EMNLP 2023 Papers — Page 7
Conference on Empirical Methods in Natural Language Processing · 1047 papers
LM vs LM: Detecting Factual Errors via Cross Examination
Roi Cohen (Tel Aviv University), Amir Globerson (Tel Aviv University)
Anomaly DetectionTransformerPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Proposes a zero-shot factuality detection method by letting two language models cross-question each other, identifying inconsistencies in their generated claims to determine factual errors.
Localizing Active Objects from Egocentric Vision with Symbolic World Knowledge
Te-Lin Wu (University of California Los Angeles), Nanyun Peng (University of California Los Angeles)
Object DetectionObject TrackingTransformerLarge Language ModelVision-Language-Action ModelContrastive LearningVideoText
🎯 What it does: Proposes an active object localization and tracking framework based on text instructions and symbolic knowledge, capable of identifying and tracking key objects (OUC) and tools (Tool) with state changes in the perspective camera view
Location-Aware Visual Question Generation with Lightweight Models
Nicholas Suwono, Shao-Hua Sun (National Taiwan University)
GenerationData SynthesisComputational EfficiencyKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a new task—Location-Aware Visual Question Generation (LocaVQG), which generates engaging, highly location-related questions using vehicle GPS coordinates and four-way street view images; meanwhile, it constructs an automatic data generation pipeline based on GPT-4 and trains a lightweight model FDT5, enabling real-time inference on mobile devices.
Log-FGAER: Logic-Guided Fine-Grained Address Entity Recognition from Multi-Turn Spoken Dialogue
Xue Han (China Mobile Research Institute), Junlan Feng (China Mobile Research Institute)
RecognitionData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: This paper studies the recognition of fine-grained address entities in multi-turn speech dialogues, and proposes a logic-based PSL regularization method called Log-FGAER, as well as an ontology-driven data augmentation framework leveraging ChatGPT to generate low-cost annotated dialogue data.
Longtriever: a Pre-trained Long Text Encoder for Dense Document Retrieval
Junhan Yang (University of Science and Technology of China), Xing Xie (Microsoft Research Asia)
RetrievalTransformerSupervised Fine-TuningText
🎯 What it does: Propose the Longtriever model for long-text retrieval;
Look-back Decoding for Open-Ended Text Generation
Nan Xu (University of Southern California), Xuezhe Ma (University of Southern California)
GenerationTransformerLarge Language ModelText
🎯 What it does: Proposed the Look-back decoding algorithm, which utilizes KL divergence to track distribution differences between the current and historical generation steps, automatically suppressing repetition and topic drift.
Lost in Translation, Found in Spans: Identifying Claims in Multilingual Social Media
Shubham Mittal (Mohammed Bin Zayed University Of Artificial Intelligence), Preslav Nakov (Mohammed Bin Zayed University Of Artificial Intelligence)
RecognitionTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper constructs a multilingual claim span identification (mCSI) dataset named X-CLAIM, proposes an automated annotation pipeline, and conducts experiments on six languages (English, Hindi, Punjabi, Tamil, Telugu, and Bengali).
M^3Seg: A Maximum-Minimum Mutual Information Paradigm for Unsupervised Topic Segmentation in ASR Transcripts
Ke Wang (Huawei IT Innovation and Research Center), Wei Peng (Huawei IT Innovation and Research Center)
SegmentationTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Propose an M Seg unsupervised topic segmentation framework that first learns paragraph representations by maximizing the mutual information between sentences and paragraphs, then detects topic boundaries by minimizing mutual information between different paragraphs, achieving linear topic segmentation for ASR corpus.
M2DF: Multi-grained Multi-curriculum Denoising Framework for Multimodal Aspect-based Sentiment Analysis
Fei Zhao (Nanjing University), Xinyu Dai (Nanjing University)
ClassificationTransformerVision Language ModelContrastive LearningMultimodality
🎯 What it does: Proposed a multi-granularity multi-course learning denoising framework (M2DF) for multi-modal aspect-based sentiment analysis tasks, which can reduce the negative impact of noisy images on model learning without relying on threshold filtering.
MADNet: Maximizing Addressee Deduction Expectation for Multi-Party Conversation Generation
Jia-Chen Gu (University of Science and Technology of China), Cong Liu (University of Science and Technology of China)
GenerationGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Propose MADNet for multi-party dialogue generation to address the problem of missing recipient labels;
MAF: Multi-Aspect Feedback for Improving Reasoning in Large Language Models
Deepak Nathani (University of California Santa Barbara), William Wang (University of California Santa Barbara)
TransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: This paper proposes a multi-dimensional feedback (MAF) framework, which iteratively improves the reasoning chains generated by large language models by decomposing the feedback module into dedicated submodules for different types of errors.
mAggretriever: A Simple yet Effective Approach to Zero-Shot Multilingual Dense Retrieval
Sheng-Chieh Lin (University Of Waterloo), Jimmy Lin (Vectara)
RetrievalTransformerContrastive LearningText
🎯 What it does: Proposes mAggretriever, a multilingual dense retrieval model that combines semantic and lexical features, and significantly reduces GPU memory usage through two lightweight MLM approximation methods (target language prediction and self-prediction), achieving excellent zero-shot transfer on multilingual retrieval tasks with training only in English.
