Annual Meeting of the Association for Computational Linguistics Β· 356 papers
Mirror: Multiple-perspective Self-Reflection Method for Knowledge-rich Reasoning
Hanqi Yan (King's College London), Yulan He (King's College London)
CodeTransformerLarge Language ModelPrompt EngineeringTextBenchmark
π― What it does: Proposed a multi-perspective self-reflection method called Mirror, which helps large language models self-improve in knowledge-intensive reasoning tasks;
Mobile-Bench: An Evaluation Benchmark for LLM-based Mobile Agents
Shihan Deng (University of Electronic Science and Technology of China), Shuo Shang (University of Electronic Science and Technology of China)
CodeTransformerLarge Language ModelAgentic AITextBenchmark
π― What it does: Constructed Mobile-Bench, a mobile LLM agent evaluation platform supporting dual interaction modes of UI and API, and provided an evaluation dataset containing 832 tasks (SAST, SAMT, MAMT) and 103 callable APIs, further proposing the CheckPoint evaluation metric for fine-grained inspection of agent execution processes.
π― What it does: Developed MobileSpeech, a zero-shot text-to-speech (TTS) system deployable on mobile devices, leveraging parallel mask generation and fine-grained speaker prompting to achieve low-latency, high-quality synthesis.
π― What it does: Propose the MoΓ»sai text-to-music diffusion model, which can generate high-quality 48kHz stereo music lasting multiple minutes based on text descriptions.
MPCoder: Multi-user Personalized Code Generator with Explicit and Implicit Style Representation Learning
Zhenlong Dai (Zhejiang University), Jingyuan Chen (Zhejiang University)
CodeGenerationRepresentation LearningAI Code AssistantTransformerLarge Language ModelContrastive LearningText
π― What it does: This paper proposes a multi-user personalized code generation framework called MPCODER, which can generate code that aligns with individual developers' coding habits and personal styles;
MULFE: A Multi-Level Benchmark for Free Text Model Editing
Chenhao Wang (Chinese Academy of Sciences), Jun Zhao (Chinese Academy of Sciences)
CodeComputational EfficiencyKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
π― What it does: Proposed a multi-level free-text model editing benchmark (MULFE), systematically evaluated various editing methods on this benchmark, and introduced a simple and efficient method called SIDE based on context distillation.
Multi-Level Feedback Generation with Large Language Models for Empowering Novice Peer Counselors
Alicja Chaszczewicz (Stanford University), Diyi Yang (Stanford University)
CodeGenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
π― What it does: Developed a multi-layered feedback framework based on LLM to generate contextualized and structured feedback for novice peer counselors.
Multimodal ArXiv: A Dataset for Improving Scientific Comprehension of Large Vision-Language Models
Lei Li (University of Hong Kong), Qi Liu (University of Hong Kong)
CodeLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
π― What it does: Constructed the Multimodal ArXiv dataset, including ArXivCap and ArXivQA, to enhance the understanding capabilities of large vision-language models in scientific literature.
Multimodal Contextualized Semantic Parsing from Speech
Jordan Voas (University of Texas at Austin), Ray Mooney (University of Texas at Austin)
CodeExplainability and InterpretabilityRepresentation LearningData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelContrastive LearningImageMultimodalityBenchmarkAudio
π― What it does: Proposed the SPICE task for semantic parsing in multimodal context environments, constructed the VG-SPICE dataset based on vision and speech, and developed the AViD-SP model capable of dynamically updating knowledge graphs.
CodeComputational EfficiencyRepresentation LearningTransformerSupervised Fine-TuningMixture of ExpertsVision Language ModelMultimodality
π― What it does: Propose the Conditional Mixture of LoRA (MixLoRA) framework for multi-modal instruction tuning, dynamically constructing low-rank adaptation matrices to alleviate task interference.
π― What it does: This paper proposes a multi-modal Transformer framework based on prompt learning, which can handle missing modalities in sentiment analysis and emotion recognition tasks;
MultiPICo: Multilingual Perspectivist Irony Corpus
Silvia Casola (University of Turin), Davide Bernardi (Amazon)
CodeData SynthesisTransformerLarge Language ModelPrompt EngineeringTextBenchmark
π― What it does: Constructed and released MultiPICo, a multilingual, multi-perspective irony corpus containing 18,778 short dialogues from Twitter and Reddit.
