EMNLP 2024 Papers with AI Summaries
Conference on Empirical Methods in Natural Language Processing · 1268 papers
→ EMNLP 2024 papers with code (435)
Each paper below shows an AI-generated one-line summary. Get the full 6-part summary (innovation, method, data, results, limitations) and search all 1268 EMNLP 2024 papers by keyword, author or institution —
free trial on arXivSub.
‘Quis custodiet ipsos custodes?’ Who will watch the watchmen? On Detecting AI-generated peer-reviews
Sandeep Kumar (Indian Institute of Technology Patna), Asif Ekbal (Indian Institute of Technology Jodhpur)
Anomaly DetectionLarge Language ModelText
🎯 What it does: Propose two methods for detecting AI-generated peer reviews—the TF model based on term frequency and the RR model based on regeneration—and verify their effectiveness on ICLR and NeurIPS review texts.
“A good pun is its own reword”: Can Large Language Models Understand Puns?
Zhijun Xu (Fudan University), Deqing Yang (Fudan University)
RecognitionGenerationTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Systematically evaluate the capabilities of large language models (LLMs) in understanding puns, covering three tasks: pun identification, explanation, and generation.
“Flex Tape Can’t Fix That”: Bias and Misinformation in Edited Language Models
Karina Halevy (École polytechnique fédérale de Lausanne), Antoine Bosselut (École polytechnique fédérale de Lausanne)
TransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Study the impact of model editing methods on gender, racial, and geographic biases in large language models, and propose a new benchmark dataset SEESAW-CF for evaluation
“Global is Good, Local is Bad?”: Understanding Brand Bias in LLMs
Mahammed Kamruzzaman (University of South Florida), Gene Louis Kim (University of South Florida)
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: This paper constructs question-answer pairs between global/local brands and their attributes to evaluate and quantify differences in brand preferences among large language models.
“Image, Tell me your story!” Predicting the original meta-context of visual misinformation
Jonathan Tonglet (TU Darmstadt), Iryna Gurevych (KU Leuven)
ClassificationRetrievalAnomaly DetectionTransformerLarge Language ModelVision Language ModelImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: This paper proposes an automated image contextualization task, constructs the 5Pils dataset containing 1,676 fact-checking images, and implements a baseline pipeline based on retrieval and multimodal LLMs.
“In-Dialogues We Learn”: Towards Personalized Dialogue Without Pre-defined Profiles through In-Dialogue Learning
Chuanqi Cheng (Renmin University of China), Rui Yan (Renmin University of China)
GenerationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Propose the In-Dialogue Learning (IDL) framework, which directly learns personality information from dialogues without requiring predefined profiles, enabling personalized dialogue generation on large language models.
“They are uncultured”: Unveiling Covert Harms and Social Threats in LLM Generated Conversations
Preetam Prabhu Srikar Dammu (University of Washington), Tanu Mitra
Safty and PrivacyTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Generated 1,920 LLM dialogues in a recruitment context and constructed 7 CHAST indicators based on social sciences to quantitatively assess covert harms and social threats related to race and caste across 8 LLMs (including OpenAI and open-source models).
“Thinking” Fair and Slow: On the Efficacy of Structured Prompts for Debiasing Language Models
Shaz Furniturewala (BITS Pilani), Kokil Jaidka (National University of Singapore)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Studied three types of structured prompts based on psychological decision theory (prefix prompts, automatic self-correction, causal prompts) to reduce bias in large language model text generation when model internals are inaccessible.
“We Demand Justice!”: Towards Social Context Grounding of Political Texts
Rajkumar Pujari (Purdue University), Dan Goldwasser (Purdue University)
ClassificationGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes and implements two tasks related to social context semantic attribution and sentiment identification for political texts, and constructs corresponding annotated datasets.
“You Gotta be a Doctor, Lin” : An Investigation of Name-Based Bias of Large Language Models in Employment Recommendations
Huy Nghiem (University of Maryland), Hal Daumé III (University of Southern California)
Recommendation SystemTransformerLarge Language ModelPrompt EngineeringTextTabular
🎯 What it does: Investigating racial and gender bias in large language models (GPT-3.5-Turbo and Llama 3-70B-Instruct) during hiring decisions and salary recommendations caused by candidates' first names, using 320 strong indicators of race/gender first names in the US, covering 40 occupations, and evaluating bias by comparing with US labor statistics;
***YesBut***: A High-Quality Annotated Multimodal Dataset for evaluating Satire Comprehension capability of Vision-Language Models
Abhilash Nandy (Indian Institute of Technology Kharagpur), Niloy Ganguly (Indian Institute of Technology Kharagpur)
ClassificationRecognitionPrompt EngineeringVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Established a high-quality multimodal satirical understanding dataset named YesBut, and proposed three evaluation tasks: satirical image detection, understanding, and completion;
\texttt{ModSCAN}: Measuring Stereotypical Bias in Large Vision-Language Models from Vision and Language Modalities
Yukun Jiang (CISPA Helmholtz Center for Information Security), Yang Zhang (CISPA Helmholtz Center for Information Security)
Prompt EngineeringVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes the ModSCAN framework to systematically measure the stereotypical biases against gender and race in large vision-language models (LVLMs) across both visual and language modalities, and experimentally evaluates the bias levels of LLaVA-v1.5, MiniGPT-v2, and CogVLM.
1+1>2: Can Large Language Models Serve as Cross-Lingual Knowledge Aggregators?
