ACL 2023 Papers — Page 7
Annual Meeting of the Association for Computational Linguistics · 1074 papers
MatCha: Enhancing Visual Language Pretraining with Math Reasoning and Chart Derendering
Fangyu Liu (University of Cambridge), Julian Eisenschlos (Google DeepMind)
TransformerVision Language ModelImageTextMultimodality
🎯 What it does: Proposed the MATCHA pre-training method by adding two tasks on top of Pix2Struct: chart inverse rendering (recovering data tables or drawing code from images) and mathematical reasoning (training using visualized math QA data), enhancing vision-language models' performance in chart, drawing, and text interaction tasks.
Matching Pairs: Attributing Fine-Tuned Models to their Pre-Trained Large Language Models
Myles Foley (Imperial College London), Giulio Zizzo (IBM Research)
Safty and PrivacyTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringContrastive LearningText
🎯 What it does: Studied how to attribute fine-tuned LLMs to their pre-trained models to address copyright, model theft, and security risks.
MatSci-NLP: Evaluating Scientific Language Models on Materials Science Language Tasks Using Text-to-Schema Modeling
Yu Song (University of Montreal / Mila - Quebec AI), Bang Liu (University of Montreal / Mila - Quebec AI)
TransformerSupervised Fine-TuningPrompt EngineeringTextBenchmarkPhysics Related
🎯 What it does: Constructed the MatSci-NLP materials science NLP benchmark, covering seven public tasks, and proposed a unified text-to-schema multi-task learning framework.
mCLIP: Multilingual CLIP via Cross-lingual Transfer
Guanhua Chen (Southern University of Science and Technology), Wenping Wang (Texas Aamp;M University)
RetrievalKnowledge DistillationTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: This paper proposes mCLIP, a dual-stream multilingual vision-language pre-training model that aligns CLIP with a multilingual text encoder through triangular cross-modal knowledge distillation.
MDACE: MIMIC Documents Annotated with Code Evidence
Hua Cheng (3M Health Information Systems), Matthew Gormley
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkBiomedical DataElectronic Health RecordsBenchmark
🎯 What it does: Constructed and made public the MDACE dataset, which contains text evidence spans from diagnosis and surgical codes in MIMIC-III clinical records;
Measuring Consistency in Text-based Financial Forecasting Models
Linyi Yang (Westlake Institute for Advanced Study), Yue Zhang (Westlake University)
TransformerSupervised Fine-TuningTextTabularFinance Related
🎯 What it does: In financial text prediction tasks, the FinTrust evaluation tool is proposed to detect model logical consistency under input transformations that preserve meaning, combined with two evaluation settings: ① Consistency evaluation of masked word prediction in pre-trained language models (PLMs); ② Evaluation of prediction behavior after applying four types of logical consistency transformations (negation, symmetry, additivity, transitivity) to text-based financial prediction models.
Measuring Inductive Biases of In-Context Learning with Underspecified Demonstrations
Chenglei Si (University of Maryland), He He (New York University)
ClassificationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Investigated the preferences of large language models for different features during in-context learning (ICL), by constructing 'underdetermined' demonstration data to evaluate model feature bias, and attempted various intervention methods (instructions, label words, explanations, etc.) to guide models toward specific features.
Measuring Progress in Fine-grained Vision-and-Language Understanding
Emanuele Bugliarello (DeepMind), Aida Nematzadeh (DeepMind)
Vision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Evaluated and compared the performance of multiple Vision-Language Models (VLMs) on fine-grained vision-and-language understanding benchmarks, highlighting the role of explicit object modeling and specific loss functions in enhancing fine-grained capabilities.
Measuring the Effect of Influential Messages on Varying Personas
Chenkai Sun (University of Illinois Urbana-Champaign), Heng Ji (University of Illinois Urbana-Champaign)
GenerationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: The study proposes a personalized response prediction task for news media, predicting the emotional polarity, intensity, and textual responses that different individuals may generate after viewing news.
Measuring the Instability of Fine-Tuning
Yupei Du (Utrecht University), Dong Nguyen (Utrecht University)
Explainability and InterpretabilityTransformerSupervised Fine-TuningTextBenchmark
🎯 What it does: Investigated various metrics for measuring fine-tuning instability and proposed a framework for evaluating the effectiveness of these metrics.
MeetingBank: A Benchmark Dataset for Meeting Summarization
Yebowen Hu (University of Central Florida), Fei Liu (Emory University)
Data-Centric LearningTransformerSupervised Fine-TuningPrompt EngineeringVideoTextMultimodalityBenchmark
🎯 What it does: This paper constructs a city council meeting summary benchmark dataset named MeetingBank, generating approximately 6,892 segment-level summary instances through automatic transcription, paragraph segmentation, and alignment with meeting minutes, followed by evaluation of various extractive, abstractive, and GPT-3 prompt-based summarization models on this dataset.
MeetingQA: Extractive Question-Answering on Meeting Transcripts
Archiki Prasad (UNC Chapel Hill), Mohit Bansal (UNC Chapel Hill)
TransformerLarge Language ModelTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: This paper constructs an extractive question answering dataset called MEETINGQA based on meeting transcripts and conducts benchmark experiments on it.
