ACL 2025 Papers — Page 9
Annual Meeting of the Association for Computational Linguistics · 1699 papers
Instance-Selection-Inspired Undersampling Strategies for Bias Reduction in Small and Large Language Models for Binary Text Classification
Guilherme Fonseca (Universidade Federal de Minas Gerais), Leonardo Chaves Dutra Da Rocha (Universidade Federal de São João del Rei)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper addresses the bias caused by class imbalance in binary text classification tasks, focusing on small and large language models (RoBERTa, Llama3.1), and proposes two instance-based under-sampling methods (E2SC_US and UBR). These methods are compared with 19 baseline approaches across 13 datasets.
Instruction Tuning on Public Government and Cultural Data for Low-Resource Language: a Case Study in Kazakh
Nurkhan Laiyk (Mohamed bin Zayed University of Artificial Intelligence), Fajri Koto (Mohamed bin Zayed University of Artificial Intelligence)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextRetrieval-Augmented Generation
🎯 What it does: Constructed a dataset of 10,600 Kazakh instruction-response pairs covering government and cultural domains, and fully fine-tuned multiple LLMs on this dataset;
InstructPart: Task-Oriented Part Segmentation with Instruction Reasoning
Zifu Wan (Carnegie Mellon University), Katia P. Sycara
SegmentationLarge Language ModelSupervised Fine-TuningVision Language ModelMultimodalityBenchmark
🎯 What it does: Proposes a new real-world benchmark, InstructPart, for evaluating the performance of vision-language models in task-oriented part segmentation and reasoning;
Integrating Audio, Visual, and Semantic Information for Enhanced Multimodal Speaker Diarization on Multi-party Conversation
Luyao Cheng (Tongyi Lab), Wen Wang (Tongyi Lab)
SegmentationConvolutional Neural NetworkTransformerMultimodalityBenchmark
🎯 What it does: Proposes a speaker segmentation framework based on audio, visual, and semantic multimodal information, enhancing clustering performance by constructing dual constraints between visual and semantic modalities and jointly propagating them.
INTERACT: Enabling Interactive, Question-Driven Learning in Large Language Models
Aum Kendapadi (University Of North Carolina Chapel Hill), Shashank Srivastava (University Of North Carolina Chapel Hill)
Knowledge DistillationTransformerLarge Language ModelText
🎯 What it does: Proposes the INTERACT framework, enabling large language models (LLM) to undergo interactive, question-driven learning through student-teacher dialogues;
Interactive and Expressive Code-Augmented Planning with Large Language Models
Anthony Zhe Liu, Honglak Lee (University of Michigan)
AI Code AssistantTransformerLarge Language ModelText
🎯 What it does: Propose an interactive code planning framework called REPL-Plan based on large language models (LLMs), which enables LLMs to interact with the REPL environment and recursively generate subtasks (REPL) to achieve top-down dynamic planning;
Interactive Evolution: A Neural-Symbolic Self-Training Framework For Large Language Models
Fangzhi Xu (Xi'an Jiaotong University), Zhiyong Wu (Shanghai AI Lab)
TransformerLarge Language ModelReinforcement LearningContrastive LearningText
🎯 What it does: Proposed the ENVISIONS framework, enabling large language models (LLMs) to self-train without human-annotated data by generating symbolic solutions through environmental interaction and improving the model via self-reward;
Interlocking-free Selective Rationalization Through Genetic-based Learning
Federico Ruggeri (University of Bologna), Gaetano Signorelli (University of Bologna)
ClassificationData SynthesisExplainability and InterpretabilityRecurrent Neural NetworkText
🎯 What it does: Propose the GenSPP framework, which uses genetic algorithms to separate the training of the generator and predictor, achieving unsupervised selective rationalization and completely eliminating the interlocking problem in traditional select-then-predict models.
Internal and External Impacts of Natural Language Processing Papers
Yu Zhang (Texas A&M University)
Text
🎯 What it does: This paper conducts a large-scale scientometric analysis of academic internal citations (OpenAlex) and external citations (patents, media, policy documents) of papers published in ACL, EMNLP, and NAACL from 1979 to 2024, quantifying their influence across different fields based on paper topics.
Internal Value Alignment in Large Language Models through Controlled Value Vector Activation
Haoran Jin (University of Science and Technology of China), Defu Lian (University of Science and Technology of China)
Explainability and InterpretabilityAdversarial AttackTransformerLarge Language ModelContrastive LearningText
🎯 What it does: This paper proposes the ConVA framework, which achieves fine-grained alignment of model values by identifying and activating value vectors within LLMs;
Interpret and Improve In-Context Learning via the Lens of Input-Label Mappings
Chenghao Sun (University of Science and Technology of China), Jieping Ye (University of Science and Technology of China)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Researchers conducted principal component analysis (PCA) on the hidden layers of large language models and found that input-label mappings are stored in key layers as principal components. They used PC patching technology to locate the role of these mappings in attention heads, followed by fine-tuning a small number of key heads, significantly improving the model's in-context learning performance across multiple tasks.
Introducing Graph Context into Language Models through Parameter-Efficient Fine-Tuning for Lexical Relation Mining
Jingwen Sun (University of Science and Technology of China), Guangzhong Sun (University of Science and Technology of China)
ClassificationComputational EfficiencyGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningTextGraph
🎯 What it does: Construct a lexical graph, extract graph features using a relation-sensitive graph neural network (GNN), map them to linguistic contexts, and enhance performance in lexical relation classification and entailment tasks via parameter-efficient fine-tuning (GET).
