EMNLP 2023 Papers — Page 5
Conference on Empirical Methods in Natural Language Processing · 1047 papers
Fidelity-Enriched Contrastive Search: Reconciling the Faithfulness-Diversity Trade-Off in Text Generation
Wei-Lin Chen (National Taiwan University), Chung-Chi Chen (Artificial Intelligence Research Center AIST)
GenerationTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Proposed a new decoding strategy called Fidelity-Enriched Contrastive Search (FECS), which enhances the faithfulness of text generation and suppresses repetition by incorporating a semantic similarity reward into Contrastive Search.
Fighting Fire with Fire: The Dual Role of LLMs in Crafting and Detecting Elusive Disinformation
Jason Lucas (Pennsylvania State University), Dongwon Lee (Pennsylvania State University)
GenerationAnomaly DetectionTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Propose the F3 framework, leveraging LLMs for generating, purifying, and zero-shot detection of fake information, constructing an end-to-end 'fighting fire with fire' process.
Find-2-Find: Multitask Learning for Anaphora Resolution and Object Localization
Cennet Oguz (German Research Center for Artificial Intelligence), Josef van Genabith (German Research Center for Artificial Intelligence)
Object DetectionTransformerVision Language ModelVideoTextMultimodality
🎯 What it does: Propose the Find2Find dataset and design a joint multi-task learning model to address coreference resolution and object localization under multimodal visual-language ambiguity.
Finding Authentic Counterhate Arguments: A Case Study with Public Figures
Abdullah Albanyan (Prince Sattam Bin Abdulaziz University), Eduardo Blanco (University of Arizona)
ClassificationRetrievalTransformerSupervised Fine-TuningTextRetrieval-Augmented Generation
🎯 What it does: A corpus consisting of 250 hate tweets and 54,816 paragraphs was constructed, with real counter-arguments against hate annotated; a method to retrieve online articles for real counter-arguments was proposed.
Fine-grained Conversational Decoding via Isotropic and Proximal Search
Yuxuan Yao (City University of Hong Kong), Linqi Song (City University of Hong Kong)
GenerationTransformerLarge Language ModelText
🎯 What it does: Proposed a fine-grained dialogue decoding method called Isometric and Nearest Neighbor Search (IPS), which generates semantically focused and information-rich responses by simultaneously considering the clustering of already generated words (neighbors) and their separation from context sentences (isometric) during decoding.
Fine-grained Medical Vision-Language Representation Learning for Radiology Report Generation
Siyuan Wang (University of Sydney), Qi Peng (Newcastle University)
GenerationRepresentation LearningTransformerVision Language ModelContrastive LearningTextMultimodalityBiomedical Data
🎯 What it does: Propose PhenotypeCLIP, a phenotype-based contrastive learning framework that generates medical imaging reports by leveraging fine-grained image-sentence alignment.
Fine-tuned LLMs Know More, Hallucinate Less with Few-Shot Sequence-to-Sequence Semantic Parsing over Wikidata
Silei Xu (Stanford University), Monica Lam (Stanford University)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Constructed the WikiWebQuestions dataset and implemented a seq2seq semantic parser, WikiSP, based on LLaMA. By integrating entity linking and SPARQL generation, it enables fact-based question answering on Wikidata. Additionally, GPT-3 is used as an auxiliary reasoning module to reduce model hallucinations.
FinEntity: Entity-level Sentiment Classification for Financial Texts
Yixuan Tang (Hong Kong University Of Science And Technology), Justin Tang
ClassificationRecurrent Neural NetworkTransformerLarge Language ModelTextBenchmarkFinance Related
🎯 What it does: Constructed and made publicly available the FinEntity dataset for financial entity sentiment classification, providing entity-level sentiment annotations; conducted benchmark experiments on multiple pre-trained language models (BERT, FinBERT, etc.) and ChatGPT, and validated the practicality of this dataset in regulatory and investment scenarios through a cryptocurrency market case study.
FinGPT: Large Generative Models for a Small Language
Risto Luukkonen (TurkuNLP Group, University of Turku), Sampo Pyysalo (TurkuNLP Group, University of Turku)
GenerationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Developed large-scale generative models for Finnish, FinGPT and BLUUMI.
FLatS: Principled Out-of-Distribution Detection with Feature-Based Likelihood Ratio Score
Haowei Lin (Peking University), Yuntian Gu (Peking University)
Anomaly DetectionScore-based ModelText
🎯 What it does: This paper proposes an OOD detection method called FLatS based on the likelihood ratio in the feature space, achieving theoretically optimal detection strategies.
Focus Your Attention (with Adaptive IIR Filters)
Shahar Lutati (Tel Aviv University), Lior Wolf (Tel Aviv University)
Computational EfficiencyImageTextSequential
🎯 What it does: Designed a new "Focus" layer, which first preprocesses the sequence using an adaptive second-order IIR filter and then applies local attention, aiming to efficiently capture short-term and long-term dependencies.
FOCUS: Effective Embedding Initialization for Monolingual Specialization of Multilingual Models
Konstantin Dobler (University of Potsdam), Gerard de Melo (University of Potsdam)
Domain AdaptationComputational EfficiencyRepresentation LearningTransformerText
🎯 What it does: This paper proposes an embedding initialization method called FOCUS, which rapidly adapts the target language's dedicated tokenizer while keeping the structure of multilingual pre-trained models unchanged.
