NAACL 2025 论文列表

Recent evaluations of LLMs on coreference resolution have revealed that traditional output formats and evaluation metrics do not fully capture the models’ referential understanding. To address this, we introduce IdentifyMe, a new benchmark for mention resolution presented in a multiple-choice question (MCQ) format, commonly used for evaluating LLMs. IdentifyMe features long narratives and employs heuristics to exclude easily identifiable mentions, creating a more challenging task. The benchmark also consists of a curated mixture of different mention types and corresponding entities, allowing for a fine-grained model performance analysis. We evaluate both closed- and open-source LLMs on IdentifyMe and observe a significant performance gap (20-30%) between the state-of-the-art sub-10B open models vs. closed ones. We observe that pronominal mentions, which have limited surface information, are typically harder for models to resolve than nominal mentions. Additionally, we find that LLMs often confuse entities when their mentions overlap in nested structures. The highest scoring model, GPT-4o, achieves 81.9% accuracy, highlighting the strong referential capabilities of state-of-the-art LLMs while also indicating room for further improvement.

We present ITALIC, a large-scale benchmark dataset of 10,000 multiple-choice questions designed to evaluate the natural language understanding of the Italian language and culture. ITALIC spans 12 domains, exploiting public tests to score domain experts in real-world scenarios. We detail our data collection process, stratification techniques, and selection strategies. ITALIC provides a comprehensive assessment suite that captures commonsense reasoning and linguistic proficiency in a morphologically rich language. We establish baseline performances using 17 state-of-the-art LLMs, revealing current limitations in Italian language understanding and highlighting significant linguistic complexity and cultural specificity challenges. ITALIC serves as a benchmark for evaluating existing models and as a roadmap for future research, encouraging the development of more sophisticated and culturally aware natural language systems.

Large Language Models (LLMs) perform well on unseen tasks in English, but their abilities in non-English languages are less explored due to limited benchmarks and training data. To bridge this gap, we introduce the Indic-QA Benchmark, a large dataset for context-grounded question answering in 11 major Indian languages, covering both extractive and abstractive tasks. Evaluations of multilingual LLMs, including instruction fine-tuned versions, revealed weak performance in low-resource languages due to a strong English-language bias in their training data. We also investigated the Translate-Test paradigm,where inputs are translated to English for processing and the results are translated back into the source language for output. This approach outperformed multilingual LLMs, particularly in low-resource settings. By releasing Indic-QA, we aim to promote further research into LLMs’ question-answering capabilities in low-resource languages. This benchmark offers a critical resource to address existing limitations and foster multilingual understanding.

The widespread use of social networks has significantly accelerated the dissemination of information but has also facilitated the rapid spread of fake news, leading to various negative consequences. Recently, with the emergence of large language models (LLMs), researchers have focused on leveraging LLMs for automated fake news detection. Unfortunately, many issues remain to be addressed. First, the evidence retrieved to verify given fake news is often insufficient, limiting the performance of LLMs when reasoning directly from this evidence. Additionally, the retrieved evidence frequently contains substantial redundant information, which can interfere with the LLMs’ judgment. To address these limitations, we propose a Multiple Knowledge Sources Retrieval and LLM Knowledge Conversion framework, which enriches the evidence available for claim verification. We also introduce a Redundant Information Filtering Strategy, which minimizes the influence of irrelevant information on the LLM reasoning process. Extensive experiments conducted on two challenging fact-checking datasets demonstrate that our proposed method outperforms state-of-the-art fact-checking baselines. Our code is available at https://github.com/quark233/IMRRF/tree/main.

The instruction hierarchy, which establishes a priority order from system messages to user messages, conversation history, and tool outputs, is essential for ensuring consistent and safe behavior in language models (LMs). Despite its importance, this topic receives limited attention, and there is a lack of comprehensive benchmarks for evaluating models’ ability to follow the instruction hierarchy. We bridge this gap by introducing IHEval, a novel benchmark comprising 3,538 examples across nine tasks, covering cases where instructions in different priorities either align or conflict. Our evaluation of popular LMs highlights their struggle to recognize instruction priorities. All evaluated models experience a sharp performance decline when facing conflicting instructions, compared to their original instruction-following performance. Moreover, the most competitive open-source model only achieves 48% accuracy in resolving such conflicts. Our results underscore the need for targeted optimization in the future development of LMs.

