Papers with stability
Semantic shift in social networks (2021.starsem-1)
Copied to clipboard
| Challenge: | lexical semantic change manifests differently across different communities, according to a new study . social network analysis is a tool of sociolinguists studying variation and change . |
| Approach: | They use distributional methods to quantify lexical semantic change and induce a social network on communities based on interactions between members. |
| Outcome: | The proposed method is based on interactions between members and the community. |
How LLMs React to Industrial Spatio-Temporal Data? Assessing Hallucination with a Novel Traffic Incident Benchmark Dataset (2025.naacl-industry)
Copied to clipboard
Qiang Li, Mingkun Tan, Xun Zhao, Dan Zhang, Daoan Zhang, Shengzhao Lei, Anderson S. Chu, Lujun Li, Porawit Kamnoedboon
| Challenge: | Large language models (LLMs) have a high potential to digitize and enhance the health & public services industry. |
| Approach: | They propose to use a cross-lingual benchmark dataset to assess the robustness of state-of-the-art LLMs in the spatio vs temporal domain for traffic incident classification. |
| Outcome: | The proposed model performs well in the spatio-temporal domain and in the non-English context. |
Multi-Programming Language Sandbox for LLMs (2025.acl-demo)
Copied to clipboard
Shihan Dou, Jiazheng Zhang, Jianxiang Zang, Yunbo Tao, Weikang Zhou, Haoxiang Jia, Shichun Liu, Yuming Yang, Shenxi Wu, Zhiheng Xi, Muling Wu, Rui Zheng, Changze Lv, Limao Xiong, Shaoqing Zhang, Lin Zhang, Wenyu Zhan, Rongxiang Weng, Jingang Wang, Xunliang Cai, Yueming Wu, Ming Wen, Yixin Cao, Tao Gui, Xipeng Qiu, Qi Zhang, Xuanjing Huang
| Challenge: | MPLSandbox is an out-of-the-box multi-programming language sandbox designed to provide unified and comprehensive feedback from compiler and analysis tools for Large Language Models (LLMs). |
| Approach: | They propose a multi-programming language sandbox that provides unified feedback from compilers and analysis tools for Large Language Models. |
| Outcome: | The proposed multi-language sandbox can provide comprehensive feedback from compilers and analysis tools for large language models (LLMs). |
SLTEV: Comprehensive Evaluation of Spoken Language Translation (2021.eacl-demos)
Copied to clipboard
| Challenge: | Spoken Language Translation (SLT) evaluation of machine translation (MT) quality has been investigated for decades. |
| Approach: | They propose an open-source tool for assessing machine translation (MT) quality based on time-stamped transcripts and reference translations. |
| Outcome: | The proposed evaluation tool is based on time-stamped transcripts and reference translations into a target language. |
ReAL: How Can LLMs Simulate the Real Teacher? Retrieval-enhanced Agent for Adaptive Learning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Prior methods model learner-item interactions based only on ID sequences, leading to insufficient use of both learner and item information. |
| Approach: | They propose a Retrieval-enhanced Agent for Adaptive Learning powered by large language models to simulate teacher decision-making with extensive prior knowledge and teaching experience. |
| Outcome: | The proposed model outperforms existing models on three real-world datasets in both internal and external perspectives. |
Retrieve, Rerank and Rewrite: Soft Template Based Neural Summarization (P18-1)
Copied to clipboard
| Challenge: | Existing summarization systems rely on the source text to generate summaries, which tends to work unstably. |
| Approach: | They propose to use existing summaries as soft templates to guide the seq2seq model . they use a popular IR platform to Retrieve proper summary as candidate templates . |
| Outcome: | The proposed model outperforms state-of-the-art models in terms of informativeness and readability. |
Investigating the Personality Consistency in Quantized Role-Playing Dialogue Agents (2024.emnlp-industry)
Copied to clipboard
| Challenge: | Using the Big Five personality traits model, we evaluate how stable assigned personalities are for Quantized Role-Playing Dialog Agents (QRPDA) during multi-turn interactions. |
| Approach: | They propose a non-parametric method to evaluate the stability of assigned personalities in quantized large language models (LLMs) for role-playing scenarios. |
| Outcome: | The proposed method shows that it maintains consistent personality traits in QRPDA, and it is more reliable in real-world applications. |
Towards Hierarchical Multi-Step Reward Models for Enhanced Reasoning in Large Language Models (2026.findings-acl)
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
| 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. |
LoRA-MGPO: Mitigating Double Descent in Low-Rank Adaptation via Momentum-Guided Perturbation Optimization (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) adapts large language models by training only a small fraction of parameters, but as the rank of the low-rank matrices increases, LoRA exhibits an unstable “double descent” phenomenon, which delays convergence and impairs generalization by causing instability due to the attraction to sharp local minima. |
| Approach: | They propose a framework that incorporates Momentum-Guided Perturbation Optimization (MGPO) MGPO stabilizes training dynamics by mitigating double descent phenomenon and guiding weight perturbations using momentum vectors from the optimizer’s state. |
| Outcome: | The proposed framework improves performance on natural language understanding benchmarks and shows that it improves convergence and generalization. |
FewNLU: Benchmarking State-of-the-Art Methods for Few-Shot Natural Language Understanding (2022.acl-long)
Copied to clipboard
Yanan Zheng, Jing Zhou, Yujie Qian, Ming Ding, Chonghua Liao, Li Jian, Ruslan Salakhutdinov, Jie Tang, Sebastian Ruder, Zhilin Yang
| Challenge: | Existing evaluation protocols for few-shot natural language understanding (NLU) tasks are inconsistent and hinder fair comparison and measuring progress. |
| Approach: | They propose an evaluation framework that improves previous evaluation procedures in three key aspects, i.e., test performance, dev-test correlation, and stability. |
| Outcome: | The proposed framework improves evaluation procedures in three key aspects, i.e., performance, dev-test correlation, and stability. |
What Works and Doesn’t Work, A Deep Decoder for Neural Machine Translation (2022.findings-acl)
Copied to clipboard
| Challenge: | Deep learning has demonstrated performance advantages in a wide range of natural language processing tasks. |
| Approach: | They propose to deepen the decoder layer in a Transformer model to reduce the difficulty of deep learning. |
| Outcome: | The proposed method can deepen the model on both the encoder and decoder at the same time, resulting in a deeper model and improved performance. |
Improving LLM Reasoning through Interpretable Role-Playing Steering (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for role-playing rely on prompt engineering, which lacks stability and interpretability. |
| Approach: | They propose a framework that extracts latent representations from role-play prompts and constructs a steering vector that can be injected into the model's residual stream with controllable intensity. |
| Outcome: | The proposed framework extracts latent representations from role-play prompts, selects the most relevant features based on activation patterns, and constructs a steering vector that can be injected into the model’s residual stream with controllable intensity. |
Towards Singable Lyrics Translation Using Large Language Models (2026.eacl-srw)
Copied to clipboard
| Challenge: | Existing studies on lyrics translation have relied on fine-tuning open-source language models. |
| Approach: | They examine a multilingual lyrics translation dataset and apply prompting methods to large language models to evaluate singability. |
| Outcome: | The proposed methods improve singability and naturalness, compared to naive translation, the authors show . human evaluations using songs created from translated lyrics show that complex prompting strategies improve singable naturalness . |
A Simple yet Effective Learnable Positional Encoding Method for Improving Document Transformer Model (2022.findings-aacl)
Copied to clipboard
| Challenge: | Existing document Transformers lack a robust positional encoding mechanism to indicate and embed sequential order information in documents. |
| Approach: | They propose a positional encoding method that can be pre-trained on document datasets to improve document understanding. |
| Outcome: | The proposed method outperforms baselines on document understanding tasks in form, receipt, and invoice domains and is robust and stable on noisy data with incorrect order information. |
Source-Free Unsupervised Domain Adaptation for Question Answering via Prompt-Assisted Self-learning (2024.findings-naacl)
Copied to clipboard
| Challenge: | Existing SFDA methods focus on the adaptation phase, overlooking the impact of source domain training on model generalizability. |
| Approach: | They propose a source-free domain adaptation approach for Question Answering where a model trained on a domain is adapted to unlabeled target domains without additional source data. |
| Outcome: | The proposed model outperforms existing methods in managing domain gaps and demonstrating greater stability across target domains. |
SHARP: Unlocking Interactive Hallucination via Stance Transfer in Role-Playing LLMs (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing studies on social interactions neglect hallucination while struggling with poor generalizability and implicit character fidelity judgments. |