Yue Huang (University of Notre Dame), Lichao Sun (Lehigh University)
Representation LearningTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Propose a cross-lingual knowledge aggregation method that leverages low-resource knowledge detection, target language selection, and answer replacement and fusion to enhance the multilingual consistency and overall performance of large language models (LLMs).
A Bayesian Approach to Harnessing the Power of LLMs in Authorship Attribution
Zhengmian Hu (University of Maryland), Heng Huang (University of Maryland)
ClassificationExplainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes a Bayesian framework-based authorship attribution method that leverages large language models (LLMs) to infer text probabilities, enabling efficient author identification in one-shot scenarios.
A Closer Look at Multidimensional Online Political Incivility
Sagi Pendzel (University of Haifa), Einat Minkov (University of Haifa)
ClassificationTransformerSupervised Fine-TuningText
🎯 What it does: Constructed a tweet dataset containing 13.1K multidimensional political incivility (impoliteness and intolerance) labels, and used this dataset to train and evaluate a multilabel text classification model.
A Comparison of Language Modeling and Translation as Multilingual Pretraining Objectives
Zihao Li (University of Helsinki), Jörg Tiedemann (University of Helsinki)
ClassificationRecognitionRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: In a controlled experimental environment, five different architectures and pre-training objectives (MLM, CLM, TLM, BART denoising, BART machine translation) were uniformly trained, and their performance was evaluated on six languages through four downstream tasks (sentiment analysis, named entity recognition, part-of-speech tagging, natural language inference) using probe and finetune methods.
A Comprehensive Survey of Scientific Large Language Models and Their Applications in Scientific Discovery
Yu Zhang (University Of Illinois At Urbana Champaign), Jiawei Han (University Of Illinois At Urbana Champaign)
Drug DiscoveryProtein Structure PredictionTransformerLarge Language ModelImageTextGraphTabularBiomedical DataReview/Survey PaperBenchmark
🎯 What it does: This paper reviews and systematically organizes over 260 scientific domain large language models (LLMs), analyzing their cross-field and cross-modal architectures and pre-training techniques, while providing a unified classification framework and evaluation benchmark.
A Fast and Sound Tagging Method for Discontinuous Named-Entity Recognition
Caio Corro (INSA Rennes)
RecognitionComputational EfficiencyTransformerTextBiomedical Data
🎯 What it does: Propose a novel tagging scheme based on a two-layer structure, achieving efficient and unambiguous discrete named entity recognition through weighted finite state automata (WFSA).
A Generic Method for Fine-grained Category Discovery in Natural Language Texts
Chang Tian (Ku Leuven), Marie-Francine Moens (Ku Leuven)
ClassificationTransformerContrastive LearningText
🎯 What it does: Propose the STAR method, which uses bidirectional KL similarity in log space to guide the distribution of text samples in Euclidean space, enabling fine-grained category discovery without fine-grained annotations.
A Learning Rate Path Switching Training Paradigm for Version Updates of Large Language Models
Zhihao Wang (Xiamen University), Jinsong Su (Xiamen University)
Computational EfficiencyHyperparameter SearchTransformerLarge Language ModelText
🎯 What it does: This paper studies the training paradigm for version updates of large language models (LLMs), analyzes the advantages and disadvantages of pre-training from scratch (PTFS) and continuous pre-training (CPT), and proposes a learning rate path switching (LRPS) training paradigm that combines maximum learning rate pre-training on the main path with complete learning rate decay on the branch path, balancing performance and cost.
A linguistically-motivated evaluation methodology for unraveling model’s abilities in reading comprehension tasks
Elie Antoine (Aix-Marseille Université), Philippe Langlais (Université de Montréal)
TransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Designed a fine-grained reading comprehension evaluation method based on a semantic framework, categorizing examples through multi-model voting and validating seven complexity factors.
A Morphology-Based Investigation of Positional Encodings
Poulami Ghosh (IIT Bombay), Pushpak Bhattacharyya (IIT Bombay)
Representation LearningTransformerSupervised Fine-TuningText
🎯 What it does: This paper explores the relationship between morphological complexity and the importance of positional encoding through fine-tuning experiments on the BERT model across 22 different languages.
A Multi-Perspective Analysis of Memorization in Large Language Models
Bowen Chen (University of Tokyo), Yusuke Miyao (University of Tokyo)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Investigate the memory behavior of large language models (LLMs) during the generation process, quantitatively and finely analyzing phenomena such as memory intensity, frequency boundary effects, embedding space clustering, and entropy inversion effects from multiple dimensions.
A New Pipeline for Knowledge Graph Reasoning Enhanced by Large Language Models Without Fine-Tuning
Zhongwu Chen (National Key Laboratory of Parallel and Distributed Computing National University of Defense Technology), Yong Dou (National Key Laboratory of Parallel and Distributed Computing National University of Defense Technology)
Graph Neural NetworkTransformerLarge Language ModelPrompt EngineeringGraph
🎯 What it does: Proposed a three-stage pipeline for LLM-enhanced knowledge graph reasoning without fine-tuning the LLM, comprising knowledge alignment, graph reasoning, and entity re-ranking.
A Peek into Token Bias: Large Language Models Are Not Yet Genuine Reasoners
Bowen Jiang (University of Pennsylvania), Dan Roth (Argonne National Laboratory)
Data SynthesisExplainability and InterpretabilityTransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: Investigated the 'token bias' of large language models (LLMs) in logical reasoning tasks, i.e., whether models truly reason or merely answer based on vocabulary preferences.
