ACL 2023 Papers — Page 2
Annual Meeting of the Association for Computational Linguistics · 1074 papers
AV-TranSpeech: Audio-Visual Robust Speech-to-Speech Translation
Rongjie Huang (Zhejiang University), Zhou Zhao (Zhejiang University)
GenerationKnowledge DistillationTransformerContrastive LearningVideoMultimodalityAudio
🎯 What it does: Propose AV-TranSpeech, the first text-free audio-visual speech-to-speech translation model;
Back to Patterns: Efficient Japanese Morphological Analysis with Feature-Sequence Trie
Naoki Yoshinaga (University of Tokyo)
Computational EfficiencyText
🎯 What it does: This paper proposes a greedy longest matching algorithm that utilizes a dictionary and annotated data to extract POS-enhanced patterns, which are stored in a double array trie, achieving an efficient morphological analyzer for Japanese tokenization, part-of-speech tagging, and lemma generation;
Back Translation for Speech-to-text Translation Without Transcripts
Qingkai Fang (Key Laboratory of Intelligent Information Processing Institute of Computing Technology Chinese Academy of Sciences), Yang Feng (Key Laboratory of Intelligent Information Processing Institute of Computing Technology Chinese Academy of Sciences)
Data SynthesisTransformerGenerative Adversarial NetworkContrastive LearningTextAudio
🎯 What it does: This paper proposes a speech-to-text back-translation method that does not require source language transcriptions, utilizing target-side monolingual data to generate pseudo ST data.
Backdooring Neural Code Search
Weisong Sun (State Key Laboratory for Novel Software Technology Nanjing University), Bin Luo (State Key Laboratory for Novel Software Technology Nanjing University)
RetrievalAdversarial AttackTransformerText
🎯 What it does: Researched and implemented a backdoor injection attack on neural code search models, exploiting minor changes in function/variable names to elevate the ranking of vulnerable code snippets in search results.
Backpack Language Models
John Hewitt (Stanford University), Percy Liang (Stanford University)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Propose the Backpack model, which generates word representations by linearly combining multiple non-contextual sense vectors through contextualized weight combinations, constructing interpretable and controllable language models.
Balancing Lexical and Semantic Quality in Abstractive Summarization
Jeewoo Sul (Hanyang University), Yong Suk Choi (Hanyang University)
GenerationTransformerContrastive LearningText
🎯 What it does: Propose the BalSum model in the abstract summary, which adopts a two-stage re-ranking framework to balance lexical overlap and semantic similarity in order to alleviate exposure bias;
Being Right for Whose Right Reasons?
Terne Sasha Thorn Jakobsen (Copenhagen Center for Social Data Science), Anders Søgaard (University of Copenhagen)
Explainability and InterpretabilityTransformerSupervised Fine-TuningText
🎯 What it does: This paper first constructs three interpretable annotated datasets with demographic information (DynaSent, SST-2, CoS-E), and uses these datasets to evaluate the alignment between Transformer model predictions and human reasoning (rationales), exploring model fairness across different age and ethnic groups.
Benchmarking Large Language Model Capabilities for Conditional Generation
Joshua Maynez (Google DeepMind), Sebastian Gehrmann (Google Research)
GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: Evaluate the performance of pre-trained language models with different scales and architectures on multilingual conditional generation tasks, and provide benchmarks and best practices.
BERM: Training the Balanced and Extractable Representation for Matching to Improve Generalization Ability of Dense Retrieval
Shicheng Xu (Institute of Computing Technology, Chinese Academy of Sciences), Xueqi Cheng (Institute of Computing Technology, Chinese Academy of Sciences)
RetrievalDomain AdaptationContrastive LearningText
🎯 What it does: Propose a new method called BERM to enhance the generalization ability of dense retrieval in cross-domain zero-shot scenarios;
Best-k Search Algorithm for Neural Text Generation
Jiacheng Xu (Salesforce AI Research), Yingbo Zhou (Salesforce AI Research)
GenerationComputational EfficiencyTransformerText
🎯 What it does: Proposed a new deterministic decoding algorithm called Bestk Search, aiming to balance text diversity and quality in generation tasks.
Better Simultaneous Translation with Monotonic Knowledge Distillation
Shushu Wang (Zhejiang University), Zhongqiang Huang (Alibaba DAMO Academy)
GenerationKnowledge DistillationTransformerText
🎯 What it does: Propose a two-stage beam search to generate monotonic pseudo targets and train a simultaneous translation model using sequence-level knowledge distillation;
Beyond Contrastive Learning: A Variational Generative Model for Multilingual Retrieval
John Wieting (Google DeepMind), Taylor Berg-Kirkpatrick (University of California San Diego)
GenerationRetrievalRepresentation LearningTransformerAuto EncoderContrastive LearningText
🎯 What it does: Proposed a variational generative model called VMSST for learning multilingual text embeddings and enabling retrieval and scoring
Beyond English-Centric Bitexts for Better Multilingual Language Representation Learning
Barun Patra (Microsoft), Xia Song (Microsoft)
Representation LearningData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: Proposed the XY-LENT multilingual pre-training framework, leveraging multi-language (X-Y) parallel texts and improved sampling strategies to enhance representation learning efficiency;
bgGLUE: A Bulgarian General Language Understanding Evaluation Benchmark
Momchil Hardalov (AWS AI Labs), Dragomir Radev (Yale University)
ClassificationTransformerSupervised Fine-TuningTextBenchmark
🎯 What it does: Proposed bgGLUE, the first general language understanding evaluation benchmark for Bulgarian, containing nine tasks covering annotation, classification, reasoning, sentiment analysis, and question answering;
Bhasa-Abhijnaanam: Native-script and romanized Language Identification for 22 Indic languages
Yash Madhani (AI4Bharat), Anoop Kunchukuttan (AI4Bharat)
ClassificationData SynthesisTransformerSupervised Fine-TuningMixture of ExpertsTextBenchmark
🎯 What it does: Constructed a language identification benchmark named Bhasha-Abhijnaanam covering 22 languages of the Indian Constitution, and trained a language identification model named IndicLID that supports both native scripts and Romanized text.