FreeAL: Towards Human-Free Active Learning in the Era of Large Language Models
Ruixuan Xiao (Zhejiang University), Haobo Wang (Zhejiang University)
Knowledge DistillationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Propose a collaborative learning framework named FreeAL, which utilizes a large language model (LLM) as an active annotator and employs a small language model (SLM) as a weak learner to filter high-quality examples, achieving fully automated active learning without human annotation;
From Dissonance to Insights: Dissecting Disagreements in Rationale Construction for Case Outcome Classification
Shanshan Xu (Technical University of Munich), Matthias Grabmair (Technical University of Munich)
ClassificationExplainability and InterpretabilityData-Centric LearningTransformerText
🎯 What it does: This paper constructs the RAVE dataset, collecting token-level rationality annotations from two human rights law experts on the factual details of cases from the European Court of Human Rights, and systematically analyzes inter-annotator differences;
From Heuristic to Analytic: Cognitively Motivated Strategies for Coherent Physical Commonsense Reasoning
Zheyuan Zhang (University of Michigan), Joyce Chai (University of Michigan)
TransformerPrompt EngineeringTextPhysics RelatedChain-of-Thought
🎯 What it does: This paper proposes a 'Heuristic-Analytic Reasoning' (HAR) framework based on the theory of human dual cognition, combining high-level decision-making to guide low-level reasoning, thereby enhancing the coherence and reliability of pre-trained language models in physical commonsense reasoning tasks.
From Multilingual Complexity to Emotional Clarity: Leveraging Commonsense to Unveil Emotions in Code-Mixed Dialogues
Shivani Kumar (IIIT Delhi), Tanmoy Chakraborty (IIT Delhi)
RecognitionTransformerLarge Language ModelText
🎯 What it does: Studied the emotion recognition task in code-mixed dialogues, constructed the E-MASAC dataset, and proposed the COFFEE method.
From Parse-Execute to Parse-Execute-Refine: Improving Semantic Parser for Complex Question Answering over Knowledge Base
Wangzhen Guo (Sun Yat-Sen University), Jian Yin (Sun Yat-Sen University)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelTextGraphBenchmarkChain-of-Thought
🎯 What it does: Proposed a three-stage complex knowledge graph question answering framework named PER-KBQA, which first generates KoPL logical forms using semantic parsing, then obtains intermediate reasoning steps through alignment, and finally uses these steps as context to guide the generation of the final answer.
From Values to Opinions: Predicting Human Behaviors and Stances Using Value-Injected Large Language Models
Dongjun Kang (Sungkyunkwan University), JinYeong Bak (Sungkyunkwan University)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: In this study, the authors propose a Value Injection Method (VIM), which injects fine-grained value distributions into LLAMA through two tasks: argument generation and question answering. Subsequently, the model with injected values is used to predict group opinions and behaviors.
From Wrong To Right: A Recursive Approach Towards Vision-Language Explanation
Jiaxin Ge (UC Berkeley), Boyi Li (UC Berkeley)
Explainability and InterpretabilityTransformerLarge Language ModelMultimodality
🎯 What it does: Propose the ReVisE recursive visual explanation framework, which utilizes multi-step generation self-correcting explanations
G-Eval: NLG Evaluation using Gpt-4 with Better Human Alignment
Yang Liu (Microsoft Azure AI), Chenguang Zhu (Microsoft Azure AI)
GenerationTransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: Propose the G-EVAL framework, which combines large language models (GPT-4) with chain-of-thought (CoT) reasoning, further integrated with a form-filling paradigm, to perform reference-free automatic evaluation of natural language generation (NLG) text.
GATITOS: Using a New Multilingual Lexicon for Low-resource Machine Translation
Alexander Jones (Google Research), Ishank Saxena (Google Research)
Data-Centric LearningTransformerText
🎯 What it does: This paper significantly improves translation quality by incorporating bilingual dictionaries for data augmentation in low-resource and unsupervised machine translation training.
GazeVQA: A Video Question Answering Dataset for Multiview Eye-Gaze Task-Oriented Collaborations
Muhammet Ilaslan, Mike Shou
Object DetectionSegmentationTransformerVision Language ModelContrastive LearningImageVideoTextMultimodalityBenchmark
🎯 What it does: Constructed a collaborative task VQA dataset named GazeVQA containing multi-perspective videos and eye movement data, and proposed the AssistGaze model capable of answering questions with text, images, and videos.
GD-COMET: A Geo-Diverse Commonsense Inference Model
Mehar Bhatia (University of British Columbia), Vered Shwartz (University of British Columbia)
GenerationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelTextMultimodality
🎯 What it does: Proposed and trained GD-COMET, a geo-diverse version of COMET capable of generating cross-cultural common sense reasoning.
GEM: Gestalt Enhanced Markup Language Model for Web Understanding via Render Tree
Zirui Shao (Zhejiang University), Xiaozhong Liu (Worcester Polytechnic Institute)
Representation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes GEM (Gestalt Enhanced Markup Language Model), which enhances web understanding capabilities without requiring visual input by leveraging visual information from web rendering trees (style and position) during pre-training, and introducing two pre-training tasks based on Gestalt principles (Same Textual Style Prediction and Proximate Nodes Prediction).