MAGNIFICo: Evaluating the In-Context Learning Ability of Large Language Models to Generalize to Novel Interpretations
Arkil Patel (Mila and McGill University), Dzmitry Bahdanau (Mila and McGill University)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes the MAGNIFICO evaluation framework to systematically assess the ability of large language models to acquire and generalize new word meanings in context learning;
MailEx: Email Event and Argument Extraction
Saurabh Srivastava (George Mason University), Ziyu Yao (George Mason University)
RecognitionTransformerLarge Language ModelTextBenchmarkFinance Related
🎯 What it does: Propose the MAILEX dataset and the email event extraction task involving 10 event categories and 76 roles
Make Every Example Count: On the Stability and Utility of Self-Influence for Learning from Noisy NLP Datasets
Irina Bejan (Google DeepMind), Katja Filippova (Google DeepMind)
OptimizationData-Centric LearningTransformerReinforcement LearningText
🎯 What it does: Research and implementation of a training sample cleaning and dynamic curriculum learning method based on self-influence scores to remove or reweight noisy samples and improve the performance of multi-task NLP models.
Making Large Language Models Better Data Creators
Dong-Ho Lee (University of Southern California), Sujay Jauhar
Data SynthesisDomain AdaptationData-Centric LearningTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposed a unified single-sample formatted example generation pipeline that leverages instruction-following large language models (LLMs) to generate diverse, structured annotated data under given JSON structure prompts, and continuously expands the training set through a self-reference strategy.
MaNtLE: Model-agnostic Natural Language Explainer
Rakesh Menon, Shashank Srivastava (University Of North Carolina Chapel Hill)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelTextTabularBenchmark
🎯 What it does: This paper proposes MaNtLE, a model-agnostic natural language interpreter that generates natural language explanations describing the reasoning process of classifiers from a small number of classifier predictions.
MarkQA: A large scale KBQA dataset with numerical reasoning
Xiang Huang (Nanjing University), Yuzhong Qu (Nanjing University)
TransformerPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Proposes the NR-KBQA task, which requires simultaneously performing multi-hop reasoning and numerical reasoning;
MAUD: An Expert-Annotated Legal NLP Dataset for Merger Agreement Understanding
Steven Wang, Dan Hendrycks (University of California Berkeley)
TransformerLarge Language ModelSupervised Fine-TuningTextBenchmarkFinance Related
🎯 What it does: Constructed and released MAUD, an expert-annotated reading comprehension dataset based on merger agreements, and trained a Transformer baseline model on it.
MeaeQ: Mount Model Extraction Attacks with Efficient Queries
Chengwei Dai (Chinese Academy of Sciences), Wei Zhou (Chinese Academy of Sciences)
Adversarial AttackTransformerText
🎯 What it does: This paper studies NLP model extraction attacks, proposing the MeaeQ method, which achieves efficient querying through task-related filtering and clustering dimensionality reduction, thereby constructing a pirated model with performance comparable to the victim model.
MedEval: A Multi-Level, Multi-Task, and Multi-Domain Medical Benchmark for Language Model Evaluation
Zexue He (University of California San Diego), Chun-Nan Hsu (University of California San Diego)
ClassificationRecognitionTransformerLarge Language ModelSupervised Fine-TuningTextBiomedical DataElectronic Health RecordsBenchmark
🎯 What it does: Proposes MEDEVAL, a multi-level, multi-task, multi-domain medical NLP benchmark for evaluating language models on medical texts.
MediaHG: Rethinking Eye-catchy Features in Social Media Headline Generation
Boning Zhang (Zhejiang University), Yang Yang (Zhejiang University)
GenerationTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Propose the MediaHG model on the vertical interest social platform REDBook to generate attractive titles that can accurately extract content and meet the platform's 'eye-catching' requirements.
MEGA: Multilingual Evaluation of Generative AI
Kabir Ahuja (University of Washington), Sunayana Sitaram (Microsoft Corporation)
ClassificationRecognitionTransformerPrompt EngineeringTextBenchmark
🎯 What it does: Evaluate the performance of generative large models on multilingual NLP tasks, construct the MEGA benchmark, compare GPT-3.5, GPT-4, BLOOMZ, and state-of-the-art fine-tuned models, and systematically explore the impact of different prompting strategies on low-resource languages.
MemeCap: A Dataset for Captioning and Interpreting Memes
EunJeong Hwang (University of British Columbia), Vered Shwartz (University of British Columbia)
GenerationTransformerSupervised Fine-TuningVision Language ModelMultimodalityBenchmark
🎯 What it does: Create the MEMECAP dataset for evaluating and training models that can generate meme captions and explain visual metaphors.
Memorisation Cartography: Mapping out the Memorisation-Generalisation Continuum in Neural Machine Translation
Verna Dankers (University of Edinburgh), Dieuwke Hupkes (Meta AI)
GenerationExplainability and InterpretabilityData-Centric LearningTransformerText
🎯 What it does: This paper constructs a 'memory-generalization' map containing 5M NMT data points, quantifies the memorization level of each sample using the counterfactual memorisation (CM) metric, and investigates the relationship between this level and data features, model training signals, and final performance.