MEMEX: Detecting Explanatory Evidence for Memes via Knowledge-Enriched Contextualization
Shivam Sharma (Indian Institute of Technology Delhi), Tanmoy Chakraborty (Indian Institute of Technology Delhi)
RetrievalRecurrent Neural NetworkGraph Neural NetworkTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: Proposes a new task called MEME X, aiming to retrieve evidence sentences that explain memes from text associated with memes.
Memory-efficient NLLB-200: Language-specific Expert Pruning of a Massively Multilingual Machine Translation Model
Yeskendir Koishekenov (NAVER LABS Europe), Vassilina Nikoulina (NAVER LABS Europe)
Computational EfficiencyTransformerMixture of ExpertsText
🎯 What it does: Conduct expert pruning research on the NLLB-200 large-scale multilingual Mixture-of-Experts NMT model, proposing multiple pruning metrics and strategies based on gating statistics to enable efficient inference on a single 32GB GPU.
MetaAdapt: Domain Adaptive Few-Shot Misinformation Detection via Meta Learning
Zhenrui Yue (University of Illinois Urbana Champaign), Dong Wang (University of Illinois Urbana Champaign)
ClassificationDomain AdaptationMeta LearningTransformerText
🎯 What it does: Propose a meta-learning based domain adaptive few-shot fake news detection framework, MetaAdapt, which achieves cross-domain adaptation using source domain tasks and a very small number of target domain samples.
Metaphor Detection via Explicit Basic Meanings Modelling
Yucheng Li (University of Surrey), Frank Guerin (University of Surrey)
ClassificationRepresentation LearningTransformerText
🎯 What it does: Proposed the BasicBERT model, which explicitly models the basic meaning of words by using literal annotations in the training set, and detects metaphors by comparing the basic meaning with contextual meaning;
MetaVL: Transferring In-Context Learning Ability From Language Models to Vision-Language Models
Masoud Monajatipoor (UCLA), Kai-Wei Chang (UCLA)
Computational EfficiencyKnowledge DistillationRepresentation LearningMeta LearningTransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningMultimodality
🎯 What it does: This paper proposes the MetaVL model, which transfers the few-shot context learning ability trained in language models to vision-language tasks, achieving strong few-shot performance with a smaller model.
MGR: Multi-generator Based Rationalization
Wei Liu (Huazhong University of Science and Technology), Yang Qiu (Huazhong University of Science and Technology)
Explainability and InterpretabilityRecurrent Neural NetworkMixture of ExpertsText
🎯 What it does: Propose the Multi-Generator (MGR) framework, which jointly trains multiple sets of generators with a single predictor to achieve self-explanatory NLP models;
miCSE: Mutual Information Contrastive Learning for Low-shot Sentence Embeddings
Tassilo Klein (SAP AI Research), Moin Nabi (SAP AI Research)
Representation LearningTransformerContrastive LearningText
🎯 What it does: Propose a new sentence embedding method called mi CSE, which leverages mutual information in attention layers to enforce structural consistency between different dropout views, and combines Momentum Contrastive Learning for self-supervised training.
MidMed: Towards Mixed-Type Dialogues for Medical Consultation
Xiaoming Shi (Shanghai Artificical Intelligence Laboratory), Shaoting Zhang (Shanghai Artificical Intelligence Laboratory)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringTextBiomedical DataBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose a mixed-type dialogue task for medical consultation and construct a bilingual mixed-type dialogue dataset called MidMed;
MIL-Decoding: Detoxifying Language Models at Token-Level via Multiple Instance Learning
Xu Zhang (Wangxuan Institute of Computer Technology, Peking University), Xiaojun Wan (Wangxuan Institute of Computer Technology, Peking University)
ClassificationSafty and PrivacyRecurrent Neural NetworkLarge Language ModelText
🎯 What it does: This study proposes the MIL-Decoding method, which assigns toxicity scores to each candidate word during language model decoding through a multi-instance learning (MIL) network, achieving token-level detoxification by combining the original language model probabilities.
Mind the Gap between the Application Track and the Real World
Ananya Ganesh (University of Colorado Boulder), Katharina Kann (University of Colorado Boulder)
Data-Centric LearningTransformerSupervised Fine-TuningTextReview/Survey PaperAudio
🎯 What it does: Investigated the gap between motivation and experimental setup in NLP application papers, and validated the severity of this gap through a survey of ACL 2020/EMNLP 2020 application track papers and a case study on classroom dialogue understanding systems.
Minding Language Models’ (Lack of) Theory of Mind: A Plug-and-Play Multi-Character Belief Tracker
Melanie Sclar (University of Washington), Yulia Tsvetkov
Explainability and InterpretabilityRepresentation LearningGraph Neural NetworkTransformerLarge Language ModelPrompt EngineeringTextGraphRetrieval-Augmented Generation
🎯 What it does: Propose a training-free, symbolic method called SYMBOLICTOM that explicitly tracks multi-agent belief states using graph structures and leverages existing large language models for reading comprehension reasoning.
MIR-GAN: Refining Frame-Level Modality-Invariant Representations with Adversarial Network for Audio-Visual Speech Recognition
Yuchen Hu (Nanyang Technological University), Eng Siong Chng (Nanyang Technological University)
RecognitionTransformerGenerative Adversarial NetworkContrastive LearningMultimodality
🎯 What it does: Propose an MIR-GAN framework that leverages adversarial networks and mutual information maximization to refine frame-level cross-modal invariant representations, thereby enhancing audio-visual speech recognition performance.
