Tool Preferences in Agentic LLMs are Unreliable
机构 * University of Maryland（马里兰大学）
Comments Conference on Empirical Methods in Natural Language Processing (EMNLP) 2025, main
期刊&会议
EMNLP
Conference on Empirical Methods in Natural Language Processing · 会议 · Natural Language Processing
Runaway is Ashamed, But Helpful: On the Early-Exit Behavior of Large Language Model-based Agents in Embodied Environments
机构 * Southeast University（东南大学） ； The University of Sydney（悉尼大学） ； Southeast University Shenzhen Research Institute（东南大学深圳研究院） ； Institute of Computing and Intelligence, Harbin Institute of Technology, Shenzhen, China（哈尔滨工业大学计算机与智能信息学院，深圳，中国） ； College of Computing and Data Science at Nanyang Technological University, Singapore 639798（南洋理工大学计算机与数据科学学院，新加坡）
Comments EMNLP 2025 - Findings
The Pursuit of Empathy: Evaluating Small Language Models for PTSD Dialogue Support
Comments 23 pages, 3 figures. Accepted for Oral presentation at EMNLP 2025
Scaling Low-Resource MT via Synthetic Data Generation with LLMs
机构 * University of Helsinki（赫尔辛基大学） ； University of Cambridge（剑桥大学）
Comments Accepted at EMNLP 2025 Main Conference
QA-prompting: Improving Summarization with Large Language Models using Question-Answering
机构 * Georgia Institute of Technology（佐治亚理工学院）
Comments Accepted at The Fifth Workshop on New Frontiers in Summarization (NewSumm) in The 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP 2025)
LCES: Zero-shot Automated Essay Scoring via Pairwise Comparisons Using Large Language Models
机构 * Deloitte Analytics R&D（德勤分析研究与开发）
Comments Accepted to EMNLP 2025 (Main Conference)
KoACD: The First Korean Adolescent Dataset for Cognitive Distortion Analysis via Role-Switching Multi-LLM Negotiation
Comments Accepted to Findings of EMNLP 2025
How Real Are Synthetic Therapy Conversations? Evaluating Fidelity in Prolonged Exposure Dialogues
机构 * College of Information Sciences and Technology, Penn State University（宾夕法尼亚州立大学信息科学与技术学院） ； Department of Psychiatry and Behavioral Sciences, Emory University（埃默里大学精神病学与行为科学系） ； School of Interactive Computing, Georgia Tech（佐治亚理工学院交互计算学院） ； School of Psychology, Georgia Tech（佐治亚理工学院心理学学院）
Comments 10 pages, 5 tables. Accepted for Poster presentation at EMNLP 2025
LEO-MINI: An Efficient Multimodal Large Language Model using Conditional Token Reduction and Mixture of Multi-Modal Experts
机构 * University of Waterloo（滑铁卢大学）
Comments To appear at EMNLP 2025
Unmasking Deceptive Visuals: Benchmarking Multimodal Large Language Models on Misleading Chart Question Answering
机构 * The Hong Kong University of Science and Technology（香港科学与技术大学）
Comments 34 pages in total, EMNLP 2025
FastCuRL: Curriculum Reinforcement Learning with Stage-wise Context Scaling for Efficient Training R1-like Reasoning Models
机构 * Tencent Hunyuan（腾讯文英）
Comments Accepted by EMNLP 2025
Can Language Models Follow Multiple Turns of Entangled Instructions?
