ACL 2023 Papers — Page 4
Annual Meeting of the Association for Computational Linguistics · 1074 papers
Do You Hear The People Sing? Key Point Analysis via Iterative Clustering and Abstractive Summarisation
Hao Li (University of Manchester), Goran Nenadic (University of Manchester)
GenerationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposes a two-step generative keypoint analysis framework, first generating argument clusters using neural topic modeling and iterative clustering, then generating keypoints with pre-trained language models, and designing a set-level evaluation tool based on semantic similarity.
DOC: Improving Long Story Coherence With Detailed Outline Control
Kevin Yang (UC Berkeley), Yuandong Tian (Meta AI)
GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Proposed the Detailed Outline Control (DOC) framework, which enhances the coherence and interest of thousand-character-level long-form stories through detailed outline planning and fine-grained control mechanisms.
Document-Level Event Argument Extraction With a Chain Reasoning Paradigm
Jian Liu (Beijing Jiaotong University), Zhe Zhao (Tencent AI Lab)
TransformerTextChain-of-Thought
🎯 What it does: Propose a document-level event parameter extraction method based on chain reasoning, automatically generating first-order logic rules and achieving end-to-end learning through T-Norm fuzzy logic;
Document-Level Multi-Event Extraction with Event Proxy Nodes and Hausdorff Distance Minimization
Xinyu Wang (University of Warwick), Yulan He (King's College London)
Graph Neural NetworkTransformerLarge Language ModelTextFinance Related
🎯 What it does: ProCNet proposes using event proxy nodes and minimizing Hausdorff distance to directly learn multi-event extraction at the document level globally, freeing itself from traditional entity relationship modeling and step-by-step decoding processes.
Does GPT-3 Grasp Metaphors? Identifying Metaphor Mappings with Generative Language Models
Lennart Wachowiak (King's College London), Dagmar Gromann (University of Vienna)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This paper investigates whether GPT-3 can directly predict the source domain of metaphors based on sentences and target domains, proposing an evaluation method based on few-shot prompting and fine-tuning.
Don’t Forget Your ABC’s: Evaluating the State-of-the-Art in Chat-Oriented Dialogue Systems
Sarah E. Finch (Emory University), Jinho D. Choi (Emory University)
Explainability and InterpretabilityComputational EfficiencyTextBenchmark
🎯 What it does: This paper proposes a dialogue evaluation method based on binary behavioral labels (ABC-Eval), and systematically validates it across four common evaluation approaches (dialogue-level Likert, turn-level Likert, comparative evaluation, and ABC-Eval). Ultimately, eight efficient and reliable behavioral metrics are selected to measure the dialogue quality of open-domain chatbots.
Don’t Generate, Discriminate: A Proposal for Grounding Language Models to Real-World Environments
Yu Gu (Ohio State University), Yu Su (Ohio State University)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningAgentic AITextGraphBenchmark
🎯 What it does: Propose the Pangu framework, transforming language models from generators to discriminators, combining symbolic agents for controllable plan generation in real-world environments such as knowledge base question answering.
Don’t Parse, Choose Spans! Continuous and Discontinuous Constituency Parsing via Autoregressive Span Selection
Songlin Yang (ShanghaiTech University), Kewei Tu (ShanghaiTech University)
Computational EfficiencyRepresentation LearningRecurrent Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningTextSequential
🎯 What it does: Propose a unified autoregressive span selection method applicable to both continuous and discontinuous phrase structure analysis;
Don’t Retrain, Just Rewrite: Countering Adversarial Perturbations by Rewriting Text
Ashim Gupta (University of Utah), Vivek Srikumar (University of Utah)
ClassificationAdversarial AttackTransformerLarge Language ModelText
🎯 What it does: Propose ATINTER—a standalone text rewriter designed to intercept and rewrite adversarial inputs, rendering them harmless to downstream text classifiers without requiring retraining of the classifiers.
Double-Branch Multi-Attention based Graph Neural Network for Knowledge Graph Completion
Hongcai Xu (Xi'an Jiaotong University), Wenbo Liu (Xi'an Jiaotong University)
Representation LearningGraph Neural NetworkTransformerGraph
🎯 What it does: This paper proposes a dual-branch multi-attention graph neural network (MA-GNN) for knowledge graph completion, focusing on integrating global and local structural information.
Downstream Datasets Make Surprisingly Good Pretraining Corpora
Kundan Krishna (Carnegie Mellon University), Zachary Lipton (Carnegie Mellon University)
ClassificationTransformerSupervised Fine-TuningText
🎯 What it does: This paper proposes and systematically evaluates the 'self-pretraining' method, which uses data from downstream tasks for self-supervised pretraining and then fine-tunes on the same dataset, investigating whether the benefits of pretraining mainly come from the pretraining objective rather than massive external corpora;
DrBERT: A Robust Pre-trained Model in French for Biomedical and Clinical domains
Yanis Labrak (Avignon Université), Pierre-Antoine Gourraud (Nantes Université)
TransformerLarge Language ModelSupervised Fine-TuningBiomedical Data
🎯 What it does: This paper proposes DrBERT, a French medical domain pre-trained model based on RoBERTa, and publicly releases a large-scale medical corpus called NACHOS;
DSEE: Dually Sparsity-embedded Efficient Tuning of Pre-trained Language Models
Xuxi Chen (University of Texas at Austin), Yu Cheng (Microsoft Corporation)
Computational EfficiencyKnowledge DistillationTransformerText
🎯 What it does: Proposes the DSEE framework, combining sparse low-rank updates and sparse masks to achieve dual improvements in parameters and inference efficiency while maintaining performance.
