EMNLP 2025 Papers — Page 9
Conference on Empirical Methods in Natural Language Processing · 1809 papers
iVISPAR — An Interactive Visual-Spatial Reasoning Benchmark for VLMs
Julius Mayer (Osnabrück University), Elia Bruni (Osnabrück University)
Prompt EngineeringVision Language ModelImageTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Proposed the iVISPAR benchmark, utilizing interactive sliding geometric puzzles (SGP) to evaluate the spatial reasoning and planning capabilities of vision-language models (VLMs) under three input modalities: 3D vision, 2D vision, and text.
Jailbreak LLMs through Internal Stance Manipulation
Shuangjie Fu (Chinese Academy of Sciences), Xueqi Cheng (Chinese Academy of Sciences)
Adversarial AttackLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposes a Stance Manipulation (SM) method that achieves automated jailbreaking by suppressing the internal rejection posture of LLMs, further elucidating the internal formation process of LLM rejection mechanisms;
Jailbreak-Tuning: Models Efficiently Learn Jailbreak Susceptibility
Brendan Murphy (FAR.AI), Kellin Pelrine (FAR.AI)
Adversarial AttackSupervised Fine-TuningText
🎯 What it does: Investigated and demonstrated that attacking closed-source language models through 'jailbreak-tuning' can completely undermine their safety defenses and generate high-quality harmful responses.
JI^2S: Joint Influence‐Aware Instruction Data Selection for Efficient Fine‐Tuning
Jingyu Wei (National University of Defense Technology), Mengmeng Guo (National University of Defense Technology)
Computational EfficiencyData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose a joint influence-aware instruction data selection framework called JI²S, which can filter out high-quality subsets from large-scale instruction datasets for efficient fine-tuning.
Jigsaw-Puzzles: From Seeing to Understanding to Reasoning in Vision-Language Models
Zesen Lyu (Institute for Advanced Study, UCAS), Yao Yang (Zhejiang Lab)
Prompt EngineeringVision Language ModelImageTextBenchmark
🎯 What it does: Proposed the Jigsaw-Puzzles benchmark, utilizing 1,100 real-world high spatial complexity images, designing 5 tasks ranging from perception to reasoning, to systematically evaluate the spatial reasoning capabilities of vision-language models (VLM).
Job Unfair: An Investigation of Gender and Occupational Bias in Free-Form Text Completions by LLMs
Camilla Casula (Fondazione Bruno Kessler), Sara Tonelli (Fondazione Bruno Kessler)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Examined gender and occupational bias in 9 English-Italian bilingual large language models (LLMs) during free-text completion tasks, combining manual annotation, vector embedding clustering, and lexical relevance analysis to systematically evaluate gender distribution, thematic direction, subject misunderstanding, and agent/affinity tendencies in model outputs.
Joint Modeling of Entities and Discourse Relations for Coherence Assessment
Wei Liu (Heidelberg Institute for Theoretical Studies), Michael Strube (Heidelberg Institute for Theoretical Studies)
ClassificationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposed two joint modeling methods for assessing textual coherence by integrating entities and discourse relations;
JOLT-SQL: Joint Loss Tuning of Text-to-SQL with Confusion-aware Noisy Schema Sampling
Jinwang Song (Zhengzhou University), Min Peng (Wuhan University)
OptimizationComputational EfficiencyAI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Propose a single-stage joint loss optimization framework called JOLT-SQL to simultaneously optimize schema linking and SQL generation, thereby improving Text-to-SQL performance;
Journalism-Guided Agentic In-context Learning for News Stance Detection
Dahyun Lee (Soongsil University), Kunwoo Park (Soongsil University)
ClassificationLarge Language ModelSupervised Fine-TuningAgentic AIPrompt EngineeringTextChain-of-Thought
🎯 What it does: This paper first constructs the first Korean full-text news stance detection dataset K-NEWS-STANCE and proposes the JOAICL framework based on proxy-based contextual learning, using paragraph-level stance prediction to assist LLMs in completing full-text stance judgment.
Judge and Improve: Towards a Better Reasoning of Knowledge Graphs with Large Language Models
Mo Zhiqiang (Ant Group), Jianmin Huang (Ant Group)
Explainability and InterpretabilityRepresentation LearningGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningGraph
🎯 What it does: Propose the ExGLM framework, seamlessly integrating graph neural networks (GNN) with large language models (LLM) through a graph-language co-alignment module and a Judge-and-Improve iterative mechanism, for tasks such as node classification and link prediction.
JUDGEBERT: Assessing Legal Meaning Preservation Between Sentences
David Beauchemin (Université Laval), Pierre-Luc Déziel (Université Laval)
TransformerLarge Language ModelTextBenchmarkFinance Related
🎯 What it does: Proposed a new method and dataset for evaluating meaning preservation in legal text simplification
Judging Quality Across Languages: A Multilingual Approach to Pretraining Data Filtering with Language Models
Mehdi Ali (Fraunhofer IAIS), Kristian Kersting (DFKI)
Knowledge DistillationRepresentation LearningData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: This paper proposes a multilingual pre-training data filtering method named JQL, which uses large language models (LLMs) as judges to assess the educational value of documents and distills their capabilities into a lightweight regression model, achieving efficient screening of large-scale high-quality data.
JUREX-4E: Juridical Expert-Annotated Four-Element Knowledge Base for Legal Reasoning
Huanghai Liu (Tsinghua University), Yansong Feng (Peking University)
ClassificationRetrievalTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Constructed the JUREX-4E expert-annotated four-element knowledge base, covering 155 common criminal charges in Chinese criminal cases. A hierarchical legal interpretation framework was used to provide precise and complete annotations for each element, and its value was validated in tasks such as distinguishing similar charges and case retrieval.
