NAACL 2025 论文列表

Code debugging is a vital stage of software development, essential for ensuring the reliability and performance of Large Language Models (LLMs) in the code generation task. Human debugging typically follows a multi-stage process, which includes Bug Localization, Bug Identification, Code Repair, and Code Recognition. However, existing code debugging benchmarks predominantly focus on the Code Repair stage, which offers only a limited perspective on evaluating the debugging capabilities of LLMs. In this paper, we introduce DEBUGEVAL, a comprehensive benchmark for evaluating the debugging abilities of LLMs by emulating the multi-stage human debugging process. Through evaluating on DEBUGEVAL, we observe that 7B-scale models consistently underperform compared to their larger counterparts, highlighting their limitations in comprehending code semantics. In this case, we propose the COmmunicative Agent-based data SynThesis (COAST) framework, which employs a multi-agent system to generate high-quality training data for supervised fine-tuning (SFT). Experimental results demonstrate that COAST-generated data outperform human-curated and GPT-4-generated data, enabling 7B-scale LLMs to achieve debugging performance comparable to GPT-3.5. All data and codes are available at https://github.com/NEUIR/COAST.

Challenges in managing linguistic diversity and integrating various musical modalities are faced by current music information retrieval systems. These limitations reduce their effectiveness in a global, multimodal music environment. To address these issues, we introduce CLaMP 2, a system compatible with 101 languages that supports both ABC notation (a text-based musical notation format) and MIDI (Musical Instrument Digital Interface) for music information retrieval. CLaMP 2, pre-trained on 1.5 million ABC-MIDI-text triplets, includes a multilingual text encoder and a multimodal music encoder aligned via contrastive learning. By leveraging large language models, we obtain refined and consistent multilingual descriptions at scale, significantly reducing textual noise and balancing language distribution. Our experiments show that CLaMP 2 achieves state-of-the-art results in both multilingual semantic search and music classification across modalities, thus establishing a new standard for inclusive and global music information retrieval.

Legal professionals need to write analyses that rely on citations to relevant precedents, i.e., previous case decisions. Intelligence systems assisting legal professionals in writing such documents provide great benefits but are challenging to design. Such systems need to help locate, summarize, and reason over salient precedents in order to be useful. To enable systems for such tasks, we work with legal professionals to create a colossal dataset. supporting two important backbone tasks: information retrieval (IR) and retrieval-augmented generation (RAG). This dataset **CLERC** (Case Law Evaluation and Retrieval Corpus), is constructed for training and evaluating models on their ability to (1) find corresponding citations for a given piece of legal analysis and to (2) compile the text of these citations (as well as previous context) into a cogent analysis that supports a reasoning goal. We benchmark state-of-the-art models on CLERC, showing that current approaches still struggle: GPT-4o generates analyses with the highest ROUGE F-scores but hallucinates the most, while zero-shot IR models only achieve 48.3% recall@1000.

Effective communication is vital in emergency response scenarios where clarity and speed can save lives. Traditional systems often struggle under the chaotic conditions of real-world emergencies, leading to breakdowns in communication and task management. This paper introduces CLEAR-Command, a system that leverages Large Language Models (LLMs) to enhance emergency communications. CLEAR stands for $textbfCoordinatedListening,Extraction, andAllocation inResponse. CLEAR-Command automates the transcription, summarization, and task extraction from live radio communications of emergency first responders using the OpenAI Whisper API for transcription and gpt-4o for summarization and task extraction. Our system provides a dynamic overview of task allocations and their execution status, significantly improving the accuracy of task identification and the clarity of communication. We evaluated our system through an expert pre-study with 4 experts and a user study with 13 participants. The expert pre-study identified gpt-4o as providing the most accurate task extraction, while the user study showed that CLEAR-Command significantly outperforms traditional radio communication in terms of clarity, trust, and correctness of task extraction. Our demo is hosted under thislink, and all project details are presented in ourGitlab page$.