GEMINI: Controlling The Sentence-Level Summary Style in Abstractive Text Summarization
Guangsheng Bao (Zhejiang University), Yue Zhang (Westlake University)
GenerationTransformerMixture of ExpertsText
🎯 What it does: Proposed the GEMINI model, which integrates a rewriter and a generator to achieve sentence-level style control for each summary sentence, imitating the condensation and fusion writing techniques found in human summaries.
Gender Biases in Automatic Evaluation Metrics for Image Captioning
Haoyi Qiu (University of California Los Angeles), Nanyun Peng (University of California Los Angeles)
GenerationReinforcement LearningContrastive LearningMultimodality
🎯 What it does: This paper systematically investigates the gender bias in cross-modal evaluation metrics (e.g., CLIPScore, GPTScore) for image captioning tasks, constructs a large-scale PAO-EVALBIAS dataset, and analyzes how these biases affect generative models (especially those trained with reinforcement learning) and evaluation results; subsequently, it proposes a simple linear combination of metrics (CLIPScore + CIDEr) to significantly reduce bias while maintaining correlation with human judgments.
Generating and Evaluating Tests for K-12 Students with Language Model Simulations: A Case Study on Sentence Reading Efficiency
Eric Zelikman (Stanford University), Nick Haber (Stanford University)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Generate and calibrate K-12 students' sentence reading efficiency test items using large-scale language models, automatically constructing parallelizable test forms.
Generating Commonsense Counterfactuals for Stable Relation Extraction
Xin Miao (Wuhan University), Tieyun Qian (Wuhan University)
Data SynthesisTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Propose a framework for generating causal counterfactual data (CCG) that complies with common-sense constraints, aiming to enhance the stability of relation extraction models in low-resource, cross-domain, and adversarial scenarios.
Generating Data for Symbolic Language with Large Language Models
Jiacheng Ye (University of Hong Kong), Tao Yu (University of Hong Kong)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Leveraging large language models (e.g., Codex) as data generators, combined with prompt engineering and self-consistent verification to generate symbolic language data with high annotation costs, then training with a small-scale task model (e.g., T5-Large) to significantly reduce deployment and inference costs;
Generating Summaries with Controllable Readability Levels
Leonardo F. R. Ribeiro (Amazon), Markus Dreyer (Amazon)
GenerationTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This paper studies how to finely control readability levels in summary generation.
Generative Adversarial Training with Perturbed Token Detection for Model Robustness
Jiahao Zhao (Chinese Academy of Sciences), Wenji Mao (University of Chinese Academy of Sciences)
Adversarial AttackTransformerLarge Language ModelGenerative Adversarial NetworkTextBenchmark
🎯 What it does: Proposed a generative adversarial training framework called GenerAT, which enhances the robustness of text models by achieving gradient-driven generative adversarial attacks and perturbation token detection through shared embeddings.
Generative Spoken Language Model based on continuous word-sized audio tokens
Robin Algayres (ENS, INRIA, INSERM, UPEC, PSL Research University), Emmanuel Dupoux (ENS, INRIA, INSERM, UPEC, PSL Research University)
GenerationTransformerContrastive LearningAudio
🎯 What it does: Proposes a generative speech language model called tGSLM based on continuous word-level audio embeddings, which can generate speech text without discretization units.
Generative Table Pre-training Empowers Models for Tabular Prediction
Tianping Zhang (Tsinghua University), Qian Liu (Sea AI Lab)
Data SynthesisSafty and PrivacyTransformerLarge Language ModelPrompt EngineeringTabular
🎯 What it does: Propose TAPTAP, a GPT-based table pre-training model that enhances the performance of various backend models on table prediction tasks by generating high-quality synthetic tables.
GenEx: A Commonsense-aware Unified Generative Framework for Explainable Cyberbullying Detection
Krishanu Maity (Indian Institute of Technology Patna), Pushpak Bhattacharyya (Indian Institute of Technology Bombay)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Proposes GenEx, an explainable cyberbullying detection framework for code-mixed languages, and constructs the BullyExplain dataset annotated with maliciousness, emotion, target, and explanatory rationale.
GeoLM: Empowering Language Models for Geospatially Grounded Language Understanding
Zekun Li (University of Minnesota), Muhao Chen (University of California)
Representation LearningTransformerContrastive LearningTextMultimodality
🎯 What it does: This paper proposes GEOLM, a pre-trained language model capable of simultaneously processing natural language text and geospatial data, achieving alignment between language and geospatial context;
GLEN: General-Purpose Event Detection for Thousands of Types
Sha Li (University of Illinois Urbana-Champaign), Jiawei Han (University of Illinois Urbana-Champaign)
ClassificationRecognitionTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Constructed a large-scale general event detection dataset GLEN (3,465 event types, 208,000 sentences) and proposed a multi-stage event detection model CEDAR, which performs trigger word identification, sentence-level type ranking, and trigger word-level type classification separately.
GLEN: Generative Retrieval via Lexical Index Learning
Sunkyung Lee (Sungkyunkwan University), Jongwuk Lee (Sungkyunkwan University)
RetrievalTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: Proposed a generative retrieval framework called GLEN, which utilizes dynamic vocabulary-based indexing learning to directly generate document identifiers corresponding to queries, and enhances retrieval performance through two-phase training (keyword assignment and ranking-based ID fine-tuning) as well as collision-free inference.