DSRM: Boost Textual Adversarial Training with Distribution Shift Risk Minimization
SongYang Gao, Ying Shan (Tencent)
Adversarial AttackTransformerText
🎯 What it does: Proposes a text adversarial training framework based on distribution shift risk minimization (DSRM), which estimates and optimizes adversarial loss using only clean data, eliminating the need to generate adversarial examples.
DT-Solver: Automated Theorem Proving with Dynamic-Tree Sampling Guided by Proof-level Value Function
Haiming Wang (Sun Yat-sen University), Xiaodan Liang (Huawei Noah's Ark Lab)
AI Code AssistantTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Proposed DT-Solver, an automated theorem proving framework that utilizes dynamic tree sampling combined with a proof hierarchy value function.
Dual Cache for Long Document Neural Coreference Resolution
Qipeng Guo (Amazon AWS AI), Zheng Zhang (Fudan University)
Computational EfficiencyRepresentation LearningTransformerLarge Language ModelTextBenchmark
🎯 What it does: Propose a Dual Cache structure for coreference parsing in long documents, addressing the cache loss problem caused by traditional LRU cache during topic transitions.
Dual Class Knowledge Propagation Network for Multi-label Few-shot Intent Detection
Feng Zhang (Peking University), Tengjiao Wang (Peking University)
ClassificationMeta LearningGraph Neural NetworkTransformerText
🎯 What it does: This paper proposes a Dual Class Knowledge Propagation Network (DCKPN) to address the multi-label few-shot intent detection problem.
Dual-Alignment Pre-training for Cross-lingual Sentence Embedding
Ziheng Li (Peking University), Qi Zhang (Microsoft Corporation)
RetrievalRepresentation LearningTransformerContrastive LearningText
🎯 What it does: Propose a dual alignment pre-training framework (DAP), introducing simultaneously a sentence-level translation ranking task and a token-based representation translation learning (RTL) task on dual encoders to achieve better cross-sentence embeddings;
DualGATs: Dual Graph Attention Networks for Emotion Recognition in Conversations
Duzhen Zhang (Baidu Inc), Xiuyi Chen (Baidu Inc)
RecognitionGraph Neural NetworkLarge Language ModelTextGraph
🎯 What it does: This paper proposes the DualGATs model, which uses DisGAT to capture discourse hierarchical structures, SpkGAT to capture speaker dependencies, and combines the two types of information through mutual attention interaction;
DuNST: Dual Noisy Self Training for Semi-Supervised Controllable Text Generation
Yuxi Feng (University of British Columbia), Xing Xie (University of British Columbia)
GenerationTransformerLarge Language ModelSupervised Fine-TuningAuto EncoderText
🎯 What it does: Propose the DuNST framework, which integrates self-training with bidirectional variational learning, simultaneously generating pseudo text and pseudo labels in semi-supervised controllable text generation. Introduce two flexible noises: high-temperature softened softmax and soft pseudo text, enhancing the model's ability to explore and control the attribute space.
Dynamic and Efficient Inference for Text Generation via BERT Family
Xiaobo Liang (Soochow University), Min Zhang (Soochow University)
GenerationTransformerScore-based ModelText
🎯 What it does: Utilize BERT family models for non-autoregressive text generation, achieving efficient inference through joint training of a single-step CTC generator and Levenshtein editor.
Dynamic Heterogeneous-Graph Reasoning with Language Models and Knowledge Representation Learning for Commonsense Question Answering
Yujie Wang (Shanxi University), Ru Li (Shanxi University)
Graph Neural NetworkTransformerLarge Language ModelTextGraphBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposed a dynamic heterogeneous graph reasoning method (DHLK) based on language models and knowledge representation learning, which constructs a heterogeneous knowledge graph containing concept entities and dictionary synonyms, and achieves deep integration during encoding and reasoning phases;
Dynamic Regularization in UDA for Transformers in Multimodal Classification
Ivonne Monter-Aldana (Mathematics Research Center), Fernando Sanchez-Vega
ClassificationTransformerMultimodality
🎯 What it does: This paper proposes a two-stream Transformer model (Multimodal BERT-ViT, MMBV), which achieves low-cost fusion between text and image encoders through the CLS token, and extends Unsupervised Data Augmentation (UDA) to a dynamic supervised version, automatically adjusting the consistency loss coefficient λ to balance the model's generalization and specialization, thereby better leveraging weak modal information in multimodal text-image classification tasks.
Dynamic Routing Transformer Network for Multimodal Sarcasm Detection
Yuan Tian (Institute of Automation, Chinese Academy of Sciences), Wenji Mao (Institute of Automation, Chinese Academy of Sciences)
ClassificationTransformerLarge Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a dynamic routing Transformer network, DynRT-Net, to capture cross-modal contradictions between images and text, thereby achieving multimodal sarcasm detection.
