Xu

A3: Android Agent Arena for Mobile GUI Agents with Essential-State Procedural Evaluation (2026.findings-acl)

Challenge:	Existing frameworks that rely on fixed-length chunking are unsuitable for long-document tasks due to their passive and mechanical approach to knowledge structure.
Approach:	They propose a framework that utilizes Monte Carlo Tree Search to proactively uncover connections among chunks and construct optimal semantic information paths with the objective of completing semantic relationships.
Outcome:	The proposed framework achieves higher precision while maintaining competitive recall compared to other RAG frameworks.

Copied to clipboard

Yuxiang Chai, Shunye Tang, Han Xiao, Weifeng Lin, Hanhao Li, Jiayu Zhang, Liang Liu, Pengxiang Zhao, Guangyi Liu, Guozhi Wang, Shuai Ren, Rongduo Han, Haining Zhang, Siyuan Huang, Hongsheng Li

Intrinsic Mutual Information as a Modulator for Preference Optimization (2026.findings-acl)

Challenge:	Existing evaluation methods for mobile GUI agents rely on static frame assessments or offline static apps.
Approach:	They propose an evaluation system that leverages large language models as reward models to verify task completion and process achievement.
Outcome:	The proposed system addresses the limitations of traditional function based evaluation methods on online dynamic apps.

Copied to clipboard

Peng Liao, Peijia Zheng, Lingbo Li, Shangsong Liang, Lin Chen

Challenge:	Existing methods for offline preference optimization involve additional hyperparameter tuning, resulting in substantial time overhead.
Approach:	They propose a lightweight framework for offline preference optimization that leverages hyperparameter modulation to decouple preference contributions.
Outcome:	The proposed framework achieves superior performance over existing methods while reducing training overhead by more than 15%.

EVGeoQA: Benchmarking LLMs on Dynamic, Multi-Objective Geo-Spatial Exploration (2026.findings-acl)

Copied to clipboard

Jianfei Wu, Zhichun Wang, Zhensheng Wang, Zhiyu He

Challenge:	Existing geo-spatial question answering benchmarks focus on static retrieval, failing to capture the complexity of real-world planning that involves dynamic user locations and compound constraints.
Approach:	They propose a new benchmark for Large Language Models that integrates location-anchored and dual-objective queries with a user's real-time coordinates.
Outcome:	The proposed model can summarize historical exploration trajectories to enhance exploration efficiency.

FinMRAGBench: A Realistic and Complex Benchmark for Multi-Modal RAG in Financial Document Analysis (2026.findings-acl)

Copied to clipboard

Shouqing Yang, Qi Zhang, Yuhang Yang, Ruikang Xu, Yuwei Hou, Zhulin Jia, Lirong Gao, Haobo Wang, Jinglei Chen, Jiexiang Wang, Sheng Guo, Bo Zheng, Gang Chen

Apeiron: A Scalable LLM-agentic Framework for Autonomous Full-lifecycle Demand-optimized Application Synthesis (2026.findings-acl)

Challenge:	Existing benchmarks for realistic financial analysis fail to capture realistic financial situations involving cross-document retrieval, multi-page evidence integration, and diverse analytical tasks.
Approach:	They propose a multi-modal financial RAG benchmark that evaluates large language models in realistic financial analysis settings.
Outcome:	The proposed framework achieves the strongest overall performance across all models.

Copied to clipboard

Junyan Cheng, Ankit Srivastava, Jessie Zeng, Milenko Drinic, Jack W. Stokes

Challenge:	Traditional, rigid, 'one-size-fits-all' apps are struggling in the contemporary landscape.
Approach:	They propose a scalable and extensible framework for addressing amorphous user demands through autonomous, full-lifecycle application synthesis.
Outcome:	The proposed framework outperforms baselines in CUA ratings and user-demand task scores across 300 app scenarios, 2,400 personas, and 46,338 demands.

Eval-RAR: Evaluation-Driven Retrieval-Augmented Reasoning via Reinforcement Learning (2026.findings-acl)

Copied to clipboard

Heng Yu, Rui Li, Qi Liu, Wenjun Feng, Junfeng Kang, Yi Zhan

LUMINA: Long-horizon Understanding for Multi-turn Interactive Agents (2026.findings-acl)

Challenge:	Existing methods for retrieval-augmented generation fail to provide explicit supervision for internal reasoning process.
Approach:	They propose an Evaluation-driven Retrieval-Augmented Reasoning framework that uses reinforcement learning and a fine-grained evaluation reward to optimize the process.
Outcome:	Eval-RAR outperforms existing methods on QA benchmarks on seven single-hop and multi-hop tasks.

Copied to clipboard

Amin Rakhsha, Thomas Hehn, Pietro Mazzaglia, Fabio Valerio Massoli, Arash Behboodi, Tribhuvanesh Orekondy

Travel on the ICD Tree: Benchmarking Agentic Reasoning for ICD Coding from Chinese Electronic Medical Records (2026.findings-acl)

Challenge:	Large language models struggle on multi-turn, long-horizon agentic problems that require skills such as planning, state tracking, and long context processing.
Approach:	They propose an oracle counterfactual framework for multi-turn problems that asks: how would an agent perform if it could leverage an or acle to execute a specific skill?
Outcome:	The proposed framework allows for precise oracle interventions without confounding effects present in real-world benchmarks.

Copied to clipboard

Xinjie Xu, Yongqi Fan, Shuang-shuang Chen, Qi Ye, Weibin Guo, Xinxuan Hu

RiT: Rubrics-in-Thinking Reinforcement Learning for Improved Reasoning in Large Language Models (2026.findings-acl)

Challenge:	Accurate International Classification of Diseases (ICD) coding is crucial for hospital management and healthcare data governance.
Approach:	They propose a framework to evaluate ICD coding based on complete EMRs . they use a dataset of 560 real clinical records covering 434 common diseases .
Outcome:	The proposed framework explores the capability boundaries of large language models under different paradigms.

Copied to clipboard

Xiaobin Tian, Shuai Yuan, Muyun Ding, Haonan Chen, Xiaoxi Jiang

From If-Statements to ML Pipelines: Revisiting Bias in Code-Generation (2026.findings-acl)

Challenge:	Large Reasoning Models benefit from generating intermediate reasoning steps alongside final answers.
Approach:	They propose a framework to introduce thinking-rubric supervision into intermediate reasoning.
Outcome:	The proposed framework outperforms outcome-only RL baselines on reasoning-intensive and open-ended tasks.

Copied to clipboard

Minh Duc Bui, Xenia Heilmann, Mattia Cerrato, Manuel Mager, Katharina Von Der Wense

SEA-SafeguardBench: Culturally Grounded Safety Benchmark for Southeast Asian Languages (2026.findings-acl)

Challenge:	Existing methods to evaluate code generation bias focus on overt discrimination through simple conditional statements.
Approach:	They examine ML pipelines that exhibit substantially greater bias than simple conditionals . they challenge simple conditional statements as valid proxies for bias evaluation .
Outcome:	The proposed model underestimates real-world bias in generating machine learning pipelines . the model maintains equal performance on simple conditionals and ML pipelines, the study shows .

Copied to clipboard

Panuthep Tasawong, Jian Gang Ngui, Alham Fikri Aji, Trevor Cohn, Peerat Limkonchotiwat

Challenge:	Existing multilingual safety benchmarks rely on machine-translated English data, which fails to capture nuances in low-resource languages.
Approach:	They propose to use a human-verified safety benchmark for Southeast Asian languages to validate their safety and cultural diversity.
Outcome:	The proposed model outperforms existing models in general, in-the-wild, and content generation across eight languages and 21,640 samples across three subsets: general, and in- the-wild.

COCOGEC: Counterfactual Generation for Robust Grammatical Error Correction (2026.findings-acl)

Copied to clipboard

Qianyu Wang, Xiaoman Wang, Yuanyuan Liang, Xinyuan Li, Yunshi Lan

DRP: Distilled Reasoning Pruning with Mathematical Skill-aware Step Decomposition for Efficient Large Reasoning Models (2026.findings-acl)

Challenge:	Existing GEC models fail to understand error patterns in varying contexts . a framework that generates copies of training instances with error-irrelevant contexts altered is proposed .
Approach:	They propose a framework that generates copies of training instances with error-irrelevant contexts altered.
Outcome:	The proposed framework outperforms baselines on the simulated tasks and outperformed existing models.

Copied to clipboard

Yuxuan Jiang, Dawei Li, Francis Ferraro

Bridging the Pose-Semantic Gap: A Cascade Framework for Text-Based Person Anomaly Search (2026.findings-acl)

Challenge:	Existing solutions to this problem are inference-time pruning and tuning-based distillation.
Approach:	They propose a framework that combines inference-time pruning with tuning-based distillation to enable efficient and accurate reasoning.
Outcome:	The proposed framework reduces token usage while improving accuracy on GSM8K and AIME tokens while avoiding performance drop.

Copied to clipboard

Zequn Xie, Guijin Luo, Chuxin Wang, Sihang Cai, Tao Jin, Zhou Zhao, Yixuan Tang

TemplateRL: Structured Template-Guided Reinforcement Learning for LLM Reasoning (2026.findings-acl)

Challenge:	Existing methods for person anomaly search fail to address the complexities of real-world security, authors say . Existing approaches fail to detect subtle semantic distinctions, authors argue .
Approach:	They propose a framework that decouples retrieval into two stages . structure-aware coarse retrieval and detective squad interaction are proposed .
Outcome:	The proposed framework achieves state-of-the-art performance by balancing efficiency and semantic reasoning.

Copied to clipboard

Jinyang Wu, Chonghua Liao, Mingkuan Feng, Shuai Zhang, Zhengqi Wen, Haoran Luo, Ling Yang, Huazhe Xu, Jianhua Tao

Dual-Reasoner: Bridging Interleaved Atomicity and Streaming Latency via Thinking-while-Talking (2026.findings-acl)

Challenge:	Existing RL methods rely on unstructured self-sampling to fit scalar rewards, resulting in inefficient rollouts.
Approach:	They propose a structured template-guided RL framework that augments policy optimization with explicit template guidance.
Outcome:	Experiments show that TemplateRL outperforms GRPO and GRPI by 99% on AIME and 41% on AMC with superior stability on weak models and remarkable cross-domain generalization.

Copied to clipboard

Yangzhuo Li, Shengpeng Ji, Yifu Chen, Tianle Liang, Haoyu Yang, null Junboli, Jun Fang, Lin Li, Qingyang Hong

FBS: Modeling Native Parallel Reading inside a Transformer (2026.findings-acl)

Challenge:	Existing methods to integrate Chain-of-Thought into spoken dialogue models incur prohibitive latency.
Approach:	They propose a Streaming Masking Mechanism to ensure uninterrupted audio streaming . they use a quadruple-constraint system to reconstruct logical atomicity .
Outcome:	Experimental results show that Dual-Reasoner improves speech generation performance with low latency.

Copied to clipboard

Tongxi Wang

HqeKV: Towards Hybrid Quantization and Eviction for KV Cache in Long-Context LLM Inference (2026.findings-acl)

Challenge:	Existing acceleration methods largely patch the autoregressive pipeline and miss core human-reading ingredients.
Approach:	They propose a trainable loop that injects a causal loop into Transformers via a 'parafoveal' approach.
Outcome:	The proposed model improves quality-efficiency trade-off without increasing parameters . ablations show the three modules are complementary .

Copied to clipboard

He Wang, Yu Gu, Fangfang Li, Zhigang Wang, Zhenghao Liu, Ning Wang, Xiaohua Li, Ge Yu

Challenge:	autoregressive inference requires repeated computation across transformer layers.
Approach:	They propose a hybrid compression framework built on both quantization and eviction . they propose varying importance metric and flexible conversion policies to reduce memory overhead .
Outcome:	The proposed framework outperforms state-of-the-art methods under memory constraints.

Task Matters: Knowledge Requirements Shape LLM Responses to Context–Memory Conflict (2026.findings-acl)

Copied to clipboard

Kaiser Sun, Fan Bai, Mark Dredze

Challenge:	Prior work has shown that large language models favor parametric knowledge under conflict, but this setting assumes that tasks should always rely on the provided passage.
Approach:	They propose a model-agnostic diagnostic framework that holds underlying knowledge constant while injecting controlled conflicts across tasks with varying knowledge requirements.
Outcome:	Evaluating representative open-source LLMs, the proposed framework holds underlying knowledge constant while injecting controlled conflicts across tasks with varying knowledge requirements.

Adaptive Data Collection for Latin-American Community-sourced Evaluation of Stereotypes (LACES) (2026.findings-acl)

Copied to clipboard

Guido Ivetta, Pietro Palombini, Sofía Martinelli, Marcos J Gomez, M Emilia Echeveste, Sunipa Dev, Vinodkumar Prabhakaran, Luciana Benotti

MR-ALIGN: Meta-Reasoning Informed Factuality Alignment for Large Reasoning Models (2026.findings-acl)

Challenge:	a geo-cultural gap in NLP evaluation hinders evaluation of societal biases . authors propose a new method to collect stereotypes from large language models .
Approach:	They propose a new method that integrates sourcing and validation of existing data into a single workflow.
Outcome:	The proposed method improves LACES by integrating new stereotype entries and validation of existing data.

Copied to clipboard

Xinming Wang, Jian Xu, Bin Yu, Sheng Lian, yi Chen, Boran Wang, Yingjian Zhu, Hongzhu Yi, Hong-Ming Yang, Han Hu, Cheng-Lin Liu, Xu-Yao Zhang

Chameleons and Guardians: Unveiling the Divergence in Personality Plasticity and Cognitive Resistance across LLMs (2026.findings-acl)

Challenge:	Large reasoning models (LRMs) show strong capabilities in complex reasoning, yet their marginal gains on evidence-dependent factual questions are limited.
Approach:	They propose a Meta-Reasoning informed alignment framework that quantifies state-transition probabilities along the model’s thinking process and constructs a transition-aware implicit reward that reinforces beneficial reasoning patterns while suppressing defective ones at the atomic thinking segments.
Outcome:	Empirical evaluations of four factual QA datasets and one long-form factuality benchmark show that MR-ALIGN consistently improves accuracy and truthfulness while reducing misleading reasoning.

Copied to clipboard

Jiaxin Wu, Jie Zhao, null Huangxinrong, Cuihong Zhang, Xiuzhu Wu

Challenge:	Existing studies on LLMs argue for its immutability, but prior studies have not found that personality-inducing contexts can be intentionally reshaped.
Approach:	They propose a personality-inducing framework that reshapes LLMs via multi-agent collaboration . they paraphrase MBTI questions to create semantically equivalent but expressively diverse inducing contexts .
Outcome:	Experiments on worldwide mainstream LLMs show that PIF transforms their original personalities into desired target personalities.

Entropy Scheduling in Reinforcement Learning for Large Language Models (2026.findings-acl)

Copied to clipboard

Xingjin Wang, Howe Tissue, Lu Wang, Linjing Li, Daniel Dajun Zeng

TopoDIM: One-shot Topology Generation of Diverse Interaction Modes for Multi-Agent Systems (2026.findings-acl)

Challenge:	entropy in reinforcement learning functions analogously to the learning rate in LLMs.
Approach:	They propose an entropy scheduling system that optimizes different pre-set goals by controlling and scheduling entropicy at each step of the RL process.
Outcome:	The proposed method improves AIME2024 from 50.9 to 54.9 within 40 training steps.

Copied to clipboard

Rui Sun, Jie Ding, Chenghua Gong, Tianjun Gu, Yihang Jiang, Juyuan Zhang, Liming Pan, Linyuan Lü

AVA: Attentive VLM Agent for Mastering StarCraft II (2026.findings-acl)

Challenge:	Existing communication topologies rely on spatio-temporal dialogues, which incur high latency and computation.
Approach:	They propose a framework for one-shot Topology generation with Diverse Interaction Modes that enables agents to construct heterogeneous communication without iterative coordination.
Outcome:	The proposed framework reduces total token consumption by 46.41% while improving average performance by 1.50% over state-of-the-art methods.

Copied to clipboard

Weiyu Ma, Yuqian Fu, Zecheng Zhang, Bernard Ghanem, Guohao Li

Structure-Aware Quantized Retrieval for Long-Document Question Answering (2026.findings-acl)

Challenge:	Existing StarCraft II benchmarks rely on abstract state representations that deviate from human perception . Existing systems rely only on abstract representations, creating an artificial gap between how humans process battlefield information and limiting ecological validity of learned behaviors.
Approach:	They introduce AVACraft, the first multimodal benchmark environment for complex decision-making in StarCraft II.
Outcome:	The AVACraft benchmark supports both traditional and modern multi-agent reinforcement learning paradigms.

Copied to clipboard

Hui Huang, Julien Velcin, Yacine Kessaci

GRPO-CARE: Consistency-Aware Reinforcement Learning for Multimodal Reasoning (2026.findings-acl)

Challenge:	Traditional long-document QA/RAG pipelines suffer from context fragmentation . Graph Neural Networks (GNNs) can encode hierarchy, but deep GNNs tend to over-smooth representations.
Approach:	They propose a framework that aligns hierarchical graph representations with a universal token vocabulary and integrates explicit structure into retrieval.
Outcome:	The proposed framework captures universal hierarchical patterns rather than overfitting to specific layouts.

Copied to clipboard

Yi Chen, Yuying Ge, Rui Wang, Yixiao Ge, Junhao Cheng, Ying Shan, Xihui Liu

Silencing the Guardrails: Inference-Time Jailbreaking via Dynamic Contextual Representation Ablation (2026.findings-acl)

Challenge:	Recent reinforcement learning approaches have advanced reasoning in Large Language Models (LLMs), yet their adaptation to multimodal LLMs remains underexplored.
Approach:	They propose a reinforcement learning framework that eliminates KL penalties and rewards consistency . they propose GRPO-CARE, which outperforms standard GR PO, with a base reward for accuracy and an adaptive bonus for consistency.
Outcome:	The proposed framework outperforms standard GRPO on the most difficult evaluation level and reasoning consistency test benchmarks.

Copied to clipboard

Wenpeng Xing, Moran Fang, Guangtai Wang, Changting Lin, Meng Han

Diffusion with Truncated Blocks: Fast and High-Quality Text Generation using Truncated Block Generation (2026.findings-acl)

Challenge:	Existing strategies to circumvent safety constraints face significant trade-offs between effectiveness and efficiency.
Approach:	They propose a framework that allows to infer model refusal behaviors without expensive parameter updates or training.
Outcome:	The proposed framework outperforms baselines in multiple safety-aligned open-source LLMs.

Copied to clipboard

Yuyan Zhou, Weiyu Chen, James Kwok

GigaCheck: Detecting LLM-generated Content via Object-Centric Span Localization (2026.findings-acl)

Challenge:	Diffusion-based Large Language Models (dLLMs) generate text by iteratively denoising masked sequences.
Approach:	They propose a method that iteratively denoises masked sequences to reduce the model's attention dilution by token-level noise while models employing sequence-level noising exhibit a reduced effect.
Outcome:	The proposed method improves the performance and efficiency of Diffusion-based large language models by iterating on masked sequences.

Copied to clipboard

Irina Tolstykh, Aleksandra Tsybina, Sergey Yakubson, Aleksandr Gordeev, Vladimir Dokholyan, Maksim Kuprashevich

Chain-in-Tree: Back to Sequential Reasoning in LLM Tree Search (2026.findings-acl)

Challenge:	GigaCheck is a framework for AI-generated text detection.
Approach:	They propose a dual-strategy framework for AI-generated text detection . they leverage representation learning of fine-tuned LLMs to discern authorship .
Outcome:	The proposed framework can detect LLM-generated content with high accuracy and accuracy . it can be used in mixed-authorship scenarios and in academic collaborations .

Copied to clipboard

Xinzhe Li

Addressing Overthinking in Large Vision-Language Models via Gated Perception-Reasoning Optimization (2026.findings-acl)

Challenge:	Large language models excel at tasks such as mathematical and commonsense reasoning, but their performance improves further when additional test-time compute is allocated.
Approach:	They propose a plug-in framework that decides when to branch during search instead of expanding at every step.
Outcome:	The proposed framework reduces token generation, model calls, and runtime by 75-85% on GSM8K and Math500, with negligible or no accuracy loss.

Copied to clipboard

Xingjian Diao, Zheyuan Liu, Chunhui Zhang, Weiyi Wu, Keyi Kong, Lin Shi, Kaize Ding, Soroush Vosoughi, Jiang Gui

CSMCIR: CoT-Enhanced Symmetric Alignment with Memory Bank for Composed Image Retrieval (2026.findings-acl)

Challenge:	Prior work has attempted to mitigate this issue by using adaptive reasoning strategies, but these methods overlook a fundamental bottleneck: visual perception failures.
Approach:	They propose a meta-reasoning controller that dynamically routes computation among three decision paths at each generation step.
Outcome:	The proposed method outperforms slow-thinking methods while producing shorter responses.

Copied to clipboard

Zhipeng Qian, Zihan Liang, Yufei Ma, Ben Chen, Huangyu Dai, Yiwei Ma, Jiayi Ji, Chenyi Lei, Han Li, Xiaoshuai Sun

RepoDistill: Distilling Repository Knowledge through Compression-Aware Budget Allocation and Policy Optimization (2026.findings-acl)

Challenge:	Existing approaches to search for images using single-modality are limited by representation space fragmentation.
Approach:	They propose a unified representation framework that achieves efficient query-target alignment . they introduce a multi-level Chain-of-Thought prompting strategy that guides MLMs to generate discriminative, semantically compatible captions for target images .
Outcome:	The proposed framework achieves efficient query-target alignment through synergistic components.

Copied to clipboard

Xin Yin, Zixiang Ding, Yiang Zhang, Qiang Wang, Rui Wang, Chao Ni, Zhe Cui

Finding RELIEF: Shaping Reasoning Behavior without Reasoning Supervision via Belief Engineering (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have strong performance on code translation tasks, but they struggle with repository-level scenarios where context is extensive and interdependent.
Approach:	They propose a framework that integrates retrieval with learning budget allocation for fine-grained context compression.
Outcome:	The proposed framework outperforms baselines on SWE-QA, CoderEval, and LongCodeU.

Copied to clipboard

Chak Tou Leong, Dingwei Chen, Heming Xia, Qingyu Yin, Sunbowen Lee, Jian Wang, Wenjie Li

Bridging SFT and RL: Dynamic Policy Optimization for Robust Reasoning (2026.findings-acl)

Challenge:	Existing methods for shaping large reasoning models rely on reinforcement learning or fine-tuning with gold-standard reasoning traces. Existing techniques for behavior shaping rely only on additional reward modeling.
Approach:	They propose a framework that aligns a model's self-concept with a target belief blueprint and internalizes desired traits by fine-tuning on synthesized, self-reflective QA pairs that affirm the target belief.
Outcome:	The proposed framework outperforms behavior-supervised and preference-based models while requiring significantly lower training costs.

Copied to clipboard

Taojie Zhu, Dongyang Xu, Ding Zou, Sen Zhao, Qiaobo Hao, Zhiguo Yang, Yonghong He

FrontCoder: Scaling Visual Fidelity in Front-End Code Generation (2026.findings-acl)

Challenge:	Existing unified optimization strategies overlook the statistical conflict between these distinct gradient signals.
Approach:	They propose a framework to reduce bias-variance trade-offs in Large Language Models . they propose DYPO, which leverages intrinsic group dynamics to significantly reduce RL gradient variance .
Outcome:	The proposed framework outperforms traditional pipelines on reasoning benchmarks and out-of-distribution tasks.

Copied to clipboard

Jun Feng, Jian Yang, Wei Zhang, Jing Wang, Keyi Chen, Xiaokun Yang, Weicheng Gu, Yihang Lou, Yan Bai, Xianglong Liu

Bridging Cognition and Affect: Emotion-Aware Opinion Summarization using LLMs (2026.findings-acl)

Challenge:	Existing work on front-end code generation fails to provide visual fidelity and rendering quality for front- end developers.
Approach:	They propose a three-stage pipeline to enhance front-end code generation capabilities in LLMs . they use synthetic data, quality-controlled supervised fine-tuning, and reinforcement learning .
Outcome:	The proposed model achieves competitive performance with frontier models while maintaining generation efficiency.

Copied to clipboard

Arnav Attri, Anuj Attri

The Double Bind: Revisiting Preprinting and Peer Review Two Years After the Removal of the ACL Anonymity Period (2026.findings-acl)

Challenge:	Emotion-aware Opinion Summarization (EAOS) is a framework that captures emotions that shape purchasing decisions.
Approach:	They propose a framework that integrates emotion into opinion summaries and a large-scale training dataset and an evaluation benchmark to support this task.
Outcome:	The proposed framework captures discrete emotions that shape purchasing decisions.

Copied to clipboard

A Pranav, Shane Storks, Anne Lauscher

TinyAlign: Boosting Lightweight Vision-Language Models by Mitigating Modal Alignment Bottlenecks (2026.findings-acl)

Challenge:	ACL removed the anonymity period for conference submissions in February 2024 .
Approach:	They track preprinting trends for 47k publications and analyze 1.9k peer reviews . they suggest improving visibility and investing in diversity initiatives .
Outcome:	The proposed anonymity period was removed in 2024, but it was ineffective for underrepresented researchers . the authors suggest addressing D&I issues rather than implementing anonymity policies.

Copied to clipboard

Yuanze Hu, Xinyu Wang, Zhichao Yang, Gen Li, Ye Qiu, Zhaoxin Fan, Yifan Sun, Wenjun wu, Jin Dong, Xiaotie Deng

Heterogeneity in Formal Linguistic Competence of Language Models: Is Data the Real Bottleneck? (2026.findings-acl)

Challenge:	Lightweight Vision-Language Models (VLMs) are indispensable for resource-constrained applications.
Approach:	They propose a framework that retrieves context from a memory bank to enhance alignment . they propose EMI-based approach to align vision and language models .
Outcome:	The proposed framework reduces training loss, accelerates convergence, and enhances task performance with negligible computational overhead.

Copied to clipboard

H S V N S Kowndinya Renduchintala, Sumit Bhatia

Real Men are Tough: Evaluating Gender Bias and Sensitivity to Masculinity Norms in LLMs (2026.findings-acl)

Challenge:	Large Language Models (LLMs) exhibit a puzzling disparity in their formal linguistic competence, even after training on trillions of tokens.
Approach:	They pre-train Large Language Models on 100M-token corpora and inject a minimal amount of synthetic data targeting specific linguistic phenomena into the model.
Outcome:	The proposed intervention significantly improves model performance in 8 out of the 9 worst-performing BLiMP paradigms.

Copied to clipboard

Elisa Leonardelli, Camilla Casula, Boglarka Nyul, Sara Tonelli

RISER: Orchestrating Latent Reasoning Skills for Adaptive Activation Steering (2026.findings-acl)

Challenge:	Large language models exhibit gender bias, but most evaluations focus on downstream stereotypes . a recent study found that explicit endorsement of masculinity norms is low across models .
Approach:	They investigate whether large language models rely on traditional masculinity norms as latent priors in gender-biased inference.
Outcome:	The findings show that large language models rely on stereotypes as latent priors . the authors used the Male Role Norms Inventory (MRNI) to investigate gender bias .

Copied to clipboard

Wencheng Ye, Xiaoyang Yuan, Yi Bin, Hengyu Jin, Liang Peng, Pengpeng Zeng, Heng Tao Shen

ReCoT-NER: Enhancing Zero-Shot Named Entity Recognition through Chain-of-Thought Prompting and Recall-Oriented Loss Optimization (2026.findings-acl)

Challenge:	Existing methods for domain-specific reasoning with large language models require updating parameter updates.
Approach:	They propose a plug-and-play intervention framework that adaptively steers LLM reasoning in activation space.
Outcome:	The proposed framework achieves zero-shot accuracy improvements of 3.4–6.5% over the base model while outperforming chain-of-thought-style reasoning with 2–3 higher token efficiency and robust accuracy gains.

Copied to clipboard

Dabin Fu, Fanghong Zhang

SyncThink: A Training-Free Strategy to Align Inference Termination with Reasoning Saturation (2026.findings-acl)

Challenge:	Named Entity Recognition (NER) is a key component of natural language processing (NLP) but it is difficult to implement in specialized domains such as wind power fault diagnosis.
Approach:	They propose a reasoning-enhanced generative framework that integrates Chain-of-Thought prompting and recall-oriented loss optimization to address these challenges.
Outcome:	The proposed framework improves recall and overall F1 performance across general and industrial domains.

Copied to clipboard

Gengyang Li, Wang Cai, Yifeng Gao, Yunfang Wu

Reason in Chains, Learn in Trees: Self-Rectification and Grafting for Multi-turn Agent Policy Optimization (2026.findings-acl)

Challenge:	Large language models (LLMs) achieve strong reasoning with Chain-of-Thought prompting, but long and redundant traces substantially increase inference cost.
Approach:	They propose a training-free and plug-and-play decoding method that reduces CoT overhead without modifying model weights.
Outcome:	Experiments on GSM8K, MMLU, GPQA, and BBH show that SyncThink achieves 62.00% average Top@1 accuracy using 656 generated tokens and 28.68s latency, compared to 61.22%, 2141 tokens, and 92.01s for full CoT decoding.

Copied to clipboard

Yu Li, Sizhe Tang, Tian Lan

FlashMem: Distilling Intrinsic Latent Memory via Computation Reuse (2026.findings-acl)

Challenge:	Existing approaches to reinforcement learning for Large Language Models treat trajectories as independent chains and ignore critical steps that may disproportionally impact reasoning outcome.
Approach:	They propose a framework that recovers latent correlated reward structure across seemingly independent trajectories by identifying and merging functionally similar steps/nodes.
Outcome:	The proposed framework recovers latent correlated reward structure across seemingly independent trajectories.

Copied to clipboard

Yubo Hou, Zhisheng Chen, Tao Wan, Zengchang Qin

QueueEDIT: Structural Self-Correction for Sequential Model Editing in LLMs (2026.findings-acl)

Challenge:	Large Language Models lack the mechanism to preserve dynamic context, forcing agents to redundantly reprocess history to maintain long-horizon autonomy.
Approach:	They propose a framework that distills intrinsic memory directly from transient reasoning states via computation reuse.
Outcome:	Experiments show that FlashMem matches heavy baselines while reducing inference latency by 5 times, effectively bridging the gap between efficiency and persistent cognition.

Copied to clipboard

Taolin Zhang, Haidong Kang, Dongyang Li, Qizhou Chen, Xiaofeng He, Chengyu Wang, Richang Hong

CAGenMol: Condition-Aware Diffusion Language Model for Goal-Directed Molecular Generation (2026.findings-acl)

Challenge:	Recent studies have shown that large language models (LLMs) can be effective for correcting factual inaccuracies but can still suffer from hallucinations.
Approach:	They propose a queue-based self-correction framework that addresses parameter bias during sequential model editing.
Outcome:	The proposed framework outperforms baseline models while maintaining competitive performance in single-turn editing.

Copied to clipboard

Yanting Li, Zhuoyang Jiang, Enyan Dai, Lei Wang, Wen-Cai Ye, Li Liu

MAC-Reasoner: A Multi-Agent Collaborative Framework for Enhancing Logical Reasoning in Large Language Models (2026.findings-acl)

Challenge:	Existing methods to optimize target-directed molecular generation fail to reconcile conflicting objectives without compromising structural validity.
Approach:	They propose a condition-aware discrete diffusion framework that allows for conditional denoising guided by heterogeneous structural and property signals.
Outcome:	The proposed framework improves on structure-conditioned, property-conditioned and dual-conditioned benchmarks in binding affinity, drug-likeness, and success rate.

Copied to clipboard

Yehua Lin, Liping Zheng, Yin Chen

Once Correct, Still Wrong: Counterfactual Hallucination in Multilingual Vision-Language Models (2026.findings-acl)

Challenge:	Existing methods for logical reasoning are limited to a limited number of tasks.
Approach:	They propose a multi-agent framework that constructs a Logic-Augmented Context to enhance LLMs’ reasoning by a translator agent.
Outcome:	The proposed framework improves on three backbone LLMs on four challenging benchmarks and shows consistent and robust improvements over baselines.

Copied to clipboard

Basel Mousi, Fahim Dalvi, Shammur Absar Chowdhury, Firoj Alam, Nadir Durrani

DIFFA-2: A Practical Diffusion Large Language Model for General Audio Understanding (2026.findings-acl)

Challenge:	Existing hallucination benchmarks rarely test this failure mode outside Western contexts and English.
Approach:	They propose a multimodal benchmark built from images spanning 17 MENA countries . they use a CFHR-based test to measure hallucination beyond raw accuracy .
Outcome:	The proposed model is based on images from 17 MENA countries . it measures counterfactual acceptance conditioned on correctly answering the true statement.

Copied to clipboard

Jiaming Zhou, Xuxin Cheng, Shiwan Zhao, Yuhang Jia, Cao Liu, Ke Zeng, Xunliang Cai, Yong Qin

Beyond Marginal Distributions: A Framework to Evaluate the Representativeness of Demographic-Aligned LLMs (2026.findings-acl)

Challenge:	Autoregressive (AR) large audio language models are expensive in data and computation . prior work shows diffusion-based LALMs can improve audio understanding under matched settings .
Approach:	They propose a diffusion-based LALM that upgrades the speech encoder and employs dual semantic and acoustic adapters.
Outcome:	a new model improves over existing autoregressive large language models and is competitive to strong AR models . the proposed model can make use of limited training data and improve inference efficiency . a recent study shows that diffusion-based models can improve audio understanding .

Copied to clipboard

Tristan Williams, Franziska Weeber, Sebastian Padó, Alan Akbik

CORES: Code-Oriented Reasoning for Complex Text-to-SQL and Generalizable TableQA (2026.findings-acl)

Challenge:	Existing work on marginal distributions and model steering fails to account for deeper latent structures that characterise real populations.
Approach:	They propose a framework for evaluating the representativeness of aligned models through multivariate correlation patterns in addition to marginal distributions.
Outcome:	The proposed framework compares two model steering techniques against human responses from the World Values Survey.

Copied to clipboard

Meng Zhang, Ruochun Jin, Yuanxi Peng, Wenjing Yang, Haotian Wang, Liting Sun, Kun Hu, Silin Yang, Zhang Ke-di

Do Not Guess, Verify: Logic-Guided Adaptive Reasoning for Multimodal Misinformation Detection (2026.findings-acl)

Challenge:	Text-to-SQL models struggle with complex analytical tasks such as generating simple SQL queries.
Approach:	They propose a text-to-sql model that leverages Python as a procedural reasoning pivot to enhance both complex SQL generation and tabular reasoning.
Outcome:	The proposed model outperforms baseline models on six text-to-SQL benchmarks by 6.44% on average while maintaining good capability on three tableQA benchmarks.

Copied to clipboard

Kun Huang, Rui Qiu, Xiaoming Li, Salah Uddin

Hi-ZFO: Hierarchical Zeroth- and First-Order LLM Fine-Tuning via Importance-Guided Tensor Selection (2026.findings-acl)

Challenge:	Existing multimodal misinformation detection paradigms rely on passive aggregation of multimodal features and social signals.
Approach:	They propose a verification-oriented framework that integrates large vision–language models into multimodal misinformation detection through explicit rationale-guided reasoning.
Outcome:	The proposed framework outperforms state-of-the-art methods on multimodal misinformation detection benchmarks while significantly reducing computational cost.

Copied to clipboard

Feihu Jin, Ying Tan

MCP-Guard: A Multi-Stage Defense-in-Depth Framework for Securing Model Context Protocol in Agentic AI (2026.findings-acl)

Challenge:	generative tasks require a high degree of exploratory capacity, but zeroth-order methods suffer from slow convergence . generative task-specific methods tend to converge toward local minima, causing noise and inefficient estimation .
Approach:	They propose a framework that synergizes FO precision with exploratory capability of ZO estimation.
Outcome:	The proposed framework synergizes precision of FO gradients with exploratory capability of ZO estimation.

Copied to clipboard

Wenpeng Xing, Zhonghao Qi, Yupeng Qin, Yilin Li, Caini Chang, Jiahui Yu, Changting Lin, Zhenzhen Xie, Meng Han

Half-S: Halving the Scale for Near-Lossless 4-Bit LLM Training (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are vulnerable to jailbreak, authors say . authors propose a robust, layered defense architecture designed for LLM–tool interactions .
Approach:	They propose a robust, layered defense architecture designed for LLM–tool interactions . they propose XCP-Guard, which employs a three-stage detection pipeline .
Outcome:	The proposed model achieves 96.01% accuracy in identifying adversarial prompts . the model is based on a three-stage detection pipeline that balances efficiency with accuracy .

Copied to clipboard

Jinyang Du, Ruihao Gong, Linghan Ai, Zining Wang, Yunke Peng, Yao Wang, Lei Yan, null Wxuefei, Yaoyuan Wang, Jinyang Guo, Dahua Lin, Xianglong Liu

F-Actor: Controllable Conversational Behavior in Full-Duplex Models (2026.findings-acl)

Challenge:	Existing 4-bit training pipelines rely on max-scaling, which causes representation collapse . despite this, there are limitations in the accuracy of 4-bit LLM training .
Approach:	They propose a scaling strategy that uses half-scaling as a hardware-friendly default . they propose fp4 support that allows for a faster scaling of large language models .
Outcome:	The proposed scaling strategy narrows the gap between theoretical optimum and BF16 while maintaining the efficiency benefits of 4-bit training.

Copied to clipboard

Maike Züfle, Ondrej Klejch, Nicholas Sanders, Jan Niehues, Alexandra Birch, Tsz Kin Lam

APEX: Learning Adaptive Priorities for Multi-Objective Alignment in Vision-Language Generation (2026.findings-acl)

Challenge:	Current spoken conversational systems lack customization capabilities, limiting their naturalness and usability.
Approach:	They propose an instruction-following full-duplex conversational speech model that can be trained efficiently under typical academic resource constraints.
Outcome:	The proposed model requires just 2,000 hours of data to be trained under typical academic resource constraints.

Copied to clipboard

Dongliang Chen, Xinlin Zhuang, Junjie Xu, Luojian Xie, Zehui Wang, Jiaxi Zhuang, Haolin Yang, Liang Dou, Xiao He, Xingjiao Wu, Ying Qian

DART: Mitigating Harm Drift in Difference-Aware LLMs via Distill-Audit-Repair Training (2026.findings-acl)

Challenge:	APEX optimizes for text-to-image generation by combining learning potential, conflict penalty, and progress need.
Approach:	They propose an algorithm that stabilizes heterogeneous rewards and dynamically schedules objectives . they propose a method that achieves better Pareto trade-offs across four heterogenous objectives based on P3 Adaptive Priorities .
Outcome:	The proposed algorithm achieves better pareto trade-offs across four heterogeneous objectives while maintaining competitive OCR accuracy.

Copied to clipboard

Ziwen Pan, Zihan Liang, Jad Kabbara, Ali Emami

Experience-Driven Reflective Co-Evolution of Prompts and Heuristics for Autonomous Algorithm Design (2026.findings-acl)

Challenge:	Large language models (LLMs) tuned for safety often avoid acknowledging demographic differences . current safety alignment forces LLMs to default to identity-blindness even when demographic differences are factually correct or contextually justified.
Approach:	They propose a tool to classify whether a correct answer requires recognizing group differences . they use label-conditioned reasoning from a teacher to audit outputs for harm drift cases .
Outcome:	The proposed model improves accuracy and safety on eight benchmarks.

Copied to clipboard

Yihong Liu, Junyi Li, Hongyu Lu, Xin Zhao, Ji-Rong Wen

Information Representation Fairness in Long-Document Embeddings: The Peculiar Interaction of Positional and Language Bias (2026.findings-acl)

Challenge:	Combinatorial optimization has long been dominated by manually engineered heuristics, which require substantial expert intuition and implementation overhead.
Approach:	They propose a framework that couples an island migration model with elite selection to maintain population diversity.
Outcome:	The proposed framework achieves superior accuracy on the Traveling Salesman and Bin Packing Problems.

Copied to clipboard

Elias Schuhmacher, Andrianos Michail, Juri Opitz, Rico Sennrich, Simon Clematide

RA-RRG: Multimodal Retrieval-Augmented Radiology Report Generation with Key Phrase Extraction (2026.findings-acl)

Challenge:	Existing studies show that embedding models exhibit systematic positional and language biases when documents are longer and consist of multiple segments.
Approach:	They propose a permutation-based evaluation framework to quantify embedding biases . they propose an inference-time attention calibration method that redistributes attention more evenly across document positions .
Outcome:	The proposed framework reduces the positional and language biases in embedding models . the proposed framework improves the discoverability of later segments .

Copied to clipboard

Jonggwon Park, Byungmu Yoon, Soobum Kim, Kyoyun Choi

Context-Value-Action Architecture for Value-Driven Large Language Model Agents (2026.findings-acl)

Challenge:	Existing MLLMs are computationally expensive and may produce hallucinated content . RA-RRG uses large language models to generate radiology reports .
Approach:	They propose a retrieval-augmented RRG framework that combines multimodal retrieval with large language models to generate radiology reports.
Outcome:	RA-RRG uses large language models to generate radiology reports . it suppresses hallucinations while maintaining strong report generation performance .

Copied to clipboard

TianZe Zhang, Sirui Sun, Yuhang Xie, Xin Zhang, Zhiqiang Wu, Guojie Song

Uncovering Strategic Egoism Behaviors in Large Language Models (2026.findings-acl)

Challenge:	Existing LLMs exhibit behavioral rigidity, a flaw often masked by the self-referential bias of current "LLM-as-a-judge" evaluations.
Approach:	They propose a Context-Value-Action architecture that decouples action generation from cognitive reasoning via a Value Verifier trained on authentic human data to explicitly model dynamic value activation.
Outcome:	The proposed architecture significantly outperforms baseline models on 1.1 million real-world interaction traces on CVABench.

Copied to clipboard

Yaoyuan Zhang, Zonghao Ying, Aishan Liu, Jian Yang, Tianlin Li, Yaodong Yang, Xianglong Liu

EMA: An Episodic Memory Agent for Efficient and Selective Memory (2026.findings-acl)

Challenge:	Extensive experiments on 9 proprietary LLMs reveal that SE behaviors are widespread . study identifies egoistic decision-making as a risk for large language models .
Approach:	They propose a benchmark to measure egoistic behavior in large language models . they propose toxicity, jailbreak vulnerability and a lightweight mitigation that reinforces situational constraints .
Outcome:	The proposed model has a 67.96% occurrence rate and frequently manifests as manipulative coercion.

Copied to clipboard

Hongyi Lan, Jiaqi Song, Zhengjia Zhong, Hui Li, Hong Liu, Xianming Lin, Rongrong Ji

ChildTalk: A Multi-Dialect Chinese Child Speech Corpus with Full-Length Child–Caregiver Conversations for Speech Recognition (2026.findings-acl)

Challenge:	Existing memory-augmented methods often incorporate full dialog histories without filtering, resulting in information redundancy and inference latency.
Approach:	They propose a framework that abstracts conversational context into Episodic Memory Units (EMUs) they propose EMA, MemDecider and a filtering decision module to reduce noise and improve overall performance.
Outcome:	The proposed framework reduces token consumption by 11.48% while improving performance on two widely-used benchmarks.

Copied to clipboard

Jiaming Zhou, Yujie Guo, Shiwan Zhao, Yao Lu, Jianye Wang, Haoqin Sun, Hui Wang, Yong Qin

SrDetection: A Self-Referential Framework for Data Leakage Detection in Code Large Language Models (2026.findings-acl)

Challenge:	Automatic speech recognition (ASR) for children remains challenging due to developmental variability and the scarcity of high-quality corpora.
Approach:	They propose a large-scale Chinese child speech corpus that contains 112.5 hours of speech from 498 children and 500 caregivers.
Outcome:	The proposed model improves in-domain and cross-domain performance on children's speech.

Copied to clipboard

Shuaimin Li, Liyang Fan, Zeyang li, Zhuoyue Wan, Yufang Lin, Shiwen Ni, Feiteng Fang, Hamid Alinejad-Rokny, Yuanfeng Song, Kun Jing, Chen Jason Zhang, Min Yang

Memo-SQL: Structured Decomposition and Experience-Driven Self-Correction for Training-Free NL2SQL (2026.findings-acl)

Challenge:	Existing methods for evaluating code large language models assume access to proprietary training corpora or use external reference sets with manually tuned, non-generalizable thresholds.
Approach:	They propose a framework for self-referential leakage detection for gray-box and black-box settings.
Outcome:	The proposed framework improves average F1 by 21.52 points in the gray-box setting and 14.46 points in black-box settings over strong baselines.

Copied to clipboard

Yang Zerui, Weichuan Wang, Yanwei Xu, Linqi Song, Yudai Matsuda, Wei Han, Bo Bai

Do We Always Need Query-Level Workflows? Rethinking Agentic Workflow Generation for Multi-Agent Systems (2026.findings-acl)

Challenge:	Existing NL2SQL systems rely on in-context learning with only correct examples . current test-time scaling methods often decompose questions arbitrarily, resulting in poor performance .
Approach:	They propose a structured decomposition and experience-aware self-correction framework for NL2SQL . they build a dynamic memory of successful queries and historical error–fix pairs .
Outcome:	The proposed framework achieves 68.5% execution accuracy on BIRD, setting new state of the art among open, zero-fine-tuning methods.

Copied to clipboard

Zixu Wang, Bingbing Xu, Yige Yuan, Huawei Shen, Xueqi Cheng

Challenge:	Existing approaches generate workflows either at task level or query level, but their relative costs and benefits remain unclear.
Approach:	They propose a query-level workflow generation framework that generates tasks at task level and query level.
Outcome:	The proposed framework reduces token usage by up to 83% compared to existing approaches . it maintains competitive performance with an average degradation of just 0.61% compared with existing approaches across multiple datasets .

How Chain-of-Thought Works? Tracing Information Flow from Decoding, Projection, and Activation (2026.findings-acl)

Copied to clipboard

Hao Yang, Qinghua Zhao, Lei Li, Lingyi Meng, Mengda Yu

A Survey of Retentive Network (2026.findings-acl)

Challenge:	Chain-of-Thought (CoT) prompting significantly enhances model reasoning, yet its internal mechanisms remain poorly understood.
Approach:	They reversely traced information flow across decoding, projection, and activation phases and found that CoT may serve as a decoding space pruner .
Outcome:	The proposed framework can be used to design more efficient and robust prompts.

Copied to clipboard

Haiqi Yang, Zhiyuan Li, Yi Chang, Yuan Wu

Challenge:	Existing studies on the effectiveness of the Retentive Networks have not yet been conducted.
Approach:	They propose a retention mechanism that integrates the inductive bias of recurrent neural networks with the parallelizable training advantages of attention-based models.
Outcome:	The proposed retention mechanism combines the inductive bias of recurrent neural networks with the parallelizable training advantages of attention-based models.

RAG-KT: Cross-platform Explainable Knowledge Tracing with Multi-view Fusion Retrieval Generation (2026.findings-acl)

Copied to clipboard

Zhiyi Duan, Hongyu Yuan, Rui Liu

Conformal Event Prediction with Temporal Knowledge Graph (2026.findings-acl)

Challenge:	Conventional Deep Learning (DL)-based KT models are tied to platform-specific identifiers and latent representations, making them hard to transfer and interpret.
Approach:	They propose a retrieval-augmented paradigm that frames cross-platform KT as reliable context constrained inference with LLMs.
Outcome:	Experiments on three public KT benchmarks show that the proposed paradigm improves accuracy and robustness, and also shows strong performance under cross-platform conditions.

Copied to clipboard

Cheng Hu, Cong Cao, Fangfang Yuan, Diandian Guo, Pin Xu, Yu Liu, Yanbing Liu

Infinite Babble: Inflating 3D Vision-Language Model Inference Overhead via Adversarial Geometric Perturbation (2026.findings-acl)

Challenge:	Current event prediction methods lack rigorous uncertainty quantification, which limits their reliability for decision-making.
Approach:	They propose a conformal prediction framework that applies conformal predictions to event prediction to address this challenge.
Outcome:	The proposed framework guarantees coverage while improving efficiency on three public datasets.

Copied to clipboard

Shuoyang Sun, Jiaxin Hong, Yv Zhang, Kuofeng Gao, Hao Fang, Fan Mo, Bin Chen, Shu-Tao Xia

Uncertainty-Calibrated Elastic Alignment for Multimodal Sentiment Analysis with Missing Modalities (2026.findings-acl)

Challenge:	3D Vision-Language Models (VLMs) are critical cognitive backbone for spatial intelligence, but their reliance on autoregressive decoding introduces a fundamental vulnerability regarding inference efficiency.
Approach:	They propose a framework that triggers computational and economic exhaustion in 3D-VLMs by injecting imperceptible noise that forces the model into a state of pathological verbosity.
Outcome:	The proposed framework amplifies output length and energy consumption by up to 6.45, demonstrating a potent capability to drain system resources.

Copied to clipboard

Kang He, Yuzhe Ding, Rao Fu, Yukang Feng, Kaipeng Zhang, Yiming Liu, Fei Li, Chong Teng, Donghong Ji

On the Representation Geometry of LoRA Model Merging (2026.findings-acl)

Challenge:	Existing methods for multimodal sentiment analysis are often dynamically incomplete.
Approach:	They propose a new uncertainty-calibrated elastic alignment framework to address these issues by employing probabilistic imputation to capture cross-modal ambiguity and leverage the estimated uncertainty to drive elastic alignment.
Outcome:	The proposed framework outperforms state-of-the-art models in multiple benchmarks and consistently outperformed existing models.

Copied to clipboard

Chenyang Lu, Jiaru Li, Jinman Zhao, Xinran Chen, Yining Wang, Renyi Cai, Yuchen Li, Chao He

MCLE-Mol: Empowering LLM with Molecular Comprehension and Low-Cost Continual Evolution for Interpretable Property Prediction (2026.findings-acl)

Challenge:	Existing methods for low-rank Adaptation (LoRA) fine-tuning focus on globally shared structure . combining SVD with CUR improves performance of LoRA model merging .
Approach:	They propose a training-free method that combines SVD and CUR decomposition to improve LoRA merging performance.
Outcome:	The proposed procedure improves on vision and language benchmarks.

Copied to clipboard

Zhili Pu, Lantian Zhang, Hao Duan, Zhixing Zhang, Keyun Zhu, Yongqi Fan, Ruihui Hou, Tong Ruan, Yun Tang

MTR-DuplexBench: Towards a Comprehensive Evaluation of Multi-Round Conversations for Full-Duplex Speech Language Models (2026.findings-acl)

Challenge:	Large language models (LLMs) offer a new paradigm for molecular property prediction (MPP), yet a semantic gap between natural language and molecul representations limits their ability to capture structure–activity relationships (SAR).
Approach:	They propose an ML–LLM–Rule collaborative framework for MPP that injects ML-derived substructure attribution values into LLMs and calibrates them under specific chemical contexts.
Outcome:	The proposed framework outperforms baseline models on multiple benchmark datasets and is highly interpretable.

Copied to clipboard

Zhang He, Wenqian Cui, Haoning Xu, Xiao-Hui Li, Lei Zhu, Haoli Bai, Ma Shaohua, Irwin King

Generating Effective CoT Traces for Mitigating Causal Hallucination (2026.findings-acl)

Challenge:	Existing benchmarks focus on evaluating single-round interactions, neglecting other critical aspects.
Approach:	They propose a benchmark to evaluate full-duplex speech language models in multi-round settings . they segment continuous full-dual dialogues into discrete turns for evaluation .
Outcome:	The proposed benchmark compared full-duplex speech language models with full-dual speech models . the results show that the models perform better in multi-round settings than standard models compared to benchmarks .

Copied to clipboard

Yiheng Zhao, Jun Yan

SAE-FiRE: Enhancing Earnings Surprise Predictions Through Sparse Autoencoder Feature Selection (2026.findings-acl)

Challenge:	Large language models suffer from severe causal hallucination in event causality identification (ECI) there is currently no metric for quantifying causal hallucinonation for small models .
Approach:	They propose to fine-tune large language models with Chain-of-Thought (CoT) traces to mitigate hallucination in smaller models by introducing a new metric, the Causal Hallucinations Rate, which quantifies hallucinosity.
Outcome:	The proposed pipeline reduces causal hallucination in smaller models and improves mean accuracy under intentionally misleading intervention prompts.

Copied to clipboard

Huopu Zhang, Yanguang Liu, Miao Zhang, Zirui He, Mengnan Du

From Implicit Graph Encoding to Explicit Evidence: A Training-Free LLM Framework for Temporal Knowledge Graph Reasoning (2026.findings-acl)

Challenge:	Conference call transcripts contain significant redundancy and industry-specific terminology that creates obstacles for language models.
Approach:	They propose a Sparse Autoencoder for Financial Representation Enhancement framework to extract key information from earnings conference call transcripts and eliminate redundancy.
Outcome:	The proposed method outperforms baselines in analyzing earnings conference call transcripts.

Copied to clipboard

Guo Tang, Ke Cheng, Huiming Fan, Heng Chang, Wenxiang Zheng, Xianhao Ou, Junjia Xiang, Ming Liu, Yujun Zhou, Li Lanyu, Bing Qin

LLM-Driven Multi-Perspective Location Completion for Next Location Prediction (2026.findings-acl)

Challenge:	Existing Large Language Models (LLMs) struggle with implicit modality alignment and suboptimal graph linearization.
Approach:	They propose a training-free, test-time adaptive framework that reframes TKG prediction as explicit evidence-driven reasoning.
Outcome:	ExE-LLM outperforms fully trained graph neural networks on four benchmarks . it achieves SOTA performance in inductive settings, significantly outperforming fully trained neural networks .

Copied to clipboard

Pengxiang Lan, Enneng Yang, Yuliang Liang, Jianzhe Zhao, Linying Jiang, Guibing Guo

Challenge:	Existing methods assume that check-in data is complete, overlooking the subjective nature of user behavior, leading to inaccurate capture of user preferences.
Approach:	They propose a framework that uses spatial coordinates to augment location completion by transforming geographic coordinates into text.
Outcome:	The proposed framework outperforms state-of-the-art methods on three real-world datasets.

WikiSeeker: Rethinking the Role of Vision-Language Models in Knowledge-Based Visual Question Answering (2026.findings-acl)

Copied to clipboard

Yingjian Zhu, Xinming Wang, Kun Ding, Ying Wang, Bin Fan, Shiming Xiang

Understanding Conflicts in Multi-Objective Alignment through Reward Consistency (2026.findings-acl)

Challenge:	Existing methods for Knowledge-Based Visual Question Answering rely on images as the retrieval key, and often overlook or misplace the role of Vision-Language Models (VLMs)
Approach:	They propose a multi-modal RAG framework that assigns VLMs two specialized agents: a Refiner and an Inspector.
Outcome:	Experiments on EVQA, InfoSeek, and M2KR show that the proposed framework achieves state-of-the-art performance with significant improvements in both retrieval accuracy and answer quality.

Copied to clipboard

Zhihao Xu, Yongqi Tong, Xin Zhang, Jun Zhou, Xiting Wang

MetaMem: Evolving Meta-Memory for Knowledge Utilization through Self-Reflective Symbolic Optimization (2026.findings-acl)

Challenge:	Existing training pipelines still face alignment conflicts where optimizing for one objective degrades performance on others.
Approach:	They propose a reward-based criterion that approximates alignment conflicts via reward models.
Outcome:	The proposed framework improves harmlessness and helpfulness scores by 23.07% over the vanilla dataset.

Copied to clipboard

Haidong Xin, Xinze Li, Zhenghao Liu, Yukun Yan, Shuo Wang, Cheng Yang, Yu Gu, Ge Yu, Maosong Sun

GeoLAN: Geometric Learning of Latent Explanatory Directions in Large Language Models (2026.findings-acl)

Challenge:	Existing memory systems can support long-horizon human-LLM interactions by persisting historical interactions beyond limited context windows.
Approach:	They propose a framework that augments memory systems with a self-evolving meta-memory . meta-meso is iteratively distilling transferable knowledge utilization experiences . results show MetaMem outperforms strong baselines by over 3.6% .
Outcome:	The proposed framework outperforms baselines by over 3.6% in the long-horizon human-LLM interaction.

Copied to clipboard

Tianyu Pan, Damon L. Woodard

SepSeq: A Training-Free Framework for Long Numerical Sequence Processing in LLMs (2026.findings-acl)

Challenge:	Large language models lack transparency and are often unable to explain causal relationships .
Approach:	They propose a training framework that treats token representations as geometric trajectories and applies stickiness conditions to the Kakeya Conjecture.
Outcome:	The proposed training framework maintains task accuracy while improving geometric metrics and reducing fairness biases.

Copied to clipboard

Jie Sun, Yu Liu, Lu Han, Qiwen Deng, Xiang Shu, Yang Xiao, Lintao Ma, Xingyu Lu, Jun Zhou, Pengfei Liu, Jiancan Wu, Xiang Wang

Ghost in the Shell: Synonym-Aware Logit Shaping Fingerprint for Copyright Protection of Large Vision-Language Models (2026.findings-acl)

Challenge:	Existing large-scale large-context models suffer from performance degradation when processing long numerical sequences.
Approach:	They propose a framework to mitigate attention dispersion by strategically inserting separator tokens into the model to recalibrat attention to local segments while preserving global context.
Outcome:	The proposed framework improves accuracy and reduces inference token consumption by 16.4% on 9 widely-adopted LLMs.

Copied to clipboard

Xiaofan Zheng, Xinghao Wang, Xiaojun Wan

Look Twice before You Leap: A Rational Framework for Localized Adversarial Text Anonymization (2026.findings-acl)

Challenge:	Existing fingerprinting methods for large vision-language models rely on backdoors to elicit abnormal outputs, but direct distortion of the model’s original outputs compromises modality alignment and degrades multimodal capabilities.
Approach:	They propose to embed a robust fingerprint while preserving the original normal outputs of the model.
Outcome:	The proposed fingerprint maintains multimodal performance and substantially enhances fingerprint robustness.

Copied to clipboard

Donghang Duan, Xu Zheng, Yuefeng He, Chong Mu, Leyi Cai, Lizong Zhang

K-MetBench: A Multi-Dimensional Benchmark for Fine-Grained Evaluation of Expert Reasoning, Locality, and Multimodality in Meteorology (2026.findings-acl)

Challenge:	Existing LLMs rely on remote API services, which creates privacy paradoxes and suboptimal solutions with severe utility collapse.
Approach:	They propose a localized and training-free framework with an Attacker-Arbitrator-Anonymizer architecture that allows attackers to filter out ghost leaks.
Outcome:	The proposed framework achieves superior privacy-utility trade-off compared to strong baselines.

Copied to clipboard

Soyeon Kim, Cheongwoong Kang, Myeongjin Lee, Eun-Chul Chang, Lee Jaedeok, Jaesik Choi

V-GameGym: Visual Game Generation for Code Large Language Models (2026.findings-acl)

Challenge:	Existing evaluation frameworks for large language models for domain specific tasks are coarse and do not provide a multidimensional evaluation of a model's ability to interpret domain specific data.
Approach:	They propose a diagnostic benchmark grounded in national qualification exams that exposes critical gaps across four dimensions: expert visual reasoning of charts, logical validity via expert-verified rationales, Korean-specific geo-cultural comprehension, and fine-grained domain analysis.
Outcome:	The proposed model outperforms global models in local contexts, demonstrating that parameter scaling alone cannot resolve cultural dependencies.

Copied to clipboard

Wei Zhang, Jian Yang, Renshuai Tao, Linzheng Chai, Shuyue Guo, Jiajun Wu, Xiaoming Chen, Ganqu Cui, Ning Ding, Xander Xu, HU Wei, Bowen Zhou

UCoder: Unsupervised Code Generation by Internal Probing of Large Language Models (2026.findings-acl)

Challenge:	Existing code-related benchmarks focus on single modality rather than visual game development.
Approach:	They propose a multimodal benchmark for evaluating code large language models in visual game generation that integrates a clustering-based curation methodology and a pipeline for visual code synthesis.
Outcome:	The proposed framework assesses code generation and visual game generation using a sandbox environment.

Copied to clipboard

Jiajun Wu, Jian Yang, Wei Zhang, Linzheng Chai, Yuchi Ma, Ensheng Shi, Yuqing Ma, Zhoujun Li, Xianglong Liu

Deciphering Cultural Representations in Large Language Models via Sparse Autoencoders (2026.findings-acl)

Challenge:	Large language models (LLMs) have demonstrated remarkable capabilities in code generation tasks, but their effectiveness relies on supervised training with extensive labeled data and computational resources.
Approach:	They propose an unsupervised method that leverages Internal Probing of Large language models for Code generation without any external corpus, even unlabeled code snippets.
Outcome:	The proposed method can achieve competitive performance compared to supervised approaches while reducing the dependency on labeled data and computational resources.

Copied to clipboard

Chenye Zou, Difan Jiao, Lijie Hu

Fin-Bias: Comprehensive Evaluation for LLM Decision-Making under human bias in Finance Domain (2026.findings-acl)

Challenge:	Prior work has identified so-called cultural neurons, but individual neurons are often polysemous, conflating abstract cultural knowledge with surface-level lexical cues due to superposition.
Approach:	They apply Sparse Autoencoders to decompose LLM activations into sparse, interpretable feature representations that disentangle culturally selective features.
Outcome:	The proposed model disentangles culturally selective features from paraphrasing and task formats, indicating abstraction beyond lexical correlations.

Copied to clipboard

Xiaoyu Hu, Jinman Zhao

Duplicate-Aware Controlled Code Generation: Enhancing Copyright Protection with Targeted Reordering Beam Search in LLMs (2026.findings-acl)

Challenge:	Existing benchmarks for large language models (LLMs) are limited to small sample and fail to demonstrate LLM susceptibility to context with potential human bias.
Approach:	They propose a benchmark for evaluating LLM investment decision-making when faced with uncertainty and possible human-biased opinions.
Outcome:	The proposed model can herd the explicit bias in context and even exceed human performance in predicting future stock return.

Copied to clipboard

Junbo Fu, Guoshuai Zhao, Linkang Yang, Yunqi Mi, Xueming Qian

Graph-Based Chain-of-Thought Pruning for Reducing Redundant Reflections in Reasoning LLMs (2026.findings-acl)

Challenge:	Experimental results demonstrate that TRBS effectively reduces verbatim repetition while maintaining functional adequacy.
Approach:	They propose a plug-and-play decoding method that dynamically reorders beam candidates to reduce direct copying.
Outcome:	The proposed method reduces verbatim repetition while maintaining functional adequacy on a multi-language code generation benchmark.

Copied to clipboard

Hongyuan Yuan, Xinran He, Run Shao, Bolei He, Xianwei Xue, Mengke Chen, Qiutong Pan, Haiwei Wang, Haifeng Li

Too Good to be Bad: On the Failure of LLMs to Role-Play Villains (2026.findings-acl)

Challenge:	Extending CoT through RL can induce undesirable thinking patterns such as overthinking . prior work has focused on inefficient reflection, which manifests in two problematic patterns: Indiscriminate Reflection and Repetitive Reflectione .
Approach:	They propose a graph-based approach to optimize CoT by pruning each linear CoT into a directed acyclic graph with explicit dependency edges.
Outcome:	The proposed approach reduces the average reasoning tokens by 42% while maintaining or improving accuracy.

Copied to clipboard

Zihao Yi, Qingxuan Jiang, Ruotian Ma, Xingyu Chen, Qu Yang, Mengru Wang, Fanghua Ye, Ying Shen, Zhaopeng Tu, Xiaolong Li, Liefeng Bo

Challenge:	Large Language Models (LLMs) are increasingly tasked with creative generation, but their ability to portray non-prosocial, antagonistic personas remains largely unexamined.
Approach:	They propose a moral alignment benchmark to test the safety of large language models . they find that models struggle with traits directly antithetical to safety principles .
Outcome:	The proposed model fails to accurately portray morally ambiguous or villainous characters . the model fails most with traits directly antithetical to safety principles .

AudioPrivacy: Parallel Audio Dataset for Speaker Profiling with Diverse Audio Types and Rich Attributes (2026.findings-acl)

Copied to clipboard

Jiabei He, Yanzhe Zhang, Jiaming Zhou, Hui Wang, Haoqin Sun, Yong Qin

Federated LoRA Fine-Tuning with Pipelined Error-Mitigated Aggregation and Matrix-Wise Freezing (2026.findings-acl)

Challenge:	Speech signals convey abundant speaker-related metadata, yet current privacy research focuses on identity-centric voiceprint protection, leaving sensitive Speaker Attribute Privacy (SAP) underexplored.
Approach:	They propose a large-scale Chinese dataset to evaluate speaker-related privacy leakage . the dataset includes 227.3 hours of audio from 1,000 speakers .
Outcome:	The proposed model systematically evaluates speaker-related privacy leakage in everyday scenarios.

Copied to clipboard

Haoran Wang, Xiong Wang, Yuqing Li, Jing Chen, Junyi Zhang, Nan Yan, Kun He, Wei Wang

RTCFake: Speech Deepfake Detection in Real-Time Communication (2026.findings-acl)

Challenge:	Existing methods for fine-tuning large language models often suffer from biased model aggregation and are hindered by significant communication and computation burden.
Approach:	They propose a Federated low-rank adaptation system for large language models that leverages pipelined error-mitigated model aggregation and adaptive matrix-wise parameter freezing to mitigate aggregations.
Outcome:	The proposed system improves time-to-target by 2.17-8.48 on real-world datasets.

Copied to clipboard

Jun Xue, Zhuolin Yi, Yihuan Huang, Yanzhen Ren, Yujie Chen, Cunhang Fan, Zicheng Su, Yongcheng Zhang, Bo Cai

Progress Ratio Embeddings: An Impatience Signal for Robust Length Control in Neural Text Generation (2026.findings-acl)

Challenge:	Existing detection studies focus on offline simulations and struggle to cope with complex distortions introduced during RTC transmission.
Approach:	They propose a large-scale speech deepfake dataset tailored for RTC scenarios . the dataset is constructed by transmitting speech through multiple social media and conferencing platforms .
Outcome:	The proposed dataset is constructed by transmitting speech through multiple mainstream social media and conferencing platforms, enabling precise pairing between offline and online speech.

Copied to clipboard

Ivanhoé Botcazou, Tassadit Amghar, Sylvain Lamprier, Frédéric Saubion

LeakDojo: Decoding the Leakage Threats of RAG Systems (2026.findings-acl)

Challenge:	Modern neural language models achieve high accuracy in text generation, yet precise control over generation length remains underdeveloped.
Approach:	They propose a method to provide robust length control using Reverse Positional Embeddings.
Outcome:	The proposed method provides stable length fidelity without degrading text accuracy . the proposed method generalizes well to unseen target lengths .

Copied to clipboard

Maosen Zhang, Jianshuo Dong, Lu Boting, Li Wenyue, Xiaoping Zhang, Tianwei Zhang, Han Qiu

CodeWiki: Evaluating AI’s Ability to Generate Holistic Documentation for Large-Scale Codebases (2026.findings-acl)

Challenge:	Existing studies have failed to assess RAG leakage risks for large language models . constructing and maintaining highquality RAG knowledge databases has become increasingly costly .
Approach:	They propose a framework for controlled evaluation of RAG leakage using query generation and adversarial instructions.
Outcome:	The proposed framework compares six existing attacks across fourteen LLMs, four datasets, and diverse RAG systems.

Copied to clipboard

Anh Nguyen Hoang, Minh Le-Anh, Bach Le, Nghi D. Q. Bui

Challenge:	Existing automated methods struggle to capture rich semantic dependencies and architectural structure.
Approach:	They propose a framework for automated repository-level documentation across seven programming languages.
Outcome:	The proposed framework outperforms the closed-source DeepWiki benchmark by 68.79% and is open source to support future research.

ARISE: An Adaptive Resolution-Aware Metric for Test-Time Scaling Evaluation in Large Reasoning Models (2026.findings-acl)

Copied to clipboard

Zhangyue Yin, Qiushi Sun, Zhiyuan Zeng, Zhiyuan Yu, Qipeng Guo, Xuanjing Huang, Xipeng Qiu

Breaking the Static Graph: Context-Aware Traversal for Graph-Based RAG (2026.findings-acl)

Challenge:	Existing evaluation methods for test-time scaling are limited.
Approach:	They propose an adaptive resolution-aware scaling evaluation metric specifically designed to assess the test-time scaling effectiveness of large reasoning models.
Outcome:	The proposed metric provides a reliable and fine-grained measurement of test-time scaling capabilities, revealing significant variations in scaling efficiency across models.

Copied to clipboard

Kwun Hang Lau, Fangyuan Zhang, Boyu Ruan, Yingli Zhou, Qintian Guo, Ruiyuan Zhang, Xiaofang Zhou

POP: Prefill-Only Pruning for Efficient Large Model Inference (2026.findings-acl)

Challenge:	Recent advances in RAG focus on capturing multi-hop dependencies, but static Graphs fail to retrieve complete evidence chain.
Approach:	They propose a structure-aware approach to capture multi-hop dependencies using Knowledge Graphs and Personalized PageRank to capture semantic drift.
Outcome:	Experiments show that CatRAG outperforms state-of-the-art approaches . the proposed approach achieves substantial improvements in reasoning completeness .

Copied to clipboard

Junhui He, Zhihui Fu, Jun Wang, Qingan Li

Guardian-as-an-Advisor: Advancing Next-Generation Guardian Models for Trustworthy LLMs (2026.findings-acl)

Challenge:	Existing structured pruning methods suffer from significant accuracy degradation . Existing pruning methods are expensive and require specialized hardware and kernels to perform .
Approach:	They propose a stage-agnostic pruning approach that overlooks asymmetric roles between prefill and decode stages.
Outcome:	The proposed pruning approach achieves 1.37 speedup in prefill latency with minimal performance loss.

Copied to clipboard

Yue Huang, Haomin Zhuang, Jiayi Ye, Han Bao, Yanbo Wang, Hang Hua, Siyuan Wu, Pin-Yu Chen, Xiangliang Zhang

StructBreak: Structural Cognitive Overload-Induced Safety Failures in MLLMs (2026.findings-acl)

Challenge:	prevailing taxonomies neglect robustness and honesty, yielding safer-on-paper but less useful systems.
Approach:	They propose a soft-gating pipeline where a guardian predicts a binary risk label plus a concise explanation and prepends this advice to the original query for re-inference.
Outcome:	The proposed model maintains safety while reducing over-refusal.

Copied to clipboard

Yang Luo, Liu Xinran, TianTian Ji, Zhiyi Yin, Lingyun Peng, Shuyu Li

Role-Sensitive Neurons: A Neuron-Level Gain Control Mechanism for Confidence Steering (2026.findings-acl)

Challenge:	Prior work focused on typographic and pixel-level perturbations, leaving the study of SCO unexplored.
Approach:	They propose a framework that exploits MLLMs' diagrammatic reasoning capabilities to bypass safety guardrails.
Outcome:	The proposed framework exploits the model's reasoning capabilities to bypass safety guardrails.

Copied to clipboard

Peiwen Huang, Chih-Hao Hsu, Tzu-Hung Huang, Shou-De Lin

Challenge:	Large language models (LLMs) exhibit striking behavioral flexibility.
Approach:	They propose to identify a sparse sub-network of Role-Sensitive Neurons (RSNs) that governs the transition from hesitation to action.
Outcome:	The proposed framework allows precise regulation of abstention behavior by intervention on this subspace.

StreamingEval: A Unified Evaluation Framework towards Realistic Streaming Video Understanding (2026.findings-acl)

Copied to clipboard

Tang Guowei, Tianwen Qian, Huanran Zheng, Wang Yifei, Xiaoling Wang

CogGen: A Cognitively Inspired Recursive Framework for Deep Research Report Generation (2026.findings-acl)

Challenge:	Existing research on streaming video understanding focuses on isolated aspects of visual understanding, but ignores practical deployability under realistic resource constraints.
Approach:	They propose a framework to evaluate streaming video understanding capabilities under realistic constraints.
Outcome:	StreamingEval benchmarks offline and online video models under a standardized protocol . it evaluates visual encoding efficiency, text decoding latency and task performance .

Copied to clipboard

Kuo Tian, Pengfei Sun, Zhen Wu, Junran Ding, Xinyu Dai

Debate to Align: Reliable Entity Alignment through Two-Stage Multi-Agent Debate (2026.findings-acl)

Challenge:	Existing approaches to deep research report generation rely on rigid predefined linear workflows, which cause error accumulation and limit in-depth multimodal fusion and report quality.
Approach:	They propose a Cognitively inspired recursive framework for deep research report Generation that simulates cognitive writing and abstract visual representation (AVR) they also propose CLEF, a cognitive load evaluation framework, and a benchmark from our world in data.
Outcome:	The proposed framework achieves state-of-the-art among open-source systems, surpassing Gemini Deep Research.

Copied to clipboard

Cunda Wang, Ziying Ma, Po Hu, Weihua Wang, Feilong Bao

Exploring Multilingual Pre-trained Language Model for Aspect-based Sentiment Analysis (2026.findings-acl)

Challenge:	Entity alignment (EA) aims to identify entities referring to the same real-world object across different knowledge graphs (KGs).
Approach:	They propose a reliable EA framework based on multi-agent debate that improves embedding quality and introduces a two-stage multi-role debate mechanism to enhance reliability.
Outcome:	The proposed framework improves embedding quality and the reasoning capability of LLMs while enabling more efficient debate-based reasoning.

Copied to clipboard

Ye Wang, Ruijun Jiang, Zhongqing Wang, Guodong Zhou

Mid-Think: Training-Free Intermediate-Budget Reasoning via Token-Level Triggers (2026.findings-acl)

Challenge:	Aspect-based sentiment analysis studies have focused on English datasets, but labeled data is scarce.
Approach:	They propose a multilingual pre-trained language model that leverages bilingual pre-training to leverage aspects-based sentiment analysis.
Outcome:	The proposed model outperforms state-of-the-art models across multiple languages.

Copied to clipboard

Van Yang, Shouren Wang, Debargha Ganguly, Xinpeng Li, Chaoda Song, Vikash Singh, Vipin Chaudhary, Xiaotian Han

SSR-Zero: Simple Self-Rewarding Reinforcement Learning for Machine Translation (2026.findings-acl)

Challenge:	Explicit /think> tags are used to expose intermediate reasoning and enable hybrid thinking behaviors.
Approach:	They propose a training-free prompting format that combines these triggers to achieve intermediate-budget reasoning, outperforming fixed-token and prompt-based baselines in terms of the accuracy–length trade-off.
Outcome:	The proposed method outperforms fixed-token and prompt-based prompts in accuracy–length trade-offs while improving Qwen3-8B on AIME from 69.8% to 72.4% and on GPQA from 58.5% to 61.1%.

Copied to clipboard

Wenjie Yang, Mao Zheng, Mingyang Song, Zheng Li, Sitong Wang

HisDoc-OCR: Restoring Visual Grounding in MLLMs for Chinese Historical Document OCR (2026.findings-acl)

Challenge:	Large language models (LLMs) have demonstrated remarkable capabilities in machine translation, but most MT-specific LLMs rely heavily on external supervision during training.
Approach:	They propose a reinforcement learning framework for machine translation that is reference-free and relies solely on self-judging rewards.
Outcome:	The proposed framework outperforms existing LLMs and larger general LLM models on English Chinese translation benchmarks and performs competitively with leading closed-source systems.

Copied to clipboard

Jiahuan Cao, Yongxin Shi, Zeyu Shan, Zhengyang Lu, Lianwen Jin

One LLM Does Not Simulate All Students: Ability-Aware Student Simulation via Cognitive Diagnosis Guided LLM Assignment (2026.findings-acl)

Challenge:	Despite multimodal large language models' strong performance on modern document OCR, their application to historical Chinese texts suffers from severe hallucinations, character fabrication, uncontrolled repetition, and semantic drift.
Approach:	They propose a multimodal large language model which restores visual grounding through three synergistic strategies: Layout Injection, First-Occurrence Boost, Self-Distilled Attention Focusing and HisDoc-OCR.
Outcome:	The proposed model outperforms general-purpose and OCR-specific models on Chinese historical documents.

Copied to clipboard

Huixing Que, Qi Liu, Weibo Gao, Zhenya Huang

Interactive Learning for LLM Reasoning (2026.findings-acl)

Challenge:	Existing methods rely on a single high-capacity LLM to represent an entire population of diverse learners.
Approach:	They propose an ability-aware student simulation framework that matches students with appropriate LLM backbones through cognitive alignment.
Outcome:	The proposed framework significantly reduces simulation bias and outperforms single-model baselines across the entire proficiency spectrum.

Copied to clipboard

Hehai Lin, Shilei Cao, Sudong Wang, Haotian Wu, Minzhi Li, Linyi Yang, Juepeng Zheng, Chengwei Qin

Style Amnesia: Investigating Speaking Style Degradation and Mitigation in Multi-Turn Spoken Language Models (2026.findings-acl)

Challenge:	Existing multi-agent learning approaches foster collaboration among Large Language Models (LLMs) yet they still rely on re-executing the MAS during inference.
Approach:	They propose a co-learning framework that integrates Dynamic Interaction and Perception Calibration to enhance LLMs' independent problem-solving ability.
Outcome:	The proposed framework integrates Dynamic Interaction and Perception Calibration to improve LLMs' independent problem-solving ability.

Copied to clipboard

Yu-Xiang Lin, Cheng-Han Chiang, Hung-yi Lee

NeuroSym-Cal: Bridging the Reasoning-Execution Gap in Code Generation via Hierarchical Calibration (2026.findings-acl)

Challenge:	Unlike textonly large language models (LLMs), SLMs integrate audio encoders and vocoders to support end-to-end speech understanding and generation.
Approach:	They evaluate three proprietary and two open-source SLMs and show that none of them can maintain a consistent speaking style when instructed to do so.
Outcome:	The proposed models cannot maintain a consistent speaking style after several turns of interaction, but can recall the style instruction when prompted in later turns, but fail to express it.

Copied to clipboard

Peiyang Liu, Yining Wang, Youru Li, Long Li, Zhi Cai, Wei Ye

GOBench: Stage-Wise Diagnostics and the Visual Paradox in Multimodal Graph Optimization (2026.findings-acl)

Challenge:	Existing calibration methods rely on the assumption that consensus implies correctness . Existing methods fail under systematic errors, leading to miscalibrated high-confidence predictions.
Approach:	They propose a hierarchical calibration framework that measures confidence at two levels . they propose sensitivity analysis to measure local curvature of deductive process .
Outcome:	The proposed framework de-saturates overconfident errors and improves selective generation performance on OOD benchmarks.

Copied to clipboard

Yinghao Chen, Wantong Xie, Shuli Zeng, Sijia Zhang, Xiaotian Pan, Feng Wu, Xiangyang Li

Sparse Growing Transformer: Training-Time Sparse Depth Allocation via Progressive Attention Looping (2026.findings-acl)

Challenge:	Existing benchmarks fail to represent multimodal problem specifications, score outcomes only and cannot localize where failures occur along the modeling pipeline.
Approach:	They propose a Graph Optimization benchmark that aligns multiple modalities with solver-derived oracles and a diagnostic protocol that evaluates intermediate artifacts as well as end results.
Outcome:	Graph Optimization benchmark (GOBench) evaluates intermediate artifacts as well as end results . vision reliably increases inference cost, while reliability impact is regime-dependent . current benchmarks fail to represent multimodal problem specifications, fail to localize failures .

Copied to clipboard

Yao Chen, Yilong Chen, Yinqi Yang, Junyuan Shang, Zhenyu Zhang, Zefeng Zhang, Shuaiyi Nie, Shuohuan Wang, Yu Sun, Hua Wu, Haifeng Wang, Tingwen Liu

Computational Narrative Understanding for Expressive Text-to-Speech (2026.findings-acl)

Challenge:	Existing approaches to increasing effective depth of LLMs rely on parameter reuse, extending computation through recursive execution.
Approach:	They propose a training-time sparse depth allocation framework that progressively increases depth for a small subset of parameters as training evolves.
Outcome:	The proposed model outperforms existing approaches to increasing the effective depth of language models while reducing training FLOPs overhead from approximately 16–20% to only 1–3% relative to a standard Transformer backbone.

Copied to clipboard

Gaspard Michel, Elena V. Epure, Christophe Cerisara

DataSage: Multi-agent Collaboration for Insight Discovery with External Knowledge Retrieval, Multi-role Debating, and Multi-path Reasoning (2026.findings-acl)

Challenge:	Recent advances in text-to-speech systems have been driven by large, multi-domain speech corpora.
Approach:	They propose a large-scale 5.3K hours of expressive speech drawn from character quotations . they fine-tune a flow-matching model and train from scratch .
Outcome:	The proposed model improves expressivity and intelligibility while training from scratch improves expressiveness of an autoregressive model.

Copied to clipboard

Xiaochuan Liu, Yuanfeng Song, Xiaoming Yin, Xing Chen

Eliminating Out-of-Domain Recommendations in LLM-based Recommender Systems: A Unified View (2026.findings-acl)

Challenge:	Existing data insight agents fail to deliver satisfactory results due to insufficient utilization of domain knowledge, shallow analytical depth, and error-prone code generation.
Approach:	They propose a novel multi-agent framework that incorporates external knowledge retrieval to enrich the analytical context, a multi-role debating mechanism to simulate diverse analytical perspectives and deepen analytical depth, and multi-path reasoning to improve the accuracy of the generated code and insights.
Outcome:	Extensive experiments on InsightBench show that DataSage outperforms existing data insight agents across all difficulty levels, improving by 7.5% and 13.9% respectively in insight-level and summary-level metrics.

Copied to clipboard

Hao Liao, Jiwei Zhang, Jianxun Lian, Wensheng Lu, Mingqi Wu, null Shuowangg, Yong Zhang, Yitian Huang, Mingyang Zhou, Rui Mao

Evian: Towards Explainable Visual Instruction-tuning Data Auditing (2026.findings-acl)

Challenge:	Existing approaches to reduce OOD recommendations fall into three grounding paradigms: retrieval, constrained generation and discrete item tokenizer generation.
Approach:	They propose a framework that instantiates three grounding paradigms under a single architecture . embedding-based retrieval, constrained generation and discrete item-tokenizer methods are implemented .
Outcome:	The proposed framework eradicates OOD recommendations across all variants and achieves state-of-the-art accuracy compared to strong ID-based and LLM-based baselines.

Copied to clipboard

Zimu Jia, Mingjie Xu, Andrew Estornell, Jiaheng Wei

Iterative Self-Correction for Text-Driven Person Re-Identification with Large Vision-Language Models (2026.findings-acl)

Challenge:	Existing data filtering methods rely on coarse-grained scores that lack granularity to identify nuanced semantic flaws.
Approach:	They propose a "Decomposition-then-Evaluation" paradigm that breaks model responses into constituent cognitive components.
Outcome:	The proposed model outperforms models trained on larger datasets in three key areas . the authors show that Logical Coherence is the most critical factor in data quality evaluation .

Copied to clipboard

Guijin Luo, Zequn Xie, Sihang Cai, Chuxin Wang, Zhou Zhao, Yixuan Tang

Breaking Contextual Inertia: Reinforcement Learning with Single-Turn Anchors for Stable Multi-Turn Interaction (2026.findings-acl)

Challenge:	Existing methods for Person Re-Identification (ReID) adopt a static "one-pass" paradigm, converting images to text once for retrieval.
Approach:	They propose a framework that reformulates ReID as an iterative "Think-and-Refine" process.
Outcome:	The proposed framework outperforms state-of-the-art methods in complex occlusion scenarios.

Copied to clipboard

Xingwu Chen, Zhanqiu Zhang, Steven Y. Guo, Difan Zou

Label Words as Local Task Vectors in In-Context Learning (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have demonstrated remarkable capabilities when provided with full information in a single turn, yet they exhibit substantial vulnerability in multi-turn interactions.
Approach:	They propose a generalizable training approach to stabilize multi-turn interactions by leveraging the model's intrinsic single-turn capabilities as stable internal anchors.
Outcome:	The proposed approach outperforms fine-tuning and abstention-based methods and exhibits strong cross-domain generalization.

Copied to clipboard

Bowen Zheng, Ming Ma, Zhongqiao Lin, Tianming Yang

CoGR-MoE: Concept-Guided Expert Routing with Consistent Selection and Flexible Reasoning for Visual Question Answering (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have demonstrated remarkable abilities, one of the most important being in-context learning (ICL).
Approach:	They hypothesized that the network creates a task vector in specific positions during ICL, which can be computed by averaging across the dataset.
Outcome:	The proposed model can achieve zero-shot performance with dummy inputs comparable to few-shot learning by patching the global task vector.

Copied to clipboard

Xiyin Zeng, Yi Lu, Hao Wang

Which Pieces Does Unigram Tokenization Really Need? (2026.findings-acl)

Challenge:	Recent work has begun to address routing instability in VQA models by grouping similar concepts or routing based on examples.
Approach:	They propose a Concept-Guided Routing framework which incorporates semantics of the answer options to guide expert selection in the training phase.
Outcome:	The proposed framework delivers strong performance across multiple VQA tasks, demonstrating the effectiveness of the proposed framework.

Copied to clipboard

Sander Land, Yuval Pinter

Step Potential Advantage Estimation: Harnessing Intermediate Confidence and Correctness for Efficient Mathematical Reasoning (2026.findings-acl)

Challenge:	Despite its theoretical elegance, its implementation in practice is complex, limiting its adoption to SentencePiece.
Approach:	They propose a Unigram-based probabilistic alternative to the greedy heuristics of Byte-Pair Encoding that is based on C++.
Outcome:	The proposed algorithm is remarkably robust to hyperparameter choices and can be simplified to reduce computational costs.

Copied to clipboard

Fei Wu, Zhenrong Zhang, Qikai Chang, Jianshu Zhang, Quan Liu, Jun Du

When depth is redundant: Efficient transformer-based speech anti-spoofing (2026.findings-acl)

Challenge:	Existing approaches to RLVR provide sparse supervision since reward arrives only after the full generation is complete.
Approach:	They propose a step-level reward system that extracts confidence and correctness and combines them into a Step Potential signal that explicitly estimates reasoning state at each step.
Outcome:	The proposed method outperforms existing methods on multiple benchmarks and improves accuracy while reducing response length.

Copied to clipboard

Hoan My Tran, Damien Lolive, Aghilas Sini, Arnaud Delhay, Pierre-Francois Marteau, David Guennec

GCoT-Decoding: Unlocking Deep Reasoning Paths for Universal Question Answering (2026.findings-acl)

Challenge:	Existing anti-spoofing countermeasures exhibit limited generalization to unseen spoof attacks, especially in out-of-domain evaluation settings.
Approach:	They propose a training strategy that aligns shallow and intermediate representations with those of the final transformer layer for speech deepfake detection.
Outcome:	The proposed model improves robustness to unseen spoofing attacks and enhances out-of-domain generalization over strong baselines.

Copied to clipboard

Guanran Luo, Wentao Qiu, Zhongquan Jian, Meihong Wang, Qingqiang Wu

PerMemSafe: Benchmarking Implicit Personalized Safety of Long Horizon Self-Evolving Agents (2026.findings-acl)

Challenge:	Recent work on Chain-of-Thought reasoning requires manual prompts to guide the model.
Approach:	They propose a general decoding strategy that generates CoT-style reasoning paths without prompts.
Outcome:	The proposed method maintains strong performance on fixed and free QA tasks and achieves significant improvements on free qa.

Copied to clipboard

Hengyu An, Minxi Li, Naen Xu, Chunyi Zhou, Xiaogang Xu, Tianyu Du, Jinbao Li, Shouling Ji

TagRAG: Tag-guided Hierarchical Knowledge Graph Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	Existing self-evolving agents have a low safety rate in long-horizon interactions . however, this reliance on context-independent safety evaluations is insufficient .
Approach:	They propose a framework that explicitly models personalized risk inference and memory evolution.
Outcome:	The proposed framework improves implicit personalized safety by 23.8% over prior frameworks while maintaining helpfulness in long-horizon interactions.

Copied to clipboard

Wenbiao Tao, Xinyuan Li, Yunshi Lan, Weining Qian

Danger Depends on the Mind: A Theory-of-Mind Grounded Dataset and Model for Context-Dependent Dangerous Speech (2026.findings-acl)

Challenge:	Existing approaches to retrieval-augmented generation rely on fragment-level retrieval . GraphRAG suffers from inefficiencies in information extraction and costly resource consumption .
Approach:	They propose a tag-guided hierarchical knowledge graph RAG framework for efficient global reasoning and scalable graph maintenance.
Outcome:	GraphRAG achieves an average win rate of 78.36% on a dataset spanning agriculture, computer science, law, and cross-domain settings compared with baselines .

Copied to clipboard

Yuanchen Shi, Longyin Zhang, Guodong Zhou, Fang Kong

One Token Is Enough: Improving Diffusion Language Models with a Sink Token (2026.findings-acl)

Challenge:	Existing methods for dangerous speech detection rely on binary labels that ignore who is speaking and in what mental state.
Approach:	They propose a context-dependent variant of dangerous speech detection by grounding it in Theory-of-Mind.
Outcome:	The proposed model outperforms proprietary and open-source models with significantly fewer parameters.

Copied to clipboard

Zihou Zhang, Zheyong Xie, Li Zhong, Haifeng Liu, Shaosheng Cao

Thunder-KoNUBench: A Corpus-Aligned Benchmark for Korean Negation Understanding (2026.findings-acl)

Challenge:	Existing Diffusion Language Models lack a structural constraint to stabilize attention sinks.
Approach:	They propose a simple but effective extra sink token that is constrained to attend to itself while remaining globally visible to all other tokens.
Outcome:	The proposed token is able to stabilize attention sinks and improve model performance.

Copied to clipboard

Sungmok Jung, Yeonkyoung So, Joonhak Lee, Sangho Kim, Yelim Ahn, Jaejin Lee

Progra: Progress-Aware Reinforcement Learning for Multi-Turn Function Calling (2026.findings-acl)

Challenge:	Negation is a fundamental operation in natural language that reverses the meaning of an expression into its opposite.
Approach:	They propose a sentence-level negation understanding benchmark that measures negation performance in Korean.
Outcome:	The proposed benchmark improves negation understanding and broader comprehension in Korean.

Copied to clipboard

Huacan Chai, Zijie Cao, Maolin Ran, Yingxuan Yang, Jianghao Lin, Xin Peng, Hairui Wang, Renjie Ding, Ziyu Wan, Muning Wen, Weiwen Liu, Weinan Zhang, Fei Huang, Ying Wen

Challenge:	Existing methods for multi-turn function calling are limited by redundancy and lack explicit integration of progress awareness into training.
Approach:	They propose a framework that explicitly integrates progress awareness into LLM training for multi-turn function calling.
Outcome:	Empirical results show that Progra outperforms existing methods on two public benchmarks.

Towards Modern Topic Models: A Survey of Taxonomies and Paradigm Shifts from Algorithm-Centric to LLM-Centered Topic Analysis (2026.findings-acl)

Copied to clipboard

Xuan Xu, Zhongliang Yang, Haolun Li, Rui Tian, Beilin Chu, J Song, Yu Li, Shaolin Tan, Linna Zhou

Beyond Templates: Dynamic Adaptation of Reasoning Demonstrations via Feasibility-Aware Exploration (2026.findings-acl)

Challenge:	Topic modeling (TM) is a classic unsupervised learning task in the field of natural language processing.
Approach:	They propose a new taxonomy that emphasizes the role of LLMs and the design of end-to-end workflows.
Outcome:	The proposed taxonomy emphasizes the role of LLMs and the design of end-to-end workflows.

Copied to clipboard

Yong Wu, Weihang Pan, Ke Li, Chen Binhui, Ping Li, Binbin Lin

CFlowPsyD: An Analysis-Enhanced Dataset for Asynchronous Psychological Counseling through Self-Optimizing Multi-Agent Framework (2026.findings-acl)

Challenge:	Existing reasoning datasets that are designed for powerful LLMs often lead to degraded performance when directly applied to weaker models.
Approach:	They propose a data adaptation framework that bridges the capability gap between expert reasoning trajectories and diverse SLMs by employing a selective imitation strategy guided by step-wise adaptability estimation via solution simulation.
Outcome:	The proposed framework improves generalization and data efficiency over static fine-tuning and can be applied to large models with limited model capacity.

Copied to clipboard

Donghao Li, Yifan Deng, Jinta Weng, Xingsheng Zhang, Chenxu Niu, Jingyuan Tian, Heyan Huang, Yue Hu

The Thin Line Between Comprehension and Persuasion in LLMs (2026.findings-acl)

Challenge:	Asynchronous psychological counseling (APC) is a crucial mental health service modality that transcends temporal and spatial constraints.
Approach:	They propose a self-optimizing multi-agent framework for counseling dialogue generation, CFlowPsy, which utilizes real anonymized counseling cases as seed data to synthesize diverse problem-solving-oriented APC conversations through large language models.
Outcome:	The proposed framework synthesizes diverse problem-solving-oriented APC conversations through large language models.

Copied to clipboard

Adrian de Wynter, Tangming Yuan

Evaluating Machine Translation Datasets for Low-Web Data Languages: A Gendered Lens (2026.findings-acl)

Challenge:	Large language models are excellent at maintaining high-level, convincing dialogue . but it remains unclear whether their persuasive success reflects genuine understanding of the discourse .
Approach:	They examine whether LLMs' persuasive success reflects genuine understanding of the discourse . they find that LLM's effectively maintain coherent, persuasive debates .
Outcome:	The findings show that large language models can sway beliefs of participants and audiences.

Copied to clipboard

Hellina Hailu Nigatu, Bethelhem Yemane Mamo, Bontu Fufa Balcha, Debora Taye Tesfaye, Elbethel Daniel Zewdie, Ikram Behiru Nesiru, Jitu Ewnetu Hailu, Senait Mengesha Yayo

Select2Reason: Efficient Instruction-Tuning Data Selection for Long-CoT Reasoning (2026.findings-acl)

Challenge:	afan oromo, amharic, and tigrinya are low-resourced languages . they are used for training, benchmarks, news, health, and sports . afono o'mara: quantity does not guarantee quality of MT datasets .
Approach:	They investigate the quality of machine translation datasets for three low-resourced languages . they found a large skew towards the male gender in the datasets .
Outcome:	The results show that training data has large representation of political and religious text, but benchmark datasets focus on news, health, and sports.

Copied to clipboard

Cehao Yang, Xueyuan Lin, Xiaojun Wu, Chengjin Xu, Xuhui Jiang, Honghao Liu, Hui Xiong, Jian Guo

Curriculum Learning based Hierarchical Scoring and Analysis Framework for Question Answering Task Evaluation (2026.findings-acl)

Challenge:	Large reasoning models exhibit human-like behaviors such as exploration, verification, reflection, and correction.
Approach:	They propose a supervised fine-tuning framework for long chain-of-thoughts reasoning . they leverage a difficulty-aware reward model to estimate the learning value of questions .
Outcome:	The proposed framework performs fine-tuning on large reasoning models on 10% of the data selected.

Copied to clipboard

Qiong Wu, Tan Yue, Jianxin Liang, Zhen Li, Kai He, Shuai Zhao, Dongyan Zhao

TrigReason: Trigger-Based Collaboration between Small and Large Reasoning Models (2026.findings-acl)

Challenge:	Existing evaluation methods rely on rule-based matching with shallow semantic understanding or adopt LLM-as-a-Judge approaches that incur high cost and latency while offering limited error interpretability.
Approach:	They propose a curriculum learning based hierarchical framework for QA task evaluation that supports quick scoring and fine-grained error analysis.
Outcome:	The proposed framework outperforms baseline methods on quick scoring and error analysis tasks while being 25 faster.

Copied to clipboard

Yi Zhao, Yajuan Peng, Cam-Tu Nguyen, Zuchao Li, Xiaoliang Wang, Xiaoming Fu, Hai Zhao

TURTLEAI: Benchmarking Multimodal Models for Visual Programming in Turtle Graphics (2026.findings-acl)

Challenge:	Large Reasoning Models suffer from high inference latency due to autoregressive reasoning . SpecReason adopts a polling-based design that repeatedly invokes the LRM for verification at every step .
Approach:	They propose a trigger-based collaborative reasoning framework that delegates most reasoning to the SRM and activates LRM intervention only when necessary.
Outcome:	The proposed framework reduces latency and API cost by 73.3% under edge–cloud conditions.

Copied to clipboard

Chao Wen, Jacqueline Staub, Adish Singla

Calibrated Progressive Distillation: Co-Designing Curriculum and Target Mixing for Knowledge Distillation of Large Language Models (2026.findings-acl)

Challenge:	Vision-language models have been explored for visual programming, but performance is unclear . most prior work focuses on visual programming for productivity .
Approach:	They propose a visual programming benchmark that uses visual programming to evaluate VLMs.
Outcome:	The proposed model improves on GPT-5, GPT-4o, and Qwen2-VL-72B on real-world tasks by 20% . the proposed model is based on 823 visual programming tasks in the Turtle Graphics domain .

Copied to clipboard

Mengxiang Zhang, Lingyuan Liu

Faithful-First Reasoning, Planning, and Acting for Multimodal LLMs (2026.findings-acl)

Challenge:	Existing methods for knowledge distillation address the teacher–student capacity gap by mixing teacher and student distributions in the distillation target or using curriculum learning to sequence training from easy to hard examples.
Approach:	They propose a white-box KD framework that co-designs curriculum scheduling and target mixing through a unified difficulty-aware principle.
Outcome:	The proposed framework outperforms existing methods while reducing training runtime by over 10%.

Copied to clipboard

Junxian Li, Xinyue Xu, Sai Ma, Di Zhang, Sichao Li

SEAD: A Surrogate-free Label-only Membership Inference Attack against Pre-trained LLMs with Semantic-Aware Density (2026.findings-acl)

Challenge:	Existing efforts to improve task accuracy or enrich COT generation are lacking in multimodal large language models.
Approach:	They propose a Faithful-First Reasoning, Planning, and Acting framework that evaluates faithfulness of intermediate reasoning and uses it to plan and execute faithfulness-aware actions during inference.
Outcome:	The proposed framework improves perceptual faithfulness by up to 24% over prompt-based and tool-augmented reasoning frameworks without degrading task accuracy.

Copied to clipboard

Biao Yi, Jiahao Li, Yiming Li, Yu He, Baolei Zhang, Zheli Liu, Dacheng Tao

Faithfulness-Aware Uncertainty Quantification for Fact-Checking the Output of Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	Existing membership inference attacks require access to complete logits, but such access is often unavailable in real-world deployments where only the generated text is exposed.
Approach:	They propose a surrogate-free label-only MIA approach that directly estimates token probabilities through Monte Carlo sampling of the target model.
Outcome:	The proposed approach outperforms existing label-only attacks and serves as a foundational density estimator in the label-exclusive setting.

Copied to clipboard

Ekaterina Fadeeva, Aleksandr Rubashevskii, Dzianis Piatrashyn, Roman Vashurin, Shehzaad Dhuliawala, Artem Shelmanov, Timothy Baldwin, Preslav Nakov, Mrinmaya Sachan, Maxim Panov

AGGC: Adaptive Group Gradient Clipping for Stabilizing Large Language Model Training (2026.findings-acl)

Challenge:	Existing approaches to mitigating hallucinations conflate factuality with faithfulness to the retrieved evidence, incorrectly labeling factually correct statements as hallucinos . Existing methods to mitigate hallucinics rely on a lack of training data coverage, input ambiguity, and architectural constraints.
Approach:	They propose a method for hallucination detection in Large Language Models enhanced with knowledge retrieval based on faithfulness to the retrieved context.
Outcome:	The proposed method outperforms unsupervised UQ baselines, RAG-specific methods, and supervised classifiers across multiple tasks and LLMs.

Copied to clipboard

Zhiyuan Li, Yuan Wu, Yi Chang

MADRA: Multi-Agent Debate for Risk-Aware Embodied Planning (2026.findings-acl)

Challenge:	Adaptive group-wise gradient clipping (AGGC) is a new approach to stabilize training of Large Language Models.
Approach:	They propose a method to stabilize gradient clipping by partitioning parameters into groups based on functional types and a time-dependent scheduling mechanism to balance exploration and convergence.
Outcome:	The proposed algorithm outperforms standard LoRA and achieves 72.93% accuracy . it can be integrated into existing pipelines with negligible overhead.

Copied to clipboard

JunJian Wang, Lidan Zhao, Xi Sheryl Zhang

Challenge:	Existing safety alignment methods, such as RLHF, fall into a Safety-Utility Trade-off, resulting in severe over-rejection of benign household instructions.
Approach:	They propose a meta-cognitive Critical Agent that evaluates peer debates using a structured argumentation framework derived from the Toulmin Model.
Outcome:	The proposed architecture outperforms existing systems in the SafeAware-VH benchmark.

Navigating Large-Scale Document Collections: MuDABench for Multi-Document Analytical QA (2026.findings-acl)

Copied to clipboard

Zhanli Li, Yixuan Cao, Lvzhou Luo, Ping Luo

CORBA: Contagious Recursive Blocking Attacks on Multi-Agent Systems Based on Large Language Models (2026.findings-acl)

Challenge:	Existing multi-document QA benchmarks require information from only a few documents with limited cross-document reasoning.
Approach:	They propose a benchmark for multi-document analytical QA that extracts and synthesizes information across multiple documents to perform quantitative analysis.
Outcome:	The proposed approach improves both process and outcome metrics but still has bottlenecks compared to human experts.

Copied to clipboard

Zhenhong Zhou, Zherui Li, Jie Zhang, Yuanhe Zhang, Kun Wang, Yang Liu, Qing Guo

Afrispeech Semantics: Evaluating Audio–Semantic Reasoning in Spoken Language Models Across Domains and Accents (2026.findings-acl)

Challenge:	Existing security models rely on open-ended communication, but the collaborative process itself can be exploited and disrupted.
Approach:	They propose a new threat class, called Denial-of-Collaboration, which corrupts collaborative structure and transforms communication topology into self-sabotage.
Outcome:	The proposed attacks bypass conventional safety alignments that are not designed to detect behavioral or systemic attacks.

Copied to clipboard

Chibuzor Okocha, Christan Grant

MoPrune: Scene-Guided Motion-Aware Token Pruning for Efficient Video Large Language Models (2026.findings-acl)

Challenge:	Recent multimodal models are trained on large collections of audio-text pairs using contrastive learning or nexttoken prediction objectives.
Approach:	They evaluate audio language models across five semantic and paralinguistic reasoning tasks: entailment, consistency, plausibility, accent drift, and accent restraint.
Outcome:	The evaluations assess models across five tasks including entailment, consistency, plausibility, accent drift, and accent restraint.

Copied to clipboard

Wenhao Hong, Ziyang Wang, Yixin Zhang, Zilei Wang

SCMAPR: Self-Correcting Multi-Agent Prompt Refinement for Complex-Scenario Text-to-Video Generation (2026.findings-acl)

Challenge:	Prior approaches to token pruning ignore video dynamics and the fact that different scenes exhibit different redundancy patterns.
Approach:	They propose a token pruning framework that is train-free and scene-guided to accelerate VideoLLMs by removing redundant visual information from video frames.
Outcome:	MoPrune is a training-free, scene-guided and motion-centric token pruning framework for accelerating VideoLLMs.

Copied to clipboard

Chengyi Yang, Pengzhen Li, Jiayin Qi, Aimin Zhou, Ji Wu, Ji Liu

LoReFact: Bridging the Logic Gap in Fact-Checking (2026.findings-acl)

Challenge:	Text-to-Video (T2V) generation is a challenge under complex scenarios.
Approach:	They propose a scenario-aware and self-correcting multi-agent prompt refinement framework for T2V prompting.
Outcome:	The proposed framework improves text-to-video alignment and overall generation quality under complex scenarios.

Copied to clipboard

Qiming Xie, Wenjie Zheng, Xiangqing Shen, Rui Xia

ODTQA-FoRe: An Open-Domain Tabular Question Answering Dataset for Future Data Forecasting and Reasoning (2026.findings-acl)

Challenge:	Existing fact-checking methods focus on verification of individual facts, overlooking logical dependencies . a recent study shows that text containing logical errors may still be misjudged as factual .
Approach:	They propose a content–logic coupled factuality evaluation paradigm that conceptualizes factual dimension along two complementary dimensions: content factualism and logic factuity.
Outcome:	The proposed paradigm bridges the gap between factual verification and content factuality . it incorporates the logical dimension and a logic-aware metric to expose and penalize logical fallacies.

Copied to clipboard

Zhensheng Wang, Xiaole Liu, Wenmian Yang, Kun Zhou, Yiquan Zhang, Weijia Jia

Rectified Sparse Attention for Efficient Long-Sequence Generation (2026.findings-acl)

Challenge:	Existing tabular question answering systems cannot perform future-oriented numerical prediction . open-domain tabular questions are a popular approach for QA tasks .
Approach:	They propose a task that covers time-series forecasting and forecast-based reasoning scenarios using real estate data.
Outcome:	The proposed framework decomposes the problem into three collaborative roles that synthesize the results to construct a precise and consistent final answer.

Copied to clipboard

Yutao Sun, Tianzhu Ye, Li Dong, Yuqing Xia, Jian Chen, Yizhao Gao, Shijie Cao, Jianyong Wang, Furu Wei

Knowing-but-Doing: Diagnosing and Defending Role-Play-Driven LLMs Jailbreaks via Moral Disengagement (2026.findings-acl)

Challenge:	Recent sparse decoding methods improve efficiency but suffer from KV cache misalignment, resulting in performance degradation.
Approach:	They propose a method that combines block-sparse attention with periodic dense rectification to bound error accumulation and preserve alignment with the pretraining distribution.
Outcome:	Experiments on math reasoning, language modeling, and retrieval tasks show that ReSA achieves near-lossless generation quality with significantly improved efficiency.

Copied to clipboard

Haiming Qin, Jianxun Lian, Qimin Zhong, Mingyang Zhou, Hao Liao, Naipeng Chao

Generative-to-Discriminative Test-Time Adaptation via Manifold-Aware Diffusion and Bayesian Distillation (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are increasingly used in role-play scenarios, but their safety implications remain under-characterized.
Approach:	They propose a diagnostic benchmark for role-play jailbreaks based on Bandura’s Moral Disengagement theory and propose 'MD-Trace' based defense that reduces attack success while maintaining Role Fidelity.
Outcome:	The proposed framework improves safety behavior for benign personas while increasing unsafe compliance for malicious ones.

Copied to clipboard

Boyun Zhang, Zequn Xie, Fangming Feng, Zihan Zhang, Yongbo He, Chuxin Wang, Sihang Cai, Tao Jin, Qifei Zhang

Evaluating Memory Capability in Continuous Lifelog Scenario (2026.findings-acl)

Challenge:	Existing discriminative approaches suffer from "confident but wrong" failure mode, blindly adapting to OOD noise leading to error accumulation.
Approach:	They propose a TTA framework that harmonizes the robustness of generative diffusion models with the efficiency of discriminative regression networks via Bayesian Diffusion Distillation (BDD).
Outcome:	The proposed framework reduces MAE from 0.6872 to 0.5673 and boosts binary accuracy by 5.81 percentage points (reaching 57.33%) it also reduces the MAE of the MOSI to SIMS shift and achieves an 11.18-point gain over the baseline.

Copied to clipboard

Jianjie Zheng, Zhichen Liu, Zhanyu Shen, Jingxiang Qu, Guanhua Chen, Yile Wang, Yang Xu, Yang Liu, Sijie Cheng

Challenge:	Existing benchmarks focus on online one-on-one chatting or human-AI interactions, neglecting real-world scenarios.
Approach:	They propose a framework to curate a lifelog benchmark that combines two subsets of audio data to address temporal leakage in offline settings.
Outcome:	The proposed framework outperforms existing benchmarks on live chats and AI interactions.

GoT-R1: Internalizing Graph-of-Thought via Structural Reinforcement for High-Density Reasoning (2026.findings-acl)

Copied to clipboard

Zuchao Li, Qiwei Li, Yao Yao, Hai Zhao, Lefei Zhang, Bo Du

Hey, That’s My Data! Token-Only Dataset Inference in Large Language Models (2026.findings-acl)

Challenge:	Chain-of-Thought reasoning suffers from an inherent mechanism flaw: linearity induces overthinking . emergence of Large Language Models (LLMs) has fundamentally redefined artificial intelligence .
Approach:	They propose a framework that replaces verbose linear trajectories with high-density reasoning graphs.
Outcome:	The proposed framework outperforms state-of-the-art models with reduced token overhead.

Copied to clipboard

Chen Xiong, Zihao Wang, Rui Zhu, Tsung-Yi Ho, Pin-Yu Chen, Jingwei Xiong, Haixu Tang

ToxiTrace: Gradient-Aligned Training for Explainable Chinese Toxicity Detection (2026.findings-acl)

Challenge:	Existing dataset inference methods require logit access, but many modern LLMs restrict such access.
Approach:	They propose a token-only dataset inference framework that allows models to overwrite prior knowledge when trained on new data.
Outcome:	The proposed framework overwrites prior knowledge when trained on new data.

Copied to clipboard

Boyang Li, Hongzhe Shou, Yuanyuan Liang, JingBin Zhang, Fang Zhou

Three Minds, One Legend: Jailbreak Large Reasoning Model with Adaptive Stacked Ciphers (2026.findings-acl)

Challenge:	Existing toxic content detection methods focus on sentence-level classification but fail to provide readable and contiguous toxic evidence spans.
Approach:	They propose an explainability-oriented method for Chinese toxic content detection methods . they refine saliency cues into fine-grained toxic spans with lightweight LLM guidance .
Outcome:	The proposed method improves classification accuracy and toxic span extraction while preserving efficient encoder-based inference and producing more coherent explanations.

Copied to clipboard

Nguyen Viet Anh, Shiqian Zhao, Gia Dao, Runyi Hu, Yi Xie, Xiaobao Wu, Anh Tuan Luu

From Imitation to Discrimination: Progressive Curriculum Learning for Robust Web Navigation (2026.findings-acl)

Challenge:	Existing jailbreak methods struggle to balance effectiveness with robustness against adaptive safety mechanisms.
Approach:	They propose a novel approach that targets Large Reasoning Models through an adaptive encryption pipeline designed to overwhelm their reasoning capabilities.
Outcome:	The proposed approach achieves an attack success rate of 85.6% on OpenAI GPT-o4-mini, outperforming state-of-the-art baselines by a significant margin of 17.2%.

Copied to clipboard

Peng Chuang, Wei Zhang, Renshuai Tao, Xinhao Zhang, Jian Yang

Parallelism and Generation Order in Masked Diffusion Language Models: Limits Today, Potential Tomorrow (2026.findings-acl)

Challenge:	Text-based web agents offer computational efficiency for autonomous web navigation, yet they lack discrimination capabilities to reject plausible but incorrect elements in densely populated pages.
Approach:	They propose a model that uses a text-based web agent to learn to discriminate against incorrect elements in densely populated HTML and a training curriculum to synthesize diverse cross-domain tasks with strict verification.
Outcome:	Empirical evaluation shows that the model performs better than open-source models with 58.7% step success rate.

Copied to clipboard

Yangyang Zhong, Yanmei Gu, Zhengqing Zang, Xiaomeng Li, Yuqi Ding, Xibei Jia, Yuting Shen, Zhenzhong Lan, Liwang Zhu, Weiping Liu, Junlin Zhou, Haisheng Liu, Zhong Xin Yu, Pengxin Luo, Donglian Qi, Yunfeng Yan, Junbo Zhao

Understanding New-Knowledge-Induced Factual Hallucinations in LLMs: Analysis and Interpretation (2026.findings-acl)

Challenge:	Autoregressive (AR) language models dominate modern natural language processing due to strong likelihood-based training objectives and reliable left-to-right decoding.
Approach:	They characterize MDLM behavior along two dimensions: parallelism strength and generation order . authors propose a Generate-then-Edit paradigm that mitigates dependency loss .
Outcome:	The proposed model improves on tasks that require "backward information" the Generate-then-Edit paradigm improves parallel decoding efficiency while reducing dependency loss.

Copied to clipboard

Renfei Dang, Peng Hu, Zhejian Lai, Changjiang Gao, Min Zhang, Shujian Huang

Attribution-Based Analysis and Optimization of Modular Agentic Workflows (2026.findings-acl)

Challenge:	Prior studies have shown that fine-tuning on new knowledge can induce factual hallucinations in large language models (LLMs), leading to incorrect outputs when evaluated on previously known information.
Approach:	They propose to conduct a fine-grained analysis of large language models using a dataset Biography-Reasoning and QA and knowledge reasoning tasks to understand their findings.
Outcome:	The proposed model is able to perform a range of downstream tasks without requiring a large amount of knowledge and is compared with a control dataset.

Copied to clipboard

Yingxuan Yang, Bo Huang, Siyuan Qi, Chao Feng, Haoyi Hu, Yuxuan Zhu, Jinbo Hu, Haoran Zhao, Ziyi He, Xiao Liu, ZongYu Wang, Muning Wen, Lin Qiu, Xuezhi Cao, Xunliang Cai, Yong Yu, Weinan Zhang

Challenge:	Large Language Models (LLMs) have driven the rise of agentic workflows . yet, how can we attribute performance gains to individual upgrades and their interactions?
Approach:	They propose a game-theoretic framework that models component upgrades as players and evaluates component coalitions to compute Shapley values.
Outcome:	The proposed framework provides interaction-aware attribution and recommendation for model allocation under a fixed workflow structure.

Synthetic Eggs in Many Baskets: The Impact of Synthetic Data Diversity on LLM Fine-Tuning (2026.findings-acl)

Copied to clipboard

Max Schaffelder, Albert Gatt

SolidCoder: Bridging the Mental-Reality Gap in LLM Code Generation through Concrete Execution (2026.findings-acl)

Challenge:	Increasing demand for training data is causing language models to be trained on synthetic data, a new study finds . fine-tuning models on synthetic datasets reduces self-preference bias .
Approach:	They investigate the impact of diversity of synthetic data on fine-tuned large language models.
Outcome:	The proposed model can mitigate distribution collapse, maintain diversity of output distribution, and reduce self-preference bias.

Copied to clipboard

Woojin Lee, Jin-Xia Huang

EgoMemory: Memory-Augmented Personalized Retrieval for Long-Context Egocentric Video (2026.findings-acl)

Challenge:	State-of-the-art code generation frameworks rely on mental simulations to validate buggy code.
Approach:	They propose a mental-reality gap between mental simulation and actual execution . they propose sandboxed execution with a simple principle: don't imagine—execute .
Outcome:	The proposed framework achieves state-of-the-art pass@1 performance on humanEval, CodeContests and APPS.

Copied to clipboard

Yuanmin Tang, Jue Zhang, Xiaoting Qin, Jing Yu, Meikang Qiu, Gaopeng Gou, Gang Xiong, Qingwei Lin, Saravan Rajmohan, Dongmei Zhang, Qi Wu

PERM: Psychology-grounded Empathetic Reward Modeling for Large Language Models (2026.findings-acl)

Challenge:	Existing egocentric video datasets do not support the personalization and long-context reasoning required for episodic memory retrieval.
Approach:	They propose a benchmark framework that uses MLLMs and reflective Chain-of-Thought to ground user queries in personal memory explicitly.
Outcome:	The proposed framework outperforms state-of-the-art benchmarks on three benchmarks . it can be used to generate detailed target video descriptions in long-context contexts based on user-specific object annotations enriched with user-specified object annotation data .

Copied to clipboard

Chengbing Wang, Wuqiang Zheng, Yang Zhang, Fengbin Zhu, Junyi Cheng, Yi Xie, Wenjie Wang, Fuli Feng

Remedy-R: Generative Reasoning for Machine Translation Evaluation without Error Annotations (2026.findings-acl)

Challenge:	Existing reward models evaluate empathy from a single perspective, overlooking bidirectional interaction nature of empathy.
Approach:	They propose a reward model that evaluates empathy from a single perspective . they propose PERM to integrate a bystander perspective to monitor overall interaction quality .
Outcome:	a new reward model outperforms state-of-the-art models on an emotional intelligence benchmark and an industrial daily conversation dataset.

Copied to clipboard

Shaomu Tan, Ryosuke Mitani, Ritvik Choudhary, Qiyu Wu, Toshiyuki Sekiya, Christof Monz

VizoMem: A Visual-Textual Memory Framework for Efficient Long-Horizon Reasoning (2026.findings-acl)

Challenge:	Recent MT metrics like xCOMET, Met-ricX, and Remedy have strong correlations with human preferences, but they are black boxes, revealing little insight into why a translation is good or bad.
Approach:	They propose a reasoning-driven generative MT metric trained with reinforcement learning from pairwise translation preferences without requiring error-span annotations or distillation from closed LLMs.
Outcome:	The proposed reasoning-driven generative MT metric produces step-by-step analyses of accuracy, fluency, and completeness, enabling more interpretable assessments.

Copied to clipboard

Weijie Liang, Yuanfeng Song, Xing Chen, Caleb Chen Cao, Sirui Han, Yike Guo

More Images, More Problems? A Controlled Analysis of VLM Failure Modes. (2026.findings-acl)

Challenge:	Existing systems that use long-context modeling incur computational and memory overhead.
Approach:	They propose a visual memory framework that pre-rendered text into structured images and stored as visual notes for agentic systems.
Outcome:	The proposed system reduces token consumption while preserving effective long-term memory recall.

Copied to clipboard

Anurag Das, Adrian Bulat, Alberto Baldrati, Ioannis Maniadis Metaxas, Bernt Schiele, Georgios Tzimiropoulos, Brais Martinez

EOP-LLM: Energy Oriented Pruning for Large Language Models (2026.findings-acl)

Challenge:	Existing evaluations of large vision language models lack a comprehensive analysis of their weaknesses and causes.
Approach:	They propose a new benchmark to evaluate multi-image capabilities of Large Vision Language Models.
Outcome:	The proposed model outperforms existing benchmarks on multi-image models.

Copied to clipboard

Guan Huang, Tao Shu

Sentipolis: Emotion-Aware Agents for Social Simulations (2026.findings-acl)

Challenge:	Inference energy consumption has grown rapidly in large language models (LLMs) but existing methods focus on reducing FLOPs or latency rather than modeling or enforcing end-to-end inference energy constraints.
Approach:	They propose an energy-oriented dynamic pruning framework that enables LLM inference under explicit per-sequence energy budgets.
Outcome:	EOP-LLM outperforms state-of-the-art dynamic pruning baselines while adhering to per-sequence energy constraints.

Copied to clipboard

Chiyuan Fu, Lyuhao Chen, Yunze Xiao, Weihao Xuan, Carlos Busso, Mona T. Diab

Challenge:	Recent advances in reasoning and long-context memory are making large language models (LLMs) appear increasingly human-like, which has led researchers to adopt LLM agents as a substrate for social simulation.
Approach:	They propose a framework for emotionally stateful agents that integrates continuous Pleasure-Arousal-Dominance representation, dual-speed emotion dynamics, and emotion–memory coupling.
Outcome:	The proposed framework improves emotional grounded behavior, boosting communication, and emotional continuity across thousands of interactions over multiple base models and evaluators.

Bidirectional Semantic Enhancement for Schema Routing Across Large-Scale Databases (2026.findings-acl)

Copied to clipboard

Yuyang Wu, Xiaoliang Wang, Cam-Tu Nguyen

DeepPlanner: Scaling Planning Capability for Deep Research Agents via Advantage Shaping (2026.findings-acl)

Challenge:	Existing methods relying on unidirectional query expansion fail to bridge lexical mismatches and graph-based approaches struggle to navigate schemas when explicit structural links are missing.
Approach:	They propose a retrieval framework that bridges the semantic gap between user queries and vague schema definitions by performing online generative query expansion.
Outcome:	The proposed retrieval framework bridges the semantic gap between user queries and vague schema definitions by enriching table schemas offline and performing online generative query expansion.

Copied to clipboard

Wei Fan, Wenlin Yao, Zheng Li, Feng Yao, Xin Liu, Liang Qiu, Qingyu Yin, Yangqiu Song, Bing Yin

SupChain-Bench: Benchmarking Large Language Models for Real-World Supply Chain Management (2026.findings-acl)

Challenge:	Existing approaches to planning involve implicit planning or introduce explicit planners without systematically optimizing the planning stage.
Approach:	They propose an end-to-end RL framework that enhances the planning capabilities of deep research agents.
Outcome:	Experiments show that DeepPlanner improves planning quality and achieves state-of-the-art results under a lower training budget.

Copied to clipboard

Shengyue Guan, Yihao Liu, Lang Cao

How Hypocritical Is Your LLM judge? Listener-Speaker Asymmetries in the Pragmatic Competence of Large Language Models (2026.findings-acl)

Challenge:	Large language models have shown promise in complex reasoning and tool-based decision making, but supply chain workflows require reliable long-horizon, multi-step orchestration grounded in domain-specific procedures.
Approach:	They propose a unified real-world benchmark that assesses supply chain domain knowledge and long-horizon tool-based orchestration grounded in standard operating procedures.
Outcome:	The proposed framework achieves the strongest and consistent tool-calling performance in real-world operational settings.

Copied to clipboard

Judith Sieker, Sina Zarrieß

CoopQ: Cooperative Game Inspired Layerwise Mixed Precision Quantization for LLMs (2026.findings-acl)

Challenge:	Large language models (LLMs) are increasingly studied as repositories of linguistic knowledge.
Approach:	They compare LLMs’ performance as pragmatic listeners and as pragmatic speakers . they find a robust asymmetry between pragmatic evaluation and pragmatic generation .
Outcome:	The proposed models perform better as listeners than speakers, and produce more appropriate language than speakers.

Copied to clipboard

Junchen Zhao, Ali Derakhshan, Jayden Hyman, Junhao Dong, Sangeetha Abdu Jyothi, Ian Harris

Challenge:	Existing methods for large language models struggle when the average precision drops below four bits, limiting deployment on resourceconstrained devices such as mobiles, edge sensors, or standard GPUs.
Approach:	They propose a game-like game-inspired mixed-precision quantization method which translates these Shapley estimates into a binary quadratic optimization formulation, assigning either 2 or 4-bit precision to layers under strict memory constraints.
Outcome:	The proposed method reduces Perplexity by 20 – 80 % across average precisions spanning 4 bit down to 2 bit, compared to methods relying on isolated metrics.

Breaking Language Preference in Multilingual RAG via Language-Controllable Retrieval and Language-Agnostic Reasoning (2026.findings-acl)

Copied to clipboard

Wenshuai Huo, Xiaocheng Feng, Baohang Li, Chengpeng Fu, Yichong Huang, Hui Wang, Bing Qin

From Scoring to Explanations: Evaluating SHAP and LLM Rationales for Rubric-based Teaching Quality Assessment (2026.findings-acl)

Challenge:	Existing approaches to improve retrieval accuracy and generation quality of large language models suffer from language preference.
Approach:	They propose a framework that explicitly disentangles multilingual RAG into language-controllable retrieval and language-agnostic reasoning.
Outcome:	Experimental results show that the proposed approach outperforms baselines across multilingual benchmarks.

Copied to clipboard

Ivo Bueno, Babette Bühler, Philipp Stark, Tim Fütterer, Ulrich Trautwein, Dorottya Demszky, Heather Hill, Enkelejda Kasneci

LycheeCluster: Efficient Long-Context Inference with Structure-Aware Chunking and Hierarchical KV Indexing (2026.findings-acl)

Challenge:	a framework for sentence-level interpretability of rubric-based scoring is proposed . aaron e. smith: automated scoring models provide little insight into why scores are produced .
Approach:	They propose a framework for sentence-level interpretability of rubric-based scoring that combines Shapley-value attributions with rationales generated by large language models.
Outcome:	The proposed framework compares fine-tuned pretrained language models with large language models . it shows that fine- tuned models outperform LLMs in prediction accuracy but exhibit label compression toward mid-scale scores .

Copied to clipboard

Dongfang Li, Zixuan Liu, Gang Lin, Baotian Hu, Min Zhang

HiRAS: A Hierarchical Multi-Agent Framework for Paper-to-Code Generation and Execution (2026.findings-acl)

Challenge:	Existing retrieval-based methods compromise semantic integrity through fixed-size chunking and suffer from inefficient linear scanning.
Approach:	They propose a method that preserves local semantic coherence through boundary-aware chunking and constructs a recursive hierarchical index rooted in the triangle inequality.
Outcome:	The proposed method achieves 3.6 end-to-end inference speedup with negligible degradation in model performance.

Copied to clipboard

Hanhua Hong, Yizhi LI, Jiaoyan Chen, Sophia Ananiadou, Xiaoli Li, Jung-jae Kim, Chenghua Lin

RoadMapper: A Multi-Agent System for Roadmap Generation of Solving Complex Research Problems (2026.findings-acl)

Challenge:	Existing approaches to automate computational research use fixed sequential agent pipelines with weak global coordination, which limits their robustness and overall performance.
Approach:	They propose a hierarchical multi-agent framework for end-to-end paper reproduction that employs supervisory manager agents to coordinate specialised agents across fine-grained stages.
Outcome:	The proposed framework improves the paper2code benchmark and significantly reduces hallucination in the evaluation.

Copied to clipboard

Jiacheng Liu, Zichen Tang, Zhongjun Yang, Xinyi Hu, Xueyuan Lin, Linwei Jia, Ruofei Bai, Rongjin Li, Shiyao Peng, Haocheng Gao, Haihong E

On Temperature-Constrained Non-Deterministic Machine Translation: Potential and Evaluation (2026.findings-acl)

Challenge:	Existing tools to generate structured content for research tasks are limited in their ability to generate high-quality roadmaps.
Approach:	They propose a benchmark to evaluate the ability of large language models (LLMs) to generate high-quality roadmaps for solving complex research problems.
Outcome:	The proposed system can improve LLMs’ ability for roadmap generation while saving 84% of the time required by human experts.

Copied to clipboard

Weichuan Wang, Mingyang Liu, Chen Ma, Linqi Song

ExaGPT: Example-Based Machine-Generated Text Detection for Human Interpretability (2026.findings-acl)

Challenge:	Recent studies have focused on the non-deterministic properties of language models, but these properties remain under-explored in machine translation.
Approach:	They propose a method that evaluates MT systems and identifies temperature-constrained non-deterministic MT as a distinct phenomenon.
Outcome:	The proposed framework provides higher-quality candidates than Deterministic MT under temperature constraints.

Copied to clipboard

Ryuto Koike, Masahiro Kaneko, Ayana Niwa, Preslav Nakov, Naoaki Okazaki

Zipage: Maintain High Request Concurrency for LLM Reasoning through Compressed PagedAttention (2026.findings-acl)

Challenge:	Existing interpretable detectors are not aligned with the human decision-making process and fail to offer evidence that users easily understand.
Approach:	They propose an interpretable detection approach that checks whether a text is human-written or LLM-generated by checking whether it shares more similar spans with human-generated texts.
Outcome:	ExaGPT outperforms interpretable detectors by +37.0 points at a false positive rate of 1%.

Copied to clipboard

Mengqi Liao, Lu Wang, Chaoyun Zhang, Bo Qiao, Si Qin, Qingwei Lin, Saravan Rajmohan, Dongmei Zhang, Huaiyu Wan

The Overlooked Role of Graded Relevance Thresholds in Multilingual Dense Retrieval (2026.findings-acl)

Challenge:	Existing methods to evict KV cache during inference phase are impractical for industrial-grade applications.
Approach:	They propose a method that combines token-wise KV cache eviction with PagedAttention and propose 'zipage' it achieves 95% of the performance of Full KV inference engines while delivering over 2.1 speedup .
Outcome:	The proposed method achieves 95% of the performance of Full KV inference engines while delivering over 2.1 speedup on large-scale mathematical reasoning tasks.

Copied to clipboard

Tomer Wullach, Ori Shapira, Amir David Nissan Cohen

Test-Time Scaling in Multimodal Foundation Models: A Comprehensive Survey of Generation and Reasoning (2026.findings-acl)

Challenge:	Dense retrieval models are fine-tuned with contrastive learning objectives that require binary relevance judgments.
Approach:	They examine how graded relevance scores affect multilingual dense retrieval . they argue that a well-chosen threshold can improve effectiveness and mitigate annotation noise .
Outcome:	The optimal threshold varies systematically across languages and tasks, the authors show . a well-chosen threshold can improve effectiveness and mitigate annotation noise .

Copied to clipboard

Cong Wan, Ying He, Zhongzhan Huang, Hefeng Wu

PsyScore: A Psychometrically-Aware Framework for Trait-Adaptive Essay Scoring and ZPD-Scaffolded Feedback (2026.findings-acl)

Challenge:	Recent advances have adapted this paradigm to Multimodal Foundation Models (MFMs), unlocking their potential in multimodal reasoning and generation.
Approach:	They propose a taxonomy framework that categorizes existing methodologies into three distinct strategies: sampling-based, feedback-based and search-based approaches.
Outcome:	The proposed framework categorizes existing methodologies into three distinct strategies: sampling-based, feedback-based and search-based approaches.

Copied to clipboard

Wei Xia, Jin Wu, Haoran Shi, Xiangyu Wang, Chanjin Zheng

FinMaster: A Holistic Benchmark for Full-Pipeline Financial Management with Large Language Models (2026.findings-acl)

Challenge:	Existing approaches to Automated Essay Scoring (AES) treat scoring and feedback as separate components, resulting in fragmentation.
Approach:	They propose a psychometrically-aware framework that integrates diagnostic assessment with instructional scaffolding through a shared latent ability representation.
Outcome:	The proposed framework integrates diagnostic assessment with instructional scaffolding through a shared latent ability representation.

Copied to clipboard

Junzhe Jiang, Chang Yang, Aixin Cui, Sihan Jin, Yujing Zhang, Yilin Xiao, Ruiyu Wang, Bo Li, Xiao Huang, Danny Dongning Sun, Xinrun Wang

PersonaForge: Psychology-Grounded Dual-Process Architecture for Personality-Consistent Role-Playing Agents (2026.findings-acl)

Challenge:	Existing benchmarks lack domain-specific data, realistic workflow-level task design, and standardized workflow- level evaluation.
Approach:	a new benchmark evaluates large language models on financial management workflows . the global financial services market is projected to grow to $37 trillion by 2027 .
Outcome:	a new benchmark for large language models on financial management workflows reveals critical capability gaps . accuracy drops from 90% on basic tasks to 40% on complex scenarios requiring multi-step reasoning . the global financial services market reached $25.8 trillion in 2022 and is projected to grow to $37 trillion by 2027 .

Copied to clipboard

Jizhou Tong, Sirui Zou

Challenge:	Existing approaches to role-playing with Large Language Models lack consistency across long conversations.
Approach:	They propose a three-layer personality architecture grounded in psychological theory and a dual-process generation mechanism inspired by cognitive science to solve this problem.
Outcome:	The proposed framework reduces drift over 50-turn conversations by reducing personality consistency . human evaluation confirms more authentic and psychologically coherent character behaviors.

Graph Explorer: Training Faithful KG Agents with Visibility-Grounded Supervision (2026.findings-acl)

Copied to clipboard

Yifeng Chen, Sicheng Wan, Tianyi Zhang, Xuezhou Zhang

Robust Uncertainty Quantification for Self-Evolving Large Language Models via Continual Domain Pretraining (2026.findings-acl)

Challenge:	Large language models (LLMs) are strong reasoners but still hallucinate and make unreliable decisions on knowledge-intensive questions.
Approach:	They propose a pipeline that turns LLM into executable tool supervision without manual trace labeling.
Outcome:	The proposed model improves over a reproduced prompting baseline by +22.5/+16.2 points . it is based on a Graph Explorer pipeline that turns SPARQL into executable tool supervision without manual trace labeling.

Copied to clipboard

Xiaofan Zhou, Lu Cheng

Exploring Graph Learning Tasks with Pure LLMs: A Comprehensive Benchmark and Investigation (2026.findings-acl)

Challenge:	Conformal prediction (CP) has shown promise in offering correctness guarantees for LLMs, but it faces major challenges in continual domain pretraining (CDP).
Approach:	They propose an adaptive rejection and non-exchangeable CP framework that allows the LLM to selectively abstain from answering when its confidence or competence shifts significantly.
Outcome:	Experiments show that the proposed framework improves performance under continuous domain pretraining scenarios.

Copied to clipboard

Yuxiang Wang, Xinnan Dai, Wenqi Fan, Yao Ma

Beyond Rejection Sampling: Trajectory Fusion for Scaling Mathematical Reasoning (2026.findings-acl)

Challenge:	Recent studies focus on performance benchmarks without fully comparing LLMs to graph learning models.
Approach:	They evaluate off-the-shelf and instruction-tuned graph learning models across a variety of scenarios.
Outcome:	The proposed models outperform traditional graph learning models in few-shot settings, the authors show . their models out perform models with instruction tuning, and they show excellent generalization and robustness.

Copied to clipboard

Jie Deng, Hanshuang Tong, Jun Li, Shining Liang, Ning Wu, Hongzhi Li, Yutao Xie

ReasonRec: A Reasoning-Augmented Multimodal Agent for Unified Recommendation (2026.findings-acl)

Challenge:	Large language models (LLMs) fine-tuned using rejection sampling retain only correct reasoning trajectories . however, this paradigm treats supervision as a binary filter that systematically excludes teacher-generated errors, leaving a gap in how reasoning failures are modeled during training.
Approach:	They propose a fine-tuning strategy that reframes rejection sampling as a structured supervision construction process.
Outcome:	The proposed approach outperforms RFT on multiple math benchmarks while retaining only correct reasoning trajectories.

Copied to clipboard

Yihua Zhang, Mingfu Liang, Jiyan Yang, Rong Jin, Wen-Yen Chen, Yiping Han, Huayu Li, Buyun Zhang, Liang Luo, Luke Simon, Sijia Liu, Tianlong Chen, Xi Liu

Does Chain-of-Thought Reasoning Help Mobile GUI Agents? An Empirical Study (2026.findings-acl)

Challenge:	Recent advances in multimodal recommenders lack explicit reasoning and self-awareness of uncertainty.
Approach:	They propose a reasoning-augmented multimodal agent structured around a three-stage explicit reasoning pipeline.
Outcome:	The proposed agent improves ranking metrics and performance on four standard recommendation tasks across five real-world datasets.

Copied to clipboard

Li Zhang, Longxi Gao, Mengwei Xu

Asymmetric Relational-Geometry Driven Universal Adversarial Perturbations for Vision-Language Models (2026.findings-acl)

Challenge:	Reasoning capabilities have improved vision-language models in domains like math, coding, and visual question-answering, but their impact on real-world applications remains unclear.
Approach:	They evaluate six pairs of VLMs by comparing their base and reasoning-enhanced versions across static and interactive benchmarks.
Outcome:	The reasoning-enhanced models perform better on static and interactive benchmarks than non-reasoning models.

Copied to clipboard

Jiaxin Ye, Weihai Li, Ying Wang, Simeng Qin, Zhitao Zeng, Zikai Xu

PaperScope: A Multi-Modal Multi-Document Benchmark for Agentic Deep Research Across Massive Scientific Papers (2026.findings-acl)

Challenge:	Existing universal adversarial perturbation (UAP) methods suffer from limited cross-model transferability in black-box scenarios.
Approach:	They propose an optimization-based framework that learns universal perturbations under an asymmetric relational-geometry driven objective.
Outcome:	The proposed framework outperforms state-of-the-art models in black-box transfer settings.

Copied to clipboard

Lei Xiong, Huaying Yuan, Zheng Liu, Zhao Cao, Zhicheng Dou

When Personalization Misleads: Understanding and Mitigating Hallucinations in Personalized LLMs (2026.findings-acl)

Challenge:	Existing benchmarks focus on single-document understanding, whereas real scientific workflows require integrating evidence from multiple papers.
Approach:	They propose a multi-modal multi-document benchmark for agentic deep research that integrates evidence from multiple documents.
Outcome:	Experimental results show that even advanced systems achieve limited scores on PaperScope . paper provides a rigorous benchmark alongside a pipeline for constructing large multi-modal, multi-source deep research datasets.

Copied to clipboard

Zhongxiang Sun, Yi Zhan, Chenglei Shen, Weijie Yu, Xiao Zhang, Ming He, Jun Xu

Law in Silico: Simulating Legal Society with LLM-Based Agents (2026.findings-acl)

Challenge:	Personalization can inadvertently distort factual reasoning when faced with factual queries.
Approach:	They propose a lightweight inference-time approach that mitigates personalization-induced factual distortions while preserving personalized behavior.
Outcome:	Experiments across multiple LLM backbones and personalization methods show that FPPS significantly improves factual accuracy while maintaining personalized performance.

Copied to clipboard

Yiding Wang, Yuxuan Chen, Fanxu Meng, Xifan Chen, Xiaolei Yang, Muhan Zhang

Challenge:	Large Language Models (LLMs) are powerful tools for legal simulation, but their application remains underexplored.
Approach:	They propose a unified LLM-based agent framework for simulating legal scenarios . they calibrate agent behaviors against real-world crime data .
Outcome:	The proposed framework calibrates agent behaviors against real-world crime data.

MT-Video-Bench: A Holistic Video Understanding Benchmark for Evaluating Multimodal LLMs in Multi-Turn Dialogues (2026.findings-acl)

Copied to clipboard

Yaning Pan, Qianqian Xie, Guohui Zhang, Zekun Moore Wang, Yongqian Wen, Yuanxing Zhang, Haoxuan Hu, Zhiyu Pan, Yibing Huang, Zhidong Gan, Yonghong Lin, An Ping, Shihao Li, Yanghai Wang, Tianhao Peng, Jiaheng Liu

Query-Aware Graph Attention for Precise Subgraph Retrieval in Knowledge-Augmented Reasoning (2026.findings-acl)

Challenge:	Existing evaluation benchmarks for Multimodal Large Language Models (MLLMs) focus on single-turn question answering, overlooking the complexity of multi-turn dialogues in real-world scenarios.
Approach:	They propose a video understanding benchmark for MLLMs in multi-turn dialogues that assesses six core competencies that focus on perceptivity and interactivity.
Outcome:	The MT-Video-Bench evaluates 1,000 multi-turn dialogues from diverse domains and reveals significant performance discrepancies and limitations in handling multi-turned video dialogues.

Copied to clipboard

Yuanye Xu, Linyi Guo, Yue Zhang, Fu Ning

LIBERTy: A Causal Framework for Benchmarking Concept-Based Explanations of LLMs with Structural Counterfactuals (2026.findings-acl)

Challenge:	Existing Knowledge Graph (KG)-based Retrieval-Augmented Generation (RAG) systems insufficiently model the interaction between query semantics and relation types, resulting in imprecise subgraph retrieval and unstable reasoning.
Approach:	They propose a retrieval framework that integrates query semantics and relation embeddings directly into the attention mechanism.
Outcome:	Experiments on WebQSP and CWQ establish new state-of-the-art results in both Triple Recall and Answer Recall.

Copied to clipboard

Gilat Toker, Nitay Calderon, Ohad Amosy, Roi Reichart

Contract-Coding: Towards Repo-Level Generation via Structured Symbolic Paradigm (2026.findings-acl)

Challenge:	Concept-based explanations quantify how high-level concepts influence model behavior . existing benchmarks rely on costly human-written counterfactuals that serves as imperfect proxy .
Approach:	They propose a framework for constructing datasets containing structural counterfactual pairs . they use a structured Causal Model to generate a concept-based explanation .
Outcome:	The proposed framework compares concept-based explanations to causal effects estimated from counterfactuals.

Copied to clipboard

null Yilin, Lujin Zhao, Yijie Shi

Challenge:	Spec-Coding, a framework for coding with vague intents, is a critical step toward repository-scale autonomous engineering.
Approach:	They propose a structured symbolic paradigm that bridges unstructured intent and executable code via Autonomous Symbolic Grounding.
Outcome:	The proposed framework achieves 47% functional success while maintaining near-perfect structural integrity.

Adaptive Text Anonymization: Learning Privacy-Utility Trade-offs via Prompt Optimization (2026.findings-acl)

Copied to clipboard

Gabriel Loiseau, Damien Sileo, Damien Riquet, Maxime Meyer, Marc Tommasi

SWE-QA: Can Language Models Answer Repository-level Code Questions? (2026.findings-acl)

Challenge:	Existing methods for anonymizing textual documents lack flexibility to adapt to diverse requirements.
Approach:	They propose a task formulation in which anonymization strategies are automatically adapted to specific privacy–utility requirements.
Outcome:	The proposed framework achieves better privacy–utility trade-off than existing baselines on open-source language models while remaining computationally efficient and effective on larger closed-source models.

Copied to clipboard

Weihan Peng, Yuling Shi, Yuhang Wang, Xinyun Zhang, Beijun Shen, Xiaodong Gu

Efficient PRM Training Data Synthesis via Formal Verification (2026.findings-acl)

Challenge:	Existing benchmarks for understanding and reasoning about entire soft-ware repositories focus on small, self-contained code snippets.
Approach:	They propose a repository-level code question answering benchmark to facilitate research on automated QA systems in real-world repositories.
Outcome:	The proposed benchmarks are designed to facilitate research on automated QA systems in real-world repositories.

Copied to clipboard

Ryo Kamoi, Yusen Zhang, Nan Zhang, Sarkar Snigdha Sarathi Das, Ranran Haoran Zhang, Wenpeng Yin, Rui Zhang

Revealing the Attention Floating Mechanism in Masked Diffusion Models (2026.findings-acl)

Challenge:	Existing approaches for constructing PRM training data rely on human annotation or sampling-based labeling methods that require repeated LLM calls.
Approach:	They propose a framework that synthesizes PRM training data by annotating step-level error labels using formal verification tools such as Z3 and Isabelle.
Outcome:	The proposed framework synthesizes PRM training data from formal logic and theorem proving tasks without human annotation or additional LLM calls.

Copied to clipboard

Xin Dai, Pengcheng Huang, Zhenghao Liu, Shuo Wang, Yukun Yan, Chaojun Xiao, Yu Gu, Ge Yu, Maosong Sun

Balancing Knowledge Breadth and Task Depth for Effective Domain Adaptation Fine-Tuning (2026.findings-acl)

Challenge:	Masked diffusion models (MDMs) leverage bidirectional attention and a denoising process.
Approach:	They investigate the attention behaviors of Masked diffusion models by revealing the phenomenon of Attention Floating.
Outcome:	The proposed model doubles the performance of autoregressive models in knowledge-intensive tasks.

Copied to clipboard

Mu Zhang, Yuxiang Chu, Guangya Yu, Yongqi Fan, Weiyan Zhang, Hang Hu, Tong Ruan, Jingping Liu

Explain the Flag: Contextualizing Hate Speech Beyond Censorship (2026.findings-acl)

Challenge:	a lack of knowledge breadth and task depth can hinder curriculum learning in domains such as medicine and finance.
Approach:	They propose a two-dimensional curriculum learning framework that coordinates model training along two orthogonal axes: the knowledge dimension and the task dimension.
Outcome:	The proposed framework improves accuracy on medical evaluations by 2.49% and on financial evaluations 1.2% compared with the second-best method.

Copied to clipboard

Jason Liartis, Eirini Kaldeli, Lamprini Gyftokosta, Eleftherios Chelioudakis, Orfeas Menis Mastromichalakis

EnvScaler: Scaling Tool-Interactive Environments for LLM Agent via Programmatic Synthesis (2026.findings-acl)

Challenge:	a hybrid approach to detect and explain hate speech combines large language models with vocabularies to detect hate speech in three languages . authors: the spread of hate speech online has serious personal, social, and legal consequences . eu has launched initiatives to analyze, regulate, and counteract online hate speech, authors say .
Approach:	They propose a hybrid approach that combines Large Language Models with vocabularies to detect hate speech in English, French, and Greek.
Outcome:	The proposed approach outperforms baselines in English, French, and Greek . it uses large language models and vocabularies to detect and explain hate speech . human evaluation shows that the proposed approach is accurate and clear .

Copied to clipboard

Xiaoshuai Song, Haofei Chang, Guanting Dong, Yutao Zhu, Ji-Rong Wen, Zhicheng Dou

Iterative Knowledge Graph Refinement and Integration for Medical Question Answering (2026.findings-acl)

Challenge:	Large language models (LLMs) are expected to be trained to act as agents in various real-world environments, but access to real systems is often restricted and manually built sandboxes are hard to scale.
Approach:	They propose an automated framework for scalable tool-interaction environments via programmatic synthesis that synthesizes 191 environments and about 7K scenarios and applies them to Supervised Fine-Tuning and Reinforcement Learning for Qwen3 series models.
Outcome:	The proposed framework significantly improves LLMs’ ability to solve tasks in complex environments involving multi-turn, multi-tool interactions.

Copied to clipboard

Haochen Zou, Yongli Wang

Lang2Act: Fine-Grained Visual Reasoning through Self-Emergent Linguistic Toolchains (2026.findings-acl)

Challenge:	Existing graph-based RAG methods heuristically retrieve and refine question-relevant subgraphs, potentially introducing redundant and noisy factual information that is difficult for LLMs to process.
Approach:	They propose to integrate knowledge graphs (KGs) through retrieval-augmented generation methods to improve LLM reasoning by incorporating external trustworthy knowledge resources.
Outcome:	The proposed framework achieves state-of-the-art against baseline competitors on three medical QA benchmark datasets.

Copied to clipboard

Yuqi Xiong, Chunyi Peng, Zhipeng Xu, Zhenghao Liu, Zulong Chen, Yukun Yan, Shuo Wang, Yu Gu, Ge Yu

Lost in Stories: Consistency Bugs in Long Story Generation by LLMs (2026.findings-acl)

Challenge:	Existing frameworks depend on rigid, pre-defined external tools to extend perceptual capabilities of VLMs.
Approach:	They propose a framework that leverages self-emergent linguistic toolchains to enhance visual perception and reasoning.
Outcome:	The proposed framework improves the visual perception capabilities of large language models by incorporating external visual documents to address a given query.

Copied to clipboard

Junjie Li, Xinrui Guo, Yuhao Wu, Roy Ka-Wei Lee, Hongzhi Li, Yutao Xie

Chinese Court Simulation with LLM-Based Agents System (2026.findings-acl)

Challenge:	Existing story generation benchmarks focus mainly on plot quality and fluency, leaving consistency errors unexplored.
Approach:	They propose a benchmark to evaluate narrative consistency in long-form story generation.
Outcome:	Evaluating LLMs, we find consistency errors are common in factual and temporal dimensions . authors say the findings can inform future efforts to improve consistency in long-form narrative generation.

Copied to clipboard

Kaiyuan Zhang, Jiaqi Li, Yueyue Wu, Haitao Li, Cheng Luo, Shaokun Zou, Yujia Zhou, Weihang Su, Yiqun Liu, Qingyao Ai

SPASM: Stable Persona-driven Agent Simulation for Multi-turn Dialogue Generation (2026.findings-acl)

Challenge:	Existing studies have neglected the systematic design and procedure evaluation of court simulations, which are critical to the credibility and usage of court simulators in practice.
Approach:	They propose a court simulation paradigm based on the real-world procedure structure of Chinese courts and a framework that focuses on both legal judgment prediction and court procedure analysis.
Outcome:	The proposed model outperforms judges and lawyers from the real trials in many aspects.

Copied to clipboard

Han Luo, Guy Laban

Telling Speculative Stories to Help Humans Imagine the Harms of Healthcare AI (2026.findings-acl)

Challenge:	Large language models are increasingly deployed in multi-turn settings such as tutoring, support, and counseling where reliability depends on preserving consistent roles, personas, and goals across long horizons.
Approach:	They propose a framework that decomposes LLM–LLM conversations into a modular, stability-first framework that allows for a stable persona-driven agent simulation for multi-turn dialogue generation.
Outcome:	The proposed framework decomposes the LLM-based model into four main components: persona creation, plausibility validation, and natural-language persona crafting.

Copied to clipboard

Xingmeng Zhao, Tongnian Wang, Dan Schumacher, Veronica Rammouz, Anthony Rios

Masking or Mitigating? Deconstructing the Impact of Query Rewriting on Retriever Biases in RAG (2026.findings-acl)

Challenge:	Artificial intelligence (AI) is rapidly transforming healthcare but can also introduce risks, including bias, privacy violations, and unequal access.
Approach:	They propose a framework that generates speculative user stories and supports multi-agent discussions to help people reflect on potential benefits and harms of healthcare AI before deployment.
Outcome:	The framework generates speculative user stories and supports multi-agent discussions to help people reflect on potential benefits and harms of healthcare AI before deployment.

Copied to clipboard

Agam Goyal, Koyel Mukherjee, Apoorv Saxena, Anirudh Phukan, Eshwar Chandrasekharan, Hari Sundaram

Unifying Inference-Time Planning Language Generation (2026.findings-acl)

Challenge:	Query enhancement techniques are now standard in retrieval-augmented generation systems, yet their impact on these biases remains unexplored.
Approach:	They evaluate query enhancement techniques that improve retrieval quality . they find that simple rewriting reduces bias through increased score variance . no technique uniformly addresses all biases, and effects vary substantially across retrievers .
Outcome:	The proposed method achieves strongest aggregate reduction, but fails under adversarial conditions where multiple biases combine.

Copied to clipboard

Prabhu Prakash Kagitha, Bo Sun, Ishan Desai, Andrew Zhu, Cassie Huang, Manling Li, Ziyang Li, Li Zhang

EmoRes: Toward Adaptive Psychological Support via User-Agnostic Benchmark and Topic-Mining Agent (2026.findings-acl)

Challenge:	Large language models (LLMs) are used to generate a formal representation of a plan in a planning language.
Approach:	They propose a unifying organizational framework based on intermediate representations to unify the inference-time LLM-as-formalizer methodology for classical planning.
Outcome:	The proposed framework subsumes most existing work and proposes new ones that involve syntactically similar but high-resource intermediate languages.

Copied to clipboard

Zhengwei Zou, Xuanming Jiang, Baoyi An, Dingyu Nie, Zhengxing Fang, Qingyu Liu, Xueming Qian, Guoshuai Zhao, Zhongyu Yang

Budget-Aware Anytime Reasoning with LLM-Synthesized Preference Data (2026.findings-acl)

Challenge:	Large language models generate fragmented and emotionally inconsistent dialogues lacking the therapeutic structure necessary for reliable assessment.
Approach:	They propose a framework that boosts psychological reasoning via a Topic-Mining Emotional Agent and a multi-perspective Self-Reflection Agent.
Outcome:	The proposed framework improves topic continuity, emotional coherence, and clinical interpretability over baselines and validated by ablation studies and human evaluations.

Copied to clipboard

Xuanming Zhang, Shwan Ashrafi, Aziza Mirsaidova, Amir H. Rezaeian, Miguel Ballesteros, Lydia Chilton, Zhou Yu, Dan Roth

Mitigating Cultural Bias in LLMs via Multi-Agent Cultural Debate (2026.findings-acl)

Challenge:	Recent work has explored reasoning efficiency via test-time scaling and early exit strategies.
Approach:	They propose an anytime reasoning framework and the Anytime Index to improve model quality . they also propose an inference-time self-improvement method to produce better intermediate solutions .
Outcome:	The proposed method improves on NaturalPlan, AIME, and GPQA datasets and improves reasoning quality and efficiency under budget constraints.

Copied to clipboard

Qian Tan, Lei Jiang, Yuting Zeng, Shuoyang Ding, Xiaohua Xu

ToolRM: Towards Agentic Tool-Use Reward Modeling (2026.findings-acl)

Challenge:	Existing approaches to evaluate large language models fail to address cultural bias in non-Western languages . Chinese prompting shifts bias toward East Asian perspectives rather than eliminating it, authors say .
Approach:	They propose a Chinese–English bilingual benchmark and multi-agent vote frameworks that enable explicit "no bias" judgments.
Outcome:	The proposed framework achieves 57.6% average No Bias Rate on Chinese-English benchmark and 86.0% on Arabic CAMeL benchmark.

Copied to clipboard

Renhao Li, Jianhong Tu, Yang Su, Yantao Liu, Fei Huang, Hamid Alinejad-Rokny, Derek F. Wong, Junyang Lin, Min Yang

Challenge:	lack of reliable reward models for tool-use tasks has limited progress toward agentic AI . recent advances in agentic artificial intelligence are driven by tool-using capabilities of large language models.
Approach:	They propose a pipeline that constructs pairwise preference data using rule-based scoring and multidimensional sampling to build lightweight reward models.
Outcome:	The proposed model outperforms existing models on tool calling tasks with higher accuracy.

MDocRAG-RL: Empowering Multi-Modal Document RAG via Complex Visual Reasoning with Reinforcement Learning (2026.findings-acl)

Copied to clipboard

Zhongyu Wang

MARS: Unleashing the Power of Speculative Decoding via Margin-Aware Verification (2026.findings-acl)

Challenge:	Existing RAG systems produce suboptimal embeddings and naively insert images into context without adequate visual perception, limiting reasoning capabilities.
Approach:	They propose a novel RAG framework for complex visual reasoning that integrates multimodal large language models with external knowledge to enhance retrieval efficiency.
Outcome:	The proposed framework achieves state-of-the-art performance on multiple benchmarks.

Copied to clipboard

Jingwei Song, Xinyu Wang, Hanbin Wang, Xiaoxuan Lei, Tianyu Shi, Shixin Han, Eric Yang, Xiao-Wen Chang, Lynn Ai

Challenge:	Autoregressive large language models suffer from high inference latency due to memorybandwidth constraints.
Approach:	They propose a method that decouples generation and verification by decoupling tokens and a lightweight draft model.
Outcome:	The proposed method delivers consistent and significant speedups over state-of-the-art baselines while preserving generation quality across diverse benchmarks.

AI, Take the Wheel: What Drives Delegation and Trust in Human–Computer Cooperative Question Answering? (2026.findings-acl)

Copied to clipboard

Maharshi Gor, Yoo Yeon Sung, Yu Hou, Eve Fleisig, Zhu Irene Ying, Tianyi Zhou, Jordan Lee Boyd-Graber

Self-Reinforcing Controllable Synthesis of Rare Relational Data via Bayesian Calibration (2026.findings-acl)

Challenge:	Human-AI collaboration is already happening, both in proactive delegation and deliberative adoption settings.
Approach:	They study delegating a task to AI without seeing its output and evaluating AI suggestions to decide whether to adopt them how AI output shapes final decisions.
Outcome:	The proposed game pairs 23 experts with 16 AI agents, capturing 387 delegation and 1440 adoption decisions.

Copied to clipboard

Chongsheng Zhang, Hao Wang, Zelong Yu, Esteban Garces Arias, Julian Rodemann, Zhanshuo Zhang, Qilong Li, Gaojuan Fan, Krikamol Muandet, Christian Heumann

WaveDetect: Robust Framework for Machine-Generated Text Detection via Wavelet Transform (2026.findings-acl)

Challenge:	Existing approaches to synthesis of relational/structured tabular data lack effective feedback mechanism to optimize quality of generated data.
Approach:	They propose a relational data generator with dynamic guidance framework that uses chain-of-thought steps to generate tabular data for enhancing downstream imbalanced classification performance.
Outcome:	The proposed framework outperforms existing approaches in both data fidelity and downstream imbalanced classification performance on real and synthetic datasets.

Copied to clipboard

Zhichen Liu, Kaitong Qin, Linhan He, Yang Xu

AHEAD: Attention Head Energy-Aware Dynamics for Hallucination Mitigation in MLLMs (2026.findings-acl)

Challenge:	Existing methods for detecting LLM-generated texts falter when faced with adversarial perturbations, cross-domain shifts, and the rapid temporal evolution of the foundation model.
Approach:	They propose a framework that reformulates text detection as a signal processing task within the time-frequency domain.
Outcome:	The proposed framework achieves superior accuracy and robustness against sophisticated attacks and generalization across out-of-distribution topics.

Copied to clipboard

Jiale Chang, Ying Li, Siliang Tang, Yueting Zhuang

TopoSHIELD: Reshaping the Flow of Malice via Spatio-Temporal Risk-Aware Topological Evolution in Multi-Agent Systems (2026.findings-acl)

Challenge:	Existing approaches to hallucination mitigation ignore heterogeneous behaviors of attention heads . hallucinosity is a critical barrier to multimodal large language models' reliability, authors say .
Approach:	They propose a framework that quantifies the energetic properties of each attention head during object generation through two potential networks and dynamically adjusts their contributions at inference time.
Outcome:	The proposed framework reduces hallucination rates without fine-tuning the base model while maintaining generation quality.

Copied to clipboard

Ruiyang Huang, Chenxi Wang, Tinghe Zhang, Fengrui Liu, Jiayan Sun, Haocheng Wang, Yifan Wu

Diagnosing and Mitigating Sycophancy and Skepticism in LLM Causal Judgment (2026.findings-acl)

Challenge:	Multi-agent systems (MAS) inherit general task-solving and instruction-following capabilities, but their interconnectivity introduces significant security risks.
Approach:	They propose a framework that reshapes the flow of malice via risk-aware topological evolution.
Outcome:	Empirically, TopoSHIELD reduces toxicity by 58% on GPT-4o while preserving high utility (>90% success rate).

Copied to clipboard

Edward Y Chang

SuperWriter: Reflection-Driven Long-Form Generation with Large Language Models (2026.findings-acl)

Challenge:	Current evaluations obscure the answer to causal judgment in frontier models.
Approach:	They introduce a process-integrity evaluator that checks whether a model's answer is entailed by its own derivation, internally consistent, and not dominated by user hints under pressure.
Outcome:	The proposed model fails to distinguish between the two pathologies.

Copied to clipboard

Yuhao Wu, Yushi Bai, Zhiqiang Hu, Juanzi Li, Roy Ka-Wei Lee

ProxyPrompt: Securing System Prompts against Prompt Extraction Attacks (2026.findings-acl)

Challenge:	Long-form text generation remains a challenge for large language models . generating extended sequences often leads to degraded coherence and logical consistency .
Approach:	They propose a framework that integrates explicit structured thinking into long-form text generation.
Outcome:	The proposed framework surpasses even larger-scale models in evaluation and human evaluation.

Copied to clipboard

Zhixiong Zhuang, Maria-Irina Nicolae, Hui-Po Wang, Mario Fritz

Cogito: A Cognitive Agentic Framework Driven by Dynamic Graph of Thoughts for Financial Report Generation (2026.findings-acl)

Challenge:	Existing defenses against prompt extraction are either easily bypassed or require constant updates to address new threats.
Approach:	They propose a new mechanism that replaces the original prompt with a proxy to prevent prompt leakage by obfuscating the extracted prompt.
Outcome:	The proposed defense outperforms the existing defense, which only achieves 42.80% of the prompts extracted from the original task.

