EMNLP 2023 Papers — Page 11
Conference on Empirical Methods in Natural Language Processing · 1047 papers
Using Artificial French Data to Understand the Emergence of Gender Bias in Transformer Language Models
Lina Conti (Fondazione Bruno Kessler), Guillaume Wisniewski (Université Paris Cité)
Explainability and InterpretabilityData-Centric LearningTransformerText
🎯 What it does: Train a Transformer language model using a manually created French corpus based on PCFG to investigate how it learns gender information and exhibits gender bias.
Using Interpretation Methods for Model Enhancement
Zhuo Chen (ShanghaiTech University), Kewei Tu (ShanghaiTech University)
Explainability and InterpretabilityData-Centric LearningTransformerContrastive LearningText
🎯 What it does: Propose the UIMER framework, which jointly trains models using explanation methods and gold standard rationality to improve performance in low-resource NLP tasks.
Variance Matters: Detecting Semantic Differences without Corpus/Word Alignment
Ryo Nagata (Konan University), Yoshifumi Kawasaki (University of Tokyo)
Representation LearningTransformerLarge Language ModelText
🎯 What it does: Propose a method that detects semantic differences between two corpora using only the mean norm of contextualized word vectors (i.e., the variance/concentration of word vector distributions) and can extract typical instances.
VECHR: A Dataset for Explainable and Robust Classification of Vulnerability Type in the European Court of Human Rights
Shanshan Xu, Matthias Grabmair (Graduate Institute of International and Development Studies)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelTextBenchmark
🎯 What it does: Constructed and released the VECHR dataset for identifying and explaining types of vulnerability in cases from the European Court of Human Rights.
Vera: A General-Purpose Plausibility Estimation Model for Commonsense Statements
Jiacheng Liu (University of Washington), Hannaneh Hajishirzi (Allen Institute for Artificial Intelligence)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: Developed a generic explainability detection model VERA for evaluating the credibility of single-sentence common-sense statements
VIBE: Topic-Driven Temporal Adaptation for Twitter Classification
Yuji Zhang (Hong Kong Polytechnic University), Wenjie Li (Hong Kong Polytechnic University)
ClassificationDomain AdaptationTransformerText
🎯 What it does: To address time drift in social media text classification, the VIBE model is proposed, which leverages the information bottleneck and neural topic models to capture the evolution of past and future topics, achieving adaptive learning.
Vicarious Offense and Noise Audit of Offensive Speech Classifiers: Unifying Human and Machine Disagreement on What is Offensive
Tharindu Weerasooriya (Rochester Institute of Technology), Ashiqur KhudaBukhsh (Rochester Institute of Technology)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This study detects harmful language in political social media comments through large-scale noise auditing and human evaluation, exploring the consistency and differences between machines and humans in first-person and vicarious offense perception.
Vicinal Risk Minimization for Few-Shot Cross-lingual Transfer in Abusive Language Detection
Gretel De la Peña Sarracén, Simone Ponzetto
ClassificationDomain AdaptationTransformerSupervised Fine-TuningText
🎯 What it does: This paper addresses hate speech detection in low-resource languages, proposing and evaluating three data augmentation methods based on Vicinal Risk Minimization (SSMBA, MIXUP, MIXAG), and conducting few-shot cross-lingual transfer experiments on the multi-domain, multi-language XHATE-999 dataset.
Video-Helpful Multimodal Machine Translation
Yihang Li (Kyoto University), Wei Li (Google Research)
GenerationTransformerVision Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: This paper proposes the EVA dataset and the SAFA model, utilizing video information to resolve ambiguities in subtitle translation.
ViPE: Visualise Pretty-much Everything
Hassan Shahmohammadi (University of Tübingen), Hendrik Lensch
Image TranslationGenerationRetrievalKnowledge DistillationTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: Train a lightweight language model called ViPE to generate visual descriptions from any text, thereby helping text-to-image models better present metaphors and non-literal expressions.