Dynamic Transformers Provide a False Sense of Efficiency
Yiming Chen (National University Of Singapore), Haizhou Li (National University Of Singapore)
Adversarial AttackTransformerTextBenchmark
🎯 What it does: Proposed SAME, an efficiency attack framework for multi-exit Transformers, which can force the model to delay early exit by adding minimal perturbations to the input, thus significantly reducing the model's inference efficiency.
Early Discovery of Disappearing Entities in Microblogs
Satoshi Akasaki (Yahoo Japan Corporation), Masashi Toyoda (Institute of Industrial Science, University of Tokyo)
RecognitionTransformerSupervised Fine-TuningText
🎯 What it does: Studies how to early detect disappearing entities from microblogs such as Weibo/Twitter, and constructs large-scale training and testing datasets based on this, training named entity recognition models capable of identifying and classifying these disappearing entities.
Easy Guided Decoding in Providing Suggestions for Interactive Machine Translation
Ke Wang (Alibaba Group Inc), Yu Zhao (Alibaba Group Inc)
GenerationTransformerLarge Language ModelText
🎯 What it does: Proposed an unsupervised decoding algorithm called Prefix-Suffix Guided Decoding (PSGD) for the translation suggestion task in computer-assisted translation, which generates alternative translations for text segments with marked errors under prefix and suffix constraints.
EEL: Efficiently Encoding Lattices for Reranking
Prasann Singhal (University of Texas at Austin), Greg Durrett (University of Texas at Austin)
GenerationComputational EfficiencyTransformerText
🎯 What it does: Efficiently encode a directed acyclic graph (lattice) of candidate sets in text generation tasks using Transformer, and quickly extract the optimal answer from it by leveraging the token-factored reranker (TFR).
Effective Contrastive Weighting for Dense Query Expansion
Xiao Wang (University of Glasgow), Iadh Ounis (University of Glasgow)
RetrievalTransformerContrastive LearningText
🎯 What it does: Designed and trained the CWPRF (Contrastive Weighting for PRF) model, which assigns importance weights to tokens in pseudo-relevance feedback using contrastive learning, achieving efficient and more accurate query expansion within dense retrieval frameworks such as ColBERT.
Efficient Diagnosis Assignment Using Unstructured Clinical Notes
Louis Blankemeier (Stanford University), Akshay Chaudhari (Stanford University)
ClassificationTransformerSupervised Fine-TuningTextBiomedical DataElectronic Health Records
🎯 What it does: Developed the HyDE framework, combining annotation functions and refined PubMedBERT for electronic phenotyping of clinical notes.
Efficient Semiring-Weighted Earley Parsing
Andreas Opedal (ETH Zürich), Jason Eisner (Johns Hopkins University)
Computational EfficiencyText
🎯 What it does: This paper provides a systematic derivation description of the Earley algorithm and proposes multiple acceleration schemes based on it, supporting semiring weighting and prefix weight computation, while representing the grammar as a single weighted finite state automaton (WFSA) to further share computations.
Efficient Shapley Values Estimation by Amortization for Text Classification
Chenghao Yang (Unversity of Chicago), Bing Xiang (AWS AI Labs)
ClassificationExplainability and InterpretabilityTransformerSupervised Fine-TuningText
🎯 What it does: Propose an amortized model that directly predicts the Shapley value of each token without requiring additional perturbation samples, avoiding the high computational cost of traditional methods that require extensive evaluations on large models; during the training phase, stable reference Shapley values are first generated using SVS-25, followed by training a BERT-based linear layer model; during inference, explanations can be obtained with a single forward pass; meanwhile, a local adaptation method is proposed to improve the approximation quality.
Efficient Transformers with Dynamic Token Pooling
Piotr Nawrot (University of Edinburgh), Edoardo Maria Ponti
Computational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Propose a dynamic pooling Transformer that jointly learns autoregressive language modeling and variable-length token segmentation, reducing intermediate layer sequence length to improve memory and time efficiency.
Elaboration-Generating Commonsense Question Answering at Scale
Wenya Wang (University of Washington), Noah A. Smith (University of Washington)
GenerationComputational EfficiencyKnowledge DistillationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes an alternately trainable ELABOR framework that leverages a small language model to generate and utilize 'Elaboration' (background explanation) to enhance commonsense question-answering performance.
Element-aware Summarization with Large Language Models: Expert-aligned Evaluation and Chain-of-Thought Method
Yiming Wang (Shanghai Jiao Tong University), Rui Wang (Shanghai Jiao Tong University)
GenerationTransformerLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Constructed an expert-constructed news element-aware test set and evaluated LLMs and traditional pre-trained models on it; proposed a step-by-step summarization generation method based on chain-of-thought (SumCoT).
Ellipsis-Dependent Reasoning: a New Challenge for Large Language Models
Daniel Hardt (Copenhagen Business School)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Introduces a new challenge called 'ellipsis-dependent reasoning,' using paired sentences with and without ellipsis, and testing large language models' understanding and reasoning ability through yes/no questions.