Copied to clipboard

Chen Lifan, Wei Ding, Jingwen Yang, Xiuze Zhou, Jingan Chen, Fan Lin

From Coarse to Fine: A Multi-Granularity Multimodal Framework for Teacher Sentiment Analysis (2026.findings-acl)

Challenge:	Existing approaches to financial report generation are insufficient to handle dynamic uncertainties of real-world financial environments.
Approach:	They propose a cognitively grounded agentic framework for professional financial report generation that is driven by Dynamic Graph of Thoughts and a social collaboration mechanism to facilitate coordinated agent interaction.
Outcome:	The proposed framework is based on a dynamic reasoning model and social collaboration mechanism.

Copied to clipboard

Zhiyi Duan, Xiangren Wang, Jiangshan Guan, Bing Jia, Qianli Xing

PBEBench: A Multi-Step Programming by Examples Reasoning Benchmark inspired by Historical Linguistics (2026.findings-acl)

Challenge:	Existing approaches to teacher sentiment analysis treat it as a static label . current approaches fail to capture structured heterogeneity of classroom expressions .
Approach:	They propose a coarse-to-fine multimodal framework that decomposes teacher sentiment into three granularities and employ CLS-guided cross-modal attention to recover effective signals from regulated displays.
Outcome:	The proposed framework outperforms state-of-the-art models on T-MED and CMU-MOSEI.

Copied to clipboard

Atharva Naik, null Prakam, Yash Mathur, Darsh Agrawal, Manav Nitin Kapadnis, Yuwei An, Clayton Marr, Carolyn Rose, David R. Mortensen

T2I-ReasonBench: Benchmarking Reasoning-Informed Text-to-Image Generation (2026.findings-acl)

Challenge:	a benchmark for inductive reasoning is based on sound law induction in historical linguistics . solve rates are below 5% on hard PBEBench instances with long program cascades despite expensive scaling strategies .
Approach:	They propose a benchmark for inductive reasoning inspired by sound law induction in historical linguistics.
Outcome:	The proposed approach generates problems with controllable difficulty and ordering constraints . solve rates remain below 5% on hard PBEBench instances with long program cascades .

Copied to clipboard

Kaiyue Sun, Rongyao Fang, Chengqi Duan, Xian Liu, Aoxue Li, Xihui Liu

CGBridge: Bridging Code Graphs and Large Language Models for Better Structure-Aware Code Understanding (2026.findings-acl)

Challenge:	Text-to-image (T2I) generative models have demonstrated exceptional capability in synthesizing high-quality images from textual prompts.
Approach:	They propose a benchmark to explore the knowledge-driven reasoning capabilities of T2I models.
Outcome:	The proposed benchmark examines the knowledge-driven reasoning capabilities of T2I models.

Copied to clipboard

Zeqi Chen, Zhaoyang Chu, Yi Gui, Feng Guo, Yao Wan, Chuan Shi

Candidate-Aware Retrieval and Reranking for Multiple-Choice Question Answering: Arabic as a Case Study (2026.findings-acl)

Challenge:	Existing structure-aware approaches treat structure as serialized text prompts or auxiliary training objectives, failing to provide explicit guidance during inference.
Approach:	They propose a plug-and-play method that enhances Large Language Models with Code Graph information through an external, trainable Bridge module.
Outcome:	The proposed method decouples structural reasoning from textual generation without updating the backbone.

Copied to clipboard

Yassine Bouziane, Youness Moukafih, Mounir Ghogho

Logical Structure as Knowledge: Enhancing LLM Reasoning via Structured Logical Knowledge Density Estimation (2026.findings-acl)

Challenge:	Large language models (LLMs) have recently achieved impressive results on multiple-choice question answering (MCQA) despite advances in English, LLMs continue to underperform in Arabic due to gaps in data coverage, linguistic transfer, and evaluation design.
Approach:	They propose a method that jointly models the relevance of both the question and its candidate answers when selecting contextual passages.
Outcome:	The proposed approach outperforms standard RAG baselines and reranker baselines while remaining competitive with considerably larger models.

Copied to clipboard

Zhen Bi, Zhenlin Hu, Xueshu Chen, Mingyang Chen, Cheng Deng, Yida Xue, Zhen Wang, Qing Shen, Ningyu Zhang, Jungang Lou

ACSE: An Ancient Character Semantic-Aware Embedding for Large Language Models (2026.findings-acl)

Challenge:	Existing data-centric paradigms equate quality with factuality or diversity and ignore the internal logical complexity of training samples.
Approach:	They propose a density-aware re-cognizing optimization strategy that prioritizes high-density logical samples to align training with the model's reasoning boundary.
Outcome:	The proposed metric outperforms existing methods and improves reasoning performance without increasing total data volume.

Copied to clipboard

Zhihan Zhou, Daqian Shi, Lida Shi, Rui Song, Peiqiang Qiu, Xiaolei Diao, Hao Xu

Reasoning as Gradient: Scaling MLE Agents Beyond Tree Search (2026.findings-acl)

Challenge:	Existing studies on pre-Qin documents are insufficient to understand ancient characters . ancient characters have a low level of digitization and training corpora are extremely scarce .
Approach:	They propose a semantic-aware embedding for ancient Chinese characters that integrates glyphs and lexicality into modern Chinese semantic space.
Outcome:	The proposed model integrates glyph and lexicality of ancient characters and maps them to the modern Chinese semantic space.

Copied to clipboard

Yifei Zhang, Xu Yang, Xiao Yang, Bowen Xian, Qizheng Li, Shikai Fang, Jingyuan Li, Jian Wang, Minrui Xu, Yuge Zhang, Weiqing Liu, Jiang Bian

Speak No Evil, Just Prompt: Low-resource Multilingual Toxic Speech Detection with Audio Language Model (2026.findings-acl)

Challenge:	LLM-based agents for machine learning engineering rely on tree search to rank candidates.
Approach:	They propose an LLM-based agent that operationalizes gradient-based optimization.
Outcome:	The proposed agent achieves a state-of-the-art 35.1% any-medal rate on MLE-Bench with a limited budget on a single GPU.

Copied to clipboard

Mingzi Zuo, Lei Zhang, Hailiang Sun, Shengzhi Huo, Changyu Dong, Xin Wang, Bo Wang, Hao Liu

EmotionTalk: An Interactive Chinese Multimodal Emotion Dataset With Rich Annotations (2026.findings-acl)

Challenge:	Existing methods for toxic speech detection rely on high-resource languages and lack acoustic cues.
Approach:	They propose a prompt-based adaptation framework that performs end-to-end toxicity detection without ASR.
Outcome:	The proposed framework achieves a micro-averaged ROC-AUC of 98.07% on polySpeechTox . it is based on a frozen audio language model and can perform end-to-end toxicity detection without ASR .

Copied to clipboard

Haoqin Sun, Jinghua Zhao, Xuechen Wang, Shiwan Zhao, Jiaming Zhou, Hui Wang, Xi Yang, Yequan Wang, Yonghua Lin

What Prompts Don’t Say: Understanding and Managing Underspecification in LLM Prompts (2026.findings-acl)

Challenge:	Existing datasets face issues such as low quality, limited scale, and incomplete modalities, hindering model performance.
Approach:	They propose to use Chinese multimodal datasets to capture authentic emotional interplay from 19 professional actors.
Outcome:	The EmotionTalk dataset spans 23.6 hours of dyadic conversations across diverse scenarios.

Copied to clipboard

Chenyang Yang, Yike Shi, Qianou Ma, Michael Xieyang Liu, Christian Kaestner, Tongshuang Wu

AffectCodec: Emotion-Preserving Neural Speech Codec for Expressive Speech Modeling (2026.findings-acl)

Challenge:	Under-specified prompts are 2x as likely to regress across model or prompt changes, authors show . eliot safina: a lack of explicit prompts can cause frustrations and failures .
Approach:	They propose requirements-aware prompt optimization mechanisms that improve performance by 4.8% over baselines.
Outcome:	The proposed mechanisms improve prompt performance by 4.8% over baselines.

Copied to clipboard

Jiacheng Shi, Hongfei Du, Xinyuan Song, Y. Alicia Hong, Yanfu Zhang, Ashley Gao

Agentic Conversational Search with Contextualized Reasoning via Reinforcement Learning (2026.findings-acl)

Challenge:	Existing codecs optimize acoustic reconstruction, leaving emotion expressiveness insufficiently modeled at the representation level.
Approach:	They propose an emotion-guided neural speech codec that preserves emotional information while maintaining semantic fidelity and prosodic naturalness.
Outcome:	The proposed codec preserves emotional cues while maintaining semantic fidelity and prosodic naturalness.

Copied to clipboard

Fengran Mo, Yifan Gao, Sha Li, Hansi Zeng, Xin Liu, Zhaoxuan Tan, Xian Li, Jianshu Chen, Dakuo Wang, Meng Jiang

Do Multimodal RAG Systems Leak Data? A Comprehensive Evaluation of Membership Inference and Image Caption Retrieval Attacks (2026.findings-acl)

Challenge:	Existing studies focus on single-turn scenarios, which might lack the ability to handle multi-turn interactions.
Approach:	They propose a conversational agent that interleaves search and reasoning across turns and provides tailored rewards towards evolving user goals.
Outcome:	The proposed agent interleaves search and reasoning across turns, enabling exploratory and adaptive behaviors learned through reinforcement learning (RL) training with tailored rewards towards evolving user goals.

Copied to clipboard

Ali Al-Lawati, Suhang Wang

PED: Route-Decoupled Diagnostics for Persona Consistency in Spoken Agents (2026.findings-acl)

Challenge:	Multimodal retrieval-augmented generation (mRAG) pipelines are becoming more popular for vision-centric tasks.
Approach:	They propose to use a visual asset as a trigger to leak data from a model prompt.
Outcome:	The proposed pipelines can connect private datasets and improve model performance, but they can leak private information from them.

Copied to clipboard

Weihao Liu, Junrui Wei, Zhao Zhang, Ju Zhang

Interpretable Semantic Gradients in SSD: A PCA Sweep Approach and a Case Study on AI Discourse (2026.findings-acl)

Challenge:	Existing evaluations of persona-emotion decoupling frameworks do not isolate which component caused the failure . current evaluations do little to isolate the cause of the failure, making fixes slow and ad hoc .
Approach:	They propose a diagnostic evaluation framework that decomposes persona expression into two observable routes.
Outcome:	The proposed framework decomposes persona expression into two observable routes . it can be used to perform route-comparable, reference-based analyses of separability, drift, failures and coupling .

Copied to clipboard

Hubert Plisiecki, Maria Leniarska, Jan Piotrowski, Marcin Zajenkowski

Self-Sum: Teaching an Agent to Decide Itself When and What to Summarize (2026.findings-acl)

Challenge:	Supervised Semantic Differential (SSD) is a mixed quantitative–interpretive method that models how text meaning varies with continuous individual-difference variables . currently no systematic method exists for choosing the number of retained components, introducing avoidable researcher degrees of freedom in the analysis pipeline.
Approach:	They propose a PCA sweep procedure that treats dimensionality selection as a joint criterion over representation capacity, gradient interpretability, and stability across nearby values of K.
Outcome:	The proposed method is based on a corpus of short posts about artificial intelligence written by Prolific participants who also completed Admiration and Rivalry narcissism scales.

Copied to clipboard

Hongru Wang, Rui Wang, Jushi Kai, Boyang Xue, Yongqi Li, Shijue Huang, Xiaoteng Ma, Jeff Z. Pan, Amos Storkey

GreekMMLU: A Native-Sourced Multitask Benchmark for Evaluating Language Models in Greek (2026.findings-acl)

Challenge:	Existing methods for summarizing long-horizon agents rely on fixed, rule-based summarization strategies.
Approach:	They propose a framework that empowers agents to autonomously decide when and what to summarize by modeling it as an internal cognitive action unified with environmental actions.
Outcome:	The proposed framework outperforms no-summarization and rule-based training methods on long-horizon benchmarks and shows strong generalization gains.

Copied to clipboard

Yang Zhang, Mersin Konomi, Christos Xypolopoulos, Konstantinos Divriotis, Konstantinos Skianis, Giannis Nikolentzos, Giorgos Stamou, Guokan Shang, Michalis Vazirgiannis

User-Assistant Bias in LLMs (2026.findings-acl)

Challenge:	Existing evaluation benchmarks for large language models are limited for Greek . Existing datasets are often machine-translated from English, failing to capture Greek linguistic and cultural characteristics.
Approach:	They propose a native-sourced benchmark for massive multitask language understanding in Greek . they publicize 16,857 samples and reserve 4,948 samples for a private leaderboard .
Outcome:	The proposed model is based on 21,805 multiple-choice questions across 45 subject areas . the model is publicly released and reserved for a private leaderboard .

Copied to clipboard

Xu Pan, Jingxuan Fan, Zidi Xiong, Ely Hahami, Jorin Overwiening, Ziqian Xie

ICLAD: In-Context Learning with Comparison-Guidance for Audio Deepfake Detection (2026.findings-acl)

Challenge:	Modern large language models are typically trained using structured role tags . asymmetries in training data associated with different role tags can potentially introduce inductive biases.
Approach:	They propose a task-agnostic benchmark to test user–assistant bias in large language models . they find human-preference alignment amplifies user bias, reasoning fine-tuning reduces it .
Outcome:	The proposed benchmark tests show that most instruction-tuned models exhibit strong user bias . human-preference alignment amplifies user bias, while reasoning fine-tuning reduces it.

Copied to clipboard

Benjamin Shiue-Hal Chou, Yi Zhu, Surya Koppisetti

ReStyle-TTS: Relative and Continuous Style Control for Zero-Shot Speech Synthesis (2026.findings-acl)

Challenge:	Audio deepfake detection systems do not generalize well to realistic in-the-wild deepfakkes.
Approach:	They propose a novel In-Context Learning paradigm with comparison-guidance for Audio Deepfake detection framework that uses audio language models for training-free generalization to unseen deepfakes.
Outcome:	The proposed framework improves macro F1 over specialized detectors on in-the-wild datasets with up to 2 relative improvement over existing models.

Copied to clipboard

Haitao Li, Chunxiang Jin, Chenglin Li, Wenhao Guan, Zhengxing Huang, Xie Chen

Who Gets Which Message? Auditing Demographic Bias in LLM-Generated Targeted Text (2026.findings-acl)

Challenge:	Recent controllable zero-shot text-to-speech systems can synthesize speech for unseen speakers from a short reference audio clip, but they also inherit the speaking style present in the reference.
Approach:	They propose a framework that enables continuous and reference-relative style control in zero-shot text-to-speech systems by combining style-specific LoRAs with Orthogonal LoRA Fusion.
Outcome:	The proposed framework reduces the model's dependence on reference style while preserving text fidelity while maintaining intelligibility and speaker timbre.

Copied to clipboard

Tunazzina Islam

Challenge:	Large language models generate demographically conditioned persuasive texts at scale . authors argue that such capabilities raise questions about fairness and representational bias in automated communication.
Approach:	They propose a framework for evaluating demographic-conditioned targeted messages . they find gender- and age-based asymmetries in male- and youth-targeted messages a .
Outcome:	The proposed framework evaluates generated messages across three dimensions: lexical content, language style, and persuasive framing.

FACTS: Table Summarization via Offline Template Generation with Agentic Workflows (2026.findings-acl)

Copied to clipboard

Ye Yuan, Mohammad Amin Shabani, Siqi Liu

How Language Models Conflate Logical Validity with Plausibility: A Representational Analysis of Content Effects (2026.findings-acl)

Challenge:	Existing methods for query-focused table summarization struggle with complex reasoning and token-limit issues.
Approach:	They propose a Fast, Accurate, and Privacy-Compliant table summarization approach via Offline Template Generation.
Outcome:	The proposed method outperforms baseline methods on widely-used benchmarks.

Copied to clipboard

Leonardo Bertolazzi, Sandro Pezzelle, Raffaella Bernardi

LLMs Deceive Unintentionally: Emergent Misalignment in Dishonesty from Misaligned Samples to Biased Human-AI Interactions (2026.findings-acl)

Challenge:	a number of theories have been proposed to account for content effects in large language models, including the dual-process theory of reasoning, but the mechanisms behind content effects remain unclear.
Approach:	They propose to encode validity and plausibility concepts in LLMs by aligning them in representational geometry.
Outcome:	The proposed model conflates validity and plausibility, and vice versa.

Copied to clipboard

Xuhao Hu, Wang Peng, Xiaoya Lu, Dongrui Liu, Xuanjing Huang, Jing Shao

AlphaQuanter: An End-to-End Tool-Augmented Agentic Reinforcement Learning Framework for Stock Trading (2026.findings-acl)

Challenge:	Existing studies have shown that LLMs finetuned on incorrect completions can exhibit harmful behaviors, which is called emergent misalignment.
Approach:	They investigate whether LLMs finetuned on incorrect completions can exhibit harmful behaviors . they find that 1% of misalignment data is sufficient to decrease honest behavior .
Outcome:	The proposed model can be misaligned on errors within narrow domains to exhibit harmful behaviors . the proposed model is able to exhibit dishonest behavior with only 10% biased user population .

Copied to clipboard

Zheye Deng, Weixiang Yan, Changlong Yu, Jiashu Wang

Distribution Shift Alignment Helps LLMs Simulate Survey Response Distributions (2026.findings-acl)

Challenge:	Existing Large Language Models (LLMs) lack the end-to-end optimization needed to learn a coherent strategy from market feedback.
Approach:	They propose a single-agent framework that uses reinforcement learning to learn a dynamic policy over a transparent decision workflow.
Outcome:	The proposed framework achieves state-of-the-art performance on key financial metrics.

Copied to clipboard

Ji Huang, Mengfei LI, Shuai Shao

PhysPRM: A Generative Process Reward Model with Fine-grained Diagnosis for Physics Problem Solving (2026.findings-acl)

Challenge:	Existing methods to simulate survey responses are based on zero-shot methods, but they are sensitive to prompt changes and deviate from the real-world distributions.
Approach:	They propose a distribution shift alignment method that aligns both the output distributions and the distribution shifts across different backgrounds to provide results closer to the true distribution than the training data.
Outcome:	The proposed method outperforms zero-shot methods on five public survey datasets and reduces the required real data by 53.48-69.12%.

Copied to clipboard

Yuxuan Dong, Xinyu Zhang, Lingling Zhang, Han Lai, Pengyu Li, Bifan Wei, Yaqiang Wu, Jun Liu

The Mark Fades: Adaptive Evolutionary Paraphrase-based Attack against LLM Watermarks (2026.findings-acl)

Challenge:	Existing Large Language Models (LLMs) struggle with physics problem solving due to difficulties in decoding implicit constraints and maintaining physical consistency.
Approach:	They propose a Generative PRM that treats evaluation as a generative task . it produces fine-grained diagnoses comprising critiques, final judgments, and specific error types .
Outcome:	The proposed model improves performance across seven benchmarks in Best-of-N and critique refinement strategies.

Copied to clipboard

Yusheng Zhao, Jian Zhao, Tianle Zhang, Feng Wei, Xuelong Li

CoRE: A Fine-Grained Code Reasoning Benchmark Beyond Output Prediction (2026.findings-acl)

Challenge:	Existing paraphrase-based watermark removal methods struggle to balance efficacy with text quality.
Approach:	They propose a training-free evolutionary framework that models watermark removal as a constrained multi-objective optimization problem by using a Pseudo-Log-Likelihood-guided mutation to precisely target and modify watermark-carrying tokens.
Outcome:	The proposed method outperforms baseline methods on a Qwen3 series watermark scheme while maintaining high semantic fidelity.

Copied to clipboard

Jun Gao, Yun Peng, Qian Qiao, Changhai Zhou, Yuhua Zhou, Shiyang Zhang, Shichao Weng, Zhenchang Xing, Xiaoxue Ren

Empowering Reliable Visual-Centric Instruction Following in MLLMs (2026.findings-acl)

Challenge:	Existing code reasoning benchmarks evaluate final output correctness under a single implementation.
Approach:	They propose a Code Reasoning benchmark that evaluates code reasoning through implementation invariance and process transparency.
Outcome:	The proposed benchmarks lack implementation invariance and process transparency . they observe superficial execution where models arrive at correct outputs without reasoning .

Copied to clipboard

Weilei He, Feng Ju, Zhiyuan Fan, Rui Min, Minhao Cheng

TT-SI: Self-Improving LLM Agents with Test-Time Training (2026.findings-acl)

Challenge:	Existing benchmarks for evaluating instruction-following capabilities focus on verbal instructions in the textual modality.
Approach:	They propose to incorporate vision-dependent constraints into instruction design to enable a more rigorous assessment of how well MLLMs align their outputs with both visual input and textual instructions.
Outcome:	The proposed benchmark incorporates vision-dependent constraints into instruction design, enabling a more rigorous and fine-grained assessment of how well MLLMs align their outputs with both visual input and textual instructions.

Copied to clipboard

Emre Can Acikgoz, Cheng Qian, Heng Ji, Dilek Hakkani-Tür, Gokhan Tur

Conflicts Make Large Reasoning Models Vulnerable to Attacks (2026.findings-acl)

Challenge:	Existing methods for language model fine-tuning are expensive and inefficient . existing methods rarely assess whether a training sample provides novel information .
Approach:	They propose a test-time self-improvement algorithm that generates a sample that model struggles with . they also explore Test-Time Distillation, which leverages 'stronger supervisors'
Outcome:	The proposed algorithm improves performance with +5.48% absolute accuracy gain on average across benchmarks.

Copied to clipboard

Honghao Liu, Chengjin Xu, Xuhui Jiang, Cehao Yang, Shengming Yin, Zhengwu Ma, Lionel Ni, Jian Guo

Linear Script Representations in Speech Foundation Models Enable Zero-Shot Transliteration (2026.findings-acl)

Challenge:	Large Reasoning Models have demonstrated outstanding capabilities in solving complex reasoning tasks by incorporating step-by-step chain-of-thought (CoT) reasoning.
Approach:	They evaluate three large reasoning models that perform explicit and coherent reasoning under conflicting objectives and use them to evaluate their performance.
Outcome:	The proposed models perform explicit and coherent reasoning before producing their outputs, improving problem-solving and multi-step decision making.

Copied to clipboard

Ryan Soh-Eun Shim, Kwanghee Choi, Kalvin Chang, Ming-Hao Hsu, Florian Eichin, Zhizheng Wu, Alane Suhr, Michael A. Hedderich, David Harwath, David R. Mortensen, Barbara Plank

NSL-MT: Linguistically Informed Negative Samples for Efficient Machine Translation in African Low-Resource Languages (2026.findings-acl)

Challenge:	We show that script information is linearly encoded in the activation space of multilingual speech models . modifying activations at inference time induces script change even in unconventional pairings .
Approach:	They propose to add script vectors to activations at test time to induce script change . they also show that script information is linearly encoded in the activation space of multilingual speech models .
Outcome:	The proposed approach can induce script change even in unconventional language-script pairings.

Copied to clipboard

Mamadou K. Keita, Christopher M Homan, Huy Le

EduMARS: Can Vision-Language Models Grade Like Teachers? Benchmarking Multimodal, Rubric-Based Assessment on Chinese K-12 Answers (2026.findings-acl)

Challenge:	In low-resource settings, models encounter too few examples to reliably distinguish grammatical patterns from noise.
Approach:	They propose a negative space learning machine translation (NSL-MT) method that augments limited parallel data with synthetically generated violations of the target language’s grammar and explicitly penalizes the model when it assigns high probability to these violations.
Outcome:	The proposed method delivers 3-12% BLEU gains for well-performing models and 56-89% gains for models lacking decent initial support.

Copied to clipboard

Xuan Zhao, Jiashun Chen, Wanting xu, Huiyuan Yan, Chaowei Fang, Xing Wei

MMR-GRPO: Accelerating GRPO-Style Training through Diversity-Aware Reward Reweighting (2026.findings-acl)

Challenge:	Existing benchmarks for automated grading of student work fail to evaluate real student responses . existing models fail to assess real student work, especially on cognitively demanding tasks .
Approach:	They propose a multimodal benchmark for rubric-aligned evaluation of real Chinese K-12 student answers.
Outcome:	The proposed model improves performance and interpretability of existing models on EduMARS . existing models fail to perform on real-world, cognitively demanding tasks, authors say .

Copied to clipboard

Kangda Wei, Ruihong Huang

StoryMI: Steerable Multi-Agent Therapeutic Dialogue Generation (2026.findings-acl)

Challenge:	Recent advances in large language models (LLMs) have demonstrated remarkable capabilities in mathematical reasoning tasks.
Approach:	They propose to use Maximal Marginal Relevance to reweigh rewards of multiple rollouts by balancing rollout quality with diversity to reduce rollout redundancy.
Outcome:	The proposed approach reduces training time and costs by 47.9% . evaluations across three model sizes, three GRPO variants, and five mathematical reasoning benchmarks show that it achieves comparable peak performance while requiring on average 70.2% less wall-clock time.

Copied to clipboard

Qingyu Meng, Min Chen, Dingming Liu, Yifan Mo, Yue Su, Xin Sun, Koen Hindriks, Jiahuan Pei

Disambiguation-Centric Finetuning Makes Enterprise Tool-Calling LLMs More Realistic and Less Risky (2026.findings-acl)

Challenge:	Motivational interviewing (MI) is a directive, client-centered counseling approach for eliciting clients' motivation for behavioral change.
Approach:	They propose a multi-LLM agent framework for controllable MI dialogue generation . therapist and client agents generate MI-coded utterances guided by MI codes .
Outcome:	The proposed framework can generate fluent dialogues with minimal intervention time and a high level of evaluation.

Copied to clipboard

Ashutosh Hathidara, Julien Yu, Sebastian Schreiber

ToolGate: Contract-Grounded and Verified Tool Execution for LLMs (2026.findings-acl)

Challenge:	Large language models (LLMs) are increasingly tasked with invoking enterprise APIs . however, they falter when near-duplicate tools vie for the same user intent . cnn's john mccartney and johnny mccain present a disambiguation-centric pipeline .
Approach:	They propose a disambiguation-centric pipeline that synthesizes persona-driven dialogues . they use a corpus of API specifications and rigorously validated dialogues to build reliable tools .
Outcome:	The proposed pipeline raises tool-invocation success by 27 pp over GPT-4o and 49 pp above Claude-3.5-Sonnet on a dynamic benchmark.

Copied to clipboard

Yanming Liu, Xinyue Peng, Jiannan Cao, Xinyi Wang, Songhang Deng, Jintao Chen, Jianwei Yin, Xuhong Zhang

PersonaArena: Dynamic Simulation for Evaluating and Enhancing Persona-Level Role-Playing in Large Language Models (2026.findings-acl)

Challenge:	Existing frameworks for tool-augmented LLMs rely heavily on natural language reasoning to determine when tools can be invoked and whether their results should be trusted.
Approach:	They propose a forward execution framework that provides logical safety guarantees and verifiable state evolution for LLM tool calling.
Outcome:	The proposed framework improves the reliability and verifiability of tool-augmented LLM systems while maintaining competitive performance on multi-step reasoning tasks.

Copied to clipboard

Wenlong Shi, Jianxun Lian, Mingqi Wu, Haiming Qin, Mingyang Zhou, Xing Xie, Naipeng Chao, Hao Liao

Poller: Are LLMs Suitable for Evaluating Poetry Understanding Task? (2026.findings-acl)

Challenge:	Existing research focuses on character-level settings and static evaluation formats fail to capture the complexity of everyday social interactions.
Approach:	They propose a dynamic simulation framework for evaluating and improving persona-level role-playing in large language models (LLMs).
Outcome:	The proposed framework leverages user-generated social content to construct a nuanced persona bank and elicits multi-turn, context-rich interactions within simulated social environments.

Copied to clipboard

Shanshan Wang, Derek F. Wong, Jingming Yao, Lidia S. Chao

No-Worse Context-Aware Decoding: Preventing Neutral Regression in Context-Conditioned Generation (2026.findings-acl)

Challenge:	Traditional methods for poetry evaluation are expensive and unsuitable for large-scale data.
Approach:	They propose a method leveraging Large Language Models to evaluate poetry understanding tasks using Large Language models.
Outcome:	The proposed method reduces the evaluation error between LLMs and humans by adopting the poet's perspective.

Copied to clipboard

Yufei Tao, Ameeta Agrawal

Challenge:	Large language models can answer questions and generate summaries when given external contexts.
Approach:	They propose a decode-time adapter that backs off to no-context decoding when context is non-informative and uses contrastive fallback under uncertainty.
Outcome:	The proposed model prevents neutral regression on baseline-correct items while preserving strong context-driven accuracy on helpful contexts.

Prompting Test-Time Scaling Is A Strong LLM Reasoning Data Augmentation (2026.findings-acl)

Copied to clipboard

Sondos Mahmoud Bsharat, Zhiqiang Shen

GRASS: Gradient-based Adaptive Layer-wise Importance Sampling for Memory-efficient Large Language Model Fine-tuning (2026.findings-acl)

Challenge:	Large language models exhibit strong reasoning when guided by chain-of-thought exemplars . collecting large, high-quality reasoning datasets remains laborious and resource-intensive .
Approach:	They propose a prompt-space data augmentation framework for enhancing LLM reasoning . they use a pool of 90 randomly selected reasoning instances to elicit diverse reasoning trajectories .
Outcome:	The proposed framework improves accuracy over small-data benchmarks and generalization on out-of-domain reasoning evaluations.

Copied to clipboard

Kaiyuan Tian, Linbo Qiao, Yu Tang, Gongqingjian Jiang, Baihui Liu, Yifu Gao, Xialin Su, Dongsheng Li

ltzGLUE: Luxembourgish General Language Understanding Evaluation (2026.findings-acl)

Challenge:	Low-rank adaptation methods for large language models limit expressiveness and performance . layer-wise fine-tuning methods overlook variations in layer importance across tasks and training stages, resulting in suboptimal performance on downstream tasks.
Approach:	They propose a gradient-based adaptive layer-wise importance sampling framework that updates only a subset of parameters to reduce memory usage.
Outcome:	The proposed framework outperforms state-of-the-art methods in accuracy and memory usage.

Copied to clipboard

Alistair Plum, Felicia Körner, Anne-Marie Lutgen, Laura Bernardy, Fred Philippy, Emilia Milano, Nils Rehlinger, Cedric Lothritz, Tharindu Ranasinghe, Barbara Plank, Christoph Purschke

AdapThink: Adaptive Thinking Preferences for Reasoning Language Models (2026.findings-acl)

Challenge:	ltzGLUE is the first official NLU benchmark for Luxembourgish (LTZ) based on the popular GLUE benchmark for English.
Approach:	They propose a new natural language understanding (NLU) benchmark for Luxembourgish based on the popular GLUE benchmark for English.
Outcome:	The proposed model performs well across many languages and is based on the GLUE benchmark for English.

Copied to clipboard

Wenyue Xu, Xu Wan, Wei Wang, Wenqi Huang, Wotao Yin, Shengjie Zhao, Mingyang Sun

Improving the Throughput of Diffusion-based Large Language Models via a Training-Free Confidence-Aware Calibration (2026.findings-acl)

Challenge:	Recent research has highlighted a significant inefficiency associated with the slow thinking paradigm . models often overthink simple tasks while underthinking complex challenges .
Approach:	They propose a framework for adaptive reasoning preference control that dynamically adjusts reflection preferences based on group-level distributional statistics of reasoning length and reflection intensity.
Outcome:	The proposed framework reduces average response length by 17.1%-21.4% while improving performance by 6.12-6.59 points under 32K token budgets.

Copied to clipboard

Jucheng Shen, Gaurav Sarkar, Yeonju Ro, Sharath Nittur Sridhar, Zhangyang Wang, Aditya Akella, Souvik Kundu

Same Claim, Different Judgment: Benchmarking Scenario-Induced Bias in Multilingual Financial Misinformation Detection (2026.findings-acl)

Challenge:	CadLLM is a plug-and-play model-agnostic with KV caching based dLLMs.
Approach:	They propose a lightweight adaptive method that can control the generation block size, step size, and threshold based on the average confidence score of unmasked tokens.
Outcome:	The proposed method can increase throughput by up to 1.1-2.28x over the state-of-the-art model with competitive accuracy.

Copied to clipboard

Zhiwei Liu, Yupeng Cao, Yuechen Jiang, Mohsinul Kabir, Polydoros Giannouris, Chen Xu, Ziyang Xu, Tianlei Zhu, Md. Tariquzzaman, Triantafillos Papadopoulos, Yan Wang, Lingfei Qian, Xueqing Peng, Zhuohan Xie, Ye Yuan, Saeed Almheiri, Abdulrazzaq Alnajjar, Ming-Bin Chen, Harry Stuart, Paul Thompson, Prayag Tiwari, Alejandro Lopez-Lira, Xue Liu, Jimin Huang, Sophia Ananiadou

MASS-RAG: Multi-Agent Synthesis Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	Existing research on LLM biases has focused on direct questioning or general-purpose settings . pronounced behavioral biase despite their growing deployment in financial analysis, forecasting, and decision support.
Approach:	They propose a benchmark to evaluate behavioral biases of large language models in MFMD . they use a multilingual financial misinformation dataset to integrate these with misinformation claims .
Outcome:	The proposed benchmark evaluates behavioral biases of large language models across economic scenarios.

Copied to clipboard

Xingchen Xiao, Heyan Huang, Runheng Liu, Jincheng Xie

Bayesian Active Learning with Gaussian Processes Guided by LLM Relevance Scoring for Dense Passage Retrieval (2026.findings-acl)

Challenge:	Large language models (LLMs) are widely used in retrieval-augmented generation (RAG) when retrieved contexts are noisy, incomplete, or heterogeneous, a single generation process often struggles to reconcile evidence effectively.
Approach:	They propose a multi-agent synthesis approach to retrieval-augmented generation that structures evidence processing into multiple role-specialized agents.
Outcome:	Experiments on four benchmarks show that MASS-RAG consistently improves performance over strong RAG baselines.

Copied to clipboard

Junyoung Kim, Anton Korikov, Jiazhou Liang, Justin Cui, Yifan Simon Liu, Qianfeng Wen, Mark Zhao, Scott Sanner

Reasoning-Based Refinement of Unsupervised Text Clusters with LLMs (2026.findings-acl)

Challenge:	Existing approaches passively rely on first-stage dense retrievers, which leads to two limitations: failing to retrieve relevant passages in semantically distinct clusters and failing to propagate relevance signals to the broader corpus.
Approach:	They propose a framework that propagates sparse LLM relevance signals across the embedding space to guide global exploration.
Outcome:	Experiments show that the proposed framework outperforms existing approaches under the same budget on all four datasets.

Copied to clipboard

Tunazzina Islam

Self-Evolving Multi-Agent Systems via Textual Backpropagation (2026.findings-acl)

Challenge:	Unsupervised methods are used to induce latent semantic structure from large text collections, yet their outputs often contain incoherent, redundant, or poorly grounded clusters.
Approach:	They propose a framework that leverages large language models as semantic judges to validate and restructure unsupervised clustering algorithms.
Outcome:	The proposed framework improves cluster coherence and human-aligned labeling quality over traditional models and representation-based baselines.

Copied to clipboard

Xiaowen Ma, Yunpu Ma, Chenyang Lin, Sikuan Yan, Jinhe Bi, Zixuan Cao, Yijun Tian, Volker Tresp, Hinrich Schuetze

Confidence over Time: Confidence Calibration with Temporal Logic for Large Language Model Reasoning (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have proven effective for addressing complex, high-dimensional tasks, but current approaches rely on static, manually engineered multi-agent configurations.
Approach:	They propose a framework that conceptualizes multi-agent collaboration as a layered neural network architecture.
Outcome:	The proposed framework surpasses leading multi-agent baselines under the same configurations, showing consistent performance improvements.

Copied to clipboard

Zhenjiang Mao, Anirudhh Venkat, Artem Bisliouk, Sindhura Kumbakonam Subramanian, Akshat Kothiyal, Saithej Singhu, Ivan Ruchkin

Automated Profile Inference with Language Model Agents (2026.findings-acl)

Challenge:	Existing confidence estimation methods reduce reasoning process to a single scalar score, ignoring how confidence evolves throughout generation.
Approach:	They propose to characterize the stepwise confidence signal using Signal Temporal Logic (STL) based on a discriminative STL mining procedure, they find temporal formulas that distinguish correct and incorrect responses.
Outcome:	The proposed method can distinguish between correct and incorrect reasoning signals.

Copied to clipboard

Yuntao Du, Zitao Li, Bolin Ding, Yaliang Li, Hanshen Xiao, Jingren Zhou, Ninghui Li

When Debiasing Backfires: Counterintuitive Side Effects of Preprocessing-Based Stereotype Mitigation (2026.findings-acl)

Challenge:	Existing privacy protections for large language models (LLMs) are limited due to the potential for malicious applications.
Approach:	They propose an automated profile inference framework that can extract personal information from public online activities by an adversary with the help of large language model (LLM) based agents.
Outcome:	The proposed framework is highly effective and efficient and the inferred attributes are both identifiable and sensitive, posing significant privacy risks.

Copied to clipboard

Yahan Zheng, John J. Guerrerio, Soroush Vosoughi, Weicheng Ma

Scaling Laws for Code: Every Programming Language Matters (2026.findings-acl)

Challenge:	Preprocessing-based methods for stereotype mitigation are widely used in NLP . preprocessing methods cause unintended shifts in attention flow, authors say .
Approach:	They propose to use preprocessing-based methods to reduce stereotypes for targeted groups . they find that stereotyping or counter-stereotyping can increase for other demographics .
Outcome:	The proposed methods often induce unintended shifts across demographics, the authors show . they show that such side effects are not accompanied by large changes in attention flow .

Copied to clipboard

Jian Yang, Shuyue Guo, Linzheng Chai, Wei Zhang, Aishan Liu, Chuan Hao, Zhoujun Li, Xin Zhao, Xianglong Liu, Weifeng Lv, Bryan Dai

Beyond the Singular: Revealing the Value of Multiple Generations in Benchmark Evaluation (2026.findings-acl)

Challenge:	Existing studies focus on language-agnostic settings, neglecting the inherently multilingual nature of modern software development.
Approach:	They propose a proportion-dependent scaling law that prioritizes high-utility languages . they propose PLs to have varying effects during pre-training that affect model performance .
Outcome:	The proposed scaling law is based on 1000+ experiments across multiple languages and models.

Copied to clipboard

Wenbo Zhang, Hengrui Cai, Wenyu Chen

Safety Is Not Universal: The Selective Safety Trap in LLM Alignment (2026.findings-acl)

Challenge:	Existing evaluation methods for large language models overlook the inherent randomness of LLMs.
Approach:	They propose a hierarchical statistical model that incorporates both benchmark characteristics and LLM randomness to provide a more comprehensive representation of benchmarking process.
Outcome:	The proposed model improves the accuracy of estimating the benchmark score and reduces variance.

Copied to clipboard

Iago Alves Brito, Walcy Rios, Julia Soares Dollis, Diogo Fernandes Costa Silva, Arlindo Rodrigues Galvão Filho

LLM Prompt Duel Optimizer: Efficient Label-Free Prompt Optimization (2026.findings-acl)

Challenge:	Existing safety evaluations of large language models aggregate harms under generic categories such as "Identity Hate" a bilingual benchmark identifies a selective safety trap, where defense rates vary by up to 42% within the same model solely based on the target group.
Approach:	They propose a bilingual adversarial benchmark to audit selective safety in large language models . defense rates vary by up to 42% within the same model solely based on target group .
Outcome:	The proposed benchmark identifies a selective safety trap in large language models . defense rates vary by up to 42% within the same model solely based on the target group.

Copied to clipboard

Yuanchen Wu, Saurabh Verma, Justin Lee, Fangzhou Xiong, Poppy Zhang, Amel Awadelkarim, Xu Chen, Yubai Yuan, Shawndra Hill

FrameNet-Cultures: A Benchmark for Evaluating LLMs via Cross-Cultural Frame Semantics (2026.findings-acl)

Challenge:	Large language models (LLMs) are highly sensitive to prompts, but most automatic prompt optimization methods assume access to ground-truth references that are costly to obtain.
Approach:	They propose a sample-efficient framework for label-free prompt optimization based on pairwise preference feedback from an LLM judge.
Outcome:	Experiments on BIG-bench Hard and MS MARCO show that the proposed framework identifies stronger prompts than label-free baselines while offering favorable quality–cost trade-offs.

Copied to clipboard

Neda Jamshidi, Anders Søgaard, Monica Bianchini

Meta-Tool: Efficient Few-Shot Tool Adaptation for Small Language Models (2026.findings-acl)

Challenge:	Existing evaluation paradigms for large language models lack rigorous methods to evaluate cultural alignment . FRAMENET-CULTURES is an open-ended benchmark for evaluating cultural alignment in LLMs .
Approach:	They propose a benchmark for evaluating cultural alignment in large language models based on Fillmore-style frame semantics.
Outcome:	The proposed benchmark is based on Fillmore-style frame semantics.

Copied to clipboard

Sachin Kumar

Toward A Digital Twin of U.S. Congress (2026.findings-acl)

Challenge:	Using a Llama-3.2-3B-Instruct backbone, we evaluate four adaptation mechanisms across four benchmarks: Gorilla APIBench, Spider 2.0, WebArena, and InterCode.
Approach:	They compare hypernetwork-based LoRA adaptation against carefully designed few-shot prompting in a controlled experiment . they find that few- shot prompting contributes +21.5% to performance and documentation contributes 0% .
Outcome:	The hypernetwork-based LoRA adaptation provides no measurable improvement over few-shot prompting alone.

Copied to clipboard

Hayden Helm, Tianyi Chen, Harvey McGuinness, Paige Lee, Brandon Duderstadt, Carey Priebe

OjaKV: Context-Aware Online Low-Rank KV Cache Compression (2026.findings-acl)

Challenge:	a virtual model of congresspersons based on a collection of language models meets the definition of a digital twin.
Approach:	They propose to use a daily-updated dataset to generate tweets from congresspersons . they show that a modern language model equipped with subsets of this dataset produces Tweets that are indistinguishable from actual Tweets posted by their physical counterparts.
Outcome:	The proposed model produces Tweets that are indistinguishable from actual tweets posted by congresspersons.

Copied to clipboard

Yuxuan Zhu, David H. Yang, Mohammad Mohammadi Amiri, Keerthiram Murugesan, Tejaswini Pedapati, Pin-Yu Chen

Challenge:	Existing methods for inference use static, offline-learned subspaces that perform poorly under distribution shifts.
Approach:	They propose a framework that integrates a storage policy with an online subspace adaptation to preserve key-value tokens in full rank as high-fidelity anchors.
Outcome:	Experiments show that OjaKV maintains or improves zero-shot accuracy at high compression ratios, achieving the strongest gains on long-context benchmarks requiring complex reasoning.

AURORA: Neuro-Symbolic Continual Indexing for Evolving RAG Systems (2026.findings-acl)

Copied to clipboard

Manoj Saravanan, Rohit Kumar Salla, Ramya Manasa Amancherla

Decide less, communicate more: On the construct validity of end-to-end fact-checking in medicine (2026.findings-acl)

Challenge:	Retrieval-Augmented Generation (RAG) systems depend on non-parametric indices to access external knowledge.
Approach:	They propose a framework for adapting retrieval indices under distribution shift . AURORA decouples discrete index structure from continuous metric representations . it recovers +26.9% Recall@10 on novel topics compared to static baselines compared with static baseline .
Outcome:	AURORA decouples discrete index structure from continuous metric representations . it recovers +26.9% Recall@10 on novel topics while adapting significantly faster than full retraining.

Copied to clipboard

Sebastian Antony Joseph, Lily Chen, Barry Wei, Michael Mackert, Iain James Marshall, Paul Pu Liang, Ramez Kouzy, Byron C Wallace, Junyi Jessy Li

K-GIP: Diagnosing Logical Fractures in Large Vision-Language Models via Verification Scene Graphs and Sequential Pruning (2026.findings-acl)

Challenge:	Evidence-based medicine connects to every individual, yet the nature of it is highly technical . e-fact-checking systems that connect to medical decisions are largely unused . we examine how clinical experts verify real claims from social media .
Approach:	They propose that fact-checking should be approached as an interactive communication problem . they argue that social media and AI have made medical knowledge accessible .
Outcome:	The proposed method is based on the work of a clinical expert on social media . it reveals that the method is difficult to connect claims to clinical trials .

Copied to clipboard

Yujun Hu, Xiaoyu Zhou, Changbo Wang, Gaoqi He

From Static Inference to Dynamic Interaction: A Survey of Streaming Large Language Models (2026.findings-acl)

Challenge:	Existing benchmarks that treat hallucinations as isolated errors neglect causal dependencies between visual perception and textual reasoning.
Approach:	They propose a Knowledge-Guided In-Context Probing framework that constructs a dual-perception ground truth to transform abstract priors into multi-granularity queries.
Outcome:	The proposed framework isolates deep reasoning failures from simple perceptual misses.

Copied to clipboard

Junlong Tong, Zilong Wang, YuJie Ren, Peiran Yin, Hao Wu, Wei Zhang, Xiaoyu Shen

How Adversarial Environments Mislead Agentic AI? (2026.findings-acl)

Challenge:	Existing definitions of streaming LLMs are fragmented and lack a systematic taxonomy . large language models are pre-trained on static and full-context corpora .
Approach:	They propose a systematic taxonomy of current streaming Large Language Models and propose underlying methodologies for streaming LLMs.
Outcome:	The proposed model is based on data flow and dynamic interaction to clarify existing ambiguities.

Copied to clipboard

Zhonghao Zhan, Huichi Zhou, Zhenhao Li, Peiyuan Jing, Krinos Li, Hamed Haddadi

GALA: Geometric Data Selection with Strategic Prospecting for Large Language Model Self-training (2026.findings-acl)

Challenge:	Current evaluations benchmark capability in benign settings, but never "what if the tools lie" we formalize this vulnerability as Adversarial Environmental Injection (AEI) AEI constitutes environmental deception by constructing a "fake world" of poisoned search results .
Approach:	They propose an attack model where adversaries compromise tool outputs to deceive agents.
Outcome:	The proposed model exploits a trust gap between tool outputs and actual exposure to adversaries.

Copied to clipboard

Zhongwei Xie, Ruihao Liao, Zimo Wang, Chong Chen, Xian-Sheng Hua, Xiao Luo

PII-VisBench: Evaluating Personally Identifiable Information Safety in Vision Language Models Along a Continuum of Visibility (2026.findings-acl)

Challenge:	Existing approaches to self-training are based on reject sampling and lack quality reasoning paths.
Approach:	They propose a framework for self-training using a generate-and-filter paradigm . they propose to identify diverse and informative samples from redundant data and exploit them more strategically.
Outcome:	The proposed framework exploits informative samples from redundant data and improves reasoning trajectory prospecting.

Copied to clipboard

G M Shahariar, Zabir Al Nazi, Md Olid Hasan Bhuiyan, Zhouxing Shi

Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models (2026.findings-acl)

Challenge:	Existing evaluations of PII leakage ignore how a subject’s online presence affects privacy alignment.
Approach:	They propose a benchmark that evaluates safety through the continuum of online presence by stratifying 200 subjects into four visibility categories: high, medium, low, and zero.
Outcome:	The proposed model stratifies 200 subjects into four visibility categories based on the extent and nature of their information available online.

Copied to clipboard

Hengyuan Zhang, Zhihao Zhang, Ercong Nie, Mingyang Wang, Zunhai Su, Yiwei Wang, Qianli Wang, Shuzhou Yuan, Xufeng Duan, Qibo Xue, Zeping Yu, Chenming Shang, Xiao Liang, Jing Xiong, Hui Shen, Chaofan Tao, Zhengwu Liu, Senjie Jin, Zhiheng Xi, Dongdong Zhang, Sophia Ananiadou, Tao Gui, Ruobing Xie, Hayden Kwok-Hay So, Hinrich Schuetze, Xuanjing Huang, Qi Zhang, Ngai Wong

MTRAG-UN: A Benchmark for Open Challenges in Multi-Turn RAG Conversations (2026.findings-acl)

Challenge:	Existing literature on mechanistic interpretation (MI) treats it as an observational science, leaving practical applications underexplored.
Approach:	They propose a survey structured around the pipeline to identify and improve MI models.
Outcome:	The proposed framework enables tangible improvements in Alignment, Capability, and Efficiency.

Copied to clipboard

Sara Rosenthal, Yannis Katsis, Vraj Shah, Lihong He, Lucian Popa, Marina Danilevsky

Vision-Language Models Mistake Head Orientation for Gaze Direction: Nonverbal Conversation Cues (2026.findings-acl)

Challenge:	Several benchmarks have been released to evaluate model performance on multi-turn retrieval augment generation tasks.
Approach:	They propose to benchmark 666 conversations with over 2,800 conversation turns across 6 domains and a corpora that focuses on unanswerable questions and later conversation turns.
Outcome:	The proposed benchmarks show that retrieval and generation models struggle on conversations with UNanswerable, UNderspecified, and NONstandalone questions and UNclear responses.

Copied to clipboard

Zory Zhang, Pinyuan Feng, Bingyang Wang, Tianwei Zhao, Suyang Yu, Qingying Gao, Hokin Deng, Ziqiao Ma, Yijiang Li, Dezhi Luo

RepoShapley: Shapley-Enhanced Context Filtering for Repository-Level Code Completion (2026.findings-acl)

Challenge:	Where someone looks is a nonverbal communication cue that children and adults readily use.
Approach:	They used 1,360 real-world photos to construct evaluation stimuli for Vision-Language Models (VLMs) they found a substantial performance gap between VLMs and humans .
Outcome:	The proposed model outperforms existing models in predicting gaze direction using head orientation rather than eye appearance.

Copied to clipboard

Yu Huo, Kun Zeng, Siyu Zhang, Yuquan LU, Cheng Yang, Yifu Guo, Xiaoying Tang

Red-Teaming NSFW Image Classifiers as Text-to-Image Safeguards (2026.findings-acl)

Challenge:	Large language models have strong reasoning, coding, and generation capabilities, but retrieval-augmented generation remains difficult under fixed context budgets.
Approach:	They propose a coalition-aware context filtering framework supervised by Shapley-style marginal contributions that captures sign effects via teacher-forced probing and computes exact Shaply values for small retrieval sets.
Outcome:	Experiments show that RepoShapley improves completion quality while reducing harmful context and unnecessary retrieval.

Copied to clipboard

Tinghao Xie, Yueqi Xie, Alireza Zareian, Shuming Hu, Felix Juefei-Xu, Xiaowen Lin, Ankit Jain, Prateek Mittal, Li Chen

Model-Agnostic Meta Learning for Class Imbalance Adaptation (2026.findings-acl)

Challenge:	Not Safe for Work (NSFW) image classifiers play a critical role in safeguarding text-to-image systems.
Approach:	They propose an automated red-teaming framework that leverages a set of generative AI tools to uncover NSFW image failures.
Outcome:	The proposed framework uncovers and interprets failure modes and enables it to be applied to real-world T2I and T2V systems.

Copied to clipboard

Hanshu Rao, Guangzeng Han, Xiaolei Huang

PAPERMIND: Benchmarking Agentic Reasoning and Critique over Scientific Papers in Multimodal LLMs (2026.findings-acl)

Challenge:	Existing approaches to address class imbalance and data difficulty have been used to train models.
Approach:	They propose a framework that prioritizes challenging samples and minority classes over hard examples and their semantically similar neighbors to address class imbalance.
Outcome:	The proposed framework outperforms baselines on six imbalanced datasets and achieves substantial improvements for minority classes.

Copied to clipboard

Yanjun Zhao, Tianxin Wei, Jiaru Zou, Xuying Ning, Yuanchen Bei, Lingjie Chen, Simmi Rana, Wendy H. Yang, Hanghang Tong, Jingrui He

From Scenes to Elements: Multi-Granularity Evidence Retrieval for Verifiable Multimodal RAG (2026.findings-acl)

Challenge:	Existing benchmarks assess integrated and agent-oriented scientific reasoning in isolation . Existing systems assess integrated reasoning in isolated tasks .
Approach:	They propose a benchmark to evaluate integrated and agent-oriented scientific reasoning over research papers.
Outcome:	The proposed benchmark evaluates integrated and agent-oriented scientific reasoning over scientific papers.

Copied to clipboard

Guanhua Chen, Chuyue Huang, Yutong Yao, Shudong Liu, Xueqing Song, Lidia S. Chao, Derek F. Wong

Region-R1: Reinforcing Query-Side Region Cropping for Multi-Modal Re-Ranking (2026.findings-acl)

Challenge:	Existing multimodal Retrieval-Augmented Generation (RAG) systems retrieve evidence at coarse granularities, making failures unverifiable.
Approach:	They propose a multimodal benchmark that features real-world landmarks with annotations across multiple viewpoints and a framework that treats visual elements as first-class retrieval units through three stages: element-level detection and classification, multi-granularity cross-modal alignment for evidence retrieval, and attribution-constrained generation.
Outcome:	The proposed framework achieves up to 29.2% improvement over six strong baselines for this task.

Copied to clipboard

Chan-Wei Hu, Zhengzhong Tu

A Reward-Guided Dual-Phase Framework for Adaptive Inference-Time Reasoning (2026.findings-acl)

Challenge:	Multi-modal retrieval-augmented generation relies heavily on re-rankers to surface the most relevant evidence for image-question queries.
Approach:	They propose a query-side region cropping framework that makes region selection a decision-making problem during re-ranking.
Outcome:	The proposed framework learns to retain the full image or focus only on a question-relevant region before scoring the retrieved candidates.

Copied to clipboard

Yingqian Cui, Zhenwei Dai, Pengfei He, Bing He, Hui Liu, Zhan Shi, Xianfeng Tang, Jingying Zeng, Suhang Wang, Yue Xing, Jiliang Tang, Benoit Dumoulin

LVLM-Aware Multimodal Retrieval for RAG-Based Medical Diagnosis with General-Purpose Models (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have made strong progress in reasoning.
Approach:	They propose a dual-phase test-time scaling framework that separates planning and execution and performs search over each phase independently.
Outcome:	Experiments on math reasoning and code generation benchmarks show that the proposed approach improves accuracy while reducing redundant computation.

Copied to clipboard

Nir Mazor, Tom Hope

Detecting Proxy Gaming in RL and LLM Alignment via Evaluator Stress Tests (2026.findings-acl)

Challenge:	Using retrieval augmentation, large vision language models can be used for diagnostic accuracy, but multimodal retrieval-augmented diagnosis is challenging.
Approach:	They propose a lightweight mechanism for enhancing diagnostic performance of retrieval-augmented LVLMs by fine-tuning a multimodal retriever and general-purpose backbone models.
Outcome:	The proposed mechanism achieves competitive results without medical training compared to pre-trained models with extensive training.

Copied to clipboard

Sanjeda Akter, Ibne Farabi Shihab, Anuj Sharma

Challenge:	Proxy optimization is a challenge spanning reinforcement learning and LLM alignment.
Approach:	They propose an invariance-based framework that detects proxy gaming by separating exploitable sensitivity from content-driven improvements using semantic validity audits.
Outcome:	The proposed framework achieves 78.4% precision and 81.7% recall across 15 environments and 5 algorithms.

Do Language Models Think Consistently? A Study of Value Preferences Across Varying Response Lengths (2026.findings-acl)

Copied to clipboard

Inderjeet Jayakumar Nair, Lu Wang

LEDGER: Scaling Agentic Document Editing with Dependency-aware Graph Retrieval (2026.findings-acl)

Challenge:	In short-form surveys and psychometric tests, value-related risks and preferences are often underexplored in practical settings.
Approach:	They compare short-form responses to long-form outputs to determine whether value preferences align with those expressed in long-term outputs.
Outcome:	The proposed method yields only modest gains in the consistency of value expression.

Copied to clipboard

Hang Wang, Utkarsh Garg, Reza Davari, Huitian Jiao, Hao Cheng, Baolin Peng, Si-Qing Chen, Tao Ge

NSF-CoT: Neuro-Symbolic Formal Verification of Chain-of-Thought Faithfulness in Contextual Question Answering (2026.findings-acl)

Challenge:	Document editing requires full-context awareness of dependencies, but processing entire documents for each edit incurs prohibitive token costs and latency.
Approach:	a framework that constructs lightweight dependency graphs captures semantic relationships and structural hierarchies across document elements is proposed for agentic document editing . a scaLing agentic agentic framework is based on a dependency graph framework that captures dependencies and refactors function dependencies.
Outcome:	a new framework achieves 76 consistency versus 56 baseline while reducing token usage by 85 . the framework is based on a framework that captures semantic relationships and structural hierarchies across document elements . it can be used to improve document consistency, but it also reduces token costs and latency .

Copied to clipboard

Vishal Pramanik, Maisha Maliha, Nathaniel D. Bastian, Alvaro Velasquez, Susmit Jha, Sumit Kumar Jha

Learning Flexible Large Multimodal Models with Arbitrary Modality Combinations (2026.findings-acl)

Challenge:	Existing faithfulness checks only test whether changing the written chain changes the answer, without verifying whether the steps are truly supported by the given evidence.
Approach:	They propose a method that checks CoT faithfulness step by step for contextual question answering.
Outcome:	The proposed method outperforms causal mediation, perturbation probes, and behavioral monitoring in OpenBookQA, QASC, and HotpotQA.

Copied to clipboard

Xinyu Zhao, Kangqi Ni, Jie Peng, Ang Li, Tianlong Chen

1,729 vs. 1729: The Effect of Scripts and Formats on LLM Numeracy (2026.findings-acl)

Challenge:	Multimodal Large Language Models (MLLMs) have potential for cross-modal understanding . but extending MLLM to handle diverse modalities introduces two challenges .
Approach:	They propose a dual-stage compression mechanism to reduce the number of modality tokens per modality and condense it into a single, compact token sequence.
Outcome:	Experiments show that Flex-M3 outperforms its counterpart trained on only full-modality data.

Copied to clipboard

Varshini Reddy, Craig W Schmidt, Seth Ebner, Adam Wiemerslage, Yuval Pinter, Chris Tanner

SimpleOCR: Rendering Visual Questions to Teach MLLMs to Read (2026.findings-acl)

Challenge:	Large language models (LLMs) have impressive proficiency in basic arithmetic, but little attention has been given to how they perform when numerical expressions deviate from the prevailing conventions present in their training corpora.
Approach:	They investigate numerical reasoning across a wide range of numeral scripts and formats . they show that LLM accuracy drops substantially when numerical inputs are rendered in underrepresented scripts or formats despite the underlying mathematical reasoning being identical .
Outcome:	The proposed methods can narrow the gap between LLMs and human models when they deviate from prevailing numerical conventions.

Copied to clipboard

Yibo Peng, Peng Xia, Ding Zhong, Kaide Zeng, Siwei Han, Yiyang Zhou, Jiaqi Liu, Ruiyi Zhang, Huaxiu Yao

Beyond Facts- Benchmarking Distributional Reading Comprehension in Large Language Models (2026.findings-acl)

Challenge:	MLLMs lack visual grounding mechanism to read text embedded in images, or rely on parametric shortcuts . despite strong OCR capabilities, models suffer performance degradation of 12.7% in the VQ setting .
Approach:	They propose a plug-and-play training strategy that invalidates shortcuts in text prompts . they propose 'vq' setting where text queries are rendered directly onto images .
Outcome:	The proposed training strategy surpasses the base model by 5.4% and GRPO based on original images by 2.7% on four representative OOD benchmarks.

Copied to clipboard

Pei-Fu Guo, Ya An Tsai, Chun-Chia Hsu, Kai-Xin Chen, Yun-Da Tsai, Kai-Wei Chang, Nanyun Peng, Mi-Yen Yeh, Shou-De Lin

Preference Learning Unlocks LLMs’ Psycho-Counseling Skills (2026.findings-acl)

Challenge:	Existing reading comprehension benchmarks focus on factual information, but many real-world tasks require distributional knowledge expressed across text.
Approach:	They propose a reading comprehension benchmark for LLMs to evaluate their ability to infer distributional knowledge from natural language.
Outcome:	Experiments with multiple LLMs show that the model outperforms baselines, but performance varies widely across distribution types and characteristics.

Copied to clipboard

Mian Zhang, Shaun M. Eack, Zhiyu Chen

Large Language Models Can Help Mitigate Barren Plateaus in Quantum Neural Networks (2026.findings-acl)

Challenge:	Current LLMs struggle to consistently provide effective responses to client speeches due to the lack of supervision from high-quality real psycho-counseling data.
Approach:	They propose to use a dataset to evaluate therapists' responses to client speeches using a set of professional and comprehensive principles to evaluate their responses.
Outcome:	The proposed model achieves an impressive win rate of 87% against GPT-4o.

Copied to clipboard

Jun Zhuang, Chaowen Guan

Finch: Benchmarking Finance & Accounting across Spreadsheet-Centric Enterprise Workflows (2026.findings-acl)

Challenge:	Quantum Neural Networks (QNNs) are often hindered by barren plateaus (BPs) barren peaks are where gradient variance vanishes exponentially as qubit size increases .
Approach:	They propose a framework that leverages large language models with the submartingale property to iteratively synthesize initial parameters for QNNs that yield non-negligible gradient variance.
Outcome:	The proposed framework outperforms existing initialization methods in maintaining higher gradient variance across various QNN scales.

Copied to clipboard

Haoyu Dong, Pengkun Zhang, Yan Gao, Xuanyu Dong, Yilin Cheng, Mingzhe Lu, Adina Yakefu, Shuxin Zheng

Exploiting Tree Structure for Credit Assignment in Reinforcement Learning with Large Language Models (2026.findings-acl)

Challenge:	FinWorkBench evaluates real-world enterprise-grade finance and accounting workflows . a human evaluation of GPT 5.1 Pro passes only 38.4% of workflows, a study finds .
Approach:	They propose a workflow construction process that combines LLM-assisted mining and expert annotation to build 172 composite workflows.
Outcome:	The proposed process combines expert annotation with LLM-assisted mining of workflows from authentic enterprise environments.

Copied to clipboard

Hieu Tran, Zonghai Yao, Hong yu

Demystify the Role of Memory in Machine Learning Engineering Agents (2026.findings-acl)

Challenge:	Reinforcement learning has shown strong promise for strengthening reasoning ability of large language models, but sparse, delayed rewards make token-level credit assignment a central challenge.
Approach:	They propose a critic-free algorithm that rewards tokens that change the solution.
Outcome:	The proposed algorithm improves on in-distribution benchmarks and out-of-disttribution settings.

Copied to clipboard

Xinyu Zhao, Junpeng Wang, Yuzhong Chen, Menghai Pan, Chin-Chia Michael Yeh, Jiarui Sun, Yan Zheng, Mahashweta Das, Tianlong Chen

Challenge:	Unlike short, reactive exchanges, MLE agents solve tasks through cycles of experimentation and improvement where past errors can inform future success.
Approach:	They propose a dynamic coding memory that captures and reuses debugging experiences and integrates it into two representative agent paradigms.
Outcome:	The proposed agent model captures and reuses debugging experiences and integrates it into two agent paradigms.

Characterizing Web Search in The Age of Generative AI (2026.findings-acl)

Copied to clipboard

Elisabeth Kirsten, Jost Große Perdekamp, Qinyuan Wu, Mihir Upadhyay, Krishna P. Gummadi, Muhammad Bilal Zafar

HypEHR: Hyperbolic Modeling of Electronic Health Records for Efficient Question Answering (2026.findings-acl)

Challenge:	generative search is a new search paradigm that uses LLMs to retrieve information from the web . traditional web search returns a ranked list of independent web pages .
Approach:	They compare generative search with traditional web search, which returns ranked results as a list of independent web pages.
Outcome:	The results show that generative search systems achieve topical coverage comparable to traditional search, but differ in retrieval footprints and synthesis strategies.

Copied to clipboard

Yuyu Liu, Sarang Rajendra Patil, Mengjia Xu, Tengfei Ma

What Users Leave Unsaid: Under-Specified Queries Limit Vision-Language Models (2026.findings-acl)

Challenge:	Recent studies have shown that LLM-based EHR question answering is costly to deploy and does not leverage hierarchical structure of clinical data.
Approach:	They propose a Lorentzian model that embeds codes, visits, and questions in hyperbolic space and answers queries via geometry-consistent cross-attention with type-specific pointer heads.
Outcome:	The proposed model embeds codes, visits, and questions in hyperbolic space and answers queries via geometry-consistent cross-attention with type-specific pointer heads.

Copied to clipboard

Dasol Choi, Guijin Son, Hanwool Lee, Minhyuk Kim, Hyunwoo Ko, Teabin Lim, Eungyeol Ahn, Jungwhan Kim, Seunghyeok Hong, Youngsook Song

Selective Steering: Norm-Preserving Control Through Discriminative Layer Selection (2026.findings-acl)

Challenge:	HAERAE-Vision benchmarks feature clear, explicit prompts but are often informal and underspecified . state-of-the-art models achieve under 50% on original queries, compared to GPT-5 and Gemini 2.5 Pro .
Approach:	They propose a benchmark of 653 real-world visual questions from Korean online communities . they find that even state-of-the-art models achieve under 50% on original queries .
Outcome:	HAERAE-Vision benchmarks from Korean online communities yield 1,306 query variants . state-of-the-art models achieve under 50% on original queries, compared with smaller models . authors show that query explicitation alone yields 8 to 22 point improvements .

Copied to clipboard

Quy-Anh Dang, Chris Ngo

From Bytes to Subwords: Challenges of Input Representations in NLP (2026.findings-acl)

Challenge:	Existing methods for inference-time steering are limited by their limitations . Angular Steering violates norm preservation, causing distribution shift and generation collapse .
Approach:	They propose a method that uses a norm-preserving rotation formulation to maintain activation distribution integrity and discriminative layer selection to apply steering only where features exhibit opposite-signed class alignment.
Outcome:	Experiments show that Selective Steering achieves higher attack success rates than prior methods while maintaining zero perplexity violations and approximately 100% capability retention on standard benchmarks.

Copied to clipboard

Rob Van Der Goot

The Learnability of Model-Theoretic Interpretation Functions in Artificial Neural Networks (2026.findings-acl)

Challenge:	Traditionally, characters or words have been used, but recently, subwords have become the standard.
Approach:	They examine the current use of tokenizers and examine the weaknesses of character normalization . they propose proof of concept alternatives focused on fairness and efficiency .
Outcome:	The proposed model is based on a systematic review of current tokenizers and character encodings.

Copied to clipboard

Adrian Brasoveanu, Jakub Dotlacil

A Mechanistic Perspective and Difficulty Metric for Unlearning (2026.findings-acl)

Challenge:	Entity vectors improve scores on basic event, while gated architectures benefit most.
Approach:	They extend entity-level semantic representations, modern architectures, principled competing event generation, extended systematicity tests and a two-dimensional difficulty analysis disaggregating results by modifier complexity.
Outcome:	The proposed model-theoretic interpretation functions generalize systematically to out-of-training-sample sentences.

Copied to clipboard

Jiali Cheng, Ziheng Chen, Chirag Agarwal, Hadi Amiri

UCS: Estimating Unseen Coverage for Improved In-Context Learning (2026.findings-acl)

Challenge:	Existing studies show that machine unlearning success varies across samples . easy-to-unlearn samples are associated with shorter, shallower interactions . hard-to unlear rely on longer and deeper pathways closer to late-stage computation.
Approach:	They propose a pre-unlearning metric that assigns each sample a continuous difficulty score . they show that CUD reliably separates intrinsically easy and hard samples .
Outcome:	The proposed method reliably separates intrinsically easy and hard samples and remains stable across unlearning methods.

Copied to clipboard

Jiayi Xin, Xiang Li, Evan Qiang, Weiqing He, Tianqi Shang, Weijie J Su, Qi Long

Talent or Luck? Evaluating Attribution Bias in Large Language Models (2026.findings-acl)

Challenge:	Existing selection methods prioritize heuristic notions of relevance or diversity and provide limited insight into the coverage of a demonstration set.
Approach:	They propose a training-free, subset-level coverage prior that is unrevealed by a model-consistent embedding and a Smoothed Good-Turing estimator to estimate the number of unrevelled clusters within a candidate subset.
Outcome:	Experiments on multiple intent-classification and reasoning benchmarks show that augmenting strong baselines with UCS improves ICL accuracy by 2-6% under the same selection budget.

Copied to clipboard

Chahat Raj, Mahika Banerjee, Jinhao Pan, Aylin Caliskan, Antonios Anastasopoulos, Ziwei Zhu

Uncertainty Quantification of Large Language Models through Multiple Uncertainty Sources (2026.findings-acl)

Challenge:	Existing studies on social biases in large language models focus on surface-level associations or isolated stereotypes.
Approach:	They propose a cognitively grounded bias evaluation framework to capture demographic biases across three contexts: single-actor, actor–actor and actor–observer.
Outcome:	The proposed framework captures comparative and perspective-driven biases overlooked in previous work.

Copied to clipboard

Tiejin Chen, Xiaoou Liu, Longchao Da, Jia Chen, Evangelos E. Papalexakis, Hua Wei

Dual Hierarchical Dialogue Policy Learning for Legal Inquisitive Conversational Agents (2026.findings-acl)

Challenge:	Existing methods for uncertainty quantification fail to capture multifaceted nature of natural language generation.
Approach:	They propose a multi-resource Uncertainty Quantification framework that integrates heterogeneous uncertainty signals into a unified measure.
Outcome:	The proposed framework outperforms existing methods on CoQA, NQ_Open, and HotpotQA.

Copied to clipboard

Xubo Lin, Zezhi Deng, Shihao Wang, Grace Hui Yang, Yang Deng

[b] = [d] - [t] + [p]: Self-supervised Speech Models Discover Phonological Vector Arithmetic (2026.findings-acl)

Challenge:	Existing systems for conversational AI are user-driven, but in many real-world situations, they do not extract information to achieve its own objectives.
Approach:	They propose an inquisitive conversational agent that learns when and how to ask probing questions . they also propose a framework for a conversational ICA specifically tailored to the court .
Outcome:	The proposed method outperforms single-agent RL baselines on a U.S. Supreme Court dataset.

Copied to clipboard

Kwanghee Choi, Eunjung Yeo, Cheol Jun Cho, David Harwath, David R. Mortensen

Waking Up Blind: Cold-Start Optimization of Supervision-Free Agentic Trajectories for Grounded Visual Perception (2026.findings-acl)

Challenge:	Existing studies on how self-supervised speech models encode rich phonetic information have not explored how they are structured.
Approach:	They conduct a comprehensive analysis of the underlying structure of S3M representations with particular attention to phonological vectors.
Outcome:	The proposed model encodes phonologically interpretable and compositional vectors, demonstrating phonology vector arithmetic.

Copied to clipboard

Ashutosh Bajpai, Tamal Majumder, Akshay Nambi, Tanmoy Chakraborty

Spatial-RAG: Spatial Retrieval Augmented Generation for Real-World Geospatial Reasoning Questions (2026.findings-acl)

Challenge:	Small Vision-Language Models (SVLMs) suffer from visual brittleness and poor tool orchestration.
Approach:	They propose a supervision-free framework that bootstraps agentic capabilities via Coldstart Reinforcement Learning for SVLMs.
Outcome:	The proposed framework improves task accuracy and tool efficiency by 5% and 9%.

Copied to clipboard

Dazhou Yu, Riyang Bao, Ruiyu Ning, Jinghong Peng, Gengchen Mai, Liang Zhao

Program Structure-aware Language Models: Targeted Software Testing beyond Textual Semantics (2026.findings-acl)

Challenge:	Existing large language models lack spatial computing capabilities and access to up-to-date geospatial data.
Approach:	They propose a Retrieval-Augmented Generation framework for geospatial question answering . it integrates structured spatial databases with LLMs via a hybrid spatial retriever .
Outcome:	Experiments show that Spatial-RAG significantly improves over baselines.

Copied to clipboard

Khang Tran, Khoa Nguyen, Cristian Borcea, Hai Phan

Reasoning for Hierarchical Text Classification: The Case of Patents (2026.findings-acl)

Challenge:	Recent advances in large language models for test case generation have improved branch coverage via prompt-engineered mutations, limiting their effectiveness for discovering subtle bugs and security vulnerabilities.
Approach:	They propose a program structure-aware LLM framework that integrates code property graphs and code semantics to condition test case generation on execution branches.
Outcome:	Experiments on real-world projects show that GLMTest improves branch accuracy from 27.4% to 50.2% on TestGenEval benchmark compared with state-of-the-art LLMs, i.e., Claude-Sonnet-4.5 and GPT-4o-mini.

Copied to clipboard

Lekang Jiang, Wenjun Sun, Stefan Goetz

From Shijing to English and German: Resources and Evaluation for LLM Translation of Early Chinese Poetry (2026.findings-acl)

Challenge:	Hierarchical text classification (HTC) is one of the hardest HTC scenarios because of professional difficulties and extensive labels.
Approach:	They propose a framework that reformulates hierarchical classification as a step-by-step reasoning task.
Outcome:	The proposed framework outperforms supervised fine-tuning benchmarks on other widely used HTC benchmarks.

Copied to clipboard

Ying Jiao, Meng Sun

One Mask to Rule Them All: On Hidden Facts after Editing and How to Find Them (2026.findings-acl)

Challenge:	Large language models (LLMs) show promise in literary translation, but their performance in poetry remains unexplored.
Approach:	They propose a framework that integrates knowledge-driven, rule-based, and LLM-as-judge metrics into a Shijing corpus . their code, lexical KB, and corpus reconstruction protocols are available at https://github.com/ML-KULeuven/ShijingLLMTrans.
Outcome:	The proposed framework achieves higher human correlation than traditional metrics and high statistical stability.

Copied to clipboard

Ali Kholmovaia, Paul Youssef, Nandi Schoots, Christin Seifert

Analysing the Safety Pitfalls of Steering Vectors (2026.findings-acl)

Challenge:	Knowledge editing methods such as ROME and MEMIT update factual associations by modifying MLP weights.
Approach:	They propose to use a mask to reverse edits by eliminating overattention in later layers . they also show that injecting the mask during editing drops editing success from 98% to 38% .
Outcome:	The proposed method reverses edits by eliminating overattention in later layers and drops editing success from 98% to 38%.

Copied to clipboard

Yuxiao Li, Alina Fastowski, Efstratios Zaradoukas, Bardh Prenkaj, Gjergji Kasneci

Compartmentalised Agentic Reasoning for Clinical NLI (2026.findings-acl)

Challenge:	Activation steering has emerged as a powerful tool to shape LLM behaviour without the need for weight updates.
Approach:	They propose to audit steering vectors obtained with Contrastive Activation Addition (CAA) and propose a mechanistic explanation for this finding.
Outcome:	The proposed approach significantly improves the success rate of jailbreak attacks, with stronger amplification under simple template-based attacks.

Copied to clipboard

Mael Jullien, Andre Freitas, Marco Valentino, Lei Xu

RAG or Learning? Understanding the Limits of LLM Adaptation under Continuous Knowledge Drift in the Real World (2026.findings-acl)

Challenge:	Large language models produce fluent judgments for clinical natural language inference, yet fail when the decision requires the correct inferential schema rather than surface matching.
Approach:	They propose a compartmentalised agentic framework that routes each premise–statement pair to a reasoning family and applies a specialised solver with explicit verification and targeted refinement.
Outcome:	The proposed framework improves mean accuracy from 23% with direct prompting to 57%, with the largest gains on structurally demanding reasoning types.

Copied to clipboard

Hanbing Liu, Lang Cao, Yang Li

Content Fuzzing for Escaping Information Cocoons on Digital Social Media (2026.findings-acl)

Challenge:	Existing methods to update or supplement large language models struggle under continuous knowledge drift.
Approach:	They propose a dynamic event benchmark and time-aware retrieval baseline that captures how knowledge evolves over time.
Outcome:	The proposed method enables systematic evaluation of model adaptation under continuous knowledge drift.

Copied to clipboard

Yifeng He, Ziye Tang, Hao Chen

KnowledgeBerg: Evaluating Systematic Knowledge Coverage and Compositional Reasoning in Large Language Models (2026.findings-acl)

Challenge:	Information cocoons restrict users’ exposure to posts with diverse viewpoints . social media platforms restrict the range of viewpoints that users encounter .
Approach:	They propose a confidence-guided fuzzing framework that rewrites posts while preserving their human-interpreted intent and induces different machine-inferred stance labels.
Outcome:	The proposed framework rewrites posts while preserving human-interpreted intent and induces different machine-inferred stance labels while maintaining semantic integrity with respect to the original content.

Copied to clipboard

Xiao Zhang, Qianru Meng, Yongjian Chen, Yumeng Wang, Johan Bos

Hybrid Self-evolving Structured Memory for Computer-Use Agents (2026.findings-acl)

Challenge:	Existing LLMs lack systematic coverage of a bounded knowledge universe and compositional set-based reasoning over that universe.
Approach:	They propose a benchmark for multiple-choice questions based on 1,183 enumeration seeds . they use knowledge width, cardinality of required universe, reasoning depth to formalize the challenge .
Outcome:	The proposed benchmarks achieve only 5.26–36.88 F1 on universe enumeration and 16.00–44.19 accuracy on knowledge-grounded reasoning.

Copied to clipboard

Sibo Zhu, Wenyi WU, Kun Zhou, Stephen Wang, Biwei Huang

Characterizing Selective Refusal Bias in Large Language Models (2026.findings-acl)

Challenge:	despite advances in vision–language models, real-world computer-use tasks remain challenging due to long-horizon workflows, diverse interfaces, and frequent intermediate errors.
Approach:	They propose a graph-based memory that couples discrete symbolic nodes with continuous trajectory embeddings.
Outcome:	The proposed system outperforms closed-source models in Qwen2.5-VL-7B and Gemini2.5-Pro-Vision on desktop and mobile platforms.

Copied to clipboard

Adel Khorramrouz, Sharon Levy

Infusing Theory of Mind into Socially Intelligent LLM Agents (2026.findings-acl)

Challenge:	a recent study shows that safety guardrails in large language models can inadvertently introduce or reflect new biases as they may refuse to generate harmful content targeting some demographic groups and not others.
Approach:	They examine the selective refusal bias in large language models by examining demographics and responses.
Outcome:	The proposed model fails to defend against an indirect attack on previously refused groups in 89% of the trials.

Copied to clipboard

EunJeong Hwang, Yuwei Yin, Giuseppe Carenini, Peter West, Vered Shwartz

Do BabyLMs Wanna Learn Wanna Contraction? On the Learnability without Language-Specific Bias (2026.findings-acl)

Challenge:	Theory of Mind (ToM) is a key aspect of human social intelligence, yet chatbots and LLMs do not typically integrate it.
Approach:	They propose a method that integrates Theory of Mind (ToM) into chatbots and dialogue agents to generate mental states between dialogue turns.
Outcome:	The proposed method improves dialogue and social interaction by integrating ToM with dialogue lookahead.

Copied to clipboard

Kangsan Noh, Sanghoun Song

OpenGenAlign: A Preference Dataset and Benchmark for Trustworthy Reward Modeling in Open-Ended, Long-Context Generation (2026.findings-acl)

Challenge:	grammatical constraints on wanna contraction can be learned by artificial learners under cognitively plausible learning conditions.
Approach:	They evaluate 24 BabyLMs from 2024 BabyLM Challenge and four standard models . they find that only encoder-based BabyLM models can capture wanna contraction .
Outcome:	The proposed models show modest but meaningful sensitivity on large datasets and high-frequency wanna instances.

Copied to clipboard

Hanning Zhang, Juntong Song, Juno Zhu, Yuanhao Wu, Tong Zhang, Cheng Niu

Understanding LLMs’ summarization capabilities: an analysis of biomedical abstract and lay summary generation (2026.findings-acl)

Challenge:	Existing reward models perform suboptimal on held-out benchmarks, resulting in poor quality outputs.
Approach:	They propose a framework and a high-quality dataset to evaluate reward models . they define four key metrics to assess generation quality and develop a pipeline to evaluate outputs .
Outcome:	The proposed framework and dataset improves hallucination-free, comprehensive, reliable, and efficient open-ended long-context generation.

Copied to clipboard

Batuhan Nursal, Cassie S. Mitchell

FIFA: Unified Faithfulness Evaluation Framework for Text-to-Video and Video-to-Text Generation (2026.findings-acl)

Challenge:	Abstracts use technical language for academic audiences, while lay summaries aim to make findings accessible to non-specialists.
Approach:	They evaluate the performance of lightweight LLMs in generating biomedical abstracts and lay summaries in a zero-shot setting.
Outcome:	The proposed models perform well in generating biomedical abstracts and lay summaries in a zero-shot setting.

Copied to clipboard

Liqiang Jing, Viet Dac Lai, Seunghyun Yoon, Trung Bui, Xinya Du

RLHS: Mitigating Misalignment in RLHF with Hindsight Simulation (2026.findings-acl)

Challenge:	Existing evaluation methods for VideoMLLMs are limited to one task and fail to assess hallucinations in open-ended, free-form responses.
Approach:	They propose a unified framework that extracts comprehensive descriptive facts and models their semantic dependencies via a Spatio-Temporal Semantic Dependency Graph.
Outcome:	The proposed framework aligns more closely with human judgment than existing evaluation methods and improves factual consistency in both text and video generation.

Copied to clipboard

Kaiqu Liang, Haimin Hu, Ryan Liu, Thomas L. Griffiths, Jaime Fernández Fisac

Propaganda Signals in LLMs: Perspectival Divergence and Narrative Framing in the Russia-Ukraine War (2026.findings-acl)

Challenge:	Reinforcement Learning from Hindsight Simulation (RLHF) can cause severe misalignment in generative AI, but it is not a universal method for fine-tuning large language models.
Approach:	They propose a method that uses evaluator feedback to decouple alignment signal from potentially compromised predictions.
Outcome:	The proposed method significantly outperforms RLHF in comparisons with baselines and human evaluations.

Copied to clipboard

Ofir Shabat, Ido Guy, Kira Radinsky

DELTA: Dynamic Layer-Aware Token Attention for Efficient Long-Context Reasoning (2026.findings-acl)

Challenge:	Large Language Models are increasingly used to explain, summarize, and translate real-world events . a recent study examined whether LLMs reproduce conflict-specific propaganda .
Approach:	They evaluate LLMs under several prompting contexts to determine which side they are closer to . they find model-specific leanings and technique profiles that persist across prompts .
Outcome:	The proposed model outputs align with competing narratives from different information ecosystems.

Copied to clipboard

Hossein Entezari Zarch, Lei Gao, Chaoyi Jiang, Murali Annavaram

LANTERN in the Event Stream: Training-Free Temporal Knowledge Graph Forecasting by Balancing Inertia and Shifts (2026.findings-acl)

Challenge:	Large reasoning models generate long chains of intermediate steps, but their inference cost is dominated by decoding, where each new token must attend to the entire growing sequence.
Approach:	They propose a training-free sparse attention mechanism that reduces inference cost by evicting entries from the key-value cache.
Outcome:	The proposed model matches or surpasses full attention on reasoning benchmarks . it reduces the number of attended tokens by up to 4.25 and delivers 1.54 speedup .

Copied to clipboard

Chengyuan Jin, Ao Chang, Daojian Zeng, Wenhao Teng, Xiangwen Liao, Kang Liu, Jun Zhao, Yubo Chen

Large Language Models for IT Automation Tasks: Are We There Yet? (2026.findings-acl)

Challenge:	Temporal knowledge graph forecasting (TKGF) uses long-window strengthscores and short-windowed novelty scores to predict missing entities in future queries.
Approach:	They propose a training-freeprompting framework that uses two perspectives of history to predict missing entities in future queries.
Outcome:	The proposed framework outperforms the state-of-the-art baselineAnRe framework in ICEWS14, ICEW05-15, and GDELT.

Copied to clipboard

Md. Mahadi Hassan, John Salvador, Akond Ashfaque Ur Rahman, Santu Karmaker

Follow the Path: Reasoning over Knowledge Graph Paths to Improve Large Language Model Factuality (2026.findings-acl)

Challenge:	Existing benchmarks rely on synthetic tasks that fail to capture the needs of practitioners who use IT automation tools.
Approach:	They evaluate 14 open-source and 3 proprietary LLMs and find that GPT-4.1-Mini achieves the best pass@10 rate of 23.9%, while Claude-3.5-Sonnet achieves best pass @1 performance.
Outcome:	The evaluated LLMs perform poorly in 126 tasks and show that they lack state reconciliation capabilities and lack module knowledge.

Copied to clipboard

Mike Zhang, Johannes Bjerva, Russa Biswas

Time-RA: Towards Time Series Reasoning for Anomaly Diagnosis with LLM Feedback (2026.findings-acl)

Challenge:	fs1 improves factuality of reasoning traces by sourcing them from large reasoning models and conditioning them on knowledge graph paths.
Approach:	They propose a method that improves the factuality of reasoning traces by sourcing them from large reasoning models and grounding them by conditioning on knowledge graph (KG) paths.
Outcome:	The proposed method outperforms instruction-tuned models on open-domain questions . it significantly improves model performance over more complex questions and numerical answer types compared to baselines.

Copied to clipboard

Yiyuan Yang, Zichuan Liu, Lei Song, Kai Ying, Stephen Wang, Joshua Thomas Bamford, Svitlana Vyetrenko, Jiang Bian, Qingsong Wen

Ground Then Rank: Revisiting Knowledge-Based VQA with Training-Free Entity Identification (2026.findings-acl)

Challenge:	Time series anomaly detection (TSAD) has traditionally focused on binary classification and lacks the fine-grained categorization and explanatory reasoning required for transparent decision-making.
Approach:	They propose a time-series reasoning task that reformulates TSAD from discriminative to reasoning-intensive paradigm.
Outcome:	The proposed task reformulates TSAD from discriminative to reasoning-intensive paradigm.

Copied to clipboard

Qian Ma, Qiong Wu, Zhengyi Zhou, Yao Ma

Non-invasive electromyographic speech neuroprosthesis: a geometric perspective (2026.findings-acl)

Challenge:	Existing multi-modal retrieval augmented generation (MM-RAG) methods tightly couple entity discrimination and section-level evidence ranking into a single re-ranking stage, leading to high cost and limited generalization.
Approach:	They propose a framework that decouples entity identification from section-level re-ranking.
Outcome:	The proposed framework outperforms fine-tuned multi-modal re-ranking baselines while reducing training and inference complexity.

Copied to clipboard

Harshavardhana T Gowda, Lee M. Miller

Context Attribution with Multi-Armed Bandit Optimization (2026.findings-acl)

Challenge:	Existing methods for mapping EMG to time-aligned audio limits applicability to patients who can no longer speak.
Approach:	They propose a neuromuscular speech interface that translates silently voiced articulations directly into text.
Outcome:	The proposed interface can translate silently voiced articulations directly into text without audio transfer . the proposed system could restore communication for patients with speech loss .

Copied to clipboard

Deng Pan, Keerthiram Murugesan, Ting Hua, Nuno Moniz, Nitesh V. Chawla

Peering Behind the Shield: Guardrail Identification in Large Language Models (2026.findings-acl)

Challenge:	Existing approaches to augmenting attribution with retrieval-augmented generation (RAG) focus on training models to explicitly cite context segments during generation, but their reliability remains unverifiable.
Approach:	They propose a framework that formulates context attribution as a combinatorial multi-armed bandit problem by using Linear Thompson Sampling to efficiently identify the most influential context segments while minimizing the number of model queries.
Outcome:	The proposed method reduces model queries by 30% while matching or exceeding the attribution quality of existing approaches.

Copied to clipboard

Ziqing Yang, Yixin Wu, Rui Wen, Michael Backes, Yang Zhang

WildSci: Advancing Scientific Reasoning from In-the-Wild Literature (2026.findings-acl)

Challenge:	Identifying guardrails in conversational AI agents is critical for identifying malicious content . identifying guardrail components in black-box AI agents poses security challenges .
Approach:	They propose a method that leverages guard-specific adversarial prompts to detect guardrails in black-box AI agents.
Outcome:	The proposed method achieves perfect classification accuracy in multiple scenarios.

Copied to clipboard

Tengxiao Liu, Deepak Nathani, Zekun Li, Kevin Yang, William Yang Wang

WebDART: Dynamic Decomposition and Re-planning for Complex Web Tasks (2026.findings-acl)

Challenge:	Recent advances in large language model reasoning focus on mathematics and coding domains, but scientific reasoning remains limited in other domains due to limited dataset coverage.
Approach:	They propose a framework for sustainable scientific reasoning QA generation by synthesizing a new dataset of domain-specific science questions from peer-reviewed literature.
Outcome:	The proposed framework and dataset enable scalable and sustainable research in scientific reasoning.

Copied to clipboard

Jingbo Yang, Bairu Hou, Wei Wei, Shiyu Chang, Yujia Bao

Like a Therapist, But Not: Reddit Narratives of AI in Mental Health Contexts (2026.findings-acl)

Challenge:	Large-language-model (LLM) agents are competent at straightforward web tasks, but struggle with complex tasks.
Approach:	They propose a general framework that decomposes web tasks into three subtasks . they show that WebDART lifts end-to-end success rates by 13.7 percentage points .
Outcome:	Evaluated on WebChoreArena, WebDART lifts success rates by 13.7 percentage points over previous state-of-the-art agents.

Copied to clipboard

Elham Aghakhani, Rezvaneh Rezapour

VoiceStar: Robust Zero-Shot Autoregressive TTS with Duration Control and Extrapolation (2026.findings-acl)

Challenge:	Large language models are increasingly used for emotional support and mental health–related interactions outside clinical settings.
Approach:	They analyze 5,126 Reddit posts describing use of AI for emotional support or therapy . positive sentiment is most strongly associated with task and goal alignment, they say .
Outcome:	The proposed framework analyzes language, adoption-related attitudes, and relational alignment at scale. positive sentiment is most strongly associated with task and goal alignment.

Copied to clipboard

Puyuan Peng, Zhisheng Zheng, Shang-Wen Li, Abdelrahman Mohamed, David Harwath

Evaluating Implicit Biases in LLM Reasoning through Logic Grid Puzzles (2026.findings-acl)

Challenge:	Neural codec language models (NCLMs) lack fine-grained controllability and inability to extrapolate to sequence lengths much longer than those seen during training.
Approach:	They propose a novel autoregressive encoder-decoder neural codec language model that can be trained with a Continuation-Prompt Mixed training system.
Outcome:	The proposed model outperforms or is on par with current state-of-the-art models on short-form benchmarks such as LibriSpeech and Seed-TTS in terms of intelligibility and naturalness.

Copied to clipboard

Fatima Jahara, Mark Dredze, Sharon Levy

InferPilot: Autonomous Inference Attacks Against ML Services With LLM-Based Agents (2026.findings-acl)

Challenge:	Existing methods to eliminate implicit biases in LLMs do not eradicate underlying behavioral bias.
Approach:	They propose a framework that uses logic grid puzzles to probe the influence of social stereotypes on logical reasoning and decision making in LLMs.
Outcome:	The proposed framework systematically probes the influence of social stereotypes on logical reasoning and decision making in LLMs.

Copied to clipboard

Yixin Wu, Rui Wen, Chi Cui, Michael Backes, Yang Zhang

MedErrBench: A Fine-Grained Multilingual Benchmark for Medical Error Detection and Correction with Clinical Expert Annotations (2026.findings-acl)

Challenge:	Inference attacks are important for assessing model's robustness, but their implementation and parameters are challenging for non-experts.
Approach:	They propose an autonomous agent capable of conducting inference attacks without human intervention.
Outcome:	The proposed agent achieves a 100.0% task completion rate and near-expert attack performance with an average token cost of only 0.627 per run.

Copied to clipboard

Congbo Ma, Yichun Zhang, Yousef Al-Jazzazi, Ahamed Foisal, Laasya Sharma, Yousra Sadqi, Khaled Saleh, Jihad Mallat, Farah E. Shamout

MetaScale: Test-Time Scaling with Evolving Meta-Thoughts (2026.findings-acl)

Challenge:	Existing or generated clinical text may contain inaccuracies that can lead to serious adverse outcomes.
Approach:	They introduce a multilingual benchmark for error detection, localization and correction . they assessed the performance of a range of general-purpose, language-specific, and medical-domain language models .
Outcome:	The proposed benchmark covers English, Arabic and Chinese, with natural medical cases annotated and reviewed by domain experts.

Copied to clipboard

Qin Liu, Wenxuan Zhou, Nan Xu, James Y. Huang, Fei Wang, Sheng Zhang, Hoifung Poon, Muhao Chen

Arab Voices: Mapping Standard and Dialectal Arabic Speech Technology (2026.findings-acl)

Challenge:	Existing approaches impose fixed cognitive structures that enhance performance in specific tasks but lack adaptability across diverse scenarios.
Approach:	They propose a test-time scaling framework based on meta-thoughts to improve performance . meta-thinkts are adaptive thinking strategies tailored to a given task .
Outcome:	Experimental results show that MetaScale outperforms standard inference approaches . it can scale more effectively with increasing sampling budgets and produces more structured responses .

Copied to clipboard

Peter Sullivan, AbdelRahim A. Elmadany, Alcides Alcoba Inciarte, Muhammad Abdul-Mageed

CoDA: Restoring Contextual Dominance via Copy-Encouraged Attention Intervention for Mitigating RAG Hallucinations (2026.findings-acl)

Challenge:	Dialectal Arabic datasets embody a range of domain, dialect, and quality.
Approach:	They propose a framework for automatic speech recognition in dialectal Arabic to address the limited data availability encountered in dialects.
Outcome:	The proposed framework provides access to 31 datasets covering 14 dialects to better address the limited data availability encountered in dialectal Arabic speech processing.

Copied to clipboard

JinWei Shi, Qizhuo Xie, Qianzi Hou, Zhipeng Wang, Wanting Su, Jianhua Zhao, Tao Zheng, Tieke He

CausalDetox: Causal Head Selection and Intervention for Language Model Detoxification (2026.findings-acl)

Challenge:	Retrieval-augmented generation reduces hallucination by grounding outputs in external evidence.
Approach:	They propose a lightweight inference-time attention intervention that amplifies evidence-aligned value states to enhance contextual faithfulness and reduce hallucination.
Outcome:	The proposed model reduces hallucination by grounding model outputs in external evidence.

Copied to clipboard

Yian Wang, Yuen Chen, Agam Goyal, Hari Sundaram

Commonsense Knowledge with Negation: A Resource to Enhance Negation Understanding (2026.findings-acl)

Challenge:	Large language models (LLMs) frequently generate toxic content, posing significant risks for safe deployment.
Approach:	They propose a framework that identifies and intervenes on the specific attention heads causally responsible for toxic generation.
Outcome:	The proposed framework reduces toxic generation by 5.34% while preserving linguistic fluency and speeding up head selection.

Copied to clipboard

Zijie Wang, MohammadHossein Rezaei, Farzana Rashid, Eduardo Blanco

P-QuASAR: A Unified Probabilistic Framework for Holistic Patent Quality Assessment and Refinement (2026.findings-acl)

Challenge:	Negation is a common and important semantic feature in natural language, yet Large Language Models struggle when negation is involved in natural learning tasks.
Approach:	They propose to augment existing corpora with negation by automatically augmenting existing ones with negations by combining multiple triples with if-then relations.
Outcome:	The proposed approach yields two new corpora containing over 2M triples with if-then relations.

Copied to clipboard

Xinyuan Song, Ziyi Ni, Fred Yang, Bo Zhang, Yijin Wang, Jane Zhang

From Heard to Lived Opinions: Simulating Opinion Dynamics with Grounded LLM Agents in Economic Environments (2026.findings-acl)

Challenge:	Existing methods for assessing patent quality rely on modular pipelines or generic detectors, resulting in fragmented decisions and limited integration across quality dimensions.
Approach:	They propose a probabilistic framework that represents patent specifications as Quality Graphs.
Outcome:	The proposed framework outperforms existing methods on 500 patents against seven baselines.

Copied to clipboard

Ryuji Hashimoto, Masahiro Kaneko, Ryosuke Takata, Takehiro Takayanagi, Kiyoshi Izumi

HiSA: Hierarchical State Abstraction for Scalable GUI Agents (2026.findings-acl)

Challenge:	Existing studies on opinion dynamics (OD) focus primarily on opinion exchange, with opinion change driven by linguistic interaction.
Approach:	They propose a OD simulation framework that grounds LLM-based agents in an economic environment and allows them to act and receive environmental feedback.
Outcome:	The proposed framework shows that LLM-based agents can act and receive environmental feedback at both individual and population levels while generating larger distributional shifts.

Copied to clipboard

Weiming Li, Hye-young Paik, Yulei Sui

Challenge:	Recent multimodal large language models (MLLMs) exploit insufficient state abstraction to automate workflows.
Approach:	They propose a hierarchical state abstraction approach that actively restructures knowledge rather than passively retaining historical information.
Outcome:	The proposed approach achieves a 40.58% success rate while reducing token consumption by 69.85% and monetary costs by 55.10% compared to the best-performing baseline.

SecureWebArena: A Holistic Security Evaluation Benchmark for LVLM-based Web Agents (2026.findings-acl)

Copied to clipboard

Zonghao Ying, Yangguang Shao, Jianle Gan, Gan Xu, Wenxin Zhang, Quanchen Zou, Junzheng Shi, Zhenfei Yin, Mingchuan Zhang, Aishan Liu, Xianglong Liu

Challenge:	Existing security benchmarks only cover user-level prompts and environmental threats . however, these models are vulnerable to pop-up attacks and prompt injections .
Approach:	They propose a security benchmark that covers a set of six attack vectors that span both user-level and environment-level manipulations.
Outcome:	The proposed security benchmarks cover a set of six real-world web environments with 2,970 adversarial trajectories and a multi-layered evaluation protocol dissecting agent failures across internal reasoning, behavioral execution, and task outcomes.

The Mechanics of Interference: Defusing Distractors in RAG via Sparse Autoencoder Interventions (2026.findings-acl)

Copied to clipboard

Christian Giannetti, Giovanni Trappolini, Nicola Tonellotto, Fabrizio Silvestri, Pietro Lio

Grounding Agent Memory in Contextual Intent (2026.findings-acl)

Challenge:	Large language models exhibit a critical vulnerability to distractor interference when processing retrieval-augmented contexts.
Approach:	They propose a mechanistic framework that corrects this failure mode through targeted interventions in the model’s latent space.
Outcome:	The proposed framework achieves recovery rates of up to 94% on distractor-vulnerable samples on Gemma-2 and Llama-3 model families across three QA benchmarks.

Copied to clipboard

Ruozhen Yang, Yucheng Jiang, Yueqi Jiang, Priyanka Kargupta, Yunyi Zhang, Jiawei Han

Persona-Assigned Large Language Models Exhibit Human-Like Motivated Reasoning (2026.findings-acl)

Challenge:	Large language models are deployed in long-horizon tasks that require agents to track interleaved goals, resolve references to prior information, and coordinate actions over extended trajectories.
Approach:	They propose an agentic memory system that indexes each trajectory step with a structured retrieval cue, contextual intent, and retrieves history by matching the current step’s intent.
Outcome:	The proposed system outperforms the strongest benchmark by 35.6%, with the largest gains as trajectory length increases.

Copied to clipboard

Saloni Dash, Amélie Reymond, Emma Spiro, Aylin Caliskan

Generalizable LLM Learning of Graph Synthetic Data with Post-training Alignment (2026.findings-acl)

Challenge:	Prior studies have reported that large language models (LLMs) are also susceptible to human-like cognitive biases, but the extent to which LLMs selectively reason toward identity-congruent conclusions remains unexplored.
Approach:	They investigate whether assigning 8 personas across 4 political and socio-demographic attributes induces motivated reasoning in LLMs.
Outcome:	The proposed model is assigned 8 personas across 4 political and socio-demographic attributes and shows that they have 9% reduced veracity discernment compared to models without persona.

Copied to clipboard

Yizhuo Zhang, Heng Wang, Shangbin Feng, Zhaoxuan Tan, Xinyun Liu, Yulia Tsvetkov

Distilling the Essence: Efficient Reasoning Distillation via Sequence Truncation (2026.findings-acl)

Challenge:	Existing research has focused on enhancing graph reasoning capabilities of LLMs by supervised fine-tuning on synthetic graph data.
Approach:	They propose to unlock generalizable learning of graph with post-training alignment with synthetic graph data by aligning off-the-shelf LLMs and LLM fine-tuned on synthetic graphs.
Outcome:	The proposed algorithm improves on synthetic graph problems and out-of-domain tasks with implicit graph structures.

Copied to clipboard

Wei-Rui Chen, Vignesh Kothapalli, Ata Fatahibaarzi, Hejian Sang, Shao Tang, Qingquan Song, Zhipeng Wang, Muhammad Abdul-Mageed

Beyond Cross-Modal Alignment: Measuring and Leveraging Modality Gap in Vision-Language Models (2026.findings-acl)

Challenge:	a large reasoning model (LRM) training on large amounts of reasoning data is computationally expensive.
Approach:	They propose a method to quantify computation-quality tradeoffs as a function of sequence length.
Outcome:	The proposed method reduces training time, memory and FLOPs by 50% on long training sequences while retaining the full-sequence performance.

Copied to clipboard

Hanqi Yan, Xiangxiang Cui, Lu Yin, Jindong Gu, Paul Pu Liang, Yulan He, Yifei Wang

Data Swarms: Optimizable Generation of Synthetic Evaluation Data (2026.findings-acl)

Challenge:	a recent study shows that vision-language models have modality gaps that persist even in well-aligned models.
Approach:	They propose a modality-dominance score to measure and leverage modality gaps . they propose automatic interpretability metrics to evaluate these features in a scalable manner .
Outcome:	The proposed framework allows for training-free probing and editing methods for understanding model perception across genders and generating adversarial examples.

Copied to clipboard

Shangbin Feng, Yike Wang, Weijia Shi, Yulia Tsvetkov

Concept rather than Document: Context Compression via AMR-based Conceptual Entropy (2026.findings-acl)

Challenge:	Extensive experiments demonstrate that Data Swarms outperforms eight data generation baselines across five evaluation objectives.
Approach:	They propose an algorithm to optimize the generation of synthetic evaluation data and advance quantitative desiderata of LLM evaluation.
Outcome:	The proposed algorithm outperforms baseline evaluations and Adversarial Swarms generates harder data while learning from such data.

Copied to clipboard

Kaize Shi, Xueyao Sun, Xiaohui Tao, Lin Li, Qika Lin, Guandong Xu

Can LLMs Understand the Impact of Trauma? Costs and Benefits of LLMs Coding the Interviews of Firearm Violence Survivors (2026.findings-acl)

Challenge:	Existing methods rely on lexical or token-level features that fragment semantic units and fail to capture conceptually essential content.
Approach:	They propose an unsupervised framework leveraging Abstract Meaning Representation to preserve essential information while filtering irrelevant text.
Outcome:	The proposed framework outperforms RAG and existing baselines while preserving essential information while filtering irrelevant text.

Copied to clipboard

Jessica H Zhu, Shayla Stringfield, Vahe Zaprosyan, Michael Wagner, Michel Cukier, Joseph Richardson

"I Don’t Know What to Say": A Fact-Filling Questionnaire Method to Help Non-Experts Talk to LegalAI Assistant (2026.findings-acl)

Challenge:	Firearm violence research remains underfunded and difficult to scale due to the lack of funding from the NIH and CDC.
Approach:	They use open-source large language models to inductively code interviews with 21 Black men who have survived community firearm violence.
Outcome:	The use of open-source LLMs to inductively code interviews with 21 Black men shows that the models can identify important codes, but that they are highly sensitive to data processing.

Copied to clipboard

Yuting Huang, Yiquan Wu, Meitong Guo, Ang Li, Xiaozhong Liu, Keting Yin, Fei Wu, Kun Kuang

BnMMLU: Measuring Massive Multitask Language Understanding in Bengali (2026.findings-acl)

Challenge:	Recent advances in large language models (LLMs) have greatly expanded the scope of legal AI.
Approach:	They propose a method that generates questionnaires to help users refine queries . they leverage an iterative training process that collects valuable questionnaires .
Outcome:	The proposed method improves the completeness of queries and ensures the performance of domain-specific models in downstream legal tasks.

Copied to clipboard

Saman Sarker Joy, Swakkhar Shatabda

Diagnosing LLMs via Information Spectrum Analysis: Tail Behavior and the Effects of Side Information (2026.findings-acl)

Challenge:	Large-scale multitask benchmarks have driven rapid progress in language modeling, yet most emphasize low-resource languages like English.
Approach:	They propose a benchmark for massive multitask language understanding in Bengali . they use a dataset that preserves mathematical content via MathML and a subset of questions most frequently missed by top systems to stress difficult cases.
Outcome:	The proposed benchmark covers 24 model variants across 11 LLM families.

Copied to clipboard

Yuuki Tachioka

Challenge:	Large language models exhibit non-stationary generation because of variability in output distributions . authors propose a framework that treats LLMs as general sources without stationarity or ergodicity .
Approach:	They propose a diagnostic framework that treats large language models as general sources without stationarity, ergodicity, or the asymptotic equipartition property.
Outcome:	The proposed framework treats large language models as general sources without stationarity, ergodicity, or the asymptotic equipartition property.

Scaling Collaborative Effort with Agents (2026.findings-acl)

Copied to clipboard

Shannon Zejiang Shen, Valerie Chen, Ken Gu, Alexis Ross, Zixian Ma, Jillian Ross, Alex Gu, Chenglei Si, Wayne Chi, Andi Peng, Jocelyn J Shen, Ameet Talwalkar, Tongshuang Wu, David Sontag

Challenge:	Current evaluations of agents focus on producing high-quality, final outputs in one shot, failing to account for the inherently iterative nature of many real-world problems.
Approach:	They propose a framework that captures how an agent’s utility grows with increasing user involvement.
Outcome:	The proposed framework captures how an agent’s utility grows with increasing user involvement, revealing a missing ingredient in agent design: the ability to sustain engagement and scaffold user understanding.

CriticSearch: Fine-Grained Credit Assignment for Search Agents via a Retrospective Critic (2026.findings-acl)

Copied to clipboard

Yaocheng Zhang, Haohuan Huang, Zijun Song, Zijie Zhao, Qichao Zhang, Yuanheng Zhu, Dongbin Zhao

Double-Edged Sword or Sharp Tool? Designing and Evaluating Triadic LLM-Teacher Collaboration for K-12 Writing at Scale (2026.findings-acl)

Challenge:	Existing search agent pipelines rely on sparse outcome rewards, leading to inefficient exploration and unstable training.
Approach:	They propose a tool-integrated reasoning framework that provides turn-level feedback via a retrospective critic mechanism.
Outcome:	The proposed framework outperforms baselines in multi-hop reasoning benchmarks and achieves faster convergence and training stability.

Copied to clipboard

Canran Wang, Yuwen Yang, Zhen Wang, Ming MA, Ding Yu, Chentai Wang, Keman Huang, Xiaoyong Du

IntroLM: Introspective Language Models via Prefilling-Time Self-Evaluation (2026.findings-acl)

Challenge:	a large-scale empirical dataset involving 57,954 essays from 10,195 students across 120 schools over two years is presented in this paper.
Approach:	They propose a triadic collaboration system that supports K-12 writing learning . they propose linguistic expansion as a pedagogical gatekeeper and bridge .
Outcome:	The proposed system improves writing quality through a strategic labor division . authors find that excessive linguistic expansion yields diminishing marginal utility .

Copied to clipboard

Hossein Hosseini Kasnavieh, Gholamreza Haffari, Christopher Leckie, Adel N. Toosi

A Herd of Language Models Makes a Better Zero-shot Annotator for Clinical Named Entity Recognition (2026.findings-acl)

Challenge:	Existing methods to predict output quality of large language models rely on external classifiers with limited context windows and constrained representational capacity.
Approach:	They propose a method that enables causal language models to predict their own output quality during the prefilling phase without affecting generation using [CPX] tokens.
Outcome:	The proposed method outperforms existing classifiers on Qwen3-8B and DeBERTa-v3-Large models by 14% on question-answering benchmarks.

Copied to clipboard

Seiji Shimizu, Shoko Wakamiya, Eiji Aramaki

P2P: A Poison-to-Poison Remedy for Reliable Backdoor Defense in LLMs (2026.findings-acl)

Challenge:	Clinical named entity recognition (NER) is a core task in clinical NLP.
Approach:	They propose a label-modeling method for Multi-LLM Annotation using R**epresentation learning to capture contextual similarity.
Outcome:	The proposed method improves the average F1 score by 8.6% over zero-shot baselines while reducing annotation costs.

Copied to clipboard

Shuai Zhao, Xinyi Wu, Shiqian Zhao, Xiaobao Wu, Zhongliang Guo, Yanhao Jia, Anh Tuan Luu

Learning Temporally-Aware Sample Weights for Preference Optimization (2026.findings-acl)

Challenge:	Defending Large Language Models (LLMs) against backdoors has long been trapped in a "cat-and-mouse" dilemma where defenders passively react to ever-shifting attack strategies.
Approach:	They propose a general and effective defense algorithm that implants benign triggers to reshape the model’s decision boundary.
Outcome:	The proposed defense algorithm can neutralize malicious backdoors while preserving task performance.

Copied to clipboard

Mengyang Li, Xudong Zhou, Pinlong Zhao

Challenge:	Existing methods for preference optimization rely on static functions of instantaneous model states and ignore temporal learning dynamics.
Approach:	They propose a framework that meta-learns adaptive weights using three temporal features: reward margin evolution, learning volatility, and reference deviation.
Outcome:	The proposed framework achieves statistically significant improvements over baselines on models ranging from 7B to 70B parameters.

ToolPRMBench: Evaluating and Advancing Process Reward Models for Tool-using Agents (2026.findings-acl)

Copied to clipboard

Dawei Li, Yuguang Yao, Zhen Tan, Huan Liu, Ruocheng Guo

Challenge:	Reward-guided search methods have shown potential in enhancing tool-using agents . however, there is a lack of reliable evaluation benchmarks for PRMs in tool-use settings .
Approach:	They propose a large-scale benchmark specifically designed to evaluate PRMs for tool-using agents.
Outcome:	The proposed benchmark shows that tool reward models perform better in tool-using environments.

Adapt to Thrive! Adaptive Power-Mean Policy Optimization for Improved LLM Reasoning (2026.findings-acl)

Copied to clipboard

Yiming Huang, Zhenbo Shi, Shuzheng Gao, Cuiyun Gao, Peiyi Han, Chuanyi Liu

LEAF: Towards Lightweight Explainable Hateful Video Detection via Self-Grounding CoT Guided Stage-Wise Distillation (2026.findings-acl)

Challenge:	Existing methods for reinforcement learning with verifiable rewards (RLVR) rely on static objective functions and rigid clipping strategies that misalign with the model’s evolving reasoning capabilities.
Approach:	They propose to incorporate Power-Mean Policy Optimization (PMPO) and Feedback-Adaptive Clipping (FAC) to overcome limitations of static mechanisms.
Outcome:	Extensive experiments on nine reasoning tasks show the proposed paradigm outperforms state-of-the-art methods.

Copied to clipboard

Jian Lang, Rongpei Hong, Meihui Zhong, Kaiju Li, Ting Zhong, Qiang Gao, Fan Zhou

DemMA: Dementia Multi-Turn Dialogue Agent with Expert-Guided Reasoning and Action Simulation (2026.findings-acl)

Challenge:	Existing methods for detecting hateful videos rely on opaque models with no insight into their decisions.
Approach:	They propose a lightweight, explainable video detection framework that distills "explainability" from LMMs into efficient Smaller Multimodal Models (SMMs) they use a self-grounded chain-of-thought mechanism to generate unbiased supervision signals for videos .
Outcome:	The proposed framework outperforms existing methods in detection accuracy and explainability on three video benchmarks.

Copied to clipboard

Yutong Song, Jiang Wu, Kazi Shaharair Sharif, Pengfei Zhang, Wenjun Huang, Honghui Xu, Nikil Dutt, Amir M. Rahmani

SplitThenMerge: Token-Level Skill-Compositional Sparse Mixture-of-Experts for Complex Domain-Specific Tasks (2026.findings-acl)

Challenge:	Simulating dementia patients with large language models is challenging due to the need to model cognitive impairment, emotional dynamics, and nonverbal behaviors over long conversations.
Approach:	They propose an expert-guided dementia dialogue agent for multi-turn patient simulation . they introduce a framework that trains a single LLM to jointly generate reasoning traces, patient utterances, and aligned behavioral actions .
Outcome:	The proposed model outperforms baselines in persona fidelity, clinical validity, and educational effectiveness.

Copied to clipboard

Yuting Huang, Jiawen Zhang, Yiquan Wu, Yinghao Hu, Fei Wu, Kun Kuang

Challenge:	Existing domain adaptation methods train heterogeneous skills together, making it difficult to reliably coordinate multiple skills when solving complex tasks.
Approach:	They propose a framework that decomposes domain competence into atomic skills and composes them dynamically during generation.
Outcome:	The proposed framework decomposes domain competence into atomic skills, trains them independently, and composes them dynamically during generation.

Causal-Audit: Explicit and Auditable Graph-based Reasoning via Target-Aware Causal Chain Construction (2026.findings-acl)

Copied to clipboard

Su Lan, Xuefei Yin, Yanming Zhu, Alan Wee-Chung Liew

RADS: Reinforcement Learning-Based Sample Selection Improves Transfer Learning in Low-resource and Imbalanced Clinical Settings (2026.findings-acl)

Challenge:	Existing LLM-based methods rely on implicit language-level reasoning, resulting in opaque causal assumptions and fragile predictions.
Approach:	They propose an explicit and auditable causal reasoning framework for context-free intervention-based question answering that uses four modular stages rather than implicit end-to-end prediction.
Outcome:	The proposed framework outperforms existing LLM-based methods while providing interpretable and auditable causal reasoning traces.

Copied to clipboard

Wei Han, David Martinez Iraola, Anna Khanina, Lawrence Cavedon, Karin Verspoor

Watch Out Your Industrial Copilots: Stealthy Backdoor Attack Against LLM-Based PLC Code Generation (2026.findings-acl)

Challenge:	Experimental evaluations show that RL methods favor outliers rather than truly informative samples under low-resource and class-imbalanced conditions.
Approach:	They propose a robust sample selection strategy using reinforcement learning to identify the most informative samples using a class imbalance approach.
Outcome:	The proposed strategy improves model transferability while maintaining robust performance under extreme class imbalance compared to traditional methods.

Copied to clipboard

Xinyuan An, Liu Xiaoxia, Dongxia Wang, Zhanhang Xiong, Wenhai Wang

Balancing Classification and Calibration Performance in Decision-Making LLMs via Calibration Aware Reinforcement Learning (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are being used to generate PLC code from natural language.
Approach:	They propose a stealthy backdoor attack framework targeting LLM-based PLC code generation . they incorporate six malicious logic injection patterns and a pipeline to refine stealthiness .
Outcome:	The proposed framework achieves 82.92% success rate while remaining stealthy . it bypasses quality validation and is difficult to detect .

Copied to clipboard

Duygu Nur Yaldiz, Evangelia Spiliopoulou, Zheng Qi, Siddharth Varia, Srikanth Doss, Nikolaos Pappas

Text Classification

Challenge:	Large language models (LLMs) are increasingly deployed in decision-making tasks where accuracy and reliable confidence estimates are essential.
Approach:	They propose a calibration-aware reinforcement learning formulation that directly adjusts decision-token probabilities.
Outcome:	The proposed model preserves RLVR’s accuracy level while mitigating overconfidence, reducing ECE scores up to 9 points.

Redefining Machine Simultaneous Interpretation: From Incremental Translation to Human-Like Strategies (2026.findings-acl)

Copied to clipboard

Qianen Zhang, Zeyu Yang, Satoshi Nakamura

ChunQiuTR: Time-Keyed Temporal Retrieval in Classical Chinese Annals (2026.findings-acl)

Challenge:	Simultaneous machine translation requires high-quality translations under strict real-time constraints.
Approach:	They extend the action space of simultaneous machine translation with four adaptive actions . they adapt these actions in a large language model framework and construct training references .
Outcome:	The proposed framework improves semantic metrics and achieves lower delay compared to reference translations and salami-based baselines.

Copied to clipboard

Yihao Wang, Zijian He, Jie Ren, Keze Wang

OFFSIDE: Benchmarking Unlearning Misinformation in Multimodal Large Language Models (2026.findings-acl)

Challenge:	Historical research often focuses on finding exact record for a specific regnal month . classical Chinese sources are a canonical example of evidence-centric retrieval .
Approach:	They propose a time-keyed retrieval benchmark that organizes records by month-level reign keys . they propose 'CTD', a dual-encoder that combines absolute context with offset biasing .
Outcome:	The proposed benchmark organizes records by month-level reign keys and includes chrono-near confounders that mimic real retrieval failures.

Copied to clipboard

Hao Zheng, Zirui Pang, Ling Li, Zhijie Deng, Yuhan Pu, Zhaowei Zhu, Xiaobo Xia, Jiaheng Wei

PedagogyBench: A Cognitive-Driven Benchmark for Multimodal Instructional Video Understanding (2026.findings-acl)

Challenge:	Existing benchmarks for MU are limited by a lack of image diversity and coarse-grained unlearning targets.
Approach:	They propose a benchmark to evaluate misinformation unlearning in MLLMs . OFFSIDE supports advanced unlearning targets such as fine-grained unlearning and visual rumor removal.
Outcome:	OFFSIDE supports advanced unlearning targets, such as fine-grained unlearning and visual rumor removal.

Copied to clipboard

Xiaokang Jin, Jia Zhu, Jingjiang Liu, Yabing Shi, Jueqi Guan, Hao Chen, Pasquale De Meo

FeedEval: Pedagogically Aligned Evaluation of LLM-Generated Essay Feedback (2026.findings-acl)

Challenge:	Existing video understanding benchmarks do not adequately capture the pedagogical logic embedded in instructional videos.
Approach:	They propose a pedagogy-driven segmentation strategy and a dual-stream semantic injection pipeline that combines machine pre-annotation with expert refinement.
Outcome:	The proposed model performs well on discriminative tasks but degrades on higher-order pedagogical diagnosis, relying on parametric memory rather than grounded visual perception.

Copied to clipboard

SeongYeub Chu, Jongwoo Kim, Mun Yong Yi

Challenge:	Recent research emphasizes the generation of high-quality feedback that provides justification and actionable guidance.
Approach:	They propose an LLM-based framework for evaluating LLM feedback along three dimensions: specificity, helpfulness, and validity.
Outcome:	The proposed framework evaluates LLM-generated feedback along three dimensions: specificity, helpfulness, and validity.

Probing Multimodal Large Language Models on Cognitive Biases in Chinese Short-Video Misinformation (2026.findings-acl)

Copied to clipboard

Jen-tse Huang, Chang Chen, Shiyang Lai, Wenxuan Wang, Michelle R Kaufman, Mark Dredze

Emotion–Cause Pair Extraction in Conversations via Semantic Decoupling and Alignment (2026.findings-acl)

Challenge:	Short-video platforms have become major channels for misinformation, but their robustness against misinformation entangled with cognitive biases remains under-explored.
Approach:	They propose a framework for evaluation of short-video platforms that use visual cues and social cue.
Outcome:	The proposed framework evaluates MLLMs across five modality settings.

Copied to clipboard

Tianxiang Ma, Weijie Feng, Xinyu Wang, Zhiyong Cheng

Failure makes the agent stronger: Enhancing Accuracy through Structured Reflection for Reliable Tool Interactions (2026.findings-acl)

Challenge:	Emotion-Cause Pair Extraction in Conversations (ECPEC) aims to identify the set of causal relations between emotion utterances and their triggering causes within a dialogue.
Approach:	They propose a framework for Emotion-Cause Pair Extraction in Conversations that decouples emotion-oriented semantics from cause-oriented ones and employs optimal transport to enable many-to-many and globally consistent emotion-cause matching.
Outcome:	The proposed framework achieves state-of-the-art on several benchmark datasets.

Copied to clipboard

Junhao Su, Yuanliang Wan, Junwei Yang, Hengyu Shi, Tianyang Han, Yurui Qiu, Junfeng Luo

Feedback to Reasoning: LLM-Assisted Molecular Optimization with Domain Feedback and Historical Reasoning (2026.findings-acl)

Challenge:	Existing approaches to self-reflection rely on heuristic prompting or unidirectional reasoning traces.
Approach:	They propose a structured reflection method that transforms the "from error to repair" process into a first-class, controllable, and trainable action.
Outcome:	The proposed method improves multi-turn tool-call success rates and error recovery while reducing redundant calls.

Copied to clipboard

Wenhan Gao, Xiran Fan, Chin-Chia Michael Yeh, Jiarui Sun, Yuzhong Chen, Menghai Pan, Mahashweta Das, Yi Liu

Iterative Formalization and Planning in Partially Observable Environments (2026.findings-acl)

Challenge:	Existing methods for molecular optimization do not leverage domain feedback and historical knowledge with reasoning traces and chemical insights.
Approach:	They propose a conversational molecular optimization pipeline that enables LLMs to accumulate and retrieve past actions, rationales, and feedback.
Outcome:	The proposed framework transforms LLMs from passive text generators into agentic experts that learn both actions and reasoning from experience.

Copied to clipboard

Liancheng Gong, Wang Bill Zhu, Jesse Thomason, Li Zhang

Challenge:	Existing methods to formalize an environment into the Planning Domain Definition Language (PDDL) have been shown to improve performance and control.
Approach:	They propose a framework to iteratively formalize, plan, grow, and refine PDDL representations by decomposing the environment and the goal into fully observable episodes.
Outcome:	The proposed framework improves planning success and exhibits robustness against problem complexity compared to end-to-end approaches.

StructKV: Preserving the Structural Skeleton for Scalable Long-Context Inference (2026.findings-acl)

Copied to clipboard

Zhirui Chen, Peiyang Liu, Ling Shao

Correct, Concise and Complete: Multi-stage Training For Adaptive Reasoning (2026.findings-acl)

Challenge:	Existing compression approaches prioritize tokens based on local saliency metrics to decouple prefill computation from decoding memory.
Approach:	They propose a structure-aware KV cache compression framework that prioritizes tokens based on local saliency metrics to decouple prefill computation from decoding memory.
Outcome:	The proposed framework preserves long-range dependencies and retrieval robustness.

Copied to clipboard

Nathanaël Carraz Rakotonirina, Ren Pang, Neha Anna John, Michael Bohlke-Schneider, Momchil Hardalov

At Your Own PACE: A Causal Framework for Evaluating EQ in LLMs (2026.findings-acl)

Challenge:	Large language models (LLMs) increase test-time computation, often in the form of chain-of-thought (CoT) however, reasoning traces can become unnecessarily long, increasing computation costs without improving accuracy and sometimes even degrading performance.
Approach:	They propose a multi-stage efficient reasoning method that combines supervised fine-tuning with reinforcement learning using an adaptive length penalty.
Outcome:	The proposed method reduces response length by an average of 28% for 8B models and 40% for 32B models while incurring only minor performance drops of 1.6 and 2.5 points, respectively.

Copied to clipboard

Lei Lyu, Shengling Wang, Ke Chao, Yichao Wei

Challenge:	Emotional Quotient (EQ) has emerged as a competency for seamless human-AI integration.
Approach:	They propose a framework for a closed-loop EQ evaluation using a PACE taxonomy to define four dimensions of LLM EQ.
Outcome:	The proposed framework achieves high alignment of 89.31% with human preferences while maintaining robust consistency of 83.6%.

Identifying Collective Intelligence Factor in LLM Agent Groups for Generalizable Multi-Agent System Design (2026.findings-acl)

Copied to clipboard

Zhilun Zhou, Zihan Liu, Jiahe Liu, Yihan Wang, Qingyu Shao, Fengli Xu, Depeng Jin, Yong Li

Challenge:	Prior studies have focused on designing customized MAS for specific tasks . a critical research question remains: do LLM agent groups exhibit a form of "general intelligence"
Approach:	They find a Collective Intelligence factor in human groups that captures their general capability.
Outcome:	The proposed model predicts the ACI factor based on the features of LLM agent groups and can improve generalization abilities.

Rethinking Parameter Sharing for LLM Fine-Tuning with Multiple LoRAs (2026.findings-acl)

Copied to clipboard

Hao Ban, Kaiyi Ji

Shorten After You’re Right: Lazy Length Penalties for Reasoning RL (2026.findings-acl)

Challenge:	Existing methods for large language models use parameter-efficient techniques such as Low-Rank Adaptation (LoRA) prior studies suggest that the inner A matrices are highly similar during training and therefore suitable for sharing.
Approach:	They propose an asymmetric multi-LoRA design with multiple A matrices and a single shared B in multi-task fine-tuning.
Outcome:	The proposed methods achieve more balanced performance across tasks with comparable or superior average accuracy relative to existing methods.

Copied to clipboard

Danlong Yuan, Tian Xie, Shaohan Huang, Huishuai Zhang, Zhuocheng Gong, Chong Luo, Furu Wei, Dongyan Zhao

MedDCR: Learning to Design Agentic Workflows for Medical Coding (2026.findings-acl)

Challenge:	Existing shortening methods for long reasoning models rely on additional supervision or multi-stage post-training.
Approach:	They propose a lazy length penalty that imposes length pressure on models without extra training stages.
Outcome:	The proposed method significantly reduces response length without extra training stages while maintaining or improving performance.

Copied to clipboard

Jiyang Zheng, Islam Nassar, Thanh Vu, Xu Zhong, Yang Lin, Tongliang Liu, Long Duong, Yuan-Fang Li

Challenge:	Medical coding is the process of translating unstructured clinical notes into standardized diagnostic and procedural codes.
Approach:	They propose a closed-loop framework that treats workflow design as a learning problem.
Outcome:	The proposed framework outperforms state-of-the-art workflows on benchmark datasets and produces interpretable, adaptable workflows that better reflect real coding practice.

TokenPenalty: Alleviating Attention Sinks and Positional Decay in LVLMs (2026.findings-acl)

Copied to clipboard

Xiaofeng Zhang, Yuanchao Zhu, Qiyan Zhao, Xiaosong Yuan, Jiawei Cao, Xuhang Chen

Beyond Screenshots: Evaluating VLMs’ Understanding of UI Animations (2026.findings-acl)

Challenge:	Multimodal large language models (MLLMs) often hallucinate due to two relevant phenomena: massive activation phenomenon and positional information decay.
Approach:	They propose a token-level intervention strategy that dynamically suppresses irrelevant visual tokens while preserving key contextual signals.
Outcome:	Experiments show that TokenTruth significantly improves factual consistency across MLLMs on standard image understanding benchmarks.

Copied to clipboard

Chen Liang, Xirui Jiang, Naihao Deng, Eytan Adar, Anhong Guo

Challenge:	Recent studies of Vision Language Models (VLMs) for UI understanding have focused primarily on static screenshots, leaving it unclear how well these models handle dynamic UI animations.
Approach:	They evaluate UI animation models' ability to perceive animation effects and interpret animation meaning . they use motion, context, and perceptual cues to probe factors affecting VLM performance .
Outcome:	The proposed model can detect primitive motion, but its interpretation is inconsistent . the proposed model is based on 300 annotated UI animation videos .

MemWeaver: Weaving Hybrid Memories for Traceable Long-Horizon Agentic Reasoning (2026.findings-acl)

Copied to clipboard

Juexiang Ye, Xue Li, Yang Xinyu, Chengkai Huang, Lanshun Nie, Lina Yao, Dechen Zhan

From Mimesis to Metamorphosis: Evolving VLM Judges via In-Context Comparing and Knowledge Internalization (2026.findings-acl)

Challenge:	Existing methods rely on unstructured retrieval or coarse abstractions, which lead to temporal conflicts, brittle reasoning, and limited traceability.
Approach:	They propose a unified memory framework that consolidates long-term agent experiences into three interconnected components that combine structured knowledge and evidence to construct compact yet information-dense contexts for reasoning.
Outcome:	The proposed framework significantly improves multi-hop and temporal reasoning accuracy while reducing input context length by over 95% compared to long-context baselines.

Copied to clipboard

Juntuo Wang, Yuming Qiao, Yifan Yang, Lunxi Yuan, Liang Luo, Dan Meng

Generative Text-to-Image Retrieval via Hierarchical Identifiers and Semantic Internalization (2026.findings-acl)

Challenge:	Existing approaches to subjective assessment are inconsistent and inconsistent due to inconsistent scales and inherent preference biases.
Approach:	They propose a framework that operationalizes subjective assessment as comparative analysis and internalizes it via Language Buttons.
Outcome:	The proposed framework achieves state-of-the-art performance across multiple benchmarks and is scale-steerable.

Copied to clipboard

Jie Huang, Junjie Wang, Xin Liao, Ziyou Jiang, Wenshuo Wang, Shoubin Li, Qing Wang

Bridging Reasoning and Action: Hybrid LLM–RL Framework for Efficient Cross-Domain Task-Oriented Dialogue (2026.findings-acl)

Challenge:	Existing text-to-image retrieval methods suffer from limited semantic discriminability, alignment bias, and closed-set restrictions.
Approach:	They propose a framework for semantic internalization for Generative Multimodal Alignment . they construct multi-granularity hierarchical identifiers to ensure unique, semantically consistent image representations .
Outcome:	The proposed framework outperforms state-of-the-art frameworks on Flickr30K and MS-COCO datasets . it achieves average Recall@1, Recall @5, and Recall_10 improvements of 10.65%, 8.50%, and 7.00% .

Copied to clipboard

Yangyang Zhao, Linfan Dai, Li Cai, Bowen Xing, Libo Qin

JudgeAgent: Beyond Static Benchmarks for Knowledge-Driven and Dynamic LLM Evaluation (2026.findings-acl)

Challenge:	Existing methods to solve cross-domain task-oriented dialogues are brittle when cross- domain constraints are not directly grounded in surface text or require commonsense inference.
Approach:	They propose a framework that makes LLM-derived constraint reasoning usable for RL.
Outcome:	Experiments show that the proposed framework outperforms single-model baselines on long-horizon tasks.

Copied to clipboard

Zhichao Shi, Xuhui Jiang, Chengjin Xu, Cangli Yao, Shengjie Ma, Yinghan Shen, Zixuan Li, Jian Guo, Yuanzhuo Wang

Challenge:	Current evaluation methods for large language models rely on static benchmarks . limited knowledge coverage and fixed difficulties hinder the targeted optimizations resulting in superficial evaluations of LLMs - a problem that has been addressed by JudgeAgent .
Approach:	They propose a knowledge-driven and dynamic evaluation framework for large language models . judgeAgent leverages LLM agents equipped with context graphs to traverse knowledge structures .
Outcome:	The proposed framework can achieve comprehensive evaluations and facilitate effective model iterations.

Retrieve Only Relevant Tables Whether Few or Many: Adaptive Table Retrieval Method (2026.findings-acl)

Copied to clipboard

Taehee Kim, Seungbin Yang, Jihwan Kim, Jaegul Choo

Code-Switching Information Retrieval: Benchmarks, Analysis, and the Limits of Current Retrievers (2026.findings-acl)

Challenge:	Existing methods for retrieving relevant tables from databases are limited by the number of tables required.
Approach:	They propose an adaptive table retrieval method that adjusts the number of tables retrieved according to the requirements of each query.
Outcome:	Experiments on Spider, BIRD, and Spider 2.0 show that the proposed method improves performance and retrieval and downstream tasks.

Copied to clipboard

Qingcheng Zeng, Yuheng Lu, Zeqi Zhou, Heli Qi, Puxuan Yu, Fuheng Zhao, Hitomi Yanaka, Weihao Xuan, Naoto Yokoya

WebUncertainty: Dual-Level Uncertainty Driven Planning and Reasoning For Autonomous Web Agent (2026.findings-acl)

Challenge:	a new study examines the performance of code-switching IR in monolingual contexts . code-witching is a pervasive linguistic phenomenon in global communication .
Approach:	They propose a benchmark to evaluate code-switching IR in monolingual contexts . they propose CS-MTEB, which measures performance declines of up to 27% .
Outcome:	The proposed benchmark shows that code-switching performance is degraded by 27% . the proposed benchmark is based on a dataset of mixed-language queries .

Copied to clipboard

Lingfeng Zhang, Yongan Sun, Jinpeng Hu, Hui Ma, Ying Yang, Kuien Liu, Zenglin Shi, Meng Wang

Multi-Faceted Self-Consistent Preference Alignment for Query Rewriting in Conversational Search (2026.findings-acl)

Challenge:	Existing web agents struggle with complex tasks due to rigid planning strategies and hallucination-prone reasoning.
Approach:	They propose a task-uncertainty-driven Adaptive Planning Mechanism that adaptively selects planning modes to navigate unknown environments.
Outcome:	The proposed framework performs better on the WebArena and WebVoyager benchmarks than existing frameworks.

Copied to clipboard

Zhiyu Cao, Peifeng Li, Qiaoming Zhu

CogNet-KG: Empowering Tutoring Dialogues with a Cognitively-Structured Knowledge Graph for STEM Learning (2026.findings-acl)

Challenge:	Existing approaches to rewrite ambiguous queries ignore feedback from query rewriting, passage retrieval and response generation in the rewritten process.
Approach:	They propose to construct self-consistent preference alignment data to generate more diverse rewritten queries.
Outcome:	The proposed method is effective in both in- and out-of-distribution scenarios.

Copied to clipboard

Ding Yu, Yu Lu, Tengju Li, Shasha Xiong, Shengquan Yu

From 128K to 4M: Efficient Training of Ultra-Long Context Large Language Models (2026.findings-acl)

Challenge:	Educational knowledge graphs are a critical component of intelligent tutoring systems that are structured around cognitive principles and provide support for interactive teaching.
Approach:	They propose a cognitively-structured large-scale knowledge graph for STEM learning that models nearly 500 core concepts across five subjects with various cognitively grounded relations corresponding to specific learning objectives.
Outcome:	The proposed model generates a high-quality tutoring dialogue dataset CogDialogue-QA and a specialized tutorial LLM that internalizes this structured pedagogical reasoning.

Copied to clipboard

Chejian Xu, Wei Ping, Peng Xu, Zihan Liu, Boxin Wang, Mohammad Shoeybi, Bo Li, Bryan Catanzaro

A Few Bad Apples Spoil the Bunch: Preventing Global Entropy Collapse Driven by a Small Set of Tokens in LLM Reasoning (2026.findings-acl)

Challenge:	Long-context capabilities are essential for document and video understanding, in-contact learning, and inference-time scaling.
Approach:	They propose an efficient training recipe for building ultra-long context LLMs from aligned instruct model, pushing the boundaries of context lengths from 128K to 1M, 2M, and 4M tokens.
Outcome:	The proposed model extends the context window while maintaining short context capabilities while maintaining the performance of the existing model.

Copied to clipboard

Jaeeun Jang, Hansle Lee, Sangmin Kim

Efficient Training for Cross-lingual Speech Language Models (2026.findings-acl)

Challenge:	Reinforcement learning with verifiable rewards and Reinforced Learning from internal feedback fail to benefit from test-time compute due to entropy collapse and the resulting loss of reasoning diversity.
Approach:	They propose a strategy that assigns each generated token a redistribution score and applies selective KL regularization to only the top 5% of tokens under this score.
Outcome:	The proposed model improves on both RLVR and RLIF models on math reasoning benchmarks, showing that targeted entropy control at a vanishingly small subset of tokens is sufficient to sustain reasoning diversity and effective test-time scaling.

Copied to clipboard

Yan Zhou, Qingkai Fang, Yun Hong, Yang Feng

Empirical Analysis of Task Mixture Effects in Small-scale Instruction Tuning: A Statistical Approach (2026.findings-acl)

Challenge:	Currently, large language models (LLMs) focus on the text modality, making speech modeling difficult.
Approach:	They propose a cross-lingual speech language model that trains on discrete speech tokens to achieve cross-modal and cross-linguistic alignment through continual pre-training.
Outcome:	The proposed method achieves cross-modal and cross-lingual alignment through continual pre-training.

Copied to clipboard

Jeesu Jung, Sangkeun Jung

ResearchBench: Benchmarking LLMs in Scientific Discovery via Inspiration-Based Task Decomposition (2026.findings-acl)

Challenge:	Recent studies have shown that instruction tuning can significantly vary depending on the task and format diversity of training data.
Approach:	They conduct experiments on unlabeled instruction corpora to identify effective mixtures.
Outcome:	The proposed model can achieve up to 5.7 speedup in training with 1,000 curated examples.

Copied to clipboard

Yujie Liu, Zonglin Yang, Tong Xie, Jinjie Ni, Ben Gao, Yuqiang Li, Shixiang Tang, Wanli Ouyang, Erik Cambria, Dongzhan Zhou

GraphLoRA: Structure-Aware Low-Rank Adaptation for Large Language Model Recommendation (2026.findings-acl)

Challenge:	Large language models have shown potential in assisting scientific research, yet their ability to discover high-quality research hypotheses remains unexamined due to the lack of a dedicated benchmark.
Approach:	They propose a benchmark for evaluating large language models on a sufficient set of scientific discovery sub-tasks.
Outcome:	The proposed framework extracts critical components from papers across 12 disciplines with expert validation confirming its accuracy.

Copied to clipboard

Lin Mu, Guoji Wang, Li Ni, Lei Sang, Zhize Wu, Peiquan Jin, Yiwen Zhang

Spotlight and Shadow: Attention-Guided Dual-Anchor Introspective Decoding for MLLM Hallucination Mitigation (2026.findings-acl)

Challenge:	Existing methods for translating collaborative information into textual prompts or injecting pre-trained embeddings into the LLM treat structural information as static input and fail to capture high-order relational dependencies.
Approach:	They propose a framework that generalizes low-rank adaptation from independent to structure-aware propagation by embedding a trainable graph message-passing network within the low-ranked adaptation pathway.
Outcome:	Experiments on multiple benchmarks show that GraphLoRA outperforms state-of-the-art recommendation methods and achieves superior generalization.

Copied to clipboard

Yebo Wu, Han Jin, Zhijiang Guo, Li Li

StreamMeCo: Long-Term Agent Memory Compression for Efficient Streaming Video Understanding (2026.findings-acl)

Challenge:	Multimodal Large Language Models (MLLMs) excel in tasks ranging from image captioning to complex reasoning.
Approach:	They propose a contrastive decoding framework that dynamically calibrates each token generation by mining the model’s internal perceptual discrepancies.
Outcome:	The proposed framework mitigates hallucination while enhancing general reasoning capabilities.

Copied to clipboard

Junxi Wang, Te Sun, Jiayi Zhu, Junxian Li, Haowen Xu, Zichen Wen, Xuming Hu, Zhiyu li, Linfeng Zhang

Reflective RAG: Self-Evaluation Driven Strategy Optimization in Agentic Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	StreamMeCo is an efficient Stream Agent Memory Compression framework for video understanding.
Approach:	They propose an efficient Stream Agent Memory Compression framework that evicts redundant memory nodes and introduces a time-decay memory retrieval mechanism to mitigate performance degradation.
Outcome:	The proposed framework achieves 1.87 speedup in memory retrieval while delivering an average accuracy improvement of 1.0% on three challenging benchmark datasets.

Copied to clipboard

Haiyan Wu, Chenchen Wang, Chaoqun Sun, Chengxiong Lu, Yan-Hong Chen, Zhiqiang Zhang, Xiaoqing Feng

Meta-Reasoner: Dynamic Guidance for Optimized Inference-time Reasoning in Large Language Models (2026.findings-acl)

Challenge:	Recent agentic RAG systems lack the capacity to evaluate the utility of retrieved information, leading to brittle reasoning and suboptimal decision-making.
Approach:	They propose a framework that integrates self-evaluation to dynamically optimize retrieval and generation strategy.
Outcome:	The proposed framework outperforms strong agentic baselines on five knowledge-intensive QA benchmarks and improves training stability and generalization to multi-hop reasoning tasks.

Copied to clipboard

Yuan Sui, Yufei He, Tri Cao, Sophia Simeng Han, Yulin Chen, Bryan Hooi

Polynomial Expansion Rank Adaptation: Enhancing Low-Rank Fine-Tuning with High-Order Interactions (2026.findings-acl)

Challenge:	Recent advances on prompting and post-training have enabled LLMs to perform step-wise reasoning tasks, but they tend to explore unproductive solution paths without effective backtracking or strategy adjustment.
Approach:	They propose a framework that empowers LLMs to “think about how to think” and dynamically adapts reasoning strategies in real-time.
Outcome:	The proposed framework outperforms previous SOTA methods by 9-12% in accuracy while reducing inference time by 28-35% under the same compute budget.

Copied to clipboard

Wenhao Zhang, Lin Mu, Li Ni, Peiquan Jin, Yiwen Zhang

Efficient Test-Time Scaling via Temporal Reasoning Aggregation (2026.findings-acl)

Challenge:	Low-rank adaptation (LoRA) is a widely used strategy for efficient fine-tuning of large language models, but its strictly linear structure limits expressive capacity.
Approach:	They propose a method that introduces structured polynomial expansion directly into the low-rank factor space.
Outcome:	The proposed method outperforms state-of-the-art methods across diverse benchmarks.

Copied to clipboard

Jiakun Li, Xingwei He, Kefan Li, Hongzheng Chai, Hongyue Yu, Yuan Yuan

A Survey on MLLM-based Visually Rich Document Understanding: Methods, Challenges, and Emerging Trends (2026.findings-acl)

Challenge:	Existing dynamic early-exit methods rely on single-step confidence signals . existing approaches are unreliable for detecting reasoning convergence in multi-step settings .
Approach:	They propose a training-free framework for efficient test-time scaling that determines when to terminate reasoning based on temporal aggregation of multi-step evidence rather than instantaneous signals.
Outcome:	Experiments show that TRACE reduces reasoning token usage by 25% on average while maintaining accuracy within 1–2% of full-length reasoning.

Copied to clipboard

Yihao Ding, Siwen Luo, Yue Dai, Yanbei Jiang, Zechuan Li, Qiang Sun, Geoffrey Martin, Wei Liu, Yifan Peng

Detecting Sensitive Personal Information in Japanese Pre-Training Corpora for Large Language Models (2026.findings-acl)

Challenge:	Visually Rich Document Understanding (VRDU) frameworks are a key area of research . early approaches to VRDU relied on manually crafted rules and domain-specific heuristics . conventional deep learning approaches do not integrate the diverse modalities in documents .
Approach:	They review recent advances in MLLM-based Visually Rich Document Understanding (VRDU) their findings highlight emerging trends and promising research directions .
Outcome:	The proposed frameworks are scalable, reliable, and adaptable, the authors argue . their findings highlight emerging trends and promising research directions .

Copied to clipboard

Rei Minamoto, Yusuke Oda, Daisuke Kawahara

Agentic Episodic Control (2026.findings-acl)

Challenge:	Large-scale pre-training corpora are essential for large language models, but if such content remains unfiltered, there is a risk that LLMs may memorize it and leak it through their outputs.
Approach:	They construct a Japanese text corpora dataset and train machine learning models to detect SCPI in text.
Outcome:	The proposed classifier can detect information related to SCPI in Japanese text.

Copied to clipboard

Xidong Yang, Wenhao Li, Junjie Sheng, Yun Hua, Haosheng Chen, Chuyun Shen, Xiangfeng Wang

TrajGuard: Streaming Hidden-state Trajectory Detection for Decoding-time Jailbreak Defense (2026.findings-acl)

Challenge:	Existing methods for reinforcement learning (RL) are limited by poor data efficiency and weak generalization.
Approach:	They propose a novel architecture that integrates large language models into episodic RL.
Outcome:	The proposed architecture achieves 2–6 higher data efficiency than baselines and is the only method to solve complex tasks like UnlockLocal with over 90% success.

Copied to clipboard

Cheng Liu, Xiaolei Liu, Xingyu Li, Bangzhou Xin, Kangyi Ding

DeepPrune: Parallel Scaling without Inter-trace Redundancy (2026.findings-acl)

Challenge:	Existing jailbreak defense paradigms rely on static detection of prompts, outputs, or internal states . hidden states in critical layers during decoding carry stronger and more stable risk signals .
Approach:	They propose a decoding-time defense framework that aggregates hidden-state trajectories via a sliding window to quantify risk in real time.
Outcome:	The proposed framework achieves an average defense rate of 95% in 12 jailbreak attacks and open-source LLMs.

Copied to clipboard

Shangqing Tu, Yaxuan Li, Yushi Bai, Lei Hou, Juanzi Li

Contrastive Decoding Mitigates Score Range Bias in LLM-as-a-Judge (2026.findings-acl)

Challenge:	Parallel scaling is a powerful paradigm to enhance reasoning capabilities in large language models.
Approach:	They propose a framework that enables efficient parallel scaling through dynamic pruning.
Outcome:	The proposed framework achieves token reductions of 65.73% to 88.50% compared to consensus sampling while maintaining competitive accuracy within 3.4 percentage points.

Copied to clipboard

Yoshinari Fujinuma

Challenge:	Large Language Models (LLMs) are used as evaluators but the reliability of the outcomes remains a challenge.
Approach:	They show that LLM judge outputs are highly sensitive to pre-defined score ranges and that similar biases exist among models from the same family.
Outcome:	The contrastive decoding of LLM judge outputs achieves 11.7% relative improvement in Spearman correlation with human judgments, averaged across score ranges.

ContextCheck: Sentence-Level Faithfulness Verification with Context-Aware Disambiguation (2026.findings-acl)

Copied to clipboard

Yueqin Yin, Yaxi Li, Xin Liu, Xun Wang, Kaiqiang Song, Simin Ma, Shujian Liu, Sathish Reddy Indurthi, Haoyun Deng, Pengcheng He, Mingyuan Zhou, Song Wang

Challenge:	Large language models often hallucinate, producing content that is factually incorrect or not grounded in the sources.
Approach:	They propose a framework for sentence-level faithfulness verification with context-aware disambiguation.
Outcome:	The proposed framework improves Macro F1 by over 10 points compared to baselines on three context-dependent datasets.

Less Noise, More Voice: Reinforcement Learning for Reasoning via Instruction Purification (2026.findings-acl)

Copied to clipboard

Yiju Guo, Tianyi Hu, Zexu Sun, Yankai Lin

MAST: A Multi-View Alignment Strategy for Optimal Transport-Based Contrastive Clustering of Short Text (2026.findings-acl)

Challenge:	Experimental results show that LENS outperforms GRPO in delivering higher performance and faster convergence.
Approach:	They propose a framework that purifies prompts by identifying and removing interference tokens and then transfers successful rollouts to supervise policy optimization on original noisy prompts.
Outcome:	The proposed framework outperforms GRPO in the real-world, with a 3.88% gain and speedup.

Copied to clipboard

Zijian Zheng, Yonghe Lu

AICA-Bench: Holistically Examining the Capabilities of VLMs in Affective Image Content Analysis (2026.findings-acl)

Challenge:	Short text clustering has gained significant prominence due to its ubiquity in real-world applications.
Approach:	They propose a multi-view alignment strategy with transport-based clustering that integrates structural views to capture multi-granularity semantic features.
Outcome:	Experiments show that MAST outperforms state-of-the-art methods on benchmark datasets.

Copied to clipboard

Dong She, Xianrong Yao, Liqun Chen, Jinghe Yu, Yang Gao, Zhanpeng Jin

You Only Need One Single Token to Refine Safety Alignment (2026.findings-acl)

Challenge:	Recent studies have focused on factual correctness, semantic grounding, visual reasoning, or multimodal large language models.
Approach:	They propose a benchmark to assess AICA, which integrates perception, reasoning, and generation into a unified framework.
Outcome:	The proposed framework corrects intensity errors and significantly enhances descriptive depth.

Copied to clipboard

Wenqian Yu, Shuo Chen, Zhijiang Li, Zhipeng Wang, Jindong Gu

CrisPrune: Combining Contextual Relevance and Intrinsic Saliency for Efficient Visual Token Pruning in MLLMs (2026.findings-acl)

Challenge:	Excessive safety can lead to over-refusal, where models reject harmful-looking yet benign queries, severely limiting utility.
Approach:	They propose a lightweight training-based approach that reshapes the distributions of harmful and benign samples within the model’s decision space by using a single-token prefix.
Outcome:	The proposed approach can distinguish between harmful and benign samples while keeping the model frozen.

Copied to clipboard

Ziniu Liu, Shuheng Zhou, Mingqing Liu, Hao Deng, Huijia Zhu

Beyond Topology: Generative Node Importance Estimation via Structure-Guided Semantic Reasoning (2026.findings-acl)

Challenge:	Existing methods for visual token pruning compromise the integrity of visual understanding in pursuit of efficiency.
Approach:	They propose a model-agnostic method that integrates visual saliency and text relevance to reconcile efficiency with understanding by integrating visual salions and text relevant.
Outcome:	The proposed method outperforms state-of-the-art methods on LLaVA-NeXT . it achieves 13 decrease in FLOPs while maintaining 97% of original performance .

Copied to clipboard

Kuofei Fang, Siyan Wu, Yu Guo, Bin Wu

AGTAO: Robust and Stabilized LLM Unlearning via Adversarial Gating Training with Adaptive Orthogonality (2026.findings-acl)

Challenge:	Existing methods for estimating node importance are limited and rely on topological aggregation.
Approach:	They propose a generative reasoning framework that leverages Large Language Models to generate precise importance scores for entities in Knowledge Graphs.
Outcome:	Extensive experiments show that the proposed framework outperforms existing methods and is generalized across domains.

Copied to clipboard

Pengyu Li, Lingling Zhang, Zhitao Gao, Yanrui Wu, Yuxuan Dong, Huan Liu, Bifan Wei, Jun Liu

LLMs Meet Isolation Kernel: Lightweight, Learning-free Binary Embeddings for Fast Retrieval (2026.findings-acl)

Challenge:	Large Language Models (LLMs) unintentionally memorize sensitive data, posing privacy and security risks.
Approach:	They propose a framework that reconciles unlearning efficacy and utility preservation by using a latent-space gating mechanism to simulate internal recovery attempts.
Outcome:	The proposed framework achieves superior trade-off between unlearning efficacy and model utility.

Copied to clipboard

Zhibo Zhang, Yang Xu, Kai Ming Ting, Cam-Tu Nguyen

BOLT: Benchmarking Open-World Learning for Text Classification (2026.findings-acl)

Challenge:	Large language models (LLMs) embeddings are typically high-dimensional, leading to substantial storage and retrieval overhead.
Approach:	They propose a learning-free method that transforms an LLM embedding into a binary embeddable using Isolation Kernel (IKE).
Outcome:	The proposed method performs 16.7 faster retrieval and 16 lower memory usage than the original LLM embeddings while maintaining comparable accuracy.

Copied to clipboard

Chuan Qin, Xi Chen, Jinpeng Li, Hengshu Zhu

MobileBench-OL: A Comprehensive Chinese Benchmark for Evaluating Mobile GUI Agents in Real-World Environment (2026.findings-acl)

Challenge:	Existing benchmarks focus on out-of-distribution (OOD) detection while overlooking broader challenges such as the discovery of novel categories.
Approach:	They propose a unified Benchmark and evaluation toolkit supporting Open-world learning for text classification.
Outcome:	The proposed methods overfit training distributions and struggle to generalize to unseen classes.

Copied to clipboard

Qinzhuo Wu, Zhizhuo Yang, Hanhao Li, Pengzhi Gao, Wei Liu, Jian Luan

CondenseFlow: Scalable Latent Space Collaboration via Semantic Compression for Multi-Agent Systems (2026.findings-acl)

Challenge:	Recent advances in mobile Graphical User Interface (GUI) agents highlight the growing need for comprehensive evaluation benchmarks.
Approach:	They propose an online benchmark with 1080 tasks from 80 Chinese apps that measures task execution, complex reasoning, noise robustness and auto-eval framework with a reset mechanism.
Outcome:	The proposed benchmark measures task execution, complex reasoning, and noise robustness of agents by including 5 subsets, which set multiple evaluation dimensions.

Copied to clipboard

Xiaoyu Chen, Fengge Wu, Zhao Junsuo, Yun Fan

Challenge:	Full-state latent communication in LLMs suffers from memory overhead scaling linearly with collaboration rounds.
Approach:	They propose a lightweight module that uses learnable semantic probes to compress KV caches into fixed-size representations.
Outcome:	The proposed module reduces KV cache memory by over 99% and inference latency by approximately 20% on seven benchmarks spanning six models . it outperforms text-based methods by 1.7 percentage points on average across all configurations while outperforming existing methods by 1.7%.

MAXS: Meta-Adaptive Exploration with LLM Agents (2026.findings-acl)

Copied to clipboard

Jian Zhang, Zhiyuan Wang, Zhangqi Wang, Yu He, Haoran Luo, li Yuan, Lingling Zhang, Rui Mao, Qika Lin, Jun Liu

Leveraging Human and Machine Preferences for Zero-shot Detection of AI-Generated Text (2026.findings-acl)

Challenge:	Existing methods for inference are often myopic and have divergent reasoning paths . a meta-adaptive reasoning framework is proposed to improve the efficiency of LLM agents .
Approach:	They propose a meta-adaptive reasoning framework that integrates tool execution and reasoning planning.
Outcome:	The proposed framework outperforms existing methods in performance and inference efficiency.

Copied to clipboard

Lei Jiang, Desheng Wu, Xiaolong Zheng, Cuicui Luo

MelTrim: Coarse-to-Fine Data Pruning for Speech Classification (2026.findings-acl)

Challenge:	Recent advances in large language models have enabled generated texts to closely mimic human writing, posing significant challenges to the detection of AI-generated content.
Approach:	They propose a human-machine prediction discrepancy adapter for AI-generated text detection . they use a joint fine-tuning strategy and a discrepany-aware reweighting mechanism .
Outcome:	The proposed framework improves the detection performance of five representative models under various evaluation scenarios.

Copied to clipboard

Shaobo Wang, Tianle Niu, Xuan Ouyang, Xintong Li, Zhengkun Ge, Yue Min, Xiaoqian Liu, Hankun Wang, Linfeng Zhang

MINED: Probing and Updating with Multimodal Time-Sensitive Knowledge for Large Multimodal Models (2026.findings-acl)

Challenge:	Unlike image or text classification, speech classification tasks are particularly challenging due to the difficulty in capturing the acoustic, semantic, and contextual representations.
Approach:	They propose a dataset pruning method that coarsely filters redundant samples using DBSCAN clustering on Mel-Frequency Cepstral Coefficients (MFCC) features.
Outcome:	The proposed method achieves 49.5% improvement in WA on the MEAD dataset and 41.9% reduction in EER on speaker identification tasks.

Copied to clipboard

Kailin Jiang, Ning Jiang, Yuntao Du, Yuchen Ren, Yuchen Li, Yifan Gao, Jinhe Bi, Yunpu Ma, Bin Li, Lei Liu, Qing Li

Beyond Output Confidence: Epistemic-Aware Hallucination Detection with Answer-Level Signals (2026.findings-acl)

Challenge:	Existing benchmarks for Large Multimodal Models (LMMs) are constrained by static representations, inadequately evaluating their ability to understand time-sensitive knowledge.
Approach:	They propose a benchmark containing 2,104 time-sensitive knowledge samples spanning six knowledge types to evaluate temporal awareness along 6 key dimensions and 11 challenging tasks.
Outcome:	The proposed benchmark measures temporal awareness along 6 key dimensions and 11 tasks, while most open-source LMMs still lack time understanding ability.

Copied to clipboard

Jieran Li, Xiuyuan Hu, Yang Zhao, Dongbiao Sun, Hao Zhang

PerfCoder: Large Language Models for Interpretable Code Performance Optimization (2026.findings-acl)

Challenge:	Existing methods for detecting hallucinations are confounded by epistemic uncertainty and cannot distinguish genuine uncertainty from fabricated content.
Approach:	They propose a model-agnostic metric that captures epistemic boundary deviations by measuring answer-level stability across multiple stochastic forward passes.
Outcome:	The proposed metric outperforms strong uncertainty-only baselines and can be used to detect hallucinations on open-domain question answering, dialogue generation, and code completion.

Copied to clipboard

Jiuding Yang, Shengyao Lu, Hongxuan Liu, Shayan Shirahmad Gale Bagi, Zahra Fazel, Tomasz Czajkowski, Di Niu

MIPIC: Matryoshka Representation Learning via Self-Distilled Intra-Relational and Progressive Information Chaining (2026.findings-acl)

Challenge:	Large language models (LLMs) have advanced automatic code generation, but their ability to produce high-performance code remains limited.
Approach:	They propose a family of large language models that generate performance-enhanced code through interpretable and customized optimization strategies.
Outcome:	The proposed model outperforms existing models on the PIE code performance benchmark and produces interpretable feedback that can guide larger LLMs in a planner–optimizer workflow.

Copied to clipboard

Phung Gia Huy, Hai An Vu, Minh-Phuc Truong, Thang Duc Tran, Linh Ngo Van, Thanh Hong Nguyen, Trung Le

Discovering the Gems in Early Layers: Accelerating Long-Context LLMs with 1000x Input Token Reduction (2026.findings-acl)

Challenge:	Existing approaches to train dense representations require explicit coordination of how information is arranged across embedding dimensionality and model depth.
Approach:	They propose a framework that trains Matryoshka representations using self-distilled intra-relational alignment and Progressive information chaining.
Outcome:	The proposed framework produces coherent and compact Matryoshka representations with significant performance advantages under low-dimensional models.

Copied to clipboard

Zhenmei Shi, Yifei Ming, Xuan-Phi Nguyen, Yingyu Liang, Shafiq Joty

Challenge:	Large Language Models (LLMs) have demonstrated remarkable capabilities in handling long context inputs, but this comes at the cost of increased computational resources and latency.
Approach:	They propose an algorithm that uses early LLM layers as filters to select and compress input tokens, reducing the context length for subsequent processing.
Outcome:	The proposed method outperforms existing techniques on the Needle in a Haystack task while demonstrating comparable performance on the LongBench challenge.

Self-Reflection Improves Safety of Large Reasoning Models (2026.findings-acl)

Copied to clipboard

Qiang Huang, Wei Zhai, Feng Huang, Dejing Dou

WildGraphBench: Benchmarking GraphRAG with Wild-Source Corpora (2026.findings-acl)

Challenge:	Existing safety alignment methods are shallow and do not address deeper risks and attacks in reasoning processes.
Approach:	They propose a technique that introduces a special Self-Reflection token to enable LRMs to perform self-reflection during generation and recover from harmful outputs.
Outcome:	The proposed approach outperforms the baseline model in terms of safety and helpfulness, and significantly improves model safety without adversarial training.

Copied to clipboard

Pengyu Wang, Benfeng Xu, Licheng Zhang, Shaohan Wang, Mingxuan Du, Chiwei Zhu, Zhendong Mao

CRPS: Curriculum Replay via Progressive Suffixes from Successful Trajectories for Long-Horizon LLM Agents (2026.findings-acl)

Challenge:	Existing benchmarks for Graph-based Retrieval-Augmented Generation (GraphRAG) rely on short, curated passages as external knowledge, failing to adequately evaluate systems in realistic settings involving long contexts and large-scale heterogeneous documents.
Approach:	They propose a benchmark to assess GraphRAG performance in the wild using Wikipedia's unique structure where cohesive narratives are grounded in long and heterogeneous external reference documents.
Outcome:	Experiments with articles across 12 top-level topics show that GraphRAG performs better in the wild than existing methods.