Bi-Phone: Modeling Inter Language Phonetic Influences in Text
Abhirut Gupta (Google Research), Aravindan Raghuveer (Google Research)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper proposes the Bi-Phone method, which generates spelling errors influenced by L1-L2 by mining phoneme confusions across different language pairs, and synthesizes L2 text based on this; meanwhile, it constructs the FunGLUE benchmark to evaluate model robustness under phoneme errors.
BIC: Twitter Bot Detection with Text-Graph Interaction and Semantic Consistency
Zhenyu Lei (Tsinghua University), Minnan Luo (Tsinghua University)
ClassificationGraph Neural NetworkTransformerTextGraph
🎯 What it does: Propose a Twitter bot detection framework named BIC, which detects traditional and advanced bots through two mechanisms: text-graph interaction and semantic consistency.
Bidirectional Generative Framework for Cross-domain Aspect-based Sentiment Analysis
Yue Deng (DAMO Academy, Alibaba Group), Lidong Bing (DAMO Academy, Alibaba Group)
Data SynthesisDomain AdaptationTransformerLarge Language ModelText
🎯 What it does: This paper studies cross-domain aspect-based sentiment analysis (ABSA) and proposes a bidirectional generation framework called BGCA. The framework first trains a text-to-label model using annotated data from the source domain, then generates pseudo-labels on unlabeled data from the target domain. These pseudo-labels are then fed into a label-to-text model to generate natural sentences that align with the labels, producing high-quality pseudo-data. Finally, the source domain data and generated data are jointly retrained in the text-to-label direction to complete cross-domain knowledge transfer.
BIG-C: a Multimodal Multi-Purpose Dataset for Bemba
Claytone Sikasote (University of Zambia), Antonios Anastasopoulos (George Mason University)
Data SynthesisTransformerImageTextMultimodalityBenchmarkAudio
🎯 What it does: This study constructs the BIG-C dataset, which includes image-based multi-turn dialogues in the Bemba language, corresponding audio transcriptions, and English translations, covering 92,117 sentences and approximately 180 hours of audio; meanwhile, it provides benchmark experiments for ASR, machine translation, and speech translation.
Binary and Ternary Natural Language Generation
Zechun Liu (Reality Labs, Meta Inc.), Raghuraman Krishnamoorthi (Reality Labs, Meta Inc.)
GenerationCompressionKnowledge DistillationTransformerText
🎯 What it does: Perform extremely low-bit (binary or ternary) weight and activation quantization on generative pre-trained Transformers (BART, mBART), proposing statistic-driven weight quantization and adaptive activation quantization methods.
BITE: Textual Backdoor Attacks with Iterative Trigger Injection
Jun Yan (University of Southern California), Xiang Ren (University of Southern California)
ClassificationAdversarial AttackTransformerLarge Language ModelText
🎯 What it does: Propose a text backdoor attack named BITE, which achieves control over target labels by injecting trigger words into training data;
Black-box language model explanation by context length probing
Ondřej Cífka (Inria), Antoine Liutkus (Inria)
Explainability and InterpretabilityTransformerText
🎯 What it does: Propose an explanation technique called Context Length Probing for causal language models, which quantifies the importance of each context segment by tracking the model's prediction distribution across different context lengths and generates differential importance scores.
BLASER: A Text-Free Speech-to-Speech Translation Evaluation Metric
Mingda Chen (Meta AI), Marta R. Costa-jussà (Meta AI)
TransformerLarge Language ModelMultimodalityBenchmarkAudio
🎯 What it does: Propose a text-free and ASR-independent speech-to-speech translation (S2ST) evaluation metric called BLASER, which directly embeds source speech, translated speech, and reference speech into a shared vector space and calculates similarity scores;
BLEURT Has Universal Translations: An Analysis of Automatic Metrics by Minimum Risk Training
Yiming Yan (Nanjing University), Mingxuan Wang (ByteDance AI Lab)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Conduct a systematic analysis of mainstream and state-of-the-art automatic translation evaluation metrics through Minimum Risk Training (MRT), revealing that BLEURT and BARTScore suffer from general adversarial translation defects, and propose using token-level constraints or metric integration to enhance metric robustness and translation quality.
BLIND: Bias Removal With No Demographics
Hadas Orgad (Technion - Israel Institute of Technology), Yonatan Belinkov (Technion - Israel Institute of Technology)
ClassificationTransformerText
🎯 What it does: Propose the BLIND method, which identifies and reduces bias in training samples by predicting the success of the main model through an auxiliary model, thereby alleviating social bias in downstream tasks without using any demographic information.
