EMNLP 2024 Papers — Page 7
Conference on Empirical Methods in Natural Language Processing · 1268 papers
Is Child-Directed Speech Effective Training Data for Language Models?
Steven Y. Feng (Stanford University), Michael C. Frank (Stanford University)
Representation LearningData-Centric LearningTransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper investigates the effectiveness of child-directed speech (CHILDES) in language model training by pretraining on different corpora with GPT-2 and RoBERTa, and compares it with synthetic dialogue data TinyDialogues, as well as benchmark datasets such as OpenSubtitles, Wikipedia, and BabyLM; a 'controlled rearing' experiment is conducted to examine the impact of global and local corpus ordering on model performance.
Is It Good Data for Multilingual Instruction Tuning or Just Bad Multilingual Evaluation for Large Language Models?
Pinzhen Chen (University of Edinburgh), Barry Haddow (University of Edinburgh)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Investigated the differences between localized data and machine-translated data in multilingual instruction tuning and evaluation, assessing their impact on the performance of large language models.
Is It Really Long Context if All You Need Is Retrieval? Towards Genuinely Difficult Long Context NLP
Omer Goldman (Barilan University), Reut Tsarfaty (Barilan University)
ClassificationTextBenchmark
🎯 What it does: This paper proposes and argues for a long-context task classification method based on the distribution characteristics of task information, emphasizing the insufficiency of using text length alone to measure task difficulty. Through evaluation and visualization of existing long-context benchmarks, it highlights the under-explored challenges in the dual dimensions of information dispersion and information scope.
Is LLM-as-a-Judge Robust? Investigating Universal Adversarial Attacks on Zero-shot LLM Assessment
Vyas Raina (University of Cambridge), Mark Gales (University of Cambridge)
Adversarial AttackTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Investigated the robustness of zero-shot LLM evaluation methods against general adversarial attacks, proposed a generic attack statement based on simple concatenation, and verified its effectiveness across multiple LLMs (FlanT5-xl, Llama2-7B, Mistral-7B, ChatGPT).
Is Safer Better? The Impact of Guardrails on the Argumentative Strength of LLMs in Hate Speech Countering
Helena Bonaldi (Fondazione Bruno Kessler), Marco Guerini (Fondazione Bruno Kessler)
Safty and PrivacyTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper investigates the impact of safety guardrails on the argument strength of large language models (LLMs) when generating counter-speech, and compares the effects of attacking different components of hate speech (implications, dehumanizing elements, weakest points) with overall attacks.
Is This a Bad Table? A Closer Look at the Evaluation of Table Generation from Text
Pritika Ramu (Adobe Research), Sambaran Bandyopadhyay (Adobe Research)
GenerationTransformerLarge Language ModelTextTabularChain-of-Thought
🎯 What it does: Proposed the TABEVAL evaluation framework, which utilizes LLM to expand tables into natural language atomic statements, then evaluates the semantic quality of text-to-table generation through natural language reasoning, and created a new multi-domain table dataset DESCTOTTO based on this.
Is this the real life? Is this just fantasy? The Misleading Success of Simulating Social Interactions With LLMs
Xuhui Zhou, Maarten Sap
TransformerLarge Language ModelSupervised Fine-TuningAgentic AIText
🎯 What it does: Compare two social interaction simulation modes based on large language models (SCRIPT and AGENTS), evaluate the impact of information asymmetry on model goal completion rates and naturalness, and explore the effect of fine-tuning on AGENTS mode performance using dialog data generated in SCRIPT mode.
Jailbreaking LLMs with Arabic Transliteration and Arabizi
Mansour Al Ghanim (University of Central Florida), Qian Lou (University of Central Florida)
Safty and PrivacyAdversarial AttackPrompt EngineeringText
🎯 What it does: Conduct Jailbreak attack experiments on non-standard Arabic scripts (romanization and chat-style), and manually evaluate the LLM's response to harmful prompts.
Jellyfish: Instruction-Tuning Local Large Language Models for Data Preprocessing
Haochen Zhang (Osaka University), Masafumi Oyamada (NEC Corporation)
Explainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextTabularElectronic Health Records
🎯 What it does: This paper constructs a generic data preprocessing (DP) solution called Jellyfish, which can run on a single GPU, by fine-tuning a local large language model (LLM) with instructions;
Joint Pre-Encoding Representation and Structure Embedding for Efficient and Low-Resource Knowledge Graph Completion
Chenyu Qiu (Jiangnan University), Eddie-Yin-Kwee Ng (Nanyang Technological University)
Computational EfficiencyRepresentation LearningTransformerLarge Language ModelContrastive LearningTextGraphBenchmark
🎯 What it does: Proposes a knowledge graph completion method based on Pre-Encoded Masked Language Model (PEMLM), and improves the prediction effectiveness of 1-N relations through structural embedding fusion.
Jump Starting Bandits with LLM-Generated Prior Knowledge
Parand A. Alamdari (University of Toronto), Kevin H. Wilson (Borealis AI)
TransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringText
🎯 What it does: Proposed and validated a method that uses synthetic user preference data generated by large language models (LLMs) to pretrain contextual bandits (CBLI), significantly reducing cumulative regret during the early stages of online learning.