| Approach: | They propose a generalizable and explicit paradigm for uncovering interactive patterns of Large Language Models across diverse worldviews by defining interactive hallucination through stance transfer and SHARP, a benchmark built by extracting relations from commonsense knowledge graphs. |
| Outcome: | The proposed paradigm is generalizable and explicit and demonstrates its effectiveness and stability. |
Parallel Continuous Chain-of-Thought with Jacobi Iteration (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to continuous CoT rely on sequential decoding of latent thought tokens, which leads to long training time and low inference speed. |
| Approach: | They propose a parallel continuous chain-of-thought which updates latent thought tokens iteratively in parallel instead of sequentially and improves both training and inference efficiency. |
| Outcome: | The proposed method saves 50% of training and inference time while maintaining stability and robustness in training. |
ATOM: AdapTive and OptiMized dynamic temporal knowledge graph construction using LLMs (2026.findings-eacl)
Copied to clipboard
| Challenge: | Unstructured data is expanding at an unprecedented rate, and static knowledge graphs are often overlooked due to their dynamic nature and lack of time-sensitive features. |
| Approach: | They propose a few-shot approach that builds and continuously updates Temporal Knowledge Graphs (TKGs) from unstructured texts. |
| Outcome: | Empirical results show that ATOM achieves 18% higher exhaustivity, 33% better stability, and over 90% latency reduction compared to baseline methods. |
Instruction-Driven Game Engine: A Poker Case Study (2024.emnlp-demo)
Copied to clipboard
| Challenge: | a new concept allows users to create games by natural language instructions . the concept is based on the text-based game states, which are rendered to visual display . |
| Approach: | They propose an instruction-driven game engine that allows users to create games by natural language instructions. |
| Outcome: | The proposed concept allows users to create games simply by natural language instructions . initial progress lies in developing an IDGE for poker, which supports a wide range of poker variants . |
PhyVer: Physics-Grounded Material Claim Verification with Multi-Fidelity Physical Evidence (2026.acl-demo)
Copied to clipboard
| Challenge: | Existing claim verification pipelines operate over text, producing ungrounded judgments. |
| Approach: | They propose a physics-grounded material claim verification system that can be used to verify claims with physical evidence. |
| Outcome: | The proposed system reduces MAE and sign-offs with experts over text-only GPT-5.1. |
lrnnx: A library for Linear RNNs (2026.eacl-srw)
Copied to clipboard
Karan Bania, Soham Kalburgi, Manit Tanwar, null Dhruthi, Aditya Nagarsekar, Harshvardhan Mestha, Naman Chibber, Raj Deshmukh, Anish Sathyanarayanan, Aarush Rathore, Pratham Chheda
| Challenge: | Existing implementations of linear recurrent neural networks are fragmented across different software frameworks . existing implementations often require custom CUDA kernels or lack publicly available code altogether . |
| Approach: | a unified software library implements several modern LRNN architectures under a common interface. lrnnx aims to improve accessibility, reproducibility, and extensibility of LRnn research and applications. |
| Outcome: | lrnnx aims to improve accessibility, reproducibility, and extensibility of LRNN research and applications. |
Actor-Double-Critic: Incorporating Model-Based Critic for Task-Oriented Dialogue Systems (2020.findings-emnlp)
Copied to clipboard
| Challenge: | In order to improve the sample-efficiency of deep reinforcement learning, we implemented imagination augmented agent (I2A) in spoken dialogue systems (SDS). |
| Approach: | They propose to use an actor-double-critic to improve the stability and overall performance of imagination augmented agent (I2A) in spoken dialogue systems. |
| Outcome: | The proposed model-based agent (ADC) improves the stability and sample-efficiency of deep reinforcement learning (DRL) on a restaurant booking task. |
RedOne 2.0: Rethinking Domain-specific LLM Post-Training in Social Networking Services (2026.acl-industry)
Copied to clipboard
Fei zhao, Chonggang Lu, Haofu Qian, Fangcheng Shi, Zijie Meng, Jianzhao Huang, Zheyong Xie, Shaosheng Cao
| Challenge: | Social networking services (SNS) are critical infrastructure for global interaction . supervised fine-tuning (SFT) can improve in-domain performance, but it often induces a ”seesaw” trade-off with out-of-domain robustness . |
| Approach: | They propose an SNS-oriented LLM with a progressive, RL-prioritized post-training paradigm for fast and stable adaptation. |
| Outcome: | The proposed model improves over the previous 7B model by 2.41 on average . it also yields an 8.74 average gain over its Qwen3-4B base . |
Beyond Model Interpretability: On the Faithfulness and Adversarial Robustness of Contrastive Textual Explanations (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing explanations address the contrastive aspect of explanations but their extension to textual data is under-explored and there is little investigation on their vulnerabilities and limitations. |
| Approach: | They propose a novel evaluation scheme inspired by the faithfulness of explanations by extending the computation of three metrics to textual data and benchmarking POLYJUICE and MiCE on suggested metrics. |
| Outcome: | The proposed methods demonstrate that the connectedness of counterfactuals to their original counterparts is not obvious in both models. |
Class based Influence Functions for Error Detection (2023.acl-short)
Copied to clipboard
Thang Nguyen-Duc, Hoang Thanh-Tung, Quan Hung Tran, Dang Huu-Tien, Hieu Nguyen, Anh T. V. Dau, Nghi Bui
| Challenge: | Influence functions (IFs) are powerful tools for detecting anomalous examples in large scale datasets. |
| Approach: | They propose a method to explain the instability of IFs by leveraging class information to improve the stability of ifs. |
| Outcome: | The proposed method improves performance and stability while incurring no additional computational cost. |
PibE-MPP: A Play-it-by-Ear Masking Performance Plug-in for LLMs (2026.findings-acl)
Copied to clipboard
Mengwei Wang, Simin Niu, Xun Liang, Yuefeng Ma, Sensen Zhang, Jiawei Yang, Shichao Song, Hanyu Wang, Huayi Lai
| 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. |
Robust Transfer Learning with Pretrained Language Models through Adapters (2021.acl-short)
Copied to clipboard
| Challenge: | Existing approaches to transfer learning with pretrained transformer-based language models are not robust and can be adversarial. |
| Approach: | They propose a simple yet effective adapter-based approach to fine-tune language models on downstream tasks. |
| Outcome: | The proposed approach improves stability and adversarial robustness in transfer learning to various downstream tasks. |
Self-Influence Guided Data Reweighting for Language Model Pre-training (2023.emnlp-main)
Copied to clipboard
| Challenge: | Language Models (LMs) pre-trained with selfsupervision on large text data are the default starting point for developing models for various downstream tasks. |
| Approach: | They propose a method for jointly reweighting samples by leveraging self-influence scores as an indicator of sample importance and pre-training. |
| Outcome: | The proposed method promotes novelty and stability for model pre-training. |
How Do Large Language Models Perform on PDE Discovery: A Coarse-to-fine Perspective (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods to identify partial differential equations using large language models suffer from performance degradation under extreme data scarcity. |
| Approach: | They propose a framework to use large language models to identify underlying partial differential equations out of very limited observations of a physical system. |
| Outcome: | The proposed framework is based on a coarse-to-fine paradigm to discover PDEs out of very limited observations of a physical system. |
Model Performance-Guided Evaluation Data Selection for Effective Prompt Optimization (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing prompt engineering methods rely on randomly selected evaluation subsets, leading to suboptimal prompts. |
| Approach: | They propose an iterative evaluation data selection approach for effective prompt optimization using real time model performance. |
| Outcome: | The proposed approach improves effectiveness by 1.6% to 3.1% and stability by 50% to 55.5% on two datasets BIG-bench and LIAR and two models GPT-3.5 and GPT-4o-mini. |
AudioSAE: Towards Understanding of Audio-Processing Models with Sparse AutoEncoders (2026.eacl-long)
Copied to clipboard
Georgii Aparin, Tasnima Sadekova, Alexey Rukhovich, Assel Yermekova, Laida Kushnareva, Vadim Popov, Kristian Kuznetsov, Irina Piontkovskaya
| Challenge: | Feature steering reduces Whisper’s false speech detections by 70% with negligible WER increase, demonstrating real-world applicability. |
| Approach: | They train Sparse Autoencoders across all encoder layers of Whisper and HuBERT and evaluate their stability, interpretability, and practical utility. |
| Outcome: | The proposed models capture general acoustic and semantic information as well as specific events, including environmental noises and paralinguistic sounds, and disentangle them effectively. |