Vision-Enhanced Semantic Entity Recognition in Document Images via Visually-Asymmetric Consistency Learning
Hao Wang (Shanghai University), Chenhui Chu (Kyoto University)
RecognitionData SynthesisTransformerVision Language ModelImageTextMultimodality
🎯 What it does: Propose a visual symmetric consistency learning framework VANCL leveraging color priors for semantic entity recognition in visually rich form documents
ViSoBERT: A Pre-Trained Language Model for Vietnamese Social Media Text Processing
Nam Nguyen (University of Information Technology), Kiet Nguyen (University of Information Technology)
ClassificationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Trained and released ViSoBERT, a monolingual PLM pre-trained specifically for Vietnamese social media text.
ViStruct: Visual Structural Knowledge Extraction via Curriculum Guided Code-Vision Representation
Yangyi Chen (University Of Illinois Urbana Champaign), Heng Ji (University Of Illinois Urbana Champaign)
Representation LearningTransformerVision Language ModelImageTextMultimodality
🎯 What it does: Propose the ViStruct framework, which uses a programming language (Python) to uniformly encode visual structural knowledge (concepts, attributes, relations, events), and gradually trains visual language models to extract multi-grained structural information through a hierarchical curriculum learning (Curriculum Pyramid).
Visually-Situated Natural Language Understanding with Contrastive Reading Model and Frozen Large Language Models
Geewook Kim (NAVER Cloud AI), Seunghyun Park (NAVER Cloud AI)
Computational EfficiencyRepresentation LearningTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Proposes the Contrastive Reading Model (Cream), a multimodal framework integrating a visual encoder, auxiliary encoder, and contrastive learning to enhance understanding of text-rich images and achieve soft visual prompting fusion with large language models (LLMs).
ViT-TTS: Visual Text-to-Speech with Scalable Diffusion Transformer
Huadai Liu (Zhejiang University), Zhou Zhao (Zhejiang University)
GenerationSafty and PrivacyTransformerVision Language ModelDiffusion modelContrastive LearningImageTextMultimodalityAudio
🎯 What it does: Studied and proposed the first visual-text to speech (ViT-TTS) model, which generates high-quality speech that matches the target environment by leveraging visual information along with text.
VivesDebate-Speech: A Corpus of Spoken Argumentation to Leverage Audio Features for Argument Mining
Ramon Ruiz-Dolz (University of Dundee), Javier Iranzo-Sánchez (Universitat Politècnica de València)
ClassificationSegmentationTransformerSupervised Fine-TuningTextAudio
🎯 What it does: Constructed the VivesDebate-Speech speech corpus and implemented and evaluated an argument unit (ADU) segmentation model based on audio features on this corpus, exploring the effectiveness of end-to-end and cascade approaches.
VLIS: Unimodal Language Models Guide Multimodal Language Generation
Jiwan Chung (Yonsei University), Youngjae Yu (Yonsei University)
GenerationLarge Language ModelVision Language ModelMultimodality
🎯 What it does: Propose a framework named VLIS, which combines visual language models (VLM) with vision-free text language models during the inference phase to enhance language understanding and visual alignment capabilities in multimodal text generation.
We Are What We Repeatedly Do: Inducing and Deploying Habitual Schemas in Persona-Based Responses
Benjamin Kane (University of Rochester), Lenhart Schubert (University of Rochester)
GenerationTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes generating 'habitual event patterns (schema)' from simple self-knowledge facts, and using retrieved patterns to guide large language models (LLM) in generating dialogue responses consistent with character personas.
We are Who We Cite: Bridges of Influence Between Natural Language Processing and Other Academic Fields
Jan Philip Wahle (National Research Council Canada), Saif Mohammad
Text
🎯 What it does: Quantify the citation impact between natural language processing (NLP) and 23 other disciplines, revealing that NLP's interdisciplinary engagement has significantly declined over the past few decades.
We Need to Talk About Reproducibility in NLP Model Comparison
Yan Xue (Shanxi University), Jihong Li (Shanxi University)
Text
🎯 What it does: This paper theoretically analyzes the relationship between reproducibility and the performance estimator SNR in NLP model comparisons, and proposes an evaluation scheme combining 3×2 block cross-validation with a hybrid estimator.