Global Voices, Local Biases: Socio-Cultural Prejudices across Languages
Anjishnu Mukherjee (George Mason University), Antonios Anastasopoulos (George Mason University)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Build and use WEATHub to execute Word Embedding Association Test (WEAT) across 24 languages, introduce five new social bias dimensions (toxicity, disability discrimination, sexual orientation, education, immigration), conduct systematic comparisons between multilingual and monolingual models and different contextual embedding methods, and perform in-depth bias analysis for seven Indian languages.
GlobalBench: A Benchmark for Global Progress in Natural Language Processing
Yueqi Song (Carnegie Mellon University), Graham Neubig (George Mason University)
TextBenchmark
🎯 What it does: Propose GlobalBench, a continuously expanding multilingual multitask benchmark and leaderboard, tracking the performance, utility, and fairness of NLP systems across all languages, and incentivizing improvements for under-served languages through a reward mechanism.
GNAT: A General Narrative Alignment Tool
Tanzir Pial (Stony Brook University), Steven Skiena (Stony Brook University)
RetrievalRepresentation LearningTransformerText
🎯 What it does: Propose a generic narrative text alignment tool called GNAT, which utilizes the Smith-Waterman algorithm and various text similarity metrics to achieve local alignment of texts of different lengths and domains, and calculate statistical significance.
Goal-Driven Explainable Clustering via Language Descriptions
Zihan Wang (University of California San Diego), Ruiqi Zhong (University of California Berkeley)
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: Proposed a new text clustering task called GOALEX, which requires clustering results to align with user-specified goals and provide natural language explanations for the meaning of each cluster.
GPT-RE: In-context Learning for Relation Extraction using Large Language Models
Zhen Wan (Kyoto University), Sadao Kurohashi (Kyoto University)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose GPT-RE, which leverages a large language model (GPT-3) for relation extraction based on in-context learning (ICL), and enhances example quality through task-aware retrieval and gold-standard guided reasoning.
GPTAraEval: A Comprehensive Evaluation of ChatGPT on Arabic NLP
Md Tawkat Islam Khondaker (University of British Columbia), Muhammad Abdul-Mageed
ClassificationRecognitionGenerationTransformerSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: Conduct a large-scale evaluation of ChatGPT on Arabic natural language processing (NLP) tasks, covering 44 tasks and over 60 datasets, comparing the performance of ChatGPT, GPT-4, BLOOMZ, and two specially fine-tuned Arabic models (MARBERTV2, AraT5). Analyze the performance of the two models on modern standard Arabic (MSA) and various dialects (DA), and further validate the quality of generated outputs through both human and GPT-4 evaluations.
GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints
Joshua Ainslie (Google Research), Sumit Sanghai (Google Research)
Computational EfficiencyTransformerSupervised Fine-TuningText
🎯 What it does: The study converts multi-head attention models into multi-query/grouped query attention models to improve decoding speed.
Gradient-based Gradual Pruning for Language-Specific Multilingual Neural Machine Translation
Dan He (Zoom Video Communications), Marco Turchi (Zoom Video Communications)
Computational EfficiencyTransformerText
🎯 What it does: This paper proposes a gradient-based progressive pruning method aimed at extracting optimal sub-networks for each language pair from multi-lingual NMT models, thereby alleviating parameter interference and improving translation quality.
GradSim: Gradient-Based Language Grouping for Effective Multilingual Training
Mingyang Wang (Bosch Center for Artificial Intelligence), Hinrich Schuetze (LMU Munich)
OptimizationRepresentation LearningTransformerSupervised Fine-TuningText
🎯 What it does: Studied a language grouping method based on gradient similarity to help select sets of languages that mutually positively influence each other during multilingual training.
Grammar-Constrained Decoding for Structured NLP Tasks without Finetuning
Saibo Geng (EPFL), Robert West (EPFL)
GenerationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposes the Grammar-Constrained Decoding (GCD) framework, which uses formal grammar to constrain the generation process of large language models (LLMs), enabling them to perform various structured NLP tasks without fine-tuning.
Granularity Matters: Pathological Graph-driven Cross-modal Alignment for Brain CT Report Generation
Yanzhao Shi (Beijing University of Technology), Ying Liu (Peking University Third Hospital)
GenerationGraph Neural NetworkContrastive LearningImageTextMultimodalityComputed Tomography
🎯 What it does: Propose a cross-modal alignment model called PGCA based on pathological graphs, using fine-grained graph convolution and contrastive learning to enhance brain CT report generation.
Graph vs. Sequence: An Empirical Study on Knowledge Forms for Knowledge-Grounded Dialogue
Yizhe Yang (Beijing Institute of Technology), Yang Gao (Beijing Institute of Technology)
GenerationTransformerLarge Language ModelTextGraph
🎯 What it does: Conduct systematic comparative experiments on the effects of knowledge graphs and text as two forms of knowledge in knowledge-driven dialogues, exploring the mutual adaptability between models and knowledge as well as few-shot performance.
GreedyCAS: Unsupervised Scientific Abstract Segmentation with Normalized Mutual Information
Yingqiang Gao (University of Zurich and ETH Zurich), Richard Hahnloser (University of Zurich and ETH Zurich)
SegmentationTransformerTextBiomedical Data
🎯 What it does: This paper proposes an unsupervised method called GreedyCAS for dividing scientific paper abstracts into premise segments and conclusion segments. The method treats abstracts as cyclic sentence sequences, uses normalized mutual information (NMI) as the optimization objective, and employs greedy search (including a nearest-neighbor-based batch variant) to determine two segmentation boundaries.
