Papers by Liang Lu
Consecutive Batch Model Editing with HooK Layers (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing models that retrain are time- and resource-consuming, but they lack the memory to support sequential and batch editing. |
| Approach: | They propose a model editing method that supports sequential and batch editing . they use a small amount of memory to store several hook layers that remain unchanged over time . |
| Outcome: | The proposed method is memory-friendly and can store hook layers that remain unchanged over time. |
Aegis:An Advanced LLM-Based Multi-Agent for Intelligent Functional Safety Engineering (2024.emnlp-industry)
Copied to clipboard
| Challenge: | Aegis is an advanced LLM-based multi-agent for intelligent functional safety engineering that can perform all phases of a vehicle's lifecycle, including design, development, production, operation, and decommissioning. |
| Approach: | They introduce Aegis: An Advanced LLM-Based Multi-Agent for Intelligent Functional Safety Engineering. |
| Outcome: | The proposed solution can perform Hazard Analysis and Risk Assessment (HARA), document Functional Safety Requirements (FSR), and plan test cases for Automatic Emergency Braking (AEB) systems. |
Immediate Inference: The Missing Foundation in Large Language Model Logical Reasoning (2026.acl-long)
Copied to clipboard
| Challenge: | Recent work on LLMs has focused on fine-grained skill decomposition and consistency probing at the propositional level. |
| Approach: | They propose a benchmark evaluating immediate inference that evaluates elemental operations over categorical propositions and proposes a model that uses immediate inferential reasoning. |
| Outcome: | The proposed benchmark demonstrates that models lack robust operator grounding, oscillating between structural reasoning and surface pattern matching, inconsistent handling of quantifiers and negation. |
SocAoG: Incremental Graph Parsing for Social Relation Inference in Dialogues (2021.acl-long)
Copied to clipboard
| Challenge: | Existing studies focus on identifying entities' relations from the semantics of dialogues-they utilize either the attention mechanism or a refined token graph to locate informative words. |
| Approach: | They propose a sequential structure prediction task to incrementally parse SocAoG for dynamic inference upon any incoming utterance. |
| Outcome: | Empirical results show that the proposed model infers social relations more accurately than the state-of-the-art methods. |
Improved Neural Protoform Reconstruction via Reflex Prediction (2024.lrec-main)
Copied to clipboard
| Challenge: | comparative method allows linguists to infer protoforms from their reflexes based on sound change . authors argue that this approach ignores one of the most important aspects of the comparative approach . |
| Approach: | They propose a comparative method that allows linguists to infer protoforms from their reflexes . they propose to use a system where candidate protoform from a reconstruction model are reranked by a reflex prediction model. |
| Outcome: | The comparative method surpasses state-of-the-art methods on Chinese and Romance datasets. |
RepoAgent: An LLM-Powered Open-Source Framework for Repository-level Code Documentation Generation (2024.emnlp-demo)
Copied to clipboard
Qinyu Luo, Yining Ye, Shihao Liang, Zhong Zhang, Yujia Qin, Yaxi Lu, Yesai Wu, Xin Cong, Yankai Lin, Yingli Zhang, Xiaoyin Che, Zhiyuan Liu, Maosong Sun
| Challenge: | Xia et al., 2018) demonstrate that a large language model can generate and maintain high-quality code documentation. |
| Approach: | They propose a large language model powered open-source framework for generating, maintaining, and updating code documentation. |
| Outcome: | The proposed framework generates high-quality documentation for the entire project. |
SDiaReward: Modeling and Benchmarking Spoken Dialogue Rewards with Modality and Colloquialness (2026.acl-long)
Copied to clipboard
Jingyu Lu, Yuhan Wang, Fan Zhuo, Xize Cheng, Changhao Pan, Xueyi Pu, Yifu Chen, Chenyuhao Wen, Tianle Liang, Zhou Zhao
| Challenge: | SDiaReward is an end-to-end spoken dialogue system that integrates paralinguistic nuances and spontaneous nature of human conversation. |
| Approach: | They propose an end-to-end multi-turn reward model trained on SDiaReward-Dataset . it is a collection of episode-level preference pairs targeting modality and colloquiality gaps . |
| Outcome: | The proposed model outperforms general-purpose audio LLMs in episode-level evaluation. |
A Closer Look at Few-Shot Out-of-Distribution Intent Detection (2022.coling-1)
Copied to clipboard
| Challenge: | Existing methods for few-shot out-of-distribution (OOD) intent detection are not adequate . despite its importance, few- shot OOD intent detection is a challenging problem . |
| Approach: | They propose a latent representation generation and self-supervision approach to solve few-shot OOD intent detection problem. |
| Outcome: | The proposed approach is highly effective and could improve state-of-the-art methods for few-shot OOD intent detection. |
SPARD: Self-Paced Curriculum for RL Alignment via Integrating Reward Dynamics and Data Utility (2026.acl-long)
Copied to clipboard
Xuyang Zhi, Peilun Zhou, Chengqiang Lu, Hang Lv, Yiwei Liang, Rongyang Zhang, Yan Gao, null Yiwu, Yao Hu, Hongchao Gu, Defu Lian, Hao Wang, Enhong Chen
| Challenge: | Large Language Models (LLMs) are shifting the focus from single verifiable tasks toward complex, open-ended real-world scenarios. |
| Approach: | They propose a framework that automatically adjusts reward weights and data importance to synchronize learning intent with data utility for optimal performance. |
| Outcome: | The proposed framework improves model capabilities across all domains and scales. |
Data Interpreter: An LLM Agent for Data Science (2025.findings-acl)
Copied to clipboard
