EMNLP 2024 Papers — Page 2
Conference on Empirical Methods in Natural Language Processing · 1268 papers
AppBench: Planning of Multiple APIs from Various APPs for Complex User Instruction
Hongru Wang (Chinese University of Hong Kong), Kam-Fai Wong (Chinese University of Hong Kong)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: This paper constructs the AppBench benchmark to evaluate large language models (LLMs) in planning and executing multiple API calls from different apps to satisfy complex user instructions, covering graphical execution order and permission restrictions.
APPLS: Evaluating Evaluation Metrics for Plain Language Summarization
Yue Guo (University of Illinois Urbana-Champaign), Lucy Lu Wang (University of Washington)
GenerationTransformerLarge Language ModelTextBiomedical DataBenchmark
🎯 What it does: Constructed the APPLS meta-evaluation test platform, systematically perturbing four evaluation dimensions (information, simplification, coherence, faithfulness) in the plain language summary (PLS) task, and used this platform to evaluate 14 existing evaluation metrics;
Applying Contrastive Learning to Code Vulnerability Type Classification
Chen Ji (Hangzhou Institute of Technology), Yuqing Zhang (Hangzhou Institute of Technology)
ClassificationTransformerContrastive LearningText
🎯 What it does: Propose a hierarchical contrastive learning framework for multi-class code vulnerability type identification, combining geometric diffusion and max pooling techniques to enhance representation quality.
Applying Intrinsic Debiasing on Downstream Tasks: Challenges and Considerations for Machine Translation
Bar Iluz (Hebrew University of Jerusalem), Gabriel Stanovsky (Hebrew University of Jerusalem)
GenerationRepresentation LearningTransformerText
🎯 What it does: This paper systematically evaluates the effectiveness of three intrinsic debiasing techniques (Hard-Debiasing, INLP, LEACE) in neural machine translation (NMT), exploring the impact of debiased embedding layers, tokenization strategies, and differences in target languages.
Are Data Augmentation Methods in Named Entity Recognition Applicable for Uncertainty Estimation?
Wataru Hashimoto (Nara Institute of Science and Technology), Taro Watanabe (Nara Institute of Science and Technology)
RecognitionExplainability and InterpretabilityTransformerText
🎯 What it does: This paper investigates the impact of data augmentation on confidence calibration and uncertainty estimation in named entity recognition (NER), systematically evaluating the effectiveness of various augmentation methods in cross-domain and cross-lingual scenarios.
Are Large Language Models Capable of Generating Human-Level Narratives?
Yufei Tian (University of California Los Angeles), Nanyun Peng
GenerationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Examined the capabilities of large language models in narrative generation and understanding, and proposed a quantitative analysis framework based on story arcs, plot points, and emotional dimensions.
Are Large Language Models Good Classifiers? A Study on Edit Intent Classification in Scientific Document Revisions
Qian Ruan (Technical University of Darmstadt), Iryna Gurevych (Technical University of Darmstadt)
ClassificationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Built and evaluated a large language model (LLM) fine-tuning framework tailored for classification tasks, with a focus on edit intent classification (EIC) in scientific document revisions, and used this framework to automatically annotate and generate a significantly larger Re3-Sci2.0 dataset.
Are Large Language Models In-Context Personalized Summarizers? Get an iCOPERNICUS Test Done!
Divya Patel (KDMLab, Dhirubhai Ambani Institute of Information and Communication Technology), Tanmoy Chakraborty (Indian Institute of Technology Delhi)
TransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Designed and evaluated the iCOPERNICUS framework to examine the In-Context Personalization Learning (ICPL) capability of large language models (LLMs) in context-aware personalized summarization tasks.
Are LLMs Good Zero-Shot Fallacy Classifiers?
Fengjun Pan (Nanyang Technological University), Anh Tuan Luu (Nanyang Technological University)
ClassificationTransformerPrompt EngineeringTextChain-of-Thought
🎯 What it does: This paper studies the use of large language models (LLMs) for zero-shot fallacy classification, proposing single-round and multi-round prompting strategies and conducting large-scale experiments.
ARES: Alternating Reinforcement Learning and Supervised Fine-Tuning for Enhanced Multi-Modal Chain-of-Thought Reasoning Through Diverse AI Feedback
Ju-Seung Byun (Ohio State University), Andrew Perrault (Ohio State University)
Reinforcement Learning from Human FeedbackLarge Language ModelSupervised Fine-TuningReinforcement LearningMultimodalityChain-of-Thought
🎯 What it does: A two-stage algorithm alternates between reinforcement learning (RL) and supervised fine-tuning (SFT), leveraging sentence-level scoring and correction feedback from an AI teacher to enhance the quality of multi-modal chain-of-thought reasoning.
Argument Relation Classification through Discourse Markers and Adversarial Training
Michele Luca Contalbo (University of Modena and Reggio Emilia), Matteo Paganelli (University of Modena and Reggio Emilia)
ClassificationRecognitionAdversarial AttackTransformerSupervised Fine-TuningText
🎯 What it does: In the Argument Relation Classification (ARC) task, the DISARM model is proposed by jointly training argument claim identification with the ARC task and using adversarial learning to align the representation spaces of the two tasks;
ARM: An Alignment-and-Replacement Module for Chinese Spelling Check Based on LLMs
Changchun Liu, Enhong Chen (China University of Mining)
TransformerLarge Language ModelText
🎯 What it does: Designed and implemented an ARM module that leverages LLM for alignment and replacement, enhancing the detection and correction performance of the Chinese Spelling Correction (CSC) system.