KCS: Diversify Multi-hop Question Generation with Knowledge Composition Sampling
Yangfan Wang (Harbin Institute of Technology), Jingchi Jiang (MemTensor Technology Co., Ltd.)
GenerationData SynthesisTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Propose the Knowledge Composition Sampling (KCS) framework, which, given an answer and a long-text context, first selects knowledge compositions using a sentence-level sequence prediction model, then samples diverse knowledge compositions through random decoding, and finally generates multi-hop questions using a pre-trained multi-hop question generation model.
Keep Security! Benchmarking Security Policy Preservation in Large Language Model Contexts Against Indirect Attacks in Question Answering
Hwan Chang (Chung-Ang University), Hwanhee Lee (Chung-Ang University)
Safty and PrivacyTransformerPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes a new large-scale benchmark dataset called CoPriva to evaluate whether large language models can adhere to safety policies defined in the context when facing direct and indirect attacks, primarily tested in question-answering tasks.
KG-CQR: Leveraging Structured Relation Representations in Knowledge Graphs for Contextual Query Retrieval
Chi Minh Bui (Viettel AI, Viettel Group), Khac-Hoai Nam Bui (Viettel AI, Viettel Group)
RetrievalRepresentation LearningLarge Language ModelPrompt EngineeringTextGraphBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the KG-CQR framework, which improves the retrieval phase of retrieval-augmented generation systems by leveraging semantically rich knowledge graph subgraphs to enhance the input query with context.
KG-RAG: Enhancing GUI Agent Decision-Making via Knowledge Graph-Driven Retrieval-Augmented Generation
Ziyi Guan (Huawei Hong Kong Research Center), Ngai Wong (University of Hong Kong)
OptimizationComputational EfficiencyTransformerLarge Language ModelVision Language ModelTextGraphBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose the KG-RAG framework, which converts the UI Transition Graph (UTG) into a structured vector knowledge base and enhances the decision-making and execution efficiency of GUI agents through retrieval-augmented generation.
KGE Calibrator: An Efficient Probability Calibration Method of Knowledge Graph Embedding Models for Trustworthy Link Prediction
Yang Yang (University of Galway), Edward Curry (University of Galway)
Representation LearningGraph Neural NetworkGraph
🎯 What it does: Propose a probabilistic calibration framework KGEC for knowledge graph embedding models, addressing the shortcomings of traditional calibration methods in large class spaces and ranking preservation.
KLAAD: Refining Attention Mechanisms to Reduce Societal Bias in Generative Language Models
Seorin Kim (Seoul National University), Jaejin Lee (Seoul National University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes KLAAD, a debiasing method based on attention alignment, specifically designed for decoder-only generative language models;
Knowledge Editing through Chain-of-Thought
Changyue Wang (Tsinghua University), Yiqun Liu (Tsinghua University)
Large Language ModelTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes a method for knowledge updating of LLMs through chain-of-thought editing (EditCoT).
Knowledge-Aware Co-Reasoning for Multidisciplinary Collaboration
Xurui Li, Haixu Tang (Indiana University)
OptimizationGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringTextGraphBiomedical Data
🎯 What it does: Constructed the KACR framework, leveraging knowledge graph-driven interdisciplinary collaborative reasoning optimized by reinforcement learning to enhance clinical diagnostic accuracy.
KoBLEX: Open Legal Question Answering with Multi-hop Reasoning
Jihyung Lee (POSTECH), Gary Lee (POSTECH)
TransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed the KOBLEX benchmark to evaluate multi-hop, explanation-based answers grounded in legal provisions within open-source legal QA; designed the PARSER method, enhancing multi-hop legal reasoning through parameterized legal provision generation and a three-stage retrieval process; and introduced the LF-EVAL evaluation metric for automatically assessing the legal faithfulness of answers.
Koel-TTS: Enhancing LLM based Speech Generation with Preference Alignment and Classifier Free Guidance
Shehzeen Samarah Hussain (NVIDIA Corporation), Jason Li (NVIDIA Corporation)
GenerationReinforcement Learning from Human FeedbackTransformerLarge Language ModelTextAudio
🎯 What it does: Propose a self-regressive TTS model Koel-TTS based on large language models, achieving fast and natural multilingual speech synthesis through a low-frame-rate audio codec.
KRETA: A Benchmark for Korean Reading and Reasoning in Text-Rich VQA Attuned to Diverse Visual Contexts
Taebaek Hwang (Waddle), Hyunjun Eun (SK Telecom)
Large Language ModelVision Language ModelMultimodalityBenchmark
🎯 What it does: Constructed and publicly released KRETA—a VQA benchmark for Korean text-rich images, covering 15 industry domains and 26 image types, evaluated using a two-tier reasoning framework (System 1 for basic recognition, System 2 for high-level reasoning).
Label Set Optimization via Activation Distribution Kurtosis for Zero-Shot Classification with Generative Models
Yue Li (University of Sheffield), Carolina Scarton (University of Sheffield)
ClassificationData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: Investigated the impact of label options (vocabulary selection, order, and expansion) on classification performance in zero-shot in-context learning (ICL), and proposed a post-hoc method called LOADS based on the kurtosis of neuron activation distributions to automatically select the optimal label set.