Sirui Hong, Yizhang Lin, Bang Liu, Bangbang Liu, Binhao Wu, Ceyao Zhang, Danyang Li, Jiaqi Chen, Jiayi Zhang, Jinlin Wang, Li Zhang, Lingyao Zhang, Min Yang, Mingchen Zhuge, Taicheng Guo, Tuo Zhou, Wei Tao, Robert Tang, Xiangtao Lu, Xiawu Zheng, Xinbing Liang, Yaying Fei, Yuheng Cheng, Yongxin Ni, Zhibin Gou, Zongze Xu, Yuyu Luo, Chenglin Wu
| Challenge: | Large Language Models (LLMs) excel in various domains but face challenges when applied to data science workflows due to their complex, multi-stage nature. |
| Approach: | They propose a hierarchical graph-based agent that represents complexity and a progressive strategy for step-by-step verification, refinement, and consistent context management. |
| Outcome: | The proposed agent surpasses state-of-the-art baselines on the MATH dataset and performs better on InfiAgent-DABench. |
PclGPT: A Large Language Model for Patronizing and Condescending Language Detection (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Patronizing and condescending language is an essential branch of toxic language . pre-trained language models perform poorly in detecting PCL due to its implicit toxicity traits . |
| Approach: | They propose a novel LLM benchmark for patronizing and condescending language . they use a dataset to analyze the toxicity of patronizing condescending languages . |
| Outcome: | The proposed model can detect patronizing and condescending language (PCL) the model can be used to analyze the toxicity of the language and to improve the detection. |
HELENE: Hessian Layer-wise Clipping and Gradient Annealing for Accelerating Fine-tuning LLM with Zeroth-order Optimization (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) face memory challenges due to the high cost of backpropagation. |
| Approach: | They propose a zeroth-order (ZO) optimization that matches memory usage to inference . they propose scalable and memory-efficient zeroth order (ZE) optimizer that integrates annealed A-GNB gradients with diagonal Hessian estimation and layer-wise clipping as a second-order pre-conditioner. |
| Outcome: | The proposed algorithm outperforms state-of-the-art methods with an average speedup of 20 over MeZO on RoBERTa-large and OPT-1.3B. |
MinerU2.5: A Decoupled Vision-Language Model for Efficient High-Resolution Document Parsing (2026.acl-industry)
Copied to clipboard
Junbo Niu, Zheng Liu, Zhuangcheng Gu, Bin Wang, Linke Ouyang, Zhiyuan Zhao, Tao Chu, Tianyao He, Fan Wu, Qintong Zhang, Zhenjiang Jin, Guang Liang, Rui Zhang, Wenzheng Zhang, Yuan Qu, Zhifei Ren, Yuefeng Sun, Zirui Tang, Boyu Niu, Yuanhong Zheng, Dongsheng Ma, Ziyang Miao, Hejun Dong, Siyi Qian, Junyuan Zhang, Fangdong Wang, Jingzhou Chen, Xiaomeng Zhao, Liqun Wei, Wei Li, Shasha Wang, RuiLiang Xu, Yuanyuan Cao, Lu Chen, Qianqian Wu, Huaiyu Gu, Lindong Lu, Dechen Lin, null Shenguanlin, Xuanhe Zhou, Linfeng Zhang, Yuhang Zang, Xiaoyi Dong, Jiaqi Wang, Bo Zhang, Lei Bai, Pei Chu, Weijia Li, Jiang Wu, Lijun Wu, Zhenxiang Li, Guangyu Wang, Zhongying Tu, Chao Xu, Kai Chen, Bowen Zhou, Dahua Lin, Wentao Zhang, Conghui He
| Challenge: | Document images are characterized by higher resolutions, denser content, and more complex structural layouts. |
| Approach: | They propose a 1.2B-parameter document parsing vision-language model that decouples layout analysis from local content recognition. |
| Outcome: | The proposed model surpasses general-purpose and domain-specific models on multiple benchmarks while maintaining significantly lower computational overhead. |
TransBench: Breaking Barriers for Transferable Graphical User Interface Agents in Dynamic Digital Environments (2025.findings-acl)
Copied to clipboard
Yuheng Lu, Qian Yu, Hongru Wang, Zeming Liu, Wei Su, Yanping Liu, Yuhang Guo, Maocheng Liang, Yunhong Wang, Haifeng Wang
| Challenge: | Existing GUI agents struggle to adapt to dynamic and interconnected nature of real-world digital environments, authors show . |
| Approach: | They propose a benchmark to evaluate the transferability of GUI agents across three key dimensions . transBench includes 15 app categories with diverse functionalities . |
| Outcome: | The proposed benchmark shows that existing GUI agents struggle to adapt to dynamic, interconnected environments. |
Parameter Importance is Not Static: Evolving Parameter Isolation for Supervised Fine-Tuning (2026.acl-long)
Copied to clipboard
Zekai Lin, Chao Xue, Di Liang, Xingsheng Han, Peiyang Liu, Xianjie Wu, Lei Jiang, Yu Lu, Bob Simons, Shuang Liang, Minlong Peng
| Challenge: | Recent approaches to fine-tuning of large language models suffer from task interference and catastrophic forgetting. |
| Approach: | They propose a fine-tuning framework that adapts isolation decisions based on online estimates of parameter importance. |
| Outcome: | The proposed framework reduces interference and forgetting while releasing outdated parameters to recover plasticity. |
DialogGen: Multi-modal Interactive Dialogue System with Multi-turn Text-Image Generation (2025.findings-naacl)
Copied to clipboard
Minbin Huang, Yanxin Long, Xinchi Deng, Ruihang Chu, Jiangfeng Xiong, Xiaodan Liang, Hong Cheng, Qinglin Lu, Wei Liu
| Challenge: | Text-to-image (T2I) generation models have advanced in recent years, but effective interaction with these models is challenging for average users due to the need for specialized prompt engineering knowledge and the inability to perform multi-turn image generation. |
| Approach: | They propose to use off-the-shelf MLLMs and T2I models to build a multi-modal interactive dialogue system (MIDS) that can generate correct output modalities and coherence of output images. |
| Outcome: | The proposed pipeline can generate correct output modalities and coherent multi-modal outputs compared with other state-of-the-art models. |
Why Supervised Fine-Tuning Fails to Learn: A Systematic Study of Incomplete Learning in Large Language Models (2026.acl-long)
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