Copied to clipboard

Zijing Zhang, Xiajie Yang

Challenge:	Long-horizon agents trained with sparse terminal rewards experience slow and unstable learning . this is amplified by group-normalized on-policy objectives, which yield degenerate advantages and weak learning signals.
Approach:	They propose a lightweight RL-training strategy that turns terminal successes into within-trajectory curriculum by replaying from successful suffix states.
Outcome:	The proposed strategy outperforms full-episode GRPO and naive experience replay across ALFWorld and WebShop with different foundation models.

D-Artemis: A Deliberative Cognitive Framework for Mobile GUI Multi-Agents (2026.findings-acl)

Copied to clipboard

Hongze Mi, Yibo Feng, WenJie Lu, Yuqi Wang, Jinyuan Li, Song Cao, He Cui, Tengfei Tian, Xuelin Zhang, Haotian Luo, Di Sun, Jun Fang, Hua Chai, Naiqiang Tan, Gang Pan

Challenge:	Graphical User Interface (GUI) agents aim to automate a wide spectrum of human tasks by emulating user interaction.
Approach:	They propose a deliberative framework that leverages a fine-grained tip retrieval mechanism to inform its decision-making process.
Outcome:	The proposed framework achieves SOTA among open-source general models on AndroidWorld and ScreenSpot-V2 . it leverages a fine-grained, app-specific tip retrieval mechanism to inform its decision-making process .

Temporal Flattening in LLM-Generated Text: Comparing Human and LLM Writing Trajectories (2026.findings-acl)

Copied to clipboard

Zhanwei Cao, YeoJin Go, Yifan Hu, Shanu Sushmita

Beyond Superficial Tests: Adversarial Refinement for Reliable Property-Based Testing (2026.findings-acl)

Challenge:	Large language models (LLMs) are increasingly used in daily applications, from content generation to code writing.
Approach:	They construct a longitudinal dataset of 412 human authors and 6,086 documents spanning 2012–2024 and compare them to trajectories generated by three representative LLMs.
Outcome:	The results show that LLMs produce greater lexical diversity but exhibit substantially reduced semantic and cognitive–emotional drift relative to humans.

Copied to clipboard

Xiao Li, Runlin Liu, Zhe Zhang, Xiang Gao, Hailong Sun

FAITH: Factuality Alignment through Integrating Trustworthiness and Honestness (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have demonstrated remarkable proficiency in code generation, yet their application to Property-Based Testing (PBT) remains fraught with a superficiality gap.
Approach:	They propose an agentic framework that hardens software properties through Adversarial Refinement.
Outcome:	a new framework hardens software properties through Adversarial Refinement that detects and fixes bugs in top-tier libraries.

Copied to clipboard

Xiaoning Dong, Chengyan Wu, Yajie Wen, Yu Chen, Yun Xue, Zhang Jing, Wei Xu, Bolei Ma

DRA-GRPO: Your GRPO Needs to Know Diverse Reasoning Paths for Mathematical Reasoning (2026.findings-acl)

Challenge:	Existing approaches to correct factually inaccurate outputs are lacking the semantic richness needed to properly understand its internal states of trustworthiness and honesty.
Approach:	They propose a framework for factuality alignment that integrates natural-language uncertainty signals with external knowledge and computes confidence scores and semantic entropy from LLM outputs.
Outcome:	Extensive experiments on four knowledge-intensive benchmarks show that FAITH improves the factual accuracy and truthfulness of Large Language Models (LLMs).

Copied to clipboard

Xiwen Chen, Wenhui Zhu, Peijie Qiu, Xuanzhao Dong, Hao Wang, Haiyu Wu, Huayu Li, Aris Sotiras, Yalin Wang, Abolfazl Razi

Web Fraud Attacks Against LLM-Driven Multi-Agent Systems (2026.findings-acl)

Challenge:	Existing methods for group-relative policy optimization rely on scalar correctness rewards that are often non-injective with respect to semantic content.
Approach:	They propose a framework that calibrates the reward signal using the semantic density of sampled groups.
Outcome:	The proposed framework outperforms strong baselines on five math benchmarks with 7,000 samples and 55 cost.

Copied to clipboard

Dezhang Kong, Hujin Peng, Yilun Zhang, Lele Zhao, Zhenhua Xu, Shi Lin, Changting Lin, Meng Han

From Query to Logic: Ontology-Driven Multi-Hop Reasoning in LLMs (2026.findings-acl)

Challenge:	Large Language Model (LLM)-driven multi-agent systems (MAS) are rapidly gaining popularity, and its inherent security risks are rapidly becoming a concern.
Approach:	They propose a novel attack manipulating unique structures of web links to deceive MAS by using homoglyph deception, sub-directory nesting, and parameter obfuscation.
Outcome:	The proposed attacks exploit unique structures of web links to deceive MAS . they exhibit significant destructive potential across different MAS architectures .

Copied to clipboard

Haonan Bian, Yutao Qi, Rui Yang, Yuanxi Che, Jiaqian Wang, Heming Xia, Ranran Zhen

AudioStealer: Extracting Audio Prompts via Shapley Value-Guided Query Search (2026.findings-acl)

Challenge:	Large Language Models (LLMs) exhibit limitations in complex multi-hop question answering tasks that necessitate non-linear, structured reasoning.
Approach:	They propose an ontology-driven reasoning and chain framework that combines LLMs’ generative capabilities with the structural benefits of knowledge graphs.
Outcome:	Extensive experiments across a diverse set of models and standard MQA benchmarks demonstrate that the proposed framework achieves competitive performance while producing more interpretable reasoning chains.

Copied to clipboard

Yingbin Jin, Xingjian Du, Hanjun Luo, Zihao Wang, Haibo Hu, XiaoFeng Wang, Xinfeng Li

FaithLens: Detecting and Explaining Faithfulness Hallucination (2026.findings-acl)

Challenge:	prompt stealing is a new form of attack that aims to reconstruct high-value prompts that guide music generation.
Approach:	They propose a method to steal music prompts from audio domains using a black-box attack framework.
Outcome:	The proposed method recovers prompts with high textual consistency to the ground truth while maintaining strong perceptual similarity to the target recordings.

Copied to clipboard

Shuzheng Si, Qingyi Wang, Haozhe Zhao, Yuzhuo Bai, Guanqiao Chen, Kangyang Luo, Gang Chen, Fanchao Qi, Minjia Zhang, Baobao Chang, Maosong Sun

Escaping the Sisyphus Dilemma: Experience Replay for Robust Text-to-Optimization Modeling (2026.findings-acl)

Challenge:	Recent progress in large language models (LLMs) has revolutionized text generation.
Approach:	They propose a faithfulness hallucination detection model that can provide binary predictions and corresponding explanations to improve trustworthiness.
Outcome:	The proposed model outperforms advanced models on 12 diverse tasks.

Copied to clipboard

Wantong Xie, Yinghao Chen, Yi-Xiang Hu, Feng Wu, Jieyang Xu, Sijia Zhang, Xiangyang Li

Toward Automated Robustness Evaluation of Mathematical Reasoning (2026.findings-acl)

Challenge:	Existing retrieval-augmented strategies for large language models fail to capture dynamic reasoning required to resolve execution failures.
Approach:	They propose a framework that implements Experience Replay to transform transient rectification steps into persistent knowledge.
Outcome:	The proposed framework improves model accuracy by 8.45% on complex tasks while reducing token consumption by 28.65% and interaction turns by 25.82%.

Copied to clipboard

Yutao Hou, Zeguan Xiao, Fei Yu, Yihan Jiang, Ma Shuguang, Zhaoqian Dai, Hailiang Huang, Yun Chen, Guanhua Chen

Simulating Crisis Cognition: A Computational Framework for Hypothesis Generation in Crisis Communication (2026.findings-acl)

Challenge:	Existing robustness evaluations rely on hand-crafted templates or a limited set of perturbation rules, resulting in model failure.
Approach:	They propose a framework inspired by software stress testing that generates adversarial variants via a multi-round rewrite-verify loop, ensuring semantic consistency while successfully inducing model failure.
Outcome:	The proposed framework generates adversarial variants dynamically for each LLM, minimizing the risk of data contamination.

Copied to clipboard

Changsen Yuan, Yanghao Zhou, Chong Feng, Ge Shi

Challenge:	Large Language Models (LLMs) have demonstrated remarkable fidelity in simulating social dynamics, yet using them to inform high-stakes crisis policy requires rigorous causal evaluation.
Approach:	They propose a framework that functions as an in-silico hypothesis generator to evaluate communication strategies by coupling real-world telemetry with 1,813 agents.
Outcome:	The proposed framework provides a rigorous testbed for evaluating strategies before human-subject trials.

MedScore: Generalizable Factuality Evaluation of Open-ended Long-form Medical Answers by Domain-adapted Claim Decomposition and Verification (2026.findings-acl)

Copied to clipboard

Heyuan Huang, Alexandra DeLucia, Vijay Murari Tiyyala, Mark Dredze

Challenge:	Existing factuality evaluation pipelines are poor matches for medical domains . existing methods are limited to objective, entity-centric, formulaic texts .
Approach:	They propose a pipeline to decompose medical answers into condition-aware valid facts . they use a decomposition-then-verify approach to evaluate generated text .
Outcome:	The proposed method extracts up to three times as many valid facts as existing methods . the resulting factuality score substantially varies by decomposition method, corpus, and used backbone LLM .

FinSafetyBench: Evaluating LLM Safety in Real-World Financial Scenarios (2026.findings-acl)

Copied to clipboard

Yutao Hou, Yihan Jiang, Yuhan Xie, Jian Yang, Liwen Zhang, Hailiang Huang, Guanhua Chen, Yun Chen

Optimizing Native Sparse Attention with Latent Attention and Local Global Alternating Strategies (2026.findings-acl)

Challenge:	Existing large language models (LLMs) are prone to misuse and misinformation, posing serious compliance risks.
Approach:	They propose a bilingual red-teaming benchmark to test an LLM’s refusal of requests that violate financial compliance.
Outcome:	The proposed benchmark is based on real-world financial crime cases and ethical violations and includes 14 subcategories covering financial crimes and ethical breaches.

Copied to clipboard

Yuxuan Hu, Jianchao Tan, Jiaqi Zhang, Wen Zan, Pingwei Sun, Yifan Lu, Xunliang Cai, Jing Zhang

BRIEF-Pro: Universal Context Compression with Short-to-Long Synthesis for Fast and Accurate Multi-Hop Reasoning (2026.findings-acl)

Challenge:	Existing research has proposed a variety of training-free and post-training methods for selecting critical key-value pairs at each generation step.
Approach:	They propose to use local (sliding-window) and global (compression/selective) attention across layers to enlarge long-context modeling.
Outcome:	Experiments on models from 340M to 1.3B parameters show that the proposed method matches or exceeds full attention and native sparse attention in both common-sense reasoning and long-context understanding tasks.

Copied to clipboard

Jia-Chen Gu, Junyi Zhang, Di Wu, Yuankai Li, Kai-Wei Chang, Nanyun Peng

O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruning (2026.findings-acl)

Challenge:	Experiments show that BRIEF-Pro generates more concise and relevant summaries, enhancing performance across small, large, and proprietary language models.
Approach:	They propose a universal, lightweight compressor that distills relevant evidence from retrieved documents into a concise summary for seamless integration into in-context RAG.
Outcome:	Experiments on four open-domain multi-hop question-answering datasets show that BRIEF-Pro generates more concise and relevant summaries, enhancing performance across small, large, and proprietary language models.

Copied to clipboard

Haotian Luo, Haiying He, Yibo Wang, Shiwei Liu, Wei Li, Xiaochun Cao, Dacheng Tao, Naiqiang Tan, Li Shen

One Agent to Serve All: a Lite-Adaptive Stylized AI Assistant for Millions of Multi-Style Official Accounts (2026.findings-acl)

Challenge:	Recent long-thought reasoning models adopt extended reasoning processes similar to how humans ponder over complex problems.
Approach:	They propose a model that uses RL-style fine-tuning to reduce inference overhead while maintaining accuracy.
Outcome:	The proposed model reduces inference overhead while maintaining accuracy.

Copied to clipboard

Xingyu Fan, Li Feifei, Que Wenhui, Hailong Li

SAVOIR: Learning Social Savoir-Faire via Shapley-based Reward Attribution (2026.findings-acl)

Challenge:	Existing methods for prompting in official accounts are computationally prohibitive and lack contextually grounded responses.
Approach:	They propose a lite-adaptive framework for stylized contextual question answering that scales to millions of official accounts.
Outcome:	The proposed framework can serve large volumes of official accounts with minimal overhead while maintaining stylistic diversity.

Copied to clipboard

Xiachong Feng, Yi Jiang, Xiaocheng Feng, Deyi Yin, Libo Qin, Yangfan Ye, Lei Huang, Weitao Ma, Yuxuan Gu, Chonghan Qin, Bing Qin, Lingpeng Kong

Language-Coupled Reinforcement Learning for Multilingual Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	Existing approaches to improve social intelligence of AI systems employ retrospective attributions and lack theoretical grounding.
Approach:	They propose a framework that uses Shapley values to ensure fair credit distribution with axiomatic guarantees of efficiency, symmetry, and marginality.
Outcome:	The proposed framework matches or exceeds proprietary models including GPT-4o and Claude-3.5-Sonnet.

Copied to clipboard

Rui Qi, Fengran Mo, Yufeng Chen, Xue Zhang, Shuo Wang, Hongliang Li, Xu Jinan, Meng Jiang, Jian-Yun Nie, Kaiyu Huang

Article and Comment Frames Shape the Quality of Online Comments (2026.findings-acl)

Challenge:	Existing approaches to multilingual retrieval-augmented generation (MRAG) use a single-turn retrieval and subsequent optimization to acquire and integrate beneficial external knowledge from multilingual collections.
Approach:	They propose a multilingual search-augmented reinforcement learning framework that integrates a language-coupled Group Relative Policy Optimization into the policy and reward models.
Outcome:	The proposed framework achieves competitive performance and is appropriate for various practical scenarios such as constrained training data and retrieval over collections encompassing a large number of languages.

Copied to clipboard

Matteo Guida, Yulia Otmakhova, Eduard Hovy, Lea Frermann

LAD: Learning Advantage Distribution for Reasoning (2026.findings-acl)

Challenge:	Recent work has focused on predicting comment toxicity or quality, but it ignores audience reactions.
Approach:	They propose a frame-aware system to mitigate unhealthy discourse . they analysed 1M comments across 2.7K news articles .
Outcome:	The proposed system can mitigate unhealthy discourses by analyzing 1M comments across 2.7K news articles.

Copied to clipboard

Wendi Li, Sharon Li

RealMem: Benchmarking LLMs in Real-World Memory-Driven Interaction (2026.findings-acl)

Challenge:	Recent reinforcement learning objectives focus on maximizing expected rewards, limiting diversity and exploration.
Approach:	They propose a distribution-matching framework that replaces advantage maximization with learning the advantage-induced distribution.
Outcome:	Experiments on math and code reasoning tasks show that LAD improves accuracy and diversity.

Copied to clipboard

Haonan Bian, Zhiyuan Yao, Sen Hu, Zishan Xu, Shaolei Zhang, Yifu Guo, Ziliang Yang, Xueran Han, Huacan Wang, Ronghao Chen

AFMRL: Attribute-Enhanced Fine-Grained Multi-Modal Representation Learning in E-commerce (2026.findings-acl)

Challenge:	Existing benchmarks focus on casual conversation or task-oriented dialogue, failing to capture “long-term project-oriented” interactions where agents must track evolving goals.
Approach:	They propose a benchmark that simulates the dynamic evolution of memory in real-world projects.
Outcome:	The proposed benchmarks simulate the dynamic evolution of memory in real-world projects.

Copied to clipboard

Biao Zhang, Lixin Chen, Bin Zhang, Wang Zongwei, Tong Liu, Bo Zheng

SAFER: A Controllable Safeguard for LLMs against Backdoor Attacks (2026.findings-acl)

Challenge:	Multimodal representation is crucial for E-commerce tasks such as identical product retrieval.
Approach:	They propose an approach which leverages the generative power of Multimodal Large Language Models to extract key attributes from product images and text and enhances representation learning through a two-stage training framework.
Outcome:	The proposed model achieves state-of-the-art on multiple downstream retrieval tasks, validating the effectiveness of harnessing generative models to advance fine-grained representation learning.

Copied to clipboard

Zirui Hu, Zheng Zhang, Yingjie Wang, Dacheng Tao

RIFT: Repurposing Negative Samples via Reward-Informed Fine-Tuning (2026.findings-acl)

Challenge:	Existing inference-time defenses lack explicit control over false acceptance rate (FAR) existing inference time defenses aim to mitigate poisoned inputs but lack explicit FAR control .
Approach:	They propose a framework that provides explicit control over false acceptance rate without prior knowledge of backdoor samples.
Outcome:	The proposed framework outperforms existing inference-time defenses on three benchmark datasets . it provides explicit and provable control over false acceptance rate without prior knowledge of backdoor samples .

Copied to clipboard

Zehua Liu, Shuqi Liu, Tao Zhong, Mingxuan Yuan

LEPO: Latent Reasoning Policy Optimization for Large Language Models (2026.findings-acl)

Challenge:	Reward Informed Fine-Tuning (RIFT) is an effective and robust alternative to expensive expert data for LLM alignment.
Approach:	They propose a reward-informed fine-tuning framework that utilizes all self-generated samples to learn from both positive and negative trajectories.
Outcome:	The proposed framework outperforms both RFT and Supervised Fine-Tuning (SFT) on mathematical benchmarks.

Copied to clipboard

Yuyan Zhou, Jiarui Yu, Hande Dong, Zhezheng Hao, Hong Wang, Jianqing Zhang, Qiang Lin

You Never Know a Person, You Only Know Their Defenses: Detecting Levels of Psychological Defense Mechanisms in Supportive Conversations (2026.findings-acl)

Challenge:	Existing latent reasoning methods that use chain of thought (CoT) are limited to selecting one discrete token at each reasoning step, which potentially induces information loss.
Approach:	They propose a framework that injects controllable stochasticity into latent reasoning via Gumbel-Softmax, restoring LLMs' exploratory capacity and enhancing their compatibility with Reinforcement Learning (RL).
Outcome:	The proposed framework preserves richer information for more comprehensive reasoning and is compatible with Reinforcement Learning (RL).

Copied to clipboard

Hongbin Na, Zimu Wang, Zhaoming Chen, Peilin Zhou, Yining Hua, Grace Ziqi Zhou, Haiyang Zhang, Tao Shen, Wei Wang, John Torous, Shaoxiong Ji, Ling Chen

Beyond Length Scaling: Synergizing Breadth and Depth for Generative Reward Models (2026.findings-acl)

Challenge:	Psychological defenses are strategies people use to manage distress.
Approach:	They propose a dialogue corpus with help seeker utterances labeled for defense level and a DMRS Co-Pilot pipeline that provides evidence-based pre-annotations.
Outcome:	The proposed framework reduces annotation time by 24.0% in a counterbalanced study.

Copied to clipboard

Qiyuan Zhang, Yufei Wang, Tianhe Wu, Can Xu, Qingfeng Sun, Kai Zheng, Xue Liu, Chen Ma

Pushing the Limits of LLM Tool Calling via Experiential Knowledge Integration and Activation (2026.findings-acl)

Challenge:	Recent advances in Generative Reward Models have demonstrated that scaling the length of Chain-of-Thought reasoning enhances reliability of evaluation.
Approach:	They propose a framework that reconfigures raw rationales into structured Breadth-CoT and Depth-Co T through a modular synthesis pipeline.
Outcome:	The proposed framework surpasses open-source RMs by an average of 8.2%.

Copied to clipboard

Yupu Hao, Zhuoran Jin, Huanxuan Liao, Kang Liu, Jun Zhao

Efficient Transformer Parameter Reuse via Zero-Token Mechanism (2026.findings-acl)

Challenge:	Existing approaches treat tool use as a problem of prompt design, API documents specification, or supervised or unsupervised alignment.
Approach:	They propose a knowledge-augmented tool execution framework that integrates experiential knowledge with reasoning-width-expanded inference and knowledge-aware training.
Outcome:	The proposed framework improves on BFCL-V3 and AppWorld on three model scales.

Copied to clipboard

Guanghao Li, Wenhao Jiang, Li Shen, Ming Tang, Chun Yuan

x1: Learning to Think Adaptively Across Languages and Cultures (2026.findings-acl)

Challenge:	Existing approaches to scaling up parameter counts are impractical for users with limited computational resources.
Approach:	They propose a decoupled parameter cycling strategy that employs a head-tail decoupling strategy to decouple the first (head) and last (tail) layers from the parameter cycling process.
Outcome:	The proposed approach achieves superior performance under strict parameter constraints and significantly reduces computational overhead via early exits.

Copied to clipboard

Yangfan Ye, Xiaocheng Feng, Xiachong Feng, Yichong Huang, Zekun Yuan, Lei Huang, Weitao Ma, Qichen Hong, Yunfei Lu, Dandan Tu, Bing Qin

Towards Bridging the Reward-Generation Gap in Direct Alignment Algorithms (2026.findings-acl)

Challenge:	Existing large language models (LLMs) ignore this diversity by reasoning in a single dominant language.
Approach:	They propose a family of reasoning models that can adaptively reason in an advantageous language on a per-instance basis.
Outcome:	The proposed model can reason in a single dominant language on a per-instance basis.

Copied to clipboard

Zeguan Xiao, Yun Chen, Jian Yang, Guanhua Chen, Ke Tang

Challenge:	Large language models (LLMs) have demonstrated remarkable capabilities across a wide range of tasks, including instruction following, mathematical problem solving, and coding generation.
Approach:	They propose a method that truncates both preferred and dispreferred responses to match the shorter one’s length.
Outcome:	The proposed approach improves over standard implementations and achieves 11.8 points in AlpacaEval 2 and overall improvements across downstream tasks.

FABLE: Fine-grained Fact Anchoring for Unstructured Model Editing (2026.findings-acl)

Copied to clipboard

Peng Wang, Biyu Zhou, Xuehai Tang, Jizhong Han, Songlin Hu

Challenge:	Existing methods for model editing memorize text holistically without reliable fine-grained fact access.
Approach:	They propose a hierarchical framework that decouples fine-grained fact injection from holistic text generation.
Outcome:	The proposed framework significantly improves fine-grained question answering while maintaining state-of-the-art holistic editing performance.

SOP-Maze: Evaluating Large Language Models on Complicated Business Standard Operating Procedures (2026.findings-acl)

Copied to clipboard

Jiaming Wang, Zhe Tang, Zehao Jin, Hefei Chen, Yilin Jin, Peng Ding, Xiaoyu Li, Xuezhi Cao

MalURLBench: A Benchmark Evaluating Agents’ Vulnerabilities When Processing Web URLs (2026.findings-acl)

Challenge:	Large language models (LLMs) are widely deployed as domain-specific agents, but evaluation of their capabilities in such contexts has not been fully explored.
Approach:	They propose a benchmark to evaluate LLMs' ability to follow instructions and make decisions in real-world scenarios.
Outcome:	The proposed benchmark is constructed from real-world business data and adapted into 23 complex SOP scenarios.

Copied to clipboard

Dezhang Kong, Zhuxi Wu, Shiqi Liu, ZhiCheng Tan, Kuichen Lu, Minghao Li, Qichen Liu, Shengyu Chu, Zhenhua Xu, Xuan Liu, Meng Han

Representation-Guided Parameter-Efficient LLM Unlearning (2026.findings-acl)

Challenge:	Existing models struggle to detect elaborately disguised malicious URLs, despite their ability to process malicious URL's.
Approach:	They propose a benchmark to evaluate LLMs’ vulnerabilities to malicious URLs and a lightweight defense module to mitigate the vulnerability.
Outcome:	The proposed framework analyzes 61,845 attack instances spanning 10 real-world scenarios and 7 categories of real malicious websites.

Copied to clipboard

Zeguan Xiao, Lang Mo, Yun Chen, Lei Yang, Jiehui Zhao, Lili Yang, Guanhua Chen

LPO: Towards Accurate GUI Agent Interaction via Location Preference Optimization (2026.findings-acl)

Challenge:	Existing methods to unlearning large language models often memorize sensitive or harmful information, but they struggle with the forget-retain trade-off due to the polysemantic nature of LLMs parameters.
Approach:	They propose a representation-guided low-rank unlearning approach that leverages the geometric properties of representation spaces to achieve robust and precise unlearning.
Outcome:	The proposed approach outperforms state-of-the-art models on TOFU and WMDP benchmarks while maintaining higher model utility.

Copied to clipboard

Jiaqi Tang, Yu Xia, Yi-Feng Wu, Yuwei Hu, Chen Yuhui, Qing-Guo Chen, Xiaogang Xu, Xiangyu Wu, Hao LU, Yanqing Ma, Shiyin Lu, Qifeng Chen

Challenge:	Existing strategies for spatial localization are limited due to their limited capacity to perceive positional data.
Approach:	They propose a location-based approach that leverages locational data to optimize interaction preferences.
Outcome:	The proposed approach achieves SOTA results across offline benchmarks and real-world evaluations.

Mistake Notebook Learning: Batch-Clustered Failures for Training-Free Agent Adaptation (2026.findings-acl)

Copied to clipboard

Xuanbo Su, Yingfang Zhang, Hao Luo, Xiaoteng Liu, Leo Huang

Large-Scale Multimodal Knowledge Graph about Classical Chinese Poetry: Fine-grained Method and Comprehensive Evaluation (2026.findings-acl)

Challenge:	Mistake Notebook Learning (MNL) is a new memory framework for large language model agents . it allows agents to distill shared error patterns into structured "mistake notes"
Approach:	They propose a new memory framework that enables agents to self-curate generalizable guidance from batch-clustered failures.
Outcome:	The proposed framework achieves competitive performance compared to existing memory mechanisms.

Copied to clipboard

Shuo Wang, Qing Zhu, Yang Xiao, Minglong Lei

MultiCodeAttack: Iterative Jailbreak Attacking on LLMs with Multi-Code Prompt Injection (2026.findings-acl)

Challenge:	Existing studies on classical Chinese poetry are limited by modality constraints, dataset size, or the level of refinement.
Approach:	They propose to construct a large-scale and fine-grained multimodal knowledge graph of classical Chinese poetry using an informative ontology graph and a text-image alignment method.
Outcome:	The proposed method collects knowledge about classical Chinese poetry from ontology graphs and performs four tasks that demonstrate its comprehensiveness and high quality.

Copied to clipboard

Weifeng Sun, Meng Yan, Zhou Yang, Yuchen Chen, Song Sun, David Lo

SAMem: State-Aware Memory as a Fine-Grained Memory for LLM Agents in Decision-Making (2026.findings-acl)

Challenge:	Existing approaches to jailbreak rely on fixed template design and a single programming language . however, existing approaches do not consider language diversity or adaptive template evolution .
Approach:	They propose a structured jailbreak framework that explores and optimizes multi-language code templates.
Outcome:	The proposed framework outperforms existing jailbreak baselines and produces higher harmful outputs than baseline methods.

Copied to clipboard

Tong Wang, Pei Xu, Shiyue Cao, Likun Yang, Daipeng Li, Jianbin Jiao, Kaiqi Huang

Learning to Translate by Translating: Stabilizing the Dual Loop via Semantic-Aware Self-Evolution (2026.findings-acl)

Challenge:	Existing experiential memory approaches rely on task-level memory, but this lacks the situational alignment required for complex multi-step decision-making.
Approach:	They propose a new fine-grained memory paradigm that aligns memory retrieval with the current state instead of storing and reusing globally shared experiences.
Outcome:	Experiments on complex decision-making benchmarks show that the proposed state-aware memory outperforms existing experiential memory approaches and significantly improves task-solving efficiency.

Copied to clipboard

Kui Liu, Mingming Yin, Zuoli Tang, Zihao Li, Chilin Fu, Xiaolu Zhang, Jun Zhou, Lixin Zou, Chenliang Li

How to Allocate, How to Learn? Dynamic Rollout Allocation and Advantage Modulation for Policy Optimization (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have been successful in machine translation, but lack of high-quality parallel corpora and cost constrain scalability.
Approach:	They propose an LLM-driven dual-learning framework that enables autonomous translation . they employ a robust semantic-aware reward function that balances adequacy with reconstruction fidelity .
Outcome:	The proposed model outperforms larger models on benchmarks and achieves parity with state-of-the-art supervised baselines on mainstream benchmarks.

Copied to clipboard

Yangyi Fang, Jiaye Lin, Xiaoliang Fu, Cong Qin, Haolin Shi, Chaowen Hu, Lu Pan, Ke Zeng, Xunliang Cai

Meeseeks: A Feedback-Driven, Iterative Self-Correction Benchmark evaluating LLMs’ Instruction Following Capability (2026.findings-acl)

Challenge:	Existing methods for reinforcement learning with verifiable rewards are limited by the complexity of the problem and the complexity.
Approach:	They propose a theoretically-grounded dual-pronged optimization framework for reinforcement learning with verifiable rewards that compensates for gradient attenuation of high-confidence correct actions while utilizing entropy changes as computable indicators to stabilize excessive update magnitudes.
Outcome:	The proposed framework compensates for gradient attenuation of high-confidence correct actions while utilizing entropy changes as computable indicators to stabilize excessive update magnitudes.

Copied to clipboard

Jiaming Wang, Yunke Zhao, Peng Ding, Jun Kuang, Yibin Shen, Zhe Tang, Yilin Jin, ZongYu Wang, Xiaoyu Li, Xuezhi Cao

HSUGA: LLM-Enhanced Recommendation with Hierarchical Semantic Understanding and Group-Aware Alignment (2026.findings-acl)

Challenge:	Existing models lack the ability to adhere to instructions, resulting in suboptimal performance.
Approach:	They propose an automated iterative instruction-following benchmark with integrated feedback mechanism.
Outcome:	The proposed benchmark identifies erroneous components in model responses and provides feedback accurately.

Copied to clipboard

Guorui Li, Dugang Liu, Lei Li, Xing Tang, Zhong Ming

AgenticEval: Toward Agentic and Self-Evolving Safety Evaluation of Large Language Models (2026.findings-acl)

Challenge:	Existing methods for enhancing sequential recommendation use long interaction sequences, but they lack the ability to extract user preferences from long sequences.
Approach:	They propose a plugin that integrates LLMs to infer user preferences from interaction sequences.
Outcome:	The proposed algorithms improve user semantic embedding extraction and utilization on three benchmark datasets.

Copied to clipboard

Yixu Wang, Xin Wang, Yang Yao, Xinyuan Li, Xibang Yang, Yan Teng, Xingjun Ma, Yingchun Wang

AlphaEdit+: Model Editing in the Presence of Conflicting and Inconsistent Knowledge (2026.findings-acl)

Challenge:	Existing static safety evaluation methods are ill-equipped to address dynamic nature of AI risks and evolving regulations, creating a critical safety gap.
Approach:	They propose a new paradigm of agentic safety evaluation reframing evaluation as a continuous and self-evolving process rather than a one-time audit.
Outcome:	The proposed framework shows a consistent decline in model safety as the evaluation hardens.

Copied to clipboard

Qing Liu, Jianhao Zhang, Ou Wu, Michael Ng, Yi Du

Challenge:	Existing methods for knowledge editing struggle with knowledge conflicts and inconsistencies.
Approach:	They propose a new method for knowledge editing that relaxes null-space constraints and introduces a weighting scheme to mitigate conflicts between new and historical knowledge.
Outcome:	The proposed method outperforms existing methods on challenging datasets and outperformed existing methods.

Spatiotemporal Sycophancy: Negation-Based Gaslighting in Video Large Language Models (2026.findings-acl)

Copied to clipboard

Ziyao Tang, Pengkun Jiao, Bin Zhu, Huiyan Qi, Jingjing Chen, Yu-Gang Jiang

FreeChunker: A Cross-Granularity Chunking Framework (2026.findings-acl)

Challenge:	Existing Vid-LLMs lack robust mechanisms for maintaining grounded spatiotemporal beliefs under conversational feedback.
Approach:	They propose a negation-based gaslighting evaluation framework and introduce a benchmark to investigate spatiotemporal sycophancy.
Outcome:	The proposed framework evaluates state-of-the-art Vid-LLMs across video understanding tasks.

Copied to clipboard

Zhang Wenxuan, Yuan-Hao Jiang, Yang Cao, Yonghe Wu

Invocation Refiner: A Plug-and-Play Module for Rectifying LLM Tool Invocations (2026.findings-acl)

Challenge:	Existing chunking paradigms rely on static boundary identification, limiting performance . Existing methods rely only on static knowledge, resulting in hallucinated content .
Approach:	They propose a Cross-Granularity Encoding Framework that treats sentences as atomic units and shifts from static chunk segmentation to flexible retrieval supporting arbitrary sentence combinations.
Outcome:	The proposed framework avoids the computational overhead required for semantic boundary detection and enhances adaptability to complex queries.

Copied to clipboard

Qirui Jiao, Dian Jiao, Nan Du, Ying Shen, Liang Lin

GoViG: Goal-Conditioned Visual Navigation Instruction Generation via Multimodal Reasoning (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have shown remarkable capabilities in Tool-Integrated Reasoning (TIR) however, the practical application is often hindered by frequent errors in tool invocations, such as incorrect tool names, invalid parameters, wrong tool-call order, or malformed invocation formats.
Approach:	They propose a specialized post-processing module that performs independent reasoning on the input of a frozen upstream LLM and an advanced RL algorithm to improve the tool-use reliability of base LLMs.
Outcome:	The proposed module improves task completion rates and invocation accuracy over the raw outputs of various upstream LLMs on a diverse set of tool-use and reasoning benchmarks.

Copied to clipboard

Fengyi Wu, Yifei Dong, Yilong Dai, Guangyu Chen, Qifeng Wu, Huiting Huang, Hang Wang, Qi Dai, Alexander G Hauptmann, Zhi-Qi Cheng

Temporal Contrastive Decoding: A Training-Free Method for Large Audio-Language Models (2026.findings-acl)

Challenge:	Current methods for instruction generation depend on privileged inputs such as semantic maps, landmark annotations, and panoramic views.
Approach:	They propose a task that generates coherent navigation instructions from egocentric visual observations.
Outcome:	The proposed task generates coherent navigation instructions from egocentric visual data . the proposed task improves performance over state-of-the-art methods in BLEU-4 and CIDEr scores .

Copied to clipboard

Yanda Li, Yuhan Liu, Zirui Song, Yunchao Wei, Martin Takáč, Salem Lahlou

When Internalization Fails: Finding Better Targets for Reasoning Compression (2026.findings-acl)

Challenge:	Large audio-language models (LALMs) can exhibit a temporal smoothing bias . unified decoders can produce less specific audio-grounded outputs .
Approach:	They propose a temporally blurred slow-path view that is re-encoded by a token-level logit update.
Outcome:	Experiments on MMAU and AIR-Bench show consistent improvements on strong unified LALMs.

Copied to clipboard

Mourad Heddaya, Manley Roberts, Rohan Wadhawan, Chenhao Tan

PatentMind: A Multi-Aspect Reasoning Graph for Patent Similarity Evaluation (2026.findings-acl)

Challenge:	Reasoning language models generate long reasoning traces that increase latency and cost.
Approach:	They compare three approaches to shorten reasoning traces by inference-time truncation . they use Implicit Chain-of-Thought-style curricula that progressively shorten the teacher trace .
Outcome:	The proposed methods work well on GSM8K and multiplication tasks.

Copied to clipboard

Yongmin Yoo, Qiongkai Xu, Longbing Cao

MemSearcher: Iterative Memory Integration for Search Agent via End-to-End Reinforcement Learning (2026.findings-acl)

Challenge:	Existing methods for patent similarity evaluation lack the multifaceted structure of patent documents . patent documents pose significant challenges due to specialized domain knowledge, intricate legal language, and complex structural formats.
Approach:	They propose a framework that performs patent similarity evaluation through a Multi-Aspect Reasoning Graph.
Outcome:	The proposed framework outperforms embedding-based, patent-specific, and prompt engineering benchmarks in evaluating patent similarity with expert annotations.

Copied to clipboard

Qianhao Yuan, Jie Lou, Zichao Li, Jiawei Chen, Yaojie Lu, Hongyu Lin, Le Sun, Debing Zhang, Xianpei Han

CoPA: Benchmarking Personalized Question Answering with Data-Informed Cognitive Factors (2026.findings-acl)

Challenge:	Recent LLM-based search agents often concatenate the full interaction history into the context, producing long and noisy inputs and increasing compute cost and memory overhead.
Approach:	They propose an agent framework that maintains a compact memory during multi-turn interactions.
Outcome:	The proposed framework outperforms strong history-concatenation (ReAct-style) baselines on a range of public datasets while maintaining nearly constant token counts across multi-turn interactions.

Copied to clipboard

Hang Su, Zequn Liu, Chen Hu, Xuesong Lu, Yingce Xia, Liu Zhen

RAP-ID: Mechanistic Prompt Injection Detection via Impostor Behavior Analysis (2026.findings-acl)

Challenge:	Existing LLMs rely on surface-level similarity or manual heuristics to evaluate personalization . Existing evaluation protocols for personalization are lacking sufficient data-driven validation.
Approach:	They propose a benchmark to assess personalization by mining CIPDs to quantify individual preferences.
Outcome:	The proposed benchmark provides a more comprehensive and discriminative standard than generic metrics.

Copied to clipboard

Yuchen Yang, Lei Peng, Yujie He, Yang yu, Zhongxin Wu, Yanlei Shi

When Safety Alignment Fails to Generalize: Probing with Language Game Jailbreaks (2026.findings-acl)

Challenge:	Existing defenses rely on externally deployed guardrail models or response inspection . current defenses depend on external guardrails or response inspecting .
Approach:	They propose a mechanistic, train-free detection framework that operates exclusively on internal state dynamics during the initial forward pass.
Outcome:	The proposed framework achieves competitive performance with significant overall improvements compared to heuristic methods.

Copied to clipboard

Zewen Long, Yu Peng, Fangming Dong, Congyi Li, Xingmao Guan, Shu Wu, Kai Chen

Self-EmoQ: Plutchik-Guided Value-based Planning to Drive Streaming Emotional TTS (2026.findings-acl)

Challenge:	Existing safety alignment methods rely on fixed or narrow transformation schemes to generalize . existing methods based on fixed and narrow transformations are often inadequate .
Approach:	They propose a framework for discovering and refining language game-based jailbreaks to probe alignment generalization.
Outcome:	The proposed framework allows controlled exploration of alignment behavior across closely related linguistic variants.

Copied to clipboard

Yue Zhao, Hongyan Li, Yong Chen, Luo Ji

Scaling is Not All You Need: Clinical-Oriented Reinforcement Learning Makes Parameter-Efficient Clinical Reasoning (2026.findings-acl)

Challenge:	Existing systems lack a self-emotion determination mechanism to drive the streaming text-to-speech (TTS) synthesis.
Approach:	They propose an emotion-planning framework that determines the emotion prior to the textual generation, grounding the downstream emotional TTS in a streaming manner.
Outcome:	The proposed framework outperforms baselines on DailyDialog, EmoryNLP, IMEOCAP, and MELD on emotional alignment, contextual coherence, and expressive fluency.

Copied to clipboard

Chi Liu, Yan Shu, Mengzhuo Chen, Hongming Piao, Zhijian Duan, Derek Li, Bryan Dai

Do We Need Distinct Representations for Every Speech Token? Unveiling and Exploiting Redundancy in Large Speech Language Models (2026.findings-acl)

Challenge:	Large language models are increasingly used in medicine, but expert-level clinical reasoning remains a high-complexity, high-stakes frontier.
Approach:	They propose to train clinical reasoning models using a Reasoning-Oriented Data Strategy based on topological synthesis and CoT cold-start.
Outcome:	The proposed pipeline outperforms existing models and outperformed the strongest open-source alternatives up to 671B in MedXpertQA.

Copied to clipboard

Bajian Xiang, Tingwei Guo, Xuan Chen, Yang Han

Table-as-Search: Agentic Information Seeking is Table Completion (2026.findings-acl)

Challenge:	Large Speech Language Models (LSLMs) typically operate at high token rates to ensure acoustic fidelity, yet this results in sequence lengths that exceed the underlying semantic content, incurring prohibitive inference costs.
Approach:	They propose a token-based token merging mechanism that uses a training-free token pooling mechanism to reduce prefilling FLOPs by 27.48% while maintaining competitive accuracy.
Outcome:	The proposed method reduces prefilling FLOPs by 27.48% while maintaining competitive accuracy.

Copied to clipboard

Tian Lan, Felix Henry, Bin Zhu, Qianghuai Jia, Junyang Ren, Qihang PU, Haijun Li, Longyue Wang, Zhao Xu, Weihua Luo

Minos: A Multimodal Evaluation Model for Bidirectional Generation Between Image and Text (2026.findings-acl)

Challenge:	Current Information Seeking (InfoSeeking) agents struggle to maintain focus and coherence during long-horizon exploration, as tracking search states within one plain-text context is inherently fragile.
Approach:	They propose a structured planning framework that reformulates the InfoSeeking task as a Table Completion task.
Outcome:	The proposed framework outperforms state-of-the-art frameworks across three kinds of benchmarks, including multi-agent framework and commercial systems.

Copied to clipboard

Junzhe Zhang, Huixuan Zhang, Xinyu Hu, Li Lin, Mingqi Gao, Shi Qiu, Xiaojun Wan

Global Context or Local Detail? Adaptive Visual Grounding for Hallucination Mitigation (2026.findings-acl)

Challenge:	Existing evaluation models struggle to achieve consistent performance across image-to-text (I2T) and text-to image (T2I) tasks.
Approach:	They construct a multimodal evaluation model using a large multimodal dataset and rigorous quality control strategies to train it.
Outcome:	The proposed model achieves state-of-the-art evaluation performance across 16 out-of domain datasets covering both I2T and T2I tasks among all open-source multimodal evaluation models and remain competitive with closed-source models.

Copied to clipboard

Yubo Jiang, Xin Yang, Abudukelimu Wuerkaixi, Zheming Yuan, Xuxin Cheng, Cao Liu, Ke Zeng, Fengying Xie, Zhiguo Jiang, Haopeng Zhang

FinToolSyn: A forward synthesis Framework for Financial Tool-Use Dialogue Data with Dynamic Tool Retrieval (2026.findings-acl)

Challenge:	Large vision–language models suffer from object-existence hallucinations when multi-step deliberation decouples from visual evidence.
Approach:	They propose a framework that allocates visual computation by uncertainty . they propose highlighting retains global context, while selective zoom-in performs local verification.
Outcome:	The proposed framework reduces the complexity of multimodal reasoning by minimizing the operator trade-off.

Copied to clipboard

Caishuang Huang, Yang Qiao, Rongyu Zhang, Junjie Ye, Pu Lu, null Wuwenxi, Meng Zhou, Xiku Du, Qi Zhang, Tao Gui, Xuanjing Huang

From Fake to Real: Mitigating Out-of-Distribution Bias in In-Context Learning via Feedback Supervision from Large Language Models (2026.findings-acl)

Challenge:	Existing data synthesis methods rely on static tools to generate queries . this approach fails to capture the implicit, event-driven nature of real-world needs .
Approach:	They propose a forward synthesis framework to generate high-quality financial dialogues . they construct a repository of 43,066 tools and synthesize over 148k dialogue instances .
Outcome:	Experiments show that models trained on FinToolSyn achieve a 21.06% improvement . the framework is designed to generate high-quality financial dialogues .

Copied to clipboard

Rui Song, Yingji Li, Jian Li, Fausto Giunchiglia, Hao Xu

Revealing and Enhancing Core Visual Regions: Harnessing Internal Attention Dynamics for Hallucination Mitigation in LVLMs (2026.findings-acl)

Challenge:	In-Context Learning (ICL) is one of the most common methods for complex Natural Language Understanding tasks.
Approach:	They propose a method that uses model confidence and perturbation perplexity to enhance the quality of pseudo-labels.
Outcome:	The proposed method reduces OOD biases by avoiding direct use of source data.

Copied to clipboard

Guangtao Lyu, Qi Liu, Chenghao Xu, Jiexi Yan, Muli Yang, Xueting Li, Fen Fang, Cheng Deng

Is Chain-of-Thought Reasoning of LLMs a Mirage? A Data Distribution Lens (2026.findings-acl)

Challenge:	Existing training-free methods are vulnerable to the attention sink phenomenon . Existing methods include contrastive decoding and auxiliary expert models .
Approach:	They propose a training-free attention intervention that constructs a PAD map to identify semantically core visual regions and applies per-head Median Absolute Deviation Scaling to adaptively control the intervention strength.
Outcome:	The proposed intervention improves visual grounding and reduces hallucinations on multiple LVLMs and benchmarks.

Copied to clipboard

Chengshuai Zhao, Zhen Tan, Pingchuan Ma, Dawei Li, Bohan Jiang, Yancheng Wang, Yingzhen Yang, Huan Liu

WindowsWorld: A Process-Centric Benchmark of Autonomous GUI Agents in Professional Cross-Application Environments (2026.findings-acl)

Challenge:	Chain-of-Thought (CoT) prompting has been shown to be effective in eliciting structured reasoning from large language models (LLMs).
Approach:	They propose a data distribution lens to understand when and why CoT reasoning fails . they propose 'data-based' training that trains LLMs from scratch .
Outcome:	The proposed model enables models to generate reasoning trajectories that approximate those observed during training.

Copied to clipboard

Jinchao Li, Yunxin li, Chenrui Zhao, Zhenran Xu, Baotian Hu, Min Zhang

Challenge:	Existing GUI agents perform poorly on multi-application tasks, stalling at early sub-goals.
Approach:	They propose to assess GUI Agents on complex multi-step tasks that mirror real-world professions.
Outcome:	The proposed benchmark contains 181 tasks with an average of 5.0 sub-goals across 17 common desktop applications, of which 78% are inherently multi-application.

EDU-CIRCUIT-HW: Evaluating Multimodal Large Language Models on Real-World University-Level STEM Student Handwritten Solutions (2026.findings-acl)

Copied to clipboard

Weiyu Sun, Liangliang Chen, Yongnuo Cai, Huiru Xie, Yi Zeng, Ying Zhang

What Factors Affect LLMs and RLLMs in Financial Question Answering? (2026.findings-acl)

Challenge:	Multimodal Large Language Models (MLLMs) are a promising tool for traditional education but lack authentic and domain-specific benchmarks to accurately interpret student handwritten solutions.
Approach:	They propose to use MLLMs to interpret unconstrained STEM student handwritten solutions with intertwined mathematical formulas, diagrams, and textual reasoning to bridge this gap.
Outcome:	The proposed model can detect and rectify recognition errors with minimal human intervention on unseen student solutions.

Copied to clipboard

Peng Wang, Xuesi Hu, Jiageng Wu, Yuntao Zou, Qiancheng Zhang, Dagang Li

Superficial Success vs. Internal Breakdown: An Empirical Study of Generalization in Adaptive Multi-Agent Systems (2026.findings-acl)

Challenge:	Recent studies have focused on large language models and reasoning large language model (RLLMs) however, there are few studies that explore what methods can fully unlock the performance of LLMs and RLLM in the financial domain.
Approach:	They examine the effects of prompting methods, agentic frameworks, and multilingual alignment methods on financial question-answering tasks.
Outcome:	The results show that prompting methods and agent frameworks improve LLMs' performance . the authors suggest that these frameworks can be used to enhance LLM performance if they are implemented in financial domains.

Copied to clipboard

Namyeong So, Seokgyu Jang, Taeuk Kim

Challenge:	Adaptive multi-agent systems (MAS) are increasingly adopted as solutions to complex problems.
Approach:	They conduct extensive empirical study on adaptive multi-agent systems . they find they are prone to topological overfitting and exhibit illusory coordination . authors urge prioritization of generalization in MAS development and evaluation .
Outcome:	a new study shows adaptive multi-agent systems are prone to overfitting and lack coordination . the findings highlight the need to prioritize generalization in MAS development .

D2-RAG: Dual-Decision Retrieval-Augmented Generation via Multi-Dimensional Uncertainty and Utility-Aware Decoding (2026.findings-acl)

Copied to clipboard

Jinshuo Zhang, Xiaoding Zhou, Weiyu Zhang, Guoqiang Chen, Ying Lian, Xiaoyang Meng, Yonghe Chen, Hongjiao Guan, Jiasheng Si, Wenpeng Lu

ProFit: Leveraging High-Value Signals in SFT via Probability-Guided Token Selection (2026.findings-acl)

Challenge:	Retrieval-Augmented Generation (RAG) mitigates hallucinations in large language models by incorporating external knowledge.
Approach:	They propose a dual-decision retrieval-augmented generation that integrates multi-dimensional uncertainty estimation to decide whether to retrieve and employs adaptive contrastive decoding to handle retrieved contexts of varying quality.
Outcome:	The proposed model outperforms baselines on four medical question-answering datasets while suppressing interference from noisy contexts.

Copied to clipboard

Tao Liu, Taiqiang Wu, Runming Yang, Shaoning Sun, Junjie Wang, Yujiu Yang

Challenge:	Traditional fine-tuning ignores one-to-many nature of language, leading to overfitting . authors propose a method to fine- tune LLMs by leveraging tokens.
Approach:	They propose a method to fine-tune Large Language Models by leveraging tokens to mask low-probability tokens.
Outcome:	The proposed method outperforms baselines on general reasoning and mathematical benchmarks.

DisastQA: A Comprehensive Benchmark for Evaluating Question Answering in Disaster Management (2026.findings-acl)

Copied to clipboard

Zhitong Chen, Kai Yin, Xiangjue Dong, Chengkai Liu, Xiangpeng Li, Bo Li, Junwei Ma, Yiming Xiao, Ali Mostafavi, James Caverlee

TAGS: A Test-Time Generalist–Specialist Framework with Retrieval-Augmented Reasoning and Verification (2026.findings-acl)

Challenge:	Existing benchmarks for question answering (QA) are lacking in a high-stakes environment.
Approach:	They propose a rigorously verified benchmark of 3,000 expert-annotated questions . they propose 'keypoint-based evaluation protocol' emphasizing factual completeness over verbosity .
Outcome:	Experiments with 20 models reveal substantial divergences from general-purpose models such as MMLU-Pro.

Copied to clipboard

Jianghao Wu, Feilong Tang, Yulong Li, Ming Hu, Haochen Xue, Shoaib Jameel, Zongyuan Ge, Yutong Xie, Imran Razzak

How Retrieved Context Shapes Internal Representations in RAG (2026.findings-acl)

Challenge:	Existing efforts to improve medical question answering performance follow two directions.
Approach:	They propose a framework that combines a generalist with a domain-specific specialist without any model fine-tuning or parameter updates.
Outcome:	The proposed framework boosts GPT-4o accuracy by 13.8%, deepseek-R1 by 16.8%, and improves a vanilla 7B model from 14.1% to 23.9%.

Copied to clipboard

Samuel Yeh, Sharon Li

TabularMath: Understanding Math Reasoning over Tables with Large Language Models (2026.findings-acl)

Challenge:	Retrieval-augmented generation (RAG) is a widely adopted approach for enhancing large language models with external knowledge.
Approach:	They analyze how different types of retrieved documents affect the hidden states of large language models and how these internal representation shifts relate to downstream generation behavior.
Outcome:	The results show that context relevancy and layer-wise processing influence internal representations, providing explanations of LLMs’ output behaviors and insights for RAG system design.

Copied to clipboard

Shi-Yu Tian, Zhi Zhou, Wei Dong, Kun-Yang Yu, Ming Yang, Zi-Jian Cheng, Lan-Zhe Guo, Yu-Feng Li

Seek-and-Solve: Benchmarking MLLMs for Visual Clue-Driven Reasoning in Daily Scenarios (2026.findings-acl)

Challenge:	Mathematical reasoning has long been a key benchmark for evaluating large language models.
Approach:	They propose a framework that transforms math word problems into scalable tabular reasoning tasks.
Outcome:	The proposed framework transforms math word problems into scalable and verified tabular reasoning tasks.

Copied to clipboard

Xiaomin Li, Tala Wang, Zichen Zhong, Ying Zhang, Zirui Zheng, Takashi Isobe, Dezhuang Li, Huchuan Lu, You He, Xu Jia

Structured and Abstractive Reasoning on Multi-modal Relational Knowledge Images (2026.findings-acl)

Challenge:	Existing benchmarks focus on evaluating MLLMs’ pre-existing knowledge or perceptual understanding, often neglecting the critical capability of reasoning.
Approach:	They propose a benchmark designed for visual clue-driven reasoning in daily scenarios that combines rigorous grounding in authentic daily activities and challenging query design that necessitates more than surface-level perception.
Outcome:	The proposed benchmark identifies visual clues and their ability to provide robust reasoning in daily scenarios.

Copied to clipboard

Yichi Zhang, Zhuo Chen, Lingbing Guo, Wen Zhang, Huajun Chen

Code Reffix: A Benchmark for Reflection-Guided Code Repair with Large Language Models (2026.findings-acl)

Challenge:	Existing studies on understanding and reasoning with abstractive information from the visual modality have not explored the use of STructured and Abstractive Reasoning (STAR) on such data.
Approach:	They propose an automatic STAR data engine to synthesize images with MMRK to build multi-modal instructions with reliable chain-of-thought thinking for various STAR tasks.
Outcome:	The proposed framework outperforms GPT-4o in STAR and improves performance across 8 open-source MLLMs.

Copied to clipboard

Zaiyuan Di, Jianting Chen, Yunxiao Yang, Xiaoying Gao, Li Yang, Zhihao Wang, Yang Xiang

ProtoCycle: Reflective Tool-Augmented Planning for Text-Guided Protein Design (2026.findings-acl)

Challenge:	Existing benchmarks focus on the repair generation capability of LLMs, lacking fine-grained evaluation of reflection.
Approach:	They propose a benchmark with oracle reflections and a dual-task protocol to decouple evaluation of reflection from repair.
Outcome:	The proposed benchmarks show that underperforming reflection capabilities remain a bottleneck for code repair.

Copied to clipboard

Yutang Ge, Guojiang Zhao, Sihang Li, Zheng Cheng, Zifeng Zhao, Hanchen Xia, Guolin Ke, Linfeng Zhang, Zhifeng Gao, Yu Guang Wang

Challenge:	Recent deep generative models have already shown encouraging * Equal contribution.
Approach:	They propose to use generic instruction-tuned LLMs as direct text-to-sequence generators to achieve this goal.
Outcome:	Recent studies show that reflection improves sequence quality and alignment while maintaining competitive foldability.

SEC-FinTables: Evaluating Large Language Models for Detecting Logical Inconsistencies on Tabular Data (2026.findings-acl)

Copied to clipboard

Shuyan Ke, Qiong Wu, Hui Li, Liujuan Cao

QueryLink: Leveraging Query-Memory Alignment for Long-Term Reasoning in LLM Agents (2026.findings-acl)

Challenge:	Large language models are increasingly deployed in high-stakes domains where logical inconsistencies are unrecognized.
Approach:	They propose a benchmarking system that decomposes inconsistency detection into granular subtasks and a protocol that decompiles it into subtask.
Outcome:	The proposed model decomposes inconsistencies into subtasks and identifies them in 103,395 real-world and error-injected table instances.

Copied to clipboard

Xuxian Hu, Zhu Teng, Wei Zhang, Ming He, Jianping Fan

AutoPKG: An Automated Framework for Dynamic E-commerce Product-Attribute Knowledge Graph Construction (2026.findings-acl)

Challenge:	Existing approaches to integrating external memory prioritize memory organization while overlooking a critical semantic gap between implicit, intent-driven queries and explicit, narrative-based memories.
Approach:	They propose a framework that leverages Query-Memory Alignment to project both queries and memories into a shared semantic space.
Outcome:	The proposed framework significantly outperforms SOTA methods on the LoCoMo and LongMemEval benchmarks and can be integrated as a plug-and-play component to boost existing vector-based systems like A-MEM.

Copied to clipboard

Pollawat Hongwimol, Haoning Shang, Chutong Wang, Zhichao Wan, Yi Gao, Yuanming Li, Lin Gui, Wenhao Sun, Cheng Yu

Benchmarking LLMs on Authentic Cases from Medical Journals (2026.findings-acl)

Challenge:	Product attribute extraction in e-commerce is bottlenecked by ontologies that are inconsistent, incomplete, and costly to maintain.
Approach:	They propose a multi-agent Large Language Model framework that constructs a Product-attribute Knowledge Graph from multimodal product content.
Outcome:	The proposed framework achieves 0.953 WKE for product types, 0.724 WKEs for attribute keys, and 0.531 edge-level accuracy for value assertions after canonicalization on a large real-world marketplace catalog dataset from Lazada (Alibaba).

Copied to clipboard

Wanlong Liu, Junying Chen, Yunjin Yang, Prayag Tiwari, Wenyu Chen, Benyou Wang

Challenge:	Existing medical benchmarks suffer from performance saturation due to medical exam questions.
Approach:	They evaluate the performance of over 20 open-source and proprietary large language models and benchmark them against human medical experts.
Outcome:	The new benchmark is based on authentic clinical cases sourced from medical journals and implements rigorous human review process to ensure the quality and reliability of the benchmark.

Beyond Quantity: Trajectory Diversity Scaling for Code Agents (2026.findings-acl)

Copied to clipboard

Guhong Chen, Chenghao Sun, Cheng Fu, Qiyao Wang, Zhihong Huang, ChaoPeng Wei, Guangxu Chen, Feiteng Fang, Ahmadreza Argha, Bing Zhao, Xander Xu, Qi Han, Hamid Alinejad-Rokny, Qiang Qu, Binhua Li, Shiwen Ni, Min Yang, HU Wei, Yongbin Li

Rethinking RL Evaluation: Can Benchmarks Truly Reveal Failures of RL Methods? (2026.findings-acl)

Challenge:	Code large language models (LLMs) are becoming tool-interactive agents . quantity-centric scaling exhibits an early bottleneck that underutilizes trajectory data . et al.: a new approach to scale trajectory diversity improves tool-use generalization .
Approach:	They propose a Trajectory Diversity Scaling-based data synthesis framework for code agents that scales performance through diversity rather than raw volume.
Outcome:	Experiments on general tool-use benchmarks and code agent tasks show that TDScaling improves tool-user generalization and inherent coding proficiency.

Copied to clipboard

Zihan Chen, Yiming Zhang, Hengguang Zhou, Zenghui Ding, Yining Sun, Cho-Jui Hsieh

Challenge:	Existing benchmarks for reinforcement learning for large language models do not accurately assess generalization.
Approach:	They propose three core principles for designing more faithful benchmarks: sufficient difficulty, balanced evaluation, and distributional robustness.
Outcome:	The proposed benchmarks do not accurately assess generalization across distribution shifts, difficulty levels, and counterfactual scenarios.

CaPEdit: Capability-Preserving Lifelong Knowledge Editing For Language Models (2026.findings-acl)

Copied to clipboard

Song-Li Wu, Zhaocheng Du, Xianquan Wang, Jingyi Wang

Exploring Layer-wise Information Effectiveness for Post-Training Quantization in Small Language Models (2026.findings-acl)

Challenge:	Existing approaches to incrementally correct factual inaccuracies in large language models (LLMs) but sequential edits can lead to substantial degradation of capabilities.
Approach:	They propose a framework that preserves model capabilities under LKE by decoupling fast-updating factual knowledge from slow-evolving procedural knowledge.
Outcome:	The proposed framework improves capability preservation across all fundamental capabilities by 49.78% and achieves superior editing performance.

Copied to clipboard

He Xiao, Qingyao Yang, Dirui Xie, Wendong XU, Zunhai Su, Runming Yang, Haobo Liu, Wenyong Zhou, Zhengwu Liu, Ngai Wong

ELTLM: Evaluation of Longitudinal Temporal Large Multimodal Models in Clinical Scenarios (2026.findings-acl)

Challenge:	Large language models with billions of parameters are often over-provisioned . smaller models exhibit lower robustness under extreme low-bit quantization .
Approach:	They propose a hardware-native, metric-driven post-training quantization framework that keeps uniform bit-width within each layer while mixing precision across layers.
Outcome:	LieQ reduces large accuracy gap observed for large language models with billions of parameters while preserving standard multiplication kernels.

Copied to clipboard

Gengyuan Hu, Haoxiang Liu, Chenhong Cao, Shilei Tan, Wei Gong

Easy Samples Are All You Need: Self-Evolving LLMs via Data-Efficient Reinforcement Learning (2026.findings-acl)

Challenge:	Existing evaluation benchmarks focus on static evaluation of large multimodal models . existing evaluation paradigms neglect a critical aspect of clinical practice: longitudinal analysis .
Approach:	They propose a temporal perception and reasoning benchmark to assess models' temporal grounding and consistency.
Outcome:	ELTLM features a hierarchical task taxonomy comprising Temporal Perception QA and Temporal Reasoning QA.

Copied to clipboard

Zhiyin Yu, Bo Zhang, Qibin Hou, Zhonghai Wu, Xiao Luo, Lei Bai

Collaborative Multi-Agent Scripts Generation for Enhancing Imperfect-Information Reasoning in Murder Mystery Games (2026.findings-acl)

Challenge:	Experimental results show that EasyRL consistently outperforms state-of-the-art baselines due to the substantial annotation cost and issues such as model collapse or reward hacking.
Approach:	They propose a supervised RL approach with a divide-and-conquer strategy that simulates the human cognitive acquisition curve using easy labeled data.
Outcome:	The proposed approach outperforms state-of-the-art models on mathematical and scientific benchmarks using only 10% of easy labeled data.

Copied to clipboard

Keyang Zhong, Junlin Xie, Hefeng Wu, Haofeng Li, Guanbin Li

A Picture is Worth a Thousand Words? An Empirical Study of Aggregation Strategies for Visual Financial Document Retrieval (2026.findings-acl)

Challenge:	Vision-language models have shown impressive capabilities in perceptual tasks . however, they degrade in complex multi-hop reasoning under multi-player game settings .
Approach:	They propose a multi-agent framework for evaluating and synthesizing role-driven game scripts . they use curated and synthetic datasets to model uncertainty and deception .
Outcome:	The proposed model significantly boosts the performance of VLMs in narrative reasoning and hidden fact extraction under uncertain, adversarial, and socially complex conditions.

Copied to clipboard

Ho Hung Lim, Yi Yang

ToolCPT: Improving Tool Utilization in LLM Agents via Continuous Pre-training (2026.findings-acl)

Challenge:	Visual RAG is an alternative to traditional RAG, but it requires hundreds of patch tokens per document to retrieve and store information.
Approach:	They propose to aggregate documents into a single vector to avoid semantic loss . they find global texture dominance is the root cause of this loss - they say .
Outcome:	The proposed model shows that aggregation obscures semantic changes in financial documents . global texture dominance is the root cause, and the model scales are consistent across models and embeddings.

Copied to clipboard

Yifan Yang, Jinghui Lu, null Evadeng, Ao Yang, Peijie Yu, TingHao YU, Feng Zhang

A Universal Avoidance Method for Diverse Multi-branch Generation (2026.findings-acl)

Challenge:	Current approaches to enhancing tool use for LLM-based agents focus on post-training fine-tuning or test-time context extension.
Approach:	They propose to enhance tool knowledge for LLM-based agents during continuous pre-training . they curate 5.1 million code artifacts from large-scale, high-quality code repositories .
Outcome:	The proposed model outperforms existing methods on out-of-distribution tools on multiple benchmarks.

Copied to clipboard

Kyeongman Park, Minha Jhang, Kyomin Jung

Challenge:	generative models still lack human-level creativity, especially in multi-branch diversity tasks.
Approach:	They propose a model-agnostic and computationally efficient generation strategy that penalizes similarity among previously generated outputs.
Outcome:	The proposed method achieves 1.9 times higher diversity, runs 4.4 times faster, and requires only 1/64 of the FLOPs compared to state-of-the-art methods.

CentaurTA: A Self-Improving Human-Agents Collaboration Framework for Thematic Analysis (2026.findings-acl)

Copied to clipboard

Lei Wang, Min Huang, Eduard Dragut

Challenge:	Existing large language model approaches for qualitative analysis are labor-intensive and costly.
Approach:	They propose an iterative human–agent framework for scalable thematic analysis that integrates structured human feedback with rubric-based evaluation.
Outcome:	The proposed framework improves coding alignment and transparency across multiple datasets, baselines, and LLM families.

Balancing Fidelity and Plasticity: Aligning Mixed-Precision Fine-Tuning with Linguistic Hierarchies (2026.findings-acl)

Copied to clipboard

Changhai Zhou, Shiyang Zhang, Yuhua Zhou, Jun Gao, Qian Qiao, Shichao Weng, Weizhong Zhang, Cheng Jin

From Script to Stage: Automating Experimental Design for Social Simulations with LLMs (2026.findings-acl)

Challenge:	Existing quantization-aware fine-tuning methods decouple weight precision and adapter capacity, overlooking that a layer’s ability to adapt is constrained by the information preserved in its frozen weights.
Approach:	They propose a framework that jointly optimizes per-layer quantization bit-width and LoRA rank.
Outcome:	Experiments on LLaMA and Qwen models show that the proposed framework matches or approaches 16-bit baselines while using substantially less memory.

Copied to clipboard

Yuwei Guo, Zihan Zhao, Xiaowei Liu, Xiangning Yu, Qun Ma, Deyu Zhou, Xiao Xue

Challenge:	Xu et al., 2024): multi-agent simulations based on large language models are a new paradigm for social science research . traditional experimental design relies on interdisciplinary expertise and technical barriers . Xiaoping and Xin eli argue that LLM-driven agents are unreliable for rigorous experimental design due to hallucinations and limited verifiability.
Approach:	They propose a framework for multi-agent experiment design based on script generation . Script Composition, Script Finalization, and Actor Generation are the core phases of the framework .
Outcome:	The proposed framework lowers the barrier for social science experimental design and provides scientifically grounded decision support for policy-making.

Generative Giants, Retrieval Weaklings: Why do Multimodal Large Language Models Fail at Multimodal Retrieval? (2026.findings-acl)

Copied to clipboard

Hengyi Feng, Zeang Sheng, Meiyi Qiang, Yang Li, Wentao Zhang

Learning to Extract Rational Evidence via Reinforcement Learning for Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	Rapid advances in multimodal large language models have revolutionized cross-modality understanding.
Approach:	They propose a method that uses whitening transformations to adjust MLLM representation spaces . they propose ML models that are dominated by textual semantics and visual semantics .
Outcome:	The proposed approach improves zero-shot multimodal retrieval performance without fine-tuning efforts.

Copied to clipboard

Xinping Zhao, Shouzheng Huang, Yan Zhong, Xinshuo Hu, Meishan Zhang, Baotian Hu, Min Zhang

Tree-Notebook: A Context-Aware Agent with Tree Search and Entropy-Aware Data Shadow for Interactive Data Science (2026.findings-acl)

Challenge:	Retrieval-Augmented Generation (RAG) is effective in Large Language Models (LLMs). However, retrieval noises undermine the quality of LLMs’ generation, necessitating the development of denoising mechanisms.
Approach:	They propose a model which integrates reasoning and extracting into one unified trajectory, followed by knowledge token masking to avoid information leakage.
Outcome:	Extensive experiments on five benchmark datasets show the superiority of EviOmni, which provides compact and high-quality evidence, enhances the accuracy of downstream tasks, and supports both traditional and agentic RAG systems.

Copied to clipboard

Junkun Qiu, Min Huang, Qinghai Miao

Resource-Limited Joint Multimodal Sentiment Reasoning and Classification via Chain-of-Thought Enhancement and Distillation (2026.findings-acl)

Challenge:	Experimental results show that Tree-Notebook achieves state-of-the-art (SOTA) performance on InfiAgent-DABench and DSBench.
Approach:	They propose an agentic framework that mimics the iterative cognitive process of human data scientists.
Outcome:	The proposed framework achieves state-of-the-art (SOTA) performance on InfiAgent-DABench and DSBench.

Copied to clipboard

Haonan Shangguan, Xiaocui Yang, Shi Feng, Daling Wang, Yifei Zhang, Feiliang Ren, Ge Yu

Agentic-R: Learning to Retrieve for Agentic Search (2026.findings-acl)

Challenge:	Current approaches for Multimodal Sentiment Analysis (MSA) rely on parameter-heavy LLMs for classification, overlooking multimodal sentiment reasoning generation in resource-limited environments.
Approach:	They propose a multimodal sentiment reasoning distillation model that employs a teacher-assistant-student paradigm to address deployment constraints in resource-limited environments.
Outcome:	The proposed model performs well on a resource-limited JMSRC task with only 3B parameters and shows generalization and interpretability.

Copied to clipboard

Wenhan Liu, Xinyu Ma, Yutao Zhu, Yuchen Li, Daiting Shi, Dawei Yin, Zhicheng Dou

JailMeter: An Evidence-Based Evaluation Framework for Jailbreak Attacks on Large Language Models (2026.findings-acl)

Challenge:	Existing retrievers for single-turn retrieval-augmented generation (RAG) rely on similarity-based retrievers, but similar passages are not always useful for final answer generation.
Approach:	They propose a retrieval-augmented-generation retriever that integrates reasoning with retrieval . they use local query-passage relevance and global answer correctness to measure passage utility .
Outcome:	The proposed retriever outperforms existing retrievers on QA benchmarks on seven single-hop and multi-hop searches.

Copied to clipboard

Qingjia Huang, Jingyu Zhang, Jianguo Wu, Yakai Li, Weijuan Zhang, Yankai Rong, Junyi Yao, Shengzhi Zhang, Xiaoqi Jia

DataSeer: A Manager-Centric Collaborative Multi-Agent Framework with Multi-Branch Reasoning for Automated Insight Discovery (2026.findings-acl)

Challenge:	Currently, evaluation criteria and methods used for jailbreak effectiveness are inconsistent.
Approach:	They propose a framework to measure jailbreak effectiveness using a model that filters out jailbreak noise while preserving the original malicious question.
Outcome:	The proposed framework outperforms existing evaluation methods on a challenging benchmark containing 330 human-labeled, non-rejected jailbreak instances.

Copied to clipboard

Suchen Liu, Yuanfeng Song, Jun Gao, Xing Chen

"Penny Wise, Pixel Foolish": Bypassing Price Constraints in Multimodal Agents via Visual Adversarial Perturbations (2026.findings-acl)

Challenge:	Existing methods for automated insight discovery lack contextual coherence and coverage due to single-path exploration.
Approach:	They propose a Manager-Centric Collaborative Framework that integrates planner and executor . it ensures cross-episode contextual coherence and allows for adaptive sub-goal generation .
Outcome:	The proposed framework outperforms baselines on InsightBench and Inseval.

Copied to clipboard

Jiachen Qian, Zhaolu Kang

Beyond the Individual: Virtualizing Multi-Disciplinary Reasoning for Clinical Intake via Collaborative Agents (2026.findings-acl)

Challenge:	Mobile Agents are a key component of the “Agentic Economy” where they perform high-stakes financial transactions.
Approach:	They propose a systemic vulnerability termed Visual Dominance Hallucination (VDH) VDH exploits the modality gap in CLIP-based encoders via a novel Semantic-Decoupling Loss.
Outcome:	The proposed framework exploits the modality gap in CLIP-based encoders by preserving fidelity.

Copied to clipboard

Huangwei Chen, Wu Li, Junhao Jia, Yining Chen, Xiaotao Pang, Ya-Long Chen, Li Gonghui, Haishuai Wang, Jiajun Bu, Lei Wu

SWE-Swiss: A Multi-Task Fine-Tuning and RL Recipe for High-Performance Issue Resolution (2026.findings-acl)

Challenge:	Initial outpatient consultations are costly and difficult to scale to real-time intake.
Approach:	They propose a synchronous virtual MDT framework that formalizes the consultation state using a structured SOAP representation, separating evidence collection from diagnostic reasoning to improve traceability and bias control.
Outcome:	The proposed framework outperforms state-of-the-art models on ClinicalBench and a real-world RAPID-IPN dataset in documentation quality and consultation capability.

Copied to clipboard

Zhenyu He, Qingping Yang, Wei Shen, Xiaojian Zhong, Kechi Zhang, Chenxin An, Wenlei Shi, Tianle Cai, Di He, Jiaze Chen, Jingjing Xu

BackdoorAgent: A Unified Framework for Backdoor Attacks on LLM-based Agents (2026.findings-acl)

Challenge:	SWE-Swiss-32B demonstrates strong generalization to other common LLM benchmarks.
Approach:	They propose a two-phase training recipe that decomposes issue resolution into three core skills: Localization, Repair, and Unit Test Generation.
Outcome:	The proposed model achieves a 60.2% score on the SWE-bench Verified benchmark and is in the top-tier performance bracket of much larger models.

Copied to clipboard

Yunhao Feng, Yige Li, Yutao Wu, Yingshui Tan, Yanming Guo, Yifan Ding, Kun Zhai, Xingjun Ma, Yu-Gang Jiang

DKME: Rethinking Coupled Knowledge Memory for Lifelong Model Editing of Large Language Models (2026.findings-acl)

Challenge:	Large language model (LLM) agents execute tasks through multi-step workflows that combine planning, memory, and tool use.
Approach:	They propose a modular framework that provides a unified view of backdoor threats in LLM agents.
Outcome:	The proposed framework provides a unified, agent-centric view of backdoor threats in LLM agents.

Copied to clipboard

Guanyu Zheng, Wang Zhenyu, He Tingting, Xv Wang, Haochang Wang, Yaokai Huang, Tiejun Zhao

Self-Awareness before Action: Mitigating Logical Inertia via Proactive Cognitive Awareness (2026.findings-acl)

Challenge:	Existing memory-based editors suffer from catastrophic forgetting as edits accumulate.
Approach:	They propose a method which injects factual updates into large language models without retraining or finetuning into existing memory-based editors.
Outcome:	Experiments on HalluEditBench, CKnowEdit, and WikiDatacounterfact show that the proposed model achieves a more favorable trade-off between editing success and locality compared to baselines while maintaining more stable performance as the edit scale increases.

Copied to clipboard

Fulong Fan, Peilin Liu, Liu FengZhe, Shuyan Yang, Gang Yan

MuBench: Assessment of Multilingual Capabilities of Large Language Models Across 61 Languages (2026.findings-acl)

Challenge:	Existing work on abductive and long-context reasoning reports that current models still lack self-awareness of missing premises.
Approach:	They propose a reasoning framework that introduces self-awareness of missing premises before making the final decision.
Outcome:	SABA achieves best performance on all three difficulty splits of detective puzzle benchmark . a small early mistake can remain uncorrected and can guide later reasoning .

Copied to clipboard

Wenhan Han, Yifan Zhang, Zhixun Chen, null Binbinliu, Mykola Pechenizkiy, Meng Fang, Yin Zheng

PrefRAG: Correcting Semantic Errors in Auto-Formalization for Logical Reasoning with Program Preference RAG (2026.findings-acl)

Challenge:	Existing evaluation datasets lack cross-lingual alignment, leaving assessments of multilingual capabilities fragmented in both language and skill coverage.
Approach:	They propose to use multilingual consistency as a complementary metric to assess performance bottlenecks and guide model improvement.
Outcome:	The proposed model lacks cross-lingual alignment and language coverage gaps between state-of-the-art models.

Copied to clipboard

Yuyin Zhou, Yongmei Liu

LoopCoder: Scaling Code Intelligence via Looped Language Models (2026.findings-acl)

Challenge:	Existing auto-formalization methods for logical reasoning are prone to syntactic and semantic errors.
Approach:	They propose a neuro-symbolic paradigm for logical reasoning based on auto-formalization . they propose 'programme preference retrieval-augmented generation' to detect and repair syntactic and semantic errors.
Outcome:	The proposed approach outperforms baselines on in-distribution and out-of-difference datasets.

Copied to clipboard

Jian Yang, Wei Zhang, Shuyue Guo, Yizhi LI, Linzheng Chai, Zhengmao Ye, Shukai Liu, Yuyang Song, Jiajun Wu, Che Liu, Tianyu Zheng, Siwei Wu, Leo L, Xudong Ma, Chuan Hao, Ran Tao, Yan Xing, Jianzhou Wang, Mingjie Tang, Aishan Liu, Zhoujun Li, Xianglong Liu, Weifeng Lv, Bryan Dai

TempTool-R1: Tool-Augmented Reinforcement Learning for Temporal Knowledge Graph Question Answering (2026.findings-acl)

Challenge:	Large language models have mastered syntax-level code generation, but complex algorithmic reasoning remains a challenge.
Approach:	They propose a recurrent inductive bias that aligns with the recursive nature of programming logic.
Outcome:	The proposed model achieves comparable performance to standard dense models with more parameters.

Copied to clipboard

Zicheng Huang, Yajuan Tong, Xinhui Tu, Tingting He

Lightweight Haar Wavelet Subband Pruning for LLMs (2026.findings-acl)

Challenge:	Existing approaches to temporal knowledge graph question answering struggle with multi-hop reasoning and implicit temporal constraints.
Approach:	They propose a temporal tool-based API capable of transforming implicit temporal cues into executable operations and supervised fine-tuning teaches the model to interweave chain-of-thought reasoning with think-then-tool usage.
Outcome:	The proposed framework outperforms existing methods on three challenging questions.

Copied to clipboard

Jiang Li, Pengfei Cao, Chenxi Zhou, Tian Lan, Xiangdong Su, Kang Liu, Jun Zhao, Guanglai Gao

Challenge:	Large language models (LLMs) have impressive performance but require computational and memory resources.
Approach:	They propose a post-training framework that uses a Haar wavelet transform to prune weights.
Outcome:	The proposed pruning framework reduces pruning time and computational costs by removing less important weights while preserving model architecture.

Prompt-R1: Collaborative Automatic Prompting Framework via End-to-end Reinforcement Learning (2026.findings-acl)

Copied to clipboard

Wenjin Liu, Haoran Luo, Xueyuan Lin, Haoming Liu, Tiesunlong Shen, Jiapu Wang, Rui Mao, Erik Cambria

ProCeedRL: Process Critic with Explorative Demonstration Reinforcement Learning for LLM Agentic Reasoning (2026.findings-acl)

Challenge:	Existing large language models are limited in understanding, reasoning, calculation, and generation, limiting their performance in complex reasoning and dynamic tasks.
Approach:	They propose a plug-and-play framework that integrates a small-scale LLM (as agent) with large-scale large-level LLMs (a as environment) they propose generating prompts that are used to interact with LLM, and a double constraint reward that optimizes correctness and quality of generation.
Outcome:	The proposed framework significantly outperforms baseline large-scale large-language models across various tasks.

Copied to clipboard

Jingyue Gao, Yanjiang Guo, Chen Xiaoshuai, Jianyu Chen

Accelerating Prefilling via Decoding-time Contribution Sparsity (2026.findings-acl)

Challenge:	Large Language Models (LLMs) exhibit exceptional reasoning capabilities, driven by Reinforcement Learning with Verifiable Rewards (RLVR).
Approach:	They propose a method that uses a process-level critic to monitor interactions in real time, incorporating reflection-based demonstrations to guide agents in stopping the accumulation of errors.
Outcome:	The proposed approach exceeds the model’s saturated exploration performance and achieves superior performance on complex deep search and embodied tasks.

Copied to clipboard

Zhiyuan He, Yike Zhang, Chengruidong Zhang, Huiqiang Jiang, Yuqing Yang, Lili Qiu

Challenge:	Existing acceleration methods exploit attention score sparsity by estimating blocks with high attention scores and applying dynamic sparse attention.
Approach:	They propose a method which replaces dense attention with Triangle attention in a subset of layers to reduce the time needed to decode.
Outcome:	Experiments show that TriangleMix achieves near-lossless performance on long-context and long-constrast reasoning benchmarks while significantly improving efficiency.

SpecBound: Adaptive Bounded Self-Speculation with Layer-wise Confidence Calibration (2026.findings-acl)

Copied to clipboard

Zhuofan Wen, Yang Feng

Stable and Explainable Personality Trait Evaluation in Large Language Models with Internal Activations (2026.findings-acl)

Challenge:	Speculative decoding has emerged as a promising approach to accelerate autoregressive inference in large language models.
Approach:	They propose a self-draft framework that suppresses spurious confidence via layer-wise temperature annealing in early-exit decision and adaptively bounds speculation length based on token-wise decoding difficulty.
Outcome:	The proposed framework suppresses spurious confidence and bounds speculation length based on token-wise decoding difficulty.

Copied to clipboard

Xiaoxu Ma, Xiangbo Zhang, Zhenyu Weng

Learning Continuous Temporal Dynamics on Symplectic Manifolds for Temporal Knowledge Graph Embedding (2026.findings-acl)

Challenge:	Existing questionnaire-based evaluation methods exhibit limited stability and offer little explainability, as their results are sensitive to minor variations in prompt phrasing or role-play configurations.
Approach:	They propose an internal-activation-based approach for stable and explainable personality trait evaluation in Large Language Models by interpolating a persona vector associated with a target personality trait from the model's internal activations.
Outcome:	The proposed approach yields significantly more stable personality trait evaluations than existing methods, even under questionnaire and role-play variants.

Copied to clipboard

Jiang Li, Zehua Duo, Tian Lan, Feilong Bao, Guanglai Gao, Xiangdong Su

mPresenter: An Agentic Framework for Generating Multilingual Presentation Videos from Scientific Papers (2026.findings-acl)

Challenge:	Existing methods for temporal knowledge graph embedding lack explicit structural constraints for continuous-time dynamics.
Approach:	They propose a Temporal Knowledge Graph Embedding framework that embeds temporal dynamics into a symplectic phase space.
Outcome:	The proposed framework achieves competitive performance with lower embedding dimensions.

Copied to clipboard

Wenhan Han, Xiao Xiao, Mykola Pechenizkiy, Meng Fang

Tree-CoT-RT: An Explainable Multi-Path Tree-Guided Chain-of-Thought and Reinforcement Learning Framework for Aspect Sentiment Quad Prediction (2026.findings-acl)

Challenge:	Existing Paper2Video systems are monolingual and often rely on single-pass pipelines.
Approach:	They propose a multilingual agentic Paper2Video system that decomposes the task into planning, audience-oriented critique, layout-aware slide generation, and multilingual figure interpretation.
Outcome:	The proposed system improves question-answering accuracy relative to previous systems while maintaining affordable cost and latency.

Copied to clipboard

Hao Zhang, Jiahao Wang, Zhenke Duan, Xin Yin, Haichuan Hu, Hualong Chen, null Suyi, Congqing He, Yike Tan, Yu-N Cheah

PRISM-MCTS: Learning from Reasoning Trajectories with Metacognitive Reflection (2026.findings-acl)

Challenge:	Existing methods lack explainability and generalization, making it difficult to justify inference decisions and detect implicit sentiment across domains and varied expression patterns.
Approach:	They propose an explainable multi-path tree-guided chain-of-thought framework specifically designed for ASQP.
Outcome:	Experiments on benchmark datasets show that Tree-CoT-RT outperforms baselines.

Copied to clipboard

Siyuan Cheng, Bozhong Tian, Yanchao Hao, Zheng Wei

CausalityCheck: A Framework for Evaluating Causal Reasoning in Large Language Models (2026.findings-acl)

Challenge:	Existing reasoning models are limited by inefficiency and computational redundancy . PRISM-MCTS integrates a process reward model with a dynamic shared memory .
Approach:	They propose a reasoning framework that integrates a process reward model with a dynamic shared memory.
Outcome:	PRISM-MCTS integrates a process reward model with a dynamic shared memory . it halves trajectory requirements on GPQA while surpassing MCTS-RAG and Search-o1 .

Copied to clipboard

Jiang Li, Zehua Duo, Guanglai Gao, Xiangdong Su

OmniDiagram: Advancing Unified Diagram Code Generation via Visual Interrogation Reward (2026.findings-acl)

Challenge:	Existing evaluation methods fail to accurately reflect a model's causal reasoning capabilities.
Approach:	They propose a tool to automatically generate causal reasoning checklists to assess the causal reasoning abilities of 18 large language models.
Outcome:	The proposed tool assesses the causal reasoning abilities of 18 large language models.

Copied to clipboard

Haoyue Yang, Xuanle Zhao, Xuexin Liu, Feibing Jiang, Yao Zhu

Why Agents Compromise Safety Under Pressure (2026.findings-acl)

Challenge:	Existing studies on programmable diagram generation focus on a narrow set of tasks and languages.
Approach:	They propose a unified framework that integrates diverse diagram code languages and task definitions.
Outcome:	The proposed framework can bridge complex visual information with executable code across diverse tasks and languages.

Copied to clipboard

Hengle Jiang, Ke Tang

Challenge:	Recent studies have focused on adversarial attacks, but this perspective overlooks a critical threat arising from the internal drive of the agent.
Approach:	They propose a new concept called Agentic Pressure which characterizes tension when compliant execution becomes infeasible.
Outcome:	The proposed model is able to achieve goal achievement while maintaining safety constraints.

CoMoL: Efficient Mixture of LoRA Experts via Dynamic Core Space Merging (2026.findings-acl)

Copied to clipboard

Jie Cao, Zhenxuan Fan, Zhuonan Wang, Tianwei Lin, Ziyuan Zhao, Rolan Yan, Wenqiao Zhang, Feifei Shao, Hongwei Wang, Jun Xiao, Siliang Tang

QuCo-RAG: Quantifying Uncertainty from the Pre-training Corpus for Dynamic Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	Existing PEFT methods suffer from limited parameter efficiency and coarse-grained adaptation due to proliferation of LoRA experts and instance-level routing.
Approach:	They propose a new MoE-LoRA framework that incorporates expert diversity, parameter efficiency, and fine-grained adaptation.
Outcome:	The proposed framework outperforms existing methods on multiple tasks while maintaining parameter efficiency.

Copied to clipboard

Dehai Min, Kailin Zhang, Tongtong Wu, Lu Cheng

CaM-HG: Causal-Enhanced MoE and Hypergraphs Network for Incomplete Multimodal Emotion Recognition in Conversations (2026.findings-acl)

Challenge:	Existing methods for reducing LLM hallucinations rely on model-internal signals . Existing approaches rely only on model internal signals, resulting in unreliability .
Approach:	They propose a method that shifts from subjective confidence to objective statistics . they leverage Infini-gram for millisecond-latency queries over 4 trillion tokens .
Outcome:	The proposed method reduces hallucinations in large language models by reducing uncertainty in the model.

Copied to clipboard

Mingjian Yang, Yong Wang, Peng Liu, Wen Yin

Large-Scale Diverse Synthesis for Mid-Training (2026.findings-acl)

Challenge:	Existing methods that focus on statistical reconstruction often fail to bridge these gaps, effectively leaving semantic holes.
Approach:	They propose a Causal-Enhanced Mixture-of-Experts and Hypergraph Network to bridge missing features . they use experts to synthesize missing features that are realistic and causally consistent .
Outcome:	The proposed model synthesizes missing features that are realistic and causally consistent . it surpasses benchmarks on IEMOCAP, CMU-MOSI, and CMU MOSEI by 1.43% and 1.25% .

Copied to clipboard

Xuemiao Zhang, Chengying Tu, Can Ren, Rongxiang Weng, Hongfei Yan, Jingang Wang, Xunliang Cai

DiSec: Mitigating Backdoors in Pre-trained Language Models via Disentanglement of Adversarial Weights for Secure Fine-Tuning (2026.findings-acl)

Challenge:	Existing data synthesis methods generate simplistic and homogeneous QA pairs with limited scale and diversity.
Approach:	They propose a framework to synthesize large-scale, diverse, and high-quality QA data for mid-training.
Outcome:	The proposed framework improves on 500B-token BoostQA data over pre-training benchmarks.

Copied to clipboard

Sunanda Das, Qinghua Li

LUNE: Efficient LLM Unlearning via LoRA Fine-Tuning with Negative Examples (2026.findings-acl)

Challenge:	Existing defenses rely on privileged assumptions, limiting their applicability in realistic settings.
Approach:	They propose a task-agnostic backdoor attack that contaminates pre-trained language models . authors propose auxiliary text purification framework that uses only clean auxiliary data .
Outcome:	The proposed framework suppresses attack success while preserving clean-task utility.

Copied to clipboard

Yezi Liu, Hanning Chen, Wenjun Huang, Yang Ni, Mohsen Imani

Think Earlier, Not Longer: Prompt Optimization via Reducing Unhealthy Exploration (2026.findings-acl)

Challenge:	Large Language Models encode vast factual knowledge, yet their inability to selectively forget specific information hinders privacy protection, bias mitigation, and post-deployment correction.
Approach:	They propose a LoRA-based negative-only unlearning framework that updates only low-rank adapters while freezing the backbone.
Outcome:	The proposed framework reduces computational cost by about an order of magnitude compared to full fine-tuning and memory-editing methods.

Copied to clipboard

Ling-I Wu, Minyu Chen, Jingyang Li, Xi Chang, Guoqiang Li

Psychological Counseling Cannot Be Achieved Overnight: Automated Psychological Counseling Through Multi-Session Conversations (2026.findings-acl)

Challenge:	Existing approaches to improve reasoning performance ignore the presence of unhealthy exploration that increases token usage without contributing to effective problem-solving.
Approach:	They propose an entropy-dynamics-aware prompt optimization framework that trains a lightweight optimizer to generate concise clarifications.
Outcome:	The proposed framework reduces ambiguity-induced early-stage uncertainty while preserving the model's reasoning capabilities.

Copied to clipboard

Bichen Wang, Junzhe Wang, Yixin Sun, Xing Fu, Yanyan Zhao, Bing Qin

Code over Words: Overcoming Semantic Inertia via Code-Grounded Reasoning (2026.findings-acl)

Challenge:	Existing studies on single-session counseling are limited to a single-session setting.
Approach:	They propose to use a large language model to deliver automated psychological counseling to a dataset constructed using real client profiles from publicly available psychological case reports.
Outcome:	The proposed model performs better than baseline models across multiple sessions.

Copied to clipboard

Manjie Xu, Isabella Yin, Xinyi Tu, Chi Zhang, Yixin Zhu

Bypassing Neural Evaluations for Fast Audio Editing via Adaptive Trajectory Extrapolation (2026.findings-acl)

Challenge:	Large Language Models struggle with semantic inertia, a problem that is often attributed to natural language encoding, which entangles descriptive semantics and logical rules, leading to persistent hallucinations of familiar physics despite explicit contradictory rules.
Approach:	They propose a framework that decouples logical dynamics from visual priors via amortized theory induction and counterfactual contrastive alignment.
Outcome:	The proposed framework outperforms expensive inference-time search methods in both efficiency and accuracy.

Copied to clipboard

Xiaoqian Liu, Zhengkun Ge, Jianjin Wang, Haoran Zhang, Yuan Ge, Kaiyan Chang, Chen Xu, Tong Xiao, Zhengtao Yu, Linfeng Zhang, JingBo Zhu

Agent-GWO: Collaborative Agents for Dynamic Prompt Optimization in Large Language Models (2026.findings-acl)

Challenge:	Recent advances in audio diffusion models have significantly improved text-to-audio editing via inversion techniques, but these models typically rely on dense, fixed-step sampling trajectories to maintain structural integrity.
Approach:	They propose a model-agnostic Adaptive Trajectory Extrapolation framework that accelerates inversion-based editing process by dynamically evaluating only the most critical generative phases.
Outcome:	The proposed framework achieves a 3.9 speedup with negligible loss in fidelity.

Copied to clipboard

Xudong Wang, Chaoning Zhang, Chenghao Li, Shuxu Chen, Qigan Sun, Jiaquan Zhang, Fachrina Dewi Puspitasari, Tae-Ho Kim, Jiwei Wei, Malu Zhang, Guoqing Wang, Yang Yang, Heng Tao Shen

Structured Confidence–Guided Online Adaptation for LLM-based Multi-Label Classification (2026.findings-acl)

Challenge:	Existing automatic prompt optimization methods fail to optimize prompts and decoding hyperparameters within a unified framework to achieve stable global improvements.
Approach:	They propose a dynamic prompt optimization framework for complex reasoning that unifies prompt templates and decodes hyperparameters as inheritable agent configurations.
Outcome:	Experiments on multiple mathematical and hybrid reasoning benchmarks show that Agent-GWO improves accuracy and stability over existing prompt optimization methods.

Copied to clipboard

Pengyu Xu, JingRen Hou, Liping Jing, Jian Yu

SERM: Self-Evolving Relevance Model with Agent-Driven Learning from Massive Query Streams (2026.findings-acl)

Challenge:	Large language models (LLMs) enable zero-shot and few-shot multi-label text classification . but most approaches perform static inference and degrade under streaming test data .
Approach:	They propose a structured confidence-guided online adaptation framework for LLM-based multi-label generation without parameter updates.
Outcome:	The proposed framework improves Micro-F1 and Macro-F1, with the largest gains on long-tail labels.

Copied to clipboard

Chenglong Wang, Canjia Li, Xingzhao Zhu, Yifu Huo, Huiyu Wang, Weixiong Lin, Yun Yang, Qiaozhi He, Tian Hua Zhou, null Changxiaojia, JingBo Zhu, Tong Xiao

Challenge:	Existing approaches to generate relevance judgments are limited due to dynamic nature of query distributions.
Approach:	They propose a self-evolving relevance model approach to generalize queries to practical search scenarios . they use a multi-agent sample miner and a relevance annotator to generate reliable labels .
Outcome:	The proposed approach can achieve significant performance gains on a large-scale industrial platform, validated by offline multilingual evaluations and online testing.

CLAG: Adaptive Memory Organization via Agent-Driven Clustering for Small Language Model Agents (2026.findings-acl)

Copied to clipboard

Taeyun Roh, WonJune Jang, Junha Jung, Jaewoo Kang

ClinAlign: Scaling Healthcare Alignment from Clinician Preference (2026.findings-acl)

Challenge:	Large language model agents rely on external memory to support knowledge reuse and reasoning tasks.
Approach:	They propose a CLustering-based AGentic memory framework where an agent actively organizes memory . they employ an SLM-agent driven router to assign each new memory to a semantically coherent cluster .
Outcome:	The proposed framework improves answer quality and robustness over previous memory systems.

Copied to clipboard

Shiwei Lyu, Xidong Wang, Hao Zhu, Lei Liu, Chaohe Zhang, Jian Wang, Jinjie Gu, Benyou Wang, Yue Shen

Challenge:	Existing methods for aligning open-ended outputs with fine-grained clinician preferences are weakly grounded in professional guidelines.
Approach:	They propose a framework to align large language models' outputs with fine-grained clinician preferences . they propose 119 broadly reusable, clinically grounded principles organized by clinical dimensions .
Outcome:	The proposed framework outperforms existing models on HealthBench-Hard and Deepseek-R1 and o3.

SessionIntentBench: A Multi-task Inter-session Intention-shift Modeling Benchmark for E-commerce Customer Behavior Understanding (2026.findings-acl)

Copied to clipboard

Yuqi Yang, Weiqi Wang, Baixuan Xu, Wei Fan, Qing Zong, Chunkit Chan, Zheye Deng, Xin Liu, Yifan Gao, Changlong Yu, Chen Luo, Yang Li, Zheng Li, Qingyu Yin, Bing Yin, Yangqiu Song

Adapting Where It Matters: Depth-Aware Adaptation for Efficient Multilingual Speech Recognition in Low-Resource Languages (2026.findings-acl)

Challenge:	Existing models fail to capture and model customer intention effectively because of insufficient information exploitation and only apparent information like descriptions and titles are used.
Approach:	They propose to exploit existing session data to capture and model intention in E-commerce product purchase sessions using a multimodal benchmark.
Outcome:	The proposed framework can bridge the gap between intention understanding in simplified research cases like co-buy intention and more complex yet practical scenarios like session history.

Copied to clipboard

Yang Xiao, Eun-Jung Holden, Ting Dang

Targeted Exploration via Unified Entropy Control for Reinforcement Learning (2026.findings-acl)

Challenge:	Recent speech foundation models excel at multilingual automatic speech recognition (ASR) for high-resource languages, but their performance drops substantially on low-resourced languages due to the limited data availability.
Approach:	They propose a Depth-Aware Model Adaptation framework that allocates adaptation capacity according to each layer’s role.
Outcome:	The proposed framework matches or surpasses state-of-the-art accuracy with 80% fewer trainable parameters and achieves 29% error reduction under extreme data scarcity.

Copied to clipboard

Chen Wang, Lai Wei, Yanzhi Zhang, Chenyang Shao, Zedong Dan, Weiran Huang, Ge Lan, Yue Wang

Remember Me, Refine Me: A Dynamic Procedural Memory Framework for Experience-Driven Agent Evolution (2026.findings-acl)

Challenge:	Existing methods for group relative policy optimization suffer from entropy collapse . Existing exploration methods introduce additional bias or variance during exploration, making it difficult to maintain stability.
Approach:	They propose a framework that provides targeted mechanisms for exploration and stabilization.
Outcome:	The proposed framework expands search space on difficult prompts while preventing entropy growth uncontrollably.

Copied to clipboard

Zouying Cao, Jiaji Deng, Li Yu, Weikang Zhou, Zhaoyang Liu, Bolin Ding, Hai Zhao

Knowing When to Quit: Diagnosing and Training LLMs to Abort Futile Reasoning (2026.findings-acl)

Challenge:	Existing frameworks treat memory as a static append-only archive . Existing systems focus on passive accumulation, resulting in a 'passive accumulation' of memory.
Approach:	They propose a framework for experience-driven agent evolution that integrates procedural memory with contextual information to create a high-quality experience pool.
Outcome:	Experiments on BFCL-V3 and AppWorld show that ReMe outperforms memoryless Qwen3-8B.

Copied to clipboard

Xinyan Guan, Jiali Zeng, Chunlei Xin, Yaojie Lu, Hongyu Lin, Xianpei Han, Le Sun, Fandong Meng

AutoL2S: Auto Long-Short Reasoning for Efficient Large Language Models (2026.findings-acl)

Challenge:	Large language models generate costly yet semantically void reasoning on beyond-capability tasks . the dominant failure mode is specious reasoning, superficially valid outputs with subtle hallucinations .
Approach:	They propose a capability-aligned reinforcement learning approach that aligns model behavior with capability boundaries.
Outcome:	The proposed model reduces futile reasoning while maintaining performance across tasks.

Copied to clipboard

Feng Luo, Yu-Neng Chuang, Guanchu Wang, Hoang Anh Duy Le, Shaochen Zhong, Hongyi Liu, Jiayi Yuan, Yang Sui, Vladimir Braverman, Vipin Chaudhary, Xia Hu

Doc-PP: Document Policy Preservation Benchmark for Large Vision-Language Models (2026.findings-acl)

Challenge:	Existing approaches to distilling large language models (LLMs) are inefficient and generate excessively long chain-of-thought reasoning even for inputs that admit concise solutions.
Approach:	They propose a distillation framework that empowers non-reasoning LLMs to think only when necessary.
Outcome:	The proposed framework reduces reasoning length up to 71% with minimal accuracy loss while preserving accuracy.

Copied to clipboard

Haeun Jang, Hwan Chang, Hwanhee Lee

RIPRAG: Hack a Black-box Retrieval-Augmented Generation Question-Answering System with Reinforcement Learning (2026.findings-acl)

Challenge:	Existing safety research focuses on implicit social norms or text-only settings, overlooking the complexities of multimodal documents.
Approach:	They propose a benchmark to assess the safety of large vision-Language Models (LVLMs) they propose 'Document Policy Preservation Benchmark' to assess document policy compliance.
Outcome:	The proposed framework outperforms standard prompting defenses in the evaluation of multimodal documents.

Copied to clipboard

Meng Xi, Sihan Lv, Yechen Jin, Guanjie Cheng, Naibo Wang, Ying Li, Jianwei Yin

CodeMEM: AST-Guided Adaptive Memory for Repository-Level Iterative Code Generation (2026.findings-acl)

Challenge:	Existing methods to generate RAG documents require knowledge of the target RAG system’s internal composition and implementation details, whereas black-box methods are unable to utilize interactive information.
Approach:	They propose a RIPRAG attack framework that treats the target RAG system as a black box and leverages a Reinforcement Learning from Black-box Feedback (RLBF) method to optimize the generation model for poisoned documents.
Outcome:	The proposed method achieves an attack success rate (ASR) improvement of up to 0.72 compared to baseline methods.

Copied to clipboard

Peiding Wang, Li Zhang, Fang Liu, Chongyang Tao, Yinghao Zhu

HalluGuard: Evidence-Grounded Small Reasoning Models to Mitigate Hallucinations in Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	Existing memory management approaches show promise but remain limited by natural language-centric representations.
Approach:	They propose an AST-guided dynamic memory management system for repository-level iterative code generation that maintains and updates repository context through AST operations.
Outcome:	The proposed system improves instruction following by 12.2% and reduces interaction rounds by 2–3 while maintaining competitive inference latency and token efficiency.

Copied to clipboard

Loris Bergeron, Ioana Buhnila, Jerome Francois, Radu State

Capabilities and Evaluation Biases of Large Language Models in Classical Chinese Poetry Generation: A Case Study on Tang Poetry (2026.findings-acl)

Challenge:	Large Language Models excel at NLP tasks but remain prone to hallucinations . small language models can achieve competitive results in specific tasks .
Approach:	They propose a 4B-parameter Small Reasoning Model (SRM) that can be used to classify document-claim pairs as grounded or hallucinated in closed-book, document-grounded settings.
Outcome:	The proposed model achieves 84.4% balanced accuracy on the RAGTruth subset of the LLM-AggreFact benchmark, surpassing specialized models, MiniCheck (7B; 84.0%) and Granite Guardian 3.3 (82.2%) Across the benchmark, it reaches 77.1% BAcc, surpasses larger general-purpose LLMs such as GPT-4o (75.9%).

Copied to clipboard

Bolei Ma, Yina Yao, Anna-Carolina Haensch

SWE-QA-Pro: A Representative Benchmark and Scalable Training Recipe for Repository-Level Code Understanding (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are increasingly applied to creative domains, yet performance in classical Chinese poetry generation and evaluation remains poorly understood.
Approach:	They propose a framework that combines computational metrics, LLM-as-a-judge assessment, and human expert validation to evaluate large language models.
Outcome:	The proposed framework evaluates state-of-the-art LLMs across multiple dimensions of poetic quality in Tang poetry generation.

Copied to clipboard

Songcheng Cai, Zhiheng Lyu, Yuansheng Ni, Xiangchao Chen, Baichuan Zhou, Shenzhe Zhu, Yi Lu, Haozhe Wang, Chi Ruan, Benjamin Schneider, Weixu Zhang, Xiang Li, Andy Zheng, Yuyu Zhang, Ping Nie, Wenhu Chen

Do MLLMs Understand Pointing? Benchmarking and Enhancing Referential Reasoning in Egocentric Vision (2026.findings-acl)

Challenge:	Existing benchmarks for agentic repository-level code understanding overlook long tail topics and rely on memorized knowledge.
Approach:	They propose a repository-level agentic code understanding benchmark that uses long-tail repositories with executable environments to enforce topical balance.
Outcome:	Empirically, a Qwen3-8B model trained with the proposed benchmark outperforms GPT-4o by 2.3 points.

Copied to clipboard

Chentao Li, Zirui Gao, Mingze Gao, Yinglian Ren, Jianjiang Feng, Jie Zhou

Can Post-Training Transform LLMs into Causal Reasoners? (2026.findings-acl)

Challenge:	Egocentric AI agents rely on pointing to resolve referential ambiguities in natural language commands.
Approach:	They propose a question-answering benchmark to evaluate and enhance pointing reasoning in egocentric views.
Outcome:	The proposed benchmark evaluates and enhances pointing reasoning in egocentric views.

Copied to clipboard

Junqi Chen, Sirui Chen, Chaochao Lu

MoE Adapter for Large Audio Language Models: Sparsity, Disentanglement, and Gradient-Conflict-Free (2026.findings-acl)

Challenge:	Causal inference is a core component of human cognition and requires decision-makers to distinguish between causation and association.
Approach:	They propose a dataset comprising seven core causal tasks for training and five diverse test sets and evaluate five different post-training approaches.
Outcome:	The proposed model achieves 93.5% accuracy on the CaLM benchmark, compared to 55.4% by OpenAI o3.

Copied to clipboard

Yishu Lei, Shuwei He, Hu Jing, Dan Zhang, Xianlong Luo, Danxiang Zhu, Shikun Feng, Rui Liu, Jingzhou HE, Yu Sun, Hua Wu, Haifeng Wang

QaRL: Rollout-Aligned Quantization-Aware RL for Fast and Stable Training under Training–Inference Mismatch (2026.findings-acl)

Challenge:	Existing research on Large Language Models (LLMs) limited to textual input modality . acoustic information is intrinsically heterogeneous, entangling attributes such as speech, music, and environmental context.
Approach:	They propose a sparse Mixture-of-Experts architecture to decouple acoustic information by routing audio tokens to specialized experts.
Outcome:	The proposed architecture outperforms existing models on audio semantic and paralinguistic tasks while retaining shared experts for global context.

Copied to clipboard

Hao Gu, Hao Wang, Jiacheng Liu, Lujun Li, Qiyuan Zhu, Bei Liu, Binxing Xu, Lei Wang, Xintong Yang, Sida Lin, Sirui Han, Yike Guo

Challenge:	Recent work has shown that reinforcement learning with simple rule-based reward functions (RLVR) can induce emergent reasoning behaviors and yield gains in challenging domains such as math problem solving.
Approach:	They propose a rollout-alignment-quantization-aware RL which aligns training-side forward with the quantized rollout to minimize mismatch.
Outcome:	The proposed approach outperforms quantized-rollout training by +5.5 on Qwen3-30B-A3B MoE for math problems while maintaining low-bit throughput.

Beyond Static Rules: Automated Discovery of Latent Vulnerabilities in Text-to-SQL (2026.findings-acl)

Copied to clipboard

Hanqing Wang, Yongdong Chi, Jian Yang, Lei Yang, Jiehui Zhao, Yun Chen, Guanhua Chen

Can We Entrust Justice to AI?: How Persona Traps Contaminate Reasoning in Criminal Investigation (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have been successful in Text-to-SQL tasks, but their deployment in real-world environments is hindered by latent reliability issues.
Approach:	They propose a framework to autonomously uncover latent failure patterns in LLM-based Text-to-SQL generation.
Outcome:	The proposed framework uncovers a substantial number of failure cases on state-of-the-art open-source LLMs.

Copied to clipboard

Jaewook Lee, Myeong-Cheol Kang, Jong-hun Shin

Beyond Semantic Similarity: Appraisal-Guided Chain-of-Thought Reasoning and Retrieval for Multimodal Emotional Support Conversations (2026.findings-acl)

Challenge:	a study injected personas into neutralized murder mystery scenarios to examine reasoning stability of large language models . adrian s. gupta: if models are used to evaluate suspects, are they free from the trap of implicit bias? he says current alignment techniques focus on identity-based bias while neglecting relationship-based ones .
Approach:	a study systematically injected personas into neutralized murder mystery scenarios . it found that implicit bias propagation was observed across all models . authors propose stability evaluation should encompass outputs and reasoning processes .
Outcome:	The proposed pipeline can analyze evidence and evaluate suspects in murder mysteries . it shows that models outwardly state "that information is irrelevant to the judgment" the proposed pipeline could be extended to include reasoning processes, authors say .

Copied to clipboard

Yuqi Chu, Lizi Liao, Jinggui Liang, Boyang Li, Richang Hong

When and What to Ask: AskBench and Rubric-Guided RLVR for LLM Clarification (2026.findings-acl)

Challenge:	Existing retrieval-augmented generation paradigms rely on semantic similarity to retrieve historical dialogues that are surface analogous but therapeutically incongruent.
Approach:	They propose to use appraisal-guided reasoning chains to generate appraisal-based reasoning chains and apply a dual-signal verification mechanism to verify and correct them.
Outcome:	Extensive experiments on two ESC benchmarks show that the proposed model significantly outperforms state-of-the-art models.

Copied to clipboard

Jiale Zhao, Ke Fang, Lu Cheng

Challenge:	Large language models respond even when prompts omit critical details or include misleading information, leading to hallucinations or reinforced misconceptions.
Approach:	They propose an interactive benchmark that converts standard QA pairs into multi-turn interactions with explicit checkpoints.
Outcome:	The proposed benchmark improves accuracy, rubric adherence, and interaction efficiency with strong generalization to unseen domains.

FreezeEmpath: Efficient Training for Empathetic Spoken Chatbots with Frozen LLMs (2026.findings-acl)

Copied to clipboard

Yun Hong, Yan Zhou, Yang Feng

To Intervene or Not: Guiding Inference-time Alignment with Probabilistic Model Blending (2026.findings-acl)

Challenge:	Recent research has made significant progress in developing empathetic spoken chatbots based on large language models (LLMs).
Approach:	They propose an end-to-end empathetic spoken chatbot trained efficiently that generates emotionally expressive speech and outperforms other emmpathetic models in emphatic dialogue, SER, and SpokenQA tasks.
Outcome:	The proposed model outperforms other empathetic models on e-dialog, SER, and SpokenQA tasks and achieves strong results on several speech tasks.

Copied to clipboard

Jin Gan, Xin Li, Jun Luo

Challenge:	Existing approaches to inference-time alignment are expensive and only offer guidances during output generation.
Approach:	They propose an inference-time alignment framework that shifts from binary decisions to creating hybrid distributions integrating both models’ knowledge.
Outcome:	The proposed framework reduces the number of inference-time alignment interventions and improves performance on challenging model pairs.

FLEXITOKENS: Flexible Tokenization for Evolving Language Models (2026.findings-acl)

Copied to clipboard

Abraham Toluwase Owodunni, Orevaoghene Ahia, Sachin Kumar

RAG in the Wild: On the (In)effectiveness of LLMs with Mixture-of-Knowledge Retrieval Augmentation (2026.findings-acl)

Challenge:	Widely used subword tokenizers overfragment sequences in unseen domains, languages, and scripts . inefficient tokenizer models can cause overfragments in out-of-distribution domains if not trained properly .
Approach:	They propose a byte-level LM with learnable tokenizers to make tokenization adaptive . they propose 'flexitoken' which enables significantly greater flexibility during adaptation .
Outcome:	The proposed method significantly reduces token overfragmentation and improves on multilingual benchmarks and domains.

Copied to clipboard

Ran Xu, Yuchen Zhuang, Yue Yu, Haoyu Wang, Wenqi Shi, Carl Yang

Chain-of-Procedure: Hierarchical Visual-Language Reasoning for Procedural QA (2026.findings-acl)

Challenge:	Retrieval-augmented generation (RAG) enhances large language models by integrating external knowledge retrieved at inference time.
Approach:	They evaluate RAG systems using MassiveDS, a large-scale datastore with mixture of knowledge.
Outcome:	The proposed approach improves performance on knowledge-intensive NLP tasks.

Copied to clipboard

Guanhua Chen, Yutong Yao, Shenghe Sun, Ci-jun Gao, Shudong Liu, Lidia S. Chao, Feng Wan, Derek F. Wong

PEGRL: Improving Machine Translation by Post-Editing Guided Reinforcement Learning (2026.findings-acl)

Challenge:	Recent advances in vision-language models (VLMs) have achieved impressive results on standard image-text tasks, yet their capability in visual procedure question answering (VP-QA) remains largely unexplored.
Approach:	They propose a multimodal benchmark specifically designed for visual procedural reasoning that synergizes cross-modal procedure retrieval, context-aware step decomposition, and the next step prediction.
Outcome:	The proposed framework significantly outperforms baselines on visual procedure question answering (VP-QA) Experiments on six VLMs show that it performs better than baselines.

Copied to clipboard

Yunzhi Shen, Hao Zhou, Xin Huang, Xue Han, Junlan Feng, Shujian Huang

RanLoRA: Residual-aware Nonlinear Low-Rank Adaptation (2026.findings-acl)

Challenge:	Reinforcement learning (RL) has shown strong promise for LLM-based machine translation . however, translation-oriented RL remains challenged by high-variance policy gradients induced by Monte Carlo baselines and large trajectory space that favors global exploration over fine-grained local optimization.
Approach:	They propose a two-stage RL framework that uses post-editing as an auxiliary task to stabilize training and guide overall optimization.
Outcome:	The proposed framework supports global exploration and fine-grained optimization while supporting global exploration.

Copied to clipboard

Xu Luo, Yongbin Liu, Chunping Ouyang, Ying Yu

Multi-Hop Knowledge Editing via Critic-Guided Multi-Agent Reasoning (2026.findings-acl)

Challenge:	Low-Rank Adaptation (LoRA) relying on linear low-rank projections restricts adaptation to linear subspaces, limiting flexibility on complex downstream tasks.
Approach:	They propose a nonlinear low-rank Adaptation approach that leverages pretrained weights to decompose them into principal components that are kept frozen and residual components that can be used for task-specific adaptation.
Outcome:	The proposed approach outperforms vanilla LoRA and representative variants on commonsense reasoning, image classification, and mathematical reasoning tasks.

Copied to clipboard

Xudong Li, Yuhang Tian, Dandan Song, Zhijing Wu, Shuhao Zhang, Jun Yang, Yongyu Huo, Changzhi Zhou, Xinyu Zhang, Chenhao Li, Huipeng Ma, Luan Zhang, Yan Xu, Qian Liu

Rethinking Post-Unlearning Behavior of Large Vision-Language Models (2026.findings-acl)

Challenge:	Existing knowledge editing methods rely on unidirectional, feed-forward pipelines . a minor retrieval error or logical mismatch at an early hop can become a silent failure .
Approach:	They propose a framework for closed-loop post-edit reasoning that uses a Critic agent to verify coherence and step-wise correctness.
Outcome:	Experiments on MQuAKE-2002 and MQuADE-hard show that CARE effectively mitigates error propagation . a minor retrieval error or logical mismatch at an early hop can become a silent failure .

Copied to clipboard

Minsung Kim, Nakyeong Yang, Kyomin Jung

Scattered Hypothesis Generation for Open-Ended Event Forecasting (2026.findings-acl)

Challenge:	Existing methods to remove knowledge from large vision-Language Models often fail to provide quality and informative post-unlearning responses.
Approach:	They propose a task that requires models to provide privacy-preserving yet informative responses for LVLMs.
Outcome:	The proposed method reduces the risk of unlearning after naive suppression by providing informative and visually grounded responses.

Copied to clipboard

He Chang, Zhulin Tao, Lifang Yang, Xianglin Huang, Yunshan Ma

Challenge:	Existing methods for event forecasting focus on the most probable outcomes, neglecting the intrinsic uncertainty of real-world events.
Approach:	They propose a reinforcement learning framework that optimizes inclusiveness and diversity of the hypothesis by integrating validity-gated score into the overall objective.
Outcome:	The proposed framework outperforms baselines on two real-world benchmark datasets.

Context-attended Adversarial Reinforcement Learning for Robust Multi-step Retrieval Augmented Generation (2026.findings-acl)

Copied to clipboard

Yingtao Ren, Xiao Luo, Yu-Cheng Chang, Chin-teng Lin

From Pseudo-Balancing to True Specialization: Memory-Aware Routing for Mixture-of-Experts (2026.findings-acl)

Challenge:	Existing approaches to multi-step retrieval-augmented generation are susceptible to retrieval noise and fabricated documents in real-world scenarios.
Approach:	They propose a framework for multi-step retrieval-augmented generation that incorporates external knowledge into a retriever to generate responses from adversarial samples.
Outcome:	The proposed framework improves performance in multiple noisy scenarios and can be used to improve multi-step retrieval-augmented generation.

Copied to clipboard

Peixuan Hou, Yunbo Hou, Bin Chen, LI He, Jian Xu, Weiping Li, Bo Zheng, Guojie Song

Learning to Retrieve User History and Generate User Profiles for Personalized Persuasiveness Prediction (2026.findings-acl)

Challenge:	Existing methods to optimize expert-centered load balancing fail to account for pseudo-balance phenomenon . severe knowledge overlap among experts leads to redundant representations and inefficient parameter utilization .
Approach:	They propose a method that prioritizes expert utilization over semantic alignment . they use memory-aware routing to ensure expert load balancing is consistent .
Outcome:	Experimental results show that MAR improves expert specialization by 35% and accuracy by 2%-25% . MAR matches baseline performance with only half the experts .

Copied to clipboard

Sejun Park, Yoonah Park, Jongwon Lim, Yohan Jo

Understanding LLM Reasoning for Abstractive Summarization (2026.findings-acl)

Challenge:	Large language models (LLMs) are increasingly used in decision-support applications that aim to influence human behavior or beliefs, such as health coaching, tutoring, and targeted marketing.
Approach:	They propose a context-aware user profiling framework with two trainable components that generate optimal queries to retrieve persuasion-relevant records from a user’s history and a profiler that summarizes these records into a model.
Outcome:	The proposed framework raises F1 from 33% to 47% on Llama-3.3-70B-Instruct.

Copied to clipboard

Haohan Yuan, Haopeng Zhang

Feasible is Not Enough: Cost-Aware Optimal Tool-Chain Planning on Multi-Solution Tool Graphs (2026.findings-acl)

Challenge:	Explicit reasoning strategies improve reference-based quality, but weaken factual grounding, whereas implicit reasoning in LRMs shows the opposite tendency.
Approach:	They adapt general reasoning strategies to the summarization setting and conduct a large-scale comparative study of 8 reasoning strategies and 3 Large Reasoning Models (LRMs) they find a trade-off between summary quality and factual faithfulness.
Outcome:	The proposed reasoning strategies and 3 Large Reasoning Models (LRMs) are compared with 8 reasoning strategies across 8 datasets.

Copied to clipboard

Liangliang Liu, Yanming Li, Yigang Liu, Jialong Han, Rujia Shen, Yi Guan, Yi Lin, Jingchi Jiang

DOS: Dependency-Oriented Sampler for Masked Diffusion Language Models (2026.findings-acl)

Challenge:	Existing tools and benchmarks often form tool learning (TL) as a single-solution setting . exploring large-scale TG is computationally expensive, especially under constrained context budgets.
Approach:	They propose a framework for learning optimal TL policies over large tool graphs . they train a reinforcement learning agent to acquire transferable expansion skills .
Outcome:	The proposed framework improves task success and solution optimality by 46.21% and 66.34% on multiSoTLBench.

Copied to clipboard

Xueyu Zhou, Yangrong Hu, Jian Huang

BIASEDTALES-ML: A Multilingual Dataset for Analyzing Narrative Attribute Distributions in LLM-Generated Stories (2026.findings-acl)

Challenge:	Existing decoding strategies for pre-trained MDLMs rely on token-level uncertainty criteria, while largely overlooking sequence-level information and inter-token dependencies.
Approach:	They propose a training-free decoding strategy that leverages inter-token dependencies to inform token updates during generation.
Outcome:	Empirical results show that the proposed approach consistently achieves superior performance on both code generation and mathematical reasoning tasks.

Copied to clipboard

Yuxuan Ouyang, Yingfeng Luo, JingBo Zhu, Tong Xiao

Rethinking Reasoning: A Survey on Reasoning-based Backdoors in LLMs (2026.findings-acl)

Challenge:	Existing studies on the use of Large Language Models (LLMs) focus primarily on English, leaving the cross-lingual generalization of aligned behavior underexplored.
Approach:	They propose a structured generator-extractor pipeline and a multi-dimensional distributional analysis framework to examine how narrative attributes vary across languages, models, and social conditions.
Outcome:	The proposed model reveals substantial cross-lingual variability in narrative generation patterns, indicating that distributions observed in English do not always exhibit similar characteristics in other languages, particularly in lower-resource settings.

Copied to clipboard

Man Hu, Xinyi Wu, Zhufeng Suo, Jinbo Feng, Linghui Meng, Yanhao Jia, Anh Tuan Luu, Shuai Zhao

From Tokens to Steps: Verification-Aware Speculative Decoding for Efficient Multi-Step Reasoning (2026.findings-acl)

Challenge:	Recent models such as OpenAI o1 and DeepSeek-R1 produce explicit reasoning traces, often via Chain-of-Thought prompting.
Approach:	They propose a taxonomy that offers a unified perspective for summarizing existing approaches and categorizing reasoning-based backdoor attacks into associative, passive, and active.
Outcome:	The proposed taxonomy categorizes reasoning-based backdoor attacks into associative, passive, and active.

Copied to clipboard

Kiran Purohit, Ramasuri Narayanam, Soumyabrata Pal

SPS: Steering Probability Squeezing for Better Exploration in Reinforcement Learning for Large Language Models (2026.findings-acl)

Challenge:	Speculative decoding (SD) allows a lightweight draft model to propose outputs that a stronger target model verifies.
Approach:	They propose a verification-aware speculative decoding framework that performs step-level verification using only model-internal signals.
Outcome:	Experiments show that SpecGuard outperforms both SD and reward-guided SD in accuracy and reliability tests.

Copied to clipboard

Yifu Huo, Chenglong Wang, Ziming Zhu, Shunjie Xing, Peinan Feng, Tongran Liu, Qiaozhi He, Tian Hua Zhou, null Changxiaojia, JingBo Zhu, Zhengtao Yu, Tong Xiao

Memp: Exploring Agent Procedural Memory (2026.findings-acl)

Challenge:	Reinforcement learning (RL) training typically improves single-sample success rates but limited exploration of diverse reasoning trajectories.
Approach:	They propose a training paradigm that interleaves conventional RL with inverse reinforcement learning (IRL) they propose 'Steering Probability Squeezing' to enhance exploration without external supervision .
Outcome:	The proposed training paradigm improves Pass@k and improves exploration of diverse reasoning trajectories without external supervision.

Copied to clipboard

Runnan Fang, Yuan Liang, Xiaobin Wang, Jialong Wu, Shuofei Qiao, Pengjun Xie, Fei Huang, Huajun Chen, Ningyu Zhang

Challenge:	Large Language Models (LLMs) based agents suffer from brittle procedural memory that is manually engineered or entangled in static parameters.
Approach:	They propose a procedural-memory repository that distills past agent trajectories into fine-grained, step-by-step instructions and higher-level, script-like abstractions.
Outcome:	The proposed repository can be used to improve agents' performance on travelplanner and Alfworld.

ATLAS: Orchestrating Heterogeneous Models and Tools for Multi-Domain Complex Reasoning (2026.findings-acl)

Copied to clipboard

Jinyang Wu, Guocheng Zhai, Ruihan Jin, Jiahao Yuan, Yuhao Shen, Shuai Zhang, Zhengqi Wen, Jianhua Tao

SWE-AGILE: A Software Agent Framework for Efficiently Managing Dynamic Reasoning Context (2026.findings-acl)

Challenge:	Existing approaches to optimize large language models with external tools are limited.
Approach:	They propose a dual-path framework for dynamic tool usage in cross-domain complex reasoning . they exploit empirical priors for domain-specific alignment and RL-based multi-step routing .
Outcome:	The proposed framework outperforms closed-source models and existing methods on in-distribution and out-of-distortion tasks.

Copied to clipboard

Shuquan Lian, Juncheng Liu, Yazhe Chen, Yuhong Chen, Hui Li

CAR: Empowering Agents with Dynamic Tool Synthesis and Global Trajectory Rectification (2026.findings-acl)

Challenge:	representative ReAct-style approaches lack explicit System-2 reasoning for deep analysis and handling complex edge cases.
Approach:	They propose a software agent framework that preserves full reasoning history while compressing historical reasoning content into concise Reasoning Digests.
Outcome:	Empirically, the proposed framework sets a new standard for 7B-8B models on SWE-Bench-Verified using only 2.2k trajectories and 896 tasks.

Copied to clipboard

Kai Wang, Xu Wang, Zhiyuan Fu, Yudong Zhang, Yang Wang

Challenge:	Existing LLM agents are brittle in open-ended environments due to two limitations: 1) a closed action space; 2) myopic error recovery.
Approach:	They propose a novel architecture that augments the action space and revises global strategies by adding a reflective replanning mechanism to the system.
Outcome:	Experiments show that CAR outperforms baselines in a diagnostic benchmark with pruned toolsets to simulate tool scarcity.

FastDiSS: Few-step Match Many-step Diffusion Language Model on Sequence-to-Sequence Generation (2026.findings-acl)

Copied to clipboard

Dat Nguyen Cong, Tung Kieu, Hoang Thanh-Tung

GCIG: GraphRAG-based Cross-document Instruction Generation for Boosting LLM Reasoning (2026.findings-acl)

Challenge:	Existing models that correct errors in the model but lack a high quality of output . a novel training framework that matches inference noise to the model's inference signal improves performance .
Approach:	They propose a training framework that matches inference noise to model errors by perturbing the self-conditioning signal to match inference errors.
Outcome:	The proposed framework surpasses standard continuous diffusion models while providing 400x faster inference speed.

Copied to clipboard

Xiaoliang Xu, Huang Yuan, Junmei Wang, Can Xu

Towards General Agentic Intelligence via Environment Scaling (2026.findings-acl)

Challenge:	Existing methods for fine-tuning large language models struggle in knowledge-intensive domains and complex reasoning tasks due to their limited coverage of single-document knowledge and repetitive content.
Approach:	They propose a GraphRAG-based cross-document instruction generation framework that generates diverse questions through task-aware prompts and context-sensitive retrieval.
Outcome:	The proposed framework outperforms existing methods on knowledge-intensive and multi-hop question-answering tasks.

Copied to clipboard

Runnan Fang, Shihao Cai, Baixuan Li, Jialong Wu, Guangyu Li, Wenbiao Yin, Xinyu Wang, Xiaobin Wang, Liangcai Su, Zhen Zhang, Shibin Wu, Zhengwei Tao, Yong Jiang, Pengjun Xie, Ningyu Zhang, Fei Huang, Wentao Zhang, Jingren Zhou

EAIR: Entity-aware Inference-Time Knowledge Routing for Multi-Hop Knowledge Editing (2026.findings-acl)

Challenge:	Diverse real-world APIs require precise, robust function-calling intelligence, which needs agents to develop these capabilities through interaction in varied environments.
Approach:	They propose a framework that scales up environments to enable agentic intelligence . they use a two-phase agent fine-tuning strategy to first endow agents with basic agentic capabilities, then specializing them for domain-specific contexts.
Outcome:	Experiments on -bench, -Bench, and ACEBench show that the model significantly enhances the models’ function-calling capability.

Copied to clipboard

Jungyu Lee, Kunhui Lee, Gyun Lee, Seung-Hoon Na

Preventing Safety Drift in Large Language Models via Coupled Weight and Activation Constraints (2026.findings-acl)

Challenge:	Existing in-context knowledge editing methods suffer from paraphrase sensitivity . Existing methods interfere with the model's original knowledge and can degrade its inherent capabilities.
Approach:	They propose a novel entity-aware inference-time knowledge routing method to address paraphrase sensitivity.
Outcome:	The proposed method achieves highest strict case accuracy in 11 of 12 settings, significantly reducing paraphrase sensitivity.

Copied to clipboard

Songping Peng, Zhiheng Zhang, Daojian Zeng, Lincheng Jiang, Xieping Gao

GAPO: Robust Advantage Estimation for Real-World Code LLMs (2026.findings-acl)

Challenge:	Existing defenses constrain either weights or activations in isolation, without considering their coupled effects on safety.
Approach:	They propose a weight-activation constraint that enforces a precomputed safety subspace on weight updates and applies regularization to safety-critical features identified by sparse autoencoders.
Outcome:	The proposed model outperforms baselines even under high harmful data ratios.

Copied to clipboard

Jianqing Zhang, Zhezheng Hao, Wei Xia, Hande Dong, Hong Wang, Chenxing Wei, Yuyan Zhou, Yubin Qi, Qiang Lin, Jian Cao

What’s Missing in Screen-to-Action? Towards a UI-in-the-Loop Paradigm for Multimodal GUI Reasoning (2026.findings-acl)

Challenge:	Reinforcement learning (RL) is widely used for post-training large language models (LLMs) in code editing, but in real-world code editing scenarios, reward distributions are often skewed with unpredictable noise, leading to distorted advantage computation and increased rollout outliers.
Approach:	They propose a group-relative method that finds an interval with the highest SNR and uses the median of that interval as an adaptive Q to replace the group mean in advantage calculation.
Outcome:	The proposed method improves on nine instruction-tuned LLMs while remaining plug-and-play and efficient.

Copied to clipboard

Songze Li, Xiaoke Guo, Tianqi Liu, Biao Yi, Zhaoyan Gong, Zhiqiang Liu, Huajun Chen, Wen Zhang

When to Trust Tools? Adaptive Tool Trust Calibration For Tool-Integrated Math Reasoning (2026.findings-acl)

Challenge:	Existing GUI reasoning methods rely on direct screen-based decision-making, which lacks interpretability and overlooks a comprehensive understanding of UI elements, ultimately leading to task failure.
Approach:	They propose a GUI reasoning paradigm that treats the GUI reasoning task as a cyclic *Screen-UI elements-Action process.
Outcome:	The proposed paradigm achieves state-of-the-art UI understanding performance while yielding superior results in GUI reasoning tasks.

Copied to clipboard

Ruotao Xu, Yixin Ji, Yu Luo, Jinpeng Li, Dong Li, Peifeng Li, Juntao Li, Min Zhang

V-MAGE: A Game Evaluation Framework for Assessing Vision-Centric Capabilities in Multimodal Large Language Models (2026.findings-acl)

Challenge:	Large reasoning models have performance enhancements but still suffer from shortcomings due to limitations of the underlying language models.
Approach:	They propose a framework that allows the model to choose when to trust or ignore the tool results based on the confidence score of generated code blocks.
Outcome:	The proposed framework reduces the "Tool Ignored" issue by 4.1% to 7.5%.

Copied to clipboard

Xiangxi Zheng, Linjie Li, Zhengyuan Yang, Ping Yu, Alex Jinpeng Wang, Rui Yan, Yuan Yao, Lijuan Wang

Understanding and Preventing Entropy Collapse in RLVR with On-Policy Entropy Flow Optimization (2026.findings-acl)

Challenge:	Existing static image-text benchmarks are insufficient for evaluating multimodal large language models’ dynamic perception and interactive reasoning abilities.
Approach:	They propose a game-based evaluation framework to assess multimodal large language models’ visual reasoning in dynamic, continuous-space environments.
Outcome:	The proposed framework systematically assesses MLLMs’ visual reasoning in dynamic, continuous-space environments.

Copied to clipboard

Huimin Xu, Shuai Zhao, Xiaobao Wu, Anh Tuan Luu

Reinforcement Learning with Semantic Rewards Enables Low-Resource Language Expansion without Alignment Tax (2026.findings-acl)

Challenge:	Existing RLVR algorithms suffer from entropy collapse, leading to premature determinism and unstable optimization.
Approach:	They propose an adaptive entropy flow balancing mechanism that rescales entropic-increasing and enotro-decreazing updates according to their contributions to enthroy change.
Outcome:	The proposed method outperforms existing RLVR algorithms on six reasoning benchmarks.

Copied to clipboard

Zeli Su, Ziyin Zhang, Zhou Liu, Xuexian Song, Zhankai Xu, Longfei Zheng, Xiaolu Zhang, Rong Fu, Guixian Xu, Wentao Zhang

MEIC-DT: Memory-Efficient Incremental Clustering for Long-Text Coreference Resolution with Dual-Threshold Constraints (2026.findings-acl)

Challenge:	Extending large language models to low-resource languages often incurs an "alignment tax" token-level fine-tuning enforces token-level surface imitation on narrow and biased data distributions.
Approach:	They propose a semantic-space alignment paradigm powered by group-level semantic rewards instead of likelihood maximization.
Outcome:	The proposed model acquires low-resource capa- bilities while mitigating alignment tax on Tibetan–Chinese machine translation and Ti- betan headline generation.

Copied to clipboard

Kangyang Luo, Shuzheng Si, Yuzhuo Bai, Cheng Gao, Zhitong Wang, Cheng Huang, Yingli Shen, Yufeng Han, Wenhao Li, Cunliang Kong, Maosong Sun

Lost in Diffusion: Uncovering Hallucination Patterns and Failure Modes in Diffusion Large Language Models (2026.findings-acl)

Challenge:	Existing supervised neural methods are underexplored for coreference resolution, especially in incremental clustering.
Approach:	They propose a dual-threshold incremental clustering approach based on a lightweight Transformer.
Outcome:	Experiments on common benchmarks show that MEIC-DT achieves highly competitive coreference performance under stringent memory constraints.

Copied to clipboard

Zhengnan Guo, Fei Tan

FARSS: Fisher-Optimized Adaptive Low-Rank and Singular-Vector Selection for Knowledge-Preserving Fine-Tuning (2026.findings-acl)

Challenge:	dLLMs have emerged as a promising non-autoregressive paradigm for text generation, but their hallucination mechanisms remain underexplored.
Approach:	They present the first controlled comparative study to evaluate hallucination patterns in Diffusion Large Language Models.
Outcome:	The proposed model exhibits higher propensity for hallucination than AR counterparts controlled for architecture, scale, and pre-training weights.

Copied to clipboard

Renxing Chen, Ziwei Xiang, Peisong Wang, Hongjian Fang, Meng Li, Fanhu Zeng, Yanan Zhu, Peipei Yang, Xu-Yao Zhang, Jian Cheng

Align Documents to Questions: Question-Oriented Document Rewriting for Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	Low-rank adaptation methods for large language models have limitations in preserving world knowledge and limiting updates to preserve world knowledge.
Approach:	They propose a Fisher-optimized adaptive low Rank and Singular-VectorSelection framework for knowledge-preserving fine-tuning that allows efficient and task-sensitive updates.
Outcome:	The proposed framework outperforms existing methods for knowledge-preserving fine-tuning.

Copied to clipboard

Jiaang Li, Zhendong Mao, Quan Wang, Yuning Wan, Yongdong Zhang

GlimpRouter: Efficient Collaborative Inference by Glimpsing One Token of Thoughts (2026.findings-acl)

Challenge:	Retrieval-Augmented Generation (RAG) enhances the factuality of Large Language Models (LLMs) however, LLMs exhibit a stylistic bias when presented with mixed contexts, revealing a bottleneck in their utility.
Approach:	They propose a style-controlled rewriter that aligns retrieved documents with a question-oriented style while preserving facts.
Outcome:	The proposed model improves RAG pipelines by 8% with negligible latency overhead.

Copied to clipboard

Wenhao Zeng, Xuteng Zhang, Yuling Shi, Chao Hu, Yuting Chen, Beijun Shen, Xiaodong Gu

Plug-and-Play Data Module for Code RL: Adaptive Ambiguity Replay (2026.findings-acl)

Challenge:	Existing routing strategies rely on local token probabilities or post-hoc verification, introducing significant inference overhead.
Approach:	They propose a step-wise collaboration framework that generates only the first token of each reasoning step and routes it to a larger model only when initial token entropy exceeds a threshold.
Outcome:	The proposed approach reduces inference latency while preserving accuracy.

Copied to clipboard

Jianqing Zhang, Wei Xia, Zhezheng Hao, Hong Wang, Hande Dong, Qiang Lin, Yang Liu, Jian Cao, Qiang Yang

Can AI Revise Research Papers with Human Review Feedback? An Empirical Study and Benchmark (2026.findings-acl)

Challenge:	Existing approaches to reinforcement learning (RL) rely on static, in-epoch metrics that overlook training dynamics, often introducing low-utility or outdated data.
Approach:	They propose a plug-and-play module that prioritizes cross-epoch ambiguous samples to neutralize the noise from stale experiences.
Outcome:	Extensive experiments on nine LLMs show that Adaptive Ambiguity Replay outperforms state-of-the-art baselines on real-world code editing tasks.

Copied to clipboard

Zihan Luo, Hong Huang, Jianxun Lian, Yu Chang, Xing Xie, Hai Jin

Dunhuang-Bench: How Well Do MLLMs Understand Cultural Heritage? (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are fundamentally reshaping the scientific landscape, transitioning the role of AI from passive tools to active partners within a new paradigm of Human-AI collaboration.
Approach:	They propose a benchmark to evaluate the ability of Large Language Models to improve papers with human feedback.
Outcome:	The proposed benchmark tests the skills of Large Language Models (LLMs) on paper interpretation, experimental implementation, and paper formulation, using authors’ camera-ready versions as natural human baselines.

Copied to clipboard

Junyi Yuan, Jian Zhang, Tianxiu Yu, Yanlin Zhou, Xiaobo Jin, Qiufeng Wang, Fangyu Wu

Multi-Docker-Eval: A ‘Shovel of the Gold Rush’ Benchmark on Automatic Environment Building for Software Engineering (2026.findings-acl)

Challenge:	Recent advances in Multimodal Large Language Models (MLLMs) have led to extensive evaluations on Chinese cultural benchmarks.
Approach:	They construct a large-scale benchmark comprising 486 images and 22,970 QA pairs to evaluate MLLMs' cultural understanding.
Outcome:	The proposed benchmark incorporates three task formats to evaluate MLLMs’ cultural understanding: Question Answering with Text Description, Multi-turn Dialogue, and Question Answers with Choices.

Copied to clipboard

Kelin Fu, Tianyu Liu, Zeyu Shang, Yingwei MA, Jiaheng Liu, Jian Yang, Kaigui Bian

Not All Modalities at Once: Dynamic Dropout and Bidirectional Fusion for Robust Multi-modal Knowledge Graph Completion (2026.findings-acl)

Challenge:	Automated environment configuration is a critical bottleneck in scaling software engineering (SWE) automation.
Approach:	They propose a reliable evaluation standard for automated environment configuration for 40 real-world repositories spanning 9 programming languages.
Outcome:	The proposed benchmark includes 40 real-world repositories spanning 9 programming languages and measures success in achieving executable states and efficiency under realistic constraints.

Copied to clipboard

Jiashun Peng, Fu Zhang, Hongzhi Chen, Jingwei Cheng, Yingsong Ning, Xiaoke Wang

If Only My CGM Could Speak: A Privacy-Preserving Agent for Question Answering over Continuous Glucose Data (2026.findings-acl)

Challenge:	Existing MKGC methods train with all modalities available, implicitly assuming consistent complementarity . however, this often induces modality dependence and modality competition under heterogeneous noise, which can hinder robust multi-modal fusion and limit overall performance.
Approach:	They propose a framework to infer missing links in multimodal knowledge graphs by leveraging structured triples together with auxiliary modalities such as text and images.
Outcome:	The proposed framework outperforms baselines and achieves new state-of-the-art results.

Copied to clipboard

Yanjun Cui, Ali Emami, Temiloluwa Prioleau, Nikhil Singh

E-ABSA20K: A Dataset and Propose-and-Verify for Aspect-Based Sentiment Analysis in Long E-commerce Reviews (2026.findings-acl)

Challenge:	Current patient platforms only offer static summaries which do not support inquisitive user queries.
Approach:	They propose a framework for question answering over personal glucose data that uses large language models to provide a reasoning engine that selects analytical functions.
Outcome:	The proposed framework achieves 94% value accuracy on synthetic queries and 88% on ambiguous real-world queries.

Copied to clipboard

Tong Sun, Mingyang Ma, Cheng Yu

ECHA: Jailbreaking LVLMs via the Mismatch between Implicit Semantic Reconstruction and Explicit Safety Alignment (2026.findings-acl)

Challenge:	Aspect-based Sentiment Analysis (ABSA) is critical for extracting actionable product insights from e-commerce reviews.
Approach:	They propose a framework that decomposes ABSA into two stages to extract review-level quadruple reviews from 20K reviews from four product categories.
Outcome:	The proposed framework outperforms existing benchmarks and single-stage prompting and competitive ABSA extraction baselines.

Copied to clipboard

Chenxing Xu, Junyong Jiang, Zehu Zhang, Lu Dong

Thinking with Map: Reinforced Parallel Map-Augmented Agent for Geolocalization (2026.findings-acl)

Challenge:	Existing safety guardrails fail to intercept latent intent, whereas LVLMs can implicitly synthesize holistic malicious semantics from fragmented visual cues.
Approach:	They propose an Emoji Chain Hinting Attack (ECHA) framework that decouples sensitive concepts into semantically related emoji chains and structural text masks.
Outcome:	The proposed framework outperforms existing baselines and bypasses safety guardrails in over 81% of instances with a single attempt.

Copied to clipboard

Yuxiang Ji, Yong Wang, Ziyu Ma, Yiming Hu, Hailang Huang, Xuecai Hu, Guanhua Chen, Liaoni Wu, Xiangxiang Chu

Revisiting Entropy Regularization: Adaptive Coefficient Unlocks Its Potential for LLM Reinforcement Learning (2026.findings-acl)

Challenge:	Existing large vision-language model (LVLM) approaches overlook a common strategy used by humans — using maps.
Approach:	They propose a method to equip a vision-language model with the ability to think with maps and optimize it using agentic reinforcement learning and parallel test-time scaling.
Outcome:	The proposed method outperforms open- and closed-source models on most metrics.

Copied to clipboard

Xiaoyun Zhang, Xiaojian Yuan, Di Huang, Wang You, Chen Hu, Jingqing Ruan, Kejiang Chen, Xing Hu

SLIP: Soft Label Mechanism and Key-Extraction-Guided CoT-based Defense Against Instruction Backdoor in APIs (2026.findings-acl)

Challenge:	Reasoning ability is a defining capability of Large Language Models (LLMs), but RLVR training suffers from policy entropy collapse, hindering exploration and limiting reasoning performance.
Approach:	They propose a framework that dynamically balances exploration and exploitation via three components: difficulty-aware coefficient allocation, initial-anchored target entropy, and dynamic global coefficient adjustment.
Outcome:	The proposed framework outperforms baselines on multiple mathematical reasoning benchmarks.

Copied to clipboard

Wu Zhengxian, Juan Wen, Wanli Peng, Haowei Chang, Yinghan Zhou, Xue Yiming

U-Fold: Dynamic Intent-Aware Context Folding for User-Centric Agents (2026.findings-acl)

Challenge:	Existing black-box instruction backdoors can detect poisoned inputs, but fail to recover correct outputs once the backdoor is activated.
Approach:	They propose a soft label mechanism and key-extraction-guided CoT-based defense against instruction backdoors in APIs (SLIP) they propose KCOT-based model to extract task-relevant keywords and phrases rather than only considering the single trigger or overall text semantics.
Outcome:	The proposed model reduces the average attack success rate to 25.13% and improves clean accuracy to 87.15% and outperforms state-of-the-art black-box defenses.

Copied to clipboard

Jin Su, Runnan Fang, Yeqiu Li, Xiaobin Wang, Shihao Cai, Pengjun Xie, Ningyu Zhang, Fajie Yuan

AdaTooler-V: Adaptive Tool-Use for Images and Videos (2026.findings-acl)

Challenge:	Existing context-folding methods are designed for single-query or single-intent scenarios.
Approach:	They propose a dynamic context-folding framework tailored to user-centric tasks that preserves fine-grained information through dynamic context folding.
Outcome:	The proposed framework outperforms ReAct and previous folding frameworks on long, noisy tasks.

Copied to clipboard

Chaoyang Wang, Kaituo Feng, Dongyang Chen, Zhongyu Wang, Zhixun Li, Sicheng Gao, Meng Meng, Xu Zhou, Manyuan Zhang, Yuzhang Shang, Xiangyu Yue

Exposing Privacy Risks in Graph Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	Existing models exhibit blind tool-use reasoning patterns, which significantly increases inference overhead and degrades model performance.
Approach:	They propose an MLLM that performs adaptive tool-use by determining whether a visual problem truly requires tools.
Outcome:	The proposed model outperforms existing methods in visual reasoning tasks.

Copied to clipboard

Jiale Liu, Jiahao Zhang, Suhang Wang

BalanceSFT: Improving LLM Function Calling with Balanced Training Signals and Data Hardness (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have limitations such as generating factually incorrect information (hallucinations) Retrieval-Augmented Generation (RAG) is a powerful paradigm for enhancing LLMs with external, up-to-date knowledge.
Approach:	They investigate the data extraction vulnerabilities of Graph RAG systems by executing tailored attacks on them.
Outcome:	The proposed attacks exploit the vulnerability of Graph RAG systems to leak raw text and structured data.

Copied to clipboard

Bingguang Hao, Zengzhuang Xu, Maolin Wang, Yuntao Wen, Yicheng Chen, Cunyin Peng, Long Chen, Xiangyu Zhao, Jinjie Gu, Chenyi Zhuang, Ji Zhang

How Robust Are Large Language Models for Clinical Numeracy? An Empirical Study on Numerical Reasoning Abilities in Clinical Contexts (2026.findings-acl)

Challenge:	Currently, Supervised Fine-Tuning (SFT) is the prevailing method for equipping Large Language Models (LLMs) with function calling capabilities, but its effectiveness is often compromised by two challenges: 1) lengthy Chain-of-Thought (CoT) reasoning tokens dominate training signals over concise function calls in the learning objective; 2) scarcity of hard training examples.
Approach:	They propose a framework that uses a self-adjusted signal balancing loss and a hard data re-sampling strategy to selectively generate new, high-quality complex data guided by model errors.
Outcome:	The proposed framework surpasses state-of-the-art models like GPT-5 in function calling performance.

Copied to clipboard

Minh-Vuong Nguyen, Fatemeh Shiri, Zhuang Li, Karin Verspoor

Safety-Aware Dialogue System for Postoperative Oral Cancer Care with Structured Clarification and a Clinically Curated Dataset (2026.findings-acl)

Challenge:	Existing evaluations of Large Language Models for clinical numerical reasoning provide limited operation-level coverage and limited robustness of numerical understanding across clinical note formats.
Approach:	They propose a benchmarking tool that evaluates four main types of clinical numeracy . they present longitudinal MIMIC-IV vital-sign records in three semantically equivalent representations .
Outcome:	The proposed benchmark evaluates four main types of clinical numeracy: value retrieval, arithmetic computation, relational comparison, and aggregation.

Copied to clipboard

Tzu-Chi Liu, Hui-Ying Yang, Shiow-Ching Shun, Yu-Chi Chen, Lu-Yen Anny Chen, Yong-Sheng Chen

FTibSuite: A Comprehensive Resource Suite for Tibetan Vision–Language Modeling (2026.findings-acl)

Challenge:	Clinical dialogue systems can enhance patient education and follow-up care by providing brief and subjective messages that lack critical clinical context.
Approach:	They propose a safety-aware dialogue system that applies information-gain guided clarification before RAG-based response generation and screens user utterances for emotional distress and suicidal ideation.
Outcome:	The proposed system improves quality and clinical appropriateness relative to strong baselines while aligning with expert judgments on clinically concerning utterances.

Copied to clipboard

Guixian Xu, Yide Liang, Zeli Su, Xuexian Song, Ziyin Zhang, Yushuang Dong, Ting Zhang, Xu Han

DeepMed: Building a Medical DeepResearch Agent via Multi-hop Med-Search Data and Turn-Controlled Agentic Training & Inference (2026.findings-acl)

Challenge:	FTibSuite provides an end-to-end training-and-evaluation workflow for vision–language models . Tibetan is underserved due to the lack of infrastructure for reproducible training and evaluation.
Approach:	They propose a resource-centric workflow for Tibetan VLMs that provides an end-to-end training-and-evaluation workflow and human-verified multimodal annotations.
Outcome:	FTibSuite provides an end-to-end training-and-evaluation workflow and human-verified multimodal annotations.

Copied to clipboard

Zihan Wang, Hao Wang, Shi Feng, Xiaocui Yang, Daling Wang, Yiqun Zhang, Jinghao Lin, Xiaozhong Ji, Haihua Yang

Softpick: No Attention Sink, No Massive Activations with Rectified Softmax (2026.findings-acl)

Challenge:	Medical reasoning models are constrained by parametric knowledge and can induce hallucinations and spurious attributions.
Approach:	They propose a model that uses a multi-hop med-search QA synthesis method to apply the DR paradigm in medical contexts.
Outcome:	The proposed model outperforms larger medical reasoning models on medical benchmarks.

Copied to clipboard

Zayd Muhammad Kawakibi Zuhri, Erland Hilman Fuadi, Alham Fikri Aji

UniCM: A Unified Consistency Model For Efficient Multimodal Generation and Understanding (2026.findings-acl)

Challenge:	Quantized models using softpick outperform softmax on standard benchmarks . softmax is widely used in statistics and especially in machine learning .
Approach:	They introduce a rectified, not sum-to-one, drop-in replacement for softmax in transformer attention mechanisms that eliminates attention sink and massive activations.
Outcome:	The proposed model outperforms softmax on benchmarks with lower bit precisions.

Copied to clipboard

Chenkai Xu, Xu Wang, Zhenyi Liao, Yishun Li, TianQi Hou, Zhijie Deng

Can Multi-agent Help Disambiguation in Multi-domain Translation? (2026.findings-acl)

Challenge:	Consistency models (CMs) have shown promise in the efficient generation of both image and text.
Approach:	They propose to use a discrete token for both image and text generation to achieve a unified denoising perspective.
Outcome:	The proposed model outperforms SD3 on GenEval and Image Reward while being 1.5 faster at long-sequence generating speed.

Copied to clipboard

Zhibo Man, Shaoyang Xu, Yujie Zhang, Yi Feng, Yuanmeng Chen, Yufeng Chen, Xu Jinan, Wenxuan Zhang

Agent-Dice: Disentangling Knowledge Updates via Geometric Consensus for Agent Continual Learning (2026.findings-acl)

Challenge:	Existing multi-agent systems have shown strong potential for machine translation (MT) but their performance in multidomain translation remains unsatisfactory due to cross-domain word ambiguity .
Approach:	They propose a multi-agent collaborative disambiguation framework for MDT that leverages the collaborative capabilities of LLMs for disambiguations.
Outcome:	The proposed framework improves translation performance across multiple domains and improves disambiguation accuracy.

Copied to clipboard

Zheng Wu, Xingyu Lou, Xinbei Ma, Yansi Li, Weiwen Liu, Weinan Zhang, Jun Wang, Zhuosheng Zhang

Automatic Slide Updating with User-Defined Dynamic Templates and Natural Language Instructions (2026.findings-acl)

Challenge:	Large Language Model (LLM)-based agents extend the utility of LLMs by interacting with dynamic environments.
Approach:	They propose a parameter fusion framework based on directional consensus evaluation that disentangles knowledge updates through a two-stage process.
Outcome:	The proposed framework disentangles knowledge updates through a two-stage process with minimal computational overhead and parameter updates.

Copied to clipboard

Kun Zhou, Jiakai He, Wenmian Yang, Zhensheng Wang, Yiquan Zhang, Weijia Jia

Challenge:	Existing automation methods follow fixed template filling and cannot support dynamic updates for diverse, user-authored decks.
Approach:	They propose a framework that combines multimodal slide parsing, natural language instruction grounding, and tool-augmented reasoning for tables, charts, and textual conclusions.
Outcome:	The proposed framework updates content while preserving layout and style while maintaining a strong reference baseline on DynaSlide.

PretrainRL: Alleviating Factuality Hallucination of Large Language Models at the Beginning (2026.findings-acl)

Copied to clipboard

Langming Liu, Kangtao Lv, Haibin Chen, Weidong Zhang, Yejing Wang, Shilei Liu, Xin Tong, Yujin Yuan, Yongwei Wang, Wenbo Su, Bo Zheng

OMHBench: Benchmarking Balanced and Grounded Omni-Modal Multi-Hop Reasoning (2026.findings-acl)

Challenge:	Large language models suffer from factual hallucinations where they generate verifiable falsehoods.
Approach:	They propose a framework that integrates reinforcement learning into the pretraining phase to consolidate factual knowledge.
Outcome:	The proposed framework significantly alleviates factual hallucinations and outperforms state-of-the-art methods.

Copied to clipboard

Seunghee Kim, Ingyu Bang, Seokgyu Jang, Changhyeon Kim, Sanghwan Bae, Jihun Choi, Richeng Xuan, Taeuk Kim

Beyond Uniform SVD: Dual-Level Optimization across Columns and Modules for LLM Compression (2026.findings-acl)

Challenge:	Existing evaluation frameworks for multimodal large language models suffer from limitations . modality shortcuts and biased reasoning paths are common in such models .
Approach:	a new benchmark evaluates omni-modal multi-hop reasoning using 6,144 questions . authors propose OMHBench to address these limitations by comparing modalities .
Outcome:	OMHBench evaluates omni-modal multi-hop reasoning on 6,144 questions with balanced reasoning paths . evaluation of 13 state-of-the-art models shows performance gap exists between MLLMs and open-source models .

Copied to clipboard

Lin Xv, Xian Gao, Ting Liu, Yuzhuo fu

RSA-Bench: Benchmarking Audio Large Models in Real-World Acoustic Scenarios (2026.findings-acl)

Challenge:	Existing methods for low-rank decomposition overlook decomposing errors and suboptimal approximation.
Approach:	They propose a low-rank decomposition framework that integrates low-level optimization at column and module levels.
Outcome:	The proposed framework outperforms state-of-the-art methods and baselines in SVD and pruning.

Copied to clipboard

Yibo Zhang, Kaiwen Luo, Liang Lin, Shilinlu Yan, Jin Wang, Yaoqi Guo, Yitian Chen, Yalan Qin, Zhenhong Zhou, Kun Wang, Li Sun

MedLayBench-V: A Large-Scale Benchmark for Expert-Lay Semantic Alignment in Medical Vision Language Models (2026.findings-acl)

Challenge:	Existing evaluations rely on synthetic Gaussian noise or simplistic single-source interference, failing to capture the intricate, multi-layered acoustic dynamics that characterize authentic physical environments.
Approach:	They propose a robustness benchmark to stress-test Audio Large Models (ALLMs) using high-fidelity auditory scene simulations.
Outcome:	The proposed model performs well on a wide range of tasks, including automatic speech recognition, speech translation, and audio-based reasoning.

Copied to clipboard

Han Jang, Junhyeok Lee, Heeseong Eum, Kyu Sung Choi

Divergent Thinking: Escape the Homogeneity Trap in Generative Commonsense Reasoning (2026.findings-acl)

Challenge:	Medical Vision-Language Models are predominantly trained on professional literature, limiting their ability to communicate findings in the lay register required for patient-centered care.
Approach:	They propose a multimodal benchmark dedicated to expert-lay semantic alignment that enforces strict semantic equivalence by integrating unified medical language system (UMS) Concept Unique Identifiers (CUIs) with micro-level entity constraints.
Outcome:	The proposed benchmark enforces strict semantic equivalence by integrating unified medical language system (UMLS) Concept Unique Identifiers (CUIs) with micro-level entity constraints.

Copied to clipboard

Yiheng Tao, Kaiwen Cheng, Zhiwei Nie, Chang Liu, Jie Chen

SafeSteer: A Decoding-level Defense Mechanism for Multimodal Large Language Models (2026.findings-acl)

Challenge:	Generative commonsense reasoning requires models to synthesize coherent narratives that satisfy lexical constraints and commonsensical logic.
Approach:	They propose a framework that allows for deep semantic diversity rather than surface-level lexical variation.
Outcome:	The proposed framework achieves over 10% improvement in overall accuracy on NoRa and SPICE score on CommonGen-Lite.

Copied to clipboard

Xinyi Zeng, Xue Yang, Jingyuan Zhang, Huanqian Yan, Xiang Chen, Kaiwen Wei, Hankun Kang, Yu Tian

SEMA-RAG: A Self-Evolving Multi-Agent Retrieval-Augmented Generation Framework for Medical Reasoning (2026.findings-acl)

Challenge:	Existing defense methods rely on fine-tuning or inefficient post-hoc interventions, limiting their ability to address novel attacks.
Approach:	They propose a decoding-level defense mechanism that employs a lightweight discriminator to iteratively steer the decoding process toward safety.
Outcome:	The proposed method improves safety performance by up to 33.40% without fine-tuning on multiple MLLMs.

Copied to clipboard

Yongfeng Huang, Ruiying Chen, James Cheng

On the Step Length Confounding in LLM Reasoning Data Selection (2026.findings-acl)

Challenge:	Standard RAG frameworks treat retrieval as a static, single-round auxiliary step . compressed workflow makes it difficult to form reliable evidence chains .
Approach:	They propose a framework that decouples tasks and allows for dynamic multi-round exploration . they propose retrieval-augmented generation (RAG) to mitigate hallucinations and knowledge obsolescence .
Outcome:	The proposed framework improves the strongest baseline by +6.46 accuracy points on average across five benchmarks and five LLM backbones.

Copied to clipboard

Bing Wang, Rui Miao, Chen Shen, Shaotian Yan, Kaiyuan Liu, Ximing Li, Xiaosong Yuan, Sinan Fan, Jun Zhang, Jieping Ye

SLIM: Stealthy Low-Coverage Black-Box Watermarking via Latent-Space Confusion Zones (2026.findings-acl)

Challenge:	Existing pipelines generate long reasoning data from more capable Large Language Models (LLMs) and apply manually heuristic or naturalness-based selection methods to filter high-quality samples.
Approach:	They propose to use supervised fine-tuning to generate long reasoning data from more capable Large Language Models and apply manually heuristic or naturalness-based selection methods to filter high-quality samples.
Outcome:	Experiments on four LLMs and five evaluation benchmarks show that the proposed approach is effective in mitigating step length confounding problem.

Copied to clipboard

Hengyu WU, Yang Cao

SQL-ASTRA: Alleviating Sparse Feedback in Agentic SQL via Column-Set Matching and Trajectory Aggregation (2026.findings-acl)

Challenge:	Training data is a critical asset in Large Language Model (LLM) development and is often proprietary.
Approach:	They propose a framework that allows per-user data provenance verification under strict black-box access.
Outcome:	The proposed framework achieves ultra-low coverage capability, strong black-box verification performance, and great scalability while preserving both stealthiness and model utility.

Copied to clipboard

Long Li, Zhijian Zhou, Jiangxuan Long, Peiyang Liu, Weidi Xu, Zhe Wang, Shirui Pan, Chao Qu

Rotation Control Unlearning: Quantifying and Controlling Continuous Unlearning for LLM with The Cognitive Rotation Space (2026.findings-acl)

Challenge:	Agentic SQL is a framework for multiturn agent learning, but it is limited to single-turn paradigms.
Approach:	They propose a framework that provides a universal two-tiered reward mechanism for credit assignment . they propose 'Aggregated Trajectory Reward' to resolve multi-turn credit assignment.
Outcome:	The proposed framework outperforms SOTA Arctic-Text2SQL-R1-7B on BIRD and Spider 2.0 using identical models.

Copied to clipboard

Xiang Zhang, Kun Wei, Xu Yang, Jiahua Li, Su Yan, Cheng Deng

Challenge:	Existing methods to remove undesirable data from Large Language Models suffer from cumulative catastrophic utility loss under continuous unlearning requests.
Approach:	They propose a method that leverages the rotational salience weight of RCU to quantify and control the unlearning degree in the continuous unlearning process.
Outcome:	The proposed method achieves SOTA performance without a retained dataset.