MIReAD: Simple Method for Learning High-quality Representations from Scientific Documents
Anastasiia Razdaibiedina (University of Toronto), Aleksandr Brechalov (University of Toronto)
ClassificationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningTextBiomedical Data
🎯 What it does: Fine-tune SciBERT using a journal classification task, learning high-quality scientific document representations based solely on paper titles and abstracts.
MISGENDERED: Limits of Large Language Models in Understanding Pronouns
Tamanna Hossain (University of California, Irvine), Sameer Singh (University of California, Irvine)
TransformerPrompt EngineeringTextBenchmark
🎯 What it does: This paper constructs the MISGENDERED framework to systematically evaluate gendered errors in large language models when using non-binary pronouns (including singular they, neo-pronouns, etc.).
Mitigating Label Biases for In-context Learning
Yu Fei (UC Irvine), Antoine Bosselut (EPFL)
ClassificationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper investigates the label bias that arises in in-context learning (ICL) and proposes a novel bias calibration method called Domain-Context Calibration (DC), which estimates label bias using random in-domain vocabulary and corrects it during inference.
MixCE: Training Autoregressive Language Models by Mixing Forward and Reverse Cross-Entropies
Shiyue Zhang (Bloomberg), David Rosenberg (UNC Chapel Hill)
GenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Studied a new training objective MIXCE, combining forward and backward cross-entropy to improve the generation quality of autoregressive language models, avoiding overgeneralization and low-quality text.
Mixture-of-Domain-Adapters: Decoupling and Injecting Domain Knowledge to Pre-trained Language Models’ Memories
Shizhe Diao (Hong Kong University of Science and Technology), Tong Zhang (Hong Kong University of Science and Technology)
Domain AdaptationTransformerSupervised Fine-TuningMixture of ExpertsText
🎯 What it does: This paper proposes the MixDA method, which separates the original pre-trained FFN in the Transformer's FFN structure from newly added domain adapters. By employing a two-stage training process (first learning domain knowledge on unlabeled domain data, then learning task knowledge on labeled task data), it dynamically fuses outputs from different domain adapters through a Mixture-of-Adapters gate, achieving efficient domain adaptation across multiple tasks.
MM-SHAP: A Performance-agnostic Metric for Measuring Multimodal Contributions in Vision and Language Models & Tasks
Letitia Parcalabescu (Heidelberg University), Anette Frank (Heidelberg University)
Explainability and InterpretabilityTransformerSupervised Fine-TuningVision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Propose MM-SHAP, a performance-independent metric based on Shapley values, to quantitatively evaluate the contribution of each modality in vision-and-language models across different tasks, datasets, and instances;
MMDialog: A Large-scale Multi-turn Dialogue Dataset Towards Multi-modal Open-domain Conversation
Jiazhan Feng (Peking University), Qingwei Lin (Microsoft Corporation)
GenerationData SynthesisRetrievalTransformerVision Language ModelDiffusion modelContrastive LearningImageTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Construct a large-scale multimodal open-domain dialogue dataset called MMDialog, define multimodal generation and retrieval tasks based on this dataset, propose the MM-Relevance evaluation metric, and provide baseline models.
Modality Adaption or Regularization? A Case Study on End-to-End Speech Translation
Yuchen Han (Northeastern University), Jingbo Zhu (Northeastern University)
Domain AdaptationRepresentation LearningData-Centric LearningTransformerSupervised Fine-TuningTextMultimodalityAudio
🎯 What it does: Through a case study on the E2E ST pre-training and fine-tuning framework, this paper investigates the impact of modal adaptation and regularization on performance, and proposes an auxiliary branch method with adjustable modal adaptation (TAB).
Model-Based Simulation for Optimising Smart Reply
Benjamin Towle (University of Nottingham), Ke Zhou (University of Nottingham)
GenerationOptimizationTransformerWorld ModelText
🎯 What it does: Propose a Smart Reply method (SIMSR) based on Model-Based Simulation, directly optimizing the diversity and relevance of the reply set;
Model-Generated Pretraining Signals Improves Zero-Shot Generalization of Text-to-Text Transformers
Linyuan Gong (University of California Berkeley), Xia Song (Microsoft)
Representation LearningMeta LearningTransformerSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: Proposed the METRO-T0 model, which improves the pre-training of T5-style text-to-text Transformers by utilizing pre-training signals generated by the model (RTD+CLM), thereby enhancing zero-shot generalization capabilities.