| Challenge: | Incomplete learning is widespread and heterogeneous in large language models . authors identify five recurrent sources of incomplete learning: missing prerequisite knowledge, conflicts between SFT supervision and pre-training knowledge, internal inconsistencies within SFT data, left-side forgetting during sequential fine-tuning, and insufficient optimization for rare or complex patterns. |
| Approach: | They propose a diagnostic-first framework that maps incomplete learning to causes . they identify five recurrent sources of incomplete learning: missing prerequisite knowledge, conflicts between supervision and pre-training knowledge, internal inconsistencies, left-side forgetting during sequential fine-tuning, and insufficient optimization for rare or complex patterns. |
| Outcome: | The proposed framework maps incomplete learning to causes using observable training and inference signals. |
SAGE : A Top-Down Bottom-Up Knowledge-Grounded User Simulator for Multi-turn Agent Evaluation (2026.findings-eacl)
Copied to clipboard
| Challenge: | Existing evaluation methods rely on static benchmarks or narrow task-specific datasets that fail to capture the open-ended nature of real-world interactions. |
| Approach: | They propose a user Simulation framework for multi-turn AGent Evaluation that integrates top-down knowledge from business contexts and bottom-up knowledge from agent infrastructure. |
| Outcome: | The proposed framework produces interactions that are more realistic and diverse while identifying up to 33% more agent errors. |
Out-of-Scope Intent Detection with Self-Supervision and Discriminative Training (2021.acl-long)
Copied to clipboard
| Challenge: | Existing methods for out-of-scope intent detection rely on strong assumptions on data distribution and confidence threshold selection. |
| Approach: | They propose a method to train an out-of-scope intent classifier in a fully end-to-end manner by simulating the test scenario in training. |
| Outcome: | The proposed method improves on four benchmark dialogue datasets and improves over state-of-the-art methods. |
LayoutDIT: Layout-Aware End-to-End Document Image Translation with Multi-Step Conductive Decoder (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods struggle to capture the visual layout in complex document images. |
| Approach: | They propose to integrate layout knowledge into document image translation by using a layout-aware encoder and a multi-step conductive decoder to achieve the translation step by step. |
| Outcome: | The proposed model outperforms state-of-the-art methods with better parameter efficiency. |
Is LLM an Overconfident Judge? Unveiling the Capabilities of LLMs in Detecting Offensive Language with Annotation Disagreement (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing studies focus on evaluating large language models' ability to handle disagreement cases. |
| Approach: | They evaluate the performance of large language models in detecting offensive language at varying levels of agreement. |
| Outcome: | The proposed model improves detection accuracy and model alignment with human judgment by using disagreement samples in training. |
BUMP: A Benchmark of Unfaithful Minimal Pairs for Meta-Evaluation of Faithfulness Metrics (2023.acl-long)
Copied to clipboard
Liang Ma, Shuyang Cao, Robert L Logan IV, Di Lu, Shihao Ran, Ke Zhang, Joel Tetreault, Alejandro Jaimes
| Challenge: | Existing benchmarks measure the correlation with human judgements of faithfulness on model-generated summaries, but they are insufficient for diagnosing whether metrics are consistent, effective on human-written texts, and sensitive to different error types. |
| Approach: | They propose to use unfaithful minimal pairs to measure the consistency of automatic faithfulness metrics by comparing human-written summary pairs with a dataset of 889 human-writing, minimally different summary pairs. |
| Outcome: | The proposed benchmarks show that the most discriminative metrics tend not to be the most consistent, and that the best performing metrics are sensitive to errors. |
Multi-Level Cross-Modal Alignment for Speech Relation Extraction (2024.emnlp-main)
Copied to clipboard
Liang Zhang, Zhen Yang, Biao Fu, Ziyao Lu, Liangying Shao, Shiyu Liu, Fandong Meng, Jie Zhou, Xiaoli Wang, Jinsong Su
| Challenge: | Existing studies use synthetic speech to train and evaluate SpeechRE models, hindering their development . modality gap issue limits performance of existing models, limiting future researches . |
| Approach: | They propose to use speech data to train and evaluate SpeechRE models by using real speech . they propose to train a cross-modal alignment model to bridge the modality gap . |
| Outcome: | The proposed model can train to bridge the modality gap between speech encoder and text decoder . the proposed model is based on two real SpeechRE datasets . |
SLAM-Omni: Timbre-Controllable Voice Interaction System with Single-Stage Training (2025.findings-acl)
Copied to clipboard
Wenxi Chen, Ziyang Ma, Ruiqi Yan, Yuzhe Liang, Xiquan Li, Ruiyang Xu, Zhikang Niu, Yanqiao Zhu, Yifan Yang, Zhanxun Liu, Kai Yu, Yuxuan Hu, Jinyu Li, Yan Lu, Shujie Liu, Xie Chen
| Challenge: | a new spoken dialogue system with single-stage training is demonstrating its low latency and high quality . SLAM-Omni achieves zero-shot timbre control by modeling spoken language with semantic tokens . |
| Approach: | They propose a timbre-controllable, end-to-end voice interaction system with single-stage training. |
| Outcome: | The proposed system outperforms previous models on 4 GPUs with limited data. |
CLUE: A Chinese Language Understanding Evaluation Benchmark (2020.coling-main)
Copied to clipboard
Liang Xu, Hai Hu, Xuanwei Zhang, Lu Li, Chenjie Cao, Yudong Li, Yechen Xu, Kai Sun, Dian Yu, Cong Yu, Yin Tian, Qianqian Dong, Weitang Liu, Bo Shi, Yiming Cui, Junyi Li, Jun Zeng, Rongzhao Wang, Weijian Xie, Yanting Li, Yina Patterson, Zuoyu Tian, Yiwen Zhang, He Zhou, Shaoweihua Liu, Zhe Zhao, Qipeng Zhao, Cong Yue, Xinrui Zhang, Zhengliang Yang, Kyle Richardson, Zhenzhong Lan
