EMNLP 2024 Papers — Page 4
Conference on Empirical Methods in Natural Language Processing · 1268 papers
Dependency Graph Parsing as Sequence Labeling
Ana Ezquerro (Universidade da Coruña), Carlos Gómez-Rodríguez (Universidade da Coruña)
Representation LearningRecurrent Neural NetworkTransformerTextGraph
🎯 What it does: Convert dependency graphs (e.g., semantic dependencies and enhanced UD) into a sequence labeling task, proposing a linearized encoding that can handle reentrancy, cycles, and multiple child nodes.
Detecting Errors through Ensembling Prompts (DEEP): An End-to-End LLM Framework for Detecting Factual Errors
Alex Chandler (University of Texas at Austin), Hui Su (Fidelity Investments)
ClassificationAnomaly DetectionTransformerLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Design and implement the DEEP framework, generating binary features through multiple GPT-4/Turbo prompts, and detecting factual errors in text summaries using ensemble learning and calibration methods.
Detecting Online Community Practices with Large Language Models: A Case Study of Pro-Ukrainian Publics on Twitter
Kateryna Kasianenko (Queensland University of Technology), Axel Bruns (Queensland University of Technology)
ClassificationLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: The study proposes a method for identifying online community practices through large-scale language models, using the case of NAFO and Eurovision Twitter communities supported by Ukrainian enthusiasts. It constructs and annotates a gold standard dataset to evaluate model performance.
Detecting Subtle Differences between Human and Model Languages Using Spectrum of Relative Likelihood
Yang Xu (Southern University of Science and Technology), Yongyuan Li (Southern University of Science and Technology)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Studied a method that uses spectral features based on relative likelihood values to distinguish human-generated text from model-generated text.
Detection and Measurement of Syntactic Templates in Generated Text
Chantal Shaib (Northeastern University), Byron C Wallace (Northeastern University)
GenerationExplainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: Studied syntactic templates (based on part-of-speech sequences) appearing in the generated text of large language models (LLMs), and proposed methods to extract, count, and evaluate these templates; through experiments across multiple models, tasks, and datasets, explored the frequency, sources, and roles of these templates in detecting model memory and style replication.
DetoxLLM: A Framework for Detoxification with Explanations
Md Tawkat Islam Khondaker (University of British Columbia), Laks V. S. Lakshmanan
Data SynthesisSafty and PrivacyExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: Constructed a cross-platform pseudo-parallel toxic text corpus and trained a detoxification framework incorporating explanations and synonym detection.
Development of Cognitive Intelligence in Pre-trained Language Models
Raj Sanjay Shah (Georgia Institute of Technology), Sashank Varma (Georgia Institute of Technology)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Investigate the cognitive and developmental trajectories of large pre-trained language models across four psychological dimensions (numerical, linguistic, conceptual, and fluid reasoning)
DGLF: A Dual Graph-based Learning Framework for Multi-modal Sarcasm Detection
Zhihong Zhu (Shanghai University), Yefeng Zheng (Tencent YouTu Lab)
ClassificationGraph Neural NetworkTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Construct a dual-graph learning framework DGLF for multimodal sarcasm detection;
Dialog2Flow: Pre-training Soft-Contrastive Action-Driven Sentence Embeddings for Automatic Dialog Flow Extraction
Sergio Burdisso (Idiap Research Institute), Petr Motlicek (Idiap Research Institute)
Representation LearningTransformerContrastive LearningText
🎯 What it does: Map dialog sentences to a latent space of action clusters through a pre-trained sentence embedding model (Dialog2Flow), enabling automatic extraction of task-oriented dialogue workflows.
Direct Multi-Turn Preference Optimization for Language Agents
Wentao Shi (University of Science and Technology of China), Fuli Feng (University of Science and Technology of China)
OptimizationReinforcement Learning from Human FeedbackSupervised Fine-TuningReinforcement LearningAgentic AITextSequential
🎯 What it does: Propose the DMPO (Direct Multi‑Turn Preference Optimization) loss, directly optimizing the reinforcement learning objective in multi-turn language agent tasks.
DISCERN: Decoding Systematic Errors in Natural Language for Text Classifiers
Rakesh R Menon (University of North Carolina at Chapel Hill), Shashank Srivastava (University of North Carolina at Chapel Hill)
ClassificationExplainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: Propose a framework named DISCERN, which leverages large language models to generate precise natural language descriptions, identify and explain systematic errors in text classifiers, and thereby improve models through data augmentation or active learning.
Discovering Biases in Information Retrieval Models Using Relevance Thesaurus as Global Explanation
Youngwoo Kim (University of Massachusetts Amherst), James Allan (University of Massachusetts Amherst)
RetrievalExplainability and InterpretabilityKnowledge DistillationTransformerText
🎯 What it does: Built a 'relevant dictionary' to globally explain the matching behavior of Transformer-based retrieval models.
Discovering Knowledge-Critical Subnetworks in Pretrained Language Models
Deniz Bayazit (EPFL), Antoine Bosselut (EPFL)
Explainability and InterpretabilityComputational EfficiencyKnowledge DistillationRepresentation LearningTransformerTextGraph
🎯 What it does: Discover and utilize sparse subnetworks to precisely remove certain relational knowledge from pre-trained language models.