Introducing Verification Task of Set Consistency with Set-Consistency Energy Networks
Mooho Song (Seoul National University), Jay-Yoon Lee (Seoul National University)
ClassificationAnomaly DetectionRepresentation LearningTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Proposed the set consistency verification task and designed the Set-Consistency Energy Network (SC-Energy) to detect logical inconsistencies among multiple statements.
Intuitive Fine-Tuning: Towards Simplifying Alignment into a Single Process
Ermo Hua, Bowen Zhou (Tsinghua University)
Reinforcement Learning from Human FeedbackSupervised Fine-TuningReinforcement LearningText
🎯 What it does: This paper explains the similarities and differences between supervised fine-tuning (SFT) and preference optimization (PO) by unifying them into a Markov decision process (MDP) framework, and proposes the Intuitive Fine-Tuning (IFT) method, which integrates the advantages of SFT and PO into a single process, achieving model alignment using only target answer data.
InvestAlign: Overcoming Data Scarcity in Aligning Large Language Models with Investor Decision-Making Processes Under Herd Behavior
Huisheng Wang (Tsinghua University), H. Vicky Zhao (Tsinghua University)
TransformerLarge Language ModelSupervised Fine-TuningTextFinance Related
🎯 What it does: This study proposes and verifies the InvestAlign framework, which generates high-quality supervised fine-tuning data using theoretical solutions of simple problems, trains LLMs to align investors' decisions under group behavior, and demonstrates its superior performance in complex investment scenarios.
Investigating and Enhancing the Robustness of Large Multimodal Models Against Temporal Inconsistency
Jiafeng Liang (Harbin Institute of Technology), Bing Qin (National University of Singapore)
Reinforcement Learning from Human FeedbackTransformerVideoTextMultimodalityBenchmark
🎯 What it does: This paper systematically studies the robustness of large-scale multi-modal models (LMM) when facing temporal inconsistencies, constructs a new benchmark TEMROBBENCH, and proposes a multi-modal PanoDPO method to enhance temporal robustness.
Investigating and Extending Homans’ Social Exchange Theory with Large Language Model based Agents
Lei Wang (Renmin University of China), Xu Chen (Renmin University of China)
Large Language ModelAgentic AIText
🎯 What it does: This paper utilizes large language model (LLM) agents to construct a virtual society, conducts social exchange games, and verifies and extends Homans' social exchange theory.
INVESTORBENCH: A Benchmark for Financial Decision-Making Tasks with LLM-based Agent
Haohang Li (Stevens Institute of Technology), Qianqian Xie (Fin AI)
TransformerLarge Language ModelReinforcement LearningAgentic AITextMultimodalityBenchmarkFinance Related
🎯 What it does: Proposed INVESTORBENCH, a comprehensive benchmark framework for evaluating LLM-driven financial decision-making agents;
IOPO: Empowering LLMs with Complex Instruction Following via Input-Output Preference Optimization
Xinghua Zhang (Alibaba Group), Yongbin Li (Alibaba Group)
TransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmark
🎯 What it does: Proposes the TRACE benchmark and the IOPO alignment method, aiming to enhance large language models' ability to follow complex instructions under multiple constraints.
IPO: Your Language Model is Secretly a Preference Classifier
Shivank Garg (Indian Institute of Technology Roorkee), Paras Chopra (Lossfunk)
OptimizationLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmark
🎯 What it does: Implicitly determine the quality of answers using the output probabilities ('Yes/No') of large language models (LLMs) themselves, and build a preference dataset based on this for direct preference optimization (DPO), enabling model self-improvement.
iQUEST: An Iterative Question-Guided Framework for Knowledge Base Question Answering
Shuai Wang (Chalmers University of Technology), Yinan Yu (Chalmers University of Technology)
Graph Neural NetworkTransformerLarge Language ModelGraphBenchmarkChain-of-Thought
🎯 What it does: Proposed the iQUEST framework, which achieves multi-hop knowledge graph reasoning by iteratively generating sub-questions and using graph neural networks for two-hop neighbor retrieval.
IRIS: An Iterative and Integrated Framework for Verifiable Causal Discovery in the Absence of Tabular Data
Tao Feng (Monash University), Gholamreza Haffari (Monash University)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextBiomedical DataAlzheimer's DiseaseRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes the IRIS framework to achieve real-time causal discovery from unstructured text to structured data.
IRIS: Interpretable Retrieval-Augmented Classification for Long Interspersed Document Sequences
Fengnan Li (Duke University), Matthew M. Engelhard (Duke University)
ClassificationExplainability and InterpretabilityTextBiomedical DataElectronic Health RecordsRetrieval-Augmented Generation
🎯 What it does: Proposed a lightweight, interpretable retrieval-augmented long text classification framework named IRIS, which classifies extremely long documents by aggregating document segments using retrieval query vectors.
Iron Sharpens Iron: Defending Against Attacks in Machine-Generated Text Detection with Adversarial Training
Yuanfan Li (Xi'an Jiaotong University), Xiaoming Liu (Xi'an Jiaotong University)
Adversarial AttackTransformerText
🎯 What it does: Proposes the GREATER framework, which enhances the robustness of machine-generated text (MGT) detectors through adversarial training. The framework includes an adversarial generator, GREATER-A (which identifies important tokens using a proxy model, performs gradient ascent in the embedding space, and then generates stealthy and efficient attacks via greedy search and pruning), and a detector, GREATER-D (which simultaneously learns adversarial samples from the generator and updates together during training).