KARL: Knowledge-Aware Retrieval and Representations aid Retention and Learning in Students
Matthew Shu (Yale University), Jordan Lee Boyd-Graber (University of Maryland)
RetrievalRepresentation LearningData-Centric LearningTransformerLarge Language ModelTextSequentialRetrieval-Augmented Generation
🎯 What it does: Proposed KAR-L3, a content-aware flashcard scheduler that uses deep knowledge tracing, retrieval, and BERT to predict student memory.
KB-Plugin: A Plug-and-play Framework for Large Language Models to Induce Programs over Low-resourced Knowledge Bases
Jiajie Zhang (Tsinghua University), Juanzi Li (Tsinghua University)
Data-Centric LearningAI Code AssistantTransformerLarge Language ModelTextGraph
🎯 What it does: Propose the KB-Plugin framework, enabling large language models to perform program induction on low-resource knowledge bases through pluggable modules.
KidLM: Advancing Language Models for Children – Early Insights and Future Directions
Mir Tafseer Nayeem (University of Alberta), Davood Rafiei (University of Alberta)
Safty and PrivacyData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: This study develops a child language model called KidLM, focusing on constructing child-friendly datasets and improving pre-training methods;
Kiss up, Kick down: Exploring Behavioral Changes in Multi-modal Large Language Models with Assigned Visual Personas
Seungjong Sun (Sungkyunkwan University), Jang Hyun Kim (Sungkyunkwan University)
Explainability and InterpretabilityTransformerLarge Language ModelVision Language ModelDiffusion modelImageMultimodality
🎯 What it does: This paper constructs a dataset of 5,185 fictional avatar images and uses these images as visual personalized identifiers for multi-modal large language models (LLMs). Subsequently, the behavior changes of LLMs are evaluated in an ultimatum game, particularly focusing on how the models perceive and respond to aggression displayed in their own and opponents' avatars.
KNN-Instruct: Automatic Instruction Construction with K Nearest Neighbor Deduction
Jianshang Kou (University of Science and Technology of China), Zhendong Mao (University of Science and Technology of China)
Data SynthesisData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposes a method called KNN‑INSTRUCT for automatically constructing large-scale SFT datasets, utilizing KNN reasoning to generate high-quality, scalable single-turn dialogues.
Knowledge Conflicts for LLMs: A Survey
Rongwu Xu (Tsinghua University), Wei Xu (Tsinghua University)
TransformerLarge Language ModelTextReview/Survey Paper
🎯 What it does: Reviews knowledge conflicts arising in large language models when integrating contextual knowledge and parameter knowledge, identifies three types of conflicts (context-memory, inter-context, intra-memory), and analyzes their causes, model behaviors, and existing mitigation strategies.
Knowledge Graph Enhanced Large Language Model Editing
Mengqi Zhang (Shandong University), Zhumin Chen (Shandong University)
Graph Neural NetworkTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: Proposes the GLAME method, which enhances the editing process of large language models using a knowledge graph
Knowledge Planning in Large Language Models for Domain-Aligned Counseling Summarization
Aseem Srivastava, Md Shad Akhtar
GenerationRecurrent Neural NetworkGraph Neural NetworkTransformerLarge Language ModelTextBiomedical DataBenchmark
🎯 What it does: A PIECE framework is proposed for mental health counseling dialogues, generating more professionally compliant counseling summaries through a planning-then-generation approach, utilizing knowledge filtering, knowledge scaffolding, and structured knowledge (sheaf learning).
Knowledge Verification to Nip Hallucination in the Bud
Fanqi Wan (Sun Yat-sen University), Shuming Shi (Tencent)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper studies the hallucination problem caused by inconsistencies between external knowledge and pre-training knowledge in LLM alignment data, proposing a method based on Knowledge Consistency Alignment (KCA).
Knowledge-Centric Hallucination Detection
Xiangkun Hu (Amazon AWS AI), Zheng Zhang (Amazon AWS AI)
ClassificationTransformerLarge Language ModelTextBenchmark
🎯 What it does: Built the REfCHECKER framework, which uses knowledge triplets to extract and verify assertions from LLM-generated text, detecting hallucinations.
KnowledgeSG: Privacy-Preserving Synthetic Text Generation with Knowledge Distillation from Server
Wenhao Wang (Zhejiang University), Yanfeng Wang (Shanghai Jiao Tong University)
Data SynthesisFederated LearningSafty and PrivacyKnowledge DistillationTextFinance Related
🎯 What it does: Propose a client-server framework KnowledgeSG that generates high-quality synthetic text by locally fine-tuning models with differential privacy and performing knowledge distillation through server-side specialized models;
KnowTuning: Knowledge-aware Fine-tuning for Large Language Models
Yougang Lyu (Shandong University), Zhaochun Ren (Leiden University)
Data-Centric LearningReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningTextBiomedical Data
🎯 What it does: Knowledge-aware fine-tuning of large language models through a two-stage approach combining fine-grained knowledge enhancement and coarse-grained knowledge comparison, improving completeness, factual accuracy, and logical consistency in knowledge-intensive question answering.
Label Confidence Weighted Learning for Target-level Sentence Simplification
Xin Ying Qiu, Jingshen Zhang (Guangdong University of Foreign Studies)
GenerationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed Label Confidence Weighted Learning (LCWL), which leverages weakly supervised pseudo labels and incorporates label confidence weighting into the encoder-decoder loss to address label noise in multi-level sentence simplification.