Memory-Based Invariance Learning for Out-of-Domain Text Classification
Chen Jia (SI-TECH Information Technology Fudan University), Yue Zhang (Westlake University)
ClassificationDomain AdaptationMeta LearningTransformerText
🎯 What it does: Propose a domain-invariant representation learning method based on memory networks, leveraging meta-learning in multi-source domain training to learn memory gains and enhance text classification performance on unknown target domains.
Merging Experts into One: Improving Computational Efficiency of Mixture of Experts
Shwai He (University of Maryland), Dacheng Tao (University of Sydney)
Computational EfficiencyTransformerMixture of ExpertsText
🎯 What it does: Propose an MEO (Merging Experts into One) method that merges multiple expert parameters into a single expert, preserving the representational diversity of multiple experts in the Mixture of Experts model while significantly reducing computational costs.
Merging Generated and Retrieved Knowledge for Open-Domain QA
Yunxiang Zhang (University of Michigan), Lu Wang (University of Illinois at Chicago)
GenerationRetrievalRepresentation LearningTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes the COMBO framework, which improves the accuracy of open-domain question answering by pairing retrieved text with LLM-generated text through compatibility matching.
Meta-Learning Online Adaptation of Language Models
Nathan Hu (Stanford University), Chelsea Finn (Stanford University)
Meta LearningTransformerSupervised Fine-TuningText
🎯 What it does: Propose a meta-learning method called CaMeLS for unsupervised online adaptation of language models, which learns importance weights for each token and dynamically weights the loss during online fine-tuning to enhance the model's ability to memorize factual information in new documents and improve subsequent question-answering performance.
MILDSum: A Novel Benchmark Dataset for Multilingual Summarization of Indian Legal Case Judgments
Debtanu Datta (Indian Institute of Technology Kharagpur), Saptarshi Ghosh (Indian Institute of Technology Kharagpur)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Constructed the MILDSum dataset, containing 3,122 English legal judgments along with their English and Hindi summaries, and conducted benchmark evaluations of multiple summarization and translation methods
MingOfficial: A Ming Official Career Dataset and a Historical Context-Aware Representation Learning Framework
You-Jun Chen (National Central University), Richard Tsai
RecognitionRepresentation LearningGraph Neural NetworkTransformerLarge Language ModelTextMultimodalityGraph
🎯 What it does: Constructed a multimodal dataset containing career records of officials from the Ming Dynasty and related texts, and utilized graph neural networks to learn official embeddings, identifying subtle characteristics of officials.
Mirages. On Anthropomorphism in Dialogue Systems
Gavin Abercrombie (Heriot Watt University), Zeerak Talat (Mohamed Bin Zayed University Of Artificial Intelligence)
Review/Survey Paper
🎯 What it does: This paper reviews the linguistic factors that lead to anthropomorphism in dialogue systems, discusses the associated risks, and proposes normative recommendations to mitigate anthropomorphism.
Mirror: A Universal Framework for Various Information Extraction Tasks
Tong Zhu (Soochow University), Min Zhang (Soochow University)
ClassificationRecognitionTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed the Mirror framework, unifying various information extraction tasks into multi-slot tuples and mapping them to multi-span cyclic graphs, employing non-autoregressive graph decoding to achieve one-time extraction of all spans.
Mitigating Backdoor Poisoning Attacks through the Lens of Spurious Correlation
Xuanli He (University College London), Trevor Cohn (University of Melbourne)
Anomaly DetectionAdversarial AttackTransformerSupervised Fine-TuningText
🎯 What it does: This paper proposes treating backdoor triggers as pseudo-relevant, and filters out training instances containing triggers by calculating z-scores of lexical and syntactic features.
Mitigating Over-Generation for Unsupervised Keyphrase Extraction with Heterogeneous Centrality Detection
Mingyang Song (Beijing Jiaotong University), Liping Jing (Beijing Jiaotong University)
Representation LearningGraph Neural NetworkLarge Language ModelTextBenchmark
🎯 What it does: This paper proposes an unsupervised keyword extraction method called CentralityRank, which evaluates the importance of candidate phrases by leveraging implicit and explicit centrality in heterogeneous graphs.
Mitigating Temporal Misalignment by Discarding Outdated Facts
Michael Zhang, Eunsol Choi (University of Texas at Austin)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper introduces the 'fact duration prediction' task, which uses models to predict how long a fact will remain correct in the future, thereby calibrating the confidence of question-answering systems under temporal misalignment to avoid providing outdated answers.