IRT-Router: Effective and Interpretable Multi-LLM Routing via Item Response Theory
Wei Song, Runze Wu (University Of Science And Technology Of China)
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: Propose an IRT-based multi-LLM routing framework called IRT-Router, which can efficiently match the most suitable model according to query difficulty and LLM capabilities;
Is linguistically-motivated data augmentation worth it?
Ray Groshan (University of Colorado), Alexis Palmer (University of Colorado)
RecognitionGenerationTransformerText
🎯 What it does: This paper systematically compares two categories of data augmentation methods for low-resource languages, evaluating their effectiveness on machine translation and linear annotation tasks.
Is That Your Final Answer? Test-Time Scaling Improves Selective Question Answering
William Jurayj, Benjamin Van Durme (Johns Hopkins University)
Computational EfficiencyLarge Language ModelTextChain-of-Thought
🎯 What it does: Explore introducing confidence thresholds when scaling large language models at test time, enabling selective answering and evaluating their impact on confidence and computational budget.
ISR: Self-Refining Referring Expressions for Entity Grounding
Zhuocheng Yu (Peking University), Zhonghui He (Peking University)
RecognitionRetrievalTransformerLarge Language ModelReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes an Iterative Self-Optimization (ISR) training framework that leverages multimodal large language models (MLLM) to automatically generate and iteratively refine referential expressions (RE), thereby enhancing the performance of the entity grounding task.
It’s Not a Walk in the Park! Challenges of Idiom Translation in Speech-to-text Systems
Iuliia Zaitova (Saarland University), Tania Avgustinova (Saarland University)
Explainability and InterpretabilityTransformerLarge Language ModelTextAudio
🎯 What it does: Systematically evaluate the performance of German→English and Russian→English speech translation (SLT), text translation (MT), and large language models (LLM) on idioms, and reveal the shortcomings of SLT in idiom processing through hierarchical analysis.
It’s Not Bragging If You Can Back It Up: Can LLMs Understand Braggings?
Jingjie Zeng (Dalian University of Technology), Hongfei Lin (Dalian University of Technology)
RecognitionGenerationTransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: This paper systematically studies the understanding and generation of self-praise behavior in large language models (LLMs), constructing three tasks (identification, explanation, generation) and designing corresponding evaluation metrics.
Jailbreak Large Vision-Language Models Through Multi-Modal Linkage
Yu Wang (Chinese Academy of Sciences), Tianxing He (Tsinghua University)
Adversarial AttackPrompt EngineeringVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Propose a Multimodal Link (MML) framework that achieves exploitation on vision-language models by encrypting malicious text in images and guiding VLMs to decrypt it.
Jailbreaking? One Step Is Enough!
Weixiong Zheng (Guangdong University of Technology), Yongmei Zhou (Guangdong University of Foreign Studies)
Safty and PrivacyAdversarial AttackTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Propose a reverse embedding defense attack (REDA) method to achieve one-time, cross-model LLM jailbreaking.
JailbreakRadar: Comprehensive Assessment of Jailbreak Attacks Against LLMs
Junjie Chu (CISPA Helmholtz Center for Information Security), Yang Zhang (CISPA Helmholtz Center for Information Security)
Safty and PrivacyAdversarial AttackTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper systematically evaluates 17 representative jailbreak attacks, proposes a six-class attack taxonomy, constructs a unified prohibited questions dataset with 160 questions and 16 violation categories, and assesses attack effectiveness and multiple defense methods on nine alignment LLMs, revealing attack patterns and defense challenges.
JoPA: Explaining Large Language Model’s Generation via Joint Prompt Attribution
Yurui Chang (Pennsylvania State University), Lu Lin (Pennsylvania State University)
Explainability and InterpretabilityTransformerPrompt EngineeringText
🎯 What it does: Propose a Joint Prompt Attribution (JoPA) framework that identifies the most influential subset of prompts by performing counterfactual search on combinations of input prompt text.
Just a Scratch: Enhancing LLM Capabilities for Self-harm Detection through Intent Differentiation and Emoji Interpretation
Soumitra Ghosh (Fondazione Bruno Kessler), Asif Ekbal (Indian Institute of Technology Jodhpur)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper improves the performance of large language models (LLMs) in detecting self-harm on social media by constructing the CESM-100, a contextual emoji explanation matrix containing 100 scenarios, and the SHINES dataset for self-harm identification and intent extraction;
Just Go Parallel: Improving the Multilingual Capabilities of Large Language Models
Muhammad Reza Qorib (National University of Singapore), Hwee Tou Ng (National University of Singapore)
TransformerLarge Language ModelText
🎯 What it does: The study systematically evaluates and quantifies the role of parallel data in enhancing the multilingual capabilities of large language models, focusing on translation and multilingual commonsense reasoning.
JuStRank: Benchmarking LLM Judges for System Ranking
Ariel Gera (IBM Research), Asaf Yehudai (IBM Research)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper studies the performance of large language models (LLMs) as reviewers in system-level ranking tasks and constructs the first large-scale reviewer benchmark, JuStRank.