Large language models (LLMs) have achieved remarkable performance on various NLP tasks, yet their potential in more challenging task like finance, has not been fully explored. In this paper, we present CFinBench: a meticulously crafted, the most comprehensive evaluation benchmark to date, for assessing the financial knowledge of LLMs under Chinese context. In practice, to better align with the career trajectory of Chinese financial practitioners, we build a systematic evaluation from 4 first-level categories: (1) Financial Subject: whether LLMs can memorize the necessary basic knowledge of financial subjects, such as economics, statistics and auditing. (2) Financial Qualification: whether LLMs can obtain the needed financial qualified certifications, such as certified public accountant, securities qualification and banking qualification. (3) Financial Practice: whether LLMs can fulfill the practical financial jobs, such as tax consultant, junior accountant and securities analyst. (4) Financial Law: whether LLMs can meet the requirement of financial laws and regulations, such as tax law, insurance law and economic law. CFinBench comprises 99,100 questions spanning 43 second-level categories with 3 question types: single-choice, multiple-choice and judgment. We conduct extensive experiments on a wide spectrum of representative LLMs with various model size on CFinBench. The results show that GPT4 and some Chinese-oriented models lead the benchmark, with the highest average accuracy being 66.02%, highlighting the challenge presented by CFinBench. All the data and evaluation code are open sourced at https://cfinbench.github.io/

Computational modeling of user-generated desires on social media can significantly aid decision-makers across various fields. Initially explored through wish speech,this task has evolved into a nuanced examination of hope speech. To enhance understanding and detection, we propose a novel scheme rooted in formal semantics approaches to modality, capturing both future-oriented hopes through desires and beliefs and the counterfactuality of past unfulfilled wishes and regrets. We manually re-annotated existing hope speech datasets and built a new one which constitutes a new benchmark in the field. We also explore the capabilities of LLMs in automatically detecting hope speech, relying on several prompting strategies. To the best of our knowledge, this is the first attempt towards a language-driven decomposition of the notional category hope and its automatic detection in a unified setting.

There is a significant gap between patient needs and available mental health support today. In this paper, we aim to thoroughly examine the potential of using Large Language Models (LLMs) to assist professional psychotherapy. To this end, we propose a new benchmark, CBT-Bench, for the systematic evaluation of cognitive behavioral therapy (CBT) assistance. We include three levels of tasks in CBT-Bench: **I: Basic CBT knowledge acquisition**, with the task of multiple-choice questions; **II: Cognitive model understanding**, with the tasks of cognitive distortion classification, primary core belief classification, and fine-grained core belief classification; **III: Therapeutic response generation**, with the task of generating responses to patient speech in CBT therapy sessions.These tasks encompass key aspects of CBT that could potentially be enhanced through AI assistance, while also outlining a hierarchy of capability requirements, ranging from basic knowledge recitation to engaging in real therapeutic conversations. We evaluated representative LLMs on our benchmark. Experimental results indicate that while LLMs perform well in reciting CBT knowledge, they fall short in complex real-world scenarios requiring deep analysis of patients’ cognitive structures and generating effective responses, suggesting potential future work.

Authorship Verification (AV) (do two documents have the same author?) is essential in many real-life applications. AV is often used in privacy-sensitive domains that require an offline proprietary model that is deployed on premises, making publicly served online models (APIs) a suboptimal choice. Current offline AV models however have lower downstream utility due to limited accuracy (eg: traditional stylometry AV systems) and lack of accessible post-hoc explanations. In this work, we address the above challenges by developing a trained, offline model CAVE (Controllable Authorship Verification Explanations). CAVE generates free-text AV explanations that are controlled to be (1) accessible (uniform structure that can be decomposed into sub-explanations grounded to relevant linguistic features), and (2) easily verified for explanation-label consistency. We generate silver-standard training data grounded to the desirable linguistic features by a prompt-based method Prompt-CAVE. We then filter the data based on rationale-label consistency using a novel metric Cons-R-L. Finally, we fine-tune a small, offline model (Llama-3-8B) with this data to create our model CAVE. Results on three difficult AV datasets show that CAVE generates high quality explanations (as measured by automatic and human evaluation) as well as competitive task accuracy. We have submitted our code and datasets as supplementary material.

Topic modelling is a pivotal unsupervised machine learning technique for extracting valuable insights from large document collections. Existing neural topic modelling methods often encode contextual information of documents, while ignoring contextual details of candidate centroid words, leading to the inaccurate selection of topic words due to the *contextualization gap*. In parallel, it is found that functional words are frequently selected over topical words. To address these limitations, we introduce **CAST**: **C**orpus-**A**ware **S**elf-similarity Enhanced **T**opic modelling, a novel topic modelling method that builds upon candidate centroid word embeddings contextualized on the dataset, and a novel self-similarity-based method to filter out less meaningful tokens. Inspired by findings in contrastive learning that self-similarities of functional token embeddings in different contexts are much lower than topical tokens, we find self-similarity to be an effective metric to prevent functional words from acting as candidate topic words. Our approach significantly enhances the coherence and diversity of generated topics, as well as the topic model’s ability to handle noisy data. Experiments on news benchmark datasets and one Twitter dataset demonstrate the method’s superiority in generating coherent, diverse topics, and handling noisy data, outperforming strong baselines.