Ladder: A Model-Agnostic Framework Boosting LLM-based Machine Translation to the Next Level
Zhaopeng Feng (Zhejiang University), Zuozhu Liu (Zhejiang University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose MT-Ladder, a model-agnostic, cost-effective method that leverages intermediate translations generated by existing LLMs and reference translations to construct pseudo-refinement triplets, performing instruction-based fine-tuning of LLMs to significantly improve translation quality.
Language Concept Erasure for Language-invariant Dense Retrieval
Zhiqi Huang (Capital One), James Allan (University of Massachusetts Amherst)
RetrievalTransformerContrastive LearningText
🎯 What it does: Through the multi-task training framework LANCER, language concept elimination techniques are employed to reduce language-related signals in multilingual dense retrieval models, achieving language-agnostic retrieval performance.
Language is Scary when Over-Analyzed: Unpacking Implied Misogynistic Reasoning with Argumentation Theory-Driven Prompts
Arianna Muti (Università di Bologna), Tommaso Caselli (University of Groningen)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: This paper treats the detection of implicit misogyny as an argumentation reasoning task, utilizing large language models to generate implicit premises (warrants) of messages, thereby determining whether the text contains misogynistic tendencies.
Language models and brains align due to more than next-word prediction and word-level information
Gabriele Merlin (MPI for Software Systems Saarbrücken), Mariya Toneva (MPI for Software Systems Saarbrücken)
TransformerLarge Language ModelSupervised Fine-TuningTextBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Two perturbations (input word order shuffling and fine-tuning with specific stimulus text) were applied to the pre-trained GPT-2 language model, and multi-level comparative experiments were constructed to disentangle the contributions of next-word prediction, word-level information, and multi-word information to brain-model alignment.
Language Models as Compilers: Simulating Pseudocode Execution Improves Algorithmic Reasoning in Language Models
Hyungjoo Chae (Yonsei University), Jinyoung Yeo (Yonsei University)
AI Code AssistantTransformerLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: This paper proposes the THINK-AND-EXECUTE framework, which first uses an LLM to discover task-agnostic logic and write it as pseudocode, then simulates the execution of the pseudocode on each instance to obtain the answer.
Language Models Learn Rare Phenomena from Less Rare Phenomena: The Case of the Missing AANNs
Kanishka Misra (University of Texas at Austin), Kyle Mahowald (University of Texas at Austin)
Representation LearningTransformerLarge Language ModelText
🎯 What it does: Studies how language models learn extremely rare English grammatical structures (AANN: Article+Adjective+Numeral+Noun) under limited data, and explores the potential mechanisms by which they induce such structures from more common related constructions.
Language-to-Code Translation with a Single Labeled Example
Kaj Bostrom (University of Texas at Austin), Jacob Andreas (Microsoft)
AI Code AssistantLarge Language ModelText
🎯 What it does: Propose a semi-supervised framework named ICIP (In-Context Inverse Programming), which leverages large language models (LLM) to automatically generate natural language commands and iteratively optimize them with only a small number of labeled or zero labeled examples, thereby constructing a language-to-code translator;
Large Language Model as an Assignment Evaluator: Insights, Feedback, and Challenges in a 1000+ Student Course
Cheng-Han Chiang (National Taiwan University), Hung-yi Lee (National Taiwan University)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This study first deployed GPT-4 as an automatic homework evaluation tool (LLM TA) in a generative AI course with over 1,000 students, and collected user experience, acceptance, and cases of errors and prompt-hacking through student questionnaires;
Large Language Models Are Involuntary Truth-Tellers: Exploiting Fallacy Failure for Jailbreak Attacks
Yue Zhou (University of Illinois Chicago), Yang Zhang (MIT-IBM Watson AI Lab IBM Research)
Safty and PrivacyAdversarial AttackTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: The paper investigates the shortcomings of LLMs when generating fallacious reasoning and proposes a 'Fallacy Failure Attack (FFA)' to bypass LLM security barriers without requiring internal model information.
Large Language Models Are Poor Clinical Decision-Makers: A Comprehensive Benchmark
Fenglin Liu (University of Oxford), David A. Clifton (University of Oxford)
Drug DiscoveryTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBiomedical DataElectronic Health RecordsBenchmark
🎯 What it does: Construct the ClinicBench benchmark, covering 3 scenarios (reasoning, generation, understanding), 11 tasks, 17 datasets (including 6 newly constructed clinical datasets), and evaluate the performance of 22 LLMs (7B-70B) under zero-shot and few-shot conditions.
Large Language Models as Foundations for Next-Gen Dense Retrieval: A Comprehensive Empirical Assessment
Kun Luo (Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences), Kang Liu (Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences)
RetrievalTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: Compare LLMs with traditional pre-trained models as dense retrieval backbone encoders, systematically evaluating their performance on six key dimensions (in-domain accuracy, data efficiency, zero-shot generalization, long retrieval, instruction retrieval, multi-task learning).
Large Language Models Can Be Contextual Privacy Protection Learners
Yijia Xiao (University of California, Los Angeles), Wei Cheng (NEC Laboratories America)
Safty and PrivacyData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBiomedical Data
🎯 What it does: Propose Contextual Privacy Protection Language Models (CPPLM), which fine-tune large language models using multiple methods to protect context-sensitive PII while injecting domain knowledge.