GROOViST: A Metric for Grounding Objects in Visual Storytelling
Aditya K Surikuchi (University of Amsterdam), Raquel Fernández (University of Amsterdam)
Vision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Proposed a new no-reference evaluation metric called GROOViST to measure the visual groundedness in visual story generation.
Grounding Visual Illusions in Language: Do Vision-Language Models Perceive Illusions Like Humans?
Yichi Zhang (University of Michigan), Joyce Chai (University of Michigan)
Large Language ModelVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Constructed the GVIL dataset, covering five types of visual illusions, and designed four benchmark tasks (SameDiffQA, RefQA, AttrQA, RefLoc) to evaluate the perception and language grounding capabilities of visual language models (VLM) regarding illusions.
Guideline Learning for In-Context Information Extraction
Chaoxu Pang (Key Lab of Intelligent Information Processing of Chinese Academy of Sciences Institute of Computing Technology), Ping Luo (Key Lab of Intelligent Information Processing of Chinese Academy of Sciences Institute of Computing Technology)
Data-Centric LearningMeta LearningTransformerLarge Language ModelPrompt EngineeringTextFinance RelatedRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose a rule learning framework that automatically generates and retrieves rules using error cases to enhance the in-context learning performance of large models in information extraction.
Hallucination Detection for Generative Large Language Models by Bayesian Sequential Estimation
Xiaohua Wang (Fudan University), Xuanjing Huang (Fudan University)
RetrievalAnomaly DetectionTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Propose a hallucination detection framework based on Bayesian sequential analysis, dynamically determining the number of external evidence retrievals and performing hierarchical judgment.
Hallucination Mitigation in Natural Language Generation from Large-Scale Open-Domain Knowledge Graphs
Xiao Shi (University of Texas at Arlington), Chengkai Li (University of Texas at Arlington)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningTextGraph
🎯 What it does: Proposed a new large-scale open-domain knowledge graph to text generation dataset called GraphNarrative, and designed a sentence trimming algorithm to mitigate information hallucination.
HalOmi: A Manually Annotated Benchmark for Multilingual Hallucination and Omission Detection in Machine Translation
David Dale (Meta), Marta R. Costa-jussà (Meta)
Anomaly DetectionTransformerLarge Language ModelTextBenchmark
🎯 What it does: Released a manually annotated dataset containing 18 language pairs for detecting hallucinations and omissions in machine translation
HaluEval: A Large-Scale Hallucination Evaluation Benchmark for Large Language Models
Junyi Li (Gaoling School of Artificial Intelligence, Renmin University of China), Ji-Rong Wen (Gaoling School of Artificial Intelligence, Renmin University of China)
TransformerLarge Language ModelTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Constructed HaluEval, an evaluation benchmark for large language models containing 35,000 hallucination samples generated both by human annotation and automatic generation, used to test models' ability to detect hallucinations.
Harnessing Black-Box Control to Boost Commonsense in LM’s Generation
Yufei Tian (University of California, Los Angeles), Nanyun Peng (University of California, Los Angeles)
GenerationTransformerLarge Language ModelText
🎯 What it does: Propose the BOOST framework, which leverages a frozen pre-trained language model (PTLM) and an auxiliary NADO head, guided by a reference-free general knowledge evaluator (O-Scorer), to generate more commonsense-aligned text.
Have LLMs Advanced Enough? A Challenging Problem Solving Benchmark For Large Language Models
Daman Arora (Microsoft Research), Mausam (IIT Delhi)
TransformerLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Proposed JEEBENCH, a challenging question bank consisting of 515 high-difficulty mathematics, physics, and chemistry problems from 8 years of the Indian IIT JEE-Advanced exams, to evaluate LLMs' long-range reasoning and professional knowledge application capabilities.
Hi Guys or Hi Folks? Benchmarking Gender-Neutral Machine Translation with the GeNTE Corpus
Andrea Piergentili (University of Trento), Luisa Bentivogli (Fondazione Bruno Kessler)
GenerationData SynthesisTransformerLarge Language ModelTextBenchmark
🎯 What it does: Investigated gender-neutral translation from English to Italian, constructed the GeNTE benchmark dataset, and explored automatic evaluation methods.
Hi-ArG: Exploring the Integration of Hierarchical Argumentation Graphs in Language Pretraining
Jingcong Liang (Fudan University), Zhongyu Wei (Fudan University)
Representation LearningGraph Neural NetworkTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Proposed a hierarchical argumentation graph Hi-ArG and built an automated construction process, followed by further pretraining of language models using the GreaseArG multimodal model and custom pretraining tasks
HiddenTables and PyQTax: A Cooperative Game and Dataset For TableQA to Ensure Scale and Data Privacy Across a Myriad of Taxonomies
William Watson (J.P. Morgan AI Research), Manuela Veloso (J.P. Morgan AI Research)
Safty and PrivacyLarge Language ModelPrompt EngineeringTabularChain-of-Thought
🎯 What it does: Proposed the HiddenTables collaborative game, utilizing code-generating LLMs to interact with a secure Oracle to protect table data privacy and generating the PyQTax dataset.
