EMNLP 2024 Papers — Page 13
Conference on Empirical Methods in Natural Language Processing · 1268 papers
Varying Sentence Representations via Condition-Specified Routers
Ziyong Lin (National Key Laboratory of General Artificial Intelligence, Bigai), Zilong Zheng (National Key Laboratory of General Artificial Intelligence, Bigai)
Computational EfficiencyRepresentation LearningTransformerText
🎯 What it does: Proposes a conditional sentence representation framework named CSR based on a conditionally specified router to enhance the performance of a three-encoder model in conditional semantic textual similarity and knowledge graph completion tasks.
Verba volant, scripta volant? Don’t worry! There are computational solutions for protoword reconstruction
Liviu P Dinu, Laurentiu Zoicas (University of Bucharest)
Recurrent Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper constructs the largest ProtoRom database of Romance language cognates and etymologies (19,222 cognate sets) and conducts benchmark experiments on the task of automatic reconstruction of protoforms.
Verifiable, Debuggable, and Repairable Commonsense Logical Reasoning via LLM-based Theory Resolution
Armin Toroghi (University of Toronto), Scott Sanner (University of Toronto)
Explainability and InterpretabilityTransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: Propose the LLM-TRes framework, embedding LLM as a theoretical reasoner into a verifiable reasoning process to address reasoning errors and hallucinations;
Verification and Refinement of Natural Language Explanations through LLM-Symbolic Theorem Proving
Xin Quan (University Of Manchester), Andre Freitas (University Of Manchester)
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: Studied the collaboration between large language models and theorem provers to automatically verify and improve natural language explanations in natural language inference (NLI).
VerifyMatch: A Semi-Supervised Learning Paradigm for Natural Language Inference with Confidence-Aware MixUp
Seo Yeon Park (Hanyang University), Cornelia Caragea (University of Illinois Chicago)
ClassificationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose VerifyMatch, a semi-supervised NLI learning framework that combines LLM-generated pseudo-labels with SSL verifiers, utilizing MixUp to denoise mismatched or low-confidence samples;
VGBench: Evaluating Large Language Models on Vector Graphics Understanding and Generation
Bocheng Zou (University of Wisconsin-Madison), Yong Jae Lee (University of Wisconsin-Madison)
GenerationData SynthesisLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningTextGraphBenchmarkChain-of-Thought
🎯 What it does: Proposes the VGBench benchmark, systematically evaluating large language models (LLMs) in understanding and generating vector graphics (SVG, TikZ, Graphviz), and collects 4,279 QA pairs and 5,845 generated samples.
VHASR: A Multimodal Speech Recognition System With Vision Hotwords
Jiliang Hu, Hai Zhao (Wuhan University)
RecognitionRecurrent Neural NetworkTransformerContrastive LearningImageMultimodalityAudio
🎯 What it does: Propose a dual-stream multimodal speech recognition system called VHASR, which enhances ASR performance by utilizing visual hotwords (vision hotwords).
Video-LLaVA: Learning United Visual Representation by Alignment Before Projection
Bin Lin (Peking University Shenzhen Graduate School), Li Yuan (Peking University Shenzhen Graduate School)
Representation LearningTransformerLarge Language ModelVision Language ModelContrastive LearningImageVideoTextMultimodality
🎯 What it does: Pre-align the visual features of images and videos to the language feature space, then input unified visual representations into large language models (LLMs) via a shared projection layer, achieving a vision-language model that processes images and videos simultaneously in one go.
Video-Text Prompting for Weakly Supervised Spatio-Temporal Video Grounding
Heng Zhao (Agency for Science, Technology and Research), Joey Tianyi Zhou (Agency for Science, Technology and Research)
RetrievalTransformerPrompt EngineeringVision Language ModelContrastive LearningVideoText
🎯 What it does: Propose two weakly supervised spatiotemporal video localization methods, Video-Text Prompting (VTP) and Contrastive Video-Text Prompting (CVTP), by drawing visual prompts (e.g., red circles) on video frames and inserting corresponding text prompts into query texts, preserving global context information and enhancing candidate box feature representation.
VideoCLIP-XL: Advancing Long Description Understanding for Video CLIP Models
Jiapeng Wang (South China University of Technology), Lianwen Jin (South China University of Technology)
RetrievalRepresentation LearningTransformerLarge Language ModelVision Language ModelContrastive LearningVideoTextMultimodalityBenchmark
🎯 What it does: Propose the VideoCLIP-XL model and construct the VILD dataset using an automated data collection system to enhance the understanding of long descriptions by video CLIP.