A Probability–Quality Trade-off in Aligned Language Models and its Relation to Sampling Adaptors
Naaman Tan (National University of Singapore), Ryan Cotterell (ETH Zürich)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This paper studies the relationship between probability and quality (reward) in alignment language models (e.g., RLHF) from both theoretical and experimental perspectives, proving that a probability-quality trade-off exists under large samples and proposing a global normalization sampling adapter to control this trade-off.
A Simple and Effective L\_2 Norm-Based Strategy for KV Cache Compression
Alessio Devoto (Sapienza University of Rome), Pasquale Minervini (University of Edinburgh)
CompressionTransformerLarge Language ModelText
🎯 What it does: Propose a KV cache compression strategy based on the L2 norm of key vectors
A Simple LLM Framework for Long-Range Video Question-Answering
Ce Zhang (University of North Carolina at Chapel Hill), Gedas Bertasius (University of North Carolina at Chapel Hill)
TransformerLarge Language ModelPrompt EngineeringVideoText
🎯 What it does: Proposes a two-stage framework called LLoVi based on LLM, which first segments long videos with short-term visual captions and then uses large language models for long-term reasoning to complete question answering.
A Simple yet Effective Training-free Prompt-free Approach to Chinese Spelling Correction Based on Large Language Models
Houquan Zhou (Soochow University), Min Zhang (Soochow University)
TransformerLarge Language ModelText
🎯 What it does: This paper proposes a Chinese spelling correction method based on large language models (LLM) that requires neither training nor prompting.
A SMART Mnemonic Sounds like “Glue Tonic”: Mixing LLMs with Student Feedback to Make Mnemonic Learning Stick
Nishant Balepur (University of Maryland), Jordan Lee Boyd-Graber
OptimizationComputational EfficiencyData-Centric LearningReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Designed and implemented SMART, a keyword mnemonic generator aligned with student feedback, which fine-tunes LLaMA-2 70B and collects three types of student preferences (comparisons, Likert scales, and learning effectiveness) in a Flashcard application. Finally, a Bayesian model synthesizes effectiveness signals and aligns the model using Direct Preference Optimization (DPO).
A Study of Nationality Bias in Names and Perplexity using Off-the-Shelf Affect-related Tweet Classifiers
Valentin Barriere (Universidad de Chile), Sebastian Cifuentes (CENIA)
ClassificationRecognitionData SynthesisTransformerLarge Language ModelText
🎯 What it does: Leverage person name entities automatically identified from Twitter data, introducing minor perturbations using common names from different countries within the same corpus to generate counterfactual samples, evaluating the preference of sentiment, emotion, hate speech, and offensive text classifiers for country names.
A Survey of AMR Applications
Shira Wein (Amherst College), Juri Opitz (University of Zurich)
TextReview/Survey Paper
🎯 What it does: Reviewed and categorized over 100 papers that use AMR for downstream tasks, systematically summarizing their application scope, technical approaches, and future development directions.
A Survey of Ontology Expansion for Conversational Understanding
Jinggui Liang (Singapore Management University), Lizi Liao (Singapore Management University)
Recurrent Neural NetworkTransformerLarge Language ModelPrompt EngineeringContrastive LearningTextReview/Survey Paper
🎯 What it does: This paper is a review article that systematically organizes and summarizes the current research status of ontology expansion (Ontology Expansion, OnExp) in conversational understanding. It proposes a unified classification framework consisting of three categories: new intent discovery (NID), new slot value discovery (NSVD), and joint ontology expansion (Joint OnExp). The paper also compiles and evaluates related datasets, evaluation metrics, benchmark experiments, and GitHub resources.
A Survey on In-context Learning
Qingxiu Dong (Peking University), Zhifang Sui (Peking University)
TransformerLarge Language ModelReinforcement LearningPrompt EngineeringTextReview/Survey PaperRetrieval-Augmented Generation
🎯 What it does: This paper provides a systematic review of in-context learning (ICL) in large language models, defining ICL, summarizing model training, prompt design, scoring functions, analysis methods, and application scenarios, and discussing challenges and future directions.
A Systematic Analysis of Large Language Models as Soft Reasoners: The Case of Syllogistic Inferences
Leonardo Bertolazzi (University of Trento), Raffaella Bernardi (University of Trento)
TransformerLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: Systematically analyze the soft reasoning capabilities of large language models in reasoning, focusing on syllogistic inference in deductive reasoning, and evaluate the quality of their conclusions through generative multiple-choice tasks;
A Systematic Survey and Critical Review on Evaluating Large Language Models: Challenges, Limitations, and Recommendations
Md Tahmid Rahman Laskar (York University), Jimmy Xiangji Huang (York University)
TransformerPrompt EngineeringTextReview/Survey PaperBenchmark
🎯 What it does: Systematically reviews and critically evaluates the entire process of evaluating large language models (LLMs), identifying and summarizing the main challenges and limitations in terms of reproducibility, reliability, and robustness during the evaluation process.
A Thorough Examination of Decoding Methods in the Era of LLMs
Chufan Shi (Tsinghua University), Wai Lam (Chinese University of Hong Kong)
GenerationComputational EfficiencyTransformerLarge Language ModelTextReview/Survey Paper
🎯 What it does: Conducted a systematic multi-dimensional evaluation of decoding methods for large language models, including performance, robustness, speed, hyperparameter sensitivity, and performance under quantization conditions.