ArMeme: Propagandistic Content in Arabic Memes
Firoj Alam (Qatar Computing Research Institute), Maram Hasanain (University of New Brunswick)
ClassificationConvolutional Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningMultimodalityBenchmark
🎯 What it does: This paper constructs the first dataset containing approximately 6,000 Arabic internet memes annotated with propagandistic content, exploring the multimodal propaganda detection task;
ArxivDIGESTables: Synthesizing Scientific Literature into Tables using Language Models
Benjamin Newman (University of Washington), Kyle Lo (University of Washington)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringTextTabularBenchmark
🎯 What it does: Constructed a large-scale, high-quality literature review table dataset called ARXIVDIGESTABLES, and proposed an automatic evaluation framework named DECONTEXTEVAL to measure the similarity between tables automatically generated by language models and manually written tables.
ASETF: A Novel Method for Jailbreak Attack on LLMs through Translate Suffix Embeddings
Hao Wang (Beihang University), Lei Sha (Beihang University)
Computational EfficiencyAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed the ASETF framework, which first performs gradient optimization on the attack suffix in the continuous word embedding space, then converts the optimized vectors into coherent, readable text suffixes through an embedding translation model, thereby achieving 'jailbreak' attacks on LLMs.
ASL STEM Wiki: Dataset and Benchmark for Interpreting STEM Articles
Kayo Yin (University of California, Berkeley), Danielle Bragg (Microsoft Research)
RecognitionData-Centric LearningGraph Neural NetworkTransformerContrastive LearningVideoTextBenchmark
🎯 What it does: Constructed and released the ASL STEM Wiki dataset, containing 254 STEM Wikipedia articles translated into English-to-ASL continuous videos (totaling 315.84 hours), along with benchmark experiments for finger spelling detection and alignment.
Assessing “Implicit” Retrieval Robustness of Large Language Models
Xiaoyu Shen (Digital Twin Institute, Eastern Institute of Technology), Wei Zhang (Digital Twin Institute, Eastern Institute of Technology)
RetrievalTransformerSupervised Fine-TuningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Evaluate the retrieval robustness of large language models under implicit retrieval enhancement (without explicitly judging the relevance of retrieved contexts) and improve robustness through fine-tuning methods.
Assessing and Verifying Task Utility in LLM-Powered Applications
Negar Arabzadeh (University of Waterloo), Julia Kiseleva (Microsoft Research)
TransformerLarge Language ModelAgentic AIText
🎯 What it does: Built and evaluated the AgentEval framework, using three types of LLM agents (CriticAgent, QuantifierAgent, VerifierAgent) to automatically generate evaluation criteria, quantify application performance, and verify the robustness of the criteria.
AssistantBench: Can Web Agents Solve Realistic and Time-Consuming Tasks?
Ori Yoran (Tel Aviv University), Jonathan Berant (Tel Aviv University)
TransformerLarge Language ModelAgentic AIPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: This paper proposes the ASSISTANTBENCH benchmark, which includes 214 real and time-consuming web tasks, and designs a new SPA (SeePlan-Act) web agent to address these tasks.
ATAP: Automatic Template-Augmented Commonsense Knowledge Graph Completion via Pre-Trained Language Models
Fu Zhang (Northeastern University), Jingwei Cheng (Northeastern University)
Representation LearningData-Centric LearningRecurrent Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringGraph
🎯 What it does: This study proposes the ATAP framework, which combines automatically generated continuous prompts with pre-trained language models (PLMs) to jointly complete the commonsense knowledge graph (CKG) completion task (chain prediction), addressing the low efficiency of traditional manual/discrete prompts and their poor performance on long-tail entities.
ATM: Adversarial Tuning Multi-agent System Makes a Robust Retrieval-Augmented Generator
Junda Zhu (Beihang University), Lei Sha (Beihang University)
GenerationAdversarial AttackData-Centric LearningTransformerSupervised Fine-TuningAgentic AITextRetrieval-Augmented Generation
🎯 What it does: Propose an adversarial tuning multi-agent system (ATM), which enhances the robustness and generation quality of retrieval-augmented generation models when facing noisy retrieval results through alternating iterative training between an attacker and a generator.
Atomic Inference for NLI with Generated Facts as Atoms
Joe Stacey (Imperial College London), Marek Rei (Imperial College London)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: This paper proposes an atomic reasoning framework that utilizes facts generated by LLMs as atomic units for natural language inference. The model is trained to make entailment/contradiction predictions at the fact level without using fact labels, and instance-level predictions are obtained through deterministic rules.
Atomic Self-Consistency for Better Long Form Generations
Raghuveer Thirukovalluru (Duke University), Bhuwan Dhingra (Duke University)
GenerationLarge Language ModelContrastive LearningText
🎯 What it does: Propose the Atomic Self-Consistency (ASC) method, which generates more complete and higher recall long-text answers by clustering, filtering, and merging atomic facts from multiple generated texts.