Towards Stable Natural Language Understanding via Information Entropy Guided Debiasing (2023.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to debiase Natural Language Understanding models use dataset biases instead of learning the intended task. |
| Approach: | They propose a debiasing framework that detects and purifies dataset biases using information entropy. |
| Outcome: | The proposed framework improves the stability of performance on out-of-distribution datasets for a set of widely adopted NLU models. |
Revisiting Parameter-Efficient Tuning: Are We Really There Yet? (2022.emnlp-main)
Copied to clipboard
| Challenge: | Pretrained language models (PLMs) are used as backbones to be combined with additional parameters and finetuned on downstream tasks in an end-to-end manner. |
| Approach: | They propose to use a fraction of parameters to tune pretrained language models (PLMs) this is the first comprehensive investigation into the training and evaluation of PETuning methods. |
| Outcome: | The proposed methods have been validated and tested with a rigorous evaluation protocol and have shown that they are unstable and inconsistent. |
Dynamic Prefix as Instructor for Incremental Named Entity Recognition: A Unified Seq2Seq Generation Framework (2025.findings-acl)
Copied to clipboard
| Challenge: | Named Entity Recognition (NER) is a fundamental problem in information extraction. |
| Approach: | They propose a parameter-efficient method for Incremental Named Entity Recognition (INER) task aimed at updating a model to extract entities from an expanding set of entity type candidates by employing a dynamic prefix as a task instructor to guide the generative model. |
| Outcome: | Empirical results show that the proposed method preserves task-invariant knowledge while adapting to new entities with minimal parameter updates. |
Measuring and Mitigating Local Instability in Deep Neural Networks (2023.findings-acl)
Copied to clipboard
| Challenge: | Uncertain details like random initialization can change the outputs of a trained system with potentially disastrous consequences. |
| Approach: | They propose a model stability problem by studying how the predictions of a deep neural network change as a consequence of stochasticity in the training process. |
| Outcome: | The proposed method outperforms data-agnostic methods and is 90% cheaper than the gold standard. |
ExCAR: Event Graph Knowledge Enhanced Explainable Causal Reasoning (2021.acl-long)
Copied to clipboard
| Challenge: | Existing work infers the causation between events based on knowledge from annotated causal event pairs, but additional evidence information is unexploited. |
| Approach: | They propose an Event graph knowledge enhanced explainable CAusal Reasoning framework that acquires additional evidence information from a large-scale causal event graph as logical rules for causal reasoning. |
| Outcome: | The proposed framework outperforms state-of-the-art methods in human evaluation and in animal models. |
On the Sensitivity and Stability of Model Interpretations in NLP (2022.acl-long)
Copied to clipboard
| Challenge: | Recent years have witnessed the emergence of post-hoc interpretations that aim to uncover how NLP models make predictions. |
| Approach: | They propose two new criteria that provide complementary notions of faithfulness to removal-based criteria. |
| Outcome: | The proposed methods overcome limitations of gradient-based methods on removal-based criteria and overcome limitations in the proposed methods. |
Factors Influencing the Surprising Instability of Word Embeddings (N18-1)
Copied to clipboard
| Challenge: | Word embeddings are low-dimensional, dense vector representations that capture semantic properties of words. |
| Approach: | They examine the stability of word embeddings by examining their properties and analyzing their effects on downstream tasks. |
| Outcome: | The results show that even high frequency words exhibit substantial instability, which can have implications for downstream tasks. |
B2T Connection: Serving Stability and Performance in Deep Transformers (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to prevent the vanishing gradient problem in deep neural networks are not effective. |
| Approach: | They propose a method that can equip both higher stability and effective training by a simple modification from Post-LN. |
| Outcome: | The proposed method outperforms Pre-LN and Post-Ln on a wide range of tasks. |
From Implicit Exploration to Structured Reasoning: Guideline and Refinement for LLMs (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing models rely on implicit exploration, which leads to unstable reasoning paths and lack of error correction. |
| Approach: | They propose a framework that shifts from implicit exploration to structured reasoning through guideline and refinement. |
| Outcome: | The proposed model outperforms strong baselines on the Big-Bench Hard benchmark. |
TemplateRL: Structured Template-Guided Reinforcement Learning for LLM Reasoning (2026.findings-acl)
Copied to clipboard
Jinyang Wu, Chonghua Liao, Mingkuan Feng, Shuai Zhang, Zhengqi Wen, Haoran Luo, Ling Yang, Huazhe Xu, Jianhua Tao
| 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. |
Bridging SFT and RL: Dynamic Policy Optimization for Robust Reasoning (2026.findings-acl)
Copied to clipboard
| 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. |
From Alignment to Assignment: Frustratingly Simple Unsupervised Entity Alignment (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for cross-lingual entity alignment rely on lexical matching and probability reasoning, but they inherit poor interpretability and low efficiency from neural networks. |
| Approach: | They propose a simple but effective unsupervised entity alignment method without neural networks that can be used to find the equivalent entities between crosslingual KGs. |
| Outcome: | Extensive experiments show that the proposed method beats advanced supervised methods across all datasets while having high efficiency, interpretability, and stability. |
mABC: Multi-Agent Blockchain-inspired Collaboration for Root Cause Analysis in Micro-Services Architecture (2024.findings-emnlp)
Copied to clipboard
Wei Zhang, Hongcheng Guo, Jian Yang, Zhoujin Tian, Yi Zhang, Yan Chaoran, Zhoujun Li, Tongliang Li, Xu Shi, Liangfan Zheng, Bo Zhang
| Challenge: | Root cause analysis (RCA) in Micro-services architectures with escalating complexity is challenging due to fault propagation and circular dependencies among nodes. |
| Approach: | They propose a framework where multiple agents follow Agent Workflow and collaborate in blockchain-inspired voting to ensure the reliability of root cause analysis. |
| Outcome: | The proposed framework reduces the number of steps and standardizes task processing through Agent Workflow. |
Text is All You Need: LLM-enhanced Incremental Social Event Detection (2025.acl-long)
Copied to clipboard
| Challenge: | Existing state-of-the-art (SOTA) SED models rely on graph neural networks (GNNs) Existing SED frameworks rely heavily on GNNs, which require complex graph construction and time-consuming training processes. |
| Approach: | They propose a framework that leverages the rich background knowledge of large language models to formalize and disambiguate short texts by completing abbreviations and summarizing informal expressions. |
| Outcome: | The proposed framework outperforms existing models on two challenging real-world datasets. |
Self-Evolution Fine-Tuning for Policy Optimization (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Recent years have showcased the remarkable capabilities and performance of large language models (LLMs) across a broad range of tasks. |
| Approach: | They propose supervised fine-tuning (SEFT) for LLM alignment to eliminate the need for annotated samples while retaining the stability and efficiency of SFT. |
| Outcome: | The proposed method eliminates the need for annotated samples while maintaining the stability and efficiency of SFT. |
Learning Invariant Representation Improves Robustness for MRC Models (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to improve machine reading comprehension models are vulnerable and not robust to adversarial examples. |
| Approach: | They propose to construct positive example pairs which have same answer by augmentation and then introduce stability and contrastive loss to improve invariance of representation. |
| Outcome: | The proposed approach boosts the robustness of QA models across different tasks and attack sets significantly and consistently. |
Half-S: Halving the Scale for Near-Lossless 4-Bit LLM Training (2026.findings-acl)
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
| 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. |
Beyond the First Error: Process Reward Models for Reflective Mathematical Reasoning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for training effective PRMs focus on the first incorrect step and all preceding steps, assuming that all subsequent steps are incorrect. |
| Approach: | They propose a data annotation method specifically designed to score the long CoT reasoning process by using an LLM-based judger for annotation. |
| Outcome: | The proposed method improves PRMs' ability to identify effective self-correction behaviors and reasoning based on erroneous steps. |
Noise Stability Regularization for Improving BERT Fine-tuning (2021.naacl-main)
Copied to clipboard
| Challenge: | Recent studies show that fine-tuning pre-trained language models is unstable when there are only a small number of training samples available. |
| Approach: | They propose to use a method to regularize noise in deep nets to improve fine-tuning on NLP tasks. |
