Annual Meeting of the Association for Computational Linguistics · 356 papers
Enhancing Dialogue State Tracking Models through LLM-backed User-Agents Simulation
Xingguang Wang (University of Illinois Urbana-Champaign), Cheng Niu (University of Illinois Urbana-Champaign)
CodeData SynthesisData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningAgentic AIPrompt EngineeringText
🎯 What it does: This paper uses GPT-4 to simulate user-agent dialogues, generating synthetic multi-turn dialogues with dialogue state tags, and performs two-stage fine-tuning on LLaMA-2 using this data to improve dialogue state tracking performance.
🎯 What it does: Studied the robustness of retrieval-augmented language models against retrieval noise and proposed an adaptive adversarial training method called RAAT.
Enhancing Reinforcement Learning with Label-Sensitive Reward for Natural Language Understanding
Kuo Liao (Tencent), Chengguo Yin (Tencent)
CodeReinforcement Learning from Human FeedbackSupervised Fine-TuningReinforcement LearningTextChain-of-Thought
🎯 What it does: This paper proposes an improved reinforcement learning framework, RLLR, which enhances the performance of large language models in natural language understanding tasks by utilizing label-sensitive rewards;
🎯 What it does: Built an error-preserving automatic speech recognition system for elementary school English learners, collecting and manually annotating 85 hours of child spoken language data containing 45,004 sentences with error annotations.
ESCoT: Towards Interpretable Emotional Support Dialogue Systems
Tenggan Zhang (Renmin University of China), Qin Jin (Renmin University of China)
CodeGenerationExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningTextSequentialChain-of-Thought
🎯 What it does: Proposed an explainable generation framework named ESCoT for emotion support dialogue systems, and constructed the first emotion support dialogue dataset with chain-of-thought reasoning, ESD-CoT (1,708 dialogues), based on this framework.
EUROPA: A Legal Multilingual Keyphrase Generation Dataset
Olivier Salaün (Universite De Montreal), Philippe Langlais (Universite De Montreal)
CodeGenerationTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Constructed and publicly released the multilingual (24 languages) keyphrase generation dataset EUROPA for EU court judgments, covering cases from 1957 to 2023 with approximately 280,000 instances;
EWEK-QA : Enhanced Web and Efficient Knowledge Graph Retrieval for Citation-based Question Answering Systems
Mohammad Dehghan (University of Waterloo), Mehdi Rezagholizadeh (Huawei Noah's Ark Lab)
CodeRetrievalComputational EfficiencyRepresentation LearningTransformerLarge Language ModelTextGraphRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed a reference-based QA system EWEK-QA that simultaneously leverages web text and knowledge graphs, enhancing answer quality and efficiency through adaptive web retrieval and efficient KG triplet extraction.
EXAMS-V: A Multi-Discipline Multilingual Multimodal Exam Benchmark for Evaluating Vision Language Models
Rocktim Das (Mohamed bin Zayed University of Artificial Intelligence), Preslav Nakov (Mohamed bin Zayed University of Artificial Intelligence)
CodeRecognitionTransformerVision Language ModelImageTextMultimodalityTabularBenchmark
🎯 What it does: Proposed and constructed EXAMS-V, a college-level multi-subject examination benchmark containing 20,932 questions, spanning 20 subjects, 11 languages, and incorporating multimodal information such as images, tables, and charts, designed to evaluate the comprehensive reasoning capabilities of vision-language models.
Expedited Training of Visual Conditioned Language Generation via Redundancy Reduction
Yiren Jian (Dartmouth College), Hongxia Yang (ByteDance Inc)
CodeGenerationTransformerLarge Language ModelVision Language ModelImageVideoText
🎯 What it does: Proposed the EVLGen framework, which employs a frozen ViT and LLM, and achieves single-stage pre-training through vision-language alignment using TomeFormer;
Experiential Co-Learning of Software-Developing Agents
Chen Qian (Tsinghua University), Maosong Sun (Tsinghua University)
CodeAI Code AssistantTransformerLarge Language ModelAgentic AITextGraphRetrieval-Augmented Generation
🎯 What it does: Proposes an Experiential Co-Learning framework enabling software development LLM agents to collect and leverage experiences from past tasks through a two-phase collaboration (instructor + assistant), achieving higher quality and efficiency in code generation when facing new tasks.
Explainability and Hate Speech: Structured Explanations Make Social Media Moderators Faster
Agostina Calabrese (University of Edinburgh), Francesco Barbieri (Snap Inc)
CodeExplainability and InterpretabilityText
🎯 What it does: This study evaluates the impact of different types of explanations (no explanation, generic explanation, structured explanation) on the decision speed of social media content moderators, with a comparison conducted among 25 professional moderators under three experimental settings.