VideoScore: Building Automatic Metrics to Simulate Fine-grained Human Feedback for Video Generation
Xuan He (University of Waterloo), Wenhu Chen (University of Waterloo)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoTextMultimodality
🎯 What it does: Constructed a large-scale human multi-dimensional evaluation dataset named VIDEOFEEDBACK, and trained an automatic video evaluation model named VIDEOSCORE based on it;
VIEWS: Entity-Aware News Video Captioning
Hammad Ayyubi (Columbia University), Shih-Fu Chang (Columbia University)
GenerationTransformerLarge Language ModelVision Language ModelVideoTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Addressing the task of generating descriptive captions for news videos that include entities (entity-aware video captioning), and propose a modular framework: first use a visual model to detect named entities in the video (Entity Perceiver), then use a large language model to extract relevant background from an external knowledge base (Knowledge Extractor), and finally integrate visual information, entities, and background into the caption generation model.
VIMI: Grounding Video Generation through Multi-modal Instruction
Yuwei Fang (Snap Inc), Sergey Tulyakov (Snap Inc)
GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelDiffusion modelImageVideoTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Construct a retrieval-augmented multi-modal video generation dataset, and adopt two-stage pre-training with multi-modal instruction fine-tuning to achieve high-quality video generation under text+image prompts.
Virtual Personas for Language Models via an Anthology of Backstories
Suhong Moon (University of California Berkeley), David M. Chan (University of California Berkeley)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringTextTabular
🎯 What it does: Proposed and validated the Anthology method, which utilizes open-ended life narratives (backstories) generated by LLMs as prefixes to modulate the outputs of large language models, creating representative, diverse, and consistent virtual human personalities for behavioral research simulations.
Vision-Language Model Fine-Tuning via Simple Parameter-Efficient Modification
Ming Li (University of Tokyo), Masashi Sugiyama (RIKEN Center for Advanced Intelligence Project)
ClassificationComputational EfficiencyKnowledge DistillationSupervised Fine-TuningVision Language ModelImage
🎯 What it does: Propose CLIPFit, which achieves efficient fine-tuning by only fine-tuning specific internal parameters of CLIP without adding external parameters;
Visual Prompting in LLMs for Enhancing Emotion Recognition
Qixuan Zhang (Australian National University), Zhenyue Qin (Yale University)
ClassificationRecognitionTransformerPrompt EngineeringImage
🎯 What it does: This paper proposes the Set-of-Vision (SoV) visual prompting method, which precisely marks each face on images using bounding boxes, numbering, and facial key points, directly inputting the full image into Vision-Language Large Models (VLLM) for zero-shot emotion recognition; simultaneously, it provides an overlapping face processing algorithm and a text-visual hybrid prompting strategy; experiments validate SoV's performance across multiple models (GPT-4V, LLaVA, MiniGPT-4, Video-LLaVA), significantly improving emotion recognition accuracy; it also compares various visual prompting methods, demonstrating SoV's advantages in different difficulty scenarios.
Visual Text Matters: Improving Text-KVQA with Visual Text Entity Knowledge-aware Large Multimodal Assistant
Abhirama Subramanyam Penamakuri (Indian Institute of Technology Jodhpur), Anand Mishra (Indian Institute of Technology Jodhpur)
RetrievalExplainability and InterpretabilityComputational EfficiencyLarge Language ModelVision Language ModelMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Studied the Text-Knowledge-Aware Visual Question Answering (Text-KVQA) task, and proposed an integrated framework that combines visual text entity linking with knowledge retrieval.
VIVA: A Benchmark for Vision-Grounded Decision-Making with Human Values
Zhe Hu (Hong Kong Polytechnic University), Yu Yin (Case Western Reserve University)
Large Language ModelPrompt EngineeringVision Language ModelMultimodalityBenchmark
🎯 What it does: Constructed the VIVA benchmark, collecting 1,240 images annotated with multiple action choices, corresponding human values, and reasons, to evaluate the decision-making ability of vision-language models in visual contexts.
VLEU: a Method for Automatic Evaluation for Generalizability of Text-to-Image Models
Jingtao Cao (Chinese University of Hong Kong), Kam-Fai Wong (Chinese University of Hong Kong)
GenerationTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Proposed the VLEU evaluation metric to automatically measure the generalization capability of text-to-image models; samples visual-textual pairs using LLM and calculates KL divergence scores by assessing semantic consistency with CLIP.