| Outcome: | The proposed method improves fine-tuning on natural language processing tasks by incorporating noise to the input and demonstrating generalizability and stability. |
IDEAW: Robust Neural Audio Watermarking with Invertible Dual-Embedding (2024.emnlp-main)
Copied to clipboard
| Challenge: | Traditional methods for embedding watermarks into audio have low capacity and unsatisfactory imperceptibility. |
| Approach: | They propose a dual-embedding wa- termarking model for efficient locating and a model that can withstand attacks. |
| Outcome: | The proposed model can withstand attacks with higher capacity and more efficient locating ability compared to existing methods. |
StableToolBench-MirrorAPI: Modeling Tool Environments as Mirrors of 7,000+ Real-World APIs (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing tool environments face challenges in balancing stability, scale, and realism, especially for benchmarking purposes. |
| Approach: | They propose a framework that trains specialized LLMs to accurately simulate real API responses by supervised fine-tuning and chain-of-thought reasoning. |
| Outcome: | The proposed framework achieves superior accuracy and stability compared to state-of-the-art methods on the newly constructed MirrorAPI-Bench and its integration into StableToolBench. |
Large Language Models are Not Yet Human-Level Evaluators for Abstractive Summarization (2023.findings-emnlp)
Copied to clipboard
| Challenge: | ChatGPT and GPT-4 are popular as evaluation metric for complex generative tasks . however, they are not ready as human replacements due to significant limitations . |
| Approach: | They conduct extensive analysis to examine the stability and reliability of LLMs as automatic evaluators for abstractive summarization. |
| Outcome: | The proposed methods outperform the commonly used automatic metrics but are not ready for human evaluation due to significant limitations. |
Critic-Guided Decoding for Controlled Text Generation (2023.findings-acl)
Copied to clipboard
| Challenge: | Recent work has demonstrated reinforcement learning and weighted decoding as effective approaches to achieve a higher level of language control and quality with pros and cons. |
| Approach: | They propose a method that combines reinforcement learning and weighted decoding to train a critic from reward models. |
| Outcome: | The proposed method generates more coherent and well-controlled texts than previous methods on three controlled generation tasks, topic control, sentiment control, and detoxification. |
SCOUT: Selective Coupling via Optimal Unbalanced Transport for Interpretable Text Classification (2026.acl-long)
Copied to clipboard
| Challenge: | Standard interpretable models often rely on scalar similarities that obscure the true evidentiary basis of a prediction. |
| Approach: | They propose a new paradigm that grounds prototype reasoning in the selective correspondence of discriminative fragments. |
| Outcome: | The proposed model outperforms rationale extraction and post-hoc attribution methods on seven benchmarks. |
Understanding Cross-Domain Adaptation in Low-Resource Topic Modeling (2025.acl-long)
Copied to clipboard
| Challenge: | Existing topic modeling models struggle in low-resource settings where data is limited . et al., 2003: domain adaptation for low-source topic modeling is challenging in low resources . |
| Approach: | They propose a domain adaptation framework that disentangles domaininvariant and domain-specific components to improve topic adaptation. |
| Outcome: | The proposed model outperforms state-of-the-art methods on low-resource datasets on diverse datasets. |
CollabKG: A Learnable Human-Machine-Cooperative Information Extraction Toolkit for (Event) Knowledge Graph Construction (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing IE tools lack multi-task support and automatic updates for KG and EKG construction. |
| Approach: | They propose a human-machine-cooperative IE toolkit for KG and EKG construction that unifies different IE subtasks and integrates LLMs as the assistant machine. |
| Outcome: | The proposed tool improves annotation quality, efficiency, and stability simultaneously. |
AlignMMBench: Evaluating Chinese Multimodal Alignment in Large Vision-Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Existing benchmarks focus on basic abilities using nonverbal methods, such as yes-no and multiple-choice questions. |
| Approach: | They propose a benchmark that provides more nuanced evaluations of alignment capabilities for large Vision-Language Models (VLMs) they use a rule-calibrated evaluator that exceeds GPT-4's evaluation ability and a “alignment score” to assess the robustness and stability of models across diverse prompts. |
| Outcome: | The proposed benchmark covers 13 tasks across three categories and includes both single-turn and multi-turn dialogue scenarios. |
Energy-Based Reranking: Improving Neural Machine Translation Using Energy-Based Models (2021.acl-long)
Copied to clipboard
Sumanta Bhattacharyya, Amirmohammad Rooshenas, Subhajit Naskar, Simeng Sun, Mohit Iyyer, Andrew McCallum
| Challenge: | Autoregressive neural machine translation (NMT) uses a tractable likelihood computation and efficient sampling. |
| Approach: | They propose to use an energy-based model to mimic the behavior of the task measure and use it to train an energy based re-ranking algorithm. |
| Outcome: | The proposed model improves on the samples drawn from the NMT with a higher BLEU score than the experimental model and the energy-based re-ranking algorithm. |
A Multi-Document Coverage Reward for RELAXed Multi-Document Summarization (2022.acl-long)
Copied to clipboard
| Challenge: | Multi-document summarization models are limited by limited references and with maximum-likelihood objectives. |
| Approach: | They propose to fine-tune an MDS baseline with a reward that balances a reference-based metric such as ROUGE with coverage of the input documents. |
| Outcome: | The proposed model improves on the Multi-News and WCEP datasets with a low-variance estimator . the proposed model also improves the coverage of the input documents . |
DiPT: Enhancing LLM Reasoning through Diversified Perspective-Taking (2025.findings-naacl)
Copied to clipboard
| Challenge: | Existing work on improving language model reasoning typically explores a single solution path, which can be prone to errors. |
| Approach: | They propose a novel approach that incorporates diversified viewpoints into existing reasoning methods to improve their reasoning performance. |
| Outcome: | The proposed approach can be flexibly integrated into existing models that focus on a single reasoning approach, enhancing their reasoning performance and stability when presented with paraphrased problems. |
Is Word Segmentation Child’s Play in All Languages? (P19-1)
Copied to clipboard
| Challenge: | Existing word learning strategies for infants are cross-linguistically robust . infants do not know which language(s) will be found in their environment at the beginning of development . |
| Approach: | They propose to use 11 conceptually diverse algorithms to learn word-like units in infants . they propose to employ cross-linguistically robust algorithms that can be used by all infants. |
| Outcome: | The proposed algorithms perform above chance on 8 different languages . the results show that some of the algorithms are cross-linguistically valid . |
Disentangling Text Representation With Counter-Template For Unsupervised Opinion Summarization (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches for unsupervised opinion summarization are based on reconstruction model, but selection is too coarse as not all information in each input is equally essential for the summary. |
| Approach: | They propose a framework for unsupervised opinion summarization based on text representation disentanglement with counter-template. |
| Outcome: | The proposed framework outperforms the state-of-the-art models on quality and stability on two benchmark datasets. |
Natural Logic-guided Autoregressive Multi-hop Document Retrieval for Fact Verification (2022.emnlp-main)
Copied to clipboard
| Challenge: | Recent evidence retrieval approaches rely on heuristics and assume hyperlinks between documents. |
| Approach: | They propose a retrieval method that combines a retriever and a proof system that reranks documents and reorders them . |
| Outcome: | The proposed method exceeds or is on par with the current state-of-the-art on FEVER, HoVer and FEVEROUS-S while using 5 to 10 times less memory than competing systems. |
Residual Prompt Tuning: improving prompt tuning with residual reparameterization (2023.findings-acl)
Copied to clipboard
| Challenge: | Prompt tuning is one of the most parameter-efficient approaches for parameter-effective tuning of pre-trained language models. |
| Approach: | They propose to reparameterize soft prompt embeddings using a shallow network with a residual connection and use it to tune prompt embeds P. |
| Outcome: | The proposed method outperforms prompt tuning on SuperGLUE, T5-Base and BERT-Bass models and can reduce the prompt length by 10 times without hurting performance. |
More Tokens, Lower Precision: Towards the Optimal Token-Precision Trade-off in KV Cache Compression (2025.findings-emnlp)
Copied to clipboard
Jiebin Zhang, Dawei Zhu, Yifan Song, Wenhao Wu, Chuqiao Kuang, Xiaoguang Li, Lifeng Shang, Qun Liu, Sujian Li
| Challenge: | storing more tokens in the KV cache at lower precision can enhance the long-context performance of large language models. |
| Approach: | They propose a token-precision trade-off strategy to optimize KV cache compression . they also propose storing more tokens in the KV at lower precision . |