Explicating the Implicit: Argument Detection Beyond Sentence Boundaries
Paul Roit (Bar Ilan University), Ido Dagan (Bar Ilan University)
CodeRecognitionTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Reformulate the cross-sentence semantic argument detection problem as a text entailment task: first extract local arguments using a sentence-level QA-SRL parser, construct concise hypothesis sentences, and then use an NLI model to determine whether the hypothesis is entailed by the document, thereby identifying implicit arguments of predicates that appear across sentences in the document.
Exploiting Intrinsic Multilateral Logical Rules for Weakly Supervised Natural Language Video Localization
Zhe Xu (Xidian University), Cheng Deng (Xidian University)
CodeObject DetectionGraph Neural NetworkTransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningVideoText
🎯 What it does: Propose a framework for weakly supervised natural language video localization, which utilizes relation-guided prompts to generate an intrinsic temporal relation graph (ITRG), and designs multi-party temporal logic rules based on this to constrain model training, thereby improving the logical consistency and accuracy of localization.
Exploring Collaboration Mechanisms for LLM Agents: A Social Psychology View
Jintian Zhang (Zhejiang University), Shumin Deng (National University Of Singapore)
CodeTransformerLarge Language ModelAgentic AITextBenchmark
🎯 What it does: Experimentally and theoretically analyze the cooperative mechanisms in multi-agent LLM societies, constructing multi-agent societies with diverse personalities and thinking patterns and assessing their impact on task performance.
🎯 What it does: Propose a conditional variational mechanism-based pinyin input method model (CV-IME), which enhances candidate diversity and accuracy in low-resource environments by combining continuous and discrete latent variables.
Exploring Precision and Recall to assess the quality and diversity of LLMs
Florian Le Bronnec (Université Paris-Dauphine), Alexandre Allauzen (Université Paris-Dauphine)
CodeGenerationRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: This paper proposes a distributed evaluation framework based on Precision and Recall to measure the quality and diversity of text generated by large language models (LLMs), enabling comprehensive assessment of LLMs without relying on aligned corpora.
CodeTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper systematically evaluates the zero-shot and few-shot performance of large language models in computational argumentation tasks (argument mining and argument generation), and proposes a new conversational counter-argument generation benchmark;
🎯 What it does: Proposed a large-scale English zero-shot stance detection dataset called EZ-STANCE, and converted the stance detection task into a natural language inference task.
CodeTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes and systematically evaluates multiple methods for estimating the factual confidence of large language models, constructing a unified experimental framework and conducting comparisons across various models and datasets.
Faithful Logical Reasoning via Symbolic Chain-of-Thought
Jundong Xu (National University of Singapore), Wynne Hsu (National University of Singapore)
CodeOptimizationExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Proposes a fully LLM-based symbolic chain-of-thought reasoning framework called SymbCoT, which uses hybrid symbolic expressions combining natural language and first-order logic/constraint optimization for translation, planning, solving, and verification;
FanOutQA: A Multi-Hop, Multi-Document Question Answering Benchmark for Large Language Models
Andrew Zhu (University of Pennsylvania), Chris Callison-Burch (University of Pennsylvania)
CodeLarge Language ModelTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes the FanOutQA dataset, specifically designed to evaluate the reasoning capabilities of large models on multi-hop, multi-document, and fan-out questions, and provides three evaluation settings (closed-book, open-book, and given evidence).
🎯 What it does: Significantly accelerate open-domain question answering inference by adding sentence selection in the FiD framework and adopting two-phase training;
Few-shot Transfer Learning for Knowledge Base Question Answering: Fusing Supervised Models with In-Context Learning
Mayur Patidar (TCS Research), Indrajit Bhattacharya (Indian Institute of Technology)
CodeRetrievalDomain AdaptationTransformerLarge Language ModelPrompt EngineeringTextGraphRetrieval-Augmented Generation
🎯 What it does: To address the few-shot transfer learning problem in knowledge base question answering, the FuSIC-KBQA architecture is proposed, combining technologies such as retrievers trained on the source domain, LLM re-ranking, few-shot context learning, and execution feedback to generate logical forms.
