ACL 2025 论文列表

Language models are often miscalibrated, leading to confidently incorrect answers. We introduce GRACE, a benchmark for language model calibration that incorporates comparison with human calibration. GRACE consists of question-answer pairs, in which each question contains a series of clues that gradually become easier, all leading to the same answer; models must answer correctly as early as possible as the clues are revealed. This setting permits granular measurement of model calibration based on how early, accurately, and confidently a model answers. After collecting these questions, we host live human vs. model competitions to gather 1,749 data points on human and model teams’ timing, accuracy, and confidence. We propose a metric, CalScore, that uses GRACE to analyze model calibration errors and identify types of model miscalibration that differ from human behavior. We find that although humans are less accurate than models, humans are generally better calibrated. Since state-of-the-art models struggle on GRACE, it effectively evaluates progress on improving model calibration.

This work contributes to the scarce empirical literature on LLM-based interactive homework in real-world educational settings and offers a practical, scalable solution to improve homework in schools. Homework is an important part of education in schools across the world, but to maximize benefit, it must be accompanied by feedback and follow-up questions. We developed a prompting strategy that enables GPT-4 to conduct interactive homework sessions for high school students learning English as a second language. Our strategy requires minimal effort in content preparation, one of the key challenges of alternatives such as home tutors or ITSs. We carried out a Randomized Controlled Trial (RCT) in four high-school classes, replacing traditional homework with GPT-4 homework sessions for the treatment group. We found that the treatment group had higher levels of satisfaction and desire to keep using the system among the students. This occurred without compromising learning outcomes, and one group even showed significantly better learning gains.

Large Language Models (LLMs) often succumb to adversarial prompts, a phenomenon popularly known as “jailbreaking.” While jailbreaking primarily targets short-term noncompliance with predefined policies, we argue that a deeper vulnerability lies in altering an LLM’s fundamental axiomatic beliefs, such as mathematical or philosophical truths. In this work, we introduce GoodLiar, a reinforcement learning (RL)-based framework that generates deceptive contexts to systematically rewrite an LLM’s core logical or philosophical understandings. By incentivizing an RL agent to produce persuasive and coherent arguments, GoodLiar aims to induce persistent belief shifts, rather than merely influencing immediate judgments of factual truthfulness. %rather than one-off policy breaches. Our approach introduces DA-ILQL, a novel offline RL method that extends ILQL by integrating on-policy data and language exploration to enhance the language discovery and optimization. Through extensive evaluations on multiple LLMs, we show that deceptive contexts discovered by GoodLiar consistently outperform simple multi-turn prompting methods.

Large language models (LLMs)-based query expansion for information retrieval augments queries with generated hypothetical documents with LLMs. However, its performance relies heavily on the scale of the language models (LMs), necessitating larger, more advanced LLMs. This approach is costly, computationally intensive, and often has limited accessibility. To address these limitations, we introduce GOLFer - Smaller LMs-Generated Documents Hallucination Filter & Combiner - a novel method leveraging smaller open-source LMs for query expansion. GOLFer comprises two modules: a hallucination filter and a documents combiner. The former detects and removes non-factual and inconsistent sentences in generated documents, a common issue with smaller LMs, while the latter combines the filtered content with the query using a weight vector to balance their influence. We evaluate GOLFer alongside dominant LLMs-based query expansion methods on three web search and ten low-resource datasets. Experimental results demonstrate that GOLFer consistently outperforms other methods using smaller LMs, and maintains competitive performance against methods using large-size LLMs, demonstrating its effectiveness.

***Video Comment Art*** enhances user engagement by providing creative content that conveys humor, satire, or emotional resonance, requiring a nuanced and comprehensive grasp of cultural and contextual subtleties. Although Multimodal Large Language Models (MLLMs) and Chain-of-Thought (CoT) have demonstrated strong reasoning abilities in STEM tasks (e.g. mathematics and coding), they still struggle to generate creative expressions such as resonant jokes and insightful satire. Moreover, existing benchmarks are constrained by their limited modalities and insufficient categories, hindering the exploration of comprehensive creativity in video-based Comment Art creation. To address these limitations, we introduce **GODBench**, a novel benchmark that integrates video and text modalities to systematically evaluate MLLMs’ abilities to compose Comment Art. Furthermore, inspired by the propagation patterns of waves in physics, we propose **Ripple of Thought (RoT)**, a multi-step reasoning framework designed to enhance the creativity of MLLMs. Extensive experiments on GODBench reveal that existing MLLMs and CoT methods still face significant challenges in understanding and generating creative video comments. In contrast, RoT provides an effective approach to improving creative composing, highlighting its potential to drive meaningful advancements in MLLM-based creativity.