| Outcome: | The proposed method achieves an optimal point within the Information Bottleneck compared to standalone KV pruning or KV quantization. |
Interpretable Semantic Gradients in SSD: A PCA Sweep Approach and a Case Study on AI Discourse (2026.findings-acl)
Copied to clipboard
| 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. |
VidCapBench: A Comprehensive Benchmark of Video Captioning for Controllable Text-to-Video Generation (2025.findings-acl)
Copied to clipboard
Xinlong Chen, Yuanxing Zhang, Chongling Rao, Yushuo Guan, Jiaheng Liu, Fuzheng Zhang, Chengru Song, Qiang Liu, Di Zhang, Tieniu Tan
| Challenge: | Existing studies have not identified a link between video caption evaluation and T2V generation. |
| Approach: | They propose a video caption evaluation scheme specifically designed for T2V generation that integrates video annotation with caption evaluation. |
| Outcome: | The proposed system is agnostic to any particular caption format and can be used for training. |
Inverse-Q*: Token Level Reinforcement Learning for Aligning Large Language Models Without Preference Data (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Reinforcement Learning from Human Feedback (RLHF) relies on complex methodologies like Proximal Policy Optimization (PPO) that require extensive hyper-parameter tuning and pose challenges in sample efficiency and stability. |
| Approach: | They propose an innovative framework that leverages direct preference optimization techniques but extends them by estimating the conditionally optimal policy directly from the model’s responses. |
| Outcome: | The proposed framework matches and exceeds the effectiveness of Proximal Policy Optimization (PPO) in terms of convergence speed and alignment of model responses with human preferences. |
Efficient Shapley Values Estimation by Amortization for Text Classification (2023.acl-long)
Copied to clipboard
| Challenge: | Shapley Values are often estimated with a small number of stochastic model evaluations, but this can only be mitigated by aggregating thousands of model evaluation. |
| Approach: | They propose to combine a model with thousands of model evaluations to estimate Shapley Values without additional model evaluation. |
| Outcome: | The proposed model estimates Shapley Values accurately with up to 60 times speedup compared to traditional methods and does not suffer from stability issues as inference is deterministic. |
The Lawyer That Never Thinks: Consistency and Fairness as Keys to Reliable AI (2025.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are increasingly used in high-stakes domains like law and research. |
| Approach: | They evaluate six leading Large Language Models on rationality, stability, and ethical fairness through reasoning tests, legal challenges, and bias-sensitive scenarios. |
| Outcome: | The models perform well on reasoning tests, legal challenges, and bias-sensitive scenarios. |
Revisiting Demonstration Selection Strategies in In-Context Learning (2024.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) have shown an impressive ability to perform a wide range of tasks using in-context learning (ICL). |
| Approach: | They propose a data- and model-dependent method to select models using in-context learning, TopK + ConE, and propose unified explanations for the effectiveness of previous methods. |
| Outcome: | The proposed method improves language understanding and generation tasks with different model scales. |
Active Learning for Abstractive Text Summarization via LLM-Determined Curriculum and Certainty Gain Maximization (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Abstractive text summarization (ATS) requires laborious data annotation and time-consuming model training. |
| Approach: | They propose a novel active learning framework that asks large language models to rate difficulty of instances and then uses certainty gain maximization to select instances with a distribution that aligns well with the overall distribution. |
| Outcome: | The proposed framework improves stability, effectiveness, and efficiency of abstractive text summarization backbones. |
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
| 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. |
Provably Safe Offline-to-Online RL: Decoupling Learning from Data-Driven Safety Enforcement (2026.acl-long)
Copied to clipboard
| Challenge: | Hybrid offline–online reinforcement learning (O2O RL) promises both sample efficiency and robust exploration, but suffers from instability due to distribution shift between offline and online data. |
| Approach: | They propose a framework that decouples policy optimization from safety enforcement . they propose dynamic curricula that gradually extend temporal horizons and anneal offline–online data mixing . |
| Outcome: | The proposed framework preserves the exploratory value of online interactions without collapsing to conservative policies. |
Enhancing Translation Ability of Large Language Models by Leveraging Task-Related Layers (2024.lrec-main)
Copied to clipboard
| Challenge: | Experimental validation shows that adjusting task-related layers significantly improves performance on translation tasks while maintaining stability and accuracy on other tasks. |
| Approach: | They propose to adjust task-related layers in large models to better harness their machine translation capabilities by revealing the structure and characteristics of attention weights through singular value decomposition. |
| Outcome: | The proposed method reduces computational resource consumption and catastrophic forgetting while maintaining stability and accuracy on other tasks. |
Knowledge Editing through Chain-of-Thought (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing knowledge editing methods focus on multi-hop QA tasks and require frequent retraining. |
| Approach: | They propose a new knowledge editing framework that updates large language models with new information to maintain their world knowledge without retraining. |
| Outcome: | The proposed method achieves state-of-the-art performance while offering superior generalization, effectiveness, and stability compared to existing methods. |
StablePT : Towards Stable Prompting for Few-shot Learning via Input Separation (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing studies on prompt tuning have shown that language models can be effective few-shot learners with prompting. |
| Approach: | They propose to treat the hard prompt and soft prompt as separate inputs to mitigate noise brought by prompt initialization. |
| Outcome: | Experimental results show that the proposed method outperforms state-of-the-art methods by 6.97% in accuracy and reduces the standard deviation by 1.92 on average. |
Using Context-to-Vector with Graph Retrofitting to Improve Word Embeddings (2022.acl-long)
Copied to clipboard
| Challenge: | Contextualized embeddings are expensive and resource-demanding, hence environmentally unfriendly. |
| Approach: | They propose a method to convert contextualized embeddings from pre-trained models into static embeddables using synonym knowledge and weighted vector distribution. |
| Outcome: | The proposed method outperforms baseline embeddings by a large margin through extrinsic and intrinsic tasks. |
ARXSA: A General Negative Feedback Control Theory in Vision-Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | a new approach to the self-attention mechanism is proposed for integrating data from multiple batches. |
| Approach: | They propose an autoregressive with exogenous inputs approach for the Transformer model . the proposed method transforms the Encoder block into a negative feedback predictive control system . |
| Outcome: | The proposed method is validated through comparative evaluations. |
Stable Language Model Pre-training by Reducing Embedding Variability (2024.emnlp-main)
Copied to clipboard
| Challenge: | Stable pre-training is essential for achieving better-performing language models, but tracking pre-train stability is impractical due to high computational costs. |
| Approach: | They propose to use Token Embedding Variability as a proxy to estimate pre-training stability. |
| Outcome: | The proposed method improves stability and lowers perplexities even at deeper layer counts. |
ALDEN: Reinforcement Learning for Active Navigation and Evidence Gathering in Long Documents (2026.acl-long)
Copied to clipboard
| Challenge: | Visually rich documents (VRDs) combine text, tables, and figures within complex, semantically structured layouts. |
| Approach: | They propose a multi-turn reinforcement learning framework that fine-tunes VLMs as interactive agents capable of actively navigating long, visually rich documents. |
| Outcome: | The proposed framework achieves state-of-the-art on five long-document benchmarks. |
Extracting Social Determinants of Health from Pediatric Patient Notes Using Large Language Models: Novel Corpus and Methods (2024.lrec-main)
Copied to clipboard
Yujuan Fu, Giridhar Kaushik Ramachandran, Nicholas J. Dobbins, Namu Park, Michael Leu, Abby R. Rosenberg, Kevin Lybarger, Fei Xia, Özlem Uzuner, Meliha Yetisgen
| Challenge: | Social determinants of health (SDoH) are often studied in the electronic health record (EHR) however, there are difficulties in documenting SDoH in a tabular format due to the lack of a comprehensive SDoh tool. |
| Approach: | They propose to annotate social history sections from 1,260 clinical notes from pediatric patients within the University of Washington (UW) hospital system. |
| Outcome: | The proposed corpus captures ten distinct health determinants including living and economic stability, prior trauma, education access, substance use history, and mental health with an overall annotator agreement of 81.9 F1. |