Modeling Appropriate Language in Argumentation
Timon Ziegenbein (Leibniz University Hannover), Henning Wachsmuth (Leibniz University Hannover)
ClassificationTransformerLarge Language ModelText
🎯 What it does: This study systematically models appropriate language in online arguments, proposing a 14-dimensional hierarchical classification and manually annotating 2191 arguments;
Modeling Instance Interactions for Joint Information Extraction with Neural High-Order Conditional Random Field
Zixia Jia (Beijing Institute for General Artificial Intelligence), Kewei Tu (ShanghaiTech University)
TransformerText
🎯 What it does: Propose a high-order conditional random field framework CRFIE for joint entity, relation, and event extraction, explicitly modeling second-order and third-order dependencies between nodes and edges;
Modeling Structural Similarities between Documents for Coherence Assessment with Graph Convolutional Networks
Wei Liu (Heidelberg Institute for Theoretical Studies), Michael Strube (Heidelberg Institute for Theoretical Studies)
ClassificationGraph Neural NetworkTransformerTextGraphBenchmark
🎯 What it does: Construct sentence graphs and subgraph sets, build a document-subgraph heterogeneous graph using shared subgraphs, then encode the graph with GCN for text coherence evaluation and automatic writing scoring.
Modeling User Satisfaction Dynamics in Dialogue via Hawkes Process
Fanghua Ye (University College London), Emine Yilmaz (University College London)
TransformerText
🎯 What it does: This paper proposes a new user satisfaction estimator called ASAP, which models the dynamic changes in satisfaction over time in dialogues using a Hawkes process, thereby achieving more accurate automatic evaluation;
Modeling What-to-ask and How-to-ask for Answer-unaware Conversational Question Generation
Xuan Long Do (Institute for Infocomm Research), Ai Ti Aw (Institute for Infocomm Research)
GenerationTransformerLarge Language ModelText
🎯 What it does: Proposed a two-stage answer-agnostic dialogic question generation framework named SG-CQG, which can generate coherent and diverse dialogic questions and answers without knowing the answers.
Modular Visual Question Answering via Code Generation
Sanjay Subramanian (University of California Berkeley), Dan Klein (University of California Berkeley)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Generating Python programs using large language models (e.g., code-davinci-002) to decompose visual tasks into calling pre-trained visual modules (query, get_pos, find_matching_image) and combining them through code logic to accomplish visual question answering;
mOKB6: A Multilingual Open Knowledge Base Completion Benchmark
Shubham Mittal (Indian Institute of Technology Delhi), Mausam (Indian Institute of Technology Delhi)
Representation LearningTransformerContrastive LearningTextGraphBenchmark
🎯 What it does: Constructed the first multilingual open knowledge base benchmark mOKB6, which includes six languages (English, Hindi, Telugu, Spanish, Portuguese, and Chinese), and improved the coreference and open IE pipeline;
MolXPT: Wrapping Molecules with Text for Generative Pre-training
Zequn Liu (Peking University), Tie-Yan Liu (Microsoft Research AI4Science)
Drug DiscoveryTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBiomedical Data
🎯 What it does: Designed and pre-trained MolXPT, a unified generative pre-trained Transformer capable of processing scientific text, SMILES, and their wrapped sequences simultaneously.
MoralDial: A Framework to Train and Evaluate Moral Dialogue Systems via Moral Discussions
Hao Sun (Tsinghua University), Minlie Huang (Xiaomi AI Lab)
GenerationExplainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: Propose the MORALDIAL framework, which constructs moral discussion data to train and evaluate the morality of open-domain dialogue systems; build four subtasks (moral answer, explanation, revision, reasoning), and adopt multi-task learning to train on DialoGPT/Blenderbot; design a no-reference evaluation method based on answer-RoT consistency scoring.
More than Classification: A Unified Framework for Event Temporal Relation Extraction
Quzhe Huang (Peking University), Dongyan Zhao (Peking University)
TransformerText
🎯 What it does: Convert event time relations into logical expressions of start and end time points, and propose a unified framework to predict event time relations
Morphological Inflection with Phonological Features
David Guriel (Barilan University), Reut Tsarfaty (Barilan University)
Recurrent Neural NetworkText
🎯 What it does: In the morphological inflection task, the authors explored methods to introduce phoneme-level phonological features into the model and conducted experiments on eight shallow orthography languages.
Morphological Inflection: A Reality Check
Jordan Kodner (Stony Brook University), Zoey Liu (University of Florida)
Data-Centric LearningText
🎯 What it does: Systematically evaluate assessment methods for morphological inflection tasks, propose a data sampling and evaluation strategy based on frequency weighting and overlap awareness, and test the generalization ability of existing models across multiple languages.
MOSPC: MOS Prediction Based on Pairwise Comparison
Kexin Wang (ByteDance), Mingxuan Wang (ByteDance)
Audio
🎯 What it does: Propose a MOS prediction framework MOSPC based on pairwise comparisons, and enhance its generalization performance through C-Mixup;
Movie101: A New Movie Understanding Benchmark
Zihao Yue (Renmin University of China), Qin Jin (Renmin University of China)
TransformerVision Language ModelContrastive LearningVideoTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Constructed a large-scale Chinese movie understanding benchmark named Movie101, and proposed two tasks: Movie Clip Narration (MCN) and Temporal Narration Localization (TNG), along with a new evaluation metric, MNScore, which shows high correlation with human assessments.
