ACL 2024 Papers — Page 9
Annual Meeting of the Association for Computational Linguistics · 940 papers
Synergistic Interplay between Search and Large Language Models for Information Retrieval
Jiazhan Feng (Peking University), Daxin Jiang (Beihang University)
RetrievalTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: The InteR framework is proposed to achieve more accurate zero-shot retrieval by iteratively refining information through combining retrieval models (RM) with large language models (LLM).
Synthesizing Text-to-SQL Data from Weak and Strong LLMs
Jiaxi Yang (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences), Chang Zhou (Alibaba Group)
Data SynthesisAI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextTabularBenchmark
🎯 What it does: Propose a text-to-SQL (text-to-SQL) model called SENSE based on open-source large language models (LLMs), leveraging 'strong data' generated by large models to enhance multi-domain generalization, and combining 'weak data' generated by small models with executor feedback for preference learning, enabling the model to learn from errors.
Systematic Task Exploration with LLMs: A Study in Citation Text Generation
Furkan Şahinuç (Technical University of Darmstadt), Iryna Gurevych (Technical University of Darmstadt)
GenerationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes a systematic framework that leverages large language models (LLMs) to explore the reference text generation task. By systematically varying input components, prompt templates, and evaluation metrics, the study constructs a new reference text dataset and conducts multi-dimensional assessments of model outputs.
T-Eval: Evaluating the Tool Utilization Capability of Large Language Models Step by Step
Zehui Chen (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
Large Language ModelAgentic AIPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes T-Eval, a benchmark for evaluating LLM's tool usage capability based on fine-grained steps.
T2S-GPT: Dynamic Vector Quantization for Autoregressive Sign Language Production from Text
Aoxiong Yin (Zhejiang University), Yueting Zhuang (Zhejiang University)
GenerationTransformerLarge Language ModelAuto EncoderVideoTextMultimodality
🎯 What it does: Proposes a two-stage text-to-sign language generation framework: first, a Dynamic Vector Quantized Variational Autoencoder (DVQ-VAE) encodes sign language sequences into variable-length discrete codes, then a GPT-style autoregressive model (T2S-GPT) generates discrete code sequences and their durations based on text, enabling automatic text-to-sign language generation.
Talk With Human-like Agents: Empathetic Dialogue Through Perceptible Acoustic Reception and Reaction
Haoqiu Yan (University of Science and Technology of China), Linli Xu (Tencent Youtu Lab)
GenerationTransformerLarge Language ModelSupervised Fine-TuningAuto EncoderTextMultimodalityAudio
🎯 What it does: Built a multi-modal dialogue system named PerceptiveAgent, which captures the speaker's acoustic information through a speech captioner, understands multi-modal context and generates empathetic responses using an LLM, and generates expressive speech replies via an MSMA-Synthesizer.
TAMS: Translation-Assisted Morphological Segmentation
Enora Rice (University of Colorado Boulder), Katharina von der Wense (University of Colorado Boulder)
SegmentationRecurrent Neural NetworkTransformerLarge Language ModelText
🎯 What it does: Propose a model called TAMS that utilizes translation information to assist in canonical segmentation for low-resource languages.
TaPERA: Enhancing Faithfulness and Interpretability in Long-Form Table QA by Content Planning and Execution-based Reasoning
Yilun Zhao (Yale University), Chen Zhao (NYU Shanghai)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTabularChain-of-Thought
🎯 What it does: Proposes the TAPERA framework for long-form table question answering (LFTQA), achieving modular reasoning and generation through QA-based content planning, table reasoning with executable Python programs, and answer generation.
TaSL: Continual Dialog State Tracking via Task Skill Localization and Consolidation
Yujie Feng (Hong Kong Polytechnic University), Xiao-Ming Wu (Hong Kong Polytechnic University)
Computational EfficiencyTransformerSupervised Fine-TuningText
🎯 What it does: Propose a continuous dialogue state tracking framework named TaSL based on task skill localization and integration, achieving continuous learning without memory replay;
TasTe: Teaching Large Language Models to Translate through Self-Reflection
Yutong Wang (Harbin Institute of Technology), Min Zhang (Harbin Institute of Technology)
GenerationLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: Propose the TASTE framework, which achieves machine translation through two-stage self-reflection using LLM;
TaxoLLaMA: WordNet-based Model for Solving Multiple Lexical Semantic Tasks
Viktor Moskvoretskii (HSE University), Irina Nikishina (Universität Hamburg)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: An LLM instruction-tuned on WordNet, named TaxoLLaMA, is proposed, which unifies solving four types of lexical semantic tasks (Hypernym Discovery, Taxonomy Enrichment, Lexical Entailment, Taxonomy Construction).
Tell Me More! Towards Implicit User Intention Understanding of Language Model Driven Agents
Cheng Qian (Tsinghua University), Maosong Sun (Tsinghua University)
TransformerLarge Language ModelSupervised Fine-TuningAgentic AITextBenchmark
🎯 What it does: Proposed the IN3 benchmark to evaluate language model-driven agents' understanding of users' implicit intentions, and trained the Mistral-Interact model as a pre-interaction module to enhance agents' task understanding and execution efficiency.
Temperature-scaling surprisal estimates improve fit to human reading times – but does it do so for the “right reasons”?