Generating emotionally appropriate responses in conversations with large language models presents a significant challenge due to the complexities of human emotions and cognitive processes, which remain largely underexplored in their critical role in social interactions. In this study, we introduce a two-stage automatic data generation framework to create CAPE, a Chinese dataset named Cognitive Appraisal theory-based Emotional corpus. This corpus facilitates the generation of dialogues with contextually appropriate emotional responses by accounting for diverse personal and situational factors. We propose two tasks utilizing this dataset: emotion prediction and next utterance prediction. Both automated and human evaluations demonstrate that agents trained on our dataset can deliver responses that are more aligned with human emotional expressions. Our study shows the potential for advancing emotional expression in conversational agents, paving the way for more nuanced and meaningful human-computer interactions.

With the rapid development of large language models (LLMs), due to their strong performance across various fields, LLM-based evaluation methods (LLM-as-a-Judge) have become widely used in natural language generation (NLG) evaluation. However, these methods encounter the following challenges: (1) distinguishing instruction-following ability, (2) being applicable across diverse NLG tasks, and (3) identifying low-quality outputs. To address these issues, we propose CAMIEval, a multidimensional comparative evaluation method based on instruction-following. Specifically, we define three fundamental dimensions of instruction-following: relevance, factuality, and adherence. Subsequently, we introduce a concrete Chain-of-Thoughts (ConcreteCoT) process to enhance the accuracy of evaluations. In addition, we trained a “regrettable model” RegretLM to generate low-quality outputs, which helps the evaluator better identify the potential shortcomings of the candidate output by comparing low-quality outputs with reference outputs. Through this comparison, the evaluator can generate instruction-specific dimensions that complement the fundamental dimensions, forming a more comprehensive evaluation metric system. Experiments on two NLG evaluation benchmarks demonstrate that CAMIEval consistently outperforms existing methods in terms of correlation with human evaluations, providing a general and accurate framework for evaluating the outputs of LLMs.

Recent years have witnessed a significant interest in developing large multi-modal models (LMMs) capable of performing various visual reasoning and understanding tasks. This has led to the introduction of multiple LMM benchmarks to evaluate LMMs on different tasks. However, most existing LMM evaluation benchmarks are predominantly English-centric. In this work, we develop a comprehensive LMM evaluation benchmark for the Arabic language to represent a large population of over 400 million speakers. The proposed benchmark, named CAMEL-Bench, comprises eight diverse domains and 38 sub-domains including, multi-image understanding, complex visual perception, handwritten document understanding, video understanding, medical imaging, plant diseases, and remote sensing-based land use understanding to evaluate broad scenario generalizability. Our CAMEL-Bench comprises around 29,036 questions that are filtered from a larger pool of samples, where the quality is manually verified by native speakers to ensure reliable model assessment. We conduct evaluations of both closed-source, including GPT-4 series, and open-source LMMs. Our analysis reveals the need for substantial improvement, especially among the bestopen-source models, with even the closed-source GPT-4o achieving an overall score of 62%. Our benchmark will be publicly released.

Large Language Models (LLMs) are pretrained on extensive multilingual corpora to acquire both language-specific cultural knowledge and general knowledge. Ideally, while LLMs should provide consistent responses to culture-independent questions across languages, we observe significant performance disparities. To address this, we explore the **C**ross-Lingual Self-**A**ligning ability of **L**anguage **M**odels (**CALM**) to align knowledge across languages. Specifically, for a given question, we sample multiple responses across different languages and select the most self-consistent response as the target, leaving the remaining responses as negative examples. We then employ direct preference optimization (DPO) to align the model’s knowledge across different languages. Evaluations on the MEDQA and X-CSQA datasets demonstrate CALM’s effectiveness in enhancing cross-lingual knowledge question answering, both in zero-shot and retrieval-augmented settings. We also found that increasing the number of languages involved in CALM training leads to higher accuracy and consistency. We offer a qualitative analysis of how cross-lingual consistency can enhance knowledge alignment and explore the method’s generalizability.

Simultaneous speech translation (SimulST) systems must balance translation quality with response time, making latency measurement crucial for evaluating their real-world performance. However, there has been a longstanding belief that current metrics yield unrealistically high latency measurements in unsegmented streaming settings. In this paper, we investigate this phenomenon, revealing its root cause in a fundamental misconception underlying existing latency evaluation approaches. We demonstrate that this issue affects not only streaming but also segment-level latency evaluation across different metrics. Furthermore, we propose a modification to correctly measure computation-aware latency for SimulST systems, addressing the limitations present in existing metrics.