BLOOM+1: Adding Language Support to BLOOM for Zero-Shot Prompting
Zheng Xin Yong (Brown University), Vassilina Nikoulina (NAVER LABS Europe)
ClassificationDomain AdaptationComputational EfficiencyRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Building upon the existing 176B-parameter BLOOM model, this work adapts eight uncovered languages (German, Russian, Bulgarian, Thai, Turkish, Greek, Korean, and Guarani) using continued pre-training, MAD-X language adapters, and IA³, among other parameter-efficient methods. The performance is evaluated on zero-shot prompt tasks (XNLI, KLUE-NLI, AmericasNLI, XCOPA, XStoryCloze, XWinograd, PAWS-X). Additionally, the adaptation strategy is migrated to BLOOMZ (a multi-task instruction-tuned version), exploring the effects of incorporating new languages in multi-task fine-tuning.
BOLT: Fast Energy-based Controlled Text Generation with Tunable Biases
Xin Liu (University of Michigan), Lu Wang (University of Michigan)
GenerationComputational EfficiencyText
🎯 What it does: Propose BOLT, a tunable bias-based energy model for achieving fast controlled text generation while maintaining autoregressive decoding.
Bootstrapping Neural Relation and Explanation Classifiers
Zheng Tang (University of Arizona), Mihai Surdeanu (University of Arizona)
ClassificationExplainability and InterpretabilityTransformerSupervised Fine-TuningText
🎯 What it does: Proposed a self-training neural-symbolic relation extraction method, achieving semi-supervised learning by converting model explanation outputs into rules and using these rules to generate new annotations in unlabelled text.
BREAK: Breaking the Dialogue State Tracking Barrier with Beam Search and Re-ranking
Seungpil Won (LG AI Research), Kyomin Jung (Seoul National University)
TransformerLarge Language ModelPrompt EngineeringTextSequential
🎯 What it does: Propose the BREAK framework, which first generates k dialogue state candidates using beam search, and then selects the most contextually appropriate state using a re-ranker;
Breeding Machine Translations: Evolutionary approach to survive and thrive in the world of automated evaluation
Josef Jon (Charles University), Ondřej Bojar (Charles University)
OptimizationTransformerText
🎯 What it does: This paper proposes an automated method based on genetic algorithms (GA) and minimum Bayes risk (MBR) decoding to search, improve, and generate new translation candidates from the n-best list produced by machine translation (MT) systems, thereby enhancing translation quality or revealing blind spots in evaluation metrics.
Bridging the Domain Gaps in Context Representations for k-Nearest Neighbor Neural Machine Translation
Zhiwei Cao (Xiamen University), Jinsong Su (Xiamen University)
RetrievalDomain AdaptationRepresentation LearningTransformerSupervised Fine-TuningText
🎯 What it does: Propose an offline key representation corrector that reconstructs key vectors in kNN-MT data storage to reduce the representation gap between the source domain and downstream domains and improve retrieval quality;
Bridging the Gap between Decision and Logits in Decision-based Knowledge Distillation for Pre-trained Language Models
Qinhong Zhou (Tsinghua University), Yang Liu (Tsinghua University)
Knowledge DistillationTransformerTextBenchmark
🎯 What it does: Propose a decision-based knowledge distillation method that estimates the teacher model's logits by bridging the information gap between decisions and logits through test-time data augmentation and theoretical orthogonal angle probability inference; subsequently, traditional KL distillation training is applied to the student model using the estimated logits.
Bridging The Gap: Entailment Fused-T5 for Open-retrieval Conversational Machine Reading Comprehension
Xiao Zhang (Beijing Institute of Technology), Xian-Ling Mao (Beijing Institute of Technology)
GenerationRetrievalComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningTextRetrieval-Augmented Generation
🎯 What it does: Proposed a one-stage end-to-end Entailment Fused-T5 (EFT) framework for simultaneously performing decision-making (Yes/No/Inquire) and subsequent question generation in open-retrieval conversational machine reading comprehension (OCMRC) tasks, bridging the information gap between decision-making and generation through shared entailment representations.
Bring More Attention to Syntactic Symmetry for Automatic Postediting of High-Quality Machine Translations
Baikjin Jung (Pohang University of Science and Technology), Yunsu Kim (Pohang University of Science and Technology)
Data-Centric LearningTransformerText
🎯 What it does: In the task of automatic post-editing for high-quality English-German machine translation, the authors propose a regularization method that encourages the MT encoder's self-attention to achieve syntactic symmetry;
BUCA: A Binary Classification Approach to Unsupervised Commonsense Question Answering
Jie He (University of Edinburgh), Jeff Pan
ClassificationRepresentation LearningTransformerLarge Language ModelContrastive LearningTextGraph
🎯 What it does: Convert knowledge graph triplets into positive and negative question-answer pairs, train a binary classification model to evaluate sentence合理性, and use it for unsupervised common sense question answering.
BUMP: A Benchmark of Unfaithful Minimal Pairs for Meta-Evaluation of Faithfulness Metrics
Liang Ma (Dataminr Inc), Alejandro Jaimes (Dataminr Inc)
GenerationData-Centric LearningTextBenchmark
🎯 What it does: Propose the BUMP benchmark dataset, using 889 human-edited single error pairs to evaluate summary faithfulness metrics.