ELQA: A Corpus of Metalinguistic Questions and Answers about English
Shabnam Behzad (Georgetown University), Amir Zeldes (Georgetown University)
ClassificationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: This paper constructs the ELQA dataset, collecting and organizing over 70,000 metalinguistic questions and answers about English from the English Language & Usage and English Language Learners forums, and proposes a closed-book free-form question answering task based on this dataset;
EM Pre-training for Multi-party Dialogue Response Generation
Yiyang Li (Shanghai Jiao Tong University), Hai Zhao (Shanghai Jiao Tong University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningTextSequential
🎯 What it does: Pre-train multi-party dialogue response generation using the EM framework, first generating implicit recipient labels via the expectation step, then training the response generation model through the maximization step; no recipient annotations are required during pre-training, but real annotations can be used in subsequent fine-tuning.
Empowering Cross-lingual Behavioral Testing of NLP Models with Typological Features
Ester Hlavnova (Google Research), Sebastian Ruder (Google Research)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Proposed the Multi-lingual Morphology Checklist (M2C) framework for generating behavioral tests tailored to the morphological features of different languages and assessing the generalization ability of large-scale language models.
End-to-end Knowledge Retrieval with Multi-modal Queries
Man Luo (Arizona State University), Chitta Baral (Arizona State University)
RetrievalTransformerVision Language ModelContrastive LearningMultimodalityBenchmark
🎯 What it does: Proposed an end-to-end visual-language retrieval model called ReViz, which can directly use mixed image and text queries to retrieve knowledge; simultaneously constructed a new dataset named ReMuQ for evaluating such tasks; and designed a weakly supervised pre-training task called VL-ICT, leveraging the WiT dataset for multi-modal retrieval pre-training.
Enhancing Dialogue Generation via Dynamic Graph Knowledge Aggregation
Chen Tang (University of Surrey), Frank Guerin (University of Surrey)
GenerationGraph Neural NetworkTransformerLarge Language ModelTextGraphRetrieval-Augmented Generation
🎯 What it does: Proposed a dialogue generation framework based on dynamic graph knowledge aggregation (SaBART), which constructs pseudo nodes to dynamically generate subgraphs and aggregates knowledge at multiple levels, achieving seamless integration of language models with knowledge graphs.
Enhancing Event Causality Identification with Counterfactual Reasoning
Feiteng Mu (Hong Kong Polytechnic University), Wenjie Li (Hong Kong Polytechnic University)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: This paper introduces counterfactual reasoning into Event Causality Identification (ECI), estimating and eliminating two types of biases by using either event-masking context or event pairs exclusively, thereby achieving bias-free reasoning.
Enhancing Grammatical Error Correction Systems with Explanations
Yuejiao Fei (Westlake University), Shuming Shi (Tencent)
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: Constructed an explainable grammar error correction dataset named EXPECT, which includes annotations of evidence words and error types, and proposed multiple models based on tags, interaction, and generation for explainable GEC.
Enhancing Language Representation with Constructional Information for Natural Language Understanding
Lvxiaowei Xu (Zhejiang University), Tianxiang Wang (Zhejiang University)
Representation LearningGraph Neural NetworkTransformerSupervised Fine-TuningText
🎯 What it does: Propose a HyCxG framework that combines basic construction syntax with a relation hypergraph attention network, enhancing the representation capability of pre-trained language models (PLMs) through construction information and applying it to natural language understanding tasks.
Enhancing Personalized Dialogue Generation with Contrastive Latent Variables: Combining Sparse and Dense Persona
Yihong Tang, Yuexian Hou (China Mobile Communication Group Tianjin Co Ltd)
GenerationTransformerAuto EncoderContrastive LearningText
🎯 What it does: This paper proposes a dialogue generation model named CLV, which integrates sparse (e.g., gender, age) and dense (text description) personality information. It uses a self-separation algorithm to implicitly cluster dense text into sparse categories in the hidden space, then employs a conditional variational autoencoder (CVAE) and contrastive learning to encode dialogues, queries, and personality information. A Decider selects the most suitable latent variables, which are then fed into the GPT-2 generator to produce more consistent, coherent, and diverse personalized responses.
Entailment as Robust Self-Learner
Jiaxin Ge (Peking University), James Glass (Massachusetts Institute of Technology)
ClassificationTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper utilizes a text entailment model for pre-training and achieves zero-shot adaptation by converting different NLU tasks into entailment forms (supposition). Subsequently, the authors propose a SimPLE algorithm based on pseudo-label self-training, which generates multi-labels using dropout, filters uncertain samples based on proximity similarity, and re-labels through majority voting to enhance self-training quality.
Entity Tracking in Language Models
Najoung Kim (Boston University), Sebastian Schuster (Saarland University)
Object TrackingTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This paper designs a box state tracking task to investigate the ability of large language models (LLMs) to track changes in entity states within natural language text.
Environmental Claim Detection
Dominik Stammbach (ETH Zurich), Markus Leippold (University of Zurich)
ClassificationTransformerLarge Language ModelText
🎯 What it does: Proposed and implemented the environmental declaration detection task, released an expert-annotated dataset and model.