VLFeedback: A Large-Scale AI Feedback Dataset for Large Vision-Language Models Alignment
Lei Li (University of Hong Kong), Qi Liu (University of Hong Kong)
Reinforcement Learning from Human FeedbackLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: Proposed the VLFeedback large-scale AI-annotated visual language feedback dataset, and trained the aligned LVLM Silkie using Direct Preference Optimization (DPO) based on this dataset;
Voices in a Crowd: Searching for clusters of unique perspectives
Nikolas Vitsakis (Heriot Watt University), Ioannis Konstas (Heriot Watt University)
Representation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose a framework that does not require prior annotator metadata: first, use a supervised model to predict individual labels for each annotator on each text, obtaining behavioral embeddings; then perform dimensionality reduction and unsupervised clustering to form different 'voices' groups, and verify whether the clustering corresponds to minority or cross-minority perspectives through post-hoc internal/external metrics and qualitative analysis.
Voices Unheard: NLP Resources and Models for Yorùbá Regional Dialects
Orevaoghene Ahia (University of Washington), Yulia Tsvetkov (University of Washington)
Domain AdaptationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextMultimodalityBenchmarkAudio
🎯 What it does: Constructed the YORÙLECT corpus, providing high-quality parallel text and speech data for four Yoruba dialects (Standard, Ifè, Ìlàje, Ìjèbú), and publicly released it.
VPTQ: Extreme Low-bit Vector Post-Training Quantization for Large Language Models
Yifei Liu (Microsoft), Mao Yang (Microsoft)
CompressionComputational EfficiencyTransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper proposes an extremely low-bit weight quantization method based on Vector Post-Training Quantization (VPTQ), which can compress LLMs to 2-4 bits while maintaining high accuracy.
Walking in Others’ Shoes: How Perspective-Taking Guides Large Language Models in Reducing Toxicity and Bias
Rongwu Xu, Han Qiu (Tsinghua University)
Safty and PrivacyLarge Language ModelPrompt EngineeringText
🎯 What it does: Propose a black-box prompting method based on the social psychology perspective-taking (Perspective-Taking Prompting, PET), which reduces harmful and biased content generated by large language models (LLMs) by first imagining the feelings of different audiences and then performing self-correction before text generation.
Watch Every Step! LLM Agent Learning via Iterative Step-level Process Refinement
Weimin Xiong (Peking University), Sujian Li (Peking University)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningAgentic AIContrastive LearningBenchmark
🎯 What it does: Proposed and implemented the IPR framework, which refines training of LLM agents through an iterative step-wise process, enabling agents to obtain more detailed process supervision in interactive tasks and significantly improving action decision quality.
Waterfall: Scalable Framework for Robust Text Watermarking and Provenance for LLMs
Gregory Kang Ruey Lau (National University of Singapore), Bryan Kian Hsiang Low (National University of Singapore)
Safty and PrivacyTransformerLarge Language ModelText
🎯 What it does: Proposes a training-agnostic scalable text watermarking framework called WATERFALL to protect intellectual property in texts such as articles and code under scenarios involving LLM generation, plagiarism, and unauthorized training.
Weak Reward Model Transforms Generative Models into Robust Causal Event Extraction Systems
Italo Luis Da Silva (King's College London), Yulan He (King's College London)
GenerationData-Centric LearningReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: This paper proposes a complete framework for causal event extraction evaluation and reinforcement learning. First, a human-annotated DeBERTa-Valid model is trained to align closely with manual evaluations. Subsequently, this evaluation model is used as a reward model, and PPO is applied to fine-tune the FLAN-T5 generative model, improving alignment with human preferences. Finally, a weak-to-strong supervision workflow is introduced, enabling the training of a reward model with equivalent performance using only a small amount of labeled data.
What are the Generator Preferences for End-to-end Task-Oriented Dialog System?