Dissecting Fine-Tuning Unlearning in Large Language Models
Yihuai Hong (South China University of Technology), Haiqin Yang (International Digital Economy Academy (IDEA))
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmark
🎯 What it does: Studied methods for large models to learn through fine-tuning, using activation patches and parameter recovery experiments to explore internal mechanisms, revealing that these methods do not truly eliminate knowledge in model parameters but adjust model behavior by changing the MLP's last layer coefficients.
Distilling Knowledge from Text-to-Image Generative Models Improves Visio-Linguistic Reasoning in CLIP
Samyadeep Basu (University of Maryland), Soheil Feizi (University of Maryland)
Computational EfficiencyKnowledge DistillationTransformerDiffusion modelScore-based ModelContrastive LearningImageTextMultimodality
🎯 What it does: Propose a lightweight, sample-efficient distillation method called SDS-CLIP, which fine-tunes CLIP using the denoising diffusion loss from Stable Diffusion to enhance its vision-language reasoning capabilities.
Distract Large Language Models for Automatic Jailbreak Attack
Zeguan Xiao (Shanghai University of Finance and Economics), Yun Chen (Shanghai University of Finance and Economics)
Adversarial AttackTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Propose a black-box automated 'Distraction based Adversarial Prompts (DAP)' framework to generate generalizable jailbreak templates that can induce LLMs to produce harmful outputs without explicitly containing specific malicious instructions.
Distractor Generation in Multiple-Choice Tasks: A Survey of Methods, Datasets, and Evaluation
Elaf Alhazmi (Macquarie University), Ahoud Alhazmi (Macquarie University)
GenerationTransformerLarge Language ModelPrompt EngineeringTextReview/Survey PaperRetrieval-Augmented Generation
🎯 What it does: Reviews the distractor generation (DG) task, datasets, methods, and evaluation metrics in English multiple-choice and fill-in-the-blank questions.
Distributional Properties of Subword Regularization
Marco Cognetta (Tokyo Institute of Technology), Naoaki Okazaki (Tokyo Institute of Technology)
TransformerText
🎯 What it does: Investigate the distributional bias caused by BPE and MaxMatch Dropout, propose a uniform sampling subword tokenizer, and significantly improve performance on multilingual MT tasks.
Diversity Over Size: On the Effect of Sample and Topic Sizes for Topic-Dependent Argument Mining Datasets
Benjamin Schiller (summetix GmbH), Iryna Gurevych (Technical University of Darmstadt)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper experimentally investigates the impact of the composition of the Topic-Related Argument Mining (TDAM) dataset (number of samples and topics) on model performance, and creates a new FS150T-Corpus dataset consisting of 150 topics with 144 samples per topic based on the UKP Corpus;
DiVERT: Distractor Generation with Variational Errors Represented as Text for Math Multiple-choice Questions
Nigel Fernandez (University of Massachusetts Amherst), Andrew Lan (Eedi)
Data SynthesisRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningAuto EncoderText
🎯 What it does: Proposed DiVERT, which leverages a variational autoencoder framework to learn interpretable error text representations and generates corresponding mathematical MCQ distractors based on these errors.
Divide and Conquer Radiology Report Generation via Observation Level Fine-grained Pretraining and Prompt Tuning
Yuanpin Zhou (Zhejiang University HiThink Research), Huogen Wang (HiThink Research)
GenerationTransformerPrompt EngineeringVision Language ModelContrastive LearningMultimodalityBiomedical DataElectronic Health Records
🎯 What it does: This paper proposes a radiology report generation model called DCRRG based on a 'divide and conquer' strategy, which first performs fine-grained pre-training at the observation level, then uses contrastive learning to align cross-modal features, and finally generates complete reports in the decoding stage through prompt tuning;
DKEC: Domain Knowledge Enhanced Multi-Label Classification for Diagnosis Prediction
Xueren Ge (University of Virginia), Homa Alemzadeh (University of Virginia)
ClassificationRepresentation LearningGraph Neural NetworkTransformerLarge Language ModelTextBiomedical DataElectronic Health RecordsChain-of-Thought
🎯 What it does: Propose a diagnostic prediction method based on automatically constructing a heterogeneous medical knowledge graph and employing a label-level attention mechanism in multi-label text classification.
Do great minds think alike? Investigating Human-AI Complementarity in Question Answering with CAIMIRA
Maharshi Gor (University of Maryland), Jordan Lee Boyd-Graber
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: This paper investigates the complementarity between humans and large language models in question-answering tasks, proposing the CAIMIRA framework that utilizes Item Response Theory (IRT) to quantitatively compare the QA capabilities of humans and AI.
Do Large Language Models Know How Much They Know?
Gabriele Prato (Chandar Research Lab), Sarath Chandar (Mila - Quebec AI Institute)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Built a benchmark based on diary entry recall to detect LLM's perception of its own knowledge volume
Do LLMs Know to Respect Copyright Notice?
Jialiang Xu (Stanford University), Denghui Zhang (Stevens Institute of Technology)
Safty and PrivacyTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Studied whether large language models (LLMs) comply with copyright regulations when user inputs contain copyright information, and constructed a benchmark dataset consisting of 43,200 simulated queries, conducting experiments across different content types, copyright notice types, and text lengths.
Do LLMs learn a true syntactic universal?