Envisioning Future from the Past: Hierarchical Duality Learning for Multi-Turn Dialogue Generation
Ang Lv (Gaoling School of Artificial Intelligence, Renmin University of China), Rui Yan (Wangxuan Institute of Computer Technology, Peking University)
GenerationTransformerText
🎯 What it does: Propose a hierarchical bidirectional learning framework (HDLD) to generate multi-turn dialogues, fully utilizing the bidirectional relationship between past and future dialogues (duality).
EPIC: Multi-Perspective Annotation of a Corpus of Irony
Simona Frenda (University of Turin), Davide Bernardi (Amazon)
ClassificationData-Centric LearningTransformerSupervised Fine-TuningTextBenchmark
🎯 What it does: Create EPIC, a multi-perspective irony annotated corpus containing social media dialogues from five English variants, annotated by binary classification from annotators in the corresponding countries.
ESCOXLM-R: Multilingual Taxonomy-driven Pre-training for the Job Market Domain
Mike Zhang (IT University of Copenhagen), Barbara Plank (IT University of Copenhagen)
Domain AdaptationRepresentation LearningTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Proposed and trained a multilingual domain-adaptive pre-trained model ESCOXLM-R based on XLM-R large, utilizing the European Skills and Occupations Taxonomy (ESCO) for domain pre-training and introducing a new relation prediction task;
Estimating the Uncertainty in Emotion Attributes using Deep Evidential Regression
Wen Wu (University of Cambridge), Philip Woodland (University of Cambridge)
RecognitionTransformerAudio
🎯 What it does: Propose the DEER method, which employs deep evidence regression to estimate the uncertainty of emotional attributes
Ethical Considerations for Machine Translation of Indigenous Languages: Giving a Voice to the Speakers
Manuel Mager (AWS AI Labs), Ngoc Thang Vu (University of Stuttgart)
TextReview/Survey Paper
🎯 What it does: This paper explores ethical issues in machine translation for Indigenous languages through a literature review and interviews and questionnaires with Indigenous language speakers;
ETHICIST: Targeted Training Data Extraction Through Loss Smoothed Soft Prompting and Calibrated Confidence Estimation
Zhexin Zhang (Tsinghua University), Minlie Huang (Tsinghua University)
Data-Centric LearningTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Propose the ETHICIST method, which achieves targeted training data extraction through soft prompts and loss smoothing techniques, and effectively activates the model's memory.
Evaluate AMR Graph Similarity via Self-supervised Learning
Ziyi Shou (Hong Kong University of Science and Technology), Fangzhen Lin (Hong Kong University of Science and Technology)
Representation LearningGraph Neural NetworkTransformerContrastive LearningTextGraph
🎯 What it does: Proposed a self-supervised learning-based AMR graph similarity metric called AMRSim, which can automatically evaluate the similarity of AMR graphs and replace traditional matching algorithms.
Evaluating Open-Domain Dialogues in Latent Space with Next Sentence Prediction and Mutual Information
Kun Zhao (Wuhan University), Xiaohui Cui (University of Sheffield)
GenerationTransformerAuto EncoderContrastive LearningText
🎯 What it does: This paper proposes an automatic evaluation metric called CMN, based on CVAEs, Next Sentence Prediction, and mutual information, for assessing the quality of open-domain dialogue generation.
Evaluating Open-Domain Question Answering in the Era of Large Language Models
Ehsan Kamalloo (University of Alberta), Davood Rafiei (University of Alberta)
TransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Conduct a systematic study on evaluation methods for large language models in open-domain question answering, highlighting that traditional word-matching evaluations significantly underestimate model performance.
Evaluating Paraphrastic Robustness in Textual Entailment Models
Dhruv Verma (Stony Brook University), Adam Poliak (Bryn Mawr College)
ClassificationTransformerLarge Language ModelTextBenchmark
🎯 What it does: Constructed 1,126 pairs of RTE examples and their rewritten versions (PaRTE), using them to evaluate the robustness of existing RTE models against sentence rewriting.
Evaluating pragmatic abilities of image captioners on A3DS
Polina Tsvilodub (University of Tübingen), Michael Franke (University of Tübingen)
GenerationData SynthesisConvolutional Neural NetworkRecurrent Neural NetworkSupervised Fine-TuningReinforcement LearningVision Language ModelContrastive LearningImageTextBenchmark
🎯 What it does: Proposed the Annotated 3D Shapes (A3DS) dataset and used it to evaluate the pragmatic performance of image captioners in reference tasks, comparing captioners fine-tuned with multi-agent reinforcement learning to baseline models.
Evaluating Zero-Shot Event Structures: Recommendations for Automatic Content Extraction (ACE) Annotations
Erica Cai (University of Massachusetts Amherst), Brendan O’Connor
Data-Centric LearningTextBenchmark
🎯 What it does: This paper investigates the issues present when evaluating zero-shot (zero-shot) event extraction methods using the ACE corpus, and proposes three improvement suggestions (coreference matching, dual head selection, and modality/event subset evaluation)
Event Extraction as Question Generation and Answering
Di Lu (Dataminr Inc), Alejandro Jaimes (Dataminr Inc)
GenerationTransformerLarge Language ModelText
🎯 What it does: Propose a context-aware event argument extraction framework QGA-EE based on question generation and answer answering, which directly generates context-aware questions and answers using a seq2seq model.