Wanshi Xu (Peking University), Yuexian Zou (Peking University)
GenerationRetrievalMixture of ExpertsTextSequential
🎯 What it does: Proposed the RPG framework, introducing a generator preference extractor, entity retriever, and gated preference modulator into end-to-end task-oriented dialogue systems, utilizing an attribute displacement matrix to achieve fine-grained filtering for entity retrieval and provide preference signals to the generator for dialogue turns;
What Are the Odds? Language Models Are Capable of Probabilistic Reasoning
Akshay Paruchuri (Google), Daniel McDuff (Google)
TransformerLarge Language ModelPrompt EngineeringTabularBenchmarkFinance Related
🎯 What it does: This paper constructs a benchmark dataset specifically for evaluating the probabilistic reasoning capabilities of language models, and systematically evaluates the performance of three large language models (Gemini 1.0 Ultra, GPT-4-Turbo, GPT-3.5-Turbo) on this dataset, exploring the performance of tasks such as percentile estimation, sampling, and probability calculation under idealized distributions and real-world distributions.
What do Large Language Models Need for Machine Translation Evaluation?
Shenbin Qian (University of Surrey), Fred Blain
TransformerLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: This paper systematically evaluates the information needs and prompting strategies of large language models (LLMs) in machine translation quality assessment, covering zero-shot, chain-of-thought (CoT), and few-shot methods;
What is “Typological Diversity” in NLP?
Esther Ploeger (Aalborg University), Johannes Bjerva (Aalborg University)
Data-Centric LearningTextReview/Survey PaperBenchmark
🎯 What it does: This paper systematically reviews the usage of 'typological diversity' in NLP literature and proposes two quantitative metrics (Average Pairwise Language Distance MPSD and Grambank Grammatical Feature Coverage) to objectively assess the linguistic sample diversity in research.
What is lost in Normalization? Exploring Pitfalls in Multilingual ASR Model Evaluations
Kavya Manohar (Digital University Kerala), Leena G Pillai (Digital University Kerala)
RecognitionExplainability and InterpretabilityTransformerSupervised Fine-TuningTextAudio
🎯 What it does: Explores and empirically demonstrates the defects and misleading performance improvements in text normalization within multilingual ASR evaluations, particularly under the Indic script.
What the Harm? Quantifying the Tangible Impact of Gender Bias in Machine Translation with a Human-centered Study
Beatrice Savoldi (Fondazione Bruno Kessler), Luisa Bentivogli (Fondazione Bruno Kessler)
GenerationTransformerText
🎯 What it does: This paper quantitatively evaluates the service quality gap and economic cost caused by gender bias for female users through post-editing experiments on machine translation (MT) outputs in real user scenarios.
What’s Mine becomes Yours: Defining, Annotating and Detecting Context-Dependent Paraphrases in News Interview Dialogs
Anna Wegmann (Utrecht University), Dong Nguyen (Utrecht University)
ClassificationData-Centric LearningTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: The study defines, annotates, and automatically detects context-dependent synonymous sentences (contextually related synonymous expressions) in news interview dialogues.
When Context Leads but Parametric Memory Follows in Large Language Models
Yufei Tao (Portland State University), Ameeta Agrawal (Portland State University)
Explainability and InterpretabilityTransformerPrompt EngineeringText
🎯 What it does: This paper conducts experiments on the responses of nine mainstream large language models under different context scales to investigate how models allocate local context knowledge and global parameter knowledge in knowledge-consistent scenarios, and evaluates their hallucination tendencies.
When Generative Adversarial Networks Meet Sequence Labeling Challenges
Yu Tong (Shantou University), Jiang Dazhi (Midea Group)
ClassificationGenerationTransformerGenerative Adversarial NetworkText
🎯 What it does: Proposed a unified SLGAN framework to address the generative adversarial issues in sequence labeling tasks.
When Is Multilinguality a Curse? Language Modeling for 250 High- and Low-Resource Languages
Tyler A. Chang (University of California San Diego), Benjamin K. Bergen (University of California San Diego)
Representation LearningTransformerLarge Language ModelText
🎯 What it does: Pretrained and systematically evaluated monolingual and multilingual GPT-2 Transformer models for over 250 languages, investigating the impact of language distribution on low-resource and high-resource languages.
When LLMs Meets Acoustic Landmarks: An Efficient Approach to Integrate Speech into Large Language Models for Depression Detection
Xiangyu Zhang (University of New South Wales), Julien Epps (University of New South Wales)
ClassificationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextMultimodalityAudio
🎯 What it does: Proposes an efficient multimodal depression detection method that combines acoustic landmarks with large language models (LLM).
