ACL 2024 Papers — Page 4
Annual Meeting of the Association for Computational Linguistics · 940 papers
Faithful Chart Summarization with ChaTS-Pi
Syrine Krichene (Google DeepMind), Julian Eisenschlos (Google DeepMind)
GenerationTransformerLarge Language ModelImageTextTabularChain-of-Thought
🎯 What it does: Propose a no-reference chart summary evaluation metric called CHATS-CRITIC, and build a self-correcting chart summary generation pipeline named CHATS-PI based on this metric, significantly improving the factual accuracy and fluency of summaries.
Faithful Logical Reasoning via Symbolic Chain-of-Thought
Jundong Xu (National University of Singapore), Wynne Hsu (National University of Singapore)
OptimizationExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Proposes a fully LLM-based symbolic chain-of-thought reasoning framework called SymbCoT, which uses hybrid symbolic expressions combining natural language and first-order logic/constraint optimization for translation, planning, solving, and verification;
FanOutQA: A Multi-Hop, Multi-Document Question Answering Benchmark for Large Language Models
Andrew Zhu (University of Pennsylvania), Chris Callison-Burch (University of Pennsylvania)
Large Language ModelTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes the FanOutQA dataset, specifically designed to evaluate the reasoning capabilities of large models on multi-hop, multi-document, and fan-out questions, and provides three evaluation settings (closed-book, open-book, and given evidence).
FastFiD: Improve Inference Efficiency of Open Domain Question Answering via Sentence Selection
Yufei Huang (Tsinghua University), Maosong Sun (Tsinghua University)
RetrievalComputational EfficiencyTransformerSupervised Fine-TuningTextRetrieval-Augmented Generation
🎯 What it does: Significantly accelerate open-domain question answering inference by adding sentence selection in the FiD framework and adopting two-phase training;
Favi-Score: A Measure for Favoritism in Automated Preference Ratings for Generative AI Evaluation
Pius Von Däniken (ZHAW School of Engineering), Mark Cieliebak (ZHAW School of Engineering)
GenerationText
🎯 What it does: Propose a new preference scoring metric called Favi-Score to quantify the preference tendency of automated evaluation metrics toward different generation systems, and demonstrate that this preference can lead to incorrect system rankings.
Feature-Adaptive and Data-Scalable In-Context Learning
Jiahao Li (University Of Science And Technology Of China), Zhendong Mao (University Of Science And Technology Of China)
ClassificationTransformerLarge Language ModelText
🎯 What it does: Propose the FADS-ICL framework, which utilizes out-of-context samples as supervision. It first extracts general features individually using LLMs, then trains a lightweight task-specific modulator to adapt the general features for specific tasks, thereby achieving feature adaptation and data scalability in In-Context Learning.
Few-shot Transfer Learning for Knowledge Base Question Answering: Fusing Supervised Models with In-Context Learning
Mayur Patidar (TCS Research), Indrajit Bhattacharya (Indian Institute of Technology)
RetrievalDomain AdaptationTransformerLarge Language ModelPrompt EngineeringTextGraphRetrieval-Augmented Generation
🎯 What it does: To address the few-shot transfer learning problem in knowledge base question answering, the FuSIC-KBQA architecture is proposed, combining technologies such as retrievers trained on the source domain, LLM re-ranking, few-shot context learning, and execution feedback to generate logical forms.
FinanceMATH: Knowledge-Intensive Math Reasoning in Finance Domains
Yilun Zhao (Yale University), Arman Cohan (Yale University)
Large Language ModelPrompt EngineeringTextTabularBenchmarkFinance RelatedRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Constructed the FinanceMATH benchmark to evaluate LLMs' knowledge-intensive mathematical reasoning capabilities in the financial domain, and created a financial terminology knowledge base containing 864 terms.
Fine-Grained Image-Text Alignment in Medical Imaging Enables Explainable Cyclic Image-Report Generation
Wenting Chen (City University of Hong Kong), Yixuan Yuan (Chinese University of Hong Kong)
GenerationExplainability and InterpretabilityTransformerLarge Language ModelVision Language ModelContrastive LearningMultimodalityBiomedical Data
🎯 What it does: Propose the AdaMatch model to achieve fine-grained alignment between adaptive patches and text words, and provide explicit interpretability within a cyclic CXR-report generation framework.
Fine-Grained Modeling of Narrative Context: A Coherence Perspective via Retrospective Questions
Liyan Xu (WeChat AI), Jie Zhou (WeChat AI)
ClassificationRetrievalTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Constructed a fine-grained narrative context graph called NARCO, leveraging LLM to automatically generate retroactive question edges, supporting multiple downstream narrative understanding tasks.
Fine-Tuned Machine Translation Metrics Struggle in Unseen Domains
Vilém Zouhar (ETH Zürich), Brian Thompson (AWS AI Labs)
Domain AdaptationTransformerLarge Language ModelSupervised Fine-TuningBiomedical DataBenchmark
🎯 What it does: This paper first constructs a biomedical domain dataset covering 11 language pairs with approximately 25,000 paragraph-level MQM annotations and publicly releases it; subsequently, it compares this dataset with the existing WMT MQM data to explore the robustness of MT evaluation metrics when fine-tuned on unseen domains (bio).
Fine-Tuning Pre-Trained Language Models with Gaze Supervision
Shuwen Deng (University of Potsdam), Lena Jäger
Representation LearningData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Introduce an eye-tracking module during the fine-tuning of pre-trained language models, enhancing text representation through auxiliary loss using synthetic scan paths.