K/DA: Automated Data Generation Pipeline for Detoxifying Implicitly Offensive Language in Korean
Minkyeong Jeon (Korea University), Byung-Jun Lee (Korea University)
GenerationData SynthesisLarge Language ModelSupervised Fine-TuningTextRetrieval-Augmented Generation
🎯 What it does: Proposes K/DA—a pipeline for automatically generating paired Korean offensive language datasets, focusing on trending slang and implicit aggression;
KatFishNet: Detecting LLM-Generated Korean Text through Linguistic Feature Analysis
Shinwoo Park (Yonsei University), Yo-Sub Han (Yonsei University)
ClassificationTextBenchmark
🎯 What it does: This paper creates a benchmark dataset for detecting LLM-generated text in Korean, named KatFish, and proposes a novel detection method called KatFishNet based on Korean-specific linguistic features.
KazMMLU: Evaluating Language Models on Kazakh, Russian, and Regional Knowledge of Kazakhstan
Mukhammed Togmanov (Mohamed bin Zayed University of Artificial Intelligence), Fajri Koto (Mohamed bin Zayed University of Artificial Intelligence)
TransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Proposes KazMMLU—a multitask language understanding benchmark tailored for the Kazakhstan context, comprising approximately 23,000 multiple-choice questions covering high school and university levels, STEM, humanities, social sciences, and professional disciplines, with both Kazakh and Russian versions provided.
KERL: Knowledge-Enhanced Personalized Recipe Recommendation using Large Language Models
Fnu Mohbat (Rensselaer Polytechnic Institute), Mohammed J Zaki
Recommendation SystemTransformerLarge Language ModelTextGraphBenchmarkRetrieval-Augmented Generation
🎯 What it does: The study proposes the KERL system, achieving unified recommendation, recipe generation, and micronutrient calculation based on FoodKG through multi-task LLM.
Keys to Robust Edits: From Theoretical Insights to Practical Advances
Jianhao Yan (Zhejiang University), Yue Zhang (Westlake University)
TransformerLarge Language ModelContrastive LearningText
🎯 What it does: This paper studies knowledge editing in large language models, pointing out that existing locate-edit methods using model internal representations as keys and values can lead to insufficient robustness and specificity;
KG-Agent: An Efficient Autonomous Agent Framework for Complex Reasoning over Knowledge Graph
Jinhao Jiang (Renmin University of China), Ji-Rong Wen (Renmin University of China)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningAgentic AIGraphRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes the KG-Agent framework, utilizing a small LLM, toolbox, KG executor, and knowledge memory to enable autonomous multi-step reasoning over knowledge graphs.
KiRAG: Knowledge-Driven Iterative Retriever for Enhancing Retrieval-Augmented Generation
Jinyuan Fang (University of Glasgow), Craig MacDonald
GenerationRetrievalLarge Language ModelContrastive LearningTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the KiRAG model, which enhances multi-hop question answering by splitting documents into knowledge triplets, performing knowledge-driven iterative retrieval, sorting documents using retrieved triplets, and generating answers with LLMs, thereby improving both retrieval and generation effectiveness.
Know You First and Be You Better: Modeling Human-Like User Simulators via Implicit Profiles
Kuang Wang (Chinese University of Hong Kong), Haizhou Li (Chinese University of Hong Kong)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningContrastive LearningText
🎯 What it does: This paper proposes a user simulation framework called USP based on implicit user profiling, which can extract latent features from human-computer dialogues and generate personalized, coherent user responses.
Know Your Mistakes: Towards Preventing Overreliance on Task-Oriented Conversational AI Through Accountability Modeling
Suvodip Dey (University of Illinois Urbana-Champaign), Dilek Hakkani-Tür (University of Illinois Urbana-Champaign)
Anomaly DetectionTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: In task-oriented dialogue systems, a large language model (LLM) architecture incorporating an 'accountability head' is proposed to detect and correct false positives and false negatives in dialogue state tracking (DST), thereby reducing users' over-reliance on AI.
Knowledge Boundary of Large Language Models: A Survey
Moxin Li (National University Of Singapore), Yang Deng (Singapore Management University)
TransformerLarge Language ModelPrompt EngineeringTextReview/Survey PaperRetrieval-Augmented Generation
🎯 What it does: A systematic review of the knowledge boundaries of large language models (LLMs), proposing a unified formal definition and four types of knowledge classification (universal knowledge boundaries, parameterized knowledge boundaries, outward knowledge boundaries), and summarizing and commenting on existing methods and datasets around three research questions: 'why study', 'how to identify', and 'how to alleviate'.
Knowledge Decoupling via Orthogonal Projection for Lifelong Editing of Large Language Models
Haoyu Xu (Northeastern University), Xingwei Wang (Northeastern University)
TransformerLarge Language ModelText
🎯 What it does: This paper proposes a lifelong knowledge editing framework named Knowledge Decoupling Editing (KDE), designed to continuously update the factual knowledge of large language models (LLMs) without compromising their general capabilities. The framework employs editable memory, a switching mechanism, knowledge decoupling cache, and two-phase training, utilizing orthogonal projection to eliminate interference from subsequent edits on previously edited knowledge (knowledge coupling).