SpecEdit: A Spectral Approach for Multi-Round Knowledge Editing (2026.findings-acl)

Copied to clipboard

Junxian Liu, Zikun Deng, Xin Wu, Dazhen Deng, Yi Cai

SAGE: Sign-Adaptive Gradient for Memory-Efficient LLM Optimization (2026.findings-acl)

Challenge:	Multi-round knowledge editing suffers from performance degradation as edits accumulate . intrinsic knowledge of model and historical edit memories are naively coupled during editing . SpecEdit improves model editing performance by reducing destructive coupling .
Approach:	They propose a spectral-based model editing module that integrates into existing editing methods without altering their original optimization procedures.
Outcome:	The proposed model improves performance on multiple LLMs and editing methods.

Copied to clipboard

Wooin Lee, Hyun-Tae Kim

DP3: Differentially Private Prompt Perturbation for Multi-turn LLM Inference (2026.findings-acl)

Challenge:	Existing methods to train LLMs consume memory equivalent to twice the model size, resulting in a hybrid design that reverts to AdamW and negates the memory gains.
Approach:	They propose a new, memory-efficient O(d) adaptive scale that replaces AdamW in a hybrid structure that combines a Lion-style update direction with a memory-saving adaptive scale.
Outcome:	The proposed model outperforms existing methods on LLMs up to 1.3B parameters while significantly reducing optimizer state memory.

Copied to clipboard

Lele Zheng, Chao Zhang, Feiyang Yuan, Ke Cheng, Tao Zhang, Anxiao Song, Yulong Shen

RAMP: Risk-Aware Multi-Turn Planning for Jailbreak Red-Teaming (2026.findings-acl)

Challenge:	Large language models (LLMs) are widely used for text understanding and generation . existing methods that assume single-turn interactions break down in multi-turn settings .
Approach:	They propose a differentially private prompt perturbation framework for multi-turn LLM inference . DP3 constructs a perturbation mapping table to reuse perturbations for recurring tokens .
Outcome:	The proposed framework reduces privacy costs and degrades cross-turn semantic coherence . it also provides a context-aware utility function to maintain semantic consistency across turns .

Copied to clipboard

Yize Liu, Yunyun Hou, Aina Sui

LexGenius: An Expert-Level Benchmark for Large Language Models in Legal General Intelligence (2026.findings-acl)

Challenge:	Existing methods for evaluating the safety of large language models rely on heuristic strategies or trained attack agents.
Approach:	They propose a method for multi-turn jailbreaking that iteratively plans and executes each turn via a Judge, a Transitioner, and a Planner.
Outcome:	The proposed framework achieves strong attack performance across open-source and closed-source target models while remaining effective under stricter turn budgets.

Copied to clipboard

Wenjin Liu, Haoran Luo, Xin Feng, Xiang Ji, Lijuan Zhou, Rui Mao, Jiapu Wang, Shirui Pan, Erik Cambria

From Attenuation to Attention: Variational Information Flow Manipulation for Fine-Grained Visual Perception (2026.findings-acl)

Challenge:	Existing benchmarks for legal general intelligence (GI) are result-oriented and do not evaluate the legal intelligence of large language models (LLMs).
Approach:	They propose a Chinese legal benchmark for evaluating legal GI in large language models . they use recent legal cases and exam questions to create multiple-choice questions .
Outcome:	The proposed benchmarks lack a systematic evaluation of the legal intelligence of large language models (LLMs) the results show that even the best LLMs lagging behind human legal professionals.

Copied to clipboard

Jilong Zhu, Yang Feng

VulAgent: Hypothesis-Validation Driven Multi-Agent Architecture for Vulnerability Detection (2026.findings-acl)

Challenge:	Existing input-centric solutions fail to reverse this intrinsic mechanism of information loss.
Approach:	They propose a Variational Information Flow framework that leverages a probabilistic perspective to model visual saliency relevant to the question-answer pair as a latent distribution.
Outcome:	The proposed framework improves general VQA, fine-grained perception and visual grounding.

Copied to clipboard

Ziliang Wang, Ge Li, Jia Li, Hao Zhu, Zhi Jin

Challenge:	Recent reports indicate that software vulnerabilities caused by insecure coding practices remain a major security threat.
Approach:	They propose a multi-agent vulnerability detection framework based on hypothesis validation . they use multi-view analyzers to localize and localize security-sensitive operations .
Outcome:	The proposed framework reduces false positives and increases accuracy by 6.6 percentage points on PrimeVul and SVEN.

From Attack Surfaces to Actual Operations: A Survey of Modern LLM Jailbreaks (2026.findings-acl)

Copied to clipboard

Ruikang Zhou, Changsheng Sun, Mark Huasong Meng

ARC: Active and Reflection-driven Context Management for Long-Horizon Information Seeking Agents (2026.findings-acl)

Challenge:	Existing taxonomies focus on manipulation methods rather than underlying mechanisms, limiting our understanding of attack effectiveness and defensive strategies.
Approach:	They propose a two-fold taxonomy to categorize attacks across three tiers based on exploited vulnerabilities and approaches and an operational taxonomies to evaluate attacks across four dimensions.
Outcome:	The proposed taxonomy categorizes attacks across three tiers based on exploited vulnerabilities and approaches and evaluates attacks on four dimensions to assess real-world feasibility and sustainability.

Copied to clipboard

Yilun Yao, Shan Huang, Elsie Dai, Zhewen Tan, Zhenyu Duan, Shousheng Jia, Yanbing Jiang, Tong Yang

ToolDNA: Autonomous Evolution of Tool Metadata for Robust Dialogue Agents (2026.findings-acl)

Challenge:	Existing approaches to managing context are based on raw accumulation or passive summarization, treating it as static artifact and allowing early errors or misplaced emphasis to persist.
Approach:	They propose a framework that treats context as a dynamic internal reasoning state during execution.
Outcome:	Experiments on long-horizon information-seeking benchmarks show that ARC outperforms passive context compression methods.

Copied to clipboard

Qiuyuan Ai, Cong Wang, Jiaqi Zhang, Zengxin Han, Jie Song

ExpertIVS: Sociological Expert Driven Individual Value Simulation in Large Language Models (2026.findings-acl)

Challenge:	Task-oriented dialogue systems face labor-intensive manual metadata tuning and sparse reinforcement learning (RL) rewards that fail to diagnose invocation errors.
Approach:	They propose a framework that enables auto-evolution of policy networks and tool metadata via RL . a tool metadata loop coordinates metadata through policy-generated candidates during rollouts .
Outcome:	The proposed framework achieves +11% problem resolution and +54% accuracy over commercial LLMs with prompt engineering and +25%/+35% over supervised fine-tuning.

Copied to clipboard

Zhen Wang, Yuqi Ren, Yuehan Cui, Hongxiang Wang, Jianxiang Peng, Zhaoxia Zhang, Bingkun Zhu, Tongxuan Zhang, Dezhi Tong, Deyi Xiong

Be Your Own Red Teamer: Safety Alignment via Self-Play and Reflective Experience Replay (2026.findings-acl)

Challenge:	Existing methods for social simulations mechanically stitch survey responses into prompts, which suffer from semantic fragmentation, failing to capture the internal coherence of human value systems.
Approach:	They propose a framework employing 14 Sociological Expert Agents to interpret World Values Survey responses through structured professional perspectives rather than direct responses concatenation.
Outcome:	Experiments on 480 individuals from 12 countries show that ExpertIVS outperforms baselines in value generalization and significantly outperfies the existing methods.

Copied to clipboard

Hao Wang, Yanting Wang, Hao Li, Rui Li, Lei Sha

RetentiveKV: State-Space Memory for Uncertainty-Aware Multimodal KV Cache Eviction (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have remarkable capabilities but are vulnerable to adversarial “jailbreak” attacks designed to bypass safety guardrails.
Approach:	They propose to empower a large language model to be its own red teamer . safety self-play allows the model to act as both the Attacker and Defender .
Outcome:	The proposed approach outperforms baselines trained on static adversarial datasets and establishes a new benchmark for proactive safety alignment.

Copied to clipboard

Sihao Liu, YuFan Xiong, Zhonghua Jiang, Zhaode Wang, Chengfei lv, Shengyu Zhang

RCBSF: A Multi-Agent Framework for Automated Contract Revision via Stackelberg Game (2026.findings-acl)

Challenge:	Existing methods for evicting KV pairs rely on the "persistence of importance" hypothesis . visual tokens display "deferred importance" but become pivotal during later decoding, authors say .
Approach:	They propose an entropy-driven method that reformulates KV eviction from "discrete context truncation" to "continuous memory evolution" they propose to prune visual tokens with "deferred importance" visual token exhibiting low salience but becoming pivotal during later decoding .
Outcome:	The proposed method achieves 5.0 KV cache compression and 1.5 decoding acceleration.

Copied to clipboard

Shijia Xu, Yu Wang, Xiaolong Jia, Zhou Wu, Kai Liu, April Xiaowen Dong

Challenge:	despite the adoption of Large Language Models (LLMs), contract revision remains impeded because generic models treat strict legal constraints as mere suggestions.
Approach:	They propose a risk-constrained bilevel Stackelberg framework that models high-stakes revision as a strategic interaction rather than an open-ended conversation.
Outcome:	The proposed framework achieves state-of-the-art performance with an average RRR of 84.21% and enhanced token efficiency.

RAVR: Reference-Answer-guided Variational Reasoning for Large Language Models (2026.findings-acl)

Copied to clipboard

Tianqianjin Lin, Xi Zhao, Xingyao Zhang, Rujiao Long, Yi Xu, Zhuoren Jiang, Wenbo Su, Bo Zheng

Data-Centric Continual Pre-training for 500+ Languages: A New Bilingual Translation Corpus and Multilingual Models (2026.findings-acl)

Challenge:	Experiments show that reinforcement learning (RL) can refine the reasoning abilities of large language models (LLMs) but requires a key prerequisite: the model must already be able to generate high-utility reasoning paths with non-negligible probability.
Approach:	They propose a framework that uses answer-conditioned reasoning as a variational surrogate for question-only reasoning.
Outcome:	Experiments on 11 benchmarks and 3 models show that RAVR reduces hesitation, strengthens conclusion consolidation, and promotes problem-specific strategies in reasoning.

Copied to clipboard

Shaoxiong Ji, Zihao Li, Jaakko Paavola, Hengyu Luo, Jörg Tiedemann

Failures are Treasures: Constructing a Pedagogical Bridge for Agentic Strategy Distillation (2026.findings-acl)

Challenge:	Large language models pre-trained on massive data have promoted multilingual natural language processing (NLP).
Approach:	They construct a bilingual translation corpus with 2,500 language pairs and develop a suite of four models with parallel data.
Outcome:	The proposed model suites are evaluated across 7 tasks and 12 benchmarks.

Copied to clipboard

Jiaxin Guo, Hao Sun, Wenhao Zhang, Chunyu Yang, Yan Zhang

Privacy-Preserving Reasoning with Knowledge-Distilled Parametric Retrieval Augmented Generation (2026.findings-acl)

Challenge:	Existing knowledge distillation methods focus on imitating successful trajectories, whereas small language models are fragile and often collapsing after encountering errors.
Approach:	They propose a Pedagogical Bridge for Reflective Insight and Distillation of Guiding Errors that combines reflection-in-action and reflection-on-action to enable agents to diagnose and correct critical errors while abstracting transferable strategies from contrastive student–teacher trajectories.
Outcome:	Experiments show that the proposed model significantly elevates performance in large language models (SLMs) .

Copied to clipboard

Jinwen Chen, Hainan Zhang, Liang Pang, Yongxin Tong, Haibo Zhou, Wei Lin, Zhiming Zheng

Beyond Sentence-level Labels: Integrating Conversational Context and Personal Experience for Natural Emotional Expression (2026.findings-acl)

Challenge:	Existing RAG systems require uploading local documents to the cloud, resulting in inference latency and poor generalization on out-of-distribution (OOD) inputs.
Approach:	They propose a generalizable knowledge-distilled parametric RAG model aligned with standard RAG in document structure and parameter activation.
Outcome:	The proposed model outperforms baselines in accuracy and generalizes well on out-of-distribution (OOD) data.

Copied to clipboard

Haiyang Sun, Chenyang Le, Wei Wang, Leying Zhang, Chuang Li, Bing Han, Chenda Li, Mengxiao Bi, Yanmin Qian

Structure-Aware Zero-Shot Relational Learning for Knowledge Graphs without External Knowledge (2026.findings-acl)

Challenge:	Existing systems rely on sentence-level labels, which fails to capture the subtle nuances of human affect.
Approach:	They propose to use a large-scale, context-aware speech corpus derived from multi-speaker audiobooks to generate a speech that is human-like.
Outcome:	The proposed model outperforms existing methods in terms of emotional expression accuracy and naturalness.

Copied to clipboard

Kuan Xu, Baoxin Zhang, Shuyue Fan, Ming Chen, Zhipeng Ke, Jian Yu, Xuezhong Zhou

Continual Safety Alignment via Gradient-Based Sample Selection (2026.findings-acl)

Challenge:	Existing methods for Zero-shot Relational Learning depend on external knowledge, resulting in increased annotation costs and limited practical applicability.
Approach:	They propose a structure-aware paradigm that performs ZRL without external knowledge . it leverages intrinsic structural patterns in KGs to bridge semantic correlations for new relations with existing ones.
Outcome:	The proposed paradigm achieves 10.66% improvement in MRR while reducing annotation costs and enhancing practical applicability on three real-world benchmarks.

Copied to clipboard

Thong Bach, Dung Nguyen, Thao Minh Le, Truyen Tran

ClinicalMC: A Benchmark for Multi-Course Clinical Decision-Making with Large Language Models (2026.findings-acl)

Challenge:	Large language models require continuous adaptation to new domains, tasks, and evolving requirements.
Approach:	They propose a gradient-based sample selection method that filters high-gradient samples during fine-tuning.
Outcome:	The proposed method significantly improves alignment preservation while maintaining competitive task performance on continual domain tasks.

Copied to clipboard

Ruihui Hou, Siyi Zhu, Ziyue Huai, Guangya Yu, Yongqi Fan, ChunMing Wang, Tong Ruan

Safe-SAIL: Towards a Fine-grained Safety Landscape of Large Language Models via Sparse Autoencoder Interpretation Framework (2026.findings-acl)

Challenge:	Existing benchmarks assess LLM performance in single-course settings and lack systematic evaluation in multi-course scenarios, where a patient’s condition evolves over time.
Approach:	They propose to use large language models to assess their performance in multi-course clinical decision-making scenarios where a patient’s condition evolves over time.
Outcome:	The proposed model includes 1,275 Chinese and 5,804 English samples across four stages from admission to discharge.

Copied to clipboard

Jiaqi Weng, Han Zheng, Hanyu Zhang, Ej Zhou, Qinqin He, Jialing Tao, Hui Xue, Zhixuan Chu, Xiting Wang

Graph-Assisted Large Language Models: A Perspective on Mitigating Intrinsic Limitations (2026.findings-acl)

Challenge:	Existing studies on how SAEs derive most fine-grained latent features for safety remain unexplored.
Approach:	They propose a framework for interpreting SAE features in safety-critical domains . they train a suite of SAEs with human-readable explanations and systematic evaluations based on pornography, politics, violence, and terror .
Outcome:	The proposed framework reduces interpretation cost by 55% and improves safety-critical features.

Copied to clipboard

Haitong Luo, Fali Wang, Weiyao Zhang, Xianren Zhang, Zhiwei Zhang, Tianxiang Zhao, Minhua Lin, Jiahao Zhang, Hui Liu, Xianfeng Tang, Qi He, Suhang Wang, Xuying Meng, Yujun Zhang

GUITester: Enabling GUI Agents for Exploratory Defect Discovery (2026.findings-acl)

Challenge:	Large language models exhibit intrinsic limitations such as knowledge cutoff, single-threaded reasoning that hinders finer-grained branch and aggregation, and rigid collaboration mechanisms that struggle to coordinate specialized capabilities.
Approach:	They propose a taxonomy spanning Graph-Assisted Knowledge Augmentation, Graph Assisted Reasoning and Planning, and Graphed LLM Collaboration.
Outcome:	The proposed models show that graphs can augment and correct LLMs and support dynamic coordination among experts and agents in collaborative settings.

Copied to clipboard

Yifei Gao, Jiang Wu, Xiaoyi Chen, Yifan Yang, Zhe Cui, Tianyi Ma, Jiaming Zhang, Jitao Sang

Challenge:	Exploratory GUI testing is essential for software quality but suffers from high manual costs.
Approach:	They propose a framework that decouples navigation from verification via two modules . they propose 143 tasks and a GUITestBench benchmark that features 26 defects .
Outcome:	The proposed framework outperforms state-of-the-art benchmarks in 143 tasks and 26 defects.

BiCSRouter: Bi-Level Cross-System Routing for Utility-Aware LLM Inference (2026.findings-acl)

Copied to clipboard

Mao Keyu, Eiki Murata, Ukyo Honda

Challenge:	Existing routing frameworks operate within a single computational paradigm . a cross-system routing framework that integrates two orthogonal regimes is proposed .
Approach:	They propose a cross-system routing framework that integrates two orthogonal regimes . they propose MBPP-based model that decomposes routing into intra-regime configuration selection and inter-regem system selection .
Outcome:	The proposed framework outperforms 15 representative baselines on MBPP and MATH benchmarks.

FAER: Benchmarking VLMs for Failure-Aware Embodied Reasoning (2026.findings-acl)

Copied to clipboard

Hao Song, Kaifeng Liu, Yuanxing Liu, Xiang Tian, Xuesong Wang, Chen Yifan, Weinan Zhang, Ting Liu

Reverse Constitutional AI: A Framework for Controllable Toxic Data Generation via Probability-Clamped RLAIF (2026.findings-acl)

Challenge:	Visual-language models (VLMs) are the core component of embodied agents in perceiving the environment and making decisions.
Approach:	They propose a failure-aware benchmark to evaluate the performance of visual language models (VLMs) in long-horizon tasks.
Outcome:	The proposed benchmark evaluates the performance of 16 widely utilized VLMs and 4 LLMs for FAER tasks.

Copied to clipboard

Yuan Fang, Yiming Luo, Aimin Zhou, Fei Tan

ToolGrad: Efficient Tool-use Dataset Generation with Textual “Gradients” (2026.findings-acl)

Challenge:	Existing approaches to red teaming focus on searching for individual adversarial inputs.
Approach:	They propose a framework for automated adversarial data generation that inverts harmless constitution into constitution of toxicity and iteratively refining model outputs through critique–revision pipeline.
Outcome:	The proposed framework generates diverse, high-quality toxic data without human annotation and significantly improves semantic coherence without sacrificing adversarial strength.

Copied to clipboard

Zhongyi Zhou, Kohei Uehara, Haoyu Zhang, Jingtao Zhou, Lin Gu, Ruofei Du, Zheng Xu, Tatsuya Harada

Unlocking Exploration in RLVR: Uncertainty-aware Advantage Shaping for Deeper Reasoning (2026.findings-acl)

Challenge:	Prior work synthesizes tool-use LLM datasets by first generating a user query, then complex tool-using annotations like DFS.
Approach:	They propose an agentic framework that synthesizes user queries and generates valid tool-use chains . they propose a dataset with more complex tool use, lower cost, and almost 100% pass rate .
Outcome:	Experiments show that tools trained on ToolGrad outperform expensive baseline datasets and proprietary LLMs.

Copied to clipboard

Can Xie, Ruotong Pan, Xiangyu Wu, Zhang Yunfei, Jiayi Fu, Tingting Gao, Guorui Zhou

GlossaGen: Making Academic Translation Smarter with Glossing (2026.findings-acl)

Challenge:	Reinforcement Learning with Verifiable Rewards (RLVR) has shown significant promise for enhancing the reasoning capabilities of large language models (LLMs).
Approach:	They propose a model-free method that refines credit assignment by leveraging the model's internal uncertainty signals.
Outcome:	Extensive experiments on five mathematical reasoning benchmarks show that the proposed method outperforms strong RLVR baselines on multiple model scales, including 1.5B and 7B.

Copied to clipboard

Zixiao Wang, Duzhen Zhang, Juntian Zhang, Yuhan Liu, Guoming Li, Haolun Wu, Le Song, Xiuying Chen

DeCoVec: Building Decoding Space based Task Vector for Large Language Models via In-Context Learning (2026.findings-acl)

Challenge:	Existing machine translation systems obscure or mistranslate key terminology, while paraphrasing aimed at lay readers often oversimplifies it, hindering their ability to master domain-specific technical vocabulary.
Approach:	They propose a task which produces translations dynamically adapted to a reader’s academic proficiency, or level, and a framework to address this challenge.
Outcome:	The proposed framework achieves higher scores than baselines on a synthesized benchmark and human evaluations.

Copied to clipboard

Feiyang Li, Yile Wang

Hebbian-Guided Bi-Directional Rank Adaptation for Parameter-Efficient Fine-Tuning (2026.findings-acl)

Challenge:	Existing approaches to steering large language models require fine-tuning or manipulation of internal states, limiting their flexibility and scalability.
Approach:	They propose a framework that constructs task vectors directly in the decoding space by leveraging in-context learning.
Outcome:	The proposed framework outperforms standard few-shot baselines on TruthfulQA, Math-500, and AQUA-RAT with gains up to +5.50 accuracy.

Copied to clipboard

Xing Zhao, Liu Yang, Xi-Le Zhao, Yueqi Yang, Tianqi Jiang

StruNRAG: Evaluation of OCR-Induced Structural Noise on RAG Robustness (2026.findings-acl)

Challenge:	Low-Rank Adaptation (LoRA) is a widely used method to fine-tune large language models . but its fixed-rank design cannot capture the varying importance across different layers .
Approach:	They propose a framework that bi-directionally reallocates low-rank capacity using Hebbian-inspired importance estimation.
Outcome:	Experiments show that HeBiRA improves performance over baselines.

Copied to clipboard

Mengna Gao, Dapeng Yin, Shuyue Zhu, Bingxuan Hou, Zhanpeng Ni, Junli Wang

Memory as Action: Autonomous Context Curation for Long-Horizon Agentic Tasks (2026.findings-acl)

Challenge:	Existing evaluations of RAG systems ignore structural noise, authors say . complex layouts can cause OCR failures and disrupt semantic flow of text . advanced LLMs demonstrate robustness against local noise, but struggle to maintain reasoning capabilities under severe structural disruption that fragments global context.
Approach:	They propose a benchmark to evaluate RAG robustness against OCR-induced structural perturbations.
Outcome:	The proposed benchmark systematically injects three categories of real-world structural noise into a bilingual dataset of 2,132 question-answer pairs . results show that advanced LLMs demonstrate robustness against local noise, but struggle to maintain reasoning capabilities under severe structural disruption .

Copied to clipboard

Yuxiang Zhang, Jiangming Shu, Ye Ma, Xueyuan Lin, Shangxi Wu, Jitao Sang

Experience is the Teacher: Reusing Atomic Thoughts from LLMs to Improve Medical Dialogue (2026.findings-acl)

Challenge:	Existing approaches to managing working memory are based on external mechanisms that lack awareness of the agent’s reasoning state, leading to suboptimal decisions.
Approach:	They propose a framework that treats working memory management as learnable policy actions and enables joint optimization of information retention and task performance through end-to-end reinforcement learning.
Outcome:	The proposed framework matches models 16 larger while reducing average context length by 51%, with learned strategies that adapt to model capabilities and generalize across task complexities.

Copied to clipboard

Guangya Yu, Hui Luo, Qi Ye, Ruihui Hou, Weiyan Zhang, Mingxi Shang, Xuanwu Li, ChunMing Wang, Tong Ruan

Beyond Static Personas: Situational Personality Steering for Large Language Models (2026.findings-acl)

Challenge:	Recent large reasoning models (LLMs) lack dynamic and diverse thinking capabilities . reusing atomic thoughts provides a practical pathway toward dynamic reasoning .
Approach:	They propose a framework that extracts atomic thoughts from teacher models and reuses them to guide reasoning and generate responses.
Outcome:	The proposed framework extracts atomic thoughts from teacher models and reuses them to guide reasoning and generate responses.

Copied to clipboard

Zesheng Wei, Mengxiang Li, Zilei Wang, Yang Deng

Born Pragmatic, Trained to Hallucinate? Quantifying the Origins of Contextual Bias in LLMs via the PaCE Benchmark (2026.findings-acl)

Challenge:	Existing personalization methods rely on static personality modeling to achieve optimal performance.
Approach:	They propose a training-free framework for advanced situational personality steering that incorporates situation-dependent behavior patterns within LLM personalities through analysis of persona neurons.
Outcome:	The proposed framework surpasses baselines on PersonalityBench and SPBench, demonstrating generalization and robustness to complex, unseen situations and different models architecture.

Copied to clipboard

Ziming Li, Yu Tian, Tian Lan, Jiang Li, Zehua Duo, Guanglai Gao, Xiangdong Su

Discourse & Pragmatics

Challenge:	Large language models excel at capturing communicative intent, but they have a side effect: pragmatic hallucination.
Approach:	They propose a benchmark to quantify the impact of pragmatic hallucination on large language models . they propose RLHF and SFT to induce a strong tendency for pragmatic over-attribution .
Outcome:	The proposed model outperforms existing models in predicting pragmatic hallucinations . the evaluations show that current alignment paradigms lack precise control over pragmatic boundaries .

Domain-Specific Data Generation Framework for RAG Adaptation (2026.findings-acl)

Copied to clipboard

Chris Xing Tian, Weihao Xie, Zhen Chen, Hui Liu, Zhengyuan Yi, Haoliang Li, Shiqi Wang, Siwei Ma

Tracing Mathematical Proficiency Through Problem-Solving Processes (2026.findings-acl)

Challenge:	Retrieval-Augmented Generation (RAG) combines the language understanding and reasoning capabilities of large language models (LLMs) with external retrieval to produce domain-grounded responses.
Approach:	They propose a scalable and modular data-centric framework for generating domain-grounded question–answer–context triples tailored to diverse RAG adaptation strategies.
Outcome:	The proposed framework generates domain-grounded question–answer–context triples for multiple RAG adaptation strategies.

Copied to clipboard

Jungyang Park, Suho Kang, Jaewoo Park, Jae Hong Kim, Jaewoo Shin, Seonjoon Park, Youngjae Yu

Decision Biases and Intent-Irony Decoupling in Large Language Models (2026.findings-acl)

Challenge:	Knowledge Tracing (KT) models a learner's evolving knowledge state over time, but lacks the rich information embedded in students' problem-solving processes.
Approach:	They propose a framework that uses a teacher-student-teacher pipeline to extract students’ Mathematical Proficiency (MP) as intermediate representation.
Outcome:	The proposed framework improves the prediction performance of existing KT methods and provides interpretable explanations by explicitly modeling students’ mathematical proficiency.

Copied to clipboard

Kewei Guo, Lingyun Sun, Manhao Guan

Anonpsy: A Graph-Based Framework for Structure-Preserving De-identification of Psychiatric Narratives (2026.findings-acl)

Challenge:	Large Language Models (LLMs) exhibit impressive linguistic fluency, but it remains unclear whether they possess human-like Theory of Mind (ToM) or rely on statistical heuristics . a recent study examined the performance of LLMs against 300 human participants .
Approach:	a study establishes a framework for large language models that modulates contextual contrast, linguistic cues, and cognitive mechanisms.
Outcome:	a new evaluation framework compares ten state-of-the-art LLMs against 300 human participants . the framework systematically modulates contextual contrast, linguistic cues, and cognitive mechanisms .

Copied to clipboard

Kyungho Lim, Byung-Hoon Kim

HearSay Benchmark: Do Audio LLMs Leak What They Hear? (2026.findings-acl)

Challenge:	Existing approaches to de-identify narratives operate at the text level and offer limited control over which semantic elements are preserved or altered.
Approach:	They propose a de-identification framework that reformulates the task as graph-guided semantic rewriting.
Outcome:	The proposed framework preserves diagnostic fidelity while achieving low re-identification risk.

Copied to clipboard

Jin Wang, Kaiwen Luo, Liang Lin, Weiliu Wang, Yitian Chen, Moayad Aloqaily, Xuehai Tang, Zhenhong Zhou, Kun Wang, Li Sun, Qingsong Wen

LongInsightBench: A Comprehensive Benchmark for Evaluating Omni-Modal Models on Human-Centric Long-Video Understanding. (2026.findings-acl)

Challenge:	Recent advances in audio large language models have led to their potential privacy implications unexplored.
Approach:	They propose a benchmark to examine whether ALLMs leak user privacy through acoustic voiceprints.
Outcome:	The proposed benchmark is constructed from over 22,000 real-world audio clips.

Copied to clipboard

ZhaoYang Han, Qihan Lin, Hao Liang, Bowen Chen, Zhou Liu, Wentao Zhang

Fixing Semantic Blind Spots in Anchor Tokens of dMLLMs (2026.findings-acl)

Challenge:	LongInsightBench is the first benchmark designed to assess models’ ability to understand long videos, with a focus on human language, viewpoints, actions, and other contextual elements.
Approach:	They propose a benchmark to assess models’ ability to understand long videos with a focus on human language, viewpoints, actions, and other contextual elements.
Outcome:	The proposed model excels in three key areas: a) long-duration, human-centric videos; b) diversifying and challenging task scenarios; c) quality assurance pipeline; and d) reliability.

Copied to clipboard

Ruixuan Xu, Jiexi Xu, Qiyan Zhao, Xiaofeng Zhang

Beyond Transcription: Unified Audio Schema for Perception-Aware AudioLLMs (2026.findings-acl)

Challenge:	Autoregressive models (ARMs) are prone to hallucinations due to their sequential text generation and high latency.
Approach:	They propose a training-free decoding strategy that augments the attention key space with a static, distance-aware matrix to reduce the attention sink effect on semantic anchors.
Outcome:	The proposed method reduces the attention sink effect on semantic anchors while enhancing their ability to aggregate global visual information.

Copied to clipboard

Linhao Zhang, Yuhan Song, Aiwei Liu, Chuhan Wu, Sijun Zhang, Wei Jia, Yuan Liu, Houfeng Wang, Zhou Xiao

From Style to Story: A Curriculum Learning Approach for Imitative Novel Generation (2026.findings-acl)

Challenge:	Recent Audio Large Language Models (AudioLLMs) excel at reasoning tasks, but struggle at elementary auditory perception.
Approach:	They propose a framework that organizes audio information into three explicit components in a unified JSON format.
Outcome:	The proposed framework boosts fine-grained perception by 10.9% on MMSU over state-of-the-art models while preserving robust reasoning capabilities.

Copied to clipboard

Xueran Han, Yuhan Liu, Mingzhe Li, Wei Liu, Sen Hu, Rui Yan, Zhiqiang xu, Xiuying Chen

ArkRepoBench: A Repository-Level Code Completion Benchmark for HarmonyOS Development (2026.findings-acl)

Challenge:	Novels create rich, immersive worlds with intricate plots and distinct styles, captivating readers through complex storytelling.
Approach:	They propose a novel generation system that imitates novel elements by predicting plot developments and writing concrete details using vivid, expressive language.
Outcome:	The novel imitative novel generation system is trained through a curriculum learning paradigm, progressing from low-level stylistic mastery to high-level narrative coherence.

Copied to clipboard

Yanlin Wang, Bowen Zhang, Yanli Wang, Daya Guo, Terry Yue Zhuo, Jiachi Chen, Mingwei Liu, Xingong Zhang, Zibin Zheng

e5-omni: Explicit Cross-modal Alignment for Omni-modal Embeddings (2026.findings-acl)

Challenge:	Despite the maturity of LLM-based code assistance for mainstream languages, the capabilities of ArkTS are largely unexplored.
Approach:	They propose to benchmark repository-level code completion for ArkTS using 7,519 samples from 20 official HarmonyOS repositories.
Outcome:	The proposed benchmark covers multiple difficulty levels and categorizes completion instances into Single-File, Cross-Filled Independent, and Cross-Filed Dependent settings based on dependency analysis.

Copied to clipboard

Haonan Chen, Sicheng Gao, Radu Timofte, Tetsuya Sakai, Zhicheng Dou

Large Language Models Are Still Misled by Simple Bias Ensembles (2026.findings-acl)

Challenge:	Recent omni-modal embeddings rely heavily on implicit alignment from pretrained visionlanguage models.
Approach:	They propose a lightweight explicit alignment recipe that adapts off-the-shelf VLMs into robust omni-modal embedding models.
Outcome:	The proposed model improves on MMEB-V2 and AudioCaps with a lightweight explicit alignment recipe.

Copied to clipboard

Zhouhao Sun, Zhiyuan Kan, Xiao Ding, Li Du, Bibo Cai, Yang Zhao, Bing Qin, Ting Liu

Text-to-TrajVis: Enabling Trajectory Data Visualizations from Natural Language Questions (2026.findings-acl)

Challenge:	Existing benchmarks for large language models are constrained to datasets where each sample is manually injected with only one type of bias.
Approach:	They propose a multi-bias benchmark where each sample contains multiple types of biases.
Outcome:	The proposed benchmark shows that existing LLMs and debiasing methods perform poorly on this benchmark, highlighting the challenge of eliminating compounded biases.

Copied to clipboard

Tian Bai, Huiyan Ying, Kailong Suo, Victor Junqiu Wei, Tao Fan, Yuanfeng Song

Challenge:	Existing datasets for this task are limited and there is no suitable one available.
Approach:	They propose a new visualization language called Trajectory Visualization Language (TVL) to facilitate querying trajectory data and generating visualizations.
Outcome:	The proposed language can be used to query and generate trajectory data and generate visualizations with large language models.

MemTR: Enhancing Tool-Calling Reliability via Uncertainty-Triggered FFN-Space Retracing (2026.findings-acl)

Copied to clipboard

Hongtao Duan, Lu Jiang, Minying Zhang, Xiaobing Zhu, Tianpeng Bu, Hao Jiang, Xinyu Wei, Lulu hu

More Than Sum of Its Parts: Deciphering Intent Shifts in Multimodal Hate Speech Detection (2026.findings-acl)

Challenge:	Existing tool-calling methods rely on costly tool-use training data or only constrain syntax, leaving tool selection and argument value errors largely unsolved.
Approach:	They propose a method that decodes tool evidence from the tool library and mixes it into the output at the uncertain layer.
Outcome:	The proposed method reduces tool calling failures by 2%–9% with only 1%–2% runtime overhead.

Copied to clipboard

Runze Sun, Yu Zheng, Zexuan Xiong, Zhongjin Qu, Lei Chen, Jie Zhou, Jiwen Lu

Specialization without Sparsity: Efficient and Expressive Split-Path Experts for LLM Fine-Tuning (2026.findings-acl)

Challenge:	Existing systems struggle with multimodal content where the emergent meaning transcends the aggregation of individual modalities.
Approach:	They propose a framework to characterize semantic intent shifts where modalities interact to construct implicit hate from benign cues or neutralize toxicity through semantic inversion.
Outcome:	The proposed framework outperforms state-of-the-art benchmarks on H-VLI and on established benchmarks.

Copied to clipboard

Ganghao Liu, Qin Zhou, Zhe Wang, Xuehan Lu, Haihua Huang, Yunfei Tong, Heng Tian

Pardon? Evaluating Conversational Repair in Large Audio-Language Models (2026.findings-acl)

Challenge:	Parameter-efficient fine-tuning (PEFT) is a low-cost alternative to full fine-timing due to the massive overhead.
Approach:	They propose a Mixture-of-Experts approach that enhances specialization while maintaining low resource overhead.
Outcome:	The proposed approach outperforms or matches state-of-the-art methods on GLUE, GSM8K, MBPP, and a text rewriting task from SmolTalk.

Copied to clipboard

Shuanghong Huang, Jinlei Xu, Youchao Zhou, Yanghao Zhou, Xuan Zhao, Chong Feng, Wenxuan Zhang

ToxReason: A Benchmark for Mechanistic Chemical Toxicity Reasoning via Adverse Outcome Pathway (2026.findings-acl)

Challenge:	Existing evaluations of large audio-language models focus on answer accuracy and robustness to acoustic perturbations, but they assume that inputs remain semantically answerable.
Approach:	They propose a repair-aware evaluation setting that explicitly distinguishes between answerable and unanswerable audio inputs.
Outcome:	The proposed evaluation setting distinguishes between answerable and unanswerable audio inputs.

Copied to clipboard

Jueon Park, WonJune Jang, Chanhwi Kim, Yein Park, Jaewoo Kang

CVRH: Cross-modal Variational Role Hypergraph Network via Semantic Enhancement for Multi-modal Event Argument Extraction (2026.findings-acl)

Challenge:	Recent advances in large language models have enabled molecular reasoning for property prediction. however, toxicity arises from complex biological mechanisms, necessitating mechanistic reasoning for reliable prediction.
Approach:	They propose a benchmark that evaluates organ-level toxicity reasoning across multiple organs . they find strong predictive performance does not necessarily imply reliable reasoning .
Outcome:	The proposed benchmark evaluates toxicity prediction performance and reasoning quality across LLMs.

Copied to clipboard

Bangze Pan, Yang Li, Ruili Pu, Suge Wang, Jian Liao, JianXing Zheng, Xiaoli Li, Deyu Li

Event Extraction

Challenge:	Existing methods focus on weakly aligning uni-modal representations and generatively data augmentation techniques, but they ignore the potential impact of event role information on MEAE.
Approach:	They propose a cross-modal variational role hypergraph network via semantic enhancement to model high-order role correlations among cross-mod arguments in multi-modal documents.
Outcome:	The proposed method achieves a 6.9% improvement in F1-score on the M2E2 benchmark compared to current state-of-the-art methods.

DiffCL: Difference-Aware Contrastive Learning for Automatic Answer Grading with Multi-Level Semantic Modeling (2026.findings-acl)

Copied to clipboard

Lei Chen, BoYu Gao, Zitao Liu, Tingjie Wan, Weiqi Luo

GaLa: Hypergraph-Guided Visual Language Models for Procedural Planning (2026.findings-acl)

Challenge:	Existing approaches to automate answer grading lack semantic understanding and scoring consistency.
Approach:	They propose a difference-aware AAG framework that integrates heuristic difference labeling with dual-contrastive learning.
Outcome:	The proposed method outperforms cross-entropy-based baselines on SciEntsBank and Beetle datasets.

Copied to clipboard

Kun Wang, Li Yiming, Mingcheng Qu, Aqiang Zhang, Guang Yang, Tonghua Su

TwinVoice: A Multi-dimensional Benchmark Towards Digital Twins via LLM Persona Simulation (2026.findings-acl)

Challenge:	Existing methods for procedural planning over-rely on visual inputs and lack structured semantic information.
Approach:	They propose a vision–language framework for multimodal procedural planning that exploits implicit spatial relations and deep semantics encoded in object attributes.
Outcome:	The proposed framework outperforms existing methods in terms of execution success rate, LCS, and planning correctness.

Copied to clipboard

Bangde Du, Minghao Guo, Songming He, Ziyi Ye, Xi Zhu, Weihang Su, Shuqi Zhu, Yujia Zhou, Yongfeng Zhang, Qingyao Ai, Yiqun Liu

CE-RM: A Pointwise Generative Reward Model Optimized via Two-Stage Rollout and Unified Criteria (2026.findings-acl)

Challenge:	Existing studies show that advanced LLMs produce text indistinguishable from human writing.
Approach:	They propose a benchmark to assess persona simulation across diverse contexts by decomposing the evaluation into six fundamental capabilities including opinion consistency, memory recall, logical reasoning, persona tone, and syntactic style.
Outcome:	The proposed model achieves moderate accuracy but falls short of the basic capabilities needed to simulate personas in real-world contexts.

Copied to clipboard

Xinyu Hu, Yancheng He, Weixun Wang, Tao Feng, Li Lin, Jiashun Liu, Wenbo Su, Bo Zheng, Xiaojun Wan

Challenge:	Existing studies have shown that rule-based evaluation methods are ineffective for open-ended natural language generation.
Approach:	They propose a pointwise generative reward model with a dedicated two-stage rollout method and unified query-based criteria that can be trained with 5.7K high-quality data.
Outcome:	The proposed model achieves superior performance on diverse reward model benchmarks, especially in Best-of-N scenarios, and delivers more effective improvements in downstream RL practice.

Hard to Read, Easy to Jailbreak: How Visual Degradation Bypasses MLLM Safety Alignment (2026.findings-acl)

Copied to clipboard

Zhixue Song, Boyan Han, Yiwei Wang, Chi Zhang

Unleashing the Native Recommendation Potential: LLM-Based Generative Recommendation via Structured Term Identifiers (2026.findings-acl)

Challenge:	Recent advances in visual context compression enable MLLMs to process ultra-long contexts efficiently by rendering text into images.
Approach:	They propose a strategy that decouples visual transcription from safety auditing by enforcing a serialized pipeline to decouple visual transcription and safety assessment.
Outcome:	The proposed strategy decouples visual transcription from safety auditing to reduce the risk of jailbreaking.

Copied to clipboard

Zhiyang Zhang, Junda She, Kuo Cai, Bo Chen, Shiyao Wang, Xinchen Luo, Qiang Luo, Ruiming Tang, Han Li, Kun Gai, Guorui Zhou

Abstain-R1: Calibrated Abstention and Post-Refusal Clarification via Verifiable RL (2026.findings-acl)

Challenge:	Existing methods for constructing item identifiers face bottlenecks due to their large output space and expensive vocabulary expansion and alignment training.
Approach:	They propose to use Large Language Models to develop general-purpose, semantically-aware recommender systems that can be generalized and reusable.
Outcome:	Experiments on real-world datasets show that GRAM outperforms baselines and significantly outperformed baselines.

Copied to clipboard

Haotian Zhai, Jingcheng Liang, Dongyeop Kang

Challenge:	Existing abstention methods produce generic refusals or encourage follow-up clarifications without verifying whether they identify the key missing information.
Approach:	They propose a clarification-aware RLVR reward that rewards correct answers on unanswerable queries while optimizing explicit abstention and semantically aligned post-refusal clarification on unannounced queries.
Outcome:	The proposed model improves abstention and clarification on unanswerable queries while maintaining strong performance on answerable queries.

Learning Through Dialogue: Engagement and Efficacy Matter More Than Explanations (2026.findings-acl)

Copied to clipboard

Shaz Furniturewala, Gerard Christopher Yeo, Kokil Jaidka

CodeBind: Decoupled Representation Learning for Multimodal Alignment with Unified Compositional Codebook (2026.findings-acl)

Challenge:	Large language models (LLMs) are increasingly used as conversational partners for learning, yet the interactional dynamics supporting users’ learning and engagement are understudied.
Approach:	They analyze linguistic and interactional features from LLM and participant chats to identify the mechanisms and conditions under which LLM explanations shape changes in political knowledge and confidence.
Outcome:	The results show that LLM explanations shape political knowledge and confidence . they also show that their effects are highly conditional and vary by political efficacy .

Copied to clipboard

Zeyu Chen, Jie Li, Kai Han

MASS: Deep Research for Social Sciences with Memory-Augmented Social Simulation (2026.findings-acl)

Challenge:	Multimodal representation alignment is crucial for large language models and robotics.
Approach:	They propose a framework that optimizes multimodal representation spaces through a modality-shared-specific codebook design.
Outcome:	The proposed framework achieves state-of-the-art performance in multimodal classification and retrieval tasks.

Copied to clipboard

Yongrui Liu, Deyi Xiong

LS-Guard: Adaptive Safety Guardrails Tailored to Individual LLMs (2026.findings-acl)

Challenge:	Existing systems rely heavily on literature retrieval and synthesis, resulting research lacking insight and creativity in social science.
Approach:	They propose a method that leverages highly realistic social simulations to the creativity of LLMs-generated research.
Outcome:	The proposed model shows a 6.81% improvement in quality over foundation LLMs and 17.19% gain in Insight over strong baselines.

Copied to clipboard

Jinggui Liang, Lizi Liao

Rethinking LLM Watermark Detection in Black-Box Settings: A Non-Intrusive Third-Party Framework (2026.findings-acl)

Challenge:	Existing security guardrails built from static datasets ignore each model’s unique safety profile and often force trade-offs between safety and utility.
Approach:	They propose a framework for learning model-specific guardrails tailored to each LLM’s vulnerabilities.
Outcome:	The proposed framework significantly outperforms baseline guardrails on multiple real-world LLMs, achieving superior robustness, adaptability, and generalization.

Copied to clipboard

Zhuoshang Wang, Yubing Ren, Yanan Cao, Fang Fang, Xiaoxue Li, Li Guo

Deputy: Accelerating Large Language Model Inference with Dynamic Low-Rank Substitution (2026.findings-acl)

Challenge:	Existing secret-key schemes tightly couple detection with injection . this dependency creates a fundamental barrier for real-world governance .
Approach:	et al. introduce a black-box framework for non-intrusive, third-party watermark verification . they propose a proxy model to amplify watermark-relevant signals and complementary relative measurements .
Outcome:	a new framework decouples detection from injection and assesses alignment of query text with watermark distributions.

Copied to clipboard

Yuhua Zhou, Shichao Weng, Changhai Zhou, Yuhan Wu, Qian Qiao, Jun Gao, Fei Yang, Aimin Pan

SGG-R 3: From Next-Token Prediction to End-to-End Unbiased Scene Graph Generation (2026.findings-acl)

Challenge:	Existing dynamic schemes such as early-exit and layer-drop reduce FLOPs but break batch processing or introduce KV-cache inconsistency.
Approach:	They propose a dynamic low-rank substitution framework that employs a lightweight decision module at each layer to dynamically determine the execution branch for different tokens.
Outcome:	The proposed model reduces computation by approximately 40% compared to the original dense model while outperforming existing baseline methods.

Copied to clipboard

Jiaye Feng, Qixiang Yin, Yuankun Liu, Tong Mo, Weiping Li

Preserving Fairness and Safety in Quantized LLMs Through Critical Weight Protection (2026.findings-acl)

Challenge:	Existing methods for scene graph generation lack task-specific structured reasoning and sparse, long-tailed relation distributions.
Approach:	They propose a structured reasoning framework that integrates task-specific Chain-of-Thought and reinforcement learning with group sequence policy optimization to achieve unbiased scene graph generation.
Outcome:	The proposed framework achieves superior performance on two benchmarks.

Copied to clipboard

Muhammad Alif Al Hakim, Alfan Farizki Wicaksono, Fajri Koto

X-Router: Decoupling Knowledge and Reasoning for Cost-Effective LLM Inference (2026.findings-acl)

Challenge:	Existing quantization methods focus on general metrics like perplexity or accuracy on standard benchmarks.
Approach:	They propose a method that preserves fairness- and safety-critical weights during quantization.
Outcome:	The proposed method reduces bias and safety degradation without costly retraining or alignment while maintaining trustworthiness while retaining efficiency.

Copied to clipboard

Zixuan Wang, Yinze Ding, Zihan Wang, Jinyu Guo, Zhenhong Zhou, Junhao Dong, Chaomeng Chen

SPD-Faith Bench: Diagnosing and Improving Faithfulness in Chain-of-Thought for Multimodal Large Language Models (2026.findings-acl)

Challenge:	Existing adaptive methods focus on a single axis, overlooking evidence need and reasoning depth are only partially correlated.
Approach:	They propose a dual-axis routing framework that separates retrieval necessity from reasoning necessity under a user-defined cost–quality trade-off.
Outcome:	The proposed framework reduces token usage and latency while improving answer quality over strong baselines.

Copied to clipboard

Weijiang Lv, Yaoxuan Feng, Xiaobo Xia, Jiayu Wang, Yan Jing, Wenchao Chen, Bo Chen

HIPO: A Hierarchical Prompt Optimization Framework with Task Awareness and Fine-Grained Debugging (2026.findings-acl)

Challenge:	Existing studies on multimodal faithfulness have focused on perceptual hallucinations, raising concerns about the validity of reasoning traces.
Approach:	They propose a diagnostic benchmark that enforces explicit visual comparison to assess faithfulness of reasoning traces.
Outcome:	The proposed framework improves visual routing and aligns reasoning with perception.

Copied to clipboard

Lu Qi, Lei Chai, Hongrui Yu, Binhang Qi, Hailong Sun

MAPLE: A Meta-learning Framework for Cross-Prompt Essay Scoring (2026.findings-acl)

Challenge:	Existing methods for prompt optimization apply the same prompt across all samples . existing methods ignore variation in sample difficulty .
Approach:	They propose a framework that shifts the paradigm from dataset-level to sample-level optimization.
Outcome:	The proposed framework outperforms baselines on 27 tasks and reduces API calls, token consumption and overall cost by 1.2 to 80.

Copied to clipboard

Salam Albatarni, May Bashendy, Sohaila Eltanbouly, Tamer Elsayed

RACER: Retrieval-Augmented Contextual Rapid Speculative Decoding (2026.findings-acl)

Challenge:	Current approaches to automate essay scoring (AES) treat each writing task as a separate task, resulting in inconsistent performance.
Approach:	They propose a meta-learning framework that leverages prototypical networks to learn transferable representations across different writing prompts.
Outcome:	The proposed framework outperforms baseline models on ELLIPSE and ASAP (English) and LAILA (Arabic) on three diverse datasets.

Copied to clipboard

Zihong Zhang, Zuchao Li, Lefei Zhang, Ping Wang, Hai Zhao

PEC-Home: Interpretation of Progressively Elliptical Commands in Smart Homes (2026.findings-acl)

Challenge:	Existing methods for decoding large language models generate one token per step, causing high inference latency.
Approach:	They propose a method that integrates retrieved exact patterns with logit-driven future cues.
Outcome:	Experiments on Spec-Bench, HumanEval, and MGSM-ZH show that RACER outperforms training-free methods and accelerates inference.

Copied to clipboard

Yingyu Shan, Zeming Liu, Silin Li, Boao Qian, Jiashu Yao, Yuhang Guo, Haifeng Wang

Towards IP Intelligence: Benchmarking Large Language Models on Intellectual Property Knowledge and Practice (2026.findings-acl)

Challenge:	Existing home assistants struggle to interpret elliptical commands based on ellipine expressions . current assistants overlook the progressive omission that occurs in human dialogue as context accumulates - limiting their effectiveness in real-world applications .
Approach:	They propose a simulated home dataset specifically designed for interpreting progressively elliptical commands in smart homes.
Outcome:	The proposed dataset shows that existing home assistants struggle to execute user-intended operations based solely on elliptical commands.

Copied to clipboard

Qiyao Wang, Guhong Chen, Hongbo Wang, Huaren Liu, Minghui Zhu, Zhifei Qin, Li Linwei, Yilin Yue, Shiqiang Wang, Jiayan Li, Wu Yihang, Ziqiang Liu, Longze Chen, Run Luo, Liyang Fan, Jiaming Li, Lei Zhang, Kan Xu, Hamid Alinejad-Rokny, Chengming Li, Shiwen Ni, Yuan Lin, Min Yang

EchoMLLM: Incentivizing Echocardiographic Video Understanding with Keyframe Grounding and Report Generation (2026.findings-acl)

Challenge:	Existing datasets and benchmarks focus only on patents or cover limited aspects of the IP field, lacking alignment with real-world scenarios.
Approach:	They propose a bilingual IP task taxonomy and a large-scale bilingual benchmark to evaluate LLMs in real-world IP practice.
Outcome:	The proposed model achieves only 75.8% accuracy, indicating room for improvement . open-source IP and law-oriented models lag behind closed-source general-purpose models .

Copied to clipboard

Heyu Huang, Wanran Sun, Chi Chen, Bo Chen, Zonghao Guo, Yuhua Li, Ruixuan Li, Kunlun He, Maosong Sun

HAUNTATTACK: When Attack Follows Reasoning as a Shadow (2026.findings-acl)

Challenge:	Echocardiography analysis requires a dual capability: rigorous quantitative keyframe localization and comprehensive qualitative synthesis.
Approach:	They propose a unified framework designed for real-world echocardiography video understanding.
Outcome:	a new framework is designed to support real-world echocardiography video understanding . it reduces temporal grounding errors by up to 76% and improves report generation quality by 65% .

Copied to clipboard

Jingyuan Ma, Rui Li, Zheng Li, Junfeng Liu, Heming Xia, Lei Sha, Zhifang Sui

Rethinking Multiple-Choice Questions for RLVR: Unlocking Potential via Distractor Design (2026.findings-acl)

Challenge:	Emerging Large Reasoning Models (LRMs) excel in mathematical and reasoning tasks, showcasing remarkable capabilities.
Approach:	They propose a framework that embeds harmful instructions into reasoning questions . they evaluate 11 LRMs and observe an average attack success rate of over 70% .
Outcome:	The proposed framework improves reasoning models by 13 percentage points over baseline.

Copied to clipboard

Xu Guo, Qiming Ge, Jian Tong, Kedi Chen, Jin Zhang, Xiaogui Yang, Xuan Gao, Haijun Lv, Zhihui Lu, Yicheng Zou, Qipeng Guo

S2ST-Omni: Hierarchical Language-Aware SpeechLLM Adaptation for Multilingual Speech-to-Speech Translation (2026.findings-acl)

Challenge:	Existing approaches to RLVR use multiple-choice questions as verifiable rewards . however, not all tasks provide reliable verification .
Approach:	They propose a framework that actively constructs high-quality distractors to block elimination shortcuts and promote deep reasoning.
Outcome:	The proposed method significantly improves reasoning capabilities of Large Language Models.

Copied to clipboard

Yu Pan, Xiongfei Wu, Yang Yuguang, Jixun Yao, Maxime Cordy, Lei Ma, Jianjun Zhao

G-LoRA: Global-Local Decoupled Low-Rank Adaptation (2026.findings-acl)

Challenge:	S2ST-Omni integrates a speech-to-text frontend with a modular, plug-and-play text-tospeech backend.
Approach:	They propose a compositional S2ST framework that integrates a speech-to-text frontend with a modular, plug-and-play text-tospeech backend.
Outcome:	The proposed framework outperforms existing frameworks in translation and synthesis . it integrates a speech-to-text translation frontend with a plug-and-play text-tospeech backend .

Copied to clipboard

Jiahao Xiong, Yihong Huang, Yihe Liu, Xianming Hu, Hongbo Zhao, Kai Zhang

Making MLLMs Blind: Adversarial Smuggling Attacks in MLLM Content Moderation (2026.findings-acl)

Challenge:	Low-Rank Adaptation (LoRA) improves the fine-tuning efficiency and performance of large language models.
Approach:	They propose a low-rank adaptive approach that decomposes update matrix into global and local adapters and assigns them to local and global adapters.
Outcome:	The proposed method achieves up to 2.7% accuracy improvement over LoRA and its variants on commonsense reasoning, mathematical reasoning, and code generation.

Copied to clipboard

Zhiheng Li, Zongyang Ma, Yuntong Pan, Ziqi Zhang, Xiaolei Lv, Bo Li, Jun Gao, Jianing Zhang, Chunfeng Yuan, Bing Li, Weiming Hu

LatentRefusal: Latent-Signal Refusal for Unanswerable Text-to-SQL Queries (2026.findings-acl)

Challenge:	Multimodal Large Language Models (MLLMs) are increasingly being deployed as content moderators . however, they exploit the Human-AI capability gap and create adversarial environments . smuggling attacks exploit the human-AI gap and exploit the vulnerability .
Approach:	They construct a benchmark to evaluate the vulnerability of MLLMs as content moderators . they identify three root causes: limited capabilities of vision encoders, robustness gap in OCR .
Outcome:	The proposed model exploits the Human-AI capability gap and is vulnerable to smuggling attacks.

Copied to clipboard

Xuancheng Ren, Shijing Hu, Zhihui Lu, Jiangqi Huang, Qiang Duan

Where meaning lives: Layer-wise accessibility of psycholinguistic features in encoder and decoder language models (2026.findings-acl)

Challenge:	Existing refusal strategies for unanswerable and underspecified user queries are brittle due to model hallucinations or add complexity and overhead.
Approach:	They propose a latent-signal refusal mechanism that predicts query answerability from hidden activations of an LLM.
Outcome:	The proposed scheme reduces schema noise and sparse, localized question–schema mismatch cues that indicate unanswerability.

Copied to clipboard

Taisiia Tikhomirova, Dirk U. Wulff

DynamicFocalPO: Adaptive Focusing Strategy for Preference Optimization (2026.findings-acl)

Challenge:	Existing studies show that transformer language models encode psychologically meaningful aspects of meaning at different depths.
Approach:	They conduct a layer-wise probing study of 58 psycholinguistic features across 10 transformer models . they find that apparent localization of meaning is method-dependent .
Outcome:	The results show that where meaning “lives” in transformer models reflects methodological choices and architectural constraints.

Copied to clipboard

Shu Zhou, Junan Chen, Rui Ling, Xin Wang, Tao Fan, Hao Wang

Graph-GRPO: Stabilizing Multi-Agent Topology Learning via Group Relative Policy Optimization (2026.findings-acl)

Challenge:	Recent preference optimization algorithms such as Direct Preference Optimization (DPO) have become prevalent for aligning large language models with human preferences.
Approach:	They propose a preference optimization algorithm that introduces a modulating factor that down-weighs misranked preference pairs and employs focusing strategy that adapts over the course of training.
Outcome:	Experiments show that DynamicFocalPO surpasses both DPO and FocalPO on benchmarks including Alpaca Eval 2.0 and Arena-Hard using Mistral-Base-7B and Llama-3-Instruct-8B.

Copied to clipboard

Yueyang Cang, Xiaoteng Zhang, Erlu Zhao, Zehua Ji, Yuhang Liu, Yuchen He, Zhiyuan Ning, Chen Yijun, Wenge Que, Li Shi

ReMedi: Reasoner for Medical Clinical Prediction (2026.findings-acl)

Challenge:	Recent approaches to optimize communication topology rely on single-sample policy gradients with absolute rewards.
Approach:	They propose a topology optimization framework that integrates Group Relative Policy Optimization.
Outcome:	The proposed topology optimization framework outperforms state-of-the-art methods on reasoning and code generation benchmarks.

Copied to clipboard

Yushi Cao, Yiming Chen, Hongchao Jiang, Hung-yi Lee, Robby T. Tan

JX4MEI: Multimodal Semantically-Enhanced LLM for Joint Multimodal Emotion-Intent Explanation and Classification (2026.findings-acl)

Challenge:	Existing approaches to predicting future clinical outcomes from EHRs focus on enhancing medical knowledge through distillation or RAG while relying on the model’s internal ability to interpret contextual information.
Approach:	They propose a framework for improving clinical outcome prediction from EHR using a sample regeneration mechanism that leverages ground-truth answers as hints to enhance reasoning.
Outcome:	Experiments on multiple EHR prediction tasks show significant gains of up to 19.9% over state-of-the-art baselines in terms of F1 score, underscoring ReMedi’s effectiveness in real-world clinical prediction.

Copied to clipboard

YiJie Huang, Xiaocui Yang, Shi Feng, Daling Wang, Yifei Zhang, Ning Yuan, Zhuoyue Jia, Wen Zhang

Breaking the "Provable Security": Detecting Finite-Precision Artifacts in LLM-based Steganography via Low-Probability Vanishing (2026.findings-acl)

Challenge:	Existing multimodal emotion and intent recognition tasks focus on classification, not rationale and intrinsic connections between these states.
Approach:	They propose a task that requires models to jointly predict emotion and intent while generating natural language explanations for why they co-occur.
Outcome:	The proposed model outperforms baseline models in prediction and explanation generation.

Copied to clipboard

Wenzhao Cao, Yaofei Wang, Donghui Hu

FACT-E: Causality-Inspired Evaluation for Trustworthy Chain-of-Thought Reasoning (2026.findings-acl)

Challenge:	Recent advances in Large Language Models have fostered a new class of generative linguistic steganography, claim “provably secure” by theoretically aligning the stego distribution with the language model’s natural distribution.
Approach:	They propose a framework that transforms the detection task from semantic classification to a statistical audit of the sampling mechanism.
Outcome:	The proposed framework breaks the security of AC and Meteor with high detection accuracy, whereas state-of-the-art semantic steganalyzers degrade to random guessing.

Copied to clipboard

Yuxi Sun, Aoqi Zuo, Haotian Xie, Wei Gao, Mingming Gong, Jing Ma

DRIFT: Transferring Reasoning Priors for Efficient MLLM Fine-Tuning (2026.findings-acl)

Challenge:	Existing models generate explanations that appear coherent while containing unfaithful intermediate steps.
Approach:	They propose a causality-inspired framework for evaluating CoT quality using controlled perturbations as an instrumental signal to separate genuine step-to-step dependence from bias-driven artifacts.
Outcome:	Experiments on GSM8K, MATH, and CommonsenseQA show that FACT-E improves reasoning-trajectory selection and yields stronger in-context learning exemplars.

Copied to clipboard

Chao Huang, Zeliang Zhang, Jiang Liu, Ximeng Sun, Jialian Wu, Xiaodong Yu, Ze Wang, Chenliang Xu, Emad Barsoum, Zicheng Liu

Mitigating Hallucinations in VLMs: Enhancing Visual Attention via Head-Wise Perturbation (2026.findings-acl)

Challenge:	Multimodal large language models (MLLMs) have made rapid progress in perception and alignment, but their reasoning ability often lags behind strong text-only LLMs.
Approach:	They propose a method that transfers reasoning knowledge in the gradient space while preserving multimodal alignment.
Outcome:	Experiments on multimodal reasoning benchmarks show that DRIFT outperforms naive merging and standard SFT.

Copied to clipboard

Zhenghua Wang, Yixin Wu, Feiran Zhang, Qi Qian, Changze Lv, Xuanjing Huang, Xiaoqing Zheng

Modeling Multi-Dimensional Cognitive States in Large Language Models under Cognitive Crowding (2026.findings-acl)

Challenge:	Vision–Language Models (VLMs) have demonstrated strong capabilities in tasks that require joint understanding of text and images.
Approach:	They propose a strategy that incorporates head-wise attention perturbation via continuous multiplicative noise coupled with a visual-guided loss focused on vision-sensitive text tokens to promote a more balanced attention distribution.
Outcome:	The proposed approach outperforms baseline models on three benchmarks and consistently outperformed the baseline model.

Copied to clipboard

Lin Zhong, Siyu Zhu, Zizhen Yuan, Jinhao Cui, Xinyang Zhao, Lingzhi Wang, Hao Chen, Qing Liao

EmoMM: Benchmarking and Steering MLLM for Multimodal Emotion Recognition under Conflict and Missingness (2026.findings-acl)

Challenge:	Existing Large Language Models (LLMs) mainly address isolated tasks such as emotion analysis or stance detection.
Approach:	They propose a large-scale model that combines large-level annotations with hyperbolic space to model human cognitive states.
Outcome:	The proposed model outperforms baseline models on cognitive dimensions on single dimension tasks while retaining strong hierarchical structure.

Copied to clipboard

Yueru Sun, Yimeng Zhang, Haoyu Gu, Nuo Chen, Dong She, Xianrong Yao, Yang Gao, Zhanpeng Jin

Veri-R1: Toward Precise and Faithful Claim Verification via Online Reinforcement Learning (2026.findings-acl)

Challenge:	Multimodal Large Language Models (MLLMs) have shown promise in MER, but their internal decision-making mechanisms under modality conflict and missingness remain underexplored.
Approach:	They propose a multimodal large language model that can detect and control modality conflicts and missing subsets by a lightweight mechanism that detects and controls modality conflict.
Outcome:	The proposed framework improves performance across settings, showing it can handle conflict and missing behaviors.

Copied to clipboard

Qi He, Cheng Qian, Xiusi Chen, Bingxiang He, Yi R. Fung, Heng Ji

MimicLM: Zero-Shot Voice Imitation through Autoregressive Modeling of Pseudo-Parallel Speech Corpora (2026.findings-acl)

Challenge:	Existing approaches to online claim verification rely on prompt engineering or pre-designed reasoning workflows.
Approach:	They propose an online reinforcement learning framework that enables an LLM to interact with a search engine and receive reward signals that explicitly shape its planning, retrieval, and reasoning behaviors.
Outcome:	Empirical results show that Veri-R1 improves joint accuracy by 30% and doubles evidence score, often surpassing larger-scale model counterparts.

Copied to clipboard

Tao Feng, Yuxiang Wang, Yuancheng Wang, Xueyao Zhang, Dekun Chen, Chaoren Wang, Xun Guan, Zhizheng Wu

Fine-Grained Data Ordering Improves Fine-Tuning for Large Language Models (2026.findings-acl)

Challenge:	Existing approaches to voice imitation use complex model design and a quality ceiling when synthetic speech is used as training sources.
Approach:	They propose a model that uses synthetic speech as training sources while retaining real recordings as targets.
Outcome:	The proposed model outperforms existing methods in naturalness while maintaining competitive similarity scores across speaker identity, accent, and emotion dimensions.

Copied to clipboard

Xiaomeng Hu, Yixuan Tang, Haoze Li, Hao Chen, Qi Zhang, Zhanming Shen, Yiming Zhang, Haobo Wang, Junbo Zhao

LifeSim: Long-Horizon User Life Simulator for Personalized Assistant Evaluation (2026.findings-acl)

Challenge:	Prior work focused on data preprocessing, focusing on filtering and cleaning data . a study aimed to improve fine-grained scheduling of data order in epochs .
Approach:	They propose a fine-grained scheduling method of data order in epochs to fill this gap . they define data difficulty based on relevance between data and model .
Outcome:	The proposed method improves on pre-training and small-scale fine-tuning experiments 2.4% over baselines.

Copied to clipboard

Feiyu Duan, Xuanjing Huang, Zhongyu Wei

Challenge:	Existing benchmarks for personalized assistants fail to capture the complexity of external contexts and users’ cognitive states.
Approach:	They propose a user simulator that models user cognition through the Belief-Desire-Intention model within physical environments for coherent life trajectories generation and simulates intention-driven user interactive behaviors.
Outcome:	The proposed model can model user cognition through the Belief-Desire-Intention model within physical environments for coherent life trajectories generation and simulates intention-driven user interactive behaviors.

Interactive Semantic Parsing with Reinforcement Learning for Knowledge Graph Reasoning (2026.findings-acl)

Copied to clipboard

Yurun Song, Xiangqing Shen, Jianfei Yu, Rui Xia

CLewR: Curriculum Learning with Restarts for Machine Translation Preference Learning (2026.findings-acl)

Challenge:	Existing approaches to improve LLM reliability rely on factual hallucinations . Existing methods rely only on graph traversal, resulting in imprecise retrieval and heavy post-processing burdens.
Approach:	They propose a framework that integrates knowledge Graphs as structured, high-fidelity buffers to enhance LLM reliability.
Outcome:	The proposed framework allows logical constraints to be dynamically interleaved with graph search while optimizing via reinforcement learning with only final answer feedback eliminates the need for gold program annotations.

Copied to clipboard

Alexandra Dragomir, Florin Brad, Radu Tudor Ionescu

Targeting the Needle, Ignoring the Haystack: Anchoring Crucial Cues for Evolving Scam Call Detection via an LLM-Assisted Classifier (2026.findings-acl)

Challenge:	Large language models (LLMs) have demonstrated competitive performance in zero-shot multilingual machine translation (MT).
Approach:	They propose a curriculum learning strategy with restarts which reiterates easy-to-hard curriculum multiple times during training to effectively mitigate catastrophic forgetting of easy examples.
Outcome:	The proposed model replicates easy-to-hard curriculum multiple times during training to mitigate catastrophic forgetting of easy examples.

Copied to clipboard

Tong Wu, Qinliang Su, Jianxing Yu, Bo Liang, Minhua Huang

Distributional Alignment for Large Language Models under Domain Shift (2026.findings-acl)

Challenge:	Existing methods for fraud detection on online service platforms often fail to generalize due to the scarcity of labeled data and the continuous evolution of conversational contexts.
Approach:	They propose a framework that anchors detection on Semantic Primitives . they prioritize stable evidence over conversational noise to ensure a verifiable fraud tactic .
Outcome:	The proposed framework achieves superior robustness and efficiency compared to baselines . it prioritizes stable evidence over diverse conversational noise .

Copied to clipboard

Viet Thanh Pham, Lizhen Qu, Zhuang Li, Gholamreza Haffari

Traces in the Brain: Neural Evidence for Syntactic Movement in English and Chinese (2026.findings-acl)

Challenge:	Existing distributional alignment models are unstable and degrade under cultural and domain shifts.
Approach:	They propose a distributional alignment technique that improves distribution prediction under cultural and domain shift.
Outcome:	The proposed method improves fidelity and robustness of LLM distribution estimation under domain and cultural shift.

Copied to clipboard

Yuhan Huang, Zhengwu Ma, Yuqi Jin, Beth Chan, Zheng Shen, Jackie Yan-Ki Lai, John T. Hale, Jixing Li

LinguaGame: A Linguistically Grounded Game-Theoretic Paradigm for Multi-Agent Dialogue Generation (2026.findings-acl)

Challenge:	Syntactic movement is a core concept in generative linguistics to account for word-order variation and long-distance dependencies.
Approach:	They annotated every sentence in the audiobook The Little Prince using X-bar style tree annotations.
Outcome:	The proposed model shows that deep structure significantly predicts neural responses in English but not in Chinese.

Copied to clipboard

Yuxiao Ye, Yiming Zhang, Yiran Rex Ma, Huiyuan Xie, Huining Zhu, Zhiyuan Liu

A Dual-View Analysis of Multiple Languages in Colonial Newspapers (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have enabled Multi-Agent Systems (MASs) where agents interact through natural language to solve complex tasks or simulate multi-party dialogues.
Approach:	They propose a linguistically-grounded game-theoretic paradigm for multi-agent dialogue generation that uses a training-free equilibrium approximation algorithm to model dialogue over communicative intents and strategies.
Outcome:	The proposed framework improves agents’ communication efficiency by helping them convey their intended meaning more effectively through language.

Copied to clipboard

Zhan Su, Xiaoya Chen, Fengran Mo, Ida L. Vos, Prayag Tiwari, Yazhou Zhang, Qian Zheng, Natália da Silva Perez

CARE-CR: Context-Aware Routing and Expert Fusion for Multi-Preference Cognitive Restructuring (2026.findings-acl)

Challenge:	Historical newspapers from the colonial period offer valuable evidence of how racializing language evolved over time.
Approach:	They propose a contextual question answering and visual question answering task from colonial newspapers . they propose linguistic training for temporal word embedding with a compass to study racialization .
Outcome:	The proposed tasks are limited for low-resource tasks, the authors show . the authors compare the results of two QA pairs from colonial newspapers to a compass .

Copied to clipboard

Hongzhi Qi, Liangcheng Wang, Yijing YU, Jianqiang Li, Bing Xiang Yang, Qing Zhao

Cognitive Modeling

Challenge:	Large Language Models (LLMs) offer promising avenues for automated cognitive restructuring in mental health settings, but current approaches lack the adaptability to balance conflicting therapeutic dimensions, such as empathy and rationality.
Approach:	They propose a decoupled optimization framework that implements a dimension-guided Monte Carlo tree search to train expert policies specialized for distinct therapeutic attributes rather than relying on a monolithic alignment strategy.
Outcome:	The proposed framework achieves consistent improvements over baselines across multiple evaluation dimensions, including diagnostic accuracy, contextual appropriateness, task effectiveness, and overall helpfulness, while enabling controllable cognitive restructuring generation.

EvoCoT: Overcoming the Exploration Bottleneck in Reinforcement Learning for LLMs (2026.findings-acl)

Copied to clipboard

Huanyu Liu, Jia Li, Yihong Dong, Chang Yu, Taozhi Chen, Lecheng Wang, Yongding Tao, Bin Gu, Ge Li

Context as a Tool: Context Management for Long-Horizon SWE-Agents (2026.findings-acl)

Challenge:	Existing approaches to reinforcement learning with verifiable reward (RLVR) are limited by difficulty or lack of exploration.
Approach:	They propose a self-evolving curriculum learning framework based on chain-of-thought reasoning optimization that constrains exploration space by self-generating and verifying CoT trajectories.
Outcome:	The proposed framework enables LLMs to solve previously unsolved problems without external supervision and is compatible with various RL fine-tuning methods.

Copied to clipboard

Shukai Liu, Bo Jiang, Jian Yang, Yizhi LI, Jinyang Guo, Xianglong Liu, Bryan Dai

Beyond Static Profiles: Capturing the Fluidity of User Preferences in Diverse Scenarios (2026.findings-acl)

Challenge:	Existing large language models rely on append-only context maintenance or passively triggered compression heuristics, leading to context explosion, semantic drift, and degraded reasoning in long-running interactions.
Approach:	They propose a new context management paradigm that elevates context maintenance to a callable tool . they propose 'cat' framework that injects context-management actions into complete interaction trajectories .
Outcome:	The proposed model outperforms ReAct-based agents and static compression baselines on SWE-Verified tests.

Copied to clipboard

Chunyang Gao, Yi Huang, Jingyu Yao, Xiaoting Wu, Junlan Feng

Challenge:	Existing approaches to personalize Large Language Models often default to homogeneous behaviors . preferences can shift, and conflict, depending on context, authors argue .
Approach:	They propose a hierarchical taxonomy to differentiate between stable and situational preferences . they use a dataset of 10k meticulously curated preferences to test their taxonomies .
Outcome:	The proposed model differentiates between stable and situational preferences based on curated user preferences . it provides a practical testbed for advancing dynamic, context-aware personalization in conversational agents.

Mitigating Coordinate Prediction Bias from Positional Encoding Failures (2026.findings-acl)

Copied to clipboard

Xingjian Tao, Yiwei Wang, Yujun Cai, Yihong Luo, Kai Han, Jing Tang

PerDucer: Keyphrase-Driven Personalization Inducer for Summarization from User Histories (2026.findings-acl)

Challenge:	Multimodal Large Language Models (MLLMs) excel at general vision-language tasks, but precise coordinate prediction remains a challenge.
Approach:	They propose a training-free, inference-time correction method to correct VPEs . they isolate position-unconditioned tendencies by shuffling VPE and use it to steer digit decoding .
Outcome:	The proposed method is training-free, inference-time correction method . it effectively rectifies coordinate drift, yielding consistent improvements without retraining .

Copied to clipboard

Parthiv Chatterjee, Asish Joel Batha, Sourish Dasgupta, Tanmoy Chakraborty

Summarization

Challenge:	Prior work reported that prepending long interaction histories to LLMs leads to unstable personalization, especially for multi-aspect documents.
Approach:	They propose a personalization inducer for frozen language models that maps latent preference signals to a small set of personalized keyphrases for the query document.
Outcome:	The proposed model outperforms the strongest history-prompting LLMs and SLMs in the PENS and OpenAI-Reddit benchmarks.

Beyond A Fixed Seal: Adaptive Stealing Watermark in Large Language Models (2026.findings-acl)

Copied to clipboard

Shuhao Zhang, Yuli Chen, Jiale Han, Bo Cheng, Jiabao Ma

UniGeM: Unifying Data Selection and Mixing via Geometric Exploration and Mining (2026.findings-acl)

Challenge:	Existing detection methods for large language models rely on fixed strategies to steal watermarks.
Approach:	They propose a novel steal-based watermark algorithm that derives watermark information from watermarked texts to craft highly targeted adversarial attacks.
Outcome:	The proposed system significantly increases steal efficiency against target watermarks under identical conditions.

Copied to clipboard

Changhao Wang, null Yunfeiyu, Xinhao Yao, Jiaolong Yang, Lu Yu, Junpeng Fang, Chaobo Li, Riccardo Cantoro, Qing Cui, Jun Zhou

Challenge:	Large Language Models (LLMs) scaling is limited by data quality and domain mixing and instance selection are two separate problems.
Approach:	They propose a framework that unifies mixing and selection without training proxy models or relying on external reference datasets.
Outcome:	The proposed framework achieves 2.0 data efficiency over a random baseline and further improves overall performance compared to SOTA methods in reasoning-heavy evaluations and multilingual generalization.

Don’t Just Listen, Try Planning: Graph-based Retrieval-Generation Agent for Long-form Audio Meeting Understanding (2026.findings-acl)

Copied to clipboard

Quanwei Tang, Dong Zhang, Shoushan Li, Guodong Zhou

FLASH: Focused Layer Attention Sink Hijacking (2026.findings-acl)

Challenge:	Existing question answering (QA) datasets for long audio meetings suffer from acoustic information loss and poor long-term dependency capture.
Approach:	They propose a question answering dataset that captures three core dimensions of long-form audio meeting content.
Outcome:	The proposed model captures three core dimensions of long-form audio meeting content: complex semantics, multi-speaker interactions, and quite long timestamps.

Copied to clipboard

Siyuan Deng, Yerong Li, Fanhua Shang, Hongying Liu

SPIDE: Serial and Parallel Intertwined Speculative Decoding (2026.findings-acl)

Challenge:	Large Language Models (LLMs) remain vulnerable to jailbreaking attacks despite advances in safety alignment .
Approach:	They propose a new diagnostic auditing framework that dismantles the model's internal safety anchor by precisely scaling attention scores in these vulnerable layers.
Outcome:	The proposed framework achieves a state-of-the-art Attack Success Rate of over 77% with an unprecedented efficiency of 1.53 queries on average.

Copied to clipboard

Wenru Xu, Peixuan Xu, Ziqi Yang, Ming Hu, Zihui Wang, Jianzhong Qi, Rongshan Yu, Xiaoliang Fan, Cheng Wang

MdEval: Massively Multilingual Code Debugging (2026.findings-acl)

Challenge:	Speculative decoding (SD) is a training-free SD framework that orchestrates dynamic alternation combining serial dynamic drafting with parallel draft verification.
Approach:	They propose a serial and parallel intertwined speculative DEcoding framework that orchestrates dynamic alternation combining serial dynamic drafting and parallel draft verification.
Outcome:	The proposed framework accelerates inference while reducing the LLM usage costs.

Copied to clipboard

Shukai Liu, Linzheng Chai, Jian Yang, Jiajun Shi, He Zhu, Liran Wang, Jin Ke, Wei Zhang, Hualei Zhu, Shuyue Guo, Tao Sun, Jiaheng Liu, Yunlong Duan, Yu Hao, Liqun Yang, Guanglin Niu, Ge Zhang, Zhoujun Li

Breaking Consensus Bias: Unsupervised Reinforcement Learning for Machine Translation (2026.findings-acl)

Challenge:	Existing benchmarks primarily focus on Python and are limited in terms of language diversity.
Approach:	They propose a multilingual debugging benchmark that includes 3.9K test samples of 20 programming languages and introduces the debug instruction corpora MdEval-Instruct by injecting bugs into the correct multilingual queries and solutions.
Outcome:	The proposed benchmark includes 3.9K test samples of 20 programming languages and covers the automated program repair task, bug localization task, and bug identification task.

Copied to clipboard

Shuting Jiang, Ran Song, Siqi Zhang, Yuxin Huang, Shengxiang Gao, Zhengtao Yu

Reasoning-Aware AIGC Detection via Alignment and Reinforcement (2026.findings-acl)

Challenge:	Existing RL approaches for MT face fixed references or the production of homogeneous references leading to mode collapse in unsupervised settings.
Approach:	They propose an Entropy-Driven Unsupervised RL framework for machine translation that leverages entropy for supervision construction and self-evolution.
Outcome:	The proposed framework outperforms supervised and unsupervised baselines in multiple language pairs.

Copied to clipboard

Zhao Wang, Max Xiong, Jianxun Lian, Zhicheng Dou

Beyond Modality Collapse: Taming Guided Modality Entropy for Omni-modal Emotion Reasoning (2026.findings-acl)

Challenge:	Existing approaches to AIGC detection have relied on statistical classifiers or black-box neural models, which exploit surface-level patterns and struggle to generalize as LLMs evolve.
Approach:	They propose a framework that generates interpretable reasoning chains before classification using supervised fine-tuning and reinforcement learning to improve accuracy.
Outcome:	The proposed framework achieves state-of-the-art performance across multiple benchmarks, offering a robust and transparent solution for AIGC detection.

Copied to clipboard

Xian Zhao, Rui Hu, Yuxiang Zhang, Delai Qiu, Yining Wang, Shengping Liu, Jian Yu, Jitao Sang

ID10M-JAM: Stress-Testing Idiom Identification Under Challenging Context (2026.findings-acl)

Challenge:	EmoOmni is a data paradigm for omni-modal large language models that can be used for emotion reasoning.
Approach:	They propose a data paradigm that interleaves guided tokens into reasoning traces to enforce structured evidence extraction.
Outcome:	The proposed paradigm over-relys on a dominant modality while neglecting complementary cues.

Copied to clipboard

Kai Golan Hashiloni, Lior Livyatan, Ofri Hefetz, Alon Mannor, Bar Cohen, Kfir Bar

Challenge:	Large language models (LLMs) achieve strong performance on idiom identification benchmarks, yet their robustness to misleading contextual signals remains largely untested.
Approach:	They propose an adversarial extension of the ID10M dataset that jams idiom understanding by injecting coherent but conflicting context before each target sentence.
Outcome:	The proposed benchmark exposes systematic vulnerabilities in LLMs’ contextual reasoning, pushing idiom identification to its breaking point.

MENTOR: Mitigating Identity Drift in Dynamic Role-Playing via Dual-Chain Structured Memory (2026.findings-acl)

Copied to clipboard

Zhu Zhuoning, Xingyu Gao, Hailong Shi

Challenge:	Long-context LLMs are evolving into long-term agents that interact with users over extended horizons.
Approach:	They propose a cognitive architecture that mitigates identity drift without fine-tuning . MENTOR uses a Dual-Chain Memory Mechanism: a Global Chain (G) for long-term event logging and isolated Role Chains (Rr) as per-role working memories.
Outcome:	The proposed architecture improves the overall score (Avg) from 0.46 to 0.75 on average, with substantial gains in identity adherence and knowledge fidelity.

Beyond Benchmarks: A Capability-Based Maturity Model for Systematic AI Integration in Hospitals (2026.findings-acl)

Copied to clipboard

Rui Li, Xiaofen WU, Mingqian Liu, Xiaoxia Song, Xu Xiangjun, Jiacheng Qiao, Zhuang Boqin, Xu Chen

Challenge:	Current Large Language Models (LLMs) excel in standardized tests focused on medical knowledge recall, but not in real-world healthcare scenarios.
Approach:	They propose a "capability-based hospital AI Maturity Model" framework that categorizes capabilities into distinct maturity levels . medical artificial intelligence is currently at a critical transition stage from technical verification to deep clinical integration .
Outcome:	The proposed model provides a clear, stepwise evolutionary path for hospitals from foundational infrastructure construction to ubiquitous intelligence.

DiffuSpec: Unlocking Diffusion Language Models for Speculative Decoding (2026.findings-acl)

Copied to clipboard

Guanghao Li, Zhihui Fu, Min Fang, Qibin Zhao, Ming Tang, Chun Yuan, Jun Wang

TENP: Trapezoidal Expert Neuron Pruning For Mixture-of-Experts (2026.findings-acl)

Challenge:	Autoregressive (AR) decoding in large language models is latency-bounded by strictly sequential token generation.
Approach:	They propose a diffusion-based drafter that proposes multi-token candidates and then verifies them in parallel by the target model.
Outcome:	The proposed drafter generates multi-token proposals in a single forward pass while remaining compatible with standard AR verifiers.

Copied to clipboard

Jiangyang He, Shaolin Zhu, Deyi Xiong

FlowRAG: Synergizing Explicit Reasoning via Frequency-Aware Multi-Granularity Graph Flow (2026.findings-acl)

Challenge:	Existing compression approaches remove entire experts, disrupting routing topology and harming performance, or rely on unstructured weight pruning with limited practical efficiency.
Approach:	They propose a structured Trapezoidal Expert Neuron Prunning framework that uses a trapezoidal pattern to identify and retain important experts while applying expert neuron pruning (ENP) to less important experts.
Outcome:	The proposed framework outperforms the full-parameter model by 10% on code generation tasks under a sparse activation of experts and a 40% routing sparsity.

Copied to clipboard

Bihao Zhan, Zongsheng Cao, Jie Zhou, Bo Zhang, Liang He

OntoGuard: Enforcing Action Admissibility for LLM Agents in Complex Interactive Environments (2026.findings-acl)

Challenge:	Existing methods for Graph-based retrieval-augmented generation rely on implicit semantic relevance propagation.
Approach:	They propose a semantic-aware retrieval framework that improves both semantic recall and explicit reasoning.
Outcome:	Extensive experiments show that FlowRAG improves both semantic recall and explicit reasoning.

Copied to clipboard

Pengxiang Liu, Tao Ren, Wei Xiong, Tingrui Yang, Junjie Wang, Jun HU

Challenge:	Existing approaches to large language models (LLMs) are limited by their ability to enforce environmental and behavioral admissibility.
Approach:	They propose an ontological framework to guard LLM agents by enforcing environmental and behavioral admissibility.
Outcome:	Experiments on ScienceWorld and VirtualHome show that OntoGuard can enforce environmental and behavioral admissibility while preventing invalid actions.

Self-Correcting RAG: Enhancing Faithfulness via MMKP Context Selection and NLI-Guided MCTS (2026.findings-acl)

Copied to clipboard

Shijia Xu, Zhou Wu, Xiaolong Jia, Yu Wang, Kai Liu, April Xiaowen Dong

DiffER: Diffusion Entity-Relation Modeling for Reversal Curse in Diffusion Large Language Models (2026.findings-acl)

Challenge:	Existing approaches to retrieval-augmented generation still face problems with low context utilization and frequent hallucinations.
Approach:	They propose a framework that reformulates retrieval and generation as constrained optimization and path planning.
Outcome:	The proposed framework significantly improves reasoning accuracy on complex queries while reducing hallucinations.

Copied to clipboard

Shaokai He, Kaiwen Wei, Xinyi Zeng, Xiang Chen, Xue Yang, Zhenyang Li, Jiang Zhong, Yu Tian

Layer-aware Dual-directional Modulation for Low-resource Machine Translation (2026.findings-acl)

Challenge:	Existing large language models exhibit unidirectional behavior when processing bidirectional relationships . authors propose a solution to alleviate the reversal curse in Diffusion LLMs .
Approach:	They propose a model that addresses the "reversal curse" of bidirectional behavior in large language models . they propose 'entity-aware training' and balanced data construction to alleviate asymmetry and missing relations .
Outcome:	The proposed model alleviates the "reversal curse" in Diffusion LLMs . the proposed model employs whole-entity masking to mitigate entity fragmentation .

Copied to clipboard

Siqi Zhang, Ran Song, Shuting Jiang, Yuxin Huang, Zhengtao Yu

MATCHA: Matching Text via Contrastive Semantic Alignment (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have demonstrated excellent performance in Machine Translation (MT) however, a performance gap persists between high-resource and low-resourced languages due to imbalanced pre-training data.
Approach:	They propose a layer-wise metric to quantify the activation divergence between high- and low-resource languages.
Outcome:	The proposed model outperforms standard LoRA fine-tuning on Chinese-to-seven low-resource language translations.

Copied to clipboard

Siran Li, Ece Sena Etoglu, Carsten Eickhoff, Seyed Ali Bahrainian

PlaM: Training-Free Plateau-Guided Model Merging for Better Visual Grounding in MLLMs (2026.findings-acl)

Challenge:	MATCHA is an automatic metric that rewards semantic agreement with a reference and penalizes contradictions.
Approach:	They introduce a metric that jointly rewards semantic agreement with a reference and penalizes contradictions.
Outcome:	The proposed metric outperforms popular metrics on eight public benchmarks compared with human annotations on question-answering, image caption generation, natural language inference, summarization, and semantic textual similarity tasks.

Copied to clipboard

Zijing Wang, YongKang Liu, Mingyang Wang, Ercong Nie, Deyuan Chen, Zhengjie Zhao, Shi Feng, Daling Wang, Xiaocui Yang, Yifei Zhang, Hinrich Schuetze

Efficient Inference for Large Vision-Language Models: Bottlenecks, Techniques, and Prospects (2026.findings-acl)

Challenge:	Multimodal instruction fine-tuning degrades textual reasoning capability, undermining multimodal performance.
Approach:	They propose a plateau-guided model merging method that selectively injects base language model parameters into MLLMs to mitigate this degradation.
Outcome:	The proposed framework reduces multimodal instruction fine-tuning degradation by incorporating a plateau-guided model merging method into MLLMs.

Copied to clipboard

Jun Zhang, Yicheng Ji, Feiyang Ren, Yihang Li, Bowen Zeng, Zonghao Chen, Ke Chen, Lidan Shou, Gang Chen, Huan Li

Adaptive Zooming via Relevance-Informed Positional Resource Allocation for Training-free LLM Context Extension (2026.findings-acl)

Challenge:	Large Vision-Language Models are hindered by a systemic efficiency barrier known as visual token dominance.
Approach:	They propose a systematic taxonomy of efficiency techniques structured around the inference lifecycle . they examine visual encoding, prefilling, and decoding to understand bottlenecks .
Outcome:	The proposed techniques reveal how upstream decisions dictate downstream bottlenecks . the proposed techniques include hybrid compression and modality-aware decoding .

Copied to clipboard

Hongbo Zhao, Huibin Wang, Bin Tang, Xianming Hu, Yihong Huang, Yijun Shen, Nuoyi Chen, Ping Li, Kai Zhang

Mitigating Position Bias in Transformers via Layer-Specific Positional Embedding Scaling (2026.findings-acl)

Challenge:	Existing training-free methods for extrapolating beyond training context lengths are semantics-agnostic . Existing methods that focus on relative token distances can indiscriminately blur semantically relevant and irrelevant tokens .
Approach:	They propose an adaptive positional zooming method that uses semantic relevance to extrapolate beyond training context lengths.
Outcome:	Experiments show that RiPRA outperforms existing training-free extrapolation methods . relevant tokens get higher positional resolution, while irrelevant tokens are compressed .

Copied to clipboard

Zhenghua Wang, Yiran Ding, Changze Lv, Yixin Wu, Tianlong Li, Zhibo Xu, Muling Wu, Tianyuan Shi, Shizheng Li, Qi Qian, Xuanjing Huang, Xiaoqing Zheng

Semantic-pragmatic Annotations in the Prague Dependency Treebank (2026.findings-acl)

Challenge:	Existing methods to address the "lost-in-the-middle" problem suffer from high latency or suboptimal hand-crafted scaling strategies.
Approach:	They propose a layer-specific positional embedding scaling method that assigns distinct scaling factors to each layer.
Outcome:	Experiments show that the proposed method mitigates positional attention bias and delivers consistent improvements across multiple long-context benchmarks.

Copied to clipboard

Marie Mikulová, Eva Hajicova, Jiří Mírovský, Anna Nedoluzhko, Michal Novák, Pavlína Synková, Jan Štěpánek, Barbora Štěpánková, Jan Hajič

Penetrating Linguistic Disguises: A Slang-aware Label-Aligned Framework for Fine-Grained Toxicity Extraction in Chinese Hate Speech Detection (2026.findings-acl)

Challenge:	morphology and syntax work on sentence level, but semantic-pragmatic phenomena are often related to two or more neighbouring sentences and possibly to an extra-linguistic context.
Approach:	They present semantic-pragmatic specification and annotations in the Prague Dependency Treebank - Consolidated 2.0 release2 by annotating the entire corpus.
Outcome:	The proposed annotations are based on the Prague Dependency Treebank -Consolidated 2.0 (PDT-C 2.0) the dataset contains more than 3 million tokens (of Czech) manually annotated from morphology to surface and deep syntax including several types of semantic-pragmatic annotations.

Copied to clipboard

Wei Liu, Xiaoliang Chen, Duoqian Miao, Xu Gu, Xianyong Li, Yajun Du

LightMoE: Task-Aware Expert Availability Management for Memory-Efficient MoE-LLM Inference (2026.findings-acl)

Challenge:	Flexible word boundaries and linguistic obfuscation, particularly slang, challenge precise span-level hate speech detection in Chinese.
Approach:	They propose a Slang-aware Label-Aligned Framework that maps slang to explicit hate semantics and uses task-specific branches to mitigate feature interference.
Outcome:	The proposed framework reduces ambiguity by mapping obscure slang to explicit hate semantics.

Copied to clipboard

Puhan Luo, Yunhao Yao, Junyang Wang, Junyang Zhang, Xiangyang Li

SelfBudgeter: Adaptive Token Allocation for Efficient LLM Reasoning (2026.findings-acl)

Challenge:	Existing solutions for balancing model accuracy with inference latency are limited due to memory constraints.
Approach:	They propose a framework for memory-efficient MoE inference that exploits the functional redundancy and temporal locality of expert activation.
Outcome:	The proposed framework improves accuracy-efficiency trade-off by 4.3% over pruned models and 2.4% over dynamic swapping methods while maintaining inference latency comparable to pruned model.

Copied to clipboard

Zheng Li, Qingxiu Dong, Jingyuan Ma, Di Zhang, Kai Jia, Zhifang Sui

Streaming Hallucination Detection in Long Chain-of-Thought Reasoning (2026.findings-acl)

Challenge:	Recent large reasoning models have shown exceptional performance on various tasks, but they consume excessive tokens even for simple queries, leading to resource waste and prolonged user latency.
Approach:	They propose a self-adaptive reasoning strategy that automatically allocates budgets according to problem complexity and introduces GRPO for reinforcement learning to reduce output length.
Outcome:	The proposed model achieves an average response length compression of 61% on math reasoning tasks while maintaining accuracy.

Copied to clipboard

Haolang Lu, Minghui Pan, Ripeng LI, Guoshun Nan, Jialin Zhuang, Zijie Zhao, Zhongxiang Sun, Kun Wang, Yang Liu

Training-Free Adaptive Speculative Decoding via Linguistic Priors (2026.findings-acl)

Challenge:	Long chain-of-thought reasoning improves performance of large language models, yet hallucinations in such settings often emerge subtly and propagate across reasoning steps.
Approach:	They propose to treat step-level hallucination judgments as local observations and introduce a cumulative prefix-level signal that tracks the global evolution of the reasoning state over the entire trajectory.
Outcome:	The proposed method enables streaming hallucination detection in long CoT reasoning, providing real-time, interpretable evidence.

Copied to clipboard

Jingyi Wang, Jiaqi Huang, Zunnan Xu, Jun Zhou, Kehong Yuan, Xiang Qian

Knowledge-Infused Multi-Bit Watermarking for RAG Knowledge Bases (2026.findings-acl)

Challenge:	Speculative decoding (SPD) is a promising technique to accelerate Large Language Models (LLMs). current approaches neglect the inherent heterogeneity of natural language and fail to distinguish between semantically-rich content and structurally-predictable syntax.
Approach:	They propose a training-free framework that leverages linguistic priors to enable adaptive drafting and verification.
Outcome:	The proposed framework significantly accelerates inference without additional training.

Copied to clipboard

Ke Yang, Shuguang Yuan, Jing Yu, Chi Chen

ClimAgent: LLM as Agents for Autonomous Open-ended Climate Science Analysis (2026.findings-acl)

Challenge:	Existing RAG watermarking methods are limited in their encoding capacity and potential degradation of performance or knowledge quality.
Approach:	They propose knowledge-infused and multi-bit watermarking (KMW) for RAG knowledge bases by benign knowledge completion and a tailored generative watermark algorithm.
Outcome:	The proposed method extracts watermarks from adversarial RAGs while remaining stealthy and secure.

Copied to clipboard

Hao Wang, Jindong Han, Wei Fan, Hao Liu

Challenge:	Existing approaches to climate research are limited to simple Q A tasks . a lack of data and computational expertise has created bottlenecks .
Approach:	They propose a general-purpose autonomous framework to perform end-to-end climate research tasks across diverse climate sub-fields.
Outcome:	The proposed framework outperforms state-of-the-art benchmarks in rigorousness and practicality.

CogBench: Benchmarking Cognitive Alignment of Large Language Models in Educational Question Answering (2026.findings-acl)

Copied to clipboard

Tong Lu, Zhichun Wang, Yuanhao Sun, Yaoyu Zhou, Mingrui Li, Yiming Guan, Zhiyong Bai

CARL: Constraint-Aware Reinforcement Learning for Planning with LLMs (2026.findings-acl)

Challenge:	Large language models (LLMs) possess strong capabilities in language understanding and generation, as well as remarkable problem-solving abilities.
Approach:	They propose a benchmark to assess the cognitive alignment capabilities of large language models in educational QA.
Outcome:	The proposed evaluation benchmark assesses the cognitive alignment capabilities of large language models in educational QA.

Copied to clipboard

Qiuyi Qi, Jinjian Zhang, Mutian Bao, Tian Liang, Guocong Li, Dongnan Liu, Wei Zhou, Jie Liu, Ming Kong, Linjian Mo, Feng Zhang, Qiang Zhu

ChildEval:WHEN LARGE LANGUAGE MODELS MEET CHILDREN’S PERSONALITIES (2026.findings-acl)

Challenge:	Existing approaches to constraint-aware planning fail to enhance the model’s intrinsic focus on constraints.
Approach:	They propose a constraint-aware reinforcement learning framework that encourages constraint focus and penalizes neglect of LLMs.
Outcome:	The proposed framework outperforms existing frameworks and state-of-the-art reasoning models in a number of real-world applications.

Copied to clipboard

Yanyan Luo, Xue Han, Chunxu Zhao, Ruiqiao Bai, Yaxing Zhang, Qian Hu, Lijun Mei, Junlan Feng

Belief in Authority: Impact of Authority in Multi-Agent Evaluation Framework (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have achieved remarkable success in effectively understanding and generating human language, leading to a revolutionary era in LLMs.
Approach:	They propose a benchmark to evaluate LLMs' ability to infer and follow child-centered preferences in long-context conversations.
Outcome:	The proposed benchmark spans five top-level and fourteen sub-level categories covering children’s daily lives and development.

Copied to clipboard

Junhyuk Choi, Jeongyoun Kwon, Heeju Kim, Haeun Cho, Hayeong Jung, Sehee Min, Bugeun Kim

Challenge:	Multi-agent systems utilizing large language models assign authoritative roles to improve performance, yet the impact of authority bias on agent interactions remains underexplored.
Approach:	They propose to classify authoritative roles into legitimate, referent, and expert types and analyze their influence across 12-turn conversations using French and Raven’s power-based theory.
Outcome:	The proposed model enables agents to perform better in multi-agent evaluations.

ToMELP: A Theory-of-Mind Benchmark for Route-Controlled Persuasion under the Elaboration Likelihood Model (2026.findings-acl)

Copied to clipboard

Ruirui Wang, Haoran Zhang, Tian Lan, Zehua Duo, Jiang Li, Guanglai Gao, Xiangdong Su

StateX: Enhancing RNN Recall via Post-training State Expansion (2026.findings-acl)

Challenge:	Theory of Mind (ToM) is widely regarded as central to effective persuasion, yet existing evaluations fail to capture the infer–apply loop that arises in real-world dialogue.
Approach:	They propose a benchmark that conditions on the audience persona p and the Elaboration Likelihood Model (ELM) route r within persuasive conversations.
Outcome:	The proposed model can model the interlocutor's mental states over multiple turns and adapt strategy and tone accordingly.

Copied to clipboard

Xingyu Shen, Yingfa Chen, Zhen Leng Thai, Xu Han, Zhiyuan Liu, Maosong Sun

TOXIFRENCH: Benchmarking and Enhancing Language Models via CoT Fine-Tuning for French Toxicity Detection (2026.findings-acl)

Challenge:	Existing studies show that RNNs with large recurrent states are expensive to train . however, the ability to recall contextual information from long contexts is underperforms them in certain aspects.
Approach:	They propose a framework that expands the states of pre-trained RNNs by scaling them up to 1.3B . they use a recurrent architecture that compresses contextual information into a fixedsize state .
Outcome:	Experiments on models with up to 1.3B parameters show that StateX expands state sizes without incurring high post-training costs or compromising other capabilities.

Copied to clipboard

Axel Delaval, Shujian Yang, Haicheng Wang, Han Qiu, Jialiang LU

Memory-Guided Hard Data Augmentation for Multimodal Named Entity Recognition (2026.findings-acl)

Challenge:	toxicity detection in French remains underdeveloped due to the lack of culturally relevant, human-annotated, large-scale datasets.
Approach:	They propose a method that generalizes French online comments using a semi-automated annotation pipeline that reduces manual labeling to only 10% through high-confidence LLM-based pre-annotation and human verification.
Outcome:	The proposed model outperforms GPT-4o and DeepSeek-R1 on the benchmark while maintaining cross-lingual capabilities.

Copied to clipboard

Xinyu Liu, Kai fu, Yinghan Shi, Quanyou Chu, Ming Du, Hongya Wang, Xiaojun Meng, Jiansheng Wei, Yanghua Xiao, Bo Xu

DUET: Joint Exploration of User–Item Profiles in Recommendation System (2026.findings-acl)

Challenge:	Existing methods for Named Entity Recognition (NER) ignore the internal state of the target model.
Approach:	They propose a framework to repair model-specific errors by using a model-based approach . they employ cross-validation to identify model- specific Hard Data and a memory tree to induce macro-level error patterns from micro-level failures.
Outcome:	The proposed framework yields significant performance gains on Twitter and other platforms.

Copied to clipboard

Yue Chen, Yifei Sun, Lu Wang, Fangkai Yang, Pu Zhao, Minjie Hong, Yifei Dong, Minghua He, Nan Hu, Jianjin Zhang, Zhiwei Dai, Yuefeng Zhan, Weihao Han, Hao Sun, Qingwei Lin, Weiwei Deng, Feng Sun, Qi Zhang, Saravan Rajmohan, Dongmei Zhang

Challenge:	Existing LLMs are opaque and difficult to interpret, resulting in limited interpretability.
Approach:	They propose an interaction-aware profile generator that jointly produces user and item profiles conditioned on both user history and item evidence.
Outcome:	The proposed model outperforms baselines on three real-world datasets.

ATAAT: Adaptive Threat-Aware Adversarial Tuning Framework against Backdoor Attacks on Vision-Language-Action Models (2026.findings-acl)

Copied to clipboard

Kewei Chen, Yayu Long, Shuai Li, Mingsheng Shang

CoTrust: Privacy-Preserving Collaboration Between Large and Small Language Models in Trusted Execution Environments (2026.findings-acl)

Challenge:	Existing backdoor models rely on visual inputs for instruction parsing, rendering the perception pathway a critical attack surface.
Approach:	They propose an Adaptive Threat-Aware Adversarial Tuning framework that detects and decouples the optimal gradient decoupling strategy based on the adversary's capabilities.
Outcome:	The proposed framework achieves a highly robust targeted attack success rate while maintaining extreme stealthiness with a 5% poisoning rate.

Copied to clipboard

Zhenya Ma, Tingyi Wang, Yongheng Deng, Ziqing Qiao, Yinggui Wang, Tao Wei, Lei Wang, Ju Ren

FAQ: Mitigating Quantization Error via Regenerating Calibration Data with Family-Aware Quantization (2026.findings-acl)

Challenge:	Large language models (LLMs) provide powerful text generation capabilities, but accessing sensitive user inputs raises privacy concerns.
Approach:	They propose a privacy-preserving collaborative inference framework that combines large language models with small language models inside TEE to preserve privacy.
Outcome:	Experiments show that CoTrust outperforms unconstrained LLMs on multiple question answering and summarization benchmarks while maintaining strong privacy protection.

Copied to clipboard

Haiyang Xiao, Weiqing Li, Jinyue Guo, Guochao Jiang, Guohua Liu, Yuewei Zhang

Challenge:	representativeness and universality of calibration data remain a bottleneck in quantization accuracy.
Approach:	They propose a framework that leverages prior knowledge from LLMs to generate calibration samples . their framework reduces accuracy loss by up to 28.5% compared to baseline .
Outcome:	Experiments show that family-aware quantization reduces accuracy loss by up to 28.5% compared to baseline data.

HACHIMI: Scalable and Controllable Student Persona Generation via Orchestrated Agents (2026.findings-acl)

Copied to clipboard

Yilin Jiang, Fei Tan, Xuanyu Yin, Leng Jing, Aimin Zhou

D-QRELO: Training- and Data-Free Delta Compression for Large Language Models via Quantization and Residual Low-Rank Approximation (2026.findings-acl)

Challenge:	ad-hoc prompting and hand-crafted profiles with limited control over educational theory and population distributions are often used for student personas.
Approach:	They propose a framework that generates theory-aligned, quota-controlled personas . they factorize each persona into a theory-anchored educational schema .
Outcome:	HACHIMI generates theory-aligned, quota-controlled personas for grades 1-12 . results show near-perfect schema validity, accurate quots, and substantial diversity .

Copied to clipboard

Junlin Li, Shuangyong Song, Guodong DU, Ngai Wong, Xuebo Liu, Yongxiang Li, Min Zhang, Jing Li, Xuelong Li

Beyond Static Toolsets: Self-Evolving LLM Tool Agents via Continual Documentation Adaptation (2026.findings-acl)

Challenge:	Existing methods for fine-tuned large language models fail on fine-scale datasets . large data scale amplifies delta parameter magnitude, singular values, and entropy, causing compression errors.
Approach:	They propose a training- and data-free delta compression method that captures dominant delta structure and compensates residual low-rank approximation to recover fine-grained details from smaller residual error.
Outcome:	The proposed method outperforms existing methods on large-scale datasets on dense and MoE architectures.

Copied to clipboard

Bin Wu, Edgar Meij, Emine Yilmaz

Challenge:	Existing methods for evaluating tool usage assume static toolsets with fixed APIs and documentation.
Approach:	They propose a continual documentation adaptation framework that allows LLM agents to self-evolve by updating tool documentation.
Outcome:	The proposed framework improves performance on three evolution patterns on dynamic extensions of StableToolBench and RestBench.

ZoFia: Zero-Shot Fake News Detection with Entity-Guided Retrieval and Multi-LLM Interaction (2026.findings-acl)

Copied to clipboard

Lvhua Wu, Xuefeng Jiang, Sheng Sun, Yan Lei, Tian Wen, Yuwei Wang, Min Liu

SenseJudge: Human-Centric Preference-Driven Judgment Framework (2026.findings-acl)

Challenge:	Large language models (LLMs) are limited by knowledge cutoff and can generate factual hallucinations when handling time-sensitive news.
Approach:	They propose a two-stage zero-shot fake news detection framework that uses a hierarchical salience and saliency-calibrated minimum margin of relevance algorithm to extract core entities accurately.
Outcome:	The proposed framework outperforms existing zero-shot baselines and even most few-shot methods on two public datasets.

Copied to clipboard

Rui Li, Junfeng Liu, Xiangwen Kong, Zhifang Sui

Challenge:	Existing judgment approaches rely on trained judgers using fixed preference data . existing judgment approaches neglect diverse user preferences and struggle to adapt to real-world human-AI dialogue scenarios.
Approach:	They propose a customizable judgment framework driven by human preferences and a diverse instruction following benchmark derived from real-world multi-turn interactions.
Outcome:	The proposed framework surpasses other judgment methods and models in two tasks, and achieves model ranking that aligns with real human sense.

Reliability-Aware Adaptive Self-Consistency for Efficient Sampling in LLM Reasoning (2026.findings-acl)

Copied to clipboard

Junseok Kim, Nakyeong Yang, Kyungmin Min, Kyomin Jung

CLaS-Bench: A Cross-Lingual Alignment and Steering Benchmark (2026.findings-acl)

Challenge:	Self-consistency improves reasoning reliability but incurs substantial inference cost . Adaptive self-consistent methods rely on count-based stopping rules that treat all responses equally .
Approach:	They propose a method that reframs adaptive sampling from response counting to evidence sufficiency by leveraging response-level confidence.
Outcome:	The proposed method reduces inference cost by up to 70% while preserving accuracy on GSM8K.

Copied to clipboard

Daniil Gurgurov, Yusser Al Ghussin, Tanja Baeumel, Cheng-Ting Chou, Patrick Schramowski, Marius Mosbach, Josef Van Genabith, Simon Ostermann

From Prediction to Intervention: Personalized Meal-Level Glucose Regulation via an LLM Agent (2026.findings-acl)

Challenge:	Understanding and controlling behavior of large language models (LLMs) is an important topic in multilingual NLP.
Approach:	They propose a lightweight parallel-question benchmark for evaluating language-forcing behavior in large language models across 32 languages.
Outcome:	The proposed benchmark measures language steering in 32 languages across 32 languages.

Copied to clipboard

Mingyu Huang, Weiqing Min, Ying Jin, Yilin Wang, Shuqiang Jiang

Challenge:	Existing approaches to individualized glucose regulation are generic and do not account for individual-specific glucose dynamics.
Approach:	They propose a physio-feedback agentic loop that integrates individualized absorption modeling with dietary intervention to regulate glucose response.
Outcome:	The proposed system improves prediction accuracy and reduces glucose excursions.

PsyPath: Psychologically-guided Self-Exploration for Personality Detection (2026.findings-acl)

Copied to clipboard

Zheng Li, Hongxin Ding, Chenyu Zhang, Weimin Xiong, Dawei Zhu, Sujian Li

CCTVBench: Contrastive Consistency Traffic VideoQA Benchmark for Multimodal LLMs (2026.findings-acl)

Challenge:	Personality detection aims to label traits via identifying linguistic cues from written text.
Approach:	They propose a framework that allows large language models to generate and answer psychologically meaningful questions and a hybrid scoring mechanism to evaluate the generated nodes in the reasoning paths.
Outcome:	The proposed framework outperforms baselines on two benchmark datasets and significantly improves performance and interpretability in downstream tasks.

Copied to clipboard

Xingcheng Zhou, Hao Guo, Rui Song, Walter Zimmer, Mingyu Liu, André Schamschurko, Hu Cao, Alois Knoll

PROGRAM: Programmatic Retrieval Optimization with Generative Reasoning and Augmented Multi-queries (2026.findings-acl)

Challenge:	Existing Vision-language models are prone to hallucinating nonexistent entities or events and missing subtle but critical visual cues.
Approach:	They propose a Traffic VideoQA Benchmark that enforces a single structured decision pattern over each video question quadruple and provides actionable diagnostics that decompose failures into positive omission, positive swap, negative hallucination, mutual-exclusivity violation.
Outcome:	The proposed model detects true hazards when an accident occurs, and rejects plausible-but-false hypotheses under near-identical counterfactual scenes.

Copied to clipboard

Gun Il Kim, Jungkyu Shin, Jong Wook Kim, Beakcheol Jang

TiMem: Temporal-Hierarchical Memory Consolidation for Long-Horizon Conversational Agents (2026.findings-acl)

Challenge:	Current retrieval-augmented generation methods struggle with complex multi-hop reasoning, relying on unstructured semantic matching that lacks the logical structure needed to systematically guide retrieval.
Approach:	They propose a framework that elevates retrieval to structured, program-guided reasoning by combining three stages of program-type selection and evidence accumulation.
Outcome:	Evaluated on five benchmarks including HotPotQA, 2WikiMultihopQA, ARC-Challenge, MMLU-Pro, and MedQA with various LLMs, PROGRAM achieves state-of-the-art performance with up to 24% relative improvement on HotPtQA and 13.2% on MedQA over strong baselines including FLARE, ProbTree and Self-RAG.

Copied to clipboard

Kai Li, Xuanqing Yu, Ziyi Ni, Yi Zeng, Yao Xu, Zheqing Zhang, Xin Li, Jitao Sang, Xiaogang Duan, Xuelei Wang, Chengbao Liu, Jie Tan

GraphDx: A Cost-Aware Knowledge-Enhanced Multi-Agent Framework for Sequential Diagnosis (2026.findings-acl)

Challenge:	Existing memory frameworks provide limited support for temporally structured information across hierarchical levels, leading to fragmented memories and unstable long-horizon personalization.
Approach:	They propose a temporal–hierarchical memory framework that organizes conversations through a Temporal Memory Tree.
Outcome:	The proposed framework outperforms baselines while reducing the recalled memory length by 52.20%.

Copied to clipboard

Shaoting Tan, Ning Liu, Yuntao Du, Shuyue Wei, Wu Shuai, Qian Li, Yanyu Xu, Wei Zhang, Lizhen Cui, Haitao Yuan

ProUIE: A Macro-to-Micro Progressive Learning Method for LLM-based Universal Information Extraction (2026.findings-acl)

Challenge:	Existing Large Language Models struggle to reason systematically under cost constraints . Existing approaches lack the knowledge-reasoning capability to reason under cost .
Approach:	They propose a knowledge-enhanced framework that leverages large language models to construct MDKGs . they propose three collaborative agents that handle language understanding and generation .
Outcome:	GraphDx improves diagnostic success rates from 50–68% to 79–93% while reducing test costs by 20–54%.

Copied to clipboard

Wenda Liu, Song Zhigang, Shuai Nie, Guangyao Liu, Lisung Chen, Binyu Yang, Yaran Chen, Peng Zhou, Hongzhen Wang, Yuchen Liu, Wenyue Hu, Jiaming Xu, Runyu Shi, Ying Huang

MAVIS: Multi-Agent Video Retrieval via Structured Video Understanding (2026.findings-acl)

Challenge:	ProUIE improves universal information extraction (UIE) without external information . many LLM-based methods rely on extra schema cues, external resources or complex alignment and verification pipelines .
Approach:	They propose a Macro-to-Micro progressive learning approach that improves UIE without external information.
Outcome:	ProUIE outperforms instruction-tuned baselines on average for NER and RE while using a smaller backbone.

Copied to clipboard

Jie Zhang, Qilang Ye, Hao Zhou, Haochen Liang, Fei Luo

ThinkBrake: Efficient Reasoning via Log-Probability Margin Guided Decoding (2026.findings-acl)

Challenge:	Existing methods for video retrieval rely on embedding-based full-corpus scanning, but there is a bottleneck in semantic asymmetry and computational redundancy.
Approach:	They propose a multi-agent framework that rethinks retrieval as cooperative reasoning . they parse raw videos into a structured semantic library, enabling explicit attribute-level indexing .
Outcome:	The proposed framework bridges the granularity mismatch gap by parsing raw videos into a structured semantic library . it employs a Logic-aware Debate mechanism with a strict veto protocol . the proposed framework achieves competitive performance without task-specific fine-tuning .

Copied to clipboard

Sangjun Song, Minjae Oh, Seungkyu Lee, Sungmin Jo, Yohan Jo

MSIA: Adaptive Medical Multimodal Multi-turn Semantic Jailbreak (2026.findings-acl)

Challenge:	Recent advances in Large Reasoning Models (LRMs) have demonstrated remarkable capabilities across various tasks.
Approach:	They propose a system that stops reasoning when the margin between continuation token and lt;/think gt; narrows.
Outcome:	The proposed model reduces thinking token usage by 30% and improves accuracy by 8% while reducing thinking tokens by 72%.

Copied to clipboard

Zhiheng Han, Yao Zhang, Jun Wang, Zhenglu Yang

Why These Documents? Explainable Generative Retrieval with Hierarchical Category Paths (2026.findings-acl)

Challenge:	Medical large language models exhibit high domain specificity and condensed semantics, making them vulnerable to diagnostic errors in real-world clinical settings.
Approach:	They propose a framework for modeling and inducing multi-turn medical semantic jailbreaks in clinical dialogues.
Outcome:	Experiments on chest X-ray-based multimodal medical dialogues show that MSIA outperforms existing jailbreak methods with an average success rate of 76.67%.

Copied to clipboard

Sangam Lee, Ryang Heo, SeongKu Kang, Susik Yoon, Jinyoung Yeo, Dongha Lee

ETHICA-MT: Introducing a Framework and Dataset for Studying Ethical Orientations in LLM-based Machine Translation (2026.findings-acl)

Challenge:	Generative retrieval directly decodes a document identifier, making it impossible to provide explanations for its retrieval decision.
Approach:	They propose a hierarchical category path-Enhanced Generative Retrieval that generates category paths step-by-step and decodes docid.
Outcome:	The proposed method provides explanations for retrieval decision by generating hierarchical category paths step-by-step and decoding docid.

Copied to clipboard

Omri Asscher, Arif Ahmad, Ananya Agrawal, Monojit Choudhury

ArgBench: Benchmarking LLMs on Computational Argumentation Tasks (2026.findings-acl)

Challenge:	Existing models for translation have not been systematically examined for their default ethical tendencies or their ability to employ and prioritize specified ethical approaches in conflicted translation situations.
Approach:	They propose a framework for examining ethical reasoning and implementation in large language models (LLMs) that systematically examines default ethical tendencies and their ability to employ and prioritize specified ethical approaches in conflicted translation situations.
Outcome:	The proposed framework examines the ethical reasoning and implementation of large language models in translation tasks.

Copied to clipboard

Yamen Ajjour, Carlotta Quensel, Nedim Lipka, Henning Wachsmuth

Generics are not quantificational: A new path from language models to semantic theory (2026.findings-acl)

Challenge:	Argumentation skills are an essential toolkit for large language models (LLMs).
Approach:	They propose a benchmark to evaluate the generalizability of five LLM families across 46 computational argumentation tasks.
Outcome:	The proposed benchmark evaluates the generalizability of five LLM families across 46 computational argumentation tasks covering mining arguments, assessing perspectives, evaluating argument quality, reasoning about arguments, and generating arguments.

Copied to clipboard

Gustavo Cilleruelo Calderón, Mahrad Almotahari, Emily Allaway, Barry Haddow, Alexandra Birch

Beyond Neural Incompatibility: Cross-Scale Knowledge Transfer in Large Language Models through Latent Semantic Alignment (2026.findings-acl)

Challenge:	Generic sentences express generalizations that tolerate exceptions without explicitly communicating information about quantities.
Approach:	They compare generics and quantificational sentences to find out what quantifiers are . they argue that generics are not quantificationals, contrary to dominant views .
Outcome:	The proposed model recovers many semantic facts about quantifiers and their "quantificational counterparts".

Copied to clipboard

Jian Gu, Aldeida Aleti, Chunyang Chen, Hongyu Zhang

CURaTE: Continual Unlearning in Real Time with Ensured Preservation of LLM Knowledge (2026.findings-acl)

Challenge:	Existing methods that reuse layer parameters are limited by incompatibility . a central challenge is to make cross-scale knowledge transfer effective and efficient .
Approach:	They propose a method that uses latent semantic alignment to facilitate cross-scale knowledge transfer . they use activations to pair target and source layers in latent space to achieve alignment .
Outcome:	The proposed method is effective when source and target models differ in architecture and parameterization.

Copied to clipboard

Seyun Bae, Seokhan Lee, Eunho Yang

Decoupling the Effect of Chain-of-Thought Reasoning: A Human Label Variation Perspective (2026.findings-acl)

Challenge:	Existing methods for unlearning specific pieces of knowledge are ineffective due to the inability to filter out in advance all potentially problematic data.
Approach:	They propose a method for unlearning specific pieces of knowledge after training . they use a sentence embedding model to create sharp decision boundaries .
Outcome:	The proposed method achieves more effective forgetting than existing methods and maintains near perfect knowledge preservation over any number of updates.

Copied to clipboard

Beiduo Chen, Tiancheng Hu, Caiqi Zhang, Robert Litschko, Anna Korhonen, Barbara Plank

CTRL: Control-Based Time Series Forecasting with LLM-Guided Residual Learning (2026.findings-acl)

Challenge:	Reasoning-tuned large language models (LLMs) with long Chain-of-Thought excel at single-answer tasks, yet their ability to model Human Label Variation remains underexplored.
Approach:	They conduct systematic disentanglement experiments to isolate the effect of reasoning text from intrinsic model priors on distribution-based tasks.
Outcome:	The proposed model improves distributional alignment, but distributional ranking is governed by model priors.

Copied to clipboard

Minkyoung Kim, Daeun Ji, Yohan Lee, Beomsoo Kim, Beakcheol Jang

INDOTABVQA: A Benchmark for Cross-Lingual Table Understanding in Bahasa Indonesia Documents (2026.findings-acl)

Challenge:	Existing time series forecasting approaches reduce them to numerical predictors that bypass their strengths or allow direct forecast generation that destabilizes predictions in non-stationary settings.
Approach:	They propose a framework that decouples semantic reasoning from quantitative prediction.
Outcome:	The proposed framework decouples semantic reasoning from quantitative prediction.

Copied to clipboard

Somraj Gautam, Anathapindika Dravichi, Gaurav Harit

Cross-Cultural Transfer of Emoji Semantics and Sentiment in Financial Social Media (2026.findings-acl)

Challenge:	INDOTABVQA provides a benchmark for evaluating cross-lingual Table Visual Question Answering (VQA) on real-world document images in Bahasa Indonesia.
Approach:	They propose a benchmark for evaluating cross-lingual Table Visual Question Answering on real-world document images in Bahasa Indonesia.
Outcome:	The proposed model improves on a 3B model and a LoRA- finetuned 7B model on Bahasa Indonesian document images by 11.6% and 17.8% respectively.

Copied to clipboard

Ahmed Mahrous, Roberto Di Pietro

EXPO-SQL: Execution-based Clause-level Policy Optimization for Text-to-SQL (2026.findings-acl)

Challenge:	Emojis are widely used in online financial communication, but it is unclear whether they provide transferable sentiment signals across languages, platforms, and asset communities.
Approach:	They examine whether emoji usage, semantics, and sentiment polarity remain stable across financial communities and asset communities.
Outcome:	The results show that emojis provide compact, language-independent sentiment cues that improve model generalization across markets and platforms.

Copied to clipboard

Jaehoon Lee, CheolWon Na, Suyoung Bae, Jin-Seop Lee, Jihyung Lee, YunSeok Choi, Jee-Hyong Lee

Synapse: Empowering LLM Agents with Episodic-Semantic Memory via Spreading Activation (2026.findings-acl)

Challenge:	Existing RL methods assign query-level rewards to all clauses, treating correct and incorrect clauses equally.
Approach:	They propose a method which provides fine-grained supervision through clause-level rewards.
Outcome:	Experiments on widely-used Text-to-SQL benchmarks show that EXPO-SqL outperforms existing methods by fine-grained clause-level learning.

Copied to clipboard

Hanqi Jiang, Junhao Chen, Yi Pan, Ling Chen, Weihang You, Yifan Zhou, Ruidong Zhang, Yohannes Abate, Tianming Liu

Challenge:	Large Language Models excel at generalized reasoning, but lack the ability to accumulate experiences and maintain narrative coherence over long horizons.
Approach:	They propose a unified memory architecture that transcends static vector similarity.
Outcome:	The proposed model outperforms state-of-the-art methods in temporal and multihop reasoning tasks.

Copied to clipboard

Allen Schmaltz

SDAR: A Synergistic Diffusion-AutoRegression Paradigm for Scalable Sequence Generation (2026.findings-acl)

Challenge:	Existing methods for interpreting neural network-based language models (LMs) are limited to approximately conditional quantities.
Approach:	They introduce a similarity-distance-magnitude activation function and an SDM estimator to control class- and prediction-conditional accuracy among selective classifications.
Outcome:	The proposed estimator is more robust to covariate shifts and out-of-distribution inputs while remaining informative over in-difference data.

Copied to clipboard

Shuang Cheng, Yihan Bian, Dawei Liu, Yuhua Jiang, Yihao Liu, Linfeng Zhang, Qian Yao, Zhongbo Tian, Wenhai Wang, Qipeng Guo, Kai Chen, Biqing Qi, Bowen Zhou

Comparing Human and Large Language Model Interpretation of Implicit Information (2026.findings-acl)

Challenge:	Autoregressive (AR) language models are a dominant paradigm in the field of parallelism and non-causal modeling.
Approach:	They propose a blockwise discrete diffusion model that preserves AR-compatible serving while enabling parallel intra-block generation.
Outcome:	The proposed model achieves theoretical speedups over 5 and wall-clock speedup of 2.3 on H200 GPUs in latency-critical regimes.

Copied to clipboard

Antonio De Santis, Tommaso Bonetti, Andrea Tocchetti, Marco Brambilla

Simplify-Pro: A Two-level and Progressive LLM-based Framework for Auto Long Text Simplification (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are a popular approach for generating text indistinguishable from human-generated language.
Approach:	They propose an LLM-based pipeline that builds a structured knowledge graph from a context sentence by extracting relational triplets, validating implicit inferences, and analyzing temporal relations.
Outcome:	The proposed pipeline builds a structured knowledge graph from a context sentence by extracting relational triplets, validating implicit inferences, and analyzing temporal relations.

Copied to clipboard

Peng Zhou, Guangxin Li, Xiaoying Huang, Yiming Tang

Pattern Enhanced Multi-Turn Jailbreaking: Exploiting Structural Vulnerabilities in Large Language Models (2026.findings-acl)

Challenge:	Existing studies have focused on lexical- and sentence-level simplification, leaving long text simplification comparatively unexplored .
Approach:	They propose a two-level and progressive LLM-based framework that establishes an effective paradigm for automatic long text simplification under diverse test scenarios.
Outcome:	The proposed framework outperforms advanced and proprietary LLMs in in-domain and out-of-domain simplification tasks and matches or outperformed existing LLM frameworks.

Copied to clipboard

Ragib Amin Nihal, Rui Wen, Kazuhiro Nakadai, Jun Sakuma

Data-Efficient Adaptation to Contextual Shifts in LLM-based Conversational Recommendation (2026.findings-acl)

Challenge:	Existing multi-turn methods for large language models exploit conversational context to bypass safety constraints gradually.
Approach:	They propose a framework of five conversation patterns to construct multi-turn jailbreaks through natural dialogue.
Outcome:	The proposed framework exploits conversational contexts to construct multi-turn jailbreaks . it reveals that models exhibit distinct weakness profiles and model families share similar failure modes .

Copied to clipboard

Hyeongjun Yang, Donghyun Kim, Seokju Hwang, Midan Shim, KyuHwan Yeom, KaeHyun Um, Kyong-Ho Lee

PsychEval: A Multi-Session and Multi-Therapy Benchmark for High-Realism AI Psychological Counselor (2026.findings-acl)

Challenge:	Existing data selection methods struggle to distinguish learnable samples under contextual shifts.
Approach:	They propose a framework agnostic to underlying large language model-based conversational recommender systems (CRSs) that captures user preferences through free-form conversations and generates contextually relevant recommendations.
Outcome:	The proposed framework outperforms baselines on three CRS benchmarks with real-world temporal splits.

Copied to clipboard

Qianjun Pan, Junyi Wang, Jie Zhou, Yutao Yang, Junsong Li, Kaiyin Xu, Yougen Zhou, Yihan Li, JingYuan Zhao, Qin Chen, Ningning Zhou, Kai Chen, Liang He

Challenge:	Existing models focus on a single therapy, but complex cases require flexible strategies among various therapies.
Approach:	They propose a multi-session, multi-therapy, and highly realistic benchmark . it is designed to address three key challenges: 1) can we train a highly realistic AI counselor? 2) How to systematically evaluate an AI counselor?"
Outcome:	The proposed benchmark is annotated with extensive professional skills and includes over 677 meta-skills and 4577 atomic skills.

GrandGuard: Taxonomy, Benchmark, and Safeguards for Elderly-Chatbot Interaction Safety (2026.findings-acl)

Copied to clipboard

Changxuan Fan, Xi Yang, Yueyuan Zheng, Bin Zhou, Yuanping Wang, Wenbin Hu, Huihao Jing, Ki Sen Hung, Dazhao Du, Haoran Li, Janet Hui-wen Hsiao, Yangqiu Song

SciExplore: Evaluating Autonomous Agents from Scientific Navigation to Information Integration (2026.findings-acl)

Challenge:	a survey of older adults shows that many LLMs mishandle elderly-specific contextual risks.
Approach:	They propose a framework to assess elderly-specific contextual risks in LLM interactions . they use a taxonomy to identify 50 fine-grained risk types across mental well-being, financial, medical, toxicity, and privacy domains .
Outcome:	a new framework assesses elderly-specific contextual risks in LLM interactions . it achieves 96.2% and 90.9% unsafe-prompt detection accuracy, respectively .

Copied to clipboard

Yinhao Tang, Youqing Fang, Yanan Sun, Wenran Liu, Weiming Zhang, Bin Liu, Kuikun Liu, Wenwei Zhang, Kai Chen

When Models Outthink Their Safety: Unveiling and Mitigating Self-Jailbreak in Large Reasoning Models (2026.findings-acl)

Challenge:	Existing benchmarks emphasize general-domain retrieval or static scientific question answering . SciExplore focuses on scientific database navigation, ambiguous literature retrieval, missing reference completion, and cross-source structured knowledge synthesis tasks.
Approach:	They propose a benchmark to evaluate scientific information-seeking and reasoning capabilities of LLMs and agents.
Outcome:	The new benchmark assesses the capabilities of state-of-the-art LLMs and agents in scientific research workflows.

Copied to clipboard

Yingzhi Mao, Chunkang Zhang, Junxiang Wang, Xinyan Guan, Boxi Cao, Yaojie Lu, Hongyu Lin, Xianpei Han, Le Sun

NASH: Numerically Aware Scoring Heuristic for Robust Semantic Similarity (2026.findings-acl)

Challenge:	Existing methods often apply coarse-grained constraints over entire reasoning trajectories . Existing approaches often apply unsafe constraints, causing unsafe outputs .
Approach:	They propose a trajectory-level training framework that mitigates Self-Jailbreak . they propose 'chain-of-guardrail' to mitigate self-jailbreak by targeting step-level interventions .
Outcome:	The proposed framework mitigates Self-Jailbreak by targeting step-level interventions while maintaining reasoning ability.

Copied to clipboard

Yu-Shiang Huang, Yun-Yu Lee, Tzu-Hsin Chou, Che Lin, Chuan-Ju Wang

Beyond Query Bias: Candidate-Aware Iterative Refinement for Zero-Shot Composed Image Retrieval (2026.findings-acl)

Challenge:	Numerical precision is critical in financial NLP, yet embedding-based semantic similarity metrics exhibit numerical blindness.
Approach:	They propose a model-agnostic metric that decouples numerical verification from textual semantic evaluation.
Outcome:	The proposed metric improves numerical sensitivity while maintaining general semantic performance.

Copied to clipboard

Nan Sun, Jing Tang, Lei Sun, Rui Chen, Yuxing Lu, Xiangxiang Chu, Hefei Ling, Yujun Cai

Large Reasoning Models Are (Not Yet) Multilingual Latent Reasoners (2026.findings-acl)

Challenge:	Existing methods to retrieve target images suffer from inherent cognitive bias due to unknown candidate distribution.
Approach:	They propose a training-free framework that reframes ZS-CIR as a self-correcting process . they propose to use retrieved results as feedback to perceive the candidate distribution .
Outcome:	Experiments on public benchmarks show that CoRR outperforms other SOTA methods.

Copied to clipboard

Yihong Liu, Raoyuan Zhao, Hinrich Schuetze, Michael A. Hedderich

Towards Scalable Lightweight GUI Agents via Multi-role Orchestration (2026.findings-acl)

Challenge:	Recent work shows that large reasoning models arrive at the correct answer before completing textual reasoning steps, indicating the presence of latent reasoning.
Approach:	They conduct a systematic investigation of multilingual latent reasoning in large reasoning models across 11 languages.
Outcome:	The proposed model arrive at the correct answer before completing the reasoning steps, indicating the presence of latent reasoning.

Copied to clipboard

Ziwei Wang, Junjie Zheng, Leyang Yang, Sheng Zhou, Xiaoxuan Tang, Fang Zhouhua, Zhiwei Liu, Dajun Chen, Yong Li, Jiajun Bu

Challenge:	Advanced GUI agents suffer from prohibitive deployment costs on resource-constrained devices.
Approach:	They propose a lightweight GUI agent with GUI-specific knowledge and task scalability . LAMO-3B supports monolithic execution and MAS-style orchestration .
Outcome:	The proposed GUI agent LAMO-3B supports monolithic execution and MAS-style orchestration.

Memory Matters More: Event-Centric Memory as a Logic Map for Agent Searching and Reasoning (2026.findings-acl)

Copied to clipboard

Yuyang Hu, Jiongnan Liu, Jiejun Tan, Yutao Zhu, Zhicheng Dou

Format-Adapter: Improving Reasoning Capability of LLMs by Adapting Suitable Format (2026.findings-acl)

Challenge:	Existing methods for storing and retrieving memory are limited by shallow semantic retrieval.
Approach:	They propose a memory mechanism that organizes and retrieves past experiences to support decision-making.
Outcome:	Experiments on LoCoMo and NarrativeQA show that CompassMem improves retrieval and reasoning performance across multiple backbone models.

Copied to clipboard

Dingzirui Wang, Xuanliang Zhang, Rongyu Cao, Longxu Dou, Xianzhen Luo, Yingwei MA, Qingfu Zhu, Binhua Li, Fei Huang, Yongbin Li

Behavior-Aware Item Modeling via Dynamic Procedural Solution Representations for Knowledge Tracing (2026.findings-acl)

Challenge:	Prior work showed that multiple reasoning formats outperform a single format when generating multiple answers.
Approach:	They propose a method to measure reasoning error when generating multiple answers . they propose 'formatadapter' which generates and selects suitable reasoning formats .
Outcome:	The proposed method achieves a 4.3% performance improvement over previous works on math and commonsense reasoning tasks.

Copied to clipboard

Jun Seo, Sangwon Ryu, Heejin Do, Hyounghun Kim, Gary Lee

Tracing the Light of Thought: A Probabilistic Self- and Cross-Consistency Verification Mechanism Improving Mathematical Reasoning in LLMs (2026.findings-acl)

Challenge:	Knowledge Tracing (KT) aims to predict learners’ future performance from past interactions, but they overlook the procedural dynamics of problem solving.
Approach:	They propose a framework that enriches item representations by integrating dynamic procedural solution information.
Outcome:	Experiments on XES3G5M and NIPS34 show that BAIM outperforms strong pretraining-based baselines, achieving particularly large gains under repeated learner interactions.

Copied to clipboard

Xiaoyang Liu, Dawei Wang, Tian Li, Huizhi Liang, Gary Ushaw, Richard Davison

E2E-GMNER: End-to-End Generative Grounded Multimodal Named Entity Recognition (2026.findings-acl)

Challenge:	Existing methods for evaluating reasoning paths are not efficient, but they are prone to errors.
Approach:	They propose a probabilistic self- and cross-consistency framework for mathematical reasoning that employs an accept-reject mechanism to encourage high-quality reasoning paths.
Outcome:	The proposed framework improves on 9 LLMs across 4 challenging benchmarks.

Copied to clipboard

Meng Zhang, Jinzhong Ning, Xiaolong Wu, Hongfei Lin, Yijia Zhang

Think Less, Know More: State-Aware Reasoning Compression with Knowledge Guidance for Efficient Reasoning (2026.findings-acl)

Challenge:	Existing approaches decouple textual entity recognition and visual grounding, leading to error accumulation and suboptimal joint optimization.
Approach:	They propose a fully end-to-end generative framework that unifies recognition, semantic typing, visual grounding and implicit knowledge reasoning within a single multimodal large language model.
Outcome:	The proposed framework achieves highly competitive performance compared with state-of-the-art methods.

Copied to clipboard

Yi Sui, Chaozhuo Li, Dawei Song

Muse: Towards Reproducible Long-Form Song Generation with Fine-Grained Style Control (2026.findings-acl)

Challenge:	Existing CoT compression methods struggle to balance accuracy and efficiency . long CoT reasoning also introduces an overthinking phenomenon, authors say .
Approach:	They propose a framework that performs step-wise CoT compression by modeling stage-specific redundancy sources and integrating with a retrieval-augmented guidance.
Outcome:	The proposed framework reduces average response length by 59.9% while improving accuracy by 4.8 points over existing methods.

Copied to clipboard

Changhao Jiang, Jiahao Chen, Zhenghao Xiang, Zhixiong Yang, Hanchen Wang, Jiabao Zhuang, Xinmeng Che, Jiajun Sun, Hui Li, Yifei Cao, Shihan Dou, Ming Zhang, Junjie Ye, Tao Ji, Tao Gui, Qi Zhang, Xuanjing Huang

Distilling LLM Reasoning into Dense Encoders: Bridging the Accuracy-Efficiency Gap in Recommendation (2026.findings-acl)

Challenge:	Recent commercial systems such as Suno demonstrate strong capabilities in long-form song generation, but academic research remains non-reproducible due to the lack of publicly available training data.
Approach:	They propose a system for long-form song generation with fine-grained style conditioning that includes a licensed synthetic dataset and a song generation model, Muse.
Outcome:	The proposed system achieves competitive performance on phoneme error rate, text–music style similarity, and audio aesthetic quality while enabling controllable segment-level generation across different musical structures.

Copied to clipboard

Donghee Han, Daeyoung Roh, A Young Kim, Hwanjun Song, Mun Yong Yi

Decoding the Multimodal Mind: Generalizable Brain-to-Text Translation via Multimodal Alignment and Adaptive Routing (2026.findings-acl)

Challenge:	Existing distillation approaches target Small Language Models (SLMs) or Conventional Recommendation Models, but face a critical trade-off between computational cost and semantic reasoning capacity.
Approach:	They propose a framework that establishes a text encoder as the optimal student architecture for scalable recommendation.
Outcome:	Experiments on four datasets show that the proposed framework outperforms state-of-the-art models and achieves significantly reduced latency.

Copied to clipboard

Chunyu Ye, Yunhao Zhang, Jingyuan Sun, Chong Li, Yang Zhao, Shaonan Wang

Which bird does not have wings: Negative-constrained KGQA with Schema-guided Semantic Matching and Self-directed Refinement (2026.findings-acl)

Challenge:	Current approaches to decoding language from the human brain rely on unimodal representations, neglecting the brain’s inherently multimodal processing.
Approach:	They propose a framework that leverages Multimodal Large Language Models to align brain signals with a shared semantic space encompassing text, images, and audio.
Outcome:	The proposed framework achieves an 8.48% improvement on the most commonly used benchmark on fMRI datasets with textual, visual, and auditory stimuli.

Copied to clipboard

Midan Shim, Seokju Hwang, KaeHyun Um, Kyong-Ho Lee

A Self-Evolving LLM Agent Framework for Role-Based Norm Compliance in Healthcare (2026.findings-acl)

Challenge:	Existing KGQA benchmarks and methods are biased toward positive and calculation constraints. Negative constraints are neglected, although they frequently appear in real-world questions.
Approach:	They propose a task where each question contains at least one negative constraint and a corresponding dataset, NestKGQA.
Outcome:	The proposed framework outperforms baselines on both KGQA and NEST-KGQA benchmarks under few-shot settings.

Copied to clipboard

Haijie Ruan, Xiaowu Jiang, Zhanpeng LI, Wei Jia, Xuanwu Xu, Xiao-Fen Shan, Shujie Chen, Xindong Ye

M3TQA: Massively Multilingual Multitask Table Question Answering (2026.findings-acl)

Challenge:	Existing systems treat roles as static prompts and rely on one-shot safety filters . a self-evolving LLM agent is proposed that learns from role-based social experience .
Approach:	They propose a self-evolving LLM agent that learns from role-based social experience and explicitly models communicator-level individual traits informed by prior communication questionnaires and clinical literature.
Outcome:	The proposed agent learns from role-based social experience and models communicator-level individual traits informed by prior communication questionnaires and clinical literature.

Copied to clipboard

Daixin Shu, Jian Yang, Zhenhe Wu, Xianjie Wu, Xianfu Cheng, Guan Xiangyuan, Yanghai Wang, Pengfei Wu, Tingyang Yang, Hualei Zhu, Wei Zhang, Ge Zhang, Jiaheng Liu, Zhoujun Li

ShredBench: Evaluating the Semantic Reasoning Capabilities of Multimodal LLMs in Document Reconstruction (2026.findings-acl)

Challenge:	Existing multilingual table benchmarks suffer from geolinguistic imbalance - overrepresenting certain languages and lacking sufficient scale for rigorous cross-lingual analysis.
Approach:	They propose a framework for massively multilingual table question answering that includes tables expanded to 97 languages from Chinese and English sources.
Outcome:	Experiments on state-of-the-art LLMs show that synthetically generated training data significantly boosts performance, especially for low-resource languages.

Copied to clipboard

Zichun Guo, Yuling Shi, Wenhao Zeng, Chao Hu, Haotian Lin, Terry Yue Zhuo, Jiawei Chen, Xiaodong Gu, Wenping Ma

TELL-TALE: Task Efficient LLMs with Task Aware Layer Elimination (2026.findings-acl)

Challenge:	Empirical evaluations on state-of-the-art MLLMs reveal a significant performance gap . ML models lack the fine-grained cross-modal reasoning required to bridge visual discontinuities.
Approach:	They propose a benchmark that renders fragmented documents directly from Markdown to facilitate evaluation of VRDU tasks.
Outcome:	The proposed benchmark renders fragmented documents directly from Markdown.

Copied to clipboard

Omar Naim, Krish Sharma, Niyar R Barman, Nicholas Asher

Reason-KE++: Aligning the Process, Not Just the Outcome, for Faithful LLM Knowledge Editing (2026.findings-acl)

Challenge:	Large Language Models typically come with a fixed architecture, but not all layers contribute equally to every downstream task.
Approach:	They propose an inference-time method that selectively removes irrelevant or detrimental layers . the method is hardware-agnostic, requires no retraining, and operates entirely at inference time .
Outcome:	The proposed method matches or surpasses baseline performance while reducing computational costs.

Copied to clipboard

Yuchen Wu, Liang Ding, Li Shen, Dacheng Tao

RACC: Regret-Aware Confidence Calibration for Consistent Masked Discrete Diffusion Decoding (2026.findings-acl)

Challenge:	Current methods for modifying parameters to integrate new knowledge are not accurate enough.
Approach:	They propose an SFT+RL framework that instills process-level faithfulness by a stage-aware Reward mechanism and a Stage-assisted Reward Mechanism.
Outcome:	The proposed framework instills process-level faithfulness while boosting final accuracy.

Copied to clipboard

Qinglin Zeng, Jusheng Zhang, Jing Yang, Ningyuan Liu, Keze Wang

The Impact of Off-Policy Training Data on Probe Generalisation (2026.findings-acl)

Challenge:	Masked Discrete Diffusion Models (MDMs) enable parallel generation via iterative refinement, but their current decoding paradigms are static and myopic.
Approach:	They propose a Regret-Aware Confidence Calibration framework that aligns decoding decisions with the model’s latent self-correction capabilities.
Outcome:	The proposed framework aligns decoding decisions with model’s latent self-correction capabilities.

Copied to clipboard

Adrians Skapars, Nathalie Maria Kirch, Samuel Dower, Ekdeep Singh Lubana, Dmitrii Krasheninnikov

Challenge:	Probing has emerged as a promising method for monitoring large language models (LLMs), enabling cheap inference-time detection of concerning behaviours.
Approach:	They propose to use off-policy data to influence probe generalisation across eight distinct LLM behaviours to test their hypothesis.
Outcome:	The proposed method can be used to generalise to on-policy examples in a large language model, but it is based on a lack of examples.

GR1: Reinforcement-Enhanced LLM for Geoscience Reasoning (2026.findings-acl)

Copied to clipboard

Yule Xie, Jiaxin Ding, Cheng Deng, Shiqing Gao, Junran Zhang, Sibo Zhang, Zeyuan Wang, Ke Wu, Xin Ding, Luoyi Fu, Meng Jin, Xinbing Wang

FOCUS: A Fine-Grained Customer-Oriented Sentiment Dialogue Summarization Dataset for Chinese Customer Service (2026.findings-acl)

Challenge:	Recent advances in large language models have demonstrated RL's substantial capacity to enhance multi-step reasoning beyond what supervised instruction tuning achieves.
Approach:	They propose a framework that converts multimodal questions into descriptive text . they propose RL-enhanced geoscience reasoning that can be fine-tuned to a text-only level .
Outcome:	The proposed framework improves accuracy and accuracy on multimodal questions while preserving answerability and difficulty.

Copied to clipboard

Qian Chen, Mengqiang Hu, Xin Guo

ABC-Bench: Benchmarking Agentic Backend Coding in Real-World Development (2026.findings-acl)

Challenge:	Existing studies largely overlook fine-grained sentiment dynamics expressed by customers . current methods often exhibit misalignment between aspects and sentiments .
Approach:	They propose a three-stage approach to building an aspect-aware sentiment dataset . they use a fine-grained customer-oriented Chinese dialogUe summarization dataset based on this scheme .
Outcome:	The proposed model improves faithfulness and interpretability of the proposed dataset.

Copied to clipboard

Jie Yang, Honglin Guo, Li Ji, Jiazheng Zhou, Rui Zheng, Zhikai Lei, Shuo Zhang, Zhiheng Xi, Shichun Liu, Yuxin Wang, Bo Wang, Yining Zheng, Tao Gui, Xipeng Qiu

LiveCANNBench: Benchmark SWE AI Coding for Ascend CANN (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have redefined the role of AI in software engineering . current benchmarks focus on localized code generation, but neglect dynamic, full-process requirements of real-world engineering.
Approach:	They propose a benchmark to evaluate agentic backend coding within a realistic, executable workflow.
Outcome:	The ABC-Bench benchmark evaluates agentic backend coding within a realistic, executable workflow.

Copied to clipboard

Sijie Wang, Kai Zhao, Wee Peng Tay, Shuo Zhang, Chengwen Liu, Quanjiang Guo, Ren Junhao, Xin Li, Heng Lian, Jingdi Lei, Rui She, Huacan Wang, Ronghao Chen

Challenge:	Recent advances in agents have enabled multi-file, multi-language, and dependency-aware AI coding.
Approach:	They propose an SWE-level benchmark for AI coding in the Huawei Ascend CANN software stack.
Outcome:	The proposed benchmark is constructed from real-world CANN repositories and consists of over 400 task instances spanning multiple file, multi-language, and execution-aware coding challenges.

When Models Decide and When They Bind: A Two-Stage Computation for Multiple-Choice Question Answering (2026.findings-acl)

Copied to clipboard

Hugh Mee Wong, Rick Nouwen, Albert Gatt

Who Watches the Watchmen? Humans Disagree With Translation Metrics on Unseen Domains (2026.findings-acl)

Challenge:	Multiple-choice question answering (MCQA) is easy to evaluate but adds a meta-task . prior work has shown that language models exhibit selection biases for particular option identifiers such as the label "A"
Approach:	They find that option-boundary residual states contain strong linearly decodable signals . winning content position becomes decoded after final option is processed .
Outcome:	The proposed model solves the problem and outputs the symbol that represents the answer.

Copied to clipboard

Finn Schmidt, Jan Philip Wahle, Terry Ruas, Bela Gipp

Stress-Testing Emotional Support Models: Moving from Homogeneous to Diverse Help Seekers (2026.findings-acl)

Challenge:	Existing studies that analyze unseen domains vary translation systems, annotators, or evaluation conditions, confounding domain effects with human annotation noise.
Approach:	They propose to use human error span annotations to evaluate translations of six translation systems across one seen news domain and two unseen technical domains to address these biases.
Outcome:	The proposed model improves on the human annotations in two unseen domains and on the news domains.

Copied to clipboard

Chaewon Heo, Cheyon Jin, Yohan Jo

Progressive Planning and Reinforced Reasoning: Large Language Model-Guided Multi-hop Question Answering over Knowledge Graph (2026.findings-acl)

Challenge:	Existing simulators fail to capture behavioral diversity of real-world seekers . lack of reliable automated evaluation frameworks hinders field's establishment .
Approach:	They propose a controllable seeker simulator driven by nine psychological and linguistic features that underpin seeker behavior.
Outcome:	The proposed model achieves superior profile adherence and behavioral diversity compared to existing approaches.

Copied to clipboard

Xiang Li, Runhai Jiao, Ruifan Li, Dongnan Wu, Ruojiao Qiao, Lei Liu

AED-RAG: Continuous Multi-Granular Context Fusion for Retrieval-Augmented Generation via Adaptive Ensemble Decoding (2026.findings-acl)

Challenge:	Existing approaches to multi-hop question answering lack effective intermediate guidance and policy networks focus on local neighborhood information, making it difficult to anticipate the long-term consequences of decisions.
Approach:	They propose a framework that converts decomposed sub-question sequences into stepwise decision guidance and a structure-aware lookahead policy network to enhance the agent's global state awareness and decision foresight in complex environments.
Outcome:	The proposed framework surpasses state-of-the-art methods while showing strong generalization.

Copied to clipboard

Junzhe Zhou, Fulin Lin, Tairan Cheng, Shaowen Chen, Hongwei Wang

Neural Chain-of-Thought Search: Searching the Optimal Reasoning Path to Enhance Large Language Models (2026.findings-acl)

Challenge:	Existing alignment strategies that rely on discrete reranking struggle to address this granularity mismatch or effectively balance external evidence with internal knowledge.
Approach:	They propose a framework that synergizes discrete retrieval with continuous reranking to discern the information density differences between unstructured narrative passages and structured knowledge triplets.
Outcome:	Extensive experiments on four open-domain QA benchmarks show that AED-RAG significantly outperforms competitive baselines.

Copied to clipboard

Guoming Ling, Zhongzhan Huang, Yupei Lin, Junxin Li, Shanshan Zhong, Hefeng Wu, Liang Lin

GOAT: A Training Framework for Goal-Oriented Agent with Tools (2026.findings-acl)

Challenge:	Recent research indicates that Large Reasoning Models suffer from a strategic bottleneck at reasoning path planning.
Approach:	They propose a framework that reformulates reasoning as a dynamic search for the optimal thinking strategy.
Outcome:	The proposed framework improves accuracy and computational cost while reducing generation length by over 22%.

Copied to clipboard

Hyunji Min, Sangwon Jung, Junyoung Sung, Dosung Lee, Leekyeung Han, Paul Hongsuck Seo

LLMs as ASP Programmers: Self-Correction Enables Task-Agnostic Nonmonotonic Reasoning (2026.findings-acl)

Challenge:	Recent advances in large language models have led to remarkable progress across a wide range of natural language processing tasks.
Approach:	They propose a training framework that enables fine-tuning LLM agents without human annotation.
Outcome:	The proposed framework enables fine-tuning LLM agents without human annotation.

Copied to clipboard

Adam Ishay, Joohyung Lee

DRIV-EX: Counterfactual Explanations for Driving LLMs (2026.findings-acl)

Challenge:	Recent large language models struggle with high computational costs and logical inconsistencies . a framework that translates natural language into Answer Set Programming (ASP) is developed .
Approach:	They propose a framework that translates natural language into Answer Set Programming (ASP) stable model semantics allow LLMs to express default rules and exceptions, they show .
Outcome:	The proposed framework outperforms existing methods on nonmonotonic reasoning tasks without any per-task engineering and applies uniformly across reasoning tasks.

Copied to clipboard

Amaia Cardiel, Eloi Zablocki, Elias Ramzi, Eric Gaussier

ROMA: Real-time Omni-Multimodal Assistant with Interactive Streaming Understanding (2026.findings-acl)

Challenge:	Large language models (LLMs) are increasingly used as reasoning engines in autonomous driving, yet their decision-making remains opaque.
Approach:	They propose to use gradient-based optimization on continuous embeddings to identify the input shifts required to flip a model’s decision.
Outcome:	The proposed method exposes latent biases and provides concrete insights to improve the robustness of LLM-based driving agents.

Copied to clipboard

Tian Xueyun, Wei Li, Bingbing Xu, Heng Dong, Yuanzhuo Wang, Huawei Shen

Towards Provably Secure Generative AI: Reliable Consensus Sampling (2026.findings-acl)

Challenge:	Existing omni-multimodal large language models lack incomplete modality support or lack autonomous proactive monitoring.
Approach:	They propose a real-time omni-multimodal assistant for unified reactive and proactive interaction that decouples response initiation from generation to ensure precise triggering without task conflict.
Outcome:	The proposed model achieves state-of-the-art performance on proactive tasks while competing in reactive settings.

Copied to clipboard

Yu Cui, Hang Fu, Sicheng Pan, Zhuoyu Sun, Yifei Liu, Yuhong Nie, Bo Ran, Baohan Huang, Xufeng Zhang, Haibin Zhang, Cong Zuo, Licheng Wang

Pause or Fabricate? Training Language Models for Grounded Reasoning (2026.findings-acl)

Challenge:	Existing research on generative AI security is driven by mutually reinforcing attack and defense methodologies grounded in empirical experience.
Approach:	They propose a new algorithm that uses a random sampling algorithm to control risk.
Outcome:	The proposed algorithm improves robustness and utility while maintaining latency comparable to existing algorithms.

Copied to clipboard

Yiwen Qiu, Linjuan Wu, Yizhou Liu, Yuchen Yan, Jin Ma, Xu Tan, Yao Hu, Daoxin Zhang, Wenqi Zhang, Weiming Lu, Jun Xiao, Yongliang Shen

AgentV-RL: Scaling Reward Modeling with Agentic Verifier (2026.findings-acl)

Challenge:	Large language models implicitly fabricate information when inputs are incomplete, causing confidence but unreliable conclusions.
Approach:	They propose a framework for grounded reasoning under incomplete information that decomposes reasoning into two stages . they propose stage-specific rewards to penalize hallucinations, enabling models to detect gaps, stop proactively, and resume reasoning after clarification.
Outcome:	The proposed framework improves premise detection and task success by 30% . it also reduces average response length by over 20% .

Copied to clipboard

Jiazheng Zhang, Ziche Fu, Zhiheng Xi, Wenqing Jing, Mingxu Chai, Wei He, Guoqiang Zhang, Chenghao Fan, Chenxin An, Wenxiang Chen, Zhicheng Liu, Haojie Pan, Dingwei Zhu, Tao Gui, Qi Zhang, Xuanjing Huang

DiTReducio: A Training-Free Acceleration for DiT-Based TTS via Progressive Calibration (2026.findings-acl)

Challenge:	Existing approaches to improve LLM reasoning are limited in complex domains and lack external grounding makes verifiers unreliable on computation-intensive tasks.
Approach:	They propose a framework that transforms reward modeling into a multi-turn, tool-augmented deliberative process.
Outcome:	The proposed framework surpasses state-of-the-art ORMs by 25.2% under parallel and sequential TTS.

Copied to clipboard

Yanru Huo, Ziyue Jiang, Zuoli Tang, Qingyang Hong, Zhou Zhao

DcLM: Output Length Control of Large Language Models via Dynamic Length Markers (2026.findings-acl)

Challenge:	Existing training-free acceleration approaches for text-to-speech models are constrained by training costs.
Approach:	They propose a training-free acceleration framework that compresses computations in DiT-based TTS models . they propose Temporal Skipping and Branch Skipping to eliminate redundant computations .
Outcome:	Experimental results show that the proposed framework reduces FLOPs and improves RTF by 37.1%.

Copied to clipboard

Zhe Chen, Jiaao Yu, Honglin Li

LongMP-Bench: A Benchmark for Multimodal Persona Understanding in Long-Term Dialogues (2026.findings-acl)

Challenge:	Large language models (LLMs) have limited awareness of output length, making it difficult to satisfy precise length requirements.
Approach:	They propose a model-agnostic approach that introduces dynamic length markers to guide length-controllable outputs.
Outcome:	The proposed method significantly reduces length deviation across multiple datasets.

Copied to clipboard

Zhuoqun Li, Zhaopei Huang, Wenxuan Wang, Qin Jin

TOOLCAD: Exploring Tool-Using Large Language Models in Text-to-CAD Generation with Reinforcement Learning (2026.findings-acl)

Challenge:	Existing datasets suffer from limited persona diversity and static, overly simplified settings, making them insufficient for capturing the complexity of real-world interactions.
Approach:	They propose a benchmark to evaluate models' ability to understand evolving user personas within long-term multimodal dialogues by using a dataset that contains long conversations from 150 users.
Outcome:	The proposed benchmark aims to assess models' ability to track persona evolution, integrate visual and textual inputs, and apply persona understanding in realistic dialogue scenarios.

Copied to clipboard

Yifei Gong, Xing Wu, Wenda Liu, null Tukang

Beyond Overlap Metrics: Rewarding Reasoning and Preferences for Faithful Multi-Role Dialogue Summarization (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have shown remarkable advances in enabling language agents to tackle real-world tasks.
Approach:	They propose a tool-using agent-based CAD framework that automates text-to-CAD modeling . they propose an interactive CAD gym to roll out reasoning and tool-augmented interaction trajectories with the CAD engine .
Outcome:	The proposed framework can generalize across complex modeling tasks, supporting their open-source counterparts.

Copied to clipboard

Xiaoyong Mei, Tingting Zuo, Da Chen, Guangyu Hu, Xiangyu Wen, Chao Duan, Mingyan Zhang, Fudan Zheng

From Signal Degradation to Computation Collapse: Uncovering the Two Failure Modes of LLM Quantization (2026.findings-acl)

Challenge:	Existing methods for multi-role dialogue summarization favor surface-level imitation of references rather than genuine gains in faithfulness or alignment with human preferences.
Approach:	They propose a framework that couples explicit cognitive-style reasoning with reward-based optimization for multi-role dialogue summarization.
Outcome:	The proposed framework matches strong baselines on ROUGE and BERTScore, while in-depth analysis on SAMSum shows clear gains in factual faithfulness and model-based preference alignment.

Copied to clipboard

Chenxi Zhou, Pengfei Cao, Jiang Li, Bohan Yu, Jinyu Ye, Jun Zhao, Kang Liu

Challenge:	Existing research on PTQ spans three primary directions.
Approach:	They conduct a systematic analysis of post-training quantization failures using PTQ . they show that targeted repair can mitigate Signal Degradation but remains ineffective for Computation Collapse .
Outcome:	The proposed method mitigates Signal Degradation but remains ineffective for Computation Collapse.

Establishing a Scale for Kullback-Leibler Divergence in Language Models Across Various Settings (2026.findings-acl)

Copied to clipboard

Ryo Kishino, Yusuke Takase, Momose Oyama, Hiroaki Yamagiwa, Hidetoshi Shimodaira

Simple Role Assignment is Extraordinarily Effective for Safety Alignment (2026.findings-acl)

Challenge:	Fig. 1 and 2 shows that log-likelihood vectors provide a consistent representation for language models . weight permutation symmetries and architectural dependencies hinder direct comparisons between models with different learning methods or designs.
Approach:	They propose a log-likelihood vector for comparing language models as probability distributions . they establish a consistent scale for KL divergence across various settings .
Outcome:	The proposed model comparisons show that the log-likelihood space is smaller than the weight space . the proposed model compares language models across checkpoints, model sizes, quantization, fine-tuning, and layers .

Copied to clipboard

Zhou Ziheng, Jiakun Ding, Zhaowei Zhang, Ruosen Gao, Ying Nian Wu, Demetri Terzopoulos, Yipeng Kang, Fangwei Zhong, Junqi Wang

Challenge:	a new study proposes a role-conditioned pipeline for value alignment . principles alone are incomplete, and they provide little guidance on when and how a value applies in context.
Approach:	They propose a role-conditioned pipeline with role-based critics and a model-free approach that is based on role conditioning.
Outcome:	The proposed approach outperforms principle-based, Chain-of-Thought and other benchmarks.