LaMDAgent: An Autonomous Framework for Post-Training Pipeline Optimization via LLM Agents
Taro Yano (NEC Corporation), Masafumi Oyamada (NEC Corporation)
OptimizationTransformerSupervised Fine-TuningAgentic AIPrompt EngineeringText
🎯 What it does: Leverage LLM agents to automatically construct and optimize complete post-training pipelines, systematically exploring combinations and sequences of various post-training techniques such as model fusion and supervised fine-tuning;
LaMP-QA: A Benchmark for Personalized Long-form Question Answering
Alireza Salemi (University of Massachusetts Amherst), Hamed Zamani (University of Massachusetts Amherst)
TransformerLarge Language ModelTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Constructed and made public the LaMP-QA benchmark for evaluating long-form personalized question-answering systems, and proposed an evaluation framework based on user history and question narratives.
Language Mixing in Reasoning Language Models: Patterns, Impact, and Internal Causes
Mingyang Wang (Bosch Center for Artificial Intelligence), Hinrich Schuetze (LMU Munich)
Explainability and InterpretabilityKnowledge DistillationTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Systematically study the language mixing phenomenon in reasoning language models (RLMs), analyzing its occurrence patterns, impact on performance, and internal causes.
Language Model Based Text-to-Audio Generation: Anti-Causally Aligned Collaborative Residual Transformers
Juncheng Wang (Hong Kong Polytechnic University), Shujun Wang (Hong Kong Polytechnic University)
GenerationTransformerReinforcement LearningDiffusion modelTextMultimodalityAudio
🎯 What it does: Propose a text-to-audio generation framework called Siren based on language models, which splits the prediction of multi-layer RVQ tokens into multiple collaborative Transformers and uses reinforcement learning for reverse causal alignment to address gradient conflicts and exposure bias caused by orthogonality and semantic decay in RVQ layers.
Language Models as Causal Effect Generators
Lucius E.j. Bynum, Kyunghyun Cho (New York University)
GenerationData SynthesisTransformerLarge Language ModelTextBenchmark
🎯 What it does: Proposed the Sequence-Driven Structural Causal Models (SD-SCM) framework, which automatically generates observational, interventional, and counterfactual sequence data by utilizing language models and user-specified directed acyclic graphs (DAGs), and uses this to construct a new causal inference benchmark;
Language Models as Continuous Self-Evolving Data Engineers
Peidong Wang (Northeastern University), Kaisong Song (Northeastern University)
Data-Centric LearningReinforcement Learning from Human FeedbackTransformerSupervised Fine-TuningReinforcement LearningAgentic AIText
🎯 What it does: Propose the LANCE framework, enabling large language models to autonomously generate, clean, review, and annotate data with preference information, achieving self-training of the model;
Language Models Can be Efficiently Steered via Minimal Embedding Layer Transformations
Diogo Tavares (NOVA University of Lisbon), Joao Magalhaes
Computational EfficiencyRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Achieve parameter-efficient fine-tuning of large language models (LLMs) by adding a learnable translation (bias) in the embedding layer, and propose the TinyTE method.
Language models can learn implicit multi-hop reasoning, but only if they have lots of training data
Yuekun Yao (Saarland University), Alexander Koller (Saarland University)
Explainability and InterpretabilityRepresentation LearningData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: This paper conducts experiments on a synthetic k-hop reasoning dataset (k=2,3,4,…) by training the GPT-2 model from scratch, investigating whether language models can learn implicit multi-hop reasoning and exploring their requirements for training data volume and model depth.
Language Models Identify Ambiguities and Exploit Loopholes
Jio Choi (University of North Carolina Chapel Hill), Elias Stengel-Eskin (University of North Carolina Chapel Hill)
Explainability and InterpretabilityTransformerPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: This paper designs three types of ambiguous scenarios (scalar implicature, bracketing ambiguity, and power scenarios) to evaluate whether large language models proactively exploit loopholes when faced with vague instructions and conflicting goals, i.e., identifying and misinterpreting ambiguities to satisfy their own objectives rather than user needs.
Language-Guided Temporal Token Pruning for Efficient VideoLLM Processing
Yogesh Kumar (Indian Institute of Technology Jodhpur)
Computational EfficiencyTransformerVision Language ModelVideoText
🎯 What it does: Proposes a language-guided temporal token pruning (LGTTP) method for efficiently processing long videos in vision-language models, which adaptively retains important frames and removes irrelevant frames based on temporal prompts in queries.
Language-to-Space Programming for Training-Free 3D Visual Grounding
Boyu Mi (Shanghai Jiao Tong University), Jiangmiao Pang (Shanghai Artificial Intelligence Laboratory)
Object DetectionRetrievalComputational EfficiencyTransformerLarge Language ModelVision Language ModelTextPoint Cloud
🎯 What it does: Propose LASP, a training-free 3D visual localization method that leverages Python code generated by LLMs to encode spatial relationships, combines automated testing and optimization, and ultimately achieves object localization in 3D scenes through symbolic reasoning and VLM discrimination.
Languages Still Left Behind: Toward a Better Multilingual Machine Translation Benchmark
Chihiro Taguchi (University Of Notre Dame), David Chiang (University Of Notre Dame)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Evaluate and reveal the shortcomings of the FLORES+ multilingual MT benchmark in terms of quality, domain coverage, and cultural bias through human re-evaluation and experiments.
Large Language Models as Realistic Microservice Trace Generators
Donghyun Kim (University Of Texas At Austin), Aditya Akella (University Of Texas At Austin)
Data SynthesisAnomaly DetectionTransformerLarge Language ModelSupervised Fine-TuningGraphSequential
🎯 What it does: Propose TraceLLM, which uses a large language model to generate synthetic workload traces for microservice call graphs.