ByGPT5: End-to-End Style-conditioned Poetry Generation with Token-free Language Models
Jonas Belouadi (Bielefeld University), Steffen Eger (Bielefeld University)
GenerationTransformerLarge Language ModelText
🎯 What it does: Developed an end-to-end style-conditioned poetry generation model named ByGPT5, capable of generating four-line poems that comply with aesthetic constraints such as meter, foot, and alliteration without requiring manual rules or post-processing.
Byte-Level Grammatical Error Correction Using Synthetic and Curated Corpora
Svanhvít Lilja Ingólfsdóttir (Miðeind), Vésteinn Snæbjarnarson (Miðeind)
Data SynthesisTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: For Icelandic grammatical error correction (GEC), this paper compares byte-level (ByT5) and subword-level (mT5, mBART) models, jointly fine-tunes them using synthetic error data and human-corrected corpora, ultimately achieving high-quality corrections on real text;
C-STANCE: A Large Dataset for Chinese Zero-Shot Stance Detection
Chenye Zhao (University of Illinois at Chicago), Cornelia Caragea (University of Illinois at Chicago)
ClassificationRecurrent Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Constructed and released the Chinese zero-shot stance detection dataset C-STANCE, containing 48,126 Weibo-target pairs covering two types of targets: noun phrases and propositions; proposed two zero-shot stance detection subtasks: target-level and domain-level; conducted experiments on multiple baseline models, demonstrating that Transformer models outperform RNN models, with RoBERTa achieving 78.5% F1 macro average on subtask A; simultaneously performed cross-lingual zero-shot experiments, showing that the Chinese dataset is more challenging.
CAME: Confidence-guided Adaptive Memory Efficient Optimization
Yang Luo (National University of Singapore), Yang You (National University of Singapore)
OptimizationComputational EfficiencyText
🎯 What it does: Proposes CAME, a confidence-guided adaptive memory-efficient optimizer for training large-scale language models.
Can Language Models Make Fun? A Case Study in Chinese Comical Crosstalk
Jianquan Li (Chinese University of Hong Kong Shenzhen), Benyou Wang (University of Manchester)
GenerationTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Constructed the largest Chinese cross-talk (xiangsheng) script dataset and evaluated the humor generation performance of multiple pre-trained language models based on this dataset.
Can Large Language Models Be an Alternative to Human Evaluations?
Cheng-Han Chiang (National Taiwan University), Hung-yi Lee (National Taiwan University)
ClassificationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes using large language models (LLMs) for text quality assessment as an alternative to traditional human evaluation.
Can LMs Learn New Entities from Descriptions? Challenges in Propagating Injected Knowledge
Yasumasa Onoe (University of Texas at Austin), Eunsol Choi
TransformerSupervised Fine-TuningPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Investigate whether language models can apply new entity definitions in reasoning tasks after receiving them, propose the entity knowledge propagation task, and construct two evaluation datasets.
Can NLI Provide Proper Indirect Supervision for Low-resource Biomedical Relation Extraction?
Jiashu Xu (Harvard University), Muhao Chen (University of Southern California)
RecognitionTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringContrastive LearningBiomedical DataBenchmark
🎯 What it does: Reformulate the biomedical relation extraction task as a natural language inference problem, utilizing natural language descriptions of relations and entity type masks for indirect supervision.
CASE: Aligning Coarse-to-Fine Cognition and Affection for Empathetic Response Generation
Jinfeng Zhou (Tsinghua University), Minlie Huang (Tsinghua University)
GenerationGraph Neural NetworkTransformerTextGraph
🎯 What it does: This study proposes the CASE model for aligning cognition and emotion in empathetic dialogues, by constructing consensus knowledge graphs and emotional concept graphs, and employing a coarse-to-fine alignment mechanism to generate more empathetic responses.
CASN:Class-Aware Score Network for Textual Adversarial Detection
Rong Bao (Fudan University), Dacheng Tao (University of Sydney)
Anomaly DetectionTransformerScore-based ModelContrastive LearningTextStochastic Differential Equation
🎯 What it does: Constructed a text adversarial sample detection framework based on a class-aware score network (CASN), which determines whether a sample is an adversarial sample by matching denoising scores, estimating the gradient of the text feature distribution through Langevin dynamics, and utilizing feature drift during the denoising process.
CAT: A Contextualized Conceptualization and Instantiation Framework for Commonsense Reasoning
Weiqi Wang (Hong Kong University of Science and Technology), Lei Chen (Hong Kong University of Science and Technology)
TransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Propose a semi-supervised framework named CAT that integrates event conceptualization and instantiation to automatically generate context-related abstract common-sense knowledge on large knowledge bases.
CATS: A Pragmatic Chinese Answer-to-Sequence Dataset with Large Scale and High Quality
Liang Li (Chinese Academy of Sciences), Yongbin Li (Alibaba Group)
GenerationData SynthesisGraph Neural NetworkTransformerLarge Language ModelTextGraphBenchmark
🎯 What it does: Constructed a large-scale, high-quality Chinese answer-to-sequence dataset named CATS, and proposed a Unified Graph Transformation (UGT) and Node Segment Embedding (NSE) method to convert SQL and result tables into a joint graph, thereby transforming the answer-to-sequence task into a graph-to-text task.