PrExMe! Large Scale Prompt Exploration of Open Source LLMs for Machine Translation and Summarization Evaluation (2024.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) are useful for low-resource scenarios and time-restricted applications. |
| Approach: | They propose a large-scale evaluation tool for large language models that uses prompts . they evaluate 720 prompt templates for open-source LLM-based metrics on MT and summarization datasets a 6.6M evaluations. |
| Outcome: | The proposed model evaluates 720 prompt templates on machine translation and summarization datasets. |
PanicToCalm: A Proactive Counseling Agent for Panic Attacks (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing models for training such models are limited due to ethical and logistical issues. |
| Approach: | They propose a dataset that includes high-distress episodes constructed from first-person narratives and structured around the principles of Psychological First Aid. |
| Outcome: | The proposed model outperforms baseline models in counselor-side metrics and client affect improvement. |
Predicting Embedding Reliability in Low-Resource Settings Using Corpus Similarity Measures (2022.lrec-1)
Copied to clipboard
| Challenge: | a paper aims to evaluate embedding similarity, stability and reliability in low-resource settings . it uses corpus similarity measures before training to predict properties of embeddables . |
| Approach: | They use corpus similarity measures before training to predict properties of embeddings . they then apply the same measures to low-resource settings by modelling reliability . authors hope to use this method to evaluate low-source languages with limited corpus size . |
| Outcome: | The paper shows that it is possible to predict downstream embedding similarity using upstream corpus similarity measures . the main finding is that the measures remain robust on small amounts of training data . |
Sensitivity-LoRA : Low-Load Sensitivity-Based Fine-Tuning for Large Language Models (2025.findings-emnlp)
Copied to clipboard
Hao Zhang, Bo Huang, Zhenjia Li, Xi Xiao, Hui Yi Leong, Zumeng Zhang, Xinwei Long, Tianyang Wang, Hao Xu
| Challenge: | Low-Rank Adaptation (LoRA) is a promising approach to adapting LLMs to specialized tasks . existing rank allocation techniques remain computationally inefficient and unstable . |
| Approach: | They propose a low-rank adapted model that approximates model weight updates using low-ranked decomposition. |
| Outcome: | The proposed method is limited by its uniform rank allocation to each incremental matrix . it leverages the second-order derivatives of the loss function to capture weight sensitivity . |
Self-training Language Models for Arithmetic Reasoning (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Recent work improves the reasoning capabilities of language models by scaling training data to more diverse or complex collections, but reaching further improvements becomes exceedingly expensive. |
| Approach: | They propose to use implicit feedback to improve models' reasoning capabilities by training from implicit feedback. |
| Outcome: | The proposed model can reach a correct result in +13.9% and +25.9% more cases than previous models, underlining the importance of actuality of self-training feedback. |
A Crosslingual Investigation of Conceptualization in 1335 Languages (2023.acl-long)
Copied to clipboard
Yihong Liu, Haotian Ye, Leonie Weissweiler, Philipp Wicke, Renhao Pei, Robert Zangenfeind, Hinrich Schütze
| Challenge: | Conceptualizer is a method that creates a bipartite directed alignment graph between source language concepts and sets of target language strings. |
| Approach: | They propose a method that creates a bipartite directed alignment graph between source language concepts and sets of target language strings. |
| Outcome: | The proposed method has good alignment accuracy across all languages and on 32 Swadesh concepts. |
Exploring Quantization for Efficient Pre-Training of Transformer Language Models (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Quantization has proven to be effective after pre-training and during fine-tuning, but its effects on pre-trainer performance have remained unexplored. |
| Approach: | They propose a linear quantization strategy to be applied during the pre-training of Transformers to improve model efficiency and stability. |
| Outcome: | The proposed method improves model efficiency, stability, and performance while maintaining language modeling ability. |
IAEval: A Comprehensive Evaluation of Instance Attribution on Natural Language Understanding (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Instance attribution (IA) aims to identify the training instances leading to the prediction of a test example. |
| Approach: | They propose a systematic and comprehensive evaluation scheme covering four significant requirements: sufficiency, completeness, stability and plausibility. |
| Outcome: | The proposed evaluation scheme covers four significant requirements: sufficiency, completeness, stability and plausibility. |
Beware of Model Collapse! Fast and Stable Test-time Adaptation for Robust Question Answering (2023.emnlp-main)
Copied to clipboard
| Challenge: | Pre-trained language models (PLMs) have achieved great success in question answering, but their robustness is insufficient to support their practical applications. |
| Approach: | They propose a method which regularizes the model's output and an efficient side block to reduce its inference time. |
| Outcome: | The proposed method achieves comparable or better results than previous TTA methods at a speed close to vanilla forward propagation, which is 1.8 to 4.4 speedup compared to previous methods. |
Editing Across Languages: A Survey of Multilingual Knowledge Editing (2025.emnlp-main)
Copied to clipboard
| Challenge: | Knowledge Editing is a growing subdomain of model editing focused on ensuring factual edits generalize across languages. |
| Approach: | They present a taxonomy of multilingual knowledge editing methods and benchmarks . authors summarize key findings on method effectiveness and transfer patterns . |
| Outcome: | The proposed methods are compared against available benchmarks and benchmark datasets. |
LIFBench: Evaluating the Instruction Following Performance and Stability of Large Language Models in Long-Context Scenarios (2025.acl-long)
Copied to clipboard
| Challenge: | Existing benchmarks rarely focus on instruction-following in long-context scenarios or stability on different inputs. |
| Approach: | They propose a scalable dataset to evaluate LLMs’ instruction-following capabilities and stability across long contexts. |
| Outcome: | The proposed method evaluates LLMs’ instruction-following capabilities and stability across long contexts. |
Agent-GWO: Collaborative Agents for Dynamic Prompt Optimization in Large Language Models (2026.findings-acl)
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
| 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. |
CARMA: Enhanced Compositionality in LLMs via Advanced Regularisation and Mutual Information Alignment (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) struggle with compositional generalisation, limiting their ability to systematically combine learned components to interpret novel inputs. |
| Approach: | They propose a method that enhances stability and robustness of compositional reasoning in large language models (LLMs) by regularising and preserving layer-wise stability constraints. |
| Outcome: | The proposed method reduces variability, stabilizes token representations, and improves compositional reasoning. |
QaRL: Rollout-Aligned Quantization-Aware RL for Fast and Stable Training under Training–Inference Mismatch (2026.findings-acl)
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. |
Can We Entrust Justice to AI?: How Persona Traps Contaminate Reasoning in Criminal Investigation (2026.findings-acl)
Copied to clipboard
| 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 . |
Probing the “Creativity” of Large Language Models: Can models produce divergent semantic association? (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models possess remarkable capacity for processing language, but it remains unclear whether they can further generate creative content. |
| Approach: | They utilize the divergent association task (DAT) to examine the creative thinking of large language models through a cognitive perspective. |
| Outcome: | The proposed model outperforms the greedy search strategy while outperforming the average human level. |
Dovetail: A CPU/GPU Heterogeneous Speculative Decoding for LLM inference (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) are demanding more memory and computational resources . however, these devices typically feature weaker GPUs and stronger CPUs . |
| Approach: | They propose a lossless inference acceleration method that leverages the characteristics of heterogeneous devices and the advantages of speculative decoding. |
| Outcome: | The proposed method achieves speedups ranging from 1.79 to 10.1 across different devices . it uses a draft model on the GPU to perform preliminary predictions, while a target model on CPU validates these outputs . |
Less is More: Improving LLM Reasoning with Minimal Test-Time Intervention (2026.acl-long)
Copied to clipboard
| Challenge: | Recent advances in large language models (LLMs) have focused on test-time scaling to improve reasoning quality but at the cost of efficiency. |