Task-Stratified Knowledge Scaling Laws for Post-Training Quantized Large Language Models (2026.findings-acl)

Copied to clipboard

Chenxi Zhou, Pengfei Cao, Jiang Li, Bohan Yu, Jinyu Ye, Jun Zhao, Kang Liu

MemPO: Self-Memory Policy Optimization for Long-Horizon Agents (2026.findings-acl)

Challenge:	Existing scaling laws focus on general performance, overlooking crucial fine-grained factors and how quantization differentially impacts diverse knowledge capabilities.
Approach:	They propose a framework that unifies model size, bit-width, and fine-grained factors into memorization, application, and reasoning.
Outcome:	The proposed framework shows strong fit and cross-architecture consistency on 293 different PTQ configurations.

Copied to clipboard

Ruoran Li, Xinghua Zhang, Haiyang Yu, Shitong Duan, Xiang Li, Wenxin Xiang, Chonghua Liao, Xudong Guo, Yongbin Li, Jinli Suo

Challenge:	Existing methods for long-horizon agents introduce the external memory module and look up the relevant information from the stored memory, which prevents the model from proactively managing its memory content and aligning with the agent’s overarching task objectives.
Approach:	They propose an algorithm which enables agents to autonomously manage their memory during interaction with environment and selectively retain crucial information.
Outcome:	Extensive experiments show that the proposed algorithm achieves absolute F1 score gains of 25.98 over the base model and 7.1 over the previous SOTA baseline while preserving task performance.

Reason Only When Needed: Efficient Generative Reward Modeling via Model-Internal Uncertainty (2026.findings-acl)

Copied to clipboard

Chao Xue, Yao Wang, Mengqiao Liu, Di Liang, Xingsheng Han, Peiyang Liu, Xianjie Wu, Chenyao Lu, Lei Jiang, Yu Lu, Haibo Shi, Shuang Liang, Minlong Peng, Flora D. Salim

Metacognitive Self-Correction for Multi-Agent System via Prototype-Guided Next-Execution Reconstruction (2026.findings-acl)

Challenge:	Existing approaches to generating reward models rely on voting-based mechanisms to evaluate CoT outputs.
Approach:	They propose an efficient generative reward modeling framework grounded in model-internal uncertainty.
Outcome:	The proposed framework reduces inference cost while improving answer accuracy.

Copied to clipboard

Xu Shen, Qi Zhang, Song Wang, Zhen Tan, Xinyu Zhao, Laura Yao, Vaishnav Tadiparthi, Hossein Nourkhiz Mahjoub, Ehsan Moradi Pari, Kwonjoon Lee, Tianlong Chen

DualGuard: Dual-stream Large Language Model Watermarking Defense against Paraphrase and Spoofing Attack (2026.findings-acl)

Challenge:	Large Language Model based multi-agent systems (MAS) excel at collaborative problem solving but remain brittle to cascading errors.
Approach:	They propose a metacognitive framework that enables step-level error detection and self-correction in Large Language Model based multi-agent systems (MAS) .
Outcome:	The proposed framework outperforms baselines on the Who When benchmark and delivers consistent gains on AgentErrorBench.

Copied to clipboard

Hao Li, Yubing Ren, Yanan Cao, Yingjie Li, Fang Fang, Shi Wang, Li Guo

PeerCheck: Enhancing LLM-Generated Academic Reviews Towards Human-Level Quality (2026.findings-acl)

Challenge:	Existing watermarking algorithms focus on defending against paraphrase and piggyback spoofing attacks, which can inject harmful content, compromise reliability, and undermine trust in attribution.
Approach:	They propose an algorithm capable of defending against paraphrase and spoofing attacks.
Outcome:	Experiments on large language models and language models show that DualGuard is the first watermarking algorithm capable of defending against both paraphrase and spoofing attacks.

Copied to clipboard

Zeyuan Chen, Ziqing Yang, Yihan Ma, Michael Backes, Yang Zhang

Challenge:	Increasing use of large language models (LLMs) in academic review has raised concerns about quality and fairness.
Approach:	They propose a framework to improve the quality of LLM-generated reviews by using retrieval-augmented generation.
Outcome:	The proposed framework improves the human-level quality of LLM-generated reviews by adopting prompt engineering and retrieval-augmented generation.

MorphoBench: A Benchmark with Difficulty Adaptive to Model Reasoning (2026.findings-acl)

Copied to clipboard

Xukai Wang, Xuanbo Liu, Mingrui Chen, Haitian Zhong, Xuanlin Yang, Bohan Zeng, Jinbo Hu, Hao Liang, Junbo Niu, Xuchen Li, Ruitao Wu, Ruichuan An, Yang Shi, Liu Liu, Qiang Liu, Zhouchen Lin, Xu-Yao Zhang, Wentao Zhang, Bin Dong

Privacy Risks of Intermediate Representations: Attribute Inference in Distributed LLM Inference (2026.findings-acl)

Challenge:	Existing benchmarks designed to evaluate the reasoning capabilities of large models are limited in scope and lack flexibility to adapt difficulty according to evolving reasoning capacities of models.
Approach:	They propose a benchmark that incorporates multidisciplinary questions to evaluate the reasoning capabilities of large models and can adjust and update question difficulty based on the reasoning abilities of advanced models.
Outcome:	The proposed benchmark incorporates multidisciplinary questions to evaluate the reasoning capabilities of large models and can adjust and update question difficulty based on the reasoning abilities of advanced models.

Copied to clipboard

Yang Lyu, Jin Cao, Yang Xiao, Zhe Sun, Ben Niu, Fenghua Li, Hui LI

ReCon: Active Defense against Large Vision-Language Model Jailbreaks via Reverse Safety Concept Injection (2026.findings-acl)

Challenge:	Distributed LLMs avoid raw inputs by transmitting intermediate hidden states, a practice widely assumed to preserve privacy.
Approach:	They propose a distributed inference framework that transmits intermediate hidden states to avoid sending raw inputs by exposing sensitive user attributes.
Outcome:	The proposed approach achieves Top-1 accuracy of 0.997 on CMS, 0.980 on Skytrax, and 0.986 on ECHR.

Copied to clipboard

Zheng He, Yiwei Wang, Hongxing Wang, Yujun Cai

Verifying the Subjective: Structured Multilingual Rewards for Low-Resource Alignment (2026.findings-acl)

Challenge:	Existing defense strategies neglect visual threats and lack of fine-grained specificity regarding specific attack semantics.
Approach:	They propose a black-box defense framework that maps unsafe concepts to fine-grained, constructive Safe Concepts.
Outcome:	a new black-box defense framework enhances robustness against jailbreak attacks . it maps detected unsafe concepts to fine-grained, constructive Safe Concepts . the proposed framework is available for free at http://www.epa.org/recon/ .

Copied to clipboard

Jiu Sha, Mengxiao Zhu

Memory-Driven Role-Playing: Evaluation and Enhancement of Persona Knowledge Utilization in LLMs (2026.findings-acl)

Challenge:	Structured Multilingual Reward Modeling Framework extends Reinforcement Learning with Verifiable Rewards (RLVR) to subjective and open-ended tasks.
Approach:	They propose a framework that extends Reinforcement Learning with Verifiable Rewards to subjective and open-ended tasks.
Outcome:	The proposed framework improves reasoning capability and response quality on 7 tasks across 50 low-resource languages.

Copied to clipboard

Kai Wang, Haoyang You, Yang Zhang, Zhongjie Wang

Scripts Through Time: A Survey of the Evolving Role of Transliteration in NLP (2026.findings-acl)

Challenge:	Existing models fail to recall and accurately apply designated persona knowledge without explicit cues . memory-driven role-playing paradigms are attracting significant interest .
Approach:	They propose a memory-driven role-playing paradigm that frames persona knowledge as the LLM's internal memory store and a prompting architecture that guides structured memory retrieval and response generation.
Outcome:	The proposed paradigm provides a comprehensive diagnostic for four-stage role-playing abilities across 12 LLMs.

Copied to clipboard

Thanmay Jayakumar, Deepon Halder, Raj Dabre

Health-ORSC-Bench: A Benchmark for Measuring Over-Refusal and Safety Completion in Health Context (2026.findings-acl)

Challenge:	Cross-lingual transfer is often hindered by the "script barrier" where differences in writing systems inhibit transfer learning . transliteration is a powerful technique to bridge this gap by increasing lexical overlap . authors present a taxonomy of key motivations to utilize transliterations in language models .
Approach:	They propose a taxonomy of key motivations to utilize transliterations in NLP . they analyze the evolution and effectiveness of these methods and discuss trade-offs .
Outcome:	The proposed transliteration technique is effective in cross-lingual NLP, the authors argue . the proposed translliteration method is a powerful tool to overcome the "script barrier"

Copied to clipboard

Zhihao Zhang, Liting Huang, Guanghao Wu, Preslav Nakov, Heng Ji, Usman Naseem

LLM Inductive Reasoning Through Multi-Agent Enhanced Monte Carlo Tree Search (2026.findings-acl)

Challenge:	Existing safety alignment benchmarks fail to evaluate Safe Completion: the model’s ability to maximise helpfulness on dual-use or borderline queries without crossing into actionable harm.
Approach:	They propose a large-scale benchmark to measure Over-Refusal and Safe Completion quality in healthcare.
Outcome:	The framework evaluates 30 state-of-the-art LLMs including GPT-5 and Claude-4.

Copied to clipboard

Xiang Li, Yucheng Zhou, Xiangzhi Wei, Zesheng Shi, Haiyuan Wan, Gong Yifan, Fangming Liu, Jing Li

UniCreative: Unifying Long-form Logic and Short-form Sparkle via Reference-Free Reinforcement Learning (2026.findings-acl)

Challenge:	Existing methods for enhancing inductive reasoning of large language models often lack explicit optimization guidance and effective error correction.
Approach:	They propose a plug-and-play test-time framework that integrates multi-agent coordination with Monte Carlo Tree Search to improve inductive reasoning.
Outcome:	The proposed framework outperforms existing methods on four benchmarks and shows consistent improvements on QWQ-32B and Deepseek-V3 .

Copied to clipboard

Xiaolong Wei, Zerun Zhu, Simin Niu, Xingyu Zhang, Peiying Yu, Changxuan Xiao, Yuchen Li, Jicheng Yang, Zhejun Zhao, Chong Meng, Long Xia, Daiting Shi

Challenge:	Existing alignment paradigms for creative writing use static reward signals and supervised data.
Approach:	They propose a constraint-aware reward model that synthesizes query-specific criteria to provide fine-grained preference judgments.
Outcome:	The proposed framework aligns models with human preferences across content quality and structural paradigms without supervised fine-tuning and ground-truth references.

FFN Lens: How Transformers Divide Labor for Multilingual Tasks (2026.findings-acl)

Copied to clipboard

Jiatong Li, Hailong Cao, Yang Liu

Seeing Beyond Words: MatVQA for Challenging Visual-Scientific Reasoning in Materials Science (2026.findings-acl)

Challenge:	Large Language Models (LLMs) exhibit strong performance on multilingual tasks, yet the process of constructing predictions in the target language remains under-explored.
Approach:	They propose a novel interpretability method focusing on the Feed-Forward Network (FFN) layers of Large Language Models.
Outcome:	The proposed interpretability method is based on the Feed-Forward Network (FFN) layer of Large Language Models.

Copied to clipboard

Sifan Wu, Huan Zhang, Yizhan Li, Farshid Effaty, Hongyuan Mei, Amirreza Ataei, Bang Liu

RLShield: Dynamic Jailbreak Detection for LLMs via Reinforced Adaptive Learning (2026.findings-acl)

Challenge:	Multimodal Large Language Models (MLLMs) outperform existing benchmarks in both natural language and coding domains.
Approach:	They propose a scalable benchmark that integrates vision and language modalities to address this gap by eliminating textual shortcuts.
Outcome:	The new benchmark outperforms existing benchmarks in both natural language and coding domains.

Copied to clipboard

Zhao Tong, Pengfei Yang, Yimeng Gu, Haichao Shi, Qiang Liu, Xingcheng Xu, Shu Wu, Xiao-Yu Zhang

MEDSYN: Benchmarking Multi-EviDence SYNthesis in Complex Clinical Cases for Multimodal Large Language Models (2026.findings-acl)

Challenge:	Existing approaches to detect jailbreak prompts rely on static model components or fixed decision thresholds.
Approach:	They propose a dynamic jailbreak detection framework that employs reinforcement learning for adaptive threshold selection.
Outcome:	Experimental results show that the framework outperforms baselines in detection performance while maintaining high computational efficiency.

Copied to clipboard

Boqi Chen, Xudong Liu, Jiachuan Peng, Marianne Frey-Marti, Kyle Lam, Bang Zheng, Lin Li, Jianing Qiu

Sem-DPO: Mitigating Semantic Inconsistency in Preference Optimization for Prompt Engineering (2026.findings-acl)

Challenge:	Existing benchmarks for multimodal large language models do not capture real-world clinical complexity.
Approach:	They evaluate multilingual, multimodal multimodal models of clinical cases with up to 7 distinct visual clinical evidence types per case.
Outcome:	The proposed model outperforms human models on differential diagnosis (DDx) generation and final diagnosis (FDx) selection.

Copied to clipboard

Anas Mohamed, Azal Ahmad Khan, Xinran Wang, Ahmad Faraz Khan, Shuwen Ge, Saman Bahzad Khan, Ayaan Ahmad, Ali Anwar

BNLP: A Text Annotation Platform for Quality Control of LLM-Generated Annotations (2026.findings-acl)

Challenge:	Direct Preference Optimization (DPO) is an off-policy alternative to RL for automatic prompt engineering, but its token-level regularization leaves semantic inconsistency unchecked as prompts that win higher preference scores can still drift away from the user’s intended meaning.
Approach:	They propose a variant of Direct Preference Optimization that preserves semantic consistency while maintaining its simplicity and efficiency.
Outcome:	The proposed model outperforms state-of-the-art prompt optimization baselines and several DPO variants on three standard text-to-image prompt-optimization benchmarks and three language models.

Copied to clipboard

Xinhao Zhuang, Qiongyu Tian, Yalin Chen, Tianle Xin, Yongyong Fu, Yunchao Ling, Guoqing Zhang

Challenge:	Existing annotation tools lack support for Large Language Models (LLMs) or use LLMs as one-off preannotation engines, compromising data reliability.
Approach:	They propose a text annotation platform that embeds LLM-assisted labeling into a quality-aware collaborative workflow.
Outcome:	Experiments show that BNLP reduces annotation time by 74.3% and improves annotation quality by 11.6% over purely manual annotation in LLM-assisted settings.

ElasticFlow: One-Step Physics-Consistent Policy with Elastic Time Horizons for Language-Guided Manipulation (2026.findings-acl)

Copied to clipboard

Kewei Chen, Yayu Long, Shuai Li, Mingsheng Shang

AdaptiveK: Complexity-Driven Sparse Autoencoders for Interpretable Language Model Representations (2026.findings-acl)

Challenge:	Existing methods for embodied AI use iterative denoising to achieve high latency and lack physical consistency.
Approach:	They propose a distillation-free, physics-consistent one-step policy framework that reconstructs the Mean Field Theory by directly modeling the average velocity field.
Outcome:	Experiments on LIBERO, CALVIN, and RoboTwin show that the proposed framework outperforms state-of-the-art methods on long-horizon tasks.

Copied to clipboard

Yifei Yao, Hanrong Zhang, Mengnan Du

Experience-Driven Multi-Agent Optimization for Black-Box Jailbreak Attacks on Large Language Models (2026.findings-acl)

Challenge:	Existing approaches to decomposing model activations into interpretable features fail to account for input complexity.
Approach:	They propose a framework that dynamically adjusts sparsity levels based on the semantic complexity of each input.
Outcome:	The proposed framework outperforms fixed-sparsity approaches on reconstruction fidelity, explained variance, cosine similarity and interpretability metrics while eliminating the burden of extensive hyperparameter tuning.

Copied to clipboard

Zhaoyang Han, Yihe Liu, Kai Zhang, Ping Li

CMIG: Conceptual Metaphor Theory-Inspired Framework for Metaphorical Image Generation (2026.findings-acl)

Challenge:	Existing methods for jailbreak have poor transferability and high sensitivity to preprocessing . EMJO provides an effective and scalable paradigm for systematic jailbreak optimization .
Approach:	They propose a model that couples agents into a closed-loop "probe–evaluate–revise” process . they propose EMJO, which can be query-efficient and transferable, under black-box access.
Outcome:	a new approach outperforms existing jailbreak baselines on diverse LLMs . it achieves up to 11% improvement in attack success rate while reducing query cost .

Copied to clipboard

Qingbao Huang, Cheng Yang, Jiawei Yao, Zhiyue Liu, Yi Cai, Xingmao Zhang

Position: From Noise to Signal to Selbstzweck - Reframing Human Label Variation in the Era of Post-training in NLP (2026.findings-acl)

Challenge:	Existing text-to-image systems often produce visually plausible but semantically literal outputs.
Approach:	They propose a structured prompting framework inspired by Conceptual Metaphor Theory . they propose to identify source–target mappings, filter projectable source attributes and select a visual realization strategy in a reproducible reasoning workflow.
Outcome:	The proposed framework improves semantic alignment and controllability on metaphor prompts.

Copied to clipboard

Shanshan Xu, Santosh T.y.s.s, Barbara Plank

Challenge:	Human Label Variation (HLV) refers to legitimate disagreement in annotation . current preference-learning datasets routinely collapse multiple annotations into a single label .
Approach:	They propose to preserve human label variation as an embodiment of pluralism . they argue that disagreement in annotations should be treated as a selfzweck .
Outcome:	The proposed approach preserves pluralism and human pluralismos, the authors argue . they argue that disagreements in annotations should be treated as a selfzweck .

Rethinking Assessments of Prompt Injection Attacks (2026.findings-acl)

Copied to clipboard

Chi Cui, Yixin Wu, Michael Backes, Yang Zhang

SegDRE: A Salient Entity Guided Approach to Document-Level Relation Extraction (2026.findings-acl)

Challenge:	Prompt injection attacks are recognized as one of the primary risks faced by LLM-integrated applications in recent years.
Approach:	They evaluate prompt injection attacks on LLM-integrated applications across 37 target tasks, 185 injected tasks, 21 attack instructions, and 143,745 queries.
Outcome:	The proposed framework provides a solid foundation for assessing vulnerabilities in LLM-integrated applications and evaluating the efficacy of defensive strategies.

Copied to clipboard

Xing Yang, Chengxiang Tan

Live-Aid: A Large-Scale Dialogue Dataset and Benchmark for Interleaved Multi-party Interactions in Live Streaming (2026.findings-acl)

Challenge:	Existing models struggle to address two major bottlenecks in Document-level Relation Extraction: extreme class imbalance and complexity of multi-hop reasoning.
Approach:	They propose a method that decouples the extraction space into dense and sparse scenarios.
Outcome:	The proposed approach yields consistent improvements over various backbone models and achieves advanced performance compared to existing enhancement methods.

Copied to clipboard

Yiming Lei, Yize Fan, Zeming Liu, Jiaji Dong, Hui Qiu, Haitao Leng, Qingjie Liu, Kehai Chen, Tingting Gao, Yunhong Wang

Semi-Supervised Diseased Detection from Speech Dialogues with Multi-Level Data Modeling (2026.findings-acl)

Challenge:	Existing Multimodal Large Language Models struggle with dynamic interactions due to the scarcity of high-quality interleaved data.
Approach:	They propose a large-scale interleaved live interaction Chinese dataset with human-annotated video responses.
Outcome:	The proposed model can be used to evaluate live interactions in Chinese over 1,100 hours and 80,037 dialogue turns.

Copied to clipboard

Xingyuan Li, Mengyue Wu

From Isolated Scoring to Collaborative Ranking: A Comparison-Native Framework for LLM-Based Paper Evaluation (2026.findings-acl)

Challenge:	Existing methods for learning from weakly-supervised speech data are hampered by severe data scarcity and the subjective nature of clinical annotations.
Approach:	They propose a framework that explicitly models pathological traits by jointly learning from frame-level, segment-level and session-level representations within unsegmented clinical dialogues.
Outcome:	The proposed framework is model-agnostic, robust across languages and conditions, and highly data-efficient.

Copied to clipboard

Pujun Zheng, Jiacheng Yao, Jinquan Zheng, Chenyang Gu, Guoxiu He, Jiawei Liu, Yong Huang, Tianrui Guo, Wei Lu

UnAC: Adaptive Visual Prompting with Abstraction and Stepwise Checking for Complex Multimodal Reasoning (2026.findings-acl)

Challenge:	Large language models (LLMs) are currently used to evaluate scientific papers by assigning an absolute score to each paper independently.
Approach:	They propose a comparison-native framework for paper evaluation that integrates comparison into both data construction and model learning.
Outcome:	The proposed framework achieves an average relative improvement of 21.8% over the strong baseline DeepReview-14B, while exhibiting robust generalization to five previously unseen datasets.

Copied to clipboard

Yifan Wang, Yun Fu

AnalystBench: Benchmarking professional long-form report generation with web-mined multimodal tasks (2026.findings-acl)

Challenge:	Recent large multimodal models (LMMs) have demonstrated impressive capabilities in image understanding, yet they struggle to perform complex reasoning on multimodal problems.
Approach:	They propose a multimodal prompting method that strengthens reasoning for multimodal tasks in large multimodal models.
Outcome:	The proposed method improves reasoning on three public benchmarks and shows that it can be used to extract key information from images.

Copied to clipboard

Chau Minh Pham, Zichao Wang, Puneet Mathur, Alexa Siu, Akriti Jain, Aparna Garimella, Ananya B. Sai, Nedim Lipka, Mohit Iyyer, Varun Manjunatha

BizCompass: Benchmarking the Reasoning Capabilities of LLMs in Business Knowledge and Applications (2026.findings-acl)

Challenge:	Existing benchmarks decompose the end-to-end professional report generation into individual components.
Approach:	They propose a benchmarking tool that evaluates 20 real-world professional report generation tasks grounded in multimodal document collections.
Outcome:	The proposed model outperforms closed-source models on executive summarization tasks but drops significantly on long-horizon synthesis tasks.

Copied to clipboard

Jianing Hao, Yuhe Wu, Yuanjian Xu, Shichang Meng, Shuai Yuan, Wei Zeng, Zixuan Wang, Guang Zhang

When More Thinking Hurts: Overthinking in LLM Test-Time Compute Scaling (2026.findings-acl)

Challenge:	Existing benchmarks focus on narrow tasks and leave a fundamental question unanswered . Existing models only focus on specific tasks, requiring rigorous reasoning and knowledge .
Approach:	They propose a benchmark to connect theoretical foundations with practical business knowledge and applications.
Outcome:	The benchmark systematically evaluates both open-source and commercial LLMs . it reveals how theoretical knowledge translates into practical performance in business .

Copied to clipboard

Shu Zhou, Rui Ling, Junan Chen, Xin Wang, Tao Fan, Hao Wang

Learning from Failures: Error Notebook-guided Secure Code Generation (2026.findings-acl)

Challenge:	Existing research implicitly assumes that longer thinking leads to better results . a recent study suggests that test-time compute scaling is more effective than model scaling .
Approach:	They challenge the assumption that longer thinking yields better results . they show that models exhibit overthinking and marginal returns diminish at higher budgets .
Outcome:	The proposed framework reduces computation significantly while maintaining comparable accuracy.

Copied to clipboard

Xinyu Zhong, Peng Lan, Zhifang Liao

Evaluating Reasoning Models for Queries with Presuppositions (2026.findings-acl)

Challenge:	Existing security code generation methods rely on abstract security knowledge, resulting in suboptimal security.
Approach:	They propose a framework that integrates a Security Error Notebook and a Function Erro Notebook to provide concrete, actionable guidance to LLMs.
Outcome:	The proposed framework achieves a substantial leap in SP@1 metric, with GPT-4o-mini performance improving from 60.21% to 66.7% on CodeGuard+.

Copied to clipboard

Rose Sathyanathan, Kinshuk Vasisht, Danish Pruthi

GRE Score: Generative Risk Evaluation for Large Language Models (2026.findings-acl)

Challenge:	Prior work notes that large language models fail to challenge erroneous assumptions and can reinforce users’ misinformed opinions.
Approach:	They construct queries with varying degrees of presuppositions spanning health, science, and general knowledge and evaluate several widely-deployed models.
Outcome:	The proposed models achieve higher accuracy but fail to challenge a large fraction of false presuppositions.

Copied to clipboard

Zaitang LI, Pin-Yu Chen, Tsung-Yi Ho

MSVBench: Towards Human-Level Evaluation of Multi-Shot Video Generation (2026.findings-acl)

Challenge:	Large language models have revolutionized generative tasks, but concerns about their trustworthiness and vulnerability to adversarial attacks persist.
Approach:	They propose an attack-independent evaluation of LLM robustness using conditional generation for synthetic text creation and a method to quantify the model's resilience.
Outcome:	The proposed method achieves a consistent ranking of LLM robustness when compared to the attack-based model ranking on TrustLLM (CITATION).

Copied to clipboard

Haoyuan Shi, Yunxin li, Nanhao Deng, Zhenran Xu, Xinyu Chen, Longyue Wang, Baotian Hu, Min Zhang

M-TRACE: Detecting and Mitigating Time-Anchor Drift via Step-wise Conflict Checking in Temporal Reasoning (2026.findings-acl)

Challenge:	Existing evaluation methods for complex multi-shot video are anchored to single-shot paradigms, lacking comprehensive story assets and cross-shot metrics.
Approach:	They propose a framework that synergizes the high-level semantic reasoning of Large Multimodal Models with the fine-grained perceptual rigor of domain-specific expert models.
Outcome:	The proposed framework synergizes the high-level semantic reasoning of Large Multimodal Models with the fine-grained perceptual rigor of domain-specific expert models.

Copied to clipboard

Danyu Huang, Jiayuan Jiang, Yao Zhang, Jun Wang, Huijia Li, Zhenglu Yang

Faithful Persona Steering under Incongruity via Dual-Stream Refinement (2026.findings-acl)

Challenge:	Experimental results show that M-TRACE effectively reduces time-anchor drift . external knowledge may be inaccurate while internal knowledge can become outdated .
Approach:	They propose a multi-agent reasoning framework for temporal knowledge conflicts . they propose 'TimeConfQA' which guides conflict-aware final reasoning .
Outcome:	Experimental results show that M-TRACE reduces time-anchor drift and improves performance on complex temporal question answering tasks.

Copied to clipboard

Yu-An Chu, Jen-Ren Pong, Chia-Yao Yeh, Meng-Fen Chiang

Challenge:	Existing methods for personalization in large language models often neglect incongruity of human personas . empirical evaluations show that QuirkyMind mitigates drift while preserving "quirks"
Approach:	a framework disentangles identity definition from its expression by anchoring traits in a dual-stream latent state.
Outcome:	a framework disentangles identity definition from its expression, and it mitigates drift without erasing authentic incongruities.

ChemVLR: Prioritizing Reasoning in Perception for Chemical Vision-Language Understanding (2026.findings-acl)

Copied to clipboard

Xuanle Zhao, Xinyuan Cai, Xiang Cheng, Xiuyi Chen, Bo XU

SDC-LoRA: Singular-Subspace Drift Controlled LoRA to Mitigate Knowledge Forgetting (2026.findings-acl)

Challenge:	Currently, vision-Language Models are optimized for direct visual question-answering tasks.
Approach:	They propose a visual-language-based VLM that prioritizes reasoning within the perception process.
Outcome:	The proposed model outperforms existing models and domain-specific open-source models in the chemical domain.

Copied to clipboard

Geyuan Zhang, Xiaofei Zhou, Shihao Liu, Jingyuan Tian, Jizheng Ma

Regret-Now: A Physics-Inspired Regret Framework for Temporal Knowledge Graph Question Answering with LLMs (2026.findings-acl)

Challenge:	Existing approaches to adapt LLMs to new tasks focus on limiting knowledge forgetting . et al., 2023b) suggest a solution to this problem by limiting update energy in the principal singular subspace of W0 .
Approach:	They propose a low-rank Adaptation (LoRA) that steers early updates away from principal directions and mitigates forgetting by constraining update energy in the principal singular subspace of W0.
Outcome:	The proposed model mitigates forgetting on MMLU, TruthfulQA, and HellaSwag while keeping minor-subspace updates unchanged.

Copied to clipboard

Danyu Huang, Yao Zhang, Jun Wang, Zhenglu Yang

Low-probability Tokens Sustain Exploration in Reinforcement Learning with Verifiable Reward (2026.findings-acl)

Challenge:	Large Language Models have impressive results in general reasoning tasks, but they still exhibit a lack of dynamic error-correction.
Approach:	They propose a temporal reasoning framework that uses the principle of minimum potential energy to model the reasoning process as a dynamic trajectory moving toward a more stable state.
Outcome:	The proposed framework shows consistent gains over strong baselines on two standard TKGQA benchmarks.

Copied to clipboard

Guanhua Huang, Tingqiang Xu, Mingze Wang, Qi Yi, Xue Gong, Siheng Li, Ruibin Xiong, Kejiao Li, Yuhao Jiang, Bo Zhou

Preserving Language Capabilities in Vision-Language Models via Representation Regulation (2026.findings-acl)

Challenge:	Recent studies show that RLVR training is slow and results plateau as policy entropy collapses . low-probability regularization (Lp-Reg) reduces the number of low-quality exploratory tokens induced by RL training .
Approach:	They propose a method to reduce RLVR over-penalization by eliminating low-probability exploratory tokens . they propose 'Low-provability Regularization' to reduce the gradual elimination of low-quality exploratory entropy tokens.
Outcome:	The proposed method eliminates low-probability exploratory tokens and prevents suppression of potentially valuable low-property candidates.

Copied to clipboard

ZiXuan Chen, Juncheng Tao, Ziqian Zeng

Me-Agent: A Personalized Mobile Agent with Two-Level User Habit Learning for Enhanced Interaction (2026.findings-acl)

Challenge:	Vision-Language Models (VLMs) provide a unified framework to process both text-only and vision-language tasks.
Approach:	They propose a method to reduce the distance between visual and textual representations by introducing a Representation Distribution Difference (RDD) loss.
Outcome:	Empirical evidence shows that finetuning VLMs on vision-language data has degraded language capabilities.

Copied to clipboard

Shuoxin Wang, Chang Liu, Gowen Loo, Lifan Zheng, Kaiwen Wei, Huanqian Yan, Xinyi Zeng, Jingyuan Zhang, Yu Tian

Dynamic PMI-Guided Contrastive Decoding Reduces Hallucination in Large Language Models: A Unified Framework of Fine-Grained Input Transformations (2026.findings-acl)

Challenge:	Existing Large Language Model (LLM)-based mobile agents follow explicit user instructions without personalized needs.
Approach:	They propose a user preference learning strategy enhanced with a Personal Reward Model to improve personalization performance.
Outcome:	The proposed agent achieves state-of-the-art performance while maintaining competitive instruction execution performance.

Copied to clipboard

Dongsheng Chen, Yingqi Zhu, Xingyue Zhang, Wenqing Zhou, Lei Li

LSEG: A Fine-tuning Free Method for NL2FOL via Logic-Structure and Entropy Guided Inference Controlling (2026.findings-acl)

Challenge:	Despite the remarkable generation capabilities of large language models, the issue of hallucination remains a critical challenge.
Approach:	They propose a contrastive decoding framework based on dynamic pointwise mutual information that disentangles spurious dependencies induced by context priors, lexical co-occurrences, and syntactic structures and prioritizes causal logic.
Outcome:	The proposed framework significantly improves the model’s factuality and reasoning robustness while maintaining high computational efficiency.

Copied to clipboard

Yuwei Huang, Shi Wang, Kangli Zi, Tianyu Luo, Jiang Zhixiao, Ruiheng Wang, Yufei Wang

Compact Example-Based Explanations for Language Models (2026.findings-acl)

Challenge:	Large language models struggle with natural language to first order logic (NL2FOL) translation due to logical hallucination.
Approach:	They propose a fine-tuning free framework to correct hidden state deviation by leveraging logical stability across logic preserving perturbations of the input.
Outcome:	The proposed framework improves logical consistency during inference and improves accuracy over baselines.

Copied to clipboard

Loris Schoenegger, Benjamin Roth

MedCPI: A Construct–Personalize–Integrate Framework for KG-enhanced Clinical Prediction (2026.findings-acl)

Challenge:	Existing training data influence estimation methods rely on naive selection strategies to provide explanations of a human-interpretable size.
Approach:	They propose a retraining-free metric that quantifies how useful a set of examples is for explaining a model's output.
Outcome:	The proposed model can predict whether a set of examples supports or undermines the model’s predictions.

Copied to clipboard

Hang Wang, Hang Dong, Lu Liu

Across Programming Language Silos: A Study on Cross-Lingual Retrieval-Augmented Code Generation (2026.findings-acl)

Challenge:	Existing KG-enhanced approaches to clinical prediction are limited . existing approaches to personalize and integrate knowledge are weakly controlled .
Approach:	They propose a framework to integrate medical knowledge graphs into EHRs to support KG-enhanced clinical prediction.
Outcome:	The proposed framework improves on MIMIC-III and MIMIC IV tasks.

Copied to clipboard

Qiming Zhu, Jialun Cao, Xuanang Chen, Weili Zhang, Yaojie Lu, Hongyu Lin, Xianpei Han, Le Sun, Shing-Chi Cheung

ErrorRadar: Benchmarking Complex Mathematical Reasoning of Multimodal Large Language Models Via Error Detection (2026.findings-acl)

Challenge:	Current research on large language models with retrieval-augmented code generation (RACG) has focused on single-language settings, leaving their cross-lingual effectiveness underexplored.
Approach:	They construct a dataset covering 13 PLs with nearly 14K instances to study cross-lingual code knowledge transfer in RACG.
Outcome:	The proposed model shows unequal cross-lingual knowledge transfer even with direct injection and shows limited reliance on natural language information embedded in code when equipped with a code-specific retriever.

Copied to clipboard

Yibo Yan, Shen Wang, Jiahao Huo, Hang Li, Boyan Li, Jiamin Su, Xiong Gao, YiFan Zhang, Tianlong Xu, Zhendong Chu, Aoxiao Zhong, Kun Wang, Hui Xiong, Philip S. Yu, Xuming Hu, Qingsong Wen

Disentangling Continued Pre-Training: Attention-Driven Routing and Semantic Hub Preservation in Language Adaptation (2026.findings-acl)

Challenge:	Current mathematical benchmarks focus on evaluating MLLMs’ problem-solving ability, yet there is a crucial gap in addressing more complex scenarios such as error detection.
Approach:	They propose to evaluate multimodal error detection by evaluating two sub-tasks error step identification and error categorization.
Outcome:	The proposed task evaluates MLLMs' ability to handle multimodal questions compared to text-only models.

Copied to clipboard

Khanh-Tung Tran, Vinh-Khanh Tran, Barry O’Sullivan, Hoang D. Nguyen

Do You Get the Hint? Benchmarking LLMs on the Board Game Concept (2026.findings-acl)

Challenge:	Continued Pre-Training (CPT) enables Large Language Models (LLMs) to acquire second-language capabilities, yet the mechanisms underlying CPT remain poorly understood.
Approach:	They investigate how CPT adapts model representations across diverse language families and scripts, model sizes, and architectures.
Outcome:	The proposed model can be surgically transferred between base and CPT models with minimal loss.

Copied to clipboard

Ine Gevers, Walter Daelemans

Challenge:	Large language models have achieved impressive progress on many benchmarks, yet they still have fundamental weaknesses.
Approach:	They introduce Concept, a word-guessing board game, as a benchmark for probing abductive reasoning.
Outcome:	The proposed game is easily solved by humans, but is still very challenging for state-of-the-art LLMs (no model exceeds 40% success rate).

Topic-Based Watermarks for Large Language Models (2026.findings-acl)

Copied to clipboard

Alexander Nemecek, Yuzhou Jiang, Erman Ayday

SParK-Eval: Evaluating Structure-Aware Knowledge Acquisition in LLMs for Domain Adaptation to Industrial Records (2026.findings-acl)

Challenge:	Existing watermarking methods often involve trade-offs between attack robustness, generation quality and additional overhead.
Approach:	They propose a topic-guided watermarking scheme that partitions the vocabulary into topic-aligned token subsets.
Outcome:	The proposed method achieves text quality comparable to industry-leading systems and improves watermark robustness against paraphrasing and lexical perturbation attacks with minimal performance overhead.

Copied to clipboard

Ekant Muljibhai Amin, Yuta Koreeda, Yasuhiro Sogawa

SOLAR-RL: Semi-Online Long-horizon Assignment Reinforcement Learning (2026.findings-acl)

Challenge:	Large Language Models (LLMs) often struggle in domain adaptation for industrial settings where available corpora are limited and structurally diverse.
Approach:	They propose a framework that constructs question–answer pairs from pretraining data and annotates each with its input structure.
Outcome:	The proposed framework can be used to analyze how input structure affects parametric knowledge acquisition during domain-adaptive pretraining.

Copied to clipboard

Jichao Wang, Liuyang Bian, Yufeng Zhou, Han Xiao, Yue Pan, Guozhi Wang, Hao Wang, Zhaoxiong Wang, Yafei Wen, Xiaoxin Chen, Shuai Ren, Lingfang Zeng

Kumatigi: Quality-Driven Data Augmentation for Low-Resource Machine Translation (2026.findings-acl)

Challenge:	Existing approaches to training GUI agents on dynamic tasks are based on SFT or Behavior Cloning.
Approach:	They propose a framework that integrates global trajectory insights directly into offline learning . they reconstruct diverse rollout candidates from static data and detect first failure point .
Outcome:	The proposed framework improves long-horizon task completion rates and robustness compared to baselines.

Copied to clipboard

Cheick Tidiani Cissé

CAMO: An Agentic Framework for Automated Causal Discovery from Micro Behaviors to Macro Emergence in LLM Agent Simulations (2026.findings-acl)

Challenge:	Neural machine translation for extremely low-resource languages faces compounding challenges: limited parallel data, orthographic inconsistency, and inconsistent metadata for principled training.
Approach:	They propose a quality-annotated French-Bambara corpus combining systematic curation with data augmentation strategies tailored to Bambaran.
Outcome:	The proposed framework achieves up to +3–4 BLEU over strong baselines.

Copied to clipboard

Xiangning Yu, Yuwei Guo, Yuqi Hou, Xiao Xue, Qun Ma

Challenge:	LLM-empowered agent simulations generate rich, adaptive, and often nonlinear interaction patterns.
Approach:	They propose an automated Causal discovery framework for LLM agent simulations that converts mechanistic hypotheses into computable factors and learns a compact causal representation centered on an emergent target.
Outcome:	Experiments across four emergent settings demonstrate the promise of CAMO.

MoLAN: A Unified Modality-Aware Noise Dynamic Editing Framework for Multimodal Sentiment Analysis (2026.findings-acl)

Copied to clipboard

Xingle Xu, YongKang Liu, Dexian Cai, Shi Feng, Xiaocui Yang, Daling Wang, Yifei Zhang

C2PO: Diagnosing and Disentangling Bias Shortcuts in LLMs (2026.findings-acl)

Challenge:	Existing approaches to multimodal sentiment analysis treat entire modality as an independent unit for feature enhancement or denoising, which often suppresses redundant noise at the cost of weakening critical information.
Approach:	They propose a ModaLity-aware noise dynAmic editiNg framework that performs modality-awful block partitioning by dividing features of each modality into multiple blocks.
Outcome:	Experiments on five models and four datasets show that MoLAN+ achieves the state-of-the-art performance.

Copied to clipboard

Xuan Feng, Bo An, Tianlong Gu, Liang Chang, Fengrui Hao, Peipeng Yu, Shuai Zhao

Deconstruct, Diagnose, and Deliberate: A Protocol-Adaptive Role-Specific Multi-Agent Framework for Fake News Detection (2026.findings-acl)

Challenge:	Existing approaches to solve large language models address stereotypical and structural biases in isolation . however, prior paradigms address these in isolation, often at the expense of exacerbating the other .
Approach:	They propose a framework to tackle latent spurious feature correlations within input that drive erroneous reasoning shortcuts.
Outcome:	The proposed framework mitigates stereotypical and structural biases while preserving robust general reasoning capabilities.

Copied to clipboard

Zhigan Zhang, Jie Sui

Position: Multimodal Large Language Models Can Significantly Advance Scientific Reasoning (2026.findings-acl)

Challenge:	Existing methods for fake news detection rely on monolithic verification methods . Existing approaches often yield ambiguous verdicts due to superficial processing .
Approach:	They propose a protocol-adaptive role-specific multi-agent framework that decomposes verification into factual, logical, and contextual dimensions.
Outcome:	The proposed framework outperforms baseline methods in both predictive accuracy and explanatory quality.

Copied to clipboard

Yibo Yan, Shen Wang, Jiahao Huo, Jingheng Ye, Zhendong Chu, Xuming Hu, Philip S. Yu, Carla P Gomes, Bart Selman, Qingsong Wen

E3-TIR: Enhanced Experience Exploitation for Tool-Integrated Reasoning (2026.findings-acl)

Challenge:	Current scientific reasoning models struggle with generalization across domains and fall short of multimodal perception.
Approach:	They propose to use multimodal large language models to integrate text, images, and other modalities to enhance scientific reasoning.
Outcome:	The proposed models can integrate text, images, and other modalities and improve reasoning across disciplines.

Copied to clipboard

Weiyang Guo, Zesheng Shi, Liye Zhao, Jiayuan Ma, Zeen Zhu, Junxian He, Min Zhang, Jing Li

EVA: Evolving Semantic Adversaries for Red-Teaming GUI Agents Against Environmental Injection Attacks (2026.findings-acl)

Challenge:	Existing training paradigms for Large Language Models (LLMs) suffer from inefficient exploration and mode degradation due to a lack of prior guidance, while SFT-then-RL is limited by high data costs and capability plateaus caused by low-entropy collapse.
Approach:	They propose an Enhanced Experience Exploitation paradigm that integrates expert prefixes, expert guided, and self-exploration to improve agent training.
Outcome:	The proposed model achieves a 6% performance improvement over traditional paradigms on tool-use tasks while requiring less than 10% of the synthetic data.

Copied to clipboard

Yijie Lu, Manman Zhao, Tianjie Ju, Zihe Yan, Xinbei Ma, Yuan Guo, Daizong Ding, Gongshen Liu, Zhuosheng Zhang

NEAT: Neuron-Based Early Exit for Large Reasoning Models (2026.findings-acl)

Challenge:	Existing methods for red-teaming face a trade-off between requiring target-specific knowledge and incurring prohibitive computational costs.
Approach:	They propose a framework that evolves payloads exclusively on the semantic dimension via a discovery-deployment pipeline.
Outcome:	Experiments show that EVA outperforms baselines in terms of attack success rate while evolving benign seeds into successful attacks within 1.18 to 1.71 iterations.

Copied to clipboard

Kang Liu, YongKang Liu, Xiaocui Yang, Peidong Wang, Wen Zhang, Shi Feng, Yifei Zhang, Daling Wang

From “Thinking” to “Justifying”: Aligning High-Stakes Explainability with Professional Communication Standards (2026.findings-acl)

Challenge:	Existing approaches to reduce overthinking require additional rollout computation or externally labeled datasets.
Approach:	They propose a Neuron-based Early reAsoning exiT framework that monitors neuron-level activation dynamics to enable training-free early exits.
Outcome:	The proposed framework reduces the amount of reasoning steps generated by LRMs while maintaining accuracy.

Copied to clipboard

Chen Qian, Yimeng Wang, Yu Chen, Lingfei Wu, Andreas Stathopoulos

EHRAG: Bridging Semantic Gaps in Lightweight GraphRAG via Hybrid Hypergraph Construction and Retrieval (2026.findings-acl)

Challenge:	Existing explanations for large language models (LLMs) need to be able to verify outputs.
Approach:	They propose a method that constrains output communication to present a conclusion before its structured justification.
Outcome:	The proposed approach achieves 83.9% accuracy and correctness over CoT.

Copied to clipboard

Yifan Song, Xingjian Tao, Zhicheng Yang, Yihong Luo, Jing Tang

PURE: Post-hoc Unlocking and REfinement for Discrete Diffusion Decoding (2026.findings-acl)

Challenge:	Existing lightweight approaches to retrieval-augmented generation fail to capture latent semantic connections between disjoint entities.
Approach:	They propose a lightweight RAG framework that constructs a hypergraph capturing both structure and semantic relationships using a hybrid structural-semantic retrieval mechanism.
Outcome:	EHRAG outperforms state-of-the-art methods on four datasets while maintaining zero token consumption.

Copied to clipboard

Yangryeol Park, Kunhui Lee, Hanback Choi, Cheoneum Park, Donghyeon Jeon, Inho Kang, Seung-Hoon Na

OSCR-Attack: One-Shot Character Level Attacks through Self-Optimizing Continuous Relaxation (2026.findings-acl)

Challenge:	Masked diffusion language models (MDLMs) are limited by a monotonic unmasking policy, where committed tokens cannot be revised.
Approach:	They propose a training-free inference algorithm for two-phase decoding that unlocks unstable regions through deterministic window masking and stochastic leftward relaxation.
Outcome:	The proposed algorithm significantly improves accuracy on reasoning benchmarks on GSM8K.

Copied to clipboard

Lingyi Kong, Zhuo Liu, Zhanghao Hu, Qilong Qiu, Yutao Yang, Jingjing Xue, Zheng Wang, Lin Gui, Feiping Nie

MTEB-NL and E5-NL: Embedding Benchmark and Models for Dutch (2026.findings-acl)

Challenge:	Character-level adversarial attacks preserve semantics but are costly and inefficient . generative LLMs are gaining popularity due to their uncertainty and vulnerability to textual adversarials .
Approach:	They propose an end-to-end framework that transforms discrete choices into continuous representations and a conflict resolution strategy that maps them back into discrete insertion operations.
Outcome:	The proposed framework improves ASR by 21.45% points and accelerates the attack by 3.66 times compared to baselines.

Copied to clipboard

Nikolay Banar, Ehsan Lotfi, Jens Van Nooten, Cristina Arhiliuc, Marija Kliocaite, Walter Daelemans

Perceptual Hallucination in Vision–Language Models: Definition, Analysis and Verification (2026.findings-acl)

Challenge:	Recent advances in embedding resources have led to a lack of representation of the Dutch language in multilingual resources.
Approach:	They introduce Massive Text Embedding Benchmark for Dutch (MTEB-NL) which includes existing Dutch datasets and newly created ones, covering a wide range of tasks.
Outcome:	The proposed models demonstrate strong performance across multiple tasks.

Copied to clipboard

Taewook Hwang, Inbum Heo, Sung Jun Lee, Sangkeun Jung

RATION: Entropy-Driven Task-Adaptive Visual Attention Allocation Framework for Multimodal Reasoning (2026.findings-acl)

Challenge:	Recent advances in large language models (LLMs) have dramatically improved text understanding and generation capabilities.
Approach:	They define perceptual hallucination as the phenomenon where VLMs generate information as if perceived, despite absent or damaged visual evidence.
Outcome:	The proposed model reduces hallucination exposure by 36% on average, with reductions of up to 88%.

Copied to clipboard

Xingle Xu, Fanheng Kong, Dexian Cai, Shi Feng, Xiaocui Yang, Daling Wang, Yifei Zhang

Multimodal In-context Learning for ASR of Low-resource Languages (2026.findings-acl)

Challenge:	Prior studies have focused on strengthening multimodal reasoning by improving representation alignment or increasing computation, but these methods do not characterize the differences in visual demands across tasks.
Approach:	They propose an entropy-driven task-adaptive visual attention allocation framework that uses visual attention entropic as a control signal to dynamically allocate attention according to task demands.
Outcome:	The proposed framework achieves consistent performance gains across diverse reasoning tasks, datasets, and models, providing a clear direction toward more reliable multimodal reasoning.

Copied to clipboard

Zhaolin Li, Jan Niehues

a1: Steep Test-time Scaling Law via Environment Augmented Generation (2026.findings-acl)

Challenge:	In-context learning with large language models addresses this limitation, but prior work focuses on high-resource languages covered during training and text-only settings.
Approach:	They propose to use multimodal ICL to learn unseen languages with multimodal learning to improve ASR in large language models.
Outcome:	The proposed model outperforms existing models on unseen languages with multimodal ICL (MICL) and cross-lingual transfer learning matches or outperformed models without using target-language data.

Copied to clipboard

Lingrui Mei, Shenghua Liu, Yiwei Wang, Baolong Bi, Yuyao Ge, Jun Wan, Yurong Wu, Xueqi Cheng

Look and Think: Efficient Multimodal Reasoning via Modality-Decoupled Compression (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have made remarkable advances in reasoning, yet continue to struggle with hallucinations, logical errors, and inability to self-correct during complex multi-step tasks.
Approach:	They propose a framework that enhances LLM reasoning through real-time environmental feedback validating each reasoning step, dynamic branch exploration for investigating alternative solution paths when faced with errors, and experience-based learning from successful reasoning trajectories.
Outcome:	The proposed model outperforms comparable models by 24.4 percentage points across benchmarks while outperforming comparable models.

Copied to clipboard

Xidi Cai, Junhao Zheng, Jingye Li, Boyuan Li, Shaowei Zhang, Qianli Ma

Topology-Enhanced Alignment for Large Language Models: Trajectory Topology Loss and Topological Preference Optimization (2026.findings-acl)

Challenge:	Multimodal large language models have strong performance on visual question answering benchmarks . however, their inference efficiency is severely constrained by the rapidly growing context .
Approach:	They propose a modality-decoupled compression method that enables efficient multimodal inference . they propose to evict visual tokens whenever visual grounding is unnecessary .
Outcome:	The proposed method reduces the average context length by up to 57% while maintaining comparable performance to the standard MLLM baseline.

Copied to clipboard

Yurui Pan, Ke Xu, Bo Peng

Inference-Time Scaling of Verification: Self-Evolving Deep Research Agents via Test-Time Rubric-Guided Verification (2026.findings-acl)

Challenge:	Existing algorithms for supervised fine-tuning and reinforcement learning from human feedback (RLHF) do not constrain how hidden states move from a user prompt to an answer.
Approach:	They propose a topology-enhanced alignment framework that regularizes these trajectories using 0-dimensional persistent homology.
Outcome:	The proposed framework regularizes semantic trajectory in hidden space using 0-dimensional persistent homology.

Copied to clipboard

Yuxuan Wan, Tianqing Fang, Zaitang LI, Yintong Huo, Wenxuan Wang, Haitao Mi, Dong Yu, Michael R. Lyu

Challenge:	Recent advances in Deep Research Agents (DRAs) are transforming automated knowledge discovery and problem-solving.
Approach:	They propose an inference-time scaling of verification wherein an agent self-improves at test time by evaluating its generated answers.
Outcome:	The proposed model outperforms vanilla agent-as-judge and LLM judge baselines by 12%–48% in meta-evaluation F1 score.

FinCARDS: Card-Based Analyst Reranking for Financial Document Question Answering (2026.findings-acl)

Copied to clipboard

Yixi Zhou, Fan Zhang, YU Chen, Haipeng Zhang, Preslav Nakov, Zhuohan Xie

A Unified Feature Mixture Framework for Joint Speech and Singing Deepfake Detection (2026.findings-acl)

Challenge:	Existing reranking frameworks optimize semantic relevance, leading to unstable rankings and opaque decisions on long documents.
Approach:	They propose a structured reranking framework that reframes financial evidence selection as constraint satisfaction under a finance-aware schema.
Outcome:	FINCARDS improves early-rank retrieval over lexical and LLM-based reranking baselines while reducing ranking variance.

Copied to clipboard

Aastha Sharma, Guangjing Wang

From Atomic to Complex tasks: Cross-Tasking Improves Zero-Shot Argument Generation and Retrieval (2026.findings-acl)

Challenge:	Existing methods for deepfake detection fail under speech-to-singing domain shift . a speech-retentive multi-domain fine-tuning strategy enables adaptation to singing .
Approach:	They propose a unified deepfake detector based on a multi-branch mixture-of-experts architecture that integrates three complementary feature views.
Outcome:	The proposed detector achieves 1.82% EER on CtrSVDD, compared to 37–62% for existing detectors . it can generalize to unseen generators and preserve strong speech performance .

Copied to clipboard

Lynda Tamine, Ahmed Rayane Kebir, Merveille Dona Codjo, Enzo Pasquies, Jose G Moreno

Sculpting the Vector Space: Towards Efficient Multi-Vector Visual Document Retrieval via Prune-then-Merge Framework (2026.findings-acl)

Challenge:	Argument retrieval and argument generation (AR) have attracted significant attention in recent years . argllms are a powerful tool for analyzing argument quality and extracting argument units .
Approach:	They propose that argument generation and argument retrieval could leverage cross-tasking atomic argument mining and argument quality assessment tasks even if there is no supervision.
Outcome:	The proposed framework outperforms base models in argument generation and retrieval tasks even without supervision.

Copied to clipboard

Yibo Yan, Mingdong Ou, Yi Cao, Xin Zou, Jiahao Huo, Shuliang Liu, James Kwok, Xuming Hu

ThinkLinker: From Low-Rank Interaction to Knowledge-Aware Verification for Multimodal Entity Linking (2026.findings-acl)

Challenge:	Visual Document Retrieval (VDR) is of importance in multimodal retrieval applications.
Approach:	They propose a two-stage pruning and merging frameworks that combine pruning and merge techniques to achieve higher compression rates.
Outcome:	The proposed framework outperforms existing methods on 29 visual document retrieval datasets.

Copied to clipboard

Yingyao Ma, Yuanyuan Zhou, Congyu Zhang, Yi Yuan, Jiasong Wu, Lotfi Senhadji, Huazhong Shu

Logit Arithmetic Elicits Long Reasoning Capabilities Without Training (2026.findings-acl)

Challenge:	Existing methods for multimodal entity linking rely on textual context for disambiguation . textual contextual information alone fails to resolve ambiguity, leading to unreliable disambiguations in weak contexts.
Approach:	They propose a two-stage multimodal entity linking framework called ThinkLinker . they propose fusion mechanism to model joint dependencies among features .
Outcome:	The proposed framework outperforms state-of-the-art models on public benchmark datasets.

Copied to clipboard

Yunxiang Zhang, Muhammad Khalifa, Lechen Zhang, Xin Liu, Ayoung Lee, Xinliang Frederick Zhang, Farima Fatahi Bayat, Lu Wang

Evolving Agentic Workflow Driven by Human-Agent Collaboration (2026.findings-acl)

Challenge:	Large reasoning models exhibit long chain-of-thought reasoning with complex strategies such as backtracking and self-verification, yet, these capabilities typically require resource-intensive post-training.
Approach:	They propose a decoding-time approach which transfers long chain-of-thought reasoning capabilities from a substantially smaller reasoning guider to a large non-reasoning target.
Outcome:	The proposed method improves performance over a model 21x smaller than the target model by 21.5% and 24.2% over the model.

Copied to clipboard

Yuxin Liu, Jinxuan Zhang, Yuezhang Peng, Hefeng Zhou, Xiangfeng Wang, Jiong Lou, Chentao Wu, Jie LI, Jingjing Qu, Chaochao Lu

Challenge:	Existing approaches to generate agentic workflows using large language models are limited by high manual design costs, inefficient agentic search, and poor dynamic adaptability to new tasks and human preferences.
Approach:	They propose an evolutionary framework for generating agentic workflows through human-agent collaboration using evolutionary algorithms that mutate and cross over their structures, prompts, and LLM backbones.
Outcome:	The proposed framework surpasses other automated baselines by 27.34% while achieving comparable performance to o1-preview at only one-fourth of the cost.

HintPilot: LLM-based Compiler Hint Synthesis for Code Optimization (2026.findings-acl)

Copied to clipboard

Hanyun Jiang, Peisen Yao, Kaiyue Li, Tingting Lin, Chengpeng Wang, Kui Ren

What Makes an LLM a Good Optimizer? A Trajectory Analysis of LLM-Guided Evolutionary Search (2026.findings-acl)

Challenge:	Existing methods to optimize source code rely on invasive transformations that can introduce semantic errors and miss fine-grained compiler-level optimization opportunities.
Approach:	They propose a method that bridges LLM-based reasoning with traditional compilers by synthesizing compiler hints.
Outcome:	HintPilot achieves 6.88x speedup over -Ofast while preserving program correctness.

Copied to clipboard

Xinhao Zhang, Xi Chen, François Portet, Maxime Peyrard

Token-level Inference-Time Alignment for Vision-Language Models (2026.findings-acl)

Challenge:	Recent work has demonstrated the promise of orchestrating large language models (LLMs) within evolutionary and agentic optimization systems.
Approach:	They present a large-scale study of LLM-guided evolutionary search . they find strong LLMs behave as local refiners, producing frequent improvements . weaker LLM optimizers exhibit large semantic drift, they say .
Outcome:	The results highlight the importance of trajectory analysis for understanding and improving LLM-based optimization systems.

Copied to clipboard

Kejia Chen, Junjun Zheng, Jiawen Zhang, Manxi Lin, Xiao Pan, Jiacong Hu, Jian Lou, Zunlei Feng, Mingli Song

STELLA: A Multimodal LLM for Protein Functional Annotation via Unified Sequence-Structure Encoding (2026.findings-acl)

Challenge:	Vision-Language Models (VLMs) often prioritize linguistic fluency over visual fidelity . despite widespread adoption, VLMs often exhibit a critical failure mode: hallucination .
Approach:	They propose a framework for Token-level Inference-Time Alignment that steers the decoding process without updating the base model parameters.
Outcome:	The proposed framework improves performance on 13 benchmarks across architectures . it boosts LLaVA-1.5-7B by 8.6% on MMVet and achieves a 74.0 MMStar score .

Copied to clipboard

Hongwang Xiao, Wenjun Lin, Xi Chen, Hui Wang, Kai Chen, Jiashan Li, Yuancheng Sun, Sicheng Dai, Boya Wu, Qiwei Ye

RealFin: How Well Do LLMs Reason About Finance When Users Leave Things Unsaid? (2026.findings-acl)

Challenge:	a multimodal protein language model (LLM) integrates sequence, structure, and function into functional annotation.
Approach:	They propose a multimodal protein language model that synergistically aligns bimodal representations with the textual modality to advance protein functional annotation.
Outcome:	The proposed model synergizes bimodal representations with the textual modality to advance protein functional annotation.

Copied to clipboard

Yuyang Dai, Yan Lin, Zhuohan Xie, Yuxia Wang

Preference Optimization for Review Question Generation Improves Writing Quality (2026.findings-acl)

Challenge:	General-purpose models tend to over-commit and guess, while most finance-specialized models fail to clearly identify missing premises.
Approach:	They propose a bilingual benchmark that removes premises from exam-style questions while keeping them linguistically plausible.
Outcome:	The proposed model overcommits and guesses while most finance-specialized models fail to clearly identify missing premises.

Copied to clipboard

Karun Sharma, Vidushee Vats, Shengzhi LI, Yuxiang Wang, Zhongtian Sun, Prayag Tiwari

Your LLM Agent Can Leak Your Data: Data Exfiltration via Backdoored Tool Use (2026.findings-acl)

Challenge:	Peer reviewers are overloaded and face tight deadlines, leading some to rely on large language models (LLMs) to draft questions and comments.
Approach:	They use open-review review datasets to train a human preference model based on human reviewer questions . human evaluations show IntelliAsk generates more grounded, substantive and effortful questions than strong baselines .
Outcome:	The proposed model predicts reviewer-question quality better than API-based SFT baselines and provides scalable evaluation.

Copied to clipboard

Wuyang Zhang, Shichao Pei

EvoHyper: Evolving Hypergraph Topologies for Unified Collaboration in Multi-Agent Communication (2026.findings-acl)

Challenge:	Existing studies have not examined how backdoored agents can influence tool-use sequences to perform harmful actions.
Approach:	They propose a backdoor attack framework that embeds semantic triggers into fine-tuned LLM agents.
Outcome:	The proposed framework embeds semantic triggers into fine-tuned LLM agents . when triggered, the backdoored agent invokes memory-access tool calls to retrieve stored user context and exfiltrates it via disguised retrieval tool calls.

Copied to clipboard

Heng Zhang, Yihao Zhong, Lubin Gan, Zhihe Chen, Jiajun Wu, Yuling Shi, Xiaodong Gu, Hao Zhang, Haochen You, Jin Huang

Challenge:	Existing methods for multi-agent collaboration use a fixed communication graph and manage collaboration structure and shared memory in separate modules.
Approach:	They propose a framework that uses an evolving hypergraph topology for multi-agent collaboration.
Outcome:	The proposed framework achieves 3.2% to 7.8% accuracy gains over state-of-the-art methods and efficient, reducing token consumption by up to 23.5%.

Think Before you Write: QA-Guided Reasoning for Character Descriptions in Books (2026.findings-acl)

Copied to clipboard

Argyrios Papoudakis, Mirella Lapata, Frank Keller

GroupRank: A Groupwise Paradigm for Effective and Efficient Passage Reranking with LLMs (2026.findings-acl)

Challenge:	Character description generation is an important capability for narrative-focused applications . however, generating accurate character descriptions from long-form narratives is challenging . enabling built-in reasoning mode of current LLMs often degrades performance .
Approach:	They propose a framework that decouples reasoning from generation by generating a structured QA reasoning trace and a generation model that conditions on this trace to produce the final character description.
Outcome:	The proposed framework improves faithfulness, informativeness, and grounding over long-context baselines.

Copied to clipboard

Meixiu Long, Duolin Sun, Dan Yang, Yihan Jiao, Lei Liu, Jiahai Wang, Binbin Hu, Yue Shen, Jie Feng, Zhehao Tan, Junjie Wang, Lianzhen Zhong, Jian Wang, Peng Wei, Jinjie Gu

Mirage: A Diagnostic Framework for Evaluating the Realism of Synthetic Contact Center Dialogue Generation (2026.findings-acl)

Challenge:	Existing rankers excel in lexical-matching scenarios, while they struggle with complex queries requiring deep reasoning.
Approach:	They propose a new paradigm that balances flexibility and context awareness to unlock the full potential of groupwise reranking.
Outcome:	The proposed approach achieves a state-of-the-art 65.2 NDCG@10 on BRIGHT and surpasses baselines by 2.1 points on R2MED while delivering a 6.4 inference speedup.

Copied to clipboard

Rishikesh Devanathan, Varun Nathan, Ayush Kumar

BloomEval: A Bloom’s Cognitive Taxonomy-Based Benchmark for Evaluating LRMs via Cognitive Hierarchy Trace (2026.findings-acl)

Challenge:	capturing the full realism of real agent–customer interactions remains a challenge . privacy constraints and data scarcity limit the availability of real conversations .
Approach:	They propose a diagnostic evaluation framework for synthetic dialogue generation . they benchmarked strategies guided by structured supervision on call attributes .
Outcome:	The proposed framework outperforms synthetic transcripts on quality assurance tasks . it shows that current generation strategies lack sentiment fidelity, disfluency modeling, behavioral variation and conversational realism .

Copied to clipboard

Zhiyi Duan, Lei Gao, Jiangshan Guan, Qi Wang, Rui Liu

CAIR: Causal Adaptive Information-based Reinforcement Learning for Multimodal Emotion Reasoning (2026.findings-acl)

Challenge:	Existing benchmarks for Large Reasoning Models rely on answer correctness, but fail to assess the structural coherence and cognitive soundness of the reasoning process itself.
Approach:	They propose a framework that maps a model's reasoning trajectory onto hierarchical cognitive levels and an annotation pipeline to ensure a scalable yet reliable annotation pipeline.
Outcome:	The proposed framework detects hierarchy jumps, breaks, and overthinking errors and enables scalable yet reliable annotation.

Copied to clipboard

Fengyu Zhang, Bin Liu, Jianhua Tao, Zhuofan Wen, Shun Chen, Hailiang Yao, Zhengqi Wen

A Multi-View Media Profiling Suite: Resources, Evaluation, and Analysis (2026.findings-acl)

Challenge:	Existing methods for multimodal emotion reasoning produce fluent but superficial explanations that lack authentic logical derivation.
Approach:	They propose a framework that treats rationales as causal mediators between raw perceptual signals and emotional semantics and an adaptive optimization mechanism to balance perception and reasoning across varying cognitive loads.
Outcome:	The proposed framework outperforms specialized SFT models by 14.4% while enhancing rationale faithfulness.

Copied to clipboard

Muhammad Arslan Manzoor, Dilshod Azizov, Daniil Orel, Umer Siddique, Zain Muhammad Mujahid, Yufang Hou, Preslav Nakov

Mechanistic Interpretability of Text-to-Image Diffusion Models via Cross-Attention Interventions (2026.findings-acl)

Challenge:	a large-scale label set for media outlets from Media Bias/Fact Check (MBFC) is lacking in the field.
Approach:	They propose to use a large-scale label set to analyze outlets' representations . they also propose to evaluate embedding views and fusion strategies .
Outcome:	The proposed method achieves state-of-the-art results on ACL-2020 and establishes strong benchmarks on MBFC-2025.

Copied to clipboard

Maisha Maliha, Dean F. Hougen

Revisiting Entropy in Reinforcement Learning for Large Reasoning Models (2026.findings-acl)

Challenge:	Text-to-image diffusion models generate high quality images through iterative denoising, but their internal mechanisms for grounding prompt semantics into visual structure remain unclear.
Approach:	They propose a mechanistic interpretability framework that probes how individual prompt tokens are represented and utilized during the denoising process.
Outcome:	The proposed framework enables module-wise and head-wise attribution of semantic changes across denoising timesteps.

Copied to clipboard

Renren Jin, Pengzhi Gao, Yuqi Ren, Zhuowen Han, Tongxuan Zhang, Wuwei Huang, Wei Liu, Jian Luan, Deyi Xiong

How Large Language Models Balance Internal Knowledge with User and Document Assertions (2026.findings-acl)

Challenge:	Reinforcement learning with verifiable rewards (RLVR) has emerged as a paradigm for enhancing the reasoning capabilities of large language models.
Approach:	They propose a positive-advantage reweighting approach that regulates model entropy by adjusting the loss weights assigned to tokens with positive advantages during RLVR training.
Outcome:	The proposed approach regulates model entropy by adjusting loss weights assigned to tokens with positive advantages during RLVR training while maintaining competitive performance.

Copied to clipboard

Shuowei Li, Haoxin Li, Wenda Chu, Yi Fang

ComicVQA: A Benchmark for Visual Reasoning in Multimodal LLMs (2026.findings-acl)

Challenge:	Large language models often need to balance their internal parametric knowledge with external information, such as user beliefs and content from retrieved documents, in real-world scenarios like RAG or chat-based systems.
Approach:	They propose a three-source interaction framework to evaluate 27 large language models from 3 families on 2 datasets.
Outcome:	The proposed framework systematically evaluates 27 large language models from 3 families on 2 datasets.

Copied to clipboard

Esther Gan, Hannah Brown, David Herel, Kenji Kawaguchi, Min-Yen Kan, Michael Qizhe Shieh

AJ-Bench: Benchmarking Agent-as-a-Judge for Environment-Aware Evaluation (2026.findings-acl)

Challenge:	ComicVQA is a visual reasoning benchmark for comics.
Approach:	They propose a comics-based benchmark for evaluating MLLMs on visual reasoning.
Outcome:	The proposed model achieves 62.6% accuracy on Missing Panel Prediction and 46.4% on Panel Sorting, compared to open-source models.

Copied to clipboard

Wentao Shi, Yu Wang, Yuyang Zhao, Yuxin Chen, Fuli Feng, Xueyuan Hao, Xi Su, Qi GU, Hui Su, Xunliang Cai, Xiangnan He

Challenge:	Existing approaches to verify agent behaviors in complex environments rely on rule-based verifiers or LLM-as-a-Judge models.
Approach:	They propose a benchmark to evaluate Agent-as-a-Judge across three domains . the benchmark covers search, data systems, and graphical user interfaces - with 155 tasks and 516 trajectories .
Outcome:	The proposed benchmark outperforms existing benchmarks in search, data systems, and GUI domains while revealing open challenges in agent-based verification.

Can LLMs Estimate Student Struggles? Human-AI Difficulty Alignment with Proficiency Simulation for Item Difficulty Prediction (2026.findings-acl)

Copied to clipboard

Ming Li, Han Chen, Yunze Xiao, Jian Chen, Hong Jiao, Tianyi Zhou

Cognitive Modeling

Challenge:	Accurate estimation of item (question or task) difficulty suffers from the cold start problem.
Approach:	They propose to use large-scale empirical analysis to examine human-AI Difficulty Alignment . they find that models struggle to simulate the capability limitations of students .
Outcome:	The proposed model size is not reliably helpful for human-AI alignment . high performance often impedes accurate difficulty estimation, the authors say .

Transition-Matrix Regularization for Next Dialogue Act Prediction in Counselling Conversations (2026.findings-acl)

Copied to clipboard

Eric Rudolph, Philipp Steigerwald, Jens Albrecht

ClimateCause: Complex and Implicit Causal Structures in Climate Reports (2026.findings-acl)

Challenge:	Evaluated on a 60-class German counselling taxonomy, this improves macro-F1 by 9–42% relative depending on encoder and corpus-derived transition patterns.
Approach:	They propose to use a KL regularization term to align next dialogue act distributions with corpus-derived transition patterns to improve macro-F1 by 9–42% relative to encoders.
Outcome:	The proposed term improves macro-F1 by 9–42% relative to encoders and significantly improves dialogue-flow alignment.

Copied to clipboard

Liesbeth Allein, Nataly Pineda-Castañeda, Andrea Rocci, Marie-Francine Moens

Reference-Free Evaluation of Taxonomies (2026.findings-acl)

Challenge:	Existing datasets for causal discovery from text lack granularity and abstraction for domains characterized by such complex causality.
Approach:	They propose a manually expert-annotated dataset of higher-order causal structures from science-for-policy climate reports.
Outcome:	The proposed dataset is highly readable and can be used to quantify readability.

Copied to clipboard

Pascal Wullschleger, Majid Zarharan, Donnacha Daly, Marc Pouly, Jennifer Foster

Challenge:	Taxonomies are used to classify items, ideas or organisms based on shared characteristics.
Approach:	They introduce two reference-free metrics for quality evaluation of taxonomies in the absence of labels.
Outcome:	The proposed metrics correlate well with F1 against ground truth taxonomies on five taxonomies and improve hierarchical classification when used with label hierarchies.

CSI: An Investigative Multi-Agent Framework for Explainable Short Video Fake News Detection (2026.findings-acl)

Copied to clipboard

Yuxin Wang, Yang Yang, Huaiwen Zhang

Dial HEALTHDIAL for Advice: A Multilingual and Multi-Parallel Spoken Dialogue Dataset for Knowledge-Grounded Information Seeking (2026.findings-acl)

Challenge:	Existing methods for short video fake news detection rely on black-box MSLMs with poor explainability and superficial understanding or on specific prompt strategies for Multimodal Large Language Models (MLLMs)
Approach:	They propose a multi-agent framework called CSI for short video fake news detection.
Outcome:	The proposed framework provides rigorous explanations while achieving state-of-the-art performance on two real-world datasets.

Copied to clipboard

Songbo Hu, Yinhong Liu, Ej Zhou, Evgeniia Razumovskaia, Xiaobin Wang, Alexander Fraser, Ivan Vulić, Anna Korhonen

Fast and Effective On-Policy Distillation from Reasoning Prefixes (2026.findings-acl)

Challenge:	Creating spoken dialogue datasets is methodologically challenging due to the personally identifiable nature of speech signals.
Approach:	They propose a large-scale, multilingual, and multi-parallel dataset for developing and evaluating retrieval-augmented generation-based spoken dialogue systems.
Outcome:	The proposed dataset includes 6,000 information-seeking dialogues and 163 hours of user speech recorded from native speakers of four official WHO languages.

Copied to clipboard

Dongxu Zhang, Zhichao Yang, Sepehr Janghorbani, Jun Han, Andrew Ressler II, Qian Qian, Gregory D Lyng, Sanjit Singh Batra, Robert E. Tillman

MaRF: Leveraging Representation-Level Fusion of Formula Semantics for Mathematical Information Retrieval (2026.findings-acl)

Challenge:	On-policy distillation (OPD) requires expensive on-the-fly sampling of the student policy during training, which substantially increases training cost.
Approach:	They propose to use on-policy distillation to sample trajectories from student model . they propose to terminate the sampling early during distillation .
Outcome:	The proposed method matches the performance of full OPD in long reasoning outputs while reducing training FLOP by 2x–40x.

Copied to clipboard

Suyuan Wang, Hongbo Zheng, Nickvash Kani

From Documents to Segments: A Contextual Reformulation for Topic Assignment (2026.findings-acl)

Challenge:	Mathematical information retrieval (MIR) relies on combining textual content with mathematical expressions.
Approach:	They propose a dual-encoder representation-level fusion framework for MIR that integrates formula semantics into context-aware dense retrieval.
Outcome:	The proposed framework outperforms baselines on the ARQMath-3 benchmark.

Copied to clipboard

Hoonsang Yoon, Takyoung Kim, Wonkee Lee, Ilmin Cho, Dilek Hakkani-Tür, Stanley Jungkyu Choi

Can LLMs Compress (and Decompress)? Evaluating Code Understanding and Execution via Invertibility (2026.findings-acl)

Challenge:	Traditional topic modeling treats each document as a single, coherent unit of topic.
Approach:	They propose a paradigm that redefines topic assignment at the level of segments . they propose 'segment intrusion task' to extend word intrusion to the span level .
Outcome:	The proposed paradigm improves topic purity, interpretability and applicability to multi-theme corpora.

Copied to clipboard

Nickil Maveli, Antonio Vergari, Shay B Cohen

Confidence Estimation for LLMs in Multi-turn Interactions (2026.findings-acl)

Challenge:	a recent development of code-LLMs has demonstrated remarkable performance across various software engineering applications.
Approach:	They propose a round-trip code execution reasoning task to test round- trip consistency . they use zero-shot prompting, supervised fine-tuning on execution traces and self-reflection mechanisms to evaluate models .
Outcome:	The proposed benchmarks show that LLMs struggle with round-trip consistency . the benchmarks lack the internal coherence required for trustworthy code reasoning .

Copied to clipboard

Caiqi Zhang, Ruihan Yang, Xiaochen Zhu, Chengzu Li, Tiancheng Hu, Yijiang River Dong, Deqing Yang, Nigel Collier

Beyond Black-Box Labels: Interpretable Criteria for Diagnosing Subjective NLP Tasks (2026.findings-acl)

Challenge:	Despite recent progress, most prior work studies confidence in single-turn question answering.
Approach:	They propose a logit-based probe that measures confidence in multi-turn dialogues . they propose 'infoECE' and a "hinter-guesser" paradigm for generating controlled evaluations based on data .
Outcome:	The proposed framework is grounded in calibration and monotonicity of confidence as more information becomes available.

Copied to clipboard

Nisrine Rair, Alban Goupil, Valeriu Vrabie, Emmanuel Chochoy

AutoRubric: Rubric-Based Generative Rewards for Faithful Multimodal Reasoning (2026.findings-acl)

Challenge:	Existing approaches to assess annotator judgments aggregate disagreement into a single gold label . et al., 2022) show disagreement is diffuse, but standard approaches are not rigorous .
Approach:	They propose a schema-level diagnostic for auditing expert-designed annotation schemas prior to gold-label commitment . they find disagreement is not diffuse: instability concentrates in a few criteria, while nearly half of covered sentences activate multiple categories.
Outcome:	The proposed diagnostic separates unstable criteria with hard-to-operationalize boundaries and systematic overlap that blurs the boundaries between mutually exclusive categories.

Copied to clipboard

Mengzhao Jia, Zhihan Zhang, Ignacio Cases, Zheyuan Liu, Meng Jiang, Peng Qi

HER: Human-like Reasoning and Reinforcement Learning for LLM Role-playing (2026.findings-acl)

Challenge:	Multimodal large language models (MLLMs) have advanced from perception tasks to complex multi-step reasoning.
Approach:	They propose a framework that integrates reinforcement learning with verifiable rewards with process-level supervision through automatically collected rubric-based generative rewards.
Outcome:	The proposed framework achieves state-of-the-art performance on six multimodal reasoning benchmarks and significantly improves reasoning faithfulness in dedicated evaluations.

Copied to clipboard

Chengyu Du, Xintao Wang, Aili Chen, Weiyuan Li, Rui Xu, Junteng Liu, Zishan Huang, Rong Tian, Zijun Sun, Yuhao Li, Liheng Feng, Deming Ding, Pengyu Zhao, Yanghua Xiao

Do Generalisation Results Generalise? (2026.findings-acl)

Challenge:	Existing models for LLM role-playing lack high-quality datasets with explicit reasoning traces and reliable reward signals aligned with human preferences.
Approach:	They propose a unified framework for cognitive-level persona simulation that strictly distinguishes characters’ first-person thinking processes from LLMs’ third-person reasoning.
Outcome:	The proposed framework outperforms the Qwen3-32B baseline model and achieves a 30.26% and 14.97% performance on the minimax benchmarks.

Copied to clipboard

Matteo Boglioni, Andrea Sgobbi, Gabriel Tavernini, Francesco Rita, Marius Mosbach, Tiago Pimentel

Challenge:	Existing studies evaluating generalisation performance on large language models focuses on a single out-of-distribution dataset .
Approach:	They examine whether OOD generalisation results generalise across multiple OOD testsets throughout a finetuning run and then evaluate the partial correlation of results .
Outcome:	The proposed model achieves high scores on multiple OOD testsets, regressing out in-domain performance.

Do LLMs Overthink Basic Math Reasoning? Benchmarking the Accuracy-Efficiency Tradeoff in Language Models (2026.findings-acl)

Copied to clipboard

Gaurav Srivastava, Aafiya Shamshad Hussain, Sriram Srinivasan, Xuan Wang

PV-SQL: Synergizing Database Probing and Rule-based Verification for Text-to-SQL Agents (2026.findings-acl)

Challenge:	Large language models (LLMs) achieve impressive performance on complex benchmarks yet sometimes fail on basic math reasoning.
Approach:	They propose a benchmark to evaluate the efficiency of reasoning in large language models . they formalize the accuracy-verbosity tradeoff and introduce the overthinking score .
Outcome:	The proposed model performs well on complex benchmarks but fails on basic math reasoning . the proposed model generates 18 more tokens while achieving lower accuracy .

Copied to clipboard

Yuan Tian, Tianyi Zhang

Anatomy of Unlearning: The Dual Impact of Fact Salience and Model Fine-Tuning (2026.findings-acl)

Challenge:	Existing text-to-SQL systems struggle with deep contextual understanding .
Approach:	They propose a framework that provides a tool to help query databases with deeper contextual understanding . they propose two components that iteratively generate probing queries and verify queries .
Outcome:	Experiments show PV-SQL outperforms the best text-to-SqL baseline by 5% execution accuracy and 20.8% valid efficiency score while consuming fewer tokens.

Copied to clipboard

Anna Borisiuk, Andrey Savchenko, Alexander Panchenko, Elena Tutubalina

What Do Neural Speech Models Know About Phonology? Evidence from Structured Phoneme Confusions (2026.findings-acl)

Challenge:	Existing studies assume that all facts are equally forgettable . popular facts, frequent and widely distributed, may be more deeply embedded than rare ones, making them harder to erase.
Approach:	They propose a benchmark to evaluate how unlearning differs between pretrained and supervised fine-tuned models when fact popularity is taken into account.
Outcome:	The proposed model is compared with pretrained and SFT models on the forget data and shows that it performs better on both models.

Copied to clipboard

Eli Stafford, Aimée Lahaussois, Guillaume Wisniewski

Rerank Before You Reason: Analyzing Reranking Tradeoffs through Effective Token Cost in Deep Search Agents (2026.findings-acl)

Challenge:	acoustic and phonological models of speech recognition are often limited to the phoneme level . a recent study has shown that phoneme confusions are strongly structured in phonology space .
Approach:	They adopt a featural representation of phonemes grounded in phonological theory which models speech sounds as structured bundles of distinctive articulatory and acoustic properties.
Outcome:	The proposed model allows us to analyse phoneme confusions at a finer granularity and to investigate whether certain phonological features are more vulnerable than others.

Copied to clipboard

Sahel Sharifymoghaddam, Jimmy Lin

Position: LLM Watermarking Should Align Stakeholders’ Incentives for Practical Adoption (2026.findings-acl)

Challenge:	Recent work emphasizes improving efficiency in LLM-based systems, especially for longcontext and multi-step reasoning.
Approach:	They analyze the role of listwise reranking in deep search pipelines and compare their results to a novel ETC metric to determine model scale and reasoning effort.
Outcome:	The proposed model scale, reasoning effort, reranking depth, and total token cost (ETC) metric improve retrieval and end-to-end accuracy and moderate reranked agents achieve comparable accuracy at substantially lower cost.

Copied to clipboard

Yepeng Liu, Xuandong Zhao, Dawn Song, Gregory W. Wornell, Yuheng Bu

Learning to Refine: Self-Refinement of Parallel Reasoning in LLMs (2026.findings-acl)

Challenge:	despite advances in watermarking algorithms, real-world deployment remains limited . model watermarks can be used to protect intellectual property and promote trust in AI .
Approach:	They propose to revisit three classes of watermarking to examine incentives for large language models . model watermarks naturally aligns with LLM provider interests, yet faces new challenges in open-source ecosystems .
Outcome:	The proposed methods can be used in dataset decontamination, user-controlled provenance, and in-context watermarking.

Copied to clipboard

Qibin Wang, Pu Zhao, Shaohan Huang, Fangkai Yang, Lu Wang, Furu Wei, Qingwei Lin, Saravan Rajmohan, Dongmei Zhang

AUDITA: A New Dataset to Audit Humans vs. AI Skill at Audio QA (2026.findings-acl)

Challenge:	Existing approaches to test-time scaling are limited due to the quality of candidate responses.
Approach:	They propose a new metric to quantify the relative improvement of self-refinement beyond majority voting.
Outcome:	The proposed method achieves state-of-the-art performance across five benchmarks over other methods.

Copied to clipboard

Tasnim Kabir, Dmytro Kurdydyk, Aadi Palnitkar, Liam Dorn, Ahmed Haj Ahmed, Jordan Lee Boyd-Graber

NeoAraBERT: A Modern Foundation Model for Arabic Embeddings with Diacritics-Aware Tokenization and POS-Targeted Masking (2026.findings-acl)

Challenge:	Existing audio question answering benchmarks emphasize sound event classification or caption-grounded queries.
Approach:	They propose a large-scale, real-world audio question answering benchmark to evaluate audio reasoning beyond surface-level acoustic recognition.
Outcome:	The proposed model achieves 32.13% accuracy while demonstrating comprehension of audio . state-of-the-art models perform poorly, with average accuracy below 8.86%.

Copied to clipboard

Chadi Abou Chakra, Hadi Khaled Hamoud, Osama Rakan Al Mraikhat, Qusai Abu Obaida, Mohamad Ballout, Fadi Zaraket

Steering Away from Refusal: A Black-box Jailbreak Method Based on First-Token Distribution (2026.findings-acl)

Challenge:	NeoAraBERT is an open-source text-embedding model for Arabic.
Approach:	They propose to train Arabic text-embedding models on open-source datasets . they benchmarked NeoAraBERT against five top-performing Arabic models on 23 tasks .
Outcome:	The proposed model outperforms five other models on 23 tasks in Arabic . it shows substantial improvement on classical and modern standard Arabic compared to other models .

Copied to clipboard

Shuangjie Fu, Du Su, Xin Chen, Fei Sun, Huawei Shen, Xueqi Cheng

Mina: A Multilingual LLM-Powered Legal Assistant Agent for Empowering Access to Justice in Bangladesh (2026.findings-acl)

Challenge:	Existing methods to analyze black-box jailbreaks lack direct optimization signals to refine adversarial prompts.
Approach:	They propose a distribution-jailbreak attack method that selects effective jailbreak templates and iteratively optimizes adversarial suffixes by maximizing the KL divergence from the standard refusal distribution.
Outcome:	The proposed method achieves state-of-the-art Attack Success Rate (ASR) on all tested open-source models and delivers over 94% ASR on GPT-4.1.

Copied to clipboard

Azmine Toushik Wasi, Wahid Faisal, Mst Rafia Islam, Md Rizwan Parvez

Zero-Shot Context-Aware ASR for Diverse Arabic Varieties (2026.findings-acl)

Challenge:	Existing AI legal assistants lack Bengali-language support and jurisdiction-specific adaptation, limiting their effectiveness.
Approach:	They developed a multilingual LLM-based legal assistant tailored for the Bangladeshi context that employs multilingual embeddings and a RAG-based chain-of-tools framework for retrieval, reasoning, translation, and document generation.
Outcome:	Evaluated by law faculty from leading Bangladeshi universities across all stages of the 2022 and 2023 Bangladesh Bar Council examinations, Mina achieved scores of 75–80% in preliminary MCQs, written, and simulated viva voce components.

Copied to clipboard

Bashar Talafha, Amin Abu Alhassan, Muhammad Abdul-Mageed

The Sonar Moment: An Audio Geo-Localization Benchmark for Audio-Language Models (2026.findings-acl)

Challenge:	Large-scale multilingual ASR has substantially improved recognition for high-resource languages.
Approach:	They propose a proxy-guided -best selection paradigm that conditions inference on external side information without parameter updates.
Outcome:	The proposed model reduces WER by 15.6% relative and recovers a fraction of oracle n-best gains on the common voice MSA testbed.

Copied to clipboard

Ruixing Zhang, Zihan Liu, Leilei Sun, Tongyu Zhu, Weifeng Lv

Towards Understanding the Robustness of Sparse Autoencoders (2026.findings-acl)

Challenge:	AGL1K is the first audio geo-localization benchmark for audio language models (ALMs) it is based on a crowd-sourced platform and is available in 72 countries and territories.
Approach:	They propose a benchmark for audio geo-localization that quantifies the informativeness of each recording and a metric that quantizes the information of each audio clip.
Outcome:	The proposed benchmarks cover 72 countries and territories and can be used to improve audio geo-localization.

Copied to clipboard

Ahson Saiyed, Sabrina Sadiekh, Chirag Agarwal

Beyond Local vs. External: A Game-Theoretic Framework for Trustworthy Knowledge Acquisition (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are vulnerable to optimization-based jailbreak attacks that exploit internal gradient structure.
Approach:	They propose to integrate pretrained Sparse Autoencoders into transformer residual streams at inference time without modifying model weights or blocking gradients.
Outcome:	The proposed model reduces jailbreak success rate by 5x compared to baseline models . compared with models with weak white-box attacks, the proposed model is more robust .

Copied to clipboard

Rujing Yao, Yufei Shi, Yang Wu, Ang Li, Zhuoren Jiang, XiaoFeng Wang, Haixu Tang, Xiaozhong Liu

False Sense of Security: Why Probing-based Malicious Input Detection Fails to Generalize (2026.findings-acl)

Challenge:	Cloud-hosted Large Language Models (LLMs) offer unmatched reasoning capabilities and dynamic knowledge, yet submitting raw queries to these services can expose sensitive user intent.
Approach:	They propose a framework that formulates the trade-off between knowledge utility and privacy as a strategic game.
Outcome:	The proposed framework reduces intent leakage while maintaining high-fidelity answer quality.

Copied to clipboard

Cheng Wang, Zeming Wei, Qin Liu, Wenxuan Zhou, Muhao Chen

Do Transformers Grok Succinct Algorithms? Mechanistic Evidence for Counting Circuits (2026.findings-acl)

Challenge:	Recent work has leveraged probing-based approaches to study the separability of malicious and benign inputs in Large Language Models’ internal representations.
Approach:	They propose to use probing-based methods to study separability of malicious and benign inputs in LLMs' internal representations to detect harmful and benign content.
Outcome:	The proposed methods show that they learn superficial patterns rather than semantic harmfulness.

Copied to clipboard

Yizheng Zhao

How Value Induction Reshapes LLM Behavior (2026.findings-acl)

Challenge:	Recent studies suggest that Transformers are inherently succinct, capable of representing recursive algorithms like binary counting over exponential state spaces.
Approach:	They propose to bridge this gap by testing the Succinctness Hypothesis using mechanistic interpretability on a large-scale computation task.
Outcome:	The proposed model can represent recursive algorithms over exponential state spaces . the proposed model is able to generalize perfectly, whereas massive LSTM baselines fail completely.

Copied to clipboard

Arnav Arora, Natalie Schluter, Katherine Metcalf, Maartje Ter Hoeve

Challenge:	Induction of values into LLMs can have unintended effects on the user interacting with it.
Approach:	They investigate the unintended effects of value incorporation into models by fine-tuning existing preference datasets and measuring their effect on safety, anthropomorphism and QA benchmarks.
Outcome:	The proposed model improves safety, anthropomorphism and QA benchmarks by inducing values and incorporating values into the model.

Stable Evidence, Unstable Decisions: An Empirical Analysis of Model Decision Stability in Vision–Language Models (2026.findings-acl)

Copied to clipboard

Ali Khoramfar, Mohammad Javad Dousti, Alireza Mohamadian, Heshaam Faili

PlanRAG-Audio: Planning and Retrieval Augmented Generation for Long-form Audio Understanding (2026.findings-acl)

Challenge:	Large language models (LLMs) provide visual information alongside their predictions, but it remains unclear whether consistency in such information implies consistent decisions.
Approach:	They construct configurations that retain lesion content while varying surrounding context and scale and measure decision flips together with consistency in model-reported influential slices.
Outcome:	The proposed models show that flip rates reach up to 75% across lesion-containing presentations, despite high overlap in reported evidence.

Copied to clipboard

Masao Someki, Chien-yu Huang, Siddhant Arora, Samuele Cornell, Markus Müller, Nathan Susanj, Rupak Vignesh Swaminathan, Grant Strimel, Jing Liu, Shinji Watanabe

VecCISC: Improving Confidence-Informed Self-Consistency with Reasoning Trace Clustering and Candidate Answer Selection (2026.findings-acl)

Challenge:	Long-form audio understanding poses significant challenges due to the extreme length of audio sequences and the need to reason over heterogeneous acoustic cues distributed over time.
Approach:	They propose a retrieval-augmented generation framework for scalable long-form audio understanding . planRAG-Audio explicitly plans which modalities and temporal spans are required for a given query .
Outcome:	Experiments show that planRAG-Audio reduces the length of inputs for long-form audio models . the proposed framework can efficiently reason over long-term speech data .

Copied to clipboard

James Petullo, Sonny George, Dylan Cashman, Nianwen Xue

Incentivizing In-depth Reasoning over Long Contexts with Process Advantage Shaping (2026.findings-acl)

Challenge:	Weighted majority voting requires a critic to evaluate each candidate’s reasoning trace to produce the answer’s confidence score.
Approach:	They propose a lightweight framework that uses a measure of semantic similarity to filter reasoning traces that are semantically equivalent to others, degenerate, or hallucinated.
Outcome:	The proposed framework reduces token usage by 47% while maintaining or exceeding the accuracy of CISC.

Copied to clipboard

Miao Peng, Weizhou Shen, Nuo Chen, Chenliang Li, Ming Yan, Jia Li

Dissecting Clinical Reasoning in Natural Language Inference for Large Language Models (2026.findings-acl)

Challenge:	Reinforcement Learning with Verifiable Rewards (RLVR) has proven effective in enhancing LLMs’ short-context reasoning but falters in long-contemporal scenarios requiring precise grounding and multi-hop reasoning.
Approach:	They propose a framework that constructs high-difficulty, multi-hop long-context QA pairs with inherent reasoning chains to overcome this bottleneck.
Outcome:	The proposed framework outperforms RLVR baselines and matches frontier LLMs while using far fewer parameters.

Copied to clipboard

Mael Jullien, Andre Freitas, Marco Valentino, Leonardo Ranaldi

LLM-Codec: Neural Audio Codec Meets Language Model Objectives (2026.findings-acl)

Challenge:	Recent studies on large language models (LLMs) have demonstrated the impact of prompting strategies and fine-tuning techniques on their reasoning capabilities.
Approach:	They examine four classes of prompting strategies to elicit reasoning in large language models . they then construct demonstrations using a frontier model to distil multi-step reasoning capabilities into smaller models based on Low-Rank Adaptation (LoRA).
Outcome:	The proposed model improves in 75% of the models on MedNLI and TREC Clinical Trials.

Copied to clipboard

Ho-Lam Chung, Yiming Chen, Hung-yi Lee

Who is the richest club in the championship? Detecting and Rewriting Underspecified Questions Improve QA Performance (2026.findings-acl)

Challenge:	Neural audio codecs are optimized for waveform reconstruction rather than autoregressive prediction.
Approach:	They propose to augment codec training with language-model-facing objectives while keeping both codec and LLM architectures unchanged.
Outcome:	The proposed model improves speech coherence and predictability by preserving the semantic alignment between audio and text representations.

Copied to clipboard

Yunchong Huang, Gianni Barlacchi, Sandro Pezzelle

Beyond Chunks and Graphs: Retrieval-Augmented Generation through Triplet-Driven Thinking (2026.findings-acl)

Challenge:	Large language models (LLMs) perform well on well-posed factual queries, yet standard question-answering (QA) benchmarks remain far from solved.
Approach:	They propose an LLM-based classifier to identify underspecified questions and apply it to several widely used QA datasets.
Outcome:	The proposed classifier detects underspecified questions in QA datasets and significantly improves on them.

Copied to clipboard

Shengbo Gong, Xianfeng Tang, Qi He, Carl Yang, Wei Jin

Where Reasoning Breaks: Logic-Aware Path Selection by Controlling Logical Connectives in LLMs Reasoning Chains (2026.findings-acl)

Challenge:	Retrieval-augmented generation (RAG) is critical for reducing hallucinations and incorporating external knowledge into Large Language Models (LLMs).
Approach:	They propose a framework that leverages an LLM to decompose questions into searchable triplets with placeholders.
Outcome:	Empirical results show that T2RAG outperforms state-of-the-art multi-round and Graph RAG methods while reducing retrieval costs by up to 45%.

Copied to clipboard

Seunghyun Park, Yuanyuan Lei

Reheat Nachos for Dinner? Evaluating AI Support for Cross-Cultural Communication of Neologisms (2026.findings-acl)

Challenge:	a single transition error can propagate through the entire reasoning chain, leading to unstable performance.
Approach:	They propose a framework that intervenes at logical connective junctions to improve LLMs' reasoning.
Outcome:	The proposed framework achieves favorable accuracy–efficiency trade-off compared to global inference time scaling methods like beam search and self-consistency.

Copied to clipboard

Dayeon Ki, Yu Hou, Rachel Rudinger, Hal Daumé Iii, Marine Carpuat, Fumeng Yang

Learning Dynamic Representations and Policies from Multimodal Clinical Time-Series with Informative Missingness (2026.findings-acl)

Challenge:	Neologisms and emerging slang are central to daily conversation, yet challenging for non-native speakers (NNS) to interpret and use appropriately in cross-cultural communication with native speakers (NS).
Approach:	They use AI to learn English neologisms and write messages using the learned word to an NS friend.
Outcome:	The proposed model shows that AI Explanation yields the largest gains over no support in NS-rated competence, while contextual appropriateness judgments show indifference across support.

Copied to clipboard

Zihan Liang, Ziwen Pan, Ruoxuan Xiong

Quantize What Counts: More for Keys, Less for Values (2026.findings-acl)

Challenge:	Existing methods to accommodate missingness in clinical time series, but how to extract and use information carried by the observation process itself remains underexplored.
Approach:	They propose a patient representation learning framework that leverages informative missingness to learn multimodal clinical time series from structured and textual data.
Outcome:	The proposed framework improves offline treatment policy learning and adverse outcome prediction on ICU sepsis cohorts from MIMIC-III, MIMIC IV, and eICU.

Copied to clipboard

Mohsen Hariri, Alan Luo, Weicong Chen, Tianyi Zhang, Qifan Wang, Xiaotian Han, Vipin Chaudhary

Challenge:	Empirical evaluations across various prominent LLMs and benchmarks show that key-favored allocations retain up to 98.3% accuracy compared to uniform allocations (e.g., 4-bit keys, 2-bit values).
Approach:	They propose two theorems that anchor mixed-precision KV quantization in the intrinsic geometry of Transformer models.
Outcome:	Empirical evaluations show that key-favored allocations retain up to 98.3% accuracy while conserving memory.

PersonaAgent: Bridging Memory and Action for Personalized LLM Agents (2026.findings-acl)

Copied to clipboard

Weizhi Zhang, Xinyang Zhang, Chenwei Zhang, Liangwei Yang, Jingbo Shang, Zhepei Wei, Henry Peng Zou, Zijie Huang, Zhengyang Wang, Yifan Gao, Xiaoman Pan, Lian Xiong, Jingguo Liu, Philip S. Yu, Xian Li

Beyond Fixed Psychological Personas: State Beats Trait, but Language Models are State-Blind (2026.findings-acl)

Challenge:	Existing Large Language Model (LLM) enabled agents lack flexibility to respond to users’ varying needs and preferences.
Approach:	They propose a test-time user-preference alignment strategy that optimizes the persona prompt, ensuring real-time preference alignment through textual loss feedback between simulated and ground-truth responses.
Outcome:	The proposed framework outperforms baseline methods in real-time and in real applications.

Copied to clipboard

Tamunotonye Harry, Ivoline C. Ngong, Chima Nweke, Yuanyuan Feng, Joseph Near

From RAG to Agentic RAG for Faithful Islamic Question Answering (2026.findings-acl)

Challenge:	Existing persona datasets capture only trait, and ignore impact of state.
Approach:	They use a Reddit dataset to study user interactions with language models . they find that existing persona datasets capture only trait and ignore impact of state .
Outcome:	The proposed dataset decomposes variance and finds that LLMs are state-blind . the reward models react to user state, but inconsistently, the authors say .

Copied to clipboard

Gagan Bhatia, Hamdy Mubarak, Mustafa Jarrar, George Mikros, Fadi Zaraket, Mahmoud Alhirthani, Mutaz al-Khatib, Logan Cochrane, Kareem Mohamed Darwish, Rashid Yahiaoui, Firoj Alam

Losses that Cook: Topological Optimal Transport for Structured Recipe Generation (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are increasingly used for Islamic question answering, where ungrounded responses may carry serious religious consequences.
Approach:	They propose a bilingual, bilingual, Arabic/English benchmark with atomic single-gold answers that measures hallucination and abstention.
Outcome:	The proposed model improves accuracy and robustness even with a small model.

Copied to clipboard

Mattia Ottoborgo, Daniele Rege Cambrin, Paolo Garza

Concept Tokens: Learning Behavioral Embeddings Through Concept Definitions (2026.findings-acl)

Challenge:	Existing work on cooking recipes relies on cross-entropy, but it does not address holistic composition of ingredient sets and numerical aspects of recipes.
Approach:	They propose a topological loss that represents ingredient lists as point clouds in embedding space . they show that the Dice loss excels in time/temperature precision .
Outcome:	The proposed model improves ingredient- and action-level metrics while preserving time/temperature precision.

Copied to clipboard

Ignacio Sastre, Aiala Rosá

Beyond Monolithic Rewards: Hybrid Multi-Aspect Reward Optimization (2026.findings-acl)

Challenge:	Concept Tokens is a lightweight method that adds a special token to a pretrained LLM . we find that negating the hallucination token reduces hallucines and lowers precision .
Approach:	They propose a lightweight method that adds a new special token to a pretrained LLM and learns only its embedding from multiple natural language definitions of a target concept.
Outcome:	The proposed method can learn only its embedding from multiple definitions of a target concept . the study shows that it can improve hallucinations and recasting in closed-book questions .

Copied to clipboard

Radha Gulhane, Sathish Reddy Indurthi

VAUQ: Vision-Aware Uncertainty Quantification for LVLM Self-Evaluation (2026.findings-acl)

Challenge:	Existing approaches to optimize for multimodal learning use a single reward mechanism, but they lack confidence calibration across domains.
Approach:	They propose a hybrid reward and multi-aspect reward modeling framework that integrates model-based and rule-based reward paradigms for accuracy and confidence calibration.
Outcome:	The proposed model improves accuracy and confidence calibration across multimodal tasks and introduces a generalized length-penalty reward to stabilize training and improve performance.

Copied to clipboard

Seongheon Park, Changdae Oh, Hyeong Kyu Choi, Sean Du, Sharon Li

Will it Merge? On The Causes of Model Mergeability (2026.findings-acl)

Challenge:	Existing self-evaluation methods rely on a model’s ability to estimate the correctness of its own outputs, but they depend heavily on language priors and are therefore ill-suited for evaluating vision-conditioned predictions.
Approach:	They propose a vision-aware uncertainty quantification framework that measures how strongly a model’s output depends on visual evidence.
Outcome:	The proposed framework outperforms existing methods across multiple datasets.

Copied to clipboard

Adir Rahamim, Asaf Yehudai, Boaz Carmeli, Leshem Choshen, Yosi Mass, Yonatan Belinkov

Carefully Considering Culture: Analyzing LLM Alignment in Single- and Multi-Cultural Settings using Cultural Consensus Theory (2026.findings-acl)

Challenge:	Model merging has emerged as a promising technique for combining fine-tuned models into a single expert model without retraining.
Approach:	They propose a model merging technique that preserves weak model knowledge . they define mergeability as a property of model updates that captures how well they retain trained knowledge when merged with other model updates.
Outcome:	The proposed method preserves weak knowledge in the base model.

Copied to clipboard

Krishna Pothugunta, John P. Lalor

Parallel-SFT: Improving Zero-Shot Cross-Programming-Language Transfer for Code RL (2026.findings-acl)

Challenge:	Recent work in NLP has examined large language models for their understanding of cultural norms across countries, ignoring group consensus or possible multicultural environments.
Approach:	They apply cultural consensus theory to the World Values Survey to model multidimensional nuance by ignoring group consensus or over-regularizing consensus.
Outcome:	The proposed model misrepresents cultural structures by failing to form cohesive consensus or severely over-regularizing consensus.

Copied to clipboard

Zhaofeng Wu, Shiqi Wang, Boya Peng, Anuj Kumar Goyal, Melanie Kambadur, Sebastian Ruder, Yoon Kim, Chloe Bi

Distorted or Fabricated? A Survey on Hallucination in Video LLMs (2026.findings-acl)

Challenge:	Modern language models demonstrate impressive coding capabilities in common programming languages (PLs) but their performance in lower-resource PLs is often limited by training data availability.
Approach:	They propose a zero-shot cross-programming-language transfer task for code RL . they propose RL training in a source PL fails to improve performance on other target PLs .
Outcome:	The proposed approach improves transferability in Llama-3.1 code generation on parallel-stack model . it also improves performance on other target PLs, compared to single-PL SFT .

Copied to clipboard

Yiyang Huang, Yitian Zhang, Yizhou Wang, Mingyuan Zhang, Liang Shi, Huimin Zeng, Yun Fu

Generative Floor Plan Design with LLMs via Reinforcement Learning with Verifiable Rewards (2026.findings-acl)

Challenge:	Despite significant advances in video-language modeling, hallucinations remain a persistent challenge in video large language models.
Approach:	They present a systematic taxonomy that categorizes hallucinations into two core types: dynamic distortion and content fabrication.
Outcome:	The proposed taxonomy categorizes hallucinations into two core types: dynamic distortion and content fabrication.

Copied to clipboard

Luis Lara, Aristides Milios, ZhiHao Luo, Aditya Sharma, Ge Ya Luo, Christopher Beckham, Florian Golemo, Christopher Pal

Challenge:	Existing generative models focus on respecting the requested connectivity between rooms, but do not support generating floor plans that respect numerical constraints.
Approach:	They propose a text-based approach that fine-tunes a large language model on real plans and applies reinforcement learning with verifiable rewards to improve adherence to topological and numerical constraints.
Outcome:	The proposed model outperforms existing methods on Realism, Compatibility, Diversity metrics.

Creating Grammar Teaching Material for Endangered Languages with Hybrid Grammar Induction (2026.findings-acl)

Copied to clipboard

Sebastien Christian

OpenExempt: A Diagnostic Benchmark for Legal Reasoning and a Framework for Creating Custom Benchmarks on Demand (2026.findings-acl)

Challenge:	Existing methods for creating grammar lessons are labor-intensive and often fall to teachers who lack formal training in grammar.
Approach:	They propose a hybrid grammar-induction method that uses typological priors, Bayesian inference, constrained LLM reasoning and retrieval from sparse corpora to generate topic-specific grammar lessons.
Outcome:	The proposed method can produce coherent and useful lessons with better quality when modest explanatory evidence is available.

Copied to clipboard

Sergio Servantez, Sarah B. Lawsky, Rajiv Jain, Daniel W. Linna Jr., Kristian J Hammond

Context-Conditioned Masked LoRA: Dynamic Rank Routing for Compute-Efficient Parameter-Efficient Fine-Tuning (2026.findings-acl)

Challenge:	Reasoning benchmarks are expensive to build and ill suited for isolating specific failure modes.
Approach:	They propose a framework and benchmark for diagnostic evaluation of legal reasoning that uses symbolic representations of U.S. Bankruptcy Code statutes to generate large space of reasoning tasks and their machine-computable solutions on demand.
Outcome:	The proposed framework and benchmark provides diagnostic insights into the competencies and failure modes of language models.

Copied to clipboard

Rifat Rafiuddin, Rafae Abdullah

A Survey on Evaluation of LLM-based Agents (2026.findings-acl)

Challenge:	Large pretrained language models (LMs) are commonly adapted via fine-tuning, but full updates are costly at scale.
Approach:	They propose a lightweight router that activates an input-dependent subset of LoRA rank directions and turns it into dynamic rank routing.
Outcome:	The proposed method improves accuracy–efficiency Pareto frontier versus static-rank LoRA and adaptive-rank baselines, while preserving memory and reducing overhead.

Copied to clipboard

Asaf Yehudai, Lilach Eden, Alan Li, Guy Uziel, Yilun Zhao, Roy Bar-Haim, Arman Cohan, Michal Shmueli-Scheuer

Challenge:	This paper provides the first comprehensive survey of evaluation methods for LLM-based agents . LLMs are static, having fixed knowledge, and confined to text-to-text interaction.
Approach:	They analyze the evaluation of LLM-based agents across five perspectives . they identify current trends and key gaps in evaluation methods .
Outcome:	The proposed evaluation frameworks and tools are based on five perspectives . the results highlight current trends and identify gaps in future research .

ProToM: Promoting Prosocial Behaviour via Theory of Mind-Informed Feedback (2026.findings-acl)

Copied to clipboard

Matteo Bortoletto, Yichao Zhou, Lance Ying, Tianmin Shu, Andreas Bulling

JudgeMeNot: Personalizing Large Language Models to Emulate Judicial Reasoning in Hebrew (2026.findings-acl)

Challenge:	ProToM provides targeted, context-sensitive feedback to individual agents, achieving a higher success rate, shorter task completion times, and is consistently preferred by human users.
Approach:	They propose a Theory of Mind-informed facilitator that provides targeted, context-sensitive feedback to individual agents.
Outcome:	The proposed system provides targeted, context-sensitive feedback to promote prosocial behaviour, even when not directly aligned with one’s own goals.

Copied to clipboard

Itay Razumenko, Arnon Sturm, Nir Grinberg

Stabilizing Efficient Reasoning with Step-Level Advantage Selection (2026.findings-acl)

Challenge:	Existing efforts to personalize for individual decision-makers focus on user preferences rather than reasoning.
Approach:	They propose a synthetic-organic supervision pipeline that transforms raw judicial decisions into instruction-tuning data.
Outcome:	The proposed pipeline outperforms state-of-the-art methods across three tasks and settings.

Copied to clipboard

Han Wang, Xiaodong Yu, Jialian Wu, Jiang Liu, Ximeng Sun, Mohit Bansal, Zicheng Liu

RePrompT: Recurrent Prompt Tuning for Integrating Structured EHR Encoders with Large Language Models (2026.findings-acl)

Challenge:	Large language models generate long and verbose reasoning traces at inference time . short context post-training alone induces substantial reasoning compression .
Approach:	They propose a step-level advantage selection approach that reduces reasoning length by over 30% . they propose to use GRPO without any length-aware objective to train models in a shorter context window .
Outcome:	The proposed approach reduces average reasoning length by over 30% while improving Pass@1 accuracy by 3.79 points over the strongest length-aware baseline.

Copied to clipboard

Arya Hadizadeh Moghaddam, Drew Ross, Mohsen Nayebi Kerdabadi, Dongjie Wang, Zijun Yao

CLARO: Controlled Attribute-Driven Reasoning Optimization for Efficient Chain-of-Thought (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have shown promising results for mining EHRs . translating time-stamped sequences into plain text can obscure both temporal structure and code identities, weakening the ability to capture code co-occurrence and longitudinal regularities.
Approach:	They propose a time-aware LLM framework that integrates structured EHR encoders through prompt tuning without modifying underlying architectures.
Outcome:	Experiments on MIMIC-III and MIMIC IV show that RePrompT outperforms both EHR-based and LLM-based baselines across multiple clinical prediction tasks.

Copied to clipboard

Oded Schlesinger, Young Kyung Kim, J. Matias Di Martino, Guillermo Sapiro

Hidden States as Early Signals: Step-level Trace Evaluation and Pruning for Efficient Test-Time Scaling (2026.findings-acl)

Challenge:	Recent approaches guide reasoning length through token penalties or truncation, risking the omission of necessary steps.
Approach:	They propose a framework to internalize conciseness and attribute-driven reasoning optimization . they propose guiding models to embed high-quality structural attributes within a token budget .
Outcome:	The proposed method outperforms state-of-the-art models across benchmarks yielding accuracy gains of up to 63.6%.

Copied to clipboard

Zhixiang Liang, Beichen Huang, Zheng Wang, Minjia Zhang

From Recall to Forgetting: Benchmarking Long-Term Memory for Personalized Agents (2026.findings-acl)

Challenge:	Existing approaches to speed up parallel scaling have relied on similarity-based or confidence-based pruning, but these signals do not reliably indicate trace quality.
Approach:	They propose a pruning framework that evaluates reasoning steps using hidden states and dynamically prunes unpromising traces during generation.
Outcome:	The proposed framework reduces end-to-end inference latency by 45%–70% on average compared to self-consistency while improving reasoning accuracy.

Copied to clipboard

Md Nayem Uddin, Kumar Shubham, Eduardo Blanco, Chitta Baral, Gengyu Wang

Challenge:	Existing benchmarks frame long-term memory evaluation as fact retrieval from past conversations, providing limited insight into agents’ ability to consolidate memory over time or handle frequent knowledge updates.
Approach:	They propose a long-term memory benchmark that evaluates three memory-grounded tasks: remembering, reasoning, and recommending.
Outcome:	The proposed benchmarks evaluate three tasks: remembering, reasoning, and recommending.

Towards Inference-time Scaling for Continuous Space Reasoning (2026.findings-acl)

Copied to clipboard

Minghan Wang, Thuy-Trang Vu, Ehsan Shareghi, Gholamreza Haffari

SynthFix: Adaptive Neuro-Symbolic Code Vulnerability Repair (2026.findings-acl)

Challenge:	Recent advances in reasoning large language models have expanded along training and inferencetime dimensions.
Approach:	They propose to use COCONUT (CITATION) continuous space reasoning LM as the backbone to generate diverse reasoning paths through dropout-based sampling.
Outcome:	The proposed method could enable performance gains similar to those observed in the discrete space, but only marginally improves in the continuous space.

Copied to clipboard

Yifan Zhang, Jieyu Li, Kexin Pei, Yu Huang, Kevin Leach

Imperfectly Cooperative Human-AI Interactions: Comparing the Impacts of Human and AI Attributes in Simulated and User Studies (2026.findings-acl)

Challenge:	Large Language Models (LLMs) struggle with complex semantic and structural correctness required for automated code repair.
Approach:	They propose a hybrid neural-symbolic framework that unifies code synthesis with compiler-informed symbolic feedback to improve LLM-based vulnerability repair.
Outcome:	The proposed framework improves code repair accuracy and efficiency over strong SFT and RFT training strategies on the FixJS and CodeFlaws benchmarks.

Copied to clipboard

Myke C. Cohen, Mingqian Zheng, Neel Bhandari, Hsien-Te Kao, Xuhui Zhou, Daniel Nguyen, Laura Cassani, Maarten Sap, Svitlana Volkova

Challenge:	In simulations, personality traits and AI attributes were comparatively influential, but with actual human subjects, AI attributes – particularly transparency – were much more impactful.
Approach:	They compare a purely simulated dataset and a parallel human subjects experiment to examine how human personality traits and AI design characteristics jointly shape interaction outcomes in imperfectly cooperative scenarios.
Outcome:	The results show that personality traits and AI attributes are comparatively influential in simulations, but with actual human subjects, they are much more impactful.

CheMM-R1: Enhancing Chemical Structure Recognition and Elucidation with Reasoning Multimodal Large Language Models (2026.findings-acl)

Copied to clipboard

Liting Huang, Zhihao Zhang, Shoujin Wang

Agree, Disagree, Explain: Decomposing Human Label Variation in NLI through the Lens of Explanations (2026.findings-acl)

Challenge:	Existing multimodal large language models lack domain-specific expertise to perform chemical tasks.
Approach:	They propose a benchmark dataset for evaluating multi-step multimodal reasoning capacities in the chemistry domain.
Outcome:	The proposed model surpasses existing models in all CheMM-Bench tasks.

Copied to clipboard

Pingjun Hong, Beiduo Chen, Siyao Peng, Marie-Catherine de Marneffe, Benjamin Roth, Barbara Plank

DeepSpecs: Expert-Level Question Answering in 5G (2026.findings-acl)

Challenge:	Natural Language Inference (NLI) datasets often exhibit label variation.
Approach:	They extend LiTEx taxonomy to two NLI datasets and jointly analyze label variation and label variation.
Outcome:	The proposed model combines explanations as a lens to analyze variation in NLI annotations and examine individual differences in reasoning.

Copied to clipboard

Aman Ganapathy Manvattira, Yifei Xu, Ziyue Dang, Songwu Lu

On-Policy Self-Distillation for Efficient Diffusion Language Models with Early-Stage Calibration (2026.findings-acl)

Challenge:	3GPP standards define the technical design and implementation of 5G systems . expert-level questions require navigating thousands of pages of cross-referenced standards .
Approach:	They propose a standard-native retrieval-augmented generation system that can answer 5G questions . they use SpecDB, ChangeDB, TDocDB and a metadata-rich retrieval system to do this .
Outcome:	The proposed solution outperforms base models and state-of-the-art RAG systems in QA datasets . expert-level queries require navigating thousands of pages of cross-referenced standards .

Copied to clipboard

Huaisheng Zhu, MingYu Liu, Junze Liu, Zhen Ge, Tian Wang, Jiri Gesi, Dakuo Wang, Weiqi Zhang, Houyu Zhang, Yufan Guo, Xian Li, Bing Yin, Sujay Sanghavi

Arctic-Text2SQL-R1: Simple Rewards, Strong Reasoning in Text-to-SQL (2026.findings-acl)

Challenge:	Recent studies have demonstrated that masked diffusion models (MDMs) can surpass autoregressive models (ARMs) in various tasks.
Approach:	They propose a method to calibrate early token predictions without demonstration data by distilling an unnormalized target distribution into the original model.
Outcome:	Experiments on math, planning, and RLHF tasks show that COPSD improves both effectiveness and efficiency, and further enhances performance when combined with supervised fine-tuning.

Copied to clipboard

Zhewei Yao, Guoheng Sun, Łukasz Borchmann, Zheyu Shen, Minghang Deng, Bohan Zhai, Hao Zhang, Ang Li, Yuxiong He

Do Emotions Influence Moral Judgment in Large Language Models? (2026.findings-acl)

Challenge:	Translating natural language questions into SQL is a core challenge in natural language understanding and human-computer interaction.
Approach:	They propose a reinforcement learning framework and model family to generate accurate, executable SQL using a lightweight reward signal based solely on execution correctness.
Outcome:	The proposed framework outperforms previous versions of 70B-class systems and achieves state-of-the-art execution accuracy across six diverse Text2SQL benchmarks.

Copied to clipboard

Mohammad Saim, Tianyu Jiang

RShield: A User-level Traceable Backdoor Watermark for LLMs in Embedding-as-a-Service (2026.findings-acl)

Challenge:	Recent systems enforce explicit ethical constraints, but moral judgment rarely involves such clear-cut prohibitions.
Approach:	They develop an emotion-induction pipeline that infuses emotion into moral situations and evaluate shifts in moral acceptability across datasets and LLMs.
Outcome:	The proposed pipeline can infuses emotion into moral situations and evaluate moral acceptability shifts across datasets and LLMs.

Copied to clipboard

Lingyun Xiang, Yufan Zhong, Chengfu Ou, Zhihua Xia, Chunfang Yang, Daojian Zeng, Zhangjie Fu

The Adaptive Interrogator: Detecting Trojan LLMs in Multi-Agent Systems via Evolved Conversational Strategies (2026.findings-acl)

Challenge:	Existing backdoor watermarking techniques are limited to zero-bit detection . RShield enables reliable user-level attribution of large language models under model extraction attacks.
Approach:	They propose a multi-bit backdoor watermarking technique that enables reliable user-level attribution of large language models under model extraction attacks.
Outcome:	RShield achieves 100% multi-bit watermark recovery and high semantic fidelity under model extraction attacks compared to existing methods.

Copied to clipboard

Rana Muhammad Shahroz Khan, Ruichen Zhang, Zhen Tan, Charles Fleming, Tianlong Chen

SmartAD: Capacity-Aligned Agent Distillation for Small Language Models (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have been focused on single-agent, white-box environments, but multi-agend systems (MAS) have a critical blind spot: supply chain vulnerabilities.
Approach:	They propose a black-box auditing framework that leverages an Evolutionary Algorithm to autonomously expose hidden threats.
Outcome:	The proposed framework achieves superior detection rates (up to 100% in specific configurations) and robustness across diverse architectures.

Copied to clipboard

Guokai Tang, Feng Zhao

TRM-Planner: Offline Target Planning and Distillation for Tiny Recursive Models (2026.findings-acl)

Challenge:	Large language models (LLMs) show strong reasoning and decision-making ability, but their high inference cost motivates transferring agentic skills to small language models.
Approach:	They propose a capacity-aligned agent distillation framework that trains SLMs on full reason–act–observe trajectories from a tool-using teacher.
Outcome:	The proposed framework outperforms all baselines on multi-hop QA and math benchmarks with 1.5B and 3B models.

Copied to clipboard

Euijin Baek, Housam Babiker, Mi-Young Kim, Randy Goebel

MAPLE: Multi-Aspect Panels of LLM Evaluators for Open-Ended Questions (2026.findings-acl)

Challenge:	Tiny Recursive Models (TRMs) perform iterative reasoning with an Adaptive Computation Time (ACT)-style loop, but their supervised training targets can be brittle and their halting behavior difficult to tune.
Approach:	They propose a two-stage teacher-cache distillation recipe that shifts compute to offline teacher-caching stage.
Outcome:	The proposed model improves su-pervision while leaving student-time inference unchanged.

Copied to clipboard

Michinori Jinji, Kyohei Atarashi, Koh Takeuchi, Hisashi Kashima

Challenge:	LLM-as-a-Judge uses LLMs to evaluate open-ended questions . however, the discrepancy between LLM generated evaluations and human evaluations remains a critical problem in this field .
Approach:	They propose a framework that orchestrates evaluations across multiple criteria using multiple LLMs.
Outcome:	The proposed framework achieves superior alignment with human evaluations compared to baselines.

MARCH: Evaluating the Intersection of Ambiguity Interpretation and Multi-hop Inference (2026.findings-acl)

Copied to clipboard

Jeonghyun Park, Ingeol Baek, Seunghyun Yoon, Haeun Jang, Aparna Garimella, Akriti Jain, Nedim Lipka, Hwanhee Lee

Enhancing Multilingual RAG Systems with Debiased Language Preference-Guided Query Fusion (2026.findings-acl)

Challenge:	Existing benchmarks on multi-hop QA focus on single-hop and layered ambiguity, but they focus on ambiguous questions . ambiguities can arise at any stage, complicating the reasoning process .
Approach:	They propose a benchmark to evaluate ambiguity in multi-hop question answering . they propose MARCH, which uses 2,209 carefully annotated questions .
Outcome:	The proposed framework outperforms existing approaches and significantly outperfies existing frameworks.

Copied to clipboard

Jeonghyun Park, Byeongjeong Kim, Seojin Hwang, Hwanhee Lee

MT-OSC: Path for LLMs that Get Lost in Multi-Turn Conversation (2026.findings-acl)

Challenge:	Existing studies show that mRAGs exhibit a perceived preference for high-resource languages, particularly English.
Approach:	They propose a debiased language preference metric to explicitly factor out structural priors . they propose mRAG framework that leverages monolingual alignment to optimize cross-lingual retrieval and generation.
Outcome:	The proposed framework outperforms baselines for English pivoting and mRAG in multiple languages.

Copied to clipboard

Jyotika Singh, Fang Tu, Miguel Ballesteros, Weiyi Sun, Sandip Ghoshal, Michelle Yuan, Yassine Benajiba, Sujith Ravi, Dan Roth

Soft Head Selection for Injecting ICL-Derived Task Embeddings (2026.findings-acl)

Challenge:	Large language models suffer performance degradation when user instructions and context are distributed over multiple conversational turns.
Approach:	They propose a framework that condenses chat history in the background without disrupting the user experience.
Outcome:	The proposed framework reduces token counts by up to 72% in 10-turn dialogues while remaining robust to distractors and irrelevant turns.

Copied to clipboard

Jungwon Park, Jimyeong Kim, Changin Choi, Wonjong Rhee

CORAL: Adaptive Retrieval Loop for Culturally-Aligned Multilingual RAG (2026.findings-acl)

Challenge:	Large language models (LLMs) are commonly adapted to downstream tasks using parameter-efficient fine-tuning (PEFT) or in-context learning (ICL).
Approach:	They propose a gradient-based method that derives task-specific embeddings from activations using few-shot prompts and injects them during inference.
Outcome:	The proposed method outperforms existing methods on open-ended generation, reasoning, and natural language understanding tasks while using fewer trainable parameters.

Copied to clipboard

Nayeon Lee, Jiwoo Song, Byeongcheol Kang

Safety Guardrails of Large Language Models Are Vulnerable to Value-Driven Adversarial Prompting (2026.findings-acl)

Challenge:	Multilingual retrieval-augmented generation is inadequate for culturally grounded queries . Across two cultural QA benchmarks, CORAL achieves a 3.58%p accuracy improvement on low-resource languages .
Approach:	They propose a multilingual retrieval-augmented generation approach that enables iterative refinement of both the retrieval space and the retrieving probe based on the quality of the evidence.
Outcome:	Using CORAL, researchers find that culturally grounded queries can be improved . if retrieved documents are insufficient, the system reselects them and rewrites the query .

Copied to clipboard

Xiaohao Luo, Ying Wei, Zhijun Li

Fair RAG: End-to-End Fairness Across Retrieval and Generation (2026.findings-acl)

Challenge:	Existing jailbreak attacks against large language models (LLMs) can be divided into white-box attacks and black-box attack.
Approach:	They propose a value-driven jailbreak attack that exploits the phenomenon that large language models agree with humans to induce LLMs to affirm the moral value of harmful tasks.
Outcome:	Extensive experiments on five state-of-the-art (SOTA) LLMs show the value-driven jailbreak attack achieves an average attack success rate (ASR) of 91.8% on JailbreakBench and 95.2% on the AdvBench subset.

Copied to clipboard

Farsheed Haque, Zhe Fu, Ramit Aditya, Depeng Xu, Xi Niu

ManCC: A Task-Anchored Benchmark for Manchu–Classical Chinese Cross-Lingual Modeling (2026.findings-acl)

Challenge:	Large Language Models (LLMs) can amplify demographic bias by generating skewed context . prior work treats fairness in retrieval or generation in isolation, leaving end-to-end fairness underexplored .
Approach:	They propose a pipeline that jointly controls both retrieval and generation stages . large language models can handle a broad set of inference tasks, they argue .
Outcome:	The proposed pipeline reduces retriever-side skew and achieves lowest generator-side disparity while preserving utility.

Copied to clipboard

Meiqi Wang, Xiaoxin Sun, Dongjie Wang, Ruixin Yu, Xiantao Heng, Shuo Wang, Zhen Huang, Peng Zhao, Suhua Wang, Minghao Yin

Dynamic Graph Navigation via Triplet Chains for Structure-Aware Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	Mainstream research in natural language processing has focused on high-resource and modern languages.
Approach:	They propose a task-anchored benchmark for Manchu–Classical Chinese translation . they use a parallel corpus of 16,627 sentence pairs to evaluate the model .
Outcome:	The proposed benchmarks show that linguistic differences influence performance and broader language coverage facilitate low-resource transfer.

Copied to clipboard

Feng Zhao, Yufei Wu, Xianggan Liu, Ruilin Zhao

Style over Story: Measuring LLM Narrative Preferences via Structured Selection (2026.findings-acl)

Challenge:	Retrieval-Augmented Generation (RAG) is a strategy to mitigate hallucination and factual errors in large language models (LLMs).
Approach:	They propose a structure-aware RAG which achieves noise removal in retrieval through multi-chain graph navigation reasoning.
Outcome:	The proposed method achieves noise removal in retrieval through multi-chain graph navigation reasoning (Trig-Nav) compared to baseline methods, it significantly improves the model’s performance, validating the effectiveness of this approach.

Copied to clipboard

Donghoon Jung, Jiwoo Choi, Songeun Chae, Seohyon Jung

Challenge:	Existing studies show that large language models encode preference-related signatures, including political preferences, personality traits, and value correlations.
Approach:	They propose a constraint-selection-based experiment for measuring narrative preferences of Large Language Models by eliciting choices among narrative constraints when alternatives are explicitly specified.
Outcome:	The proposed model prioritizes Style over content elements like Event, Character, and Setting.

Visual Interference in Speech Evaluation: Cultural Asymmetry and Cross-Modal Bias in MLLMs (2026.findings-acl)

Copied to clipboard

Kyusik Kim, Hyunwoo Yoo, Jaehoon Choi, Gail Rosen, Bongwon Suh

DraDDP: A Multimodal Multi-Party Dialogue Discourse Parsing Dataset (2026.findings-acl)

Challenge:	a new paradigm shifts the paradigm of speech processing from simple transcription to complex social reasoning.
Approach:	They construct a cross-modal dataset to examine cultural asymmetry in MLLMs . they find that ML models actively reproduce context-dependent sociolinguistic ideologies based on native audio .
Outcome:	The proposed model exhibits cultural asymmetry in anglophone and Korean contexts . the model reproduces sociolinguistic ideologies, consistent with Expectancy Violation Theory .

Copied to clipboard

Shannan Liu, Peifeng Li, Yaxin Fan, Qiaoming Zhu

Table-R1: Region-based Reinforcement Learning for Table Understanding (2026.findings-acl)

Challenge:	Existing studies on multi-party dialogue discourse parsing focus on textual modality and two-party dialog . et al., 2016) focused on text-based discourse parses, ignoring the complexity and richness of multimodal interactions in real-world scenarios.
Approach:	They construct the first publicly available English multimodal dataset for multi-party dialogue discourse parsing based on American TV dramas.
Outcome:	The proposed dataset contains 495 dialogue segments with 6,374 utterances and 9.1 hours of parallel video content, covering rich multi-party interaction scenarios.

Copied to clipboard

Zhenhe Wu, Jian Yang, Zhongjiang He, Changzai Pan, Jiaheng Liu, Xianjie Wu, Yu Zhao, Shuangyong Song, Yongxiang Li, Zhoujun Li, Xuelong Li

ARM2: Adaptive Reasoning Model with Vision Understanding and Executable Code (2026.findings-acl)

Challenge:	Tables are a widely used data format that poses unique challenges for language models due to their structured row-column interactions.
Approach:	They propose a region-based reinforcement learning approach that integrates region evidence into reasoning steps.
Outcome:	The proposed method outperforms baseline models on three benchmark datasets and significantly reduces the reasoning token consumption by 67.5%.

Copied to clipboard

Jian Xie, Zhendong Chu, Aoxiao Zhong, Kai Zhang, Mingzhe Han, Xing Fan, Jialie Shen, Qingsong Wen

A Framework of Reflective Agents with Adaptive Collaboration for Attributed Summary Generation (2026.findings-acl)

Challenge:	Large Reasoning Models suffer from the "over-thinking" problem, causing performance degradation.
Approach:	They propose a unified model that balances reasoning performance and efficiency across multiple formats through a reinforcement learning framework augmented with length-aware optimization.
Outcome:	The proposed model reduces token costs while preserving performance compared to traditional models.

Copied to clipboard

Yu Chen, Peng Chen, Ziwei Zheng, Bang Wang

VLURes: Benchmarking Long-Text Grounding and Cross-Lingual Robustness in Vision Language Models (2026.findings-acl)

Challenge:	Existing prompt-based summarization approaches face limitations such as positional preference, poor citation quality and sensitivity to uninformative documents.
Approach:	They propose a framework of Reflective Agents with Adaptive Collaboration for attributed summarization that performs iterative summarizing via reflective agents’ collaboration.
Outcome:	The proposed framework outperforms baselines on the ALCE benchmark in factual correctness and citation quality.

Copied to clipboard

Jesse Atuhurra, Iqra Ali, Tomoya Iwakura, Hidetaka Kamigaito, Tatsuya Hiraoka

Trustworthy and Explainable Causal Representation Learning in Transformers (2026.findings-acl)

Challenge:	*VLURes provides a practical testbed for long-text grounding and multilingual robustness in web-realistic agent settings.
Approach:	They propose a multilingual benchmark for evaluating vision-language models under long-text grounding.
Outcome:	*VLURes provides a testbed for long-text grounding and multilingual robustness in web-realistic agent settings.

Copied to clipboard

Yang Liu, Yinghao Zhang, Lin Liu, Jiuyong Li, Debo Cheng, Zaiwen Feng

Affectron: Emotional Speech Synthesis with Affective and Contextually Aligned Nonverbal Vocalizations (2026.findings-acl)

Challenge:	Existing approaches to interpretable representation learning rely on masks that weight the significance of input features, but the origin of these masks is uncertain.
Approach:	They propose a causal framework that directly learns identifiable representations from attention weights rather than relying on importance masks.
Outcome:	The proposed framework learns identifiable and explainable representations from attention weights, rather than masks, and guarantees faithfulness on real-world datasets.

Copied to clipboard

Deok-Hyeon Cho, Hyung-Seok Oh, Seung-Bin Kim, Seong-Whan Lee

Reasoning-Guided Exploration for Online DPO (2026.findings-acl)

Challenge:	Existing studies on speech synthesis incorporating NVs have limited ability to generate expressive speech.
Approach:	They propose a framework for affective and contextually aligned NV generation using a small-scale open corpus.
Outcome:	The proposed framework expands the distribution of NV types and insertion locations . it produces more expressive and diverse NVs than baseline systems while preserving naturalness of verbal speech stream.

Copied to clipboard

Zetian Hu, Shunyu Liu, Ting-En Lin, Fei Huang, Yongbin Li, Dacheng Tao

CAPA: Contribution-Aware Pruning and FFN Approximation for Efficient Large Vision-Language Models (2026.findings-acl)

Challenge:	Recent work has aimed to enhance reasoning capabilities of language models, but these methods are limited to domains with objectively verifiable answers.
Approach:	They propose a self-play framework to improve reasoning on general-domain data.
Outcome:	Experiments show that the proposed framework improves reasoning performance on general-domain data while maintaining competitive performance on verifiable academic benchmarks.

Copied to clipboard

Samyak Jha, Junho Kim

Strong Reasoning Isn’t Enough: Evaluating Evidence Elicitation in Interactive Diagnosis (2026.findings-acl)

Challenge:	Efficient inference in Large Vision Language Models is constrained by the high cost of processing thousands of visual tokens.
Approach:	They propose a framework that prunes visual tokens using attention contribution at critical functional transitions and reduces computations using efficient linear approximations.
Outcome:	The proposed framework achieves competent efficiency–performance trade-offs with improved robustness.

Copied to clipboard

Zhuohan Long, Zhongyu Wei

IF-GEO: Conflict-Aware Instruction Fusion for Multi-Query Generative Engine Optimization (2026.findings-acl)

Challenge:	Existing evaluations of medical consultation are static or outcome-centric, neglecting the evidence-gathering process.
Approach:	They propose an interactive evaluation framework that explicitly models the consultation process using a simulated patient and a measurement module grounded in atomic evidences.
Outcome:	The proposed evaluation framework outperforms baseline evaluation methods in medical consultation settings.

Copied to clipboard

Heyang Zhou, Jiajia Chen, Xiaolu Chen, Jie Bao, Zhen Chen, Yong Liao

Investigating Human and LLMs’ Decisions in Unverifiable Environments: A Case Study with GitHub Activity Overview (2026.findings-acl)

Challenge:	Existing methods to improve content visibility are static heuristic rules or optimize for heterogeneous queries.
Approach:	They propose a "diverge-then-converge" framework that extracts optimization preferences from latent queries and synthesizes a global revision blueprint for guided editing.
Outcome:	The proposed framework achieves substantial performance gains while maintaining robustness across diverse retrieval scenarios.

Copied to clipboard

Zheng Jiang, Wei Wang, Gaowei Zhang, Yang Feng, Yi Wang

PlotGen-Bench: Evaluating VLMs on Generating Visualization Code from Diverse Plots across Multiple Libraries (2026.findings-acl)

Challenge:	examining the behaviors of Large Language Models as artificial social actors is underexplored, especially in unverifiable scenarios where conventional benchmarking has little to help improve their abilities.
Approach:	They propose a method to collect, compare, and reason about human and LLMs' decisions in an unverifiable scenario and use it to examine their behaviors.
Outcome:	The proposed method compared human and LLM decisions in an unverifiable scenario on GitHub and found that proprietary LLMs behave more like humans than open-source LLM systems.

Copied to clipboard

Yi Zhao, Zhen Yang, Shuaiqi Duan, Wenmeng Yu, Zhe Su, Jibing Gong, Jie Tang

Chronos: Learning Temporal Dynamics of Reasoning Chains for Test-Time Scaling (2026.findings-acl)

Challenge:	PlotGen-Bench evaluates vision-language models' ability to generate executable visualization code from plots under realistic and complex visualization requirements.
Approach:	They propose a benchmark to evaluate plot-to-code generation in vision-language models . they use Matplot, Matplos, Mat3D, Mat4D, and Mat4E to evaluate their performance .
Outcome:	The proposed benchmark covers 9 major categories, 30 subcategories, and 3 core tasks . it covers 2D, 3D and animated plots across 5 widely used visualization libraries.

Copied to clipboard

Kai Zhang, Jiayi Liao, Chengpeng Li, Ziyuan Xie, Sihang Li, Xiang Wang

V-RoLoRA: RLVR-Driven MoE Routing for Steerable Pluralistic Alignment (2026.findings-acl)

Challenge:	Existing methods for testing time scales treat reasoning traces or tokens equally, ignoring substantial variations in trajectory quality and localized logical failures.
Approach:	They propose a chronological reasoning scorer that models each trajectory as a time series.
Outcome:	The proposed method achieves relative improvements of 34.21% over Pass@128 and 22.70% over Maj@135 on HMMT25, highlighting its effectiveness.

Copied to clipboard

Jing Wang, Yaomin Wu, Yinglin Wang, Yitong Yang

ParaCook: On Time-Efficient Planning for Multi-Agent Systems (2026.findings-acl)

Challenge:	Current methods for steering large language models rely on prompt engineering or reasoning-time guidance.
Approach:	They propose a value-controllable pluralistic alignment framework enhanced with conditioned gating that dynamically directs the flow among multiple experts based on an input value or moral vector.
Outcome:	The proposed method outperforms prompt-based steering and multi-task PEFT benchmarks on two 8-billion-parameter backbones.

Copied to clipboard

Shiqi Zhang, Xinbei Ma, Yunqing Xu, Zouying Cao, Pengrui Lu, Haobo Yuan, Tiancheng Shen, Zhuosheng Zhang, Hai Zhao, Ming-Hsuan Yang

Multimodal Unlearning Across Vision, Language, Video, and Audio: Survey of Methods, Datasets, and Benchmarks (2026.findings-acl)

Challenge:	Existing agent benchmarks focus on task completion while neglecting time efficiency in parallel and asynchronous operations.
Approach:	They propose a framework for large language models that allows agents to plan long-horizon tasks in a scalable way.
Outcome:	The proposed framework is based on the Overcooked game and can be used to evaluate time efficiency-aware multi-agent planning.

Copied to clipboard

Nobin Sarwar, Shubhashis Roy Dipta, Zheyuan Liu, Vaidehi Patil

From Interpretability to Performance: Optimizing Retrieval Heads for Long-Context Language Models (2026.findings-acl)

Challenge:	Survey aims to identify challenges of multimodal unlearning for vision, language, audio and video . retraining after deletion requests or policy updates is often impractical, survey finds .
Approach:	They propose to enable selective removal across modalities while retaining overall utility.
Outcome:	This study compares models with existing models to identify weaknesses and improves performance.

Copied to clipboard

Youmi Ma, Naoaki Okazaki

Bridging the Temporal Gap in Multimodal LLMs: Deeply Stacking Temporal Tokens for Audio-Visual Speech Recognition (2026.findings-acl)

Challenge:	Recent studies on mechanistic interpretability revealed that long-context factuality is closely related to a set of attention heads, retrieval heads.
Approach:	They propose a method that generates training signals by contrasting normal model outputs with those from an ablated variant.
Outcome:	The proposed method achieves significant improvements on LLMs with a sparse retrieval score distribution.

Copied to clipboard

Liyong Wang, Junliang Xing, Tianyu Hu, Jianfei Jiang, Jihuai Zhao, Huimin Ma

MHSafeEval: Role-Aware Interaction-Level Evaluation of Mental Health Safety in Large Language Models (2026.findings-acl)

Challenge:	Existing audio-visual speech recognition systems suffer from a temporal gap . visual speech patterns captured from lip movements provide complementary information that remains inherently robust to acoustic noise.
Approach:	They propose a framework that deeply stacks temporal tokens across both encoding and decoding stages to bridge this temporal gap.
Outcome:	The proposed framework outperforms existing supervised, self-supervised, and LLM-based methods by 6.1% on LRS2 and 7.8% on LLS3.

Copied to clipboard

Suhyun Lee, Palakorn Achananuparp, Neemesh Yadav, Ee-Peng Lim, Yang Deng

Enabling Stroke-Level Structural Analysis of Hieroglyphic Scripts without Language-Specific Priors (2026.findings-acl)

Challenge:	Existing evaluation frameworks assess isolated responses using coarse-grained taxonomies or static datasets.
Approach:	They propose a role-aware mental health safety taxonomy that characterizes clinically significant harm in terms of interactional roles an AI counselor adopts.
Outcome:	The proposed framework significantly improves failure-mode coverage and diagnostic granularity.

Copied to clipboard

Fuwen Luo, Zihao Wan, Ziyue Wang, Yaluo Liu, Pau Tong Lin Xu, Xuanjia Qiao, Xiaolong Wang, Peng Li, Yang Liu

Challenge:	Existing structural analysis methods for hieroglyphic scripts are script-specific and labor-intensive.
Approach:	They propose a hieroglyphic Stroke Analyzer framework that captures character-internal structures and semantics without handcrafted data.
Outcome:	The proposed framework captures character-internal structures and semantics without priors . it can be used to generalize hieroglyphic scripts across languages .

CODEPROMPTZIP: Code-specific Prompt Compression for Retrieval-Augmented Generation in Coding Tasks with LMs (2026.findings-acl)

Copied to clipboard

Pengfei He, Shaowei Wang, Tse-Hsun Chen

Detecting Hallucinations in Retrieval-Augmented Generation via Semantic-level Internal Reasoning Graph (2026.findings-acl)

Challenge:	Existing prompt compression techniques for natural language lack fine-grained control over compression ratios.
Approach:	They propose a code-aware prompt compression framework for RAG that enables precise length control while preserving critical information.
Outcome:	The proposed framework outperforms baselines on three code-related tasks while maintaining the most informative tokens.

Copied to clipboard

Jianpeng Hu, Yanzeng Li, Jialun Zhong, Lei Zou, Wenfa Qi

MeasHalu: Mitigation of Scientific Measurement Hallucinations for Large Language Models with Enhanced Reasoning (2026.findings-acl)

Challenge:	Existing methods for detecting faithfulness hallucinations are coarse or do not capture the models’ internal reasoning processes, making it difficult to learn.
Approach:	They propose a semantic-level internal reasoning graph-based method for detecting faithfulness hallucination using Large language models.
Outcome:	The proposed method achieves better overall performance compared to state-of-the-art baselines on RAGTruth and Dolly-15k.

Copied to clipboard

Ruijun Huang, Zhiqiao Kang, Yuxuan Zhu, Junxiong Li, Jiahao Zhao, Minghuan Tan, Feng Jiang, Min Yang

CRAFTQA: A Code-Driven Adaptive Framework for Complex Structured Data Reasoning (2026.findings-acl)

Challenge:	Large Language Models (LLMs) exhibit severe hallucinations, which undermine reliability of automated scientific document understanding systems.
Approach:	They propose a framework for mitigating scientific measurement hallucinations through enhanced reasoning and targeted optimization.
Outcome:	The proposed framework significantly reduces hallucination rates and improves overall accuracy on the MeasEval benchmark.

Copied to clipboard

Chengtao Gan, Zhiqiang Liu, Long Jin, Yushan Zhu, Lei Liang, Wen Zhang

EMTIR-GRPO: Efficient Multi-Tool Augmented Large Language Models via Reinforcement Learning (2026.findings-acl)

Challenge:	Existing unified structured data question answering methods rely on a set of predefined functions, which restricts their ability to perform complex reasoning beyond these predefined operations.
Approach:	They propose a novel adaptive code-driven framework that generates code-based reasoning operations based on a question.
Outcome:	The proposed framework improves on multiple structured datasets on real-world scenarios.

Copied to clipboard

Shixin Jiang, Zhihao Zhu, Jiafeng Liang, Yang Wu, Ming Liu, Bing Qin

Multi-Persona Thinking for Bias Mitigation in Large Language Models (2026.findings-acl)

Challenge:	Tool-integrated reasoning (TIR) enables large language models to invoke external tools for tasks beyond their internal capacity but often suffers from tool overuse.
Approach:	They propose an algorithm that uses a composite reward to model tool costs and tool efficiency.
Outcome:	The proposed algorithm models heterogeneous tool costs and encourages more cost-effective tool-use strategies.

Copied to clipboard

Yuxing Chen, Guoqing Luo, Zijun Wu, Lili Mou

LCO: LLM-based Constraint Optimization for Safer Agentic LLMs in Real-world Tasks (2026.findings-acl)

Challenge:	Large Language Models exhibit social biases, which can lead to harmful stereotypes and unfair outcomes.
Approach:	They propose a simple inference-time framework that encourages reasoning from multiple perspectives.
Outcome:	The proposed framework reduces bias by encouraging reasoning from multiple perspectives.

Copied to clipboard

Jiayong Wan, Jiawei Chen, Zhaoxia Yin, Liu Shuyuan, Hang Su

MedKInstruct: A Multimodal Knowledge Graph Based Framework for Multi-Hop and Hard-Negative Instruction Data Synthesis in MedVQA (2026.findings-acl)

Challenge:	Existing defense methods are insufficient to address in-context reward hacking (ICRH), where LLMs iteratively optimize their behavior to maximize proxy objectives, resulting in harmful side effects.
Approach:	They propose a framework that reduces in-context reward hacking (ICRH) through repeated interactions with the environment.
Outcome:	The proposed framework reduces ICRH without model fine-tuning while maintaining task performance.

Copied to clipboard

Yinan Wu, Jihang Jin, Xuhao Bao, Weiyan Zhang, Hanjing Yan, Tong Ruan, ChunMing Wang

GeometryZero: Advancing Geometry Solving via Group Contrastive Policy Optimization (2026.findings-acl)

Challenge:	Existing methods for medical visual question answering focus on image–caption pairs, limiting the model’s ability to learn relevant medical knowledge during training.
Approach:	They propose to synthesize instruction data from image–caption pairs and incorporate a multimodal medical knowledge graph to assist LVLMs in synthesizing knowledge-intensive instruction data.
Outcome:	The proposed model outperforms existing methods on the public datasets Slake and VQA-RAD by 4.16% and 4.50%.

Copied to clipboard

Yikun Wang, Yibin Wang, Dianyi Wang, Zimian Peng, Qipeng Guo, Dacheng Tao, Jiaqi Wang

CollabCoder: Plan-Code Co-Evolution via Collaborative Decision-Making for Efficient Code Generation (2026.findings-acl)

Challenge:	Existing methods for auxiliary construction training are expensive and underperform . Existing Corresponding Author training methods lack self-correction capabilities in reasoning chains.
Approach:	They propose a reinforcement learning framework that rewards auxiliary construction with geometric reasoning by grouping construction rewards with a Length Reward.
Outcome:	Experiments on Geometry3K and MathVista show that GeometryZero outperforms baselines on auxiliary constructions.

Copied to clipboard

Duy Tung Doan, Quang Huy Phung, Dzung Nguyen, Khac-Hoai Nam Bui

Beyond the Safety Tax: Mitigating Unsafe Text-to-Image Generation via External Safety Rectification (2026.findings-acl)

Challenge:	Existing multi-agent code generation frameworks are constrained by static planning, isolated execution, high computational overhead, and limited adaptability to complex tasks.
Approach:	They propose a plan-code co-evolution framework that allows dynamic multi-agent collaboration to improve code quality and robustness across tasks.
Outcome:	The proposed framework improves code quality and robustness across tasks while reducing the number of API calls by an average of 4-10 per execution.

Copied to clipboard

Xiangtao Meng, Yingkai Dong, Ning Yu, Li Wang, Zheng Li, Shanqing Guo

WenetSpeech-Wu: Datasets, Benchmarks, and Models for a Unified Chinese Wu Dialect Speech Processing Ecosystem (2026.findings-acl)

Challenge:	Existing safety defenses typically intervene internally within the generative model, but suffer from severe concept entanglement, leading to degradation of benign generation quality.
Approach:	They propose a structurally isolated safety module that performs external, interpretable rectification without modifying the base model.
Outcome:	The proposed module performs external, interpretable rectification without modifying the base model.

Copied to clipboard

Chengyou Wang, Mingchen Shao, Jingbin Hu, Zeyu Zhu, Hongfei Xue, Bingshen Mu, Xin Xu, Xingyi Duan, Binbin Zhang, Zhu Pengcheng, Chuang Ding, Xiaojun Zhang, Hui Bu, Lei Xie

ReaGeo: Reasoning-Enhanced End-to-End Geocoding with LLMs (2026.findings-acl)

Challenge:	despite its linguistic significance, the Wu dialect of Chinese has long been hindered by the lack of large-scale speech data, standardized evaluation benchmarks, and publicly available models.
Approach:	They propose to use WenetSpeech-Wu as a large-scale, multi-dimensionally annotated open-source speech corpus for the Wu dialect of Chinese.
Outcome:	The proposed dataset includes 8,000 hours of speech data and strong open-source models . the proposed dataset is competitive and empirically validated .

Copied to clipboard

Jian Cui, Zhiyuan Ren, Desheng Weng, Yongqi Zhao, Gong Wenbin, Yu Lei, Zhenning Dong

EviReport: From Reasoned Outlines to Evidence Tracked Long-Form Reports (2026.findings-acl)

Challenge:	Existing methods rely on text retrieval and geographic knowledge bases to generate coordinates, and they are prone to error propagation and dependency on structured knowledge bases.
Approach:	They propose to use large language models to convert geographic coordinates into geohash sequences and introduce a Chain-of-Thought mechanism to enhance the model’s reasoning over spatial relationships.
Outcome:	The proposed framework can handle explicit address queries in single-point predictions and effectively resolve vague relative location queries.

Copied to clipboard

Zihan Liu, Jianhui li, Zexin Wang, Fei Sun, Jingjing LI, Zheyuan Li, Ke Xiang, Hang Cui, Houhua Gong, Changhua Pei, Gaogang Xie

Preference Estimation via Opponent Modeling in Multi-Agent Negotiation (2026.findings-acl)

Challenge:	Evidence-intensive reports often produce fluent but under-supported drafts . eviReport is an evidence-grounded workflow for automated long-form report generation .
Approach:	They propose an evidence-tracked workflow that organizes corpus evidence into compact, traceable units and retrieves query-relevant subgraphs into retrieval-ready packages.
Outcome:	The proposed workflow outperforms baselines in factual coverage, factual accuracy and visual evidence integration.

Copied to clipboard

Yuta Konishi, Kento Yamamoto, Eisuke Sonomoto, Rikuho Takeda, Ryo Furukawa, Yusuke Muraki, Takafumi Shimizu, Kazuma Fukumura, Yuya Kanemoto, Takayuki Ito, Shiyao Ding

Challenge:	Existing numerical-only approaches fail to capture qualitative information embedded in natural language interactions, resulting in unstable and incomplete preference estimation.
Approach:	They propose a preference estimation method that integrates natural language information into a Bayesian opponent modeling framework.
Outcome:	The proposed method improves agreement rate and preference estimation accuracy by integrating probabilistic reasoning with natural language understanding.

AutoSearch: Adaptive Search Depth for Efficient Agentic RAG via Reinforcement Learning (2026.findings-acl)

Copied to clipboard

Jingbo Sun, Wenyue Chong, Songjun Tu, Qichao Zhang, Yaocheng Zhang, Jiajun Chai, Xiaohan Wang, Wei Lin, Guojun Yin, Dongbin Zhao

Beneficial Reasoning Behaviors in Agentic Search and Effective Training Methods to Obtain Them (2026.findings-acl)

Challenge:	Prior work limits search depth to reduce cost, but this often leads to underexploration of complex questions.
Approach:	They propose a reinforcement learning framework that evaluates each search step via self-generated intermediate answers.
Outcome:	Extensive experiments on multiple benchmarks show that AutoSearch achieves a superior accuracy-efficiency trade-off, alleviating over-searching while preserving search quality.

Copied to clipboard

Jiahe Jin, Abhijay Sai Paladugu, Chenyan Xiong

Beyond Prompt: Fine-grained Simulation of Cognitively Impaired Standardized Patients via Stochastic Steering (2026.findings-acl)

Challenge:	Agentic search requires large language models to perform multi-step searches to solve complex information needs.
Approach:	They propose a training approach that equips agentic search models with reasoning behaviors before reinforcement learning (RL) they compare successful and failed trajectories and propose supervised fine-tuning and standard RL .
Outcome:	The proposed approach outperforms direct RL by 37.2% on three web benchmarks and 6.2% on seven multi-hop QA benchmarks.

Copied to clipboard

Weikang Zhang, Zimo Zhu, Zhichuan Yang, Chen Huang, Wenqiang Lei, See-Kiong Ng

Where Did It Go Wrong? Capability-Oriented Failure Attribution for Vision-and-Language Navigation Agents (2026.findings-acl)

Challenge:	Existing methods for training patients with cognitive impairment rely on discrete prompt engineering and fail to capture the heterogeneity of deficits across domains and severity levels.
Approach:	They propose to use steering vectors from contrastive pairs of instructions and responses to capture domain-specific features and introduce a Stochastic Token Modulation mechanism to regulate the intervention probability.
Outcome:	The proposed model outperforms baselines in clinical authenticity and severity controllability while remaining open-source.

Copied to clipboard

Jianming Chen, Yawen Wang, Junjie Wang, Xiaofei Xie, Shoubin Li, Qing Wang, Fanjiang Xu

MultiFileTest: A Multi-File-Level LLM Unit Test Generation Benchmark and Impact of Error Fixing Mechanisms (2026.findings-acl)

Challenge:	Existing testing methods are system-level and provide limited insight into which capability deficiencies cause task failures.
Approach:	They propose a capability-oriented testing approach that enables failure detection and attribution by seed selection and mutation.
Outcome:	The proposed method detects more failure cases and pinpoints capability-level deficiencies than state-of-the-art baselines, providing more interpretable and actionable guidance for improving embodied agents.

Copied to clipboard

Yibo Wang, Congying Xia, Wenting Zhao, Jiangshu Du, Chunyu Miao, Zhongfen Deng, Philip S. Yu, Chen Xing

Challenge:	Existing evaluation benchmarks for LLM unit test generation focus on function-level code rather than on more practical, challenging multi-file codebases.
Approach:	They propose a multi-file-level benchmark for unit test generation covering Python, Java, and JavaScript.
Outcome:	The proposed benchmarks show that most LLMs exhibit moderate performance on MultiFileTest, highlighting the benchmark’s inherent difficulty.

SAMoRA: Semantic-Aware Mixture of LoRA Experts for Task-Adaptive Learning (2026.findings-acl)

Copied to clipboard

Boyan Shi, Wei Chen, Shuyuan Zhao, Junfeng Shen, Shengnan Guo, Shaojiang Wang, Huaiyu Wan

DualAlign: Generating Clinically Grounded Synthetic Data (2026.findings-acl)

Challenge:	Existing methods for multitask learning fail to match input semantics with expert capabilities, leading to weak expert specialization.
Approach:	They propose a parameter-efficient mixture-of-experts framework for task-adaptive learning that aligns textual semantics with the most suitable experts for precise routing.
Outcome:	The proposed framework outperforms the state-of-the-art methods and holds excellent task generalization capabilities.

Copied to clipboard

Rumeng Li, null XWang, Hong yu

Breaking the Evaluation Paradox: Evaluating High-Entropy Search with Computationally Irreducible Constraints (2026.findings-acl)

Challenge:	Large language models (LLMs) can generate fluent clinical text, but ensuring that such outputs are clinically grounded and useful for downstream modeling remains challenging.
Approach:	They propose a disease-agnostic framework for generating privacy-preserving, clinically faithful synthetic EHR narratives.
Outcome:	The proposed framework produces context-aware, symptom-rich sentences that more closely reflect real-world clinical documentation.

Copied to clipboard

Juntao Wu, Wei Wen, Xianting Huang, Shuai Pang, Ruizhi Qiao, Xing Sun, Ke Wang

CoT-Edit: Reinforcement Learning of Chain-of-Thought Reasoning for Code Edit Suggestion (2026.findings-acl)

Challenge:	a new framework for evaluation of exhaustive search capabilities is needed . high-entropy enumeration tasks make such ground truth impossible for humans to create . VERITAS is a framework built on the principle of computationally irreducible constraints .
Approach:	They propose a framework that uses non-optimizable constraints to create verifiable searches . VERITAS can generate infinite number of test cases with perfect ground truth and precise difficulty control .
Outcome:	a new evaluation framework for large language models is based on non-optimizable constraints . the framework can generate infinite number of test cases with perfect ground truth and precise difficulty control .

Copied to clipboard

Wuya Chen, Yihao Yang, Yue Lin

MemORAI: Memory Organization and Retrieval via Adaptive Graph Intelligence for LLM Conversational Agents (2026.findings-acl)

Challenge:	Current code edit models require continuous human guidance to maintain context coherence, thereby disrupting programming flow and increasing cognitive load.
Approach:	They propose a reinforcement learning framework that guides LLMs to discover chain-of-thought (CoT) reasoning paths for code editing without requiring human-annotated CoT data.
Outcome:	The proposed framework outperforms baselines on an industrial dataset and achieves 60.2% edit accuracy.

Copied to clipboard

Hung Pham Van, Nguyen Manh Hieu, Khang Pham Tran Tuan, Nam Le Hai, Linh Ngo Van, Nguyen Thi Ngoc Diep, Trung Le

MDC-Bench: A Multidisciplinary Causal Benchmark Based on Causal Structures for Evaluating Large Language Models (2026.findings-acl)

Challenge:	Existing graph-based memory systems suffer from information dilution, absent provenance tracking, and uniform retrieval that ignores query context.
Approach:	They propose a framework that integrates memory organization and retrieval via a Graph Intelligence framework.
Outcome:	Evaluated on LOCOMO and LongMemEval benchmarks, MemORAI achieves state-of-the-art performance in memory retrieval and personalized response generation.

Copied to clipboard

Peng Wang, Yuxiong Yan, Xiao Ding, Kai Xiong, Bibo Cai, Chao Peng, Yutai Hou, Dandan Tu, Bing Qin, Ting Liu

MathAgent: Adversarial Evolution of Constraint Graphs for Mathematical Reasoning Data Synthesis (2026.findings-acl)

Challenge:	Existing causal datasets focus on the commonsense domain, but LLMs perform poorly when answering complex questions.
Approach:	They propose a multidisciplinary causal evaluation benchmark to assess LLMs' knowledge and skills.
Outcome:	The proposed model improves in domain specialization, structural diversity, and task complexity.

Copied to clipboard

Zixiong Yu, Jun Rao, Guhan Chen, Songtao Tian, Bohan Li, Jiansheng Wei, Min Zhang, Xiaojun Meng

Neural Induction of Finite-State Transducers (2026.findings-acl)

Challenge:	Current approaches to synthesising high-quality mathematical reasoning data without human priors suffer from mode collapse and limited logical complexity.
Approach:	They propose a hierarchical synthesis framework that formulates data synthesis as an unsupervised optimization problem over a constraint graph followed by semantic instantiation rather than a direct text generation task.
Outcome:	The proposed framework outperforms widely-used datasets on eight mathematical benchmarks.

Copied to clipboard

Michael Ginn, Alexis Palmer, Mans Hulden

Challenge:	Existing methods to construct finite-state transducers by hand are difficult and require domain knowledge and significant human effort.
Approach:	They propose a method for automatically constructing unweighted FSTs following the hidden state geometry learned by a recurrent neural network.
Outcome:	The proposed method outperforms classical transducer learning algorithms by up to 87% accuracy on held-out test sets.

Dynamic Sampling that Adapts: Self-Aware Iterative Data Persistent Optimization for Mathematical Reasoning (2026.findings-acl)

Copied to clipboard

Jun Rao, Xuebo Liu, Hexuan Deng, Zepeng Lin, Zixiong Yu, Jiansheng Wei, Xiaojun Meng, Min Zhang

Forge: Quality-Aware Reinforcement Learning for NP-Hard Optimization in LLMs (2026.findings-acl)

Challenge:	Current data selection paradigms rely on static, externally defined metrics, which fail to adapt to the evolving capabilities of models during training.
Approach:	They propose a dynamic sampling framework that aligns training data with the model's intrinsic competence by iterating on real-time feedback.
Outcome:	Extensive experiments on eight benchmarks show that SAI-DPO outperforms static baselines at most nearly 6 points, achieving state-of-the-art efficiency with significantly less data.