We’re Afraid Language Models Aren’t Modeling Ambiguity
Alisa Liu (University of Washington), Yejin Choi (University of Washington)
ClassificationRecognitionData SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: This paper constructs the first benchmark dataset, AMBIENT, specifically targeting ambiguity in natural language inference (NLI), and evaluates the understanding and processing capabilities of pre-trained language models and fine-tuned multi-label NLI models for ambiguity through a series of experiments.
Weakly Supervised Semantic Parsing with Execution-based Spurious Program Filtering
Kang-il Lee (Seoul National University), Kyomin Jung (Seoul National University)
World ModelTextTabularRetrieval-Augmented Generation
🎯 What it does: Propose a voting filtering mechanism based on program execution results to remove spurious programs in weakly supervised semantic parsing.
Weakly-Supervised Learning of Visual Relations in Multimodal Pretraining
Emanuele Bugliarello (Google DeepMind), Lisa Hendricks (University of Copenhagen)
RetrievalRepresentation LearningVision Language ModelContrastive LearningMultimodality
🎯 What it does: Propose two weakly supervised pre-training methods based on scene graphs (Verbalised Scene Graphs and Masked Relation Classification), leveraging a small amount of manually annotated visual relationships to enhance the fine-grained visual-linguistic understanding capability of multi-modal pre-training models.
Well Begun is Half Done: Generator-agnostic Knowledge Pre-Selection for Knowledge-Grounded Dialogue
Lang Qin (Nankai University), Zhenglu Yang (Nankai University)
GenerationRetrievalGraph Neural NetworkTransformerLarge Language ModelReinforcement LearningPrompt EngineeringTextGraphRetrieval-Augmented Generation
🎯 What it does: This paper proposes GATE, a model-agnostic knowledge pre-selection method that first unifies text and knowledge graphs into a graph structure, uses a graph attention network to score nodes, and adaptively selects varying amounts of knowledge through reinforcement learning, providing high-quality context for subsequent dialogue generation.
What Comes Next? Evaluating Uncertainty in Neural Text Generators Against Human Production Variability
Mario Giulianelli (University of Amsterdam), Barbara Plank (LMU Munich)
GenerationExplainability and InterpretabilityTransformerText
🎯 What it does: This paper evaluates whether the uncertainty of neural text generators across different tasks matches human-generated variability by performing instance-level multi-dimensional distance analysis (lexical, syntactic, semantic) on multi-reference datasets for four natural language generation tasks.
What do Deck Chairs and Sun Hats Have in Common? Uncovering Shared Properties in Large Concept Vocabularies
Amit Gajbhiye (Cardiff University), Steven Schockaert (Cardiff University)
ClassificationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a modeling method based on shared concept attributes, representing each concept as a set of attributes it satisfies, and leveraging this representation to enhance performance in ultra-fine entity type annotation tasks.
What Else Do I Need to Know? The Effect of Background Information on Users’ Reliance on QA Systems
Navita Goyal (University of Maryland), Hal Daumé III (U.S. Army Research Lab)
Explainability and InterpretabilityTransformerSupervised Fine-TuningText
🎯 What it does: Examined how users rely on model predictions when question-answering systems lack background information, and explored the impact of providing necessary background information on user dependency and confidence.
What to Read in a Contract? Party-Specific Summarization of Legal Obligations, Entitlements, and Prohibitions
Abhilasha Sancheti (University of Maryland), Rachel Rudinger (Adobe Research)
GenerationTransformerLarge Language ModelSupervised Fine-TuningTextFinance Related
🎯 What it does: Proposes a specific summarization task targeting contract parties, automatically extracting key obligations, rights, and prohibitions of both parties in lease contracts.