Multistage Collaborative Knowledge Distillation from a Large Language Model for Semi-Supervised Sequence Generation
Jiachen Zhao (University Of Massachusetts Amherst), Andrew McCallum (Ibm Research Ai)
CodeGenerationKnowledge DistillationTransformerLarge Language ModelText
π― What it does: Propose a multi-stage collaborative knowledge distillation method (MCKD), which generates pseudo labels using a small amount of labeled data and a few examples prompted by a large language model (LLM). Subsequently, through cross-partition and multi-round distillation, the student model is gradually improved and ultimately achieves high performance on low-resource sequence generation tasks.
Navigate through Enigmatic Labyrinth A Survey of Chain of Thought Reasoning: Advances, Frontiers and Future
Zheng Chu (Harbin Institute of Technology), Ting Liu (Harbin Institute of Technology)
CodeLarge Language ModelPrompt EngineeringTextMultimodalityReview/Survey PaperBenchmarkChain-of-Thought
π― What it does: A systematic review of Chain-of-Thought (CoT) and its extensions (XoT), summarizing related methods, benchmarks, frontiers, and future research directions.
Navigating the Shadows: Unveiling Effective Disturbances for Modern AI Content Detectors
Ying Zhou (University of Chinese Academy of Sciences), Le Sun (University of Chinese Academy of Sciences)
CodeClassificationAnomaly DetectionAdversarial AttackTransformerLarge Language ModelText
π― What it does: Systematically evaluate the detection of ChatGPT-generated text and design 12 black-box perturbation methods to explore the robustness of detectors
Never Lost in the Middle: Mastering Long-Context Question Answering with Position-Agnostic Decompositional Training
Junqing He (International Digital Economy Academy), Jiaxing Zhang (International Digital Economy Academy)
CodeGenerationRetrievalTransformerLarge Language ModelSupervised Fine-TuningTextRetrieval-Augmented GenerationChain-of-Thought
π― What it does: To address the 'middle failure' issue of large language models in long-text multi-document question answering, this paper proposes a position-agnostic multi-step QA (PAM QA) training method. By decomposing tasks into question repetition, index prediction, and answer summarization, the method enhances the model's attention focusing and information retrieval capabilities in long contexts.
Pragya Srivastava (Microsoft Research), Amit Sharma (Microsoft Research)
CodeClassificationGenerationOptimizationTransformerLarge Language ModelPrompt EngineeringText
π― What it does: The study investigates the interaction between task instruction refinement and in-context example (ICE) optimization in large language models, proposing a metric called NICE to measure the importance of ICE, and systematically evaluates the impact of different tasks and instructions on model performance.
NounAtlas: Filling the Gap in Nominal Semantic Role Labeling
Roberto Navigli (Sapienza University of Rome), Alessandro Scirè (Sapienza University of Rome)
CodeTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringContrastive LearningText
π― What it does: This paper constructs NounAtlas, a large-scale noun predicate lexicon, and generates the first silver-annotated noun semantic role labeling (SRL) dataset based on it. Training and validation on this dataset demonstrate a unified SRL method for both nouns and verbs.
OceanGPT: A Large Language Model for Ocean Science Tasks
Zhen Bi (Zhejiang University), Huajun Chen (Zhejiang University)
CodeRobotic IntelligenceTransformerLarge Language ModelAgentic AIPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
π― What it does: We constructed the first pre-trained language model for ocean science, OceanGPT, and generated ocean domain instruction data through the multi-agent collaborative DOINSTRUCT framework, followed by creating the OCEANBENCH benchmark evaluation.
Timothy Ossowski (University of Wisconsin), Junjie Hu (University of Wisconsin)
CodeDomain AdaptationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelImageTextBiomedical DataRetrieval-Augmented Generation
π― What it does: Propose the OLIVE method, which directly injects object-level visual embeddings into large language models, supporting object-level reasoning, zero-shot transfer, and multi-image context prompting.
π― What it does: Systematically investigate the impact of calibration data on the post-training quantization and pruning compression effects of large language models.
π― What it does: This paper formalizes chain-of-thought (CoT) within a probabilistic framework, elucidating the representational capabilities of CoT language models (LMs), and proves that CoT-trained RNNs and Transformers are weakly equivalent to probabilistic Turing machines (PTMs) at different precision levels, thereby demonstrating their theoretical Turing completeness.
CodeTransformerLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
π― What it does: Created the SUMMEVAL-OP evaluation dataset and proposed OP-I-PROMPT and OP-PROMPTS for multi-dimensional opinion summary evaluation.