| Challenge: | Existing language evaluation benchmarks for English are limited to English . lack of such benchmarks makes it difficult to replicate success in other languages . |
| Approach: | They introduce a large-scale Chinese language understanding evaluation benchmark . the benchmark uses a set of current state-of-the-art pre-trained Chinese models . |
| Outcome: | The first large-scale Chinese Language Understanding Evaluation (CLUE) benchmark is released . the benchmark evaluates models across a wide range of tasks on original Chinese text . existing language evaluation benchmarks are mostly limited to English . |
STATE ToxiCN: A Benchmark for Span-level Target-Aware Toxicity Extraction in Chinese Hate Speech Detection (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing studies on Chinese hate speech detection lack span-level fine-grained annotations. |
| Approach: | They construct a Span-level target-aware Toxicity Extraction dataset and evaluate existing models for Chinese hateful slang. |
| Outcome: | The proposed dataset is the first span-level Chinese hate speech dataset and evaluates the ability of existing models to understand hate semantics. |
Too Correct to Learn: Reinforcement Learning on Saturated Reasoning Data (2026.acl-short)
Copied to clipboard
| Challenge: | Strong base models saturate benchmarks, resulting in weaker performance, a paradox . a new approach to Reinforcement Learning (RL) is needed to improve performance . |
| Approach: | They propose a method that uses constrained uniform top-k sampling to flatten the local optimization landscape by sampling uniformly from constrained high-confidence candidates. |
| Outcome: | Experiments show that the proposed approach prevents policy degeneration and boosts out-of-domain generalization. |
VersaTune: An Efficient Data Composition Framework for Training Multi-Capability LLMs (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing work focuses on domain-specific enhancements during fine-tuning, the challenge of which lies in catastrophic forgetting of knowledge across other domains. |
| Approach: | They propose a data composition framework that allows LLMs to enhance their multi-domain capabilities during supervised fine-tuning. |
| Outcome: | The proposed framework improves multi-domain fostering performance by 29.77% compared to uniform weights. |
Logical Consistency is Vital: Neural-Symbolic Information Retrieval for Negative-Constraint Queries (2025.findings-acl)
Copied to clipboard
Ganlin Xu, Zhoujia Zhang, Wangyi Mei, Jiaqing Liang, Weijia Lu, Xiaodong Zhang, Zhifei Yang, Xiaofeng Ma, Yanghua Xiao, Deqing Yang
| Challenge: | Current dense retrieval methods compute similarities between dense vectors but overlook the real query intents. |
| Approach: | They propose a neuro-symbolic information retrieval method that leverages first-order logic to optimize the embeddings of naive natural language by considering the logical consistency between queries and documents. |
| Outcome: | The proposed method outperforms existing methods on negative-constraint queries under zero-shot and low-resource retrieval tasks. |
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
| 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. |
Towards Quantifiable Dialogue Coherence Evaluation (2021.acl-long)
Copied to clipboard
| Challenge: | Existing automatic dialogue coherence evaluation metrics are expensive and high-latency, which cannot meet the requirements of a dialogue system. |
| Approach: | They propose a framework to train a quantifiable dialogue coherence metric that can reflect actual human rating standards. |
| Outcome: | Experimental results show that the model trained by QuantiDCE presents stronger correlations with human judgements than the other state-of-the-art metrics. |
What If Consensus Lies? Selective-Complementary Reinforcement Learning at Test Time (2026.acl-long)
Copied to clipboard
| Challenge: | Existing TTRL methods rely on positive pseudo-labeling strategies to enhance reasoning capabilities. |
| Approach: | They propose a test-time reinforcement learning framework that mitigates label noise amplification by deriving pseudo-rewards from majority voting consensus. |
| Outcome: | The proposed framework mitigates label noise amplification by implementing selective positive pseudo-labeling and entropy-gated negative p-labeled pruning. |
Inter-GPS: Interpretable Geometry Problem Solving with Formal Language and Symbolic Reasoning (2021.acl-long)
Copied to clipboard
| Challenge: | Existing methods for solving geometric problems are either small in scale or not publicly available. |
| Approach: | They propose a large-scale benchmark for geometric problem solving using formal language and symbolic reasoning. |
| Outcome: | The proposed approach parses geometry problems into formal language and performs symbolic reasoning step by step. |
OCNLI: Original Chinese Natural Language Inference (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Recent efforts to extend natural language understanding to other languages have focused on (automatically) translating existing English datasets. |
| Approach: | They propose to use a Chinese dataset to generate annotated sentences from native speakers specializing in linguistics to elicit annotations. |
| Outcome: | The proposed dataset does not rely on automatic translation or non-expert annotation. instead, it elicits annotations from native speakers specializing in linguistics. |
UniGeo: Unifying Geometry Logical Reasoning via Reformulating Mathematical Expression (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing work on geometry problem solving treats calculation and proving as two specific tasks hindering a deep model to unify reasoning ability on multiple math tasks. |
| Approach: | They propose a large-scale Unified Geometry problem benchmark to unify geometry on multiple math tasks. |
| Outcome: | The proposed framework outperforms the existing model with 5.6% and 3.2% accuracies on calculation and proving problems. |
LearnAct: Few-Shot Mobile GUI Agent with a Unified Demonstration Benchmark (2026.findings-acl)