MMNMT: Modularizing Multilingual Neural Machine Translation with Flexibly Assembled MoE and Dense Blocks
Shangjie Li (Tianjin University), Deyi Xiong (Tianjin University)
TransformerMixture of ExpertsText
🎯 What it does: Propose the MMNMT framework, flexibly assembling dense and Mixture-of-Experts (MoE) modules to achieve efficient capacity and low-resource robustness for multilingual translation;
Model-tuning Via Prompts Makes NLP Models Adversarially Robust
Mrigank Raman (Carnegie Mellon University), Danish Pruthi (Indian Institute of Science)
ClassificationAdversarial AttackTransformerPrompt EngineeringText
🎯 What it does: Proposes using prompt tuning (MVP) as an alternative to traditional MLP head fine-tuning methods to achieve higher adversarial robustness in downstream classification tasks
Modeling Conceptual Attribute Likeness and Domain Inconsistency for Metaphor Detection
Yuan Tian (Institute of Automation, Chinese Academy of Sciences), Daniel Zeng
ClassificationGraph Neural NetworkTransformerLarge Language ModelContrastive LearningText
🎯 What it does: This paper proposes AIDIL, a metaphor detection framework for word pairs based on Conceptual Metaphor Theory, which can simultaneously model the similar attributes and domain inconsistency between source and target words;
Modeling Empathic Similarity in Personal Narratives
Jocelyn Shen (Massachusetts Institute of Technology), Cynthia Breazeal (Massachusetts Institute of Technology)
RetrievalRecommendation SystemTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Construct and evaluate the 'empathy similarity' task, develop the EMPATHICSTORIES dataset, and train a retrieval model, demonstrating its superiority over traditional semantic similarity models in helping users find emotionally resonant stories.
Modeling Legal Reasoning: LM Annotation at the Edge of Human Agreement
Rosamond Thalken (Cornell University), Matthew Wilkens (Cornell University)
ClassificationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextChain-of-Thought
🎯 What it does: Construct and utilize an expert-annotated dataset of legal reasoning types (formal vs. macroscopic) to evaluate the performance of large language models on this challenging classification task.
Models See Hallucinations: Evaluating the Factuality in Video Captioning
Hui Liu (Peking University), Xiaojun Wan (Peking University)
GenerationData SynthesisTransformerSupervised Fine-TuningVision Language ModelContrastive LearningVideoTextBenchmark
🎯 What it does: This paper constructs two video caption truthfulness annotation datasets to systematically evaluate factual errors in video captions and proposes a fact consistency evaluation metric called FactVC based on CLIP.
MolCA: Molecular Graph-Language Modeling with Cross-Modal Projector and Uni-Modal Adapter
Zhiyuan Liu (National University of Singapore), Tat-Seng Chua (National University of Singapore)
Drug DiscoveryGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningTextGraph
🎯 What it does: This study proposes MolCA, which utilizes a cross-modal projector (Q-Former) to map molecular 2D graphs into the text space, enabling a large language model (Galactica) to understand and generate text based on molecular graphs.
MoPe: Model Perturbation based Privacy Attacks on Language Models
Marvin Li (Harvard), Seth Neel (Harvard)
Safty and PrivacyTransformerLarge Language ModelImageText
🎯 What it does: Proposed and validated a membership inference attack called MoPe based on model parameter perturbation, which can detect the membership of training data in language models by approximating the Hessian trace.
More Than Spoken Words: Nonverbal Message Extraction and Generation
Dian Yu (Tencent AI Lab), Dong Yu (Tencent AI Lab)
ClassificationGenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This study proposes and implements the task of extracting non-linguistic information (NM) from Chinese novel texts, as well as automatically generating NM for sentences in dialogue generation.
MoT: Memory-of-Thought Enables ChatGPT to Self-Improve
Xiaonan Li (Fudan University), Xipeng Qiu (Fudan University)
RetrievalExplainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the MoT (Memory-of-Thought) framework, enabling large language models to enhance their reasoning capabilities by using pre-thought reasoning paths as external memory without labeled data or parameter updates, and retrieving relevant memories during testing.
MProto: Multi-Prototype Network with Denoised Optimal Transport for Distantly Supervised Named Entity Recognition
Shuhui Wu (Zhejiang University), Weiming Lu (Zhejiang University)
ClassificationRecognitionText
🎯 What it does: Proposed a multi-prototype network called MProto to address the noise problem in remote supervision named entity recognition;
MQuAKE: Assessing Knowledge Editing in Language Models via Multi-Hop Questions
Zexuan Zhong (Princeton University), Danqi Chen (Princeton University)
TransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed a multi-hop question evaluation benchmark, MQUAKE, to assess the performance of knowledge editing methods in multi-hop reasoning, and introduced MeLLo, an editing method based on external memory.
mRedditSum: A Multimodal Abstractive Summarization Dataset of Reddit Threads with Images
Keighley Overbay (Seoul National University), Gunhee Kim (Seoul National University)
TransformerLarge Language ModelVision Language ModelMultimodalityBenchmark
🎯 What it does: This paper constructs the MREDDITSUM dataset, which contains 3033 Reddit discussion threads with images and text, and provides human-generated summaries that cover image information.
MT2: Towards a Multi-Task Machine Translation Model with Translation-Specific In-Context Learning
Chunyou Li (Beijing Jiaotong University), Ming Zhou (Beijing Lanzhou Technology Co Ltd)
Domain AdaptationTransformerPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Propose the MT2 multi-task machine translation model, which leverages translation-specific context learning to unify tasks such as translation memory, terminology constraints, and document-level translation.