Explainable Recommendation with Personalized Review Retrieval and Aspect Learning
Hao Cheng (Shenzhen University), Hao Liao (Shenzhen University)
Recommendation SystemExplainability and InterpretabilityTransformerTextRetrieval-Augmented Generation
🎯 What it does: Proposed a retrieval-augmented and aspect-based explainable recommendation model (ERRA) that simultaneously predicts ratings and generates personalized explanations.
Explaining How Transformers Use Context to Build Predictions
Javier Ferrando (TALP Research Center, Universitat Politècnica de Catalunya), Marta R. Costa-jussà (Meta AI)
Explainability and InterpretabilityTransformerText
🎯 What it does: Propose an interpretability method combining residual flow and attention decomposition to analyze the contribution of context in Transformer language generation models when generating each word.
ExplainMeetSum: A Dataset for Explainable Meeting Summarization Aligned with Human Intent
Hyun Kim (Electronics and Telecommunications Research Institute), Seung-Hoon Na (Jeonbuk National University)
GenerationExplainability and InterpretabilityTextBenchmark
🎯 What it does: Constructed the ExplainMeetSum dataset and proposed a multi-extractor guided summarization model called Multi-DYLE, along with an interpretable evidence extraction task named E3
Explanation-based Finetuning Makes Models More Robust to Spurious Cues
Josh Magnus Ludan (University of Pennsylvania), Chris Callison-Burch (University of Pennsylvania)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes an 'explanation-based fine-tuning' method that, during the fine-tuning of large language models, simultaneously prompts the model to generate free-text explanations supporting its predictions, thereby reducing the model's reliance on irrelevant features (pseudo-relevance) in the training set.
Explicit Syntactic Guidance for Neural Text Generation
Yafu Li (Zhejiang University), Yue Zhang (Westlake University)
GenerationTransformerText
🎯 What it does: Proposed a syntax-based generation framework that recursively generates text from the root node of a syntax tree, splitting the generation process into two steps: filling in syntax context and syntax expansion.
Exploiting Biased Models to De-bias Text: A Gender-Fair Rewriting Model
Chantal Amrhein (University of Zurich), Samuel Läubli (University of Zurich)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes using existing gender bias models to generate biased text, then training a neural rewrite model to achieve gender fair language generation.
Exploring and Verbalizing Academic Ideas by Concept Co-occurrence
Yi Xu (Shanghai Jiao Tong University), Chenghu Zhou (IGSNRR Chinese Academy Of Sciences)
GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Construct evolutionary concept co-occurrence graphs and co-occurrence citation quintuples datasets, and design a temporal link prediction based on masked language models and idea generation based on pre-trained language models, integrated into a real-time research assistant system.
Exploring Better Text Image Translation with Multimodal Codebook
Zhibin Lan (Xiamen University), Jinsong Su (Xiamen University)
Image TranslationTransformerVision Language ModelAuto EncoderContrastive LearningImageTextMultimodality
🎯 What it does: Propose a novel text-image translation (TIT) method and release the first publicly available Chinese-English text-image translation dataset OCRMT30K;
Exploring Continual Learning for Code Generation Models
Prateek Yadav (University of North Carolina Chapel Hill), Bing Xiang (AWS AI Labs)
GenerationAI Code AssistantTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes a continuous learning benchmark called CODETASK-CL for code generation and systematically evaluates the performance of various continuous learning methods on this benchmark.
Exploring How Generative Adversarial Networks Learn Phonological Representations
Jingyi Chen (Ohio State University), Micha Elsner (Ohio State University)
Explainability and InterpretabilityRepresentation LearningConvolutional Neural NetworkGenerative Adversarial NetworkAudio
🎯 What it does: This paper uses a generative adversarial network (ciwGAN) to study the acoustic and phonetic representations of nasal sounds in English and French, examining the correspondence between hidden variables and nasal features.
Exploring Large Language Models for Classical Philology
Frederick Riemenschneider (Heidelberg University), Anette Frank (Heidelberg University)
ClassificationRecognitionTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Constructed and evaluated four pre-trained language models for classical languages (Ancient Greek and Latin), including monolingual and multilingual versions of RoBERTa and T5, and conducted systematic benchmarking on tasks such as part-of-speech tagging, dependency syntax, and lemmatization.
Exploring Lottery Prompts for Pre-trained Language Models
Yulin Chen (Tsinghua University), Pengjun Xie (Alibaba Group)
ClassificationTransformerPrompt EngineeringText
🎯 What it does: Investigated and verified that in pre-trained language models (PLM), nearly every instance has a 'lottery prompt' that enables the model to make correct predictions, and explored the search for these prompts, discovery of strong prompts, and generalization to unseen data through parameter-free ensemble methods.