MPCHAT: Towards Multimodal Persona-Grounded Conversation
Jaewoo Ahn (Seoul National University), Gunhee Kim (Seoul National University)
RetrievalTransformerVision Language ModelContrastive LearningMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Constructed the first multimodal personalized dialogue dataset MPCHAT, and designed three retrieval-based multimodal personalized dialogue tasks (next response prediction, context-based persona prediction, speaker identification) for evaluation.
mPMR: A Multilingual Pre-trained Machine Reader at Scale
Weiwen Xu (Chinese University of Hong Kong), Lidong Bing (DAMO Academy, Alibaba Group)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Developed a multilingual machine reading comprehension pre-training framework called mPMR, which inherits and enhances the natural language understanding capabilities of existing multilingual pre-training models (e.g., XLM-R) by generating MRC-style training data on Wikipedia links.
MS-DETR: Natural Language Video Localization with Sampling Moment-Moment Interaction
Wang Jing, Xiaoli Li (Nanyang Technological University)
RetrievalTransformerVision Language ModelVideoText
🎯 What it does: Proposed an MS-DETR based on the proposal-based DETR framework for natural language video localization, combining a multi-scale vision-language encoder with an anchor-guided temporal decoder to achieve efficient temporal interaction.
Multi-CLS BERT: An Efficient Alternative to Traditional Ensembling
Haw-Shiuan Chang (Amazon), Andrew McCallum (University of Massachusetts)
Computational EfficiencyRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Achieve model ensemble without additional computational cost by introducing multiple CLS tokens into BERT;
Multi-Document Summarization with Centroid-Based Pretraining
Ratish Surendran Puduppully (Institute for Infocomm Research, A STAR), Mark Steedman (University of Edinburgh)
GenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes a pre-training objective based on cluster center documents (Centrum) for multi-document summarization tasks.
Multi-Grained Knowledge Retrieval for End-to-End Task-Oriented Dialog
Fanqi Wan (Sun Yat-sen University), Wei Bi (Tencent AI Lab)
RetrievalKnowledge DistillationTransformerSupervised Fine-TuningTextRetrieval-Augmented Generation
🎯 What it does: Propose Multi-grAined KnowlEdge Retriever (MAKER), which decouples knowledge retrieval from generation, achieving multi-grained retrieval through entity and attribute selectors, and trains the retriever via knowledge distillation from the generator.
Multi-granularity Temporal Question Answering over Knowledge Graphs
Ziyang Chen (National University of Defense Technology), Xiang Zhao (National University of Defense Technology)
Representation LearningTransformerLarge Language ModelGraph
🎯 What it does: This paper introduces the concept of Multi-granularity Temporal Knowledge Graph Question Answering (Multi-granularity Temporal Question Answering), constructs a large-scale MULTITQ dataset based on ICEWS05-15, and designs a Transformer-based baseline model called MultiQA to handle multi-granularity temporal questions.
Multi-Level Knowledge Distillation for Out-of-Distribution Detection in Text
Qianhui Wu (Microsoft), Chin-Yew Lin (Microsoft)
Anomaly DetectionKnowledge DistillationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Adopt multi-layer knowledge distillation to transfer knowledge from the fine-tuned teacher model to the student model trained from scratch, aiming to enhance unsupervised text OoD detection performance.
Multi-modal Action Chain Abductive Reasoning
Mengze Li (Zhejiang University), Fei Wu (Hikvision Research Institute)
RecognitionExplainability and InterpretabilityConvolutional Neural NetworkGraph Neural NetworkTransformerVision-Language-Action ModelVideoTextMultimodalityChain-of-Thought
🎯 What it does: Studied the multimodal action chain abductive reasoning task (MAR), and proposed a complete framework for target event localization and subsequent action chain inference.
Multi-Row, Multi-Span Distant Supervision For Table+Text Question Answering
Vishwajeet Kumar (IBM Research India), Feifei Pan
RetrievalTransformerTextTabular
🎯 What it does: Proposed the MITQA system, which first retrieves table rows and then locates answers, specifically designed for multi-row and multi-span weakly supervised table+text QA;
Multi-source Semantic Graph-based Multimodal Sarcasm Explanation Generation
Liqiang Jing (Shandong University), Liqiang Nie (Shandong University)
GenerationGraph Neural NetworkTransformerLarge Language ModelVision Language ModelMultimodalityRetrieval-Augmented Generation
🎯 What it does: Proposed the TEAM model, aiming to automatically generate explanatory sentences for multimodal sarcastic social posts (image + caption);
Multi-Source Test-Time Adaptation as Dueling Bandits for Extractive Question Answering
Hai Ye (National University of Singapore), Hwee Tou Ng (National University of Singapore)
Domain AdaptationTransformerReinforcement LearningContrastive LearningText
🎯 What it does: This paper studies online adaptation of extractive QA models using user feedback in multi-source test-time adaptation (TTA), performing model selection and updates through multi-armed bandit and dueling bandit frameworks.
Multi-target Backdoor Attacks for Code Pre-trained Models
Yanzhou Li (Nanyang Technological University), Yang Liu (Nanyang Technological University)
Adversarial AttackAI Code AssistantTransformerLarge Language ModelText
🎯 What it does: A multi-objective backdoor attack on code pre-training models, where the backdoor is implanted during the pre-training phase and can be triggered in downstream code understanding and code generation tasks.