FineSurE: Fine-grained Summarization Evaluation using LLMs
Hwanjun Song (Korea Advanced Institute of Science and Technology), Saab Mansour (AWS AI Labs)
ClassificationTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Propose FineSurE, a fine-grained multi-dimensional automatic evaluation framework that utilizes LLMs for fact-checking and key fact alignment, providing scores for faithfulness, completeness, and conciseness;
FinTextQA: A Dataset for Long-form Financial Question Answering
Jian Chen (HSBC Lab), Junwei Liang (Hong Kong University of Science and Technology)
TransformerTextBenchmarkFinance RelatedRetrieval-Augmented Generation
🎯 What it does: This paper constructs the FinTextQA long-form financial question-answering dataset and implements and evaluates the LFQA system based on the Retrieval-Augmented Generation (RAG) framework.
FLEUR: An Explainable Reference-Free Evaluation Metric for Image Captioning Using a Large Multimodal Model
Yebin Lee (Seoul National University), Myungjoo Kang (Seoul National University)
Explainability and InterpretabilityLarge Language ModelVision Language ModelMultimodality
🎯 What it does: Proposed FLEUR, an interpretable, reference-free image description evaluation metric based on a large-scale multimodal model, which directly compares candidate descriptions with images to provide scores and explanations.
Focus on Your Question! Interpreting and Mitigating Toxic CoT Problems in Commonsense Reasoning
Jiachun Li (University of Chinese Academy of Sciences), Jun Zhao (University of Chinese Academy of Sciences)
Explainability and InterpretabilityComputational EfficiencyTransformerTextChain-of-Thought
🎯 What it does: Identified and analyzed the Toxic CoT issue that arises when large language models use Chain-of-Thought prompts, and proposed the RIDERS method to address this problem
FOFO: A Benchmark to Evaluate LLMs’ Format-Following Capability
Congying Xia (Salesforce Research), Caiming Xiong (Salesforce Research)
TransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Constructed and evaluated the FOFO benchmark to test LLMs' ability in complex, domain-specific format following.
FollowBench: A Multi-level Fine-grained Constraints Following Benchmark for Large Language Models
Yuxin Jiang (Hong Kong University of Science and Technology), Wei Wang (Huawei)
Large Language ModelTextBenchmark
🎯 What it does: Propose the FollowBench benchmark, which uses multi-layer fine-grained constraints to evaluate the instruction-following capability of LLMs
Fora: A corpus and framework for the study of facilitated dialogue
Hope Schroeder (Massachusetts Institute of Technology), Jad Kabbara (Massachusetts Institute of Technology)
ClassificationRecognitionData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Constructed and made publicly available the Fora corpus, containing 262 transcripts of semi-structured facilitative dialogues hosted by non-profit organizations, with approximately 10,000 utterances manually annotated for shared behaviors and facilitation strategies.
Forgetting before Learning: Utilizing Parametric Arithmetic for Knowledge Updating in Large Language Models
Shiwen Ni (Shenzhen Institutes Of Advanced Technology Chinese Academy Of Sciences), Min Yang (Shenzhen Institutes Of Advanced Technology Chinese Academy Of Sciences)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose the F-Learning two-phase forgetting-learning framework, which updates LLM knowledge through parameter arithmetic operations.
Fortify the Shortest Stave in Attention: Enhancing Context Awareness of Large Language Models for Effective Tool Use
Yuhan Chen (Gaoling School of Artificial Intelligence, Renmin University of China), Rui Yan (Gaoling School of Artificial Intelligence, Renmin University of China)
AI Code AssistantTransformerLarge Language ModelTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: This paper discovers that LLMs exhibit waveform patterns in attention allocation, where key information located in waveform troughs is easily overlooked. It proposes the Attention Buckets method, which enhances the model's focus on key information during inference by parallel processing with multiple base RoPE and weighted fusion.
Framing in the Presence of Supporting Data: A Case Study in U.S. Economic News
Alexandria Leto (University of Colorado Boulder), Maria Leonor Pacheco (University of Colorado Boulder)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextTabularFinance Related
🎯 What it does: Developed a computational framework that utilizes supporting data (such as economic indicators) to automatically identify news reporting frameworks—judging the overall economic sentiment, indicator type, and direction at the article level, and determining the metric attribution and sentiment polarity of each numerical value; this framework was used for large-scale selection and analysis of economic reporting from major US media.
FreeCtrl: Constructing Control Centers with Feedforward Layers for Learning-Free Controllable Text Generation
Zijian Feng (Nanyang Technological University), Zixiao Zhu (Nanyang Technological University)
GenerationTransformerLarge Language ModelText
🎯 What it does: A training-free controllable text generation method called FreeCtrl is proposed by dynamically adjusting the value vector weights in the FFN layer of large language models.
From Moments to Milestones: Incremental Timeline Summarization Leveraging Large Language Models
Qisheng Hu (National University of Singapore), Hwee Tou Ng (National University of Singapore)
GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Propose an incremental timeline summarization method called LLM-TLS based on large language models, capable of simultaneously handling event and topic timelines;
From Sights to Insights: Towards Summarization of Multimodal Clinical Documents
Akash Ghosh (Indian Institute of Technology Patna), Setu Sinha (Indira Gandhi Institute of Medical Sciences)
TransformerLarge Language ModelVision Language ModelMultimodalityBiomedical Data
🎯 What it does: Propose a multimodal clinical document summarization model, EDI-Summ, which can generate more accurate summaries by leveraging both text and images.