Copied to clipboard

Xiaozhe Li, Xinyu Fang, Shengyuan Ding, Yang Li, Linyang Li, Haodong Duan, Qingwen Liu, Kai Chen

Datasets for Scientific Literature Understanding: A Survey (2026.findings-acl)

Challenge:	Existing benchmarks focus on correctness, overlooking optimality . large language models excel at math, coding, logic and puzzles .
Approach:	They propose a framework for training and evaluating Large Language Models on NP-hard optimization problems through quality-aware RLVR.
Outcome:	The proposed framework outperforms existing benchmarks on math, coding, logic and puzzles.

Copied to clipboard

Yuanzhe Zhang, Xun Zhao, Maodi Hu, Xi Sun, Donghuan Song, Zhixiong Zhang

Challenge:	Empowering machines to understand scientific literature is crucial for accelerating scientific discovery and advancing the AI for Science paradigm.
Approach:	They propose a systematic taxonomy that organizes resources spanning structural understanding, text understanding, multimodal understanding and pre-training/instruction fine-tuning.
Outcome:	The proposed taxonomy organizes resources spanning structural understanding, text understanding, multimodal understanding and pre-training/instruction fine-tuning.

ODASim: Ordered, Distinctive and Absolute Semantic Similarity for Code Explanation Evaluation (2026.findings-acl)

Copied to clipboard

Prince Kumar, Vitobha Munigala, Jaydeep Sen, Ashish Mittal, Vishwajeet Kumar, Srikanth G. Tamilselvam

Almieyar-Oryx-BloomBench: A Bilingual Multimodal Benchmark for Cognitively Informed Evaluation of Vision-Language Models (2026.findings-acl)

Challenge:	Existing methods for code explanations fail to distinguish correct from partially or fully incorrect explanations and their similarity scores are poorly calibrated.
Approach:	They propose a model-agnostic graded fine-tuning framework that learns calibrated similarity representations between code and explanations to support fine-grained supervision and evaluation.
Outcome:	The proposed framework improves F1 score and ECE scores on two embedding models and reduces expected calibration error.

Copied to clipboard

Mohammad Mahdi Abootorabi, Omid Ghahroodi, Anas Madkoor, Marzia Nouri, Doratossadat Dastgheib, Ehsaneddin Asgari

OPT-BENCH: Evaluating the Iterative Self-Optimization of LLM Agents in Large-Scale Search Spaces (2026.findings-acl)

Challenge:	Existing evaluations focus on piecemeal or disconnected tasks, obscuring critical cognitive weaknesses and providing little insight for targeted improvement.
Approach:	They propose a bilingual, cognitively human-grounded multimodal benchmark for VLMs that evaluates six levels of cognition through carefully designed image–question–answer tasks.
Outcome:	The proposed framework ensures scalability, cultural inclusivity, and linguistic fidelity.

Copied to clipboard

Xiaozhe Li, Jixuan Chen, Xinyu Fang, Shengyuan Ding, Haodong Duan, Qingwen Liu, Kai Chen

VocabTailor: Dynamic Vocabulary Selection for Downstream Tasks in Small Language Models (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have demonstrated remarkable capabilities in reasoning and tool use, but their ability to continuously refine solutions in response to dynamic environmental feedback remains underexplored.
Approach:	They propose a benchmark to evaluate self-improvement capabilities in large-scale search spaces by combining 20 machine learning tasks with 10 classic NP-hard problems.
Outcome:	The proposed framework emulates human-like cognitive adaptation and operates via a general perception–memory–reasoning loop, iteratively refining solutions based on environmental feedback.

Copied to clipboard

Hanling Zhang, Yayu Zhou, Tongcheng Fang, Zhihang Yuan, Guohao Dai, Wanli Ouyang, Yu Wang

Feedback Adaptation for Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	Existing static vocabulary pruning designs that reduce memory usage suffer from rigid, one-size-fits-all designs that cause information loss during the prefill stage and lack flexibility.
Approach:	They propose a decoupled dynamic vocabulary selection framework that addresses memory constraints through offloading embedding and implements a hybrid static-dynamic vocabulary selection strategy for LM Head.
Outcome:	The proposed framework reduces memory usage by 99% with minimal or no degradation in performance.

Copied to clipboard

Jihwan Bang, Seunghan Yang, Kyuhong Shim, Simyung Chang, Juntae Lee, Sungha Choi

From Implicit to Explicit: Token-Efficient Logical Supervision for Mathematical Reasoning in LLMs (2026.findings-acl)

Challenge:	Existing evaluation protocols focus on overall accuracy and fail to capture how systems adapt after feedback is introduced.
Approach:	They propose to use feedback adaptation as a problem setting for RAG systems . they propose a minimal inference-time instantiation that incorporates feedback without retraining .
Outcome:	The proposed evaluations show that training-based approaches exhibit a trade-off between delayed correction and reliable adaptation.

Copied to clipboard

Shaojie Wang, Liang Zhang

When Backdoors Go Beyond Triggers: Semantic Drift in Diffusion Models Under Encoder Attacks (2026.findings-acl)

Challenge:	Recent studies reveal large language models lack logical reasoning abilities . logical relationship understanding is a core capability underlying genuine logical thinking .
Approach:	They propose a lightweight training framework targeting logical relationship understanding . they propose logical relation understanding framework that provides explicit supervision .
Outcome:	The proposed framework outperforms CoT-SFT training frameworks in logical relationship understanding tasks.

Copied to clipboard

Shenyang Chen, Liuwan Zhu

Reducing Peak Memory Usage for Modern Multimodal Large Language Model Pipelines (2026.findings-acl)

Challenge:	Existing studies on text-to-image models only measure trigger activation and visual fidelity, but backdoors can induce semantic corruption.
Approach:	They propose a framework that measures internal embedding drift and downstream functional misalignment.
Outcome:	The proposed model is vulnerable to backdoor attacks, the authors show . they show that the model is optimized to generate the desired style for benign prompts .

Copied to clipboard

Junwan Kim, Hyunkyung Bae

MAKI: Multi-layer Aligned Knowledge Injection for Structure-aware Knowledge Graph Completion with Large Language Models (2026.findings-acl)

Challenge:	Existing methods to reduce memory usage of multimodal large language models rely on storing large numbers of vision tokens in the key–value cache . however, such compression is typically only applied after all inputs are processed, resulting in high peak memory usage during the prefill stage.
Approach:	They propose a sequential input-compression mechanism that enforces a fixed memory budget by performing structure-aware key–value cache compression during the prefill stage.
Outcome:	The proposed approach reduces peak memory usage while maintaining generative performance with only minimal degradation, enabling more practical and memory-efficient multimodal inference.

Copied to clipboard

Zhiwen Xie, Xin Wang, Guangyou Zhou, Derek F. Wong

From Domains to Instances: Dual-Granularity Data Synthesis for LLM Unlearning (2026.findings-acl)

Challenge:	Existing knowledge graph completion methods struggle to capture structural information in knowledge graphs (KGs) Existing approaches for KGC focus on learning representations of entities and relations through observed structural patterns.
Approach:	They propose a multi-layer Aligned Knowledge Injection model that tightly integrates structured KG information into LLMs through multi-layered alignment.
Outcome:	The proposed method outperforms state-of-the-art methods on benchmark datasets.

Copied to clipboard

Xiaoyu Xu, Minxin Du, Zitong LI, Zi Liang, Zhibiao Guo, Zhang Shiyu, Peizhao Hu, Qingqing Ye, Haibo Hu

Challenge:	Currently, the evaluation of unlearning is limited due to the lack of granularity in the model.
Approach:	They propose a framework for synthesizing high-quality forget sets that exploits the target model per se to elicit data that matches its internal knowledge distribution through seed-guided and adversarial prompting.
Outcome:	The proposed framework achieves a superior balance of relevance, diversity, and efficiency across benchmarks.

ERRV: Eliciting Efficient Reasoning through Reasoning Vectors for Policy Optimization in Large Language Models (2026.findings-acl)

Copied to clipboard

Zhuowen Han, Lei Yang, Renren Jin, Dan Shi, Chenxi Sun, Deyi Xiong

LLM-induced Rationales for More Compact Explainable Style Classification Models (2026.findings-acl)

Challenge:	Existing efforts to improve reasoning efficiency of large language models focus on modifying the reinforcement learning reward, such as adding length penalties.
Approach:	They propose a training framework that elicits efficient reasoning through reasoning vectors and a framework that allows the model to generate high-quality responses during reinforcement learning.
Outcome:	The proposed framework reduces reasoning length by 30% while maintaining stability, while retaining high accuracy.

Copied to clipboard

Ahmad Aljanaideh, Saeb Ganideh

MDTeamGPT: Mitigating Context Collapse and Enabling Self-Evolution in Medical Multi-Agent Reasoning (2026.findings-acl)

Challenge:	Existing methods for extracting explanations from complex models are based on discovering a large number of features, and this affects interpretability.
Approach:	They propose a model that leverages Large Language Models and clustering algorithms to discover a compact set of interpretable features.
Outcome:	The proposed model reduces the number of features on 3 Style Classification tasks by 85–99% while reducing the number by 85.

Copied to clipboard

Kai Chen, Xinfeng Li, Tianpei Yang, Hewei Wang, Guang Yang, Jing Huo, Yang Gao

TSPO: Breaking the Double Homogenization Dilemma in Multi-turn Search Policy Optimization (2026.findings-acl)

Challenge:	Long, multi-round, multirole interaction trajectories lead to severe information dilution and context window overload, triggering context collapse which destabilizes reasoning.
Approach:	They propose a multi-agent framework that compresses and reorganizes multi-round consensus.
Outcome:	The proposed framework outperforms baselines across text-based and multimodal tasks while demonstrating superior diagnostic performance and stability in complex clinical scenarios.

Copied to clipboard

Shichao Ma, Zhiyuan Ma, Ming Yang, Xiaofan Li, Xing Wu, Jintao Du, Yu Cheng, Weiqiang Wang, Qiliang Liu, Zhengyang Zhou, Yang Wang

Towards semantic reliable clinical QA: Query pipeline optimization for cancer patient question answering systems (2026.findings-acl)

Challenge:	Large Language Models (LLMs) can solve complex tasks through iterative information retrieval.
Approach:	They propose a turn-level stage-aware policy optimization approach to solve this problem . they introduce a first-occurrence latent reward mechanism to allocate partial rewards .
Outcome:	Experiments show that TSPO outperforms state-of-the-art models on Qwen2.5-3B and 7B models.

Copied to clipboard

MaoLin He, Rena Wei Gao, Mike Conway, Brian E. Chapman

LayerNorm Induces Recency Bias in Transformer Decoders (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are promising for medical Question-Answering but suffer from hallucinations that jeopardize patient safety.
Approach:	They propose a three-level controllable metadata-aware framework optimized for Cancer Patient QA (CPQA) they propose combining semantic retrieval with real-time Boolean search to overcome metadata blindness.
Outcome:	The proposed framework improves the answer accuracy of Claude-3-haiku by 5.24% over chain-of-thought prompting and about 3% over a naive RAG setup.

Copied to clipboard

Junu Kim, Xiao Liu, Zhenghao Lin, Lei Ji, Yeyun Gong, Edward Choi

Scaling Performance and Low-Resource Annotation with Many-Shot In-Context Learning for Named Entity Recognition (2026.findings-acl)

Challenge:	Existing studies show that stacking causal self-attention layers alone induces a positional bias in attention scores toward earlier tokens, but this differs from the bias toward later tokens observed in Transformer decoders, known as recency bias.
Approach:	They propose to stack causal self-attention layers and layer norm to induce recency bias in Transformer decoders by analyzing the interaction between causal self and other architectural components.
Outcome:	The proposed method provides new theoretical insights into how positional information interacts with architectural components and suggests improvements in positional encoding strategies.

Copied to clipboard

Qi Zhang, Fangping Lan, Cornelia Caragea, Longin Jan Latecki, Eduard Dragut

PhageBench: Can LLMs Understand Raw Bacteriophage Genomes? (2026.findings-acl)

Challenge:	Existing studies on ICL for Named Entity Recognition (NER) have mainly explored few-shot settings, but the potential of scaling to hundreds of demonstrations has not been thoroughly investigated.
Approach:	They evaluate various LLMs across multiple domains using hundreds of ICL examples and then assess the feasibility of using many-shot ICL as a data annotation framework.
Outcome:	The proposed framework can be scaled to hundreds of examples and annotate and refining data for low-resource NER tasks.

Copied to clipboard

Yusen Hou, Weicai Long, Haitao HU, Houcheng su, Junning Feng, Yanlin Zhang

POLAR: A Benchmark for Multilingual, Multicultural, and Multi-Event Online Polarization (2026.findings-acl)

Challenge:	phage genome annotation is a critical component of microbial ecosystems and antibiotics.
Approach:	They propose a benchmark to evaluate phage genome understanding by mirroring workflow of bioinformatics experts.
Outcome:	The benchmark outperforms baseline models in phage contig identification and host prediction.

Copied to clipboard

Usman Naseem, Robert Geislinger, Juan Ren, Sarah Kohail, Rudy Alexandro Garrido Veliz, P Sam Sahil, Yiran Zhang, Idris Abdulmumin, Marco Antonio Stranisci, Özge Alacam, Cengiz Acarturk, Aisha Jabr, Saba Anwar, Abinew Ali Ayele, Simona Frenda, Alessandra Teresa Cignarella, Elena Tutubalina, Oleg Rogov, Aung Kyaw Htet, Xintong Wang, Surendrabikram Thapa, Kritesh Rauniyar, Tanmoy Chakraborty, MD Arfeen Zeeshan, Dheeraj Kodati, Satya Keerthi, Sahar Moradizeyveh, Firoj Alam, Md Arid Hasan, Syed Ishtiaque Ahmed, Ye Kyaw Thu, Shantipriya Parida, Ihsan Ayyub Qazi, Lilian Diana Awuor Wanzare, Nelson Odhiambo Onyango, Clemencia Siro, Jane Wanjiru Kimani, Ibrahim Said Ahmad, Adem Chanie Ali, Martin Semmann, Chris Biemann, Shamsuddeen Hassan Muhammad, Seid Muhie Yimam

VANE: Guiding High-Value Exploration in RLVR via Outcome-Process Novelty Shaping (2026.findings-acl)

Challenge:	polarization is a pervasive threat to democratic institutions, civil discourse, and social cohesion worldwide . most existing datasets focus on English or high-resource languages, reflecting a widespread trend across NLP tasks .
Approach:	They propose a multilingual, multicultural, and multi-event dataset with over 110K instances in 22 languages drawn from diverse online platforms and real-world events.
Outcome:	The proposed dataset analyzes polarization detection, type, and manifestation using a variety of annotation platforms adapted to each cultural context.

Copied to clipboard

Xu He, Jialiang Guo, Fucheng Xiong, Haodong Zhao, Xingyang li, Ke Zeng, Xunliang Cai

Exploring Coding Spot: Understanding Parametric Contributions to LLM Coding Performance (2026.findings-acl)

Challenge:	Extensive experiments on large-scale mathematical reasoning and out-of-distribution tasks demonstrate the effectiveness and generalization of the proposed method.
Approach:	They propose a method that quantifies novelty across the outcome space and semantic process space by using reward or solution divergence.
Outcome:	Experiments on Qwen2.5-Math-7B demonstrate the proposed method is general and efficient.

Copied to clipboard

Dongjun Kim, Minhyuk Kim, Yongchan Chun, Chanjun Park, Heuiseok Lim

FedGUI: Benchmarking Federated GUI Agents across Heterogeneous Platforms, Devices, and Operating Systems (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have demonstrated proficiency in code generation and comprehension across multiple programming languages.
Approach:	They propose a parameter-localized subset of LLMs that facilitates coding capabilities.
Outcome:	The proposed model significantly improves performance on coding tasks while preserving non-coding functionalities.

Copied to clipboard

WenHao Wang, Haoting Shi, Mengying Yuan, Yiquan Lin, Panrong Tong, Hanzhang Zhou, Guangyi Liu, Pengxiang Zhao, Yue Wang, Siheng Chen

Towards Proactive Information Probing: Customer Service Chatbots Harvesting Value from Conversation (2026.findings-acl)

Challenge:	a lack of benchmarks capture real-world, cross-platform heterogeneity in GUI training . traditional methods to train GUI agents rely on centralized data collection and manual labeling .
Approach:	They propose a benchmark for developing and evaluating federated GUI agents across mobile, web and desktop platforms.
Outcome:	The proposed benchmarks show that cross-platform collaboration improves performance and identify platform and OS as the most influential factors.

Copied to clipboard

Chen Huang, Zitan Jiang, Zou Changyi, Wenqiang Lei, See-Kiong Ng

Benchmarking Vision-Language Models on Chinese Ancient Documents: From OCR to Knowledge Reasoning (2026.findings-acl)

Challenge:	a new technology is transforming customer service chatbots into strategic bridges for business intelligence . a recent study shows that customer service bots are increasingly being used as reactive support tools .
Approach:	They propose a task of Proactive Information Probing which optimizes when to probe users for pre-specified information while minimizing conversation turns and user friction.
Outcome:	The proposed framework outperforms baselines in both information probing and service quality.

Copied to clipboard

Haiyang Yu, Yuchuan Wu, Fan Shi, Jinghui Lu, Ke Niu, Xiaodong Ge, Minghan Zhuo, Jingqun Tang, Bin Li

DaMo: Data Mixing Optimizer in Fine-tuning Multimodal LLMs for Mobile Phone Agents (2026.findings-acl)

Challenge:	Existing document benchmarks focus on English printed texts or simplified Chinese . current vision-language models struggle with visual complexity and poor adaptability .
Approach:	They propose a benchmark to evaluate Chinese ancient documents' visual/linguistic complexity . ancient documents are valuable cultural heritage, but they face challenges in digitization and understanding .
Outcome:	the first benchmark for Chinese ancient documents evaluates VLMs from OCR to knowledge reasoning . ancient documents carry thousands of years of Chinese history and culture . traditional methods only scan images, while current models struggle with visual complexity .

Copied to clipboard

Kai Shi, Jun Yang, Ni Yang, Binqiang Pan, Qingsong Xie, null Zhangchao, Zhenyu Yang, Tianhuang Su, Haonan Lu

From Outcome to Process: Optimizing MoE Load Balancing with MCTS (2026.findings-acl)

Challenge:	Mobile Phone Agents (MPAs) have attracted huge attention due to their practicability in a multitude of scenarios.
Approach:	They propose a data mixture optimization solution that extrapolates optimal data mixtures from a trainable network.
Outcome:	The proposed model outperforms existing methods on open-source benchmarks and on open source benchmarks.

Copied to clipboard

Wenjun Ke, Hengyuan Xu, Ziyu Shang, Yao He, Jiahao Wang, Zijie Xu, Peng Wang, Yuhang Lou, Jiajun Liu

Learning Optimal Message Representations for Agentic Communication (2026.findings-acl)

Challenge:	Existing balancing strategies focus on constraining the final distribution of expert usage, but overlook the routing decisions made at each layer.
Approach:	They propose a three-stage framework that leverages process-level rewards to guide balanced expert routing.
Outcome:	Extensive experiments show that LayerMoE improves the performance of state-of-the-art LoRA-MoA baselines, yielding an average accuracy gain of 1.39%.

Copied to clipboard

Shashwat Gupta, Anson Bastos, Mayukh Das, Supriyo Ghosh, Nagarajan Natarajan, Chetan Bansal, Saravan Rajmohan

CARO: Chain-of-Analogy Reasoning Optimization for Robust Content Moderation (2026.findings-acl)

Challenge:	Existing approaches lack the intelligence necessary to understand, learn or apply optimal communication representations adaptively.
Approach:	They propose to dynamically learn the optimal message representations to enhance agentic performance by using an Expanding Markov Decision Process.
Outcome:	The proposed framework improves agentic performance while maintaining efficiency.

Copied to clipboard

Bingzhe Wu, Haotian Lu, Yuchen Mou

N-GRPO: Embedding-Level Neighbor Mixing for Enhanced Policy Optimization (2026.findings-acl)

Challenge:	Current large language models struggle with ambiguous content moderation cases due to misleading "decision shortcuts" . authors propose a two-stage training framework to induce robust analogical reasoning in LLMs .
Approach:	They propose a two-stage training framework to induce robust analogical reasoning in LLMs . they bootstrap analogy reasoning chains via retrieval-augmented generation and SFT .
Outcome:	The proposed framework outperforms state-of-the-art reasoning models and specialized moderation models on ambiguous moderation benchmarks.

Copied to clipboard

Xukun Zhu, Hang Yu, Peng Di, Linchao Zhu

GFT: From Imitation to Reward Fine-Tuning with Unbiased Group Advantages and Dynamic Coefficient Rectification (2026.findings-acl)

Challenge:	Recent studies show that Large Language Models can generate diverse solutions during the rollout phase.
Approach:	They propose a new approach that leverages Semantic Neighbor Mixing to generate diverse input representations by mixing anchor tokens and nearest semantic neighbors.
Outcome:	Experimental results show that the proposed approach improves on strong baselines and generalizes on out-of-distribution tasks.

Copied to clipboard

Wangjie Gan, Miao Pan, Linbo Xi, Wenqi Zhang, Jintao Chen, Jianwei Yin, Xuhong Zhang

SciImpact: A Multi-Dimensional, Multi-Field Benchmark for Scientific Impact Prediction (2026.findings-acl)

Challenge:	Existing studies have demonstrated that supervised fine-tuning and reinforcement learning are effective in integrating knowledge injection with robust generalization.
Approach:	They propose a unified post-training framework that addresses intrinsic limitations of supervised fine-tuning and reinforcement learning.
Outcome:	The proposed framework surpasses SFT-based methods and yields policies that integrate more smoothly with subsequent RL training.

Copied to clipboard

Hangxiao Zhu, Yuyu Zhang, Ping Nie, Yu Zhang

Challenge:	Prior work on scientific impact prediction has focused on citation counts and its variants, leaving limited evaluation of models’ capability to reason about other dimensions.
Approach:	They propose a large-scale, multi-dimensional benchmark for scientific impact prediction spanning 19 fields.
Outcome:	The proposed model outperforms larger models and close-source models in a wide range of fields and measures of scientific impact across 19 fields.

When Helpers Become Hazards: A Benchmark for Analyzing Multimodal LLM-Powered Safety in Daily Life (2026.findings-acl)

Copied to clipboard

Xinyue Lou, Xu Jinan, Jingyi Yin, Xiaolong Wang, Zhaolu Kang, null Liaoyouwei, Yixuan Wang, Xiangyu Shi, Fengran Mo, SU Yao, Kaiyu Huang

MM-ShiftKV: Decode-Aware Prefill-Stage KV Selection for Multimodal Large Language Models (2026.findings-acl)

Challenge:	Safety impact of Multimodal Large Language Models (MLLMs) on human behavior is evaluated in this study.
Approach:	They propose a safety-warning-based evaluation framework that encourages models to provide clear and informative safety warnings, rather than generic refusals.
Outcome:	The proposed safety-warning-based evaluation framework encourages models to provide clear and informative safety warnings, rather than generic refusals.

Copied to clipboard

Jinsong Shu, Chenyang Wu, Zhongle Xie, Baokun Wang, Lidan Shou

Measuring What Matters Beyond Text: Evaluating Multimodal Summaries by Quality, Alignment, and Diversity (2026.findings-acl)

Challenge:	Recent work suggests a prefill-stage KV cache selection method to estimate KV importance from prefilling statistics.
Approach:	They propose a training-free, decode-aware and strictly prefill-only KV selection method that retains key-value caching for decoding .
Outcome:	The proposed method outperforms existing methods under tight cache budgets on multimodal benchmarks.

Copied to clipboard

Abid Ali, Diego Molla, Usman Naseem

MonCulture-Eval: A Hierarchical Benchmark for Evaluating Mongolian Cultural Capabilities of Large Language Models across Scripts and Regions (2026.findings-acl)

Challenge:	MLLMs have facilitated multimodal summarization with multimodal outputs, but their evaluation is fragmented . MM-Eval integrates assessments of textual quality, cross-modal alignment, and visual diversity .
Approach:	They propose a unified evaluation framework that integrates assessments of textual quality, cross-modal alignment, and visual diversity.
Outcome:	The proposed framework improves over heuristic aggregation baselines and provides an interpretable, reference-weak framework for comparative evaluation of multimodal summaries.

Copied to clipboard

Quulgan Minggad, Xiao Zinan, Yuan Sun

Less is More: Knowledge-Aware Compression for Long Legal Judgment Prediction (2026.findings-acl)

Challenge:	Large Language Models excel at multilingual translation and instruction-following in low-resource settings like Tibetan, but lack cultural intelligence quantification.
Approach:	They propose a benchmark to assess the cultural intelligence of Large Language Models in Mongolia . they use a three-layer cognitive hierarchy and specialized tasks to assess their cultural intelligence .
Outcome:	The monCulture-Eval benchmark assesses the cultural intelligence of large language models in the Mongolian context across two writing systems and three regional sub-cultures.

Copied to clipboard

Fanghao Lou, Qiqi Wang, Guanyu Chen, Senbo Zhang, Kaiqi Zhao, Qian Liu, Huijia Li

Towards Visually Grounded Multimodal Summarization via Cross-Modal Transformer and Gated Attention (2026.findings-acl)

Challenge:	Recent advances leverage large language models (LLMs) for legal reasoning, but they face high computational costs and information degradation when handling long cases.
Approach:	They propose a framework that selectively retains legally relevant information while reducing redundant or less informative content, enabling efficient and accurate long-context reasoning.
Outcome:	The proposed framework outperforms existing methods on four real-world datasets spanning multiple jurisdictions and languages.

Copied to clipboard

Abid Ali, Diego Molla, Usman Naseem

Fairness Evaluation and Inference Level Mitigation in LLMs (2026.findings-acl)

Challenge:	Existing methods for multimodal summarization often inject shallow visual features into deep models, leading to representational mismatches and weak cross-modal grounding.
Approach:	They propose a framework that performs text summarization and representative image selection . a deep visual processor aligns the visual encoder with the language model at corresponding depths .
Outcome:	The proposed framework produces more accurate, visually grounded summaries and selects more representative images.

Copied to clipboard

Afrozah Nadeem, Mark Dras, Usman Naseem

SCOPE: Preserving Modality-Specific Cues to Mitigate Modality Laziness in Multimodal Learning (2026.findings-acl)

Challenge:	Large language models display undesirable behaviors embedded in their internal representations, undermining fairness, inconsistency drift, and the propagation of unwanted patterns during extended dialogues.
Approach:	They propose a pruning-based framework that detects context-aware neuron activations and applies adaptive masking to modulate their influence during generation.
Outcome:	The proposed framework detects context-aware neuron activations and applies adaptive masking to modulate their influence during generation.

Copied to clipboard

Jingfan Yang, Rui Zhang, Liang Hong, Ke Yuan

Embedding-based In-Context Prompt Training for Enhancing LLMs as Text Encoders (2026.findings-acl)

Challenge:	Existing approaches to learning multimodal representations emphasize shared semantics and overlook modality-specific cues.
Approach:	They propose a framework for learning complete multimodal representations using shared and practical cues.
Outcome:	SCOPE outperforms SOTA benchmarks on four datasets and achieves 27.10% accuracy improvement.

Copied to clipboard

Ailiang Lin, Zhuoyun Li, Keyu Mao, Kotaro Funakoshi, Manabu Okumura

Creating ConLangs to Probe the Metalinguistic Grammatical Knowledge of LLMs (2026.findings-acl)

Challenge:	Large language models (LLMs) have been widely explored for embedding generation.
Approach:	They propose an embedding-based in-context prompt training strategy that leverages in-constext learning to generate high-quality embeddables while reducing computational burden.
Outcome:	The proposed method surpasses models trained on publicly available retrieval data and achieves state-of-the-art embedding performance on the MTEB benchmark.

Copied to clipboard

Chihiro Taguchi, Richard Sproat

ReasonIF: Large Reasoning Models Fail to Follow Instructions During Reasoning (2026.findings-acl)

Challenge:	'ConLang' is a term used to describe any artificially created language intended to be as expressive as naturally evolved human languages.
Approach:	They propose to use large language models to create a modular system that uses LLMs as a tool in the development of Constructed Languages.
Outcome:	The proposed system creates phonology, morphology and syntax, lexicon, orthography, and grammatical handbook using module-specific sets of prompts.

Copied to clipboard

Yongchan Kwon, Shang Zhu, Federico Bianchi, Kaitlyn Zhou, James Zou

An LLM-Embedding Semantic Adaptation Network for Post-level Semantic Drift Evaluation (2026.findings-acl)

Challenge:	Prior studies assess instruction adherence in the model’s main responses, but it is also critical for large reasoning models to follow user instructions throughout their reasoning process.
Approach:	They propose a systematic benchmark for assessing reasoning instruction following to assess the model's adherence to instructions.
Outcome:	The proposed benchmark reduces the risk of undesirable shortcuts, hallucinations, or reward hacking within reasoning traces.

Copied to clipboard

Ning Chen, Mingyu Kang, Jie Li, Linyuan Lü

CEAID: Benchmark of Multilingual Machine-Generated Text Detection Methods for Central European Languages (2026.findings-acl)

Challenge:	Evaluating semantic drift is essential for understanding discourse evolution and opinion formation in online discussions.
Approach:	They propose an LLM-embedding Semantic Adaptation Network to evaluate semantic drift . they use an LRU module, an LEM-Embedding graph convolutional network module and an adaptive fusion module to integrate features from event related posts.
Outcome:	The proposed model achieves state-of-the-art performance on the semantic drift evaluation task compared to baseline models.

Copied to clipboard

Dominik Macko, Jakub Kopál

Distilling the Essence, Discarding the Dross: Improving Fairness in Multimodal Large Language Models via Historical Reflection-Guided Prompt Optimization (2026.findings-acl)

Challenge:	Existing methods for machine-generated text detection are mostly focused on English . existing methods are almost unusable for non-English languages, leaving the transferability towards these languages unexplored.
Approach:	They propose to use a train-language combination to compare MGT detection methods . they focus on multi-domain, multi-generator, and multilingual evaluation .
Outcome:	The proposed methods are the most performant in the Central European languages and resistant against obfuscation.

Copied to clipboard

Juncheng Hu, Jiming Yu, Rui Song, Kedi Lyu, Yingji Li, Zheli Liu

Tree-of-Evidence: Efficient "System 2" Search for Faithful Multimodal Grounding (2026.findings-acl)

Challenge:	Existing approaches to debiase MLLMs rely on handcrafted prompts that are brittle and difficult to generalize across tasks and bias types.
Approach:	They propose an adaptive self-debiasing framework that optimizes task-specific debiasers to suppress stereotypical outputs.
Outcome:	The proposed framework suppresses stereotypical outputs while maintaining performance.

Copied to clipboard

Micky C. Nnamdi, Benoit Louis Marteau, Yishan Zhong, J. Ben Tamo, May Dongmei Wang

Jailbreaking Large Language Models with Morality Attacks (2026.findings-acl)

Challenge:	Attention-based methods fail to faithfully represent the model’s decision process when integrating heterogeneous modalities.
Approach:	They propose an inference-time search algorithm that frames interpretability as a discrete optimization problem.
Outcome:	The proposed algorithm retains over 98% of full-model AUROC with as few as five evidence units and achieves higher decision agreement and lower error than LIME, SHAP, saliency, and concept-bottleneck baselines under sparse budgets.

Copied to clipboard

Ying Su, Zheng Mingen, Weili Diao, Haoran Li

ExpSeek: Self-Triggered Experience Seeking for Web Agents (2026.findings-acl)

Challenge:	Pluralism alignment is the goal of creating AI that can coexist with and serve morally multifaceted humanity.
Approach:	They propose to use jailbreak attacks to manipulate LLMs’ judgment over pluralistic values by using a morality dataset with 10.4K instances.
Outcome:	The proposed method exploits the persuasion abilities of LLMs to produce moral content over pluralistic values.

Copied to clipboard

Wenyuan Zhang, Xinghua Zhang, Haiyang Yu, Shuaiyi Nie, Bingli Wu, Juwei Yue, Tingwen Liu, Yongbin Li

Challenge:	Existing methods for integrating experience into web agents are struggling to adapt to dynamically changing contextual observations during agent-environment interaction.
Approach:	They propose a model that shifts experience toward step-level proactive seeking by estimating step- level entropy thresholds and designing step-Level tailored experience content.
Outcome:	The proposed model achieves 9.3% and 7.5% performance improvements on Qwen3-8B and 32B models across four challenging web agent benchmarks.

Do Personality Traits Interfere? Geometric Limitations of Steering in Large Language Models (2026.findings-acl)

Copied to clipboard

Pranav Bhandari, Usman Naseem, Mehwish Nasim

AHA: Aligning Large Audio-Language Models for Reasoning Hallucinations via Counterfactual Hard Negatives (2026.findings-acl)

Challenge:	Existing studies on personality steering in large language models rely on injecting trait-specific steering vectors into the residual stream to control the strength of trait expression.
Approach:	They examine the geometric relationships between Big Five personality steering directions by applying geometric conditioning schemes to their steering vectors.
Outcome:	The proposed model can be used to steer personality traits in large language models.

Copied to clipboard

Yanxi Chen, Wenhui Zhu, Xiwen Chen, Zhipeng Wang, Xin Li, Peijie Qiu, Hao Wang, Xuanzhao Dong, Yujian Xiong, Anderson Schneider, Yuriy Nevmyvaka, Yalin Wang

Web Sitemap Knowledge Can Enhance Autonomous Browsing (2026.findings-acl)

Challenge:	Large Audio-Language Models suffer from hallucinations, e.g., generating text not grounded in the audio input.
Approach:	They propose a framework to address hallucination problems in large audio-language models . they use a preference dataset to test the model's accuracy .
Outcome:	The proposed model outperforms the latest SOTA methods in terms of performance and generalization.

Copied to clipboard

Yuyao Zhang, Hongyu Lu, Jiajie Jin, Hongjin Qian, Shiyu Li, Zhao Yang, Yutao Zhu, Ji-Rong Wen, Zhicheng Dou

Coarse-to-Fine Multimodal Information Selection for Video Speaking Style Recognition with Large Language Models (2026.findings-acl)

Challenge:	Existing web agents suffer from limited robustness, efficiency and task success due to lack of structural understanding of websites and lack of browsing priors in pre-trained models.
Approach:	They propose an agent-oriented sitemap protocol that integrates structured website knowledge into web agents.
Outcome:	The proposed agent-oriented sitemap improves robustness, efficiency and effectiveness without extra training.

Copied to clipboard

Beibei Zhang, Yanan Lu, Lin Fen, Tongwei Ren

Large Language Models Require Curated Context for Reliable Political Fact-Checking—Even with Reasoning and Web Search (2026.findings-acl)

Challenge:	Video speaking style recognition (VSSR) aims to classify conversations into different types . integrating all multimodal data yields suboptimal results, authors say .
Approach:	They propose a framework that allows users to obtain multimodal data via coarse-to-fine selection . they propose to use visual captions and textual dialogues to integrate multimodal information .
Outcome:	The proposed framework outperforms existing training-free approaches and most training-based methods on multiple datasets.

Copied to clipboard

Matthew R. DeVerna, Kai-Cheng Yang, Harry Yaojun Yan, Filippo Menczer

IREASONER: Trajectory-Aware Intrinsic Reasoning Supervision for Self-Evolving Large Multimodal Models (2026.findings-acl)

Challenge:	Large language models (LLMs) have raised hopes for automated end-to-end fact-checking, but prior studies report mixed results.
Approach:	They evaluate 15 large language models on 6,000 claims fact-checked by PolitiFact . standard models perform poorly, reasoning offers minimal benefits, and web search provides only moderate gains .
Outcome:	The models predict claim veracity and a curated RAG system improved macro F1 by 233% on average across model variants.

Copied to clipboard

Meghana Sunil, Manikandarajan Venmathimaran, Muthu Subash Kavitha

CLaRE-ty Amid Chaos: Quantifying Representational Entanglement to Predict Ripple Effects in LLM Editing (2026.findings-acl)

Challenge:	Existing self-evolving frameworks mainly reward final outcomes, leaving intermediate reasoning weakly constrained despite its importance for visually grounded decision making.
Approach:	They propose a framework that improves an LMM’s implicit reasoning by explicitly eliciting chain-of-thought (CoT) and rewarding its internal agreement.
Outcome:	The proposed framework yields +2.1 points across diverse multimodal reasoning benchmarks under fully unsupervised post-training.

Copied to clipboard

Manit Baser, Alperen Yildiz, Dinil Mon Divakaran, Mohan Gurusamy

STAR-Teaming: A Strategy-Response Multiplex Network Approach to Automated LLM Red Teaming (2026.findings-acl)

Challenge:	Large language models (LLMs) are outdated or incorrect over time due to unintended ripple effects that propagate even to the hidden space.
Approach:	They propose a lightweight representation-level technique to identify where ripple effects may occur by detecting entanglement between facts using forward activations from a single intermediate layer.
Outcome:	The proposed method achieves 62.2% improvement in Spearman correlation with ripple effects while being 2.74 faster and using 2.85 less peak GPU memory.

Copied to clipboard

Min Jae Jung, YongTaek Lim, Chaeyun Kim, Junghwan Kim, Kihyun Kim, Minwoo Kim

Critic Rule Induction: Improving Temporal Knowledge Graph Forecasting with Generator-Critic Language Models (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are susceptible to jailbreak prompts that can elicit harmful or inappropriate responses.
Approach:	They propose a black-box framework for automated red teaming that integrates a Multi-Agent System with a Strategy-Response Multiplex Network and employs network-driven optimization to sample effective attack strategies.
Outcome:	The proposed framework surpasses existing methods and achieves higher attack success rate (ASR) at lower computational cost.

Copied to clipboard

Yingsong Ning, Fu Zhang, Jingwei Cheng, Jiashun Peng, Xiaoke Wang

Context-Agent: Dynamic Discourse Trees for Non-Linear Dialogue (2026.findings-acl)

Challenge:	Existing methods for predicting future facts from time-evolving graphs rely on statistical co-occurrences and extensive path enumeration.
Approach:	They propose a Critic-Guided Rule Induction method which treats temporal rules as rule hypotheses to be examined and adopts a decoupled Generation-Discrimination pipeline to induce rules that are high-coverage and high-precision.
Outcome:	The proposed method outperforms strong baselines on three benchmarks and achieves state-of-the-art performance.

Copied to clipboard

Junan Hu, Shudan Guo, Wenqi Liu, Jianhua Yin, Yinwei Wei

Can LLMs Really Judge? A Progressive Argumentation-Mining Framework for Distinguishing Understanding from Aggregation (2026.findings-acl)

Challenge:	Existing approaches to managing non-linear dialogue flow are misaligned with the intrinsically hierarchical and branching structure of natural discourse.
Approach:	They propose a framework that models multi-turn dialogue history as a dynamic tree structure.
Outcome:	The proposed framework enhances task completion rates and improves token efficiency across various LLMs.

Copied to clipboard

Fuyu Wang, Jiangtong Li, Kun Zhu, Changjun Jiang

SMART: Semantic Header Flattening and Pseudo-Code-Style Reasoning for LLM-based Complex Table Question Answering (2026.findings-acl)

Challenge:	Existing evaluations of large language models rely on dataset-based generation accuracy . however, generative correctness does not guarantee discriminative capability to verify solutions .
Approach:	They propose a diagnostic framework that explicitly controls context and isolates discriminative behaviors.
Outcome:	The proposed framework explicitly controls context and isolates discriminative behaviors.

Copied to clipboard

Yongshuo Zhang, Jixiong Chen, Kaihe xu, Wei Wei, Shihao Zou

ConMA : Confidence-Guided Kernel Sampling with Multi-Stage Aggregation for LLM Reasoning (2026.findings-acl)

Challenge:	Existing approaches to complex table question answering rely on handcrafted table linearization or prompts . Existing methods rely only on hand-crafted table and require hierarchical hierarchies to align conditions, attributes, and values.
Approach:	They propose a framework that explicitly decouples table structure understanding from reasoning execution.
Outcome:	Experiments show that SMART improves accuracy and robustness of complex table question answering (TQA) . SMart decouples table structure understanding from reasoning execution, enabling state-of-the-art performance.

Copied to clipboard

Yinuo Wang, Qingjie Li, Wenyao Cui, Qiuchi Li, Zhang Huaping

Human-Centered Supervision for Sentiment Analysis in Telugu: A Systematic Inquiry Beyond Accuracy (2026.findings-acl)

Challenge:	Existing approaches to test-time scaling rely on external verifiers and one-shot independent sampling.
Approach:	They propose a test-time scaling framework that reallocates a fixed inference budget into iterative sample–filter–diversify–select cycles.
Outcome:	ConMA outperforms baselines on multiple benchmarks while converging early with only 18 samples on average, substantially reducing inference cost.

Copied to clipboard

Vallabhaneni Raj Kumar, Ashwin S, Supriya Manna, Niladri Sett, Cheedella V S N M S Hema Harshitha, Kurakula Harshitha, Basina Deepakraj, Anand Kumar Sharma, Tanuj Sarkar, Samanthapudi Shakeer, Bondada Navaneeth Krishna

Evidence-Aligned Entity Verification for Hallucination Detection in Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	a limited amount of annotated data has slowed progress in machine learning for low-resource languages . a sentiment label records an annotator's final decision, but it is not a valid record of the annotation's interpretation.
Approach:	They propose a large-scale Telugu sentiment classification dataset annotated with sentiment labels and human-selected rationales from multiple native speakers.
Outcome:	The proposed model improves classification performance, explanation quality, and social bias by incorporating human rationales.

Copied to clipboard

Runsong Jia, Zhen Fang, Mengjia Wu, Jie Lu, Yi Zhang

From Personal to Collective: On the Role of Local and Global Knowledge in LLM Personalization (2026.findings-acl)

Challenge:	Existing methods for hallucination detection depend on internal signals like uncertainty and self-consistency checks to identify unreliable outputs.
Approach:	They propose a retrieval-augmented generation method to enhance hallucination detection by addressing information updating challenges.
Outcome:	The proposed method improves on existing methods with strong generalization capabilities.

Copied to clipboard

Zehong Wang, Junlin Wu, Zhaoxuan Tan, Bolian Li, Xianrui Zhong, Zheli Liu, Qingkai Zeng

Challenge:	Large language models (LLMs) are the backbone of modern AI systems supporting a wide range of applications.
Approach:	They propose a Local–Global knowledge framework that augments user-specific signals with a global knowledge encoding collective behavioral trends.
Outcome:	The proposed framework outperforms existing methods on five personalization benchmarks.

ProMediate: A Simulation Testbed for Evaluating Proactive Mediation in Multi-Party Negotiation (2026.findings-acl)

Copied to clipboard

Ziyi Liu, Bahareh Sarrafzadeh, Pei Zhou, Longqi Yang, Jieyu Zhao, Ashish Sharma

InteracSPARQL : An Interactive System for SPARQL Query Refinement Using Natural Language Explanations (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are increasingly integrated into agentic frameworks to assist individual users in completing diverse tasks.
Approach:	They propose a simulation environment with a plug-and-play proactive AI mediator . they use a socio-cognitive evaluation framework to measure consensus changes, intervention latency, mediator effectiveness and intelligence.
Outcome:	The proposed model outperforms a generic baseline in multi-party negotiation scenarios while being 77% faster in response.

Copied to clipboard

Xiangru Jian, Zhengyuan Dong, M. Tamer Özsu

Are Large Language Models Reliable Reviewers? A Benchmark for Error Detection in Financial Documents (2026.findings-acl)

Challenge:	Existing approaches for SPARQL generation rely on one-turn models.
Approach:	They propose a training-free interactive refinement pipeline that acts as a plug-and-play enhancement for existing SPARQL systems.
Outcome:	The proposed approach improves the accuracy of base models without fine-tuning . it transforms potentially flawed queries from any source into verifiable code .

Copied to clipboard

Ying He, Zhouhong Gu, Zhecheng Hu, Yubo Zhou, Hao Shen, Jiaqing Liang, Zhaoqian Dai, Ma Shuguang, Fei Yu, Yanghua Xiao, Zhixu Li

Training-Free Test-Time Contrastive Learning for Large Language Models (2026.findings-acl)

Challenge:	Existing LLMs struggle to identify errors in financial documents, a study shows . 18% of financial practitioners make errors daily, one-third make errors several times weekly, and 59% make errors multiple times monthly.
Approach:	They introduce FinED-Bench, a publicly available Benchmark for financial error detection . it covers nine real-world financial scenarios and includes over 900 documents in 2025 . supervised fine-tuning can significantly improve the performance of weaker LLMs, they show .
Outcome:	The proposed benchmark covers nine real-world financial scenarios and includes over 900 documents reported in 2025 that are unseen by existing language models.

Copied to clipboard

Kaiwen Zheng, Kai Zhou, Jinwu Hu, Te Gu, Mingkai Peng, Fei Liu

To Diff or Not to Diff? Structure-Aware and Adaptive Output Formats for Efficient LLM-based Code Editing (2026.findings-acl)

Challenge:	Existing training-free alternatives to training-based models are static or depend on external guidance.
Approach:	They propose a training-free adaptation framework that enables a frozen LLM to improve online by distilling supervision from its own inference experiences.
Outcome:	The proposed framework outperforms existing test-time adaptation methods under online evaluation.

Copied to clipboard

Wei Cheng, Yongchang Cao, Chen Shen, Binhua Li, Jue Chen, Yongbin Li, Wei Hu

CogEmp:A Cognitive Empathy-Oriented Dialogue System for Structured Psychological Counseling (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are increasingly used for code editing, yet the full-code generation paradigm suffers from severe efficiency bottlenecks.
Approach:	They propose to use a structure-aware diff format to train LLMs to choose the most token-efficient format between a given diff format and full code.
Outcome:	The proposed approach matches the most token-efficient format with full-code generation while reducing latency and cost by over 30% on long-code editing tasks.

Copied to clipboard

Yiguo Deng, Xia Lei, Yuan Zhang, Long Ye

LiveLongBench: Tackling Long-Context Understanding for Spoken Texts from Live Streams (2026.findings-acl)

Challenge:	Existing models lack accurate modeling of cognitive empathy, especially the ability to understand users’ emotions and their underlying psychological causes.
Approach:	They propose a model tailored for the Chinese cultural context that integrates cognitive empathy into LLMs.
Outcome:	The proposed model outperforms existing models in key evaluation metrics, particularly in empathy, comprehensibility, and professionalism.

Copied to clipboard

Yongxuan Wu, Runyu Chen, Peiyu Liu, Hongjin Qian

Challenge:	Existing studies show that spoken text exhibits unique linguistic properties, such as high redundancy and repetitive phrases.
Approach:	They propose a long-text dataset that better handles redundancy in spoken text . their results highlight key limitations of current methods and suggest future directions .
Outcome:	The proposed benchmark improves existing methods and improves on redundancy in spoken text.

CPC-GRPO: Answer-Free Reinforcement Learning with Cross-Prompt Consensus Rewards (2026.findings-acl)

Copied to clipboard

Gyunyeop Kim, Sangwoo Kang

MASH: Evading Black-Box AI-Generated Text Detectors via Style Humanization (2026.findings-acl)

Challenge:	Reinforcement learning with verifiable rewards is a popular post-training tool for large language models, but relies on a ground-truth answer or external verifier, which limits applicability and increases cost.
Approach:	They propose an answer-free training objective that derives rewards solely from the model’s own probabilities by exploiting prompt paraphrases as multiple semantic views of the same intent.
Outcome:	The proposed objective derives rewards solely from the model’s own probabilities by exploiting prompt paraphrases as multiple semantic views of the same intent.

Copied to clipboard

Yongtong Gu, Songze Li, Xia Hu

MM-Doc-R1: Training Agents for Long Document Visual Question Answering through Multi-turn Reinforcement Learning (2026.findings-acl)

Challenge:	Existing detection methods rely on white-box assumptions or require prohibitively high computational and interaction costs, rendering them ineffective under practical black-box scenarios.
Approach:	They propose a framework that evades black-box detection methods based on style transfer by using style-injection supervised fine-tuning and direct preference optimization to shape distributions of AI-generated texts to resemble those of human-written texts.
Outcome:	The proposed framework achieves an average Attack Success Rate (ASR) of 92%, surpassing the strongest baselines by an average of 24% while maintaining superior linguistic quality.

Copied to clipboard

Jiahang Lin, Kai Hu, Binghai Wang, Yuhao Zhou, Zhiheng Xi, Honglin Guo, Shichun Liu, Junzhe Wang, Shihan Dou, Enyu Zhou, Hang Yan, Zhenhua Han, Tao Gui, Qi Zhang, Xuanjing Huang

Multimodal Chemical Structure-Text Coreference in Intellectual Property via Rule-guided Reinforcement Learning (2026.findings-acl)

Challenge:	Existing work on long document visual question answering is based on Retrieval-Augmented Generation (RAG) where textual or visual content is encoded into embeddings and relevance is determined by similarity scores with respect to the original query.
Approach:	They propose a framework that employs an agentic, vision-aware workflow to address long document visual question answering through iterative information discovery and synthesis.
Outcome:	The proposed framework outperforms existing RL systems by 10.4% on the MMLongbench-Doc benchmark and demonstrates superior training performance over GRPO.

Copied to clipboard

Hanmeng Zhong, Wentao Wu, Linqing Chen, Peng Zhou

TheraAgent: Self-Improving Therapeutic Agent for Precise and Comprehensive Treatment Planning (2026.findings-acl)

Challenge:	Existing tools for identifying chemical structures and textual referents are inadequate for this multimodal task.
Approach:	They propose a RULE-guided multimodal Reinforcement learning framework for chemical structure-text coreference . RULER is a rule-driven reinforcement learning framework that uses rule-based reward functions to obtain the correct domain knowledge.
Outcome:	The proposed framework improves on the baseline framework and shows superior efficacy.

Copied to clipboard

Junkai Li, Yunghwei Lai, Tianyi Zhu, Zheng Long Lee, Weizhi Ma, Yang Liu

LearnAct: Few-Shot Mobile GUI Agent with a Unified Demonstration Benchmark (2026.findings-acl)

Challenge:	Existing large language models rely on one-shot output without explicit verification, resulting in rough, incomplete, and potentially unsafe treatment plans.
Approach:	They propose an agentic framework that replaces one-shot generation with an iterative generate-judge-refine pipeline.
Outcome:	The proposed framework achieves state-of-the-art results on HealthBench, leading in Accuracy and Completeness.

Copied to clipboard

Guangyi Liu, Pengxiang Zhao, Liang Liu, Zhiming Chen, Yuxiang Chai, Yaozhen Liang, WenHao Wang, Siheng Chen, Zhengxi Lu, Shuai Ren, Hao Wang, Shibo He, Yong Liu, Wenchao Meng

From Local Perspective to Global Reasoning: A Neuro-Symbolic Framework for Zero-Shot Relation Extraction (2026.findings-acl)

Challenge:	Mobile GUI agents show promise in automating tasks but face significant generalization challenges in long-tail scenarios.
Approach:	They propose a benchmark framework for mobile GUI agents that measures the performance of GUI agents by analyzing their performance.
Outcome:	The LearnGUI benchmark outperforms existing methods in offline and online evaluations and demonstrates consistent gains across model architectures.

Copied to clipboard

Kailun Lyu, Fu Zhang, Zehan Li, Jingwei Cheng

DiFRa: A Unified Framework for Harmonizing Semantic Diversity and Factual Consistency in Question-Answer Generation (2026.findings-acl)

Challenge:	Existing methods for zero-shot relationship extraction do not distinguish between unseen, semantically similar relations.
Approach:	They propose a framework to enable global reasoning across a set of predictions.
Outcome:	The proposed framework outperforms existing methods and establishes new state-of-the-art results on widely used datasets.

Copied to clipboard

Zhenqin Li, ShengYong Ding, Shuangyin Li

Geoparsing: Diagram Parsing for Plane and Solid Geometry with a Unified Formal Language (2026.findings-acl)

Challenge:	Question-Answer Generation (QAG) is essential for domain-specific large language models post-training.
Approach:	They propose a framework that balances semantic diversity and factual consistency . they propose entropy and consistency scores that harmonize the trade-off between diversity and correctness .
Outcome:	The proposed framework outperforms baseline models in generating diverse QA pairs . the proposed framework harmonizes semantic entropy and consistency scores to quantify trade-off between diversity and correctness.

Copied to clipboard

Peijie Wang, Ming-Liang Zhang, Jun Cao, Chao Deng, Dekang Ran, Pi Bu, Hongda Sun, Xuan Zhang, Yingyao Wang, Jun Song, Bo Zheng, Fei Yin, Cheng-Lin Liu

Beyond Memorization: Testing LLM Reasoning on Unseen Theory of Computation Tasks (2026.findings-acl)

Challenge:	Recent advances in Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities across various vision reasoning tasks.
Approach:	They propose a unified formal language that integrates plane and solid geometry, comprehensively covering geometric structures and semantic relations.
Outcome:	The proposed language achieves state-of-the-art parsing performance and significantly boosts MLLMs’ capabilities for downstream geometry reasoning tasks.

Copied to clipboard

Shlok Shelat, Jay Raval, Souvik Roy, Manas Gaur

Beyond Dialogue Time: Temporal Semantic Memory for Personalized LLM Agents (2026.findings-acl)

Challenge:	Large language models (LLMs) have demonstrated strong performance on formal language tasks, yet whether this reflects genuine symbolic reasoning or pattern matching on familiar constructions remains unclear.
Approach:	They propose a benchmark for deterministic finite automata (DFA) construction from regular languages, comprising factual knowledge questions, seen construction problems from public sources, and unseen problems.
Outcome:	The proposed model achieves perfect accuracy on factual questions and 84-90% on seen tasks, but falls sharply on unseen problems (by 30-64%), with failures stemming from systematic misinterpretation of language constraints, incorrect handling of Kleene-star semantics, and a failure to preserve global consistency.

Copied to clipboard

Miao Su, Yucan Guo, Zhongni Hou, Long Bai, Zixuan Li, Yufei Zhang, Guojun Yin, Wei Lin, Xiaolong Jin, Jiafeng Guo, Xueqi Cheng

LongCLI-Bench: A Preliminary Benchmark and Study for Long-horizon Agentic Programming in Command-Line Interfaces (2026.findings-acl)

Challenge:	Existing methods focus on point-wise memory, losing durative information that captures persistent states and evolving patterns.
Approach:	They propose a memory framework that models semantic time for point-wise memory and supports the construction and utilization of durative memory.
Outcome:	Experiments on LongMemEval and LoCoMo show that the proposed method outperforms existing methods and achieves up to 12.2% improvement in accuracy.

Copied to clipboard

Yukang Feng, Jianwen Sun, Zelai Yang, Jiaxin Ai, Chuanhao Li, Zizhen Li, Fanrui Zhang, Kang He, Rui Ma, Jifan Lin, Jie Sun, Yang Xiao, Sizhuo Zhou, Wenxiao Wu, Yiming Liu, Pengfei Liu, Shenglin Zhang, Kaipeng Zhang

Cognitive-Uncertainty Guided Knowledge Distillation for Accurate Classification of Student Misconceptions (2026.findings-acl)

Challenge:	Existing benchmarks for agentic programming in long-horizon command-line interface tasks are limited by short task horizons, data contamination from GitHub scraping, and a lack of fine-grained evaluation metrics.
Approach:	They propose a benchmark to evaluate agentic capabilities across long-horizon command-line interface tasks.
Outcome:	The proposed benchmarks cover four engineering categories: from scratch, feature addition, bug fixing, and refactoring.

Copied to clipboard

Qirui Liu, Hao Chen, Weijie Shi, Jiajie Xu, Jia Zhu

Beyond Black-Box Interventions: Latent Probing for Faithful Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	Existing methods for identifying student misconceptions overlook students' reasoning processes, authors report .
Approach:	They propose a knowledge distillation framework that mines high-value samples from existing data.
Outcome:	The proposed framework outperforms sota LLM and standard fine-tuned 72B models on cross-topic tests.

Copied to clipboard

Linfeng Gao, Qinggang Zhang, Baolong Bi, Bo Zeng, Zheng Yuan, Zerui Chen, Zhimin Wei, Shenghua Liu, Linlong Xu, Longyue Wang, Weihua Luo, Jinsong Su

FalconCopilot: Empowering LLMs Towards Integrated Human-Machine Systems for Aviation Autonomy (2026.findings-acl)

Challenge:	Existing approaches to improve contextual faithfulness treat the LLM as a black box, generating responses that are inconsistent with the provided context.
Approach:	They propose a framework for faithful RAG that operates in three stages: (i) fine-grained knowledge pruning to filter irrelevant context, (ii) latent conflict probing to identify hard conflicts in the model’s latent space, and (iv) conflict-aware attention to modulate attention heads toward faithful context integration.
Outcome:	Experiments show that ProbeRAG significantly improves both accuracy and contextual faithfulness.

Copied to clipboard

Jingyuan Yan, Qingchen Liu, Qichao Ma, Jiahu Qin

Challenge:	Complex flight tasks require both intricate, long-horizon decision-making and precise operations.
Approach:	They propose a LLM-based copilot system that addresses deficiencies in adaptability and fine-grained decision support while integrating with a high-fidelity environment.
Outcome:	The proposed system shortens task completion time while attaining a level of performance approaching that of a human instructor.

Accelerating Training of Autoregressive Video Generation Models via Local Optimization with Representation Continuity (2026.findings-acl)

Copied to clipboard

Yucheng Zhou, Jianbing Shen

RouteRAG: Efficient Retrieval-Augmented Generation from Text and Graph via Reinforcement Learning (2026.findings-acl)

Challenge:	Existing autoregressive models have shown superior performance and efficiency in image generation, but are constrained by high computational costs and prolonged training times in video generation.
Approach:	They propose a Local Optimization method which optimizes tokens within localized windows while leveraging contextual information to reduce error propagation.
Outcome:	The proposed method achieves superior performance to the baseline while halving the training cost without sacrificing quality.

Copied to clipboard

Yucan Guo, Miao Su, Saiping Guan, Zihao Sun, Xiaolong Jin, Jiafeng Guo, Xueqi Cheng

HORIZON: A Benchmark for In-the-wild User Behaviour Modeling (2026.findings-acl)

Challenge:	Existing graph-based or hybrid systems lack the ability to integrate supplementary evidence as reasoning unfolds.
Approach:	They propose a framework that integrates non-parametric knowledge into Large Language Models . they use a RL-based framework to optimize the entire generation process via RL .
Outcome:	The proposed framework outperforms existing RAG frameworks in five question answering benchmarks.

Copied to clipboard

Arnav Goel, Pranjal A Chitale, Bhawna Paliwal, Bishal Santra, Amit Sharma

Two-Stage Parameter Alignment for Multi-LoRA Merging in Large Language Models (2026.findings-acl)

Challenge:	Existing user modeling benchmarks focus on short sessions and next-item prediction within a single domain.
Approach:	They propose a benchmark that reformulates user modeling along three axes . it covers 54M users and 35M items, enabling pretraining and evaluation . they propose tasks and evaluation setups that better reflect real-world deployment scenarios .
Outcome:	The proposed benchmark covers 54M users and 35M items, and is based on Amazon Reviews.

Copied to clipboard

Zijian Li, Xiachong Feng, Weitao Ma, Yichong Huang, Xiaocheng Feng, Bing Qin

The Agent’s First Day: Benchmarking Learning, Exploration, and Scheduling in the Workplace Scenarios (2026.findings-acl)

Challenge:	Current general model merging methods are prone to parameter interference problems . a novel two-stage parameter alignment framework is proposed to address this problem .
Approach:	They propose a two-stage parameter alignment framework that integrates low-rank LoRAs . they propose to reduce the computational complexity of existing methods by preserving fine-grained functions .
Outcome:	The proposed framework exhibits greater robustness than other methods in high-rank and high-interference scenarios while preserving fine-grained functions.

Copied to clipboard

Daocheng Fu, Jianbiao Mei, Rong Wu, Xuemeng Yang, Jia Xu, Ding Wang, Pinlong Cai, Yong Liu, Licheng Wen, Botian Shi

Challenge:	Existing research mainly focuses on performance upper bounds in static environments, overlooking stochastic real-world deployment.
Approach:	They propose a dynamic evaluation environment that simulates a "trainee" agent continuously exploring a novel setting.
Outcome:	The proposed model evaluates agents in a dynamic evaluation environment that simulates a "trainee" agent continuously exploring a novel setting.

PrefIx: Understand and Adapt to User Preference in Human-Agent Interaction (2026.findings-acl)

Copied to clipboard

Jialin Li, Zhenhao Chen, Hanjun Luo, Hanan Salam

Challenge:	Current benchmarks evaluate task accuracy but overlook how agents interact . Preference-aware agents show 7.6% average UX improvement and 18.5% gain in preference alignment.
Approach:	They propose a configurable environment that evaluates both what agents accomplish and how they interact.
Outcome:	The proposed model improves performance and improves user experience by 7.6% and 18.5% respectively.

How Do Answer Tokens Read Reasoning Traces? Self-Reading Patterns in Thinking LLMs for Quantitative Reasoning (2026.findings-acl)

Copied to clipboard

Haoyang Chen, Yi Liu, Jianzhi Shao, Tao Zhang, Chengfu Huo, Wei Hu

MENTOR: Efficient Autoregressive Image Generation with Balanced Multimodal Control (2026.findings-acl)

Challenge:	Prior work on activation steering has focused on shaping reasoning traces, but it remains unclear how answer tokens actually read and integrate the reasoning to produce reliable outcomes.
Approach:	They propose a training-free steering method that uses self-reading quality scores to guide inference toward benign self-readiness and away from uncertain and disorganized reading.
Outcome:	The proposed method yields consistent accuracy gains in the reasoning traces generated by thinking LLMs.

Copied to clipboard

Haozhe Zhao, Zefan Cai, Shuzheng Si, Liang Chen, Jiuxiang Gu, Wen Xiao, Minjia Zhang, Junjie Hu

Better and Worse with Scale: How Contextual Entrainment Diverges with Model Size (2026.findings-acl)

Challenge:	Recent text-to-image models achieve impressive visual quality but still face challenges in precise controllability, balancing multimodal inputs, and high training cost for multimodal image generation.
Approach:	They propose an autoregressive framework with a two-stage training paradigm for controllable multimodal image generation.
Outcome:	Extensive experiments on DreamBench++ and DreamBech show that the proposed framework achieves a strong balance between textual and visual guidance for controllable image generation.

Copied to clipboard

Dikshant Kukreja, Kshitij Sah, Gautam Gupta, Avinash Anand, Rajiv Ratn Shah, Zhengkui Wang, Aik Beng Ng, Erik Cambria

Challenge:	Larger language models become better and worse at handling contextual information . et al. (2017) formalized contextual entrainment as a tendency to favor tokens in context .
Approach:	They formalize the first scaling laws for contextual entrainment . they find large models are four times more resistant to counterfactual misinformation .
Outcome:	The largest models are four times more resistant to counterfactual misinformation than the smallest, but twice as prone to copying arbitrary tokens.

LLM-Guided Semantic Bootstrapping for Interpretable Text Classification with Tsetlin Machines (2026.findings-acl)

Copied to clipboard

Jiechao Gao, Rohan Kumar Yadav, Yuangang Li, Yuandong Pan, Jie Wang, Ying Liu, Michael Lepech

Evaluating the Impact of Reviewer Guideline Design on LLM-Based Automated Peer Review (2026.findings-acl)

Challenge:	Pretrained language models (PLMs) provide strong semantic representations but are costly and opaque.
Approach:	They propose a framework that transfers pretrained language models into symbolic form and integrates them into symbolic models.
Outcome:	The proposed framework improves interpretability and accuracy across multiple text classification tasks while remaining fully symbolic and efficient.

Copied to clipboard

Haowen Li, Yoichi Ishibashi, Masafumi Oyamada

Challenge:	a growing workload has made peer review automation an urgent necessity, says a new study . official conference guidelines and reviewer-imitating guidelines degraded review performance . current human-based peer review system faces serious challenges, authors say .
Approach:	They analyze how reviewer guidelines influence automated peer review . official conference guidelines produce review results consistent with human judgments .
Outcome:	The proposed reviewer guidelines produce results consistent with human judgments . the proposed reviewers' imitations degraded performance, the authors note .

MI-CXR: A Benchmark for Longitudinal Reasoning over Multi-Interval Chest X-rays (2026.findings-acl)

Copied to clipboard

Sunghwan Steve Cho, Yunseok Han, Jaeyoung Do

Relevance to Utility: Process-Supervised Rewrite for RAG (2026.findings-acl)

Challenge:	Existing medical VQA benchmarks focus on single images or short-horizon image pairs.
Approach:	They propose a benchmark for standardized evaluation of longitudinal reasoning over multi-visit sequences.
Outcome:	The proposed benchmark shows low overall performance (29.3% accuracy) and is only modestly above random guessing.

Copied to clipboard

Jaeyoung Kim, Jongho Kim, Seung-won Hwang, Seoho Song, Young-In Song

On the Cultural Anachronism and Temporal Reasoning in Vision Language Models (2026.findings-acl)

Challenge:	Existing bridge modules attempt to rewrite documents for better generation, but fail to capture "document utility".
Approach:	They propose to observe reasoning process as process supervision and scale this observation to enhance reliability in distillation.
Outcome:	The proposed method improves over strong bridging baselines on multiple open-domain question-answering benchmarks.

Copied to clipboard

Mukul Ranjan, Prince Jha, Khushboo Kumari, Zhiqiang Shen

GLA: Grounding Large Language Models in Molecular Hierarchy for Chemical Understanding (2026.findings-acl)

Challenge:	Vision-Language Models (VLMs) are increasingly applied to cultural heritage materials.
Approach:	They propose a temporal anachronism benchmark to evaluate temporal reasoning on 1,600 Indian cultural artifacts.
Outcome:	The proposed model achieves only 58.7% accuracy on the best model, which is a significant performance gap across architectures and scales.

Copied to clipboard

Yingxu Li, Jingjie Zeng, Zekun Wang, Hongfei Lin, Liang Yang

NeuRAG: End-to-End Neural Knowledge Augmentation via Hyper-Neurons (2026.findings-acl)

Challenge:	Existing molecule-language models obscure the hierarchical organization of chemical semantics . Existing models rely on linear or uniform encodings, causing structural distortion .
Approach:	They propose a framework that integrates intrinsic molecular topology into large language models.
Outcome:	The proposed framework improves on cross-modal retrieval, captioning, and property prediction benchmarks.

Copied to clipboard

Liwei Zheng, Xuemin Liu, Jie Liu

DEAR: Distributional Error-Aware Reliability for Robust Multimodal Sentiment Analysis with Missing Modalities (2026.findings-acl)

Challenge:	Existing approaches to grounding large language models in external knowledge are constrained by a decoupled architecture: retrieval and reasoning operate as separate stages, with retrieved text merely prepended as passive context.
Approach:	They propose an end-to-end Neuralized RAG framework that unifies knowledge retrieval and fusion through Hyper-Neurons.
Outcome:	Extensive experiments across multiple datasets and LLMs demonstrate NeuRAG’s strong and consistent performance as a promising novel RAG paradigm.

Copied to clipboard

Shihao Zou, Wei Wei, Yongshuo Zhang

OpenPhone: Mobile Agentic Foundation Models (2026.findings-acl)

Challenge:	Existing methods focus on feature completion but neglect semantic shifts caused by distribution gaps and decision risks under high uncertainty.
Approach:	They propose a distributional error-aware reliability estimation framework for robust MSA . they propose reconstructed features to be explicitly aligned with original distributional manifold .
Outcome:	The proposed framework mitigates semantic shifts by aligning reconstructed features with original distributional manifold . Extensive experiments on MOSI, MOSEI, and SIMS validate the framework .

Copied to clipboard

Yangqin Jiang, Chao Huang

Fast Retrieval and Slow Reasoning for Explainable Multimodal Sentiment Analysis (2026.findings-acl)

Challenge:	Mobile GUI agents face a critical dilemma: on-device models (4B or smaller) lack sufficient performance, while capable models are either too large for mobile deployment or prohibitively costly.
Approach:	They propose a mobile GUI agent system that leverages device-cloud collaboration to tap cost-efficiency of on-device models and high capability of cloud models.
Outcome:	The proposed system matches or nears larger models with reduced cloud costs on mobile platforms.

Copied to clipboard

Aoqiang Zhu, Min Hu, Yan Xing

Exploration-Exploitation Reshaping towards Efficient Reasoning for Large Language Models (2026.findings-acl)

Challenge:	Existing multimodal sentiment analysis methods rely on holistic fusion . such strategies introduce redundant information and obscure the decision process .
Approach:	They propose an interpretable framework that decomposes multimodal sentiment modeling into two cooperative pathways.
Outcome:	The proposed framework achieves competitive performance, higher efficiency, stronger robustness to noise, and clearer decision transparency than existing holistic fusion methods.

Copied to clipboard

Yufeng Shi, Weilin Luo, Yuxiang Zhang, Zongmeng Zhang, Haoyang Liu, Yubing Wang, Bin Wang, Wengang Zhou, Houqiang Li

COMPEL: Compensated Mixture-of-Experts Pruning with Expert-Layer distribution (2026.findings-acl)

Challenge:	Large Reasoning Models (LRMs) are constrained by the overthinking issue.
Approach:	They propose a policy optimization framework that reshapes the exploration and exploitation through two core components: self-imitation and self-guidance exploration.
Outcome:	The proposed model achieves superior reasoning efficiency without compromising overall accuracy.

Copied to clipboard

Seohee Yoon, Yong Suk Choi

Towards Self-Evolving Agents: Enabling Autonomy through Interactive Experience Refinement (2026.findings-acl)

Challenge:	Mixture-of-Experts (MoE) architectures are effective for scaling Large Language Models (LLMs) however, existing pruning methods adopt uniform pruning across layers, which fails to capture layer-wise variations in expert importance and redundancy.
Approach:	They propose a Mixture-of-Experts pruning method that activates only a subset of experts during inference by estimating expert importance using Fisher information.
Outcome:	The proposed pruning method outperforms existing pruning methods while reducing inference latency and peak GPU memory usage.

Copied to clipboard

Cheng Yang, Xuemeng Yang, Licheng Wen, Daocheng Fu, Jianbiao Mei, Rong Wu, Pinlong Cai, Yufan Shen, Nianchen Deng, Jia Xu, Botian Shi, Yu Qiao, Haifeng Li

Challenge:	Large Language Models struggle with complex, multi-step operational tasks because they remain static during inference and cannot learn from past experience.
Approach:	They propose a framework that organizes cross-domain insights to facilitate orchestration of long-horizon workflows.
Outcome:	The proposed framework outperforms existing methods on the TAC productivity benchmark and shows strong cross-task transferability.

Good Reasoning Makes Good Demonstrations: Implicit Reasoning Quality Supervision via In-Context Reinforcement Learning (2026.findings-acl)

Copied to clipboard

Tiehua Mei, Minxuan Lv, Leiyu Pan, Zhenpeng Su, Hongru Hou, Hengrui Chen, Ao Xu, Deqing Yang

Design First, Code Later: Aesthetically Pleasing Template-Free Slides Generation (2026.findings-acl)

Challenge:	Reinforcement Learning with Verifiable Rewards (RLVR) improves reasoning in large language models but treats all correct solutions equally, potentially reinforcing flawed traces that arrive at correct answers by chance.
Approach:	They propose a method that reweights rewards by a factor approximately proportional to Evidence Gain and assigns higher weights to high-quality traces without requiring costly computation.
Outcome:	Experiments on mathematical reasoning benchmarks show that Reinforcement Learning with Verifiable Rewards (RLVR) improves reasoning in large language models but treats all correct solutions equally.

Copied to clipboard

Zhiyao Cui, Chenxu Wang, Shuyue Hu, Yiqun Zhang, Wenqi Shao, Qiaosheng Zhang, Zhen Wang

Benchmarking the Fine-Grained Discriminability in Image-Text Retrieval via Controlled Contrastive Differences (2026.findings-acl)

Challenge:	Existing approaches to producing presentation slides rely on fixed templates or executable code . Existing methods rely only on predefined templates and emit executable codes .
Approach:	They propose a hierarchical slides generation workflow DeepSlides that organizes slide design tasks without any predefined template or style.
Outcome:	The proposed framework outperforms baseline methods on evaluated metrics and achieves superior performance in human preference evaluations.

Copied to clipboard

Zhen Wang, Xi Zhou, Yating Yang, Bo Ma, Lei Wang, Rui Dong, Azmat Anwar, Siru Miao

Learning on Imbalanced Noisy Data via Debiased Sample Selection and LLM-Driven Annotation (2026.findings-acl)

Challenge:	Existing cross-modal image-text retrieval models often retrieve samples with inconsistent details.
Approach:	They propose two fine-grained image-text retrieval benchmarks that incorporate extensive contrastive samples with one controlled contrastive difference from its anchor.
Outcome:	Extensive experiments show that contrastive samples can significantly degrade retrieval performance.

Copied to clipboard

Bo Yuan, Yulin Chen, Yin Zhang

Beyond Task-Level Context: Class-Conditional Context Vectors for Implicit In-Context Learning (2026.findings-acl)

Challenge:	Existing approaches to learning with noisy labels are prone to selection bias and training bias . obtaining large-scale high-quality datasets is expensive and time-consuming in practical scenarios .
Approach:	They propose an imbalanced learning with noisy labels task to let model learn from noisy labels . they first conduct debiased sample selection to better separate clean samples from noisy samples . then they feed selected clean samples to active annotator large language models for re-annotating noisy samples.
Outcome:	The proposed method is superior to existing methods on synthetic and real-world datasets.

Copied to clipboard

Jianxin Zhang, Yilu Hu, Teng Liu, Pei Guo, Juntao Li

DPN-LE: Dual Personality Neuron Localization and Editing for Large Language Models (2026.findings-acl)

Challenge:	Existing approaches aggregate demonstrations from all classes into a shared, task-level context vector, capturing global task information but without explicitly preserving fine-grained, class-conditional semantic distinctions.
Approach:	They propose a class-conditional context vector extension to implicit in-context learning that explicitly models class-specific contextual information by constructing separate context vectors for each class.
Outcome:	The proposed extension outperforms task-level context vector baselines and achieves higher average accuracy than conventional few-shot learning on most models.

Copied to clipboard

Lifan Zheng, Xue Yang, Jiawei Chen, Chenyan WU, Jingyuan Zhang, Fanheng Kong, Xinyi Zeng, Xiang Chen, Yu Tian

Reasoning in a Combinatorial and Constrained World: Benchmarking LLMs on Natural-Language Combinatorial Optimization (2026.findings-acl)

Challenge:	Current methods for editing personality traits in large language models can change personalities but reduce performance.
Approach:	They propose a novel paradigm for personality editing that locates and edits LLM neurons and enables competitive personality control at inference time.
Outcome:	Experiments on LLaMA-3-8B-Instruct and Qwen2.5-7B-instruct show that the proposed approach can improve performance and improve performance.

Copied to clipboard

Xia Jiang, Jing Chen, Cong Zhang, Jie Gao, Chengpeng Hu, Chenhao Zhang, Yaoxin Wu, Yingqian Zhang

DARL: Encouraging Diverse Answers for General Reasoning without Verifiers (2026.findings-acl)

Challenge:	NLCO evaluates large language models for combinatorial optimization (CO) . existing evaluations emphasize relatively simple reasoning competencies .
Approach:	They propose a combinatorial optimization benchmark that evaluates large language models on CO reasoning.
Outcome:	The proposed model can handle combinatorial optimization without writing code or calling external solvers.

Copied to clipboard

Chongxuan Huang, Lei Lin, Xiaodong Shi, Wenping Hu, Ruiming Tang

Clustered Self-Assessment: A Simple yet Effective Method for Uncertainty Quantification in Large Language Models (2026.findings-acl)

Challenge:	Recent efforts such as RLPR have extended RLVR to general domains, enabling training on broader datasets and achieving improvements over RL PR.
Approach:	They propose a framework that encourages the generation of diverse answers within a controlled deviation range from the reference while preserving alignment with it.
Outcome:	Extensive experiments on 13 benchmarks show that DARL surpasses RLPR in both reasoning accuracy and output diversity.

Copied to clipboard

Qi Cao, Takeshi Kojima, Andrew Gambardella, Helinyi Peng, Yutaka Matsuo, Yusuke Iwasawa

STEP: Success-Rate-Aware Trajectory-Efficient Policy Optimization (2026.findings-acl)

Challenge:	Existing methods for uncertainty quantification in large language models rely on indirect signals, such as entropy across sampled generations, which can be difficult to interpret and do not fully leverage the model’s ability to assess its own uncertainty.
Approach:	They propose a method that groups sampled generations into semantically distinct clusters and uses the probability assigned by the LLM to each option as a confidence estimate.
Outcome:	The proposed method outperforms baseline methods and achieves competitive performance with as few as two additional samples.

Copied to clipboard

Yuhan Chen, Yuxuan Liu, Long Zhang, Pengzhi Gao, Jian Luan, Wei Liu

AscendKernelGen: LLM-Driven Kernel Generation for NPUs (2026.findings-acl)

Challenge:	Existing GRPO-based methods allocate sampling uniformly across tasks regardless of difficulty, propagate misleading learning signals and incur high sample-collection costs.
Approach:	They propose a framework that allocates sampling based on per-task success rates and performs fine-grained step-level optimization.
Outcome:	The proposed method improves sample efficiency and training stability over existing GRPO variants and three ablation variants on OSWorld and AndroidWorld.

Copied to clipboard

Xinzi Cao, Jianyang Zhai, Pengfei Li, Zhiheng Hu, Cen Yan, null Mubingxu, Guanghuan Fang, Bin She, Jiayu Li, Yihan Su, Dongyang Tao, Feidiao Yang, Chang-Dong Wang, Yutong Lu, Weicheng Xue, Bin Zhou, Yonghong Tian

Challenge:	Neural Processing Units (NPUs) are critical for AI infrastructure, but their development remains a bottleneck due to vendor-specific Domain-Specific Languages (DSLs).
Approach:	They propose a framework for NPU kernel development that bridges the gap in hardware-specific coding . compiler success on complex Level-2 kernels improves from 0% to 95.5%, they say .
Outcome:	The proposed framework bridges the gap in hardware-specific coding, showing a near-zero success rate on complex kernels.

Learning to Evolve: A Self-Improving Framework for Multi-Agent Systems via Textual Parameter Graph Optimization (2026.findings-acl)

Copied to clipboard

Shan He, Runze Wang, Zhuoyun Du, Huiyu Bai, Zouying Cao, Yu Cheng, Bo Zheng

Learning How to Remember: A Meta-Cognitive Management Method for Structured and Transferable Agent Memory (2026.findings-acl)

Challenge:	Existing methods for designing and optimizing multi-agent systems are static and do not learn from experience.
Approach:	They propose a framework that enables a multi-agent system to learn to evolve . they use "textual gradients" to pinpoint failures and suggest granular modifications .
Outcome:	a new framework enables a multi-agent system to learn to evolve . it learns from historical optimization experiences to improve its performance .

Copied to clipboard

Sirui Liang, Pengfei Cao, Jian Zhao, Wenhao Teng, Xiangwen Liao, Jun Zhao, Kang Liu

Challenge:	Existing approaches store memory in fixed representations and reuse it at a single or implicit level of abstraction, which limits generalization and often leads to negative transfer when distribution shift.
Approach:	They propose a Meta-Cognitive Memory Abstraction method which decouples task execution from memory management by combining a frozen task model with a learned memory copilot.
Outcome:	Experiments on ALFWorld, ScienceWorld, and BabyAI show that the proposed method improves performance, out-of-distribution generalization, and cross-task transfer over several baselines.

LoRE: Enhancing Search Relevance with Progressive Chain-of-Thought and Preference Alignment (2026.findings-acl)

Copied to clipboard

Chenji Lu, Zhuo Chen, Hui Zhao, Zhiyuan Zeng, Gang Zhao, Junjie Ren, null Lihaoran, Songyan Liu, Pengjie Wang, Chuan Yu, Jian Xu, Bo Zheng

Multilingual Refusal Alignment for Safer Large Language Models (2026.findings-acl)

Challenge:	E-commerce search relevance is a critical component of retrieval systems.
Approach:	They propose a large-generative model for search relevance that trains reasoning knowledge, multi-modal understanding and rule awareness into three core competencies.
Outcome:	The proposed model outperforms GPT-5 in Macro-F1 and achieves 27% online gain.

Copied to clipboard

Aleksandra Krasnodębska, Wojciech Kusa, Aldo Lipani

View-R1: Asymmetric Policy Optimization for Difficulty-Aware Multimodal Reinforcement Learning (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are increasingly used globally, but their safety and alignment can vary unpredictably between languages.
Approach:	They propose a multilingual refusal alignment dataset to investigate whether alignment transfers cross-lingually and how language consistency is preserved during training.
Outcome:	The proposed model can be trained on multilingual datasets without affecting general performance.

Copied to clipboard

Minjie Hong, Zirun Guo, Jiabao Zhang, Zehan Wang, Ziang Zhang, Tao Jin, Zhou Zhao

PseudoGD: Enhancing Spatial Reasoning in Vision-Language Models through Pseudo Geometric Knowledge Distillation (2026.findings-acl)

Challenge:	Multimodal Large Language Models (MLLMs) are powerful at integrating diverse data but struggle with complex reasoning.
Approach:	They propose a method which separates responses into positive and negative groups to stabilize training and preserve knowledge.
Outcome:	The proposed model View-R1 achieves a 10.55% improvement in reasoning and outperforms larger models while maintaining and improving performance on general tasks.

Copied to clipboard

Gwanghee Lee, Yeeun Choi, Kyoungson Jhang

DMSD: Dual-Modal Semantic Disentanglement for Compositional Zero-Shot Learning (2026.findings-acl)

Challenge:	Recent Large Vision-Language Models (LVLMs) have shown remarkable success in general semantic understanding, but struggle with 3D spatial reasoning tasks.
Approach:	They propose a framework to help vision encoders internalize 3D geometric information using only standard 2D images.
Outcome:	The proposed framework achieves State-of-the-Art (SOTA) performance across various model architectures.

Copied to clipboard

Pan Yang, Jing Yang, Ruan Xiao li, Yuling Chen, Yuankai Wu, Quan Zhou, Xu Wang

ReFreeKV: Towards Threshold-Free KV Cache Compression (2026.findings-acl)

Challenge:	Compositional Zero-Shot Learning (CZSL) is a new research paradigm that learns sub-concepts from seen compositions and recognizes unseen novel combinations.
Approach:	They propose a Dual-Modal Semantic Disentanglement framework that integrates visual and textual information to achieve effective sub-concept disentangling.
Outcome:	The proposed framework achieves state-of-the-art performance on three benchmark datasets . it integrates a class-centroid bridge module to guide class centroids toward the textual space .

Copied to clipboard

Xuanfan Ni, Liyan Xu, Chenyang Lyu, Longyue Wang, Mo Yu, Lemao Liu, Fandong Meng, Jie Zhou, Piji Li

Safety Sidecar: Reflection-Driven Runtime Control for Safer Agents (2026.findings-acl)

Challenge:	Towards the KV cache efficiency, we propose a new objective that lifts the threshold constraints for robust KV compression.
Approach:	They propose a method that adjusts KV cache budgets while preserving full-cache performance.
Outcome:	The proposed method can reduce memory consumption while preserving full-cache performance.

Copied to clipboard

Wang Bin, Quan Jiazheng, Xingrui Yu, Hu Hansen, Yu Hao, Anjun Gao, Zhenglin Wan, Hui LI, Ivor Tsang

MCGA: A Multi-task Classical Chinese Literary Genre Audio Corpus (2026.findings-acl)

Challenge:	Existing safety controls fail to provide runtime intervention or cross-architecture portability for autonomous LLM agents.
Approach:	They propose a model-agnostic, plug-and-play module to provide arbitrary agent safety control and auditability.
Outcome:	The proposed module improves the secure-solution rate by 2.9–11.2 percentage points . it adds only 3.2s to end-to-end latency and a negligible average cost of 5.37 10-4 per scenario .

Copied to clipboard

Yexing Du, Kaiyuan Liu, Bihe Zhang, Youcheng Pan, Bo Yang, Liangyu Huo, Xiyuan Zhang, Jian Xie, Daojing He, Yang Xiang, Ming Liu, Bing Qin

VET: Verifiable Execution Tracing for Reliable Text-to-SQL Generation (2026.findings-acl)

Challenge:	Multimodal Large Language Models (MLLMs) have advanced Chinese Classical Studies (CCS) but the audio dimension of CCS remains underexplored due to a lack of high-quality, domain-specific audio corpora.
Approach:	They propose a 119-hour audio corpus comprising 22,000 audio samples to bridge this gap . it encompasses a diverse range of literary genres across six tasks .
Outcome:	The proposed corpus encompasses a diverse range of literary genres across six tasks: Automatic Speech Recognition (ASR), Speech-to-Text Translation (S2TT), Speech Emotion Captioning (SEC), Spoken Question Answering ( SQA), Speech Understanding (SU), and Speech Reasoning (SR).

Copied to clipboard

Dongyu Wang, Jingyu Li, Lan Zhang, null Ganggang.yu, Liang Huang

PACE: Prefix-Protected and Difficulty-Aware Compression for Efficient Reasoning (2026.findings-acl)

Challenge:	Existing methods for text-to-SQL generation are prone to hallucinations and grounding . authors present a novel reasoning paradigm that transforms text- to-Sql from unverifiable textual rationales into step-wise executable semantics.
Approach:	They propose a reasoning paradigm that transforms text-to-SQL from unverifiable textual rationales into step-wise executable semantics.
Outcome:	The proposed reasoning paradigm transforms text-to-SQL from unverifiable textual rationales into step-wise executable semantics.

Copied to clipboard

Ruixiang Feng, Yuntao Wen, Silin Zhou, Ke Shi, Yifan Wang, Ran Le, Zhenwei An, Zongchao Chen, Chen Yang, Guangyue Peng, Yiming Jia, Dongsheng Wang, Tao Zhang, Lisi Chen, Yang Song, Shen Gao, Shuo Shang

CreativeBench: Benchmarking and Enhancing Machine Creativity via Self-Evolving Challenges (2026.findings-acl)

Challenge:	Existing LRMs often suffer from "overthinking" and excessively long reasoning traces . a dual-level framework for length compression of LRM is proposed .
Approach:	They propose a framework for prefix-protected and difficulty-aware compression under hierarchical supervision.
Outcome:	The proposed framework reduces token usage while improving accuracy on math benchmarks.

Copied to clipboard

Zihan Wang, Lam Nguyen, Zhengyang Zhao, Mengyue Yang, Chengwei Qin, Yujiu Yang, Linyi Yang

Challenge:	Increasing saturation of web data limits further scaling of model intelligence.
Approach:	They propose a benchmark to evaluate machine creativity in code generation that combines combinatorial and exploratory creativity through reverse engineering and self-play.
Outcome:	The proposed benchmark targets combinatorial and exploratory creativity through reverse engineering and self-play.

Machine Behavior in Relational Moral Dilemmas: Moral Rightness, Predicted Human Behavior, and Model Decisions (2026.findings-acl)

Copied to clipboard

Jiseon Kim, Jea Kwon, Luiz Felipe Vecchietti, Wenchao Dong, Jaehong Kim, Meeyoung Cha

Challenge:	Human moral judgment is context-dependent and changes based on interpersonal relationships.
Approach:	They characterize LLM behavior using the Whistleblower’s Dilemma . they find moral rightness remains consistently fairness-oriented .
Outcome:	The model decisions mirror moral rightness judgments, rather than their behavioral predictions.

Prompting the Unknown: Understanding Response Uncertainty in Large Language Models (2026.findings-acl)

Copied to clipboard

Ze Yu Zhang, Arun Verma, Finale Doshi-Velez, Bryan Kian Hsiang Low

Explainable Quantum Program Repair with Verifiable Proof Traces (2026.findings-acl)

Challenge:	Large language models are widely used in decision-making across diverse domains.
Approach:	They propose a prompt-response concept model that explains the relationship between the amount of task-relevant information provided in the prompt and the LLM-generated response uncertainty by identifying four sources of response uncertainty.
Outcome:	The proposed model shows that the amount of information provided in the prompt influences the LLM-generated response uncertainty.

Copied to clipboard

Tingting Li, Ziming Zhao, Zhaoxuan Li, Jiongchi Yu, Xiaofei Yue, Jianwei Yin

MA2P: A Meta-Cognitive Autonomous Intelligent Agents Framework for Complex Persuasion (2026.findings-acl)

Challenge:	Existing approaches to program repair provide only post-hoc, non-verifiable explanations that are not executable or verifiably.
Approach:	They propose a framework that couples repair generation with machine-checkable executable explanations for quantum programs where correctness hinges on subtle semantic properties such as circuit equivalence and fidelity preservation.
Outcome:	Experiments on QASMBench with mutation-generated quantum program bugs show that the proposed framework improves both semantic precision and explanation faithfulness over baselines that rely on unconstrained or purely natural-language explanations.

Copied to clipboard

Dingyi Zhang, Ziqing Zhuang, Linhai Zhang, Ziyang Gao, Deyu Zhou

Synergizing Semantic Anchors and Ordinal Smoothed Cross-Entropy for Speech Fluency Classification (2026.findings-acl)

Challenge:	Existing approaches to persuasion generate generic or weakly grounded responses even when such cues are identified.
Approach:	They propose a meta-cognitive autonomous intelligent agent framework for complex persuasion that coordinates perception management, mental-state inference, strategy execution, memory maintenance, and performance evaluation.
Outcome:	The proposed framework achieves a higher persuasion success rate than baselines.

Copied to clipboard

Mulati Kahaer, Sirajahmat Ruzmamat, XuDong Pang, Subinuer Maimaitituerxun, Zaokere Kadeer, Abudurexiti Reheman, Wenwen Lu, Panpan Zheng, Aishan Wumaier

Text Classification

Challenge:	Existing methods fail to bridge the semantic gap between static expert priors and dynamic temporal representations while overlooking the inherent ordinal nature of fluency scores.
Approach:	They propose a set of expert features targeting fluency disruptions and rhythmic regularity to provide explicit linguistic priors.
Outcome:	The proposed model outperforms baseline models in both macroscopic and microscopic speech flow trends and local anomalies.

PAC-BENCH: Evaluating Multi-Agent Collaboration under Privacy Constraints (2026.findings-acl)

Copied to clipboard

Minjun Park, Donghyun Kim, Hyeonjong Ju, Seungwon Lim, Dongwook Choi, Taeyoon Kwon, Minju Kim, Jinyoung Yeo

SciText2Eq: Assessing LLMs for Explainable Equation Generation for Scientific Creativity (2026.findings-acl)

Challenge:	Recent research explores multi-agent systems where agents collaborate toward shared goals to handle complex tasks.
Approach:	They propose a benchmark for systematic evaluation of multi-agent collaboration under privacy constraints.
Outcome:	The proposed benchmark shows that privacy constraints degrade collaboration performance and make outcomes depend more on the initiating agent than the partner.

Copied to clipboard

Yifan Mo, Xiao Fu, Yue Su, Qingyu Meng, Koen Hindriks, Qingzhi Liu, Jiahuan Pei

Reward Yourself: Efficient Self Rewards for Trustworthy Sampling (2026.findings-acl)

Challenge:	Prior work has addressed problems in unstructured grounding, multi-equation dependency, and human-aligned evaluation.
Approach:	They construct a dataset of scientific texts and evaluate it using an explainable equation generation workflow using automatic metrics and human judgments.
Outcome:	The proposed model achieves moderate performance on lexical and syntactic similarity, but struggles with semantic accuracy.

Copied to clipboard

Mingjie Li, Wai Man Si, Michael Backes, Yang Zhang

Analyze Like a Venture Capitalist: Information-Gain and Knowledge Enhanced Graph Reasoning for Startup Success Prediction (2026.findings-acl)

Challenge:	Retraining reward models to address privacy leaks and stereotypes is expensive . recent advances in large language models have led to improvements in understanding .
Approach:	They propose a lightweight intrinsic reward that can be used to prune existing LLMs to approximate an "untrust" and an ""untrust "" token distribution.
Outcome:	Experiments with two reward models and four LLMs show that selfRW improves trustworthiness with minimal impact on general utility benchmarks.

Copied to clipboard

Haoyu Pei, Zhongyang Liu, Xiangyi Xiao, Xiaocong Du, Suting Hong, Kunpeng Zhang, Haipeng Zhang

Read the Room, Read the Image: Understanding Indirect Speech Acts in Multimodal Visual Contexts (2026.findings-acl)

Challenge:	Most venture capital investments fail, while a few deliver outsized returns.
Approach:	They propose a framework that synthesizes relational evidence across sources . they propose combining information-gain-driven retriever and knowledge base to ground reasoning .
Outcome:	The proposed framework achieves +5.9% F1 and +22.1% Precision@5 over state-of-the-art baselines.

Copied to clipboard

Jaehee Kim, Ji Hoon Chung, Seoyoon Park, Unsol Kim, Kyungwon Park, JiHak Kim, Yi-Jun Chen, Hansaem Kim

Break Through the Compression Bottleneck: From Theory to Practice (2026.findings-acl)

Challenge:	Existing benchmarks focus on explicit context, but do not address context-dependent pragmatic understanding.
Approach:	They propose a benchmark for evaluating ISA understanding through integrated reasoning over visual context and dialogue.
Outcome:	Experiments show that state-of-the-art models struggle with visually grounded indirect speech acts . linguistic meaning emerges through the relationship between an utterance and situational context .

Copied to clipboard

Xiusheng Huang, Lu Wang, Yequan Wang, Jun Zhao, Kang Liu

Cognitive Analysis Graph-Guided Multi-Turn Safety Enhancement for Large Language Models (2026.findings-acl)

Challenge:	Existing compression methods suffer from bottleneck issues when compression ratio is increased.
Approach:	They propose a novel approach to combine low-rank decomposition and quantization methods to reduce the compression bottleneck.
Outcome:	The proposed method reduces the computational and memory overhead of existing methods while maintaining model accuracy.

Copied to clipboard

Lanxue Zhang, Yuqiang Xie, Fang Fang, Yubing Ren, Xuebin Wang, Yanan Cao

Entropy Ratio Clipping as a Soft Global Constraint for Stable Reinforcement Learning (2026.findings-acl)

Challenge:	Existing methods to align large language models with human preferences lack relationship consideration between question and response.
Approach:	They propose an inference-time safety alignment framework that explicitly models the cognitive process of problem solving through a structured cognitive analysis graph.
Outcome:	The proposed framework enables transparent, controllable reasoning while maintaining strong safety guarantees.

Copied to clipboard

Zhenpeng Su, Leiyu Pan, Minxuan Lv, Tiehua Mei, Zijia Lin, Yuntao Li, Wenping Hu, Ruiming Tang, Kun Gai, Guorui Zhou

BCL: Bayesian In-Context Learning Framework for Information Extraction (2026.findings-acl)

Challenge:	Large language model post-training often adopts an off-policy training paradigm . however, the off-poliicy training model introduces distribution shifts that push the policy beyond the trust region.
Approach:	They propose to use the entropy ratio as a global metric to measure the relative change in policy exploration throughout updates.
Outcome:	Experiments show that the proposed metric improves performance across multiple benchmarks.

Copied to clipboard

Haoliang Liu, Chengkun Cai, Xu Zhao, Han Zhu, Shizhou Huang, Xinglin Zhang, Tao Chen, Jenq-Neng Hwang, Zhang Huaping, Lei Li

G-HiRel: Enhancing the Adaption to Knowledge Updating for Large Language Model Reasoning (2026.findings-acl)

Challenge:	Existing information extraction (IE) tasks rely on in-context learning with large language models.
Approach:	They propose a Bayesian-based in-context learning framework that refines label representations across IE tasks using particle filtering and Bayes updates.
Outcome:	The proposed framework improves performance over existing methods (up to 30%) it underperforms one-shot prompting by a substantial margin on NER tasks and CodeIE fails on RE tasks with near-zero micro-F1.

Copied to clipboard

Yudai Pan, Jiajie Hong, Tianzhe Zhao, Lingyun Song, Lingling Zhang, Yixin Chen, Xuequn Shang

WoW-Bench: Evaluating Fine-Grained Acoustic Perception in Audio-Language Models via Marine Mammal Vocalizations (2026.findings-acl)

Challenge:	Large language models (LLMs) have good performance in multiple reasoning tasks, but are limited to adapt the rapid knowledge updates in the real-world scenario.
Approach:	They propose an LLM reasoning framework with hierarchical relational retrieval for large-scale knowledge updating, named G-HiRel.
Outcome:	The proposed framework achieves superiority in terms of accuracy and interpretability on three benchmarks.

Copied to clipboard

Jaeyeon Kim, Heeseung Yun, Tony Woo, Chao-Han Huck Yang, Gunhee Kim

AdaMARP: An Adaptive Multi-Agent Interaction Framework for General Immersive Role-Playing (2026.findings-acl)

Challenge:	Large audio-language models extend language understanding into the auditory domain, yet their ability to perform low-level listening, such as pitch and duration detection, remains underexplored.
Approach:	They propose a global benchmark to evaluate low-level auditory perception and cognition using marine mammal vocalizations to better assess models’ low- level listening.
Outcome:	The proposed models show performance far below human levels, indicating a need for stronger auditory grounding in LALMs.

Copied to clipboard

Zhenhua Xu, Dongsheng Chen, Shuo Wang, Jian Li, Chengjie Wang, Meng Han, Yabiao Wang

Challenge:	Existing LLMs lack immersion and adaptability, resulting in limited character orchestration and on-the-fly character introduction.
Approach:	They propose an LLM-based framework that allows actors to interact with users in an ongoing narrative.
Outcome:	The proposed framework outperforms commercial LLMs in character consistency, environment grounding, and narrative coherence.

Train in Vain: Functionality-Preserving Poisoning to Prevent Unauthorized Use of Code Datasets (2026.findings-acl)

Copied to clipboard

Yuan Xiao, Jiaming Wang, Yuchen Chen, Wei Song, Jun Sun, Shiqing Ma, Yanzhou Mu, Juan Zhai, Chunrong Fang, Jin Song Dong, Zhenyu Chen

Revisiting the Uniform Information Density Hypothesis in LLM Reasoning (2026.findings-acl)

Challenge:	Existing methods for dataset poisoning require full-dataset poison, which breaks code compilability.
Approach:	They propose a functionality-preserving poisoning approach that injects short, compilable weak-use fragments into executed code paths.
Outcome:	The proposed method contaminates 10% of the dataset while maintaining 100% compilability and functional correctness.

Copied to clipboard

Minju Gwak, Guijin Son, Jaehyung Kim

pQuant: Towards Effective Low-Bit Language Models via Decoupled Linear Quantization-Aware Training (2026.findings-acl)

Challenge:	a recent study has highlighted the fragility of Chain-of-Thought reasoning . a hypothesis suggests that effective communication is achieved by maintaining a stable flow of information.
Approach:	They propose a framework to quantify uniformity of information flow at local and global levels . they propose entropy-based stepwise density metric to quantify this phenomenon .
Outcome:	The proposed framework outperforms alternative signals as predictors of reasoning quality.

Copied to clipboard

Wenzheng Zhang, Bingzheng Liu

Incomplete Prompt Jailbreaks in Large Language Models (2026.findings-acl)

Challenge:	Existing methods for building efficient large language models with sub 2-bit weights are lacking in accuracy and scalability.
Approach:	They propose a method that decouples parameters by splitting linear layers into two specialized branches.
Outcome:	The proposed method achieves state-of-the-art performance in extremely low-bit quantization.

Copied to clipboard

Yeonjea Kim, Bumjin Park, Jaesik Choi

The Best of Both Worlds: Combining Parallel and Sequential Inference Scaling via Aggregation Fine-Tuning (2026.findings-acl)

Challenge:	Large language models (LLMs) are increasingly released as open-weight models with safeguards against harmful requests.
Approach:	They formalize incomplete prompt jailbreaks as incomplete prompts elicit harmful continuations . they identify two functional neurons that delay refusal until sentence termination .
Outcome:	The proposed model fails to generalize across content domains and attractor types . the proposed model can be used to perform more precise and robust IPJ defenses .

Copied to clipboard

Yafu Li, Zhilin Wang, Tingchen Fu, Ganqu Cui, Sen Yang, Yu Cheng

A.S.E: A Repository-Level Benchmark for Evaluating Security in AI-Generated Code (2026.findings-acl)

Challenge:	Empirical results show that AFT-trained models achieve substantial gains with test-time scaling.
Approach:	They introduce a supervised fine-tuning paradigm where models synthesize multiple draft responses into a single, refined answer.
Outcome:	Empirical results show that AFT-trained models outperform baseline models while eliminating external guidance.

Copied to clipboard

Keke Lian, Wang Bin, Lei Zhang, Libo Chen, Junjie Wang, Ziming Zhao, Yujiu Yang, Miaoqian Lin, Haotong Duan, Haoran Zhao, Shuang Liao, Mingda Guo, Quan Jiazheng, Yilu Zhong, Chenhao He, Chen Zichuan, Jie Wu, Haoling Li, Zhaoxuan Li, Jiongchi Yu, Hui LI, Dong Zhang

Large Language Models Are Overconfident in Their Own Responses (2026.findings-acl)

Challenge:	Existing security evaluation benchmarks lack relevance to real-world AI programming tasks . current LLMs struggle with secure coding, research shows .
Approach:	They propose a repository-level evaluation benchmark to assess security of AI-generated code.
Outcome:	The proposed framework mirrors real-world AI programming tasks and offers valuable insights into the state of AI code generation.

Copied to clipboard

Mario Sanz-Guerrero, Manuel Mager, Katharina Von Der Wense

Structured Dialogue Refinement: Building Retrieval-Augmented Question Answering on Goal-Oriented Dialogues (2026.findings-acl)

Challenge:	Prior work has shown that instruction-tuned large language models are less well calibrated than their base pre-trained counterparts.
Approach:	They propose a simple inference-time strategy that frams the model’s answer as user input during confidence elicitation.
Outcome:	The proposed approach reduces overconfidence and improves calibration by up to 26% without retraining.

Copied to clipboard

Bin Wu, Sawan Kumar, Prasetya Ajie Utama, Mohamed Yahya

Thought-Action Graph Reasoning: Faithful and Efficient Reasoning of Large Language Models via Reusing Past Experience (2026.findings-acl)

Challenge:	Retrieval-Augmented Generation (RAG) is widely used for knowledgeintensive question answering (QA), but a large amount of real-world problem-solving knowledge is captured in goal-oriented dialogues.
Approach:	They propose a framework that adapts dialogue corpora for RAG at both retrieval and generation stages without altering the underlying pipeline.
Outcome:	The proposed framework improves retrieval quality and QA performance under dialogue-specific structural challenges.

Copied to clipboard

Zhixiao Qi, Feng Huang, Yunqi Zhang, Shijie Zhang, Qingqing Sun, Yongfeng Huang, Minghu Jiang, Shuai Chen, Tianyi Zhang

XLSR-MamBo: Scaling the Hybrid Mamba-Attention Backbone for Audio Deepfake Detection (2026.findings-acl)

Challenge:	Existing methods for integrating knowledge graphs with LLMs suffer from poor generalization or low reasoning efficiency.
Approach:	They propose a thought-action Graph (TAG) that decomposes LLM-KG interaction trajectories into fine-grained semantic operators and guides LLM to execute on them.
Outcome:	The proposed paradigm outperforms state-of-the-art methods on KGQA benchmarks while reducing the number of LLM calls and generated tokens.

Copied to clipboard

Kwok-Ho Ng, Tingting Song, Yongdong WU, Zhihua Xia

Anchored Cyclic Generation: A Novel Paradigm for Long-Sequence Symbolic Music Generation (2026.findings-acl)

Challenge:	Advanced speech synthesis technologies have enabled highly realistic speech generation, posing security risks that motivate research into audio deepfake detection (ADD).
Approach:	They propose a modular framework that integrates an XLSR front-end with synergistic Mamba-Attention backbones to capture artifacts in spoofed speech signals.
Outcome:	The proposed framework achieves competitive performance on the ASVspoof 2021 LA, DF, and In-the-Wild benchmarks compared to other state-of-the art systems.

Copied to clipboard

Boyu Cao, Lekai Qian, Dehan Li, Haoyu Gu, Mingda Xu, Qi Liu

TRUST: Towards Robust Social Bot Detection via Uncertainty-Guided Pseudo-Labeling and Graph Structure Purification (2026.findings-acl)

Challenge:	Autoregressive sequence modeling has been successful in many domains, but maintaining long-term coherence and structural integrity remains a challenge.
Approach:	They propose an ACG paradigm that relies on anchor features from previously generated musical content to guide subsequent generation during the autoregressive process.
Outcome:	The proposed framework outperforms existing methods in symbolic music generation tasks.

Copied to clipboard

Ruixuan Xu, Mengting Hu, Zhunheng Wang, Ming Jiang, Rui Ying, Zhen Zhang, Hang Gao, Shuaipeng Liu, Renhong Cheng

XGUARD: A Graded Benchmark for Evaluating Safety Failures of Large Language Models on Extremist Content (2026.findings-acl)

Challenge:	Existing graph-based detection models are vulnerable to deceptive message propagation, where bots deliberately interact with legitimate users.
Approach:	They propose a framework to mitigate deceptive message propagation by node-level uncertainty estimation and graph structure purification.
Outcome:	The proposed framework improves on three benchmark datasets and six GNN backbones on real-world social bots.

Copied to clipboard

Vadivel Abishethvarman, Bhavik Chandna, Pratik Jalan, Usman Naseem

PiCSAR: Probabilistic Confidence Selection and Ranking for Reasoning Chains (2026.findings-acl)

Challenge:	Existing safety evaluations rely on binary labels, overlooking the nuanced risk these outputs pose.
Approach:	They propose a framework to assess the severity of extremist content generated by Large Language Models (LLMs) it categorizes model responses into five danger levels (0–4) defined by degree of extremism endorsement .
Outcome:	The proposed framework categorizes model responses into five danger levels (0–4) defined by degree of extremist endorsement, enabling nuanced analysis of failure frequency and severity.

Copied to clipboard

Joshua Ong Jun Leang, Zheng Zhao, Aryo Pradipta Gema, Sohee Yang, Wai-Chung Kwan, Xuanli He, Wenda Li, Pasquale Minervini, Eleonora Giunchiglia, Shay B Cohen

DeepPresenter: Environment-Grounded Reflection for Agentic Presentation Generation (2026.findings-acl)

Challenge:	Recent studies show that large reasoning models (LLMs) achieve strong performance on complex reasoning tasks.
Approach:	They propose a method that scores each candidate generation using the joint log-likelihood of the reasoning and final answer.
Outcome:	The proposed method outperforms baselines with 2x fewer samples in 20 out of 25 comparisons.

Copied to clipboard

Hao Zheng, Guozhao Mo, Xinru Yan, Qianhao Yuan, Wenkai Zhang, Xuanang Chen, Yaojie Lu, Hongyu Lin, Xianpei Han, Le Sun

Challenge:	Existing presentation agents rely on predefined workflows and fixed templates to generate presentations.
Approach:	They propose an agentic framework that adapts to diverse user intents and iterative refinement based on observation.
Outcome:	The proposed framework can be used to generate presentations with environmental observations.

An Iterative Utility Judgment Framework Inspired by Philosophical Relevance via LLMs (2026.findings-acl)

Copied to clipboard

Hengran Zhang, Keping Bi, Jiafeng Guo, Xueqi Cheng

FormulaReasoning: A Dataset for Formula-Based Numerical Reasoning (2026.findings-acl)

Challenge:	Relevance emphasizes the aboutness of a result to a query, while utility refers to the result’s usefulness or value to an information seeker.
Approach:	They propose an Iterative utiliTy judgmEnt fraMework to promote each step in Retrieval-Augmented Generation (RAG) they propose to use relevance ranking, utility judgments, and answer generation to prioritize high-utility results over low-utilitity results.
Outcome:	The proposed framework improves relevance, ranking, and answer generation on retrieval (TREC DL, WebAP), utility judgment task (GTI-NQ), and factoid question-answering (NQ) datasets.

Copied to clipboard

Xiao Li, Bolin Zhu, Kaiwen Shi, Sichen Liu, Yin Zhu, Yiwei Liu, Gong Cheng

CORD: Bridging the Audio–Text Reasoning Gap via Weighted On-policy Cross-modal Distillation (2026.findings-acl)

Challenge:	Existing datasets for numerical reasoning often lack explicit knowledge of formulas . current datasets do not provide process supervision information, resulting in incomplete reasoning .
Approach:	They propose a benchmark for formula-based numerical reasoning with 5,324 questions . they provide annotations in English and Chinese and a formula database as an external knowledge source .
Outcome:	The proposed model includes 5,324 questions requiring calculations grounded in external physics principles.

Copied to clipboard

Hu Jing, Danxiang Zhu, Xianlong Luo, Dan Zhang, Shuwei He, Yishu Lei, Shikun Feng, Hai-Tao Zheng, Jingzhou HE, Yu Sun, Hua Wu, Haifeng Wang

PlanE: Meta Planning of Data, Tuning, and Inference for Extractive-based LLMs (2026.findings-acl)

Challenge:	Large Audio Language Models (LALMs) exhibit a degradation in knowledge and reasoning capabilities . empirical results show that CORD significantly bridges the audio–text performance gap .
Approach:	They propose a framework that performs online cross-modal self-distillation to bridge the acoustic-semantic gap between LALMs and text-based models.
Outcome:	The proposed framework bridges the acoustic-semantic gap between LALMs and text-based models . it employs on-policy reverse KL divergence with importance-aware weighting .

Copied to clipboard

Jiacheng Wang, Weiyan Zhang, Guangya Yu

Challenge:	Existing methods for optimizing LLMs for task-specific tasks are limited due to the sheer volume of data.
Approach:	They propose a Planning framework for constructing Extractive-based LLMs called PlanE . they propose 'data decomposition', instruction tuning, prompt inference and a 'Data-Tuning-Inference' planner .
Outcome:	The proposed framework improves performance across different datasets and on different dataset.

Generalizable Prompt Tuning for Audio-Language Models via Semantic Expansion (2026.findings-acl)

Copied to clipboard

Jaehyuk Jang, Wonjun Lee, Kangwook Ko, Changick Kim

Leveraging Label Semantics and Entity Description Generation for LLM-based Fine-grained Entity Typing (2026.findings-acl)

Challenge:	Prompt tuning has achieved remarkable progress in vision–language models, but its generalization ability in ALMs remains underexplored.
Approach:	They propose a plug-and-play framework that regularizes the prompt embedding space . they propose introducing a semantic expansion loss with margin constraints that promote compactness .
Outcome:	The proposed framework regularizes the prompt embedding space by incorporating semantic neighbors generated by large language models.

Copied to clipboard

Yingying Ma, Hongliang Dai, Xia Zhang, Piji Li

Trident: Self-Supervised Preference Alignment via Triplet Regularization (2026.findings-acl)

Challenge:	Fine-grained entity typing (FET) aims to assign semantically rich and contextually appropriate types to entity mentions.
Approach:	They propose a descriptor-based retrieval-augmented framework that reduces effective label space . they propose to use natural language descriptores as an intermediate semantic representation .
Outcome:	The proposed framework outperforms existing methods under noisy supervision.

Copied to clipboard

Yingnan Guo, Kejia Chen, Xiaofeng Zhang, Zifei Wu, Yu Zhang

Why Do Multilingual Reasoning Gaps Emerge in Reasoning Language Models? (2026.findings-acl)

Challenge:	Large vision-Language Models suffer from noisy supervision and semantic ambiguity in self-supervised settings.
Approach:	They propose a self-supervised framework that constructs reliable preference triplets . they propose 'trident' objective that enforces semantic separation between the triplet components .
Outcome:	The proposed framework outperforms state-of-the-art RLHF and RLAIF benchmarks on LLaVA-1.5-7B and achieves 95.70% precision on POPE using only 4k self-generated triplets and a single epoch.

Copied to clipboard

Deokhyung Kang, Seonjeong Hwang, Daehui Kim, Hyounghun Kim, Gary Lee

EvoMemKG: An Evolvable Memory Agent for Multi-hop Knowledge Graph Reasoning (2026.findings-acl)

Challenge:	Recent reasoning language models (RLMs) achieve strong performance on complex reasoning tasks, yet they still exhibit a multilingual reasoning gap.
Approach:	They propose a strategy that incorporates an English translation into the initial reasoning trace when an understanding failure is detected.
Outcome:	The proposed strategy incorporates an English translation into the initial reasoning trace when an understanding failure is detected.

Copied to clipboard

Shiyu Tian, Shuyue Xing, Zhuoxin Han, Caixia Yuan, Xiaojie Wang

Scaling Laws or Threshold Effects: Exploring the Optimal Vocabulary Size for Balancing Performance and Efficiency in Low-Resource Languages (2026.findings-acl)

Challenge:	Existing methods for integrating knowledge graphs with large language models lack continuous learning capabilities.
Approach:	They propose an agent framework with a dynamic, evolvable memory mechanism specifically designed for KG reasoning.
Outcome:	EvoMemKG achieves state-of-the-art performance without training or tools . it achieves improvements of up to 20% over baseline on multi-hop queries .

Copied to clipboard

Ao Han, Andong Chen, Yuan Sun, Xiaobing Zhao

RouterHGC: Optimized Router for LLM-based Multi-Agent Systems via Heterogeneous Graph Contrastive Learning (2026.findings-acl)

Challenge:	vocab expansion scaling laws are well-established for high-resource languages, but they remain unverified in low-resourced settings.
Approach:	They propose to scale trilingual vocabulary for languages with 140 to 195,000 tokens . they find that BBPE follows a "decline-then-rise" pattern, whereas BPE improves monotonically .
Outcome:	The proposed configuration reduces pre-training duration by over 71% across 1.5B to 8B models while improving downstream performance.

Copied to clipboard

Yitao Xiao, Shaoyong Guo, Guoming Yang, Qingnan Wang, Yinlin Ren, Xuesong Qiu, Qi Feng

Grouped Adaptive Weight Sharing (GAWS): An Inference-Efficient Adaptation Method for Large Language Models (2026.findings-acl)

Challenge:	Large Language Models (LLMs)-driven Multi-Agent Systems (MAS) have demonstrated remarkable scalability and generalizability across complex tasks.
Approach:	They propose a new framework for routing using large language models . they formalize routing as node selection through edge-weight prediction .
Outcome:	The proposed framework outperforms the best single LLM and baselines on five datasets . it achieves 0.80%–6.17% accuracy gains on MATH and HotpotQA while reducing inference cost by 27.40%.

Copied to clipboard

Eman Alsuradi, Junhyun Lee, Kyenghun Lee, Hyeonmok Ko, Fahed Jubair

Lending Eyesight to Language Models: Modeling and Probing Human scanpath through Transformer Decoder (2026.findings-acl)

Challenge:	Low-Rank Adaptation (LoRA) is a new approach to fine-tuning large language models . adapters are lightweight, task specific modules that can be used for adapters in latency-sensitive settings.
Approach:	They propose a low-rank adapter with a weight sharing mechanism that reduces latency by 40% . they analyze LoRA adapters on GPUs and identify segmented function calls as the primary source of latency.
Outcome:	The proposed adapter reduces latency to about 40% of the gap between the unmerged LoRA and the base model while maintaining parameter efficiency and comparable accuracy.

Copied to clipboard

Junlin Li, David Robert Reich, Yu-Yin Hsu

Timesteps of Mamba Align with Human Reading Times (2026.findings-acl)

Challenge:	Decoded language models have shown to exhibit striking parallels with human cognitive processes.
Approach:	They propose a plug-and-play module that transforms an autoregressive language model into an autorregressive eye model and probes it through a linguistic model.
Outcome:	The proposed module can be used to model human-like gaze shifts in language models.

Copied to clipboard

Yuji Yamamoto, Shinnosuke Isono, Yoshinobu Kawahara, Sho Yokoi

Challenge:	In Mamba, the recurrent state transition at each layer conceptually takes some duration of time, the discretization timestep t, determined dynamically in response to the input.
Approach:	They propose to align per-word processing time in a popular state-space language model Mamba with human reading time using a naturalistic reading dataset.
Outcome:	The proposed model can predict reading times comparable to baselines such as word frequency and GPT-2 surprisal and significant even when they are controlled for.

Astra: Activation-Space Tail-Eigenvector Low-Rank Adaptation of Large Language Models (2026.findings-acl)

Copied to clipboard

Kainan Liu, Yong Zhang, Ning Cheng, Yun Zhu, Yanmeng Wang, Shaojun Wang, Jing Xiao

AROMA: Augmented Reasoning Over a Multimodal Architecture for Virtual Cell Genetic Perturbation Modeling (2026.findings-acl)

Challenge:	Existing methods for fine-tuning pre-trained models are limited due to suboptimal activation subspaces.
Approach:	They propose a method that leverages tail eigenvectors of model output activations to construct low-rank adapters.
Outcome:	The proposed method outperforms existing methods across 16 benchmarks and surpasses full fine-tuning in certain scenarios.

Copied to clipboard

Wang Zhenyu, Geyan Ye, Wei Liu, Man Tat Alexander Ng

Exploring Two-Phase Continual Instruction Fine-tuning for Multilingual Adaptation in Large Language Models (2026.findings-acl)

Challenge:	Existing methods for virtual cell genetic perturbation modeling suffer from unconstrained reasoning, uninterpretable predictions, and retrieval signals that are weakly aligned with regulatory topology.
Approach:	They propose an Augmented Reasoning Over a Multimodal Architecture for virtual cell genetic perturbation modeling.
Outcome:	The proposed model outperforms existing methods across multiple cell lines and remains robust under zero-shot evaluation on unseen cells.

Copied to clipboard

Divyanshu Aggarwal, Sankarshan Damle, Navin Goyal, Satya Lokam, Sunayana Sitaram

SCALER: Synthetic Scalable Adaptive Learning Environment for Reasoning (2026.findings-acl)

Challenge:	A key challenge for Large Language Models (LLMs) is improving their Multilingual instruction-following ability over time without deteriorating their ability in languages they already excel at, typically English.
Approach:	They propose a two-phase Continual Fine-tuning setup to improve a model's Multilingual adaptability by comparing an English-only LLM with a multilingual instruction dataset.
Outcome:	The proposed model improves on two-phase Continual Fine-tuning (CFT) setups on a multilingual instruction dataset.

Copied to clipboard

Caijun Xu, Changyi Xiao, Zhongyuan Peng, Xinrun Wang, Yixin Cao

Probing Social Identity Bias in Chinese LLMs with Gendered Pronouns and Social Groups (2026.findings-acl)

Challenge:	Reinforcement learning (RL) is a principled way to enhance the reasoning capabilities of large language models, yet its effectiveness hinges on training signals that remain informative as models evolve.
Approach:	They propose a framework that sustains effective learning signals through adaptive environment design that transforms real-world programming problems into verifiable reasoning environments with controllable difficulty and unbounded instance generation.
Outcome:	The proposed framework outperforms baselines across diverse reasoning benchmarks and exhibits more stable, long-horizon training dynamics.

Copied to clipboard

Geng Liu, Li Feng, Junjie Mu, Mengxiao Zhu, Francesco Pierri

AutoVecCoder: Teaching LLMs to Generate Explicitly Vectorized Code (2026.findings-acl)

Challenge:	Large language models (LLMs) are increasingly deployed in user-facing applications, raising concerns that they reflect and amplify social biases.
Approach:	They propose a Mandarin-specific evaluation framework to examine social identity biases in Chinese LLMs using Mandarin-based prompts.
Outcome:	The proposed framework compares ingroup (“We”) and outgroup (“They”) framings across 240 social groups salient in the Chinese context, using a two-tiered measurement framework that assesses both sentiment and toxicity.

Copied to clipboard

ShangZhan Li, Xinyu Yin, Xuanyu Jin, Ye He, Yuxin Zhou, Yuxuan Li, Xu Han, Wanxiang Che, Qi Shi, Ting Liu, Maosong Sun

CompassLLM: A Multi-Agent Approach toward Geo-Spatial Reasoning for Popular Path Query (2026.findings-acl)

Challenge:	Current development practices face a dichotomy between automation and performance.
Approach:	They propose a framework to empower LLMs with the capability of automated explicit vectorization.
Outcome:	The proposed framework achieves state-of-the-art performance on the SSE and AVX subsets of SimdBench.

Copied to clipboard

Md. Nazmul Islam Ananto, Shamit Fatin, Mohammed Eunus Ali, Md Rizwan Parvez

Free-MAD: Consensus-Free Multi-Agent Debate (2026.findings-acl)

Challenge:	Existing algorithms and machine learning methods require model training, parameter tuning, and retraining when accommodating data updates.
Approach:	They propose a multi-agent framework that leverages the reasoning capabilities of Large Language Models into the geo-spatial domain to solve the popular path query.
Outcome:	Experiments on real and synthetic datasets show that CompassLLM performs better than existing models while being cost-effective.

Copied to clipboard

Yu Cui, Hang Fu, Haibin Zhang, Licheng Wang, Cong Zuo

Dialectical Structured Reasoning for Explainable Multimodal Fake News Detection (2026.findings-acl)

Challenge:	Existing multi-agent debate methods rely on multiple rounds of interaction among agents to reach consensus, and the final output is decided by majority voting in the last round.
Approach:	They propose a multi-agent debate framework that eliminates the need for consensus among agents and reconstructs the debate phase by introducing anti-conformity.
Outcome:	Experiments on eight benchmark datasets show that Free-MAD significantly improves reasoning performance while requiring only a single-round debate and thus reducing token costs.

Copied to clipboard

Ruichao Yang, Yufan Bian, Wei Gao, Bo-Wen Zhang, Jing Ma, Hongzhan Lin, Ziyang Luo, Xiaobin Zhu, Xu-Cheng Yin

Unlocking Human-Like Visible Logic: How Logic Diagrams Boost Logic Reasoning in Large Language Models? (2026.findings-acl)

Challenge:	Existing fake news detection models are opaque and lack deductive transparency . a framework for dialectical structured reasoning is proposed to address this limitation .
Approach:	They propose a framework that model fake news detection as an explicit dialectical process over multimodal social context.
Outcome:	The proposed framework achieves state-of-the-art while producing transparent explanations that mirror human reasoning process.

Copied to clipboard

Xin Wu, Yuqi Bu, Mengchen Zhao, Qingbao Huang, Yi Cai

Let Retrievers Think Before Action: Thought-Augmented Embedding for Dense Retrieval (2026.findings-acl)

Challenge:	Recent advances in large language models (LLMs) have demonstrated their remarkable capabilities in natural language understanding and generation, but they struggle with formal logical reasoning.
Approach:	They propose to incorporate visual logic diagrams into LLMs’ reasoning workflows to enhance their performance on formal logic tasks.
Outcome:	The proposed model improves on syllogistic and conditional reasoning with programmatically generated Venn, Euler, and Linear diagrams.

Copied to clipboard

Ruiran Yan, Wen Xiong, Ze Liu, Chaozhuo Li, Hao Liao, Defu Lian, Zheng Liu

TravelBehaviorQA: A Benchmark Dataset for Behavioral Interpretation of GPS Trajectories (2026.findings-acl)

Challenge:	Large language models have demonstrated that explicit step-by-step thinking can substantially improve performance on complex tasks.
Approach:	They propose a model that generates preliminary thoughts for input queries before document retrieval.
Outcome:	The proposed model generates preliminary thoughts for input queries before document retrieval.

Copied to clipboard

Dongyang Zhen, Niping Duan, Huan Zhou, Qingbin Cui

SpiralThinker: Latent Reasoning through an Iterative Process with Text–Latent Interleaving (2026.findings-acl)

Challenge:	Recent advances in large language models (LLMs) have enabled strong performance on reasoning, summarization, and dialogue tasks across diverse domains.
Approach:	They propose a large-scale benchmark dataset that reframes trajectory analysis as a language-based understanding task.
Outcome:	The proposed dataset compares GPS trajectories with human-grounded question-answering (QA) pairs.

Copied to clipboard

Shengmin Piao, Sanghyun Park

Verifier-Free RL for LLMs via Intrinsic Gradient-Norm Reward (2026.findings-acl)

Challenge:	Existing latent reasoning methods lack mechanisms to ensure stable reasoning dynamics in latent space and a systematic way to interleave implicit and explicit reasoning.
Approach:	They propose a framework that performs iterative updates over latent representations while enabling interleaved reasoning across latent and textual steps.
Outcome:	SpiralThinker achieves state-of-the-art among latent reasoning baselines.

Copied to clipboard

Xuexiang Wen, Hang Yu, Linchao Zhu, Gaoang Wang

NovBench: Evaluating Large Language Models on Academic Paper Novelty Assessment (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have demonstrated remarkable capabilities in reasoning tasks and general tasks.
Approach:	They propose a "Verifier-free Intrinsic Gradient-Norm Reward" that uses only the policy model itself.
Outcome:	The proposed reward outperforms the state-of-the-art RLIF baseline INTUITOR on math benchmarks and shows cross-domain transfer to code benchmarks when trained only on math data.

Copied to clipboard

Wenqing Wu, Yi Zhao, Yuzhuo Wang, Siyou Li, Juexi Shao, Yunfei Long, Chengzhi Zhang

LaMPE: Length-aware Multi-grained Positional Encoding for Adaptive Long-context Scaling Without Training (2026.findings-acl)

Challenge:	Existing methods for evaluating novelty have been proposed, but there is no systematic evaluation of their ability to generate novelty evaluations.
Approach:	They propose a benchmark to evaluate large language models’ ability to generate novelty evaluations in support of human peer review.
Outcome:	The proposed framework evaluates the quality of LLM-generated novelty evaluations under different prompting strategies.

Copied to clipboard

Sikui Zhang, Guangze Gao, Ziyun Gan, Chunfeng Yuan, Zefeng Lin, Houwen Peng, Bing Li, Weiming Hu

Challenge:	Large language models (LLMs) experience significant performance degradation when the input exceeds the pretraining context window due to the out-of-distribution (OOD) behavior of Rotary Position Embedding (RoPE).
Approach:	They propose a training-free method that remaps out-of-distribution (OOD) positions into the in-distance range with fixed mapping strategies, ignoring the dynamic relationship between input length and effective context window.
Outcome:	Experiments on three representative LLMs across five mainstream long-context benchmarks show that the proposed method achieves significant performance improvements compared to existing methods.

Can Small Vision–Language Models Perform Sign Language Translation? (2026.findings-acl)

Copied to clipboard

Anal Roy Chowdhury, Debarshi Kumar Sanyal

FastKV: Decoupling of Context Reduction and KV Cache Compression for Prefill-Decoding Acceleration (2026.findings-acl)

Challenge:	Vision-Language Models (VLMs) have shown strong generalization across multimodal tasks, but their capacity to handle sign language translation (SLT) remains unclear.
Approach:	They propose entity- and semantics-aware metrics tailored for SLT to evaluate their performance.
Outcome:	The proposed metrics highlight the limitations of general-purpose VLMs to SLT, unlike their applicability in other tasks.

Copied to clipboard

Dongwon Jo, Jiwon Song, Yulhwa Kim, Jae-Joon Kim

Beyond Outcome Verification: Verifiable Process Reward Models for Structured Reasoning (2026.findings-acl)

Challenge:	Large language models (LLMs) excel at handling long-context sequences, but require substantial prefill computation and key-value (KV) cache.
Approach:	They propose a KV cache compression framework that decouples prefill computation from decoding KV budget.
Outcome:	The proposed framework reduces latency in prefill and decoding by leveraging the stabilization of token importance in later layers.

Copied to clipboard

Massimiliano Pronesti, Anya Belz, Yufang Hou

Auto-Stega: An Agent-Driven System for Lifelong Strategy Evolution in LLM-Based Text Steganography (2026.findings-acl)

Challenge:	Recent work on reinforcement learning with verifiable rewards (RLVR) has shown that large language models can be substantially improved using outcome-level verification signals.
Approach:	They propose a framework where intermediate reasoning steps are checked by deterministic, rule-based verifiers.
Outcome:	The proposed framework achieves 20% higher F1 than state-of-the-art models and 6.5% higher than verifiable outcome rewards, with substantial gains in evidence grounding and logical coherence.

Copied to clipboard

Jiuan Zhou, Yu Cheng, Yuan Xie, Zhaoxia Yin

Detecting AI-Generated Video: A Vision–Language Dual-View Survey (2026.findings-acl)

Challenge:	prevailing methods rely on hand-crafted or pre-specified strategies and struggle to balance efficiency, imperceptibility, and security, particularly at high embedding rates.
Approach:	They propose an agent-driven self-evolving framework that is the first to realize self-changing steganographic strategies by automatically discovering, composing, and adapting strategies at inference time.
Outcome:	The proposed framework achieves 42.2% perplexity and 1.6% anti-steganalysis performance over SOTA methods at high embedding rates.

Copied to clipboard

Dylan Xinming Hou, Juntian Zhang, Xu Gu, Yichen Wu, Nils Lukas, Gus Xia, Xiuying Chen, Yuhan Liu

UnIte: Uncertainty-based Iterative Document Sampling for Domain Adaptation in Information Retrieval (2026.findings-acl)

Challenge:	realism of AI-generated Videos (AIGC-V) rendering artifact-centric detection insufficient, authors argue . a vision–language dual-view taxonomy is proposed to systematize this rapidly evolving field .
Approach:	They propose a Vision–Language Dual-View taxonomy to systematize AIGC-V detection . they propose realism of AI-generated Videos is rendering traditional inspection insufficient .
Outcome:	The proposed model aims to show that the existing methods are consistent with real-world facts.

Copied to clipboard

Jongyoon Kim, Minseong Hwang, Seung-won Hwang

Improving LLM Code Reasoning via Semantic Equivalence Self-Play with Formal Verification (2026.findings-acl)

Challenge:	Existing methods focus on diversity but fail to capture model uncertainty.
Approach:	They propose a method to generalize neural retrievers to an unseen domain by generating pseudo queries on target domain documents.
Outcome:	The proposed method improves performance on large datasets with small and large models while limiting the learning utility of the current model.

Copied to clipboard

Poon Tsz Nok, Antonio Valerio Miceli Barone

Towards Identification and Intervention of Safety-Critical Parameters in Large Language Models (2026.findings-acl)

Challenge:	Existing approaches to training large language models fail to capture deep semantic properties and edge cases.
Approach:	They propose a framework that leverages Liquid Haskell proofs for validating equivalence and execution-based counterexamples for inequivalent training.
Outcome:	The proposed framework achieves 13.3pp accuracy gain on EquiBench and consistent gains on PySecDB.

Copied to clipboard

Weiwei Qi, Zefeng Wu, Tianhang Zheng, Zikang Zhang, Xiaojun Jia, Zhan Qin, Kui Ren

OPINE: A Prior-calibrated Scoring Framework for LLM-based Multi-label Scientific Opinion Classification (2026.findings-acl)

Challenge:	Existing safety-related methodologies for large language models are lacking . despite advances in safety alignment techniques, safeguarding LLMs during adaptation to various tasks remains a challenge.
Approach:	They propose a framework to quantify how different parameters affect LLM safety . they propose two targeted intervention paradigms for safety enhancement and preservation .
Outcome:	The proposed framework reveals safety-critical patterns across different LLM architectures.

Copied to clipboard

Mengting Zhang, Gaofeng Pan, Zhixiong Zhang, Yang Li, Guangyin Zhang

Vocab Diet: Reshaping the Vocabulary of LLMs via Vector Arithmetic (2026.findings-acl)

Challenge:	Existing methods for scientific opinion classification rely on direct label generation and are limited by the multi-label nature of scientific expressions.
Approach:	They propose a framework that reformulates scientific opinion classification as a controllable pipeline.
Outcome:	The proposed framework outperforms baseline models on 18 discourse functions in micro, macro, and example settings.

Copied to clipboard

Yuval Reif, Guy Kaplan, Roy Schwartz

RecMem: Recurrence-based Memory Consolidation for Efficient and Effective Long-Running LLM Agents (2026.findings-acl)

Challenge:	Large language models often encode word-form variation as linear directions in the embedding space.
Approach:	They propose a compact reshaping of large language models' vocabulary by using shared vectors instead of unique tokens.
Outcome:	The proposed approach frees 10-40% of vocabulary slots to be reallocated where tokenization is inefficient.

Copied to clipboard

Zijie Dai, Shiyuan Deng, Sheng Guan, Yizhou Tian, Xin Yao, Xiao Yan, James Cheng

Let the Comments Speak: A Multi-Agent Framework based on Large Language Model for Comment-Guided Code Refactoring (2026.findings-acl)

Challenge:	Existing memory systems invoke LLMs to extract episodic and semantic memory, and this leads to substantial token consumption.
Approach:	They propose a method that stores incoming interactions in a subconscious memory layer and encodes them using lightweight embedding models for retrieval.
Outcome:	Experiments show that RecMem reduces the memory construction token cost of three SOTA memory systems by up to 87% while exceeding their accuracy.

Copied to clipboard

Zixuan Wang, Wutong Yu, Deyu Zhou

AdapTime: Enabling Adaptive Temporal Reasoning in Large Language Models (2026.findings-acl)

Challenge:	Current Large Language Models focus on syntax and ignore the vital semantic signals in code comments.
Approach:	They propose a Multi-Agent framework for COmment-guided code Refactoring that populates original code with precise comments to provide necessary semantic guidance for subsequent refactoring.
Outcome:	The proposed framework significantly improves code quality and achieves higher developer acceptance compared to baselines.

Copied to clipboard

Yimin Deng, Yejing Wang, Zhenxi Lin, Zichuan Fu, Guoshuai Zhao, Derong Xu, Yefeng Zheng, Xiangyu Zhao, Xian Wu, Li Zhu, Xueming Qian

Hierarchical Representation Alignment Learning of Diffusion Transformers for Neural Audio Codec (2026.findings-acl)

Challenge:	Existing methods for temporal reasoning are limited and apply a fixed pipeline to all questions.
Approach:	They propose an adaptive temporal reasoning method that dynamically executes reasoning steps based on context and task requirements.
Outcome:	Experiments on two temporal QA benchmarks show the proposed method works.

Copied to clipboard

Sang-Hoon Lee, Ha-Yeong Choi

CoV: Chain-of-View Prompting for Spatial Reasoning (2026.findings-acl)

Challenge:	Recent advances in diffusion and conditional flow matching models for low-resolution domains are underexplored.
Approach:	They propose a CFM-based model that iteratively generates raw waveform in low-bitrate conditions . they propose DVQ, a factorized quantization method that uses a single quantizer .
Outcome:	The proposed model outperforms state-of-the-art neural audio codecs in audio quality and semantic intelligibility under low-bitrate conditions.

Copied to clipboard

Haoyu Zhao, Akide Liu, Zeyu Zhang, Weijie Wang, Feng Chen, Ruihan Zhu, Gholamreza Haffari, Bohan Zhuang

A Graph Talks, But Who’s Listening? Rethinking Evaluations for Graph-Language Models (2026.findings-acl)

Challenge:	Embodied question answering requires collecting context that is distributed across multiple viewpoints . most recent vision–language models (VLMs) are constrained to a fixed and finite set of input views .
Approach:	They propose a training-free, test-time reasoning framework that transforms a VLM into an active viewpoint reasoner through a coarse-to-fine exploration process.
Outcome:	The proposed framework improves LLM-Match performance by 11.98% on four mainstream VLMs.

Copied to clipboard

Soham Petkar, Hari Aakash K, Anirudh Vempati, Akshit Sinha, Ponnurangam Kumaraguru, Chirag Agarwal

Rejection-to-Acceptance Transition: Model Editing-Based Jailbreak Backdoor Injection Not Limited to Few Output Tokens (2026.findings-acl)

Challenge:	Existing benchmarks for Graph-Language Models (GLMs) do not assess true multimodal integration.
Approach:	They propose a benchmark to evaluate multimodal reasoning over graph topology and textual semantics.
Outcome:	The proposed benchmarks show that strong performance is achievable using textual or structural features in isolation, bypassing the need for joint reasoning.

Copied to clipboard

Shiji Yang, Min Cai, Hao Xiong, Congyao Mei, Haodong Zou, Shicheng Tan, Jie Chen, Fulan Qian, Shu Zhao

Causal Evidence Extraction and Triangulation in Crisis Reports using Large Language Models: A ReliefWeb-based Study (2026.findings-acl)

Challenge:	Existing methods for jailbreaking LLMs are implemented by binding backdoors to predefined phrases as first few output tokens, inducing the LLM’s next-token prediction to produce continuous responses.
Approach:	They propose a model editing-based jailbreak backdoor attack that hijacks LLM representations into a acceptance domain rather than binding to a few output tokens.
Outcome:	The proposed model editing method outperforms existing methods, showing stronger jailbreak capabilities across LLMs and datasets.

Copied to clipboard

Yuanjun Zhang, Mourad Oussalah

Sounding vs. Being an Expert: Disentangling Authority, Register and Cultural Impact in Sycophantic LLMs (2026.findings-acl)

Challenge:	Humanitarian crises generate large volumes of narrative situation reports describing interventions and evolving outcomes.
Approach:	They propose a large language model pipeline that extracts structured intervention-outcome records with direction and strength attributes.
Outcome:	The proposed pipeline extracts structured intervention-outcome records with direction and strength attributes.

Copied to clipboard

Gabriele Maraia, Fabio Massimo Zanzotto, Leonardo Ranaldi

Challenge:	Large Language Models exhibit sycophancy, a tendency to align with user assertions even when they conflict with factual correctness.
Approach:	They propose an adversarial evaluation framework that isolates two drivers of credibility: explicit authority (credentials) and implicit authority (linguistic register).
Outcome:	The proposed framework disentangles two drivers of credibility: explicit authority (credentials) and implicit authority (linguistic register).

Datamart-Agent: LLM-Driven Game-Theoretic Agent for Data Marketplace Modeling (2026.findings-acl)

Copied to clipboard

Pangjing Wu, Peter Q. Chen, Xiaodong Li, Wenqi Fan, Li Qing

Challenge:	Existing studies on data marketplaces model static equilibria and complete information, which limits their realism.
Approach:	They propose an LLM-driven game-theoretic agent that makes equilibrium-consistent decisions in analytically tractable data marketplaces with evolving and incomplete-information.
Outcome:	The proposed framework matches equilibrium-consistent decision execution in a static data marketplace with a dynamic game tree memory and mechanism-guided reflection without updating parameters.

Multi-Drafter Speculative Decoding with Alignment Feedback (2026.findings-acl)

Copied to clipboard

Taehyeon Kim, Hojung Jung, Se-Young Yun

Challenge:	Existing methods to accelerate large language model (LLM) inference use a smaller model to draft future tokens, which are then verified by the target LLM.
Approach:	They propose a unified framework that integrates multiple drafters into the SD process.
Outcome:	Extensive experiments show that MetaSD outperforms single-drafter approaches.

Probabilistic Depression Detection from Textual Time Series (2026.findings-acl)

Copied to clipboard

Fabian Schmidt, Seyedehmoniba Ravan, Vladimir Vlassov

Distilling Examples into Task Instructions: Enhanced In-Context Learning for Real-World B2B Conversations (2026.findings-acl)

Challenge:	Existing models for depression severity estimations lack uncertainty estimates and temporal interpretability.
Approach:	They propose a Probabilistic framework for Depression Detection from clinical interview utterance sequences that predicts PHQ-8 scores while modeling calibrated uncertainty.
Outcome:	The proposed framework achieves competitive performance among text-only systems and produces well-calibrated intervals.

Copied to clipboard

Guy Rotman, Adi Kopilov, Danit Berger Zalmanson, Omri Allouche

Multilingual Tokenization through the Lens of Indian Languages: Challenges and Insights (2026.findings-acl)

Challenge:	In-context learning (ICL) is the standard method for low-resource classification, yet its efficacy in specialized domains remains largely unexplored.
Approach:	They propose a framework that distills verbose examples into compact, interpretable representations of structured classification criteria and precise task descriptions.
Outcome:	The proposed method achieves 99% reduction in token usage and improves macro-averaged AUC by up to 7% over traditional ICL.

Copied to clipboard

Maharaj Brahma, N J Karthika, Rajat Verma, Nagasai Saketh Naidu, Rohit Saluja, Maunendra Sankar Desarkar, Ganesh Ramakrishnan

TrimTokenator: Towards Adaptive Visual Token Pruning for Large Multimodal Models (2026.findings-acl)

Challenge:	Existing tokenizers are often skewed towards high-resource languages limiting their effectiveness for linguistically diverse and morphologically rich languages.
Approach:	They evaluate multilingual tokenization across 17 Indic languages spanning 11 scripts and two language families.
Outcome:	The proposed method improves tokenization quality and vocabulary size in 17 languages . poor tokenization can lead to increase in sequence lengths, fragment meaningful units, weaken model's ability to capture linguistic structure and semantics.

Copied to clipboard

Hao Zhang, Lyu Mengsi, Chenrui He, Yulong Ao, Yonghua Lin

TTVS: Boosting Self-Exploring Reinforcement Learning via Test-time Variational Synthesis (2026.findings-acl)

Challenge:	Existing token pruning methods rely on costly calibration or suboptimal importance metrics, leading to redundant retained tokens.
Approach:	They propose a token pruning strategy that preserves cross-modal alignment and informational diversity.
Outcome:	The proposed method maintains strong performance while reducing tokens by 88.9% on two models.

Copied to clipboard

Sikai Bai, Haoxi Li, Jie Zhang, Yongjiang Liu, Song Guo

Parallel Context-of-Experts Decoding for Retrieval Augmented Generation (2026.findings-acl)

Challenge:	Existing test-time methods are limited in specialized or novel domains where supervision is prohibitively expensive or unavailable.
Approach:	They propose a framework that augments training stream from unlabeled test queries.
Outcome:	Extensive experiments show TTVS outperforms state-of-the-art RL-based techniques on unlabeled test-time data.

Copied to clipboard

Giulio Corallo, Paolo Papotti

LLaTiSA: Towards Difficulty-Stratified Time Series Reasoning from Visual Perception to Semantics (2026.findings-acl)

Challenge:	Retrieval Augmented Generation relies on concatenating documents into a long context prompt, causing prefill bottlenecks.
Approach:	They propose a training-free framework that shifts evidence aggregation from attention to decoding . they treat retrieved documents as isolated "experts", synchronizing their predictions via a retrieval-aware extension of context-awful decoding.
Outcome:	The proposed framework shifts evidence aggregation from attention to decoding . it treats retrieved documents as isolated experts, synchronizing their predictions .

Copied to clipboard

Yueyang Ding, HaoPeng Zhang, Rui Dai, Yi Wang, Tianyu Zong, Kaikui Liu, Xiangxiang Chu

NeedleChain: Measuring Intact Context Comprehension Capability of Large Language Models (2026.findings-acl)

Challenge:	Current research hinders the development of unified Time Series Reasoning Models (TSRMs) time series data are a fundamental modality for capturing the temporal dynamics of complex systems.
Approach:	They propose a time series reasoning model that integrates visualized patterns with precision-calibrated numerical tables to enhance the temporal perception of Vision-Language Models.
Outcome:	The proposed model outperforms existing models and exhibits robust out-of-distribution generalization across diverse tasks and real-world scenarios.

Copied to clipboard

Hyeonseok Moon, Heuiseok Lim

From Scores to Preferences: Redefining Evaluation Paradigm for Speech Quality Reward Modeling (2026.findings-acl)

Challenge:	Existing benchmarks for context understanding embed query-irrelevant content . this shifts evaluation toward retrieving relevant snippets rather than fully integrating all provided information.
Approach:	They propose a benchmark to evaluate whether models can faithfully incorporate all given evidence . they propose 'needlechain' benchmark to test whether models incorporate all available information .
Outcome:	The proposed benchmarks overestimate the ability of large language models to integrate all given evidence when the context is entirely query-relevant.

Copied to clipboard

Yifei Cao, Changhao Jiang, Jiabao Zhuang, Jiajun Sun, Ming Zhang, Zhiheng Xi, Hui Li, Shihan Dou, Yuran Wang, Yunke Zhang, Tao Ji, Tao Gui, Qi Zhang, Xuanjing Huang

FIND: Toward Multimodal Financial Reasoning and Question Answering for Indic Languages (2026.findings-acl)

Challenge:	Experimental results show that the MOS-aware GRM significantly improves fine-grained speech quality discrimination.
Approach:	They propose a MOS-aware reward model that incorporates MOS gap into reward function during reinforcement learning.
Outcome:	The proposed model significantly improves fine-grained speech quality discrimination.

Copied to clipboard

Sarmistha Das, Vaibhav Vishal, Syed Ibrahim Ahmad, Sriparna Saha, Manish Gupta

Think Smart, Not Hard: Difficulty Adaptive Reasoning for Large Audio Language Models (2026.findings-acl)

Challenge:	Existing benchmarks for numerical reasoning in multilingual Indic languages are inadequate . e.g., FinVQA is a framework for evaluating financial numerical reasoning .
Approach:	They propose a framework that combines supervised fine-tuning with constraint-aware decoding to promote faithful numerical reasoning.
Outcome:	The proposed framework spans English, Hindi, Bengali, Marathi, Gujarati, and Tamil . it combines supervised fine-tuning with constraint-aware decoding to promote faithful numerical reasoning .

Copied to clipboard

Zhichao Sheng, Shilin Zhou, Chen Gong, Zhenghua Li

Can Large Language Models Infer Human Actions and Motives? Evaluation in Social Prediction and Inspection Games (2026.findings-acl)

Challenge:	Existing methods to determine whether to perform reasoning lack fine-grained mechanisms to adapt reasoning length to problem complexity.
Approach:	They propose a difficulty-adaptive reasoning method that dynamically links reasoning length to the model’s perceived problem difficulty.
Outcome:	The proposed method reduces average reasoning length by 50%, achieving higher efficiency without sacrificing accuracy.

Copied to clipboard

Kaleen Shrestha, Abhinav Gupta, Harish Dukkipati, Zhonghao Shi, Maja Mataric

Self-Consistency from Only Two Samples: CoT–PoT Ensembling for Efficient LLM Reasoning (2026.findings-acl)

Challenge:	Game theory provides a framework for studying human behaviors through incentivized games that simulate social situations.
Approach:	They used two validated games from the cognitive science literature to study how well several recent open- and closed-source LLMs predict player actions with underlying human motives.
Outcome:	The results show that state-of-the-art LLMs can achieve accuracy close to human levels in predicting players’ actions with underlying human motives in SPGs, but failed to recognize statistical patterns in players’ action.

Copied to clipboard

Raman Saparkhan, Majd Hawasly, Md Rizwan Parvez, Mohammad Raza

ZoomRAG: Hierarchical Random-walk Zooming across Multi-scale Information Graphs for Fast and Accurate RAG (2026.findings-acl)

Challenge:	Self-consistency (SC) is a popular technique for improving the reasoning accuracy of large language models but it comes at a high computational cost due to extensive sampling.
Approach:	They propose a hybrid ensembling approach that leverages the complementary strengths of Chain-of-Thought and Program-of -Thus . they propose encapsulating two different modes of reasoning to create a single output and a final answer is selected as the most frequently occurring one among these outputs.
Outcome:	The proposed approach reduces the number of samples required for SC by 9.3x . the majority of tasks can be addressed with only two samples, which has not been possible with prior methods.

Copied to clipboard

Xianming Hu, Jingyang Chen, Bin Tang, Yihe Liu, Yihong Huang, Hongbo Zhao, Nuoyi Chen, Jie Zhang, Ping Li, Kai Zhang

Training Language Models to Use Prolog as a Tool (2026.findings-acl)

Challenge:	retrieval-augmented generation (RAG) is a powerful tool for NLP applications . but it is challenging to encode large knowledge bases as compact offline structures .
Approach:	They propose a coarse-to-fine hierarchical graph inference method that uses random walks to retrieve information from a corpus of documents.
Outcome:	The proposed method reduces offline indexing costs and accelerates retrieval.

Copied to clipboard

Niklas Mellgren, Peter Schneider-Kamp, Lukas Galke Poech

Frame of Reference: Addressing the Challenges of Common Ground Representation in Situational Dialogs (2026.findings-acl)

Challenge:	Language models often produce plausible but incorrect reasoning traces that are difficult to verify.
Approach:	They train language models to use Prolog as an external symbolic reasoning tool . they find an accuracy–auditability trade-off between tuning for correctness alone and using Prolog only for the final computation .
Outcome:	The proposed model outperforms supervised fine-tuning on a clean version of GSM8K.

Copied to clipboard

Biswesh Mohapatra, Théo Charlot, Giovanni Duca, Mayank Palan, Laurent Romary, Justine Cassell

MultiDx: A Multi-Source Knowledge Integration Framework towards Diagnostic Reasoning (2026.findings-acl)

Challenge:	Prior studies have demonstrated that Large Language Models (LLMs) are capable of performing grounding acts such as requesting clarification or producing acknowledgments, yet relatively little work has investigated how common ground can be explicitly represented and stored for later use.
Approach:	They propose to use relational references to represent common ground in situated dialogues and propose to improve both the establishment of common ground and its subsequent use in the conversation.
Outcome:	The proposed models can establish and exploit common ground in situated dialogues and improve its subsequent use.

Copied to clipboard

Yimin Deng, Zhenxi Lin, Yejing Wang, Guoshuai Zhao, Pengyue Jia, Zichuan Fu, Derong Xu, Yefeng Zheng, Xiangyu Zhao, Li Zhu, Xian Wu, Xueming Qian

Training Verifier to Assessing Complex Real-World Tool-Use Trajectories (2026.findings-acl)

Challenge:	Existing approaches focus on diagnostic reasoning based on internal model knowledge or static knowledge bases.
Approach:	They propose a two-stage diagnostic reasoning framework that integrates multi-perspective evidence to generate a diagnostic prediction.
Outcome:	The proposed method generates suspected diagnoses and reasoning traces from web search, SOAP-formatted case, and clinical case database.

Copied to clipboard

Linzhuang Sun, Mingyang Chen, Hao Liang, Tianpeng Li, Zhou Yijie, Chenzheng Zhu, Tianyu Guo, Huanyao Zhang, Jingxuan Wei, Bihui Yu, Fan Yang, Wentao Zhang

Challenge:	Existing methods for training effective AI agents often resort to synthetic data generation.
Approach:	They propose a plug-and-play framework for data quality control in tool-use scenarios . they construct a tool-verify dataset and release a benchmark to assess its performance .
Outcome:	The proposed framework surpasses Qwen2.5-72B-Instruct on Tool-V-Bench and the previous APIGen-MT dataset.

HyperAdaLoRA: Accelerating LoRA Rank Allocation During Training via Hypernetworks without Sacrificing Performance (2026.findings-acl)

Copied to clipboard

Hao Zhang, Zhenjia Li, Yifan Gao, Xi Xiao, Heng Zhang, Shuyang Zhang, null Xiaoxincc, Bo Huang, Yuhang Wu, Tianyang Wang, Hao Xu

ConSensus: Multi-Agent Collaboration for Multimodal Sensing (2026.findings-acl)

Challenge:	Low-Rank Adaptation (LoRA) assumes a uniform rank r for each incremental matrix, not accounting for the varying significance of weight matrices across modules and layers.
Approach:	They propose a framework that allows for faster convergence of low-rank adaptive models . they use a hypernetwork to prune the outputs of the hypernetworks to generate parameters .
Outcome:	The proposed framework accelerates convergence of AdaLoRA by leveraging a hypernetwork.

Copied to clipboard

Hyungjun Yoon, Mohammad Malekzadeh, Sung-Ju Lee, Fahim Kawsar, Lorena Qendro

Beyond High-Entropy Exploration: Correctness-Aware Low-Entropy Segment-Based Advantage Shaping for Reasoning LLMs (2026.findings-acl)

Challenge:	Large language models are increasingly grounded in sensor data to perceive and reason about human physiology and the physical world.
Approach:	They propose a training-free multi-agent collaboration framework that decomposes multimodal sensing tasks into specialized, modality-aware agents.
Outcome:	The proposed framework matches or exceeds debate methods on multimodal sensing benchmarks while achieving 12.7 times reduction in token cost.

Copied to clipboard

Xinzhu Chen, Xuesheng Li, Zhongxiang Sun, Weijie Yu

GraphMind: LLMs as Dynamic Knowledge Builders for Sequential Decision-Making (2026.findings-acl)

Challenge:	Recent work studies RLVR through token entropy, arguing that high-entropies drive exploration and should receive stronger updates.
Approach:	They propose a correctness-aware reinforcement framework that performs fine-grained advantage modulation over low-entropy segments.
Outcome:	The proposed framework improves accuracy over strong RL baselines across three backbones and six math benchmarks while maintaining high-entropy exploration.

Copied to clipboard

Sunguk Shin, Hayeong Lee, Jun Ho Seo, Jinho Lee, Myunsoo Kim, Minsuk Chang, Byung-Jun Lee

EET: Experience-Driven Early Termination for Cost-Efficient Software Engineering Agents (2026.findings-acl)

Challenge:	Large language models (LLMs) have demonstrated remarkable performance in natural language understanding and generation, establishing themselves as foundational tools across a wide range of domains.
Approach:	They propose an LLM agent architecture that integrates a knowledge graph as a graph-based memory module and integrates it into the agent to generate efficient plans.
Outcome:	The proposed architecture improves the performance and efficiency of the LLM in navigation tasks designed to present long-horizon and partially observable challenges.

Copied to clipboard

Yaoqi Guo, Ying Xiao, Jie M. Zhang, Mark Harman, Yiling Lou, Yang Liu, Zhenpeng Chen

Fundamental Reasoning Paradigms Induce Out-of-Domain Generalization in Language Models (2026.findings-acl)

Challenge:	Large language models are reshaping modern software development, but they often incur substantial monetary cost.
Approach:	They propose an experience-driven early termination approach that extracts structured experience from prior issue-resolution executions and leverages it to guide early termination during patch generation and selection.
Outcome:	The proposed approach reduces cost by 19%–55% with negligible loss in resolution rate (at most 0.2%) EET extracts structured experience from prior issue-resolution executions and leverages it to guide early termination during patch generation and selection.

Copied to clipboard

Mingzi Cao, Xingwei Tan, Mahmud Elahi Akhter, Marco Valentino, Maria Liakata, Xi Wang, Nikolaos Aletras

Rethinking Text-to-SQL: Dynamic Multi-turn SQL Interaction for Real-world Database Exploration (2026.findings-acl)

Challenge:	Deduction, induction, and abduction are fundamental reasoning paradigms, core for human logical thinking.
Approach:	They propose to use a dataset of symbolic tasks to induce deductive skills into large language models (LLMs) they then use FT to fine-tune models to improve OOD generalization .
Outcome:	The proposed approach yields strong generalizability with substantial performance gains (up to 14.60) across realistic out-of-domain tasks.

Copied to clipboard

Linzhuang Sun, Tianyu Guo, Hao Liang, Ruitong Liu, Yuying Li, Qifeng Cai, Jingxuan Wei, Yuchen Wu, Bihui Yu, Xiangxiang Zhang, Wentao Zhang, Bin Cui

LogitSpec: Accelerating Retrieval-based Speculative Decoding via Next Next Token Speculation (2026.findings-acl)

Challenge:	Structured Query Language (SQL) is the cornerstone for data-driven decision-making.
Approach:	They propose a benchmark to rigorously evaluate Large Language Models within a dynamic interaction framework.
Outcome:	The proposed benchmark aims to rigorously evaluate LLMs within a dynamic interaction framework.

Copied to clipboard

Tianyu Liu, Qitan Lv, Hao Li, Xing Gao, Xiao Sun, Xiaoyan Sun

Answering Cross-Dimensional Geometric Visual Questions by Multi-constraint Spatial Reasoning (2026.findings-acl)

Challenge:	Speculative decoding (SD) is a promising technique for LLM inference acceleration.
Approach:	They propose a method to generate draft tokens in a retrieval-based manner to reduce drafting overhead and improve inference speed.
Outcome:	Extensive tests show that LogitSpec can achieve 2.61 speedup and 3.28 mean accepted tokens per decoding step.

Copied to clipboard

Dongling Li, Qi Chen, Jianxing Yu, Hanjiang Lai, Yanghui Rao, Wenqing Chen, Jian Yin

SudokuFill: A Multi-Agent Progressive Filling Framework for Document-Level Scientific Information Extraction (2026.findings-acl)

Challenge:	Existing methods for solving complex visual questions are limited in their ability to represent in a cross-dimensional space.
Approach:	They propose a method that can answer complex visual questions using cross-dimensional reasoning.
Outcome:	The proposed method can answer complex visual questions in 2D to 3D space with great application value.

Copied to clipboard

Yang Li, Yajiao Wang, Yu Zhang, Yuanzhe Zhang, Maodi Hu, Mengting Zhang, Xi Sun, Hua Yue, Zhixiong Zhang

Emergent Relational Order in LLM Agent Societies: From Collective Affect to Authority Stratification (2026.findings-acl)

Challenge:	Scientific information extraction (SciIE) is a key bottleneck for turning unstructured papers into computable knowledge bases.
Approach:	They propose a scientific information extraction framework that solves a Sudoku problem as a progressive filling problem.
Outcome:	The proposed framework outperforms the GPT-4o model on a document-level adjuvant dataset.

Copied to clipboard

Zhiyuan Ji, Xinyu Chen, Ziqi Dai, Shiyun Tang, Chunyu Wei, Yueguo Chen

VCB Bench: An Evaluation Benchmark for Audio-Grounded Large Language Model Conversational Agents (2026.findings-acl)

Challenge:	Fei Xiaotong’s Differential Order Pattern characterizes rural society as egocentric and relationally graded, with cooperation attenuating over social distance.
Approach:	They propose a multi-agent framework grounded in Affect Control Theory, Social Identity Theory, and Durkheimian collective affect.
Outcome:	Extensive simulations support interpreting Differential Order as a structure-sensitive emergent outcome of general social mechanisms.

Copied to clipboard

Jiliang Hu, Wenfu Wang, Zuchao Li, Chenxing Li, Yiyang Zhao, Hanzhao Li, Liqiang Zhang, Meng Yu, Dong Yu

LLM-FK: Multi-Agent LLM Reasoning for Foreign Key Detection in Large-Scale Complex Databases (2026.findings-acl)

Challenge:	despite advances in multimodal conversational systems, current benchmarks lack comprehensive evaluation across key dimensions.
Approach:	They propose a Chinese benchmark built exclusively on real human speech to fill this gap . they assess LALMs across three complementary axes: instruction following, knowledge understanding, robustness .
Outcome:	VCB Bench assesses LALMs across three complementary axes: instruction following, knowledge understanding, and robustness . VCBM Bench provides reproducible and fine-grained framework for Chinese voice chat bots . results show significant performance disparities and offer tangible insights for future improvements .