When Parts Are Greater Than Sums: Individual LLM Components Can Outperform Full Models
Ting-Yun Chang (University of Southern California), Robin Jia (University of Southern California)
ClassificationExplainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: This paper analyzes the in-context learning (ICL) mechanism of large language models (LLMs) by decomposing the Transformer into independent components such as attention heads and MLPs, and identifies high-performing, inefficient, and label-biased components.
When Reasoning Meets Information Aggregation: A Case Study with Sports Narratives
Yebowen Hu (University of Central Florida), Fei Liu (Emory University)
GenerationData SynthesisLarge Language ModelTextChain-of-Thought
🎯 What it does: This paper analyzes step-by-step descriptions of NBA basketball games to study the performance of large language models in information aggregation and reasoning tasks. It proposes a game narrative generation method called SPORTSGEN and evaluates the model's score calculation ability under different segmentation strategies (batch splitting, player splitting, holistic processing).
Where Am I From? Identifying Origin of LLM-generated Content
Liying Li (Hong Kong Polytechnic University), Minhao Cheng (Pennsylvania State University)
TransformerLarge Language ModelText
🎯 What it does: Built an LLM content tracing framework that embeds secret watermarks in generated text to enable author tracking.
Where am I? Large Language Models Wandering between Semantics and Structures in Long Contexts
Seonmin Koo (Korea University), Heuiseok Lim (Korea University)
TransformerLarge Language ModelTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Systematically evaluate the alignment between large language models (LLMs) in question answering (QA) and evidence selection tasks under long context environments, constructing a diverse test set through semantic relevance and structural diversity, and exploring error sources between the two tasks.
Where is the signal in tokenization space?
Renato Geh, Guy Van Den Broeck
Representation LearningTransformerLarge Language ModelText
🎯 What it does: Studied the probability distribution of large language models (LLMs) in non-canonical tokenization spaces, proving that finding the most probable tokenization and computing marginal probabilities are respectively NP-hard and #P-hard problems, and estimated marginal probabilities via importance sampling. Experiments showed that non-canonical tokenization improves performance in question-answering tasks.
Which Programming Language and What Features at Pre-training Stage Affect Downstream Logical Inference Performance?
Fumiya Uchiyama (University of Tokyo), Yutaka Matsuo (University of Tokyo)
Data-Centric LearningAI Code AssistantTransformerLarge Language ModelTextBenchmark
🎯 What it does: This study begins with the pre-training phase, systematically comparing the impact of different programming languages (ten in total) and natural languages (Wikipedia, FineWeb, C4) on downstream logical reasoning performance when training large language models. It uses from-scratch GPT-2 and LLaMA models, and evaluates 3-shot proximal learning on FLD and bAbi logical reasoning tasks.
Which questions should I answer? Salience Prediction of Inquisitive Questions
Yating Wu (University of Texas at Austin), Junyi Jessy Li (University of Texas at Austin)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextChain-of-Thought
🎯 What it does: Built and trained the QSALIENCE model to predict the significance of curiosity-driven questions and analyzed its association with answerability and summary quality.
Whispers that Shake Foundations: Analyzing and Mitigating False Premise Hallucinations in Large Language Models
Hongbang Yuan (Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences), Jun Zhao (Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences)
OptimizationExplainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: This paper conducts a systematic analysis of error reasoning caused by false premises in large language models, and proposes the FAITH method to reduce such hallucinations by constraining a small number of critical attention heads.
Whiteboard-of-Thought: Thinking Step-by-Step Across Modalities
Sachit Menon (Columbia University), Carl Vondrick (Columbia University)
RecognitionPrompt EngineeringVision Language ModelImageTextMultimodalityChain-of-Thought
🎯 What it does: Proposes the Whiteboard-of-Thought method, enabling multi-modal large models to generate code for drawing intermediate visual results and re-inputting them into the model to achieve cross-modal visual reasoning
Who is better at math, Jenny or Jingzhen? Uncovering Stereotypes in Large Language Models
Zara Siddique (Cardiff University), Luis Espinosa-Anke (AMPLYFI)
TransformerLarge Language ModelText
🎯 What it does: This paper constructs a large-scale GlobalBias dataset (876,000 sentences, covering 40 gender-ethnicity combinations), and systematically evaluates the bias in internal representations and generated outputs of large language models on gender and ethnic cross-stereotypes using adjusted perplexity (APX) and zero-shot generation tasks.