Attention Score is not All You Need for Token Importance Indicator in KV Cache Reduction: Value Also Matters
Zhiyu Guo (Nara Institute of Science and Technology), Taro Watanabe (Nara Institute of Science and Technology)
Computational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Propose a Token Pruning method that simultaneously utilizes attention scores and value vector norms to evaluate token importance in LLM KV cache compression
Attribute Diversity Determines the Systematicity Gap in VQA
Ian Berlot-Attwell (University of Toronto Vector Institute), Naomi Saphra (Kempner Institute at Harvard University)
TransformerVision Language ModelMultimodalityBenchmark
🎯 What it does: This paper systematically investigates the systematicity gap of visual question answering models on unseen attribute combinations by constructing the CLEVR-HOPE diagnostic dataset.
Attribute or Abstain: Large Language Models as Long Document Assistants
Jan Buchmann (Technical University of Darmstadt), Iryna Gurevych (Technical University of Darmstadt)
RetrievalTransformerLarge Language ModelTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Constructed the LAB benchmark and investigated the citation/reasoning capabilities of LLMs in long document contexts.
AudioVSR: Enhancing Video Speech Recognition with Audio Data
Xiaoda Yang (Zhejiang University), Tao Jin (Zhejiang University)
RecognitionData SynthesisTransformerSupervised Fine-TuningContrastive LearningVideoMultimodalityAudio
🎯 What it does: Utilize audio data for data augmentation (generating videos corresponding to speech) and train a cross-lingual Visual Speech Recognition model, AudioVSR, achieving VSR in zero-shot, full-sample, and cross-lingual scenarios.
Automated Essay Scoring: A Reflection on the State of the Art
Shengjie Li (University of Texas at Dallas), Vincent Ng (University of Texas at Dallas)
Explainability and InterpretabilityLarge Language ModelTextReview/Survey Paper
🎯 What it does: Systematically reflect on the current state of the field of automatic essay scoring (AES) and propose seven recommendations for future research directions.
Automatic Instruction Evolving for Large Language Models
Weihao Zeng (Microsoft), Weizhu Chen (Microsoft)
OptimizationData-Centric LearningTransformerLarge Language ModelTextBenchmark
🎯 What it does: Propose Auto Evol-Instruct, which utilizes large language models to automatically design, analyze, and optimize instruction evolution methods, achieving instruction dataset evolution without human intervention.
Automatic sentence segmentation of clinical record narratives in real-world data
Dongfang Xu (Cedars Sinai Medical Center), Graciela Gonzalez Hernandez (Cedars Sinai Medical Center)
SegmentationTransformerLarge Language ModelTextBiomedical DataElectronic Health RecordsBenchmark
🎯 What it does: Propose a sentence segmentation method based on BERT sequence labeling combined with a dynamic sliding window, and manually annotate 90 clinical notes from MIMIC-III to create the first clinical text sentence segmentation dataset.
Automatically Generated Definitions and their utility for Modeling Word Meaning
Francesco Periti (University of Milan), Nina Tahmasebi (University of Gothenburg)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This study fine-tunes Llama 2-chat and Llama 3-instruct through instruction tuning to construct a model called LlamaDictionary, which can generate lexical definitions. The generated definitions are used as intermediate representations to evaluate the model's performance on tasks such as lexical generation, lexical context recognition, lexical induction, and lexical change detection.
AutoPersuade: A Framework for Evaluating and Explaining Persuasive Arguments
Till Raphael Saenger (Princeton University), Brandon M. Stewart (Princeton University)
Explainability and InterpretabilityLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes the AutoPersuade three-step workflow: ① Collect diverse persuasive arguments and audience feedback; ② Build a supervised semi-non-negative matrix factorization (SUN) topic model to extract latent features that explain persuasiveness; ③ Use the model to predict the persuasiveness of new arguments and estimate the causal effects of these features; and validate the process in a case study on vegetarianism.
Autoregressive Multi-trait Essay Scoring via Reinforcement Learning with Scoring-aware Multiple Rewards
Heejin Do (POSTECH), Gary Lee (POSTECH)
ClassificationTransformerReinforcement LearningText
🎯 What it does: Propose a reinforcement learning-based automatic writing scoring method called SaMRL, which directly optimizes scoring quality using multiple rewards (QWK and MSE) within a self-attention multi-feature scoring framework.
Autoregressive Pre-Training on Pixels and Texts
Yekun Chai (Baidu), Hua Wu (Baidu)
Representation LearningTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: Propose two pixel-based autoregressive pre-training models, PixelGPT and DualGPT, which are pre-trained on visual-text images and text modalities respectively, and explore the effects of joint pre-training across both modalities.
AutoScraper: A Progressive Understanding Web Agent for Web Scraper Generation
Wenhao Huang (Fudan University), Zulong Chen (Alibaba Holding-Aicheng Technology-Enterprise)
GenerationAI Code AssistantTransformerLarge Language ModelAgentic AIText
🎯 What it does: This paper proposes the AUTOSCRAPER framework, which generates executable web scrapers via LLM to support unified extraction across multiple web pages.
Back to School: Translation Using Grammar Books
Jonathan Hus (George Mason University), Antonios Anastasopoulos (George Mason University)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Improve machine translation for 16 low-resource languages by incorporating a dictionary, grammar book, and a small number of parallel sentences into the prompt using GPT-4-turbo.