Tong Liu (LMU Munich), Vera Demberg (Saarland University)
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: This paper investigates the application of temperature scaling to the surprisal of predictions from large language models (GPT-2), and explores its impact on predicting human reading duration.
Temporal Knowledge Question Answering via Abstract Reasoning Induction
Ziyang Chen (National University of Defense Technology), Min Zhang (Harbin Institute of Technology (Shenzhen))
TransformerLarge Language ModelPrompt EngineeringTextGraphBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed and implemented the Abstract Reasoning Induction (ARI) framework, which employs constructivist ideas to divide time knowledge reasoning into two stages: knowledge-irrelevant and knowledge-related. It generates abstract methods through historical reasoning samples to guide LLMs in proactive and continuous learning and reasoning.
Text Embedding Inversion Security for Multilingual Language Models
Yiyi Chen (Aalborg University), Johannes Bjerva (Aalborg University)
Safty and PrivacyRepresentation LearningAdversarial AttackTransformerLarge Language ModelContrastive LearningTextBenchmark
🎯 What it does: Investigate multilingual embedding inversion attacks, explore the feasibility of black-box multilingual and cross-lingual text recovery, and propose a mask-based defense method.
Text-like Encoding of Collaborative Information in Large Language Models for Recommendation
Yang Zhang (University of Science and Technology of China), Xiangnan He (University of Science and Technology of China)
Recommendation SystemTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Propose a recommendation method called BinLLM, which converts collaborative information into a binary sequence (compressible as dot-decimal notation) and directly inputs it as text into LLMs.
Text-to-Song: Towards Controllable Music Generation Incorporating Vocal and Accompaniment
Zhiqing Hong (Zhejiang University), Zhimeng Zhang (Zhejiang University)
GenerationTransformerPrompt EngineeringGenerative Adversarial NetworkContrastive LearningTextAudio
🎯 What it does: Proposed and implemented the Text-to-Song task, using a two-stage Melodist model to first generate vocals and then accompaniment, ultimately synthesizing a complete song.
That’s Optional: A Contemporary Exploration of “that” Omission in English Subordinate Clauses
Ella Rabinovich (Academic College of Tel Aviv-Yaffo)
TransformerLarge Language ModelText
🎯 What it does: Explored the grammatical phenomenon of optionally omitting the conjunction 'that' in English subordinate clauses and verified the impact of the Uniform Information Density (UID) hypothesis on this grammatical choice.
The Belebele Benchmark: a Parallel Reading Comprehension Dataset in 122 Language Variants
Lucas Bandarkar (Meta AI), Madian Khabsa (Meta AI)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Constructed a parallel multiple-choice reading comprehension benchmark called BELEBELE, which includes 122 languages, and generated 900 questions and answers through manual translation and strict quality control of the FLORES-200 text;
The Dawn After the Dark: An Empirical Study on Factuality Hallucination in Large Language Models
Junyi Li (Renmin University of China), Ji-Rong Wen (Renmin University of China)
TransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBiomedical DataBenchmarkRetrieval-Augmented Generation
🎯 What it does: Developed a new factuality hallucination evaluation benchmark called HaluEval 2.0, proposed an LLM-based extraction-validation framework for hallucination detection, and systematically studied the sources of hallucinations and mitigation methods (such as RLHF, retrieval-augmented, and advanced decoding) during pre-training, SFT, RLHF, and inference stages.
The Earth is Flat because...: Investigating LLMs’ Belief towards Misinformation via Persuasive Conversation
Rongwu Xu (Tsinghua University), Han Qiu (Tsinghua University)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextSequential
🎯 What it does: Constructed the Farm dataset and tested the susceptibility of LLMs to factual misinformation through multi-turn persuasive dialogues
The Echoes of Multilinguality: Tracing Cultural Value Shifts during Language Model Fine-tuning
Rochelle Choenni (University of Amsterdam), Ekaterina Shutova (University of Amsterdam)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper investigates how the language and data sources during the fine-tuning of multilingual language models (MLMs) influence the cultural values encoded in the models, and evaluates the alignment of value changes with human values through Cloze probing using questions from the World Values Survey (WVS).
The Fine-Tuning Paradox: Boosting Translation Quality Without Sacrificing LLM Abilities
David Stap (Amazon AGI), Ke Tran (Amazon AGI)
GenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Investigate which existing translation advantages of large language models (LLMs) are weakened after fine-tuning with parallel corpora, and propose a hybrid fine-tuning strategy combining monolingual and parallel corpora to mitigate these declines.
The Heuristic Core: Understanding Subnetwork Generalization in Pretrained Language Models
Adithya Bhaskar (Princeton University), Danqi Chen (Princeton University)
Explainability and InterpretabilityComputational EfficiencyTransformerText
🎯 What it does: Structured pruning of pre-trained language models to extract subnetworks that maintain the original model's in-domain performance, and discovering that all subnetworks share a set of attention heads, forming a so-called 'heuristic core' to explain differences in out-of-domain generalization among different subnetworks.
The Hidden Space of Transformer Language Adapters
Jesujoba Alabi, Mor Geva (Tel Aviv University)
Domain AdaptationRepresentation LearningTransformerText
🎯 What it does: Studied the mechanism of how Transformer language adapters internally evolve and adapt to new languages when freezing the pre-trained model, analyzing the distribution and role of the adaptation process across different layers.