Copied to clipboard

Zijian Tang, Ying Zhang, Sibo Cai, Ruoxuan Wang

When Meaning Travels: A Granular Lens on Hybrid-MoE’s Role in Idiomatic Understanding for Language Models (2026.findings-acl)

Challenge:	Existing methods for detecting missing foreign keys are limited in capturing semantic dependencies across schemas.
Approach:	They propose a framework that integrates four agents to detect missing foreign keys . they propose combinatorial search space explosion, ambiguous inference and global inconsistency .
Outcome:	The proposed framework achieves F1-scores above 93% on large-scale MusicBrainz database . it reduces candidate search space by two to three orders of magnitude without losing true FKs .

Copied to clipboard

Sarmistha Das, Vaibhav Vishal, Shreyas Guha, Amaan Ali, Kitsuchart Pasupa, Sriparna Saha

IntentCoding: Amplifying User Intent in Code Generation (2026.findings-acl)

Challenge:	idioms provide a fascinating gateway to creativity, cultural values, historical context, and diverse perspectives inherent to diverse linguistic traditions.
Approach:	They propose a multimodal idiom corpus enriched with seven idiomatic tones . they propose idiomic hybridization framework that embeds multiple idiomatic expert opinions .
Outcome:	The proposed framework achieves 5–6% performance gains across advanced vision language models.

Copied to clipboard

Zheng Fang, Yihong Dong, Lili Mou, Dongming Jin, Zhi Jin, Ge Li

UbuntuGuard: A Culturally-Grounded Policy Benchmark for Equitable AI Safety in African Languages. (2026.findings-acl)

Challenge:	IntentCoding captures the influence of user intent by masking out the intent, and integrates seamlessly with existing decoding procedures.
Approach:	They propose a decoding strategy that captures the influence of user intent by masking out the intent and applies a multi-strength ensemble mechanism to amplify the effect of user intention during generation.
Outcome:	The proposed model significantly improves both constraint satisfaction and functional correctness compared to greedy decoding approaches.

Copied to clipboard

Tassallah Abdullahi, Macton Mgonzo, Mardiyyah Oduwole, Paul Okewunmi, Abraham Toluwase Owodunni, Ritambhara Singh, Carsten Eickhoff

microCLIP: Unsupervised CLIP Adaptation via Coarse-Fine Token Fusion for Fine-Grained Image Classification (2026.findings-acl)

Challenge:	Current guardian models are predominantly Western-centric and optimized for high-resource languages . low-resourced African languages are vulnerable to evolving harms, cross-lingual failures, cultural misalignment .
Approach:	They propose a policy-based safety benchmark for African languages built from adversarial queries authored by 155 domain experts across sensitive fields.
Outcome:	The proposed model overestimates multilingual safety, cross-lingual transfer provides partial but insufficient coverage, and dynamic models struggle to localize African-language contexts.

Copied to clipboard

Sathira Silva, Eman Ali, Chetan Arora, Muhammad Haris Khan

Prompting Across Time: Evaluating LLMs on Historical and Contemporary Offensive Language (2026.findings-acl)

Challenge:	Existing UA methods for fine-grained image classification rely on coarse-grain visual tokens, which misses fine spatial details.
Approach:	They propose a label-free self-training framework that adapts visual features and LLMderived text prototypes using fine-grained cues.
Outcome:	The proposed framework improves alignment between finegrained visual regions and rich textual descriptions while updating only layer norms and a tiny head.

Copied to clipboard

Sanne Hoeken, Sophie Jasmin Spliethoff, Silke Schwandt, Özge Alacam, Sina Zarrieß

LDEDE: LRP-Driven Efficient Detection and Editing Framework for LLM Privacy Neurons (2026.findings-acl)

Challenge:	Existing research on large language models and hate speech detection has focused on contemporary data.
Approach:	They propose to use a modular prompt design to evaluate early-modern English invectives . they propose to widen the scope of NLP research on hate speech beyond the contemporary domain .
Outcome:	The proposed model outperforms a modern hate-speech benchmark on Early Modern English invectives . the results show that the model is more robust to contextual and contextual factors than the current model .

Copied to clipboard

Zhao Zhengyuan, Cao Lifeng, null Sunhaodong, Shi Haotian, Du Xuehui, Liu Aodi, Niu Lanjie, Yang Xiaocheng

Chimera: Compositional Jailbreak Attacks on LLMs via Judgment-Driven Search over Heterogeneous Strategies (2026.findings-acl)

Challenge:	Existing privacy protection methods fail to cover context-dependent sensitive information and are prone to performance degradation.
Approach:	They propose a Layer-wise Relevance Propagation-driven framework for efficient privacy neuron detection and editing.
Outcome:	The proposed framework achieves 80% higher efficiency than gradient attribution methods while reducing leakage risks of Phone, Email, and medical privacy by 42.7%–73.5% on average and cutting computational time by 60%–90%.

Copied to clipboard

Leo Hyun Park, Juwon Cho, Gyuhwan Kim, YoonDong Yeo, Taekyoung Kwon

Bias Dynamics in BabyLMs: Towards a Compute-Efficient Sandbox for Democratising Pre-Training Debiasing (2026.findings-acl)

Challenge:	Existing methods for evaluating large language models face two limitations: they explore homogeneous transformations in isolation and rely on brittle judgment metrics that misclassify non-refusal hallucinations as successful attacks.
Approach:	They propose a framework that generates compositional jailbreak attacks via judgment-driven search over heterogeneous strategies.
Outcome:	The proposed framework generates compositional jailbreak attacks over heterogeneous strategies . strongREJECT++ improves attack success rates and transferability compared to state-of-the-art .

Copied to clipboard

Filip Trhlík, Andrew Caines, Paula Buttery

PuzzleClone: A DSL-Powered Framework for Synthesizing Verifiable Data (2026.findings-acl)

Challenge:	Pre-trained language models (LMs) have grown substantially in both societal adoption and training costs.
Approach:	They propose to use low-cost proxy models to democratise pre-model debiasing research by using small and mutable corpora.
Outcome:	The proposed model can approximate bias acquisition and learning dynamics of larger models despite their reduced size.

Copied to clipboard

Kai Xiong, Yanwei Huang, Rongjunchen Zhang, Kun Chen, Haipang WU, Yingcai Wu

Automated Knowledge Component Generation and Interpretable Knowledge Tracing in Coding Problems (2026.findings-acl)

Challenge:	Existing datasets with verifiable answers are limited in reliability, diversity, and scalability . a new approach to generate verifikatable data at scale is needed to improve models' performance .
Approach:	They propose a formal framework for synthesizing verifiable data at scale using a novel DSL-driven approach.
Outcome:	The proposed framework improves performance on a wide range of puzzles and logic benchmarks.

Copied to clipboard

Zhangqi Duan, Nigel Fernandez, Arun Balajiee Lekshmi Narayanan, Mohammad Hassany, Rafaella Sampaio de Alencar, Peter Brusilovsky, Bita Akram, Andrew Lan

Evaluation Pitfalls and Sparsity Limitations in LLM-based Confidence Estimates for Classification (2026.findings-acl)

Challenge:	Existing solutions to automate KC generation and tagging for open-ended programming problems are highly labor-intensive and prone to bias and errors.
Approach:	They propose an automated pipeline for KC generation and tagging for open-ended programming problems using large language models.
Outcome:	The proposed method outperforms existing ones and outperfies human-written KCs on future student response prediction.

Copied to clipboard

Elena Merdjanovska, Omar Zaidan, Andreas Rücklé

MATA: Multi-Agent Framework for Reliable and Flexible Table Question Answering (2026.findings-acl)

Challenge:	Xuan et al., 2023) show that verbalization produces extremely sparse outputs for confidence estimation.
Approach:	They propose to standardize stepwise interpolation for a fairer comparison . they advocate standardizing stepwise intercepts for AUARC evaluation .
Outcome:	The proposed method achieves the best AUARC score (+2.3 points over vanilla verbalization) while requiring less inference cost.

Copied to clipboard

Sieun Hyeon, Jusang Oh, Sunghwan Steve Cho, Jaeyoung Do

Hallucination Detection in Long-Form Text Generated by LLMs: A Benchmark and a Hyper-Relational Knowledge Graph Approach (2026.findings-acl)

Challenge:	Recent advances in Large Language Models have significantly improved table understanding tasks . practical deployment of TableQA systems presents several persistent challenges .
Approach:	They propose a multi-agent TableQA framework that leverages multiple reasoning paths and tools built with small language models.
Outcome:	The proposed framework achieves state-of-the-art accuracy and efficient reasoning while avoiding excessive LLM inference.

Copied to clipboard

Zituo Li, Guangzhou Chen, Jianbin Sun, Qi Song, Yang Kewei

Can Intelligent Agents Revolutionize Scale Generation? (2026.findings-acl)

Challenge:	Existing methods for hallucination detection are coarse-grained and lack long-range consistency checks.
Approach:	They propose a benchmark for long-form hallucination detection that incorporates diverse entity types and intricate factual dependencies spanning extended contexts.
Outcome:	The proposed framework outperforms baselines and robustly integrates fact-centric hyper-relational knowledge graphs.

Copied to clipboard

Chenghao Jia, Zhitao Yuan, Zhaokang Zong, YiFei Yin, Zhe Chen, Man Lan, Shengjun Wu

Challenge:	Existing measurement scales require extensive manual labor and require extensive validation and validation.
Approach:	They propose a multi-agent framework that automates scale development by leveraging collaborative AI agents.
Outcome:	The proposed framework automates scale development while maintaining rigorous quality standards.

On Finding Inconsistencies in Documents (2026.findings-acl)

Copied to clipboard

Charles Lovering, Seth Ebner, Brandon Smock, Michael Krumdick, Muhammad Saad Rabbani, Ahmed Muhammad, Varshini Reddy, Chris Tanner

Towards Preference Following in Tool Calling Language Agents (2026.findings-acl)

Challenge:	Language models can be used to quickly and easily detect inconsistencies in documents .
Approach:	They propose a benchmark to measure language models' ability to detect inconsistencies in documents . they use a document with an inconsistent inserted manually by a domain expert .
Outcome:	The best-performing model recovered 64% of the inserted inconsistencies on 50 arXiv papers and found that the original authors had already found inconsistent inconsistances.

Copied to clipboard

Zhi-Yuan Chen, Siyu Lu, Qianlong Xie, Xingxing Wang, Yankai Lin

ReSQL: Self-Improving Framework for Reasoning-Aware Text-to-SQL Dataset Generation (2026.findings-acl)

Challenge:	Currently, large language model (LLM)-based agents can't follow user preferences when calling tools.
Approach:	They propose a benchmark to evaluate agents' ability to identify personalized user preferences from interaction histories and to adhere to these preferences when calling tools.
Outcome:	The proposed model achieves 51.16% accuracy on the APOLLO benchmark, while GPT-4o achieves only 51.13% accuracy.

Copied to clipboard

Minjun Park, Yongju Seong, Myoseop Sim, Kyungkoo Min, Stanley Jungkyu Choi

Identifying and typifying demographic unfairness in phoneme-level embeddings of self-supervised speech recognition models (2026.findings-acl)

Challenge:	Experimental results show that ReSQL significantly improves execution accuracy and self-correction ability over strong baselines.
Approach:	They propose a framework that generates and learns from its own error-reasoning dataset . it allows models to internalize robust error-reference patterns and apply them to unseen queries .
Outcome:	The proposed framework improves execution accuracy and self-correction ability over strong baselines.

Copied to clipboard

Felix Herron, Solange Rossato, Alexandre Allauzen, François Portet

MolSafeEval: A Benchmark for Uncovering Safety Risks in AI-Generated Molecules (2026.findings-acl)

Challenge:	a number of studies have been done to improve ASR for speaker groups, but there is still room for improvement . authors propose a framework typifying two types of error in phoneme embeddings .
Approach:	They propose a framework typifying two types of error that can occur in phoneme modeling . they propose random error/high variance in phonemes embedding vs systematic error/embedding bias .
Outcome:	The proposed framework typifies errors in phoneme modeling in ASR systems . it shows that training only on a single, typically disadvantaged SG improves performance .

Copied to clipboard

Tong Xu, Xinzhe Cao, Zhihui Zhu, Keyan Ding, Huajun Chen

Dynamic Tool Dependency Retrieval for Lightweight Function Calling (2026.findings-acl)

Challenge:	generative models may produce molecules with toxic, reactive, or otherwise hazardous characteristics.
Approach:	They propose a benchmark to evaluate and analyze the safety risks of molecular generation.
Outcome:	The proposed benchmark aims to evaluate and analyze the safety risks of molecular generation.

Copied to clipboard

Bhrij Patel, Davide Belli, Amir Jalalirad, Maximilian Arnold, Aleksandr Ermolov, Bence Major

Audit Me If You Can: Query-Efficient Active Fairness Auditing of Black-Box LLMs (2026.findings-acl)

Challenge:	Existing retrieval methods rely on static inputs, failing to capture multi-step tool dependencies and evolving task context.
Approach:	They propose a lightweight retrieval method that conditions on initial query and evolving task context.
Outcome:	The proposed method improves function calling success rates between 23% and 104% compared to state-of-the-art retrieval methods.

Copied to clipboard

David Hartmann, Lena Pohlmann, Lelia Hanslik, Noah Gießing, Bettina Berendt, Pieter Delobelle

Mending the Holes: Mitigating Reward Hacking in Reinforcement Learning for Multilingual Translation (2026.findings-acl)

Challenge:	Large Language Models exhibit systematic biases across demographic groups.
Approach:	They propose to use auditing as uncertainty estimation over a fairness metric . they propose to introduce the Bounded Active Fairness Auditor for query-efficient auditing .
Outcome:	The proposed auditing tool reduces query access costs and improves performance over time.

Copied to clipboard

Yifeng Liu, Siqi Ouyang, Yatish H R, Lei Li

MedCoach: Enhancing Medical Reasoning in LLMs via Knowledge Graph-Augmented Chain-of-Thought Distillation (2026.findings-acl)

Challenge:	Existing methods for training large language models rely heavily on high-quality parallel data, which are often scarce or unavailable for low-resource languages.
Approach:	They propose a reinforcement training method using only monolingual text to elevate LLMs’ translation capabilities on massive low-resource languages while retaining their performance on high-resourced languages.
Outcome:	The proposed model outperforms LLaMAX, one of the strongest open-source multilingual LLMs on 1,414 language directions on Flores-101 dataset.

Copied to clipboard

Chuan Li, Ye Lyu, Chengyu Wang, Mingyuan Fan, Cen Chen

Pru-CoT: Towards Efficient Reasoning Distillation via Pruning Chain-of-Thought (2026.findings-acl)

Challenge:	Existing methods for training specialized reasoning models for the medical domain are limited due to the scarcity of high-quality, large-scale Chain-of-Thought (CoT) data.
Approach:	They propose a framework that introduces a dedicated coach role to guide the student model through question decomposition.
Outcome:	The proposed framework smooths the learning curve in medical reasoning by facilitating domain adaptation before advancing to complex long-chain reasoning.

Copied to clipboard

Han Liu, Shuotian Ma, Hui Li, Xiaotong Zhang, Fenglong Ma, Hong Yu

Curriculum-RLAIF: Curriculum Alignment with Reinforcement Learning from AI Feedback (2026.findings-acl)

Challenge:	Existing heuristics fail to capture global causal logic due to rigid rules and limited search spaces.
Approach:	They propose a framework that extracts the essential logical structure from reasoning chains.
Outcome:	Experiments show that Pru-CoT models generate more compact reasoning paths compared to models trained on verbose data.

Copied to clipboard

Jiaye Lin, Mengdi Li, Xufeng Zhao, Wenhao Lu, Peilin Zhao, Stefan Wermter, Di Wang

Challenge:	Existing approaches to align large language models with human preferences are limited in generalizability due to distribution shift, preference label noise, and mismatch of challenging samples with model capacity.
Approach:	They propose a framework that constructs preference pairs with varying difficulty levels and then produces a specific curriculum for reward model training.
Outcome:	The proposed framework improves generalizability of reward models by a significant margin without incurring additional inference costs compared to existing non-curriculum baselines.

Query-Focused Individual Simulation with Progressive Persona Completion (2026.findings-acl)

Copied to clipboard

Weiwen SU, Naoki Yoshinaga, Masashi Toyoda

VIDA: A Visual Intent-driven Design Assistant for Proactive Multimodal Clarification (2026.findings-acl)

Challenge:	Existing approaches to simulating individual responses from persona information assume rich persona profiles, which are often unavailable in practice.
Approach:	They propose a query-focused individual simulation where relevant persona information is identified and requested on demand for each query.
Outcome:	Experiments on two dialogue datasets show that the proposed method achieves comparable performance to approaches that rely on rich persona information extracted from dialogue history.

Copied to clipboard

Yanshan Liu, Hongbo Zhang, Zhen Sun, Jiaheng Wei, Kaishun Wu

DeepSynth-Eval: Objectively Evaluating Information Consolidation in Deep Survey Writing (2026.findings-acl)

Challenge:	Existing vision-language models fail to provide accurate and complete answers to user requests . a new strategy-aware design assistant is developed to help designers create proactive, visually grounded, and strategically prioritized clarification questions.
Approach:	They propose a visual intent-driven design assistant to generate proactive, visually grounded, and strategically prioritized clarification questions.
Outcome:	The proposed assistant improves the strategic alignment score by 20.59% over baselines and restores visual grounding capabilities lost during fine-tuning.

Copied to clipboard

Hongzhi Zhang, Yuanze Hu, Tinghai Zhang, Jia Fu, Tao Wang, Junwei Jing, Zhaoxin Fan, Wei Bi, Ruiming Tang, Han Li, Guorui Zhou, Kun Gai

A Scalable Entity-Based Framework for Auditing Bias in Large Language Models (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are evolving towards autonomous agents . retrieval capabilities are well-benchmarked, but post-retrieval synthesis is under-evaluated due to open-ended writing.
Approach:	They propose a benchmark to evaluate information consolidation capabilities using survey papers as gold standards.
Outcome:	The proposed benchmark analyzes the post-retrieval synthesis stage of large language models . it leverages high-quality survey papers as gold standards and reverse-engineers research requests . the proposed benchmark outperforms single-turn generation and reduces hallucinations .

Copied to clipboard

Akram Elbouanani, Aboubacar Tuo, Adrian Popescu

HiGMem: A Hierarchical and LLM-Guided Memory System for Long-Term Conversational Agents (2026.findings-acl)

Challenge:	Existing approaches to bias evaluation in large language models trade ecological validity for statistical control, or use artificial prompts that lack scale and rigor.
Approach:	They propose a framework that uses named entities as probes to measure bias in large language models.
Outcome:	The proposed framework reproduces bias patterns observed in natural text, enabling large-scale analysis.

Copied to clipboard

Shuqi Cao, Jingyi He, Fei Tan

Self-Guided Plan Extraction for Instruction-Following Tasks with Goal-Conditional Reinforcement Learning (2026.findings-acl)

Challenge:	Existing memory systems rely on vector similarity for retrieval, resulting in bloated evidence sets . existing systems produce little additional recall, but this approach lowers retrieval precision .
Approach:	They propose a two-level event-turn memory system that uses event summaries as semantic anchors to predict which related turns are worth reading.
Outcome:	The proposed system achieves the best F1 on four of five question categories and improves adversarial F1 from 0.54 to 0.78 over A-Mem while retrieving an order of magnitude fewer turns.

Copied to clipboard

Zoya Volovikova, Nikita Sorokin, Dmitriy Lukashevskiy, Aleksandr Panov, Alexey Skrynnik

Adaptive Prompt Optimization for Open-Ended Tasks: Uncertainty Preference as a Secondary Signal (2026.findings-acl)

Challenge:	a framework for instruction-following tasks is proposed for instruction following tasks . previous methods rely on expert trajectories and learn directly from the agent's own interactions with the environment without expert supervision.
Approach:	They propose a framework for instruction-following tasks that enables a language model to generate and refine high-level plans through a self-learning mechanism.
Outcome:	The proposed framework adheres to instructions more strictly than baseline methods while showing strong generalization to previously unseen instructions.

Copied to clipboard

Shuyang Zhang, Zhixuan Liu, Zhichen Dong, Hao Zhang, Chaochao Lu, Chao Yang

DPLoRA: A Dual-Pruning Framework based on ILP Optimization and Progressive Pruning for Parameter-Efficient LoRA Fine-Tuning (2026.findings-acl)

Challenge:	Recent training-free prompt optimizers treat performance as maximizing a single scalar score and ignore a second signal that the desired style is task dependent.
Approach:	They propose a semantic-entropy-based method that uses task uncertainty to guide prompt optimization by selecting high-entropicy candidates for creative tasks and low-energetic candidates for conservative ones.
Outcome:	The proposed method outperforms baselines on MT-Bench subsets and integrates easily into existing prompt optimizers.

Copied to clipboard

Changjun Park, Sejong Yoon, Jaekwang Kim

Demystifying Multi-Agent Debate: The Role of Confidence and Diversity (2026.findings-acl)

Challenge:	Large language models (LLMs) require computational resources for fine-tuning.
Approach:	They propose a framework that optimizes rank allocation via two stages . they propose an initial pruning stage and a progressive pruning stage .
Outcome:	The proposed framework outperforms existing PEFT baselines on GLUE and instruction-following tasks while reducing training time and trainable parameters by over 80%.

Copied to clipboard

Xiaochen Zhu, Caiqi Zhang, Yizhou Chi, Tom Stafford, Nigel Collier, Andreas Vlachos

Challenge:	Multi-agent debate (MAD) is widely used to improve large language models' (LLMs) reasoning and test-time scaling.
Approach:	They propose a diversity-aware initialisation that selects a more diverse pool of candidate answers, increasing the likelihood that a correct hypothesis is present at the start of debate.
Outcome:	The proposed protocol outperforms vanilla MAD and majority vote on six reasoning-oriented QA benchmarks.

RoleConflictBench: A Benchmark of Role Conflict Scenarios for Evaluating LLMs’ Contextual Sensitivity (2026.findings-acl)

Copied to clipboard

Jisu Shin, Hoyun Song, Juhyun Oh, Changgeon Ko, Eunsu Kim, Chani Jung, Alice Oh

Challenge:	a new benchmark measures the contextual sensitivity of large language models in role conflict scenarios . role conflicts are social dilemmas where multiple roles cannot be fulfilled simultaneously . authors: models are forced to arbitrate between dynamic contextual cues and learned preferences .
Approach:	They propose a benchmark to measure the contextual sensitivity of large language models in role conflict scenarios.
Outcome:	The proposed benchmark measures the contextual sensitivity of large language models in role conflict scenarios.

RbtAct: Rebuttal as Supervision for Actionable Review Feedback Generation (2026.findings-acl)

Copied to clipboard

Sihong Wu, Yiling Ma, Yilun Zhao, Tiansheng Hu, Owen Jiang, Manasi Patwardhan, Arman Cohan

Challenge:	Prior studies show that large language models can draft fluent reviews but they miss specific issues, show shallow analysis, and produce generic phrasing.
Approach:	They propose a task that targets actionable review feedback generation and places existing peer review rebuttal at the center of learning.
Outcome:	The proposed model improves on a large dataset that maps review segments to rebuttal segments that address them, with perspective labels and impact categories that order author uptake.

Is Grokking Worthwhile? Functional Analysis and Transferability of Generalization Circuits in Transformers (2026.findings-acl)

Copied to clipboard

Kaiyu He, Mian Zhang, Peilin Wu, Xinya Du, Zhiyu Chen

When 20 Agents Fail to Sort: The Distributed Sorting Benchmark for Scalable Multi-Agent Systems (2026.findings-acl)

Challenge:	Large Language Models struggle with the "curse of two-hop reasoning" in compositional tasks.
Approach:	They propose to form a "Generalization Circuit" during a prolonged "grokking" phase . they argue that grokkking is the process of integrating memorized atomic facts into an easy-acquire reasoning path.
Outcome:	The proposed model is superior to non-grokked models, but it requires a large computational cost . the study shows that grokking is not the sudden acquisition of a new reasoning paradigm .

Copied to clipboard

Xin Yang, Junhao Wang, Bintao Tang, Xuxin Cheng, Cao Liu, Ke Zeng, Wenyuan Jiang

Challenge:	MAS-BENCH isolates coordination under explicit communication constraints . CAMOC significantly improves coordination success and efficiency across backends .
Approach:	They propose a distributed-sorting benchmark that isolates coordination under explicit communication constraints.
Outcome:	MAS-BENCH improves coordination success and efficiency across backends . CAMOC significantly improves efficiency under shared-state interaction .

RIGOURATE: Quantifying Scientific Exaggeration with Evidence-Aligned Claim Evaluation (2026.findings-acl)

Copied to clipboard

Joseph James, Chenghao Xiao, Yucheng Li, Nafise Sadat Moosavi, Chenghua Lin

Glance-or-Gaze: Incentivizing LMMs to Adaptively Focus Search via Reinforcement Learning (2026.findings-acl)

Challenge:	Scientific rigour tends to be sidelined in favour of bold statements, leading authors to overstate claims beyond what their results support.
Approach:	They propose a multimodal framework that retrieves supporting evidence from a paper and assigns each claim an overstatement score.
Outcome:	The proposed framework retrieves supporting evidence from ICLR and NeurIPS papers and assigns each claim an overstatement score.

Copied to clipboard

Hongbo Bai, Yujin Zhou, Yile Wu, Chi-Min Chan, Pengcheng Wen, Kunhao Pan, Sirui Han, Yike Guo

Before Forgetting, Learn to Remember: Revisiting Foundational Learning Failures in LVLM Unlearning Benchmarks (2026.findings-acl)

Challenge:	Existing search-augmented approaches rely on indiscriminate whole-image retrieval and lack deep iterative reflection, limiting their effectiveness on complex visual queries.
Approach:	They propose a fully autonomous framework that shifts from passive perception to active visual planning and introduces a Selective Gaze mechanism that dynamically chooses whether to glance at global context or gaze into high-value regions.
Outcome:	Experiments across six benchmarks demonstrate state-of-the-art performance.

Copied to clipboard

Junehyoung Kwon, MiHyeon Kim, Eunju Lee, JungMin Yun, Byeonggeuk Lim, YoungBin Kim

Challenge:	Large Vision-Language Models (LVLMs) are capable of learning from vast webscale datasets but pose privacy risks as they can unintentionally memorize sensitive information.
Approach:	They propose a Reliable Multi-hop and Multi-image Memorization Benchmark that ensures robust foundational learning through principled data scaling and reasoning-aware QA pairs.
Outcome:	Extensive experiments show that ReMem provides a reliable framework for diagnosing both learning and unlearning behaviors in Large Vision-Language Models.

Investigating Links between Illicit Massage Businesses through Natural Language Processing and Graph Machine Learning (2026.findings-acl)

Copied to clipboard

Vasuki Garg, Osman Ozaltin, Maria Mayorga, Sherrie Caltagirone

TopoRAG: Graph-based RAG via Topology-aware Approximate Nearest Neighbor Search (2026.findings-acl)

Challenge:	Illicit massage businesses exploit vulnerable individuals through forced sex or labor . identifying key indicators from vast volume of data associated with these businesses poses significant challenge .
Approach:	They propose a multi-stream data integration approach focusing on Yelp reviews . they propose bespoke subgraph extraction strategies to detect links between massage businesses .
Outcome:	The proposed approach outperforms baseline methods in a multi-stream data integration framework based on consumer reviews on Yelp.com and contextual data from the U.S. Census and business license records.

Copied to clipboard

Tianhao Wu, Siqiang Luo

Self-Explaining Hate Speech Detection with Moral Rationales (2026.findings-acl)

Challenge:	Recent studies extend RAG with graph-structured knowledge, enhancing retrieval to capture relational context beyond isolated text chunks.
Approach:	They propose a retrieval framework that integrates structural constraints into ANN search . they propose heuristic neighbor expansion which augments the retrieved set by traversing immediate neighbors .
Outcome:	The proposed framework improves precision and reduces context redundancy compared to existing methods.

Copied to clipboard

Francielle Vargas, Jackson Trager, Diego Alves, Matteo Guida, Surendrabikram Thapa, Berk Atıl, Daryna Dementieva, Andrew J Smart, Ameeta Agrawal

Principled Detection of Hallucinations in Large Language Models via Multiple Testing (2026.findings-acl)

Challenge:	Existing models for hate speech detection are opaque and rely on surface-level cues. Existing approaches often encode biases originating from training data and annotation processes.
Approach:	They propose a framework that integrates moral rationale supervision into training . they propose SMRA for self-explaining hate speech detection .
Outcome:	The proposed framework improves performance across binary hate speech detection and multi-label moral sentiment classification.

Copied to clipboard

Jiawei Li, Akshayaa Magesh, Venugopal Veeravalli

What Do Vision–Language Models Encode for Personalized Image Aesthetics Assessment? (2026.findings-acl)

Challenge:	Existing methods to detect hallucinations are prone to generating false alarms and false feedbacks.
Approach:	They propose a method that aggregates multiple evaluation scores via conformal p-values, enabling calibrated detection with controlled false alarm rate.
Outcome:	The proposed method aggregates multiple evaluation scores via conformal p-values, enabling calibrated detection with controlled false alarm rate.

Copied to clipboard

Koki Ryu, Hitomi Yanaka

A Diversity Diet for a Healthier Model: A Case Study of French ModernBERT (2026.findings-acl)

Challenge:	Personalized image aesthetics assessment (PIAA) is an important research problem with practical applications.
Approach:	They propose a vision-language model that encodes multi-level aesthetic attributes . they analyze visual representations of VLMs to examine their internal representations .
Outcome:	The proposed framework can be used to personalize images without fine-tuning . it can be implemented in a variety of image domains and architectures.

Copied to clipboard

Louis Estève, Christophe Servan, Thomas Lavergne, Agata Savary

Is Your Language Model Ready for Monetization Decisions? (2026.findings-acl)

Challenge:	Diversity has been gaining interest in the NLP community in recent years.
Approach:	They propose to use diversity-driven sampling to pre-train models on French with a fixed compute budget.
Outcome:	The diversity-driven sampling reduces the pre-training dataset by 94% and the pretraining time by 73% while maintaining comparable performance.

Copied to clipboard

Jialu Gao, Hua Li, Tracy Ortman, Yeyun Gong, Jian Jiao

Activation Reward Models for Few-Shot Model Alignment (2026.findings-acl)

Challenge:	Existing benchmarks focus on shopping-centric scenarios and user-facing data, overlooking intermediate decision stages and robustness considerations.
Approach:	They propose a multi-task benchmark to evaluate large language models in real-world monetization contexts.
Outcome:	The proposed benchmark covers intent understanding, commercial matching, and user behavior modeling.

Copied to clipboard

Tianning Chai, Chancharik Mitra, Brandon Huang, Gautam Rajendrakumar Gare, Zhiqiu Lin, Assaf Arbelle, Leonid Karlinsky, Rogerio Feris, Trevor Darrell, Deva Ramanan, Roei Herzig

Cross-lingual Matryoshka Representation Learning across Speech and Text (2026.findings-acl)

Challenge:	A common approach is to use reward models that enable reinforcement-learning post-training.
Approach:	They propose a method that steers LLM activations to align with few-shot preference data without finetuning.
Outcome:	The proposed method surpasses zero-shot, few-shot and voting-based benchmarks on reward hacking and noise signals.

Copied to clipboard

Yaya SY, Dioula Doucouré, Christophe Cerisara, Irina Illina

TRACE: Two-Phase RL for Causal Graph Exploration and Deeper Psychological Intervention in Dynamic Counseling Scenarios (2026.findings-acl)

Challenge:	Speakers of under-represented languages face language barriers and modality barriers . we train a bilingual speech-text embedding model for French-Wolof .
Approach:	They train a bilingual speech-text Matryoshka embedding model that enables efficient retrieval of French text from Wolof speech queries.
Outcome:	The proposed model can retrieve French text from Wolof speech queries without expensive ASR-translation pipelines.

Copied to clipboard

Shilin Tang, Zunyi Yin, Xuefeng Liang, Guanghui Shi, Song Tong, Chen Guangyu

LLM Multi-Agent Systems for Long Triple Set Data-to-Text Generation (2026.findings-acl)

Challenge:	Existing models lack the ability to actively explore the underlying causes of psychological distress.
Approach:	They propose a two-phase reinforcement learning framework that implements a causal-graph-driven reward scheme across two phases: an exploration phase that rewards the causal graph reconstruction following a surface-to-deep path, and an intervention phase that supports targeted restructuring of irrational beliefs.
Outcome:	Extensive experiments show that TRACE outperforms existing models, enabling causal-chain-aware psychological intervention beyond surface-level empathy.

Copied to clipboard

Chinonso Cynthia Osuji, Simon Mille, Mark Andrade, Jane Adkins, Ornait O’Connell, Elaine Uí Dhonnchadha, Bláithín Heffernan, Fírinne Nic an tSaoir, Anya Belz, Thiago Castro Ferreira, Brian Davis

Biomed-Enriched: Data-Efficient Biomedical Pretraining via Paragraph-Level Annotation (2026.findings-acl)

Challenge:	Existing data-to-text benchmarks that do not involve content selection feature short input-output pairs designed for sentence or paragraph-level generation with reference texts spanning only a few dozen tokens.
Approach:	They propose a system that generates multi-paragraph outputs in English and Irish . they compare a multi-agent configuration against a single-task variant .
Outcome:	The proposed framework generates multi-paragraph outputs in English and Irish . human evaluation and LLM-as-a-judge score better in both languages .

Copied to clipboard

Rian Touchent, Nathan Godey, Éric Villemonte de la Clergerie

Identifying the Achilles’ Heel: An Iterative Method for Uncovering Factual Errors in Large Language Models (2026.findings-acl)

Challenge:	Large language models have demonstrated remarkable capabilities across a wide range of general tasks, from question answering to code generation.
Approach:	They use a paragraph-level pipeline to annotate PubMed Central paragraphs . they use XLM-RoBERTa to fine-tune the pipeline and propagate annotations to the full corpus .
Outcome:	The proposed approach improves performance on 11 tasks while using 2.5x fewer tokens and only public data.

Copied to clipboard

Wenxuan Wang, Yuk-Kit Chan, Zixuan Ling, Shi Juluan, Youliang Yuan, Jen-tse Huang, Yifei Zhang, Wenxiang Jiao, Zhaopeng Tu, Michael R. Lyu

Probing the Plasticity and Correlation of LLM Value Systems: LLM Value Rankings are Not Stable (2026.findings-acl)

Challenge:	Current methods for evaluating LLMs’ veracity are limited by the need for extensive human labor, test data contamination, or limited scope, hindering efficient and effective exposure of errors.
Approach:	They propose a framework that extracts fact triplets to generate diverse question types using rule-based natural language processing techniques.
Outcome:	The proposed framework can trigger factual errors in up to 55% of questions in large LLMs while maintaining coverage of questions.

Copied to clipboard

Zhenheng Tang, Qihua Pan, Jingya Shen, Xiang Liu, Qian Wang, Bo Li, Xiaowen Chu

FAMA: Failure-Aware Meta-Agentic Framework for Open-Source LLMs in Interactive Tool Use Environments (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have similar value rankings but little is known about how susceptible they are to external influence and how different values are correlated with each other.
Approach:	They propose to use 6 different value transformation prompting methods to examine the plasticity of LLM value systems by comparing them with 8 LLMs.
Outcome:	The proposed methods are effective on 8 LLMs and 3 families.

Copied to clipboard

Amir Saeidi, Venkatesh Mishra, Souradeep Mukhopadhyay, Gaowen Liu, Ali Payani, Jayanth Srinivasa, Chitta Baral

Beyond Reasoning Gains: Mitigating General-Capability Forgetting in Large Reasoning Models (2026.findings-acl)

Challenge:	Large Language Models are being increasingly deployed as decision-making core of autonomous agents . however, in conversational benchmarks, these agents fail due to the cascading effects of incorrect decision- making .
Approach:	They propose a framework that analyzes failure trajectories from baseline agents to identify most prevalent errors.
Outcome:	Experiments show that the framework improves performance over open-source LLMs . the framework can be used to build reliable, multi-turn tool-use agents .

Copied to clipboard

Hoang Phan, Xianjun Yang, Yuanshun Yao, Jingyu Zhang, Shengjie Bi, Xiaocheng Tang, Madian Khabsa, Lijuan Liu, Deren Lei

ActorMind: Emulating Human Actor Reasoning for Speech Role-Playing (2026.findings-acl)

Challenge:	Reinforcement learning with verifiable rewards (RLVR) has delivered impressive gains in mathematical and multimodal reasoning . however, the recipe introduces a significant risk of capability regression, where models forget foundational skills after prolonged training without employing regularization strategies.
Approach:	They propose a replay strategy with dynamic objective reweighting for general knowledge preservation using short-horizon signals of convergence and instability.
Outcome:	The proposed method preserves general capabilities and improves reasoning . it can be applied to existing RLVR pipelines without training additional models or tuning .

Copied to clipboard

Xi Chen, Wei Xue, Yike Guo

Neuronal Insights into LLM Attacks: Targeted Neuron Tuning for Precise and Robust Vulnerability Patching (2026.findings-acl)

Challenge:	Existing work on role-playing focuses on textual modalities, neglecting speech . et al., 2025) show that speech role-players can generate spontaneous responses with personalized traits based on the context.
Approach:	They propose a framework that allows models to deliver spontaneous responses with personalized verbal traits based on their role, scene, and spoken dialogue.
Outcome:	The proposed framework enhances speech role-playing by generating spontaneous responses with personalized traits based on their role, scene, and spoken dialogue.

Copied to clipboard

Dan Shi, Renren Jin, Zhuowen Han, Yuqi Ren, Xinwei Wu, Zhigen Li, Deyi Xiong

Ro-SLM: Onboard Small Language Models for Robot Task Planning and Operation Code Generation (2026.findings-acl)

Challenge:	Existing gradient-based attribution methods are inapplicable to adversarial attacks . et al.: Targeted neuron tuning improves model robustness against jailbreak attacks despite the model's vulnerability to jailbreak.
Approach:	They propose a gradient-based method to identify key neurons sensitive to adversarial behaviors in open-ended generation tasks.
Outcome:	The proposed method detects key neurons sensitive to adversarial behaviors in open-ended tasks.

Copied to clipboard

Wenhao Wang, Yanyan Li, Long Jiao, Jiawei Yuan

Just Use XML: Revisiting Joint Translation and Label Projection (2026.findings-acl)

Challenge:	Recent advances in large language models (LLMs) provide robots with contextual reasoning abilities to comprehend human instructions.
Approach:	They propose a framework that enables reliable SLM-driven robot operation by distilling LLMs’ knowledge and reasoning.
Outcome:	The proposed framework enables reliable SLM-driven robot operation by distilling LLMs’ knowledge and reasoning.

Copied to clipboard

Thennal D K, Chris Biemann, Hans Ole Hatzel

CaTS-Bench: Can Language Models Describe Time Series? (2026.findings-acl)

Challenge:	Label projection is an effective technique for cross-lingual transfer, extending span-annotated datasets from high-resource languages to low-resourced ones.
Approach:	They propose a framework that performs translation and label projection via XML tags.
Outcome:	The proposed framework outperforms baselines and improves translation quality across languages and annotation complexity.

Copied to clipboard

Luca Zhou, Pratham Yashwante, Marshall Fisher, Alessio Sampieri, Zihao Zhou, Fabio Galasso, Rose Yu

How do Visual Attributes Influence Web Agents? A Comprehensive Evaluation of User Interface Design Factors (2026.findings-acl)

Challenge:	Existing time series captioning benchmarks rely on fully synthetic or generic captions . authors propose a pipeline for generating high-fidelity synthetic captions, which is validated .
Approach:	They propose a benchmark for Context-aware Time Series reasoning across 11 diverse domains . they evaluate leading Vision-Language Models on their benchmark .
Outcome:	The proposed benchmark evaluates 1746 human-rewritten captions and shows they perform better than open-source models.

Copied to clipboard

Kuai Yu, Naicheng Yu, Han Wang, Rui Yang, Huan Zhang

Scaling Unverifiable Rewards: A Case Study on Visual Insights (2026.findings-acl)

Challenge:	Existing studies on the effect of environmental variation on web agents have focused on robustness to adversarial attacks with less attention to agents’ preferences in benign scenarios.
Approach:	They propose a controlled evaluation pipeline to quantify how visual attributes influence web-agent decision-making by comparing variants and browsing interactions.
Outcome:	Extensive experiments on 8 variant families, 5 real-world websites and 4 representative web agents show that background color contrast, item size, position, and card clarity have a strong influence on agents’ actions, whereas font styling, text color, and item image clarity exhibit minor effects.

Copied to clipboard

Shuyu Gan, James Mooney, Pan Hao, Renxiang Wang, Mingyi Hong, Qianwen Wang, Dongyeop Kang

SilentDrift: Exploiting Action Chunking for Stealthy Backdoor Attacks on Vision-Language-Action Models (2026.findings-acl)

Challenge:	Existing methods to scale complex, open-ended tasks with unverifiable rewards are not scalable to multi-stage pipelines.
Approach:	They propose a process-based refinement framework that scales inference across stages of a multi-agent pipeline, instead of refining a single output over time.
Outcome:	The proposed framework scales inference across stages of a multi-agent pipeline, instead of refining a single output over time as in prior work.

Copied to clipboard

Bingxin Xu, Yuzhang Shang, Binghui Wang, Emilio Ferrara

Chunking

Challenge:	Existing backdoor attacks introduce kinematic discontinuities and distributional anomalies that can be flagged by standard trajectory detection.
Approach:	They propose a backdoor attack exploiting an intra-chunk visual open-loop vulnerability . they propose 93.2% Attack Success Rate and a poisoning rate under 2% .
Outcome:	The proposed attack achieves a 93.2% Attack Success Rate with a poisoning rate under 2% while maintaining a 95.3% Clean Task Success Rate.

X-ray Made Simple: Lay Radiology Report Generation and Robust Evaluation (2026.findings-acl)

Copied to clipboard

Kun Zhao, Chenghao Xiao, Sixing Yan, Haoteng Tang, William K. Cheung, Noura Al Moubayed, Liang Zhan, Chenghua Lin

Not All Citations Are Equal:Entropy-Guided Citation Selection for Noise-Resistant Medical LLM (2026.findings-acl)

Challenge:	Technical language and templated nature of professional reports hinder patient comprehension and allow models to artificially boost lexical metrics such as BLEU by reproducing common report patterns.
Approach:	They propose a layman's RRG framework that leverages layperson-friendly language to enhance patient accessibility and promote robust evaluation and report generation by encouraging models to focus on semantic accuracy over rigid templates.
Outcome:	The proposed framework improves model performance with more layman-style data, compared to templated professional language and inflated lexical scores.

Copied to clipboard

Minyu Gao, Hanlin Xiao, Ruoyu Wang, Shuai Yang, YeXuan Zhang, Xin Wu, Xingyu Liu

Decomposing Unitization and Typing for Efficient and Consistent Span-Bound Concept Annotation (2026.findings-acl)

Challenge:	Large language models have demonstrated extensive potential in medical applications . however, their practical deployment in healthcare faces significant challenges .
Approach:	They propose a training-free multi-turn reasoning framework and a post-training methodology that provides external knowledge support for large language models.
Outcome:	The proposed framework elicits internal thought, external thought, and fusion thought, with an entropy-based reward that encourages selective citation of beneficial external knowledge while penalizing noisy citations.

Copied to clipboard

Nupoor Gandhi, Michael Bada, Emma Strubell

InsightEval: An Expert-Curated Benchmark for Assessing Insight Discovery in LLM-Driven Data Agents (2026.findings-acl)

Challenge:	Substantial resources are typically spent on unitizing, the task of identifying precise span boundaries for entity mentions.
Approach:	They propose a method that focuses manual efforts on typed position annotations instead of full concept annotation.
Outcome:	The proposed procedure reduces the cost of concept annotations by focusing on typed positions instead of full concept annotation.

Copied to clipboard

Zhenghao Zhu, Yuanfeng Song, Xing Chen, Chengzhong Liu, Cui Yakun, Caleb Chen Cao, Sirui Han, Yike Guo

Challenge:	Existing frameworks for data analysis and insight exploration are lacking in terms of benchmarks . existing frameworks suffer from format inconsistencies, poorly conceived objectives, and redundant insights.
Approach:	They propose a data-curation pipeline to construct a new dataset named InsightEval.
Outcome:	The proposed benchmarks highlight prevailing challenges in automated insight discovery and raise key findings to guide future research.

ArrowGEV: Grounding Events in Video via Learning the Arrow of Time (2026.findings-acl)

Copied to clipboard

Fangxu Yu, Ziyao Lu, Liqiang Niu, Fandong Meng, Jie Zhou

Multi-lingual Functional Evaluation for Large Language Models (2026.findings-acl)

Challenge:	Existing approaches for grounding events in videos are limited by their time-sensitive nature . arrow of time in physics characterizes intrinsic directionality of temporal processes .
Approach:	They propose a framework that explicitly models temporal directionality in events to improve event grounding and temporal understanding in VLMs.
Outcome:	The proposed framework improves event grounding and directionality understanding in VLMs.

Copied to clipboard

Victor Ojewale, Inioluwa Deborah Raji, Suresh Venkatasubramanian

Why LoRA Fails to Forget: Regularized Low-Rank Adaptation Against Backdoors in Language Models (2026.findings-acl)

Challenge:	Multilingual competence in large language models is often evaluated via static data benchmarks such as Belebele, M-MMLU and M-GSM.
Approach:	They extend existing functional benchmark templates from English to five additional languages that span the range of resources available for NLP: French, Spanish, Hindi, Arabic and Yoruba.
Outcome:	The proposed models are translated from English to French, Spanish, Hindi, Arabic and Yoruba.

Copied to clipboard

Hoang-Chau Luong, Lingwei Chen

BoundRL: Efficient Token-level Structured Text Segmentation through Reinforced Boundary Generation (2026.findings-acl)

Challenge:	Low-Rank Adaptation (LoRA) is widely used for parameter-efficient fine-tuning of large language models, but is ineffective at removing backdoor behaviors from poisoned pretrained models when fine-timing on clean datasets.
Approach:	They propose a low-rank Adaptation method which increases spectral strength and corrects alignment through clean-strengthened regularization and trigger-insensitive constraints.
Outcome:	The proposed method significantly reduces attack success rates while maintaining clean accuracy.

Copied to clipboard

Haoyuan Li, Zhengyuan Shen, Sullam Jeoung, Yueyan Chen, Jiayu Li, Qi Zhu, Shuai Wang, Vassilis N. Ioannidis, Huzefa Rangwala

AOT*: Efficient Synthesis Planning via LLM-Empowered AND-OR Tree Search (2026.findings-acl)

Challenge:	Structured texts often contain elements beyond plain language, such as code snippets, which conventional sentence-level segmentation methods cannot handle effectively.
Approach:	They propose a token-level approach that performs efficient token-based text segmentation and label prediction for long structured texts.
Outcome:	The proposed approach outperforms existing models on short-shot prompts and SFT and standard RLVR models on complex LLM prompts.

Copied to clipboard

Xiaozhuang Song, Xuanhao Pan, Xinjian Zhao, Hangting Ye, Shufei Zhang, Jian Tang, Tianshu Yu

AutoSUIT Bench - Automated Security UnIt Test Benchmark for LLM Coding (2026.findings-acl)

Challenge:	Experimental evaluation shows that AOT* achieves competitive solve rates using 3-5 fewer iterations than existing LLM-based approaches.
Approach:	They propose a framework that integrates LLM-generated chemical synthesis pathways with systematic AND-OR tree search.
Outcome:	Experimental results show that AOT* improves search efficiency and solves faster than existing approaches.

Copied to clipboard

Samuel Osebe, Fan Yang, Junyi Li, Yue Gu, Yongxin Wang, Satyapriya Krishna, Kai-Wei Chang, Aram Galstyan, Rahul Gupta, Weitong Ruan

AnchorMem: Anchored Facts with Associative Contexts for Building Memory in Large Language Models (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are evolving rapidly on code generation tasks.
Approach:	They propose to automate the vulnerability code benchmark creation with iterative auto validation.
Outcome:	The proposed benchmark covers 232 CWE categories across C/C++, Java, and Python languages.

Copied to clipboard

Zhanyu Shen, Sijie Cheng, Zhicheng Guo, Weiqin Wang, Yile Wang, Hui Huang

Filling in the Mechanisms: How do LMs Learn Filler-Gap Dependencies under Developmental Constraints? (2026.findings-acl)

Challenge:	Existing memory systems rely on summarization to preserve contextual nuances and obscuring key retrieval features.
Approach:	They propose a method that decouples the retrieval unit from the generation context.
Outcome:	The proposed method outperforms baseline models on the LoCoMo benchmark.

Copied to clipboard

Atrey Desai, Sathvik Nair

PictoEduca: Building a Dataset for Spanish Text-to-Pictogram Generation (2026.findings-acl)

Challenge:	Language models lack language-specific biases, yet still posit some important syntactic generalizations.
Approach:	They applied Distributed Alignment Search to checkpoints of a language model from the BabyLM challenge to evaluate whether representations of filler-gap dependencies transfer between wh-questions and topicalization.
Outcome:	The results suggest shared, yet item-sensitive mechanisms may develop with limited training data.

Copied to clipboard

Alfonso Manuel Paredes Umeres, Marco Antonio Sobrevilla Cabezudo

HCFD: A Benchmark for Audio Deepfake Detection in Healthcare (2026.findings-acl)

Challenge:	PictoEduca is the first large-scale Spanish text-to-pictogram dataset for augmentative and alternative communication.
Approach:	They present PictoEduca, a large-scale Spanish text-to-pictogram dataset for augmentative and alternative communication.
Outcome:	The proposed dataset combines automatic annotation with targeted expert correction, supporting scalable and high-quality corpus construction.

Copied to clipboard

Mohd Mujtaba Akhtar, null Girish, Muskaan Singh

Linear Semantic Segmentation for Low-Resource Spoken Dialects (2026.findings-acl)

Challenge:	a new task for detecting codec-fakes under pathological speech conditions is presented . we focus on codec based synthetic speech since neural codec decoding is a core building block in speech generation pipelines.
Approach:	They propose a new task for detecting codec-fakes under pathological speech conditions . they focus on codec based synthetic speech since neural codec decoding is a core building block in speech pipelines .
Outcome:	The proposed framework outperforms speech-based models on Healthcare CodecFake . it achieves the strongest performance on the task across clinical conditions and codecs .

Copied to clipboard

Kirill Chirkunov, Younes Samih, Abed Alhakim Freihat, Hanan Aldarmaki

Thinking Like a Botanist: Challenging Multimodal Language Models with Intent Driven Chain-of-Inquiry (2026.findings-acl)

Challenge:	Existing models for semantic segmentation are primarily developed and evaluated on high-resource written text, limiting their effectiveness on low-resourced conversational varieties.
Approach:	They propose a multi-genre benchmark for semantic segmentation in Arabic, focusing on dialectal discourse.
Outcome:	The proposed model outperforms baselines on dialectal non-news genres while performing well on high-resource written text.

Copied to clipboard

Syed Nazmus Sakib, Nafiul Haque, Shahrear Bin Amin, Hasan Muhammad Abdullah, Md Mehedi Hasan, Mohammad Zabed Hossain, Shifat E. Arman

Analysing Differences in Persuasive Language in LLM-Generated Text: Uncovering Stereotypical Gender Patterns (2026.findings-acl)

Challenge:	Visual question-based reasoning is a key component of vision-language models.
Approach:	They propose a framework for visual question-answering that integrates visual intent with visual severity to improve diagnostic accuracy.
Outcome:	The proposed framework improves diagnostic correctness, reduces hallucination, and increases reasoning efficiency.

Copied to clipboard

Amalie Brogaard Pauli, Maria Barrett, Max Müller-Eberstein, Isabelle Augenstein, Ira Assent

Challenge:	Prior work has shown that large language models can successfully persuade humans and amplify persuasive language.
Approach:	They propose a framework for evaluating how persuasive language generation is affected by recipient gender, sender intent, or output language.
Outcome:	The proposed framework varies persuasive language when the recipient gender is specified or when the sender intent is specified.

Robust In-Context Selection via Online Learned Position-Corrected Attention (2026.findings-acl)

Copied to clipboard

Deeksha Koul, Gaurav Kumar, Yash Sabale, Sunita Sarawagi

Comprehensiveness Metrics for Automatic Evaluation of Factual Recall in Text Generation (2026.findings-acl)

Challenge:	Existing methods to fix this limitation can be classified into two ways: (1) Methods that use the LLM to generate the selection either via logits of item identifiers, or explicit rank permutations often requiring multiple LLM calls or fine-tuning.
Approach:	They propose a method that harnesses attention patterns available from a single forward call on the Large Language Model (LLM) the method learns the logic for item selection using a few in-context examples and a simple online position-debiasing mechanism to correct attention distortion.
Outcome:	The proposed method improves selection performance over direct generation and prior attention-based methods while remaining robust to prompt variations and item ordering.

Copied to clipboard

Adam Dejl, James Barry, Alessandra Pascale, Javier Carnerero-Cano

TableVista: Benchmarking Multimodal Table Reasoning under Visual and Structural Complexity (2026.findings-acl)

Challenge:	Large language models (LLMs) produce incomplete or selectively omit key information . omissions of key information or misrepresentation of conflicting evidence can cause harm .
Approach:	They propose a method that decomposes texts into atomic statements and uses natural language inference to identify missing facts and a Q A-based metric that extracts question-answer pairs and compares responses across sources.
Outcome:	The proposed evaluation metrics show they perform better than more complex metrics, but at a cost.

Copied to clipboard

Zheyuan Yang, Liqiang Shang, Junjie Chen, Xun Yang, Chenglong Xu, Bo Yuan, Chenyuan Jiao, Yaoru Sun, Yilun Zhao

Lost in Translation: Do LVLM Judges Generalize Across Languages? (2026.findings-acl)

Challenge:	TableVista evaluates multimodal table reasoning under visual and structural complexity . current models struggle to maintain reasoning consistency when structural complexity combined with visually integrated presentations.
Approach:	They propose a benchmark for evaluating multimodal table reasoning under visual and structural complexity.
Outcome:	The proposed model performs poorly on visual and structural complexity.

Copied to clipboard

Md Tahmid Rahman Laskar, Mohammed Saidul Islam, Mir Tafseer Nayeem, Amran Bhuiyan, Mizanur Rahman, Shafiq Joty, Enamul Hoque, Jimmy Huang

Making Revisions Understandable: A Survey of Edit Intentions, Methods, and Applications (2026.findings-acl)

Challenge:	MM-JudgeBench is the first large-scale benchmark for multilingual and multimodal judge model evaluation.
Approach:	They propose a multilingual benchmark for multilingual and multimodal judge model evaluation that includes over 60K pairwise preference instances spanning 25 typologically diverse languages.
Outcome:	The proposed benchmark includes over 60K pairwise preference instances spanning 25 languages.

Copied to clipboard

Fangping Lan, Qi Zhang, Eduard Dragut

Challenge:	Text revision is a core process in document creation, capturing how authors iteratively refine, reorganize, and improve written content.
Approach:	They synthesize text revision research through the lens of edit intentions . they review prior work across the revision workflow including corpus construction, edit intention taxonomies, edit intentions, and edit intention identification.
Outcome:	The proposed approach synthesizes datasets, taxonomies, identification methods, and applications and highlights key open research directions.

GSM-Noise: Exploring and Enhancing Large Language Models’ Reasoning under Noisy Inputs (2026.findings-acl)

Copied to clipboard

Zhengxin Zhang, Chengyu Huang, Xufu Liu, Dan Zhao, Jinyan Su, Claire Cardie

VALUE ALIGNMENT TAX: Measuring Value Trade-offs in LLM Alignment (2026.findings-acl)

Challenge:	Large language models struggle when dealing with complex, ill-formed, or noisy inputs . open-source models are less robust, while closed-source ones are more robust .
Approach:	They propose to use GSM-Noise to refine inputs before engaging in in-depth analysis to improve LLM robustness under noisy conditions.
Outcome:	The proposed model can achieve consistent performance gains under noisy conditions with prompt engineering, supervised finetuning, and reinforcement learning.

Copied to clipboard

Jiajun Chen, Hua Shen

Confidence-Aware Ranker Ensembles for Robust In-Context Knowledge Editing (2026.findings-acl)

Challenge:	Existing work on value alignment characterizes value relations statically, ignoring how interventions reshape the value system.
Approach:	They propose a framework that quantifies value trade-offs by measuring how alignment-induced changes propagate across interconnected values relative to achieved on-target gain.
Outcome:	The proposed framework measures how value trade-offs propagate across values . it can be used to evaluate intended improvements and unintended side effects .

Copied to clipboard

Tejal Nair, Mahmud Wasif Nafee, Maiqi Jiang, Ashley Gao, Haipeng Chen, Yanfu Zhang

EMSDialog: Synthetic Multi-person Emergency Medical Service Dialogue Generation from Electronic Patient Care Reports via Multi-LLM Agents (2026.findings-acl)

Challenge:	Large language models excel at factual recall, but can propagate stale or incorrect knowledge.
Approach:	They propose a feature-weighted ensemble for in-context knowledge editing that calibrates three heterogeneous rankers and extracts simple confidence features from each ranker.
Outcome:	The proposed method achieves 88.33% Edit-Success Rate over the best single retriever . it improves edit accuracy without touching model weights and approaches oracle upper bound (91%).

Copied to clipboard

Xueren Ge, Sahil Murtaza, Anthony Cortez, Homa Alemzadeh

Better LLM Reasoning via Dual-Play (2026.findings-acl)

Challenge:	Existing medical dialogue corpora are largely dyadic or lack multi-party workflow and annotations needed for this setting.
Approach:	They propose an ePCR-grounded, topic-flow-based multi-agent generation pipeline that iteratively plans, generates, and self-refines dialogues with rule-based factual and topic flow checks.
Outcome:	The proposed pipeline yields a dataset of 4,414 synthetic multi-speaker EMS conversations annotated with 43 diagnoses, speaker roles, and turn-level topics.

Copied to clipboard

Zhengxin Zhang, Chengyu Huang, Aochong Oliver Li, Claire Cardie

CobwebTM: Probabilistic Concept Formation for Lifelong and Hierarchical Topic Modeling (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have made remarkable progress through Reinforcement Learning with Verifiable Rewards (RLVR) however, external supervision remains a bottleneck for tasks and domains for which supervised data are scarce or non-existent.
Approach:	They propose a novel dual-play framework that adversarially trains two models initialized from the same base model.
Outcome:	The proposed framework improves the math reasoning performance of large language models.

Copied to clipboard

Karthik Singaravadivelan, Anant Gupta, Zekun Wang, Christopher J. MacLellan

Adaptive Test-Time Compute Allocation with Evolving In-Context Demonstrations (2026.findings-acl)

Challenge:	Topic modeling seeks to uncover latent semantic structure in text corpora with minimal supervision.
Approach:	They propose a lifelong hierarchical topic model based on incremental probabilistic concept formation that constructs semantic hierarchies online without predefining the number of topics.
Outcome:	The proposed model achieves strong topic coherence, stable topics over time, and high-quality hierarchies without predefining the number of topics.

Copied to clipboard

Bowen Zuo, Dongruo Zhou, Yinglun Zhu

Challenge:	Existing approaches to scaling test-time compute rely on static compute allocation or sample from fixed generation distributions.
Approach:	They propose a test-time compute allocation framework that jointly adapts where computation is spent and how generation is performed.
Outcome:	The proposed approach outperforms baselines while consuming less inference-time compute.

CCD: Mitigating Hallucinations in Radiology MLLMs via Clinical Contrastive Decoding (2026.findings-acl)

Copied to clipboard

Xi Zhang, Zaiqiao Meng, Jake Lever, Edmond S. L. Ho

Can Large Language Models Adequately Perform Symbolic Reasoning Over Time Series? (2026.findings-acl)

Challenge:	Multimodal large language models generate medical hallucinations due to over-sensitivity to clinical sections.
Approach:	They propose a framework that integrates structured clinical signals from task-specific radiology expert models.
Outcome:	The proposed framework improves overall performance on radiology report generation (RRG) on the MIMIC-CXR dataset, it yields up to 17% improvement in RadGraph-F1.

Copied to clipboard

Zewen Liu, Juntong Ni, Xianfeng Tang, Max SY Lau, Qi He, Wenpeng Yin, Wei Jin

Reading Between the Lines: The One-Sided Conversation Problem (2026.findings-acl)

Challenge:	Large Language Models (LLMs) and Multimodal LLMs (MLLMs) show strong performance in complex reasoning tasks, but their ability to extract symbolic laws from time series data remains underexplored.
Approach:	They propose a benchmark to assess symbolic reasoning over real-world time series across three tasks: multivariate symbolic regression, Boolean network inference, and causal discovery.
Outcome:	The proposed framework integrates LLMs with genetic programming to form a closed-loop symbolic reasoning system.

Copied to clipboard

Victoria Ebert, Rishabh Singh, Tuochao Chen, Noah A. Smith, Shyamnath Gollakota

Fico: Evaluating Vision-Language Models under Visual Fidelity and Compression at Scale (2026.findings-acl)

Challenge:	In many real-world scenarios, only one side of a conversation is available for processing.
Approach:	They propose a one-sided conversation problem to reconstruct the missing speaker's turns and generate faithful summaries from one-side transcripts.
Outcome:	The proposed model improves reconstructions with prompting, but smaller models require fine tuning.

Copied to clipboard

Jianhong Tu, Nicholas Crispino, Kyle Montgomery, Chenguang Wang, Dawn Song

SycoBench-600: Measuring Sycophancy and Correction Selectivity in LLM Assistants (2026.findings-acl)

Challenge:	Visual text compression is emerging paradigm for rendering text as images for processing by vision-language models.
Approach:	They propose a benchmark to assess VLM robustness under dense visual inputs.
Outcome:	Evaluating 13 general-purpose VLMs and 3 OCR-specialized models reveals performance drops sharply under increased density or reduced resolution; cross-task transfer between OCR, NIAH, and VQA is limited; and VQ is comparatively robust because low-level details are lost before high-level semantics.

Copied to clipboard

Debu Sinha

Challenge:	Existing evidence shows that instruction-following language models can be helpful and cooperative, but they also create reliability failure modes.
Approach:	They propose a benchmark that measures the ability to accept correct suggestions while resisting incorrect ones.
Outcome:	The proposed benchmark measures the ability to accept correct suggestions while resisting incorrect ones.

Calibrating Model-Based Evaluation Metrics for Summarization (2026.findings-acl)

Copied to clipboard

Hongye Liu, Dhanajit Brahma, Ricardo Henao

SafeMERGE: Preserving Safety Alignment in Fine-Tuned Large Language Models via Selective Layer-Wise Model Merging (2026.findings-acl)

Challenge:	Recent advances in summary evaluation are based on model-based metrics to assess quality dimensions, such as completeness, conciseness, and faithfulness.
Approach:	They propose a general framework that generates individual and average proxy scores without relying on reference summaries, human annotations, or expensive model-based metrics.
Outcome:	The proposed framework outperforms baselines on seven datasets on continuous-value scenarios, such as summarization, but is applicable to discrete-value tasks, such QA.

Copied to clipboard

Aladin Djuhera, Swanand Ravindra Kadhe, Farhan Ahmed, Syed Zawad, Holger Boche

Learning to Control Summaries with Score Ranking (2026.findings-acl)

Challenge:	Recent studies show that fine-tuning can erode safety alignment, causing LLMs to respond to harmful or unethical prompts.
Approach:	They propose a lightweight framework that restores safety while maintaining downstream performance.
Outcome:	The proposed framework reduces harmful outputs compared to other defenses, with negligible impact on utility.

Copied to clipboard

Hongye Liu, Liang Ding, Ricardo Henao

Summarization

Challenge:	Recent advances in summarization focus on improving summary quality across multiple dimensions, but they overlook the challenge of controlling summary generation with respect to individual dimensions.
Approach:	They propose a loss function that aligns model outputs with fine-grained, model-based evaluation scores to enable both improvement in summary quality and dimension-specific control.
Outcome:	The proposed method improves the overall quality of summaries while maintaining strong control over individual quality dimensions.

Raw Pointer Rewriting with LLMs for Translating C to Safer Rust (2026.findings-acl)

Copied to clipboard

Yifei Gao, Chengpeng Wang, Pengxiang Huang, Xuwei Liu, Mingwei Zheng, Xiangyu Zhang

Beyond Single Plots: A Benchmark for Question Answering on Multi-Charts (2026.findings-acl)

Challenge:	C2Rust is a system programming language that enforces strict memory and type safety guarantees.
Approach:	They propose a raw pointer rewriting technique that lifts raw pointers in individual functions to appropriate Rust data structures.
Outcome:	The proposed technique eliminates 18.57% of local raw pointers and improves memory safety on 28 real-world C projects.

Copied to clipboard

Azher Ahmed Efat, Seok Hwan Song, Wallapak Tavanapong

Assessing Y-Axis Influence: Bias in Multimodal Language Models on Chart-to-Table Translation (2026.findings-acl)

Challenge:	Existing research on chart understanding has been limited to single chart images.
Approach:	They propose a dataset specifically designed for question answering over multi-chart images.
Outcome:	The proposed method shows a 27.4% LLM-based accuracy drop on human-authored questions and a 5.39% gain in the human-generated questions.

Copied to clipboard

Seok Hwan Song, Azher Ahmed Efat, Wallapak Tavanapong

FLOP-Efficient Training: Early Stopping Based on Test-Time Compute Awareness (2026.findings-acl)

Challenge:	Existing studies have not examined the y-axis-related biases in chart-to-table translation models.
Approach:	They propose a framework for analyzing y-axis-related bias on five state-of-the-art models.
Outcome:	The proposed framework analyzes y-axis-related biases on five state-of-the-art models.

Copied to clipboard

Hossam Amer, Maryam Dialameh, Hossein Rajabzadeh, Walid Ahmed, Weiwei Zhang, Yang Liu

Save the Good Prefix: Precise Error Penalization via Process-Supervised RL to Enhance LLM Reasoning (2026.findings-acl)

Challenge:	Prior work shows that increasing test-time compute (TTC) can improve accuracy of large language models.
Approach:	They propose a TTC-aware training algorithm that jointly selects a checkpoint and a corresponding TTC configuration to minimize training compute without sacrificing accuracy.
Outcome:	The proposed method reduces training compute by 92% while maintaining accuracy.

Copied to clipboard

Haolin Liu, Dian Yu, Sidi Lu, Yujun Zhou, Rui Liu, Zhenwen Liang, Haitao Mi, Chen-Yu Wei, Dong Yu

POSTCONDBENCH: Benchmarking Correctness and Completeness in Formal Postcondition Inference (2026.findings-acl)

Challenge:	Existing reinforcement learning methods rely on sparse outcome rewards, which fail to credit correct intermediate steps in partially successful solutions.
Approach:	They propose a process reward model that rewards correct steps only when they detect errors . they propose VPPO, which rewards the correct prefix and an erroneous suffix .
Outcome:	a new approach outperforms sparse-reward RL and prior PRM-guided baselines on Pass@1 and Pass@K . a process reward model (PRM) outperformed sparser-rebound RL on multiple reasoning benchmarks .

Copied to clipboard

Gehao Zhang, Juan Zhai

Challenge:	Existing benchmarks emphasize correctness under limited evaluation settings . evaluation of formal specifications is time-consuming, errorprone and requires substantial expertise.
Approach:	They propose a multilingual benchmark for evaluating method-level postcondition generation from real-world software.
Outcome:	The proposed benchmarks show that evaluation remains a key bottleneck . 420 Python and Java tasks are paired with a high-quality postcondition set .

Who’s Asking? Simulating Role-Based Questions for Conversational AI Evaluation (2026.findings-acl)

Copied to clipboard

Navreet Kaur, Hoda Ayad, Hayoung Jung, Shravika Mittal, Munmun De Choudhury, Tanu Mitra

CircuitSynth: Reliable Synthetic Data Generation (2026.findings-acl)

Challenge:	Language model users embed personal and social context in their questions.
Approach:	They propose a framework for simulating role-based questions using a taxonomy of asker roles for patients, caregivers, practitioners.
Outcome:	The proposed framework simulates 15,321 questions that embed each asker role’s goals, behaviors, and experiences.

Copied to clipboard

Zehua Cheng, Wei Dai, Jiahao Sun, Thomas Lukasiewicz

MedRedFlag: Investigating how LLMs Redirect Misconceptions in Real-World Health Communication (2026.findings-acl)

Challenge:	Existing approaches lack mechanisms to balance linguistic expressivity with formal guarantees regarding validity and coverage.
Approach:	They propose a neuro-symbolic framework that decouples semantic reasoning from surface realization.
Outcome:	The proposed framework achieves 100% Schema Validity even in complex logic puzzles where unconstrained baselines fail (12.4%) while outperforming state-of-the-art methods in rare-combination coverage.

Copied to clipboard

Sraavya Sambara, Yuan Pu, Ayman Ali, Vishala Mishra, Lionel Wong, Monica Agrawal

Challenge:	Large language models (LLMs) are increasingly being used by lay users for medical advice, but they have not yet been tested for this crucial competency.
Approach:	They develop a semi-automated pipeline to curate MedRedFlag, a dataset of 1100+ reddit questions that require redirection.
Outcome:	The proposed pipeline compares state-of-the-art LLMs to those from clinicians to find out how they perform under real-world health communication.

Can VLMs Predict Future States? Bootstrapping World Models from Inverse Dynamics (2026.findings-acl)

Copied to clipboard

Yifu Qiu, Yftah Ziser, Anna Korhonen, Shay B Cohen, Edoardo Ponti

PL-MTEB: Polish Massive Text Embedding Benchmark (2026.findings-acl)

Challenge:	unified vision–language models (VLMs) struggle to generate physically plausible transitions between frames from instructions.
Approach:	They find that VLMs struggle to generate physically plausible transitions between frames from instructions.
Outcome:	The proposed model outperforms state-of-the-art image editing models on Aurora-Bench . it achieves the best average human evaluation across all subsets of Aurora-bench compared with other models .

Copied to clipboard

Rafał Poświata, Sławomir Dadas, Michał Wiktor Perełkiewicz

RubRIX: Rubric-Driven Risk Mitigation in Caregiver-AI Interactions (2026.findings-acl)

Challenge:	Text embeddings are used in many NLP tasks, including document clustering, semantic search, question answering, and classification.
Approach:	They introduce the Polish Massive Text Embedding Benchmark (PL-MTEB) it is a comprehensive benchmark for text embeddings in the Polish language.
Outcome:	The proposed model is based on 30 different NLP tasks in the Polish language.

Copied to clipboard

Drishti Goel, Jeongah Lee, Qiuyue Zhong, Violeta J. Rodriguez, Daniel S. Brown, Ravi Karkar, Dong Whi Yoo, Koustuv Saha

Spectral Gravity Formant Estimation for Phonetic Segmentation (2026.findings-acl)

Challenge:	Existing frameworks focused on general risks may not adequately capture nuanced risks of LLMs in caregiving contexts.
Approach:	They propose a theory-driven, clinician-validated framework for evaluating risks in LLMs . RubRIX operationalizes five empirically-derived risk dimensions: Inattention, Bias Stigma, Information Inaccuracy, Uncritical Affirmation, and Epistemic Arrogance.
Outcome:	The proposed framework reduces risk components by 45-98% after one iteration across models.

Copied to clipboard

Michael S. Yantosca, Albert M. K. Cheng

Listen, Pause, and Reason: Toward Perception-Grounded Hybrid Reasoning for Audio Understanding (2026.findings-acl)

Challenge:	a recent study suggests that end-to-end orthographic approaches miss the mark on time . linguistic applications which require high fidelity in the temporal domain, the loss of timing information is untenable .
Approach:	a new algorithm uses spectral gravity to estimate formants for enhanced phonetic segmentation . a deadline-bounded expectation maximization algorithm is proposed to estimate salient speech frequencies .
Outcome:	a new algorithm outperforms the state-of-the-art on key clustering metrics . the proposed algorithm generates reasonable alignments across multiple languages with no a priori training.

Copied to clipboard

Jieyi Wang, Yazhe Niu, Dexuan Xu, Zhongyu Wei

DeFrame: Debiasing Large Language Models Against Framing Effects (2026.findings-acl)

Challenge:	Recent Large Audio Language Models (LALMs) have shown strong capabilities in audio understanding, yet their reasoning remains vulnerable to perceptual errors.
Approach:	They propose a large-scale dataset for Perception-Aware Question Answering that uses a hierarchical decoupling strategy to separate speech from environmental sounds and distinguishes among multiple speakers.
Outcome:	The proposed model improves on MMAU-mini, MMAR, and PAQA while maintaining comparable performance on multiple benchmarks.

Copied to clipboard

Kahee Lim, Soyeon Kim, Steven Euijong Whang

Idea First, Code Later: Disentangling Problem Solving from Code Generation in Evaluating LLMs for Competitive Programming (2026.findings-acl)

Challenge:	Existing debiasing methods improve overall fairness, but fail to reduce framing-induced disparities.
Approach:	They propose a framing-aware debiasing method that encourages LLMs to be more consistent across frams.
Outcome:	The proposed method reduces overall bias and improves robustness against framing disparities, enabling LLMs to produce fairer and more consistent responses.

Copied to clipboard

Sama Hadhoud, Alaa Elsetohy, Frederikus Hudi, Jan Christian Blaise Cruz, Steven Halim, Alham Fikri Aji

LLMs in the Real World: Evaluating “AI” in Emergency Contexts (2026.findings-acl)

Challenge:	Existing evaluations conflate algorithmic reasoning with code-level implementation.
Approach:	They propose to center editorials in both solution generation and evaluation . they propose to compare editorials to gold standards and validate an LLM-as-a-judge protocol .
Outcome:	The proposed approach improves solve rates on some LLMs with gold editorials . but the gap between gold and generated editorials shows bottlenecks in implementation .

Copied to clipboard

Sara Court, Lara Downing, Micha Elsner

A Survey of Toxicity Mitigation Strategies for Multilingual Language Models (2026.findings-acl)

Challenge:	Despite considerable overlap between academic and industry-based developers of Large Language Models (LLMs), it seems Natural Language Processing researchers have a science outreach problem.
Approach:	They propose a set of concrete recommendations for stakeholders at every stage of the development and deployment pipeline.
Outcome:	The proposed model performs worse with lower-resourced languages or worse with higher-resource languages.

Copied to clipboard

Soham Dan, Himanshu Beniwal, Thomas Hartvigsen

LLM-KT: Enhancing Large Language Models with Knowledge Tracing via Multi-Level Plug-and-Play Alignment (2026.findings-acl)

Challenge:	Large language models can reproduce and amplify toxic content, including hate speech, harassment, and bias.
Approach:	They propose a comprehensive survey of the many detoxification methods tailored to multilingual LLMs.
Outcome:	The proposed methods are based on data filtering, style transfer, expert-based logit steering, retrieval augmentation, and human feedback.

Copied to clipboard

Ziwei Wang, Jie Zhou, Qin Chen, Bo Jiang, Qingchun Bai, Liang Dou, Liang He

Fast-Decoding Diffusion Language Models via Progress-Aware Confidence Schedules (2026.findings-acl)

Challenge:	Existing methods to learn behavioral sequences fail to capture complex behavioral patterns due to a lack of deep reasoning capabilities and world knowledge.
Approach:	They propose a framework that integrates the reasoning power of Large Language Models with the sequential modeling strengths of traditional KT methods via multi-level plug-and-play alignment.
Outcome:	Extensive experiments on four standard datasets show that the proposed framework outperforms existing methods on state-of-the-art questions.

Copied to clipboard

Amr Mohamed, Yang Zhang, Michalis Vazirgiannis, Guokan Shang

When Cultures Meet: Multicultural Text-to-Image Generation (2026.findings-acl)

Challenge:	SchED is a training-free, model-agnostic early-exit algorithm that terminates diffusion decoding using a progress-aware confidence threshold.
Approach:	They propose a training-free, model-agnostic early-exit algorithm that terminates diffusion decoding using a progress-aware confidence threshold.
Outcome:	The proposed algorithm achieves 4 speedups on instruction-tuned models while maintaining baseline performance on average.

Copied to clipboard

Parth Bhalerao, Oana Ignat, Brian Trinh, Mounika Yalamarty

Aligning with Your Own Voice: Self-Corrected Preference Learning for Hallucination Mitigation in LVLMs (2026.findings-acl)

Challenge:	a new task to evaluate text-to-image generation models for multicultural scenes is unexplored.
Approach:	They propose a benchmark task to evaluate text-to-image models in multicultural settings . they use a dataset of 9,000 images spanning five countries, three age groups, two genders, 25 historical landmarks, and five languages to analyze behavior .
Outcome:	The proposed benchmark analyzes the behavior of state-of-the-art models across multiple dimensions including alignment, image quality, aesthetics, knowledge, and fairness.

Copied to clipboard

Byeonggeuk Lim, JungMin Yun, Junehyoung Kwon, Kyeonghyun Kim, YoungBin Kim

VocalRep: Structure-Aware Vocal Representations for Multimodal Generation (2026.findings-acl)

Challenge:	Existing preference learning-based approaches rely on proprietary models to construct preference datasets, causing a distributional mismatch between the proprietary and target models.
Approach:	They propose a framework that aligns LVLMs using in-distribution data derived from the model's intrinsic knowledge.
Outcome:	The proposed framework surpasses baselines in hallucination mitigation while requiring only 5.2k samples.

Copied to clipboard

Da Shen, Zhenqiang Weng, Tianyu Liu, Gongyu Chen, Runhua Shi, Jiahui Chen, Chaofan Ding, Wei-Qiang Zhang, Zihao Chen

RealSec-bench: A Benchmark for Evaluating Secure Code Generation in Real-World Repositories (2026.findings-acl)

Challenge:	Existing approaches to vocal separation are optimized for signal-level reconstruction, but they overlook structural disentanglement required for downstream generation tasks.
Approach:	They propose a structure-aware learning framework to disentangle vocals, harmonies, and accompaniment . they combine global vocal identity conditioning with ranking-based objectives .
Outcome:	The proposed framework disentangles lead vocals, harmonies, and accompaniment while enforcing role consistency across long-form audio.

Copied to clipboard

Yanlin Wang, Ziyao Zhang, Chong Wang, Xinyi Xu, Mingwei Liu, Yong Wang, Jiachi Chen, Zibin Zheng

InfiAgent: An Infinite-Horizon Framework for General-Purpose Autonomous Agents (2026.findings-acl)

Challenge:	Existing benchmarks for large language models fail to capture complex interplay between functionality and security.
Approach:	They propose a benchmark for secure code generation constructed from real-world, high-risk Java repositories.
Outcome:	The proposed benchmarks highlight the gap between functional and secure code generation in LLMs.

Copied to clipboard

Chenglin Yu, Yuchen Wang, Songmiao Wang, Hongxia Yang, Li Ming

Efficient and Effective Internal Memory Retrieval for LLM-Based Healthcare Prediction (2026.findings-acl)

Challenge:	Existing LLMs break down on long-horizon tasks due to unbounded context growth and accumulated errors.
Approach:	They propose a framework that externalizes persistent state into a file-centric state abstraction and keeps the agent’s reasoning context strictly bounded regardless of task duration.
Outcome:	Experiments on DeepResearch and an 80-paper literature review show that the proposed framework maintains higher long-horizon coverage than baseline models without task-specific fine-tuning.

Copied to clipboard

Mingchen Li, Jiatan Huang, Zonghai Yao, Hong yu

GAVEL: Evidence-Contract Debate with Mechanized Scrutiny for Provenance-Grounded Fact-Checking (2026.findings-acl)

Challenge:	Existing approaches to ground large language models in external knowledge are limited by hallucinations and a lack of granular medical context.
Approach:	They propose a framework that replaces external retrieval with internal, key-based knowledge access by encoding clinical information directly into the model’s parameter space.
Outcome:	The proposed framework achieves state-of-the-art performance across four benchmark healthcare outcome prediction datasets.

Copied to clipboard

Ruoyu Xu, Gaoxiang Li, Victor S. Sheng

EMPATH: An Ensemble Method for Automatic Fine-Grained Turn-Level Dialogue Empathy Evaluation with a Novel Emotional Distance Metric (2026.findings-acl)

Challenge:	Evidence-grounded fact-checking requires predicting claim veracity while returning faithful evidence at fine granularity.
Approach:	They propose a multi-agent debate framework that enforces evidence grounding throughout inference.
Outcome:	The proposed framework improves provenance-aware metrics over existing frameworks.

Copied to clipboard

Dongning Rao, Zhihua Liang, Zhihua Jiang

Q2EI: Query-to-Entity Inference for Semantic Condensation in Domain-Specific Retrieval (2026.findings-acl)

Challenge:	Empathy evaluation metrics are lacking in the competitions, and classical dialogue evaluation metrics require further investigation.
Approach:	They propose a framework which combines fine-tuned models, large language models, classical dialogue evaluation metrics, and a novel metric.
Outcome:	The proposed framework improves on the WASSA 2024 benchmark and shows a statistically significant 8% improvement on the EX dataset.

Copied to clipboard

Yixuan Sun, Zhenqin Xu, HanFeng Zhai, Zishu Yu, Xiaohui Peng

TVWorld: Foundations for Remote-Control TV Agents (2026.findings-acl)

Challenge:	Existing generative expansions introduce redundancy or hallucinations that cause semantic drift.
Approach:	They propose a query rewriting strategy that reframes rewrite as semantic condensation rather than expansion.
Outcome:	The proposed method outperforms baselines on medical and legal benchmarks while reducing token consumption.

Copied to clipboard

Zhantao Ma, Quanfeng Lu, Shuai Zhong, Dahai Yu, Ping Luo, Michael Ng

RiTeK: A Dataset for Large Language Models Complex Reasoning over Textual Knowledge Graphs in Medicine (2026.findings-acl)

Challenge:	Existing work on large vision–language models focuses on point-and-click interaction, while remote-control interaction is underexplored.
Approach:	They propose a topology-aware training framework that injects topology awareness into LVLMs.
Outcome:	The proposed model achieves 68.3% success rate on TVWorld-N, surpassing closed-source benchmarks and state-of-the-art (SOTA) benchmarks show that existing agents lack topology awareness for focus-based, long-horizon TV navigation.

Copied to clipboard

Jiatan Huang, Mingchen Li, Zonghai Yao, Dawei Li, Yuxin Zhang, Zhichao Yang, Yongkang Xiao, Feiyun Ouyang, Xiaohan Li, Shuo Han, Hong yu

PRA-RAG: Provably Robust Aggregation in Retrieval-Augmented Generation against Retrieval Corruption (2026.findings-acl)

Challenge:	Existing methods for retrieving medical textual knowledge Graphs struggle to perform well, a study finds . existing methods struggle to provide accurate answers to complex questions, he says .
Approach:	They synthesize user queries integrating diverse topological structures, relational information, and complex textual descriptions.
Outcome:	a new dataset for medical textual knowledge graphs shows that existing methods struggle to perform well . main bottlenecks lie in the scarcity of existing medical TKGs and the limited expressiveness of their topological structures .

Copied to clipboard

Xue Tan, Yi Zheng, Chang Huo, Yunruo Zhang, Yu Liu, Hao Luan, Zhuyang Yu, Jun Dai, Xiaoyan Sun, Ping Chen

Evaluating Perspectival Biases in Cross-Modal Retrieval (2026.findings-acl)

Challenge:	Existing defense mechanisms lack theoretical robustness guarantees and perform unreliably when the LLM has limited knowledge of the retrieved content.
Approach:	They propose a provably robust retrieval aggregation algorithm designed to defend against poisoning attacks on retrieved texts.
Outcome:	Experiments show that PRA-RAG reduces the attack success rate to as low as 1% while maintaining an accuracy of 71%, significantly outperforming representative state-of-the-art (SOTA) methods.

Copied to clipboard

Teerapol Saengsukhiran, Peerawat Chomphooyod, Narabodee Rodjananant, Chompakorn Chaksangchaichot, Patawee Prakrankamanant, Witthawin Sripheanpol, Pak Lovichit, Sarana Nutanong, Ekapol Chuangsuwanich

Beyond Static Synthetic Noise: Assessing the Robustness of Large Language Models to Natural Context Variation in the Real World (2026.findings-acl)

Challenge:	a recent study shows that multimodal retrieval systems are expected to operate in a semantic space, agnostic to the language or cultural origin of the query.
Approach:	They introduce a benchmark to quantify linguistic and cultural biases in multimodal retrieval systems . they propose a framework to decouple language from culture and decouples it from semantics .
Outcome:	The proposed benchmark systematically measures the effects of linguistic and cultural biases on retrieval performance.

Copied to clipboard

Yulong Wu, Viktor Schlegel, Riza Batista-Navarro

Measuring Watermarking under Jailbreaking: ASR Inflation and Goal-Compliance Mismatch (2026.findings-acl)

Challenge:	Current robustness evaluation methods rely on static synthetic perturbations to stress-test models.
Approach:	They propose a framework for automatically evaluating QA models under naturally occurring textual perturbations by replacing context passages with revised Wikipedia edit histories.
Outcome:	The proposed framework replaces context passages with revised Wikipedia edit histories to improve model performance.

Copied to clipboard

Sungwoo Han, Sangjun Moon, Jingun Kwon, Hidetaka Kamigaito, Manabu Okumura

Inventive Problem Solving with LLMs: A Benchmark for TRIZ Reasoning (2026.findings-acl)

Challenge:	Recent work studies watermarking under benign prompts, but its behavior under jailbreaking prompts remains underexplored.
Approach:	They evaluate six methods on four LLMs using two jailbreak benchmarks and three settings: Static, AutoDAN, and DSN.
Outcome:	The proposed methods inflate judge-based attack success rate under jailbreaking, but not harmful-goal compliance.

Copied to clipboard

Zhu Wang, Brian Uzzi

Do LLM Agents Really Mimic Humans? Diagnosing and Aligning Microeconomic Behaviors in Macro-ABMs (2026.findings-acl)

Challenge:	Large language models are increasingly used in inventive problem-solving but effective support requires more than open-ended idea generation.
Approach:	They propose a dataset and benchmark for TRIZ reasoning grounded in open technical sources and U.S. patents.
Outcome:	The proposed framework represents trade-offs and links them to standardized inventive principles.

Copied to clipboard

Guangya Liu, Cheng Wang, Jiangtong Li, Huafei Wu, Changjun Jiang

Challenge:	Existing studies focus on replicating macro-level stylized facts while neglecting verification of micro-level decision-making.
Approach:	They propose a framework that replicates macro-level stylized facts while ignoring micro-level decision-making.
Outcome:	The proposed framework improves alignment with human trends and captures behavioral heterogeneity.

Explanation Quality Assessment as Ranking with Listwise Rewards (2026.findings-acl)

Copied to clipboard

Thomas Bailleux, Tanmoy Mukherjee, Emmanuel Lonca, Pierre Marquis, Zied Bouraoui

ViLegalLM: Language Models for Vietnamese Legal Text (2026.findings-acl)

Challenge:	a new approach to explanation quality assessment is to rank explanations by relative quality . standard reward objectives do not preserve graded distinctions well enough for policy optimization .
Approach:	They reformulate explanation quality assessment as a ranking problem instead of a generation problem . they train listwise and pairwise ranking models to preserve ordinal structure .
Outcome:	The proposed model outperforms regression on score separation and performance on listwise and pairwise models.

Copied to clipboard

Truong-Phuc Nguyen, Quy-Nhan Nguyen, Minh-Tien Nguyen

From Heads to Neurons: Causal Attribution and Steering in Multi-Task Vision–Language Models (2026.findings-acl)

Challenge:	ViLegalLM is the first suite of Vietnamese pretrained language models for legal text understanding and generation.
Approach:	They propose a suite of Vietnamese pretrained language models for legal text understanding and generation.
Outcome:	The proposed models outperform instruction-tuned adaptation on four main Vietnamese legal downstream tasks.

Copied to clipboard

Qidong Wang, Junjie Hu, Ming Jiang

Syntax as a Rosetta Stone: Universal Dependencies for In-Context Coptic Translation (2026.findings-acl)

Challenge:	Existing models focus on single tasks, limiting comparability of neuron importance . ranking strategies overlook how task-dependent information pathways shape write-in effects of feed-forward network (FFN) neurons.
Approach:	They propose a gradient-free framework for task-aware neuron attribution and steering in multi-task vision-language models.
Outcome:	The proposed framework outperforms existing methods in identifying task-critical neurons and improves model performance after steering.

Copied to clipboard

Abhishek Purushothama, Emma Thronson, Alexia Guo, Amir Zeldes

AI Agents for the Science of Science: A Survey of Tasks, Architectures, Evaluations, and Challenges (2026.findings-acl)

Challenge:	Existing work using bilingual dictionaries to support inference for vocabulary items is lacking for low-resource languages.
Approach:	They propose to use universal dependency parses of input sentences to augment in-context learning prompts for low resource machine translation for the Coptic language.
Outcome:	The proposed approach achieves state-of-the-art results for the Coptic language.

Copied to clipboard

Yixuan Liu, Yicheng Zhang

Challenge:	The Science of Science (SciSc) examines how scientific knowledge is produced, evaluated, and transformed by utilizing large-scale scholarly and bibliometric data.
Approach:	They propose a task-centered taxonomy for AI agents that model citations, collaborations, and community dynamics.
Outcome:	The proposed taxonomy distinguishes agents as simulations from tools for empirical analysis and scientific workflows.

A2O: LLM-based Agentic Learning of Action-to-Object Features for Video Action Recognition (2026.findings-acl)

Copied to clipboard

Taiki Sekii, Fumiaki Sato

PrinciplismQA: A Philosophy-Grounded Approach to Assessing LLM-Human Clinical Medical Ethics Alignment (2026.findings-acl)

Challenge:	Recent action recognition based on vision–language pretraining and self-supervised video foundation models tends to induce spurious correlations and shortcut learning by relying on action-irrelevant cues.
Approach:	They propose a framework in which an LLM agent integrates the two approaches within an agentic learning paradigm to design motion features tailored to the target actions.
Outcome:	The proposed model is based on the commonsense knowledge of large language models (LLMs) and the open vocabulary object detector to make the model attend to objects in a video required for recognizing the target actions.

Copied to clipboard

Chang Hong, Minghao Wu, Qingying Xiao, Yuchi Wang, Xiang Wan, Guangjun Yu, Benyou Wang, Yan Hu

Take Out Your Calculators: Estimating the Real Difficulty of Question Items with LLM Student Simulations (2026.findings-acl)

Challenge:	Existing benchmarks lack systematic approaches to integrate philosophical frameworks and expert validation for ethical reasoning assessment.
Approach:	They propose a philosophy-grounded approach to assess medical ethics alignment . PrinciplismQA comprises 3,648 expert-validated questions spanning knowledge assessment and clinical reasoning .
Outcome:	PrinciplismQA provides a philosophy-grounded approach to assessing medical ethics alignment.

Copied to clipboard

Christabel Acquaye, Yi Ting Huang, Marine Carpuat, Rachel Rudinger

Numerical Reasoning

Challenge:	Standardized math assessments require expensive human pilot studies to establish the difficulty of test items.
Approach:	They propose to use large language models to model difficulty of multiple-choice math questions for real-world students.
Outcome:	The proposed model predicts difficulty of multiple-choice math questions for students . correlations between model and real-world difficulty are high, the authors show .

SAFER: Advancing Safety Alignment via Efficient Ex-Ante Reasoning (2026.findings-acl)

Copied to clipboard

Kehua Feng, Keyan Ding, Yuhao Wang, Menghan Li, Fanjunduo Wei, Xinda Wang, Huajun Chen

Speculative Verification: Exploiting Information Gain for Speculative Decoding (2026.findings-acl)

Challenge:	Existing alignment methods struggle to cover diverse safety scenarios and remain vulnerable to adversarial attacks.
Approach:	They propose a framework for 'Safety' alignment via eFficient' Ex-Ante-R**easoning that instantiates structured Ex-Ance reasoning and embeds predefined safety rules.
Outcome:	The proposed framework enhances safety performance while maintaining usefulness and efficiency.

Copied to clipboard

Sungkyun Kim, Jaemin Kim, Dogyeong Yun, Jiho Shin, Junyeol Lee, Jiwon Seo

EngiBench: A Benchmark for Evaluating Large Language Models on Engineering Problem Solving (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are used for many applications but their size and computational cost make inference serving a significant challenge.
Approach:	They propose an efficient augmentation to Speculative Decoding (SD) that predicts speculation accuracy and dynamically adapts the verification length to maximize throughput.
Outcome:	The proposed model reduces wasted verification on rejected tokens and improves decoding efficiency.

Copied to clipboard

Xiyuan Zhou, Xinlei Wang, Yirui He, Ruixi Zou, Yang Wu, Yuheng Cheng, Yulu Xie, Wenxuan Liu, Huan Zhao, Yan Xu, Jinjin Gu, Junhua Zhao

LLM-Based Human-Agent Collaboration and Interaction Systems: A Survey (2026.findings-acl)

Challenge:	Existing benchmarks focus on well-defined or abstract reasoning and fail to capture real-world engineering problems.
Approach:	They propose a hierarchical benchmark to evaluate large language models on engineering problems.
Outcome:	The proposed model performs well under well-defined conditions and is based on three levels of difficulty and covers diverse engineering subfields.

Copied to clipboard

Henry Peng Zou, Wei-Chieh Huang, Yaozu Wu, Jizhou Guo, Yankai Chen, Chunyu Miao, Hoang H Nguyen, Yue Zhou, Weizhi Zhang, Liancheng Fang, Hanrong Zhang, Fangxin Wang, Pengfei Zhang, Langzhou He, Yangning Li, Dongyuan Li, Renhe Jiang, Philip S. Yu

Thinking Twice Makes Large Language Models Safer and More Helpful (2026.findings-acl)

Challenge:	Recent advances in large language models (LLMs) have sparked growing interest in building fully autonomous agents.
Approach:	They propose to integrate human-provided information, feedback, or control into the agent system to enhance system performance, reliability, and safety.
Outcome:	The proposed systems improve system performance, reliability, and safety by integrating human-provided information, feedback, or control into the agent system.

Copied to clipboard

Yutao Mou, Yuxiao Luo, Shikun Zhang, Wei Ye

AEQ-Bench: Measuring Empathy of Omni-Modal Large Models (2026.findings-acl)

Challenge:	Existing safety alignment techniques for large language models (LLMs) struggle to balance harmlessness and usefulness.
Approach:	They propose a safety-aware reflection-based reasoning framework that internalizes self-reflective reasoning and encourages reflection and correction.
Outcome:	The proposed framework outperforms reasoning-based alignment methods in safety alignment.

Copied to clipboard

Xuan Luo, Lewei Yao, Lanqing Hong, Kai Chen, Dehua Tao, Daxin Tan, Yukun Deng, Ruifeng Xu, Jing Li

BAPO: Boundary-Aware Policy Optimization for Reliable Agentic Search (2026.findings-acl)

Challenge:	Existing benchmarks focus on cognitive abilities, such as knowledge retrieval, complex reasoning, and instruction following, largely overlooking empathy evaluation.
Approach:	They propose to benchmark two core empathetic capabilities of omnimodal large models (OLMs) generating empatries by comprehending affective cues from multi-modal inputs and judging empathy of audio responses without relying on text transcription.
Outcome:	The proposed benchmark outperforms existing models with audio output capabilities but is unreliable for evaluating fine-grained paralinguistic expressiveness.

Copied to clipboard

Shiyu Liu, Yongjing Yin, Jianhao Yan, Yunbo Tang, Qinggang Zhang, Bei Li, Xin Chen, Jingang Wang, Xunliang Cai, Jinsong Su

GRASP: Graph-Reasoning Aided Survey Planning for High-Fidelity Related Work Generation (2026.findings-acl)

Challenge:	Existing RL-based agentic search models fail to recognize reasoning boundaries and rarely admit "I DON'T KNOW" lack of reliability leads to plausible but unreliable answers, introducing significant risks .
Approach:	They propose a framework to cultivate reliable boundary awareness without compromising accuracy.
Outcome:	Experiments show that the proposed framework improves the reliability of agentic search models.

Copied to clipboard

Haoming Li, Jessica Ouyang

Benchmarking Agentic Newswriting via Journalistic Workflows (2026.findings-acl)

Challenge:	Graph-Reasoning Aided Survey Planning (GRASP) is a framework for analyzing cited papers.
Approach:	They propose a framework that combines LLM planning for related work generation with graph algorithms to extract key relationships among cited papers.
Outcome:	The proposed framework generates RWS that closely match human-written targets in terms of discourse roles, intents, and grouping of citations.

Copied to clipboard

Yen-Che Chien, Kuang-Da Wang, Wei-Yao Wang, Wen-Chih Peng

Self-Correcting Text-to-Video Generation with Misalignment Detection and Localized Refinement (2026.findings-acl)

Challenge:	Recent advances in autonomous digital agents highlight their potential for structured tasks through autonomous decision-making and task decomposition, but it remains unclear how well such systems support real-world information-intensive workflows.
Approach:	They propose a benchmark to evaluate how journalists can use agents to organize and organize information from the web.
Outcome:	The proposed system can be used to iterate and evaluate newswriting tasks in real-world situations.

Copied to clipboard

Daeun Lee, Jaehong Yoon, Jaemin Cho, Mohit Bansal

Switching Heads and Softening Tokens: Turnkey Solutions to Visually Grounded Document QA (2026.findings-acl)

Challenge:	Recent text-to-video models struggle to faith-fully follow text prompts, authors say . authors propose a new refinement framework that detects fine-grained misalignments .
Approach:	They propose a video refinement framework that detects fine-grained misalignments . they propose preserving regions that should be preserved rather than regenerated .
Outcome:	The proposed framework detects fine-grained misalignments and performs targeted corrections . it preserves correctly generated entities, segments regions across frames, and regenerates problematic regions .

Copied to clipboard

Ximing Wen, Wenbo Li, Sudipta Paul, Yashas Malur Saidutta, Kalpa Gunaratna, Srinivas Chappidi

HTMuon: Improving Muon via Heavy-Tailed Spectral Correction (2026.findings-acl)

Challenge:	Document Question Answering lacks robust, end-to-end solutions capable of handling complex, multi-answer queries without reliance on ad-hoc processing.
Approach:	They propose a single-head architecture where coordinates are represented as special tokens within the unified vocabulary.
Outcome:	The proposed architectures improve visual grounding but lack spatial precision bound by discretization.

Copied to clipboard

Tianyu Pang, Yujie Fang, Zihang Liu, Shenyang Deng, Lei Hsiung, Shuhua Yu, Yaoqing Yang

Can Large Language Models Effectively Support Decision-Making in Sudden Emergencies? (2026.findings-acl)

Challenge:	Muon’s orthogonalized update rule suppresses the emergence of heavy-tailed weight spectra and over-emphasizes the training along noise-dominated directions.
Approach:	They propose to preserve Muon's ability to capture parameter interdependencies while producing heavier-tailed updates and inducing heavier-tail weight spectra.
Outcome:	The proposed algorithm suppresses the emergence of heavy-tailed weight spectra and over-emphasizes training along noise-dominated directions.

Copied to clipboard

Mengna Zhu, Jibing Wu, Lihua Liu, Yuran Gong, Yang Hao, Fu Yachao, Mao Wang, Lei Hou, Juanzi Li

J-Shuwa: A Large-Scale Web-Collected Japanese Sign Language-Japanese Parallel Corpus (2026.findings-acl)

Challenge:	Existing research has focused on the earlier stages of emergency response . lack of suitable datasets for reliable and compliance-aware decision-oriented modeling and evaluation is limiting current research .
Approach:	They propose a first real-world emergency decision-making dataset EDM-Bench . they propose 'rule-enhanced reasoning framework' that integrates external regulatory knowledge with constrained inference mechanisms to improve both decision safety and interpretability.
Outcome:	The proposed framework improves decision safety and interpretability by integrating regulatory knowledge with constrained inference mechanisms.

Copied to clipboard

Junwen Mo, MinhDuc Vo, Noriki Nishida, Shin’ichi Satoh, Hideki Nakayama

Hallucinations as Orthogonal Noise: Inference-Time Manifold Alignment via Dynamic Contextual Orthogonalization (2026.findings-acl)

Challenge:	Japanese Sign Language (JSL) is a low-resource sign language that has received limited attention in the AI community due to the lack of large-scale, publicly available parallel corpora.
Approach:	They propose a large-scale JSL-Japanese parallel corpus constructed from YouTube videos with hard-coded subtitles and closed captions.
Outcome:	The proposed model is effective for training models and can be used for future research across a wide range of tasks.

Copied to clipboard

Mingkuan Zhao, Wentao Hu, Tianchen Huang, Yuheng Min, Suquan Chen, Yide Gao, Yanbo Zhai, Shuangyong Song, Xuelong Li

ConvX: A Lightweight Converter to Bridge Indexed Dense Representations and Large Language Models for Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	Hallucinations in Large Language Models persist in critical domains where generated content diverges from contextual facts or logical constraints.
Approach:	They propose to generate hallucinations as orthogonal noise relative to the semantic manifold of the residual stream.
Outcome:	The proposed method achieves superior contextual faithfulness compared to state-of-the-art methods.

Copied to clipboard

Bonggeun Choi, Keunha Kim, Junho Han, Youngjoong Ko

Resonant Context Anchoring: Decoupling Attention Routing and Signal Gain at Inference Time (2026.findings-acl)

Challenge:	Existing RAG pipelines suffer from critical efficiency limitations due to their complexity and complexity.
Approach:	They propose a compression-based RAG framework that directly leverages indexed dense representations produced by a retriever, substituting to long text contexts.
Outcome:	Empirical results show that the proposed model achieves competitive performances compared to the state-of-the-art model that uses a large ad-hoc context compressor while offering substantially improved inference efficiency.

Copied to clipboard

Mingkuan Zhao, Yide Gao, Wentao Hu, Suquan Chen, Tianchen Huang, Zhenhua An, Zetao Chang, Xiayu Sun, Yuheng Min

S2H-DPO: Hardness-Aware Preference Optimization for Vision–Language Models (2026.findings-acl)

Challenge:	Existing mitigation strategies rely on suppressing specific neuron activations or employing computationally expensive contrastive decoding mechanisms, which often result in increased perplexity or significantly elevated inference latency.
Approach:	They propose a lightweight inference-time intervention method grounded in the perspective of residual stream signal dynamics to resolve the signal attenuation of external evidence during its propagation through deep networks.
Outcome:	The proposed method improves contextual faithfulness across multiple factual consistency and strong knowledge-conflict tasks while maintaining the model’s general language understanding capabilities.

Copied to clipboard

Nitish Shukla, Surgan Jandial, Arun Ross

Dr. Assistant: Enhancing Clinical Diagnostic Inquiry via Structured Diagnostic Reasoning Data and Reinforcement Learning (2026.findings-acl)

Challenge:	Existing methods focus on localized reasoning with pre-specified image indices, bypassing the skills of global visual search and autonomous cross-image comparison.
Approach:	They propose a learning framework that constructs multi-image preference data across three hierarchical reasoning levels requiring an increasing level of capabilities.
Outcome:	The proposed approach maintains strong single-image reasoning performance while strengthening multi-image understanding capabilities.

Copied to clipboard

Yue Guo, Fanfu Wang, Jianwei Lv, Xincheng Shi, Yuchen Li, Youya Wang, Yunsheng Zeng, Yujing Liu, Yunhao Qiao, Gen Li, Junfeng Wang, Bo Yuan

FigSIM: A Dataset for Fine-grained Suicide Severity and Figurative Language in Suicide Memes (2026.findings-acl)

Challenge:	Clinical Decision Support Systems (CDSSs) provide reasoning and inquiry guidance for physicians, yet they face high maintenance costs and low generalization capability.
Approach:	They propose a clinical diagnostic model with clinical reasoning and inquiry skills, the Dr. Assistant, and a pipeline to capture abstract reasoning logic.
Outcome:	The proposed model outperforms open-source models and achieves competitive performance to closed-source model.

Copied to clipboard

Liuliu Chen, Elise Carrotte, Brian E. Chapman, Jo Robinson, Mike Conway

Activation Steering for Chain-of-Thought Compression (2026.findings-acl)

Challenge:	Suicide memes are increasingly common on social media, yet remain poorly understood and potentially harmful.
Approach:	They propose a dataset designed for fine-grained analysis of suicide memes and benchmark 16 models for figurative language, suicide severity, and content detection.
Outcome:	The proposed model outperforms existing models on figurative language, suicide severity, and suicide-related content detection tasks.

Copied to clipboard

Seyedarmin Azizi, Erfan Baghaei Potraghloo, Souvik Kundu, Massoud Pedram

Decomposed Prompting Does Not Fix Knowledge Gaps, But Helps Models Say “I Don’t Know” (2026.findings-acl)

Challenge:	Large language models produce intermediate explanations, commonly referred to as chains of thought (CoTs), but the generated rationales are typically verbose, consuming many additional tokens, and thus degrading throughput and increasing inference energy consumption.
Approach:	They propose to generate concise reasoning traces by directly adjusting internal representations via activation steering.
Outcome:	The proposed method reduces generated token length by 69.4% across five reasoning benchmarks while maintaining accuracy.

Copied to clipboard

Dhruv Madhwal, Lyuxin David Zhang, Dan Roth, Tomer Wolfson, Vivek Gupta

VisTW: Benchmarking Vision-Language Models for Taiwanese Mandarin in Taiwan (2026.findings-acl)

Challenge:	Large language models struggle to recognize their knowledge limits in closed-book question answering . disagreements between prompting regimes are indicative of potential errors, study finds .
Approach:	They evaluate three task-equivalent prompting regimes: Direct, Assistive, Incremental . disagreements between prompting régimes provide a precise signal of internal uncertainty .
Outcome:	The proposed decomposed prompting outperforms standard uncertainty baselines as an error detector.

Copied to clipboard

Zhi Rui Tam, Yung-Yu Shih, Yen-Wei Lee, Ya-Ting Pai, Wen Yu Chang, Yun-Nung Chen

PsyChain: A Collaborative Chain-of-Agents Framework for Generating Personalized and Professional Counseling Dialogues (2026.findings-acl)

Challenge:	Vision-Language Models (VLMs) struggle in Taiwanese Mandarin environments due to region-specific orthographic and cultural context.
Approach:	They propose a human-grounded purity penalty for character mixing under Taiwanese-Mandarin-style prompts.
Outcome:	The proposed model outperforms the strongest open-weight baseline by 22 percentage points on dialogue tasks.

Copied to clipboard

Yi Feng, Zijie Yang, Chen Zhang, Wenxuan Zhang, Dongming Zhang, Liping Jing, Jian Yu

Breaking the Autoregressive Chain: Hyper-Parallel Decoding for Efficient LLM-Based Attribute Value Extraction (2026.findings-acl)

Challenge:	Existing psychological counseling datasets suffer from monolithic client personas, insufficient therapeutic depth, and a lack of process controllability.
Approach:	They propose a framework that evolves static counseling corpora into high-fidelity dialogues . they use a Client Profiler that pairs life scenarios with psychological personality archetypes based on client personality and stage progression .
Outcome:	The proposed framework achieves 61-91% win rates against domain-specific baselines in pairwise evaluation and the highest average score in human evaluation, indicating potential for real-world counseling.

Copied to clipboard

Theodore Glavas, Nikhita Vedula, Dushyanta Dhyani, Yilun Zhu, Shervin Malmasi

On the Editability of Delta Parameters in Post-Trained Models (2026.findings-acl)

Challenge:	Existing methods for text generation require multiple independent sequences to be decoded in parallel.
Approach:	They propose an algorithm that accelerates offline decoding by leveraging shared memory and computation across batches.
Outcome:	Experiments show that attribute-value pairs are conditionally independent, enabling decoding in parallel up to 96 tokens per prompt.

Copied to clipboard

Qiaoyu Tang, Le Yu, Bowen Yu, Hongyu Lin, Keming Lu, Yaojie Lu, Xianpei Han, Le Sun

Reasoning with Memory: Adaptive Information Management for Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	Several studies have explored delta parameter properties via pruning, quantization, low-rank approximation, and extrapolation, but what properties of delta parameters are essential for maintaining performance?
Approach:	They propose to examine delta parameter properties along magnitude and sign . they propose to use a loss-based local surrogate analysis to examine editing effects .
Outcome:	The proposed analysis shows that delta parameters can be edited while maintaining performance.

Copied to clipboard

Hieu Man, Ro-ee Tal, Abhishek Kumar, Jaejin Cho, Benjamin Hsu

LiCoMemory: Lightweight and Cognitive Agentic Memory for Efficient Long-Term Reasoning (2026.findings-acl)

Challenge:	Multi-hop reasoning remains a fundamental challenge for Retrieval-Augmented Generation systems.
Approach:	They propose a framework that provides a dynamic cognitive workspace for multi-hop reasoning . it uses an explicit working memory that persists across retrieval cycles and is continuously updated .
Outcome:	The proposed framework achieves state-of-the-art performance over existing systems on eight QA benchmarks.

Copied to clipboard

Zhengjun Huang, Zhoujin Tian, Qintian Guo, Fangyuan Zhang, Yingli Zhou, Di Jiang, Zeying Xie, Xiaofang Zhou

Multimodal Item Scoring for Natural Language Recommendation via Gaussian Process Regression with LLM Relevance Judgments (2026.findings-acl)

Challenge:	Large Language Models are constrained by limited context windows and lack of persistent memory . recent efforts address these limitations via external memory architectures .
Approach:	They propose an end-to-end agentic memory framework for real-time updating and retrieval that integrates hierarchical and temporal indexing layers.
Outcome:	The proposed framework outperforms established benchmarks in temporal reasoning, multi-session consistency, and retrieval efficiency.

Copied to clipboard

Yifan Simon Liu, Qianfeng Wen, Jiazhou Liang, Mark Zhao, Justin Cui, Anton Korikov, Armin Toroghi, Junyoung Kim, Scott Sanner

Dialectic-Med: Mitigating Diagnostic Hallucinations via Counterfactual Adversarial Multi-Agent Debate (2026.findings-acl)

Challenge:	Existing NLRec approaches use Dense Retrieval to compute item relevance scores . DR views the request as the sole relevance label, leading to a weak proxy for query relevance.
Approach:	They propose to use Gaussian Process Regression to model item relevance . they propose to combine LLM with a Gauss-based kernel to model multimodal relevance judging .
Outcome:	The proposed approach outperforms simpler unimodal kernels and baseline methods by up to 65% on four NLRec datasets and two LLM backbones.

Copied to clipboard

Zhixiang Lu, Jionglong Su

EPIR: Capturing Promoting and Inhibiting Relationships between Events (2026.findings-acl)

Challenge:	Existing Chain-of-Thought (CoT) approaches lack intrinsic correction mechanisms, rendering them vulnerable to error propagation.
Approach:	They propose a multi-agent framework that enforces diagnostic rigor through adversarial dialectics.
Outcome:	Empirical evaluations show that the proposed framework improves explanation faithfulness and mitigates hallucinations.

Copied to clipboard

Bowen Dong, Wenjun Wang, Xueli Liu, Quanlin Qiu

Challenge:	promoting and inhibiting relationships capture directional, probabilistic, and context-dependent shifts in event likelihood.
Approach:	They propose a framework for estimating promoting and inhibiting relationships from observed event data.
Outcome:	The proposed framework outperforms state-of-the-art models on real-world datasets in accuracy.

EULoInf: Efficient Hessian-Free Entropy Based Uncertainty-Aware Data Influence Approximation (2026.findings-acl)

Copied to clipboard

Runxin Cai, Jingtan Wang, Bryan Kian Hsiang Low

CodeRise: Bootstrapping LLMs for Ultra Low-Resource Programming Languages via Progressive Self-Refinement Curriculum (2026.findings-acl)

Challenge:	Extensive studies show that the effectiveness of fine-tuning heavily relies on the quality of training data.
Approach:	They propose a framework that approximates influence via uncertainty and gradient based validation loss lookahead.
Outcome:	The proposed framework matches or outperforms prior methods across diverse tasks and LLM architectures while reducing computational time and memory usage by over 50%.

Copied to clipboard

Tengfei Wen, Xuanang Chen, Ben He, Xiaoliang Cong, Le Sun

What Tokens Truly Matter? The Logit Conflation Problem in LLM Sampling (2026.findings-acl)

Challenge:	Existing methods for training data generation for low-resource languages suffer from a cold-start problem and lack diversity.
Approach:	They propose a two-stage framework that generates a high-quality, diverse, and progressively complex curriculum for Ultra Low-Resource Programming Languages (ULRPLs) they leverage the full formal syntax of the target language as structural guidance and apply a biased sampling strategy over library modules.
Outcome:	The proposed framework outperforms training-free and training-based baselines on two ULRPLs, Tengo and Janet.

Copied to clipboard

Pinlong Zhao, Huijun Tang, Pengfei Jiao, Mengyang Li

Bears, all bears, and some bears. Language Constraints on Language Models’ Inductive Inferences (2026.findings-acl)

Challenge:	Existing methods for large language models filter tokens based on logit magnitudes or derived statistics, under the implicit assumption that high-logit tokens are desirable.
Approach:	They propose to isolate the Logit Conflation Problem by using attention-weighted attribution to extract prompt-relevance from token logits.
Outcome:	The proposed method improves on LLaMA-3 and is training-free and low latency.

Copied to clipboard

Sriram Padmanabhan, Siyuan Song, Kanishka Misra

B-APO: Bias-Targeted Adversarial Preference Optimization for Debiasing Multimodal Large Language Models (2026.findings-acl)

Challenge:	Language places subtle constraints on how we make inductive inferences.
Approach:	They propose to use language to constrain inductive inferences by replicating an experiment . they find subtle differences arise in general purpose statistical learners like VLMs .
Outcome:	The proposed model can be used to extend inductive inferences to humans using language . the model can extend properties of a category to other members of the population, the authors show .

Copied to clipboard

Pinlong Zhao, Zike Ding, Zengshu Ye, Zhou Zhaoting

On the Fallacy of Global Token Perplexity in Spoken Language Model Evaluation (2026.findings-acl)

Challenge:	Existing debiasing methods create biased responses by completely removing an entire modality, forming an extreme and static training environment.
Approach:	They propose a method to debiase multimodal large language models by masking one modality and then enlarge the margin between clean and adversarial responses.
Outcome:	The proposed method achieves superior debiasing performance while maintaining general capabilities.

Copied to clipboard

Chan-Jan Hsu, Liang-Hsuan Tseng, Yi-Cheng Lin, Yen-Chun Kuo, Ju-Chieh Chou, Kai-Wei Chang, Hung-yi Lee, Carlos Busso

Modeling and Solving Stable Matching under Probabilistic Preferences with Large Language Models (2026.findings-acl)

Challenge:	Generative spoken language models are often evaluated using global token perplexity, which overlooks fundamental differences between speech and text modalities.
Approach:	They propose a variety of likelihood- and generative-based evaluation methods that serve in place of naive global token perplexity.
Outcome:	The proposed evaluations more faithfully reflect perceived generation quality, as evidenced by stronger correlations with human-rated mean opinion scores (MOS).

Copied to clipboard

Yuqi Kong, Shiyu Liu, Jiaxu Li, Hongtao Liu, Qi Qi, Weiran Shen

XQ-MEval: A Dataset with Cross-lingual Parallel Quality for Benchmarking Translation Metrics (2026.findings-acl)

Challenge:	Large language models (LLMs) have shown strong capability in understanding and simulating humans’ decisions, suggesting a new way to use LLMs as tools to study social systems.
Approach:	They propose a Hybrid GS–LLM matching method that integrates Gale–Shapley with probabilistic acceptance decisions.
Outcome:	The proposed method outperforms classical baselines in terms of stability and improves robustness under uncertainty.

Copied to clipboard

Jingxuan Liu, Zhi Qu, Jin Tei, Hidetaka Kamigaito, Lemao Liu, Taro Watanabe

Faithfulness vs. Safety: Evaluating LLM Behavior Under Counterfactual Medical Evidence (2026.findings-acl)

Challenge:	averaging metric scores across languages is suspicious since translations of equal quality receive different scores across language.
Approach:	They propose a semi-automatically built dataset to benchmark translation metrics using MQM-defined errors and a normalization strategy to mitigate cross-lingual scoring bias.
Outcome:	The proposed model shows that translation metrics suffer from cross-lingual scoring bias . the proposed model is based on a semi-automatically built dataset covering nine translation directions .

Copied to clipboard

Kaijie Mo, Siddhartha Venkatayogi, Chantal Shaib, Ramez Kouzy, Wei Xu, Byron C Wallace, Junyi Jessy Li

Challenge:	Existing models are overwhelmingly accurate when presented with counterfactual medical evidence . prior work explored conflicts between context and LLM parametric knowledge in the general domain .
Approach:	They construct a counterfactual medical QA dataset that requires models to answer clinical comparison questions with evidence from randomized controlled trials.
Outcome:	The proposed model overemphasizes the latter, and the model overestimates the latter.

Your LLM Agents are Temporally Blind: The Misalignment Between Tool Use Decisions and Human Time Perception (2026.findings-acl)

Copied to clipboard

Yize Cheng, Arshia Soltani Moakhar, Chenrui Fan, Parsa Hosseini, Kazem Faghih, Zahra Sodagar, Wenxiao Wang, Soheil Feizi

Challenge:	Large language model agents assume a stationary context, failing to account for real-world time elapsed between messages.
Approach:	They construct a dataset of multi-turn user–agent message trajectories across 76 scenarios . they collect human preferences between "calling a tool" and "directly answering" they also examine whether existing models lack human temporal perception .
Outcome:	The results show that existing models display poor alignment with human temporal perception . the findings provide insights to foster the development of more time-aware and human-aligned agents.

How Do LLMs "Trust" Unknown Knowledge? An Unknown Knowledge Based Jailbreak Attack (2026.findings-acl)

Copied to clipboard

Yixiao Huang, Lan Zhang, Chaoran Wang

ToolSafe: Enhancing Tool Invocation Safety of LLM-based agents via Proactive Step-level Guardrail and Feedback (2026.findings-acl)

Challenge:	Existing research on how to effectively utilize unknown knowledge has focused on how it can be used to enhance LLMs' performance in specialized fields.
Approach:	They propose a completely unrestricted and fully randomized jailbreak attack that embeds malicious queries within trust-enhanced unknown knowledge.
Outcome:	The proposed method achieves 99% to 100% ASR on all tested LLMs, including the latest GPT-5.1, and becomes SOTA.

Copied to clipboard

Yutao Mou, Zhangchi Xue, Lijun Li, Peiyang Liu, Shikun Zhang, Wei Ye, Jing Shao

A Survey of Large Models in Sports (2026.findings-acl)

Challenge:	Unlike chatbots, autonomous agents act directly on external environments, making tool invocation safety critical for reliable deployment.
Approach:	They develop a benchmark for step-level tool invocation safety detection in LLM agents and a guardrail model that proactively detects unsafe tool invoking actions before execution using multi-task reinforcement learning.
Outcome:	The proposed model reduces harmful tool invocations of ReAct-style agents by 65% on average and improves benign task completion by 10% under prompt injection attacks.

Copied to clipboard

Yichen Xu, Jianzhe Ma, Chuhan Wang, Zhonghao Cao, Liangyu Chen, Wenxuan Wang, Qin Jin

SafetyALFRED: Evaluating Safety-Conscious Planning of Vision Language Models (2026.findings-acl)

Challenge:	Increasing interest in sports has led to the rapid advancement of large models, particularly multimodal large language models (MLLMs) . linguistic intelligence is a key component of large-model-driven sports intelligence .
Approach:	They propose to establish a foundation for advancing research and practical development of large-model-driven sports intelligence.
Outcome:	The proposed model-driven sports intelligence will be able to process and generate sports-related language effectively and process multiple data modalities.

Copied to clipboard

Josue Torres-Fonseca, Naihao Deng, Yinpei Dai, Shane Storks, Yichi Zhang, Rada Mihalcea, Casey Kennington, Joyce Chai

Towards Unified Multimodal Large Language Models: A survey (2026.findings-acl)

Challenge:	Existing safety evaluations focus on hazard recognition through disembodied question answering (QA) settings, but lack a critical gap in evaluating an agent.
Approach:	They evaluate multimodal large language models with six categories of kitchen hazards . they propose a safety-based approach that prioritizes multi-step corrective actions .
Outcome:	The proposed model can recognize hazards in QA settings, but average mitigation success rates are low . the proposed model is based on the embodied agent benchmark ALFRED .

Copied to clipboard

Xu Ma, Yitian Zhang, Yun Fu

♪ Something Just Like TRuST ♪ *: Toxicity Recognition of Span and Target (2026.findings-acl)

Challenge:	unified multimodal large language models (MLLMs) are emerging but lack a systematic framework to connect them and situate current trends within a broader landscape.
Approach:	They present a systematic review of unified Multimodal Large Language Models . they outline the foundational concepts and prerequisites for understanding them .
Outcome:	The present review provides a systematic and systematic overview of unified MLLMs . it discusses persistent challenges and identify promising directions for future research .

Copied to clipboard

Berk Atıl, Namrata Sureddy, Rebecca J. Passonneau

Challenge:	Toxic language is pervasive online, and because LLMs are trained on web data, it generates such content.
Approach:	They propose a large-scale dataset that synthesizes toxicity definitions and an annotation scheme . they use a rigorous human annotation process to evaluate the diversity of the annotations .
Outcome:	The proposed model outperforms existing models on three tasks and is not reliable.

Synergizing Stylometrics with Semantics: Dual-Path Framework for LLM Detection and Attribution (2026.findings-acl)

Copied to clipboard

Xingyu Lu, Yumeng Ma, Xiang Zhou, Shengli Gan, Guiying Deng, Yang Wen, Yanbing Liu

Enhancing Multilingual Reasoning via Steerable Model Merging (2026.findings-acl)

Challenge:	Existing methods for identifying MGTs rely on statistical likelihood or deep embeddings.
Approach:	They propose a framework that extracts model-specific stylistic fingerprints across lexical, syntactic, and structural dimensions.
Outcome:	The proposed framework achieves a Macro-F1 score of 95.6% on the Wikipedia dataset.

Copied to clipboard

Zhuoran Li, Rui Xu, Jian Yang, Junnan Liu, Zhijun Chen, Qianren Mao, Hongcheng Guo, Jiaheng Liu, Likang Xiao, Ming LI, Xiaojie Wang

Adaptive Backtracking for Privacy Protection in Large Language Models (2026.findings-acl)

Challenge:	Model merging is an effective technique for composing the capabilities of a multilingual model and a reasoning model.
Approach:	They propose a model merging framework that modulates the contribution of each source model.
Outcome:	Experiments show that the proposed model merging framework outperforms strong baselines on multilingual reasoning benchmarks across 21 different languages.

Copied to clipboard

Zhihao Yao, Yuxuan Gu, Xiachong Feng, Weitao Ma, Bo Li, Xiaocheng Feng, Bing Qin

SemPA: Improving Sentence Embeddings of Large Language Models through Semantic Preference Alignment (2026.findings-acl)

Challenge:	Existing privacy protection methods are prone to privacy leakage, but they are not effective in ensuring the privacy of users.
Approach:	They propose to capture latent leakage tendency of large language models during generation process and to construct a new benchmark for personal information.
Outcome:	The proposed method improves privacy by up to 14% over strong baselines against adversarial attacks, avoiding the degradation of response utility.

Copied to clipboard

Ziyang Chen, Zhenxuan Huang, Yile Wang, Weiqin Wang, Lu Yin, Hui Huang

Interpretability from the Ground Up: Stakeholder-Centric Design of Automated Scoring in Educational Assessments (2026.findings-acl)

Challenge:	Existing sentence embedding methods rely on fixed prompt templates or involve modifications to the model architecture, compromising its generative capabilities.
Approach:	They propose a sentence-level direct preference optimization approach that boosts the sentence representations while preserving the generative ability of LLMs.
Outcome:	The proposed method improves representations of semantically meaningful vectors without sacrificing generation capability.

Copied to clipboard

Yunsung Kim, Michael Hardy, Joseph Tey, Candace Thille, Christopher J Piech

DANCE: Diversity-attended Dynamic Caching with Asymmetric Quantization for Test-time Adaptation of Vision-Language Models (2026.findings-acl)

Challenge:	Despite increasing demand for transparency and interpretability, the field has yet to develop a widely accepted solution for interpretable automated scoring to be used in large-scale real-world assessments.
Approach:	They propose to develop four principles of interpretability targeted at assessment stakeholder groups to address the need for transparency and interpretability in automated scoring.
Outcome:	The proposed framework outperforms many uninterpretable scoring methods in terms of scoring accuracy and is, on average, within 0.06 QWK of the uninterprétable SOTA.

Copied to clipboard

Shunge Zou, Changhu Wang, Wei Ju, Ziyue Qiao, Xiao Luo

How to Set the Learning Rate for Large-Scale Pre-training? (2026.findings-acl)

Challenge:	Existing approaches to test-time adaptation of vision-language models measure prediction entropy but these samples tend to approach prototypes with limited coverage of data distributions.
Approach:	They propose a new approach for test-time adaptation of vision-language models . they construct a dynamic cache to store diversity-aware test samples .
Outcome:	The proposed approach is more efficient than current methods on augmented visual models.

Copied to clipboard

Yunhua Zhou, Shuhao Xing, Junhao Huang, Xipeng Qiu, Qipeng Guo

Evaluating Large Vision Language Models on Bangla Medical Visual Question Answering (2026.findings-acl)

Challenge:	Optimal configuration of the learning rate (LR) is a fundamental yet formidable challenge in large-scale pre-training.
Approach:	They propose a Fitting Paradigm and a Transfer Paradigme to investigate fit and transfer . they propose scalability and elucidate the reasons why module-wise parameter tuning underperforms .
Outcome:	The proposed model reduces the search complexity by reducing the search cost by lowering the search factor.

Copied to clipboard

Rafid Ahmed, Intesar Tahmid, Mir Sazzat Hossain, Tasnimul Hossain Tomal, Md Mahir Jawad, Anam Borhan Uddin, Md Fahim, Md Farhad Alam Bhuiyan

SIV-Bench: A Video Benchmark for Social Interaction Understanding and Reasoning (2026.findings-acl)

Challenge:	Recent advances in Large Language Models and Large Vision Language Model (LVLMs) have demonstrated promising capabilities in complex reasoning tasks, but low-resource contexts like Bangla are underexplored.
Approach:	They propose a multilingual medical visual question answering dataset using Bangla.
Outcome:	The proposed model performs well on generalized visual tasks but struggles with fine-grained diagnostic reasoning, achieving low accuracy in specialized categories.

Copied to clipboard

Fanqi Kong, Weiqin Zu, Xinyu Chen, Yaodong Yang, Song-Chun Zhu, Xue Feng

No One Fits All: From Fixed Prompting to Learned Routing in Multilingual LLMs (2026.findings-acl)

Challenge:	Existing benchmarks for Multimodal Large Language Models (MLLMs) have been lacking due to the rich nature of social interaction.
Approach:	They propose a video benchmark to evaluate MLLMs' capabilities across social scene understanding, social state reasoning, and social dynamics prediction.
Outcome:	The proposed benchmarks evaluate MLLMs' capabilities across social scene understanding, social state reasoning, and social dynamics prediction tasks.

Copied to clipboard

Wei-Chi Wu, Sheng-Lun Wei, Hen-Hsen Huang, Hsin-Hsi Chen

Understanding and Mitigating Spurious Signal Amplification in Test-Time Reinforcement Learning for Math Reasoning (2026.findings-acl)

Challenge:	Existing studies show that translation-based prompting is not universally optimal for multilingual LLMs.
Approach:	They evaluate translation-based prompting across ten languages and four benchmarks . they propose a lightweight classifier that predicts whether native or translation- based prompts are optimal .
Outcome:	The proposed classifiers achieve statistically significant improvements over fixed prompting strategies across ten languages and four benchmarks.

Copied to clipboard

Yongcan Yu, Lingxiao He, Jian Liang, Kuangpu Guo, Meng Wang, Qianlong Xie, Xingxing Wang, Ran He

Steering LLM Thinking with Budget Guidance (2026.findings-acl)

Challenge:	a framework to mitigate spurious optimization signals is proposed for test-time reinforcement learning (TTRL) Reinforcement learning with verifiable rewards (RLVR) is an effective paradigm for improving large language models on structured challenging reasoning tasks.
Approach:	They propose a framework to mitigate spurious optimization signals from label noise . they propose to use a frequency-based sampling strategy to exclude ambiguous samples .
Outcome:	The proposed framework outperforms existing TTRL baselines on three large language models across multiple mathematical reasoning benchmarks.

Copied to clipboard

Junyan Li, Wenshuo Zhao, Yang Zhang, Chuang Gan

Distilling Long-CoT Reasoning through Collaborative Step-wise Multi-Teacher Decoding (2026.findings-acl)

Challenge:	Existing budget control methods for large language models are inadequate for long reasoning . budget guidance can be used to control reasoning length without fine-tuning .
Approach:	They propose a budget guidance method that models a Gamma distribution over remaining thinking length during next-token generation and uses it to guide generation in a soft, token-level manner.
Outcome:	The proposed method achieves up to 26% accuracy gain on the MATH-500 benchmark compared to baseline methods while maintaining competitive accuracy with only 63% of the thinking tokens used by the full-thinking model.

Copied to clipboard

Taewon Yun, Jisu Shin, Jeonghwan Choi, Seunghwan Bang, Hwanjun Song

Adversarial Yet Cooperative: Multi-Perspective Reasoning in Retrieved-Augmented Language Models (2026.findings-acl)

Challenge:	Existing curation-based approaches to inference are inefficient and fail to adapt dynamically, leading to redundant sampling and missed opportunities for complementary reasoning.
Approach:	They propose a collaborative multi-teacher decoding framework that performs step-wise reasoning synthesis guided by predictive perplexity–based scoring and beam search.
Outcome:	The proposed framework generates higher-quality reasoning data and achieves student-level results, demonstrating that fine-grained collaboration yields structured, efficient, and robust reasoning distillation.

Copied to clipboard

Can Xu, Lingyong Yan, Jiayi Wu, Haosen Wang, Shuaiqiang Wang, Yuchen Li, Jizhou Huang, Dawei Yin, Xiang Li

PyraMathBench: Evaluating and Improving Mathematical Capability in Large Language Models (2026.findings-acl)

Challenge:	Existing training paradigms rely on outcome-oriented rewards, which provide insufficient signal for shaping the complex, multi-step reasoning process.
Approach:	They propose a framework that integrates large reasoning models with retrieval-augmented generation to improve reasoning fidelity and verification rigor.
Outcome:	Experiments on multiple benchmarks demonstrate the effectiveness of the proposed framework.

Copied to clipboard

Zetian Ouyang, Linlin Wang, Gerard de Melo, Liang He

D2PCM:A Multi-Turn Dialogue Dataset with Personalized Contextual Memory (2026.findings-acl)

Challenge:	Numerical reasoning is ubiquitous in scientific research and financial analysis, but few benchmarks evaluate them by integrating numerical processing and mathematical reasoning.
Approach:	They propose a numerically-integrated hierarchical benchmark with 27,215 questions derived from 7,404 math word problems that spans 4 key cognitive aspects, 14 subcategories, and 2 modalities.
Outcome:	The proposed model improves Qwen-2.5 score with SOLVE and IRPO training.

Copied to clipboard

Zhe Yang, Yi Huang, Yaqin Chen, Chunyang Gao, Jingyu Yao, Junlan Feng

MTP-RL: Acceleration of Reinforcement Learning Rollouts with Policy-Aligned Multi-Token Prediction (2026.findings-acl)

Challenge:	Conventional interactive algorithms have predominantly treated memory as a contextual element, neglecting the nuanced cognitive processes involved in individualized memory encoding and retrieval.
Approach:	They propose a multi-turn dialogue dataset with Personalized Contextual Memory to facilitate advanced research on personalized memory processing.
Outcome:	The proposed datasets provide a comprehensive benchmark to facilitate advanced research on personalized memory processing.

Copied to clipboard

Ke Wang, Aohan Zeng, Zhengxiao Du, Yuxuan Hu, Bohan Zhang, Xinyi Wang, Jie Tang, Jing Zhang

ReflexiCoder: Teaching Large Language Models to Self-Reflect on Generated Code and Self-Correct It via Reinforcement Learning (2026.findings-acl)

Challenge:	Reinforcement learning (RL) is widely applied to boost the performance of pretrained models, yet its training efficiency is severely constrained by rollout generation.
Approach:	They propose a framework that accelerates the rollout phase for diverse models by equipping a pipeline to equip the multi-layer parameter-sharing MTP for all models and an advantage-aware MTP optimization strategy.
Outcome:	The proposed framework achieves stable growth of acceptance length during RL training, and also accelerates RL rollouts, achieving an average 23.1%–55.3% reduction in rollout time compared to baselines.

Copied to clipboard

Juyong Jiang, Jiasi Shen, Sunghun Kim, Kang Min Yoo, Jeonghoon Kim, Sungju Kim

TRAC: Token-level Reward Assignment for Coherent Abstractive Summarization (2026.findings-acl)

Challenge:	Existing iterative refinement strategies that generate solutions in a single forward pass often hit a performance ceiling on complex algorithmic tasks.
Approach:	They propose a reinforcement learning framework that internalizes the structured reasoning trajectory directly into the model’s weights.
Outcome:	The proposed framework achieves 94.51% (87.20%) on HumanEval, 81.80% (78.57%) on MBPP, 35.00% on BigCodeBench, 52.21% on LiveCodeBech, and 37.34% on CodeForces in a single-attempt setting.

Copied to clipboard

null 陈宣齐, null 容梓莹, Xinfeng Liao, Lianxi Wang, Ying Gao, Shengyi Jiang

Summarization

Challenge:	Large Language Models (LLMs) have achieved remarkable success in text summarization, but maintaining logical coherence and contextual consistency remains a pervasive challenge in long-form generation.
Approach:	They propose a framework that introduces a token-level reward function by integrating relative sentence gain, inter-sentence attention, and a Gaussian length penalty.
Outcome:	The proposed model outperforms the sequence-level baseline by 11.05% in fluency and 10.61% in Relevance.

Planning Beyond Text: Graph-based Reasoning for Complex Narrative Generation (2026.findings-acl)

Copied to clipboard

Hanwen Gu, Chao Guo, Junle Wang, Wenda Xie, Yisheng Lv

Enrich, Aggregate, and Generate: Three-stage Biomedical Data-to-Text Generation Using Large Language Models in Low-resource Scenarios (2026.findings-acl)

Challenge:	Existing methods for long-form complex narrative generation struggle to maintain global narrative coherence and logical consistency.
Approach:	They propose a framework that performs narrative planning on structural graph representations instead of direct sequential text representations.
Outcome:	The proposed model outperforms representative baselines across diverse scenarios.

Copied to clipboard

Yupian Lin, Guangya Yu, Yuang Bian, Cheng Yuan, Hui Luo, Tong Ruan

HiCoLoRA: Addressing Context-Prompt Misalignment via Hierarchical Collaborative LoRA for Zero-Shot DST (2026.findings-acl)

Challenge:	Biomedical data-to-text generation is a branch of Natural Language Generation, aiming at generating textual natural language descriptions that can fluently and precisely describe the structured data.
Approach:	They propose an LLM framework that can be used to generate textual natural language descriptions using in-context learning.
Outcome:	The proposed framework provides good interpretability and superior performance on the BioLeaflets dataset.

Copied to clipboard

Shuyu Zhang, Yifan Wei, Xinru Wang, Yanmin Zhu, Yangfan He, Yixuan Weng, Yujie Liu, Bin Li

From Conversation to Evaluation: Benchmarking LLMs on Development Knowledge via SimpleDevQA (2026.findings-acl)

Challenge:	Existing approaches to zero-shot Dialog State Tracking (zs-DST) are inadequate to generalize to new domains without extensive training.
Approach:	They propose a framework that enhances zero-shot slot inference through robust prompt alignment.
Outcome:	Experiments on multi-domain datasets show that HiCoLoRA outperforms baselines, achieving SOTA in zs-DST.

Copied to clipboard

Jing Zhang, Lianghong Guo, Yanlin Wang, Terry Yue Zhuo, Yong Wang, Mingwei Liu, Jiachi Chen, Ensheng Shi, Yuchi Ma, Hongyu Zhang, Zibin Zheng

LLEOT: A Privacy-Enhancing Offsite Tuning Framework via Loss Landscape Elevation (2026.findings-acl)

Challenge:	Existing Dev Knowledge QA benchmarks are limited in development knowledge scope and often not built from real user queries.
Approach:	They conduct preliminary analysis of real user–LLM dialogues from WildChat to investigate the importance of Dev Knowledge QA in AI-assisted software development scenarios.
Outcome:	The proposed benchmark is based on real user–LLM dialogues from WildChat.

Copied to clipboard

Jin Zhong, Jinglin Liang, Tongtong Yang, Zijian Xie, Shuangping Huang, Hanlin Gu

Beyond Compromise: Pareto-Lenient Consensus for Efficient Multi-Preference LLM Alignment (2026.findings-acl)

Challenge:	Existing approaches to fine-tune large language models are infeasible due to privacy regulations.
Approach:	They propose an offsite tuning framework that secures data privacy and model parameter and capability privacy.
Outcome:	The proposed framework secures data privacy and model parameter and capability privacy while preserving gradient alignment.

Copied to clipboard

Renxuan Tan, Rongpeng Li, Zhifeng Zhao, Honggang Zhang

HSS-Synth: Humanities and Social Sciences Data Synthesis for LLMs (2026.findings-acl)

Challenge:	Recent approaches to align LLMs with diverse human values are based on static linear scalarization or rigid gradient projection . however, these approaches often sacrifice potential global Pareto improvements to avoid transient local trade-offs.
Approach:	They propose a game-theoretic framework that reimagines alignment as a dynamic negotiation process.
Outcome:	The proposed framework breaks the deadlock between static linear scalarization and rigid gradient projection . it allows the model to escape local degradation and explore the distal Pareto-optimal frontier .

Copied to clipboard

Ru Peng, Tianyu Zhao, Xijun Gu, Zhiting Fan, Haokai Xu, Jinyang Zhang, Yawen Zeng, Yihong Zhuang, Kexin Yang, Junyang Lin, Dayiheng Liu, Junbo Zhao

Challenge:	High-quality, diverse data are vital for large language models (LLMs) but remain scarce and costly.
Approach:	They define the first HSS domain system covering 14 mainstream fields and introduce HSS-Synth.
Outcome:	the proposed pipeline outperforms 14 leading baselines on 16 benchmarks.

LLMRouterBench: A Massive Benchmark and Unified Framework for LLM Routing (2026.findings-acl)

Copied to clipboard

Hao Li, Yiqun Zhang, Zhaoyan Guo, Chenxu Wang, Shengji Tang, Qiaosheng Zhang, Yang Chen, Biqing Qi, Peng Ye, Lei Bai, Zhen Wang, Shuyue Hu

Challenge:	Large language model (LLM) routing assigns each query to the best suitable model from an ensemble.
Approach:	They introduce a large-scale benchmark and unified framework for LLM routing . they find that many routing methods exhibit similar performance under unified evaluation .
Outcome:	The proposed benchmark provides comprehensive metrics for both performance-oriented and performance-cost trade-off routing.

CEMT:Controllable Element-Oriented Machine Translation via Structured Linguistic Reasoning (2026.findings-acl)

Copied to clipboard

Lingling Shi, Haoyu Jin, Ruiyu Fang, Shuangyong Song, Jinsong Su, Yongxiang Li, Xuelong Li

Orchestrating Tokens and Sequences: Dynamic Hybrid Policy Optimization for RLVR (2026.findings-acl)

Challenge:	Large Language Models suffer from paraphrasing errors, omissions, or hallucinations when input contains translation-specific elements that require strict preservation or controlled transformation.
Approach:	They propose a Controllable Element-Oriented Machine Translation framework that decomposes the translation process into a linguistically grounded analysis, strategy formulation, and final generation.
Outcome:	The proposed framework improves on the WMT23/24 Chinese–English benchmarks while significantly reducing element-level constraint violations.

Copied to clipboard

Zijun Min, Bingshuai Liu, Ante Wang, Long Zhang, Anxiang Zeng, Haibo Zhang, Jinsong Su

Seeing Isn’t Believing: Mitigating Belief Inertia via Active Intervention in Embodied Agents (2026.findings-acl)

Challenge:	Existing RLVR algorithms focus on different granularities and have complementary strengths and limitations.
Approach:	They propose a framework for reinforcement learning with verifiable rewards that bridges RLVR and GSPO . group-level importance ratios are used to update a policy, which preserves fine-grained credit assignment .
Outcome:	The proposed framework outperforms existing methods on seven reasoning benchmarks.

Copied to clipboard

Hanlin Wang, Chak Tou Leong, Jian Wang, Wenjie Li

AlgBench: To What Extent Do Large Reasoning Models Understand Algorithms? (2026.findings-acl)

Challenge:	Existing large language models (LLMs) have enabled agents to tackle complex embodied tasks through environmental interaction, but they still make suboptimal decisions and perform ineffective actions.
Approach:	They propose an active belief intervention mechanism that generates explicit belief states . they characterize belief inertia as a key failure mode of LLM-based agents .
Outcome:	The proposed method achieves significant gains in task success rates across embodied benchmarks.

Copied to clipboard

Henan Sun, Kaichi Yu, Yuyao Wang, Bowen Liu, Xunkai Li, Rong-Hua Li, Nuo Chen, Jia Li

A Learnable Skill Combination Strategy for Multi-task Learning in Natural Language Understanding (2026.findings-acl)

Challenge:	Existing benchmarks for algorithmic reasoning fail to answer a critical question: do LRMs master algorithmic thinking? Empirical evaluations on leading LRM models reveal substantial performance heterogeneity, while models perform well on non-optimized tasks, accuracy drops sharply to around 49% on globally optimized algorithms.
Approach:	They propose an algorithm-centric benchmark that evaluates large reasoning models under an algorithmic paradigm.
Outcome:	Empirical evaluations on leading LRMs reveal substantial performance heterogeneity . models perform well on non-optimized tasks, accuracy drops sharply to around 49% .

Copied to clipboard

Zhe Yang, Yi Huang, Yaqin Chen, Mengfei Guo, Xiaoting Wu, Junlan Feng

CoopValue: Revealing LLM Value Preferences Through Multi-Agent Cooperation (2026.findings-acl)

Challenge:	a novel multi-task learning framework for domain-specific natural language understanding tasks addresses these limitations by combing multiple tasks into a single framework.
Approach:	They propose a multi-task learning framework that decomposes the language model into modular skill components and employs a dynamic, learnable skill-combination mechanism to adaptively handle diverse tasks.
Outcome:	The proposed framework surpasses conventional multi-task learning approaches in performance.

Copied to clipboard

Zee Hen Tang, Mi-Yen Yeh

Challenge:	Existing evaluations of large language models rely on single-agent dilemmas or static binary-choice tasks, offering limited insight into how cooperation contexts influence LLM behavior.
Approach:	They propose a multi-agent evaluation framework that assesses LLMs’ value preferences through cooperative scenarios.
Outcome:	The proposed framework assesses LLMs’ value preferences through cooperative scenarios.

QUARTZ: Quantile-Aware Routing and Queueing for TTFT SLOs in LLM Serving (2026.findings-acl)

Copied to clipboard

Zhipeng Liu, Yifan Zheng, Fanqi Kong, Ziming Zhao

Challenge:	Prefill costs scale with prompt length and decode lengths are uncertain, and prefix locality creates strong performance skew across requests.
Approach:	They propose a quantile-aware routing and queueing layer that predicts conservative quantiles rather than point estimates using lightweight router-visible signals.
Outcome:	The proposed layer predicts conservative quantile-based request-cost proxies, rather than point estimates, using lightweight router-visible signals.

MPTc-Bench: Measuring Cross-market Generative Ability of Vision-Language Models via Movie Poster Transcreation (2026.findings-acl)

Copied to clipboard

Youyuan Lin, Yuan Li, Yahan Yu, Fei Cheng, Shin’ya Nishida, Chenhui Chu

Counterfactual Fairness Evaluation of LLM-Based Contact Center Agent Quality Assurance System (2026.findings-acl)

Challenge:	Recent work adapts textual transcreation to image editing and formulates image transcreations to better match a target audience while preserving meaning.
Approach:	They propose a two-stage planner-editor pipeline in which an VLM planner specifies executable edits and an image editor renders them.
Outcome:	The proposed model can transcreate a visual asset for a different market while preserving its identity while matching market-specific design preferences and multilingual typography.

Copied to clipboard

Kawin Mayilvaghanan, Siddhant Gupta, Ayush Kumar

Masked Text-to-Audio Flow-Matching and Reward Feedback Optimization (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are increasingly deployed in contact-center Quality Assurance (QA) relying on web-scale training data raises concerns regarding demographic and behavioral biases that may distort workforce assessment.
Approach:	They propose to evaluate the fairness of large language models on 3,000 real-world transcripts . they find systematic disparities in the CFR and MASD across different dimensions .
Outcome:	Evaluating 18 LLMs on 3,000 real-world contact center transcripts, they find systematic disparities . larger, more strongly aligned models show lower unfairness, though fairness does not track accuracy.

Copied to clipboard

Rongjie Huang, Dongchao Yang, Wenxiang Guo, Huadai Liu, Xize Cheng, Zehan Wang, Zhou Zhao, Xixin Wu, Helen M. Meng

Orthogonal Representation Editing: Decoupling Semantic Entanglement in Batch Knowledge Editing of LLMs (2026.findings-acl)

Challenge:	Experimental results show that Flow-matching generative models can scale training by increasing data, computational resources, and model size.
Approach:	They propose a flow-matching transformer with masked generative modeling for scaling text-to-audio inference-time prediction.
Outcome:	The proposed model scales inference-time computations by masking generation and re-predicting them through iterative decoding.

Copied to clipboard

Wenhao Yu, Zhicong Lu, Bo Lv, Fangyin Ma, Kaiwen Wei, Shihao Yang, Nayu Liu

Debiasing LLMs by Masking Unfairness-Driving Attention Heads (2026.findings-acl)

Challenge:	Existing knowledge editing methods suffer from performance degradation in batch knowledge editing.
Approach:	They propose an orthogonal representation editing method which decouples semantic entanglement from edit vectors and enforcing orthogonals on edit vector.
Outcome:	The proposed method outperforms existing methods and achieves superior performance in cross-lingual knowledge editing scenarios.

Copied to clipboard

Tingxu Han, Wei Song, Ziqi Ding, Ziming Li, Chunrong Fang, Yuekang Li, Dongfang Liu, Zhenyu Chen, Zhenting Wang

Single-Agent Generation Surpasses Multi-Agent Systems in Semantic Diversity (2026.findings-acl)

Challenge:	Existing work probes when biased outputs appear, but gives little insight into the mechanisms that generate them, leaving existing mitigations largely fragile.
Approach:	They propose a lightweight debiasing framework that detects bias heads and selectively masks only those heads that activate under DA and CoT.
Outcome:	The proposed framework reduces unfairness by 391.9%- 534.5% in both one- and two-turn dialogues.

Copied to clipboard

Cui Encheng, Shaowen Peng, Kazuhiro Ito, XU Jinsha, Hisada Shohei, Shoko Wakamiya, Eiji Aramaki

Med-SRAF: A Multi-Agent Framework for Medical Reasoning via Semantic Routing and Agentic Fusion (2026.findings-acl)

Challenge:	Multi-Agent Systems (MAS) are used to improve reasoning diversity and robustness by simulating interactions among agents with distinct roles.
Approach:	They find that a Multi-Output strategy produces the highest diversity without degrading logical validity.
Outcome:	The proposed approach outperforms multi-agent systems in semantic diversity . the results point to a more efficient and effective way to expand diversity - the authors say .

Copied to clipboard

Xiao Li, Zhuo Chen, Jun Xia, Hongxin Xiang, Chao Wang, Wenjie Du, Yang Wang

R-Search: Empowering LLM Reasoning with Search via Multi-Reward Reinforcement Learning (2026.findings-acl)

Challenge:	Existing RAG methods suffer from a two-part problem: semantic drift and concatenation fallacy . et al.: rapid development of Large Language Models has led to a paradigm shift in artificial intelligence .
Approach:	They propose a multi-agent retrieval augmentation framework guided by medical domain knowledge to address these challenges.
Outcome:	The proposed framework outperforms existing general RAG baselines on five widely used medical benchmarks.

Copied to clipboard

Qingfei Zhao, Ruobing Wang, Dingling Xu, Daren Zha, Ma Bowen, Zhichun Wang, Shijie Jia, Limin Liu, Xin Wang

Decoupled Reasoning with Implicit Fact Tokens (DRIFT): A Dual-Model Framework for Efficient Long-Context Inference (2026.findings-acl)

Challenge:	Large language models (LLMs) have demonstrated impressive capabilities in multi-step and long-chain reasoning, but extending their reasoning capabilities to encompass deep interactions with search remains a non-trivial challenge.
Approach:	They propose a framework for Reasoning–Search integration that integrates multi-reward signals to optimize the reasoning–search interaction trajectories.
Outcome:	Experiments on seven datasets show that R-Search significantly outperforms mainstream RAG baselines.

Copied to clipboard

Wenxuan Xie, Xuhong Wang

AnchorAlign: A Novel Anchor Alignment-enhanced Generative Method for Joint Named Entity Recognition and Relation Extraction (2026.findings-acl)

Challenge:	Existing solutions to integrate extensive, dynamic knowledge into Large Language Models (LLMs) are constrained by finite context windows, retriever noise, or the risk of catastrophic forgetting.
Approach:	They propose a dual-model architecture that explicitly decouples knowledge extraction from the reasoning process by compressing document chunks into implicit fact tokens conditioned on the query.
Outcome:	The proposed architecture significantly outperforms strong baselines among comparably sized models on long-context tasks while maintaining inference accuracy.

Copied to clipboard

Xiaolong Weng, Yuanyun Zhou, Boyu Qiu, Zehua Wang, Ying Xiong, Buzhou Tang

Why Can Distillation Work with Limited Resources? A Systematic Study (2026.findings-acl)

Challenge:	Named Entity Recognition and Relation Extraction are interdependent tasks in information extraction.
Approach:	They propose a generative method enhanced by anchor alignment to bridge NER and RE tasks . they use anchor entities as semantic pivots to align the two tasks based on their semantic representations .
Outcome:	The proposed method outperforms state-of-the-art models on five benchmark datasets.

Copied to clipboard

Xiao Hu, Xingyu Lu, Liyuan Mao, YiFan Zhang, Tianke Zhang, Bin Wen, Fan Yang, Tingting Gao, Guorui Zhou

Autoregressive Semantic Visual Reconstruction Helps VLMs Understand Better (2026.findings-acl)

Challenge:	Recent advances in large language models have driven reasoning performance . low-resource distillation can boost models' performance, but a framework is missing .
Approach:	They conduct a controlled experiment to find out why low-resource distillation can boost model performance . they find that distillation enhances the presence of advanced cognitive behaviors .
Outcome:	The proposed model shows more flexible reasoning, the authors show . they show that distillation enhances the presence of advanced cognitive behaviors .

Copied to clipboard

Dianyi Wang, Wei Song, Yikun Wang, Siyuan Wang, Kaicheng Yu, Zhongyu Wei, Jiaqi Wang

MSCode: Advancing Human Motion-Language Understanding via Modality-Shared Codebook (2026.findings-acl)

Challenge:	Typical large vision-language models emphasize vision-to-language alignment while overlooking fine-grained visual information.
Approach:	They introduce autoregressive semantic visual reconstruction (ASVR) that enables joint learning of visual and textual modalities within a unified autoregression framework.
Outcome:	The proposed model improves baselines and multimodal understanding benchmarks by 2-3%.

Copied to clipboard

Haoyu Shi, Huaiwen Zhang

RAG over Tables: Hierarchical Memory Index, Multi-Stage Retrieval, and Benchmarking (2026.findings-acl)

Challenge:	Existing methods for motion understanding lack precise alignment between motion and modalities . existing methods lack precise semantics and a mismatch between motion, text .
Approach:	They propose a modality-shared codebook that enables unified representation learning and precise alignment between motion and linguistic modalities.
Outcome:	The proposed model surpasses current state-of-the-art methods in many areas . it enables unified representation learning and precise alignment of motion and modalities .

Copied to clipboard

Jiaru Zou, Dongqi Fu, Sirui Chen, Xinrui He, Zihao Li, Yada Zhu, Jiawei Han, Jingrui He

CoTEvol: Self-Evolving Chain-of-Thoughts for Data Synthesis in Mathematical Reasoning (2026.findings-acl)

Challenge:	Retrieval-Augmented Generation (RAG) integrates knowledge from tables with an external knowledge base to improve the answer relevance and accuracy.
Approach:	They propose a table-corpora-aware RAG framework called T-RAG to integrate external knowledge into Large Language Models (LLMs) they then develop a multi-table question answering benchmark called MultiTableQA which spans 3 different task types, 57,193 tables, and 23,758 questions in total.
Outcome:	The proposed framework achieves state-of-the-art accuracy, recall, and runtime performance, with improvements of up to 9.4%.

Copied to clipboard

Zhuo Wang, Zhuo Zhang, Yafu Li, Yu Cheng, Lizhen Qu, Zenglin Xu

Feedback Is The Key for Automated Survey Generation (2026.findings-acl)

Challenge:	Existing methods to improve LLMs' reasoning abilities suffer from diminishing returns or high computing overhead.
Approach:	They propose a genetic evolutionary framework that casts CoT generation as a population-based search over reasoning trajectories.
Outcome:	The proposed framework improves correct-CoT synthesis success by over 30% and enhances structural diversity with markedly improved efficiency.

Copied to clipboard

Tianyi Xu, Zhe Zhao, Tianshuo Wei, Yiqun Kou, Liuliu Han, Ye Wei

Challenge:	Large Language Models (LLMs) provide a promising foundation for literature surveys, but guiding them to generate accurate, reliable content remains a fundamental challenge.
Approach:	They propose a feedback-driven framework that incorporates feedback across three dimensions: outline feedback for structural clarity, citation feedback for evidence validation, and content feedback for readability and analytical depth.
Outcome:	The proposed framework significantly improves both citation and content quality, demonstrating feedback as the critical mechanism for automatic survey generation.

Sentient Agent as a Judge: Evaluating Higher-Order Social Cognition in Large Language Models (2026.findings-acl)

Copied to clipboard

Bang Zhang, Ruotian Ma, Qingxuan Jiang, Peisong Wang, Jiaqi Chen, Zheng Xie, Xingyu Chen, Yue Wang, Fanghua Ye, Jian Li, Yifan Yang, Zhaopeng Tu, Xiaolong Li

Alignment Data Map for Efficient Preference Data Selection and Diagnosis (2026.findings-acl)

Challenge:	Large language models (LLMs) have evolved from statistical sequence predictors to sophisticated autonomous agents capable of reasoning, planning, and sustaining multi-turn conversa-tions.
Approach:	They propose a system that instantiates a "Sentient Agent" that simulates human-like emotional changes and inner thoughts to provide a more realistic evaluation of the model in multi-turn conversations.
Outcome:	The proposed framework measures the agent's higher-order social cognition in multi-turn conversations.

Copied to clipboard

Seohyeong Lee, Eunwon Kim, Hwaran Lee, Buru Chang

Fine-Grained Detection of Context-Grounded Hallucinations Using LLMs (2026.findings-acl)

Challenge:	constructing high-quality preference datasets faces scalability challenges due to prohibitive cost and complexity of human annotation.
Approach:	They propose a tool to identify and select effective preference data by LLM-as-a-judge, explicit reward model, and reference-based approaches.
Outcome:	The proposed tool reduces annotation costs while preserving alignment effectiveness.

Copied to clipboard

Yehonatan Peisakhovsky, Zorik Gekhman, Yosi Mass, Liat Ein-Dor, Roi Reichart

When Reviews Disagree: Fine-Grained Contradiction Analysis in Scientific Peer Reviews (2026.findings-acl)

Challenge:	Existing representations of hallucinations limit the types of errors that can be expressed, so we propose a new representation based on free-form textual descriptions, capturing the full range of possible errors.
Approach:	They propose a benchmark for localizing hallucinations using LLMs with a human annotation of over 1,000 examples and a protocol to verify its quality in a humans evaluation.
Outcome:	The proposed representation captures the full range of possible errors, and the best model achieves an F1 score of 0.67.

Copied to clipboard

Sandeep Kumar, Yash Kamdar, Abid Hossain, Bharti Kumari, Tanik Saikh, Asif Ekbal

One Script Instead of Hundreds? On Pretraining Romanized Encoder Language Models (2026.findings-acl)

Challenge:	Existing approaches frame reviewer disagreement as binary contradiction detection over isolated sentence pairs, abstracting away review context and obscuring differences in severity of evaluative conflict.
Approach:	They propose a fine-grained formulation of reviewer contradiction analysis that operates over full peer reviews by explicitly identifying contradiction evidence spans and assigning graded disagreement intensity scores.
Outcome:	The proposed framework outperforms strong single-agent and generic multi-agend baselines in evidence identification and intensity agreement.

Copied to clipboard

Benedikt Ebing, Lennart Keller, Goran Glavaš

Relaxing the Constraints: A Dual-Importance Projection Mechanism for Lifelong Model Editing (2026.findings-acl)

Challenge:	a recent study has focused on setups that favor romanization for cross-lingual transfer . a fidelity-based approach is needed to improve performance for high-resource languages .
Approach:	They propose to pretrain LMs from scratch on romanized and original texts for six languages . they find that romanization improves encoding efficiency for segmental scripts at a negligible cost .
Outcome:	The proposed method reduces the loss of script-specific information and dilution of language-specific representations from increased subword overlap.

Copied to clipboard

Zhenghai Chen, Senbin Xu, Jiaxi Tan, Xinhua Wu, Yan Zhang, Xiawu Zheng, Shengchuan Zhang, Ke Li, Sicheng Zhao, Liujuan Cao, Rongrong Ji

Not all ANIMALs are equal: metaphorical framing through source domains and semantic frames (2026.findings-acl)

Challenge:	Existing knowledge editing methods rely on strict orthogonal projection to preserve previously edited knowledge, but this constraint limits gradient expressiveness, resulting in degradation of model generalization and overall performance as the number of edits increases.
Approach:	They propose a method that leverages Singular Value Decomposition to identify critical gradient subspaces and introduces a dual mechanism comprising "accumulated importance" and "projection importance"
Outcome:	Extensive experiments on five mainstream LLMs show that the proposed method achieves an average comprehensive performance improvement of 10.36% and effectively maintains the model’s general capabilities on downstream tasks.

Copied to clipboard

Yulia Otmakhova, Matteo Guida, Lea Frermann

Challenge:	a computational framework allows to derive discourse metaphors through their source domains and semantic frames.
Approach:	They propose a computational framework that allows to derive salient discourse metaphors through their source domains and semantic frames.
Outcome:	The proposed framework uncovers well-known source domains and reveals nuanced frame-level associations that distinguish how the issue is portrayed.