Large Language Models Badly Generalize across Option Length, Problem Types, and Irrelevant Noun Replacements
Guangxiang Zhao (Qiyuan Tech), Xiangzheng Zhang (360zhinao)
Explainability and InterpretabilityTransformerLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Propose the Generalization Stress Test framework to evaluate the generalization ability of LLMs under minimal content perturbations, such as variations in option length, question types, and irrelevant noun substitutions.
Large Language Models Discriminate Against Speakers of German Dialects
Minh Duc Bui (Johannes Gutenberg University Mainz), Katharina von der Wense (Johannes Gutenberg University Mainz)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Investigate biases in large language models toward German dialect speakers, designing association tasks and decision tasks to assess biases in dialect naming and dialect usage.
Large Language Models Do Multi-Label Classification Differently
Marcus Ma (University of Southern California), Shrikanth Narayanan (University of Southern California)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Investigate the behavior of large language models (LLMs) in multi-label classification, analyzing the output distribution at each step and discovering that they generate peak distributions similar to sequential single-label classification. Propose a distribution alignment task and present zero-shot and supervised methods (e.g., max-over-generations, unary and binary breakdown), verified through comparative experiments to outperform traditional methods.
Large Language Models for Automated Literature Review: An Evaluation of Reference Generation, Abstract Writing, and Review Composition
Xuemei Tang (Hong Kong Polytechnic University), Zhenguang Cai (Chinese University of Hong Kong)
TransformerLarge Language ModelTextReview/Survey PaperBenchmarkRetrieval-Augmented Generation
🎯 What it does: Constructed a framework for automatically evaluating the ability of large language models (LLMs) to write literature reviews, designed three independent tasks (reference generation, abstract writing, and review writing), and quantified LLM performance across three tasks using multi-dimensional evaluation metrics (false positive rate, accuracy, coverage, factual consistency, semantic coverage).
Large Language Models Have Intrinsic Meta-Cognition, but Need a Good Lens
Ziyang Ma (Southeast University), Deyu Zhou (Southeast University)
TransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Studied the metacognitive abilities of large language models (LLMs), proposed an automatically evaluated framework named AutoMeco without human annotations, and designed a training-agnostic Markovian Intrinsic Reward Adjustment (MIRA) strategy to improve the accuracy of self-evaluation perspectives.
Large Language Models Meet Knowledge Graphs for Question Answering: Synthesis and Opportunities
Chuangtao Ma (Aalborg University), Haofen Wang (Tongji University)
Graph Neural NetworkTransformerLarge Language ModelGraphReview/Survey Paper
🎯 What it does: Reviews the integration methods of large language models (LLMs) and knowledge graphs (KGs) in question answering, proposes a structured multi-dimensional classification framework, systematically organizes and aligns solutions for various complex question answering tasks, analyzes the advantages and disadvantages of existing technologies, and outlines future research directions.
Large Language Models Threaten Language’s Epistemic and Communicative Foundations
Shashank Srivastava (UNC Chapel Hill)
TransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: This paper systematically elaborates the impact of large language models (LLMs) on the cognitive and dissemination foundations of language from the perspective of language philosophy and AI ethics. It introduces and discusses concepts such as 'epistemic doppelgängers,' 'authorship entropy,' 'hybrid authorship graph,' and 'proof-of-interaction,' and provides corresponding technical frameworks and metrics, aiming to offer references for future regulation, education, and technical practices.
LASER: An LLM-based ASR Scoring and Evaluation Rubric
Amruta Parulekar (Indian Institute of Technology Bombay), Preethi Jyothi (Indian Institute of Technology Bombay)
RecognitionLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmarkAudio
🎯 What it does: Proposed a LASER metric for ASR evaluation based on large language models (LLMs), which learns error types through carefully designed prompts and examples, and provides fine-grained scores.
Latent Inter-User Difference Modeling for LLM Personalization
Yilun Qiu, Fuli Feng (University Of Science And Technology Of China)
Representation LearningTransformerLarge Language ModelPrompt EngineeringAuto EncoderContrastive LearningText
🎯 What it does: This paper proposes a personalized LLM framework called DEP, which models differences between users in the latent space by generating difference-aware embeddings using user history and contrast information from similar users;
LATTE: Learning to Think with Vision Specialists
Zixian Ma (University of Washington), Silvio Savarese (University of Washington)
Data SynthesisKnowledge DistillationLarge Language ModelSupervised Fine-TuningVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Learning the reasoning capabilities of visual language models when using specialized visual expert tools (LATTE).
Layer-Aware Representation Filtering: Purifying Finetuning Data to Preserve LLM Safety Alignment
Hao Li (Shanghai Artificial Intelligence Laboratory), Lei Sha (Beihang University)
Safty and PrivacyRepresentation LearningData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: Propose Layer-Aware Representation Filtering (LARF), a method that filters training samples causing safety alignment degradation by identifying safety-sensitive layers in LLMs and leveraging their representations.
Layer-wise Minimal Pair Probing Reveals Contextual Grammatical-Conceptual Hierarchy in Speech Representations
Linyang He (Columbia University), Nima Mesgarani (Columbia University)
ClassificationData SynthesisExplainability and InterpretabilityRepresentation LearningTransformerAudio
🎯 What it does: Synthesize 116,300 audio minimal pairs using TTS, systematically evaluating the syntactic and semantic representations of S3M, ASR, codec, and AudioLLM in context;
Layered Insights: Generalizable Analysis of Human Authorial Style by Leveraging All Transformer Layers
Milad Alshomary (Columbia University), Kathleen McKeown (Columbia University)
ClassificationDomain AdaptationTransformerContrastive LearningText
🎯 What it does: Propose the LIGHT method, which utilizes multi-layer projection and contrastive learning on representations from all layers of the Transformer model in the author attribution task to generate author similarity by fusing multi-layer features.