John T. Hale (Google DeepMind), Miloš Stanojević (Google DeepMind)
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: Investigate whether large multilingual LLMs learn syntactic universals, with a focus on the Final-over-Final Condition (FOFC)
Do LLMs Overcome Shortcut Learning? An Evaluation of Shortcut Challenges in Large Language Models
Yu Yuan (University of Science and Technology of China), Qi Liu (University of Science and Technology of China)
Explainability and InterpretabilityTransformerPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Developed the Shortcut Suite to evaluate the robustness and generalization of large language models across various shortcuts.
Do LLMs Plan Like Human Writers? Comparing Journalist Coverage of Press Releases with LLMs
Alexander Spangher (University of Southern California), Mark Dredze (Bloomberg)
TransformerLarge Language ModelSupervised Fine-TuningTextBenchmarkFinance Related
🎯 What it does: This paper constructs a 'large-scale' corpus containing 250k press releases and their corresponding 650k news articles, and develops a contrastive summarization method based on NLI to identify effective news coverage. Subsequently, the dataset is used to compare human journalists' planning decisions (angles and sources) with planning suggestions from large language models (LLMs), and evaluate their alignment and creativity.
Do LLMs suffer from Multi-Party Hangover? A Diagnostic Approach to Addressee Recognition and Response Selection in Conversations
Nicolò Penzo (Fondazione Bruno Kessler), Marco Guerini (Fondazione Bruno Kessler)
ClassificationRecognitionSafty and PrivacyTransformerLarge Language ModelPrompt EngineeringTextGraph
🎯 What it does: Propose a diagnostic method to evaluate the zero-shot performance of large language models (LLMs) in multi-party conversation (MPC) tasks for response selection (RS) and addressee recognition (AR), focusing on the contribution of text versus structural information and prompt sensitivity.
Do Text-to-Vis Benchmarks Test Real Use of Visualisations?
Hy Nguyen (University of Sydney), Jonathan K. Kummerfeld (University of Sydney)
TextBenchmark
🎯 What it does: This paper analyzes public code and existing Text-to-Vis benchmarks to explore whether benchmark datasets truly reflect users' actual usage in four visualization libraries: Python, R, JavaScript, and Vega-Lite.
Do We Need Language-Specific Fact-Checking Models? The Case of Chinese
Caiqi Zhang (University of Cambridge), Andreas Vlachos (University of Cambridge)
ClassificationRetrievalAdversarial AttackTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: Explores whether a dedicated fact-checking model for Chinese is necessary, proposes a Chinese document-level evidence retriever (DLR), and constructs an adversarial dataset based on CHEF to test the model's robustness against language and cultural biases.
Do You Know What You Are Talking About? Characterizing Query-Knowledge Relevance For Reliable Retrieval Augmented Generation
Zhuohang Li (Vanderbilt University), Bradley A. Malin (Vanderbilt University)
GenerationAnomaly DetectionRepresentation LearningTransformerTextBiomedical DataRetrieval-Augmented Generation
🎯 What it does: Proposed a framework based on statistical significance testing to evaluate the relevance between user queries and knowledge bases in Retrieval-Augmented Generation (RAG) systems, enabling timely detection of 'out-of-knowledge-boundary' queries and identification of query distribution drift;
DocCGen: Document-based Controlled Code Generation
Sameer Pimparkhede (IIT Bombay), Pushpak Bhattacharyya (IBM Research)
AI Code AssistantTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes the DocCGen framework, which decomposes the natural language to code generation task into two steps: retrieving relevant library documents and constrained decoding based on the documents;
DocEdit-v2: Document Structure Editing Via Multimodal LLM Grounding
Manan Suri (University of Maryland), Dinesh Manocha (MBZUAI)
TransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Proposes an end-to-end multi-modal document editing framework, DocEdit-v2, which can locate the editing region based on user natural language requests, generate editing commands, reformulate the commands into formats suitable for large models, and perform localized edits on the HTML+CSS structure of documents using GPT-4V/Gemini.
DocHieNet: A Large and Diverse Dataset for Document Hierarchy Parsing
Hangdi Xing (Zhejiang University), Cong Yao (Alibaba Group)
RecognitionTransformerVision Language ModelTextMultimodalityBenchmark
🎯 What it does: Constructed a large-scale, multi-page, multi-domain, multi-layout, bilingual document hierarchy parsing dataset named DocHieNet, and proposed the DHFormer framework to accomplish document hierarchy parsing.
DocKD: Knowledge Distillation from LLMs for Open-World Document Understanding Models
Sungnyun Kim (KAIST AI), Stefano Soatto (AWS AI Labs)
Knowledge DistillationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Leverage large language models (LLM) combined with external document knowledge to generate high-quality document annotations, which are subsequently used to train a small visual document understanding (VDU) model, achieving open-source world document understanding.
Does Fine-Tuning LLMs on New Knowledge Encourage Hallucinations?
Zorik Gekhman (Technion Israel Institute Of Technology), Jonathan Herzig (Google Research)
Explainability and InterpretabilityLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Investigate the impact of introducing new factual knowledge during supervised fine-tuning on hallucinations in large language models, and propose the SliCK framework for four-level knowledge classification;
Does Large Language Model Contain Task-Specific Neurons?
Ran Song (Kunming University of Science and Technology), Zhengtao Yu (Kunming University of Science and Technology)
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: Investigate and verify the existence of task-specific neurons in LLMs, and propose a method based on causal gradient variation with special tokens to locate them.
Does Object Grounding Really Reduce Hallucination of Large Vision-Language Models?