Multi-VALUE: A Framework for Cross-Dialectal English NLP
Caleb Ziems, Diyi Yang
GenerationData SynthesisDomain AdaptationExplainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Constructed the Multi-VALUE framework, which converts standard American English (SAE) into synthetic text across 50 English dialects using controllable rules, and leveraged this text for stress testing and data augmentation in tasks such as question answering, machine translation, and semantic parsing, further releasing a gold standard CoQA dataset for Chicano and Indian English.
MultiCapCLIP: Auto-Encoding Prompts for Zero-Shot Multilingual Visual Captioning
Bang Yang (Peking University), Yuexian Zou (University of Oxford)
GenerationTransformerPrompt EngineeringVision Language ModelAuto EncoderImageText
🎯 What it does: Propose a model called MultiCapCLIP that is trained solely with text and can perform zero-shot multilingual visual description
MultiEMO: An Attention-Based Correlation-Aware Multimodal Fusion Framework for Emotion Recognition in Conversations
Tao Shi (Tsinghua University), Shao-Lun Huang (Tsinghua University)
RecognitionConvolutional Neural NetworkRecurrent Neural NetworkTransformerContrastive LearningImageTextMultimodalityAudio
🎯 What it does: Propose the MultiEMO framework to achieve multimodal emotion recognition based on text, audio, and visual modalities;
MultiInstruct: Improving Multi-Modal Zero-Shot Learning via Instruction Tuning
Zhiyang Xu (Virginia Tech), Lifu Huang (Virginia Tech)
Representation LearningTransformerSupervised Fine-TuningPrompt EngineeringVision Language ModelMultimodalityBenchmark
🎯 What it does: This paper proposes the MULTIINSTRUCT dataset and performs multi-modal instruction tuning on the OFA model, significantly improving performance on zero-shot tasks.
Multilingual Conceptual Coverage in Text-to-Image Models
Michael Saxon (University of California Santa Barbara), William Yang Wang (University of California Santa Barbara)
GenerationDiffusion modelImageTextBenchmark
🎯 What it does: Developed the CoCo-CroLa benchmark to evaluate the generation coverage of text-to-image models for tangible concepts in multilingual environments.
Multilingual Event Extraction from Historical Newspaper Adverts
Nadav Borenstein (University of Copenhagen), Isabelle Augenstein (University of Copenhagen)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper constructs a multilingual historical newspaper advertisement dataset containing English, French, and Dutch, and transforms the event attribute extraction task into an extractive QA (question answering) problem.
Multilingual Knowledge Graph Completion with Language-Sensitive Multi-Graph Attention
Rongchuan Tang (University of Chinese Academy of Sciences), Yu Zhou (University of Chinese Academy of Sciences)
Representation LearningGraph Neural NetworkGraph
🎯 What it does: Propose a unified multilingual knowledge graph completion framework that concatenates multiple language KGs into a unified graph using shared entity embeddings, and achieves cross-graph knowledge transfer through language-sensitive multi-graph attention and aggregation modules.
Multilingual LLMs are Better Cross-lingual In-context Learners with Alignment
Eshaan Tanwar (Delhi Technological University), Tanmoy Chakraborty (Indian Institute of Technology Delhi)
ClassificationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Propose a cross-lingual context learning prompt design method called X-InSTA, which leverages semantic similarity examples and manually crafted task-aligned statements to enhance the performance of multilingual LLMs in cross-lingual text classification.
Multilingual Multifaceted Understanding of Online News in Terms of Genre, Framing, and Persuasion Techniques
Jakub Piskorski (Polish Academy of Science), Preslav Nakov (Mohamed bin Zayed University of Artificial Intelligence)
ClassificationTransformerLarge Language ModelTextBenchmark
🎯 What it does: Constructed and made publicly available a multilingual, multidimensional news dataset covering six European languages, 1,612 articles, annotated with news genres, frameworks, and 23 fine-grained persuasion techniques, and conducted baseline experiments for multi-label classification on this dataset.
Multimodal Persona Based Generation of Comic Dialogs
Harsh Agrawal (IIT Delhi), Mausam (IIT Delhi)
GenerationTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a manga dialogue generation framework based on multimodal (text + image) and character features, and constructs the COMSET dataset containing 13 comics, 54K comic panels, and 200+ characters with their personality descriptions.
Multimodal Relation Extraction with Cross-Modal Retrieval and Synthesis
Xuming Hu (Tsinghua University), Philip S. Yu (University of Illinois at Chicago)
Data SynthesisRetrievalRepresentation LearningConvolutional Neural NetworkTransformerVision Language ModelMultimodalityRetrieval-Augmented Generation
🎯 What it does: Propose a cross-modal retrieval and synthesis framework that retrieves textual and visual evidence for objects, sentences, and entire graphs, and performs relation extraction by fusing multi-modal information through multi-head attention.
MultiTabQA: Generating Tabular Answers for Multi-Table Question Answering
Vaishali Pal (University of Amsterdam), Maarten de Rijke (University of Amsterdam)
TransformerLarge Language ModelSupervised Fine-TuningTabular
🎯 What it does: This paper proposes a multi-table question-answering model called MultiTabQA, which can directly generate structured table answers from natural language questions or SQL queries and multiple tables;
MultiTACRED: A Multilingual Version of the TAC Relation Extraction Dataset
Leonhard Hennig (German Research Center for Artificial Intelligence), Sebastian Möller (German Research Center for Artificial Intelligence)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This study translates the large-scale English relation extraction dataset TACRED into 12 languages via machine translation and automatic annotation projection, constructing a multilingual version called MultiTACRED;
Multitask Pre-training of Modular Prompt for Chinese Few-Shot Learning
Tianxiang Sun (Fudan University), Xuanjing Huang (Fudan University)
ClassificationMeta LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This paper proposes a multi-task pre-trained modular prompt (MP2), which is pre-trained on 38 Chinese NLP tasks using deep+wide soft prompts and achieves fast few-shot adaptation on downstream tasks.