The MERSA Dataset and a Transformer-Based Approach for Speech Emotion Recognition
Enshi Zhang (Florida International University), Christian Poellabauer (Florida International University)
RecognitionRecurrent Neural NetworkTransformerTextMultimodalityAudio
🎯 What it does: This paper first constructs a multimodal dataset named MERSA, which includes natural and scripted speech, text transcripts, physiological signals, and self-reported emotion questionnaires. It also proposes a Transformer framework based on wav2vec2.0, BERT, and LSTM, which performs early fusion of speech and text to predict emotion dimensions (arousal, valence, dominance).
The Probabilities Also Matter: A More Faithful Metric for Faithfulness of Free-Text Explanations in Large Language Models
Noah Siegel, Maria Perez-Ortiz (University College London)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Investigate the credibility of free-text explanations generated by large language models (Llama-2 series), propose Correlational Explanatory Faithfulness (CEF) metric, and implement Correlational Counterfactual Test (CCT) on Counterfactual Test, evaluating the credibility of explanations across three major NLP tasks.
The Unreasonable Effectiveness of Easy Training Data for Hard Tasks
Peter Hase (Allen Institute for AI), Sarah Wiegreffe (Allen Institute for AI)
ClassificationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: The study investigates the generalization ability of pre-trained language models after being fine-tuned with only easy data, and explores the impact of different hardness metrics and scales on the results.
Think Twice: Perspective-Taking Improves Large Language Models’ Theory-of-Mind Capabilities
Alex Wilf (Carnegie Mellon University), Louis-Philippe Morency
TransformerPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Proposes the SIMTOM two-stage prompting framework, which first performs perspective-taking and then answers Theory-of-Mind (ToM) questions to enhance the reasoning performance of large language models (LLMs).
Threads of Subtlety: Detecting Machine-Generated Texts Through Discourse Motifs
Zae Myung Kim (University of Minnesota Twin Cities), Dongyeop Kang (University of Minnesota Twin Cities)
ClassificationGraph Neural NetworkTransformerLarge Language ModelText
🎯 What it does: The paper converts text into a hierarchical RST structure, transforms it into a recursive hypergraph, and performs network motif analysis to explore differences between human and LLM-generated text in discourse structure, thereby enhancing authorship detection performance.
Through the Lens of Split Vote: Exploring Disagreement, Difficulty and Calibration in Legal Case Outcome Classification
Shanshan Xu (Technical University of Munich), Matthias Grabmair (Technical University of Munich)
ClassificationTransformerSupervised Fine-TuningText
🎯 What it does: This paper constructs an SV-ECHR case outcome classification dataset containing judges' split voting information by crawling decisions from the European Court of Human Rights (ECHR), and proposes a classification framework for identifying sources of judicial disagreement from legal, linguistic, and NLP perspectives. Subsequently, a Fine-tuned LegalBERT model is trained and evaluated on this dataset for case outcome classification (COC) tasks, including its difficulty perception for split cases, confidence alignment with human voting distributions, and human calibration.
Through the MUD: A Multi-Defendant Charge Prediction Benchmark with Linked Crime Elements
Xiao Wei (Shanghai University), Erik Cambria (Nanyang Technological University)
ClassificationExplainability and InterpretabilityTransformerContrastive LearningTextBenchmark
🎯 What it does: Proposed a benchmark for multi-defendant criminal cases called MUD, and designed the EJudge model to achieve explainable sentencing prediction based on criminal elements and legal rules.
Time is Encoded in the Weights of Finetuned Language Models
Kai Nylund (Paul G. Allen School of Computer Science & Engineering, University of Washington), Noah Smith (Paul G. Allen School of Computer Science & Engineering, University of Washington)
Representation LearningTransformerSupervised Fine-TuningText
🎯 What it does: Studied the impact of time on language model performance, introduced the concept of 'time vectors' derived by fine-tuning on a single time period and calculating the difference, and used these vectors for interpolation and analogy in the weight space to enhance the model's generalization ability for new time periods without additional training.
Time Sensitive Knowledge Editing through Efficient Finetuning
Xiou Ge (Apple), Yunyao Li (Adobe)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Investigated time-sensitive knowledge editing in large language models through parameter-efficient fine-tuning (PEFT).
TimeArena: Shaping Efficient Multitasking Language Agents in a Time-Aware Simulation
Yikai Zhang (Fudan University), Jiangjie Chen (Fudan University)
Computational EfficiencyTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Constructed TIMEARENA, a text-based simulation environment specifically designed to evaluate the execution efficiency of language models in multi-task scenarios involving time duration, action and resource consumption, as well as parallel processing constraints.
TimeBench: A Comprehensive Evaluation of Temporal Reasoning Abilities in Large Language Models
Zheng Chu (Harbin Institute of Technology), Bing Qin (Harbin Institute of Technology)
TransformerSupervised Fine-TuningReinforcement LearningPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Built TIMEBENCH, a hierarchical, multimodal temporal reasoning benchmark covering three categories (symbols, common sense, and events), comprising 10 datasets and 16 subtasks, to systematically evaluate the time reasoning capabilities of large language models.
Timeline-based Sentence Decomposition with In Context Learning for Temporal Fact Extraction
Jianhao Chen (Nanjing University), Yuzhong Qu (Nanjing University)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextSequential
🎯 What it does: This paper proposes a time-axis-based sentence decomposition method, combining it with the context learning of large language models (LLMs), and achieves high-quality extraction of temporal facts from complex sentences using a fine-tuned small pre-trained language model (PLM).