Gregor Geigle (University of Würzburg), Goran Glavaš (University of Würzburg)
GenerationTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: Systematically evaluate whether incorporating object alignment objectives (Referring Expressions and Grounded Captioning) into large vision-language models (LVLM) for open-ended image captioning can reduce hallucinations.
DogeRM: Equipping Reward Models with Domain Knowledge through Model Merging
Tzu-Han Lin (National Taiwan University), Yun-Nung Chen (National Taiwan University)
AI Code AssistantReinforcement Learning from Human FeedbackTransformerSupervised Fine-TuningText
🎯 What it does: Propose a method called DogeRM that integrates domain knowledge into a reward model by merging a pre-trained reward model with a domain-specific SFT language model through weighted averaging, thereby enhancing the domain performance of the reward model without requiring additional preference data.
Domain adapted machine translation: What does catastrophic forgetting forget and why?
Danielle Saunders (RWS Language Weaver), Steve DeNeefe (RWS Language Weaver)
Domain AdaptationTransformerSupervised Fine-TuningText
🎯 What it does: Studied catastrophic forgetting in neural machine translation during domain adaptation, analyzed the content and causes of forgetting, and proposed a new metric to quantify forgetting at the lexical level.
Don’t Forget Your Reward Values: Language Model Alignment via Value-based Calibration
Xin Mao (Nanyang Technological University), Anh Tuan Luu (SEA Group)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningContrastive LearningText
🎯 What it does: Propose a reward-based calibration method called VCB for directly aligning large language models, avoiding the limitations of traditional sequence-based calibration methods.
Don’t Just Say “I don’t know”! Self-aligning Large Language Models for Responding to Unknown Questions with Explanations
Yang Deng (Singapore Management University), Tat-Seng Chua (National University of Singapore)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Proposed a scalable self-aligned method, Self-Aligned, enabling large language models to identify and provide explanatory answers for unanswerable unknown questions.
Dual-oriented Disentangled Network with Counterfactual Intervention for Multimodal Intent Detection
Zhanpeng Chen (Peking University), Yuexian Zou (Peking University)
ClassificationTransformerContrastive LearningMultimodality
🎯 What it does: Proposed the bidirectional decoupling network DuoDN for multimodal intent detection.
Dual-Space Knowledge Distillation for Large Language Models
Songming Zhang (Beijing Jiaotong University), Jinan Xu (Beijing Jiaotong University)
Knowledge DistillationTransformerLarge Language ModelText
🎯 What it does: Designed and verified a dual-space knowledge distillation framework, DSKD, which unifies the output spaces of teacher and student models and automatically aligns tokens from different vocabularies through cross-model attention, achieving efficient distillation of large language models.
DVD: Dynamic Contrastive Decoding for Knowledge Amplification in Multi-Document Question Answering
Jing Jin (Peking University), Zhijiang Guo (Huawei)
Contrastive LearningText
🎯 What it does: Propose a decoding strategy called DVD that dynamically amplifies selected document knowledge in multi-document question answering.
Dynamic Multi-granularity Attribution Network for Aspect-based Sentiment Analysis
Yanjiang Chen (University of Science and Technology of China), Qi Liu (University of Science and Technology of China)
ClassificationExplainability and InterpretabilityGraph Neural NetworkTransformerLarge Language ModelText
🎯 What it does: Proposed a Dynamic Multi-Granularity Attribution Network (DMAN), which leverages Integrated Gradients to extract multi-step attribution information, integrates token-level and span-level attributions, and dynamically focuses on syntactic graphs to enhance Aspect-based Sentiment Analysis (ABSA) performance.
Dynamic Multi-Reward Weighting for Multi-Style Controllable Generation
Karin De Langis, Dongyeop Kang (University of Minnesota)
GenerationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Fine-tune LLaMA2 7B using reinforcement learning (PPO), achieving controllable multi-style text generation through reward signals generated by multiple style discriminators (sentiment, formality, sarcasm, emotion, toxicity).
Dynamic Rewarding with Prompt Optimization Enables Tuning-free Self-Alignment of Language Models
Somanshu Singla (University Of California San Diego), Eric P. Xing (Mohamed Bin Zayed University Of Artificial Intelligence)
OptimizationReinforcement Learning from Human FeedbackTransformerPrompt EngineeringText
🎯 What it does: Proposed the Dynamic Reward and Prompt Optimization (DRPO) method, achieving self-alignment without additional training, significantly enhancing the performance of large language models (LLMs) on alignment tasks.
DynamicER: Resolving Emerging Mentions to Dynamic Entities for RAG
Jinyoung Kim (Seoul National University), Gunhee Kim (Seoul National University)
RetrievalTransformerTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposes the DYNAMICER benchmark to evaluate the identification of emerging entity mentions in a continuously updated knowledge base and their alignment to dynamic entities; and designs a Temporal Segmented Clustering with Continuous Adaptation (TempCCA) method based on time-segmented clustering and continuous adaptation to address novel mentions;
DynaThink: Fast or Slow? A Dynamic Decision-Making Framework for Large Language Models
Jiabao Pan (Cambridge University), Haizhou Li (Chinese University of Hong Kong)
Computational EfficiencyTextBenchmarkChain-of-Thought
🎯 What it does: Proposed and implemented the DynaThink framework, enabling large language models to dynamically select 'fast thinking' or 'slow thinking' modes during reasoning tasks, thereby optimizing resource consumption while maintaining accuracy.