Full Parameter Fine-tuning for Large Language Models with Limited Resources
Kai Lv (Fudan University), Xipeng Qiu (Fudan University)
OptimizationComputational EfficiencyLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed a novel optimizer called LOMO, which integrates gradient computation with parameter updates, achieving full-parameter fine-tuning of large language models (including 65B models) on a single machine with 8 RTX 3090 GPUs with extremely low memory requirements; also explained that SGD can be used for full-parameter fine-tuning and provided relevant theoretical and practical details; and proposed alternative solutions for gradient normalization and mixed-precision training.
Fundamental Capabilities of Large Language Models and their Applications in Domain Scenarios: A Survey
Jiawei Li (Beijing Institute of Technology), Heyan Huang (Beijing Institute of Technology)
TransformerLarge Language ModelTextReview/Survey PaperBenchmark
🎯 What it does: This paper systematically reviews the four core capabilities of large language models (LLMs) (memory, reasoning, generalization, differentiation) and their applications in nine practical domains (medicine, law, computational biology, etc.), and proposes a model selection strategy based on the importance of these capabilities.
G-DIG: Towards Gradient-based DIverse and hiGh-quality Instruction Data Selection for Machine Translation
Xingyuan Pan (ByteDance Research), Shanbo Cheng (ByteDance Research)
Data-Centric LearningSupervised Fine-TuningText
🎯 What it does: Automatically selects high-quality and diverse instruction-tuning data using gradient information to enhance the performance of large language models (LLMs) in machine translation tasks.
Generalizability of Mixture of Domain-Specific Adapters from the Lens of Signed Weight Directions and its Application to Effective Model Pruning
Tuc Nguyen (Indiana University), Thai Le (Indiana University)
Computational EfficiencyKnowledge DistillationAdversarial AttackSupervised Fine-TuningMixture of ExpertsText
🎯 What it does: Comprehensively evaluate the generalizability of Mixture of Domain-Specific Adapters in in-domain reasoning, elucidate the mechanism of performance degradation by analyzing weight sign differences (FSD), and propose greedy mixing and sparse pruning strategies based on FSD.
Generalizing Conversational Dense Retrieval via LLM-Cognition Data Augmentation
Haonan Chen (Renmin University of China), Ziliang Zhao (Renmin University of China)
RetrievalData-Centric LearningLarge Language ModelPrompt EngineeringContrastive LearningText
🎯 What it does: This paper proposes CONVAUG, a multi-layer data augmentation framework based on large language models (LLMs), aimed at enhancing the robustness and generalization ability of conversational dense retrieval.
Generate-then-Ground in Retrieval-Augmented Generation for Multi-hop Question Answering
Zhengliang Shi (Shandong University), Zhaochun Ren (Leiden University)
GenerationRetrievalKnowledge DistillationTransformerPrompt EngineeringTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose a generate-then-ground framework, where the LLM first generates a single-step answer using its own knowledge, then corrects and verifies the answer through retrieved documents to complete multi-hop QA.
Generating and Evaluating Plausible Explanations for Knowledge Graph Completion
Antonio Di Mauro (NEC Laboratories Europe), Carolin Lawrence (NEC Laboratories Europe)
Explainability and InterpretabilityRepresentation LearningGraphBenchmark
🎯 What it does: Proposes the GradPath method, generating path-based explanations for Knowledge Graph Completion (KGC) that are interpretable and understandable to humans, and constructs a comprehensive human-centric interpretability evaluation framework.
Generating Coherent Sequences of Visual Illustrations for Real-World Manual Tasks
João Bordalo (NOVA School of Science and Technology), Joao Magalhaes (NOVA School of Science and Technology)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelDiffusion modelImageTextSequential
🎯 What it does: Propose a method combining a sequence context decoder with an improved latent diffusion model to generate visually coherent image sequences that match multi-step manual task instructions.
Generating Contrastive Narratives Using the Brownian Bridge Process for Narrative Coherence Learning
Feiteng Mu (Hong Kong Polytechnic University), Wenjie Li (Hong Kong Polytechnic University)
GenerationTransformerContrastive LearningText
🎯 What it does: Generate contrastive narratives through the Brownian Bridge process, then cross or replace events with the original narrative to create hard negative samples, using contrastive learning to train the narrative coherence evaluator CohEval.
Generating Harder Cross-document Event Coreference Resolution Datasets using Metaphoric Paraphrasing
Shafiuddin Rehan Ahmed (University of Colorado), James H. Martin (University of Colorado)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Utilize GPT-4 to perform constraint-based metaphor rewriting on event triggers in Event Coref Bank Plus (ECB+), generating the ECB+META dataset with enhanced lexical diversity and metaphorical richness, and evaluate cross-document event coreference resolution methods on this dataset.
Generative Cross-Modal Retrieval: Memorizing Images in Multimodal Language Models for Retrieval and Beyond
Yongqi Li, Tat-Seng Chua (National University Of Singapore)
RetrievalLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: Enable multimodal large language models (MLLMs) to remember and retrieve images by learning to generate unique identifiers corresponding to images, directly invoking images from model parameters without relying on external retrieval indexes.