One-Shot Learning as Instruction Data Prospector for Large Language Models
Yunshui Li (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences), Yongbin Li (Alibaba Group)
CodeData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
π― What it does: Propose the NUGGETS method, which filters the most valuable instruction set for subsequent instruction tuning without requiring additional annotations by evaluating a golden score for each instruction example through a single round of in-context learning from large language models.
Order-Agnostic Data Augmentation for Few-Shot Named Entity Recognition
Huiming Wang (Singapore University of Technology and Design), Lidong Bing (DAMO Academy, Alibaba Group)
CodeRecognitionTransformerLarge Language ModelPrompt EngineeringText
π― What it does: Propose the Order-Agnostic Data Augmentation (OADA) method to enhance the performance of few-shot named entity recognition (NER) models.
PAGED: A Benchmark for Procedural Graphs Extraction from Documents
Weihong Du (Sichuan University), Wenqiang Lei (Sichuan University)
CodeData SynthesisTransformerLarge Language ModelPrompt EngineeringTextGraphBenchmark
π― What it does: Propose the PAGED benchmark, construct a large-scale high-quality program graph-document pair dataset, systematically evaluate existing methods, and explore the potential of LLMs in program graph extraction.
π― What it does: Propose the PairCFR framework, which combines contrastive learning and cross-entropy during training of Counterfactually Augmented Data (CAD), pairing original samples with their adversarial counterparts to enhance the model's learning of global features.
Parrot: Enhancing Multi-Turn Instruction Following for Large Language Models
Yuchong Sun (Renmin University of China), Kun Gai (Renmin University of China)
CodeOptimizationData-Centric LearningReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmark
π― What it does: Enhance the instruction-following capability of large language models in multi-round interactions through the Parrot framework, with training and optimization focused on human-like multi-turn dialogues;
Picturing Ambiguity: A Visual Twist on the Winograd Schema Challenge
Brendan Park (Brock University), Ali Emami (Brock University)
CodeGenerationData SynthesisExplainability and InterpretabilityTransformerPrompt EngineeringVision Language ModelDiffusion modelImageTextMultimodalityBenchmark
π― What it does: Propose the WINOVIS dataset and evaluation framework for testing the pronoun disambiguation capabilities of text-to-image models in multimodal contexts
π― What it does: Propose the PITA model, which leverages generative prompt tuning through dynamic prompt templates and task interaction graphs to jointly decode three subtasks: argument component type identification, relation recognition, and relation type classification.
IΓ±igo Alonso (University of Basque Country), Mirella Lapata (University of Edinburgh)
CodeGenerationTransformerVision Language ModelImageTextTabular
π― What it does: Propose PixT3, a pixel-level table-to-text generation model that treats tables as images and is pre-trained using a self-supervised structural learning curve, applicable to open-ended, controlled, and loosely controlled generation scenarios.
Planning Like Human: A Dual-process Framework for Dialogue Planning
Tao He (Harbin Institute of Technology), Bing Qin (Harbin Institute of Technology)
CodeReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
π― What it does: Designed and implemented a dialogue planning framework based on dual-process theory (DPDP), combining a fast policy language model with a slow Monte Carlo Tree Search (MCTS), and proposed a non-parametric switching gate and two-phase training (offline reinforcement learning + MCTS-guided self-play).
PokeMQA: Programmable knowledge editing for Multi-hop Question Answering
Hengrui Gu (Jilin University), Xin Wang (Jilin University)
CodeTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
π― What it does: Proposes the PokeMQA framework, which improves multi-hop question answering models using a programmable knowledge editing method, decoupling problem decomposition from knowledge conflict detection, and enhancing reasoning quality through external memory and knowledge prompting.
Pouring Your Heart Out: Investigating the Role of Figurative Language in Online Expressions of Empathy
Gyeongeun Lee (University of Illinois at Chicago), Natalie Parde (University of Illinois at Chicago)
CodeClassificationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
π― What it does: This paper constructs the AcnEmpathize dataset for acne forum empathetic expressions and investigates the role of metaphorical language, such as metaphors, idioms, and hyperbole, in online empathy detection.
Probing Language Models for Pre-training Data Detection
Zhenhua Liu (Soochow University), Wenliang Chen (Soochow University)
CodeData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
π― What it does: This paper proposes an internal activation analysis method based on detection technology to determine whether a given text is included in the pre-training data of large language models; meanwhile, a new academic-level dataset named ArxivMIA is constructed as a more challenging benchmark for pre-training data detection.