We introduce IFIR, the first comprehensive benchmark designed to evaluate instruction-following information retrieval (IR) in expert domains. IFIR includes 2,426 high-quality examples and covers eight subsets across four specialized domains: finance, law, healthcare, and science literature. Each subset addresses one or more domain-specific retrieval tasks, replicating real-world scenarios where customized instructions are critical. IFIR enables a detailed analysis of instruction-following retrieval capabilities by incorporating instructions at different levels of complexity. We also propose a novel LLM-based evaluation method to provide a more precise and reliable assessment of model performance in following instructions. Through extensive experiments on 15 frontier retrieval models, including those based on LLMs, our results reveal that current models face significant challenges in effectively following complex, domain-specific instructions. We further provide in-depth analyses to highlight these limitations, offering valuable insights to guide future advancements in retriever development.

Materials discovery and design are essential for advancing technology across various industries by enabling the development of application-specific materials. Recent research has leveraged Large Language Models (LLMs) to accelerate this process. We explore the potential of LLMs to generate viable hypotheses that, once validated, can expedite materials discovery. Collaborating with materials science experts, we curated a novel dataset from recent journal publications, featuring real-world goals, constraints, and methods for designing real-world applications. Using this dataset, we test LLM-based agents that generate hypotheses for achieving given goals under specific constraints. To assess the relevance and quality of these hypotheses, we propose a novel scalable evaluation metric that emulates the process a materials scientist would use to evaluate a hypothesis critically. Our curated dataset, proposed method, and evaluation framework aim to advance future research in accelerating materials discovery and design with LLMs.

Answering questions that require reasoning and aggregation across both structured (tables) and unstructured (raw text) data sources presents significant challenges. Current methods rely on fine-tuning and high-quality, human-curated data, which is difficult to obtain. Recent advances in Large Language Models (LLMs) have shown promising results for multi-hop question answering (QA) over single-source text data in a zero-shot setting, yet exploration into multi-source Table-Text QA remains limited. In this paper, we present a novel Hybrid Graph-based approach for Table-Text QA that leverages LLMs without fine-tuning. Our method constructs a unified Hybrid Graph from textual and tabular data, pruning information based on the input question to provide the LLM with relevant context concisely. We evaluate our approach on the challenging Hybrid-QA and OTT-QA datasets using state-of-the-art LLMs, including GPT-3.5, GPT-4, and LLaMA-3. Our method achieves the best zero-shot performance on both datasets, improving Exact Match scores by up to 10% on Hybrid-QA and 5.4% on OTT-QA. Moreover, our approach reduces token usage by up to 53% compared to the original context.

Large Language Models (LLMs) face limitations in AI legal and policy applications due to outdated knowledge, hallucinations, and poor reasoning in complex contexts. Retrieval-Augmented Generation (RAG) systems address these issues by incorporating external knowledge, but suffer from retrieval errors, ineffective context integration, and high operational costs. This paper presents the Hybrid Parameter-Adaptive RAG (HyPA-RAG) system, designed for the AI legal domain, with NYC Local Law 144 (LL144) as the test case. HyPA-RAG integrates a query complexity classifier for adaptive parameter tuning, a hybrid retrieval approach combining dense, sparse, and knowledge graph methods, and a comprehensive evaluation framework with tailored question types and metrics. Testing on LL144 demonstrates that HyPA-RAG enhances retrieval accuracy, response fidelity, and contextual precision, offering a robust and adaptable solution for high-stakes legal and policy applications.

Resume matching assesses the extent to which candidates qualify for jobs based on the content of resumes. This process increasingly uses natural language processing (NLP) techniques to automate parsing and rating tasks—saving time and effort. Large language models (LLMs) are increasingly used for this purpose—thus, we explore their capabilities for resume matching in an observational study. We compare zero-shot GPT-4 and human ratings for 736 resumes submitted to job openings from diverse fields using real-world evaluation criteria. We also study the effects of prompt engineering techniques on GPT-4 ratings and compare differences in GPT-4 and human ratings across racial and gender groups. Our results show: LLM scores correlate minorly with humans, suggesting they are not interchangeable; prompt engineering such as CoT improves the quality of LLM ratings; and LLM scores do not show larger group differences (i.e., bias) than humans. Our findings provide implications for LLM-based resume rating to promote more fair and NLP-based resume matching in a multicultural world.