To be Continuous, or to be Discrete, Those are Bits of Questions
Yiran Wang (National Institute of Information and Communications Technology), Masao Utiyama (National Institute of Information and Communications Technology)
TransformerContrastive LearningText
🎯 What it does: In traditional deep learning models, inputs/outputs often use continuous vectors, while natural language is inherently discrete. This paper introduces binary representations (between continuous and discrete) into the model's output, proposing a structured contrastive hashing method, and directly represents syntactic trees or entity hierarchical structures with binary labels via bit-level CKY parsing.
To Distill or Not to Distill? On the Robustness of Robust Knowledge Distillation
Abdul Waheed (Mohamed bin Zayed University of Artificial Intelligence), Muhammad Abdul-Mageed (University of British Columbia)
RecognitionKnowledge DistillationTransformerAudio
🎯 What it does: This paper distills the Whisper large model into a smaller, dedicated Arabic ASR model, and evaluates it on standard benchmarks and newly collected low-resource dialect data from five dialects.
To Generate or to Retrieve? On the Effectiveness of Artificial Contexts for Medical Open-Domain Question Answering
Giacomo Frisoni (University of Bologna), Zaiqiao Meng (University of Glasgow)
TransformerLarge Language ModelPrompt EngineeringTextBiomedical DataBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose MEDGENIE, a fully generative open-domain medical question-answering framework that first generates synthetic context using a medical large language model and then reads and answers.
Token-wise Influential Training Data Retrieval for Large Language Models
Huawei Lin (Rochester Institute of Technology), Weijie Zhao (Rochester Institute of Technology)
Explainability and InterpretabilityComputational EfficiencyData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Designed and implemented RapidIn, a scalable framework for estimating the impact of training data in large language models;
ToMBench: Benchmarking Theory of Mind in Large Language Models
Zhuang Chen (Tsinghua University), Minlie Huang (Tsinghua University)
TransformerLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Proposed T MBENCH, a Theory of Mind (ToM) evaluation framework for large language models, comprising 8 tasks and 31 fine-grained capabilities, using multiple-choice question-and-answer formats and constructing an uncontaminated bilingual dataset from scratch.
ToolSword: Unveiling Safety Issues of Large Language Models in Tool Learning Across Three Stages
Junjie Ye, Xuanjing Huang (Fudan University)
Safty and PrivacyReinforcement Learning from Human FeedbackTransformerLarge Language ModelPrompt EngineeringTextMultimodality
🎯 What it does: Propose the ToolSword framework to conduct fine-grained security assessment of the tool learning process in LLMs, and construct six categories of security scenarios to test 11 LLMs across three stages: input, execution, and output.
Toward In-Context Teaching: Adapting Examples to Students’ Misconceptions
Alexis Ross (MIT), Jacob Andreas (MIT)
TransformerLarge Language ModelText
🎯 What it does: Proposed the ADAPT evaluation framework and the ATOM adaptive teaching method, studying how large language models (e.g., GPT-4) and probabilistic models can teach students to address misconceptions;
Towards Artwork Explanation in Large-scale Vision Language Models
Kazuki Hayashi (Nara Institute of Science and Technology), Taro Watanabe (Nara Institute of Science and Technology)
Explainability and InterpretabilityLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelMultimodalityBenchmark
🎯 What it does: Proposed the artwork interpretation generation task, constructed a图文 explanation dataset based on Wikipedia, and designed specialized evaluation metrics for knowledge utilization, including entity coverage, entity F1, and entity co-occurrence, to assess and instruction fine-tune multiple large vision-language models.
Towards Better Understanding of Contrastive Sentence Representation Learning: A Unified Paradigm for Gradient
Mingxin Li (Beihang University), Zhijie Nie (Beihang University)
OptimizationRepresentation LearningTransformerContrastive LearningText
🎯 What it does: This paper proposes a unified gradient paradigm by analyzing the gradient of contrastive learning loss, which includes three components: gradient dissipation, weight, and ratio. The framework is used to modify traditional non-contrastive losses, achieving results comparable to or even better than contrastive learning on sentence similarity (STS) tasks.
Towards Faithful and Robust LLM Specialists for Evidence-Based Question-Answering
Tobias Schimanski (University of Zürich), Markus Leippold (University of Regensburg)
RetrievalExplainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextRetrieval-Augmented Generation
🎯 What it does: Built an expandable synthetic data generation pipeline (SYNSCIQA) and improved data quality through quality filtering; refined the LLM to enhance citation accuracy and answer attributability in evidence-based QA; designed four test sets to evaluate model robustness in and out of distribution.
Towards Privacy-Aware Sign Language Translation at Scale
Phillip Rust (University of Copenhagen), Jean Maillard (FAIR at Meta)
Safty and PrivacyRepresentation LearningTransformerSupervised Fine-TuningAuto EncoderContrastive LearningVideoTextMultimodality
🎯 What it does: Proposed a two-stage, privacy-friendly sign language translation framework named SSVP-SLT, which first learns sign language representations using self-supervised video pre-training (Mask-AE) on large-scale unlabeled videos, and then completes English-sign language translation through supervised fine-tuning on a small amount of labeled parallel data.