Large Language Models Can Self-Correct with Key Condition Verification
Zhenyu Wu (Xi'an Jiaotong University), Meng Jiang (University of Notre Dame)
TransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Propose a zero-shot prompting method called PROCO, which enables large language models to self-correct reasoning errors without relying on external feedback through an iterative verify-then-correct process.
Large Language Models for Data Annotation and Synthesis: A Survey
Zhen Tan (Arizona State University), Huan Liu (Arizona State University)
Data SynthesisTransformerPrompt EngineeringMixture of ExpertsTextReview/Survey PaperRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: This paper reviews the application of large language models (LLM) in data annotation and synthesis, proposing a comprehensive framework covering annotation generation, evaluation, and utilization, and systematically summarizes methods, techniques, challenges, and social impacts.
Large Language Models Know What is Key Visual Entity: An LLM-assisted Multimodal Retrieval for VQA
Pu Jian (Institute of Automation, Chinese Academy of Sciences), Jiajun Zhang (Institute of Automation, Chinese Academy of Sciences)
RetrievalTransformerLarge Language ModelVision Language ModelMultimodalityRetrieval-Augmented Generation
🎯 What it does: By leveraging the reasoning capabilities of large language models (LLMs) to extract key visual entities and independently encoding these entities for multimodal retrieval, thereby enhancing the retrieval-augmented performance of visual question answering (VQA).
Latent Concept-based Explanation of NLP Models
Xuemin Yu (Dalhousie University), Hassan Sajjad (Dalhousie University)
Explainability and InterpretabilityLarge Language ModelText
🎯 What it does: Propose a local explanation method called LACOAT based on latent concepts from training data to interpret predictions of deep NLP models.
LawBench: Benchmarking Legal Knowledge of Large Language Models
Zhiwei Fei (Nanjing University), Vincent Ng (University of Texas at Dallas)
Large Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Developed the LawBench benchmark, which includes 20 Chinese civil law system-related tasks for evaluating LLMs' legal knowledge.
Layer by Layer: Uncovering Where Multi-Task Learning Happens in Instruction-Tuned Large Language Models
Zheng Zhao (University of Edinburgh), Shay B. Cohen (University of Edinburgh)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: The study investigates how pre-trained and instruction-tuned large language models encode task-related information at different layers, revealing the functions of three types of layers (shared layers, transitional layers, and refined layers) through matrix analysis methods.
Leading Whitespaces of Language Models’ Subword Vocabulary Pose a Confound for Calculating Word Probabilities
Byung-Doh Oh (New York University), William Schuler (Ohio State University)
Representation LearningTransformerLarge Language ModelText
🎯 What it does: This paper points out that subword tokenization-based language models may encounter issues where the sum of word probability distributions exceeds 1 during word probability calculation, due to leading spaces in subword vocabularies, leading to incorrect allocation of word-level surprisal values; it proposes a 'post-spacing decoding (WT)' method that reassigns the probability of trailing spaces to the preceding word, restoring consistency in word probabilities.
Learn Beyond The Answer: Training Language Models with Reflection for Mathematical Reasoning
Zhihan Zhang (University of Notre Dame), Meng Jiang (Tencent AI Lab)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Proposes the RefAug technique, appending reflection segments to each math problem in the training data to enhance the model's deep understanding and subsequent reasoning.
Learn to Refuse: Making Large Language Models More Controllable and Reliable through Knowledge Scope Limitation and Refusal Mechanism
Lang Cao (University of Illinois Urbana Champaign)
Safty and PrivacyTransformerPrompt EngineeringTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed the L2R (Learn to Refuse) system, which leverages an independent structured knowledge base and refusal mechanisms to enable large language models to proactively refuse when they cannot answer questions, thereby enhancing answer reliability.
Learning from Natural Language Explanations for Generalizable Entity Matching
Somin Wadhwa (Northeastern University), Luyang Kong
Computational EfficiencyKnowledge DistillationTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Reframe the entity matching task as a conditional generation task, using Chain-of-Thought (CoT) natural language explanations generated by large language models (LLMs) to distill a small Seq2Seq model, achieving efficient and generalizable entity matching.
Learning Interpretable Legal Case Retrieval via Knowledge-Guided Case Reformulation
Chenlong Deng (Renmin University of China), Zhicheng Dou (Renmin University of China)
RetrievalExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringContrastive LearningTextRetrieval-Augmented Generation
🎯 What it does: Built a case rewriting system called KELLER guided by legal knowledge, which leverages large language models (LLMs) to extract case sub-facts with the aid of professional legal knowledge and achieves retrieval through cross-matching of sub-facts.
Learning Personalized Alignment for Evaluating Open-ended Text Generation
Danqing Wang (Carnegie Mellon University), Yuandong Tian (Meta AI)
Explainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Designed and trained PERSE — a personalized evaluation framework based on LLMs, used to measure the consistency between generated text and individual preferences, and provide interpretable scores and explanations.