DyVo: Dynamic Vocabularies for Learned Sparse Retrieval with Entities
Thong Nguyen (University of Amsterdam), Andrew Yates (University of Amsterdam)
RetrievalTransformerLarge Language ModelTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: This paper proposes the DyVo model, which integrates Wikipedia entity embeddings into the vocabulary by dynamically incorporating entity candidates through a dynamic vocabulary head within the Learned Sparse Retrieval (LSR) framework, enabling both queries and document representations to simultaneously include word fragments and entities.
EAGLE-2: Faster Inference of Language Models with Dynamic Draft Trees
Yuhui Li (Peking University), Hongyang Zhang (University of Waterloo)
Computational EfficiencyTransformerLarge Language ModelTextBenchmark
🎯 What it does: Propose EAGLE-2, a lossless inference acceleration method that utilizes a context-aware dynamic draft tree.
ECCO: Can We Improve Model-Generated Code Efficiency Without Sacrificing Functional Correctness?
Siddhant Waghjale (Carnegie Mellon University), Daniel Fried (Carnegie Mellon University)
AI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: Proposes the ECCO benchmark to evaluate the ability of large language models to generate efficient Python code while maintaining functional correctness; simultaneously implements a reproducible cloud execution platform, JUDGE0, to stably measure runtime and memory consumption.
ECIS-VQG: Generation of Entity-centric Information-seeking Questions from Videos
Arpan Phukan (IIT Patna), Asif Ekbal (IIT Jodhpur)
GenerationTransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningVideoTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: This study proposes the task of generating entity-centric information retrieval questions (ECIS) from videos and constructs a new dataset called VIDEOQUESTIONS;
ECON: On the Detection and Resolution of Evidence Conflicts
Cheng Jiayang (Hong Kong University of Science and Technology), Zheng Zhang (Amazon AWS AI)
RetrievalTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: To address inter-evidence conflicts in information retrieval, this paper proposes a process for automatically generating high-quality, verifiable conflicting evidence pairs using large language models, along with evaluations of this process;
Effective Demonstration Annotation for In-Context Learning via Language Model-Based Determinantal Point Process
Peng Wang (Zhejiang University), Yong Jiang (Alibaba Group)
Data-Centric LearningMeta LearningTransformerLarge Language ModelText
🎯 What it does: This paper proposes a selective annotation strategy called LM-DPP for few-shot learning scenarios in large language models (LLMs), aiming to select examples with both low uncertainty and high diversity under a limited annotation budget to reduce manual labeling efforts;
Effective Synthetic Data and Test-Time Adaptation for OCR Correction
Shuhao Guan (University College Dublin), Derek Greene (University College Dublin)
RecognitionData SynthesisDomain AdaptationTransformerLarge Language ModelText
🎯 What it does: Utilize weakly supervised generation of synthetic data with multiple noise levels and improve OCR post-processing character error rate through self-correcting noise adaptation during test time (SCN-TTA).
Efficient LLM Comparative Assessment: A Product of Experts Framework for Pairwise Comparisons
Adian Liusie (University of Cambridge), Mark Gales
Computational EfficiencyLarge Language ModelMixture of ExpertsText
🎯 What it does: This paper views LLM contrastive evaluation as a Product of Experts (PoE) framework, deriving scores for candidate texts using contrastive probability information, and provides an efficient closed-form solution and greedy comparison selection strategy.
Efficient Overshadowed Entity Disambiguation by Mitigating Shortcut Learning
Panuthep Tasawong (VISTEC), Sarana Nutanong (VISTEC)
ClassificationTransformerText
🎯 What it does: This paper proposes an entity disambiguation method based on adversarial counterfactual training (CFT), which uses counterfactual samples generated by masking the surface features of entities in the input to suppress the model's 'shortcut learning' reliance on surface characteristics, thereby improving disambiguation performance for shadowed entities.
Efficient Performance Tracking: Leveraging Large Language Models for Automated Construction of Scientific Leaderboards
Furkan Şahinuç (Technical University of Darmstadt), Iryna Gurevych (Technical University of Darmstadt)
Data-Centric LearningTransformerLarge Language ModelTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: This paper proposes an automated scientific leaderboard construction framework based on large language models and builds a new manually annotated dataset called SCILEAD;
Efficient Sequential Decision Making with Large Language Models
Dingyang Chen (University Of South Carolina), Yinglun Zhu
TransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: This paper proposes an efficient method to integrate large language models (LLMs) into the contextual bandit decision process, utilizing an online model selection algorithm to dynamically balance the usage of traditional contextual bandits and LLM-driven strategies;
Efficient Temporal Extrapolation of Multimodal Large Language Models with Temporal Grounding Bridge
Yuxuan Wang (Beijing Institute for General Artificial Intelligence), Zilong Zheng (Beijing Institute for General Artificial Intelligence)
RetrievalComputational EfficiencyTransformerLarge Language ModelVision Language ModelOptical FlowVideoMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose a framework named Temporal Grounding Bridge (TGB), which integrates multimodal large language models (MLLM) with an efficient time normalization mechanism, enabling time normalization, keyframe extraction, and alignment with language queries for long videos without expanding the pre-trained context window, thereby enhancing performance in long-video question answering and temporal localization tasks.
Efficient Unseen Language Adaptation for Multilingual Pre-Trained Language Models
Po-Heng Chen (National Taiwan University), Yun-Nung Chen (National Taiwan University)
ClassificationDomain AdaptationComputational EfficiencyTransformerPrompt EngineeringText
🎯 What it does: Proposed an efficient soft prompt tuning method to adapt multilingual pre-trained language models (mPLMs) for zero-shot cross-lingual transfer tasks on unseen languages.