Multi-level Adaptive Contrastive Learning for Knowledge Internalization in Dialogue Generation
Chenxu Yang (Chinese Academy of Sciences), Weiping Wang (Chinese Academy of Sciences)
GenerationRepresentation LearningTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Proposes a multi-level adaptive contrastive learning (MACL) framework to address the knowledge echo (knowledge repetition) problem in knowledge dialogue generation, thereby improving knowledge internalization.
Multi-level Contrastive Learning for Script-based Character Understanding
Dawei Li (University of California, San Diego), Shiping Yang (Independent Researcher)
Representation LearningTransformerLarge Language ModelContrastive LearningText
🎯 What it does: To address role understanding in scripts, this paper proposes a multi-layer contrastive learning framework that leverages multi-perspective information from abstracts and dialogues to learn fine-grained and global role representations.
Multi-Source Multi-Type Knowledge Exploration and Exploitation for Dialogue Generation
Xuanfan Ni (Nanjing University of Aeronautics and Astronautics), Piji Li (Shandong University)
GenerationTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Designed and implemented a framework called KnowEE, which enhances the quality of open-domain multi-turn dialogue generation through multi-source, multi-type knowledge exploration and injection.
Multi-Source Probing for Open-Domain Conversational Understanding
Yuanxi Li (University of Illinois at Urbana-Champaign), Minlie Huang (Tsinghua University)
ClassificationTransformerPrompt EngineeringText
🎯 What it does: Propose the Multi-Source Probing (MSP) method to probe the understanding ability of open-domain dialogue models through a generative approach.
Multi-Task Knowledge Distillation with Embedding Constraints for Scholarly Keyphrase Boundary Classification
Seo Park, Cornelia Caragea (University of Illinois Chicago)
ClassificationKnowledge DistillationRepresentation LearningTransformerText
🎯 What it does: Propose a model combining multi-task knowledge distillation with cosine embedding constraints for key phrase boundary detection and classification in academic papers.
Multi-view Contrastive Learning for Entity Typing over Knowledge Graphs
Zhiwei Hu (Shanxi University), Jeff Z. Pan (University of Edinburgh)
ClassificationRepresentation LearningGraph Neural NetworkMixture of ExpertsContrastive LearningGraph
🎯 What it does: Proposed a multi-perspective contrastive learning-based knowledge graph entity type prediction model called MCLET, which extracts information from three perspectives: entity-type, entity-cluster, and cluster-type;
Multilingual estimation of political-party positioning: From label aggregation to long-input Transformers
Dmitry Nikolaev (University of Stuttgart), Sebastian Padó (University of Stuttgart)
ClassificationTransformerLarge Language ModelText
🎯 What it does: The paper positions multi-lingual party manifestos on the left-right spectrum using two methods (sentence-level label aggregation and direct prediction by long-text Transformers).
Multilingual Holistic Bias: Extending Descriptors and Patterns to Unveil Demographic Biases in Languages at Scale
Marta R. Costa-jussà (Meta), Carleigh Wood (Meta)
Explainability and InterpretabilityData-Centric LearningTransformerTextBenchmark
🎯 What it does: Expand the original HolisticBias dataset into a multi-lingual version, generating a dataset with approximately 20,459 sentences across 50 languages and 13 demographic axes, and use this dataset to evaluate bias in machine translation and sentence embedding models related to gender and other demographic features.
Multilingual k-Nearest-Neighbor Machine Translation
David Stap (University of Amsterdam), Christof Monz (University of Amsterdam)
RetrievalData-Centric LearningTransformerTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes to improve the translation quality of low-resource languages by using multilingual k-nearest neighbor retrieval (k-NN-MT), achieved through the construction of a cross-lingual and multilingual retrieval datastore.
Multilingual Large Language Models Are Not (Yet) Code-Switchers
Ruochen Zhang (Brown University), Alham Fikri Aji (MBZUAI)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This paper systematically evaluates the performance of multilingual large language models on code-switching tasks, comparing results from zero-shot, few-shot prompting, and fine-tuning.
Multilingual Pixel Representations for Translation and Effective Cross-lingual Transfer
Elizabeth Salesky (Johns Hopkins University), Matt Post (Human Language Technology Center of Excellence)
Representation LearningTransformerVision Language ModelImageText
🎯 What it does: This paper proposes and verifies a method of using pixel-level visual representations (pixel representations) instead of traditional subword embeddings in multilingual machine translation, demonstrating its effectiveness during multilingual training.
Multilingual Previously Fact-Checked Claim Retrieval
Matúš Pikuliak (Kempelen Institute of Intelligent Technologies), Maria Bielikova (Kempelen Institute of Intelligent Technologies)
RetrievalTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Proposed a cross-lingual retrieval system for previously fact-checked claims, constructed and released the multilingual dataset MultiClaim
Multilingual Simplification of Medical Texts
Sebastian Joseph (University of Texas at Austin), Junyi Jessy Li (University of Texas at Austin)
GenerationTransformerLarge Language ModelSupervised Fine-TuningTextBiomedical DataBenchmark
🎯 What it does: Constructed a multilingual medical text simplification dataset called MULTICOCHRANE, and evaluated multiple simplification models across four languages (English, Spanish, French, Persian) using both automatic and human evaluations.