Large Language Models infuse newfound vigor into the advancement of the medical domain, yet the scarcity of data poses a significant bottleneck hindering community progress. In this paper, we release the largest ever medical Question Answering (QA) dataset with 26 Million QA pairs named Huatuo-26M. We benchmark many existing approaches in our dataset in terms of both retrieval and generation. We also experimentally show the benefit of the proposed dataset in many aspects: (i) it serves as a fine-tuning data for training medical Large Language Models (LLMs); (ii) it works as an external knowledge source for retrieval-augmented generation (RAG); (iii) it demonstrates transferability by enhancing zero-shot performance on other QA datasets; and (iv) it aids in training biomedical model as a pre-training corpus. Our empirical findings substantiate the dataset’s utility in these domains, thereby confirming its significance as a resource in the medical QA landscape.

Entity matching (EM), which identifies whether two data records refer to the same real-world entity, is crucial for knowledge base construction and enhancing data-driven AI systems. Recent advances in language models (LMs) have shown great potential in resolving entities with rich textual attributes. However, their performance heavily depends on how structured entities are “talked” through serialized text. The impact of this serialization process remains underexplored, particularly for entities with complex relations in knowledge graphs (KGs). In this work, we systematically study entity serialization by benchmarking the effect of common schemes with LMs of different sizes on diverse tabular matching datasets. We apply our findings to propose a novel serialization scheme for KG entities based on random walks and utilize LLMs to encode sampled semantic walks for matching. Using this lightweight approach with open-source LLMs, we achieve a leading performance on EM in canonical and highly heterogeneous KGs, demonstrating significant throughput increases and superior robustness compared to GPT-4-based methods. Our study on serialization provides valuable insights for the deployment of LMs in real-world EM tasks.

The grammatical knowledge of language models (LMs) is often measured using a benchmark of linguistic minimal pairs, where LMs are presented with a pair of acceptable and unacceptable sentences and required to judge which is more acceptable. Conventional approaches compare sentence probabilities directly, but large language models (LLMs) provide nuanced evaluation methods using prompts and templates. We therefore investigate how to derive the most accurate acceptability judgments from LLMs to comprehensively evaluate their grammatical knowledge. Through extensive experiments in both English and Chinese, we compare nine judgment methods and demonstrate that two of them, in-template LP (a probability readout method) and Yes/No probability computing (a prompting-based method), achieve higher accuracy than the conventional approach. Our analysis reveals that the top two methods excel in different linguistic phenomena, suggesting they access different aspects of the LLMs’ grammatical knowledge. We find that ensembling the two methods achieves even higher accuracy. Consequently, we recommend these techniques, either individually or ensembled, as more effective alternatives to conventional approaches for assessing grammatical knowledge in LLMs.

In-context knowledge editing (IKE) enables efficient modification of large language model (LLM) outputs without parameter changes and at zero-cost. However, it can be misused to manipulate responses opaquely, e.g., insert misinformation or offensive content. Such malicious interventions could be incorporated into high-level wrapped APIs where the final input prompt is not shown to end-users. To address this issue, we investigate the detection and reversal of IKE-edits. First, we demonstrate that IKE-edits can be detected with high accuracy (F1 > 80%) using only the top-10 output probabilities of the next token, even in a black-box setting, e.g. proprietary LLMs with limited output information. Further, we introduce the novel task of reversing IKE-edits using specially tuned reversal tokens. We explore using both continuous and discrete reversal tokens, achieving over 80% accuracy in recovering original, unedited outputs across multiple LLMs. Our continuous reversal tokens prove particularly effective, with minimal impact on unedited prompts. Through analysis of output distributions, attention patterns, and token rankings, we provide insights into IKE’s effects on LLMs and how reversal tokens mitigate them. This work represents a significant step towards enhancing LLM resilience against potential misuse of in-context editing, improving their transparency and trustworthiness.