Towards Real-world Scenario: Imbalanced New Intent Discovery
Shun Zhang (Beihang University), Zhoujun Li (Beihang University)
ClassificationTransformerContrastive LearningTextBenchmark
🎯 What it does: This paper proposes the i-NID task, addressing unknown intent discovery under long-tailed distributions by designing the ImbaNID framework and creating the ImbaNIDBench benchmark.
Towards Real-World Writing Assistance: A Chinese Character Checking Benchmark with Faked and Misspelled Characters
Yinghui Li (Tsinghua University), Ying Shen (Sun-Yat Sen University)
RecognitionRetrievalAnomaly DetectionTransformerTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Constructed the Visual-C3 benchmark dataset, containing real handwritten sentences with forged characters and misspellings, and providing sentence-level and character-level annotations; proposed and evaluated two baseline methods (OCR-based and CLIP-based) to accomplish detection and correction tasks.
Towards Robust and Generalized Parameter-Efficient Fine-Tuning for Noisy Label Learning
Yeachan Kim (Korea University), SangKeun Lee (Korea University)
ClassificationData-Centric LearningTransformerSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: The study investigates the robustness of parameter-efficient fine-tuning (PEFT) under noisy label environments and proposes the Clean Routing (CleaR) mechanism;
Tracking the Newsworthiness of Public Documents
Alexander Spangher (University of Southern California), Jonathan May (University of California, Los Angeles)
ClassificationRetrievalTransformerLarge Language ModelSupervised Fine-TuningTextAudio
🎯 What it does: The study establishes a framework that links local government policy documents with news reports and predicts which policies will be covered.
Training Language Models to Generate Text with Citations via Fine-grained Rewards
Chengyu Huang (National University of Singapore), Wenya Wang (Nanyang Technological University)
GenerationKnowledge DistillationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextRetrieval-Augmented Generation
🎯 What it does: Propose using fine-grained reward training to train language models to automatically insert high-quality, verifiable citations when generating answers;
Transferable and Efficient Non-Factual Content Detection via Probe Training with Offline Consistency Checking
Xiaokang Zhang (Renmin University of China), Jie Tang (Tsinghua University)
ClassificationAnomaly DetectionLarge Language ModelText
🎯 What it does: Built a non-factual content detection framework named PINOSE, which uses offline self-consistency checks to generate pseudo-labels for training the internal representation detector, eliminating the dependence on manual annotations.
Transferable Embedding Inversion Attack: Uncovering Privacy Risks in Text Embeddings without Model Queries
Yu-Hsiang Huang (National Taiwan University), Shou-De Lin (National Taiwan University)
Safty and PrivacyAdversarial AttackTransformerLarge Language ModelTextBiomedical Data
🎯 What it does: Developed a transferable text embedding inversion attack method without requiring model queries to assess privacy risks from embedding leakage.
Transitive Consistency Constrained Learning for Entity-to-Entity Stance Detection
Haoyang Wen (Carnegie Mellon University), Alexander Hauptmann (Carnegie Mellon University)
ClassificationGenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes a learning method based on transfer consistency constraints for stance detection between entity pairs. It first constructs triple relationships with shared entities through sentence pair sampling, and then introduces a soft consistency loss during training to maintain consistency between direct predictions and inferred stances.
Translation-based Lexicalization Generation and Lexical Gap Detection: Application to Kinship Terms
Senyu Li (University of Alberta), Grzegorz Kondrak (University of Alberta)
GenerationTransformerLarge Language ModelTextGraphBenchmark
🎯 What it does: Proposed an algorithm based on machine translation and hierarchical semantic relationships for automatically generating lexicalizations of concepts and detecting lexical gaps.
TransliCo: A Contrastive Learning Framework to Address the Script Barrier in Multilingual Pretrained Language Models
Yihong Liu (LMU Munich), Hinrich Schuetze
Representation LearningTransformerSupervised Fine-TuningContrastive LearningText
🎯 What it does: Proposed and implemented a fine-tuning framework called TRANSLICO based on transliteration contrastive learning to eliminate representation barriers between different writing systems in multilingual pre-trained models.
Transparent and Scrutable Recommendations Using Natural Language User Profiles
Jerome Ramos (University College London), Aldo Lipani (University College London)
Recommendation SystemTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Leverage instruction-tuned LLMs to generate natural language user profiles, then fine-tune the LLM based on these profiles for rating prediction, thereby achieving a transparent and auditable recommendation system.
Tree Transformer’s Disambiguation Ability of Prepositional Phrase Attachment and Garden Path Effects
Lingling Zhou (Leiden University), Gijs Wijnholds (Leiden University)
Recurrent Neural NetworkTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Investigate the sensitivity of Transformer models to PP attachment ambiguity and garden path effects, proposing and evaluating a set of naturalized PP attachment datasets
Tree-Averaging Algorithms for Ensemble-Based Unsupervised Discontinuous Constituency Parsing
Behzad Shayegh (University of Alberta), Lili Mou (University of Alberta)
Text
🎯 What it does: Proposes an ensemble method based on tree averaging for unsupervised discrete phrase parsing.