π― What it does: Propose the Progressive Modality Freezing (PMF) method for multi-modal entity alignment, progressively freezing irrelevant modal features and fusing useful features.
π― What it does: Propose the Prompt Expansion framework, which generates diverse and high-quality images using expanded text prompts, significantly reducing the need for users to refine prompts.
ProtT3: Protein-to-Text Generation for Text-based Protein Understanding
Zhiyuan Liu (National University of Singapore), Tat-Seng Chua (National University of Singapore)
CodeGenerationRetrievalRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningTextMultimodalityBiomedical DataRetrieval-Augmented Generation
π― What it does: Propose the ProtT3 framework, which integrates protein language models and text language models through a cross-modal projector to accomplish protein sequence-to-text generation and retrieval tasks.
CodeAdversarial AttackLarge Language ModelAgentic AIText
π― What it does: This paper proposes the PsySafe framework, which attacks, evaluates, and defends multi-agent systems from a psychological perspective to enhance their security.
π― What it does: Integrating quality labels during training of Neural Machine Translation (NMT) models enables the model to self-assess translation quality and leverage this information during decoding to improve translation quality
CodeComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
π― What it does: This paper proposes the Quantization-Side Tuning (QST) framework, which first quantizes the weights of large language models to 4 bits, and then introduces an independent side network for task-specific fine-tuning through low-rank adapters and a gradient-free downsampling module, updating only the side network parameters to achieve significant memory compression and acceleration without significant performance loss.
RAID: A Shared Benchmark for Robust Evaluation of Machine-Generated Text Detectors
Liam Dugan (University of Pennsylvania), Chris Callison-Burch (University of Pennsylvania)
CodeAnomaly DetectionAdversarial AttackLarge Language ModelTextBenchmark
π― What it does: Developed RAID, the largest and most challenging shared benchmark dataset for evaluating machine-generated text detectors, and systematically evaluated 12 open-source, metric-based, and commercial detectors on this dataset.
RAM-EHR: Retrieval Augmentation Meets Clinical Predictions on Electronic Health Records
Ran Xu (Emory University), Carl Yang (Emory University)
CodeRetrievalGraph Neural NetworkTransformerLarge Language ModelBiomedical DataElectronic Health RecordsRetrieval-Augmented Generation
π― What it does: Propose RAM-EHR, a framework for EHR prediction that utilizes multi-source knowledge retrieval enhancement and consistency regularization.
RAVEL: Evaluating Interpretability Methods on Disentangling Language Model Representations
Jing Huang (Stanford University), Atticus Geiger (Pr(Ai) R Group 2)
CodeExplainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelAuto EncoderTextBenchmark
π― What it does: Developed and released the RAVEL benchmark to evaluate the disentanglement capability of explanation methods for entity attributes in language models, and conducted experimental comparisons of multiple methods (including MDAS) on Llama2-7B.
RDRec: Rationale Distillation for LLM-based Recommendation
Xinfeng Wang (University of Yamanashi), Fumiyo Fukumoto (University of Yamanashi)
CodeRecommendation SystemKnowledge DistillationTransformerLarge Language ModelTextChain-of-Thought
π― What it does: Propose the RDRec model, which uses a large language model to perform chain-of-thought reasoning on user reviews to extract user preferences and product attributes, and then trains a small model using these rationales for top-N and sequential recommendations.
Re3: A Holistic Framework and Dataset for Modeling Collaborative Document Revision
Qian Ruan (Technical University of Darmstadt), Iryna Gurevych (Technical University of Darmstadt)
CodeGraph Neural NetworkTransformerLarge Language ModelText
π― What it does: Proposed the Re3 framework and constructed the Re3-Sci dataset, systematically analyzing the collaborative peer review, revision, and response processes in academic papers.
π― What it does: In open-domain question answering tasks, the authors propose REANO, which enhances answer quality by incorporating a knowledge graph generation module into a retrieval-augmented reader.
Reasoning in Conversation: Solving Subjective Tasks through Dialogue Simulation for Large Language Models
Xiaolong Wang (Tsinghua University), Yang Liu (Tsinghua University)
CodeClassificationRecognitionExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
π― What it does: Propose a reasoning method based on dialogue simulation (RiC), which enhances the performance of large language models on subjective tasks through keyword extraction, dialogue generation, and dialogue-enhanced reasoning.