Focus on Your Question! Interpreting and Mitigating Toxic CoT Problems in Commonsense Reasoning
Jiachun Li (University of Chinese Academy of Sciences), Jun Zhao (University of Chinese Academy of Sciences)
CodeExplainability and InterpretabilityComputational EfficiencyTransformerTextChain-of-Thought
🎯 What it does: Identified and analyzed the Toxic CoT issue that arises when large language models use Chain-of-Thought prompts, and proposed the RIDERS method to address this problem
Fora: A corpus and framework for the study of facilitated dialogue
Hope Schroeder (Massachusetts Institute of Technology), Jad Kabbara (Massachusetts Institute of Technology)
CodeClassificationRecognitionData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Constructed and made publicly available the Fora corpus, containing 262 transcripts of semi-structured facilitative dialogues hosted by non-profit organizations, with approximately 10,000 utterances manually annotated for shared behaviors and facilitation strategies.
From Moments to Milestones: Incremental Timeline Summarization Leveraging Large Language Models
Qisheng Hu (National University of Singapore), Hwee Tou Ng (National University of Singapore)
CodeGenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Propose an incremental timeline summarization method called LLM-TLS based on large language models, capable of simultaneously handling event and topic timelines;
🎯 What it does: Automatically selects high-quality and diverse instruction-tuning data using gradient information to enhance the performance of large language models (LLMs) in machine translation tasks.
Generalizing Conversational Dense Retrieval via LLM-Cognition Data Augmentation
Haonan Chen (Renmin University of China), Ziliang Zhao (Renmin University of China)
CodeRetrievalData-Centric LearningLarge Language ModelPrompt EngineeringContrastive LearningText
🎯 What it does: This paper proposes CONVAUG, a multi-layer data augmentation framework based on large language models (LLMs), aimed at enhancing the robustness and generalization ability of conversational dense retrieval.
Generative Cross-Modal Retrieval: Memorizing Images in Multimodal Language Models for Retrieval and Beyond
Yongqi Li, Tat-Seng Chua (National University Of Singapore)
CodeRetrievalLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: Enable multimodal large language models (MLLMs) to remember and retrieve images by learning to generate unique identifiers corresponding to images, directly invoking images from model parameters without relying on external retrieval indexes.
GenTranslate: Large Language Models are Generative Multilingual Speech and Machine Translators
Yuchen Hu (Nanyang Technological University), Eng Siong Chng (Nanyang Technological University)
CodeGenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningTextMultimodalityAudio
🎯 What it does: Propose a novel generative translation paradigm, GenTranslate, which leverages a large language model (LLM) to integrate information from N-best translation candidates generated via beam search, producing a higher-quality single translation output.
CodeClassificationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This study explores using large language models (LLMs) to perform 'unfun' editing on existing humorous texts to generate non-humorous versions corresponding to the original humorous texts, thereby constructing a large-scale aligned humorous/non-humorous dataset; and verifies its cross-lingual generalizability on English-Indian code-mixed tweets.
🎯 What it does: This paper constructs an aggregated intrinsic evaluation suite to systematically compare the inference methods (greedy, merge, likelihood, least-tokens, etc.) of four subword tokenizers—BPE, UnigramLM, WordPiece, and SaGe—across different vocabulary sizes, and analyzes their performance in terms of morphology alignment, cognitive interpretability, and information theory metrics.
🎯 What it does: Proposed the GrowOVER dynamic QA and dialogue benchmark, and designed the Retrieval-Interactive LLM (RiLM) framework to achieve training-free knowledge adaptation.
GunStance: Stance Detection for Gun Control and Gun Regulation
Nikesh Gyawali (Kansas State University), Cornelia Caragea (University of Illinois Chicago)
CodeClassificationTransformerLarge Language ModelText
🎯 What it does: This paper constructs a novel stance detection dataset for gun control, named GUNSTANCE, and proposes a hybrid method combining semi-supervised learning with large language models to improve stance identification on gun ban and regulation issues.
Hard Prompts Made Interpretable: Sparse Entropy Regularization for Prompt Tuning with RL
Yunseon Choi (KAIST AI), Kee-Eung Kim (KAIST AI)
CodeClassificationExplainability and InterpretabilityTransformerLarge Language ModelReinforcement LearningPrompt EngineeringTextMultimodality
🎯 What it does: This paper proposes the PIN (Prompts made INterpretable) algorithm, which optimizes discrete hard prompts on black-box pre-trained models through reinforcement learning, generating prompts that achieve high task performance while being human-interpretable.
CodeComputational EfficiencyTransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: A dynamic expert routing framework is proposed in MoE models, which adaptively activates different numbers of experts based on input difficulty to improve computational efficiency and model performance.
Harnessing Toulmin’s theory for zero-shot argument explication
Ankita Gupta (University of Massachusetts Amherst), Brendan O’Connor
CodeExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes the 'Argument explication' task, which involves decomposing natural language arguments into <claim, premise, warrant> triplets, and achieving zero-shot generation by referencing the Toulmin model in prompts.