Why do objects have many names? A study on word informativeness in language use and lexical systems
Eleonora Gualdoni (Universitat Pompeu Fabra), Gemma Boleda (Universitat Pompeu Fabra)
ImageText
🎯 What it does: This paper proposes a word information measure based on the visual space (CIELAB), investigating the adaptability of lexical selection in different contexts. The measure is applied to English and Chinese color naming data, demonstrating that soft mapping (a single reference can correspond to multiple words) is the optimal lexicon structure.
Why Does New Knowledge Create Messy Ripple Effects in LLMs?
Jiaxin Qin (University of Illinois Urbana Champaign), Heng Ji (Stanford University)
Explainability and InterpretabilityLarge Language ModelText
🎯 What it does: This paper proposes and validates the GradSim metric to evaluate the effectiveness of information propagation (ripple effect) in language models after knowledge editing.
Will LLMs Replace the Encoder-Only Models in Temporal Relation Classification?
Gabriel Roccabruna (University of Trento), Giuseppe Riccardi (University of Trento)
ClassificationExplainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Evaluate and conduct interpretability analysis on the performance and decision-making processes of seven open-source and closed-source LLMs in the Temporal Relation Classification task, comparing their performance with the RoBERTa encoder model.
With Ears to See and Eyes to Hear: Sound Symbolism Experiments with Multimodal Large Language Models
Tyler Loakman (University of Sheffield), Chenghua Lin (University of Sheffield)
RecognitionGenerationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: This study designs three types of experiments—shape-sound, size-sound, and word-image-sound—to evaluate the cognitive ability of multimodal large language models (VLM/LLM) in understanding sound symbolism.
Word Alignment as Preference for Machine Translation
Qiyu Wu (University of Tokyo), Yoshimasa Tsuruoka (University of Tokyo)
GenerationOptimizationTransformerLarge Language ModelText
🎯 What it does: This paper proposes a framework based on Word Alignment Preference (WAP) to improve word alignment in large language models (LLMs) for machine translation through direct preference optimization (DPO), thereby reducing hallucination and omission phenomena.
Words Matter: Reducing Stigma in Online Conversations about Substance Use with Large Language Models
Layla Bouzoubaa (Drexel University), Rezvaneh Rezapour (Drexel University)
GenerationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: In non-drug-related Reddit communities, stigmatizing language toward people who use drugs (PWUS) was studied, and large language models (LLMs) were utilized to remove posts containing targeted stigma, generating more empathetic text.
Words Worth a Thousand Pictures: Measuring and Understanding Perceptual Variability in Text-to-Image Generation
Raphael Tang (Comcast AI Technologies), Ferhan Ture (Comcast AI Technologies)
GenerationPrompt EngineeringVision Language ModelDiffusion modelImageText
🎯 What it does: This study investigates how prompts in text-to-image generation models affect the perceptual diversity of generated images, and proposes a human-calibrated diversity metric called W1KP.
Working Memory Identifies Reasoning Limits in Language Models
Chunhui Zhang (Dartmouth College), Soroush Vosoughi (Dartmouth College)
TransformerLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: By integrating the working memory concept from cognitive science with the n-back task and the Big-bench Hard (BBH) benchmark, this work systematically evaluates the limitations of LLMs of different scales in working memory and reasoning, and proposes a CoT+ prompting improvement scheme based on error analysis;
World to Code: Multi-modal Data Generation via Self-Instructed Compositional Captioning and Filtering
Jiacong Wang (University of Chinese Academy of Sciences), Jun Xiao (University of Chinese Academy of Sciences)
Data SynthesisPrompt EngineeringImageTextMultimodality
🎯 What it does: Propose a multi-modal data construction pipeline named W2C, which utilizes existing Vision-Language Models (VLMs) to generate descriptions through self-instruction and organizes them in the form of Python code;
WorryWords: Norms of Anxiety Association for over 44k English Words
Saif M. Mohammad (National Research Council Canada)
Data-Centric LearningTextTabular
🎯 What it does: Created a worry lexicon called WorryWords containing 44,450 English terms, collected and averaged scores from over 1,000 participants to obtain real-valued anxiety association scores and classification labels for each term; translated the lexicon into over 100 languages; further used it to study the relationship between anxiety and other emotional dimensions, the association between children's vocabulary acquisition and anxiety, and to track anxiety arcs in text streams.