RecGPT: Generative Pre-training for Text-based Recommendation
Hoang Ngo (VinAI Research), Dat Quoc Nguyen (VinAI Research)
CodeRecommendation SystemTransformerLarge Language ModelSupervised Fine-TuningTextSequential
π― What it does: This study trains a domain-adapted large language model for text recommendation, named RecGPT-7B, and further obtains RecGPT-7B-Instruct through instruction fine-tuning. It also publicly releases the pre-training and fine-tune datasets to verify its performance on rating prediction and sequential recommendation tasks.
CodeExplainability and InterpretabilityTransformerLarge Language ModelAgentic AIPrompt EngineeringTextChain-of-Thought
π― What it does: Propose a multi-model, multi-agent 'roundtable meeting' framework called RECONCILE, which enhances reasoning ability through multi-round discussions among LLMs from different families
π― What it does: This paper proposes Reflect-RL, a two-player online reinforcement learning fine-tuning framework that first performs supervised fine-tuning (SFT) on language models, then conducts reinforcement learning (RLFT) in interactive environments, enabling models to achieve self-reflection and action in multi-round decision-making tasks.
CodeOptimizationReinforcement Learning from Human FeedbackTransformerSupervised Fine-TuningReinforcement LearningTextChain-of-Thought
π― What it does: Propose the ReFT (Reinforced Fine-Tuning) method, which first uses supervised fine-tuning (SFT) as a warm-up, then employs PPO reinforcement learning to sample multiple Chain-of-Thought (CoT) paths on the same training set for self-learning, thereby enhancing the performance of large language models on mathematical reasoning tasks.
RelayAttention for Efficient Large Language Model Serving with Long System Prompts
Lei Zhu (City University of Hong Kong), Rynson Lau (City University of Hong Kong)
CodeComputational EfficiencyTransformerText
π― What it does: To address computational bottlenecks caused by long system prompts in large language model services, the RelayAttention scheme is proposed to achieve efficient attention computation without additional training.
π― What it does: Proposes RepCodec, an end-to-end speech representation compression model that maps speech waveforms to low-bitrate semantic discrete codebooks for speech processing in large language models.
Respond in my Language: Mitigating Language Inconsistency in Response Generation based on Large Language Models
Liang Zhang (Renmin University of China), Furu Wei (Microsoft Research Asia)
CodeGenerationTransformerLarge Language ModelSupervised Fine-TuningText
π― What it does: This paper addresses the problem of language inconsistency caused by monolingual instruction fine-tuning in large language models. It proposes using pseudo-inconsistency penalty (PIP) during training to suppress the model's preference for English responses, and enhances multilingual instruction-following consistency during inference by combining prior-enhanced decoding (PED) with the language prior of the base model.
π― What it does: Proposes a Query-Guided Compressor (QGC) that compresses documents in the long context of LLM inputs by query guidance, preserving key information while supporting high compression ratios;
Rethinking Task-Oriented Dialogue Systems: From Complex Modularity to Zero-Shot Autonomous Agent
Heng-Da Xu (Beijing Institute of Technology), Heyan Huang (Beijing Institute of Technology)
CodeTransformerLarge Language ModelAgentic AIText
π― What it does: Proposed AutoTOD, a fully zero-shot autonomous task-oriented dialogue agent that directly utilizes instruction-following large language models (LLMs) to complete dialogues, invoke APIs, and generate responses without training any modules.
Retrieval Augmented Fact Verification by Synthesizing Contrastive Arguments
Zhenrui Yue (University of Illinois Urbana Champaign), Dong Wang (University of Illinois Urbana Champaign)
CodeClassificationRetrievalTransformerLarge Language ModelPrompt EngineeringContrastive LearningTextRetrieval-Augmented Generation
π― What it does: Proposed a retrieval-augmented fact verification framework called RAFTS, which determines the truthfulness of claims by retrieving relevant documents, constructing contrastive arguments, and prompting with a few examples.
Revealing the Parametric Knowledge of Language Models: A Unified Framework for Attribution Methods
Haeun Yu (University of Copenhagen), Isabelle Augenstein (University of Copenhagen)
CodeExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningText
π― What it does: Proposed a unified evaluation framework to compare instance attribution (IA) with neuron attribution (NA), and designed NA-Instances and IA-Neurons methods to align the two attribution results, assessing their ability to reveal model parameter knowledge.