Hidding the Ghostwriters: An Adversarial Evaluation of AI-Generated Student Essay Detection
Xinlin Peng (University of Chinese Academy of Sciences), Yingfei Sun (University of Chinese Academy of Sciences)
ClassificationAdversarial AttackTransformerLarge Language ModelTextBenchmark
🎯 What it does: Constructed the AIG-ASAP dataset to evaluate adversarial perturbations on AI-generated student essays and tested the robustness of existing AI-generated content detectors on this dataset.
Hierarchical Pretraining on Multimodal Electronic Health Records
Xiaochen Wang (Pennsylvania State University), Fenglong Ma (Pennsylvania State University)
ClassificationRecurrent Neural NetworkTransformerAuto EncoderContrastive LearningTextMultimodalityTabularTime SeriesBiomedical DataElectronic Health Records
🎯 What it does: Proposes a hierarchical multimodal pretraining framework called MEDHMP for pretraining on electronic health records (EHR) and downstream tasks;
HistAlign: Improving Context Dependency in Language Generation by Aligning with History
David Wan (University of North Carolina Chapel Hill), Mohit Bansal (University of North Carolina Chapel Hill)
GenerationTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: Proposed the HISTALIGN training method, which significantly enhances the contextual dependency of language models by aligning hidden states with historical memory in cached language models.
Holistic Inter-Annotator Agreement and Corpus Coherence Estimation in a Large-scale Multilingual Annotation Campaign
Nicolas Stefanovitch (European Commission Joint Research Centre), Jakub Piskorski (Polish Academy of Sciences)
RetrievalRepresentation LearningData-Centric LearningTransformerText
🎯 What it does: In a multilingual persuasion technique annotation project covering six languages and approximately 1,600 news articles, a systematic analysis of inter-annotator agreement and annotation difficulty was conducted, and a global consistency metric based on multilingual sentence embeddings—Holistic IAA—was proposed. Additionally, the impact of a two-round calibration process on corpus consistency was evaluated.
Homophone Disambiguation Reveals Patterns of Context Mixing in Speech Transformers
Hosein Mohebbi (Tilburg University), Afra Alishahi (Tilburg University)
RecognitionTransformerAudio
🎯 What it does: Investigate the context mixing mechanism of Transformers in speech recognition, using the syntactic ambiguity of French homonyms as a case study to verify and improve context mixing scoring methods in the text domain.
Hop, Union, Generate: Explainable Multi-hop Reasoning without Rationale Supervision
Wenting Zhao (Cornell University), Alexander Rush (Cornell University)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Propose an interpretable multi-hop question answering method HUG that does not require rationalization labels, capable of modeling multi-hop reasoning at both document-level and sentence-level while generating answers and rationalizations.
How do languages influence each other? Studying cross-lingual data sharing during LM fine-tuning
Rochelle Choenni (University of Amsterdam), Ekaterina Shutova (University of Amsterdam)
Explainability and InterpretabilityRepresentation LearningData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Use the training data attribution method TracIn to analyze the impact of multilingual language models on different language training samples during fine-tuning, revealing cross-lingual data sharing mechanisms.
How Do Large Language Models Capture the Ever-changing World Knowledge? A Review of Recent Advances
Zihan Zhang (University of Technology Sydney), Jun Wang (University College London)
ClassificationTransformerLarge Language ModelTextReview/Survey PaperRetrieval-Augmented Generation
🎯 What it does: Reviews the latest methods for keeping LLMs aligned with world knowledge without retraining, and proposes a systematic classification and comparison.
How Does Generative Retrieval Scale to Millions of Passages?
Ronak Pradeep (University of Waterloo), Vinh Tran (Google Research)
RetrievalTransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper systematically evaluates the performance of generative retrieval methods across different scales of document collections ranging from 100,000 to 8.6 million documents through large-scale empirical experiments, and conducts experiments for the first time on the complete MS MARCO 8.8M passage collection.
How to Enhance Causal Discrimination of Utterances: A Case on Affective Reasoning
Hang Chen (Xi'an Jiaotong University), Wenjing Zhu (Du Xiao Man Inc)
Data SynthesisExplainability and InterpretabilityRepresentation LearningRecurrent Neural NetworkGraph Neural NetworkAuto EncoderTextSequential
🎯 What it does: This paper addresses the causal discrimination problem in the emotional reasoning task (ARC) by constructing a structural causal model (SCM) with added i.i.d. noise, proposing the Cogn framework capable of handling variable-length dialogues, and utilizing an autoencoder to learn implicit causality (noise) as observable latent variables, followed by validation on synthetic and real dialogue data.
Human Learning by Model Feedback: The Dynamics of Iterative Prompting with Midjourney
Shachar Don-Yehiya (Hebrew University of Jerusalem), Omri Abend (Hebrew University of Jerusalem)
Data-Centric LearningReinforcement Learning from Human FeedbackConvolutional Neural NetworkTransformerPrompt EngineeringDiffusion modelImageText
🎯 What it does: Collect and analyze the iterative text-to-image (Text-to-Image) prompts publicly released by Midjourney Discord, constructing threads (prompt threads) and studying the changes in language features and convergence behavior of prompts during multi-round interactions.