Tree-of-Counterfactual Prompting for Zero-Shot Stance Detection
Maxwell Weinzierl (University of Texas at Dallas), Sanda Harabagiu (University of Texas at Dallas)
ClassificationTransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningTextMultimodalityChain-of-Thought
🎯 What it does: Proposes the 'Tree-of-Counterfactual Prompting' method for zero-shot stance detection;
Tree-of-Traversals: A Zero-Shot Reasoning Algorithm for Augmenting Black-box Language Models with Knowledge Graphs
Elan Markowitz (Amazon AGI), Aram Galstyan (Amazon AGI)
TransformerLarge Language ModelTextGraphBenchmarkChain-of-Thought
🎯 What it does: This paper proposes a Tree-of-Traversals zero-shot reasoning algorithm, enabling black-box large language models (LLMs) to perform tree search reasoning through knowledge graph (KG) interfaces, thereby augmenting model knowledge and answering question-answering tasks.
Trial and Error: Exploration-Based Trajectory Optimization of LLM Agents
Yifan Song (Peking University), Bill Yuchen Lin (Allen Institute for AI)
OptimizationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningAgentic AIContrastive LearningTextSequentialBenchmark
🎯 What it does: This paper proposes an Exploratory Trajectory Optimization (ETO) method, which enhances the performance of open-ended large language model (LLM) agents by allowing LLMs to learn through trial-and-error in environments, collecting failure trajectories, and comparing them with expert trajectories.
Triple-Encoders: Representations That Fire Together, Wire Together
Justus-Jonas Erker (Technical University of Darmstadt), Iryna Gurevych (Technical University of Darmstadt)
Representation LearningTransformerContrastive LearningText
🎯 What it does: Proposed a Triple-Encoder method that utilizes self-organized Hebbian-style co-occurrence learning to achieve unweighted contextualized representations in dialogue sequences, combining it with Curved Contrastive Learning (CCL) to form Contextualized CCL (C3L).
TruthX: Alleviating Hallucinations by Editing Large Language Models in Truthful Space
Shaolei Zhang (Key Laboratory of Intelligent Information Processing, Institute of Computing Technology, Chinese Academy of Sciences), Yang Feng (Key Laboratory of Intelligent Information Processing, Institute of Computing Technology, Chinese Academy of Sciences)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelAuto EncoderContrastive LearningText
🎯 What it does: Propose the TruthX method, which edits the internal representations of LLMs during inference to enhance the truthfulness of their answers.
TTM-RE: Memory-Augmented Document-Level Relation Extraction
Chufan Gao (University of Illinois Urbana-Champaign), Jimeng Sun (University of Illinois Urbana-Champaign)
ClassificationTransformerSupervised Fine-TuningTextBiomedical Data
🎯 What it does: Proposed a document-level relation extraction model TTM-RE that integrates a built-in learnable memory module (Token Turing Machine) with a noise-robust loss.
Tuning Large Multimodal Models for Videos using Reinforcement Learning from AI Feedback
Daechul Ahn (Yonsei University), Jonghyun Choi (Seoul National University)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelVideoTextMultimodality
🎯 What it does: Proposed a reinforcement learning framework based on AI feedback, VLM-RLAIF, for aligning multimodal large models between video and text;
Two Issues with Chinese Spelling Correction and A Refinement Solution
Changxuan Sun (East China Normal University), Xuesong Lu (East China Normal University)
Representation LearningData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: This paper addresses two major issues in the Chinese Spelling Correction (CSC) task: first, the SIGHAN13/14/15 datasets contain a large number of errors, leading to inaccurate model evaluation; second, existing models have limited room for improvement on the SIGHAN test set. To address these issues, the authors manually corrected three SIGHAN datasets and retrained and evaluated four representative CSC models (PLOME, REALISE, LEAD, SCOPE) using the corrected versions. Based on this, the authors proposed a post-processing scheme based on ChineseBERT, which significantly improves the precision, recall, and F1 scores of all models in detection and correction by masking model outputs, using BERT inference, and combining pinyin edit distance thresholds to determine whether to accept the inferred results.
UHGEval: Benchmarking the Hallucination of Chinese Large Language Models via Unconstrained Generation
Xun Liang (Renmin University of China), Haiying Deng (State Key Laboratory of Media Convergence Production Technology and Systems)
GenerationTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Constructed an unconstrained generation Chinese hallucination evaluation benchmark UHG Eval, and provided a unified, scalable evaluation framework.
UltraLink: An Open-Source Knowledge-Enhanced Multilingual Supervised Fine-tuning Dataset
Haoyu Wang (BUPT), Maosong Sun (Tsinghua University)
Data SynthesisData-Centric LearningSupervised Fine-TuningTextBenchmark
🎯 What it does: Constructed an open-source multilingual supervised fine-tuning dataset called UltraLink, balancing language-specific knowledge and language-agnostic capabilities;
UltraSparseBERT: 99% Conditionally Sparse Language Modelling
Peter Belcak (NVIDIA), Roger Wattenhofer (ETH Zürich)
Computational EfficiencyTransformerLarge Language ModelTextBenchmark
🎯 What it does: Propose the UltraSparseBERT model, which uses only 0.3% of neurons (12/4095) during inference, and achieves conditional sparsity by modifying the Feedforward layer into a Fast Feedforward Network (FFN).
Uncertainty Aware Learning for Language Model Alignment
Yikun Wang (Fudan University), Dacheng Tao (Nanyang Technological University)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Propose an adaptive label smoothing method based on sample uncertainty for alignment training of large language models.