Efficient Vision-Language pre-training via domain-specific learning for human activities
Adrian Bulat (Samsung AI Cambridge), Georgios Tzimiropoulos (Samsung AI Cambridge)
RecognitionRetrievalRepresentation LearningData-Centric LearningTransformerLarge Language ModelVision Language ModelContrastive LearningImageVideoTextRetrieval-Augmented Generation
🎯 What it does: A domain-aligned vision-language pre-training method is proposed for known downstream task domains (human activities), which automatically generates activity concept hierarchies and constructs queries using LLM, retrieves and filters samples from LAION-5B, re-annotates original captions with BLIP-2, and forms the Human-Centric Image-Text (HC-IT) dataset; a CLIP-style HC-VL model is trained on this dataset, and further enhanced with a minimal temporal attention module to obtain HC-VL+; significant improvements are demonstrated on zero-shot image/video retrieval, action recognition, and few-shot action recognition tasks.
EfficientRAG: Efficient Retriever for Multi-Hop Question Answering
Ziyuan Zhuang (Nanjing University), Qi Zhang (Microsoft)
RetrievalTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: Propose EfficientRAG, a multi-hop question answering retrieval framework that does not require multiple calls to large language models (LLMs), generating new queries iteratively and filtering irrelevant knowledge blocks through lightweight Labeler and Filter modules.
EFUF: Efficient Fine-Grained Unlearning Framework for Mitigating Hallucinations in Multimodal Large Language Models
Shangyu Xing (Nanjing University), Xinyu Dai (Nanjing University)
Large Language ModelVision Language ModelContrastive LearningMultimodality
🎯 What it does: Proposed an EFUF framework that utilizes unpaired data and gradient ascent to mitigate hallucinations in multi-modal large language models.
EH-MAM: Easy-to-Hard Masked Acoustic Modeling for Self-Supervised Speech Representation Learning
Ashish Seth (University of Maryland, College Park), Dinesh Manocha (University of Maryland, College Park)
Representation LearningTransformerAudio
🎯 What it does: Propose EH-MAM, a self-supervised speech representation learning framework that automatically identifies audio frames difficult to reconstruct through a loss predictor in the teacher network, and gradually masks them during training to create more challenging pre-training tasks.
EHRAgent: Code Empowers Large Language Models for Few-shot Complex Tabular Reasoning on Electronic Health Records
Wenqi Shi (Georgia Institute of Technology), May Dongmei Wang (Georgia Institute of Technology)
TransformerLarge Language ModelAgentic AITabularBiomedical DataElectronic Health RecordsRetrieval-Augmented Generation
🎯 What it does: Propose EHRAgent, an LLM agent capable of performing few-shot multi-table reasoning on electronic health records (EHR), which can automatically generate, execute, and debug code, enabling clinicians to directly interact in natural language to retrieve complex patient information.
Eliciting In-Context Learning in Vision-Language Models for Videos Through Curated Data Distributional Properties
Keunwoo Peter Yu (University of Michigan), Joyce Chai (University of Michigan)
Data-Centric LearningTransformerLarge Language ModelVision Language ModelVideoText
🎯 What it does: Developed an incremental contextual learning training paradigm called EILeV for video and text, utilizing distributed attributes to achieve few-shot learning for video description.
Eliminating Biased Length Reliance of Direct Preference Optimization via Down-Sampled KL Divergence
Junru Lu (University of Warwick), Xing Sun (Tencent YouTu Lab)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: This paper investigates the verbosity issue in Direct Preference Optimization (DPO) when generating text and proposes an improved method called SamPO. SamPO eliminates algorithmic bias toward response length by undersampling the KL divergence of DPO, thereby reducing the model's output verbosity and improving alignment.
Embedded Named Entity Recognition using Probing Classifiers
Nicholas Popovic, Michael Färber (TU Dresden)
RecognitionTransformerLarge Language ModelText
🎯 What it does: Propose EMBER, which achieves zero-shot streaming named entity recognition by leveraging the internal representations of pre-trained decoder language models and a probe classifier;
Embedding and Gradient Say Wrong: A White-Box Method for Hallucination Detection
Xiaomeng Hu (ZheJiang University), Junbo Zhao (Huawei Technologies Co., Ltd)
Anomaly DetectionExplainability and InterpretabilityRepresentation LearningLarge Language ModelTextBenchmark
🎯 What it does: Propose a hallucination detection method called EGH based on white-box information, which uses embedding differences and gradient features of language models to determine whether the answer depends on the input;
EmoKnob: Enhance Voice Cloning with Fine-Grained Emotion Control
Haozhe Chen (Columbia University), Julia Hirschberg (Columbia University)
GenerationRetrievalTransformerLarge Language ModelSupervised Fine-TuningTextRetrieval-Augmented GenerationAudio
🎯 What it does: Building upon existing base voice cloning models, we extract emotion vectors using a few examples and perform fine-tuning in the acoustic encoding space to achieve fine-grained control over arbitrary emotions. Meanwhile, we support voice synthesis with open-text descriptions of emotions by leveraging synthetic text generated by LLMs and retrieved corpora.