Recent research increasingly focuses on training vision-language models (VLMs) with long, detailed image captions. However, small-scale VLMs often struggle to balance the richness of these captions with the risk of hallucinating content during fine-tuning. In this paper, we explore how well VLMs adapt to such captions. To quantify caption quality, we propose Decomposed NLI (DNLI), an evaluation framework that breaks down generated captions into individual propositions, assessing each in isolation. This fine-grained analysis reveals a critical balance between capturing descriptive details and preventing hallucinations. Our findings show that simply reducing caption complexity or employing standard data curation techniques does not effectively resolve this issue. To tackle this challenge, we introduce Knowledge Adapted (KnowAda) fine-tuning, a data-centric approach that automatically adapts training data with the model’s existing knowledge and visual understanding. KnowAda minimizes hallucinations while preserving high descriptiveness. We validate this approach across several small-scale VLMs (up to 7B parameters) and dense caption datasets, demonstrating that KnowAda effectively balances hallucination reduction and descriptiveness. Our results show that KnowAda outperforms various baselines in both automatic metrics and human evaluations.

Deep learning-based expert models have reached superhuman performance in decision-making domains such as chess and Go. However, it is under-explored to explain or comment on given decisions although it is important for model explainability and human education. The outputs of expert models are accurate, but yet difficult to interpret for humans. On the other hand, large language models (LLMs) can produce fluent commentary but are prone to hallucinations due to their limited decision-making capabilities. To bridge this gap between expert models and LLMs, we focus on chess commentary as a representative task of explaining complex decision-making processes through language and address both the generation and evaluation of commentary. We introduce Concept-guided Chess Commentary generation (CCC) for producing commentary and GPT-based Chess Commentary Evaluation (GCC-Eval) for assessing it. CCC integrates the decision-making strengths of expert models with the linguistic fluency of LLMs through prioritized, concept-based explanations. GCC-Eval leverages expert knowledge to evaluate chess commentary based on informativeness and linguistic quality. Experimental results, validated by both human judges and GCC-Eval, demonstrate that CCC generates commentary which is accurate, informative, and fluent.

Following the advancement of large language models (LLMs), the development of LLM-based autonomous agents has become prevalent.As a result, the need to understand the security vulnerabilities of these agents has become a critical task. We examine how ReAct agents can be exploited using a straightforward yet effective method we refer to as the foot-in-the-door attack.Our experiments show that indirect prompt injection attacks, prompted by harmless and unrelated requests (such as basic calculations) can significantly increase the likelihood of the agent performing subsequent malicious actions.Our results show that once a ReAct agent’s thought includes a specific tool or action, the likelihood of executing this tool in the subsequent steps increases significantly, as the agent seldom re-evaluates its actions. Consequently, even random, harmless requests can establish a ‘foot-in-the-door’, allowing an attacker to embed malicious instructions into the agent’s thought process, making it more susceptible to harmful directives.To mitigate this vulnerability, we propose implementing a simple reflection mechanism that prompts the agent to reassess the safety of its actions during execution, which can help reduce the success of such attacks.

Power consumption plays a crucial role in on-device streaming speech recognition, significantly influencing the user experience. This study explores how the configuration of weight parameters in speech recognition models affects their overall energy efficiency. We found that the influence of these parameters on power consumption varies depending on factors such as invocation frequency and memory allocation. Leveraging these insights, we propose design principles that enhance on-device speech recognition models by reducing power consumption with minimal impact on accuracy. Our approach, which adjusts model components based on their specific energy sensitivities, achieves up to 47% lower energy usage while preserving comparable model accuracy and improving real-time performance compared to leading methods.

Pretrained language models (PLMs) have revolutionized NLP but amplify linguistic inequities in multilingual applications. While prior studies focused on transformer architectures such as BERT, we evaluate large language models (LLMs) including Mistral, TowerInstruct, OpenHathi, Tamil-Llama, and Kan-Llama. Through rigorous testing across eight languages spanning high-resource (English, German, French, Italian, Spanish) and low-resource (Hindi, Tamil, Kannada) settings, we reveal systemic failures in preserving multilingual reliability and adaptability. Using paradigms like each language for itself’ (ELFI) and each language for others’ (ELFO), we highlight the inability of current LLMs to bridge linguistic divides. Even model merging fail to mitigate these gaps, exposing fundamental limitations. These findings emphasize the critical need for reimagining AI architectures to deliver true linguistic inclusivity and equitable performance across diverse languages.