机构 * University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校） ； University of California San Diego（加州大学圣地亚哥分校）
Comments The 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP 2025) Findings
Context-aware Biases for Length Extrapolation
Comments Accepted at EMNLP 2025 Main Conference
From Language to Cognition: How LLMs Outgrow the Human Language Network
机构 * EPFL（苏黎世联邦理工学院） ； MIT（麻省理工学院） ； Georgia Institute of Technology（佐治亚理工学院）
Comments EMNLP 2025. Project Page at https://language-to-cognition.epfl.ch
FC-Attack: Jailbreaking Multimodal Large Language Models via Auto-Generated Flowcharts
机构 * The Hong Kong University of Science and Technology (Guangzhou)（香港科技大学（广州））
Comments Accepted to Findings of EMNLP 2025
Improving the quality of Web-mined Parallel Corpora of Low-Resource Languages using Debiasing Heuristics
机构 * Dept. of Computer Science and Engineering, University of Moratuwa（moratuwa大学计算机科学与工程系） ； School of Mathematical and Computational Sciences, Massey University（massey大学数学与计算科学学院）
Comments EMNLP 2025 Camera-ready version
All-in-one: Understanding and Generation in Multimodal Reasoning with the MAIA Benchmark
机构 * Università di Roma La Sapienza（罗马La Sapienza大学） ； Fondazione Bruno Kessler (FBK)（布鲁诺·凯斯勒基金会） ； Free University of Bozen-Bolzano（博兹纳-博尔扎诺自由大学） ； Dept. of Computer Science, University of Pisa（比萨大学计算机科学系） ； CoLing Lab, Dept. of Philology, Literature and Linguistics, University of Pisa（比萨大学语言学、文学与语言学系CoLing实验室） ； Istituto di Linguistica Computazionale "A. Zampolli" (CNR-ILC), ItaliaNLP Lab, Pisa（A. Zampolli计算语言学研究所（CNR-ILC），意大利NLP实验室，比萨）
Comments Accepted at Findings of EMNLP 2025
Does Reasoning Introduce Bias? A Study of Social Bias Evaluation and Mitigation in LLM Reasoning
机构 * Santa Clara University（圣克拉拉大学） ； Rochester Institute of Technology（罗切斯特理工学院） ； Docomo Innovations（Docomo创新）
Comments EMNLP Findings
Reasoning-to-Defend: Safety-Aware Reasoning Can Defend Large Language Models from Jailbreaking
机构 * Beihang University（北京航空航天大学） ； Baidu Inc.（百度公司） ； Zhongguancun Laboratory（中关村实验室）
Comments EMNLP 2025
Large Language Models Badly Generalize across Option Length, Problem Types, and Irrelevant Noun Replacements
机构 * Qiyuan Tech（奇源科技） ； Zhinao（360智脑） ； Peking University（北京大学）
Comments EMNLP 2025 Main Conference
Evaluating Step-by-step Reasoning Traces: A Survey
机构 * University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）
Comments EMNLP 2025 Findings
Beyond Pairwise: Global Zero-shot Temporal Graph Generation
机构 * Computer Science Department, Bar-Ilan University（巴伊兰大学计算机科学系） ； Efi Arazi School of Computer Science, Reichman University（里奇曼大学埃菲·阿拉兹计算机科学学院）
Comments Accepted to the main track of EMNLP 2025
No Need for Explanations: LLMs can implicitly learn from mistakes in-context
机构 * Imperial College London（伦敦帝国学院） ； Cohere
Comments EMNLP 2025
Efficient Beam Search for Large Language Models Using Trie-Based Decoding
机构 * Department of Computer Science, National Chengchi University（国立成功大学计算机科学系） ； Department of Computer Science, Cornell University（康奈尔大学计算机科学系） ； Institute of Information Science, Academia Sinica（中华科学研究院资讯研究所）
Comments 13 pages, accepted as a main conference paper at EMNLP 2025
NILE: Internal Consistency Alignment in Large Language Models
机构 * The Chinese University of Hong Kong（香港中文大学） ； City University of Hong Kong（香港城市大学） ； Huawei Noah’s Ark Lab（华为诺亚实验室） ； University of Edinburgh（爱丁堡大学）
Comments This work has been accepted by EMNLP 2025
Both Text and Images Leaked! A Systematic Analysis of Data Contamination in Multimodal LLM
机构 * Lehigh University（莱维大学） ； The Chinese University of Hong Kong, Shenzhen（香港中文大学（深圳））
Comments Accepted to EMNLP 2025 Findings
Group-SAE: Efficient Training of Sparse Autoencoders for Large Language Models via Layer Groups
机构 * University of Milan-Bicocca（米兰-比科卡大学） ； London School of Economics（伦敦经济学院）
Comments Accepted version at EMNLP'25
Rethinking Backdoor Detection Evaluation for Language Models
机构 * University of Southern California（南加州大学） ； University of California, Davis（加州大学戴维斯分校）
Comments Accepted to EMNLP 2025
Evaluating Fairness in Large Vision-Language Models Across Diverse Demographic Attributes and Prompts
机构 * Santa Clara University（圣克拉拉大学） ； DOCOMO Innovations, Inc.（DOCOMO创新公司） ； Rochester Institute of Technology（罗切斯特理工学院）
Comments EMNLP Findings
Temporal Scaling Law for Large Language Models
机构 * School of Software, Tsinghua University（清华大学软件学院） ； BNRist, Tsinghua University（清华大学BNRist） ； Kuaishou Technology（快手科技） ； Beihang University（北航） ； University of Chinese Academy of Sciences（中国科学院大学） ； School of Computer Science, BUPT（北京邮电大学计算机科学学院） ； Department of Automation, Tsinghua University（清华大学自动化系）
Comments Accepted by EMNLP'25 Main Conference (Oral presentation), Camera-ready version