Learning Planning-based Reasoning by Trajectories Collection and Process Reward Synthesizing
Fangkai Jiao (Nanyang Technological University), Shafiq Joty (Salesforce Research)
OptimizationReinforcement Learning from Human FeedbackLarge Language ModelReinforcement LearningTextChain-of-Thought
🎯 What it does: Propose a framework for offline trajectory collection and process reward synthesis, leveraging Direct Preference Optimization (DPO) to learn planning-based reasoning;
Learning to Correct for QA Reasoning with Black-box LLMs
Jaehyung Kim (Yonsei University), Yiming Yang (Carnegie Mellon University)
Computational EfficiencyData-Centric LearningTransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: Propose a COBB framework that maps incomplete reasoning from a black-box LLM to correct and more complete reasoning by training a small adapter model.
Learning to Extract Structured Entities Using Language Models
Haolun Wu (McGill University), Bhaskar Mitra (Microsoft Research)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes an entity-centric framework for structured entity extraction (SEE), designs a new evaluation metric called AESOP, and subsequently introduces the MuSEE multi-stage language model to accomplish the task.
Learning to Rank Salient Content for Query-focused Summarization
Sajad Sotudeh (Georgetown University), Nazli Goharian (Georgetown University)
GenerationTransformerLarge Language ModelText
🎯 What it does: Propose a model that combines learning to rank (LTR) with long-text question-answering focused summarization, utilizing a shared decoder to rank source text segments by importance during summary generation, thereby enhancing the relevance and authenticity of the summary.
Learning to Retrieve Iteratively for In-Context Learning
Yunmo Chen (Microsoft), Benjamin Van Durme (Microsoft)
RetrievalRecurrent Neural NetworkLarge Language ModelReinforcement LearningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposed an iterative retrieval framework, which trains the retriever to select example sets in multiple steps through reinforcement learning, thereby enhancing the context learning effectiveness of large language models in few-shot semantic parsing tasks.
Learning to Write Rationally: How Information Is Distributed in Non-native Speakers’ Essays
Zixin Tang (Pennsylvania State University), Janet G. van Hell (Pennsylvania State University)
Data-Centric LearningTransformerLarge Language ModelText
🎯 What it does: This study investigates the distribution patterns of textual information in non-native English writing, quantifying the writing characteristics of L2 authors using information theory metrics (surprise, entropy, unified information density)
Leave No Document Behind: Benchmarking Long-Context LLMs with Extended Multi-Doc QA
Minzheng Wang (University of Chinese Academy of Sciences), Yongbin Li (Alibaba Group)
TransformerLarge Language ModelTextBenchmarkFinance RelatedRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed the Loong benchmark to evaluate the long-context understanding capabilities of large language models through multi-document long-context question answering.
LEMoE: Advanced Mixture of Experts Adaptor for Lifelong Model Editing of Large Language Models
Renzhi Wang (Nanjing University of Aeronautics and Astronautics), Piji Li (Nanjing University of Aeronautics and Astronautics)
Computational EfficiencyMeta LearningTransformerMixture of ExpertsText
🎯 What it does: This paper proposes LEMoE, an advanced Mixture of Experts (MoE) adapter designed for lifelong model editing, achieving continuous knowledge updates for large language models (LLMs) through module insertion, KV Anchor routing, and clustering-based order planning;
Less is More: Parameter-Efficient Selection of Intermediate Tasks for Transfer Learning
David Schulte (Humboldt University of Berlin), Alan Akbik (Humboldt University of Berlin)
Domain AdaptationComputational EfficiencyTransformerText
🎯 What it does: Proposes the ESM-LogME method combining Embedding Space Maps (ESM) with LogME for efficiently selecting intermediate tasks.
Let Me Teach You: Pedagogical Foundations of Feedback for Language Models
Beatriz Borges (EPFL), Antoine Bosselut (EPFL)
Review/Survey Paper
🎯 What it does: Reviewed feedback theory in the field of learning science, proposed the Feedback Ecosystem Framework for Large Language Models (FELT), and designed a ten-dimensional classification system for natural language feedback content based on this framework.
Let the Expert Stick to His Last: Expert-Specialized Fine-Tuning for Sparse Architectural Large Language Models
Zihan Wang (DeepSeek AI), Yu Wu (DeepSeek AI)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsText
🎯 What it does: This paper proposes ESFT, a parameter-efficient fine-tuning method for sparse Mixture-of-Experts (MoE) large language models, which can fine-tune only the experts most relevant to downstream tasks, maintaining expert specialization while significantly reducing computational costs.
Let’s discuss! Quality Dimensions and Annotated Datasets for Computational Argument Quality Assessment
Rositsa V Ivanova (University of St. Gallen), Christina Niklaus (University of St. Gallen)
Data-Centric LearningTextReview/Survey Paper
🎯 What it does: Systematically reviewed the research progress in computational argument quality assessment (AQ) over the past two decades, constructed a classification system of quality dimensions based on existing literature, and conducted a comprehensive compilation and detailed evaluation of 57 related datasets;
Leveraging BERT and TFIDF Features for Short Text Clustering via Alignment-Promoting Co-Training
Zetong Li (Sun Yat-sen University), Jianxing Yu (Sun Yat-sen University)
Representation LearningTransformerAuto EncoderContrastive LearningText
🎯 What it does: This paper proposes a short text clustering framework called COTC, which jointly utilizes BERT's deep semantic features and TFIDF keyword information through dual-module collaborative training;
Leveraging Conflicts in Social Media Posts: Unintended Offense Dataset
Che-Wei Tsai (National Tsing Hua University), Yi-Shin Chen (National Tsing Hua University)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Propose an inter-conflict-based data collection framework, leveraging 'unintended offense' response cues (e.g., 'didn’t mean to offend you') from Twitter dialogues to extract context-rich offensive content, constructing the Unintended Offense Dataset;
Leveraging Context-Aware Prompting for Commit Message Generation
Zhihua Jiang (Jinan University), Guanghui Ye (Hunan University)
GenerationAI Code AssistantGraph Neural NetworkTransformerLarge Language ModelPrompt EngineeringTextGraph
🎯 What it does: This paper proposes a context-aware prompting model called COMMIT for generating GitHub commit messages.