Having Beer after Prayer? Measuring Cultural Bias in Large Language Models
Tarek Naous (Georgia Institute of Technology), Wei Xu (Georgia Institute of Technology)
CodeExplainability and InterpretabilityTransformerLarge Language ModelTextBenchmark
🎯 What it does: Constructed the CAMeL dataset and used it to evaluate the cultural bias and adaptation capabilities of large language models in Arabic environments.
Hide and Seek in Noise Labels: Noise-Robust Collaborative Active Learning with LLMs-Powered Assistance
Bo Yuan (Zhejiang University), Wei Jiang (Ant Group)
CodeClassificationTransformerLarge Language ModelText
🎯 What it does: Propose the NoiseAL framework, which collaborates small models with large language models for training to address the problem of noisy labels in text classification.
HiRoPE: Length Extrapolation for Code Models Using Hierarchical Position
Kechi Zhang (Peking University), Zhi Jin (Peking University)
CodeAI Code AssistantTransformerLarge Language ModelText
🎯 What it does: Propose HiRoPE, which transforms traditional RoPE into a hierarchical structure. It leverages abstract syntax tree (AST) information of code to split position indices into multi-level representations, enabling long-context reasoning without requiring additional training.
How Good is Zero-Shot MT Evaluation for Low Resource Indian Languages?
Anushka Singh (Nilekani Centre at AI4Bharat), Mitesh Khapra
CodeData SynthesisData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This study constructs an MQM evaluation dataset for low-resource Indian languages (Assamese, Kannada, Maithili, Punjabi) and performs zero-shot meta-evaluation of multiple machine translation evaluation metrics on this dataset, exploring the impacts of related language fine-tuning, underlying model replacement, and synthetic data.
How Johnny Can Persuade LLMs to Jailbreak Them: Rethinking Persuasion to Challenge AI Safety by Humanizing LLMs
Yi Zeng (Virginia Tech), Weiyan Shi (Northeastern University)
CodeSafty and PrivacyAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: By viewing LLMs as entities with human communication capabilities, the study leverages persuasion classification from social sciences to construct a system that rewrites ordinary harmful queries into humanized persuasive adversarial prompts (PAP), investigating their impact on LLM jailbreaking.
🎯 What it does: Propose the HyCoRec framework, which leverages hypergraphs to learn multifaceted preferences from items, entities, words, reviews, and knowledge, alleviating the Matthew effect in dialogue recommendations while improving recommendation and dialogue quality.
🎯 What it does: Propose Hyper-CL, a model that combines hypernetworks with contrastive learning to generate sentence representations tailored to different conditions.
Hypergraph based Understanding for Document Semantic Entity Recognition
Qiwei Li (Wuhan University), Hai Zhao (Shanghai Jiao Tong University)
CodeRecognitionGraph Neural NetworkTransformerVision Language ModelTextMultimodality
🎯 What it does: Propose a document semantic entity recognition method HGA based on hypergraph attention, and construct the HGALayoutLM model based on GraphLayoutLM.
🎯 What it does: Propose a high-dimensional spherical model based on multi-prototype (HMPEAE) for document-level event argument extraction, treating argument-prototype matching as an optimal transport problem;
IBSEN: Director-Actor Agent Collaboration for Controllable and Interactive Drama Script Generation
Senyu Han (Shanghai Jiao Tong University), Kai Yu (Shanghai Jiao Tong University)
CodeGenerationTransformerLarge Language ModelAgentic AIPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: By designing the IBSEN LLM framework with three roles—director, actor, and player—to achieve controllable drama script generation and interactive plot progression.
CodeData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: Constructed a large-scale bilingual (English-Chinese) information extraction instruction corpus named IEPILE, and enhanced the zero-shot generalization performance of LLMs in IE tasks through structured, semantic schema-based instruction generation.
Improving Event Definition Following For Zero-Shot Event Detection
Zefan Cai (University of Wisconsin - Madison), Nanyun Peng (University of California, Los Angeles)
CodeTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Constructed a diverse event definition dataset called DivED, and performed instruction fine-tuning on LLaMA-2-7B to enhance the event definition following capability in zero-shot event detection.