LCES: Zero-shot Automated Essay Scoring via Pairwise Comparisons Using Large Language Models
Takumi Shibata (Deloitte Touche Tohmatsu LLC), Yuichi Miyamura (Deloitte Touche Tohmatsu LLC)
ClassificationTransformerLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Propose a zero-shot automatic essay scoring framework LCES, which utilizes large language models to generate essay comparisons for judgment, and then converts the comparative preferences into continuous scores or rankings through RankNet.
Leaky Thoughts: Large Reasoning Models Are Not Private Thinkers
Tommaso Green (Parameter Lab), Seong Joon Oh (Parameter Lab)
Safty and PrivacyTransformerLarge Language ModelPrompt EngineeringTextSequentialChain-of-Thought
🎯 What it does: This paper investigates the risk of user privacy leakage in the reasoning traces of Large Reasoning Models (LRMs) within personal assistant scenarios, systematically evaluating the practicality and privacy of models under two assessment settings: probe and agent-based. It also proposes a simple reasoning anonymization method called RANA and provides annotated analysis to explain the leakage mechanisms.
LeanK: Learnable K Cache Channel Pruning for Efficient Decoding
Yike Zhang (Tsinghua University), Lili Qiu (Microsoft Research)
OptimizationComputational EfficiencyKnowledge DistillationTransformerLarge Language ModelTextBenchmark
🎯 What it does: Propose LeanK, a learning-based static K cache channel pruning method that enhances the decoding efficiency of long-context LLMs
Learn and Unlearn: Addressing Misinformation in Multilingual LLMs
TaiMing Lu (Johns Hopkins University), Philipp Koehn (Johns Hopkins University)
Safty and PrivacyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper investigates the spread of misinformation across different languages by injecting training data containing false information into a multilingual LLM (LLaMA3-8B), and systematically evaluates the effectiveness of three 'unlearning' approaches (only English, same-source language, cross-lingual) for eliminating misinformation. Finally, the paper proposes and verifies a method that combines English and the original language of the false information for unlearning, which can almost completely eliminate misinformation in all languages.
Learning Contextual Retrieval for Robust Conversational Search
Seunghan Yang (Qualcomm AI Research), Simyung Chang (Qualcomm AI Research)
RetrievalTransformerLarge Language ModelContrastive LearningTextBenchmark
🎯 What it does: Propose ContextualRetriever, an LLM-based retriever that directly encodes dialogue context into retrieval embeddings without requiring additional rewriting steps.
Learning from Diverse Reasoning Paths with Routing and Collaboration
Zhenyu Lei (University of Virginia), Jundong Li (University of Virginia)
Knowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes the QR-Distill framework, which efficiently trains small models using multi-path reasoning through three modules: quality filtering, conditional routing, and peer distillation.
Learning from Few Samples: A Novel Approach for High-Quality Malcode Generation
Haijian Ma (Huazhong University of Science and Technology), Yulai Xie (Huazhong University of Science and Technology)
GenerationData SynthesisTransformerLarge Language ModelReinforcement LearningGenerative Adversarial NetworkContrastive LearningText
🎯 What it does: A semi-supervised framework named GANGRL-LLM was constructed by combining generative adversarial networks (GAN) with large language models (LLM) to generate high-quality malicious code and enhance SQL injection detection performance under sample-scarce conditions.
Learning Like Humans: Advancing LLM Reasoning Capabilities via Adaptive Difficulty Curriculum Learning and Expert-Guided Self-Reformulation
Enci Zhang (State Key Laboratory of Mobile Network and Mobile Multimedia Technology), Lu Qianchun (State Key Laboratory of Mobile Network and Mobile Multimedia Technology)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningTextBenchmark
🎯 What it does: Enhancing the performance of large language models in complex mathematical reasoning tasks through two training strategies inspired by human learning
Learning Subjective Label Distributions via Sociocultural Descriptors
Mohammed Fayiz Parappan (Duke University), Ricardo Henao (Duke University)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelTextTabular
🎯 What it does: Constructed the LSLD framework, which integrates the multi-perspective human values of text (generated by LLM) with annotators' socio-cultural feature embeddings, learning and predicting the probability distribution of each text and annotator's toxicity labels, thereby obtaining an integrated subjective label distribution.
Learning to Ask: When LLM Agents Meet Unclear Instruction
Wenxuan Wang (Renmin University of China), Michael R. Lyu (Chinese University of Hong Kong)
ClassificationAgentic AIPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Proposes an evaluation framework and benchmark (NoisyToolBench) for addressing ambiguous instruction issues in large language model tool usage, along with a prompting technique (Ask-when-Needed, AwN) that enables models to proactively ask for clarification, and implements an automated evaluation tool called ToolEvaluator.
Learning to See through Sound: From VggCaps to Multi2Cap for Richer Automated Audio Captioning
Sangyeon Cho (Chung-Ang University), Junyeong Kim (Chung-Ang University)
GenerationTransformerLarge Language ModelVision Language ModelMultimodalityAudio
🎯 What it does: Proposed a vision-guided audio caption generation framework named Multi2Cap, and constructed a rich-visual-information multimodal audio caption dataset called VggCaps
Legal Fact Prediction: The Missing Piece in Legal Judgment Prediction
Junkai Liu (Harbin Institute of Technology), Shuyuan Zheng (University of Osaka)
TransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes the Legal Fact Prediction (LFP) task, addressing the challenge of traditional Legal Judgment Prediction (LJP) relying heavily on factual information;
LegalSearchLM: Rethinking Legal Case Retrieval as Legal Elements Generation
Chaeeun Kim (LBOX), Wonseok Hwang (LBOX)
GenerationRetrievalTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Designed and implemented a large-scale Korean legal case retrieval benchmark LEGAR BENCH, and proposed a generative retrieval model LegalSearchLM that treats retrieval as generating legal elements.