A Two-Step Approach for Data-Efficient French Pronunciation Learning
Hoyeon Lee (NAVER Cloud), Jaemin Kim
RecognitionTransformerText
🎯 What it does: Propose a two-step French pronunciation learning framework: first, train an autoregressive Transformer (G2P) using large-scale word-level pronunciation data, then use a shallow non-autoregressive Transformer (post-lexical phonetization) to handle phonetic phenomena such as liaison and elision between words.
A Usage-centric Take on Intent Understanding in E-Commerce
Wendi Zhou (University of Edinburgh), Jeff Z. Pan (University of Edinburgh)
Recommendation SystemTransformerLarge Language ModelTextGraphBenchmark
🎯 What it does: Propose a use-centered intent understanding paradigm and construct a product recovery benchmark evaluation framework, analyzing and quantifying the two major weaknesses of the existing intent knowledge graph FolkScope: attribute ambiguity and category rigidity;
A User-Centric Multi-Intent Benchmark for Evaluating Large Language Models
Jiayin Wang, Jian-Yun Nie (Universite De Montreal)
Large Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Constructed a multi-intent, multi-cultural evaluation benchmark (URS) based on real user interactions and evaluated 10 LLM services
ABLE: Personalized Disability Support with Politeness and Empathy Integration
Kshitij Mishra (Indian Institute of Technology Patna), Asif Ekbal (Indian Institute of Technology Jodhpur)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Developed ABLE, a dialogue support system based on user personality (OCEAN), gender, age, and levels of politeness and empathy, designed to provide personalized guidance for individuals with physical disabilities;
ABSEval: An Agent-based Framework for Script Evaluation
Sirui Liang (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)
Large Language ModelAgentic AIPrompt EngineeringTextBenchmark
🎯 What it does: This study proposes a script evaluation dataset named MCScript containing over 1500 scripts, and constructs a multi-agent script evaluation framework called ABSEval to automatically assess scripts generated by LLMs;
Academics Can Contribute to Domain-Specialized Language Models
Mark Dredze (Bloomberg), Sebastian Gehrmann (Bloomberg)
TransformerSupervised Fine-TuningTextBiomedical DataReview/Survey PaperFinance RelatedRetrieval-Augmented Generation
🎯 What it does: This paper reviews the development history and evaluation status of large models and domain-specific models, proposing that academia should focus on constructing and in-depth evaluation of domain-specific language models to address the shortcomings of general models in specific tasks;
Accurate and Data-Efficient Toxicity Prediction when Annotators Disagree
Harbani Jaggi, Erdem Biyik
ClassificationTransformerLarge Language ModelPrompt EngineeringTextTabular
🎯 What it does: The study aims to improve prediction accuracy by predicting individual annotators' toxicity scores in cases of annotator disagreement, and compares three methods (neural collaborative filtering, embedded architecture, in-context learning) with the role of demographic information.
ACE: A LLM-based Negotiation Coaching System
Ryan Shea (Columbia University), Zhou Yu (Columbia University)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Developed a negotiation coaching system called ACE based on large language models (LLMs), capable of simulating negotiation partners and providing personalized, targeted feedback;
ActPlan-1K: Benchmarking the Procedural Planning Ability of Visual Language Models in Household Activities
Ying Su (Hong Kong University of Science and Technology), Yangqiu Song (University of California San Diego)
TransformerLarge Language ModelVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Introduce the ActPlan-1K benchmark to evaluate the capabilities of vision-language models in multimodal and counterfactual home activity planning;
Adaptable Moral Stances of Large Language Models on Sexist Content: Implications for Society and Gender Discourse
Rongchen Guo (University of Ottawa), Svetlana Kiritchenko (National Research Council Canada)
Explainability and InterpretabilityLarge Language ModelPrompt EngineeringText
🎯 What it does: Explore the moral reasoning capabilities of large language models (LLMs) in explaining implicit gender discriminatory content, both critiquing and defending such content
Adaptation Odyssey in LLMs: Why Does Additional Pretraining Sometimes Fail to Improve?
Fırat Öncel (Concordia University), Çağatay Yıldız (University of Tübingen)
Domain AdaptationTransformerLarge Language ModelText
🎯 What it does: Systematically studied the changes in domain perplexity when additional pretraining is conducted on different text domains using already pre-trained large language models, and found that similar domains may actually lead to an increase in perplexity.
Adapters Mixup: Mixing Parameter-Efficient Adapters to Enhance the Adversarial Robustness of Fine-tuned Pre-trained Text Classifiers
Tuc Van Nguyen (Indiana University), Thai Le (Indiana University)
ClassificationAdversarial AttackTransformerSupervised Fine-TuningTextBenchmark
🎯 What it does: Propose the ADPMIXUP method, combining parameter-efficient fine-tuning (PEFT) with Mixup, dynamically mixing clean and adversarial adapter weights to enhance model robustness against known and unknown attacks.
Adaption-of-Thought: Learning Question Difficulty Improves Large Language Models for Reasoning
Mayi Xu (Wuhan University), Tieyun Qian (Wuhan University)
TransformerPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the ADOT (Adaptation-of-Thought) method, which automatically evaluates problem difficulty and dynamically constructs demonstration sets and retrieval strategies based on difficulty to enhance the performance of large language models on reasoning tasks.