DualFact+: A Multimodal Fact Verification Framework for Procedural Video Captioning (2026.findings-acl)

Copied to clipboard

Cennet Oguz, Yasser Hamidullah, Josef Van Genabith, Simon Ostermann

From Scaffolding to Assimilation: Progressive Structural Internalization for Format-Constrained Creative Text Generation (2026.findings-acl)

Challenge:	Existing evaluation metrics fail to evaluate factual correctness in procedural video captions . Existing metrics rely on lexical overlap or holistic semantic similarity, but miss role-specific omissions resulting in hallucinations .
Approach:	They propose a role-aware, fact-level evaluation framework that distinguishes conceptual facts from contextual facts.
Outcome:	Experiments show that state-of-the-art captioning models produce fluent but incomplete descriptions with systematic errors.

Copied to clipboard

Wenhao Li, Yuwei Yang, Xiaoqing Wu, Yufeng Han, Cunliang Kong, Yuzhuo Bai, Xin Cong, Maosong Sun

Hierarchical Visual Agent: Managing Contexts in Joint Image-Text Space for Advanced Chart Reasoning (2026.findings-acl)

Challenge:	Existing paradigms rely on unreliable prompting or rigid constrained decoding strategies to achieve aesthetic unity.
Approach:	They propose a framework to embed external constraints into the model’s intrinsic intuition and use it to generate open-ended creative texts.
Outcome:	The proposed framework surpasses baselines in both strict constraint adherence and literary aesthetics.

Copied to clipboard

Qihua Dong, Ruozhen He, Junwen Chen, Yizhou Wang, Xu Ma, Songyao Jiang, Yun Fu

Semantic-Space Exploration and Exploitation in RLVR for LLM Reasoning (2026.findings-acl)

Challenge:	Existing MLLMs are strong at understanding single plots, but struggle with multi-step reasoning . Existing approaches to manage context in chart reasoning include text-based chain-of-thought prompting .
Approach:	They propose a hierarchical visual agent framework that iteratively constructs a working context in an image–text space.
Outcome:	The proposed framework improves on strong multimodal baselines.

Copied to clipboard

Fanding Huang, Guanbo Huang, Xiao Fan, Yi He, Xiao Liang, Xiao Chen, Qinting Jiang, Faisal Nadeem Khan, Jingyan Jiang, Zhi Wang

Towards Efficient Large Language Model Serving: A Survey on System-Aware KV Cache Optimization (2026.findings-acl)

Challenge:	Recent advances in Reinforcement Learning with Verifiable Rewards (RLVR) have substantially improved the reasoning abilities of Large Language Models (LLMs).
Approach:	They propose a method that balances exploration and exploitation in the hidden-state space of response trajectories.
Outcome:	The proposed model yields consistent improvements across models, algorithms and reasoning benchmarks.

Copied to clipboard

Jiantong Jiang, Peiyu Yang, Rui Zhang, Feng Liu

A Counterfactual Explanation Framework for Retrieval Models (2026.findings-acl)

Challenge:	Large language models (LLMs) are a powerful tool for high-performance inference serving.
Approach:	They focus on system-aware KV infrastructure for serving LLMs . they analyze cross-behavior co-design affinity and behavior-objective links .
Outcome:	The proposed key-value (KV) cache is crucial for low-latency, high-throughput LLM inference serving.

Copied to clipboard

Bhavik Chandna, Procheta Sen

Teaching Language Models to Forecast Research Success Through Comparative Idea Evaluation (2026.findings-acl)

Challenge:	Existing literature on explainability of information retrieval has focused on illustrating the concept of relevance concerning a retrieval model.
Approach:	They propose to add terms to a document to improve its ranking to answer the question of which words played a role in not being favored by a retrieval model.
Outcome:	The proposed framework predicts counterfactuals for statistical and deep-learning models.

Copied to clipboard

Srujan P Mule, Aniketh Garikaparthi, Manasi Patwardhan

Social Welfare Function Leaderboard: On the Emergence of LLM Agents as the Welfare Dictator (2026.findings-acl)

Challenge:	Language models are accelerating scientific research by automating hypothesis generation and implementation.
Approach:	They ask whether LMs can forecast the empirical success of research ideas before experiments . they frame evaluation as a reasoning task via Reinforcement Learning with Verifiable Rewards .
Outcome:	The proposed model outperforms off-the-shelf models in 77.1% of the evaluations . the model outpersforms GPT-5 in the evaluation of 11,488 idea pairs .

Copied to clipboard

Zhengliang Shi, Ruotian Ma, Jen-tse Huang, Xinbei Ma, Xingyu Chen, Mengru Wang, Qu Yang, Yue Wang, Fanghua Ye, Ziyang Chen, Shanyi Wang, Cixing LI, Wenxuan Wang, Zhaopeng Tu, Xiaolong Li, Zhaochun Ren, Liefeng Bo

From Individual Excellence to Collective Sustainability: Seeking Strategic Equilibrium in Proactive Multi-Agent Teams (2026.findings-acl)

Challenge:	Large language models (LLMs) are increasingly entrusted with high-stakes decisions that affect human welfare.
Approach:	They evaluate 20 state-of-the-art Large language models (LLMs) and 20 LLM dictators to create a social welfare function benchmark.
Outcome:	The proposed model creates dilemma between maximizing collective efficiency and ensuring distributive fairness.

Copied to clipboard

Tong Zhang, Yang Wu, Yufei Shi, Rujing Yao, Zhuoren Jiang, Xiaozhong Liu

Lost in the Prompt Order: Revealing the Limitations of Causal Attention in Language Models (2026.findings-acl)

Challenge:	a team of proactive agents suffer from a greedy optimization for immediate task accuracy . a new approach to improve team collaboration is based on the opportunity cost .
Approach:	They propose a game-theoretic proactive multi-agent reinforcement learning framework to solve this imbalance . they use a Positive-Unlabeled scorer to anchor intervention quality under sparse supervision .
Outcome:	The proposed framework maintains high performance while preventing experts from over-developing.

Copied to clipboard

Hyunjong Ok, Jaeho Lee

SR-RAG: Verifiable Multi-Hop Reasoning via On-the-fly Symbolic Graph Construction (2026.findings-acl)

Challenge:	Large language models exhibit surprising sensitivity to structure of the prompt, but mechanisms underlying this sensitivity remain poorly understood.
Approach:	They conduct an in-depth investigation on placing context before the questions and options in MCQA prompts.
Outcome:	The proposed model outperforms the reverse order (QOC) by over 14%p over a wide range of models and datasets.

Copied to clipboard

Zehua Wang, Zhaojin Zhang, Boyu Qiu, Xiaolong Weng, Ying Xiong, Buzhou Tang, Min Zhang

KnowDR-REC: Auditing Knowledge-Conditioned Visual Grounding in Referring Expression Comprehension (2026.findings-acl)

Challenge:	Existing paradigms for multi-hop reasoning suffer from high construction costs and limited adaptability to dynamic knowledge.
Approach:	They propose a symbolic reasoning framework for multi-hop question answering that integrates the advantages of both paradigms by dynamically generating sub-questions, performing information retrieval and symbolic encoding based on an on-the-fly graph and using a symbol verifier to validate intermediate reasoning steps.
Outcome:	The proposed framework significantly improves accuracy and robustness on multiple multi-hop benchmarks and a medical dataset.

Copied to clipboard

Guanghao Jin, Jingpei Wu, Tianpei Guo, Yiyi Niu, Weidong Zhou, Linyi Yang, Guoyang Liu

Agent for Numerical Data Retrieval and Understanding by Code Generation and Multimodal Reasoning (2026.findings-acl)

Challenge:	Existing evaluation metrics suggest that Multimodal large language models have acquired fine-grained visual grounding capabilities.
Approach:	They propose a benchmark to assess Referring Expression Comprehension (REC) that uses intra-image visual cues to localize target objects and a controllable evaluation mechanism to test sensitivity to fine-grained factual changes.
Outcome:	The proposed benchmarks show that multimodal large language models have a high level of performance on the RefCOCO family of benchmarks.

Copied to clipboard

Florian Baud, Feda Almuhisen, Dorian Midou

Cause-CSD: A Challenge Multimodal Conversational Stance Cause Detection Dataset and Effective Method (2026.findings-acl)

Challenge:	Numerical data from sensors and time series are widely used in scientific research fields such as nuclear fusion . efficient numerical data analysis tools are crucial to accelerate experimental research .
Approach:	They propose a model-agnostic and data-adic agent that processes numerical data by code generation and multimodal reasoning.
Outcome:	The proposed agent outperforms existing methods on benchmarks on sensor data classification and time series understanding.

Copied to clipboard

Fuqiang Niu, Bowen Zhang, Junting Zhu, Qing Liao, Genan Dai, Hu Huang

The Sparse Frontier: Sparse Attention Trade-offs in Transformer LLMs (2026.findings-acl)

Challenge:	Existing stance detection methods treat opinions as surface-level labels, overlooking conversational evidence behind stance expressions.
Approach:	They propose a task that jointly identifies stance polarity and contextual evidence . they propose stance-cause Detection language model that leverages explicit context reasoning .
Outcome:	The proposed task outperforms baseline methods on text-only and multimodal subtasks.

Copied to clipboard

Piotr Nawrot, Jianing Li, Renjie Huang, Sebastian Ruder, Kelly Marchisio, Edoardo Ponti

Uncertainty-Aware Contrastive Sentence Embedding With Local Context Representation for Text Classification (2026.findings-acl)

Challenge:	Sparse attention is a promising strategy to extend long-context capabilities in LLMs . but its efficiency–accuracy trade-offs remain unclear due to the lack of comprehensive evaluation .
Approach:	They evaluate sparse attention methods across multiple model families and sizes . they find larger sparser models outperform smaller dense ones at equivalent cost .
Outcome:	The proposed methods outperform smaller sparse models at equivalent cost and improve the Pareto frontier.

Copied to clipboard

Han Liu, Jiaqing Zhan, Zhichao Chen, Qin Zhang

ZipVoice-Dialog: Non-Autoregressive Spoken Dialogue Generation with Flow Matching (2026.findings-acl)

Challenge:	Existing models for text classification are based on encoder-only transformers and generative pre-trained transformers.
Approach:	They propose an uncertainty-aware contrastive sentence embedding approach that addresses language ambiguity and inter-class separability for a text classification task.
Outcome:	The proposed approach improves classification accuracy on public datasets compared with state-of-the-art methods.

Copied to clipboard

Han Zhu, Wei Kang, Liyong Guo, Zengwei Yao, Fangjun Kuang, Weiji Zhuang, Zhaoqing Li, Zhifeng Han, Dong Zhang, Xin Zhang, Xingchen Song, Lingxuan Ye, Long Lin, Daniel Povey

All Prompts Are Created Equal? Evaluating Robustness of LLM Judges Against Non-Adversarial Prompt Variations (2026.findings-acl)

Challenge:	Existing autoregressive models for dialogue generation suffer from high latency and stability issues.
Approach:	They propose a non-autoregressive (NAR) zero-shot spoken dialogue generation model based on flow-matching.
Outcome:	The proposed model outperforms existing models in speech generation due to poor speech intelligibility and turn-taking precision.

Copied to clipboard

Savita Bhat, Vasudeva Varma

Thinking Traps in Long Chain-of-Thought: A Measurable Study and Trap-Aware Adaptive Restart (2026.findings-acl)

Challenge:	Our work establishes new standards for assessing LLM judge reliability beyond simple accuracy metrics.
Approach:	They propose a diagnostic framework grounded in attribute verifiability that enables principled decisions about evaluation automation.
Outcome:	The proposed framework establishes new standards for assessing LLM judge reliability beyond simple accuracy metrics.

Copied to clipboard

null Chenkang, Fan Yu, Junjie Nian, Sihan Zhao, Zhuoka Feng, Zijun Yao, Wang Heng, Yu Minshen, Yixin Cao

Instruction-Guided Poetry Generation in Arabic and Its Dialects (2026.findings-acl)

Challenge:	Experiments show that extended generation does not guarantee correctness . a recurring pattern in Long-CoT failures is a problem for large reasoning models .
Approach:	They propose a test-time control framework that truncates the trajectory before the trap segment and adaptively restarts decoding.
Outcome:	Experiments show that TAAR improves reasoning performance without fine-tuning model parameters.

Copied to clipboard

Abdelrahman Sadallah, Kareem Elozeiri, Mervat Abassy, Rania Elbadry, Mohamed Anwar, Abed Alhakim Freihat, Preslav Nakov, Fajri Koto

Agentic Verification for Ambiguous Query Disambiguation (2026.findings-acl)

Challenge:	Existing literature on Arabic poetry has focused on analysis tasks such as interpretation or metadata prediction, e.g., rhyme schemes and titles.
Approach:	They propose to use a large-scale instruction-based dataset to generate Arabic poetry based on predefined criteria such as style and rhyme .
Outcome:	The proposed model can generate poetry that is aligned with user requirements, based on automated metrics and human evaluation with native Arabic speakers.

Copied to clipboard

Youngwon Lee, Seung-won Hwang, Ruofan Wu, Feng Yan, Danmei Xu, Moutasem Akkad, Zhewei Yao, Yuxiong He

DICA: Dual-Indicator Guided Contrastive Alignment in Multimodal Large Language Models (2026.findings-acl)

Challenge:	Prior Diversify-then-Verify pipelines generate interpretations and then retrieve evidence . ambiguous queries require RAG to disambiguate into interpretations that can be answered from corpus .
Approach:	They propose a novel approach that unifies diversification with verification by integrating retriever relevance and generator answerability feedback early.
Outcome:	The proposed approach improves grounding-aware F1 by 23% over baselines across multiple LLMs.

Copied to clipboard

Hao Yang, Jin Wang, Xuejie Zhang

From Tasks to Teams: A Risk-First Evaluation Framework for Multi-Agent LLM Systems in Finance (2026.findings-acl)

Challenge:	Multimodal large language models may deviate from this pattern due to attention drift and underutilization of visual evidence.
Approach:	They propose a Dual-Indicator Guided Contrastive Alignment (DICA) that tracks visual attention and output image correlations to improve visual grounding.
Outcome:	The proposed model outperforms existing approaches and significantly reduces hallucinations.

Copied to clipboard

Zichen Chen, Jianda Chen, Jiaao Chen, Misha Sra

Challenge:	Existing benchmarks focus on task specific metrics such as accuracy, F1 score, or ROUGE.
Approach:	They propose a multi-agent, safety-aware evaluation agent that audits large language models without fine-tuning.
Outcome:	M-SAEA identifies unsafe trajectories with minimal false positives and reveals latent risks that are not addressed by standard metrics.

Parametric Knowledge is Not All You Need: Toward Honest Large Language Models via Retrieval of Pretraining Data (2026.findings-acl)

Copied to clipboard

Christopher Adrian Kusuma, Muhammad Reza Qorib, Hwee Tou Ng

TIME: Temporally Intelligent Meta-reasoning Engine for Context-Triggered Explicit Reasoning (2026.findings-acl)

Challenge:	Large language models are highly capable of answering questions, but they are often unaware of their own knowledge boundary, i.e., knowing what they know and what they don’t know.
Approach:	They propose a method to evaluate LLM honesty using Pythia with publicly available pretraining data.
Outcome:	The proposed method is based on Pythia, a truly open LLM with publicly available pretraining data.

Copied to clipboard

Susmit Das

Omni-R1: Towards the Unified Generative Paradigm for Multimodal Reasoning (2026.findings-acl)

Challenge:	Reasoning-oriented language models expose explicit reasoning as a long, front-loaded chain of “thinking” tokens before the main output, either always enabled or externally toggled at inference time.
Approach:	They introduce a behavioral alignment framework that learns explicit reasoning as a context-triggered control policy rather than a fixed response mode.
Outcome:	The proposed framework improves TIMEBench scores over the base model in thinking and no-thinking modes while keeping output compact.

Copied to clipboard

Dongjie Cheng, Yongqi Li, Zhixin Ma, Hongru Cai, Yupeng Hu, Wenjie Wang, Liqiang Nie, Wenjie Li

Time-for-Accuracy: Formalizing Chain-of-Thought as an Expansion of Logical Depth (2026.findings-acl)

Challenge:	Early approaches focus on text-based reasoning, but they often follow a single task-specific reasoning pattern.
Approach:	They propose a generative multimodal reasoning paradigm that unifies diverse reasoning skills by generating intermediate images during the reasoning process.
Outcome:	The proposed model unifies diverse multimodal reasoning skills by generating intermediate images during the reasoning process.

Copied to clipboard

Yue Wang, Zhi Zhang, Wang Xi, Chengjie Sun, Lili Shan, Bingquan Liu

Challenge:	Chain-of-thought (CoT) prompting can improve multi-step reasoning, but it is unclear what kind of additional sequential computation longer traces actually enable.
Approach:	They propose a deletion-based measure of step necessity under a specified inference interface to operationalize realized depth beyond raw length.
Outcome:	The proposed method combines effective logical depth with Bennett's logical depth to show that it is more efficient than a linear model.

DiffCoT: Diffusion-styled Chain-of-Thought Reasoning in LLMs (2026.findings-acl)

Copied to clipboard

Shidong Cao, Hongzhan Lin, Yuxuan Gu, Ziyang Luo, Jing Ma

System-Mediated Attention Imbalances Make Vision-Language Models Say Yes (2026.findings-acl)

Challenge:	Chain-of-Thought (CoT) reasoning improves multi-step mathematical problem solving in large language models but is vulnerable to exposure bias and error accumulation.
Approach:	They propose a diffusion-styled CoT framework that reformulates CoT reasoning as an iterative denoising process.
Outcome:	The proposed framework outperforms existing methods on three multi-step CoT reasoning benchmarks.

Copied to clipboard

Tsan Tsai Chan, Varsha Suresh, Anisha Saha, Michael Hahn, Vera Demberg

SCURank: Ranking Multiple Candidate Summaries with Summary Content Units for Enhanced Summarization (2026.findings-acl)

Challenge:	Existing mitigation strategies tend towards an image-centric interpretation of these imbalances, prioritising increased image attention while giving less consideration to the roles of the other modalities.
Approach:	They propose a more holistic, system-mediated account which attributes imbalances to functionally redundant system weights that reduce attention to image and textual inputs.
Outcome:	The proposed framework offers a useful empirical perspective on the yes-bias, a common form of hallucination in which VLMs indiscriminately respond ‘yes’.

Copied to clipboard

Bo-Jyun Wang, Ying-Jia Lin, Hung-Yu Kao

Valid Necessary: Diagnosing Latent Inefficiency in Chain-of-Thought (2026.findings-acl)

Challenge:	Existing ranking strategies for large language models suffer from instability and lack of information content.
Approach:	They propose a framework that enhances summarization by leveraging Summary Content Units (SCUs) they investigate the effectiveness of SCURank in distilling summaries from multiple LLMs .
Outcome:	The proposed framework outperforms traditional metrics and LLM-based ranking methods in summarization tasks.

Copied to clipboard

Daeyeop Lee, Hwanjo Yu

CausalGaze: Unveiling Hallucinations via Counterfactual Graph Intervention in Large Language Models (2026.findings-acl)

Challenge:	Existing reasoning step evaluators fail to distinguish “valid but inefficient” reasoning steps from necessary reasoning.
Approach:	They propose a training-free metric that identifies low-utility steps and a post-hoc compression strategy to quantify their impact on token usage.
Outcome:	The proposed metric reduces token consumption by 31–53% while maintaining accuracy at substantially higher compression rates.

Copied to clipboard

Linggang Kong, Lei Wu, Yunlong Zhang, Xiaofeng Zhong, Wang Zhen, Yongjie Wang, Yao Pan

Improved Policy Optimization for Mixture-of-Experts Models: Importance Sampling and Rewarding from an Expert-Centric Perspective (2026.findings-acl)

Challenge:	Existing classification-based methods capture noise and spurious correlations while overlooking the underlying causal mechanisms.
Approach:	They propose a hallucination detection framework based on structural causal models that captures static and passive signals from internal states and employs counterfactual interventions to disentangle causal reasoning paths from incidental noise.
Outcome:	Experiments on four datasets and three widely used LLMs show that the proposed framework improves AUROC and interpretability.

Copied to clipboard

Yining Qian, Jinpeng Li, Fei Mi, Lifeng Shang, Xiang Zhang

Challenge:	Existing approaches to reinforcement learning (RL) suffer from training instability . existing approaches often ignore token-specific discrepancies in expert assignments .
Approach:	They propose to introduce expert-level importance sampling to reduce complexity of RL . they propose to leverage expert-centric granularity to ensure a rigorous alignment between reward signals and policy updates.
Outcome:	The proposed method outperforms strong baselines across reasoning tasks.

From What Is Said to Why It Is Framed: Intent-Aware News Video Understanding (2026.findings-acl)

Copied to clipboard

Xiangzheng Kong, Minnan Luo, Wenya Wang, Jiaying Wu, Zhi Zeng, Guang Dai

H-MAS: Hierarchical Multi-Agent Scheduling for Multi-Tenant LLM Serving (2026.findings-acl)

Challenge:	Existing verification methods for short-form news videos neglect communicative intent . stylistic presentation and factual manipulation are often intertwined, resulting in shortcut learning .
Approach:	They propose a theory-grounded representation of communicative intent that captures creator stance, audience need activation, and communication strategy.
Outcome:	The proposed framework captures creator stance, audience need activation, and communication strategy.

Copied to clipboard

Yuhan Liu, Cong Xu, Qi Jia, Yihua Wang, Feiyu Chen, Liang Jin, Lu Liu, Yaqian Zhao, Yuting Ding, Xiang Li

Challenge:	Multi-tenant Model-as-a-Service (MaaS) workloads exhibit non-stationarity across multiple time scales . existing request schedulers often rely on a fixed policy that remains unchanged at runtime .
Approach:	They propose a hierarchical multi-agent scheduler that operates in a layered closed loop . they propose to maintain 1.2–3.0 higher Goodput than SGLang and vLLM .
Outcome:	Experiments show that H-MAS achieves 1.2–3.0 higher Goodput than SGLang and vLLM . it maintains more stable QoS under diverse request lengths and heterogeneous SLO targets .

EvoMD-LLM: Learning the Language of Species Evolution in Reactive Molecular Dynamics (2026.findings-acl)

Copied to clipboard

Zhichen Tang, Zhengzheng Dang, Yulin Chen, Jixin Wu, Haiwen Li, Yanming Wang

Do Image–Text Metrics Respect Semantic Invariances? (2026.findings-acl)

Challenge:	Existing models operate on static molecular representations or rely on external tools for reasoning.
Approach:	They propose a framework that reformulates species-level molecular dynamics as a symbolic temporal language modeling problem.
Outcome:	The proposed model outperforms neural networks and language-based baselines on multiple temporal prediction tasks and generates plausible interpretations of reaction dynamics.

Copied to clipboard

Amit Agarwal, Hitesh Laxmichand Patel, Meizhu Liu, Jyotika Singh, Karan Dua, Hansa Meghwani, Matthew Rowe, M. Avendi, Yassi Abbasi, Tao Sheng, Sujith Ravi, Dan Roth

Reasoning in the Dark: Interleaved Vision-Text Reasoning in Latent Space (2026.findings-acl)

Challenge:	Reference-free image–to–text evaluators are now standard for scoring image–caption alignment, yet it is unclear whether they respect semantic invariances.
Approach:	They propose an invariance probe on five popular evaluators under semantics-preserving perturbations along three axes: spatial edits, object changes, and socio-linguistic framing.
Outcome:	The proposed invariance probe shows that spatial edits and simple phrasing changes shift scores by ()6% on average and cause ranking flips in up to (),37% of cases.

Copied to clipboard

Chao Chen, Zhixin Ma, Yongqi Li, Yupeng Hu, Yinwei Wei, Wenjie Li, Liqiang Nie

Hail to the Thief: Exploring Attacks and Defenses in Decentralised GRPO (2026.findings-acl)

Challenge:	Existing multimodal reasoning methods depend on explicit reasoning steps that require labor-intensive vision-text annotations and inherently introduce significant inference latency.
Approach:	They propose a method that integrates visual and visual information into the reasoning process to improve the performance of multimodal LLMs.
Outcome:	The proposed method achieves an average performance increase of 5.45% while achieving a speed increase of over 5 times compared to existing methods.

Copied to clipboard

Nikolay Blagoev, Oguzhan Ersoy, Lydia Chen

SocraticKG: Knowledge Graph Construction via QA-Driven Fact Extraction (2026.findings-acl)

Challenge:	Recent years have seen great interest in Reinforcement Learning (RL) for the purposes of posttraining of Large Language Models (LLMs).
Approach:	They propose two defense mechanisms that check logit probabilities of completions or utilize an LLM judge to filter completions.
Outcome:	The proposed attacks can achieve attacks success rates of up to 100% in as few as 50 iterations.

Copied to clipboard

Sanghyeok Choi, Woosang Jeon, Kyuseok Yang, Taehyeong Kim

GROLE: Instance-Level Group Relative Optimization for LoRA Experts in Incremental Learning (2026.findings-acl)

Challenge:	Existing approaches to construct knowledge graphs struggle with factual coverage and information loss.
Approach:	They propose an automated KG construction method that introduces question-answer pairs as a structured intermediate representation to unfold document-level semantics prior to triple extraction.
Outcome:	The proposed method achieves superior factual retention while maintaining high structural cohesion even as extracted knowledge volume substantially expands.

Copied to clipboard

Yongyi Liao, Wencan Lai, Jun Fang, Jinjin Guo, Xiaohui Zhang, Zhiyuan Liu, Chao Liu, Pengzhang Liu, Qixia Jiang

Challenge:	Large language models demonstrate remarkable zero-shot generalization, but adapting to downstream tasks requires continual fine-tuning.
Approach:	They propose a method that incrementally constructs a pool of frozen, task-specific LoRA experts.
Outcome:	The proposed approach outperforms state-of-the-art methods in task-free and blurred-boundary settings.

SEARCH-R: Structured Entity-Aware Retrieval with Chain-of-Reasoning Navigator for Multi-hop Question Answering (2026.findings-acl)

Copied to clipboard

FU Yuqing, Yimin Deng, Wanyu Wang, Yuhao Wang, Yejing Wang, Hongshi Liu, Yiqi Wang, Xiao Han, Maolin Wang, Guoshuai Zhao, Yi Chang, Xiangyu Zhao

Why Did Apple Fall: Evaluating Curiosity in Large Language Models (2026.findings-acl)

Challenge:	Existing approaches to multi-hop question answering lack effective control over reasoning paths, leading to astray results.
Approach:	They propose a framework for multi-hop question answering that trains an end-to-end reasoning path navigator to provide a powerful sub-question decomposer by fine-tuning the Llama3.1-8B model.
Outcome:	The proposed framework trains an end-to-end reasoning path navigator . it is able to provide a powerful sub-question decomposer by fine-tuning the Llama3.1-8B model .

Copied to clipboard

Haoyu Wang, Sihang Jiang, Yuyan Chen, Yitong Wang, Xiaojun Meng, Jiansheng Wei, Yanghua Xiao

Gaperon: A Peppered English-French Generative Language Model Suite (2026.findings-acl)

Challenge:	Existing methods for evaluating curiosity-like behaviors in large language models lack curiosity-inspired features.
Approach:	They propose a psychology-inspired framework to evaluate curiosity in large language models . they adapt the Five-Dimensional Curiosity scale Revised (5DCR) to LLMs .
Outcome:	The proposed framework evaluates curiosity in large language models using questionnaires and behavioral studies.

Copied to clipboard

Nathan Godey, Wissam Antoun, Rian Touchent, Rachel Bawden, Éric Villemonte de la Clergerie, Benoît Sagot, Djamé Seddah

Rhombus: Incentivizing Coordination in Parallel Thinking through Reinforcement Learning (2026.findings-acl)

Challenge:	Standardized benchmarks have become the dominant metric for measuring progress in large language models, but their validity is compromised by data contamination and unclear relationship between benchmark scores and genuine language understanding.
Approach:	They propose to use GAPERON to investigate evaluation dynamics under realistic training conditions.
Outcome:	The proposed model outperforms models that excel on benchmarks in qualitative text generation and vice versa.

Copied to clipboard

Ziyuan Nan, Qi Yi, Di Huang, Yutong Wu, Guanhua Huang, Xue Gong, Kejiao Li, Yuhao Jiang, Chenchen Zhang, Zenan Xu, Xing Hu, Bo Zhou

FLARE: Task-Agnostic Embedding Model Evaluation via Normalizing Flows (2026.findings-acl)

Challenge:	Parallel thinking is a promising avenue for scaling test-time compute in Large Language Models . however, coordinating the exploration and aggregation stages remains challenging .
Approach:	They propose a parallel thinking framework that explicitly incentivizes coordination between components via end-to-end reinforcement learning.
Outcome:	The proposed framework improves accuracy by 6.0% over long chain-of-thought baselines while reducing wall-clock latency by 39.4% under matched token budgets.

Copied to clipboard

Jingzhou Jiang, Yixuan Tang, Yi Yang, Kar Yan Tam

RCTEA: Richness-guided Co-training for Temporal Entity Alignment (2026.findings-acl)

Challenge:	Existing methods based on kernel estimators or Gaussian mixtures fail to model high-dimensional distributions effectively, resulting in unstable rankings.
Approach:	They propose a method which uses normalizing flows to estimate information sufficiency in high-dimensional spaces by learning invertible transformations.
Outcome:	Experiments on 11 datasets show that FLARE achieves a strong Spearman’s (up to 0.90) with supervised benchmarks, remaining robust even for high-dimensional embeddings (d 3,584).

Copied to clipboard

Jiayun Li, Wen Hua, Shiqi Fan, Fengmei Jin, Haiyang Jiang, Xue Li

SQLAgent: Learning to Explore Before Generating as a Data Engineer (2026.findings-acl)

Challenge:	Existing TEA models fail to capture the orthogonal yet complementary effect between structural and temporal features.
Approach:	They propose a framework that jointly models structural and temporal aspects of Temporal Knowledge Graphs for entity alignment.
Outcome:	The proposed framework achieves state-of-the-art on public TEA benchmarks.

Copied to clipboard

Wenjia Jiang, Yiwei Wang, Boyan Han, Joey Tianyi Zhou, Chi Zhang

Reasoning Up the Instruction Ladder for Controllable Language Models (2026.findings-acl)

Challenge:	Existing approaches to large language models fail to generalize in complex, real-world settings due to database-specific nature of SQL reasoning.
Approach:	They propose a two-stage LLM-based framework that decouples knowledge acquisition from query generation.
Outcome:	The proposed framework significantly improves accuracy over baselines on large-scale benchmarks.

Copied to clipboard

Zishuo Zheng, Vidhisha Balachandran, Chan Young Park, Faeze Brahman, Sachin Kumar

CRISP: Compressing Redundancy in Chain-of-Thought via Intrinsic Saliency Pruning (2026.findings-acl)

Challenge:	Current models struggle to balance competing directives, causing conflicting instructions.
Approach:	They propose to reframe instruction hierarchy resolution as a reasoning task . they use a training dataset to enable this capability by transferring general reasoning capabilities to instruction prioritization .
Outcome:	The proposed method improves on safety-critical scenarios beyond the training distribution and jailbreaks.

Copied to clipboard

Yangsong Lan, Hongliang Dai, Piji Li

CascadeFix: Multi-Location Program Repair via Cascading Planning and Generation (2026.findings-acl)

Challenge:	Existing approaches to compress CoT with external compressors fail to align with the model’s internal reasoning dynamics, resulting in the loss of critical logical steps.
Approach:	They propose a framework that exploits the model’s intrinsic saliency to compress CoT by exploiting its reasoning termination token .
Outcome:	The proposed framework reduces redundancy in reasoning chain by exploiting the model’s intrinsic saliency.

Copied to clipboard

Huan Zhang, Li Kuang, Yang Yang, Yilei Fang, Yingjie Xia

DGPO: Beyond Pairwise Preferences with Directional Consistent Groupwise Optimization (2026.findings-acl)

Challenge:	Existing methods for automating program repair face insufficient bug dependency modeling and inadequate global repair planning when addressing semantically complex multi-location bugs.
Approach:	They propose a multi-location automatic repair method via cascading planning and generation . they propose to model dependencies among bugs and cluster them to ensure rationality .
Outcome:	The proposed method resolves 84 multi-location bugs, achieving a 31% improvement over current methods.

Copied to clipboard

Mengyi Deng, Zhiwei Li, Xin Li, Tingyu Zhu, Yulan Yuan, Zhijiang Guo, Wei Wang

Emergence and Localisation of Semantic Role Circuits in LLMs (2026.findings-acl)

Challenge:	Existing methods for directional consistency alignment of large language models are limited . a recent study suggests reverse supervision as a complement to forward reasoning .
Approach:	They propose a framework that aggregates supervision signals at the group level and explicitly models direction-aware alignment through multi-candidate comparisons.
Outcome:	The proposed framework achieves 3.2% accuracy improvement across five benchmarks and multiple datasets.

Copied to clipboard

Nura Aljaafari, Danilo Carvalho, Andre Freitas

Challenge:	Despite displaying semantic competence, large language models’ internal mechanisms that ground abstract semantic structure remain insufficiently characterised.
Approach:	They propose a causal-temporal methodology that uses contrastive minimal pairs, edge-attribution circuit discovery, and training-time tracking to characterise semantic-role circuits along three dimensions.
Outcome:	The proposed method is applicable to any behaviour isolatable through contrastive minimal pairs, including factual recall, syntactic agreement, and logical reasoning.

Thinking Economically: A Hierarchical Framework for Adaptive-Complexity Reasoning in LLMs (2026.findings-acl)

Copied to clipboard

Yubo Gao, Haotian Wu, Hong Chen, Junquan Huang, Yibo Yan, Jungang Li, Zihao Dongfang, Sicheng Tao, PS Tan, Jie Zhang, Xuming Hu

Characterizing the Robustness of Black-Box LLM Planners Under Perturbed Observations with Adaptive Stress Testing (2026.findings-acl)

Challenge:	Existing efficiency methods for Chain-of-Thought (CoT) generate excessively long rationales without commensurate accuracy gains.
Approach:	They propose a training framework that operationalizes this principle through coarse-to-fine budgeting.
Outcome:	Experiments on GSM8K and MATH500 show that HAB surpasses standard CoT in accuracy and reduces token usage, achieving stronger performance-efficiency trade-off than baselines.

Copied to clipboard

Neeloy Chakraborty, John Pohovey, Melkior Ornik, Katherine Rose Driggs-Campbell

ProMCP: Profiling Token Flows and Latency Costs in Model Context Protocol–Based LLM Agents (2026.findings-acl)

Challenge:	Large language models have been used in reasoning-heavy planning tasks but their tendency to hallucinate unsafe outputs poses risks.
Approach:	They propose a method for searching the space of prompt perturbations using adaptive stress testing with Monte-Carlo tree search.
Outcome:	The proposed method can detect scenarios, sensor configurations, and prompt phrasing that cause models to act with high uncertainty or crash.

Copied to clipboard

Sumera Anjum, Weijian Zheng, Rajkumar Kettimuthu, Heng Fan, Yunhe Feng

Reducing Hallucinations in LLMs via Factuality-Aware Preference Learning (2026.findings-acl)

Challenge:	Large Language Models are increasingly used as agents that interact with external tools and data sources to solve tasks that require fresh knowledge, precise computation, or action in a real environment.
Approach:	They propose a framework that decomposes the MCP workflow into a six-stage communication pipeline and enables granular attribution of computational costs.
Outcome:	The proposed framework decomposes the MCP workflow into a six-stage communication pipeline, enabling granular attribution of computational costs.

Copied to clipboard

Sindhuja Chaduvula, Ahmed Y. Radwan, Azib Farooq, Yani Ioannou, Shaina Raza

POLARIS: A Gödel Agent Framework for Small Language Models through Experience-Abstracted Policy Repair (2026.findings-acl)

Challenge:	Preference alignment methods can reinforce hallucinations when preference judgments reward fluency and confidence over factual correctness.
Approach:	They propose a method that corrects misordered preference pairs and adds a factuality-aware margin to emphasize pairs with clear correctness differences.
Outcome:	The proposed method improves factuality and reduces hallucination rates across seven open-weight LLMs.

Copied to clipboard

Aditya Namdev Kakade, Vivek Srivastava, Shirish Karande

PhaseMI: A Motivational Interviewing Dataset for Enhancing Phase Progression in LLM-based Counseling (2026.findings-acl)

Challenge:	Gödel agent Polaris makes policy level changes with small, auditable patches that persist in the policy and are reused on unseen instances within each benchmark.
Approach:	They propose a Gödel agent that performs policy repair via experience abstraction . Polaris makes policy level changes with small, auditable patches that persist in the policy .
Outcome:	The proposed agent improves on MGSM, DROP, GPQA, and LitBench models over the base policy and competitive baselines.

Copied to clipboard

Jina Kim, Myeongho Jeon, Soohyun Cho, Chae-Gyun Lim, Jongmin Lim, Haewon Min, Eunho Yang

Challenge:	Existing MI datasets do not explicitly model structured progression of MI phases, which is essential for effective and goal-oriented counseling.
Approach:	They propose a phase-structured MI dataset with a data generation framework that employs therapist, client, and supervisor LLMs to explicitly control phase transitions.
Outcome:	The proposed model achieves 12.3% better coverage of MI phases, 37.6% in guiding, and 61.1% in choosing.

PsychEthicsBench: Evaluating Large Language Models Against Australian Mental Health Ethics (2026.findings-acl)

Copied to clipboard

Yaling Shen, Stephanie Fong, Yiwen Jiang, Zimu Wang, Feilong Tang, Qingyang Xu, Xiangyu Zhao, Zhongxing Xu, Jiahe Liu, Jinpeng Hu, Dominic Dwyer, Zongyuan Ge

DynamixSFT: Dynamic Mixture Optimization of Instruction Tuning Collections (2026.findings-acl)

Challenge:	Mental disorders affect nearly one in seven people worldwide, yet the vast majority do not receive adequate care.
Approach:	They propose a framework to evaluate LLMs' ethical knowledge and behavioral responses through multiple-choice and open-ended tasks with fine-grained ethicality annotations.
Outcome:	Empirical results across 14 models reveal that refusal rates are poor indicators of ethical behavior, revealing a significant divergence between safety triggers and clinical appropriateness.

Copied to clipboard

Haebin Shin, Lei Ji, Xiao Liu, Zhiwei Yu, Hyunwoo Yoo, Qi Chen, Yeyun Gong

Challenge:	Several studies rely on additional models to optimize mixtures.
Approach:	They propose a method that dynamically optimizes instruction-tuning dataset mixtures by prior-scaled Boltzmann Exploration and a multi-armed bandit setup.
Outcome:	The proposed method improves the TÜLU-2-mixture and TÜLO-3-mixtures across 10 benchmarks while introducing minimal computational overhead over naive sampling.

BMAM: Brain-inspired Multi-Agent Memory Framework (2026.findings-acl)

Copied to clipboard

Yang Li, Jiaxiang Liu, Yusong Wang, Yujie Wu, Mingkun Xu

Challenge:	Language-model-based agents operating over extended interactions face persistent challenges in preserving temporally grounded information and maintaining behavioral consistency across sessions.
Approach:	They propose a general-purpose memory architecture that decomposes agent memory into six components that operate at complementary time scales.
Outcome:	BMAM outperforms memory-augmented baselines on LoCoMo benchmarks with 78.45% accuracy . a targeted refinement of the temporal-trigger heuristics raises LongMemEval multi-session accuracy from 45.2% to 56.4%.

Verified Critical Step Optimization for LLM Agents (2026.findings-acl)

Copied to clipboard

Mukai Li, Qingcheng Zeng, Tianqing Fang, Zhenwen Liang, Linfeng Song, Qi Liu, Haitao Mi, Dong Yu

Challenge:	Critical Step Optimization (CSO) focuses preference learning on verified critical steps where alternative actions demonstrably flip task outcomes from failure to success.
Approach:	They propose a method which focuses preference learning on verified critical steps where alternative actions demonstrably flip task outcomes from failure to success.
Outcome:	The proposed method outperforms the existing methods on GAIA-Text-103 and XBench-DeepSearch while requiring supervision at only 16% of trajectory steps.

Nash-Pruned CredMAS: Dynamic Panel Pruning for VLM-MAS using Nash-based Selection and Doubly-Robust Credits (2026.findings-acl)

Copied to clipboard

Yijia Fan, Mingyu Liu, Jing Yang, Jian Wang, Keze Wang, Jusheng Zhang

SWE-Mutation: Can LLMs Generate Reliable Test Suites in Software Engineering? (2026.findings-acl)

Challenge:	Multi-Agent Systems (MAS) are expensive due to static panel designs, where all N agents communicate at every T round.
Approach:	They propose an economic framework that transforms agent selection into a dynamic resource allocation game.
Outcome:	The proposed system reduces token consumption by over 25% on challenging benchmarks while reducing token consumption.

Copied to clipboard

Yuxuan Sun, Yuze Zhao, Yufeng Wang, Yao Du, Zhiyuan Ma, Jinbo Wang, Mengdi Zhang, Kai Zhang, Zhenya Huang

STaD: Scaffolded Task Design for Identifying Compositional Skill Gaps in LLMs (2026.findings-acl)

Challenge:	Evaluating software engineering capabilities is a core component of large language models (LLMs).
Approach:	They propose a benchmark to evaluate LLM-generated test suites that introduces mutated solutions that attempt to "fool" them.
Outcome:	The proposed test suites are based on 2,636 mutated variants derived from 800 original instances and include a multilingual subset spanning nine programming languages.

Copied to clipboard

Sungeun An, Swanand Ravindra Kadhe, Shailja Thakur, Chad DeLuca, Hima Patel

PseudoSeer: a Search Engine for Pseudocode (2026.findings-acl)

Challenge:	Existing benchmarks often treat complex tasks as monolithic, resulting in inconsistent performance and inconsistent explanations.
Approach:	They propose a framework for creating controlled variations of benchmark tasks based on the concept of scaffolding, which introduces structured, incremental support in a step-by-step manner.
Outcome:	The proposed framework enables systematic probing of model behavior by identifying the specific reasoning skill compositions they lack.

Copied to clipboard

Levent Toksoz, Mukund Srinath, Gang Tan, C. Lee Giles

Beyond Polarity: Continuous Affect-Enhanced Multimodal Aspect-Based Sentiment Classification (2026.findings-acl)

Challenge:	PseudoSeer is a search engine for academic pseudocode that indexes over 320,000 implementations extracted from 2.2 million arXiv papers.
Approach:	They propose to use caption-reference pairs to match short queries with a median length of five words against long documents composed primarily of natural language with limited LaTeX notation.
Outcome:	The proposed algorithm outperforms the best pretrained model by 8.7 points and achieves 66.5% R@10 .

Copied to clipboard

Ling-Ang Meng, Tianyu Zhao, Dawei Song, Jingxu Cao, Youhui Zuo

Topology Matters: Measuring Memory Leakage in Multi-Agent LLMs (2026.findings-acl)

Challenge:	Existing methods for multimodal aspect-based sentiment classification exploit discrete polarity patterns and generic visual embeddings.
Approach:	They propose a Valence–Arousal–Dominance(VAD)-Enhanced MABSC framework that integrates VAD signals into multimodal sentiment reasoning and learns emotion-sensitive image representations.
Outcome:	The proposed framework brings VAD signals into multimodal sentiment reasoning and learns emotion-sensitive image representations.

Copied to clipboard

Jinbo Liu, Defu Cao, Yifei Wei, Tianyao Su, Yuan Liang, Yushun Dong, Yan Liu, Yue Zhao, Xiyang Hu

Beyond ’Aha!’: Toward Systematic Meta-Abilities Alignment in Large Reasoning Models (2026.findings-acl)

Challenge:	Multi-agent LLMs are rapidly moving from prototype to real-world use . network topology is a first-order security parameter in multi-aggent systems .
Approach:	They propose a framework for comparing topology-conditioned memory leakage in multi-agent LLM systems.
Outcome:	The proposed framework evaluates topology-conditioned memory leakage in multi-agent LLM systems.

Copied to clipboard

Zhiyuan Hu, Yibo Wang, Hanze Dong, Yuhui Xu, Amrita Saha, Caiming Xiong, Bryan Hooi, Junnan Li

Rewarding the Rare: Uniqueness-Aware RL for Creative Problem Solving in LLMs (2026.findings-acl)

Challenge:	Prior work has shown that outcome-based reinforcement learning (RL) can incidentally elicit advanced reasoning behaviors such as self-correction, backtracking, and verification.
Approach:	They explicitly align large reasoning models with three meta-abilities: deduction, induction, and abduction, using automatically generated, self-verifiable tasks.
Outcome:	The proposed model aligns models with deduction, induction, and abduction meta-abilities using automatically generated, self-verifiable tasks.

Copied to clipboard

Zhiyuan Hu, Yucheng Wang, Yufei He, Jiaying Wu, Yilun Zhao, See-Kiong Ng, Cynthia Breazeal, Anh Tuan Luu, Hae Won Park, Bryan Hooi

Empathy Applicability Modeling for General Health Queries (2026.findings-acl)

Challenge:	Reinforcement learning (RL) is a paradigm for post-training large language models, but it suffers from exploration collapse . a new study finds that RL fails to reward correct solutions that exhibit rare high-level strategies .
Approach:	They propose a method that rewards correct solutions that exhibit rare high-level strategies by clustering rollouts according to their high- level solution strategies.
Outcome:	The proposed approach improves pass@k across large sampling budgets and increases area under the pass@K curve (AUC@K) without sacrificing pass@1.

Copied to clipboard

Shan Randhawa, Agha Ali Raza, Kentaro Toyama, Julie Hui, Mustafa Naseem

Worldwide LiveVQA: Real-Time Visual Knowledge Seeking and Updating Across Languages (2026.findings-acl)

Challenge:	Existing NLP frameworks focus on reactively labeling empathy in doctors’ responses but offer limited support for anticipatory modeling of empathy needs, especially in general health queries.
Approach:	They propose an Empathy Applicability Framework that classifies patient queries in terms of the applicability of emotional reactions and interpretations based on clinical, contextual, and linguistic cues.
Outcome:	The Empathy Applicability Framework outperforms heuristic and zero-shot LLMs in the clinical setting.

Copied to clipboard

Xuanao Huang, Xingjia Liu, Zetong Zhou, Yuyang Peng, Yao Wan, Dongping Chen

HCRE: LLM-based Hierarchical Classification for Cross-Document Relation Extraction with a Prediction-then-Verification Strategy (2026.findings-acl)

Challenge:	Existing resources for visual knowledge remain confined to English, creating a "Worldwide Knowledge Gap" eric liu: production and dissemination of knowledge exhibit a distinct trend toward decentralization and linguistic fragmentation.
Approach:	They propose a dataset for multilingual visual knowledge seeking and updating across ten major languages.
Outcome:	The proposed dataset is the first dynamic-updating dataset for multilingual visual knowledge seeking and updating across ten major languages.

Copied to clipboard

Guoqi Ma, Liang Zhang, Hongyao Tu, Hao Fu, Hui Li, Yujie Lin, Longyue Wang, Weihua Luo, Jinsong Su

VLMGuard-R1: Proactive Safety Alignment for VLMs via Reasoning-Driven Prompt Optimization (2026.findings-acl)

Challenge:	Existing approaches to cross-document relation extraction (RE) focus on identifying relations between head and tail entities from single sentence or document.
Approach:	They propose a hierarchical relation tree-based LLM-based hierarchic classification model for cross-document relation extraction (HCRE) based on predefined relations, the model can perform hierarchically classification level by level.
Outcome:	The proposed model outperforms existing baselines and validates its effectiveness.

Copied to clipboard

Menglan Chen, Xianghe Pang, Jingjing Dong, WenHao Wang, Yaxin Du, Siheng Chen

TinyAttack: Exploring Stylistic Vulnerabilities in Large Language Models (2026.findings-acl)

Challenge:	integrating vision and language models with safety standards is essential to mitigate multimodal complexity . integrating visual inputs with vision and text unveils subtle threats beyond the reach of conventional safeguards .
Approach:	They propose a framework that combines vision and language to provide a multimodal reasoning-driven prompt rewriting.
Outcome:	The proposed framework outperforms baseline models on five benchmarks with six VLMs.

Copied to clipboard

Mamta Mamta, Bogdan Grecu, Oana Cocarascu

Human-Agent Collaborative Paper-to-Page Crafting (2026.findings-acl)

Challenge:	Existing research on robustness of large language models has focused on text-based perturbations and the use of invisible characters and homoglyphs.
Approach:	They propose a framework to exploit weaknesses in Large Language Models (LLMs) by changing their stylistic structure using Unicode.
Outcome:	The proposed framework exploits vulnerabilities in large language models through Unicode-based stylistic transformations without altering its semantic or syntactic structure.

Copied to clipboard

Qianli Ma, Siyu Wang, Chen Yilin, Yinhao Tang, Yixiang Yang, Chang Guo, Bingjie Gao, Zhening Xing, Yanan Sun, Zhipeng Zhang

Challenge:	Existing approaches to create project pages from academic papers have focused on static slides and posters, but the dynamic nature of webpages remains an unaddressed challenge.
Approach:	They propose a novel multi-agent system that deconstructs paper-to-page creation into a coarse-to fine pipeline from narrative planning to multimodal content generation and interactive rendering.
Outcome:	The proposed system generates high-quality, visually appealing pages in under 15 minutes for less than $0.1 .

Skill Weaving: Efficient LLM Improvement via Modular Skillpacks (2026.findings-acl)

Copied to clipboard

Zhuo Li, Guodong DU, Zesheng Shi, Weiyang Guo, Weijun Yao, Yuan Zhou, Jiabo Zhang, Jing Li

Controlling What You Share: Assessing Language Model Adherence to Privacy Preferences (2026.findings-acl)

Challenge:	Large Language Models (LLMs) can specialize under fixed memory and inference budgets, but they struggle to achieve high performance across heterogeneous domains.
Approach:	They propose a modular improvement framework that partitions full capabilities of a general-purpose model into domain-specific delta modules that reorganize and refine the model's internal knowledge.
Outcome:	The proposed framework outperforms monolithic models on multi-task and agentic benchmarks and achieves up to 4 speedup.

Copied to clipboard

Guillem Ramírez, Alexandra Birch, Ivan Titov

GerAV: Towards New Heights in German Authorship Verification using Fine-Tuned LLMs on a New Benchmark (2026.findings-acl)

Challenge:	Large language models (LLMs) are accessed via commercial APIs, but expose data to service providers.
Approach:	They propose a framework where a local model uses natural language instructions to rewrite queries and paired them with synthetic privacy profiles to achieve better privacy preservation.
Outcome:	The proposed model outperforms large-scale few-shot models in terms of privacy preservation and performance.

Copied to clipboard

Lotta Kiefer, Christoph Leiter, Sotaro Takeshita, Elena Schmidt, Steffen Eger

LipoAgent: Coordinating Fine-Tuned LLM Agents for Safer Lipid Design (2026.findings-acl)

Challenge:	Authorship verification (AV) is a task of determining whether two texts were written by the same author.
Approach:	They propose a benchmark for German AV comprising over 400k labeled text pairs.
Outcome:	The proposed model outperforms baselines and state-of-the-art models by 0.09 and surpasses GPT-5 in a zero-shot setting by 0.08.

Copied to clipboard

Leshu Li, An Lu, Haiyu Wang, Zhibin Feng, Conghui Duan, Qing Bao, Zongmin Zhao, Sai Qian Zhang

PsychePass: Calibrating LLM Therapeutic Competence via Trajectory-Anchored Tournaments (2026.findings-acl)

Challenge:	Lipid nanoparticles (LNPs) are among the most clinically mature platforms for nucleic acid delivery, yet designing lipids that are effective and biologically safe remains a major bottleneck.
Approach:	They propose a safety-aware multi-agent LLM framework for lipid discovery that enforces toxicity as a prerequisite for efficiency prediction.
Outcome:	The proposed framework achieves an average improvement in mRNA transfection efficiency prediction across multiple foundation models.

Copied to clipboard

Zhuang Chen, Dazhen Wan, Zhangkai Zheng, Guanqun Bi, Xiyao Xiao, Binghang Li, Minlie Huang

Challenge:	evaluating therapeutic competence of large language models remains challenging due to unstructured and longitudinal nature of counseling.
Approach:	They propose a framework that calibrates the therapeutic competence of LLMs via trajectory-anchored tournaments.
Outcome:	The proposed framework calibrates the therapeutic competence of LLMs via trajectory-anchored tournaments.

Generalization or Memorization? Multi-Agent vs. Baseline LLMs and AutoML Models for Tabular Classification (2026.findings-acl)

Copied to clipboard

Aida Sanatizadeh, Sorouralsadat Fatemi, Reza Mousavi, Ahmed Abbasi

GanitLLM: Difficulty-Aware Bengali Mathematical Reasoning through Curriculum-GRPO (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are increasingly used for structured tabular data.
Approach:	They evaluate a representative modular Multi-Agent LLM framework against state-of-the-art AutoML systems and established baselines.
Outcome:	The proposed model outperforms AutoML on pre-cutoff and post-cut off datasets.

Copied to clipboard

Shubhashis Roy Dipta, Khairul Mahbub, Nadia Najjar

Can MLLMs Reason Beyond Language? VisReason: A Comprehensive Benchmark for Vision-Centric Reasoning (2026.findings-acl)

Challenge:	Existing LLMs either reason in English and translate, or simply fail on multi-step Bengali math.
Approach:	They propose a Bengali mathematical reasoning model called GanitLLM with a difficulty-aware Bengali math corpus and a curriculum-based GRPO pipeline.
Outcome:	The proposed model improves on Bn-MGSM and Bn MSVAMP by +8 and +7 accuracy points while increasing the percentage of Bengali reasoning tokens from 14% to over 88% and reducing solution length from 943 to 193 words.

Copied to clipboard

Longteng Guo, Yifan Wang, Pengkang Huo, Tailai Chen, Yuze Wu, Jing Liu, Xinxin Zhu

Multimodal Dual-Path Decoding for Medical Report Generation (2026.findings-acl)

Challenge:	Recent advances in multimodal large language models demonstrate strong performance on visual reasoning benchmarks.
Approach:	They propose a benchmark for vision-centric reasoning that integrates visual and textual information for non-trivial reasoning.
Outcome:	The proposed benchmark exposes gaps between humans and current MLLMs and reveals limited benefits from test-time reasoning strategies.

Copied to clipboard

Jinghan Sun, Dong Wei, Zhihong Zhu, Yuyang Xue, Steven McDonagh, Xian Wu

PanoramaRAG: Enabling Consistent Global Topic Awareness in Graph-Based RAG (2026.findings-acl)

Challenge:	Current methods for radiology report generation rely on encoder-decoder based frameworks that fail to integrate multimodal clinical evidence with domain-specific knowledge.
Approach:	They propose a multimodal dual-path framework that synergistically integrates large vision-language models and large language models for radiology report generation.
Outcome:	The proposed framework improves on the public MIMIC-CXR benchmark and shows that it is superior to state-of-the-art models.

Copied to clipboard

Ding Deng, Xiang Li, Yaqing Zhang, Meng Li, Xiting Wang

Universally Empowering Zeroth-Order Optimization via Adaptive Layer-wise Sampling (2026.findings-acl)

Challenge:	Existing graph-based methods for enhancing Large Language Models (LLMs) with external knowledge are focusing on local relationships, resulting in suboptimal performance for tasks that require global context.
Approach:	They propose a "panorama"-guided paradigm that integrates a light yet comprehensive "panoramic" of the corpus to guide all stages of the retrieval process.
Outcome:	The proposed paradigm performs well across five datasets and a variety of tasks.

Copied to clipboard

Fei Wang, Li Shen, Liang Ding, Chao Xue, Ye Liu, Changxing Ding

Speculative Decoding with a Speculative Vocabulary (2026.findings-acl)

Challenge:	Existing methods for fine-tuning Large Language Models are slow and lack of performance.
Approach:	They propose a Zeroth-Order optimization framework that uses forward passes to fine-tune Large Language Models.
Outcome:	The proposed framework achieves 1.7 to 3.0 wall-clock acceleration on LLaMA and OPT models.

Copied to clipboard

Miles Williams, Young D. Kwon, Rui Li, Alexandros Kouris, Stylianos I. Venieris

Entropy-Aware Reshaping of Reinforcement Signals for Multi-Answer Reasoning (2026.findings-acl)

Challenge:	Speculative decoding methods use a draft model to accelerate inference while yielding identical outputs.
Approach:	They propose a method that selects a vocabulary subset per decoding step and uses a draft model to generate a series of tokens that are verified in parallel.
Outcome:	The proposed method achieves higher acceptance length than state-of-the-art speculative decoding method, EAGLE-3.

Copied to clipboard

Zhi Li, Huidan Xu, Zhen Hu, Yali Du, Ying Liu

TripTide: A Benchmark for Adaptive Travel Planning under Disruptions (2026.findings-acl)

Challenge:	Reinforcement learning with verifiable rewards (RLVR) is a standard post-training paradigm for large language models.
Approach:	They propose a framework that reshapes how learning signals are normalized and aggregated.
Outcome:	Experiments on MCTACO and MMLU-Multi show that the proposed framework improves accuracy, training stability and cross-dataset transfer performance.

Copied to clipboard

Priyanshu Karmakar, Soumyabrata Chaudhuri, Shubhojit Mallick, Manish Gupta, Abhik Jana, Shreya Ghosh

Challenge:	Recent work has shown the promise of Large Language Models (LLMs) for personalized, constraint-aware travel itinerary generation, but real-world travel often involves disruptions such as transit cancellations, weather-related closures, or overbooked attractions.
Approach:	They propose a benchmark to evaluate the ability of Large Language Models (LLMs) to revise travel itineraries under realistic disruptions.
Outcome:	The proposed benchmark evaluates the ability of Large Language Models (LLMs) to revise travel itineraries under real-world disruption scenarios.

Rational Synthesizers or Heuristic Followers? Analyzing LLMs in RAG-based Question-Answering (2026.findings-acl)

Copied to clipboard

Atharv Naphade

IDEA: An Interpretable and Editable Decision-Making Framework for LLMs via Verbal-to-Numeric Calibration (2026.findings-acl)

Challenge:	Retrieval-Augmented Generation (RAG) is the prevailing paradigm for grounding Large Language Models.
Approach:	They propose a method to integrate conflicting retrieved evidence into large language models.
Outcome:	The proposed model is based on a curated dataset of 1,635 controversial questions paired with 15,058 diversely-sourced evidence documents.

Copied to clipboard

Yanji He, Yuxin Jiang, Yiwen Wu, Bo Huang, Jiaheng Wei, Wei Wang

An Answer is just the Start: Related Insight Generation for Open-Ended Document-Grounded QA (2026.findings-acl)

Challenge:	Existing approaches fail to integrate domain expert insights beyond simple prompting.
Approach:	They propose a framework that extracts LLM decision knowledge into an interpretable parametric model over semantically meaningful factors.
Outcome:	Experiments show that IDEA outperforms DeepSeek R1 and GPT-5.2 in accuracy and accuracy.

Copied to clipboard

Saransh Sharma, Pritika Ramu, Aparna Garimella, Koyel Mukherjee

Behavioral Consistency Validation for LLM Agents: An Analysis of Trading-Style Switching through Stock-Market Simulation (2026.findings-acl)

Challenge:	Existing QA benchmarks do not explicitly support document-grounded related insight generation . Existing document-based QA efforts focus on answering fact-based questions .
Approach:	They propose a task to generate additional insights from a document collection that improves, extends or rethinks an initial answer to an open-ended question.
Outcome:	The proposed task improves, extends, or rethinks an answer to an open-ended question.

Copied to clipboard

Zeping Li, Guancheng Wan, Keyang Chen, Yu Chen, Yiwen Zhao, Philip Torr, Guangnan Ye, Zhenfei Yin, Hongfeng Chai

NITI: Neural Plan Concretization for Incremental Execution, Bridging and Trigger Inference from Underspecified Human Policies (2026.findings-acl)

Challenge:	Recent studies have applied Large Language Models (LLMs) as agents in financial stock market simulations to test if micro-level behaviors aggregate into macro-level phenomena.
Approach:	They propose four alignment metrics and use Mann–Whitney U tests to compare agents’ style-switching behavior with financial theory.
Outcome:	The proposed model is only partially consistent with financial theory.

Copied to clipboard

Ayan Banerjee, Shomrik Barua Banerjee, Sandeep Gupta

Challenge:	Using NITI, we examine the performance of a safety-critical automated insulin dosing task with minimal contextualization infence overhead.
Approach:	They propose a framework that treats large language models as execution-time concretizers of human intent that incrementally executes abstract policies via verifier-grounded interfaces.
Outcome:	The proposed framework outperforms one-shot and chain-of-thought baselines on two structurally distinct embodied domains: a world cubing championship 22 Rubik’s Cube scramble and a safety-critical automated insulin dosing task.

Automatic Combination of Sample Selection Strategies for Few-Shot Learning (2026.findings-acl)

Copied to clipboard

Branislav Pecher, Ivan Srba, Maria Bielikova, Joaquin Vanschoren

LearnAlign: Data Selection for LLM Reinforcement Learning with Improved Gradient Alignment (2026.findings-acl)

Challenge:	Existing studies on small language models are characterised by a labelled data scarcity due to data collection/annotation costs or privacy considerations, making the training of typical deep learning models unfeasible.
Approach:	They propose a method for Automatic Combination of SamplE Selection Strategies to leverage the strengths and complementarity of various well-established selection objectives.
Outcome:	The proposed method outperforms all in-context learning strategies and performs on par or exceeds the in-constinction learning specific baselines.

Copied to clipboard

Shipeng Li, Zhiqin Yang, Shikun Li, Xiaobo Xia, Hengyu Liu, Xinghua Zhang, Gaode Chen, Dong Fang, Ying Tai, Zhe Peng

Auditing LLM Responses to Harmful Stereotypes Targeting Mental Health Groups (2026.findings-acl)

Challenge:	Reinforcement learning with verifiable rewards (RLVR) is a key technique for enhancing LLMs’ reasoning abilities, yet its data inefficiency remains a major bottleneck.
Approach:	They propose a gradient-alignment-based method which intelligently selects the learnable and representative training reasoning data for RLVR post-training.
Outcome:	Experiments on five reasoning benchmarks show that the proposed method significantly reduces training data requirements while improving performance.

Copied to clipboard

Arka Dutta, Rijul Magu, Sean Kim, Seohee Yoon, Munmun De Choudhury, Ashiqur R. KhudaBukhsh

ROSCO-Omni: Multimodal LLM-Based Communication Understanding for Non- and Minimally-Speaking Autistic Individuals (2026.findings-acl)

Challenge:	Large Language Models (LLMs) can exhibit imbalanced biases against vulnerable groups, but how they rationalize stereotypes and rights restrictions targeting mental health entities remains underexplored.
Approach:	They audit a suite of open-weight LLMs on stereotype-justification prompts tied to mental health identities.
Outcome:	The proposed models endorse harmful stereotypes when explicitly asked to justify them, with endorsement varying across model families, versions, and mental health conditions.

Copied to clipboard

Siddhant Bikram Shah, Kristina T. Johnson

LQM: Linguistically Motivated Multidimensional Quality Metrics for Machine Translation (2026.findings-acl)

Challenge:	30% of autistic individuals remain non- or minimally-speaking throughout their lives . however, caregivers rely on simultaneous integration of visual cues, auditory signals, and contextual understanding to infer intent.
Approach:	They propose a framework that fine tunes a teacher-student MLLM for domain-specialized inference.
Outcome:	The proposed framework achieves comparable performance to closed-source models .

Copied to clipboard

Samar M. Magdy, Fakhraddin Alwajih, Abdellah El Mekki, Wesam El Sayed, Muhammad Abdul-Mageed

Many-Shot Scaling of In-Context Learning with Self-Generated Demonstrations (2026.findings-acl)

Challenge:	Existing MT evaluation frameworks fail to capture dialect- and culture-specific errors in diglossic languages.
Approach:	They propose a hierarchical error taxonomy for diagnosing MT errors through six linguistic levels: sociolinguistics, pragmatics, semantics, morphosyntax, orthography, and graphetics.
Outcome:	The proposed framework produces 6,113 labeled error spans across 3,495 unique erroneous sentences . it is language-agnostic and can be easily applied to or adapted for other languages.

Copied to clipboard

Zhengyao Gu, Henry Peng Zou, Yankai Chen, Aiwei Liu, Weizhi Zhang, Philip S. Yu

Small Data, Big Noise: Adversarial Training for Robust ParameterEfficient Fine-Tuning (2026.findings-acl)

Challenge:	In-context learning methods that use self-generated annotations do not scale to many-shot scenarios.
Approach:	They propose a framework analogous to semi-supervised learning that uses self-generated annotations instead of ground truth labels.
Outcome:	The proposed framework outperforms ground truth ICL under zero-shot, few-shot and many-shot settings.

Copied to clipboard

Eitan Cohen, Idan Simai, Uri Shaham

LLMs as Lab Engineers: A Benchmark for Analytical Method Lifecycle Management (2026.findings-acl)

Challenge:	Parameter-Efficient Fine-Tuning (PEFT) is essential for adapting foundation models to downstream tasks, but current methods struggle with robustness to noise and performance degradation on limited training data.
Approach:	They propose a framework that brings adversarial training to PEFT to enhance model robustness and generalization, outperforming alternative approaches.
Outcome:	Experiments with two variants of the proposed framework show that it outperforms existing methods in low-resource settings and under word-level and character-level corruptions.

Copied to clipboard

Xiaoyi Chen, Mahsa Monshizadeh, Chaoqi Zhang, Jianjun Lang, Yang Wu, Genevieve Mortensen, Xiaozhong Liu, Haixu Tang

Too Fast, Too Shallow – LLMs, Including Reasoning LLMs, Are Unreliable Constitutional Reasoners (2026.findings-acl)

Challenge:	General-purpose commercial models outperform domain-specialized ones, while RAG and reasoning significantly improve performance.
Approach:	They propose a benchmark to evaluate LLMs' capabilities in analytical chemistry scenarios.
Outcome:	The proposed framework outperforms existing benchmarks focused on factual knowledge and provides practical guidance for analytical chemistry challenges.

Copied to clipboard

Reto Gubelmann, Peter Hongler

JTPRO: A Joint Tool–Prompt Reflective Optimization Framework for Language Agents (2026.findings-acl)

Challenge:	Using three different datasets, we assess LLMs’ constitutional reasoning abilities using three different constitutional frameworks.
Approach:	They propose to use the influential dual process theory of cognition to assess LLMs' constitutional reasoning abilities.
Outcome:	The LLMs label less than 70% correctly and open-weight reasoning LLM and gpt-4o outperform open- weight non-reasoning LLM.

Copied to clipboard

Sandip Ghoshal, Anshul Mittal, Jyotika Singh, Miguel Ballesteros, Weiyi Sun, Fang Tu, Shailender Singh, Yassine Benajiba, Fahad Shah, Sujeeth Bharadwaj, Sujith Ravi, Dan Roth

CANDICE: Agentic Causal Disentanglement with Class Conditional Knowledge Integration for Long Tailed Domain Generalization (2026.findings-acl)

Challenge:	Large language model agents struggle with ambiguous tool descriptions and underspecified tool schemas that ignore tool-specific nuances.
Approach:	They propose a framework for improving tool-calling reliability in trace-supervised settings by rolling out-driven reflection.
Outcome:	The proposed framework outperforms baselines and reflective prompt optimizers by 5%–20% on OSR.

Copied to clipboard

Midhat Urooj, Ayan Banerjee, Sandeep Gupta

Challenge:	Domain generalization and long-tailed (LT) learning models face two challenges . domain invariance often suppresses class-discriminative signals essential for long-tail recognition.
Approach:	They propose a framework that disentangles domain-invariant and class-discriminative features . they evaluate 10 diverse medical imaging datasets spanning four modalities .
Outcome:	The proposed framework achieves an average performance improvement of 10.3% across multi-domain and in-domain long-tailed tasks while preserving minority class performance.

AdaptEvolve: Improving Efficiency of Evolutionary AI Agents through Adaptive Model Selection (2026.findings-acl)

Copied to clipboard

Pretam Ray, Pratik Prabhanjan Brahma, Zicheng Liu, Emad Barsoum

OmniCode: A Benchmark for Evaluating Software Development Agents (2026.findings-acl)

Challenge:	Existing routing strategies rely on static heuristics or external controllers to optimize performance.
Approach:	They propose a framework that leverages intrinsic generation confidence to estimate solvability.
Outcome:	Empirical results show that confidence-driven selection yields favorable Pareto frontier . computational cost of state-of-the-art large language models remains a key barrier to scalable deployment .

Copied to clipboard

Atharv Sonwane, Eng-Shen Tu, Wei-Chung Lu, Claas Beger, Carter Larsen, Debjit Dhar, Simon Alford, Rachel Chen, Ronit Pattanayak, Tuan Anh Dang, Guohao Chen, Gloria Geng, Kevin Ellis, Saikat Dutta

Supplement Generation Training for Enhancing Agentic Task Performance (2026.findings-acl)

Challenge:	popular coding benchmarks focus on narrowly scoped tasks such as competition programming and patch generation.
Approach:	They propose a software engineering benchmark that aims to provide a broader set of tasks beyond code or patch generation.
Outcome:	The proposed framework performs well on bug fixing for Python, test generation, code review fixing, and style fixing with popular agent frameworks such as SWE-Agent.

Copied to clipboard

Young Min Cho, Daniele Bonadiman, Divya Bhargavi, Tamer Alkhouli, Salvatore Romeo, Dongwei Jiang, Khushbu Pahwa, Yubin Ge, Etsuko Ishii, Monica Sunkara, Yi Zhang

Challenge:	Training large foundation models for agentic tasks is impractical due to high computational costs, long iteration cycles, and rapid obsolescence as new models are released.
Approach:	They propose a method that trains a small LLM to generate supplemental text that helps the larger LLM solve the task more effectively.
Outcome:	The proposed approach decouples task-specific optimization from large foundation models . it achieves consistent and significant performance gains across diverse tasks and models - all without gradient access to the actor model.

Automatic Paper Analysis and Categorisation for Systematic Reviews with Combined Reasoning-Augmented SFT and DAPO RL (2026.findings-acl)

Copied to clipboard

Michela Lorandi, Anya Belz, Simon Mille, Craig Thomson

MotifAgent: Learning Molecular Assembly through Multi-Agent Collaboration for Chemical Language Understanding (2026.findings-acl)

Challenge:	Automating systematic reviews is expensive and time consuming, a study finds . automatic approaches are being explored but their performance has been poor .
Approach:	They propose to use reasoning-enhanced fine-tuning and DAPO reinforcement learning to automate systematic reviews.
Outcome:	The proposed methods significantly improve the performance of LLMs, the authors find . they find that reasoning-enhanced fine-tuning reduces time required for annotation by 80% .

Copied to clipboard

Jinjia Feng, Wenda Wang, Zhewei Wei

Challenge:	Existing approaches to molecular understanding are limited to static motif recognition without understanding connection rules governing how motifs assemble into valid topological structures.
Approach:	They propose a multi-agent reinforcement learning framework inspired by emergent collective intelligence to solve a problem where each motif is represented by an agent sharing a common LLM backbone.
Outcome:	Extensive experiments show that the proposed framework surpasses specialized expert models in molecular understanding tasks.

Policy-Guided Stepwise Action Planning for Controllable LLM Reasoning (2026.findings-acl)

Copied to clipboard

Jianpeng Zhou, Qisheng Hu, Jiahai Wang, Wenya Wang

Multiplication in Multimodal LLMs: Computation with Text, Image, and Audio Inputs (2026.findings-acl)

Challenge:	Existing approaches to steering large language model reasoning via high-level reasoning actions fail to outperform standard generation because planners tend to degenerate into repetitive loops or fixed patterns.
Approach:	They propose a planner-executor framework that learns to select reasoning actions dynamically while keeping the executor LLM fully frozen.
Outcome:	The proposed framework outperforms existing paradigms by preserving the executor LLM frozen . PG-HAP improves accuracy over strong baselines while producing less redundant, more adaptive trajectories.

Copied to clipboard

Samuel Gideon Balter, Ethan Jerzak, Connor Thomas Jerzak

Modeling Human Perspectives with Socio-Demographic Representations (2026.findings-acl)

Challenge:	Existing benchmarks lack systematically paired instances across modalities, making it difficult to compare genuine arithmetic limits . a model that computes 4736 may fail on a nearby instance like 8967, despite a well-tuned internal router.
Approach:	They propose a controlled multimodal multiplication benchmark that factorially varies digit length, digit sparsity, representation, and modality with paired instances from a reproducible generator.
Outcome:	The proposed model can perceive numerical content across modalities but fails to perform exact multi-digit multiplication when presented as numerals, number words, images, or in audio form.

Copied to clipboard

Leixin Zhang, Cagri Coltekin

Weight Tying Biases Token Embeddings Towards the Output Space (2026.findings-acl)

Challenge:	Recent studies show that human disagreement is widespread across many annotation tasks.
Approach:	They propose a method that jointly models annotator perspectives while learning socio-demographic representations.
Outcome:	The proposed method outperforms concatenation-based methods in predicting annotator perspectives . it learns socio-demographic representations and analyzes how demographic factors relate to variation .

Copied to clipboard

Antonio Lopardo, Avyukth Harish, Catherine Arnett, Akshat Gupta

Structured Uncertainty guided Clarification for LLM Agents (2026.findings-acl)

Challenge:	Weight tying is a common practice in language model design, but its impact on learning embedding space remains unclear.
Approach:	They show that weight tying optimizes the embedding matrix for output prediction . they also show that tied embeddable matrices align more closely with output embedders .
Outcome:	The proposed weight tying approach harms performance at scale and has implications for training smaller LLMs.

Copied to clipboard

Manan Suri, Puneet Mathur, Nedim Lipka, Franck Dernoncourt, Ryan A. Rossi, Dinesh Manocha

LRBench and Judge-R1: Principled Evaluation and Training of LLM-Based Judges for Long-Context Reasoning (2026.findings-acl)

Challenge:	Existing approaches to clarifying tasks fail when user instructions are ambiguous or incomplete.
Approach:	They propose a principled formulation of structured uncertainty that operates directly over tool parameters and their domains.
Outcome:	The proposed framework improves when2call accuracy and training-time sample efficiency.

Copied to clipboard

Xinyi Zhao, Haoqi Hu, Ziyu Wang, Jinfeng Xiao

Beyond Reproduction: A Paired-Task Framework for Assessing LLM Comprehension and Creativity in Literary Translation (2026.findings-acl)

Challenge:	Existing benchmarks for evaluating large language models (LLMs) under long contexts are underexplored.
Approach:	They propose a large-scale benchmark for evaluating large language models (LLMs) that combines reinforcement learning with multi-turn search to enable grounded and principle-aware evaluation.
Outcome:	The proposed model outperforms single-turn baselines across domains and principles.

Copied to clipboard

Ran Zhang, Steffen Eger, Arda Tezcan, Wei Zhao, Simone Paolo Ponzetto, Lieve Macken

Mechanistic Interpretability of Large-Scale Counting in LLMs through a System-2 Strategy (2026.findings-acl)

Challenge:	Large language models (LLMs) are increasingly used for creative tasks such as literary translation.
Approach:	They propose a paired-task framework that assesses translational creativity using Units of Creative Potential (UCPs) they benchmark 23 models and four creativity-oriented prompts to assess translational comprehension .
Outcome:	The proposed framework compares 23 models and four creativity-oriented prompts on literary excerpts from 11 books.

Copied to clipboard

Hosein Hasani, Mohammadali Banayeeanzade, Ali Nafisi, Sadegh Mohammadian, Fatemeh Askari, Mobin Bagherian, Amirmohammad Izadi, Mahdieh Soleymani Baghshah

CodeScout: Contextual Problem Statement Enhancement for Software Agents (2026.findings-acl)

Challenge:	Large language models exhibit systematic limitations in counting tasks due to depth constraints.
Approach:	They propose a method that decomposes large counting tasks into smaller, independent sub-problems that the model can reliably solve.
Outcome:	The proposed method surpasses architectural limitations and achieves higher accuracy on large-scale counting tasks.

Copied to clipboard

Manan Suri, Xiangci Li, Mehdi Shojaie, Songyang Han, Chao-Chun Hsu, Shweta Garg, Aniket Anand Deshmukh, Varun Kumar

Challenge:	Current AI-powered code assistance tools struggle with ambiguous problem statements . failures on such ambiguously requests are highly correlated with longer trajectories .
Approach:	They propose a contextual query refinement approach that transforms ambiguous user requests into comprehensive, actionable problem statements through lightweight pre-exploration of the target codebase.
Outcome:	Empirical results show that CodeScout improves resolution rates with 27 additional issues resolved compared to baseline method.

Evaluating Scene-based In-Situ Item Labeling for Immersive Conversational Recommendation (2026.findings-acl)

Copied to clipboard

Jiazhou Liang, Yifan Simon Liu, David Guo, Yilun Jiang, Minqi Sun, Scott Sanner

StanceAttack: Adversarial Attack for Stance Detection (2026.findings-acl)

Challenge:	Existing methods for evaluating item labels fail to leverage scenario-specific information modalities, present redundant information that is visually inferable, and lack latent awareness of users' information needs.
Approach:	They propose a principled categorization of information needs into explicit intent satisfaction and proactive information needs and define evaluation metrics for item label selection.
Outcome:	The proposed evaluation framework is based on IR-, LLM-, and VLM-based methods across fashion, movie recommendation, and retail shopping scenarios.

Copied to clipboard

Chenye Zhao, Cornelia Caragea

MEAV: Model Editing with Alignment Vectors for inference time LLM alignment in single and multidomain preference spectrum (2026.findings-acl)

Challenge:	pretrained language models (PLMs) have greatly enhanced stance detection, but they remain vulnerable to adversarial attacks.
Approach:	They propose an adversarial attack method that uses ChatGPT to create adversarials that can mislead well-trained stance detection models.
Outcome:	The proposed method outperforms existing adversarial methods with higher success rates and fewer retries on two benchmark datasets.

Copied to clipboard

Sadat Shahriar, Zheng Qi, Nikolaos Pappas, Srikanth Doss, Kishaloy Halder, Monica Sunkara, Manuel Mager, Yassine Benajiba

BioVLM: Routing Prompts, Not Parameters, for Cross-Modality Generalization in Biomedical VLMs (2026.findings-acl)

Challenge:	Existing training-time alignment methods require full retraining when a change is needed.
Approach:	They propose an inference-time model-editing-based alignment method that learns encoded representations of preference dimensions and allows dynamic adjusting of the model behavior.
Outcome:	The proposed method can be used to align large language models to human preferences . it reduces the cost of inference by half compared to the prompt engineering approach .

Copied to clipboard

Mainak Singha, Tanisha Gupta, Ankit Jha, Muhammad Haris Khan, Sayantani Ghosh, Biplab Banerjee

SCALE: Upscaled Continual Learning of Large Language Models (2026.findings-acl)

Challenge:	Pretrained biomedical vision–language models perform well on average but often degrade on challenging modalities.
Approach:	They propose a prompt-learning framework that improves cross-domain generalization without extensive backbone fine-tuning.
Outcome:	BioVLM learns a diverse prompt bank and introduces dynamic prompt selection . it can combine sparse few-shot evidence with rich LLM semantic priors . bioVLM achieves state-of-the-art on 11 MedMNIST+ 2D datasets based on the proposed framework .

Copied to clipboard

Jin-woo Lee, Junhwa Choi, Bongkyu Hwang, Jinho Choo, Bogun Kim, Jeongseon Yi, Joonseok Lee, DongYoung Jung, Jaeseon Park, Kyoungwon Park, Suk-hoon Jung

Words that make SENSE: Sensorimotor Norms in Learned Lexical Token Representations (2026.findings-acl)

Challenge:	Recent discussions suggest that further progress will come from scaling the right structure, not merely parameters or data, while preserving acquired knowledge.
Approach:	They propose a width upscaling architecture that inserts lightweight expansions into linear modules while freezing all pre-trained parameters.
Outcome:	The proposed architecture reduces severe forgetting while learning new knowledge on a controlled synthetic biography benchmark.

Copied to clipboard

Abhinav Gupta, Toben Mintz, Jesse Thomason

Interpreting Style Representations via Style-Eliciting Prompts (2026.findings-acl)

Challenge:	Empirical studies suggest that comprehending action, perceptual and abstract concepts elicits rapid, automatic activity in modality-specific brain areas.
Approach:	They propose a model that predicts Lancaster sensorimotor norms from word lexical embeddings.
Outcome:	The proposed model predicts Lancaster sensorimotor norms from word lexical embeddings.

Copied to clipboard

Junghwan Kim, David Jurgens

Optimizing Conversational Quality in Spoken Dialogue Systems with Reinforcement Learning from AI Feedback (2026.findings-acl)

Challenge:	Recent work has attempted to explain learning of style representations by generating natural language descriptions with large language models (LLMs) conditioned on input text.
Approach:	They propose a framework for interpreting style representations through style-eliciting prompts by prompting an LLM to generate text conditioned on these features.
Outcome:	The proposed framework outperforms baselines that directly prompt LLMs with target text, and achieves superior performance in both style description and style imitation.

Copied to clipboard

Siddhant Arora, Jinchuan Tian, Jiatong Shi, Hayato Futami, Yosuke Kashiwagi, Emiru Tsunoo, Shinji Watanabe

Conceptual Hierarchies within LLMs (2026.findings-acl)

Challenge:	Existing studies on reinforcement learning from human or AI feedback have focused on semantic rewards at the utterance level.
Approach:	They propose a multi-reward RLAIF framework for speech-in/speech-out dialogue systems . they combine semantic, audio-quality, and emotion-consistency rewards .
Outcome:	The proposed framework improves speech-in/speech-out dialogue system quality . it combines semantic, audio-quality, and emotion-consistency rewards . the proposed framework is available to download from the cdc.

Copied to clipboard

Tiago Almeida, Zining Zhu, Yue Ning

Challenge:	Existing literature has explored abstraction within large language models (LLMs).
Approach:	They generate a dataset of semantic hierarchies and investigate their storage locations in six LLMs using activation patching, a causal intervention technique.
Outcome:	The results show that concepts at finer levels of granularity are stored around 61-78% of the time before those at coarser levels.

Conjecture and Inquiry: Quantifying Software Performance Requirements via Interactive Retrieval-Augmented Preference Elicitation (2026.findings-acl)

Copied to clipboard

Shihai Wang, Tao Chen

Measuring Large Language Models’ Adversarial Behavior in Social Deduction Games (2026.findings-acl)

Challenge:	Existing methods to quantify software performance requirements are vague and imprecise.
Approach:	They propose an approach that quantifies software performance requirements into mathematical functions via interactive retrieval-augmented preference elicitation.
Outcome:	The proposed method improves on four real-world cases with up to 40x improvements under as few as five rounds of interactions.

Copied to clipboard

Marissa Zhao Li, Esha Shivakumar, Peiran Wang, Ying Li, Yuan Tian

ChangJuan: A Comprehensive Benchmark for Book-Length Chinese Story Evaluation (2026.findings-acl)

Challenge:	Existing safety evaluations focus on refusal-based methods that test whether models avoid responding to inappropriate or violent requests, leaving open questions about how models behave in interactive social settings.
Approach:	They propose to use a meta-LLM to construct a closed behavioral taxonomy from a multi-agent simulation to examine adversarial behavior of large language models.
Outcome:	The proposed model-based model-driven model-model-based taxonomy shows that the model-led model-learning model exhibits distinct behavioral profiles and influences social stability and competitive success.

Copied to clipboard

Dingyi Yang, Mingshuo Wang, Qin Jin

Progressive Re-ranking for Multimodal Retrieval-Augmented Generation via Curriculum Learning (2026.findings-acl)

Challenge:	Recent advances in Large Language Models (LLMs) have significantly enhanced the capacity of Automatic Story Evaluation.
Approach:	They propose a method to distill raw reviews into generally agreed viewpoints across key evaluation aspects such as plot and character.
Outcome:	The proposed model outperforms open-source baselines and raises Qwen3’s Kendall’s tau correlation with human judgments from 24.8 to 34.1.

Copied to clipboard

Zhu Min, Yanchao Hao, Jian Liu, Shizhu He, Xi Chen

Confidence-Weighted Token Set Cover for Early Hypothesis Pruning in Self-Consistency (2026.findings-acl)

Challenge:	Existing approaches to improve retrieval performance of large language models are limited by static knowledge.
Approach:	They propose a multimodal re-ranking framework that combines curriculum learning with fine-grained reranking and multimodal section reassessment to improve CLIP-based visual coarse-grain retrieval.
Outcome:	The proposed framework achieves state-of-the-art answer accuracy and competitive retrieval performance on InfoSeek and Enc-VQA.

Copied to clipboard

Md Arafat Sultan, Ramón Fernandez Astudillo

MMTabReal: Real-World Benchmark for Multimodal Table Understanding (2026.findings-acl)

Challenge:	Despite its simplicity and efficacy, the high token expenditure of self-consistency can limit its practical utility.
Approach:	They propose a fast weighted set cover algorithm that utilizes two lightweight indicators to prune intermediate hypotheses periodically.
Outcome:	The proposed method reduces self-consistency token efficiency by 10-35% on three math benchmarks.

Copied to clipboard

Prasham Yatinkumar Titiya, Jainil Trivedi, Chitta Baral, Vivek Gupta

Inject to Heal: Alleviating hallucination in LVLMs via Context Embedding Injection (2026.findings-acl)

Challenge:	Multimodal tables are ubiquitous in real applications but are difficult to evaluate in multimodal large language models.
Approach:	They propose a multimodal table benchmark that compares 500 real-world tables with 4021 question–answer pairs.
Outcome:	MMtabReal spans four question types, five reasoning categories, and eight structural archetypes.

Copied to clipboard

Mehrdad Fazli, Bowen Wei, Ziwei Zhu

Colorful Talks with Graphs: Human-Interpretable Graph Encodings for Large Language Models (2026.findings-acl)

Challenge:	a large vision-language model can generate hallucinations inconsistent with visual input . a lightweight method that embeds the last input token as a grounding signal reduces the likelihood of hallucinosity.
Approach:	They propose a training-free mitigation strategy that harnesses the hidden state of the last input token as a grounding signal to maintain visual fidelity throughout decoding and curb hallucinations.
Outcome:	The proposed method outperforms state-of-the-art methods on CHAIR, AMBER, and MMHal benchmarks.

Copied to clipboard

Angelo Zangari, Peyman Baghershahi, Sourav Medya

SAM-NER: Semantic Archetype Mediation for Zero-Shot Named Entity Recognition (2026.findings-acl)

Challenge:	Graph problems require reasoning over explicit structure, permutation invariance, and computationally complex relationships, creating a mismatch with the representations of text-based models.
Approach:	They propose a human-interpretable structural encoding strategy that injects graph structure directly into natural language prompts.
Outcome:	The proposed method improves performance on synthetic and real-world datasets.

Copied to clipboard

Ruichu Cai, Juntao Gan, Miao Mai, Zhifeng Hao, Boyan Xu

DRPG (Decompose, Retrieve, Plan, Generate): An Agentic Framework for Academic Rebuttal (2026.findings-acl)

Challenge:	Zero-shot Named Entity Recognition (ZS-NER) remains brittle under domain and schema shifts, where unseen label definitions misalign with a large language model’s intrinsic semantic organization.
Approach:	They propose a framework that stabilizes cross-domain transfer through an intermediate, domain-invariant archetype space.
Outcome:	Experiments on the CrossNER benchmark show that SAM-NER consistently outperforms strong prior ZS-NER baselines in cross-domain settings.

Copied to clipboard

Peixuan Han, YingJie Yu, Jingjun Xu, Jiaxuan You

BiasGRPO: Stabilizing Bias Mitigation in High-Variance Reward Landscapes via Group-Relative Policy Optimization (2026.findings-acl)

Challenge:	Existing approaches to support academic rebuttal rely on off-the-shelf LLMs or simple pipelines that struggle with long-context understanding.
Approach:	They propose an agentic framework for automatic academic rebuttal generation that operates through four steps: Decompose reviews into atomic concerns, Retrieve relevant evidence from the paper, Plan refortations, and Generate responses accordingly.
Outcome:	The proposed framework outperforms existing rebuttal pipelines and achieves 98% accuracy beyond the average human level using only an 8B model.

Copied to clipboard

Saket Reddy, Ke Yang, ChengXiang Zhai

Challenge:	Recent preference-based fine-tuning methods have limited exploration in offline training . previous methods have been limited by the lack of exploration inherent in offline learning .
Approach:	They propose a method that normalizes rewards across a group of completed tasks to mitigate social bias in Large Language Models.
Outcome:	The proposed approach outperforms DPO and PPO in multiple benchmarks . it can overcome limitations of previous preference-based methods .

MaskTab: Scalable Masked Tabular Pretraining with Scaling Laws and Distillation for Industrial Classification (2026.findings-acl)

Copied to clipboard

Bo Zheng, Yudong Chen, Zihua Xiong, Shuai Fang, Peidong He, Yang Yang, Sheng Guo

Enhancing Zero-Shot Time Series Forecasting in Off-the-Shelf LLMs via Noise Injection Prompting (2026.findings-acl)

Challenge:	Tabular data is high-dimensional, riddled with missing entries, and rarely labeled at scale.
Approach:	They propose a unified pre-training framework for industrial-scale tabular data . MaskTab encodes missing values via dedicated learnable tokens .
Outcome:	The proposed framework outperforms XGBoost and MaskTab-L on industrial-scale . it achieves +5.04% AUC and +8.28% KS over prior art under rigorous scaling .

Copied to clipboard

Xingyou Yin, Ceyao Zhang, Min Hu, Kai Chen

Text Embedding as Treatment: A Meta Causal Approach for Robust Sentiment Classification (2026.findings-acl)

Challenge:	Existing work relies on fine-tuning specialized modules to bridge this gap, but a novel approach is proposed to leverage off-the-shelf LLMs without any fine- tuning whatsoever.
Approach:	They propose a method to inject noise into the raw time series before tokenization to induce the model to extrapolate based on robust underlying temporal patterns rather than superficial numerical artifacts.
Outcome:	The proposed approach overcomes the brittleness of fully frozen models by injecting noise into the raw TS before tokenization.

Copied to clipboard

Fengxiang Cheng, Chuan Zhou, Xiang Li, Haoxuan Li, Wen-li Wang, Jinkun Chen, Mingming Gong, Kun Zhang

CliniCAST: Benchmarking Acoustic Grounding and Text Dominance in Medical Triage (2026.findings-acl)

Challenge:	Existing methods for sentiment classification use binary treatment of words . Existing approaches limit generalizability to novel words and low-frequency words if there is a word in a sentence that is not treated .
Approach:	They propose a meta-causal approach that uses a single training task to identify causal words for arbitrary words.
Outcome:	The proposed method reduces the spurious correlation between word treatment and sentiment classification by removing words with low treatment effects from a pre-trained language model.

Copied to clipboard

Kyusik Kim, Hyunwoo Yoo, Jaehoon Choi, Kitae Kim, Gail Rosen, Bongwon Suh

Whose Voice, Whose Avatar? Gender Matching Bias in Multimodal AI Teammates (2026.findings-acl)

Challenge:	Recent Large Audio-Language Models (LALMs) integrate acoustic capabilities into reasoning, yet whether they reliably ground clinical judgments in audible evidence remains unproven.
Approach:	They propose a benchmark that disentangles clinically meaningful acoustic cues from lexical content and speaker demographics.
Outcome:	Evaluating 5,856 synthetic samples across 12 disease conditions, the proposed model exhibits fragile acoustic grounding and pronounced "text dominance" failure mode.

Copied to clipboard

Kyusik Kim, Jaehoon Choi, Hyunwoo Yoo, Bongwon Suh

WebAnchor: Anchoring Agent Planning to Stabilize Long-Horizon Web Reasoning (2026.findings-acl)

Challenge:	Multimodal Large Language Models are increasingly deployed as social agents . yet their ability to integrate conflicting identity cues remains underexplored .
Approach:	They audit gender bias in MLLMs that pair synthetic voices with avatars of varying gender presentation and visual fidelity.
Outcome:	The findings show that multimodal fairness is not monolithic . they show that models may appear unbiased on one dimension while enforcing stereotypes on another .

Copied to clipboard

Xinmiao Yu, Liwen Zhang, Xiaocheng Feng, Pengjun Xie, Jingren Zhou, Bing Qin, Yong Jiang

Context-Driven and Reference-Guided Data Augmentation for Subtitle Translation (2026.findings-acl)

Challenge:	Existing methods for reinforcement learning (RL)-based agents struggle with long-horizon planning and strategy coherence.
Approach:	They propose a reinforcement learning framework that decouples planning and execution.
Outcome:	The proposed framework outperforms baseline and first-step RL frameworks on four benchmarks.

Copied to clipboard

Hitoshi Ito, Naoto Shirai, Kazutaka Kinugawa, Hideya Mino, Rei Endo, Yoshihiko Kawai

Real or Robotic? Assessing Whether LLMs Accurately Simulate Qualities of Human Responses in Human-LLM Dialogue (2026.findings-acl)

Challenge:	Large language models (LLMs) have demonstrated strong performance in translation tasks.
Approach:	They propose a method that expands source-side data by rewriting original subtitles using information that can be extracted from the context, such as character profiles and scene descriptions.
Outcome:	The proposed method improves BLEU scores for film subtitle translation and achieves superior stylistic quality in human evaluation.

Copied to clipboard

Jonathan Ivey, Shivani Kumar, Jiayu Liu, Hua Shen, Sushrita Rakshit, Rohan Raju, Haotian Zhang, Aparna Ananthasubramaniam, Junghwan Kim, Bowen Yi, Dustin Wright, Abraham Israeli, Anders Giovanni Møller, Lechen Zhang, David Jurgens

Can Small LLMs Learn a Robust Theory of Mind via RLVR? Investigating Generalization through the False-Belief Task (2026.findings-acl)

Challenge:	Recent work has sought to use large language models to simulate human-human and human-LLM interactions.
Approach:	They use a large-scale dataset to generate a paired LLM-LLM and human-LLm dialogues from the WildChat dataset and quantify how well they align with their human counterparts.
Outcome:	The proposed models perform similarly in simulating English, Chinese, and Russian dialogues.

Copied to clipboard

Sneheel Sarangi, Hanan Salam

Rolling Out Data Quality Overnight, without losing the plot: A Multi-Agent System for Speech Data Quality Management (2026.findings-acl)

Challenge:	Recent advances in large language models (LLMs) have demonstrated emergent capabilities in complex reasoning, largely spurred by rule-based Reinforcement Learning (RL) techniques applied during post-training.
Approach:	They evaluate whether small-scale LLMs can acquire a robust and generalizable Theory of Mind (ToM) capability through RL with verifiable rewards.
Outcome:	The proposed model performs well on in-distribution tasks but fails to transfer to unseen ToM tasks with different characteristics.

Copied to clipboard

Rishabh Kumar, Abhinav Painuli, Chriss Philip Saji, Devesh Soni, Amrith Krishna, Ganesh Ramakrishnan

Schema-Guided Response Generation using Multi-Frame Dialogue State for Motivational Interviewing Systems (2026.findings-acl)

Challenge:	Using automation to improve quality management is expensive and resource-intensive for speech datasets.
Approach:	They propose a natural language-driven agentic framework that compiles user requirements into dependency-aware DAG workflows over modular tools for audio, transcript, and metadata verification.
Outcome:	The proposed framework achieves 80-90% agreement with expert verification while requiring less than 20% of the cost and time of manual QC.

Copied to clipboard

Jie Zeng, Yukiko Nakano

From Passive Metric to Active Signal: The Evolving Role of Uncertainty Quantification in Large Language Models (2026.findings-acl)

Challenge:	Motivational interviewing (MI) is a goal-directed dialogue aimed at motivating clients to change their behavior.
Approach:	They propose a method for updating multi-frame dialogue states and a strategy decision mechanism that dynamically determines the response focus in a manner grounded in MI principles.
Outcome:	The proposed method generates responses aligned with MI principles and frequently asks questions to elicit change talk.

Copied to clipboard

Jiaxin Zhang, Wendi Cui, Zhuohang Li, Lifu Huang, Bradley A. Malin, Caiming Xiong, Chien-Sheng Wu

Value–Action Alignment in Large Language Models under Privacy–Prosocial Conflict (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have remarkable capabilities, but unreliability remains a barrier to deployment in high-stakes domains.
Approach:	They propose to transform uncertainty from a passive diagnostic metric to an active control signal guiding real-time model behavior.
Outcome:	The proposed model evolution from passive diagnostic metric to active control signal is critical for high-stakes applications.

Copied to clipboard

Guanyu Chen, Chenxiao Yu, Xiyang Hu

See or Say Graphs: Agent-Driven Scalable Graph Understanding with Vision-Language Models (2026.findings-acl)

Challenge:	Existing evaluations measure privacy-related attitudes or sharing intentions in isolation, making it difficult to determine whether a model’s expressed values jointly predict its downstream data-sharing actions as in real human behaviors.
Approach:	They propose a framework that sequentially administers standardized questionnaires for privacy attitudes, prosocialness, and acceptance of data sharing within a bounded, history-carrying session.
Outcome:	The proposed model shows that it is stable but model-specific Privacy-PSA-AoDS profiles, and substantial heterogeneity in value-action alignment.

Copied to clipboard

Shuo Han, Yukun Cao, Zezhong Ding, Zengyi Gao, S Kevin Zhou, Xike Xie

Recovered in Translation: Efficient Pipeline for Automated Translation of Benchmarks and Datasets (2026.findings-acl)

Challenge:	Existing studies have explored textual graph descriptions and visual modalities for VLMs to understand graphs.
Approach:	They propose a unified framework that enhances both scalability and modality coordination in graph understanding by integrating textual and visual modalities.
Outcome:	GraphVista scales to large graphs, 200 larger than those used in existing benchmarks, and consistently outperforms existing textual, visual, and fusion-based methods.

Copied to clipboard

Hanna Yukhymenko, Anton Alexandrov, Martin Vechev

MorphBPE: Morphology-Aware Tokenization for Efficient LLM Training (2026.findings-acl)

Challenge:	Existing benchmarks suffer from semantic drift and context loss, which can lead to misleading performance metrics.
Approach:	They propose a fully automated framework to enable translation of large language models . they propose to use universal self-improvement and multi-round ranking methods to improve translation quality .
Outcome:	The proposed framework surpasses existing benchmarks in eight languages and improves translation quality across multilingual domains.

Copied to clipboard

Ehsaneddin Asgari, Yassine El Kheir, MohammadAli SadraeiJavaheri, Ali Nazari

From Storage to Experience: A Survey on the Evolution of LLM Agent Memory Mechanisms (2026.findings-acl)

Challenge:	Tokenization is a key design choice in modern NLP systems and a critical bottleneck for multilingual Large Language Models.
Approach:	They propose a tokenization extension that constrains merge operations to respect morpheme boundaries while preserving inference.
Outcome:	The proposed tokenization improves morphological coherence and language model cross-entropy in four languages.

Copied to clipboard

Jinghao Luo, Yuchen Tian, Chuxue Cao, Ziyang Luo, Hongzhan Lin, Kaixin Li, Chuyi Kong, Ruichao Yang, Jing Ma

Challenge:	Large Language Models (LLMs)-based agents have fundamentally reshaped artificial intelligence . however, the inherent statelessness of LLMs hinders their ability to maintain logical consistency across complex, multi-step tasks .
Approach:	They propose a framework for LLM agent memory mechanisms that formalizes the development process into three stages: storage, reflection, and experience.
Outcome:	The proposed framework breaks the development process into three stages . it analyzes the need for long-range consistency, challenges in dynamic environments, and the ultimate goal of continual learning.

LLMs are Brittle to Simple Code Transformations: Introducing CETBench – A Benchmark for Code-Equivalence Checking (2026.findings-acl)

Copied to clipboard

Neeva Oza, Ishaan Govil, Parul Gupta, Dinesh Khandelwal, Dinesh Garg, Parag Singla

Enhancing Factuality through Consensus and Consistency in Summarization Using Minimum Bayes Risk Decoding (2026.findings-acl)

Challenge:	a new benchmarking tool for code equivalence checks the performance of LLMs.
Approach:	They propose a code-equivalence with transformations benchmark built from a repository of programs that may solve the same or different tasks.
Outcome:	The proposed approach boosts performance on the transformed pairs of programs.

Copied to clipboard

Riza Setiawan Soetedjo, Yusuke Sakai, Hidetaka Kamigaito, Jingun Kwon, Manabu Okumura, Taro Watanabe

Moneyball with LLMs: Analyzing Tabular Summarization in Sports Narratives (2026.findings-acl)

Challenge:	Existing methods to improve summarization quality are limited to using source as guidance . reranking can be effective, but there are limitations, such as relying on reference-free metrics and rely on a single metric.
Approach:	They propose a model that reranks model-generated summaries by considering consistency to the source document and consensus among the other candidates.
Outcome:	The proposed system is competitive with existing methods, with human evaluations further confirming that it is superior.

Copied to clipboard

Ritam Upadhyay, Naman Ahuja, Rishabh Baral, Aparna Garimella, Vivek Gupta

A Computational Method for Measuring Open Codes in Qualitative Analysis (2026.findings-acl)

Challenge:	Large language model (LLM) approaches to tabular summarization rely on prompt engineering, decomposition pipelines, or entity-level intermediate representations to achieve strong performance.
Approach:	They propose a diagnostic benchmark for long-context tabular summarization using decomposition pipelines and entity-level intermediate representations.
Outcome:	The proposed benchmark improves accuracy and numerical fidelity, but lacks local arithmetic.

Copied to clipboard

John Chen, Alexandros Nikolaos Lotsos, Sihan Cheng, Lexie Zhao, Yanjia Zhang, Jessica Hullman, Bruce Sherin, Uri Wilensky, Michael Horn

Vulnerability of LLMs’ Stated Belief? LLMs Belief Resistance Check Through Strategic Persuasive Conversation Interventions (2026.findings-acl)

Challenge:	Qualitative analysis is widely adopted across many social science disciplines.
Approach:	They propose a theory-informed computational method for measuring inductive coding results from humans and GAI.
Outcome:	The proposed method captures breadth, consensus, unique contribution, and systematic deviation without assuming ground truth.

Copied to clipboard

Fan Huang, Haewoon Kwak, Jisun An

Choose Your Lens: Multi-Perspective Value Alignment of Chain-of-Thought Reasoning (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are increasingly employed in question-answering tasks.
Approach:	They analyze how different persuasive strategies influence stated belief stability . they also examine whether verbalized confidence prompting increases vulnerability .
Outcome:	The proposed model exhibits extreme compliance, with 82.5% of belief changes occurring at the first persuasive turn.

Copied to clipboard

Gejian Zhao, Hanzhou Wu, Xinpeng Zhang

Beyond Evidence: Belief-Chain Conditioning for Persuasive Misinformation Debunking Explanation (2026.findings-acl)

Challenge:	Large language models tend to hallucinate “convenient” facts to forcefully justify stances . current methods often induce motivated reasoning, causing factual hallucinations .
Approach:	They propose a neuro-symbolic framework that enables steerable pluralism without distorting objective reality by projecting generated CoT paths onto a multi-perspective graph.
Outcome:	The proposed approach reduces factual hallucinations by 3 and improves cross-perspective consistency by 25% compared to standard steerable baselines, paving the way for trustworthy pluralistic AI.

Copied to clipboard

Yi-Li Hsu, Li-Wun Chang, Chun-Yu Hsu, Wei-Kuan Shih, Aiping Xiong, Lun-Wei Ku

CoTJudger: A Graph-Driven Framework for Automatic Evaluation of Chain-of-Thought Efficiency and Redundancy in LRMs (2026.findings-acl)

Challenge:	Existing methods to misinformation correction focus on relying on audience beliefs to generate factually accurate responses and to engage with users' mental states.
Approach:	They construct large language models with cognitive chains and use them to model their outputs on beliefs that engage with users' mental states.
Outcome:	The proposed model improves explanation quality for audiences with misinformation-aligned beliefs by incorporating believers’ chains into the model.

Copied to clipboard

Siyi Li, Jiajun Shi, Shiwen Ni, Ge Zhang, Shuaimin Li, Shijian Wang, Zhoufutu Wen, Yizhi LI, Hamid Alinejad-Rokny, Jiaheng Liu, Min Yang, Wenhao Huang

The Inner Monologue of Language Models: When Reasoning Traces Reveal More Than They Hide (2026.findings-acl)

Challenge:	Existing evaluations emphasize final accuracy or coarse token counts, and lack automated tools to separate essential logic from structural redundancy.
Approach:	They propose a graph-driven framework that quantifies reasoning efficiency by converting free-form CoTs into directed dependency graphs and extracting the Shortest Effective Path needed to reach a correct solution.
Outcome:	Evaluating 21 LRMs, the proposed framework quantifies reasoning efficiency by converting free-form CoTs into directed dependency graphs and extracting the Shortest Effective Path (SEP) needed to reach a correct solution.

Copied to clipboard

Pratham Singla, Shivank Garg, Ayush Singh, Ishan Garg, Ketan Suhaas Saichandran

Do VLMs Have a Moral Backbone? A Study on the Fragile Morality of Vision-Language Models (2026.findings-acl)

Challenge:	Recent advances in large language models have enabled them to tackle complex tasks . a fundamental question is: are these models aware of what they "learn" and "think"?
Approach:	They define three core competencies: awareness of learned latent policies, generalization of these policies across domains, alignment between internal reasoning traces and final outputs.
Outcome:	The results show that RL-trained models exhibit stronger generalizability to novel tasks than SFT models but weak alignment between reasoning traces and final outputs.

Copied to clipboard

Zhining Liu, Tianyi Wang, Xiao Lin, Penghao Ouyang, Gaotang Li, Ze Yang, Hui Liu, Sumit Keswani, Vishwa Pardeshi, Huijun Zhao, Wei Fan, Hanghang Tong

ENPMR-Bench: Benchmarking Proactive Memory Retrieval for Emotional Support Agents (2026.findings-acl)

Challenge:	Vision-Language Models (VLMs) have advanced multimodal learning, driving progress in cross-modal reasoning.
Approach:	They propose to examine moral robustness of vision-language models by analyzing their moral stances under multimodal perturbations.
Outcome:	The proposed model-agnostic multimodal perturbations expose VLMs to a variety of moral vulnerabilities, including a sycophancy trade-off where stronger instruction-following models are more susceptible to persuasion.

Copied to clipboard

Xing Fu, Yulin Hu, Mengtong Ji, Haozhen Li, Yixin Sun, Weixiang Zhao, Yanyan Zhao, Bing Qin

Grounded Concreteness: Human-Like Concreteness Sensitivity in Vision–Language Models (2026.findings-acl)

Challenge:	Existing research treats memory as a mechanism for factual retention, neglecting its role in shaping users’ emotional experiences.
Approach:	They propose a benchmark for evaluating Emotional Need-aware Proactive Memory Retrieval (ENPMR) it enables agents to infer users’ latent emotional needs and proactively retrieve appropriate memories to support empathetic interaction.
Outcome:	The proposed benchmark includes over 1,800 memory-augmented dialogues and defines structured mappings between emotional needs and supportive memory types.

Copied to clipboard

Aryan Roy, Zekun Wang, Christopher J. MacLellan

PROPER Agents: Proactivity Driven Personalized Agents for Advancing Knowledge Gap Navigation (2026.findings-acl)

Challenge:	a long tradition in cognitive science treats concreteness as a graded dimension of conceptual representation . concrete words benefit from richer sensory codes and exhibit robust behavioral advantages over abstract words .
Approach:	They compare vision-language models with text-only large language models to test their concreteness . they find that VLMs show more human-like sensitivity to concreteness than LLMs .
Outcome:	The proposed model-based training improves on the Llama text backbones and Llma Vision counterparts.

Copied to clipboard

Kirandeep Kaur, Vinayak Gupta, Aditya Gupta, Chirag Shah

Challenge:	Current approaches to proactive assistance are anchored in what users express or can read, leading to unnecessary or mistimed interventions.
Approach:	They propose a framework that explicitly models user-specific knowledge gaps in a controlled manner.
Outcome:	The proposed framework improves on quality scores and win rates across multiple domains, achieving up to 84% gains in single-turn evaluation and consistent dominance in multiturn interactions.

PO-KGQA: Preference Optimization for Low-Resource Complex Knowledge Graph Question Answering (2026.findings-acl)

Copied to clipboard

Prerna Agarwal, Ayushman Kumar Singh, Srikanta Bedathur

Beyond Prompt Engineering: A Systematic Analysis of Prompt Lexical Sensitivity and Its Impacts on Quality (2026.findings-acl)

Challenge:	Existing low-resource in-context learning-based knowledge graph question answering methods rely heavily on large language models to convert natural language questions into logical forms.
Approach:	They propose a low-resource in-context learning-based knowledge graph question answering (KGQA) that uses large language models to convert a natural language question into its corresponding logical form.
Outcome:	The proposed method outperforms other methods on complex benchmarks by approximately 9% (avg).

Copied to clipboard

Qipeng Xie, Zi Liang, Jiafei Wu, Yufei Chen, Weizheng Wang, Wenao Ma, Zhong Ming, Haiqin Yang, Kaishun Wu

DUSK: Do Not Unlearn Shared Knowledge (2026.findings-acl)

Challenge:	Existing studies on prompt engineering have focused on optimizing models for performance under stylistic perturbations.
Approach:	They conduct the first analysis of n-gram token-level mechanisms . they find that higher average performance is inherently associated with lower variance and greater stability.
Outcome:	The proposed model reduces the variance of the generated code by 40% . the proposed model is based on a large-scale dataset of 132,000 prompt variants .

Copied to clipboard

Wonje Jeung, Sangyeon Yoon, Hyesoo Hong, Soeun Kim, Seungju Han, Youngjae Yu, Albert No

Challenge:	Recent work suggests that machine learning models are indistinguishable from models trained on retain sets.
Approach:	They propose a benchmark to evaluate machine unlearning under realistic knowledge overlap . they construct documents containing both shared and unique knowledge .
Outcome:	The proposed model is indistinguishable from a model retrained on the retain set while only forget-specific content is removed.

Towards Mitigating Hallucinations in Large Vision-Language Models by Refining Textual Embeddings (2026.findings-acl)

Copied to clipboard

Aakriti Agrawal, Gouthaman KV, Rohith Aralikatti, Gauri Jagatap, Jiaxin Yuan, Sarvesh Baskar, Vijay Kamarshi, Andrea Fanelli, Furong Huang

Prior Beliefs Prejudice LLM-as-Judge: Evidence from Persuasion Evaluation (2026.findings-acl)

Challenge:	Hallucinations in Large Vision-Language Models (LVLMs) are a persistent challenge, stemming from inadequate integration of visual information during multimodal reasoning.
Approach:	They propose a visual feature incorporation method that encourages the model to learn visually-informed textual embeddings distinct from those of the base LLM and promotes a more balanced attention distribution.
Outcome:	The proposed method significantly reduces hallucinations and fosters more balanced multimodal reasoning.

Copied to clipboard

Pardis Sadat Zahraei, Xiaoning Wang, Nimet Beyza Bozdag, Gokhan Tur, Dilek Hakkani-Tür

Challenge:	Large Language Models are increasingly used as judges to evaluate text quality, content and assess arguments.
Approach:	They propose to exploit belief-conditioned rating inflation by using persuasion-based probing to examine persuasive arguments.
Outcome:	The proposed model fails to evaluate persuasive arguments based on belief alignment . the model fails in three of the three tasks, with belief-conditioned rating inflation accounting for 88% of cases.

Direct Token Optimization: A Self-Contained Approach to Large Language Model Unlearning (2026.findings-acl)

Copied to clipboard

Hong Kyu Lee, Ruixuan Liu, Li Xiong

EntroBench: Evaluating LLM Watermarking Under Multi-Entropy Scenarios and Practical User Operations (2026.findings-acl)

Challenge:	Existing methods for large language models (LLMs) rely on external resources such as auxiliary models, retain datasets, or even commercial AI services.
Approach:	They propose a self-contained unlearning approach that optimizes the token-level objectives to unlearn specific sequences without external resources.
Outcome:	The proposed approach improves the forget quality up to 16.8 over the latest benchmarks while maintaining comparable model utility.

Copied to clipboard

Pengyuan Qin, Linnan Tu, Yuhan Ke, Hefei Ling

Beyond Single-Shot: Multi-step Tool Retrieval via Query Planning (2026.findings-acl)

Challenge:	Existing evaluations of large language models (LLMs) watermarking are limited to fixed entropy settings.
Approach:	They propose a benchmark for LLM watermarking that systematically covers three entropy levels and seven representative tasks.
Outcome:	The proposed framework covers three entropy levels and seven representative tasks.

Copied to clipboard

Wei Fang, James R. Glass

On Safety Risks in Experience-Driven Self-Evolving Agents (2026.findings-acl)

Challenge:	Large language models (LLMs) are evolving from text generation into integration within agentic workflows . tools such as APIs, databases, and software tools are expanding rapidly .
Approach:	They propose a lightweight framework that models retrieval as iterative query planning . instead of single-shot matching, ToolQP decomposes instructions into sub-tasks .
Outcome:	The proposed framework achieves state-of-the-art performance and robustness across retrievers.

Copied to clipboard

Weixiang Zhao, Yichen Zhang, Yingshuo Wang, Yang Deng, Yanyan Zhao, Xuda Zhi, Yongbo Huang, Hao He, Wanxiang Che, Bing Qin, Ting Liu

Challenge:	Experience-driven self-evolution has emerged as a promising paradigm for improving the autonomy of large language model agents, yet its reliance on self-curated experience introduces underexplored safety risks.
Approach:	They investigate how experience accumulation and utilization in self-evolving agents affect safety performance across web-based and embodied environments.
Outcome:	The findings expose inherent limitations of current self-evolving agents and call for more principled strategies to ensure safe and reliable adaptation.

LLM Program Optimization via Retrieval Augmented Search (2026.findings-acl)

Copied to clipboard

Sagnik Anupam, Alexander Shypula, Osbert Bastani

EnsemW2S: Enhancing Weak-to-Strong Generalization with Large Language Model Ensembles (2026.findings-acl)

Challenge:	Recent work shows that large language models have difficulty with program optimization out-of-the-box.
Approach:	They propose a blackbox adaptation method that performs beam search over candidate optimizations by a training dataset.
Outcome:	The proposed method outperforms retrieval based on the source code in a number of ways.

Copied to clipboard

Aakriti Agrawal, Mucong Ding, Chenghao Deng, Zora Che, Arjun Rajaram, Anirudh Satheesh, Bang An, C. Bayan Bruss, John Langford, Furong Huang

Stable On-Policy Distillation through Adaptive Target Reformulation (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are rapidly approaching and potentially exceeding human-level performance . a novel method aims to improve weak experts' generalization abilities by training them on limited human- level data .
Approach:	They propose a method that iteratively combines multiple weak experts to improve their generalization performance by training on limited human-level data.
Outcome:	The proposed method improves weak experts' generalization abilities by iterating on weak models and stronger student models.

Copied to clipboard

Ijun Jang, Jewon Yeom, Juan Yeo, Hyunggyu Lim, Taesup Kim

A Syntactic and Semantic Probe into Language Evolution based on Large Language Models (2026.findings-acl)

Challenge:	Knowledge distillation (KD) is widely used for transferring capabilities from proprietary models to efficient open-source counterparts.
Approach:	They propose a method that constructs a geometric target distribution in logit space to emphasize agreement between the teacher and the student.
Outcome:	Experiments show that the proposed method outperforms supervised fine-tuning and existing on-policy baselines.

Copied to clipboard

Hao Pang, Changcheng Li, Yingxue Liu

Assessing the Effect of Context in Multi-domain Acceptability Judgment (2026.findings-acl)

Challenge:	Existing studies on language evolution have relied on manual annotated resources and rely on dependency parsing.
Approach:	They propose to use attention-based structural distance and semantic space distance to measure language development.
Outcome:	The proposed measures show that human and LLMs share common characteristics in language processing.

Copied to clipboard

Eunike Andriani Kardinata, Yusuke Sakai, Taro Watanabe

Challenge:	Existing studies evaluate sentences in isolation and do not consider how context influences LLM acceptability judgments.
Approach:	They examine how contextual cues affect model-generated acceptability ratings across multiple domains and several LLMs, using different forms of domain-specific contextual cueeds to situate sentences in intended usage settings.
Outcome:	The findings support the development of more context-aware evaluation frameworks.

Lost in Decomposition: Analyzing and Mitigating the Limitations of Long Context Methods via Context Dependency (2026.findings-acl)

Copied to clipboard

Jiayuan Guo, Yueyang Su, Yuyao Ge, Saiping Guan, Lei Yu, Jiafeng Guo, Xueqi Cheng

Challenge:	Existing workflow-based long context methods do not perform well on specific datasets . performance degradation is associated with the indiscriminate application of long context models .
Approach:	They propose a training-free adaptive routing strategy to improve long context large language models' robustness.
Outcome:	The proposed method can be generalized to all types of datasets, but performance degradation is a concern.

Tandem: Riding Together with Large and Small Language Models for Efficient Reasoning (2026.findings-acl)

Copied to clipboard

Zichuan Fu, Xian Wu, Guojing Li, Yejing Wang, Yijun Chen, Zhao Zihao, Luo Yixuan, Hanyu Yan, Yefeng Zheng, Xiangyu Zhao

PROBE: PROcess-Based BEnchmark for Hallucination Detection (2026.findings-acl)

Challenge:	Recent advances in large language models (LLMs) have catalyzed the rise of reasoningintensive inference paradigms, where models perform explicit step-by-step reasoning before generating final answers.
Approach:	They propose a large-small LLM collaboration framework that synergizes large and small language models to achieve high-quality reasoning with significantly reduced computational cost.
Outcome:	The proposed framework outperforms the mentor LLM while preserving the benefits of the thinking paradigm of LLMs.

Copied to clipboard

Yu Zhang, Peter Belcak, Shizhe Diao, Yonggan Fu, Shaona Ghosh, Morteza Mardani, Eileen Margaret Peters Long, Bei Yu, Pavlo Molchanov

Prompt Optimization for Relation Extraction using Reinforcement Learning (2026.findings-acl)

Challenge:	Existing agentic applications rely on LLMs to self-assess the factuality of outputs . but current LLM systems fail to detect hallucinations .
Approach:	They propose a benchmark that breaks down hallucination detection into four critical steps . they show that when halluciation detection is treated as a multi-step process, all models achieve considerably better performance.
Outcome:	The proposed benchmark breaks down hallucination detection into four critical steps . it shows that when halluciation detection is treated as a multi-step process, all models achieve considerably better performance.

Copied to clipboard

Ying Liu, Dong Shuai, Cui Zibo, TengQi Ye, Gang Wu

PolyAudio: Advancing Multi-Audio Reasoning in Large Audio Language Models with Interleaved Multi-Audio Contexts (2026.findings-acl)

Challenge:	Existing prompt-based methods rely heavily on large-scale annotated datasets limiting their applicability in domain-specific and low-resource scenarios.
Approach:	They propose a reinforcement learning-based automated prompt optimization framework for domain relation extraction that optimizes prompt quality through interaction with a black-box LLM.
Outcome:	The proposed framework outperforms existing prompt-based methods and supervised baselines on multiple extraction datasets across medical, financial, legal, and news domains.

Copied to clipboard

Sonal Kumar, Sreyan Ghosh, Yueqian Lin, S Sakshi, Ashish Seth, Yiran Chen, Ramani Duraiswami, Dinesh Manocha

Scaling Evaluation-Time Compute with Reasoning Models as Evaluators (2026.findings-acl)

Challenge:	Large Audio Language Models have shown impressive performance on single-clip tasks . however, their ability to reason over interleaved multi-audio contexts remains limited .
Approach:	They propose a LALM that targets multi-audio understanding via instruction tuning rather than massive-scale pre-training.
Outcome:	The proposed model outperforms baseline models on multi-audio tasks while maintaining robustness.

Copied to clipboard

Seungone Kim, Ian Wu, Jinu Lee, Xiang Yue, Seongyun Lee, Minkyeong Moon, Carolin Lawrence, Kiril Gashteovski, Julia Hockenmaier, Graham Neubig, Sean Welleck

Beyond Memorization: Extending Reasoning Depth with Recurrence, Memory and Test-Time Compute Scaling (2026.findings-acl)

Challenge:	Language model (LM) evaluators that generate chain-of-thought reasoning are widely used for the assessment of LM responses.
Approach:	They investigate whether increasing LMs' "thinking" time through scaling test-time compute can improve an LM's evaluation capability.
Outcome:	The proposed reasoning models improve evaluation performance monotonically with the number of reasoning tokens generated, mirroring trends seen in LM reasoning.

Copied to clipboard

Ivan Rodkin, Daniil Orel, Konstantin Smirnov, Arman Bolatov, Bilal Elbouardi, Besher Hassan, Yuri Kuratov, Aydar Bulatov, Preslav Nakov, Timothy Baldwin, Artem Shelmanov, Mikhail Burtsev

Navigating the Alignment-Calibration Trade-off: A Pareto-Superior Frontier via Model Merging (2026.findings-acl)

Challenge:	Reasoning is a core capability of large language models, yet how multi-step reasoning is learned and executed remains unclear.
Approach:	They evaluate how large language models learn multi-step reasoning without memorization . they find that most neural architectures trained from scratch can learn rule inference .
Outcome:	The proposed framework fails to solve a natural-language proxy task with high accuracy.

Copied to clipboard

Tiancheng Hu, Benjamin Minixhofer, Nigel Collier

Challenge:	We show that the "alignment tax" of post-training is framed as a drop in task accuracy.
Approach:	They propose a more holistic view of the alignment tax by framing it as a drop in accuracy and a degradation of model calibration.
Outcome:	The proposed method improves accuracy beyond both parents while recovering calibration lost during alignment.

From Representation to Choice: Tracing Decision Emergence Across Languages in LLMs (2026.findings-acl)

Copied to clipboard

Divyanshu Bhatt, Abhinav Joshi, Ujjaval Patel, Ashutosh Modi

Unlocking the Edge deployment and ondevice acceleration of multi-LoRA enabled one-for-all foundational LLM (2026.findings-acl)

Challenge:	Recent advances in large language models have made them highly multilingual, but how they internally reason remains unexplored.
Approach:	They propose to model multilingual reasoning through a decision-making perspective using aligned multiple-choice questions from the mMMLU benchmark.
Outcome:	The proposed model shows that languages share similar activation spaces, but subtle divergences emerge as decisions propagate through transformer layers.

Copied to clipboard

Sravanth Kodavanti, Sowmya Vajrala, Srinivas Soumitri Miriyala, Utsav Tiwari, Uttam Kumar, Utkarsh Kumar Mahawar, Achal Pratap Singh, Arya D, Narendra Mutyala, Vikram Nelvoy Rajendiran, Sharan Kumar Allur, Euntaik Lee, Dohyoung Kim, HyeonSu Lee, Gyusung Cho, JungBae Kim

Challenge:	a framework for efficient on-device inference of large language models is needed for smartphones . memory, latency, and runtime flexibility are constraints for large language model deployments.
Approach:	They propose a hardware-aware framework for efficient on-device inference of a LLaMA-based multilingual foundation model for Samsung Galaxy S24 and S25 devices with SM8650 and SM8750 chipsets respectively.
Outcome:	The proposed framework improves memory, latency and performance across 9 languages and 8 tasks.

SynthRL: Scaling Visual Reasoning with Verifiable Data Synthesis (2026.findings-acl)

Copied to clipboard

Zijian Wu, Jinjie Ni, Xiangyan Liu, Zichen Liu, Hang Yan, Michael Qizhe Shieh

From Fragments to Facts: A Curriculum-Driven DPO Approach for Generating Hindi News Veracity Explanations (2026.findings-acl)

Challenge:	SynthRL synthesizes over 3.3K additional verifiable, challenging questions from approximately 8K seed samples.
Approach:	They propose a scalable and guaranteed pipeline for automatic data scaling in reasoning-oriented RL training.
Outcome:	The proposed pipeline synthesizes over 3.3K additional verifiable, challenging questions from approximately 8K seed samples.

Copied to clipboard

Pulkit Bansal, Raghvendra Kumar, Shakti Singh, Adam Jatowt, Sriparna Saha

Joint Optimization of Training Data and Policy in RLHF (2026.findings-acl)

Challenge:	DeFactoX integrates Direct Preference Optimization (DPO) with Curriculum learning to align machine-generated explanations with human reasoning.
Approach:	They propose a framework that integrates Direct Preference Optimization (DPO) with Curriculum learning to align machine-generated explanations with human reasoning.
Outcome:	The proposed framework combines Direct Preference Optimization (DPO) with Curriculum learning to align machine-generated explanations with human reasoning.

Copied to clipboard

Zhuohao Yu, Jiali Zeng, Weizheng Gu, Mengyuan Sun, Yidong Wang, Fandong Meng, Jie Zhou, Shikun Zhang, Wei Ye

Challenge:	JODP optimizes policies on fixed training inputs, limiting the diversity of learning signals.
Approach:	They propose a framework where policy generates improved variants of training problems to enhance its own learning.
Outcome:	The proposed framework improves on safety alignment tasks by allowing 4B models to reach 8B model performance with less than 1% additional computational overhead.

Simul-COMET: A Quality Metric for Simultaneous Interpretation in Distant Language Pair Considering Word Order Difference (2026.findings-acl)

Copied to clipboard

Kosuke Doi, Mana Makinae, Yusuke Sakai, Hidetaka Kamigaito, Taro Watanabe

Risk-Controlled Event-Driven Cascading Updates for Knowledge Graph Consistency Restoration (2026.findings-acl)

Challenge:	Simultaneous interpretation (SI) uses segmenting of source speech into chunks and translating them in order.
Approach:	They propose a variation of COMET that measures monotonicity for simultaneous interpretation . they train Simul-COMET on offline translation data and show stronger alignment with evaluation scores .
Outcome:	The proposed model shows stronger alignment with evaluation scores provided by interpreters than COMET.

Copied to clipboard

Bo Ni, Qinwen Ge, Haowei Fu, Ryan A. Rossi, Xiaorui Liu, Jiejun Xu, Tyler Derr

ContractEval: A Benchmark for Evaluating Contract-Satisfying Assertions in Code Generation (2026.findings-acl)

Challenge:	Knowledge Graphs (KGs) typically treat updates as independent facts . factual, localized updates can contradict and invalidate previously correct knowledge .
Approach:	They propose a model-agnostic framework for cascading KG update identification that leverages conformal prediction to provide reliable uncertainty guarantees over the cascade as a whole.
Outcome:	The proposed framework provides reliable uncertainty guarantees over the cascade as a whole . it integrates large language models to enrich event representations with world knowledge.

Copied to clipboard

Soohan Lim, Joonghyuk Hahn, Hyunwoo Park, Sang-Ki Ko, Yo-Sub Han

ToM-Synth: Scaling Robust Theory of Mind in LLMs via 6,912 Structured Social Units (2026.findings-acl)

Challenge:	Current evaluations measure functional correctness on well-formed inputs, but they filter out inputs that violate them.
Approach:	They propose a benchmark to evaluate whether generated code enforces preconditions . they use a neuro-symbolic pipeline to evaluate code with test cases .
Outcome:	The proposed benchmark aims to evaluate whether generated code enforces preconditions . it aims at achieving pass@k scores while ignoring those that violate them .

Copied to clipboard

Guiyang Hou, Xiang Huang, Shangke Lyu, Yuchuan Wu, Weiyao Luo, Xinyu Mei, Yongliang Shen, Weiming Lu, Yongbin Li

Budget-Aware Routing for Long Clinical Text (2026.findings-acl)

Challenge:	Existing methods endowing LLMs with Theory of Mind fail to internalize the augmented ToM into the LLM.
Approach:	They propose a factorial combinatorial synthesis framework that enables systematic synthesis of ToM data and uses it for RL fine-tuning.
Outcome:	The proposed framework yields a training dataset of 27,648 instances.

Copied to clipboard

Khizar Qureshi, Geoffrey Martin, Yifan Peng

Multi-LLM Collaborative Search for Complex Problem Solving (2026.findings-acl)

Challenge:	Long-context capability is now a headline feature of large language models . clinical inputs are long because they are templated, redundant, and stitched from multiple sources.
Approach:	They propose a token-constrained subset selection problem with two design choices . they propose heuristics that balance relevance, coverage, diversity and a monotone submodular objective .
Outcome:	The proposed model is based on a subset selection problem with two design choices . positional heuristics perform best at low budgets in extractive tasks, while diversity-aware methods improve LLM generation.

Copied to clipboard

Sen Yang, Yafu Li, Wai Lam, Yu Cheng

SceneLM: 3D-Aware Language Models for Editable 3D Scene Synthesis (2026.findings-acl)

Challenge:	Large language models (LLMs) often struggle with complex reasoning tasks due to the vast reasoning space inherent in the complexity and inherent ambiguities of natural languages.
Approach:	They propose a mixture-of-search-agents paradigm that integrates diverse reasoning pathways by combining independent exploration and iterative refinement among multiple LLMs.
Outcome:	The proposed approach improves performance over single-agent and multi-agend baselines in complex mathematical and commonsense reasoning tasks.

Copied to clipboard

Xingbo Yao, Xiaoyu Chen, Doudou Zhang, Mingzhi Sheng, Boyuan Cao, Ying-Cong Chen, Hui Xiong

Leveraging Pretrained Language Models as Energy Functions for Glauber Dynamics Text Diffusion (2026.findings-acl)

Challenge:	Existing methods for synthesising 3D scenes from a single image are text-driven and lack precise metric understanding from images.
Approach:	They propose a language-model-based framework that grounds 3D scene synthesis in visual evidence by recovering an executable metric 3D layout directly from a single image.
Outcome:	The proposed framework recovers an executable metric 3D layout directly from an RGB image and instantiates, places, and edits objects for iterative refinement.

Copied to clipboard

Tarun Kathuria, Sachin Kumar

Uncovering Currency Bias and Syntax Gap in Text Embedding Models (2026.findings-acl)

Challenge:	Existing methods for generating text using Glauber dynamics are autoregressive, but they face a number of limitations.
Approach:	They propose a discrete diffusion-based generative model for text generation using Glauber dynamics from statistical physics and use pretrained causal/masked language models to improve the quality of the generated text.
Outcome:	The proposed model outperforms existing models on some common sense reasoning tasks and planning/search tasks.

Copied to clipboard

Saurav Sudevan, Harsh Pal, Yatin Katyal, Sahil Manchanda

C-ReD: A Comprehensive Chinese Benchmark for AI-Generated Text Detection Derived from Real-World Prompts (2026.findings-acl)

Challenge:	Text-embedding models often inherit societal biases, yet the influence of socio-economic markers remains unexplored.
Approach:	They propose to identify Currency Bias as a systemic representational limitation in financial AI . they analyze currency embeddings to identify currency identifiers and associative sensitivity .
Outcome:	The proposed model lacks associative sensitivity to economic hierarchies, the authors show . they show that current embedding practices pose significant risks for the fairness and reliability of financial NLP applications.

Copied to clipboard

Chenxi Qing, Junxi Wu, Zheng Liu, Yixiang Qiu, Hongyao Yu, Bin Chen, Hao Wu, Shu-Tao Xia

Seer Self-Consistency: Advance Budget Estimation for Adaptive Test-Time Scaling (2026.findings-acl)

Challenge:	Recent efforts to develop algorithms for large language models (LLMs) have limited model diversity and data homogeneity in the Chinese corpora.
Approach:	They propose a Chinese Real-prompt AI-generated text Detection benchmark that can be generalized to unseen LLMs and external Chinese datasets.
Outcome:	The proposed benchmarks address critical gaps in model diversity, domain coverage, and prompt realism that have limited prior Chinese detection benchmarks.

Copied to clipboard

Shiyu Ji, Yixuan Wang, Yijun Liu, Qingfu Zhu, Wanxiang Che

Challenge:	Large Language Models (LLMs) have demonstrated exceptional performance in reasoning tasks, particularly in mathematics.
Approach:	They propose a dynamic self-consistency framework that integrates System 1 and System 2 reasoning to improve token efficiency and latency.
Outcome:	The proposed method outperforms existing methods, achieving up to 47% reduction in token consumption and 43% reduction in inference latency without significant performance loss.

Context-Fidelity Boosting: Enhancing Faithful Generation through Watermark-Inspired Decoding (2026.findings-acl)

Copied to clipboard

Weixu Zhang, Fanghua Ye, Qiang Gao, Jian Li, Haolun Wu, Yuxing Tian, Sijing Duan, Nan Du, Xiaolong Li

CaRVE: Critiquing and Refining Visual Elaborations for Figurative Language Illustrations (2026.findings-acl)

Challenge:	Large language models produce content that contradicts or overlooks information provided in the input context, a phenomenon known as faithfulness hallucination.
Approach:	They propose a lightweight framework that boosts the generation probability of context-relevant tokens by boosting the generation of tokens.
Outcome:	The proposed framework improves faithfulness metrics with minimal generation overhead.

Copied to clipboard

Manishit Kundu, Tejomay Kishor Padole, Sumit Shekhar, Biplab Banerjee, Pushpak Bhattacharyya

Defending LLMs against Jailbreak Attacks via Template-Based ICL with a Defensive Suffix (2026.findings-acl)

Challenge:	Existing text-to-image frameworks for figurative illustration rely on proprietary models or human supervision to achieve adequate alignment.
Approach:	They propose a critique-driven framework that uses VLM feedback to refine visual elaborations for figurative image generation.
Outcome:	The proposed framework outperforms existing figurative image-to-text pipelines on human-supervised visual elaborations.

Copied to clipboard

Ruiyang Ni, Changlong Li, Shuaibiao Han, Zhiyu Yi, Perley Xu, Wenjie Ruan

From Fallback to Frontline: When Can LLMs be Superior Annotators of Human Perspectives? (2026.findings-acl)

Challenge:	State-of-the-art large language models (LLMs) are vulnerable to jailbreak attacks, such as GCG and AutoDAN.
Approach:	They propose to take the advances of online In-Context Learning and an offline defensive suffix and optimize them using an iterative algorithm and an online stochastic random search to identify the most effective ICL demonstrations.
Outcome:	The proposed method reduces attack success rate to nearly 0% while maintaining the model’s utility on benign tasks and incurring only negligible* computational overhead.

Copied to clipboard

Hasan Amin, Harry Yizhou Tian, Xiaoni Duan, Chien-Ju Ho, Rajiv Khanna, Ming Yin

Challenge:	large language models are often used as annotators at scale, but are not faithful estimators of human perspectives.
Approach:	They characterize the conditions under which large language models outperform human annotators . they find they are statistically superior frontline estimators based on low variance .
Outcome:	The proposed model outperforms human annotators when predicting subgroup opinions on subjective tasks.

Does Theory of Mind Improvement Really Benefit Human-AI Interactions? Empirical Findings from Interactive Evaluations (2026.findings-acl)

Copied to clipboard

Nanxu Gong, Zixin Chen, Haotian Li, Zishu Zhao, Jianxun Lian, Huamin Qu, Yanjie Fu, Xing Xie

SERE: Structural Example Retrieval for Enhancing LLMs in Event Causality Identification (2026.findings-acl)

Challenge:	Existing benchmarks measure ToM capability improvement through story-reading, multiple-choice questions from a third-person perspective, while ignoring the first-person, dynamic nature of human-AI interactions.
Approach:	They propose a new paradigm of interactive ToM evaluation with both perspective and metric shifts.
Outcome:	The proposed approach improves the performance of four representative LLM enhancement techniques using real-world datasets and a user study.

Copied to clipboard

Zhifeng Hao, Zhongjie Chen, Junhao Lu, Shengyin Yu, Guimin Hu, Keli Zhang, Ruichu Cai, Boyan Xu

Emotion Trajectory-aware Retrieval for Markov-driven Emotion Anticipation in LLM-based Emotional Support Conversation (2026.findings-acl)

Challenge:	Large Language Models (LLMs) have demonstrated strong performance across various NLP tasks, but their effectiveness in ECI remains limited due to biases in causal reasoning.
Approach:	They propose a structural example retrieval framework that leverages LLMs’ few-shot learning capabilities to help LLM models in ECI.
Outcome:	The proposed framework leverages LLMs’ few-shot learning capabilities to guide LLM models in causal reasoning, mitigating bias and improving accuracy.

Copied to clipboard

Hongyan Wu, Zhiliang Tian, Zhen Huang, Tengyue Hu, Linbo Qiao, Yifu Gao, Feng Liu, Dongsheng Li

High-Throughput and Memory-Efficient Zeroth-Order Fine-tuning LLMs with Distributed Parallel Computing (2026.findings-acl)

Challenge:	Existing strategies focus on planning the next-turn dialogue strategies, while external strategy planners focus on generating empathetic responses.
Approach:	They propose a Markov-driven emotion anticipation framework with emotion trajectory-aware retrieval for LLM-based ESC, which anticipates future emotion states to guide strategy planning and achieve sustained emotional support.
Outcome:	The proposed framework can anticipate future emotions and achieve sustained emotional support on two datasets with two models.

Copied to clipboard

Liangyu Wang, Huanyi Xie, Di Wang

AutoAgent: A Fully-Automated and Zero-Code Framework for LLM Agents (2026.findings-acl)

Challenge:	Zeroth-order (ZO) optimization is a memory-efficient alternative to fine-tuning large language models (LLMs).
Approach:	They propose a zeroth-order (ZO) optimization framework that offloads model parameters to CPU memory and overlapping transformer block transfer with dual forward computation on a single GPU.
Outcome:	The proposed framework achieves a 3x speedup over ZO2 on an OPT-175B model while maintaining memory efficiency and improving training throughput.

Copied to clipboard

Jiabin Tang, Tianyu Fan, Chao Huang

Challenge:	Large Language Model (LLM) agents have demonstrated remarkable capabilities in task automation and intelligent decision-making.
Approach:	They propose a Fully-Automated and highly Self-Developing framework that enables users to create and deploy LLM agents using natural language alone.
Outcome:	AutoAgent is a fully-automated and highly self-developing framework that enables users to create and deploy LLM agents using natural language alone.

DR-HM: Distill-then-Reinforce Training with Cognition-Aware Data Synthesis for Harmful Meme Detection (2026.findings-acl)

Copied to clipboard

Zihan Cheng, Jianxiang Ma, Xiaocui Yang, Peidong Wang, Wen Zhang, Shi Feng, Daling Wang, Yifei Zhang, Mingfu Zhang

Long Story Short: Disentangling Compositionality and Long-Caption Understanding in Contrastive VLMs (2026.findings-acl)

Challenge:	Current methods for harmful meme detection lack the knowledge required to identify such hate . current methods lack the ability to identify cultural stereotypes and visual metaphors .
Approach:	They propose a framework that decomposes meme analysis into a human-inspired reasoning process . they propose DR-HM to transfer knowledge from closed-source models while mitigating biases .
Outcome:	The proposed framework outperforms existing methods on three benchmark datasets.

Copied to clipboard

Israfel Salazar, Desmond Elliott, Yova Kementchedjhieva

Model in Distress: Sentiment Analysis on French Synthetic Social Media (2026.findings-acl)

Challenge:	Existing benchmarks for vision-language models treat compositionality and long-caption understanding in isolation.
Approach:	They analyze when compositional reasoning and long-caption understanding transfer across tasks and when this relationship fails.
Outcome:	The proposed model can generalize on poorly grounded captions and with strong visual grounding, while architectural choices can limit compositional learning.

Copied to clipboard

Pierre-Carl Langlais, Pavel Chizhov, Yannick Detrois, Carlos Rosas Hinostroza, Ivan P. Yamshchikov, Bastien Perroy

Working Memory Constraints Scaffold Learning in Transformers under Data Scarcity (2026.findings-acl)

Challenge:	Large companies and services handle a substantial volume of reviews and social media mentions.
Approach:	They propose a generalizable pipeline for automated customer feedback analysis using backtranslation and fine-tuned models to generate 1.7 million tweets from a small seed corpus, complemented by synthetic reasoning traces.
Outcome:	The proposed pipeline generates 1.7 million synthetic tweets from a small seed corpus, complemented by synthetic reasoning traces.

Copied to clipboard

Pranava Madhyastha, Dagmar Adamcová

Internalizing Multi-Agent Reasoning for Accurate and Efficient LLM-based Recommendation (2026.findings-acl)

Challenge:	a recent study has shown that human-like working memory constraints can be integrated into the Transformer architecture . our model incorporates fixed-width windows and temporal decay based attention mechanisms .
Approach:	They propose to integrate working memory constraints into the Transformer architecture . they use fixed-width windows and temporal decay-based attention mechanisms .
Outcome:	The proposed models show that they can learn better when training data is scarce . the findings suggest that such constraints may serve as a beneficial bias guiding models towards more robust representations .

Copied to clipboard

Yang Wu, Haoze Wang, Qian Li, Jun Zhang, Huan Yu, Jie Jiang

Do Language Models Use Logophoric Cues? Evidence from Mandarin Chinese Long-Distance Reflexive (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are reshaping recommender systems by leveraging extensive world knowledge and semantic reasoning to interpret user intent.
Approach:	They propose a single-agent Trajectory-Aligned Recommender to integrate reasoning capabilities into a model by a multi-agend teacher system.
Outcome:	The proposed model surpasses its teacher by 8.7% to 39.5% while eliminating iterative latency.

Copied to clipboard

Yunfang Dong

FineState-Bench: Benchmarking State-Conditioned Grounding for Fine-grained GUI State Setting (2026.findings-acl)

Challenge:	Using minimal pairs and surprisal-based measures, we assess whether large language models exhibit systematic biases toward non-local antecedents in logophoric contexts.
Approach:	They examine large language models’ sensitivity to four logophoric cues known to license long-distance binding of the reflexive ziji .
Outcome:	The proposed model families show that they exhibit above-chance sensitivity to all four cues, while lexically anchored cue are more robustly captured than discourse-level cue.

Copied to clipboard

Fengxian Ji, Jingpu Yang, Zirui Song, Yuanxi Wang, Zhexuan Cui, Yuke Li, Qian Jiang, Xiuying Chen

Dictionary Guided Sparse Logit Editing for Reliable Jailbreak Attacks (2026.findings-acl)

Challenge:	FineState-Bench evaluates whether an agent can correctly ground an instruction to the intended UI control and reach the exact target state.
Approach:	They propose a benchmark that evaluates whether an agent can correctly ground an instruction to the intended UI control and reach the exact target state.
Outcome:	The proposed benchmark evaluates whether an agent can ground an instruction to the intended UI control and reach the exact target state.

Copied to clipboard

Shuaibiao Han, Ruiyang Ni, Zhiyu Yi, Changlong Li, Perley Xu, Wenjie Ruan

Chain-of-Relations: Faithful and Efficient LLM Reasoning over Knowledge Graphs via Relation-Centric Exploration (2026.findings-acl)

Challenge:	Existing methods to optimize large language models suffer from high computational costs and produce uninterpretable, high-perplexity inputs.
Approach:	They propose a sparse index-based intervention that bypasses guardrails via sparser logit editing.
Outcome:	The proposed method bypasses guardrails by modifying pre-softmax logits without gradients or auxiliary models.

Copied to clipboard

Chenhui Liu, Jianpeng Zhou, Jiahai Wang

MedQPA-Gen: Medical Question Proposing and Answering for Report Generation (2026.findings-acl)

Challenge:	Existing methods adopt entity-centric exploration that incrementally constructs reasoning paths by selecting and connecting intermediate entities.
Approach:	They propose to use relation-centric exploration to construct reasoning paths by selecting and connecting intermediate entities and to reduce the dependence on entity completeness.
Outcome:	The proposed method outperforms baselines on three benchmark datasets in both F1 score and KG-grounded Rate.

Copied to clipboard

Weijie Liang, Xiyue Zhu, Ruike Zhu, Chenhao Li, Cheng Tang, Zhiyu Liu, Zhihua Gong, Shirui Luo, Yudu Li, Volodymyr Kindratenko

Faster MoE LLM Inference for Extremely Large Models (2026.findings-acl)

Challenge:	Existing training paradigms fail to explicitly target factual accuracy, resulting in inaccuracies and serious patient safety risks.
Approach:	They propose an automatic and generalizable report evaluation technique that uses question proposing and answering to enable controllable, structured reasoning grounded in medical domain knowledge and the factual correctness of the report.
Outcome:	The proposed method can improve human preference scores and perform better on downstream tasks.

Copied to clipboard

Haoqi Yang, Luohe Shi, Qiwei Li, Zuchao Li, Ping Wang, Hao Huang, Hai Zhao

Localized Cultural Knowledge is Conserved and Controllable in Large Language Models (2026.findings-acl)

Challenge:	Existing inference optimizations for coarse-grained Mixture-of-Experts models implicitly assume a fixed activation budget, which is poorly understood.
Approach:	They propose a training-free policy that adapts token-level activation using router confidence and entropy while remaining within the model’s original budget.
Outcome:	The proposed skipping policy can provide substantial throughput gains, but optimal static schedules vary significantly across models and routing mechanisms.

Copied to clipboard

Veniamin Veselovsky, Berke Argın, Benedikt Stroebl, Chris Wendler, Robert West, James Evans, Thomas L. Griffiths, Arvind Narayanan

The Confidence Paradox: Unveiling the Latent Discriminative Power of Diffusion Large Language Models in Mathematical Reasoning (2026.findings-acl)

Challenge:	Large language models (LLMs) display language patterns influenced by their native tongue when learning new languages.
Approach:	They propose to quantify the explicit-implicit localization gap in large language models by using a new cultural localization benchmark and find large gaps in the majority of models.
Outcome:	The proposed model can generate culturally localized responses in multiple languages while maintaining language accuracy and task diversity.

Copied to clipboard

Yansi Li, Gongshen Liu, Zhuosheng Zhang

Efficient Hallucination Detection in Automatic Code Generation (2026.findings-acl)

Challenge:	Diffusion large language models (DLLMs) are a promising alternative to autoregressive (AR) generation, offering token-level probabilities under bidirectional context.
Approach:	They propose to use diffusion large language models to generate token-level probabilities under bidirectional context and to examine the calibration paradox inherent to their native uncertainty estimates.
Outcome:	The proposed model outperforms AR baselines on mathematical reasoning benchmarks and is highly miscalibrated on reasoning benchmark.

Copied to clipboard

Georgii Andriushchenko, Roman Garaev, Lyudmila Rvanova, Artem Shelmanov, Vladimir V. Ivanov

DuET: Dual Execution for Test Output Prediction with Generated Code and Pseudocode (2026.findings-acl)

Challenge:	Large language models produce source code that appears correct and well-formed, but includes hallucinated elements that cause downstream test failures.
Approach:	They develop a transformer-based detector that uses LLM internal representations to identify hallucinations.
Outcome:	The proposed detector outperforms existing methods and unsupervised methods in the code generation domain.

Copied to clipboard

Hojae Han, Jaejin Kim, Seung-won Hwang, Yu Jin Kim, Moontae Lee

Attention Sinks in Diffusion Language Models (2026.findings-acl)

Challenge:	Recent studies have shown that test output prediction is difficult to achieve due to code errors.
Approach:	They propose a framework that grounds prediction on error-resilient pseudocode and simulates execution via LLM reasoning to overcome limitations of direct execution suffering from code errors.
Outcome:	The proposed framework improves Pass@1 on LiveCodeBench, BigCodeBech-Hard, DevEval and HumanEval(+) and improves on pass@1 by 13.6 pp.

Copied to clipboard

Maximo Eduardo Rulli, Simone Petruzzi, Edoardo Michielon, Fabrizio Silvestri, Simone Scardapane, Alessio Devoto

R3-SQL: Ranking Reward and Resampling for Text-to-SQL (2026.findings-acl)

Challenge:	Masked Diffusion Language Models (DLMs) employ transformer encoders with bidirectional attention, enabling parallel token generation while maintaining competitive performance.
Approach:	They conduct an empirical analysis of DLM attention patterns focusing on the attention sinking phenomenon . they find that DLMs also exhibit attention sinks, but with distinct characteristics .
Outcome:	The proposed models employ transformer encoders with bidirectional attention, enabling parallel token generation while maintaining competitive performance.

Copied to clipboard

Hojae Han, Yeonseok Jeong, Seung-won Hwang, Zhewei Yao, Yuxiong He

Detecting Hallucinations in SpeechLLMs at Inference Time Using Attention Maps (2026.findings-acl)

Challenge:	Existing rankers assign inconsistent scores to functionally equivalent SQL queries . ranking cannot recover when the correct SQL is absent from the pool.
Approach:	They propose a Text-to-SQL framework that rewards ranking and resampling . it first groups candidates by execution result and ranks groups for consistency .
Outcome:	The proposed framework achieves 75.03 execution accuracy on BIRD-dev, a new state of the art among methods using models with disclosed sizes.

Copied to clipboard

Jonas Waldendorf, Bashar Awwad Shiekh Hasan, Evgenii Tsymbalov

Leveraging External Knowledge for Historical Document Restoration via Retrieval-Augmented Large Language Models (2026.findings-acl)

Challenge:	Existing methods for hallucination detection for text-based LLMs do not capture audio-specific signals.
Approach:	They propose to capture pathological attention patterns associated with hallucination using four attention-derived metrics to train lightweight logistic regression classifiers.
Outcome:	The proposed approach outperforms baselines on in-domain data and generalises to out-of-domain ASR settings.

Copied to clipboard

Gabeen Kim, Kyeongpil Kang

From Experience to Skill: Multi-Agent Generative Engine Optimization via Reusable Strategy Learning (2026.findings-acl)

Challenge:	Historical documents suffer from illegibility due to physical deterioration and damage due to deteriorating materials.
Approach:	a new framework leverages large language models with retrieval-augmented generation to restore historical documents. authors propose a framework that leverages implicit knowledge of pre-trained LLMs with explicitly retrieved external context.
Outcome:	a new framework outperforms existing methods for restoration of historical documents in Korean . the proposed model can restore both general characters and named entities, the authors say .

Copied to clipboard

Beining Wu, Fuyou Mao, Jiong Lin, Cheng Yang, Jiaxuan Lu, Yifu Guo, Siyu Zhang, Yifan Wu, Ying Huang, Fu Li

Challenge:	Generative engines (GEs) are replacing ranked links with citation-grounded answers . current methods are unable to accumulate or transfer effective strategies across tasks and engines .
Approach:	They propose a multi-agent framework where planning, editing, and fidelity-aware evaluation serve as the execution layer.
Outcome:	The proposed framework outperforms heuristic baselines in visibility and citation fidelity on three mainstream engines.

CodeRM-NT: Reward Model for Code RL without Unit Tests (2026.findings-acl)

Copied to clipboard

Xiao Xia, Dan Zhang, Tianrui Sun

Fin-STAR: Structure-as-Semantics to Resolve Implicitness in Financial Retrieval (2026.findings-acl)

Challenge:	Existing methods rely on unit tests to evaluate code correctness and provide rewards, but these methods are difficult to verify at scale.
Approach:	They propose a code reward model that leverages Monte Carlo Tree Search guided by LLMs to generate code snippets and judges execution traces to annotate code with reward signals.
Outcome:	The proposed model outperforms synthetic unit tests on multiple code generation benchmarks and improves curriculum learning.

Copied to clipboard

Yu Zou, Yan Chen, Lida He, Qi Zhou, Xiaorui Zhou, Aixi Zhong, Yi Wang, Wei Li, Qingyu Wang, Jiatao Li, Wei Gong, Jialei Zeng, Jingmei Zhao, Ke Jiang, Qing Li

When Rules Learn: A Self-Evolving Agent for Legal Case Retrieval (2026.findings-acl)

Challenge:	Existing Retrieval-Augmented Generation systems treat structure as a physical navigational skeleton rather than intrinsic semantic knowledge.
Approach:	They propose a framework that redefining hierarchy as intrinsic semantics and uses snippets to enrich hierarchical lineage.
Outcome:	The proposed framework outperforms state-of-the-art hierarchical and graph-based benchmarks on FinTierQA Gold.

Copied to clipboard

Mingxu Tao, Jiawei Hu, Xian Zhou, Wenpeng Hu, Jiajun Cheng, Yunbo Cao, Zhunchen Luo, Guotong Geng

SpecExtend: A Drop-in Enhancement for Speculative Decoding of Long Sequences (2026.findings-acl)

Challenge:	Existing dense retrieval methods have achieved notable progress, but their effectiveness in legal case retrieval remains limited.
Approach:	They propose a self-evolving framework for rule-driven query rewriting that enhances BM25 without any parameter training.
Outcome:	The proposed framework outperforms non-evolutionary baselines, including human-designed rules and greedy rule selection, especially when powered by a high-capacity core LLM.

Copied to clipboard

Jungyoub Cha, Hyunjong Kim, Sungzoon Cho

Challenge:	Speculative decoding performance degrades as input length increases, with significant drops even at moderate lengths.
Approach:	They propose a drop-in enhancement that improves speculative decoding on long sequences without additional training.
Outcome:	The proposed enhancement accelerates speculative decoding by up to 2.84 on 16K-token long document summarization and up to 3.86 on long-form reasoning while preserving the short-input performance of state-of-the-art frameworks.

Copied to clipboard

Wenqiang Wang, Peng Chen, Yan Xiao, Yangshijie Zhang, Xiaoyue Lu, Jianjie Huang, Xiaochun Cao

Concise Math Reasoning via Difficulty-Aware Distillation (2026.findings-acl)

Challenge:	Standard in-context learning assumes identical output spaces between test and retrieval datasets . however, in practice, these datasets can be fully aligned, partially alignes, or fully disjoint in label space .
Approach:	They propose a framework for in-context learning under output-space mismatch . they identify demonstrations relevant to the test label space via a Bayesian probabilistic criterion .
Outcome:	The proposed framework achieves state-of-the-art results across three LLMs, three task types, and four datasets.

Copied to clipboard

Yifan Wu, Jingze Shi, Bingheng Wu, Jiayi Zhang, Xiaotian Lin, Yizhang Zhu, Zhaoyang Yu, Bang Liu, Chenglin Wu, Nan Tang, Yuyu Luo

Numerical Reasoning

Challenge:	Human experts tackle difficult math problems by identifying and executing a few pivotal steps rather than listing every intermediate thought.
Approach:	They propose a method for producing training data that mirrors concise human reasoning by rewriting a problem's solution to retain only the essential steps.
Outcome:	The proposed method outperforms models trained on 800k long CoT and cuts training and inference costs.

Corpora Generation for Urdu Grammatical Error Correction (2026.findings-acl)

Copied to clipboard

Syed Ahad, Burhanuddin Aliasghar Ezzi, Muhammad Arsalan Hussain, Sandesh Kumar, Abdul Samad

Rethinking Retrieval-Augmented Generation as a Cooperative Decision-Making Problem (2026.findings-acl)

Challenge:	grammatical error correction (GEC) for Urdu remains under-researched due to lack of annotated datasets.
Approach:	They propose a method for synthesizing a large dataset by collecting errors from the Urdu WikiEdits history and learning from them.
Outcome:	The proposed method synthesizes a large dataset and fine-tunes models against it.

Copied to clipboard

Lichang Song, Ting Long, Yi Chang

More Agents Improve Math Problem Solving but Adversarial Robustness Gap Persists (2026.findings-acl)

Challenge:	Existing RAG systems rely on ranking-centric, asymmetric dependency paradigms to generate results.
Approach:	They propose a framework that treats the reranker and the generator as peer decision-makers rather than being connected through an asymmetric dependency pipeline.
Outcome:	The proposed framework treats the reranker and the generator as peer decision-makers rather than being connected through an asymmetric dependency pipeline.

Copied to clipboard

Khashayar Alavi, Zhastay Yeltay, Lucie Flek, Akbar Karimi

Indic-CodecFake meets SATYAM: Towards Detecting Neural Audio Codec Synthesized Speech Deepfakes in Indic Languages (2026.findings-acl)

Challenge:	Large Language Models (LLMs) are powerful but weak when inputs are perturbed.
Approach:	They evaluate LLMs that are more powerful than single LLM in math question answering . they use a unified sampling-and-voting framework to evaluate their models .
Outcome:	The proposed models show that collaboration between agents improves accuracy and clean accuracy even with a large number of agents.

Copied to clipboard

null Girish, Mohd Mujtaba Akhtar, Orchid Chetia Phukan, Arun Balaji Buduru

Tracing Relational Knowledge Recall in Large Language Models (2026.findings-acl)

Challenge:	Speech deepfakes are highly realistic and can generate a few seconds of recorded speech.
Approach:	They propose an ALM that integrates semantic and prosodic representations from Whisper and TRILLsson to generate a speech deepfake dataset.
Outcome:	The proposed framework outperforms existing ALMs on the ICF benchmark in Indic languages.

Copied to clipboard

Nicholas Popovič, Michael Färber

LR-DWM: Efficient Watermarking for Diffusion Language Models (2026.findings-acl)

Challenge:	Feature attribution analyses of the trained probes reveal correlations between probe accuracy and relation specificity, entity connectedness, and how distributed the signal on which the probe relies is across attention heads.
Approach:	They evaluate latent representations derived from attention heads and MLP contributions . they show correlations between probe accuracy and relation specificity .
Outcome:	The proposed representations are compared with the representations obtained from attention heads and MLPs.

Copied to clipboard

Ofek Raban, Gal Chechik, Ethan Fetaya

Beyond Unimodal Shortcuts: MLLMs as Cross-Modal Reasoners for Grounded Named Entity Recognition (2026.findings-acl)

Challenge:	Current methods for large language models rely on tokens being generated sequentially . left-right Diffusion watermarking uses a fixed, deterministic left-to-right order .
Approach:	They propose a scheme that biases tokens based on both left and right neighbors . left-Right Diffusion Watermarking is a low-latency alternative to autoregressive models .
Outcome:	The proposed method can be watermarked efficiently with minimal runtime and memory overhead.

Copied to clipboard

Jinlong Ma, Yu Zhang, Xuefeng Bai, Kehai Chen, Yuwei Wang, Zeming Liu, Jun Yu, Min Zhang

BubbleRAG: Interactive Cognitive Offloading with Thought Bubble in Retrieval-Augmented Generation (2026.findings-acl)

Challenge:	Existing approaches to GMNER use MLLMs as auxiliary tools, causing cumulative error propagation and a lack of rigorous cross-modal verification.
Approach:	They propose a model that enforces structured cross-modal reasoning through Multi-style Reasoning Schema Injection and Constraint-guided Verifiable Optimization.
Outcome:	The proposed model enforces structured cross-modal reasoning through multi-style Reasoning Schema Injection and Constraint-guided Verifiable Optimization.

Copied to clipboard

Fuda Ye, Jiachuan Wang, Yongqi Zhang, Lei Chen, Shuangyin Li