Exploring the Capacity of Pretrained Language Models for Reasoning about Actions and Change
Weinan He (Sun Yat-sen University), Yongmei Liu (Sun Yat-sen University)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Proposed and implemented TRAC, a text-based action and change reasoning benchmark, encompassing four tasks: projection, executability, plan verification, and goal recognition;
Exploring the Impact of Layer Normalization for Zero-shot Neural Machine Translation
Zhuoyuan Mao (Kyoto University), Sadao Kurohashi (Kyoto University)
GenerationTransformerText
🎯 What it does: Investigated the impact of LayerNorm placement in Transformers on zero-shot translation performance
Extractive is not Faithful: An Investigation of Broad Unfaithfulness Problems in Extractive Summarization
Shiyue Zhang (University of North Carolina Chapel Hill), Mohit Bansal (University of North Carolina Chapel Hill)
GenerationTransformerLarge Language ModelTextBenchmark
🎯 What it does: Analyzed and systematically distilled five common types of unfaithfulness in extractive summarization, constructed a dataset with 1,600 human-annotated examples, and proposed a new evaluation metric called EXTEVAL.
Extrinsic Evaluation of Machine Translation Metrics
Nikita Moghe (University of Edinburgh), Alexandra Birch (University of Edinburgh)
TextReview/Survey PaperBenchmark
🎯 What it does: Evaluate the practicality of MT metrics at the sentence level by comparing sentences generated by machine translation (MT) with corresponding target language sentences and linking them to the success rates of downstream tasks (semantic parsing, question answering, dialogue state tracking); measure the ability of metrics to predict incorrect translations by constructing a 'fault detection' benchmark.
f-Divergence Minimization for Sequence-Level Knowledge Distillation
Yuqiao Wen (University of Alberta), Lili Mou (University of Hong Kong)
Knowledge DistillationTransformerText
🎯 What it does: proposes the f-DISTILL framework using f-divergence for sequence-level knowledge distillation, and presents four variants (KL, RKL, JS, TVD)
FAA: Fine-grained Attention Alignment for Cascade Document Ranking
Zhen Li (Peking University), Daxin Jiang (Microsoft Corporation)
RetrievalKnowledge DistillationRepresentation LearningTransformerText
🎯 What it does: Propose FAA (Fine-grained Attention Alignment), a jointly optimized cascaded document retrieval model that enhances document ranking performance by using a cross-encoder to perform fine-grained attention alignment on selected paragraphs and fusing paragraph relevance scores.
Facilitating Fine-grained Detection of Chinese Toxic Language: Hierarchical Taxonomy, Resources, and Benchmarks
Junyu Lu (Dalian University of Technology), Hongfei Lin (Dalian University of Technology)
ClassificationRecognitionRecurrent Neural NetworkTransformerSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper constructs a hierarchical toxic language classification framework called MONITOR TOXIC FRAME. Based on this framework, a fine-grained Chinese toxic language dataset named TOXICN was annotated and generated, and a multi-class insult dictionary containing implicit slurs was constructed. Subsequently, the TKE (Toxic Knowledge Enhancement) method was proposed to integrate dictionary knowledge into models.
Facilitating Multi-turn Emotional Support Conversation with Positive Emotion Elicitation: A Reinforcement Learning Approach
Jinfeng Zhou (Tsinghua University), Minlie Huang (Tsinghua University)
TransformerLarge Language ModelReinforcement LearningMixture of ExpertsText
🎯 What it does: Propose a novel paradigm for positive emotion activation in multi-turn emotional support dialogues, and design the SUPPORTER model to achieve dynamic emotion regulation and dialogue coherence;
Fact-Checking Complex Claims with Program-Guided Reasoning
Liangming Pan (University of California Santa Barbara), Preslav Nakov (MBZUAI)
RetrievalExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed the Program-Guided Fact-Checking (PROGRAMFC) framework, which leverages large language models to first generate interpretable reasoning programs, then sequentially invokes specialized subtask functions to accomplish multi-step evidence retrieval and fact verification;
FACTIFY-5WQA: 5W Aspect-based Fact Verification through Question Answering
Anku Rani (University of South Carolina), Amitava Das (University of South Carolina)
Explainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: Propose a fact-checking framework named FACTIFY‑5WQA based on the 5W (who, what, when, where, why) question-answer (QA) format, and construct a large-scale dataset containing 391,041 factual claims and their corresponding 5W QA pairs.
FactKG: Fact Verification via Reasoning on Knowledge Graphs
Jiho Kim (Korea Advanced Institute Of Science And Technology), Edward Choi (Korea Advanced Institute Of Science And Technology)
TransformerLarge Language ModelTextGraphBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Constructed a knowledge graph fact verification dataset named FACTKG and provided a baseline model based on graph reasoning.
Factual or Contextual? Disentangling Error Types in Entity Description Generation
Navita Goyal (University of Maryland), Hal Daumé III (University of Maryland)
GenerationTransformerLarge Language ModelText
🎯 What it does: Investigate the distinction between factual errors and context-related errors in entity description generation tasks, and propose an automated evaluation framework
Factually Consistent Summarization via Reinforcement Learning with Textual Entailment Feedback
Paul Roit (Bar Ilan University), Idan Szpektor (Google Research)
GenerationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Fine-tune pre-trained abstractive summarization models using reinforcement learning combined with natural language inference (NLI) rewards to improve factual consistency in generated summaries.