Adaptive Axes: A Pipeline for In-domain Social Stereotype Analysis
Qingcheng Zeng (Northwestern University), Rob Voigt
TransformerLarge Language ModelText
🎯 What it does: Propose the ADAPTIVE AXES pipeline, which encodes the context of masked target entities using text embedding models and captures domain-specific social stereotypes through adaptive semantic axes.
Adaptive Immune-based Sound-Shape Code Substitution for Adversarial Chinese Text Attacks
Ao Wang (China University of Petroleum (East China)), Weifeng Liu (China University of Petroleum (East China))
Adversarial AttackText
🎯 What it does: Propose a Chinese text adversarial attack framework that generates natural replacements using sound-shape codes and determines the replacement order through an adaptive immune algorithm.
Adaptive Query Rewriting: Aligning Rewriters through Marginal Probability of Conversational Answers
Tianhua Zhang (Chinese University of Hong Kong), Helen M. Meng
RetrievalOptimizationTransformerSupervised Fine-TuningText
🎯 What it does: Propose the AdaQR framework, which, under conditions of only a small amount of rewritten annotations and completely no paragraph labels, self-samples rewrite candidates and uses the marginal probability of answers retrieved by the retriever as a reward. The query rewriter model is further optimized using DPO to better align with the retriever's preferences.
Adaptive Question Answering: Enhancing Language Model Proficiency for Addressing Knowledge Conflicts with Source Citations
Sagi Shaier (University of Colorado Boulder), Philip V. Ogren (Oracle)
Explainability and InterpretabilityLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextChain-of-Thought
🎯 What it does: This paper proposes a new question-answering task: generating multiple answers along with corresponding source citations in the presence of knowledge conflicts, aiming to enhance system credibility and explainability.
AdaSwitch: Adaptive Switching between Small and Large Agents for Effective Cloud-Local Collaborative Learning
Hao Sun (Peking University), Dawei Yin (Baidu Inc)
Federated LearningComputational EfficiencyKnowledge DistillationLarge Language ModelText
🎯 What it does: Propose an adaptive switching framework ADASWITCH, enabling small local LLMs to collaborate with large cloud LLMs to accomplish complex reasoning tasks;
AdaZeta: Adaptive Zeroth-Order Tensor-Train Adaption for Memory-Efficient Large Language Models Fine-Tuning
Yifan Yang (University of California Santa Barbara), Zheng Zhang (University of California Santa Barbara)
OptimizationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose the AdaZeta framework for zeroth-order (ZO) fine-tuning of large language models, addressing the performance degradation and divergence issues of traditional ZO methods.
ADELIE: Aligning Large Language Models on Information Extraction
Yunjia Qi (Tsinghua University), Juanzi Li (Tsinghua University)
TransformerLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: Align and fine-tune large language models for information extraction (IE) tasks, constructing high-quality instruction tuning data IEInstruct and feedback data IEFeedback. ADELIESFT and ADELIEDPO are trained via supervised fine-tuning (SFT) and direct preference optimization (DPO), respectively, and evaluated on zero-shot and few-shot settings across multiple IE tasks, including closed IE, open IE, and on-demand IE.
Advancing Adversarial Suffix Transfer Learning on Aligned Large Language Models
Hongfu Liu (National University of Singapore), Michael Shieh (National University of Singapore)
Adversarial AttackTransformerLarge Language ModelTextBenchmark
🎯 What it does: Proposed the DeGCG framework, which decomposes adversarial suffix search into a preceding behavior-agnostic first-term search (FTS) and a subsequent behavior-related content search (CAS), and further introduced the alternating self-transfer i-DeGCG algorithm; by optimizing the first term to provide high-quality initialization, it significantly enhances the efficiency of suffix attacks on aligned LLMs;
Advancing Event Causality Identification via Heuristic Semantic Dependency Inquiry Network
Haoran Li (Sichuan University), Li Huang (Southwestern University of Finance and Economics)
ClassificationComputational EfficiencyRepresentation LearningTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Propose the SemDI model, reformulating the event causality identification task as a semantic dependency inquiry problem, generating fill-in words via a fill-in-the-blank Cloze analyzer, and verifying causality between events through cross-attention mechanisms.
Advancing Large Language Model Attribution through Self-Improving
Lei Huang (Harbin Institute of Technology), Bing Qin (Harbin Institute of Technology)
Data SynthesisOptimizationExplainability and InterpretabilityReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Propose the START framework, which uses self-generated synthetic data for preheating, and then iteratively enhances the citation capabilities of LLMs through rejection sampling and fine-grained preference optimization;
Advancing Process Verification for Large Language Models via Tree-Based Preference Learning
Mingqian He (Zhejiang University), Weiming Lu (Zhejiang University)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningContrastive LearningText
🎯 What it does: Proposed a tree-based preference learning verifier called Tree-PLV, which constructs reasoning trees through best-first tree search and generates step-level preference pairs for training, thereby enhancing the reasoning path evaluation capability of large language models (LLMs).
Advancing Semantic Textual Similarity Modeling: A Regression Framework with Translated ReLU and Smooth K2 Loss
Bowen Zhang (Tsinghua University), Chunping Li (Tsinghua University)
Representation LearningTransformerSupervised Fine-TuningContrastive LearningText
🎯 What it does: Propose a framework that converts multi-class semantic text similarity tasks into a regression problem, and design two new loss functions with zero-gradient buffers (Translated ReLU and Smooth K2 Loss) for fine-grained similarity annotation.