Generative Explore-Exploit: Training-free Optimization of Generative Recommender Systems using LLM Optimizers
Lütfi Kerem Senel (LMU Munich), Shervin Malmasi (Amazon.com, Inc.)
Recommendation SystemOptimizationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Propose an untrained generative recommendation system optimization method that leverages LLM to generate questions and iteratively refines the question pool through click-through rate (CTR) feedback.
Generative Pre-trained Speech Language Model with Efficient Hierarchical Transformer
Yongxin Zhu (University of Science and Technology of China), Dong Yu (Tencent AI Lab)
GenerationData SynthesisTransformerLarge Language ModelAudio
🎯 What it does: Designed and implemented a single-stage hierarchical Transformer model called GPST for unified generation of semantic tokens and acoustic tokens, achieving high-quality speech synthesis, speaker identity migration, and cross-lingual speech generation.
Generative Pretrained Structured Transformers: Unsupervised Syntactic Language Models at Scale
Xiang Hu (Ant Group), Kewei Tu (ShanghaiTech University)
GenerationData SynthesisComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Propose GPST, a structured language model that can be unsupervised pre-trained on large-scale raw text, capable of simultaneously generating sentences and their syntactic trees;
GenTranslate: Large Language Models are Generative Multilingual Speech and Machine Translators
Yuchen Hu (Nanyang Technological University), Eng Siong Chng (Nanyang Technological University)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningTextMultimodalityAudio
🎯 What it does: Propose a novel generative translation paradigm, GenTranslate, which leverages a large language model (LLM) to integrate information from N-best translation candidates generated via beam search, producing a higher-quality single translation output.
Getting Serious about Humor: Crafting Humor Datasets with Unfunny Large Language Models
Zachary Horvitz (Columbia University), Kathleen McKeown (Columbia University)
ClassificationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This study explores using large language models (LLMs) to perform 'unfun' editing on existing humorous texts to generate non-humorous versions corresponding to the original humorous texts, thereby constructing a large-scale aligned humorous/non-humorous dataset; and verifies its cross-lingual generalizability on English-Indian code-mixed tweets.
GLIMPSE: Pragmatically Informative Multi-Document Summarization for Scholarly Reviews
Maxime Darrin (International Laboratory on Learning Systems), Jackie Cheung (MILA - Quebec AI Institute)
GenerationTransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper proposes a discriminative multi-document summarization method called GLIMPSE for academic reviews, which can extract and summarize common and unique viewpoints from multiple reviews to generate concise yet comprehensive review overviews.
GPT is Not an Annotator: The Necessity of Human Annotation in Fairness Benchmark Construction
Virginia Felkner, Jonathan May (Information Sciences Institute University of Southern California)
Data-Centric LearningTransformerLarge Language ModelPrompt EngineeringContrastive LearningTextBenchmark
🎯 What it does: Constructed a community-driven fairness benchmark called WinoSemitism targeting antisemitism, and attempted to automatically extract harmful predicates from questionnaires using GPT-3.5-Turbo to generate the benchmark, exploring the feasibility of LLMs in sensitive tasks.
GradSafe: Detecting Jailbreak Prompts for LLMs via Safety-Critical Gradient Analysis
Yueqi Xie (Hong Kong University of Science and Technology), Neil Gong
Safty and PrivacyTransformerLarge Language ModelText
🎯 What it does: Propose GradSafe, a method that uses LLM safety-critical gradients to detect jailbreak/unsafe prompts;
Graph Language Models
Moritz Plenz (Heidelberg University), Anette Frank (Heidelberg University)
Representation LearningGraph Neural NetworkTransformerLarge Language ModelTextGraph
🎯 What it does: Propose a Graph Language Model (GLM) that transfers pre-trained language model weights into a graph Transformer for jointly encoding text and knowledge graph structures.
Greed is All You Need: An Evaluation of Tokenizer Inference Methods
Omri Uzan (Ben Gurion University), Yuval Pinter (Ben Gurion University)
TextBenchmark
🎯 What it does: This paper constructs an aggregated intrinsic evaluation suite to systematically compare the inference methods (greedy, merge, likelihood, least-tokens, etc.) of four subword tokenizers—BPE, UnigramLM, WordPiece, and SaGe—across different vocabulary sizes, and analyzes their performance in terms of morphology alignment, cognitive interpretability, and information theory metrics.
Grounding Language Model with Chunking-Free In-Context Retrieval
Hongjin Qian (Beijing Academy of Artificial Intelligence), Zhicheng Dou (Beijing Academy of Artificial Intelligence)
GenerationRetrievalTransformerSupervised Fine-TuningTextRetrieval-Augmented Generation
🎯 What it does: Proposed a block-free context retrieval method called CFIC for directly generating precise evidence text from long documents within Retrieval-Augmented Generation (RAG) systems.
GroundingGPT: Language Enhanced Multi-modal Grounding Model
Zhaowei Li (ByteDance Inc), Tao Wang (ByteDance Inc)
RecognitionTransformerLarge Language ModelVision Language ModelContrastive LearningImageVideoMultimodalityAudio
🎯 What it does: GroundingGPT proposes an end-to-end multimodal large language model capable of simultaneously processing images, videos, and audio, achieving fine-grained localization and understanding.