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
| 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. |
Fine-tuning Pre-trained Language Models for Few-shot Intent Detection: Supervised Pre-training and Isotropization (2022.naacl-main)
Copied to clipboard
| Challenge: | Recent studies show that fine-tuning pre-trained language models with a small set of labeled utterances in a supervised manner is helpful, but it yields an anisotropic feature space, which may suppress the expressive power of the semantic representations. |
| Approach: | They propose to regularize supervised pre-training towards isotropy by contrastive learning and correlation matrix regularizers. |
| Outcome: | The proposed methods improve supervised pre-training by regularizing the feature space towards isotropy. |
SlotRefine: A Fast Non-Autoregressive Model for Joint Intent Detection and Slot Filling (2020.emnlp-main)
Copied to clipboard
| Challenge: | Slot filling and intent detection are two main tasks in spoken language understanding systems. |
| Approach: | They propose a non-autoregressive slot filling model with two-pass iteration mechanism to handle uncoordinated slots problem. |
| Outcome: | The proposed model significantly outperforms previous models in slot filling task while speeding up decoding. |
WebAgent-R1: Training Web Agents via End-to-End Multi-Turn Reinforcement Learning (2025.emnlp-main)
Copied to clipboard
Zhepei Wei, Wenlin Yao, Yao Liu, Weizhi Zhang, Qin Lu, Liang Qiu, Changlong Yu, Puyang Xu, Chao Zhang, Bing Yin, Hyokun Yun, Lihong Li
| Challenge: | Existing work on reinforcement learning has focused on single-turn tasks such as solving math problems. |
| Approach: | They propose a framework that learns directly from online interactions by asynchronously generating diverse trajectories, guided by binary rewards depending on task success. |
| Outcome: | Experiments on the WebArena-Lite benchmark show that the framework outperforms state-of-the-art methods and strong proprietary models. |
Plot2Code: A Comprehensive Benchmark for Evaluating Multi-modal Large Language Models in Code Generation from Scientific Plots (2025.findings-naacl)
Copied to clipboard
| Challenge: | Multi-modal Large Language Models have shown remarkable progress in visual contexts, yet their ability to convert visual figures into executable code remains underexplored. |
| Approach: | They propose to use a set of visual coding metrics to assess MLLMs' visual . pass rate, text-match ratio, and GPT-4V rating judgement to assess the quality of generated code and rendered images. |
| Outcome: | The proposed benchmark includes 132 high-quality matplotlib plots across six plot types, as well as 150 and 86 plots from Python’s and R’s plotly libraries respectively, totaling 368 plots. |
Toward Human-Like Evaluation for Natural Language Generation with Error Analysis (2023.acl-long)
Copied to clipboard
| Challenge: | Pre-trained language models (PLMs) have been used to evaluate language generation tasks . pretrained error analysis can be used to refine the generated sentence toward higher confidence . |
| Approach: | They propose to combine pretrained language model based metrics with human-like error analysis to improve sentence confidence. |
| Outcome: | The proposed method outperforms top-scoring metrics in 19/25 settings. |
A Query-Response Framework for Whole-Page Complex-Layout Document Image Translation with Relevant Regional Concentration (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for document image translation rely on the vanilla encoder-decoder paradigm . a novel dynamic aggregation mechanism is designed to enhance the text semantics in query features toward translation. |
| Approach: | They propose a Query-Response DIT framework that reformulates the DIT task into a parallel response/translation process of multiple queries. |
| Outcome: | The proposed framework improves translation quality on four translation directions on three benchmarks. |
Save the Good Prefix: Precise Error Penalization via Process-Supervised RL to Enhance LLM Reasoning (2026.findings-acl)
Copied to clipboard
| 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 . |
Advancing SMoE for Continuous Domain Adaptation of MLLMs: Adaptive Router and Domain-Specific Loss (2025.acl-long)
Copied to clipboard
| Challenge: | Recent studies have explored Continual Instruction Tuning (CIT) in Multimodal Large Language Models (MLLMs), with a primary focus on Task-incremental CIT, where MLLM are required to continuously acquire new tasks. |
| Approach: | They propose a Sparse Mixture of Expert (SMoE) based method for domain-incremental CIT in Multimodal Large Language Models (MLLMs) . they equip the SMoA module with a domain-specific autoregressive loss (DSAL) they establish a new benchmark to evaluate the efficacy of their method . |
| Outcome: | The proposed method outperforms all baselines and is based on a Sparse Mixture of Experts (SMoE) module . |
Take Care of Your Prompt Bias! Investigating and Mitigating Prompt Bias in Factual Knowledge Extraction (2024.lrec-main)
Copied to clipboard
| Challenge: | Recent research shows that pre-trained language models suffer from “prompt bias” in factual knowledge extraction. |
| Approach: | They propose a representation-based approach to mitigate prompt bias during inference time by querying the model and removing it from its internal representations to generate debiased representations. |
| Outcome: | The proposed approach corrects the overfitted performance caused by prompt bias and significantly improves prompt retrieval capability. |
Metaphor Reasoning is Meta-reasoning (2026.acl-long)
Copied to clipboard
Qianyu He, Junting Lu, Yikai Zhang, Siyu Yuan, Xiaojun Meng, Jiansheng Wei, Jiaqing Liang, Yanghua Xiao
| Challenge: | Existing work on metaphor reasoning's impact on reasoning abilities is limited. |
| Approach: | They propose a system for synthesizing metaphorical riddles that satisfy five quality dimensions: diverse, balanced, reasoning-oriented, challenging, and verifiable. |