Revisiting Code Similarity Evaluation with Abstract Syntax Tree Edit Distance
Yewei Song (University of Luxembourg), Jacques Klein (University of Luxembourg)
CodeLarge Language ModelTextBenchmark
π― What it does: Re-evaluate code similarity evaluation metrics, extend Tree Edit Distance (TSED) to multiple programming languages, and compare it with structural similarity generated by GPT-4, BLEU, Jaccard, and other traditional metrics.
Reward-based Input Construction for Cross-document Relation Extraction
Byeonghu Na (KAIST), Il-chul Moon
CodeRecurrent Neural NetworkTransformerLarge Language ModelReinforcement LearningTextBenchmark
π― What it does: Propose a sentence selection module REIC based on reinforcement learning to select sentences most helpful for relation reasoning in cross-document relation extraction tasks, thereby effectively extracting information and constructing inputs from long documents;
π― What it does: Developed a robust singing voice note transcription model, ROSVOT, for automatically annotating SVS data and enhancing synthesis quality.
Zhengping Jiang (Johns Hopkins University), Anqi Liu (Johns Hopkins University)
CodeExplainability and InterpretabilityText
π― What it does: A robust free-text reasoning explanation evaluation method called RORA is studied, aiming to assess whether explanatory text truly provides new information rather than leaking labels.
π― What it does: Propose the S2GSL framework, which combines dual branches of segmented semantic graphs and syntactic latent graphs through adaptive fusion to enhance ABSA performance.
Safety Alignment in NLP Tasks: Weakly Aligned Summarization as an In-Context Attack
Yu Fu (University of California, Riverside), Yue Dong (University of California, Riverside)
CodeAdversarial AttackReinforcement Learning from Human FeedbackTransformerLarge Language ModelPrompt EngineeringTextBenchmark
π― What it does: Investigate the safety alignment of large language models (LLMs) across different natural language processing (NLP) tasks, and propose context attacks leveraging tasks with weaker safety alignment (e.g., summarization), leading to decreased safety in other tasks.
CodeSafty and PrivacyLarge Language ModelPrompt EngineeringTextBenchmark
π― What it does: Constructed SafetyBench, a bilingual multiple-choice safety evaluation benchmark containing 11,435 questions across 7 categories of safety issues.
Same Task, More Tokens: the Impact of Input Length on the Reasoning Performance of Large Language Models
Mosh Levy (Bar-Ilan University), Yoav Goldberg (Bar-Ilan University)
CodeTransformerTextChain-of-Thought
π― What it does: Investigate the impact of input length on the reasoning performance of large language models (LLMs), and construct the scalable FLenQA QA reasoning dataset for systematic evaluation.
SBAAM! Eliminating Transcript Dependency in Automatic Subtitling
Marco Gaido (Fondazione Bruno Kessler), Luisa Bentivogli (Fondazione Bruno Kessler)
CodeGenerationTransformerVideoTextMultimodality
π― What it does: Proposed the first fully end-to-end, transcription-free automatic caption generation model that directly outputs translation, segmentation, and timestamps.
SciMON: Scientific Inspiration Machines Optimized for Novelty
Qingyun Wang (University of Illinois at Urbana-Champaign), Tom Hope (Allen Institute for Artificial Intelligence)
CodeGenerationTransformerLarge Language ModelContrastive LearningTextGraphRetrieval-Augmented Generation
π― What it does: Propose the SCIMON framework to achieve literature-based scientific direction generation, taking contextual input and outputting novel natural language suggestions
Search-Adaptor: Embedding Customization for Information Retrieval
Jinsung Yoon (Google Cloud AI), Tomas Pfister (Google Cloud AI)
CodeRetrievalRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningTextMultimodalityBenchmark
π― What it does: Propose the Search-Adaptor method, which performs low-cost, controllable fine-tuning on text or multimodal embeddings of pre-trained LLMs to improve information retrieval performance.
SEER: Facilitating Structured Reasoning and Explanation via Reinforcement Learning
Guoxin Chen (Institute of Computing Technology, Chinese Academy of Sciences), Yiming Qian (Shanghai Artificial Intelligence Laboratory)
CodeExplainability and InterpretabilityReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningTextGraph
π― What it does: Propose the SEER method, which combines reinforcement learning with structured rewards to achieve high-quality tree/graph structure reasoning and explanation generation.
Self-Augmented In-Context Learning for Unsupervised Word Translation
Yaoyiran Li (University of Cambridge), Ivan VuliΔ (University of Cambridge)
CodeRepresentation LearningTransformerLarge Language ModelPrompt EngineeringText
π― What it does: Propose a self-incremental context learning-based unsupervised bilingual lexicon induction method called SAIL, which leverages LLMs to generate high-confidence word pairs and iteratively improves translation quality through repeated iterations.