Advancing Social Intelligence in AI Agents: Technical Challenges and Open Questions
Leena Mathur (Carnegie Mellon University), Louis-Philippe Morency (Carnegie Mellon University)
TextReview/Survey Paper
🎯 What it does: This paper proposes the technical challenges in building social AI agents and reviews existing research.
Advancing Test-Time Adaptation in Wild Acoustic Test Settings
Hongfu Liu (National University of Singapore), Ye Wang (National University of Singapore)
RecognitionDomain AdaptationTransformerAudio
🎯 What it does: This paper proposes a test-time adaptation (TTA) framework tailored for wild acoustic environments, enabling online model updates for refined acoustic base models (e.g., Wav2vec2, HuBERT, WavLM) under various domain shift scenarios such as noise, accents, and singing.
Adversarial Text Generation using Large Language Models for Dementia Detection
Youxiang Zhu (University of Massachusetts Boston), Xiaohui Liang (University of Massachusetts Boston)
ClassificationTransformerLarge Language ModelPrompt EngineeringTextAlzheimer's Disease
🎯 What it does: Propose the Adversarial Text Generation (ATG) method using dialogue-based text generation and perplexity judgment to achieve dementia detection through image description text.
African or European Swallow? Benchmarking Large Vision-Language Models for Fine-Grained Object Classification
Gregor Geigle (University of Würzburg), Goran Glavaš (University of Würzburg)
ClassificationTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextBenchmark
🎯 What it does: Proposed and implemented a new multi-choice fine-grained object classification benchmark called FOCI to evaluate the capability of large-scale vision-language models in fine-grained classification;
AgentReview: Exploring Peer Review Dynamics with LLM Agents
Yiqiao Jin (Georgia Institute of Technology), Jindong Wang (University of California Santa Barbara)
TransformerLarge Language ModelAgentic AIPrompt EngineeringText
🎯 What it does: This study constructs the AGENTREVIEW framework through large language model (LLM) agents, fully simulating the five stages of peer review (initial review, author response, reviewer discussion, meta-review, and final decision), and generating over 53,800 review documents. It systematically investigates the impact of multiple factors (reviewer expertise, commitment, intent; area chair style; author anonymity, etc.) on review outcomes.
AGRaME: Any-Granularity Ranking with Multi-Vector Embeddings
Revanth Gangi Reddy (University of Illinois at Urbana-Champaign), Saloni Potdar (Apple)
RetrievalTransformerContrastive LearningTextRetrieval-Augmented Generation
🎯 What it does: Propose the AGRAME method, which utilizes multi-vector embeddings to achieve arbitrary granularity retrieval ranking and incorporates multi-granularity contrastive loss during training; simultaneously design PROPCITE, a post-processing approach that adds citations to RAG text.
AKEW: Assessing Knowledge Editing in the Wild
Xiaobao Wu (Nanyang Technological University), Anh Tuan Luu (Nanyang Technological University)
TransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Proposes the AKEW benchmark to evaluate the knowledge editing effectiveness of language models in real-world scenarios;
AlignCap: Aligning Speech Emotion Captioning to Human Preferences
Ziqi Liang (University of Science and Technology of China), Hanhui Chen (Southern University of Science and Technology)
GenerationKnowledge DistillationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringTextMultimodalityAudio
🎯 What it does: Propose the AlignCap model, combining knowledge distillation (KD-Regularization) and preference optimization (PO-Regularization) to align speech emotion description with human preferences, improving the accuracy and authenticity of zero-shot and cross-domain emotional captioning.
Aligning Language Models to Explicitly Handle Ambiguity
Hyuhng Joon Kim (Seoul National University), Taeuk Kim (Hanyang University)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes a four-stage alignment process named APA (Alignment with Perceived Ambiguity), enabling large language models (LLMs) to explicitly identify and handle ambiguous queries by generating clarification requests;
Aligning Large Language Models with Diverse Political Viewpoints
Dominik Stammbach (Princeton University), Elliott Ash (ETH Zurich)
GenerationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Aligning large language models with political perspectives, training them to generate party positions on different issues based on Swiss candidate comments, and providing a method for generating balanced summaries.
Aligning Translation-Specific Understanding to General Understanding in Large Language Models
Yichong Huang (Harbin Institute of Technology), Bing Qin (Harbin Institute of Technology)
GenerationExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes a translation process called DUAT, aiming to eliminate the inconsistency between translation-specific understanding and general understanding in large language models during translation tasks.
Alignment-Enhanced Decoding: Defending Jailbreaks via Token-Level Adaptive Refining of Probability Distributions
Quan Liu (Beijing University of Posts and Telecommunications), Sen Su (Beijing University of Posts and Telecommunications)
Adversarial AttackLarge Language ModelText
🎯 What it does: Proposed and implemented Alignment-Enhanced Decoding (AED) for large language models to defend against jailbreak attacks, utilizing competitive index and self-assessment to dynamically adjust token probability distributions.
AlphaLoRA: Assigning LoRA Experts Based on Layer Training Quality
Peijun Qing (Dartmouth College), Soroush Vosoughi (Dartmouth College)
Computational EfficiencyTransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: In large language models, based on the Heavy-Tail Self-Regularization theory for layer quality assessment, AlphaLoRA is designed to automatically assign the number of LoRA-MoE experts per layer, reducing redundancy and improving performance.