Knowledge Graph Retrieval-Augmented Generation for LLM-based Recommendation
Shijie Wang (Hong Kong Polytechnic University), Dawei Yin (Baidu Inc)
Recommendation SystemGraph Neural NetworkTransformerPrompt EngineeringTextGraphRetrieval-Augmented Generation
🎯 What it does: Propose the K-RagRec framework, which enhances LLM recommendation systems by retrieving high-quality subgraphs from knowledge graphs.
Knowledge Image Matters: Improving Knowledge-Based Visual Reasoning with Multi-Image Large Language Models
Guanghui Ye (Hunan University), Zhihua Jiang (University of Sheffield)
TransformerLarge Language ModelVision Language ModelMultimodality
🎯 What it does: Propose Visual Knowledge Card (VKC) and its four-stage VKC-MIR framework, using multi-graph MLLM to address knowledge-based visual reasoning (KB-VR) tasks
Knowledge Tracing in Programming Education Integrating Students’ Questions
Doyoun Kim (Seoul National University), Yohan Jo (Seoul National University)
TransformerLarge Language ModelContrastive LearningTextMultimodality
🎯 What it does: This study proposes SQKT, a knowledge tracking model that integrates student questions and automatically extracts skill information to predict students' learning outcomes on subsequent problems in programming courses.
Knowledge-Augmented Multimodal Clinical Rationale Generation for Disease Diagnosis with Small Language Models
Shuai Niu (Hong Kong Baptist University), Xian Yang (University of Manchester)
Explainability and InterpretabilityKnowledge DistillationTransformerLarge Language ModelTextMultimodalityTime SeriesElectronic Health RecordsRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Generate multimodal (text + time-series lab tests) disease diagnosis and interpretable reasoning (clinical rationale) using small language models (SLM) through knowledge-enhanced attention mechanisms and sequential reasoning distillation.
KnowShiftQA: How Robust are RAG Systems when Textbook Knowledge Shifts in K-12 Education?
Tianshi Zheng (Hong Kong University of Science and Technology), Yangqiu Song (University of Tokyo)
RetrievalTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: This paper constructs a K-12 education QA dataset named KNOWSHIFTQA to evaluate the robustness of Retrieval-Augmented Generation (RAG) systems in scenarios involving changes to textbook knowledge.
KokoroChat: A Japanese Psychological Counseling Dialogue Dataset Collected via Role-Playing by Trained Counselors
Zhiyang Qi (University of Electro-Communications), Michimasa Inaba (University of Electro-Communications)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A Japanese counseling dialogue dataset named KokoroChat was constructed by having professional and trainee counselors role-play on a simulation platform, with detailed customer feedback scores provided for each conversation.
KRISTEVA: Close Reading as a Novel Task for Benchmarking Interpretive Reasoning
Peiqi Sui (McGill University), Pramit Chaudhuri (UT Austin)
Explainability and InterpretabilityTransformerLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: This study designs the KRISTEVA benchmark to evaluate the performance of large language models in literary close reading reasoning tasks, constructing and publicly sharing 1331 multiple-choice questions;
KV-Latent: Dimensional-level KV Cache Reduction with Frequency-aware Rotary Positional Embedding
Shi Luohe (Wuhan University), Hai Zhao (Wuhan University)
Computational EfficiencyKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose the KV-Latent framework, which significantly compresses the KV cache and accelerates inference by downsampling the Key/Value dimensions of Transformer attention heads into a latent space; meanwhile, performance is restored through two-stage training (layer-wise distillation + end-to-end fine-tuning).
L4Q: Parameter Efficient Quantization-Aware Fine-Tuning on Large Language Models
Hyesung Jeon (Seoul National University), Jae-Joon Kim (Seoul National University)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed L4Q, a parameter-efficient fine-tuning method that combines quantization-aware training (QAT) with LoRA, achieving fully quantized and memory-efficient LLM fine-tuning.
La Leaderboard: A Large Language Model Leaderboard for Spanish Varieties and Languages of Spain and Latin America
María Grandury (Universidad Politécnica de Madrid), Irune Zubiaga (Universidad del País Vasco)
Large Language ModelTextBenchmark
🎯 What it does: Established a community-driven, open-source LA LEADERBOARD, covering 66 diverse task datasets in Spanish, Catalan, Basque, and Galician, evaluating 50 public LLMs, aiming to promote model development in these languages.
LACA: Improving Cross-lingual Aspect-Based Sentiment Analysis with LLM Data Augmentation
Jakub Šmíd (University of West Bohemia in Pilsen), Pavel Kral (University of West Bohemia in Pilsen)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose the LACA framework, which leverages large language models (LLM) to generate high-quality pseudo-label data without translation tools, thereby improving cross-lingual ABSA performance.
LADM: Long-context Training Data Selection with Attention-based Dependency Measurement for LLMs
Jianghao Chen (Chinese Academy of Sciences), Jiajun Zhang (Chinese Academy of Sciences)
Data-Centric LearningTransformerLarge Language ModelText
🎯 What it does: Propose the LADM framework, which quantifies dependencies between long text segments using attention mechanisms and selects high-quality long-context data for continued pre-training.
LAMB: A Training-Free Method to Enhance the Long-Context Understanding of SSMs via Attention-Guided Token Filtering
Zhifan Ye (Georgia Institute of Technology), Souvik Kundu (Intel Labs)
Contrastive LearningTextBenchmark
🎯 What it does: Propose a training-free attention-guided token filtering method called LAMB to enhance the long-context understanding capability of state-space model (SSM)-based models.