Leveraging Estimated Transferability Over Human Intuition for Model Selection in Text Ranking
Jun Bai (Beihang University), Wenge Rong (Beihang University)
RetrievalRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Propose a text ranking model selection method called AiRTran based on pre-trained models, which calculates the expected ranking using sentence embeddings from pre-trained models as a transferability metric, and improves the accuracy of this metric through adaptive isotropic normalization (AdaIso).
Leveraging Large Language Models for NLG Evaluation: Advances and Challenges
Zhen Li (Peking University), Shuai Ma (Beihang University)
TransformerLarge Language ModelPrompt EngineeringTextReview/Survey PaperBenchmarkChain-of-Thought
🎯 What it does: Reviewed and systematized the applications and methods of large language models in natural language generation evaluation
Leveraging pre-trained language models for linguistic analysis: A case of argument structure constructions
Hakyung Sung (University of Oregon), Kristopher Kyle (University of Oregon)
RecognitionGenerationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This study evaluates the effectiveness of pre-trained language models in identifying argument structure constructions (ASC), comparing three methods: RoBERTa supervised training, GPT-4 prompting annotation, and GPT-4 generating training data followed by RoBERTa training.
Lexically Grounded Subword Segmentation
Jindřich Libovický (Charles University), Jindřich Helcl (Charles University)
Knowledge DistillationRepresentation LearningText
🎯 What it does: This paper proposes three improved subword segmentation schemes aimed at enhancing the segmentation's ability to capture morphology, thereby improving the performance of downstream NLP tasks.
Liar, Liar, Logical Mire: A Benchmark for Suppositional Reasoning in Large Language Models
Philipp Mondorf (MaiNLP, Center for Information and Language Processing, LMU Munich), Barbara Plank (MaiNLP, Center for Information and Language Processing, LMU Munich)
Large Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Proposed the TruthQuest benchmark, using knight-and-knave logic puzzles to evaluate the hypothetical reasoning ability of LLMs.
Lifelong Event Detection via Optimal Transport
Viet Dao (Vinai Research), Thien Huu Nguyen (Vinai Research)
ClassificationKnowledge DistillationRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Developed a continual event detection framework called LEDOT that combines optimal transport (OT) with memory replay, leveraging the word distribution of the pre-trained language model (PLM) head to align classifier outputs, thereby reducing catastrophic forgetting.
Lifelong Knowledge Editing for LLMs with Retrieval-Augmented Continuous Prompt Learning
Qizhou Chen (East China Normal University), Hui Xue’
Computational EfficiencyRepresentation LearningTransformerLarge Language ModelPrompt EngineeringContrastive LearningTextRetrieval-Augmented Generation
🎯 What it does: Propose the RECIPE framework, combining retrieval-enhanced continuous prompt learning to achieve lifelong knowledge editing in LLMs;
Linear Layer Extrapolation for Fine-Grained Emotion Classification
Mayukh Sharma (University of California San Diego), Julian McAuley (University of California San Diego)
ClassificationTransformerContrastive LearningText
🎯 What it does: Propose a dynamic contrast weight (β) selection method based on linear extrapolation to improve inter-layer contrast reasoning in Transformer for fine-grained sentiment classification.
Linguistic Bias in ChatGPT: Language Models Reinforce Dialect Discrimination
Eve Fleisig (University of California, Berkeley), Dan Klein (University of California, Berkeley)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper conducts a large-scale evaluation of ChatGPT (GPT-3.5 Turbo and GPT-4) across ten English dialects (standard American, standard British, and eight minority dialects), combining linguistic feature annotations and native speaker subjective evaluations to investigate whether the model exhibits biases or potential harms toward non-standard dialects.
Link, Synthesize, Retrieve: Universal Document Linking for Zero-Shot Information Retrieval
Dae Yon Hwang (Amazon AGI), Yaroslav Nechaev (Amazon AGI)
RetrievalTransformerLarge Language ModelSupervised Fine-TuningTextRetrieval-Augmented Generation
🎯 What it does: Propose a Universal Document Linking (UDL) algorithm that utilizes entropy-based selection of similarity models and employs Named Entity Recognition (NER) to determine document links, thereby generating cross-document synthetic queries to enhance zero-shot information retrieval;
LIONs: An Empirically Optimized Approach to Align Language Models
Xiao Yu (Columbia University), Zhou Yu (Columbia University)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper systematically evaluates and optimizes the complete three-stage training pipeline from supervised fine-tuning (SFT) to offline preference learning (DPO) and then to online preference learning (DPO), addressing the alignment issue of large language models. Based on publicly available models Gemma-2b and LLaMA-3-8b, the LION series of models were constructed.