| Outcome: | The proposed system improves reasoning abilities across six domains using only thousands of metaphorical riddles. |
Single-to-mix Modality Alignment with Multimodal Large Language Model for Document Image Machine Translation (2025.acl-long)
Copied to clipboard
| Challenge: | Document Image Machine Translation (DIMT) faces generalization challenges due to limited training data and the complex interplay between visual and textual information. |
| Approach: | They propose a single-to-mix Modality alignment framework leveraging Multimodal Large Language Models (MLLMs) this framework aligns an imageonly encoder with multimodal representations of an MLLM pre-trained on large-scale document image datasets. |
| Outcome: | The proposed framework improves translation quality in cross-domain generalization and challenging document image scenarios. |
Beyond Cross-Modal Alignment: Measuring and Leveraging Modality Gap in Vision-Language Models (2026.findings-acl)
Copied to clipboard
| 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. |
MQM-APE: Toward High-Quality Error Annotation Predictors with Automatic Post-Editing in LLM Translation Evaluators (2025.coling-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have shown significant potential as judges for Machine Translation (MT) quality assessment. |
| Approach: | They propose a framework that automatically post-edits the original translation based on each error, thereby filtering out non-impactful errors. |
| Outcome: | The proposed framework improves reliability and quality of error spans against GEMBA-MQM, across eight LLMs in both high- and low-resource languages. |
Document Image Machine Translation with Dynamic Multi-pre-trained Models Assembling (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing TIMT tasks focus on text-line-level images. |
| Approach: | They propose to extend the existing TIMT task and introduce a new framework to translate a source document image to markdown-formatted target translation. |
| Outcome: | The proposed task aims to translate a source document image with long context and complex layout structure to markdown-formatted target translation. |
MAS-Bench: A Unified Benchmark for Shortcut-Augmented Hybrid Mobile GUI Agents (2026.acl-long)
Copied to clipboard
Pengxiang Zhao, Guangyi Liu, Yaozhen Liang, Weiqing He, Zhengxi Lu, WenHao Wang, Yuehao Huang, Yuxiang Chai, Zhaolu Kang, Yaxuan Guo, Hao Wang, Kexin Zhang, Liang Liu, Yong Liu
| Challenge: | Shortcuts such as APIs and deep-links have emerged as efficient complements to flexible GUI operations, but systematic evaluation of GUI–shortcut hybrid agents remains underexplored. |
| Approach: | They propose a benchmark that evaluates GUI-shortcut hybrid agents with a specific focus on the mobile domain. |
| Outcome: | MAS-Bench evaluates agent's ability to generate shortcuts by discovering and creating reusable, low-cost workflows. |
SELF-GUARD: Empower the LLM to Safeguard Itself (2024.naacl-long)
Copied to clipboard
| Challenge: | Recent studies have investigated methods to improve the safety of large language models (LLMs) safety training involves fine-tuning the LLM with adversarial samples, which activate the LRM’s capabilities against jailbreak. |
| Approach: | They propose a safety training approach that integrates safety training and safeguards to train the LLM to perform harmfulness detection on its own outputs. |
| Outcome: | The proposed method reduces harmful output and adds a [harmful] or [harmless] tag to the end of the LLM's response. |
A Survey of Deep Learning for Mathematical Reasoning (2023.acl-long)
Copied to clipboard
| Challenge: | a survey of deep learning for mathematical reasoning examines the field . a comprehensive reading list is provided to assist readers interested in the field. |
| Approach: | They present a survey of deep learning for mathematical reasoning over the past decade . they outline directions for future research and highlight potential for further exploration . |
| Outcome: | The proposed framework is based on the results of a decade-long survey of deep learning for mathematical reasoning. |
VoxMind: An End-to-End Agentic Spoken Dialogue System (2026.acl-long)
Copied to clipboard
Tianle Liang, Yifu Chen, Shengpeng Ji, Yijun Chen, Zhiyang Jia, Jingyu Lu, Fan Zhuo, Xueyi Pu, Yangzhuo Li, Zhou Zhao
| Challenge: | Existing research on end-to-end spoken dialogue models has focused on core perception and generation, with limited exploration of tool-augmented extensions. |
| Approach: | They propose a framework to equip end-to-end spoken dialogue models with comprehensive agentic abilities by leveraging a 470-hour AgentChat dataset. |
| Outcome: | The proposed framework outperforms Gemini-2.5-Pro on spoken agent tasks while maintaining general conversational quality. |
Expanding the Boundaries of Vision Prior Knowledge in Multi-modal Large Language Models (2026.eacl-long)
Copied to clipboard
Qiao Liang, Yanjiang Liu, Weixiang Zhou, Ben He, Yaojie Lu, Hongyu Lin, Jia Zheng, Xianpei Han, Le Sun, Yingfei Sun
| Challenge: | Existing research treats MLLMs as unified systems optimized through end-to-end training, but the impact of vision encoder’s prior knowledge is seldom investigated. |
| Approach: | They propose a metric to quantify the effect of prior knowledge on MLLM performance by integrating prior knowledge at the vision encoder level into a training framework. |
| Outcome: | The proposed training framework incorporates prior knowledge at the vision encoder level, and significantly boosts visual understanding capabilities of MLLMs. |
Grammar-Based Code Representation: Is It a Worthy Pursuit for LLMs? (2025.findings-acl)
Copied to clipboard
Qingyuan Liang, Zhao Zhang, Zeyu Sun, Zheng Lin, Qi Luo, Xiao Yueyi, Yizhou Chen, Yuqun Zhang, Haotian Zhang, Lu Zhang, Chenbin Chenbin, Yingfei Xiong