Self-chats from Large Language Models Make Small Emotional Support Chatbot Better
Zhonghua Zheng (Harbin Institute of Technology), Liqiang Nie (Harbin Institute of Technology)
CodeKnowledge DistillationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
π― What it does: Utilize large language models (LLMs) as 'tutors' to iteratively expand the emotional support dialogue dataset (ExTES), and employ Diverse Response Inpainting (DRI) to generate rich response examples for fine-tuning the small ChatPal model, thereby enhancing the performance of small models in emotional support tasks.
π― What it does: Propose the Self-Distillation Fine-Tuning (SDFT) method, which bridges the gap between task data and the original model distribution during large language model fine-tuning by leveraging 'distilled' data generated by the model itself, thereby mitigating catastrophic forgetting.
π― What it does: Propose a semi-supervised spoken language to gloss translation framework called S3LG, which utilizes a large amount of monolingual data to generate pseudo labels, iteratively expanding the training data and improving model performance.
CodeExplainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelText
π― What it does: Explored how to improve uncertainty quantification in free-text LLMs by focusing on semantic relevance, and proposed the Shift Attention to Relevance (SAR) method.
Shoulders of Giants: A Look at the Degree and Utility of Openness in NLP Research
Surangika Ranathunga (Massey University), Aloka Fernando (University of Moratuwa)
CodeData-Centric LearningText
π― What it does: This paper quantifies the openness of the NLP community and the benefits brought by openness by semi-automatically extracting and manually annotating NLP papers from the ACL Anthology between 2015-2023. It systematically evaluates the reuse and publication status of research outcomes (data, code, pre-trained models), and analyzes the impact of different publication scenarios (Main, LREC, Other) and language resource levels on open practices.
Sign Language Translation with Sentence Embedding Supervision
Yasser Hamidullah (German Research Center for Artificial Intelligence), Cristina EspaΓ±a-Bonet (German Research Center for Artificial Intelligence)
CodeGenerationRepresentation LearningConvolutional Neural NetworkTransformerVision Language ModelVideoTextMultimodality
π― What it does: Propose an end-to-end sign language translation model that uses an unlabeled gloss as an intermediate representation, leveraging sentence embeddings (SEM) as a supervisory signal to achieve video-to-text translation;
π― What it does: Propose and implement a temporal knowledge graph embedding model named TCompoundE, which uses combined translation and scaling geometric operations to capture entity, relation, and temporal information to complete knowledge graph completion tasks.
π― What it does: Inject structured inductive bias into seq2seq models by first pretraining Transformers to simulate input/output relationships of finite state transducers (FST), then fine-tuning with tunable prefixes on downstream tasks.
Yao Yao (Shanghai Jiao Tong University), Hai Zhao (Shanghai Jiao Tong University)
CodeLarge Language ModelText
π― What it does: Propose SirLLM, which utilizes token entropy to filter key tokens in the KV cache, maintaining long-term memory in infinite-length dialogues without requiring fine-tuning;
SOTOPIA-Ο: Interactive Learning of Socially Intelligent Language Agents
Ruiyi Wang (Carnegie Mellon University), Hao Zhu (Carnegie Mellon University)
CodeReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningAgentic AITextBenchmark
π― What it does: Through the interactive learning method SOTOPIA-Ο, GPT-4 is utilized to automatically generate diverse social tasks and score the model's performance on these tasks. Subsequently, the model's social intelligence and safety are enhanced through two-stage training involving behavioral cloning and self-reinforcement.
Soul-Mix: Enhancing Multimodal Machine Translation with Manifold Mixup
Xuxin Cheng (Peking University), Yuexian Zou (Peking University)
CodeConvolutional Neural NetworkTransformerVision Language ModelTextMultimodality
π― What it does: Propose the Soul-Mix framework, which utilizes manifold mixup to blend the translation results of complete and missing texts, thereby enhancing the performance of multimodal machine translation.