LitSearch: A Retrieval Benchmark for Scientific Literature Search
Anirudh Ajith (Princeton University), Tianyu Gao (Princeton University)
RetrievalLarge Language ModelTextBenchmark
🎯 What it does: Created a literature retrieval benchmark named LitSearch, containing 597 real, manually reviewed scientific literature retrieval questions, matched with 64,183 ACL/ICLR papers.
LLaMA-MoE: Building Mixture-of-Experts from LLaMA with Continual Pre-Training
Tong Zhu (Soochow University), Yu Cheng (Chinese University of Hong Kong)
Computational EfficiencyTransformerMixture of ExpertsText
🎯 What it does: Split the FFN of LLaMA-2-7B into multiple experts and recover performance through continuous pre-training, resulting in the LLaMA-MoE model capable of efficient inference.
LLM See, LLM Do: Leveraging Active Inheritance to Target Non-Differentiable Objectives
Luísa Shimabucoro (Cohere For AI), Sara Hooker (Cohere For AI)
Knowledge DistillationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This work investigates the impact of using synthetic data on the passive and active inheritance properties of large language models (LLMs), proposing a strategy to explicitly guide the model to generate longer, more lexically diverse, and less toxic non-differentiable attributes during fine-tuning through an active inheritance mechanism.
LLM Task Interference: An Initial Study on the Impact of Task-Switch in Conversational History
Akash Gupta (University of Cambridge), Mario Fritz (CISPA Helmholtz Center for Information Security)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Investigated the performance interference of conversational LLMs during task switching, introducing the concept of task switching and quantifying its impact.
LLM-Based Agent Society Investigation: Collaboration and Confrontation in Avalon Gameplay
Yihuai Lan (Hong Kong University of Science and Technology), Hao Wang (Hong Kong University of Science and Technology)
TransformerLarge Language ModelAgentic AIPrompt EngineeringText
🎯 What it does: Studied the social behavior of large language models (LLM) in the social reasoning game Avalon, and proposed a multi-module framework to enhance their game performance and social interaction.
LLM-based Code-Switched Text Generation for Grammatical Error Correction
Tom Potter (University of Manchester), Zheng Yuan (King's College London)
GenerationData SynthesisTransformerLarge Language ModelText
🎯 What it does: This paper proposes a grammar error correction (GEC) system for code-switched text (CSW) and addresses the data scarcity issue by generating synthetic CSW data.
LLM-Evolve: Evaluation for LLM’s Evolving Capability on Benchmarks
Jiaxuan You (NVIDIA), Bryan Catanzaro (NVIDIA)
TextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose the LLM-Evolve framework, transforming traditional LLM benchmarks into multi-round, interactive sequential problem-solving evaluations;
LLM4Decompile: Decompiling Binary Code with Large Language Models
Hanzhuo Tan (Southern University of Science and Technology), Yuqun Zhang (Southern University of Science and Technology)
GenerationAI Code AssistantTransformerLarge Language ModelTextBenchmark
🎯 What it does: Developed the LLM4Decompile series of models, utilizing large-scale LLMs to directly convert binary code into readable C source code, and integrated with traditional decompilation tools to improve executable rates.
LLMEdgeRefine: Enhancing Text Clustering with LLM-Based Boundary Point Refinement
Zijin Feng (Chinese University of Hong Kong), Kam-Fai Wong (Chinese University of Hong Kong)
Representation LearningData-Centric LearningTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposes LLMEdgeRefine, a two-stage method that iteratively refines text clustering edge points using large language models.
LLMs Are Prone to Fallacies in Causal Inference
Nitish Joshi (New York University), He He (New York University)
Data SynthesisTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Evaluate the ability of large language models (LLMs) to infer causality from text through fine-tuning, particularly focusing on whether models can surpass explicitly memorized causal facts during pre-training;
LLMs Are Zero-Shot Context-Aware Simultaneous Translators
Roman Koshkin (Okinawa Institute of Science and Tenchnology), Satoshi Nakamura (Nara Institute of Science and Technology)
GenerationTransformerLarge Language ModelPrompt EngineeringTextAudio
🎯 What it does: This paper proposes a zero-shot, context-aware real-time translation method that leverages the open-source large language model (Llama-3-70B-Instruct) and Whisper ASR for incremental inference, achieving simultaneous translation without training through response priming and minimal background information injection.
LLMs Assist NLP Researchers: Critique Paper (Meta-)Reviewing
Jiangshu Du (University of Illinois Chicago), Wenpeng Yin (Penn State University)
TransformerLarge Language ModelPrompt EngineeringTextReview/Survey PaperBenchmark
🎯 What it does: Construct the ReviewCritique dataset to compare human-generated and LLM-generated paper reviews (and meta-reviews), annotate defects and explanations at the sentence level, and evaluate LLM performance in review and meta-review tasks.
LLMs learn governing principles of dynamical systems, revealing an in-context neural scaling law
Toni J.b. Liu, Christopher Earls
TransformerLarge Language ModelTime SeriesPhysics Related
🎯 What it does: Studies the ability of large language models to automatically extract and predict transition rules for various random, chaotic, continuous, and discrete dynamical systems under zero-shot and no-prompt in-context learning scenarios;
LLoCO: Learning Long Contexts Offline
Sijun Tan (University Of California Berkeley), Raluca Ada Popa (University Of California Berkeley)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Proposed the LLoCO method: first generate summary embeddings using a context compressor, then perform domain-specific parameter-efficient fine-tuning via LoRA, enabling the 4k token LLaMA2-7B to handle long texts up to 128k tokens.