| Challenge: | Existing research demonstrates the effectiveness of grammar-based code representations in small-scale models, showing their ability to reduce syntax errors and enhance performance. |
| Approach: | They develop a series of billion-scale grammar-based code representations that incorporate grammar rules into the code generation process. |
| Outcome: | Experiments on HumanEval and MBPP show that grammar-based representations reduce syntax errors and improve performance even in billion-scale models. |
H-MAS: Hierarchical Multi-Agent Scheduling for Multi-Tenant LLM Serving (2026.findings-acl)
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 . |
Graph-based Relation Mining for Context-free Out-of-vocabulary Word Embedding Learning (2023.acl-long)
Copied to clipboard
| Challenge: | Existing word embedding methods fail to model complex word formation well. |
| Approach: | They propose a graph-based relation mining method for OOV word embedding learning that can infer high-quality embeddables for OV words through passing and aggregating semantic attributes and relational information in the WRG. |
| Outcome: | The proposed method outperforms state-of-the-art models on both intrinsic and downstream tasks when faced with OOV words. |
Gold-Medal-Level Olympiad Geometry Solving with Efficient Heuristic Auxiliary Constructions (2026.acl-long)
Copied to clipboard
Boyan Duan, Xiao Liang, Shuai Lu, Yaoxiang Wang, Yelong Shen, Kai-Wei Chang, Ying Nian Wu, Mao Yang, Weizhu Chen, Yeyun Gong
| Challenge: | Existing methods for geometry theorem proving in Euclidean geometry are challenging and require a neural network to perform. |
| Approach: | They propose a method for adding auxiliary points in geometry that runs on CPUs without relying on neural network-based inference. |
| Outcome: | The proposed method achieves silver-medal-level human performance on IMO-30 benchmark. |
Commonality and Individuality! Integrating Humor Commonality with Speaker Individuality for Humor Recognition (2025.naacl-long)
Copied to clipboard
| Challenge: | Current methods for humor recognition focus on one aspect of humor commonalities, ignoring the multifaceted nature of humor. |
| Approach: | They propose a commonality and individuality incorporated network for humor recognition that integrates multifaceted humor commonalities with speaker individuality. |
| Outcome: | The proposed model integrates multifaceted humor commonalities with speaker individuality to deepen the understanding of humor expressions. |
Improving MLLM’s Document Image Machine Translation via Synchronously Self-reviewing Its OCR Proficiency (2025.findings-acl)
Copied to clipboard
| Challenge: | Multimodal Large Language Models (MLLMs) have shown strong performance in document image tasks, especially Optical Character Recognition (OCR). However, they struggle with Document Image Machine Translation (DIMT), which requires handling both cross-modal and cross-lingual challenges. |
| Approach: | They propose a novel fine-tuning paradigm that allows the model to generate OCR text before producing translation text, which allows it to leverage its strong monolingual OCR ability while learning to translate text across languages. |
| Outcome: | The proposed model can leverage its strong monolingual OCR ability while learning to translate text across languages. |
Multi-View Source Ablation for Faithful Summarization (2023.findings-eacl)
Copied to clipboard
| Challenge: | MuFaSSa is a metric for evaluating faithfulness of abstractive summaries . it uses different strategies to remove information from source document to form multiple ablated views . |
| Approach: | They propose a metric for evaluating faithfulness of abstractive summaries using multiple ablated views. |
| Outcome: | The proposed metric outperforms existing models on summarization tasks and human-annotated faithfulness labels. |
AGIEval: A Human-Centric Benchmark for Evaluating Foundation Models (2024.findings-naacl)
Copied to clipboard
Wanjun Zhong, Ruixiang Cui, Yiduo Guo, Yaobo Liang, Shuai Lu, Yanlin Wang, Amin Saied, Weizhu Chen, Nan Duan
| Challenge: | Traditional benchmarks for evaluating foundation models often fail to accurately represent their general abilities for human-centric tasks. |
| Approach: | They propose a bilingual benchmark to assess foundation models in the context of human-centric standardized exams such as college entrance exams, law school admission tests, and math competitions. |
| Outcome: | The proposed benchmark exceeds the average human performance on SAT, LSAT, and math competitions with 95% accuracy and 92.5% on the Chinese college entrance English exam. |
Take Its Essence, Discard Its Dross! Debiasing for Toxic Language Detection via Counterfactual Causal Effect (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing methods to mitigate lexical bias in toxic language detection (TLD) do not exploit the “useful” and “misleading” impact of the bias. |
| Approach: | They propose a counterfactual Causal Debiasing Framework to mitigate lexical bias in toxic language detection (TLD) it preserves the “useful impact” of lexical bias and eliminates the "misleading impact" they propose to use the same framework to analyze the causal effect of a sentence and bias tokens . |
| Outcome: | The proposed framework preserves the “useful impact” of lexical bias and eliminates the ‘misleading impact’ Empirical evaluations show that the proposed model outperforms current debiased models for out-of-distribution data. |
Runaway is Ashamed, But Helpful: On the Early-Exit Behavior of Large Language Model-based Agents in Embodied Environments (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Experiments with 4 different LLMs across 5 embodied environments show significant efficiency improvements, with only minor drops in agent performance. |
| Approach: | They propose an intrinsic method that injects exit instructions during generation and an extransic system that verifies task completion to determine when to halt an agent’s trial. |