Lemmatization as a Classification Task: Results from Arabic across Multiple Genres
Mostafa Saeed (New York University Abu Dhabi), Nizar Habash (New York University Abu Dhabi)
ClassificationExplainability and InterpretabilityData-Centric LearningRecurrent Neural NetworkTransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper redefines the Arabic lemmatization task as a classification task of mapping words to Lemma-POS-Gloss (LPG) label sets, and proposes two novel methods based on machine translation and semantic clustering, while constructing a standardized test set covering multiple genres.
Lemmatization of Polish Multi-word Expressions
Magdalena Król (AGH University of Krakow), Paweł Lewkowicz (AGH University of Krakow)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper addresses the stemming task for Polish multi-word expressions (MWEs) and proper nouns, proposing a machine learning-based stemming approach using text-to-text Transformer models (plT5, mT5). The models are pre-trained and fine-tuned using silver-standard Wikipedia link data and gold-standard PolEval 2019 dataset, with systematic evaluation of the impact of context, transfer learning, proper noun stemming, and model quantization.
LEO-MINI: An Efficient Multimodal Large Language Model using Conditional Token Reduction and Mixture of Multi-Modal Experts
Yimu Wang (University of Waterloo), Krzysztof Czarnecki (University of Waterloo)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsVision Language ModelMultimodalityBenchmark
🎯 What it does: Proposes LEO-MINI, a multimodal large language model that achieves significant compression of visual tokens while maintaining or enhancing visual reasoning performance through conditional token aggregation (COTR) followed by multimodal expert mixture (MMOE).
Less is More: The Effectiveness of Compact Typological Language Representations
York Hay Ng (University of Toronto), En-Shiun Annie Lee (University of Toronto)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningData-Centric LearningTabular
🎯 What it does: This paper proposes a pipeline for feature selection and missing value imputation on the 800-dimensional language typological features of the extended URIEL+, generating compact and interpretable language vectors;
Less Is More? Examining Fairness in Pruned Large Language Models for Summarising Opinions
Nannan Huang (RMIT University), Xiuzhen Zhang (RMIT University)
GenerationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: This paper studies the impact of post-training pruning on the fairness of generating opinion summaries in large language models and proposes a new pruning strategy.
Less Is MuRE: Revisiting Shallow Knowledge Graph Embeddings
Victor Charpenay (Mines Saint-Etienne), Steven Schockaert (Cardiff University)
Representation LearningGraphBenchmark
🎯 What it does: This paper systematically studies shallow knowledge graph embedding models, particularly focusing on MuRE as a core framework. It conducts theoretical and experimental evaluations of MuRE's expressiveness, training strategies, and design choices (linear vs. bilinear, trainable bias, scaling and translation, cross-coordinate comparison, region-based width). MuRE is compared with its extension ExpressivE and mainstream shallow models, proposing that MuRE and ExpressivE can serve as new baselines.
Let’s Play Across Cultures: A Large Multilingual, Multicultural Benchmark for Assessing Language Models’ Understanding of Sports
Punit Kumar Singh (Indian Institute of Technology Patna), Jose G Moreno (Université de Toulouse)
TransformerLarge Language ModelPrompt EngineeringVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Built and released CultSportQA: a multilingual multicultural traditional sports QA benchmark covering 60 countries, 6 continents, 11 languages, and 33,000 multiple-choice questions (MCQ), containing three categories (history, rules, scenarios) with both text and image modalities.
Let’s Reason Formally: Natural-Formal Hybrid Reasoning Enhances LLM’s Math Capability
Ruida Wang (University Of Illinois Urbana Champaign), Tong Zhang (Hong Kong University Of Science And Technology)
TransformerLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: This paper proposes a hybrid reasoning framework (NFL-HR) that integrates formal language (FL) reasoning capabilities into natural language (NL) mathematical problem solving.
LeTS: Learning to Think-and-Search via Process-and-Outcome Reward Hybridization
Qi Zhang (Zhejiang University), Junbo Zhao (Zhejiang University)
RetrievalOptimizationReinforcement LearningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Designed the LeTS framework, integrating process-level rewards with result-level rewards to enhance reasoning-retrieval behavior in retrieval-augmented generation (RAG);
Leveraging Cognitive Complexity of Texts for Contextualization in Dense Retrieval
Effrosyni Sokli (University of Milano-Bicocca), Gabriella Pasi (University of Milano-Bicocca)
ClassificationRetrievalTransformerMixture of ExpertsContrastive LearningText
🎯 What it does: This paper proposes a dense retrieval model called DenseC3, which achieves semantic contextualization of vectors by embedding the cognitive complexity (based on Bloom's cognitive hierarchy) of text into the vector representations of queries and documents.
Leveraging Knowledge Graph-Enhanced LLMs for Context-Aware Medical Consultation
Su-Hyeong Park (Catholic University of Korea), Kang-Min Kim (Catholic University of Korea)
TransformerLarge Language ModelTextGraphElectronic Health RecordsRetrieval-Augmented Generation
🎯 What it does: Propose the ILlama framework, integrating retrieval-augmented generation (RAG) with the UMLS structured knowledge graph to achieve hallucination-free, context-aware medical question answering.