Backward Lens: Projecting Language Model Gradients into the Vocabulary Space
Shahar Katz (Technion Israel Institute of Technology), Lior Wolf (Tel Aviv University)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Proposed a method to project the gradients of Transformer language models into the vocabulary space, revealing the low-rank properties of gradients and explaining the 'imprint and shift' mechanism;
BaitAttack: Alleviating Intention Shift in Jailbreak Attacks via Adaptive Bait Crafting
Rui Pu (Beijing University of Posts and Telecommunications), Xi Zhang (Beijing University of Posts and Telecommunications)
Adversarial AttackLarge Language ModelPrompt EngineeringText
🎯 What it does: Propose BaitAttack, which reduces intention transfer by generating baits and performing adaptive camouflage, thereby enhancing jailbreak attack effectiveness.
Bayesian Calibration of Win Rate Estimation with LLM Evaluators
Yicheng Gao (Yale University), Arman Cohan (Yale University)
Explainability and InterpretabilityData-Centric LearningLarge Language ModelTextBenchmark
🎯 What it does: Propose two calibration methods based on Bayesian inference (BWRS and Bayesian Dawid-Skene) to correct the bias in win rate estimation by LLM evaluators, further improving the reliability of LLM-generated text quality assessment.
Bayesian Example Selection Improves In-Context Learning for Speech, Text and Visual Modalities
Siyin Wang (Tsinghua University), Chao Zhang (Tsinghua University)
Representation LearningData-Centric LearningTransformerLarge Language ModelPrompt EngineeringImageTextMultimodalityAudio
🎯 What it does: Proposes a Bayesian theorem-based inverse reasoning example selection method called ByCS to enhance in-place learning effectiveness of large language models in multimodal scenarios (speech, text, image).
BC-Prover: Backward Chaining Prover for Formal Theorem Proving
Yuhang He (DAMO Academy, Alibaba group), Wotao Yin (DAMO Academy, Alibaba Group US)
TransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose BC-Prover, an interactive theorem proving framework that integrates pseudo steps, backward chaining search, and step planning;
Be Helpful but Don’t Talk too Much - Enhancing Helpfulness in Conversations through Relevance in Multi-Turn Emotional Support
Junlin Li (Hong Kong Polytechnic University), Chu-Ren Huang (Hong Kong Polytechnic University)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningAuto EncoderText
🎯 What it does: This paper proposes an emotion-support dialogue agent based on the 'cognitive relevance principle,' utilizing multi-level reinforcement learning (Optimal Relevance Learning) to balance the effectiveness of assistance with user cognitive load, thereby improving dialogue quality.
BEEAR: Embedding-based Adversarial Removal of Safety Backdoors in Instruction-tuned Language Models
Yi Zeng (Virginia Tech), Ruoxi Jia (Virginia Tech)
Safty and PrivacyAdversarial AttackTransformerLarge Language ModelText
🎯 What it does: Propose a two-layer optimization method named BEEAR to eliminate security backdoors in instruction-tuned large language models (LLMs), leveraging the unified drift features generated by backdoor triggers in the embedding space;
Belief Revision: The Adaptability of Large Language Models Reasoning
Bryan Wilie (Hong Kong University of Science and Technology), Pascale Fung
Explainability and InterpretabilityLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Proposed and evaluated the Belief-R dataset to test whether large language models can correctly revise or maintain existing beliefs after receiving new information.
Benchmarking Vision Language Models for Cultural Understanding
Shravan Nayak (Mila - Quebec AI Institute), Aishwarya Agrawal (Mila - Quebec AI Institute)
Large Language ModelVision Language ModelMultimodalityBenchmark
🎯 What it does: Constructed a cultural understanding benchmark in the form of Visual Question Answering (VQA) called CULTURALVQA, collecting 2,378 image-question pairs and 7,206 answers from 11 countries across five continents, covering five cultural dimensions including clothing, diet, beverages, rituals, and traditions.
Beyond Correlation: Interpretable Evaluation of Machine Translation Metrics
Stefano Perrella (Sapienza University of Rome), Roberto Navigli (Sapienza University of Rome)
Explainability and InterpretabilityTextBenchmark
🎯 What it does: Propose an interpretable evaluation framework to measure the performance of machine translation evaluation metrics under data filtering and translation re-ranking scenarios, using precision, recall, and F-score instead of traditional correlation coefficients.
Beyond Embeddings: The Promise of Visual Table in Visual Reasoning
Yiwu Zhong (Chinese University of Hong Kong), Liwei Wang (Chinese University of Hong Kong)
Representation LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: Proposes a new visual representation method called Visual Table, which achieves structured and interpretable representation of image content through hierarchical text descriptions encompassing scene, object categories, attributes, and knowledge;
Beyond Label Attention: Transparency in Language Models for Automated Medical Coding via Dictionary Learning
John Wu (University of Illinois Urbana-Champaign), Jimeng Sun (University of Illinois Urbana-Champaign)
Explainability and InterpretabilityTransformerLarge Language ModelAuto EncoderTextBiomedical DataElectronic Health Records
🎯 What it does: Studied using sparse autoencoders (dictionary learning) to enhance the interpretability of large language models in medical coding (ICD) tasks, and proposed the AutoCodeDL method combining label attention (LAAT).