Emotion Granularity from Text: An Aggregate-Level Indicator of Mental Health
Krishnapriya Vishnubhotla (University of Toronto), Saif M. Mohammad (National Research Council Canada)
Text
🎯 What it does: This paper constructs user emotion time series (emotion arcs) from social media texts such as tweets and Reddit posts, calculates the emotion granularity (EG) metric by computing the Spearman correlation between emotion pairs, and compares differences between various mental health condition (MHC) groups and the control group on this metric.
EmphAssess : a Prosodic Benchmark on Assessing Emphasis Transfer in Speech-to-Speech Models
Maureen de Seyssel (Meta AI Research), Emmanuel Dupoux (Meta AI Research)
ClassificationRecognitionTransformerSupervised Fine-TuningTextBenchmarkAudio
🎯 What it does: Proposed and implemented the EmphAssess evaluation framework for automatically assessing the performance of speech-to-speech (S2S) models in preserving and transferring prosody.
Empowering Backbone Models for Visual Text Generation with Input Granularity Control and Glyph-Aware Training
Wenbo Li (Xiamen University), Jinsong Su (Baidu Inc.)
GenerationTransformerVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: Improve visual text generation in diffusion models by introducing hybrid granularity input and three glyph-aware losses;
Empowering Large Language Model for Continual Video Question Answering with Collaborative Prompting
Chen Cai (Nanyang Technological University), Kim-Hui Yap (Nanyang Technological University)
RetrievalKnowledge DistillationMeta LearningTransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningMultimodality
🎯 What it does: The study investigates multi-modal video question answering under a continual learning framework, proposing Collaborative Prompting (ColPro) to help large language models retain knowledge and avoid catastrophic forgetting across multiple tasks.
Empowering Multi-step Reasoning across Languages via Program-Aided Language Models
Leonardo Ranaldi (University of Edinburgh), Alexandra Birch (University of Edinburgh)
Large Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Proposed Cross-PAL and its self-consistent version (S-Cross-PAL), enabling large language models to perform multi-step reasoning in multilingual environments through two-stage cross-lingual prompting.
Encoding and Controlling Global Semantics for Long-form Video Question Answering
Thong Thanh Nguyen (National University of Singapore), Anh Tuan Luu (Nanyang Technological University)
TransformerVision Language ModelContrastive LearningVideoTextMultimodalityBenchmark
🎯 What it does: Propose a multi-modal Transformer with a gated state space layer (GSMT) to integrate global semantic information in long video question answering, addressing the information loss caused by traditional frame/region selection methods.
Encoding Spreadsheets for Large Language Models
Haoyu Dong (Microsoft Corporation), Dongmei Zhang (Microsoft Corporation)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextTabularChain-of-Thought
🎯 What it does: Propose the SHEETENCODER framework, which compresses large tables through structured encoding and inputs them into LLMs, achieving efficient understanding and reasoning of table layouts and structures;
Encourage or Inhibit Monosemanticity? Revisit Monosemanticity from a Feature Decorrelation Perspective
Hanqi Yan (King's College London), Yulan He (King's College London)
Explainability and InterpretabilityRepresentation LearningReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Explores the significance of monosemanticity in large language models and enhances model alignment performance by incorporating feature decorrelation regularization (DecPO) into DPO training.
Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate
Tian Liang (Tsinghua University), Zhaopeng Tu (Tencent AI Lab)
TransformerLarge Language ModelAgentic AITextBenchmarkChain-of-Thought
🎯 What it does: Proposed and implemented the Multi-Agent Debate (MAD) framework, which mitigates the Degeneration-of-Thought (DoT) problem in large language model self-reflection by using two debaters and a judge through multi-round debates and an adaptive termination mechanism.
Enhanced Hallucination Detection in Neural Machine Translation through Simple Detector Aggregation
Anas Himmi (Universite Paris-Saclay), Nuno M Guerreiro
GenerationAnomaly DetectionText
🎯 What it does: Propose a simple unsupervised aggregation method called STARE, which sums the normalized scores of multiple hallucination detectors to enhance hallucination detection performance in neural machine translation.
Enhancing Advanced Visual Reasoning Ability of Large Language Models
Zhiyuan Li (University of Sydney), Weidong Cai (University of Sydney)
Large Language ModelVision Language ModelImageTextMultimodalityBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose CVR-LLM, which combines VLM with LLM. First, images are converted into text through context-aware image description (CaID) with dual-cycle self-refinement, then complex visual reasoning is performed using multimodal ICL and LLM.
Enhancing AI Assisted Writing with One-Shot Implicit Negative Feedback
Benjamin Towle (University of Nottingham), Ke Zhou (University of Nottingham)
GenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose NIFTY, which leverages implicit negative feedback from users rejecting suggestions in intelligent reply systems, enhancing the generation quality of AI writing models through classifier guidance.
Enhancing Data Quality through Simple De-duplication: Navigating Responsible Computational Social Science Research
Yida Mu (University of Sheffield), Nikolaos Aletras (University of Sheffield)
Data-Centric LearningTransformerLarge Language ModelText
🎯 What it does: Detect and evaluate duplicate/near-duplicate samples in 20 social media NLP datasets, and analyze their impact on model performance.
Enhancing High-order Interaction Awareness in LLM-based Recommender Model
Xinfeng Wang (University of Yamanashi), Yoshimi Suzuki (University of Yamanashi)
Recommendation SystemGraph Neural NetworkTransformerLarge Language ModelPrompt EngineeringTextGraph
🎯 What it does: This paper proposes an enhanced LLM recommendation model, ELMRec, which improves recommendation performance by integrating graph structural information and re-ranking methods.