The massive amounts of web-mined parallel data often contain large amounts of noise. Semantic misalignment, as the primary source of the noise, poses a challenge for training machine translation systems. In this paper, we first introduce a process for simulating misalignment controlled by semantic similarity, which closely resembles misaligned sentences in real-world web-crawled corpora. Under our simulated misalignment noise settings, we quantitatively analyze its impact on machine translation and demonstrate the limited effectiveness of widely used pre-filters for noise detection. This underscores the necessity of more fine-grained ways to handle hard-to-detect misalignment noise. By analyzing the reliability of the model’s self-knowledge for distinguishing misaligned and clean data at the token level, we propose self-correction—an approach that gradually increases trust in the model’s self-knowledge to correct the supervision signal during training. Comprehensive experiments show that our method significantly improves translation performance both in the presence of simulated misalignment noise and when applied to real-world, noisy web-mined datasets, across a range of translation tasks.

There are more than 300 documented signed languages worldwide, which are indispensable avenues for computational linguists to study cross-cultural and cross-linguistic factors that affect automatic sign understanding and generation. Yet, these are studied under critically low-resource settings, especially when examining multiple signed languages simultaneously. In this work, we hypothesize that a linguistically informed alignment algorithm can improve the results of sign-to-sign translation models. To this end, we first conduct a qualitative analysis of similarities and differences across three signed languages: American Sign Language (ASL), Chinese Sign Language (CSL), and German Sign Language (DGS). We then introduce a novel generation and alignment algorithm for translating one sign language to another, exploring Large Language Models (LLMs) as intermediary translators and paraphrasers. We also compile a dataset of sign-to-sign translation pairs between these signed languages. Our model trained on this dataset performs well on automatic metrics for sign-to-sign translation and generation. Our code and data will be available for the camera-ready version of the paper.

We analyze contextual representations in neural autoregressive language models, emphasizing long-range contexts that span several thousand tokens. Our methodology employs a perturbation setup and the metric Anisotropy-Calibrated Cosine Similarity, to capture the degree of contextualization of long-range patterns from the perspective of representation geometry. We begin the analysis with a case study on standard decoder-only Transformers, demonstrating that similar perplexity can exhibit markedly different downstream task performance, which can be explained by the difference in contextualization of long-range content. Next, we extend the analysis to other models, covering recent novel architectural designs and various training configurations. The representation-level results illustrate a reduced capacity for high-complexity (i.e., less compressible) sequences across architectures, and that fully recurrent models rely heavily on local context, whereas hybrid models more effectively encode the entire sequence structure. Finally, preliminary analysis of model size and training configurations on the encoding of long-range context suggest potential directions for improving existing language models.

Multimodal foundation models aim to create a unified representation space that abstracts away from surface features like language syntax or modality differences. To investigate this, we study the internal representations of three recent models, analyzing the model activations from semantically equivalent sentences across languages in the text and speech modalities. Our findings reveal that: 1) Cross-modal representations converge over model layers, except in the initial layers specialized at text and speech processing. 2) Length adaptation is crucial for reducing the cross-modal gap between text and speech, although current approaches’ effectiveness is primarily limited to high-resource languages. 3) Speech exhibits larger cross-lingual differences than text. 4) For models not explicitly trained for modality-agnostic representations, the modality gap is more prominent than the language gap.

The rapid evolution of large language models (LLMs) has transformed the competitive landscape in natural language processing (NLP), particularly for English and other data-rich languages. However, underrepresented languages like Cantonese, spoken by over 85 million people, face significant development gaps, which is particularly concerning given the economic significance of the Guangdong-Hong Kong-Macau Greater Bay Area, and in substantial Cantonese-speaking populations in places like Singapore and North America. Despite its wide use, Cantonese has scant representation in NLP research, especially compared to other languages from similarly developed regions. To bridge these gaps, we outline current Cantonese NLP methods and introduce new benchmarks designed to evaluate LLM performance in factual generation, mathematical logic, complex reasoning, and general knowledge in Cantonese, which aim to advance open-source Cantonese LLM technology. We also propose future research directions and recommended models to enhance Cantonese LLM development.