Beyond Reference: Evaluating High Quality Translations Better than Human References
Keonwoong Noh (Hanyang University), Woohwan Jung (Hanyang University)
TransformerLarge Language ModelText
🎯 What it does: Propose the RESUME metric to evaluate the quality of candidate translations relative to reference translations, and combine it with existing absolute metrics (e.g., BLEU, COMET) to address the reference bias problem;
Beyond the Turn-Based Game: Enabling Real-Time Conversations with Duplex Models
Xinrong Zhang (Tsinghua University), Zhiyuan Liu (Tsinghua University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose a duplex model that enables LLMs to generate responses while receiving user input, simulating real-time human conversation;
Beyond Turn-Based Interfaces: Synchronous LLMs as Full-Duplex Dialogue Agents
Bandhav Veluri (Meta AI), Shyamnath Gollakota (Meta AI)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningTextMultimodalityAudio
🎯 What it does: Designed and implemented SyncLLM, a model capable of synchronizing speech generation and reception in real-time, full-duplex voice conversations, breaking through the traditional half-duplex polling-based dialogue framework;
BiasAlert: A Plug-and-play Tool for Social Bias Detection in LLMs
Zhiting Fan (Zhejiang University), Zuozhu Liu (Zhejiang University)
ClassificationSafty and PrivacyExplainability and InterpretabilityTransformerSupervised Fine-TuningPrompt EngineeringTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Develop and release BiasAlert, a plug-and-play tool for detecting social bias in LLM open-text generation. It makes judgments by retrieving external bias knowledge bases and combining them with instruction-following to enhance internal reasoning.
BiasWipe: Mitigating Unintended Bias in Text Classifiers through Model Interpretability
Mamta (Indian Institute of Technology Patna), Asif Ekbal (Indian Institute of Technology Jodhpur)
ClassificationExplainability and InterpretabilityTransformerText
🎯 What it does: Identify and prune neuron weights by leveraging model explainability (Shapley values) to eliminate unintended bias in text classification models.
Bio-RFX: Refining Biomedical Extraction via Advanced Relation Classification and Structural Constraints
Minjia Wang (Tsinghua University), Jianyong Wang (Tsinghua University)
ClassificationRecognitionTransformerPrompt EngineeringBiomedical Data
🎯 What it does: Propose Bio-RFX, a medical text entity and relation extraction framework that first identifies the relationship types present in the sentence and then performs entity extraction through relationship-specific question-answering.
Birdie: Advancing State Space Language Modeling with Dynamic Mixtures of Training Objectives
Sam Blouir (George Mason University), Amarda Shehu (George Mason University)
RetrievalLarge Language ModelReinforcement LearningText
🎯 What it does: Propose the Birdie training method, combining bidirectional forward processing with dynamic mixed pre-training objectives to enhance SSM's performance on retrieval tasks.
BlendFilter: Advancing Retrieval-Augmented Large Language Models via Query Generation Blending and Knowledge Filtering
Haoyu Wang (SUNY Albany), Jing Gao (Purdue University)
RetrievalTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the BlendFilter framework, enhancing the performance of retrieval-augmented large language models (LLMs) through query generation hybridization and knowledge filtering.
BLSP-Emo: Towards Empathetic Large Speech-Language Models
Chen Wang (University of Chinese Academy of Sciences), Jiajun Zhang (University of Chinese Academy of Sciences)
RecognitionGenerationTransformerLarge Language ModelSupervised Fine-TuningTextAudio
🎯 What it does: This study proposes BLSP-Emo, an end-to-end emotional speech-language large model capable of understanding semantic and emotional cues in speech and generating empathetic text responses;
BMRetriever: Tuning Large Language Models as Better Biomedical Text Retrievers
Ran Xu (Emory University), Carl Yang (Emory University)
Data SynthesisRetrievalTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningTextBiomedical Data
🎯 What it does: Developed a series of scalable multi-scale biomedical retrieval models called BMRETRIEVER.
Boosting Logical Fallacy Reasoning in LLMs via Logical Structure Tree
Yuanyuan Lei (Texas A&M University), Ruihong Huang (Texas A&M University)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Construct a logical structure tree to explicitly represent the hierarchical logical relationships between connectives and arguments, and integrate it into LLMs for detecting and classifying logical fallacies
Boosting Scientific Concepts Understanding: Can Analogy from Teacher Models Empower Student Models?
Siyu Yuan (School of Data Science Fudan University), Deqing Yang (School of Data Science Fudan University)
Knowledge DistillationTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Propose the SCUA task, leveraging teacher large language models to generate analogies to assist student large language models in understanding scientific concepts and answering related multiple-choice questions.
Bootstrapped Policy Learning for Task-oriented Dialogue through Goal Shaping
Yangyang Zhao (Changsha University of Science and Technology), Shihan Wang (Utrecht University)
Reinforcement LearningText
🎯 What it does: Propose Bootstrapped Policy Learning (BPL), which dynamically generates sub-goal curricula through goal shaping to address the sparse reward problem in dialogue.
BPE Gets Picky: Efficient Vocabulary Refinement During Tokenizer Training
Pavel Chizhov (Technical University of Applied Sciences Würzburg-Schweinfurt), Ivan P. Yamshchikov (Technical University of Applied Sciences Würzburg-Schweinfurt)
Computational EfficiencyData-Centric LearningTextBenchmark
🎯 What it does: Propose PickyBPE, a tokenizer method that dynamically removes intermediate junk tokens during BPE training and achieves efficient vocabulary refinement with a fixed vocabulary size.
BPO: Staying Close to the Behavior LLM Creates Better Online LLM Alignment
Wenda Xu (University Of California Santa Barbara), Lei Li (University Of California Santa Barbara)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelText
🎯 What it does: Propose an online direct preference alignment method BPO, which improves model alignment by setting the trust region as a behavioral LLM that generates samples.