| Approach: | They propose a training-free framework that enhances reasoning accuracy and stability with minimal overhead. |
| Outcome: | The proposed framework yields consistent gains across general, coding, and STEM tasks while remaining highly efficient. |
SAGE: Sign-Adaptive Gradient for Memory-Efficient LLM Optimization (2026.findings-acl)
Copied to clipboard
| 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. |
BSFA: Leveraging the Subspace Dichotomy to Accelerate Neural Network Training (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies highlight a fundamental dichotomy in deep learning optimization: parameter updates along the top eigendirections of the loss Hessian (Dom-space) capture most of the update magnitude, while updates in the orthogonal component (Bulk-space), have smaller magnitudes but drive most learning progress. |
| Approach: | They propose a plug-and-play framework that scales update components projected onto distinct subspaces and a block-wise strategy that applies this estimation on a per-parameter-block basis. |
| Outcome: | The proposed framework accelerates training by differentially scaling update components projected onto distinct subspaces, while enhancing stability by moderating updates in dominant subspace and boosting convergence speed by amplifying updates in bulk-space. |
ReSURE: Regularizing Supervision Unreliability for Multi-turn Dialogue Fine-tuning (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods address data quality via static prefiltering, which decouples quality control from training and fails to mitigate turn-level error propagation. |
| Approach: | They propose an adaptive learning method that dynamically down-weights unreliable supervision without explicit filtering. |
| Outcome: | Experiments on single-source and mixed-quality datasets show improved stability and response quality. |
Guiding not Forcing: Enhancing the Transferability of Jailbreaking Attacks on LLMs via Removing Superfluous Constraints (2025.acl-long)
Copied to clipboard
| Challenge: | Jailbreaking attacks can effectively induce unsafe behaviors in Large Language Models (LLMs). |
| Approach: | They propose a conceptual framework to elucidate transferability of gradient-based jailbreaking methods . they identify superfluous constraints as significant barriers to improved transferability . |
| Outcome: | The proposed method increases the overall transfer attack success rate (T-ASR) across target and source models with varying safety levels from 18.4% to 50.3% while improving stability and controllability of jailbreak behaviors. |
Systematic Evaluation of Auto-Encoding and Large Language Model Representations for Capturing Author States and Traits (2025.findings-acl)
Copied to clipboard
Khushboo Singh, Vasudha Varadarajan, Adithya V Ganesan, August Håkan Nilsson, Nikita Soni, Syeda Mahwish, Pranav Chitale, Ryan L. Boyd, Lyle Ungar, Richard N Rosenthal, H. Schwartz
| Challenge: | Large Language Models (LLMs) are increasingly used in human-centered applications, yet their ability to model diverse psychological constructs is not well understood. |
| Approach: | They evaluated a range of Transformer-LMs to predict psychological variables across five major dimensions: affect, substance use, mental health, sociodemographics, and personality. |
| Outcome: | The models predict affect, substance use, mental health, sociodemographics, and personality across five major dimensions. |
Mitigating Hallucinations in LM-Based TTS Models via Distribution Alignment Using GFlowNets (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing mitigation strategies for Text-to-Speech systems require excessive training resources or inference latency. |
| Approach: | They propose a GFlOwNet-guided distribution AlignmenT framework that mitigates hallucinations without relying on massive resources or inference latency. |
| Outcome: | The proposed framework reduces over 50% character error rates and lowers uncertainty by up to 58% on challenging test cases. |
SedarEval: Automated Evaluation using Self-Adaptive Rubrics (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing evaluation paradigms rely on generic scoring rubrics that fail to consider the specificities of each question and its problem-solving process. |
| Approach: | They propose a new evaluation paradigm based on self-adaptive rubrics that mimic a human evaluator's analytical process. |
| Outcome: | The proposed evaluation paradigm achieves higher concordance rate with human graders than existing paradigms, including GPT-4. |
Preserving Fairness and Safety in Quantized LLMs Through Critical Weight Protection (2026.findings-acl)
Copied to clipboard
| 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. |
Rectifying the Emotional Flow: Aligning Priors and Dynamic Guidance for High-Arousal Text-to-Speech (2026.acl-long)
Copied to clipboard
| Challenge: | Existing systems suffer from linguistic collapse when pursuing high intensity or fail to meet target emotional levels. |
| Approach: | They propose an inference framework that introduces a neutral prosody bias and a uniform Classifier-Free Guidance that distorts the acoustic manifold, leading to artifacts. |
| Outcome: | The proposed framework achieves superior linguistic accuracy and expressiveness without model retraining. |
FFAEval: Evaluating Dialogue System via Free-For-All Ranking (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing evaluation metrics for open-domain dialogue systems show poor correlation with human assessment. |
| Approach: | They propose a free-for-all human evaluation framework that shares dialogue history with annotators for multi-turn scoring. |
| Outcome: | The proposed framework achieves a strong correlation with human assessment on English and Chinese dialogue systems. |
MUR: Momentum Uncertainty guided Reasoning for Large Language Models (2026.acl-long)
Copied to clipboard
Hang Yan, Fangzhi Xu, Rongman Xu, Yifei Li, Jian Zhang, Haoran Luo, Xiaobao Wu, Anh Tuan Luu, Haiteng Zhao, Qika Lin, Jun Liu
| Challenge: | Existing methods for optimizing reasoning quality are limited by overthinking. |
| Approach: | They propose a method that allocates thinking budgets to critical reasoning steps by tracking and aggregating step-wise uncertainty over time. |
| Outcome: | The proposed method reduces computation by over 45% on average while improving accuracy by 0.33–3.46%. |
Beyond Static Toolsets: Self-Evolving LLM Tool Agents via Continual Documentation Adaptation (2026.findings-acl)
Copied to clipboard
| 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. |
DPO Kernels: A Semantically-Aware, Kernel-Enhanced, and Divergence-Rich Paradigm for Direct Preference Optimization (2025.findings-acl)
Copied to clipboard
Amitava Das, Suranjana Trivedy, Danush Khanna, Yaswanth Narsupalli, Basab Ghosh, Rajarshi Roy, Gurpreet Singh, Vinija Jain, Vasu Sharma, Aishwarya Naresh Reganti, Aman Chadha
| Challenge: | Direct Preference Optimization (DPO) is a cornerstone for preference alignment but is constrained by fixed divergence measures and limited feature transformations. |
| Approach: | They propose a new enhancement of Direct Preference Optimization that integrates kernel methods to overcome these challenges. |
| Outcome: | The proposed model improves divergence measures and features by using kernels . the proposed model achieves state-of-the-art generalization in factuality, safety, reasoning, and instruction following . |
PROM: A Phrase-level Copying Mechanism with Pre-training for Abstractive Summarization (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing summarization strategies are abstractive and extractive, but are hard to control. |
| Approach: | They propose a PhRase-level cOpying Mechanism that enhances attention on n-grams and calculates an auxiliary loss for the copying prediction. |
| Outcome: | Empirical studies show that PROM improves copying accuracy and faithfulness on benchmarks. |
Beyond Overlap Metrics: Rewarding Reasoning and Preferences for Faithful Multi-Role Dialogue Summarization (2026.findings-acl)
Copied to clipboard
| 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. |
MM-JudgeBias: A Benchmark for Evaluating Compositional Biases in MLLM-as-a-Judge (2026.acl-long)
Copied to clipboard
| Challenge: | Multimodal Large Language Models (MLLMs) are increasingly used as automatic judges . however, their reliability and vulnerabilities to biases remain underexplored . |
| Approach: | They propose a benchmark to evaluate MLLMs that fail to integrate visual cues . they also introduce a test to evaluate the reliability of MLMLs based on a set of asymmetric evaluation tendencies. |
| Outcome: | Experiments on 26 state-of-the-art MLLMs reveal modality neglect and asymmetric evaluation tendencies . a standardized model with a benchmark enables a fine-grained diagnosis of nine bias types . |
Towards Stable and Effective Reinforcement Learning for Mixture-of-Experts (2026.acl-long)
Copied to clipboard
| Challenge: | Reinforcement learning with verifiable rewards (RLVR) training with Mixture-of-Experts policies remains fragile and prone to reward collapse. |
| Approach: | They propose a router shift-based policy optimization method that computes a per-token router-shift ratio conditioned on the previously activated experts and applies stop-gradient and a lower-bound floor. |
| Outcome: | The proposed method achieves better performance and greater stability than previous methods. |
MemPO: Self-Memory Policy Optimization for Long-Horizon Agents (2026.findings-acl)
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. |
QDMR-based Planning-and-Solving Prompting for Complex Reasoning Tasks (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing Plan-and-Solve prompting methods are difficult to implement for complex questions. |