WPO: Enhancing RLHF with Weighted Preference Optimization
Wenxuan Zhou (Zoom Video Communications), Chenguang Zhu
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningTextBenchmark
🎯 What it does: Propose Weighted Preference Optimization (WPO), which addresses distribution gaps and improves LLM alignment by weighted simulation of on-policy learning through offline preference pairs.
xCOMET-lite: Bridging the Gap Between Efficiency and Quality in Learned MT Evaluation Metrics
Daniil Larionov (University of Mannheim), Steffen Eger (University of Mannheim)
CompressionComputational EfficiencyKnowledge DistillationTransformerText
🎯 What it does: The study compresses the original large-scale xCOMET model into xCOMET-lite with only 2.6% parameters using compression techniques such as quantization, pruning, and knowledge distillation, while maintaining 92.1% of the original model's quality and outperforming COMET-22 by 6.4% on the WMT22 benchmark.
XDetox: Text Detoxification with Token-Level Toxicity Explanations
Beomseok Lee (Hanyang University), Yong Suk Choi (Hanyang University)
GenerationExplainability and InterpretabilityTransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: Propose the XDetox method, which first uses DecompX to perform token-level offensive explanations, accurately identifying and masking toxic words; then employs MARCO's filling mechanism to generate non-offensive content, and re-ranks the generated candidate sentences by recalculating DecompX's importance, selecting the sentence with the least toxicity as the final output.
XplainLLM: A Knowledge-Augmented Dataset for Reliable Grounded Explanations in LLMs
Zichen Chen (University of California Santa Barbara), Misha Sra (University of California Santa Barbara)
Explainability and InterpretabilityData-Centric LearningGraph Neural NetworkTransformerLarge Language ModelTextGraphRetrieval-Augmented Generation
🎯 What it does: Constructed a knowledge-enhanced dataset named XplainLLM and proposed an explanation framework based on knowledge graphs and graph attention networks (GAT) to generate trustworthy and fact-based LLM explanations;
You Make me Feel like a Natural Question: Training QA Systems on Transformed Trivia Questions
Tasnim Kabir (University of Maryland), Jordan Lee Boyd-Graber (University of Maryland)
GenerationKnowledge DistillationData-Centric LearningTransformerLarge Language ModelPrompt EngineeringContrastive LearningTextRetrieval-Augmented Generation
🎯 What it does: The study converts the carefully designed long puzzles from Quiz Bowl (QB) into short and natural web query-style questions, and uses the converted data to train a QA system;
ZEBRA: Zero-Shot Example-Based Retrieval Augmentation for Commonsense Question Answering
Francesco Maria Molfese (Sapienza University of Rome), Roberto Navigli (Sapienza University of Rome)
RetrievalExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposes the ZEBRA framework, which completes common-sense QA under zero-shot conditions through retrieving complete cases, case-guided knowledge generation, and knowledge-based reasoning.
Zero-shot Cross-domain Dialogue State Tracking via Context-aware Auto-prompting and Instruction-following Contrastive Decoding
Xiaoyu Dong (Hong Kong Polytechnic University), Xiao-Ming Wu (Hong Kong Polytechnic University)
Domain AdaptationTransformerLarge Language ModelPrompt EngineeringContrastive LearningText
🎯 What it does: This paper proposes the CAPID framework, which generates dynamic slot queries through context-aware automatic prompting and achieves zero-shot cross-domain dialogue state tracking using instruction-following contrastive decoding.
Zero-Shot Cross-Lingual NER Using Phonemic Representations for Low-Resource Languages
Jimin Sohn (GIST), David R Mortensen (Carnegie Mellon University)
RecognitionRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Propose using IPA phonetic representations for NER in zero-shot cross-lingual tasks without target language training data
Zero-shot Cross-Lingual Transfer for Synthetic Data Generation in Grammatical Error Detection
Gaetan Lopez Latouche (Ubisoft La Forge), Benjamin Swanson (Ubisoft La Forge)
ClassificationData SynthesisDomain AdaptationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Leveraging the zero-shot cross-lingual transfer capability of multilingual pre-trained models, first train a multilingual artificial error generation model, then generate synthetic error data for the target language, followed by two-stage fine-tuning to achieve multi-language grammar error detection without human annotations.
Zero-Shot Detection of LLM-Generated Text using Token Cohesiveness
Shixuan Ma (Beijing University of Posts and Telecommunications), Quan Wang (Beijing University of Posts and Telecommunications)
Anomaly DetectionLarge Language ModelText
🎯 What it does: Propose a zero-shot LLM text detection method called TOCSIN based on token cohesiveness, and use it as a general module to enhance the performance of existing detectors.