InCharacter: Evaluating Personality Fidelity in Role-Playing Agents through Psychological Interviews
Xintao Wang (Fudan University), Yanghua Xiao (SenseTime)
CodeTransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper proposes the INCHARACTER framework, which uses interview-based psychological assessment to evaluate the personality fidelity of role-playing agents (RPA), and constructs an RPA personality benchmark covering 14 psychological scales;
InstructProtein: Aligning Human and Protein Language via Knowledge Instruction
Zeyuan Wang (Zhejiang University), Huajun Chen (Zhejiang University)
CodeDrug DiscoveryProtein Structure PredictionLarge Language ModelSupervised Fine-TuningTextBiomedical Data
🎯 What it does: Proposed a large language model called InstructProtein, achieving bidirectional generation between human language and protein sequences;
Interpretable User Satisfaction Estimation for Conversational Systems with Large Language Models
Ying-Chun Lin (Purdue University), Jaime Teevan (Microsoft Corporation)
CodeExplainability and InterpretabilityKnowledge DistillationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Built a framework for interpretable user satisfaction estimation (SPUR) based on large language models, achieving satisfaction determination through three steps: supervised extraction, rule aggregation, and satisfaction evaluation.
Intrinsic Task-based Evaluation for Referring Expression Generation
Guanyi Chen (Central China Normal University), Kees Van Deemter
CodeGenerationTextBenchmark
🎯 What it does: This paper supplements traditional scoring methods by introducing two new meta-tasks (judging referential success and rewrite suggestions) for intrinsic task-based human evaluation of Reference Expression Generation (REG) models.
Intuitive or Dependent? Investigating LLMs’ Behavior Style to Conflicting Prompts
Jiahao Ying (Singapore Management University), Yongbin Liu (University of South China)
CodeExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Investigate the behavioral patterns of large language models when facing conflicts between prompts and internal memory, and propose a comprehensive evaluation framework to measure factual robustness and decision-making styles;
Investigating Multi-Hop Factual Shortcuts in Knowledge Editing of Large Language Models
Tianjie Ju (Shanghai Jiao Tong University), Gongshen Liu (Shanghai Jiao Tong University)
CodeExplainability and InterpretabilityTransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: Investigated the phenomenon of large language models leveraging fact shortcuts in multi-hop reasoning and assessed the risks to knowledge editing.
ItD: Large Language Models Can Teach Themselves Induction through Deduction
Wangtao Sun (Institute of Automation Chinese Academy of Sciences), Kang Liu (Institute of Automation Chinese Academy of Sciences)
CodeData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Proposes the Induction through Deduction (ItD) framework, which leverages the reasoning capabilities of large language models (LLMs) to first self-generate inductive data, then fine-tune and decode the model using Naive Bayesian inference, thereby significantly enhancing the inductive performance of LLMs.
Iterative Forward Tuning Boosts In-Context Learning in Language Models
Jiaxi Yang (Shenzhen Institute of Advanced Technology Chinese Academy of Sciences), Yongbin Li (Shenzhen Institute of Advanced Technology Chinese Academy of Sciences)
CodeTransformerLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Designed a two-stage Deep-Thinking framework to enhance in-context learning (ICL) performance in large language models through multi-round forward reasoning and iterative key-value matrix updates.
L-Eval: Instituting Standardized Evaluation for Long Context Language Models
Chenxin An (Fudan University), Xipeng Qiu (Fudan University)
CodeData-Centric LearningTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposes L-Eval, a standardized evaluation framework for long-context language models (LCLMs), comprising 20 subtasks, 508 long documents, and over 2000 human-annotated QA pairs, covering multiple domains, lengths (3k–200k tokens), and task types.
Label Augmentation for Zero-Shot Hierarchical Text Classification
Lorenzo Paletto (University of Turin), Roberto Esposito (University of Turin)
CodeClassificationTransformerLarge Language ModelText
🎯 What it does: This paper proposes HiLA, a label expansion method based on large language models, which enriches the label hierarchy in strict zero-shot hierarchical text classification by adding sub-labels at the deepest level; subsequently, the UP (Upward Score Propagation) algorithm is used for zero-shot hierarchical classification.
Landmark Embedding: A Chunking-Free Embedding Method For Retrieval Augmented Long-Context Large Language Models
Kun Luo (Institute of Automation Chinese Academy of Sciences), Kang Liu (Institute of Automation Chinese Academy of Sciences)
CodeRetrievalComputational EfficiencyRepresentation LearningTransformerLarge Language ModelContrastive LearningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose Landmark Embedding, a chunking-free, position-aware embedding method for language models to retrieve enhanced long-text context.
🎯 What it does: Fine-tune automatic speech recognition (ASR) models for 25 languages and 15 writing systems, and analyze the relationship between orthographic complexity and speech recognition accuracy.
Language Models can Exploit Cross-Task In-context Learning for Data-Scarce Novel Tasks
Anwoy Chatterjee (Indian Institute of Technology Delhi), Tanmoy Chakraborty (Indian Institute of Technology Delhi)
CodeClassificationData-Centric LearningTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Studies how to leverage the self-learning capabilities of large language models (LLM) by utilizing cross-task examples from different tasks to help models complete new, data-scarce tasks without target task examples, and explores further performance improvements through pseudo-label generation.