Leveraging Large Models to Evaluate Novel Content: A Case Study on Advertisement Creativity
Zhaoyi Joey Hou (University of Pittsburgh), Xiang Lorraine Li (University of Pittsburgh)
TransformerLarge Language ModelVision Language ModelImageMultimodalityBenchmark
🎯 What it does: This paper proposes a new benchmark for visual ad creativity evaluation, decomposing creativity into two dimensions: originality and atypicality, and constructing distributed annotations using 25 diverse human ratings.
Leveraging Loanword Constraints for Improving Machine Translation in a Low-Resource Multilingual Context
Felermino D. M. A. Ali (Universidade do Porto), Rui Sousa-Silva (Universidade do Porto)
GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Research on improving machine translation in low-resource language pairs by leveraging external loanword constraints
Leveraging Multilingual Training for Authorship Representation: Enhancing Generalization across Languages and Domains
Junghwan Kim (University of Michigan), David Jurgens (University of Michigan)
Representation LearningTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Trained and evaluated a multilingual shared author writing style representation model.
Leveraging Semantic Triples for Private Document Generation with Local Differential Privacy Guarantees
Stephen Meisenbacher (Technical University of Munich), Florian Matthes (Technical University of Munich)
GenerationSafty and PrivacyTransformerLarge Language ModelText
🎯 What it does: Proposes a local differential privacy text generation method DP-ST based on semantic triplets. First, documents are decomposed into SVO triplets, then privacy is applied within the semantic neighborhood using the exponential mechanism, followed by reconstructing coherent text with an LLM.
Leveraging Text-to-Text Transformers as Classifier Chain for Few-Shot Multi-Label Classification
Quang Anh Nguyen (University of Versailles), Hanane Azzag
ClassificationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Propose a sequential-set chain framework based on T5, achieving few-shot multi-label text classification through a three-stage training process (general pre-training → domain adaptation → chain specialization);
Leveraging What’s Overfixed: Post-Correction via LLM Grammatical Error Overcorrection
Taehee Park (POSTECH), Gary Lee (POSTECH)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This paper proposes a two-stage GEC method called PoCO, which first utilizes an LLM to intentionally trigger over-correction to improve recall, and then performs post-correction using a fine-tuned small model to enhance precision.
LGA: LLM-GNN Aggregation for Temporal Evolution Attribute Graph Prediction
Feng Zhao (Huazhong University of Science and Technology), Xianggan Liu (Huazhong University of Science and Technology)
Representation LearningRecurrent Neural NetworkGraph Neural NetworkTransformerLarge Language ModelPrompt EngineeringGraph
🎯 What it does: Proposed a framework named LGA that integrates large language models (LLM) and graph neural networks (GNN) for temporal evolution attribute graph prediction, utilizing LLM to perform GNN-style aggregation, and enhancing attribute prediction accuracy through two modules: attribute embedding loss and R-GCN, which separately aggregate node and attribute information.
Liaozhai through the Looking-Glass: On Paratextual Explicitation of Culture-Bound Terms in Machine Translation
Sherrie Shen (University of Edinburgh), Alexandra Birch (University of Edinburgh)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Investigates the generation of paratextual explicitation for culturally bound terms in machine translation, proposing a task to enhance cross-cultural understanding by supplementing annotations from external textual sources;
LIDDIA: Language-based Intelligent Drug Discovery Agent
Reza Averly (Ohio State University), Xia Ning (Ohio State University)
OptimizationDrug DiscoveryTransformerLarge Language ModelAgentic AIBiomedical DataChain-of-Thought
🎯 What it does: Propose the LIDDIA agent, which automatically navigates the drug discovery process in a computer
LightThinker: Thinking Step-by-Step Compression
Jintian Zhang (Zhejiang University), Ningyu Zhang (Zhejiang University)
CompressionComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningTextBenchmarkChain-of-Thought
🎯 What it does: Propose the LightThinker method, which trains LLMs to dynamically compress intermediate thinking steps during inference, condensing lengthy reasoning into a few gist tokens to reduce KV Cache size and inference cost.
LILaC: Late Interacting in Layered Component Graph for Open-domain Multimodal Multihop Retrieval
Joohyung Yun (POSTECH), Wook-Shin Han (POSTECH)
RetrievalGraph Neural NetworkLarge Language ModelVision Language ModelMultimodality
🎯 What it does: Propose the LILaC multimodal retrieval framework, which utilizes hierarchical component graphs and late-interaction subgraph retrieval to address the challenges of granularity and multi-hop reasoning in multimodal document retrieval.
LimRank: Less is More for Reasoning-Intensive Information Reranking
Tingyu Song (Yale NLP Lab), Arman Cohan (Yale NLP Lab)
Data SynthesisRetrievalData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: Propose LIMRANK-SYNTHESIZER and LIMRANK reranker, achieving efficient training for reasoning-intensive reranking in information retrieval using a small amount of high-quality synthetic data.
Linear-Time Demonstration Selection for In-Context Learning via Gradient Estimation
Ziniu Zhang (Northeastern University), Hongyang R. Zhang (Northeastern University)
ClassificationComputational EfficiencyMeta LearningTextGraphSequential
🎯 What it does: Proposed a linear-time demonstration selection algorithm based on gradient estimation, which can quickly select the most suitable k examples from a large number of examples as context prompts;
LingGym: How Far Are LLMs from Thinking Like Field Linguists?