SpaRC and SpaRP: Spatial Reasoning Characterization and Path Generation for Understanding Spatial Reasoning Capability of Large Language Models
Md Imbesat Rizvi (Technical University of Darmstadt), Iryna Gurevych (Technical University of Darmstadt)
CodeTransformerLarge Language ModelSupervised Fine-TuningTextBenchmarkChain-of-Thought
π― What it does: This paper studies the spatial reasoning capabilities of large language models (LLMs) and proposes the SpaRC spatial reasoning characterization framework and the SpaRP reasoning path dataset based on symbolic reasoning;
Speech language models lack important brain-relevant semantics
Subba Reddy Oota (Inria), Mariya Toneva (Max Planck Institute for Software Systems)
CodeRepresentation LearningTransformerLarge Language ModelBiomedical DataMagnetic Resonance Imaging
π― What it does: Evaluate the impact of low-level textual, auditory, and visual features in language model representations on brain alignment in text and speech models during reading and listening tasks using fMRI recordings.
Speech Sense Disambiguation: Tackling Homophone Ambiguity in End-to-End Speech Translation
Tengfei Yu (Harbin Institute of Technology), Min Zhang (Harbin Institute of Technology)
CodeTransformerContrastive LearningTextAudio
π― What it does: Constructed comprehensive homonym dictionaries for English, French, German, and Spanish, and annotated the MuST-C and CoVoST-2 end-to-end speech translation datasets based on these dictionaries; subsequently proposed the AmbigST model, which utilizes homonym masks and three-level contrastive learning to resolve speech homonym ambiguity.
CodeLarge Language ModelTextBenchmarkRetrieval-Augmented GenerationAudio
π― What it does: This paper systematically evaluates the impact of directly listening to recordings versus reading transcribed text on generating summaries, and further explores the differences in summary quality caused by transcription errors, expert versus non-expert annotations.
David Ponce (FundaciΓ³n Vicomtech, Basque Research and Technology Alliance), Harritxu Gete (FundaciΓ³n Vicomtech, Basque Research and Technology Alliance)
CodeGenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
π― What it does: This paper evaluates and improves the performance of large language models (LLMs) on the Split and Rephrase (SPRP) task, proposing multiple methods through LoRA fine-tuning and instruction fine-tuning;
SPOR: A Comprehensive and Practical Evaluation Method for Compositional Generalization in Data-to-Text Generation
Ziyao Xu (Peking University), Houfeng Wang (Peking University)
CodeGenerationData-Centric LearningLarge Language ModelSupervised Fine-TuningTextBenchmark
π― What it does: Proposes the SPOR (Systematicity, Productivity, Order invariance, Rule learnability) evaluation framework to comprehensively assess compositional generalization in data-to-text generation.
CodeAI Code AssistantReinforcement Learning from Human FeedbackLarge Language ModelSupervised Fine-TuningReinforcement LearningText
π― What it does: Improve code generation in large language models through reinforcement learning and compiler feedback, proposing the StepCoder framework.
StreamSpeech: Simultaneous Speech-to-Speech Translation with Multi-task Learning
Shaolei Zhang (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)
π― What it does: Proposed StreamSpeech, an 'integrated' end-to-end synchronous speech-to-speech translation model, capable of generating target speech in real-time while receiving continuous speech input, and providing intermediate ASR and text translation results;
StreamVoice: Streamable Context-Aware Language Modeling for Real-time Zero-Shot Voice Conversion
Zhichao Wang (Northwestern Polytechnical University), Yuping Wang (Northwestern Polytechnical University)
CodeGenerationCompressionTransformerLarge Language ModelAudio
π― What it does: Propose StreamVoice, a streaming zero-shot acoustic conversion system based on language models, which can achieve real-time speech conversion without using future information.
Striking Gold in Advertising: Standardization and Exploration of Ad Text Generation
Masato Mita (CyberAgent), Peinan Zhang (CyberAgent)
CodeGenerationTransformerLarge Language ModelSupervised Fine-TuningTextMultimodalityBenchmarkRetrieval-Augmented Generation
π― What it does: Standardize the task definition for Ad Text Generation (ATG), construct the first publicly available benchmark dataset CAMERA /camera_retro, and conduct multi-model, multi-modal baseline experiments and system-level evaluations on it.
Subtle Biases Need Subtler Measures: Dual Metrics for Evaluating Representative and Affinity Bias in Large Language Models
Abhishek Kumar (Brock University), Ali Emami (Brock University)
CodeGenerationTransformerLarge Language ModelMixture of ExpertsTextBenchmark
π― What it does: This paper constructs a Creativity-Oriented Generation Toolkit (CoGS) and introduces Representativeness Bias Score (RBS) and Affinity Bias Score (ABS) to evaluate subtle biases in large language models during the generation and evaluation processes.