Growing Trees on Sounds: Assessing Strategies for End-to-End Dependency Parsing of Speech
Adrien Pupier (Grenoble Alpes University), Benjamin Lecouteux (Grenoble Alpes University)
Graph Neural NetworkTransformerTextAudio
🎯 What it does: Proposed a graph-based parsing model for end-to-end dependency parsing directly from speech signals, and conducted comparative experiments with sequence labeling parsers and traditional transcription+parser pipeline models.
GrowOVER: How Can LLMs Adapt to Growing Real-World Knowledge?
Dayoon Ko (Seoul National University), Gunhee Kim (Seoul National University)
RetrievalComputational EfficiencyData-Centric LearningTransformerContrastive LearningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposed the GrowOVER dynamic QA and dialogue benchmark, and designed the Retrieval-Interactive LLM (RiLM) framework to achieve training-free knowledge adaptation.
GSM-Plus: A Comprehensive Benchmark for Evaluating the Robustness of LLMs as Mathematical Problem Solvers
Qintong Li (University of Hong Kong), Wei Bi (Tencent AI Lab)
TransformerPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Propose GSM-PLUS—an adversarial benchmark that applies eight categories of mathematical perturbations to GSM8K, systematically evaluating the robustness of LLMs in mathematical reasoning.
Guardians of the Machine Translation Meta-Evaluation: Sentinel Metrics Fall In!
Stefano Perrella (Sapienza University of Rome), Roberto Navigli (Sapienza University of Rome)
TransformerTextBenchmark
🎯 What it does: Propose and use 'sentinel metrics' to test the WMT meta-evaluation process, revealing biases in grouping strategies and linking calibration.
Guidance-Based Prompt Data Augmentation in Specialized Domains for Named Entity Recognition
Hyeonseok Kang (Chungnam National University), Riwoo Chung (KT Corporation)
RecognitionData SynthesisTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Propose a guided prompt-based data augmentation framework called GDA, which generates structurally diverse and semantically consistent incremental data by leveraging abstracted contextual and sentence structural information, specifically addressing the data scarcity problem in specialized domain NER tasks.
GumbelSoft: Diversified Language Model Watermarking via the GumbelMax-trick
Jiayi Fu (Fudan University), Yanghua Xiao (Fudan University)
GenerationTransformerLarge Language ModelText
🎯 What it does: Proposed the GumbelSoft watermarking method to solve the problem that traditional GumbelMax watermarking generates the same output for the same input
GunStance: Stance Detection for Gun Control and Gun Regulation
Nikesh Gyawali (Kansas State University), Cornelia Caragea (University of Illinois Chicago)
ClassificationTransformerLarge Language ModelText
🎯 What it does: This paper constructs a novel stance detection dataset for gun control, named GUNSTANCE, and proposes a hybrid method combining semi-supervised learning with large language models to improve stance identification on gun ban and regulation issues.
Handling Ambiguity in Emotion: From Out-of-Domain Detection to Distribution Estimation
Wen Wu (University of Cambridge), Phil Woodland
RecognitionAudio
🎯 What it does: Studied how to handle ambiguity in sentiment annotations by treating no majority agreement labels (NMA) as out-of-distribution (OOD) samples and quantifying uncertainty through Evidential Deep Learning (EDL), as well as shifting sentiment from single categories to distribution estimation by leveraging all annotation information rather than relying solely on majority voting.
Hard Prompts Made Interpretable: Sparse Entropy Regularization for Prompt Tuning with RL
Yunseon Choi (KAIST AI), Kee-Eung Kim (KAIST AI)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelReinforcement LearningPrompt EngineeringTextMultimodality
🎯 What it does: This paper proposes the PIN (Prompts made INterpretable) algorithm, which optimizes discrete hard prompts on black-box pre-trained models through reinforcement learning, generating prompts that achieve high task performance while being human-interpretable.
Harder Task Needs More Experts: Dynamic Routing in MoE Models
Quzhe Huang (Peking University), Yansong Feng (Peking University)
Computational EfficiencyTransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: A dynamic expert routing framework is proposed in MoE models, which adaptively activates different numbers of experts based on input difficulty to improve computational efficiency and model performance.
Harnessing the Power of Large Language Models for Natural Language to First-Order Logic Translation
Yuan Yang (Georgia Institute of Technology), Faramarz Fekri (Georgia Institute of Technology)
GenerationData-Centric LearningReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextChain-of-Thought
🎯 What it does: Constructed 28K real, diverse, human-verified natural language–first-order logic (NL-FOL) pairs; based on this dataset, LOGICLLAMA was trained using LoRA fine-tuning of LLaMA2 combined with RLHF and logical equivalence rewards to achieve NL-FOL translation and correction of GPT3.5 generated rules.
Harnessing Toulmin’s theory for zero-shot argument explication
Ankita Gupta (University of Massachusetts Amherst), Brendan O’Connor
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes the 'Argument explication' task, which involves decomposing natural language arguments into <claim, premise, warrant> triplets, and achieving zero-shot generation by referencing the Toulmin model in prompts.
Having Beer after Prayer? Measuring Cultural Bias in Large Language Models
Tarek Naous (Georgia Institute of Technology), Wei Xu (Georgia Institute of Technology)
Explainability and InterpretabilityTransformerLarge Language ModelTextBenchmark
🎯 What it does: Constructed the CAMeL dataset and used it to evaluate the cultural bias and adaptation capabilities of large language models in Arabic environments.