FairPrism: Evaluating Fairness-Related Harms in Text Generation
Eve Fleisig (University of California Berkeley), Hanna Wallach (Microsoft Research)
GenerationTransformerLarge Language ModelText
🎯 What it does: Proposed the FairPrism dataset, systematically collected and meticulously annotated 5,000 AI-generated texts for fairness harms related to gender and sexual orientation, covering stereotypes and degrading harms as well as special risks in interactive contexts;
Faithful Low-Resource Data-to-Text Generation through Cycle Training
Zhuoer Wang (Texas A&M University), Oleg Rokhlenko (Amazon)
GenerationTransformerSupervised Fine-TuningTextTabular
🎯 What it does: Utilizing cycle training in low-resource scenarios, bidirectional models for data-to-text and text-to-data are trained, significantly enhancing the consistency and credibility of the generated text with the input structured data.
Faithful Question Answering with Monte-Carlo Planning
Ruixin Hong (Tsinghua University), Changshui Zhang (Tsinghua University)
TransformerLarge Language ModelTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the FAME method to achieve trustworthy question answering based on Monte Carlo planning, generating reliable reasoning trees and providing answers through an interactive controller and modular reasoning environment.
Faithfulness Tests for Natural Language Explanations
Pepa Atanasova (University of Copenhagen), Isabelle Augenstein (University of Copenhagen)
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: Developed two testing methods to evaluate the credibility of natural language explanations: the counterfactual input editor and the input reconstruction test.
Faking Fake News for Real Fake News Detection: Propaganda-Loaded Training Data Generation
Kung-Hsiang Huang (University Of Illinois Urbana Champaign), Heng Ji (University Of Illinois Urbana Champaign)
Data SynthesisAnomaly DetectionTransformerLarge Language ModelReinforcement LearningPrompt EngineeringTextBenchmark
🎯 What it does: Automatically generate realistic fake news training samples by replacing key sentences in real news and incorporating propaganda techniques.
Fantastic Expressions and Where to Find Them: Chinese Simile Generation with Multiple Constraints
Kexin Yang (Alibaba Group), Jun Xie (Alibaba Group)
GenerationTransformerLarge Language ModelSupervised Fine-TuningTextRetrieval-Augmented Generation
🎯 What it does: Propose a controlled anthropomorphic generation task (CSG) and construct a large-scale, fine-grained annotated Chinese metaphor corpus named GraCe, while designing the Similor model to achieve controllable generation.
FC-KBQA: A Fine-to-Coarse Composition Framework for Knowledge Base Question Answering
Lingxi Zhang (Renmin University of China), Juanzi Li (Renmin University of China)
TransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: By detecting fine-grained components and applying intermediate-grained constraints, then generating executable logical expressions via seq2seq, the problem of generalization and executability in KBQA is addressed.
Federated Learning for Semantic Parsing: Task Formulation, Evaluation Setup, New Algorithms
Tianshu Zhang (Ohio State University), Huan Sun (Ohio State University)
Federated LearningTransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper proposes applying federated learning to semantic parsing tasks, designs an evaluation framework based on real heterogeneous data, and introduces the Lorar reweighting mechanism based on training loss decline;
FEDLEGAL: The First Real-World Federated Learning Benchmark for Legal NLP
Zhuo Zhang (Harbin Institute of Technology), Zenglin Xu (Harbin Institute of Technology)
Federated LearningTransformerTextBenchmark
🎯 What it does: Proposed FEDLEGAL, the first real-world legal NLP federated learning benchmark, covering five NLP tasks and a privacy attack task.
FERMAT: An Alternative to Accuracy for Numerical Reasoning
Jasivan Sivakumar, Nafise Sadat Moosavi (University of Sheffield)
TextBenchmark
🎯 What it does: This paper proposes FERMAT, a multi-perspective evaluation dataset for fine-grained assessment of language models' numerical reasoning capabilities.
Few-shot Adaptation Works with UnpredicTable Data
Jun Shern Chan (New York University), Ethan Perez (New York University)
Domain AdaptationMeta LearningTransformerLarge Language ModelSupervised Fine-TuningTabular
🎯 What it does: This paper automatically extracted internet tables to generate 413,299 few-shot tasks and fine-tuned the GPT-2 model on this data, improving its few-shot learning performance on 52 downstream NLP tasks.
Few-Shot Data-to-Text Generation via Unified Representation and Multi-Source Learning
Alexander Hanbo Li (AWS AI Labs), Bing Xiang (AWS AI Labs)
GenerationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextGraphTabular
🎯 What it does: Propose a unified representation method that maps tables, knowledge graph triples, and semantic representations into a virtual table, which is then linearized to build a general data-to-text model.
Few-Shot Document-Level Event Argument Extraction
Xianjun Yang (University of California Santa Barbara), Linda Petzold (University of California Santa Barbara)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Constructed a document-level few-shot event argument extraction benchmark (FewDocAE) and provided baseline models and evaluations.
Few-shot Event Detection: An Empirical Study and a Unified View
Yubo Ma, Aixin Sun (Nanyang Technological University)
ClassificationMeta LearningPrompt EngineeringText
🎯 What it does: This paper conducts a comprehensive empirical study on few-shot event detection (ED) methods, proposing a unified perspective and a better baseline model.