Peiyi Wang (Peking University), Zhifang Sui (Peking University)
CodeTransformerLarge Language ModelTextBenchmark
🎯 What it does: The study found significant positional information bias when using large language models (e.g., GPT-4) as evaluators, and proposes a calibration framework based on multi-evidence, balanced positions, and human-computer interaction;
Large Language Models as Zero-shot Dialogue State Tracker through Function Calling
Zekun Li (University of California Santa Barbara), Paul A. Crook (Meta)
CodeTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: By reconstructing the task-oriented dialogue state tracking task as function calls, the FNCTOD framework was constructed to achieve zero-shot dialogue state tracking across multiple large language models (LLMs).
Latxa: An Open Language Model and Evaluation Suite for Basque
Julen Etxaniz (University of the Basque Country), Aitor Soroa (University of the Basque Country)
CodeGenerationTransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper develops an open-source large language model series called Latxa for Basque, providing three pre-trained model versions (7B, 13B, 70B), a high-quality Basque corpus with up to 4.3M documents and 4.2B tokens, and four multiple-choice evaluation datasets (EusProficiency, EusReading, EusTrivia, EusExams).
Layer-Condensed KV Cache for Efficient Inference of Large Language Models
Haoyi Wu (ShanghaiTech University), Kewei Tu (ShanghaiTech University)
CodeComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Propose Layer-Condensed KV Cache to improve the Transformer decoder, enabling all layers to use only the KV from the top layer for attention, significantly reducing KV cache memory usage while maintaining model performance through warmup layers.
Learn from Failure: Fine-tuning LLMs with Trial-and-Error Data for Intuitionistic Propositional Logic Proving
Chenyang An (University of California, San Diego), Jingbo Shang (University of California, San Diego)
CodeTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This paper proposes to fine-tune large language models by leveraging information from failed search paths (trial-and-error) to enhance their reverse search and strategy generation capabilities in intuitive propositional logic proofs.
Learn or Recall? Revisiting Incremental Learning with Pre-trained Language Models
Junhao Zheng (South China University of Technology), Qianli Ma (South China University of Technology)
CodeClassificationTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: This paper re-examines the catastrophic forgetting phenomenon of pre-trained language models (PLMs) in incremental learning, systematically evaluates the memory degradation of PLMs and classifiers using probe techniques, and proposes a lightweight SEQ* baseline method;
Learning or Self-aligning? Rethinking Instruction Fine-tuning
Mengjie Ren (Chinese Information Processing Laboratory), Le Sun (Chinese Information Processing Laboratory)
CodeLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Investigate the Instruction Fine-Tuning (IFT) mechanism, proposing a knowledge intervention framework that separates knowledge injection from behavioral norm transfer, and verify through multiple experiments that IFT is primarily self-aligning rather than knowledge injection.
CodeTransformerLarge Language ModelSupervised Fine-TuningTabular
🎯 What it does: Propose an intermediate supervision framework that enables the model to learn both generating SQL-like operations and directly outputting answers by partially executing SQL execution graphs;
Learning to Decode Collaboratively with Multiple Language Models
Zejiang Shen, David Sontag (Massachusetts Institute of Technology)
CodeGenerationTransformerLarge Language ModelAgentic AIText
🎯 What it does: Train a small base language model to learn whether to let a larger or more specialized assistant model generate the next token during decoding, enabling multi-model collaborative generation.
🎯 What it does: Propose a two-stage framework called Learning to Edit (LTE) to efficiently and dynamically inject new knowledge into large language models and retrieve updated information in real-time during inference, avoiding knowledge confusion caused by traditional methods that rely solely on memory;
Constanza Fierro (University of Copenhagen), Mirella Lapata (Google DeepMind)
CodeGenerationRetrievalTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: This study explores enhancing the verifiability and authenticity of long-text answers within the Retrieval-Augmented Generation framework by leveraging a problem-based blueprint model, and implements a complete process for automatically annotating blueprints and citations.
LePaRD: A Large-Scale Dataset of Judicial Citations to Precedent
Robert Mahari (MIT), Alex Pentland
CodeRetrievalTransformerTextBenchmark
🎯 What it does: Constructed LePaRD, a large-scale judicial citation context-precedent paragraph dataset, encompassing approximately 1.8 million target paragraphs and corresponding contexts.