Causal Intervention and Counterfactual Reasoning for Multi-modal Fake News Detection
Ziwei Chen (Beijing University of Posts and Telecommunications), Liqiang Nie (Harbin Institute of Technology)
ClassificationMultimodality
🎯 What it does: Propose a de-biasing framework named CCD based on causal intervention and counterfactual reasoning for multi-modal fake news detection.
Causal-Debias: Unifying Debiasing in Pretrained Language Models and Fine-tuning via Causal Invariant Learning
Fan Zhou (University of Electronic Science and Technology of China), Ting Zhong (University of Electronic Science and Technology of China)
TransformerSupervised Fine-TuningText
🎯 What it does: This paper proposes the Causal-Debias framework, unifying debiasing tasks with downstream tasks during the fine-tuning stage of pre-trained language models, utilizing causal invariant learning to eliminate non-causal bias while maintaining performance.
Causality-aware Concept Extraction based on Knowledge-guided Prompting
Siyu Yuan (Fudan University), Rui Xie (Meituan)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes a concept extraction framework called KPCE that utilizes knowledge-guided prompting to reduce bias in pre-trained language models during concept extraction tasks.
Causality-Guided Multi-Memory Interaction Network for Multivariate Stock Price Movement Prediction
Di Luo (Renmin University of China), Rui Yan (Renmin University of China)
Recurrent Neural NetworkTransformerTextTime SeriesFinance Related
🎯 What it does: Proposed a Causal Guided Multi-Memory Interaction Network (CMIN), which simultaneously utilizes financial text and causality-enhanced stock correlations in multivariate stock price trend prediction, and achieves information fusion through a multi-memory network with multi-directional interaction;
Causes and Cures for Interference in Multilingual Translation
Uri Shaham (Tel Aviv University), Shruti Bhosale (Meta AI)
TransformerLarge Language ModelText
🎯 What it does: Systematically studied interference phenomena in multilingual machine translation models, evaluated the impact of model size, training data volume, and language pair ratios on interference, and proposed that expanding model capacity and adjusting sampling temperature can significantly alleviate interference and achieve synergistic effects.
CELDA: Leveraging Black-box Language Model as Enhanced Classifier without Labels
Hyunsoo Cho (Seoul National University), Sang-goo Lee (Seoul National University)
ClassificationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposes the CELDA framework, which utilizes black-box language models to generate pseudo-labels, trains a lightweight LDA classifier on unlabeled data after clustering entropy cleaning, to accomplish the text classification task;
CFSum Coarse-to-Fine Contribution Network for Multimodal Summarization
Min Xiao (State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences), Chengqing Zong (State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences)
GenerationRecurrent Neural NetworkTransformerVision Language ModelMultimodality
🎯 What it does: The study introduces a coarse-to-fine contribution network in multimodal summarization, first filtering out useless images and then using word-level and phrase-level supplementary modules to quantify and guide attention.
Chain-of-Skills: A Configurable Model for Open-Domain Question Answering
Kaixin Ma (Carnegie Mellon University), Jianfeng Gao (Microsoft Research)
RetrievalTransformerMixture of ExpertsContrastive LearningTextChain-of-Thought
🎯 What it does: Proposes Chain-of-Skills (COS), a modular retriever capable of learning and combining multiple retrieval skills (single retrieval, query expansion retrieval, entity span prediction, entity linking, re-ranking) to complete evidence collection for open-domain question answering.
Character-Aware Models Improve Visual Text Rendering
Rosanne Liu (Google Research), Noah Constant (Google Research)
GenerationTransformerLarge Language ModelVision Language ModelDiffusion modelImageTextMultimodalityBenchmark
🎯 What it does: This paper systematically studies and enhances the spelling ability of text generation models in visual text rendering by designing two benchmarks, WikiSpell and DrawText.
Characterization of Stigmatizing Language in Medical Records
Keith Harrigian (Johns Hopkins University), Mark Dredze (Johns Hopkins University)
ClassificationTransformerLarge Language ModelTextElectronic Health Records
🎯 What it does: Investigated stigmatizing language in medical records targeting different populations, constructed an annotation task with three categories of labels, and performed large-scale classification.
Characterizing and Measuring Linguistic Dataset Drift
Tyler A. Chang (University of California San Diego), Dan Roth (AWS AI Labs)
Domain AdaptationData-Centric LearningTransformerSupervised Fine-TuningText
🎯 What it does: Proposes a three-dimensional language data drift metric at the lexical, structural, and semantic levels, and uses them to predict NLP model performance in new domains and at the singleton level.
ChatGPT for Zero-shot Dialogue State Tracking: A Solution or an Opportunity?
Michael Heck (Heinrich Heine University Düsseldorf), Milica Gasic
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper explores the performance of ChatGPT in zero-shot dialogue state tracking (DST) and analyzes its strengths and limitations.
CHBias: Bias Evaluation and Mitigation of Chinese Conversational Language Models
Jiaxu Zhao (Eindhoven University of Technology), Mykola Pechenizkiy (University of Technology Sydney)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Constructed the Chinese dialogue model bias assessment and mitigation dataset CHBias, evaluated and mitigated gender, sexual orientation, age, and appearance biases in Chinese pre-trained dialogue models.
CHEER: Centrality-aware High-order Event Reasoning Network for Document-level Event Causality Identification
Meiqi Chen (Peking University), Zhiwei Liu (Meituan)
ClassificationGraph Neural NetworkTransformerTextGraph
🎯 What it does: This paper proposes a document-level event causality recognition model called CHEER based on graph neural networks, which can perform high-level reasoning at the document level and integrate event centrality and coreference information.