Multimodal Embodied Plan Prediction Augmented with Synthetic Embodied Dialogue
Aishwarya Padmakumar (Amazon), Dilek Hakkani-Tur (University of Illinois)
Data SynthesisData-Centric LearningRobotic IntelligenceTransformerAgentic AIVision-Language-Action ModelMultimodality
🎯 What it does: This paper proposes a multimodal plan prediction model tailored for the TEACh dataset and designs an agenda-based synthetic dialogue generation framework to train plan prediction models in the absence of real human dialogue data.
Multitask Multimodal Prompted Training for Interactive Embodied Task Completion
Georgios Pantazopoulos (Heriot-Watt University), Alessandro Suglia (Heriot-Watt University)
Robotic IntelligenceTransformerPrompt EngineeringVision Language ModelImageVideoTextMultimodalityBenchmark
🎯 What it does: Propose EMMA, a unified multi-task, multi-modal encoder-decoder framework that integrates vision, language, and action execution into a text generation task for interactive embodied task completion.
MULTITuDE: Large-Scale Multilingual Machine-Generated Text Detection Benchmark
Dominik Macko (Kempelen Institute of Intelligent Technologies), Maria Bielikova (Kempelen Institute of Intelligent Technologies)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Proposed the MULTITuDE dataset for systematic evaluation of multilingual machine-generated text detection methods
MultiTurnCleanup: A Benchmark for Multi-Turn Spoken Conversational Transcript Cleanup
Hua Shen (University of Michigan), Dirk Padfield (Google Research)
Data-Centric LearningTransformerLarge Language ModelTextBenchmarkAudio
🎯 What it does: Studied the Multi-Turn Cleanup task, constructing and annotating the MultiTurnCleanup dataset based on the Switchboard corpus.
NAIL: Lexical Retrieval Indices with Efficient Non-Autoregressive Decoders
Livio Baldini Soares (Google Deepmind), Tom Kwiatkowski (Google Deepmind)
RetrievalComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: Propose NAIL, a sparse retrieval model that precomputes document vocabulary weights using a non-autoregressive decoder, requiring only simple vocabulary lookup during queries.
NameGuess: Column Name Expansion for Tabular Data
Jiani Zhang (Amazon Web Services), George Karypis (Amazon Web Services)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTabularBenchmark
🎯 What it does: This paper proposes the NAMEGUESS task, which involves expanding abbreviated column names in tables into complete, readable logical column names, and constructs corresponding training and evaluation datasets;
Natural Disaster Tweets Classification Using Multimodal Data
Mohammad Basit (Jamia Millia Islamia), Salman Shaikh (King Abdullah University of Science and Technology)
ClassificationConvolutional Neural NetworkRecurrent Neural NetworkTransformerMultimodality
🎯 What it does: Designed and implemented a hierarchical multi-modal classification system that utilizes image and text information to perform multi-level classification of disaster-related tweets, ranging from informativeness to disaster type, structural damage severity, and humanitarian tasks.
Natural Language Decompositions of Implicit Content Enable Better Text Representations
Alexander Hoyle (University of Maryland), Philip Resnik (University of Maryland)
Explainability and InterpretabilityRepresentation LearningLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes the 'inferential decompositions' method, which generates a set of propositions related to the implied reasoning of the original text using large language models, and uses these propositions as explicit representations of the text to improve tasks such as text similarity measurement, topic discovery, and prediction of legislative voting behavior.
Navigating the Grey Area: How Expressions of Uncertainty and Overconfidence Affect Language Models
Kaitlyn Zhou (Stanford University), Tatsunori Hashimoto (Stanford University)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper constructs a linguistic marker framework expressing uncertainty and overconfidence, injects these markers into prompts for question-answering tasks, and systematically evaluates their impact on the performance of large language models.
Nearest Neighbor Machine Translation is Meta-Optimizer on Output Projection Layer
Ruize Gao (Shanghai Jiao Tong University), Rui Wang (Shanghai Jiao Tong University)
GenerationMeta LearningTransformerSupervised Fine-TuningText
🎯 What it does: This paper explains that nearest neighbor machine translation (kNN-MT) actually performs implicit gradient descent on the output projection layer (OPL), which is a meta-optimization process, through theoretical derivation and empirical experiments;
NeuSTIP: A Neuro-Symbolic Model for Link and Time Prediction in Temporal Knowledge Graphs
Ishaan Singh (Indian Institute of Technology), Mausam (Indian Institute of Technology)
Representation LearningRecurrent Neural NetworkGraph
🎯 What it does: Propose NeuSTIP, a neural symbolic model capable of simultaneously performing link prediction and time interval prediction on temporal knowledge graphs (TKG);
NL2TL: Transforming Natural Languages to Temporal Logics using Large Language Models
Yongchao Chen, Chuchu Fan
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Constructed a dataset of approximately 28K 'enhanced' natural language–temporal logic (NL–TL) pairs, trained a T5 model to achieve high-quality conversion from NL to TL, and proposed two application schemes to complete the full NL-to-TL transformation.