What’s “up” with vision-language models? Investigating their struggle with spatial reasoning
Amita Kamath (University of California, Los Angeles), Kai-Wei Chang (University of California, Los Angeles)
Representation LearningPrompt EngineeringVision Language ModelContrastive LearningImageTextBenchmark
🎯 What it does: This paper proposes three new benchmarks specifically designed to evaluate the spatial reasoning capabilities of vision-language models (What’sUp, COCO-spatial, GQA-spatial), and systematically assesses the performance of 18 mainstream models;
When are Lemons Purple? The Concept Association Bias of Vision-Language Models
Yingtian Tang (EPFL), Ilker Yildirim (EPFL)
RecognitionExplainability and InterpretabilityTransformerPrompt EngineeringVision Language ModelDiffusion modelContrastive LearningImageMultimodality
🎯 What it does: Study the concept association bias (CAB) in Vision-Language models when dealing with images containing multiple concepts, and explore the impact of this bias on zero-shot VQA performance.
When Do Decompositions Help for Machine Reading?
Kangda Wei (Texas A&M University), Orion Weller (Texas A&M University)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: The study investigates the impact of question decomposition on answer quality in machine reading comprehension tasks, and systematically evaluates performance across different models, decomposition strategies, and data scales.
When Language Models Fall in Love: Animacy Processing in Transformer Language Models
Michael Hanna (University of Amsterdam), Sandro Pezzelle (University of Amsterdam)
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: This study treats pre-trained Transformer language models (GPT-2, OPT, LLaMA) as subjects in psycholinguistic experiments, investigating their behaviors in typical and atypical animacy processing, primarily by comparing the models' prediction probabilities (surprisal) with human EEG N400 responses.
When Reviewers Lock Horns: Finding Disagreements in Scientific Peer Reviews
Sandeep Kumar (Indian Institute of Technology Patna), Asif Ekbal (Indian Institute of Technology Patna)
ClassificationTransformerLarge Language ModelText
🎯 What it does: This paper proposes the task of automatically identifying contradictions in peer review comments and constructs a large dataset of conflicting review comment pairs named ContraSciView.
When the Majority is Wrong: Modeling Annotator Disagreement for Subjective Tasks
Eve Fleisig (University of California Berkeley), Dan Klein (University of California Berkeley)
Safty and PrivacyData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextTabular
🎯 What it does: This paper proposes a model that predicts the level of offensiveness scores given by annotators based on their demographic characteristics and online content preferences, and further estimates the scores of members of the target group addressed by the text, thereby identifying cases where majority voting may be incorrect.
Where to start? Analyzing the potential value of intermediate models
Leshem Choshen (IBM Research), Yoav Katz (IBM Research)
ClassificationRepresentation LearningData-Centric LearningTransformerSupervised Fine-TuningText
🎯 What it does: This paper systematically investigates the potential benefits of using fine-tuned models (Intermediate Models) as base models for re-fine-tuning (Intertraining) in downstream tasks, analyzes the sensitivity of the target dataset to these benefits, the predictability of base model quality, and proposes a static model ranking method based on the MNLI linear probe; meanwhile, it provides practical model selection and evaluation processes.
Whispering LLaMA: A Cross-Modal Generative Error Correction Framework for Speech Recognition
Srijith Radhakrishnan (King Abdullah University of Science and Technology), Jesper N. Tegner (King Abdullah University of Science and Technology)
RecognitionTransformerLarge Language ModelPrompt EngineeringMultimodality
🎯 What it does: Propose a cross-modal generative error correction framework named Whispering LLaMA, integrating the Whisper acoustic model with the LLaMA language model, leveraging n-best stems and audio features for ASR error correction.
Why LLMs Hallucinate, and How to Get (Evidential) Closure: Perceptual, Intensional, and Extensional Learning for Faithful Natural Language Generation
Adam Bouyamourn (University Of California Berkeley)
GenerationExplainability and InterpretabilityLarge Language ModelReinforcement LearningImageTextRetrieval-Augmented Generation
🎯 What it does: Investigate the root causes of hallucinations in LLMs and propose a trustworthy generation framework centered on the concept of 'evidence closure'.
Why Should This Article Be Deleted? Transparent Stance Detection in Multilingual Wikipedia Editor Discussions
Lucie-Aimée Kaffee (Hasso Plattner Institute), Isabelle Augenstein (University of Copenhagen)
ClassificationTransformerText
🎯 What it does: This paper constructs a multilingual dataset from Wikipedia deletion discussions, proposing to jointly perform stance detection and policy prediction through multi-task learning to enhance the transparency of content moderation.