LangMark: A Multilingual Dataset for Automatic Post-Editing
Diego Velazquez (Welocalize), Roger Wechsler (Welocalize)
TransformerLarge Language ModelTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Constructed and publicly released a large-scale English-to-multilingual automatic post-editing (APE) dataset named LangMark, containing over 200,000 entries across seven languages (Brazilian Portuguese, French, German, Italian, Japanese, Russian, Spanish), and conducted an analysis of the dataset.
LangSAMP: Language-Script Aware Multilingual Pretraining
Yihong Liu (LMU Munich), Hinrich Schuetze
Representation LearningTransformerLarge Language ModelText
🎯 What it does: Introduce language and script embeddings in multilingual pre-training, adding them to the Transformer output layer before passing to the MLM head to form language-neutral context representations.
Language Constrained Multimodal Hyper Adapter For Many-to-Many Multimodal Summarization
Nayu Liu (Tiangong University), Bo Lv (University of Chinese Academy of Sciences)
Knowledge DistillationRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Propose a Language-Constrained Multimodal Hypernetwork Adapter (LCMHA) for the many-to-many (M3S) task in multimodal summarization.
Language Fusion for Parameter-Efficient Cross-lingual Transfer
Philipp Borchert (KU Leuven), Jochen De Weerdt (KU Leuven)
Domain AdaptationRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: This paper proposes a parameter-efficient cross-lingual transfer method (FLARE) that integrates source language (typically English) and target language representations within a low-rank adapter (LoRA), enhancing downstream task performance for low-resource languages through lightweight fusion at the bottleneck layer of the attention module.
Language Model Fine-Tuning on Scaled Survey Data for Predicting Distributions of Public Opinions
Joseph Suh (University of California, Berkeley), Serina Chang (University of California, Berkeley)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper fine-tunes public opinion survey data on large language models using forward KL loss to predict answer distributions for different subgroups on multiple-choice questions.
Language Model Probabilities are Not Calibrated in Numeric Contexts
Charles Lovering (Kensho Technologies), Chris Tanner
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This paper constructs multiple-choice templates containing numerical probability information to evaluate the probability calibration of large language models in numerical contexts.
Language Models can Subtly Deceive Without Lying: A Case Study on Strategic Phrasing in Legislation
Atharvan Dogra (Indian Institute of Technology Madras), Balaraman Ravindran (Indian Institute of Technology Madras)
TransformerLarge Language ModelPrompt EngineeringTextFinance Related
🎯 What it does: Established a testing platform for legislative environments to investigate how LLMs can conceal deceptive actions favoring specific companies through subtle wording.
Language Models Grow Less Humanlike beyond Phase Transition
Tatsuya Aoyama (Georgetown University), Ethan Wilcox (Georgetown University)
TransformerLarge Language ModelText
🎯 What it does: Investigate the relationship between the peak of psychological predictive power (PPP) and phase transitions during the pre-training process of language models, using correlation and causal experiments to validate the impact of phase transitions on PPP decline.
Language Models Resist Alignment: Evidence From Data Compression
Jiaming Ji (Peking University), Yaodong Yang (Peking University)
CompressionTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: This paper investigates the elasticity phenomenon exhibited by large language models when they resist further fine-tuning after alignment training, and formally derives its mechanism through compression theory.
Language Models, Graph Searching, and Supervision Adulteration: When More Supervision is Less and How to Make More More
Arvid Frydenlund (University of Toronto)
Explainability and InterpretabilityTransformerGraph
🎯 What it does: Conduct systematic experiments on the minimalist task (path-star task) for star graphs, revealing that traditional decoder-only models cannot learn this task under teacher forcing, but demonstrate that the task is learnable through various techniques (masking, future distribution, auxiliary subtasks, tree topology, query expansion), uncovering the 'supervision degradation'-induced Clever Hans cheating (CHC) phenomenon.
Language-Codec: Bridging Discrete Codec Representations and Speech Language Models
Shengpeng Ji (Zhejiang University), Zhou Zhao (Zhejiang University)
CompressionConvolutional Neural NetworkAuto EncoderGenerative Adversarial NetworkAudio
🎯 What it does: Propose Language-Codec, a discrete audio codec designed for downstream speech language models, reducing the information capacity of the first channel and improving the decoder and discriminator;
LAQuer: Localized Attribution Queries in Content-grounded Generation
Eran Hirsch (Bar-Ilan University), Ido Dagan (Bar-Ilan University)
GenerationExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Proposed the LAQuer task, allowing users to query and locate corresponding source document fragments after selecting segments in the generated text, achieving localized attribution;
Large Language and Protein Assistant for Protein-Protein Interactions Prediction
Peng Zhou (Hunan University), Xiangxiang Zeng (Hunan University)
Drug DiscoveryGraph Neural NetworkTransformerLarge Language ModelContrastive LearningTextGraphBiomedical DataRetrieval-Augmented Generation
🎯 What it does: Propose the LLaPA model for protein-protein interaction (PPI) prediction, integrating protein sequences and PPI network information into a large language model to predict multi-label PPI types and multi-protein complex affinity.