NLI4CT: Multi-Evidence Natural Language Inference for Clinical Trial Reports
Mael Jullien, Andre Freitas
TransformerTextBiomedical DataBenchmark
🎯 What it does: Created a new dataset NLI4CT containing 2400 natural language inference and evidence retrieval tasks from breast cancer clinical trial reports.
Noisy Exemplars Make Large Language Models More Robust: A Domain-Agnostic Behavioral Analysis
Hongyi Zheng (New York University), Abulhair Saparov (New York University)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: This paper systematically evaluates the robustness of large language models on multi-hop reasoning tasks, proposing four domain-agnostic perturbations (typographical errors, synonym replacement, repeated sentences, and inserted shortcuts), and conducts comparative experiments on Chain-of-Thought (COT), Zero-Shot COT (0COT), and Least-to-Most (LTM) prompting methods using the GSM8K and StrategyQA datasets on models such as GPT-3.5-Turbo and LLaMA2-7B/13B.
Non-Autoregressive Math Word Problem Solver with Unified Tree Structure
Yi Bin (National University of Singapore), Heng Shen
GenerationTransformerText
🎯 What it does: Propose a non-autoregressive mathematical word problem solver, MWP-NAS, which generates expressions based on a unified MTree structure;
Non-autoregressive Streaming Transformer for Simultaneous Translation
Zhengrui Ma (Key Laboratory of Intelligent Information Processing Institute of Computing Technology Chinese Academy of Sciences), Yang Feng (School of Future Science and Engineering Soochow University)
TransformerText
🎯 What it does: Propose a non-autoregressive streaming Transformer (NAST) for real-time machine translation, addressing the non-monotonic alignment and source information leakage issues in traditional autoregressive models.
Non-autoregressive Text Editing with Copy-aware Latent Alignments
Yu Zhang (Soochow University), Guohong Fu (Soochow University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes a non-autoregressive text editing method based on CTC, which completes the transformation from the source text to the target text by leveraging the potential alignment of three operations: copy (KEEP), delete (DELETE), and insert (ADD).
Non-Programmers Can Label Programs Indirectly via Active Examples: A Case Study with Text-to-SQL
Ruiqi Zhong (University of California, Berkeley), Jason Eisner (Microsoft Semantic Machines)
Data-Centric LearningAI Code AssistantLarge Language ModelTextTabular
🎯 What it does: Propose the APEL framework, enabling non-programmers to indirectly annotate the mapping from natural language to SQL programs by selecting program input-output results.
Norm of Word Embedding Encodes Information Gain
Momose Oyama (Kyoto University), Hidetoshi Shimodaira (Kyoto University)
Explainability and InterpretabilityRepresentation LearningText
🎯 What it does: This paper demonstrates through theoretical and experimental analysis that the squared norm of word vectors in the skip-gram negative sampling model approximately equals the mutual information (KL divergence) of the word, and further proves that this relationship also applies to contextual embeddings in language models.
NormDial: A Comparable Bilingual Synthetic Dialog Dataset for Modeling Social Norm Adherence and Violation
Oliver Li (Columbia University), Smaranda Muresan (Columbia University)
Data SynthesisTransformerLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: A bilingual (Chinese-English) two-person dialogue dataset called NORMDIAL was constructed through a human-AI collaborative generation process, with each dialogue turn annotated for compliance and violation of social norms.
NORMSAGE: Multi-Lingual Multi-Cultural Norm Discovery from Conversations On-the-Fly
Yi Fung (University of Illinois Urbana-Champaign), Heng Ji (University of Illinois Urbana-Champaign)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposes the NORMSAGE framework, leveraging GPT-3 zero-shot prompting and self-verification techniques to instantly discover, verify, and explain social norms from multilingual (English, Chinese) multicultural dialogues.
Not all Fake News is Written: A Dataset and Analysis of Misleading Video Headlines
Yoo Yeon Sung (University of Maryland), Naeemul Hassan (University of Maryland)
ClassificationAnomaly DetectionContrastive LearningVideoTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Constructed the video misleading title dataset VMH, analyzed its characteristics, and proposed a multimodal misinformation detection baseline.
Not all quantifiers are equal: Probing Transformer-based language models’ understanding of generalised quantifiers
Tharindu Madusanka (University of Manchester), Riza Batista-Navarro (University of Manchester)
Explainability and InterpretabilityTransformerSupervised Fine-TuningPrompt EngineeringTextChain-of-Thought
🎯 What it does: Investigated the impact of generalized quantifiers on transformer-based language models in model-checking tasks, evaluating the models' reasoning capabilities under both fine-tuned and zero-shot settings.
On Bilingual Lexicon Induction with Large Language Models
Yaoyiran Li (University of Cambridge), Ivan Vulić (University of Cambridge)
TransformerLarge Language ModelSupervised Fine-TuningTextRetrieval-Augmented Generation
🎯 What it does: This paper studies how to utilize multilingual text-to-text large language models (mLLMs) for bilingual lexicon induction (BLI), evaluating their performance on standard BLI datasets through zero-shot, few-shot prompting, and fine-tuning methods;
On Evaluation of Bangla Word Analogies
Mousumi Akter (Auburn University), Shubhra Kanti Karmaker Santu
Representation LearningTransformerTextBenchmark
🎯 What it does: Created two new Bengali word analogy datasets (16,678 manually constructed analogies and 10,594 translated and refined Mikolov data), and used them to evaluate various traditional and Transformer-based word vector models.