WiCE: Real-World Entailment for Claims in Wikipedia
Ryo Kamoi (University of Texas at Austin), Greg Durrett (University of Texas at Austin)
ClassificationRetrievalExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Constructed the WICE dataset by automatically splitting Wikipedia claims into sub-claims using GPT-3.5, and performed fine-grained annotations for text entailment, evidence retrieval, and unsupported terms;
WSDMS: Debunk Fake News via Weakly Supervised Detection of Misinforming Sentences with Contextualized Social Wisdom
Ruichao Yang (Hong Kong Baptist University), Zhiwei Yang (Hong Kong Baptist University)
ClassificationGraph Neural NetworkTextGraph
🎯 What it does: This paper proposes a weakly supervised multi-instance learning framework called WSDMS, which utilizes social media conversation trees to detect misleading sentences in news articles and infer the overall truthfulness of the article under the condition of having only article authenticity labels.
XLM-V: Overcoming the Vocabulary Bottleneck in Multilingual Masked Language Models
Davis Liang (Meta AI), Madian Khabsa (Meta AI)
Representation LearningLarge Language ModelText
🎯 What it does: Propose XLM-V, a multilingual model using one million subwords, which reduces subword sharing across languages and mitigates tokenization over-segmentation by improving the vocabulary construction method.
You Told Me That Joke Twice: A Systematic Investigation of Transferability and Robustness of Humor Detection Models
Alexander Baranov (HSE University), Pavel Braslavski (HSE University)
ClassificationAdversarial AttackData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Systematically trained and evaluated humor detection models based on RoBERTa, Naïve Bayes, and large language models, testing their generalization and robustness through cross-validation, adversarial attacks, and supplementary datasets.
Zero-shot Faithfulness Evaluation for Text Summarization with Foundation Language Model
Qi Jia (Shanghai Jiao Tong University), Kenny Zhu
GenerationTransformerLarge Language ModelText
🎯 What it does: Propose a zero-shot factualness evaluation metric called FFLM based on foundational language models, which directly measures the factualness of summaries by analyzing the probability changes between summaries and source documents.
Zero-Shot Multi-Label Topic Inference with Sentence Encoders and LLMs
Souvika Sarkar (Auburn University), Shubhra Kanti Karmaker Santu (Auburn University)
ClassificationTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes a real-time zero-shot multi-label topic inference framework that uses sentence encoders and large language models (LLMs) to perform unsupervised document topic classification.
Zero-shot Sharpness-Aware Quantization for Pre-trained Language Models
Miaoxi Zhu (Wuhan University), Dacheng Tao (University of Sydney)
OptimizationComputational EfficiencyTransformerLarge Language ModelGenerative Adversarial NetworkText
🎯 What it does: Propose a Zero-Sample Sharpness-Aware Quantization (ZSAQ) framework that quantizes pre-trained language models through generative adversarial learning, and introduces a feature adaptation module to avoid gradient propagation for discrete words.
ZEROTOP: Zero-Shot Task-Oriented Semantic Parsing using Large Language Models
Dheeraj Mekala (University of California San Diego), Subhro Roy (Microsoft Semantic Machines)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Propose a zero-shot task-oriented semantic parsing framework called ZEROTOP, which decomposes the parsing task into abstract and extractive question answering. It leverages large language models to generate intents and slot values, and employs a fine-tuned Abstainer to reject unanswerable slots, thereby achieving zero-shot semantic parsing.
ZGUL: Zero-shot Generalization to Unseen Languages using Multi-source Ensembling of Language Adapters
Vipul Rathore (Indian Institute of Technology), Mausam (Indian Institute of Technology)
Domain AdaptationRepresentation LearningTransformerText
🎯 What it does: Studied how to utilize multi-source language adapters (LA) for zero-shot cross-lingual transfer without labeled target language data. Proposed an architecture ZGUL that fuses multi-source LA during training and further fine-tunes attention weights via entropy minimization at test time.