Altogether: Image Captioning via Re-aligning Alt-text
Hu Xu (Meta FAIR), Christoph Feichtenhofer (Meta FAIR)
GenerationRetrievalTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: By performing multi-round manual re-alignment on existing alt-text, generating more refined and image-content-aligned descriptions, and training a lightweight captioner to automatically complete this re-alignment process on large-scale images.
ALVIN: Active Learning Via INterpolation
Michalis Korakakis (University of Cambridge), Adrian Weller (University of Cambridge)
ClassificationTransformerText
🎯 What it does: Designed and implemented an active learning method called ALVIN based on intra-class interpolation, which generates anchors by interpolating representation vectors between minority and majority groups. Then, it selects unlabeled samples near these anchors with high model confidence using KNN, thereby reducing the model's reliance on shortcuts.
AmbigNLG: Addressing Task Ambiguity in Instruction for NLG
Ayana Niwa (Megagon Labs), Hayate Iso (Megagon Labs)
GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: Propose the AmbigNLG task and the AmbigSNI NLG dataset, investigating how adding explicit sub-instructions can alleviate task ambiguity in NLG instructions, and verifying that this method improves the consistency of generated text with user expectations.
AMPO: Automatic Multi-Branched Prompt Optimization
Sheng Yang (Institute of Software, Chinese Academy of Sciences and University of Chinese Academy of Sciences), Linjun Yang (Microsoft)
OptimizationLarge Language ModelPrompt EngineeringTextBiomedical DataChain-of-Thought
🎯 What it does: Propose AMPO, an automated multi-branch prompt optimization method that iteratively constructs a multi-branch structure through failure cases to enhance LLM performance
AMR-Evol: Adaptive Modular Response Evolution Elicits Better Knowledge Distillation for Large Language Models in Code Generation
Ziyang Luo (Hong Kong Baptist University), Lidong Bing (Alibaba DAMO Academy)
Knowledge DistillationAI Code AssistantLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Propose the AMR-Evol framework, which first decomposes the complete responses generated by the teacher model into submodules and then uses existing validation modules to adaptively evolve the responses, thereby improving the quality of knowledge distillation in code generation.
An Analysis and Mitigation of the Reversal Curse
Ang Lv (Renmin University of China), Rui Yan (Renmin University of China)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Studied the 'reverse curse' phenomenon in large language models, and for the first time attributed it to training objectives, proposing a fine-tuning framework called BICO that transforms causal language models into bidirectional attention and masked filling training similar to ABI to alleviate the reverse curse;
An Analysis of Multilingual FActScore
Vu Trong Kim (KAIST), Viet Dac Lai (Adobe Research)
RetrievalLarge Language ModelSupervised Fine-TuningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Evaluate and improve the FActScore factuality scoring framework in a multilingual environment, constructing new indigenous annotated datasets in Spanish, Arabic, and Bengali, and systematically analyzing the bottlenecks and performance of four components (knowledge sources, retrieval, fact extraction, and scoring) under different language resource levels; propose and verify three mitigation strategies (expanding retrieval scope, using internet retrieval, and leveraging LLM-generated internal knowledge) to enhance the accuracy of multilingual FActScore.
An Audit on the Perspectives and Challenges of Hallucinations in NLP
Pranav Narayanan Venkit (Pennsylvania State University), Shomir Wilson (Pennsylvania State University)
TextReview/Survey Paper
🎯 What it does: This paper conducts a systematic audit of 103 NLP papers on hallucinations and surveys 171 NLP and AI professionals to explore the concept, definition, evaluation metrics, and differences in community cognition and practice regarding hallucinations.
An Effective Deployment of Diffusion LM for Data Augmentation in Low-Resource Sentiment Classification
Zhuowei Chen, Junyang Zhong (Guangdong University of Foreign Studies)
ClassificationData-Centric LearningTransformerDiffusion modelContrastive LearningText
🎯 What it does: Propose DiffusionCLS, which enhances sentiment classification data using a diffusion language model, focusing on reconstructing label-related vocabulary to improve low-resource sentiment classification performance.
An Electoral Approach to Diversify LLM-based Multi-Agent Collective Decision-Making
Xiutian Zhao (University of Edinburgh), Wei Peng (Huawei IT Innovation and Research Center)
TransformerLarge Language ModelAgentic AITextBenchmark
🎯 What it does: Proposed the GEDI module to extend the collective decision-making (CDM) method for LLM-based multi-agent systems, and audited 52 existing systems, revealing that CDM methods are extremely monotonous.
An Empirical Analysis of the Writing Styles of Persona-Assigned LLMs
Manuj Malik (Singapore Management University), Kian Ming A. Chai (DSO National Laboratories)
GenerationRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Systematically analyze the writing style of persona-assigned LLMs, comparing the stylistic differences between LLMs and human Reddit comments across various sociodemographic labels (age, occupation, location, political tendency).
An Empirical Analysis on Spatial Reasoning Capabilities of Large Multimodal Models
Fatemeh Shiri (Monash University), Yuan-Fang Li (Monash University)
Vision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: This paper constructs a new multimodal visual question answering benchmark, Spatial-MM, systematically evaluating the spatial reasoning capabilities of large multimodal models (LMMs), and conducts in-depth experiments on multi-hop reasoning and perspective differences.