Less is More: Mitigating Multimodal Hallucination from an EOS Decision Perspective
Zihao Yue (Renmin University of China), Qin Jin (Renmin University of China)
CodeGenerationData-Centric LearningSupervised Fine-TuningVision Language ModelMultimodality
🎯 What it does: This paper investigates the root cause of hallucinations in multimodal large models as excessive detail in training data, proposing two methods—Selective EOS Supervision and Scoring EOS Supervision—to enhance the model's EOS decision-making ability by leveraging the mechanism of assessing the completeness of generated text and visual information during EOS prediction, thereby significantly reducing hallucinations.
Leveraging Codebook Knowledge with NLI and ChatGPT for Zero-Shot Political Relation Classification
Yibo Hu (Georgia Institute of Technology), Vito D’Orazio
CodeClassificationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This study proposes two zero-shot political relationship classification methods, leveraging existing codebook knowledge through NLI (ZSP model) and ChatGPT (GPT-3.5/4) to automatically categorize source-target relationships in event encoding.
Leveraging Large Language Models for Learning Complex Legal Concepts through Storytelling
Hang Jiang (Massachusetts Institute of Technology), Jad Kabbara (Massachusetts Institute of Technology)
CodeTransformerLarge Language ModelText
🎯 What it does: Leverage large language models (LLMs) to generate stories about legal concepts and corresponding reading comprehension questions, constructing the LEGALSTORIES dataset, and evaluate the effectiveness of stories in helping non-professional readers learn legal concepts through randomized controlled trials (RCTs).
Leveraging Machine-Generated Rationales to Facilitate Social Meaning Detection in Conversations
Ritam Dutt (Carnegie Mellon University), Carolyn Rose
CodeClassificationRecognitionTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Generate rationales containing intent, assumptions, and implicit information using large language models (LLMs) and use them as text augmentation to identify social meanings in dialogues (emotion recognition and anti-persuasion strategies).
🎯 What it does: Proposes a method to accelerate Minimum Bayes Risk (MBR) decoding through reference aggregation, significantly reducing the computational complexity of pairwise evaluation metrics.
Jingxuan Tu (Brandeis University), James Pustejovsky (Brandeis University)
CodeData-Centric LearningTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper re-annotated the C-STS validation set, discovering that approximately 55% of instances contained annotation errors, and proposed a QA-based automatic error identification and conditional information extraction method; by decomposing C-STS into a two-step process of first generating answers and then evaluating answer similarity, model performance was significantly improved; simultaneously, a method using Typed Feature Structure (TFS) was proposed to construct more semantically interpretable conditions.
Chengyue Wu (University of Hong Kong), Ping Luo (University of Hong Kong)
CodeComputational EfficiencyRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed the Block Expansion method, inserting duplicated Transformer layers into the pre-trained LLaMA model, training only the newly added layers to build LLAMA PRO and LLAMA PRO - INSTRUCT, balancing general and domain-specific capabilities.
LLM-based Rewriting of Inappropriate Argumentation using Reinforcement Learning from Machine Feedback
Timon Ziegenbein (Leibniz University Hannover), Henning Wachsmuth (Leibniz University Hannover)
CodeGenerationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: To address inappropriate arguments in online discussions, this paper proposes a rewriting method based on reinforcement learning: first, an instruction-tuned LLM generates initial rewrites through prompting, and then utilizes existing appropriateness and semantic similarity classifiers as reward functions. The rewriting strategy is trained using PPO to ensure the generated arguments are both more appropriate and retain their original content.
LLM-Rubric: A Multidimensional, Calibrated Approach to Automated Evaluation of Natural Language Texts
Helia Hashemi (Microsoft), Chris Kedzie (Microsoft)
CodeLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: This paper constructs the LLM-RUBRIC framework by first using a large language model (LLM) to ask questions based on a manually designed multi-dimensional evaluation rubric, obtaining the probability distribution for each question, then using a small feedforward network to perform personalized calibration on these distributions, ultimately predicting each human rater's scores across all questions, including overall satisfaction.
CodeTransformerLarge Language ModelReinforcement LearningPrompt EngineeringTextBenchmark
🎯 What it does: Built a multi-agent dynamic game benchmark called LLMARENA based on PettingZoo to evaluate large language models' spatial reasoning, strategic planning, numerical reasoning, risk assessment, communication, opponent modeling, and team collaboration abilities in seven different game environments.
LLMs Learn Task Heuristics from Demonstrations: A Heuristic-Driven Prompting Strategy for Document-Level Event Argument Extraction
Hanzhang Zhou (Nanyang Technological University), Kezhi Mao (Nanyang Technological University)
CodeTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: In the document-level event argument extraction task, a heuristic-based linked analogy prompting strategy (HD-LoA) based on large language models (LLMs) is proposed, significantly improving few-shot reasoning performance by explicitly constructing heuristic examples and chain analogy reasoning.