Breaking Language Barriers: Cross-Lingual Continual Pre-Training at Scale
Wenzhen Zheng (Chinese Academy Of Sciences), Ming Zhou (Central South University)
Domain AdaptationOptimizationComputational EfficiencyRepresentation LearningData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: Study cross-lingual continual pre-training (CPT), migrating existing English LLM weights to Chinese (and other languages) for pre-training, significantly accelerating convergence and reducing loss
Breaking ReLU Barrier: Generalized MoEfication for Dense Pretrained Models
Jaeseong Lee (Seoul National University), Mingi Ji (Google)
Computational EfficiencyTransformerMixture of ExpertsText
🎯 What it does: Proposes the G-MoEfication method to convert any dense pre-trained language model into a sparse expert model (MoE).
Breaking the Curse of Multilinguality with Cross-lingual Expert Language Models
Terra Blevins (University of Washington), Luke Zettlemoyer (University of Washington)
Computational EfficiencyRepresentation LearningTransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: This paper proposes a cross-lingual expert language model (X-ELM), which alleviates the 'curse of multilingualism' in multilingual models by dividing large multilingual corpora into subsets and training specialized expert models for each subset.
Bridging Cultures in the Kitchen: A Framework and Benchmark for Cross-Cultural Recipe Retrieval
Tianyi Hu (University of Copenhagen), Daniel Hershcovich (University of Copenhagen)
RetrievalTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose the cross-cultural recipe retrieval task and construct a Chinese-English recipe retrieval benchmark dataset; propose the CARROT framework, which leverages LLM for query rewriting and re-ranking to bridge cultural gaps.
Bridging Local Details and Global Context in Text-Attributed Graphs
Yaoke Wang (Zhejiang University), Siliang Tang (Zhejiang University)
ClassificationGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningTextGraph
🎯 What it does: Design and implement the GraphBridge framework, integrating local text encoding with global graph structure aggregation, and propose a graph-aware token pruning module to enhance efficiency and scalability.
Bridging Modalities: Enhancing Cross-Modality Hate Speech Detection with Few-Shot In-Context Learning
Ming Shan Hee (Singapore University of Technology and Design), Roy Ka-Wei Lee (Singapore University of Technology and Design)
ClassificationDomain AdaptationTransformerLarge Language ModelPrompt EngineeringVision Language ModelTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: This study investigates the feasibility of transferring text-level hate speech detection knowledge to visual-language (memes) detection through few-shot context learning, and verifies that text examples can significantly enhance the performance of visual-language hate detection.
Building Resources for Emakhuwa: Machine Translation and News Classification Benchmarks
Felermino D. M. A. Ali (Universidade do Porto), Rui Sousa-Silva (Universidade do Porto)
ClassificationData SynthesisTransformerSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper constructs multilingual NLP resources for Emakhuwa, the largest language in Mozambique, including the first manually translated Portuguese-Emakhuwa news bilingual corpus, OCR-corrected data, monolingual data, and news topic classification datasets, along with benchmark experiments for machine translation and news classification tasks.
By My Eyes: Grounding Multimodal Large Language Models with Sensor Data via Visual Prompting
Hyungjun Yoon (Korea Advanced Institute of Science and Technology), Sung-Ju Lee (Korea Advanced Institute of Science and Technology)
ClassificationRecognitionTransformerLarge Language ModelPrompt EngineeringVision Language ModelTime SeriesBiomedical DataElectrocardiogram
🎯 What it does: Propose a visual prompting method that visualizes sensor data as images and inputs them into a multimodal large language model (MLLM), along with an automatically generated tool for optimal visualization;
C-LLM: Learn to Check Chinese Spelling Errors Character by Character
Kunting Li (Tsinghua University), Jie Zhou (Tencent Inc)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose C-LLM, which utilizes character-level tokenization to enable large language models to check and correct Chinese spelling errors character by character.
C3PA: An Open Dataset of Expert-Annotated and Regulation-Aware Privacy Policies to Enable Scalable Regulatory Compliance Audits
Maaz Bin Musa (University of Iowa), Rishab Nithyanand (University of Iowa)
Safty and PrivacyData-Centric LearningTransformerSupervised Fine-TuningTextBenchmark
🎯 What it does: This study created the first expert-annotated privacy policy dataset based on CCPA called C3PA, and used it to train a model for compliance checks.
Calibrating Language Models with Adaptive Temperature Scaling
Johnathan Xie (Stanford University), Chelsea Finn (Stanford University)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Proposed Adaptive Temperature Scaling (ATS) as a post-calibration method for large language models after RLHF fine-tuning.
Calibrating the Confidence of Large Language Models by Eliciting Fidelity
Mozhi Zhang (Fudan University), Xipeng Qiu (Fudan University)
Explainability and InterpretabilityLarge Language ModelText
🎯 What it does: Decompose the overconfidence issue of large language models in multiple-choice question answering, breaking down the model's confidence into uncertainty about the question and loyalty to the answer, and propose a UF Calibration method that seamlessly plugs in and out. This method infers the model's loyalty weights for each answer by sampling and using the 'All other options are wrong' strategy, then combines these weights with uncertainty entropy to obtain calibrated confidence.
Can Active Label Correction Improve LLM-based Modular AI Systems?