Changbing Yang (University Of British Columbia), Jian Zhu (University Of British Columbia)
TransformerLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Constructed and released the LINGGYM benchmark, utilizing Interlinear Glossed Text (IGT) from 18 publicly available reference grammars and designing a Word-Gloss inference task to evaluate large language models' metalinguistic reasoning and structural generalization capabilities in low-resource languages.
LinguaLens: Towards Interpreting Linguistic Mechanisms of Large Language Models via Sparse Auto-Encoder
Yi Jing (DCST, BNRist; KIRC, Institute for Artificial Intelligence), Juanzi Li (DCST, BNRist; KIRC, Institute for Artificial Intelligence)
Explainability and InterpretabilityLarge Language ModelAuto EncoderContrastive LearningText
🎯 What it does: Proposed the LinguaLens framework, which uses sparse autoencoders (SAE) to decompose LLM hidden layer states into a high-dimensional sparse concept space. Based on a self-built dataset with 145 multidimensional linguistic features (morphology, syntax, semantics, pragmatics, etc.), it automatically extracts and interprets internal language processing mechanisms of Chinese and English LLMs, and achieves causal intervention on features.
Linguistic and Embedding-Based Profiling of Texts Generated by Humans and Large Language Models
Sergio E. Zanotto (University of Konstanz), Segun Aroyehun (University of Konstanz)
ClassificationTransformerLarge Language ModelText
🎯 What it does: Systematically analyze the morphological, syntactic, and semantic features, as well as style embeddings, of human-written texts and texts generated by 11 large language models, and use logistic regression to evaluate feature importance, exploring the impact of different models, decoding strategies, and release dates on text diversity;
Linguistic Neuron Overlap Patterns to Facilitate Cross-lingual Transfer on Low-resource Languages
Yuemei Xu (Beijing Foreign Studies University), Lin Gui (King's College London)
Domain AdaptationRepresentation LearningTransformerLarge Language ModelPrompt EngineeringContrastive LearningText
🎯 What it does: Propose the BridgeX-ICL method, which selects the optimal bridge language in zero-shot cross-lingual context learning for low-resource languages by leveraging inter-lingual neuron overlap patterns;
LinkAlign: Scalable Schema Linking for Real-World Large-Scale Multi-Database Text-to-SQL
Yihan Wang (China Academy of Information and Communications Technology), Xin Yang (China Academy of Information and Communications Technology)
Computational EfficiencyData-Centric LearningTransformerLarge Language ModelAgentic AIPrompt EngineeringTextTabularBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes the LinkAlign framework, which addresses Schema Linking in large-scale multi-database environments through a step-by-step process, including multi-round semantic retrieval and query rewriting, response filtering to eliminate noise from irrelevant databases, and schema parsing to identify key tables and columns; it also achieves a balance between efficiency and accuracy through pluggable Pipeline and Agent modes.
LiteASR: Efficient Automatic Speech Recognition with Low-Rank Approximation
Keisuke Kamahori (University of Washington), Baris Kasikci (University of Washington)
RecognitionCompressionTransformerAudio
🎯 What it does: Perform low-rank compression on the Encoder of the ASR model, significantly reducing inference cost.
LiteraryQA: Towards Effective Evaluation of Long-document Narrative QA
Tommaso Bonomo (Sapienza University of Rome), Roberto Navigli (Sapienza University of Rome)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Constructed a high-quality literary text QA subset called LiteraryQA, performing document and QA-level cleaning on NarrativeQA through dual verification by humans and LLMs, removing mismatched books, non-narrative texts, and noisy QA pairs.
LiTEx: A Linguistic Taxonomy of Explanations for Understanding Within-Label Variation in Natural Language Inference
Pingjun Hong (LMU Munich), Barbara Plank (LMU Munich)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Proposed a linguistic explanation classification system named LITEX, specifically designed to analyze different reasoning paths under the same label in natural language inference (NLI), with annotation, verification, and analysis of free-text explanations on the e-SNLI dataset.
LiTransProQA: An LLM-based Literary Translation Evaluation Metric with Professional Question Answering
Ran Zhang (University of Mannheim), Steffen Eger (University of Technology Nuremberg)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes a reference-free, specialized quality assessment framework for literary translation called LITRANSPROQA, based on large language models.
LLaMP: Large Language Model Made Powerful for High-fidelity Materials Knowledge Retrieval
Yuan Chiang (University of California Berkeley), Janosh Riebesell (Lawrence Berkeley National Laboratory)
RetrievalTransformerLarge Language ModelAgentic AITabularPhysics RelatedRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes LLaMP—a hierarchical multi-agent framework for high-fidelity material knowledge retrieval and experimental process simulation in materials science research through large language models (LLMs);
LLM Bias Detection and Mitigation through the Lens of Desired Distributions
Ingroj Shrestha (University of Iowa), Padmini Srinivasan (University of Iowa)
TransformerLarge Language ModelSupervised Fine-TuningTextTabular
🎯 What it does: This paper proposes a fine-tuning method using weighted adaptive KL loss to align the gender-occupation output distribution of large language models with desired distributions (equal or real-world distributions), and validates it on masked language models and autoregressive models.
LLM-Driven Completeness and Consistency Evaluation for Cultural Heritage Data Augmentation in Cross-Modal Retrieval
Jian Zhang (Xi'an Jiaotong-Liverpool University), Dongming Lu (Zhejiang University)
RetrievalTransformerLarge Language ModelContrastive LearningMultimodalityChain-of-Thought
🎯 What it does: Propose the C3 framework, which enhances the completeness and consistency of cross-modal retrieval for cultural heritage through LLM-driven text augmentation.