Large Language and Reasoning Models are Shallow Disjunctive Reasoners
Irtaza Khalid (Cardiff University), Steven Schockaert (Cardiff University)
Explainability and InterpretabilityTransformerSupervised Fine-TuningReinforcement LearningTextBenchmarkChain-of-Thought
🎯 What it does: Investigate and compare the performance of large language models (LLM) and large reasoning models (LRM) on spatiotemporal discrete reasoning tasks (STaR), and conduct behavioral analysis of their reasoning mechanisms in multi-path and single-path reasoning scenarios.
Large Language Models are Good Relational Learners
Fang Wu (Stanford University), Jure Leskovec (Stanford University)
Graph Neural NetworkLarge Language ModelPrompt EngineeringTabularBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose Rel-LLM, which integrates graph neural networks with large language models, efficiently processing relational databases under a retrieval-augmented generation framework by leveraging structured graph prompts.
Large Language Models Struggle to Describe the Haystack without Human Help: A Social Science-Inspired Evaluation of Topic Models
Zongxia Li (University of Maryland), Jordan Boyd-Graber (University of Maryland)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This paper compares the performance and user experience of traditional topic models (LDA), two unsupervised LLM-driven topic models (TopicGPT, LLooM), and a human-supervised LLM model (BASS) on two datasets, evaluating their effectiveness in helping researchers understand large text collections and answer cross-document questions. It also proposes an assessment framework inspired by social science methods, including pre-test/post-test evaluations, answer consistency and quality assessments, and analysis of answer preferences.
Large Margin Representation Learning for Robust Cross-lingual Named Entity Recognition
Guangcheng Zhu (Zhejiang University), Junbo Zhao (Zhejiang University)
RecognitionRepresentation LearningTransformerContrastive LearningText
🎯 What it does: Proposes the MARAL framework, combining maximum margin contrastive learning with progressive cross-lingual adaptation techniques to enhance named entity recognition performance for low-resource languages.
LaTIM: Measuring Latent Token-to-Token Interactions in Mamba Models
Hugo Pitorro (Instituto de Telecomunicações), Marcos Vinicius Treviso (Instituto de Telecomunicações)
Explainability and InterpretabilityText
🎯 What it does: Propose the LATIM method to achieve token-to-token level explanation for the Mamba model.
LazyReview: A Dataset for Uncovering Lazy Thinking in NLP Peer Reviews
Sukannya Purkayastha (Technical University of Darmstadt), Iryna Gurevych (Technical University of Darmstadt)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Created and annotated the LAZYREVIEW dataset to investigate and automatically detect lazy thinking in NLP paper reviews, and evaluated its impact on improving review quality
LEANCODE: Understanding Models Better for Code Simplification of Pre-trained Large Language Models
Yan Wang (Central University of Finance and Economics), Yanan Zheng (Yale University)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Propose LEANCODE, a method that simplifies code by leveraging context attention scores, aiming to reduce the training and inference costs of large-scale language models.
Learn to Memorize: Scalable Continual Learning in Semiparametric Models with Mixture-of-Neighbors Induction Memory
Guangyue Peng (Peking University), Houfeng Wang (Peking University)
Computational EfficiencyRepresentation LearningMeta LearningTransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: In semi-parametric language models, the kNN memory is transformed into a learnable Mixture-of-Neighbor Induction Memory (MoNIM) and embedded into the Transformer, achieving a trainable memory module;
Learning Auxiliary Tasks Improves Reference-Free Hallucination Detection in Open-Domain Long-Form Generation
Chengwei Qin (Hong Kong University of Science and Technology), Hao Ma (Hong Kong University of Science and Technology)
GenerationAnomaly DetectionTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This paper systematically studies reference-free hallucination detection in open-domain long text generation, proposing to improve detection performance by refining internal state evaluation and multi-task fine-tuning.
Learning First-Order Logic Rules for Argumentation Mining
Yang Sun (Harbin Institute of Technology), Ruifeng Xu (Harbin Institute of Technology)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Designed and implemented a multi-task argument mining framework based on first-order logic rules (FOL-AM), capable of simultaneously performing argument component classification (ACTC) and argument relation classification (ARC).
Learning Sparsity for Effective and Efficient Music Performance Question Answering
Xingjian Diao (Dartmouth College), Jiang Gui (Dartmouth College)
Computational EfficiencyTransformerVision Language ModelVideoMultimodalityAudio
🎯 What it does: Proposed the Sparsify framework, which addresses the issues of redundancy and high computational costs in music performance multimodal question answering through sparse learning approaches.
Learning to Generate Structured Output with Schema Reinforcement Learning
Yaxi Lu (Tsinghua University), Maosong Sun (Tsinghua University)
GenerationReinforcement Learning from Human FeedbackLarge Language ModelReinforcement LearningPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Proposed SchemaBench benchmark, using approximately 40K diverse JSON schemas to evaluate LLMs' ability in structured output, and developed the Schema Reinforcement Learning (SRL) framework, combining Fine-grained Schema Validator and Thought of Structure (ToS) to improve JSON generation quality.
Learning to Look at the Other Side: A Semantic Probing Study of Word Embeddings in LLMs with Enabled Bidirectional Attention
Zhaoxin Feng (Hong Kong Polytechnic University), Xiaoyi Bao (Hong Kong Polytechnic University)
Representation LearningTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Conducts a semantic probing study on autoregressive large language models (LLMs) with bidirectional attention to investigate whether it can enhance the quality of word meaning representations.