Karan Taneja (Georgia Institute of Technology), Ashok Goel
Data-Centric LearningTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Use Active Label Correction (ALC) to correct noisy labels generated by LLMs (GPT-3.5), and replace the LLM with a smaller task-specific model.
Can Automatic Metrics Assess High-Quality Translations?
Sweta Agrawal (Instituto de Telecomunicações), Andre Martins
ClassificationTextBenchmark
🎯 What it does: Systematically evaluate the ability of existing machine translation evaluation metrics to identify high-quality translations with no errors (zero MQM), focusing on their performance in distinguishing subtle quality differences under the same source sentence.
Can Language Models Induce Grammatical Knowledge from Indirect Evidence?
Miyu Oba (Nara Institute of Science and Technology), Saku Sugawara (National Institute of Informatics)
Data-Centric LearningTransformerLarge Language ModelText
🎯 What it does: Study the efficiency of language models in learning grammatical knowledge under indirect evidence.
Can Large Language Models Always Solve Easy Problems if They Can Solve Harder Ones?
Zhe Yang (Peking University), Zhifang Sui (Alibaba Group)
Explainability and InterpretabilityTransformerLarge Language ModelTextBenchmark
🎯 What it does: Constructed the ConsisEval benchmark to study the hard-easy consistency issue of LLMs, i.e., whether LLMs can solve easier problems when solving harder ones, and proposed two metrics: consistency score (CS) and relative consistency score (RCS).
Can Large Language Models Enhance Predictions of Disease Progression? Investigating Through Disease Network Link Prediction
Haohui Lu (University of Sydney), Usman Naseem (Macquarie University)
Graph Neural NetworkTransformerLarge Language ModelGraphBiomedical DataRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the ComLLM framework, leveraging graph prompting and retrieval-augmented generation (RAG) to enhance the prediction of comorbid disease links
Can Large Language Models Faithfully Express Their Intrinsic Uncertainty in Words?
Gal Yona (Google Research), Mor Geva (Tel Aviv University)
Explainability and InterpretabilityLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposed a framework to evaluate how large language models (LLMs) faithfully express their intrinsic uncertainty in natural language, validated on knowledge-intensive question-answering tasks.
Can Large Language Models Learn Independent Causal Mechanisms?
Gael Gendron, Gillian Dobbie (University of Auckland)
Explainability and InterpretabilityTransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: Introduce the principle of independent causal mechanisms, designing a routable multi-module LLM architecture (ICLM) that achieves module independence through unsupervised vector quantization routing, mutual information regularization, and aggregation strategies, thereby enhancing abstract reasoning and continual learning performance in discrete out-of-distribution scenarios.
Can LLM Generate Culturally Relevant Commonsense QA Data? Case Study in Indonesian and Sundanese
Rifki Afina Putri (KAIST), Alice Oh (KAIST)
GenerationData SynthesisTransformerLarge Language ModelText
🎯 What it does: This paper constructs a culture-related commonsense question-answering dataset for Indonesian and Sundanese languages, and evaluates the generation quality of LLMs;
Can LLMs Learn Uncertainty on Their Own? Expressing Uncertainty Effectively in A Self-Training Manner
Shudong Liu (University of Macau), Min Zhang (Harbin Institute of Technology)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes the 'Uncertainty-aware Instruction Tuning (UaIT)' method, enabling large language models (LLMs) to self-perceive and express their own uncertainty in probabilistic terms;
Can LLMs replace Neil deGrasse Tyson? Evaluating the Reliability of LLMs as Science Communicators
Prasoon Bajpai (Indian Institute of Technology Delhi), Tanmoy Chakraborty (Indian Institute of Technology Delhi)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper proposes a new scientific question-answering dataset called SCiPS-QA, and evaluates the answer reliability and self-verification capability of large language models (LLMs) in the scientific domain based on this dataset.
Can Transformers Learn n-gram Language Models?
Anej Svete (ETH Zürich), Ryan Cotterell (ETH Zürich)
Representation LearningTransformerText
🎯 What it does: Studied the learnability of Transformers in learning random n-gram language models by generating different types of n-gram models (non-representation-based, sparse representation-based, dense representation-based), training Transformers, traditional n-gram estimation, log-linear, and neural n-gram baselines, and evaluating their deviation from the true model using KL divergence.
Can visual language models resolve textual ambiguity with visual cues? Let visual puns tell you!
Jiwan Chung (Yonsei University), Youngjae Yu (Yonsei University)
TransformerLarge Language ModelVision Language ModelDiffusion modelContrastive LearningMultimodalityBenchmark
🎯 What it does: Propose the UNPIE dataset and three evaluation tasks (Pun Grounding, Pun Disambiguation, Pun Reconstruction) to assess the multimodal capabilities of vision-language models in interpreting text ambiguity.
Can We Trust the Performance Evaluation of Uncertainty Estimation Methods in Text Summarization?
Jianfeng He (Virginia Tech), Chang-Tien Lu (Virginia Tech)
GenerationTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Studied uncertainty estimation methods in text summarization for reliability assessment, and proposed a benchmark covering 31 NLG metrics and 14 uncertainty estimation methods.
CareCorpus+: Expanding and Augmenting Caregiver Strategy Data to Support Pediatric Rehabilitation
Shahla Farzana (University of Illinois Chicago), Natalie Parde (University of Illinois Chicago)
ClassificationData SynthesisTransformerLarge Language ModelTextBenchmark
🎯 What it does: This study introduces a large dataset containing 3062 nursing strategies, aiming to support nursing strategy classification in pediatric rehabilitation, addressing the issue of existing data scarcity.