HD-Eval: Aligning Large Language Model Evaluators Through Hierarchical Criteria Decomposition
Yuxuan Liu (Peking University), Qi Zhang (Microsoft)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Propose the HD-EVAL framework, which aligns LLM evaluators by utilizing hierarchical criteria decomposition and human preference-guided aggregators.
HealMe: Harnessing Cognitive Reframing in Large Language Models for Psychotherapy
Mengxi Xiao (Wuhan University), Jimin Huang (Wuhan University)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes the HealMe model, which utilizes large language models to implement cognitive restructuring psychotherapy through a three-stage dialogue process (separating situations and thoughts, brainstorming alternative perspectives, and empathetic guidance) to achieve continuous empathy and guidance;
Hide and Seek in Noise Labels: Noise-Robust Collaborative Active Learning with LLMs-Powered Assistance
Bo Yuan (Zhejiang University), Wei Jiang (Ant Group)
ClassificationTransformerLarge Language ModelText
🎯 What it does: Propose the NoiseAL framework, which collaborates small models with large language models for training to address the problem of noisy labels in text classification.
HiRoPE: Length Extrapolation for Code Models Using Hierarchical Position
Kechi Zhang (Peking University), Zhi Jin (Peking University)
AI Code AssistantTransformerLarge Language ModelText
🎯 What it does: Propose HiRoPE, which transforms traditional RoPE into a hierarchical structure. It leverages abstract syntax tree (AST) information of code to split position indices into multi-level representations, enabling long-context reasoning without requiring additional training.
HOLMES: Hyper-Relational Knowledge Graphs for Multi-hop Question Answering using LLMs
Pranoy Panda (Fujitsu Research India), Prathosh Ap
Graph Neural NetworkTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Propose a multi-hop question answering framework called HOLMES, which utilizes a Hyper-Relational Knowledge Graph (Hyper-Relational KG) to extract and refine raw text as input for large language models (LLMs).
How Abilities in Large Language Models are Affected by Supervised Fine-tuning Data Composition
Guanting Dong (Alibaba Group), Jingren Zhou (Alibaba Group)
TransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper investigates the impact of multi-task data combination during the supervised fine-tuning (SFT) phase on the mathematical reasoning, code generation, and human instruction alignment capabilities of large language models (LLMs), and systematically evaluates the effects of model scale, data volume, data proportion, and training strategies on performance.
How Good is Zero-Shot MT Evaluation for Low Resource Indian Languages?
Anushka Singh (Nilekani Centre at AI4Bharat), Mitesh Khapra
Data SynthesisData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This study constructs an MQM evaluation dataset for low-resource Indian languages (Assamese, Kannada, Maithili, Punjabi) and performs zero-shot meta-evaluation of multiple machine translation evaluation metrics on this dataset, exploring the impacts of related language fine-tuning, underlying model replacement, and synthetic data.
How Johnny Can Persuade LLMs to Jailbreak Them: Rethinking Persuasion to Challenge AI Safety by Humanizing LLMs
Yi Zeng (Virginia Tech), Weiyan Shi (Northeastern University)
Safty and PrivacyAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: By viewing LLMs as entities with human communication capabilities, the study leverages persuasion classification from social sciences to construct a system that rewrites ordinary harmful queries into humanized persuasive adversarial prompts (PAP), investigating their impact on LLM jailbreaking.
How to Engage your Readers? Generating Guiding Questions to Promote Active Reading
Peng Cui (ETH Zürich), Mrinmaya Sachan (ETH Zürich)
GenerationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Constructed and analyzed the GUIDINGQ dataset to explore the distribution and function of guiding questions in text, developed models to generate such questions, and verified their positive impact on reading comprehension.
How to Handle Different Types of Out-of-Distribution Scenarios in Computational Argumentation? A Comprehensive and Fine-Grained Field Study
Andreas Waldis (Technical University of Darmstadt), Iryna Gurevych (Technical University of Darmstadt)
Domain AdaptationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: This paper systematically evaluates the performance of pre-trained language models (LMs) in different out-of-distribution (OOD) scenarios for computer argument (CA) tasks—topic shift, domain shift, and language shift;
HyCoRec: Hypergraph-Enhanced Multi-Preference Learning for Alleviating Matthew Effect in Conversational Recommendation
Yongsen Zheng, Liang Lin (Sun Yat Sen University)
Recommendation SystemGraph Neural NetworkTransformerTextGraphBenchmark
🎯 What it does: Propose the HyCoRec framework, which leverages hypergraphs to learn multifaceted preferences from items, entities, words, reviews, and knowledge, alleviating the Matthew effect in dialogue recommendations while improving recommendation and dialogue quality.
Hyper-CL: Conditioning Sentence Representations with Hypernetworks
Young Yoo, Taeuk Kim (Hanyang University)
Representation LearningContrastive LearningTextGraph
🎯 What it does: Propose Hyper-CL, a model that combines hypernetworks with contrastive learning to generate sentence representations tailored to different conditions.
Hypergraph based Understanding for Document Semantic Entity Recognition
Qiwei Li (Wuhan University), Hai Zhao (Shanghai Jiao Tong University)
RecognitionGraph Neural NetworkTransformerVision Language ModelTextMultimodality
🎯 What it does: Propose a document semantic entity recognition method HGA based on hypergraph attention, and construct the HGALayoutLM model based on GraphLayoutLM.