Uncertainty-Guided Modal Rebalance for Hateful Memes Detection
Chuanpeng Yang (Chinese Academy of Sciences), Songlin Hu (Chinese Academy of Sciences)
ClassificationTransformerVision Language ModelMultimodality
🎯 What it does: Propose a framework named Uncertainty-guided Modal Rebalance (UMR) to address modal uncertainty and modal imbalance in hate meme detection.
Uncovering the Full Potential of Visual Grounding Methods in VQA
Daniel Reich (University of Bremen), Tanja Schultz (University of Bremen)
Vision Language ModelImageTextMultimodality
🎯 What it does: The study identifies flaws in the training and testing practices of visual grounding (VG) methods in VQA evaluation and proposes a 'True Visual Grounding (TVG)' framework to address these issues.
Understanding and Addressing the Under-Translation Problem from the Perspective of Decoding Objective
Chenze Shao (Tencent Inc), Jie Zhou (Tencent Inc)
TransformerLarge Language ModelText
🎯 What it does: Analyze the root causes of under-translation in NMT and propose a method using EOS (End-of-Sentence) probability-based enhanced penalty to suppress under-translation.
Understanding Retrieval Robustness for Retrieval-augmented Image Captioning
Wenyan Li (University of Copenhagen), Desmond Elliott (University of Copenhagen)
GenerationRetrievalTransformerVision Language ModelContrastive LearningMultimodalityRetrieval-Augmented Generation
🎯 What it does: Analyze the robustness of the retrieval-augmented image captioning model SMALLCAP, finding it sensitive to noise in retrieved content, and enhance robustness and cross-domain performance by randomly sampling retrieval results during training.
Understanding the Effects of Noise in Text-to-SQL: An Examination of the BIRD-Bench Benchmark
Niklas Wretblad (Linköping University), Oskar Holmström (Linköping University)
TransformerLarge Language ModelPrompt EngineeringTextTabularBenchmarkFinance Related
🎯 What it does: Systematically analyzed the distribution and types of noise in the BIRD-Bench benchmark, and assessed the impact of noise on different Text-to-SQL models.
Uni-Dubbing: Zero-Shot Speech Synthesis from Visual Articulation
Songju Lei (Huawei Cloud), Zhou Zhao (Zhejiang University)
GenerationData SynthesisConvolutional Neural NetworkAuto EncoderContrastive LearningVideoMultimodalityAudio
🎯 What it does: Proposed a zero-shot cross-modal speech synthesis framework that maps silent videos to audio in the target language without requiring paired audio-visual data.
UniBridge: A Unified Approach to Cross-Lingual Transfer Learning for Low-Resource Languages
Trinh Pham (Ho Chi Minh City University Of Technology), Anh Tuan Luu (Nanyang Technological University)
Domain AdaptationRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Propose UniBridge, a unified cross-lingual transfer learning framework designed for low-resource languages, incorporating dynamic vocabulary search, semantic and lexical alignment-based embedding initialization, KL-regularized model adaptation, and multi-source transfer inference.
UniCoder: Scaling Code Large Language Model via Universal Code
Tao Sun (Beihang University), Zhoujun Li (Beihang University)
GenerationData SynthesisAI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Proposed a code generation framework called UNICODER, which uses UniCode as an intermediate representation, constructs a large-scale instruction dataset named UNICODER-INSTRUCT, and fine-tunes a general-purpose code LLM under a multi-task learning objective.
Unified Hallucination Detection for Multimodal Large Language Models
Xiang Chen (Zhejiang University), Huajun Chen (Zhejiang University)
Anomaly DetectionLarge Language ModelVision Language ModelMultimodalityBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed a unified multi-modal hallucination detection framework called UNIHD and constructed the MHaluBench unified evaluation benchmark for fine-grained detection of hallucinations in image-text and text-image tasks.
UNIMO-G: Unified Image Generation through Multimodal Conditional Diffusion
Wei Li (Baidu), Xinyan Xiao (Baidu)
GenerationTransformerLarge Language ModelPrompt EngineeringDiffusion modelImageMultimodality
🎯 What it does: Proposed a unified multimodal conditional diffusion framework named UNIMO-G, capable of receiving text and image interleaved prompts and generating high-quality images, with both text-driven and subject-driven generation capabilities.
Unintended Impacts of LLM Alignment on Global Representation
Michael J Ryan, Diyi Yang (Stanford University)
Representation LearningLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Analyze the impact of the LLM alignment process (SFT and preference tuning) on global English dialects, multilingual capabilities, and cross-national perspectives
Unity in Diversity: Collaborative Pre-training Across Multimodal Medical Sources
Xiaochen Wang (Pennsylvania State University), Fenglong Ma (Pennsylvania State University)
Representation LearningContrastive LearningMultimodalityElectronic Health Records
🎯 What it does: Proposed a cross-source multimodal medical pre-training framework named MEDCSP, which leverages patient information from different medical data sources for cross-source pre-training
Unlearning Traces the Influential Training Data of Language Models
Masaru Isonuma (University of Edinburgh), Ivan Titov (University of Edinburgh)
Explainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: Use gradient ascent to unlearn a pre-trained language model, directly measuring the change in model predictions on test or training sets after removing a specific training dataset, thereby evaluating the impact of that dataset.
Unlocking Data-free Low-bit Quantization with Matrix Decomposition for KV Cache Compression
Peiyu Liu (University of International Business and Economics), Ji-Rong Wen (Renmin University of China)
CompressionComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Propose DecoQuant, a data-free low-bit quantization method based on matrix decomposition, for compressing the KV cache of large language models and accelerating inference.