CITADEL: Conditional Token Interaction via Dynamic Lexical Routing for Efficient and Effective Multi-Vector Retrieval
Minghan Li (University of Waterloo), Xilun Chen (Meta AI)
RetrievalComputational EfficiencyTransformerContrastive LearningText
🎯 What it does: Propose a conditional token interaction model called CITADEL based on dynamic dictionary routing, significantly reducing the token interaction volume in multi-vector retrieval while maintaining or even improving retrieval accuracy.
CLAPSpeech: Learning Prosody from Text Context with Contrastive Language-Audio Pre-Training
Zhenhui Ye (Zhejiang University), Zhou Zhao (Zhejiang University)
GenerationRepresentation LearningConvolutional Neural NetworkTransformerContrastive LearningTextAudio
🎯 What it does: Proposes CLAPSpeech, a cross-modal contrastive pre-training framework, to explicitly learn prosody differences of the same text word in different contexts, and applies the pre-trained text encoder as a plugin to existing TTS systems.
ClarifyDelphi: Reinforced Clarification Questions with Defeasibility Rewards for Social and Moral Situations
Valentina Pyatkin (Bar-Ilan University), Chandra Bhagavatula (Allen Institute for Artificial Intelligence)
GenerationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Designed and trained the CLARIFYDELPHI system to generate clarifying questions in social and moral scenarios, aiming to obtain missing context and improve moral judgment.
Class based Influence Functions for Error Detection
Thang Nguyen-Duc (FPT Software AI Center), Nghi Bui (FPT Software AI Center)
Anomaly DetectionData-Centric LearningText
🎯 What it does: This paper proposes category-based influence functions (IFs-class) to detect erroneous labels in deep learning datasets.
Class-Incremental Learning based on Label Generation
Yijia Shao (Peking University), Bing Liu (University of Illinois at Chicago)
ClassificationGenerationTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: Under the class-incremental learning (CIL) scenario, this paper proposes transforming tasks into a continuous label generation problem, leveraging the generation capabilities of pre-trained language models to achieve continuous learning without task information.
CLCL: Non-compositional Expression Detection with Contrastive Learning and Curriculum Learning
Jianing Zhou (University of Illinois at Urbana-Champaign), Suma Bhat (University of Illinois at Urbana-Champaign)
ClassificationRecognitionTransformerContrastive LearningText
🎯 What it does: Proposes a framework combining contrastive learning and curriculum learning for non-compositional expression tasks (Chinese idiom recognition and metaphor detection);
Clinical Note Owns its Hierarchy: Multi-Level Hypergraph Neural Networks for Patient-Level Representation Learning
Nayeon Kim (Seoul National University), Sun Kim (Seoul National University)
Representation LearningGraph Neural NetworkTextElectronic Health Records
🎯 What it does: Propose a clinical note representation learning method based on a multi-layer hypergraph neural network, utilizing word, note-level, and classification-level information from patients' clinical notes to construct a multi-level hypergraph and perform hierarchical message passing, ultimately achieving prediction of in-hospital mortality rates for ICU patients.
CMOT: Cross-modal Mixup via Optimal Transport for Speech Translation
Yan Zhou (Key Laboratory of Intelligent Information Processing Institute of Computing Technology Chinese Academy of Sciences), Yang Feng (Key Laboratory of Intelligent Information Processing Institute of Computing Technology Chinese Academy of Sciences)
TransformerTextMultimodalityAudio
🎯 What it does: This study proposes a CMOT method combining cross-modal mixup and optimal transport (OT) for end-to-end speech translation tasks.
CoAD: Automatic Diagnosis through Symptom and Disease Collaborative Generation
Huimin Wang (Tencent), Yefeng Zheng (Tencent)
ClassificationTransformerElectronic Health Records
🎯 What it does: This paper proposes CoAD, an automatic diagnosis framework based on the Transformer decoder, which can collaboratively generate symptom sequences and disease labels to directly realize question-answer style diagnosis.
Code4Struct: Code Generation for Few-Shot Event Structure Prediction
Xingyao Wang (University of Illinois Urbana-Champaign), Heng Ji (University of Illinois Urbana-Champaign)
GenerationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This study proposes transforming the event structure prediction task into a code generation task, using large language models (LLMs) to generate event instances based on Python class definitions, thereby accomplishing event argument extraction.
CodeIE: Large Code Generation Models are Better Few-Shot Information Extractors
Peng Li (Fudan University), Xipeng Qiu (Fudan University)
GenerationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Propose converting named entity recognition and relation extraction tasks into code generation tasks, leveraging large code language models (e.g., Codex) for information extraction in few-shot settings.
COGEN: Abductive Commonsense Language Generation
Rohola Zandie (University of Denver), Mohammad Mahoor
GenerationTransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: Proposed the COGEN model to generate abductive reasoning under incomplete observations, achieving improvements in αNLI and αNLG tasks.
Cognitive Reframing of Negative Thoughts through Human-Language Model Interaction
Ashish Sharma (University of Washington), Tim Althoff (University of Washington)
TransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: Developed a retrieval-enhanced context learning model capable of generating negative thought restructuring sentences that conform to cognitive restructuring attributes.