| Outcome: | The proposed method injects exit instructions during generation and an exit method verifies task completion to determine when to halt an agent’s trial. |
MLAlgo-Bench: Can Machines Implement Machine Learning Algorithms? (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Currently, the top-performing models achieve a 48.8% task completion rate on realizing machine learning algorithms . |
| Approach: | They propose a benchmark to test machine learning's ability to generate ML code for humans . they propose an automatic evaluation framework with metrics such as task pass rate and time overhead . |
| Outcome: | The proposed benchmark is unique in its focus on interpreting complex human instructions and producing multi-step, high-complexity code. |
From Chaotic OCR Words to Coherent Document: A Fine-to-Coarse Zoom-Out Network for Complex-Layout Document Image Translation (2025.coling-main)
Copied to clipboard
| Challenge: | Document Image Translation (DIT) aims to translate documents in images from one language to another. |
| Approach: | They propose a novel end-to-end network called Zoom-out DIT to improve document translation by combining word positioning, sentence recognition and document organization. |
| Outcome: | The proposed network improves word positioning, sentence recognition and document organization, and improves translation quality. |
A Self-Denoising Model for Robust Few-Shot Relation Extraction (2025.acl-long)
Copied to clipboard
| Challenge: | Existing studies assume that the support set contains only accurately labeled instances, but this assumption is often unrealistic. |
| Approach: | They propose a self-denoising model for FSRE which can automatically correct noisy labels of support instances. |
| Outcome: | The proposed model outperforms all baselines on two public datasets showing that it can correct mislabeled support instances. |
WavAlign: Enhancing Intelligence and Expressiveness in Spoken Dialogue Models via Adaptive Hybrid Post-Training (2026.findings-acl)
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
| 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. |
Error Analysis Prompting Enables Human-Like Translation Evaluation in Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Recent research shows that large language models (LLMs) perform poorly at segment level. |
| Approach: | They propose a new prompting method that emulates the commonly accepted human evaluation framework . they will release their code and scripts to facilitate the community . |
| Outcome: | The proposed method is based on the human evaluation framework MQM and produces explainable and reliable MT evaluations at both the system and segment level. |
CodeV: Issue Resolving with Visual Data (2025.findings-acl)
Copied to clipboard
Linhao Zhang, Daoguang Zan, Quanshun Yang, Zhirong Huang, Dong Chen, Bo Shen, Tianyu Liu, Yongshun Gong, Huang Pengjie, Xudong Lu, Guangtai Liang, Lizhen Cui, Qianxiang Wang
| Challenge: | Large Language Models (LLMs) have expanded to more complex repository-level tasks. |
| Approach: | They propose a first approach to leveraging visual data to enhance the issue-resolving capabilities of Large Language Models (LLMs) they demonstrate the effectiveness of CodeV and provide valuable insights into leveraging visualization to resolve GitHub issues. |
| Outcome: | The proposed approach improves the issue-resolving capabilities of Large Language Models (LLMs) by using visual data. |
Competition-Level Problems are Effective LLM Evaluators (2024.findings-acl)
Copied to clipboard
Yiming Huang, Zhenghao Lin, Xiao Liu, Yeyun Gong, Shuai Lu, Fangyu Lei, Yaobo Liang, Yelong Shen, Chen Lin, Nan Duan, Weizhu Chen
| Challenge: | Large language models (LLMs) have demonstrated impressive reasoning capabilities, yet there is ongoing debate about their capabilities and the potential data contamination problem. |
| Approach: | They propose to evaluate the reasoning capabilities of large language models in solving recent competition-level programming problems in Codeforces. |
| Outcome: | The proposed model has experienced a cliff-like decline in problems after September 2021, which shows the potential data contamination and the challenges for any existing LLM to solve unseen complex reasoning problems. |
TED-TTS: Training-Free Intra-Utterance Emotion and Duration Control for Text-to-Speech Synthesis (2026.acl-long)
Copied to clipboard
| Challenge: | Existing controllable Text-to-Speech methods limited to inter-utterance-level control . utterance expressiveness remains a challenge in building human-like TTS synthesis systems . |
| Approach: | They propose a training-free controllable framework for pretrained zero-shot TTS to enable intra-utterance emotion and duration expression. |
| Outcome: | The proposed framework achieves state-of-the-art intra-utterance consistency while maintaining baseline-level speech quality. |
The Bitter Lesson of Diffusion Language Models for Agentic Workflows: A Comprehensive Reality Check (2026.acl-long)
Copied to clipboard
| Challenge: | Embodied and Tool-Calling agents are effective in planning and complex reasoning, but require causal, precise, and logically grounded reasoning mechanisms to be viable for agentic tasks. |
| Approach: | They propose a framework that integrates dLLMs as plug-and-play cognitive cores. |
| Outcome: | The proposed model breaks the sequential latency bottleneck in agentic interactions. |
Facilitating Fine-grained Detection of Chinese Toxic Language: Hierarchical Taxonomy, Resources, and Benchmarks (2023.acl-long)
Copied to clipboard
| Challenge: | Existing datasets suffer from a lack of fine-grained annotations, such as the toxic type and expressions with indirect toxicity. |
| Approach: | They propose a benchmark model to detect toxic language by incorporating lexical features into a Chinese dataset to facilitate fine-grained annotations. |
| Outcome: | The proposed model is based on insulting vocabulary containing implicit profanity and is able to detect toxic language with lexical features. |