Enhancing Language Model Alignment: A Confidence-Based Approach to Label Smoothing
Baihe Huang (University of California Berkeley), Yi Mao (University of California Berkeley)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Propose the Confidence Aware Label Smoothing (CALS) method to dynamically adjust the label smoothing parameter in Direct Preference Optimization (DPO) based on label confidence;
Enhancing Language Model Factuality via Activation-Based Confidence Calibration and Guided Decoding
Xin Liu (University of Michigan), Lu Wang (University of Michigan)
Explainability and InterpretabilityTransformerLarge Language ModelTextBenchmark
🎯 What it does: Proposed ACTCAB, an activation-based language model calibration method, and CODEC, a confidence-based decoding strategy, to enhance the factual accuracy of large language models.
Enhancing Legal Case Retrieval via Scaling High-quality Synthetic Query-Candidate Pairs
Cheng Gao (Tsinghua University), Maosong Sun (Tsinghua University)
Data SynthesisRetrievalLarge Language ModelText
🎯 What it does: This paper uses large language models to automatically extract key information from criminal verdict texts, generate short anonymized queries, and combines knowledge-driven data augmentation to construct the largest Chinese legal case retrieval dataset, LEAD.
Enhancing Post-Hoc Attributions in Long Document Comprehension via Coarse Grained Answer Decomposition
Pritika Ramu (Adobe Research), Balaji Vasan Srinivasan (Adobe Research)
RetrievalExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a post-hoc attribution method for long document question answering, which identifies factual fragments in the answer requiring attribution through question context-guided coarse-grained answer decomposition (CoG), and maps these fragments to evidence sentences in the original document using a retriever or large language model.
Enhancing Pre-Trained Generative Language Models with Question Attended Span Extraction on Machine Reading Comprehension
Lin Ai (Columbia University), Julia Hirschberg (Columbia University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: In this paper, we propose the QASE (Question-Attended Span Extraction) module, which significantly enhances the performance of pre-trained generative language models on extractive machine reading comprehension (MRC) tasks during the fine-tuning phase, enabling the generation of more accurate, contextually consistent, and factually coherent answers.
Enhancing Reinforcement Learning with Dense Rewards from Language Model Critic
Meng Cao (McGill University), Lei Meng (Google Deepmind)
TransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: Proposes using a large language model (LLM) as a critic to generate dense intermediate step rewards (Intrinsic Rewards) for reinforcement learning (RL) training, addressing the issues of sparse rewards and credit assignment in text generation tasks.
Enhancing Systematic Decompositional Natural Language Inference Using Informal Logic
Nathaniel Weir (Johns Hopkins University), Benjamin Van Durme (Johns Hopkins University)
Explainability and InterpretabilityComputational EfficiencyKnowledge DistillationData-Centric LearningTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Propose a decompositional textual entailment (RDTE) evaluation protocol based on informal logic and construct a high-quality decompositional entailment dataset with 1000 instances; train RoBERTa and ChatGPT student models using silver data generated by GPT-4 through knowledge distillation; embed these models into a new entailment tree generation engine called TREEWISE, improving the accuracy and interpretability of multi-hop reasoning.
Enhancing Training Data Attribution for Large Language Models with Fitting Error Consideration
Kangxi Wu (Key Laboratory of AI Safety, Chinese Academy of Sciences Institute of Computing Technology), Xueqi Cheng (Key Laboratory of AI Safety, Chinese Academy of Sciences Institute of Computing Technology)
Explainability and InterpretabilityLarge Language ModelContrastive LearningTextBenchmark
🎯 What it does: For the training data attribution task in large-scale language models, the Debias and Denoise Attribution (DDA) method is proposed, improving traditional influence functions to reduce the impact of training errors on attribution results.
Entity Insertion in Multilingual Linked Corpora: The Case of Wikipedia
Tomás Feith (École Polytechnique Fédérale de Lausanne), Robert West (École Polytechnique Fédérale de Lausanne)
RetrievalTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Proposed and addressed the novel task of 'entity insertion,' identifying the most suitable text segments to insert missing entity links in a multilingual Wikipedia environment.
EPO: Hierarchical LLM Agents with Environment Preference Optimization
Qi Zhao (Brown University), George Konidaris (Brown University)
OptimizationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningAgentic AIVision Language ModelVision-Language-Action ModelImageTextMultimodalityBenchmark
🎯 What it does: Propose a hierarchical large language model framework that utilizes subgoal decomposition and low-level action generation to address long-horizon decision-making problems, and automatically generates reward signals through environmental feedback;
Error Analysis of Multilingual Language Models in Machine Translation: A Case Study of English-Amharic Translation
Hizkiel Mitiku Alemayehu, Axel-Cyrille Ngonga Ngomo (Paderborn University)
GenerationTransformerLarge Language ModelText
🎯 What it does: This paper evaluates the performance of multilingual large language models in English-Amharic bidirectional translation and conducts a systematic analysis of translation errors.
ERVQA: A Dataset to Benchmark the Readiness of Large Vision Language Models in Hospital Environments
Sourjyadip Ray (Indian Institute of Technology Kharagpur), Pawan Goyal (Indian Institute of Technology Kharagpur)
Vision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Construct the ERVQA dataset and benchmark the performance of large vision-language models in hospital environments for visual question answering.