LM2: A Simple Society of Language Models Solves Complex Reasoning
Gurusha Juneja (Microsoft Research), Tanmoy Chakraborty (IIT Delhi)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmarkChain-of-Thought
🎯 What it does: Built a multi-model collaborative framework named LM 2, consisting of three components: splitter, solver, and verifier. The framework uses reinforcement learning to enable the splitter to dynamically generate sub-questions based on feedback from the solver and verifier, achieving more robust multi-step reasoning.
Local Contrastive Editing of Gender Stereotypes
Marlene Lutz (University of Mannheim), Anne Lauscher (University of Mannheim)
TransformerSupervised Fine-TuningContrastive LearningText
🎯 What it does: This paper proposes the Local Contrastive Editing method, which can locate and modify a small portion of weights in language models to control gender stereotype bias.
Locating Information Gaps and Narrative Inconsistencies Across Languages: A Case Study of LGBT People Portrayals on Wikipedia
Farhan Samir (University of British Columbia), Yulia Tsvetkov (University of Washington)
Safty and PrivacyComputational EfficiencyData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: Propose the INFOGAP method to automatically compare factual-level information differences in biographies across different language Wikipedias.
LogicAsker: Evaluating and Improving the Logical Reasoning Ability of Large Language Models
Yuxuan Wan (Chinese University of Hong Kong), Michael Lyu
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Propose the LogicAsker framework, which generates automated minimal functional tests (MFT) using 34 atomic logic rules and their 208 extended variants. The system evaluates the formal reasoning capabilities of LLMs, generates targeted examples and fine-tuning data based on error cases, thereby improving the model's reasoning accuracy.
LogicST: A Logical Self-Training Framework for Document-Level Relation Extraction with Incomplete Annotations
Shengda Fan (Beihang University), Jianwei Niu (Zhongguancun Laboratory)
Data-Centric LearningText
🎯 What it does: The paper proposes the LogicST framework, which diagnoses and corrects pseudo-labels through logical rules to address the problem of incomplete annotations in document-level relation extraction.
LONGAGENT: Achieving Question Answering for 128k-Token-Long Documents through Multi-Agent Collaboration
Jun Zhao (Fudan University), Xuanjing Huang (Fudan University)
RetrievalLarge Language ModelSupervised Fine-TuningAgentic AIPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes LONGAGENT, a long-text question-answering framework based on multi-agent collaboration, which first splits documents into small chunks processed by member agents, then the leader agent derives the final answer through multi-round instructions and discussions.
LongEmbed: Extending Embedding Models for Long Context Retrieval
Dawei Zhu (Peking University), Sujian Li (Peking University)
RetrievalTransformerLarge Language ModelTextBenchmark
🎯 What it does: Built the LONGEMBED benchmark and investigated training-agnostic context window expansion methods, enabling existing embedding models to extend from 512 tokens to a maximum of 32,768 tokens.
LongRAG: A Dual-Perspective Retrieval-Augmented Generation Paradigm for Long-Context Question Answering
Qingfei Zhao (Institute of Information Engineering Chinese Academy of Sciences), Jie Tang (Tsinghua University)
GenerationRetrievalTransformerSupervised Fine-TuningTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose LongRAG, a dual-perspective retrieval-augmented generation (RAG) framework for long-context question answering (LCQA), enhancing the ability to grasp global information and factual details through an information extractor and a CoT-guided filter.
Lookback Lens: Detecting and Mitigating Contextual Hallucinations in Large Language Models Using Only Attention Maps
Yung-Sung Chuang (Massachusetts Institute of Technology), James R. Glass (Massachusetts Institute of Technology)
Anomaly DetectionExplainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: Propose Lookback Lens, a method that detects contextual hallucinations in large language models by utilizing the 'lookback ratio' in attention weights, and reduces hallucination occurrence through Lookback Lens Guided Decoding.
LoRA-Guard: Parameter-Efficient Guardrail Adaptation for Content Moderation of Large Language Models
Hayder Elesedy (Samsung R&D Institute UK), Mete Ozay (Samsung R&D Institute UK)
ClassificationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose LoRA-Guard, a dual-path content moderation framework leveraging low-rank adapters, which can efficiently regulate large language models while maintaining generation performance.
LUQ: Long-text Uncertainty Quantification for LLMs
Caiqi Zhang (University of Cambridge), Nigel Collier (University of Cambridge)
Explainability and InterpretabilityTransformerLarge Language ModelTextBiomedical DataBenchmark
🎯 What it does: Propose the LUQ method for uncertainty quantification in long-text generation, and enhance the factual accuracy of LLMs through LUQ-ENSEMBLE and selective answering.
M^2PT: Multimodal Prompt Tuning for Zero-shot Instruction Learning
Taowen Wang (Rochester Institute of Technology), Dongfang Liu (Rochester Institute of Technology)
Computational EfficiencyRepresentation LearningMeta LearningTransformerLarge Language ModelPrompt EngineeringImageTextMultimodality
🎯 What it does: Proposed a multi-modal prompt tuning (M PT) method that achieves zero-shot instruction learning for multi-modal large language models by utilizing a small number of learnable visual and text prompts.