Unlocking the Power of Large Language Models for Entity Alignment
Xuhui Jiang (CAS Key Laboratory of AI Safety Institute of Computing Technology), Yuanzhuo Wang (CAS Key Laboratory of AI Safety Institute of Computing Technology)
Representation LearningGraph Neural NetworkTransformerLarge Language ModelPrompt EngineeringTextGraph
🎯 What it does: This paper proposes the ChatEA framework, combining large language models (LLM) with traditional knowledge representation learning (KRL) methods to enhance entity alignment (EA) performance by leveraging the background knowledge and reasoning capabilities of LLM.
Unsupervised Information Refinement Training of Large Language Models for Retrieval-Augmented Generation
Shicheng Xu (CAS Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences), Jie Zhou (WeChat AI)
GenerationRetrievalTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes INFO-RAG, a method that enhances the performance of large language models in retrieval-augmented generation (RAG) through information refinement under unsupervised conditions;
Unsupervised Multimodal Clustering for Semantics Discovery in Multimodal Utterances
Hanlei Zhang (Tsinghua University), Kai Gao (Hebei University of Science and Technology)
Representation LearningTransformerContrastive LearningMultimodality
🎯 What it does: Propose an unsupervised multi-modal clustering method UMC for multi-modal semantic discovery;
Unveiling Linguistic Regions in Large Language Models
Zhihao Zhang (Fudan University), Xuanjing Huang (Fudan University)
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: Through parameter importance analysis and experiments, this paper identifies core language regions (accounting for about 1% of parameters) and language-specific monolingual regions in large language models; removing the core region leads to complete loss of cross-lingual capabilities, while removing monolingual regions only affects the corresponding language; freezing the core region during further pre-training can mitigate catastrophic forgetting.
Using Natural Language Explanations to Improve Robustness of In-context Learning
Xuanli He (University College London), Pontus Stenetorp (University College London)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Using large language models to automatically generate natural language explanations (NLEs) to enhance robustness in in-context learning (ICL), and combining the generated NLEs with original examples as prompts to predict labels.
Using Synchronic Definitions and Semantic Relations to Classify Semantic Change Types
Pierluigi Cassotti (University of Gothenburg), Nina Tahmasebi (University of Gothenburg)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: By training a classification model based on semantic definitions, automatically detect types of semantic change (such as generalization, specialization, co-hyponymy shift, antonymy, etc.), and use this information to improve the performance of semantic change detection and WiC tasks.
ValueBench: Towards Comprehensively Evaluating Value Orientations and Understanding of Large Language Models
Yuanyi Ren (Peking University), Guojie Song (Peking University)
Large Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Propose ValueBench, a comprehensive benchmark integrating 44 psychological assessment tools and 453 value dimensions, designed to evaluate the value orientation and value understanding of large language models (LLMs);
VariErr NLI: Separating Annotation Error from Human Label Variation
Leon Weber-Genzel (LMU Munich), Barbara Plank (UCLouvain)
Explainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelTextBenchmark
🎯 What it does: Construct the VARIERR dataset by designing a two-round annotation process (first providing labels and reasons, then conducting self-evaluation and peer evaluation on label-reason pairs) to distinguish between annotation errors and human label variations.
VerifiNER: Verification-augmented NER via Knowledge-grounded Reasoning with Large Language Models
Seoyeon Kim (Yonsei University), Dongha Lee (Yonsei University)
RecognitionLarge Language ModelBiomedical DataRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes VERIFINER, a post-verification framework that leverages knowledge bases and large language models (LLMs) to fact-check and correct the factual accuracy and contextual relevance of generated named entity recognition (NER) results.
Video-ChatGPT: Towards Detailed Video Understanding via Large Vision and Language Models
Muhammad Maaz (Mohamed bin Zayed University of AI), Fahad Khan
TransformerLarge Language ModelVision Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: Propose Video-ChatGPT, a multimodal model that integrates a pre-trained visual encoder with a large language model for open-ended conversations with videos;
VIEScore: Towards Explainable Metrics for Conditional Image Synthesis Evaluation
Max Ku (University of Waterloo), Wenhu Chen (University of Waterloo)
GenerationExplainability and InterpretabilityTransformerLarge Language ModelVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Proposed a vision-instruction-based interpretable evaluation metric called VIESCORE, which leverages multimodal large language models (LLMs) to score various conditional image generation tasks and output natural language rationales along with numerical scores.
Virtual Compiler Is All You Need For Assembly Code Search
Zeyu Gao (Tsinghua University), Chao Zhang (Beijing University of Posts and Telecommunications)
Data SynthesisRetrievalTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: Train a large language model as a virtual compiler to generate assembly code and expand the assembly code retrieval dataset, thereby improving natural language retrieval performance in reverse engineering.
ViSAGe: A Global-Scale Analysis of Visual Stereotypes in Text-to-Image Generation
Akshita Jha (Virginia Tech), Sunipa Dev (Google Research)
GenerationPrompt EngineeringVision Language ModelDiffusion modelImageTextBenchmark
🎯 What it does: This paper constructs a global-scale visual stereotype assessment framework, systematically evaluating stereotypical features of 135 identity groups in Text-to-Image models (e.g., Stable Diffusion) and releases a new dataset, ViSAGe, annotated with 40,057 image-attribute pairs.