| Approach: | They propose a plan-and-solve prompting method based on Question Decomposition Meaning Representation (QDMR) it allows LLM to generate a QDMR graph to represent problem-solving logic . |
| Outcome: | The proposed method can represent and execute the problem-solving logic of complex questions more accurately than existing methods. |
Demystifying Data Organization for Enhanced LLM Training (2026.acl-long)
Copied to clipboard
Yalun Dai, Yangyu Huang, Tongshen Yang, Yonghan Wang, Xin Zhang, Wenshan Wu, Qihao Zhao, Hao Li, Yuanyuan Gao, Kim-Hui Yap, Scarlett Li
| Challenge: | Large Language Models (LLMs) have revolutionized various fields, yet their training efficiency is heavily reliant on effective data curation. |
| Approach: | They propose to reuse pre-computed sample-level scores originally generated for data efficiency and introduce two new data ordering methods to improve LLM training. |
| Outcome: | The proposed methods improve the stability and performance of LLM training. |
Generative Reward Modeling via Synthetic Criteria Preference Learning (2025.acl-long)
Copied to clipboard
| Challenge: | Generative Reward Models (GenRMs) leverage synthesized Chains of Thought (CoT) but this approach introduces risks of overoptimization due to the inability to guarantee the correctness of the CoTs. |
| Approach: | They propose a criteria-based preference tree for GenRMs that uses chain of thought to generate reasoning . they show that synthesized data can be learned using a long CoT format . |
| Outcome: | The proposed model shows significant improvements over baselines on multiple human preference benchmarks. |
Compatibility-Aware Dynamic Fine-Tuning for Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Recent work attributes optimization instability to the low probability of demonstrations being incompatible with the sample level. |
| Approach: | They propose a Dynamic Fine-Tuning extension of DFT that controls sample-level optimization variance. |
| Outcome: | The proposed model can generalize token-level stabilization to the sample level while remaining fully supervised and free of reward modeling. |
MDTeamGPT: Mitigating Context Collapse and Enabling Self-Evolution in Medical Multi-Agent Reasoning (2026.findings-acl)
Copied to clipboard
| 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. |
SAFO: Stable Adaptive Fairness Optimization for LLM-Based Social Survey Simulation (2026.acl-long)
Copied to clipboard
| Challenge: | Social survey simulations are increasingly used to improve minority performance and social-welfare metrics. |
| Approach: | They propose a dynamic utility–fairness optimization framework for LLM-based survey simulation that explicitly targets fairness and training stability. |
| Outcome: | The proposed framework improves minority performance and social-welfare metrics on three large-scale survey datasets from China, the U.S. and Europe. |
Stress Testing Factual Consistency Metrics for Long-Document Summarization (2026.acl-long)
Copied to clipboard
| Challenge: | Existing short-form summarization metrics struggle with input length limitations and long-range dependencies. |
| Approach: | They propose to evaluate the reliability of six widely used reference-free factuality metrics in the long-document setting by applying seven factually-preserving perturbations to summaries. |
| Outcome: | The proposed short-form summarization metrics struggle with long-range dependencies and input length limitations. |
Rethinking Expert Trajectory Utilization in LLM Post-training for Mathematical Reasoning (2026.acl-long)
Copied to clipboard
Bowen Ding, Yuhan Chen, Jiayang Lyu, Jiyao Yuan, Qi Zhu, Shuangshuang Tian, Dantong Zhu, Futing Wang, Heyuan Deng, Fei Mi, Lifeng Shang, Tao Lin
| Challenge: | evolving generic Large Language Models into specialized Large Reasoning Models requires effective post-training. |
| Approach: | They propose a plasticity-ceiling framework to harness expert trajectories . they establish the Sequential SFT-then-RL pipeline as the superior standard . |
| Outcome: | The proposed framework overcomes stability and premature convergence deficits in synchronized approaches. |
What Happened in LLMs Layers when Trained for Fast vs. Slow Thinking: A Gradient Perspective (2025.acl-long)
Copied to clipboard
| Challenge: | Xu et al., 2024) study shows that slow thinking can distinguish correct and irrelevant reasoning paths. |
| Approach: | They investigate how fast vs. slow thinking affects layer-wise gradients in large language models . they find that slow thinking can distinguish correct and irrelevant reasoning paths . |
| Outcome: | The results show that slow thinking can distinguish correct and irrelevant reasoning paths. |
Efficient Domain Continual pretraining by Mitigating the Stability Gap (2025.acl-long)
Copied to clipboard
| Challenge: | Continual pretraining is an important approach for Large Language Models to improve their performance in target domains, learn new topics and languages, and even boost their general capabilities. |
| Approach: | They propose a training strategy that mitigates instability by increasing the number of epochs, along with two data sampling strategies targeting data domain relevance and corpus distribution. |
| Outcome: | The proposed training strategy improves the average medical task performance of the OpenLlama-3B model from 36.2% to 40.7% using only 40% of the original training budget, while also enhancing general task performance without causing forgetting. |
Reasoning under Uncertainty: Efficient LLM Inference via Unsupervised Confidence Dilution and Convergent Adaptive Sampling (2025.emnlp-main)
Copied to clipboard
Zhenning Shi, Yijia Zhu, Yi Xie, Junhan Shi, Guorui Xie, Haotian Zhang, Yong Jiang, Congcong Miao, Qing Li
| Challenge: | Large language models suffer from overconfidence and computational inefficiency due to fixed computation budgets and miscalibrated confidence estimates. |
| Approach: | They propose a framework for computationally efficient, trustworthy reasoning under uncertainty using Diversity-Aware Self-Signal Dilution and Convergent Adaptive Weighted Sampling techniques. |
| Outcome: | The proposed framework reduces inference cost by 70% while maintaining accuracy levels while reducing inference costs. |
VCB Bench: An Evaluation Benchmark for Audio-Grounded Large Language Model Conversational Agents (2026.findings-acl)
Copied to clipboard
Jiliang Hu, Wenfu Wang, Zuchao Li, Chenxing Li, Yiyang Zhao, Hanzhao Li, Liqiang Zhang, Meng Yu, Dong Yu
| 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 . |
EMSDialog: Synthetic Multi-person Emergency Medical Service Dialogue Generation from Electronic Patient Care Reports via Multi-LLM Agents (2026.findings-acl)
Copied to clipboard
| 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. |
Beyond Single Representations: Multi-Model Embedding Fusion for Stable Text Classification (2026.acl-long)
Copied to clipboard
| Challenge: | Existing studies on embedding fusion have not evaluated the effectiveness of individual layers or the impact of combining embeddables from multiple models. |
| Approach: | They propose to combine embeddings from multiple models to improve performance across NLP tasks. |
| Outcome: | The proposed method improves performance on low-resource datasets and reduces the impact of any single model as the number of integrated models increases. |
Modeling and Solving Stable Matching under Probabilistic Preferences with Large Language Models (2026.findings-acl)
Copied to clipboard
| 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. |
From log 𝜋 to 𝜋: Taming Divergence in Soft Clipping via Bilateral Decoupled Decay of Probability Gradient Weight (2026.acl-long)
Copied to clipboard
Xiaoliang Fu, Jiaye Lin, Yangyi Fang, Chaowen Hu, Cong Qin, Zekai Shao, Binbin Zheng, Lu Pan, Ke Zeng
| Challenge: | Standard algorithms for Large Language Models (LLMs) enforce stability via "hard clipping" but relying on log-probability gradient yields divergent weights as probabilities vanish, destabilizing LLM training. |
| Approach: | They propose a decoupled gradient policy optimization that uses a decay mechanism to decouple the probability of a boundary token. |
| Outcome: | The proposed algorithm outperforms baselines on various mathematical benchmarks. |
PsychePass: Calibrating LLM Therapeutic Competence via Trajectory-Anchored Tournaments (2026.findings-acl)
Copied to clipboard
| 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. |
HeteroCache: A Dynamic Retrieval Approach to Heterogeneous KV Cache Compression for Long-Context LLM Inference (2026.acl-long)
Copied to clipboard
| Challenge: | Existing static compression methods suffer from coarse-grained caching and high I/O overhead. |
| Approach: | They propose a training-free dynamic compression framework that uses a sparse attention mechanism to categorize attention heads based on stability and similarity. |
| Outcome: | The proposed framework achieves state-of-the-art performance on long-context benchmarks and accelerates decoding by up to 3 compared to the original model with a 224K context. |
Entropy-Aware Reshaping of Reinforcement Signals for Multi-Answer Reasoning (2026.findings-acl)
Copied to clipboard
| 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. |
Beyond Prompt Engineering: A Systematic Analysis of Prompt Lexical Sensitivity and Its Impacts on Quality (2026.findings-acl)
Copied to clipboard
Qipeng Xie, Zi Liang, Jiafei Wu, Yufei Chen, Weizheng Wang, Wenao Ma, Zhong Ming, Haiqin Yang, Kaishun Wu
| 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 . |