On the Automatic Generation and Simplification of Children’s Stories
Maria Valentini (University of Colorado Boulder), Katharina von der Wense (Johannes Gutenberg University Mainz)
GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This paper studies the automatic generation of children's stories appropriate for age groups by large language models (LLMs), as well as the application of lexical simplification models in this task;
On the Benefits of Learning to Route in Mixture-of-Experts Models
Nishanth Dikkala (Google Research), Xin Wang (Google Research)
OptimizationComputational EfficiencyRepresentation LearningMixture of ExpertsImageText
🎯 What it does: Investigated the impact of router learning on model performance in Mixture-of-Experts models, combining theoretical proofs with synthetic/real data experiments.
On the Challenges of Using Black-Box APIs for Toxicity Evaluation in Research
Luiza Pozzobon, Sara Hooker
TextBenchmark
🎯 What it does: This paper evaluates the experimental reproducibility issues caused by updates to commercial black-box toxicity detection APIs (e.g., Perspective API) over time, and demonstrates their impact by re-scoring the RTP dataset, HELM benchmark, and toxicity mitigation methods.
On the Representational Capacity of Recurrent Neural Language Models
Franz Nowak (ETH Zürich), Ryan Cotterell (ETH Zürich)
Computational EfficiencyRepresentation LearningRecurrent Neural Network
🎯 What it does: This paper formally defines recurrent neural network language models (RLMs), proving their equivalence in expressive power to probabilistic Turing machines (PTMs) when using rational weights and infinite computation time, and establishes lower bounds under real-time computational constraints, clarifying which probabilistic language models RLMs can theoretically simulate.
Once is Enough: A Light-Weight Cross-Attention for Fast Sentence Pair Modeling
Yuanhang Yang (Harbin Institute of Technology), Zenglin Xu (Harbin Institute of Technology)
RetrievalComputational EfficiencyRepresentation LearningTransformerText
🎯 What it does: Propose MixEncoder, a lightweight cross-attention model for sentence pair modeling;
Once Upon a {\it Time} in {\it Graph}: Relative-Time Pretraining for Complex Temporal Reasoning
Sen Yang (Chinese University of Hong Kong), Wai Lam (Chinese University of Hong Kong)
Representation LearningGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningTextGraph
🎯 What it does: Propose the REMEMO framework, which constructs a graph structure using relative temporal relationships to enhance pre-trained language models
Oolong: Investigating What Makes Transfer Learning Hard with Controlled Studies
Zhengxuan Wu (Stanford University), Isabel Papadimitriou (Stanford University)
Domain AdaptationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper systematically investigates the impact of three factors—syntax, lexical alignment, and tokenizer quality—in cross-lingual transfer by conducting a series of controlled language transformations (e.g., word order perturbation, resetting/shuffling word embedding matrices, using non-dedicated tokenizers) on the GLUE dataset while maintaining a single corpus.
Open Information Extraction via Chunks
Kuicai Dong (Nanyang Technological University), Xiaoli Li (Nanyang Technological University)
Representation LearningGraph Neural NetworkTransformerText
🎯 What it does: Propose treating sentences as block sequences (SaC) for open information extraction, and implement an end-to-end Chunk-OIE model that can simultaneously perform chunk partitioning and triplet extraction.
Open-world Semi-supervised Generalized Relation Discovery Aligned in a Real-world Setting
William Hogan (University of California San Diego), Jingbo Shang (University of California San Diego)
ClassificationTransformerPrompt EngineeringText
🎯 What it does: This paper proposes a Generalized Relation Discovery task that simultaneously handles known and unknown relations in open-world relation extraction, while considering negative samples and long-tail distributions.
OpenAsp: A Benchmark for Multi-document Open Aspect-based Summarization
Shmuel Amar (Bar-Ilan University), Ido Dagan (Bar-Ilan University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: By extracting key information from existing multi-document summary datasets, we designed and constructed a multi-document open-ended oriented summary benchmark named OPENASP, and established an efficient annotation protocol.
Optimized Tokenization for Transcribed Error Correction
Tomer Wullach (OriginAI), Shlomo Chazan
RestorationData SynthesisOptimizationTransformerSupervised Fine-TuningTextAudio
🎯 What it does: This paper proposes a transcription error correction model trained solely on synthetic data, generating noisy text using real transcription error distributions, and balancing memory and generalization through language-specific BPE vocabulary sizes and maximum token length adjustments;
Optimizing Retrieval-augmented Reader Models via Token Elimination
Moshe Berchansky (Intel Labs), Moshe Wasserblat (Intel Labs)
RetrievalComputational EfficiencyTransformerTextRetrieval-Augmented Generation
🎯 What it does: Research and implemented a dynamic token filtering method (Token Filtering) in the FiD generative question-answering model, combining it with CALM's layer skipping technique to reduce decoder cross-attention computation and improve the speed of long text generation.