COLA: Contextualized Commonsense Causal Reasoning from the Causal Inference Perspective
Zhaowei Wang (Hong Kong University of Science and Technology), Simon See (NVIDIA)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Designed a context-aware consensus causal inference task and proposed the COLA framework to address this task.
CoLaDa: A Collaborative Label Denoising Framework for Cross-lingual Named Entity Recognition
Tingting Ma (Harbin Institute of Technology), Chin-Yew Lin (Microsoft)
RecognitionKnowledge DistillationTransformerTextBenchmark
🎯 What it does: Propose a collaborative label denoising framework named CoLaDa for cross-lingual named entity recognition, addressing label noise in both translated data and unlabeled target language data;
ColD Fusion: Collaborative Descent for Distributed Multitask Finetuning
Shachar Don-Yehiya (IBM Research), Leshem Choshen (IBM Research)
OptimizationFederated LearningTransformerSupervised Fine-TuningText
🎯 What it does: By allowing different contributors to fine-tune a shared base model on their own datasets, then uploading the fine-tuned model weights to a central repository, and subsequently fusing them through averaging to obtain a new base model, this process iteratively achieves distributed multi-task fine-tuning and continuous improvement in a closed loop without sharing data.
Cold-Start Data Selection for Better Few-shot Language Model Fine-tuning: A Prompt-based Uncertainty Propagation Approach
Yue Yu (Georgia Institute of Technology), Chao Zhang (Georgia Institute of Technology)
ClassificationData-Centric LearningMeta LearningTransformerPrompt EngineeringContrastive LearningText
🎯 What it does: Propose a sample selection method called PATRON for few-shot fine-tuning of pre-trained language models in a cold-start scenario with no labeled data available at the beginning.
Combo of Thinking and Observing for Outside-Knowledge VQA
Qingyi Si (Institute of Information Engineering, Chinese Academy of Sciences), Weiping Wang (Institute of Information Engineering, Chinese Academy of Sciences)
TransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodalityRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes a framework called TwO, combining a multimodal encoder, text encoder, and answer decoder, leveraging visual features and four types of knowledge (explicit text, explicit multimodal, implicit text, implicit multimodal) to address the OK-VQA task.
Comparative evaluation of boundary-relaxed annotation for Entity Linking performance
Gabriel Herman Bernardim Andrade (Nara Institute of Science and Technology), Eiji Aramaki (Nara Institute of Science and Technology)
RetrievalData-Centric LearningRecurrent Neural NetworkTransformerSupervised Fine-TuningTextBenchmark
🎯 What it does: The study investigates the impact of relaxing entity boundary annotations on the performance of an entity linking system by randomly expanding entity boundaries on the AIDA CoNLL-YAGO dataset to generate noisy data; training is conducted on three models (HPAELDC, LUKE, and a custom VanillaNER+LUKE), and the performance of the NER and ED stages is evaluated on the original test set; average experiments are conducted across different noise ratios, analyzing matching types and vocabulary dependencies.
Composition-contrastive Learning for Sentence Embeddings
Sachin Chanchani (Texas A&M University), Ruihong Huang (Texas A&M University)
Representation LearningTransformerContrastive LearningText
🎯 What it does: Propose a contrastive learning approach based on sentence component combination, constructing positive samples by splitting sentences into left and right phrases while retaining the training framework of SimCSE;
Compositional Data Augmentation for Abstractive Conversation Summarization
Siru Ouyang (University of Illinois Urbana Champaign), Diyi Yang (Stanford University)
GenerationData SynthesisGraph Neural NetworkTransformerLarge Language ModelTextGraphBenchmarkRetrieval-Augmented Generation
🎯 What it does: Developed a compositional data augmentation method (COMPO) based on dialog substructures, which generates diverse and high-quality dialogue-summary pairs by recombining dialogues at the substructure level through topic splitting and action triplets as incremental units.
Compositional Generalization without Trees using Multiset Tagging and Latent Permutations
Matthias Lindemann (University of Edinburgh), Ivan Titov (University of Edinburgh)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Propose a two-stage sequence-to-sequence model: first predict an output multiset for each input word, then arrange the words in order using a differentiable sparse permutation model.
Compounding Geometric Operations for Knowledge Graph Completion
Xiou Ge (University of Southern California), C.-C. Jay Kuo (University of Southern California)
Representation LearningGraphBenchmark
🎯 What it does: Proposes the CompoundE model, which utilizes the combination of three geometric transformations (translation, rotation, scaling) to achieve knowledge graph embedding
Concise Answers to Complex Questions: Summarization of Long-form Answers
Abhilash Potluri (University of Texas at Austin), Eunsol Choi (University of Texas at Austin)
GenerationTransformerLarge Language ModelText
🎯 What it does: Investigated how to compress long-form answers into concise responses, constructed a long-form answer summarization dataset, and conducted user evaluations
CONE: An Efficient COarse-to-fiNE Alignment Framework for Long Video Temporal Grounding
Zhijian Hou (City University of Hong Kong), Nan Duan (Microsoft Research Asia)
RetrievalComputational EfficiencyRepresentation LearningVision Language ModelContrastive LearningVideoMultimodality
🎯 What it does: Propose an efficient COarse-to-fiNE alignment framework (CONE) for the temporal localization task in long videos;
ConFEDE: Contrastive Feature Decomposition for Multimodal Sentiment Analysis
Jiuding Yang (University of Alberta), Yu Xu (Tencent)
ClassificationTransformerContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: Propose the ConFEDE framework, which enhances the representation capability for video multimodal sentiment analysis by jointly training with contrastive feature decomposition and contrastive learning.