🎯 What it does: This paper proposes the LogicBench dataset, systematically evaluating the ability of large language models to perform single logical reasoning rules (totaling 25 types, covering propositional logic, first-order logic, and non-monotonic logic) in natural language reasoning, and designs two task formats: BQA and MCQA;
Long Context is Not Long at All: A Prospector of Long-Dependency Data for Large Language Models
Longze Chen (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences), Min Yang (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences)
CodeData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: The ProLong framework filters documents with strong semantic long-range dependencies by measuring long dependency (LDS) on training samples, and uses these high-quality long texts for fine-tuning large language models (LLMs).
LooGLE: Can Long-Context Language Models Understand Long Contexts?
Jiaqi Li (National Key Laboratory of General Artificial Intelligence, BIGAI), Muhan Zhang (National Key Laboratory of General Artificial Intelligence, BIGAI)
CodeLarge Language ModelTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Constructed the LooGLE benchmark for systematic evaluation of large language models' comprehension and reasoning capabilities on ultra-long texts.
LRQuant: Learnable and Robust Post-Training Quantization for Large Language Models
Jiaqi Zhao (Harbin Institute of Technology), Liqiang Nie (Harbin Institute of Technology)
CodeComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Proposes LR Quant — a post-training quantization (PTQ) framework for large language models, leveraging learnable smooth parameters and negative log cosine similarity loss to achieve efficient, robust low-bit quantization.
M^3AV: A Multimodal, Multigenre, and Multipurpose Audio-Visual Academic Lecture Dataset
Zhe Chen (Shanghai JiaoTong University), Yanfeng Wang (Shanghai JiaoTong University)
CodeTransformerLarge Language ModelVideoTextMultimodalityBenchmarkRetrieval-Augmented GenerationAudio
🎯 What it does: Constructed a multimodal, multi-genre, cross-purpose academic lecture video dataset M3AV, providing high-quality manually annotated speech transcripts, slide OCR, and paper text.
MAGE: Machine-generated Text Detection in the Wild
Yafu Li (Zhejiang University), Yue Zhang (Westlake University)
CodeAnomaly DetectionTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Constructed a large-scale multi-domain and multi-model benchmark for detecting machine-generated text (MAGE), and systematically evaluated the performance of various detection methods in real-world scenarios on this benchmark.
Making Long-Context Language Models Better Multi-Hop Reasoners
Yanyang Li (Chinese University of Hong Kong), Liwei Wang (Chinese University of Hong Kong)
CodeExplainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the 'Reasoning with Attributions' method, enabling long-context LMs to provide the source of each step during multi-hop reasoning through chain citation (CoC) or citation (CoQ), thereby improving the handling of noisy contexts and reasoning accuracy.
Masked Thought: Simply Masking Partial Reasoning Steps Can Improve Mathematical Reasoning Learning of Language Models
Changyu Chen (Renmin University of China), Yongbin Li (Alibaba Group)
CodeTransformerLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: Proposed Masked Thought Fine-Tuning (MFT), enhancing large language models' mathematical reasoning ability by randomly masking partial steps in chain-of-thought reasoning.
CodeClassificationRecognitionRepresentation LearningData-Centric LearningTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Built and publicly released the largest, quality-focused open corpus MC² for four minority languages in China (Tibetan, Uyghur, Kazakh, Mongolian), and conducted continuous pre-training on it to obtain two usable models: MC² XLMR-large and MC² Llama-13B.
CodeComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsText
🎯 What it does: Proposed a MEFT (Memory-Efficient Fine-Tuning) framework based on sparse adapters, leveraging FFN activation sparsity and MoE structures. It places large-scale trainable parameters on the CPU and dynamically loads only a small number of parameters related to the input to the GPU during forward/backward propagation, achieving efficient fine-tuning of large-scale adapters.
MentalManip: A Dataset For Fine-grained Analysis of Mental Manipulation in Conversations
Yuxin Wang (Dartmouth College), Soroush Vosoughi (Dartmouth College)
CodeClassificationTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper creates and publicly releases the MENTALMANIP dataset, which contains 4,000 multi-turn fictional dialogues, for fine-grained detection of psychological manipulation.
MindMap: Knowledge Graph Prompting Sparks Graph of Thoughts in Large Language Models
Yilin Wen (University of Illinois Urbana Champaign), Jimeng Sun (University of Illinois Urbana Champaign)
CodeExplainability and InterpretabilityLarge Language ModelPrompt EngineeringGraphBiomedical DataRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes the MindMap framework, which utilizes knowledge graphs (KG) as an external knowledge source, enabling large language models (LLM) to understand graph structures and generate answers along with their 'mind maps' through schematic reasoning, thereby enhancing transparency and accuracy.