CARER - ClinicAl Reasoning-Enhanced Representation for Temporal Health Risk Prediction
Tuan Dung Nguyen (Hanoi University of Science and Technology), Phi Le Nguyen (Hanoi University of Science and Technology)
Representation LearningRecurrent Neural NetworkTransformerLarge Language ModelMultimodalityTime SeriesBiomedical DataElectronic Health RecordsRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the CARER framework, integrating clinical reasoning generated by large language models (LLMs) with multimodal deep learning models for time-series health risk prediction.
Casablanca: Data and Models for Multidialectal Arabic Speech Recognition
Bashar Talafha (University of British Columbia), Muhammad Abdul-Mageed (University of British Columbia)
RecognitionTransformerSupervised Fine-TuningBenchmarkAudio
🎯 What it does: This work constructs a large multi-annotated dataset named Casablanca, containing 48 hours of speech with eight Arabic dialects, gender, and code-switching labels, and evaluates multilingual ASR models on it.
CasiMedicos-Arg: A Medical Question Answering Dataset Annotated with Explanatory Argumentative Structures
Ekaterina Sviridova (Université Côte d'Azur), Rodrigo Agerri (University of Basque Country UPV/EHU)
Explainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextBiomedical DataElectronic Health RecordsBenchmark
🎯 What it does: Constructed the first multilingual medical question-answering dataset, CasiMedicos-Arg, containing 558 clinical cases and physician-written explanations, with manual annotations of claims, premises, and their support/attack relationships.
CaT-Bench: Benchmarking Language Model Understanding of Causal and Temporal Dependencies in Plans
Yash Kumar Lal (Stony Brook University), Ray Mooney
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextSequentialBenchmarkChain-of-Thought
🎯 What it does: Proposes the CAT-BENCH benchmark to evaluate large language models (LLMs) in understanding and reasoning about the temporal dependencies of steps in natural language cooking plans, along with corresponding explanations.
CELLO: Causal Evaluation of Large Vision-Language Models
Meiqi Chen (Peking University), Chaochao Lu (Shanghai Artificial Intelligence Laboratory)
Explainability and InterpretabilityData-Centric LearningLarge Language ModelVision Language ModelImageMultimodalityGraphBenchmarkChain-of-Thought
🎯 What it does: Construct a fine-grained unified causal definition and create a visual-language causal reasoning dataset named CELLO, containing 14,094 cross-four-level causal questions.
Chain and Causal Attention for Efficient Entity Tracking
Erwan Fagnou (Université Paris Dauphine-PSL), Alexandre Allauzen (Université Paris Dauphine-PSL)
Object TrackingComputational EfficiencyTransformerTextSequential
🎯 What it does: This paper investigates the bottleneck of Transformers in entity tracking tasks. Theoretical proof shows that at least log₂(n+1) layers are required to handle n state changes. Subsequently, the paper proposes ChaCAL (Chain and Causal Attention Layer), an improved attention mechanism, enabling a single layer to efficiently capture long-range dependencies and complete entity state tracking.
Chain-of-Dictionary Prompting Elicits Translation in Large Language Models
Hongyuan Lu (Chinese University of Hong Kong), Furu Wei (Microsoft Corporation)
GenerationTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Propose Chain-of-Dictionary Prompting (COD), leveraging multilingual dictionary chains to enhance large language models (LLM) machine translation capabilities for low-resource languages within prompts.
Chain-of-Note: Enhancing Robustness in Retrieval-Augmented Language Models
Wenhao Yu (Tencent AI Lab), Dong Yu
TransformerSupervised Fine-TuningTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed and implemented the CHAIN-OF-NOTE (CON) framework, which generates reading notes from retrieved documents one by one, systematically evaluates the relevance between documents and questions, and generates answers accordingly; simultaneously uses GPT-4 to automatically generate 10K training samples and performs fine-tuning on LLaMA-2-7B; conducted experiments on four open-domain question answering benchmarks: NQ, TriviaQA, WebQ, and RealTimeQA.
ChatGPT Doesn’t Trust Chargers Fans: Guardrail Sensitivity in Context
Victoria R Li, Naomi Saphra (Harvard University)
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: Studied ChatGPT (gpt-3.5-turbo)'s guardrail sensitivity when facing different user identities (gender, age, race, political orientation, sports fan status), using automatically generated user profiles and sensitive requests for large-scale experiments.
ChatRetriever: Adapting Large Language Models for Generalized and Robust Conversational Dense Retrieval
Kelong Mao (Renmin University of China), Zhicheng Dou (Renmin University of China)
RetrievalRepresentation LearningTransformerContrastive LearningTextChain-of-Thought
🎯 What it does: Utilizing large language models (LLM) for dialogue retrieval, proposes ChatRetriever, adopting a dual learning framework CSIT to enhance the representation and retrieval performance for multi-turn dialogues.
CHESS: Optimizing LLM Inference via Channel-Wise Thresholding and Selective Sparsification
Junhui He (Wuhan University), Qingan Li (Wuhan University)
Computational EfficiencyTransformerText
🎯 What it does: To address the inference efficiency of large language models on edge devices, the CHESS method is proposed to achieve activation sparsification by implementing channel-level threshold pruning and selective sparsification on the feed-forward network (FFN) and attention modules.