Human Raters Cannot Distinguish English Translations from Original English Texts
Shira Wein (Georgetown University)
RecognitionText
🎯 What it does: This paper conducted a human evaluation to test whether English reviewers can distinguish between original English texts and translated English texts;
HutCRS: Hierarchical User-Interest Tracking for Conversational Recommender System
Mingjie Qian (Sun Yat-sen University), Liang Lin (Sun Yat-sen University)
Recommendation SystemGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: Designed a multi-round conversational recommendation framework called HutCRS based on a hierarchical interest tree, which operates in two phases: first identifying the aspects of user interest, then asking about attributes related to these aspects or directly recommending items.
Hybrid Inverted Index Is a Robust Accelerator for Dense Retrieval
Peitian Zhang (Renmin University of China), Jing Yao (Microsoft Research Asia)
RetrievalKnowledge DistillationText
🎯 What it does: Propose a Hybrid Inverted Index (HI2) that unifies document embedding clustering and significant term indexing into a single inverted file to accelerate dense retrieval;
HyperNetwork-based Decoupling to Improve Model Generalization for Few-Shot Relation Extraction
Liang Zhang (Xiamen University), Jie Zhou (Tencent Inc)
ClassificationMeta LearningTransformerSupervised Fine-TuningContrastive LearningText
🎯 What it does: This paper proposes a decoupling method based on HyperNetwork, utilizing an encoder, network generator, and generated relation classifier to achieve rapid adaptation and generalization in few-shot relation extraction (FSRE).
Hyperpolyglot LLMs: Cross-Lingual Interpretability in Token Embeddings
Andrea W Wen-Yi (Cornell University), David Mimno (Cornell University)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Investigate the input word embedding layer of multilingual LLMs, exploring their cross-lingual interpretability and geometric structure, and discover differences between models (XLM-RoBERTa and mT5) in encoding language information and semantic similarity.
HyperRank: Hyperbolic Ranking Model for Unsupervised Keyphrase Extraction
Mingyang Song (Beijing Jiaotong University), Liping Jing (Beijing Jiaotong University)
RetrievalRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Proposed HyperRank, an unsupervised keyword extraction ranking model in hyperbolic space.
HyperRouter: Towards Efficient Training and Inference of Sparse Mixture of Experts
Truong Giang Do (Institute for Infocomm Research A STAR), Steven Hoi (Institute for Infocomm Research A STAR)
Computational EfficiencyTransformerMixture of ExpertsText
🎯 What it does: Studied the routing problem in Sparse Mixture-of-Experts (SMoE) and proposed HyperRouter, which dynamically generates routing parameters to balance the advantages and disadvantages of training and frozen routing.
IAG: Induction-Augmented Generation Framework for Answering Reasoning Questions
Zhebin Zhang (Huawei), Zhao Cao (Huawei)
GenerationRetrievalKnowledge DistillationTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Propose a framework named IAG (Induction-Augmented Generation), which integrates retrieval-augmented generation with inductive reasoning, leveraging inductive knowledge generated by large language models and retrieved documents to jointly assist in answering implicit reasoning questions.
IBADR: an Iterative Bias-Aware Dataset Refinement Framework for Debiasing NLU models
Xiaoyue Wang (Xiamen University), Hua Wu (Baidu Inc)
Data-Centric LearningTransformerLarge Language ModelText
🎯 What it does: Designed an iterative dataset refinement framework named IBADR without predefined bias features, which uses a shallow model to quantify sample bias, iteratively generates low-bias pseudo-samples, adds them to the sample pool, and finally retrains the NLU model.
IC3: Image Captioning by Committee Consensus
David M. Chan (University of California Berkeley), John Canny (University of California Berkeley)
GenerationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageText
🎯 What it does: By performing diversity sampling on pre-trained image captioning models and then using a large language model (LLM) for summarization, generating a single, information-complete image description.
Identification of Multimodal Stance Towards Frames of Communication
Maxwell Weinzierl, Sanda Harabagiu
ClassificationData SynthesisGraph Neural NetworkTransformerVision Language ModelImageTextMultimodality
🎯 What it does: Proposed the first multimodal stance annotation dataset MMVAX-STANCE for 113 COVID-19 vaccine communication frameworks, generated 46,000 synthetic multimodal instances by inferring text-image relationships, and constructed a stance detection model.
Identifying Informational Sources in News Articles
Alexander Spangher (University of Southern California), Jonathan May (University of Southern California)
ClassificationRetrievalTransformerLarge Language ModelSupervised Fine-TuningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Constructed the largest annotated news source dataset and proposed a source prediction task to study the combination patterns of sources in news.
Identifying Statements Crucial for Awareness of Interpretive Nonsense to Prevent Communication Breakdowns
Tomoyuki Maekawa (Keio University), Michita Imai (Keio University)
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: Proposed and validated a method to prevent understanding bias in online conversations by identifying key sentences (SCAIN) that cause explanation errors.
Ideology Takes Multiple Looks: A High-Quality Dataset for Multifaceted Ideology Detection
Songtao Liu (Huazhong University of Science and Technology), Bang Wang (Huazhong University of Science and Technology)
ClassificationTransformerTextBenchmark
🎯 What it does: Constructed a multidimensional ideological detection task and proposed the MITweet dataset along with a 12-dimensional multi-faceted ideological pattern.
IDTraffickers: An Authorship Attribution Dataset to link and connect Potential Human-Trafficking Operations on Text Escort Advertisements
Vageesh Saxena (Law & Tech Lab Maastricht University), Gerasimos Spanakis (Law & Tech Lab Maastricht University)
RecognitionRetrievalExplainability and InterpretabilityTransformerContrastive LearningTextBenchmark
🎯 What it does: Constructed and released a trafficker author attribution dataset named IDTraffickers, trained and evaluated author identification and verification models using this dataset to help law enforcement link potential traffickers in online prostitution ads.