Multitask Pretraining with Structured Knowledge for Text-to-SQL Generation
Robert Giaquinto (AWS AI Labs), Xiaofei Ma (AWS AI Labs)
GenerationRepresentation LearningTransformerLarge Language ModelTextTabular
🎯 What it does: Propose a multi-task pre-training framework called STAMP, specifically designed for the text-to-SQL generation task, incorporating joint learning of tables, SQL code, and natural language within an encoder-decoder structure.
MultiTool-CoT: GPT-3 Can Use Multiple External Tools with Chain of Thought Prompting
Tatsuro Inaba (Kyoto University), Sadao Kurohashi (Kyoto University)
Drug DiscoveryTransformerLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Proposed the MultiTool-CoT framework, which utilizes chain-of-thought prompting to enable GPT-3 to interactively invoke multiple external tools (calculator, chemical reaction predictor, molar mass list) during reasoning, and completes inference through iterative interaction between tool triggers and tool results.
Multiview Identifiers Enhanced Generative Retrieval
Yongqi Li (Hong Kong Polytechnic University), Wenjie Li (Hong Kong Polytechnic University)
RetrievalTransformerLarge Language ModelText
🎯 What it does: Propose the MINDER framework, which leverages autoregressive language models to generate multi-view identifiers (title, substring, pseudo queries) and achieve generative retrieval;
MUSTIE: Multimodal Structural Transformer for Web Information Extraction
Qifan Wang (Meta AI), Dongfang Liu (Rochester Institute of Technology)
TransformerVision Language ModelMultimodality
🎯 What it does: Proposed a multi-modal structural Transformer (MUST) that jointly encodes text, images, and OCR text on the HTML DOM tree, leveraging structural attention to capture relationships between different nodes for web information extraction.
MVP-Tuning: Multi-View Knowledge Retrieval with Prompt Tuning for Commonsense Reasoning
Yongfeng Huang (Chinese University of Hong Kong), Liwei Wang (International Digital Economy Academy)
ClassificationRetrievalTransformerPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Proposed a Commonsense QA model named MVP-Tuning that integrates multi-perspective knowledge retrieval with prompt tuning.
MvP: Multi-view Prompting Improves Aspect Sentiment Tuple Prediction
Zhibin Gou (Tsinghua University), Yujiu Yang (Tsinghua University)
ClassificationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: For the multi-aspect sentiment tuple prediction task, the MVP method is proposed, which utilizes prompts with different element orders to enable generative language models to generate sentiment tuples from multiple perspectives, and then aggregates the final results through voting.
My side, your side and the evidence: Discovering aligned actor groups and the narratives they weave
Pavan Holur (University of California Los Angeles), Vwani Roychowdhury (University of California Los Angeles)
Representation LearningGraph Neural NetworkTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes a two-step framework based on large models (INCANT + TAMPA) that automatically identifies aligned actors and their composed aligned actor groups in news texts, thereby revealing the core role network hidden within multi-narrative structures.
Naamapadam: A Large-Scale Named Entity Annotated Data for Indic Languages
Arnav Mhaske (Indian Institute of Technology Madras), Anoop Kunchukuttan (Indian Institute of Technology Madras)
RecognitionTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Constructed a large-scale multilingual named entity recognition dataset named Naamapadam, covering 11 major Indo-Aryan languages, and generated manually annotated test sets; simultaneously trained and released a multilingual BERT-based NER model named IndicNER.
NarrowBERT: Accelerating Masked Language Model Pretraining and Inference
Haoxin Li (University of Washington), Noah A. Smith (University of Washington)
Computational EfficiencyTransformerText
🎯 What it does: Proposed two variants of NarrowBERT, leveraging query sparsification and hierarchical rearrangement to perform computations only on masked tokens, significantly improving pretraining and inference speeds.
NatLogAttack: A Framework for Attacking Natural Language Inference Models with Natural Logic
Zi’ou Zheng, Xiaodan Zhu (Queen's University)
Adversarial AttackTransformerLarge Language ModelText
🎯 What it does: Proposes NatLogAttack, an adversarial attack framework based on natural logic, for generating label-preserving and label-flipping attack samples in natural language inference (NLI) tasks.
Natural Language to Code Generation in Interactive Data Science Notebooks
Pengcheng Yin (Google Inc.), Charles Sutton (Google Inc.)
AI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextTabularBenchmark
🎯 What it does: This paper proposes the ARCADE benchmark for evaluating models that generate pandas code based on natural language in interactive data science notebooks, and trains and evaluates the large-scale code language model PACHINCO on this benchmark, exploring the impact of prompting strategies on code quality and diversity.