Conflicts, Villains, Resolutions: Towards models of Narrative Media Framing
Lea Frermann (University of Melbourne), Gosia Mikolajczak (Australian National University)
ClassificationExplainability and InterpretabilityTransformerText
🎯 What it does: Propose a narrative-based media framework model, and manually annotate 428 English climate change news articles with multi-label frames (conflict, resolution, human interest, morality, economics) and narrative roles (hero, villain, victim) to construct the Narrative Frames Corpus; simultaneously design a Retrieval-based Frame Prediction (RBF) model that combines sentence retrieval with Longformer to achieve multi-label frame prediction.
Conjunct Lengths in English, Dependency Length Minimization, and Dependency Structure of Coordination
Adam Przepiórkowski (ICS Polish Academy of Sciences), Michał Woźniak (University of Warsaw)
Explainability and InterpretabilityComputational EfficiencyText
🎯 What it does: Investigates length differences in English binary coordinate structures, verifying that the left conjunct is often shorter, and explores how this phenomenon is influenced by the governor's (subject) position; further uses the Dependency Length Minimization (DLM) theoretical framework to explain the interpretability of different dependency structure models (Prague, London, etc.) for this phenomenon; simultaneously evaluates the advantages and limitations of various dependency structures in explaining empirical results.
Conjunct Resolution in the Face of Verbal Omissions
Royi Rassin (Bar-Ilan University), Reut Tsarfaty (Bar-Ilan University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes a unified task for resolving verbal omission in parallel structures, adopting a text-to-text split rewriting approach to convert sentences with omissions into a set of independent complete sentences.
Connective Prediction for Implicit Discourse Relation Recognition via Knowledge Distillation
Hongyi Wu (East China Normal University), Yadong Zhang (East China Normal University)
RecognitionKnowledge DistillationTransformerPrompt EngineeringText
🎯 What it does: Proposes a framework for conjunction prediction through knowledge distillation to improve the accuracy of implicit discourse relation recognition.
Considerations for meaningful sign language machine translation based on glosses
Mathias Müller (University of Zurich), Sarah Ebling (University of Zurich)
TransformerVideoReview/Survey Paper
🎯 What it does: This paper provides a review of gloss-based sign language machine translation research over the past five years, evaluating methods, datasets, evaluation metrics, and limitations, while offering improvement suggestions.
Consistency Regularization Training for Compositional Generalization
Yongjing Yin (Zhejiang University), Yue Zhang (Tencent Inc)
Representation LearningTransformerContrastive LearningTextBenchmark
🎯 What it does: Proposed a consistency regularization training method to enhance the compositional generalization ability of Transformers, particularly in semantic parsing and machine translation tasks.
Consistent Prototype Learning for Few-Shot Continual Relation Extraction
Xiudi Chen (Xiamen University), Xiaodong Shi (Xiamen University)
ClassificationTransformerPrompt EngineeringText
🎯 What it does: Propose the N-way K-shot continual relation extraction (NK-CRE) task and design the ConPL method to alleviate catastrophic forgetting and similar class confusion under few-shot conditions.
Constrained Tuple Extraction with Interaction-Aware Network
Xiaojun Xue (Beijing Institute of Technology), Zhendong Niu (Beijing Institute of Technology)
Graph Neural NetworkTransformerLarge Language ModelText
🎯 What it does: Proposes the Constrained Triplet Extraction (CTE) task and realizes the extraction of knowledge triplets with spatiotemporal and conditional constraints from text through an Interaction-Aware Network (IAN).
Context-Aware Transformer Pre-Training for Answer Sentence Selection
Luca Di Liello (University of Trento), Alessandro Moschitti (Amazon Alexa AI)
RetrievalRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This work designs three pre-training objectives (SSP) that align with answer sentence selection (AS2), conducts continuous pre-training on RoBERTa and ELECTRA, and then fine-tunes on multiple public and industrial AS2 datasets, ultimately enhancing the model's ability to leverage contextual information.
Contextual Distortion Reveals Constituency: Masked Language Models are Implicit Parsers
Jiaxi Li (Singapore University of Technology and Design), Wei Lu (Singapore University of Technology and Design)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Propose a parameter-free update, unsupervised syntactic parsing method based on masked language models. By applying three linguistic perturbations (substitution, decontextualization, moving) to sentence spans and calculating the distortion of context representations, the compositional nature is evaluated, followed by obtaining syntactic trees through chart parsing.
Contextual Knowledge Learning for Dialogue Generation
Wen Zheng (University of Nottingham), Ke Zhou (University of Nottingham)
GenerationTransformerLarge Language ModelText
🎯 What it does: Propose the Contextual Knowledge Learning (CKL) model to simultaneously perform fine-grained weighting of context and external knowledge in dialogue generation, thereby improving the quality of generated responses.
Continual Contrastive Finetuning Improves Low-Resource Relation Extraction
Wenxuan Zhou (University Of Southern California), Hoifung Poon (Microsoft Research)
TransformerSupervised Fine-TuningContrastive LearningTextBiomedical Data
🎯 What it does: This paper proposes a continuous contrastive learning fine-tuning framework to enhance document-level relation extraction (RE) performance under low-resource conditions.