IEKG: A Commonsense Knowledge Graph for Idiomatic Expressions
Ziheng Zeng (University of Illinois Urbana-Champaign), Suma Bhat (University of Illinois Urbana-Champaign)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningTextGraph
🎯 What it does: Constructed an idiom knowledge graph IEKG based on ATOMIC, and utilized it to enhance the understanding and reasoning capabilities of pre-trained language models for idioms.
IfQA: A Dataset for Open-domain Question Answering under Counterfactual Presuppositions
Wenhao Yu (Tecent AI Seattle Lab), Ashish Sabharwal (Allen Institute for AI)
RetrievalTransformerLarge Language ModelTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Constructed an open-domain question-answering dataset named IfQA, specifically designed to evaluate models' capabilities in counterfactual reasoning tasks involving 'if' assumptions.
Ignore This Title and HackAPrompt: Exposing Systemic Vulnerabilities of LLMs Through a Global Prompt Hacking Competition
Sander Schulhoff (University of Maryland), Jordan Boyd-Graber (University of Maryland)
Safty and PrivacyAdversarial AttackTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Collected and publicly released over 600,000 human-generated adversarial prompts through the Global Prompt Hacking competition to test the security of large language models (LLMs).
Image Manipulation via Multi-Hop Instructions - A New Dataset and Weakly-Supervised Neuro-Symbolic Approach
Harman Singh (Indian Institute Of Technology Delhi), Parag Singla (Indian Institute Of Technology Delhi)
GenerationExplainability and InterpretabilityGraph Neural NetworkVision Language ModelGenerative Adversarial NetworkImageTextMultimodality
🎯 What it does: Propose a weakly supervised neuro-symbolic image editing framework called NEUROSIM, which can perform add, delete, and modify operations on multi-object images based on multi-step natural language instructions;
Impressions: Visual Semiotics and Aesthetic Impact Understanding
Julia Kruk, Diyi Yang
GenerationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelMultimodalityBenchmark
🎯 What it does: Construct and release the Impressions dataset for studying the semi-semantic, emotional, and aesthetic impacts of images.
Improved Pseudo Data for Machine Translation Quality Estimation with Constrained Beam Search
Xiang Geng (Nanjing University), Shujian Huang (Nanjing University)
Data SynthesisTransformerLarge Language ModelText
🎯 What it does: Utilize constrained Beam Search (CBSQE) during NMT decoding to retain high-probability words from the reference, generating more accurate pseudo QE data;
Improved Unsupervised Chinese Word Segmentation Using Pre-trained Knowledge and Pseudo-labeling Transfer
Hsiu-Wen Li (National Cheng Kung University), Hung-Yu Kao (National Cheng Kung University)
SegmentationComputational EfficiencyRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed a Chinese word segmentation framework that utilizes an unsupervised segmentation model to generate pseudo labels and fine-tunes a pre-trained BERT model based on these pseudo labels, improving segmentation performance while significantly reducing training time.
Improving Bias Mitigation through Bias Experts in Natural Language Understanding
Eojin Jeon (Korea University), SangKeun Lee (Korea University)
ClassificationTransformerSupervised Fine-TuningMixture of ExpertsImageText
🎯 What it does: This paper proposes introducing a binary classifier called Bias Experts between the main model and auxiliary model to enhance bias mitigation in natural language understanding tasks.
Improving Biomedical Abstractive Summarisation with Knowledge Aggregation from Citation Papers
Chen Tang (University of Surrey), Chenghua Lin (University of Sheffield)
GenerationTransformerLarge Language ModelTextBiomedical DataRetrieval-Augmented Generation
🎯 What it does: Propose an attention-based reference aggregation model that incorporates the abstracts of cited papers as external knowledge to perform self-summarization of biomedical texts in conjunction with the main paper content; meanwhile, construct a large-scale abstract dataset containing citation information.
Improving Chinese Pop Song and Hokkien Gezi Opera Singing Voice Synthesis by Enhancing Local Modeling
Peng Bai (Xiamen University), Xiaodong Shi (Xiamen University)
GenerationData SynthesisTransformerAudio
🎯 What it does: Study the singing voice synthesis of Mandarin pop songs and Minnan opera, proposing two techniques to enhance local modeling, significantly improving the quality of synthesized audio.
Improving Dialogue Discourse Parsing via Reply-to Structures of Addressee Recognition
Yaxin Fan (Soochow University), Qiaoming Zhu (Soochow University)
RecognitionRecurrent Neural NetworkTransformerReinforcement LearningText
🎯 What it does: The study proposes a multi-task framework that jointly learns dialogue discourse parsing and addressee recognition to leverage partially overlapping structural information between the two tasks for improved parsing performance.
Improving Diversity of Demographic Representation in Large Language Models via Collective-Critiques and Self-Voting
Preethi Lahoti (Google Research), Jilin Chen (Google Research)
GenerationData-Centric LearningTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Propose a method named Collective-Critique and Self-Voting (CCSV) by enabling large language models (LLMs) to self-criticize, rewrite, and vote, significantly enhancing the representation of population and cultural diversity in their generated text.