An Empirical Study of Multilingual Reasoning Distillation for Question Answering
Patomporn Payoungkhamdee (VISTEC), Sarana Nutanong (VISTEC)
Knowledge DistillationTransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: This paper studies the reasoning distillation method for small multilingual models and proposes a novel d-CoT-nR distillation scheme that utilizes positive and negative reasoning paths to enhance reasoning performance.
An Experimental Analysis on Evaluating Patent Citations
Rabindra Nath Nandi (Hishab Singapore Pte Ltd), Sourav Medya (University of Illinois Chicago)
Explainability and InterpretabilityGraph Neural NetworkTransformerLarge Language ModelTextGraph
🎯 What it does: Construct a semantic graph using patent texts and predict the number of citations a patent will receive in the next three, five, and ten years through graph neural networks, further revealing the neighborhood characteristics of highly cited patents using a graph interpreter.
An image speaks a thousand words, but can everyone listen? On image transcreation for cultural relevance
Simran Khanuja (Carnegie Mellon University), Graham Neubig (Carnegie Mellon University)
Image TranslationTransformerLarge Language ModelVision Language ModelDiffusion modelImageTextRetrieval-Augmented Generation
🎯 What it does: Propose and evaluate the novel task of 'image cross-cultural recreation,' constructing three end-to-end, caption+LLM-based, retrieval-style generation pipelines and conducting experiments on a self-made evaluation dataset.
An Inversion Attack Against Obfuscated Embedding Matrix in Language Model Inference
Yu Lin (Bytedance), Bing Duan (Bytedance)
Safty and PrivacyAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes an element difference-based nearest neighbor attack (EDNN) targeting the embedding matrix encryption scheme in language model inference, and proves that the existing glide-reflection encryption method is ineffective in protecting user privacy, allowing the complete recovery of the original input text.
An L* Algorithm for Deterministic Weighted Regular Languages
Clemente Pasti (ETH Zurich), Ryan Cotterell (ETH Zurich)
🎯 What it does: Propose a weighted version of Angluin's L◦ algorithm to learn deterministic weighted finite state automata (WDFSA) and precisely generate a given weighted regular language.
An LLM Feature-based Framework for Dialogue Constructiveness Assessment
Lexin Zhou (University of Cambridge), Andreas Vlachos (University of Cambridge)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: Proposed a method that combines LLM with traditional features to evaluate the constructiveness of dialogues;
An Unsupervised Approach to Achieve Supervised-Level Explainability in Healthcare Records
Joakim Edin (University of Copenhagen), Tuukka Ruotsalo (University of Copenhagen)
Explainability and InterpretabilityTransformerBiomedical DataElectronic Health Records
🎯 What it does: An unsupervised method is proposed for the automatic ICD coding task on medical records, generating highly interpretable and credible feature attribution explanations through adversarial robustness training and a novel AttInGrad interpreter.
AnaloBench: Benchmarking the Identification of Abstract and Long-context Analogies
Xiao Ye (Johns Hopkins University), Daniel Khashabi (Johns Hopkins University)
RetrievalTransformerLarge Language ModelTextBenchmark
🎯 What it does: Proposed the ANALOBENCH benchmark, constructed two tasks using 340 high-quality analogy stories written by humans: T1 (selecting the most similar story from a small story library) and T2 (retrieving the top 10 most similar stories from a large story library). Additionally, expanded single-sentence analogies into 10-sentence and 30-sentence long versions using GPT-4 to study the impact of model scale and context length on analogy recognition.
Analysis of Plan-based Retrieval for Grounded Text Generation
Ameya Godbole (University of Southern California), Manzil Zaheer (Google DeepMind)
GenerationRetrievalExplainability and InterpretabilityTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: This paper aims to reduce model hallucinations and enhance text attributability by first prompting the LLM to generate a plan (outline + queries), then using these queries for secondary retrieval, and finally generating text guided by the retrieval results.
Analyzing Key Factors Influencing Emotion Prediction Performance of VLLMs in Conversational Contexts
Jaewook Lee (Konkuk University), Harksoo Kim (Konkuk University)
ClassificationRecognitionTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodalityChain-of-Thought
🎯 What it does: Reconstructed the MELD dataset based on the Friends TV series, incorporating complete scenes, character images, facial images, as well as character personality and speaking style information. The emotional prediction performance of vision-language large models (VLLM) was systematically evaluated through three subtasks (overall emotional tone prediction, character emotion prediction, and context-appropriate emotional expression selection). The study also conducted in-depth analysis of factors such as model architecture, LLM backbone, image scope, personality information, chain-of-thought reasoning, gender bias, and regional bias.
Annotation alignment: Comparing LLM and human annotations of conversational safety
Rajiv Movva (Cornell Tech), Emma Pierson (Cornell Tech)
Safty and PrivacyTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Compare the consistency between LLMs and diverse human reviewers in evaluating chatbot safety, and explore the impact of human group differences on LLM assessments.
Annotator-Centric Active Learning for Subjective NLP Tasks
Michiel van der Meer (Idiap Research Institute), Enrico Liscio (Leiden University)
Data-Centric LearningTransformerText
🎯 What it does: Proposes an annotator-centric active learning (ACAL) framework that integrates annotator selection strategies into traditional active learning to more efficiently acquire diverse annotation information.
ApiQ: Finetuning of 2-Bit Quantized Large Language Model
Baohao Liao (University of Amsterdam), Christof Monz (University of Amsterdam)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes ApiQ, an efficient fine-tuning method for low-bit quantized LLMs that achieves joint quantization of model weights and LoRA component initialization.