Neural Machine Translation for Mathematical Formulae
Felix Petersen (Stanford University), Bela Gipp (University of Göttingen)
GenerationConvolutional Neural NetworkTransformerText
🎯 What it does: Translate ambiguous LaTeX expressions into unambiguous Mathematica or semantic LaTeX expressions using a convolutional sequence-to-sequence network.
Neural Machine Translation Methods for Translating Text to Sign Language Glosses
Dele Zhu (Technical University of Berlin), Eleftherios Avramidis (German Research Center for Artificial Intelligence)
TransformerVideoText
🎯 What it does: Studies how to translate spoken text into sign language glosses and improve the quality of this translation.
Neural Unsupervised Reconstruction of Protolanguage Word Forms
Andre He (University of California, Berkeley), Dan Klein (University of California, Berkeley)
RestorationRecurrent Neural NetworkText
🎯 What it does: This paper proposes an unsupervised neural network-based method for reconstructing the morphology of original language words.
NEUROSTRUCTURAL DECODING: Neural Text Generation with Structural Constraints
Mohaddeseh Bastan (Stony Brook University), Niranjan Balasubramanian (Stony Brook University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed NEUROSTRUCTURAL DECODING, an algorithm that integrates syntactic constraints into beam search decoding, and implemented an adaptive parser to handle incomplete sentences.
Nichelle and Nancy: The Influence of Demographic Attributes and Tokenization Length on First Name Biases
Haozhe An (University of Maryland), Rachel Rudinger (University of Maryland)
Explainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: This study systematically evaluates the impact of name race/ethnicity, gender, and token length on model bias through replacement experiments with first names in social commonsense reasoning tasks, and conducts fine-grained analysis using SODAPOP framework-generated massive replacement samples;
NLG Evaluation Metrics Beyond Correlation Analysis: An Empirical Metric Preference Checklist
Iftitahu Nimah, Mykola Pechenizkiy (Eindhoven University of Technology)
GenerationTextBenchmark
🎯 What it does: Designed and evaluated a multi-level assessment framework called Metric Preference Checklist to measure the effectiveness of automatic evaluation metrics in three types of natural language generation tasks (summarization, dialogue, and controlled generation).
NLP Reproducibility For All: Understanding Experiences of Beginners
Shane Storks (University of Michigan), Joyce Chai (University of Michigan)
TextReview/Survey Paper
🎯 What it does: In an NLP course, 93 students attempted to reproduce three recent ACL papers, and their background, learning process, time consumption, difficulty perception, and feedback on reproducibility practices were collected.
NLPeer: A Unified Resource for the Computational Study of Peer Review
Nils Dycke (Technical University of Darmstadt), Iryna Gurevych (Technical University of Darmstadt)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Constructed NLPEER, a multi-domain, multi-temporal, open-licensed peer review corpus covering 5k papers and 11k review reports.
NLPositionality: Characterizing Design Biases of Datasets and Models
Sebastin Santy (University of Washington), Maarten Sap (Carnegie Mellon University)
Explainability and InterpretabilityData-Centric LearningTransformerText
🎯 What it does: Propose the NLPositionality framework, which quantifies design bias and bias localization by collecting diverse annotations on LabintheWild and calculating correlations with model/dataset labels;
No clues, good clues: Out of context Lexical Relation Classification
Lucía Pitarch (University of Zaragoza), Jorge Gracia (University of Zaragoza)
ClassificationExplainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Investigated the effectiveness of using minimal or null prompts for fine-tuning pre-trained language models on lexical relation classification (LRC) and graded lexical entailment (LE) tasks, verifying their performance in the absence of human-provided context.
Node Placement in Argument Maps: Modeling Unidirectional Relations in High & Low-Resource Scenarios
Iman Jundi (University of Stuttgart), Gabriella Lapesa (University of Stuttgart)
RetrievalRepresentation LearningTransformerSupervised Fine-TuningContrastive LearningText
🎯 What it does: This paper proposes the 'node placement' task, which helps find suitable parent nodes for new arguments in argument trees of online discussions;
NollySenti: Leveraging Transfer Learning and Machine Translation for Nigerian Movie Sentiment Classification
Iyanuoluwa Shode (Montclair State University), Anna Feldman (Montclair State University)
ClassificationDomain AdaptationTransformerLarge Language ModelText
🎯 What it does: Constructed the NollySenti movie review sentiment dataset, conducting cross-domain and cross-lingual sentiment classification experiments using transfer learning and machine translation.
Non-Sequential Graph Script Induction via Multimedia Grounding
Yu Zhou (University of California Los Angeles), Heng Ji (University of Illinois Urbana Champaign)
GenerationTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: By aligning wikiHow linear scripts with video multimodal information, automatically generate non-sequential graph scripts that capture relationships between optional and interchangeable steps.
Nonlinear Structural Equation Model Guided Gaussian Mixture Hierarchical Topic Modeling
HeGang Chen, Yanghui Rao (Sun Yat-sen University)
Representation LearningAuto EncoderText
🎯 What it does: Propose a deep hierarchical topic model called NSEM-GMHTM, which explicitly models the hierarchical and symmetric dependencies of topics through a nonlinear structural equation model (NSEM) and Gaussian Mixture Prior (GMM), thereby enhancing topic coherence and the rationality of hierarchical structures.