Learning to Reason from Feedback at Test-Time
Yanyang Li (Chinese University of Hong Kong), Liwei Wang (Chinese University of Hong Kong)
OptimizationReinforcement Learning from Human FeedbackText
🎯 What it does: Proposed a paradigm called FTTT that utilizes feedback for optimization during testing, and designed a learnable test-time optimizer named OPTUNE;
Learning to Reason Over Time: Timeline Self-Reflection for Improved Temporal Reasoning in Language Models
Adrián Bazaga (University of Cambridge), Adrià de Gispert (Amazon AGI)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextChain-of-Thought
🎯 What it does: Proposes the Temporal Self-Reflective Prompting (TISER) framework, dividing the LLM reasoning process into four stages (initial reasoning → timeline construction → self-reflection → answer generation), and extending the reasoning chain through test-time scaling; meanwhile, constructs a synthetic dataset containing intermediate reasoning trajectories for model fine-tuning.
Learning to Rewrite: Generalized LLM-Generated Text Detection
Wei Hao (Columbia University), Chengzhi Mao (Rutgers University)
Anomaly DetectionTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Propose the L2R (Learning to Rewrite) framework, which fine-tunes LLMs to generate extensive edits when rewriting human text, while making minimal changes to LLM-generated text, creating a clear distinction in edit distance for detecting LLM-generated text.
Learning Together to Perform Better: Teaching Small-Scale LLMs to Collaborate via Preferential Rationale Tuning
Sohan Patnaik (Media and Data Science Research Lab, Adobe), Balaji Krishnamurthy (Media and Data Science Research Lab, Adobe)
OptimizationKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextChain-of-Thought
🎯 What it does: This paper proposes a framework called COLLATE, which enhances the performance of downstream tasks by enabling a small LLM to internally 'collaborate' and generate and select the most useful reasoning processes without relying on larger models.
LED-Merging: Mitigating Safety-Utility Conflicts in Model Merging with Location-Election-Disjoint
Qianli Ma (Shanghai Jiao Tong University), Jing Shao (Shanghai Jiao Tong University)
Large Language ModelText
🎯 What it does: Proposed the LED-Merging three-stage (Location–Election–Disjoint) model merging framework to address the safety-efficiency trade-off during large language model (LLM) merging.
LegalAgentBench: Evaluating LLM Agents in Legal Domain
Haitao Li (Tsinghua University), Minlie Huang (Central South University)
TransformerLarge Language ModelAgentic AITextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose LegalAgentBench—a Chinese legal domain LLM agent evaluation benchmark containing 17 real legal corpora, 37 external tools, and 300 multi-difficulty tasks;
LegalReasoner: Step-wised Verification-Correction for Legal Judgment Reasoning
Weijie Shi (Hong Kong University of Science and Technology), Yike Guo (Hong Kong University of Science and Technology)
ClassificationTransformerLarge Language ModelTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed LegalReasoner, which first identifies points of contention and decomposes cases, then performs step-by-step reasoning, and detects logical errors in each step through a process validator. Errors are corrected using expert attribution strategies, thereby enhancing the reliability of legal judgment prediction.
Length Controlled Generation for Black-box LLMs
Yuxuan Gu (Harbin Institute of Technology), Tat-Seng Chua (National University of Singapore)
GenerationTransformerLarge Language ModelText
🎯 What it does: Achieve length-controlled generation on black-box LLMs by proposing an iterative sampling framework based on Metropolis-Hastings and importance sampling.
Length-Induced Embedding Collapse in PLM-based Models
Yuqi Zhou (Renmin University of China), Jun Xu (Renmin University of China)
RetrievalRepresentation LearningTransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper investigates the 'length collapse' phenomenon observed in text embeddings based on pre-trained language models when text length increases, and proposes a method called TempScale to alleviate this issue through temperature scaling.
LESA: Learnable LLM Layer Scaling-Up
Yifei Yang (Shanghai Jiao Tong University), Hai Zhao (Central South University)
Computational EfficiencyRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Proposed a learnable LLM layer depth expansion method called LESA, which achieves model depth expansion by predicting intermediate layer parameters between adjacent layers.
Less for More: Enhanced Feedback-aligned Mixed LLMs for Molecule Caption Generation and Fine-Grained NLI Evaluation
Dimitris Gkoumas (Queen Mary University of London), Maria Liakata (Queen Mary University of London)
Drug DiscoveryTransformerLarge Language ModelReinforcement LearningTextBiomedical Data
🎯 What it does: Achieved significant performance improvements in molecular title generation tasks using model fusion and alignment fine-tuning with minimal data.
Less is More: Explainable and Efficient ICD Code Prediction with Clinical Entities
James C. Douglas, Jonathan K. Kummerfeld (University of Sydney)
ClassificationExplainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningTextBiomedical DataElectronic Health Records
🎯 What it does: Proposed a method combining Named Entity Recognition (NER) and Assertion Classification (AC) to filter content relevant to clinical coding, thereby improving the efficiency and interpretability of ICD coding.
Less Mature is More Adaptable for Sentence-level Language Modeling
Abhilasha Sancheti (University of Maryland), Maha Elbayad (FAIR at Meta)
ClassificationRetrievalRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Research on sentence-level models and systematic evaluation of different sentence encoders, pooling strategies, and their impacts on downstream tasks (sentence ranking, sentence-level classification, natural language inference).