HyperMoE: Towards Better Mixture of Experts via Transferring Among Experts
Hao Zhao (Beijing University of Posts and Telecommunications), Jie Fu (Hong Kong University of Science and Technology)
TransformerMixture of ExpertsTextBenchmark
🎯 What it does: Propose the HyperMoE model, where HyperExperts generated by the hypernetwork transfer unused expert knowledge to selected experts, maintaining sparsity while enhancing expert knowledge availability.
Hyperspherical Multi-Prototype with Optimal Transport for Event Argument Extraction
Guangjun Zhang (Shanxi University), Jiye Liang (Shanxi University)
TransformerContrastive LearningText
🎯 What it does: Propose a high-dimensional spherical model based on multi-prototype (HMPEAE) for document-level event argument extraction, treating argument-prototype matching as an optimal transport problem;
I am a Strange Dataset: Metalinguistic Tests for Language Models
Tristan Thrush (Stanford University), Douwe Kiela (Stanford University)
Explainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Propose and evaluate a new dataset for meta-language self-referential capabilities in large language models (LLMs) called 'I am a Strange Dataset,' designing two subtasks: generation (completion) and verification (truth judgment); using manually constructed contrastive sentences to test whether models can understand and process self-referential semantics and syntax; simultaneously providing non-self-referential control samples to distinguish self-referential difficulty from general meta-language difficulty.
IAPT: Instance-Aware Prompt Tuning for Large Language Models
Wei Zhu (East China Normal University), Guotong Xie (East China Normal University)
Computational EfficiencyRepresentation LearningTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes a parameter-efficient fine-tuning method called Instruction-Aware Prompt Tuning (IAPT), which dynamically generates soft prompts for each input instruction using four soft tokens;
IBSEN: Director-Actor Agent Collaboration for Controllable and Interactive Drama Script Generation
Senyu Han (Shanghai Jiao Tong University), Kai Yu (Shanghai Jiao Tong University)
GenerationTransformerLarge Language ModelAgentic AIPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: By designing the IBSEN LLM framework with three roles—director, actor, and player—to achieve controllable drama script generation and interactive plot progression.
ICLEF: In-Context Learning with Expert Feedback for Explainable Style Transfer
Arkadiy Saakyan (Columbia University), Smaranda Muresan (Columbia University)
GenerationExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Propose a human-AI collaborative framework ICL-EF, which generates interpretable text style transfer data by combining LLM's self-critique and context learning with expert feedback;
Identifying while Learning for Document Event Causality Identification
Cheng Liu (Huazhong University of Science and Technology), Bang Wang (Huazhong University of Science and Technology)
ClassificationRepresentation LearningGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Studied the document-level event causality recognition task, proposing the 'identifying while learning' pattern, which utilizes an iterative process to continuously update event representations and causal graphs during identification, achieving more refined causal direction judgment.
IEPile: Unearthing Large Scale Schema-Conditioned Information Extraction Corpus
Honghao Gui (Zhejiang University), Huajun Chen (Zhejiang University)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: Constructed a large-scale bilingual (English-Chinese) information extraction instruction corpus named IEPILE, and enhanced the zero-shot generalization performance of LLMs in IE tasks through structured, semantic schema-based instruction generation.
IL-TUR: Benchmark for Indian Legal Text Understanding and Reasoning
Abhinav Joshi (IIT Kanpur), Ashutosh Modi (IIT Kanpur)
ClassificationRecognitionGenerationGraph Neural NetworkTransformerLarge Language ModelTextBenchmark
🎯 What it does: Proposed the IL-TUR benchmark, aggregating eight legal text understanding and reasoning tasks covering nine languages (English and Indian languages), and released public data and a leaderboard.
IMBUE: Improving Interpersonal Effectiveness through Simulation and Just-in-time Feedback with Human-Language Model Interaction
Inna Lin, Tim Althoff (University of Washington)
TransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Simulate personalized communication scenarios using a language model and provide instant expert-level feedback to help users practice and improve interpersonal skills in Dialectical Behavior Therapy (DBT) DEAR MAN;
IMO: Greedy Layer-Wise Sparse Representation Learning for Out-of-Distribution Text Classification with Pre-trained Models
Tao Feng (Monash University), Reza Haf
ClassificationDomain AdaptationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed a single-source domain generalization method called IMO, which utilizes layer-wise sparse masks and word-level attention from pre-trained Transformers to learn domain-invariant features, thereby enhancing out-of-distribution (OOD) performance in text classification.
Impacts of Misspelled Queries on Translation and Product Search
Greg Hanneman (Amazon), Taichi Nakatani (Amazon)
RetrievalRecommendation SystemTransformerSupervised Fine-TuningText
🎯 What it does: Study the impact of spelling errors on machine translation (MT) and e-commerce search results, and propose robustness evaluation methods and improvement strategies for this scenario.
Improving Conversational Abilities of Quantized Large Language Models via Direct Preference Alignment
Janghwan Lee (Hanyang University), Jungwook Choi (Hanyang University)
Computational EfficiencyReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Align low-precision (4-bit) quantized LLMs to eliminate token-flipping and improve dialogue quality.
Improving Event Definition Following For Zero-Shot Event Detection
Zefan Cai (University of Wisconsin - Madison), Nanyun Peng (University of California, Los Angeles)
TransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Constructed a diverse event definition dataset called DivED, and performed instruction fine-tuning on LLaMA-2-7B to enhance the event definition following capability in zero-shot event detection.
Improving Hateful Meme Detection through Retrieval-Guided Contrastive Learning
Jingbiao Mei (University of Cambridge), Marcus Tomalin (University of Cambridge)
ClassificationRetrievalTransformerVision Language ModelContrastive LearningMultimodality
🎯 What it does: Propose Retrieval-Guided Contrastive Learning (RGCL), which enhances the sensitivity of the embedding space to hate memes by adding contrastive learning on a CLIP-based vision-language encoder, and implements retrieval-based detection through K-Nearest-Neighbor (KNN) voting;
Improving Large Language Models in Event Relation Logical Prediction
Meiqi Chen (Peking University), Dongsheng Li (Microsoft Research Asia)
TransformerLarge Language ModelSupervised Fine-TuningTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: This paper systematically evaluates the shortcomings of large language models in event relation logical reasoning and proposes three methods: generative, retrieval-based, and fine-tuning approaches, significantly improving their logical consistency and event relation extraction performance.
Improving Text Embeddings with Large Language Models
Liang Wang (Microsoft Corporation), Furu Wei (Microsoft Corporation)
RetrievalRepresentation LearningTransformerLarge Language ModelPrompt EngineeringContrastive LearningTextBenchmark
🎯 What it does: Generate hundreds of thousands of multilingual synthetic retrieval task data using large language models, and fine-tune the open-source decoder LLM (Mistral-7B) with contrastive learning to obtain a text embedding model.
In-context Mixing (ICM): Code-mixed Prompts for Multilingual LLMs
Bhavani Shankar (Indian Institute of Technology Bombay), Pushpak Bhattacharyya (Indian Institute of Technology Bombay)
GenerationTransformerPrompt EngineeringTextChain-of-Thought
🎯 What it does: Investigated a technique for inserting code mixing in multilingual large language model prompts, which randomly replaces content words with their English translations in a few examples to enhance the performance of cross-lingual sequence-to-sequence tasks.
InCharacter: Evaluating Personality Fidelity in Role-Playing Agents through Psychological Interviews
Xintao Wang (Fudan University), Yanghua Xiao (SenseTime)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper proposes the INCHARACTER framework, which uses interview-based psychological assessment to evaluate the personality fidelity of role-playing agents (RPA), and constructs an RPA personality benchmark covering 14 psychological scales;
IndicGenBench: A Multilingual Benchmark to Evaluate Generation Capabilities of LLMs on Indic Languages
Harman Singh (Google DeepMind), Partha Talukdar (Google DeepMind)
GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: Released INDICGENBENCH, covering 29 Indian languages and 13 writing systems across 5 generation tasks (cross-lingual summarization, machine translation, multilingual question answering, cross-lingual question answering), and expanded existing benchmarks with human translations.
IndicIRSuite: Multilingual Dataset and Neural Information Models for Indian Languages
Saiful Haq (IIT Bombay), Pushpak Bhattacharyya (IIT Bombay)
RetrievalCompressionRepresentation LearningData-Centric LearningTransformerLarge Language ModelTextBenchmark
🎯 What it does: Constructed the machine-translated version of the MS MARCO dataset for 11 Indian languages, named INDIC-MARCO, and trained the corresponding monolingual ColBERT model, Indic-ColBERT.
IndicLLMSuite: A Blueprint for Creating Pre-training and Fine-Tuning Datasets for Indian Languages
Mohammed Safi Ur Rahman Khan (Nilekani Centre at AI4Bharat), Mitesh M. Khapra (Nilekani Centre at AI4Bharat)
Data SynthesisData-Centric LearningTransformerLarge Language ModelTextAudio
🎯 What it does: Proposed INDICLLMSUITE, which includes the pre-trained corpus SANGRAHA for 22 Indian languages, the cleaning pipeline SETU, instruction fine-tuning data INDICALIGN (comprising INSTRUCT and TOXIC sections), along with related tools and open-source resources;
Inducing Systematicity in Transformers by Attending to Structurally Quantized Embeddings
Yichen Jiang (UNC Chapel Hill), Mohit Bansal (UNC Chapel Hill)
Representation LearningRecurrent Neural NetworkTransformerText
🎯 What it does: Propose SQ-Transformer to enhance the systematic generalization ability of Transformers through structured quantization of word embeddings and systematic attention mechanisms;
Inference to the Best Explanation in Large Language Models
Dhairya Dalal (University of Galway), Paul Buitelaar (University of Galway)
Explainability and InterpretabilityTransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: Propose an interpretable framework called IBE-Eval based on the philosophical concept of 'Optimal Explanation Reasoning' (IBE), designed to automatically evaluate natural language explanations generated by large language models (LLMs).
InfoLossQA: Characterizing and Recovering Information Loss in Text Simplification
Jan Trienes (University of Duisburg-Essen), Junyi Jessy Li (University of Texas at Austin)
RestorationTransformerLarge Language ModelPrompt EngineeringTextBiomedical Data
🎯 What it does: Investigated the phenomenon of information loss during medical text simplification, proposing the INFOLOSSQA framework that identifies and complements deleted or ambiguous information through QA pair generation and answering.
Insert or Attach: Taxonomy Completion via Box Embedding
Wei Xue, Weiming Lu (Zhejiang University)
Representation LearningGraph Neural NetworkGraph
🎯 What it does: Propose the TAXBOX framework, which utilizes box embedding to complete existing taxonomies (inserting parent nodes or attaching child nodes), and designs two probabilistic scorers for insertion and attachment to eliminate the pseudo-leaf problem.