Retrieval-augmented generation (RAG) in Knowledge Graph Question Answering (KGQA) enhances the context of Large Language Models (LLMs) by incorporating information retrieved from the Knowledge Graph (KG). Most recent approaches rely on costly LLM calls to generate executable relation paths or traverse the KG, which is inefficient in complex KGQA tasks, such as those involving multi-hop or multi-entity questions. We introduce the GNN-RAG framework, which utilizes lightweight Graph Neural Networks (GNNs) for effective and efficient graph retrieval. The GNN learns to assign importance weights to nodes based on their relevance to the question, as well as the relevance of their neighboring nodes. This enables the framework to effectively handle context from deeper parts of the graph, improving retrieval performance. GNN-RAG retrieves the shortest paths connecting question entities to GNN answer candidates, providing this information as context for the LLM. Experimental results show that GNN-RAG achieves effective retrieval on two widely used KGQA benchmarks (WebQSP and CWQ), outperforming or matching GPT-4 performance with a 7B tuned LLM. Additionally, GNN-RAG excels on multi-hop and multi-entity questions outperforming LLM-based retrieval approaches by 8.9–15.5% points at answer F1. Furthermore, it surpasses long-context inference while using 9× fewer KG tokens. The code is provided in https://github.com/cmavro/GNN-RAG.

Document-level relation extraction (DocRE) identifies relations between entities across an entire document. However, as the number and complexity of entities and entity-pair relations grow, the problem space expands quadratically, causing incomplete annotations and frequent false negatives, especially in biomedical datasets due to high construction costs. This leads to low recall in real-world scenarios. To address this, we propose GLiM, a novel framework that reduces the problem space using a graph-enhanced Transformer-based model and leverages large language models (LLMs) for reasoning. GLiM employs a cascaded approach: first, a graph-enhanced Transformer processes entity-pair relations with finer granularity by dynamically adjusting the graph size based on the number of entities; then, LLM inference handles challenging cases. Experiments show that GLiM boosts average recall and F1 scores by +6.34 and +4.41, respectively, outperforming state-of-the-art models on biomedical benchmarks. These results demonstrate the effectiveness of combining graph-enhanced Transformers with LLM inference for biomedical DocRE. Code will be released at https://github.com/HaoFang10/GLiM.

Knowledge Graph Completion (KGC), which aims to infer missing or incomplete facts, is a crucial task for KGs. However, integrating the vital structural information of KGs into Large Language Models (LLMs) and outputting predictions deterministically remains challenging. To address this, we propose a new method called GLTW, which encodes the structural information of KGs and merges it with LLMs to enhance KGC performance. Specifically, we introduce an improved Graph Transformer (iGT) that effectively encodes subgraphs with both local and global structural information and inherits the characteristics of language model, bypassing training from scratch. Also, we develop a subgraph-based multi-classification training objective, using all entities within KG as classification objects, to boost learning efficiency. Importantly, we combine iGT with an LLM that takes KG language prompts as input. Our extensive experiments on various KG datasets show that GLTW achieves significant performance gains compared to SOTA baselines.

Large Vision-Language Models (LVLMs) have recently gained attention due to their distinctive performance and broad applicability. While it has been previously shown that their efficacy in usage scenarios involving non-Western contexts falls short, existing studies are limited in scope, covering just a narrow range of cultures, focusing exclusively on a small number of cultural aspects, or evaluating a limited selection of models on a single task only. Towards globally inclusive LVLM research, we introduce GIMMICK, an extensive multimodal benchmark designed to assess a broad spectrum of cultural knowledge across 144 countries representing six global macro-regions. GIMMICK comprises six tasks built upon three new datasets that span 728 unique cultural events or facets on which we evaluated 20 LVLMs and 11 LLMs, including five proprietary and 26 open-weight models of all sizes. We systematically examine (1) regional cultural biases, (2) the influence of model size, (3) input modalities, and (4) external cues. Our analyses reveal strong biases toward Western cultures across models and tasks and highlight strong correlations between model size and performance, as well as the effectiveness of multimodal input and external geographic cues. We further find that models have more knowledge of tangible than intangible aspects (e.g., food vs. rituals) and that they excel in recognizing broad cultural origins but struggle with a more nuanced understanding.

Parameter Efficient Fine-Tuning (PEFT) methods have gained popularity and democratized the usage of Large Language Models (LLMs). Recent studies have shown that a small subset of weights significantly impacts performance. Based on this observation, we introduce a novel PEFT method, called Gaussian noise Injected Fine Tuning of Salient Weights (GIFT-SW). Our method updates only salient columns, while injecting Gaussian noise into non-salient ones. To identify these columns, we developed a generalized sensitivity metric that extends and unifies metrics from previous studies. Experiments with LLaMA models demonstrate that GIFT-SW outperforms full fine-tuning and modern PEFT methods under the same computational budget. Moreover, GIFT-SW offers practical advantages to recover performance of models subjected to mixed-precision quantization with keeping salient weights in full precision.

Publicly significant images from events carry valuable contextual information with applications in domains such as journalism and education. However, existing methodologies often struggle to accurately extract this contextual relevance from images. To address this challenge, we introduce GETREASON (Geospatial Event Temporal Reasoning), a framework designed to go beyond surfacelevel image descriptions and infer deeper contextual meaning. We hypothesize that extracting global event, temporal, and geospatial information from an image enables a more accurate understanding of its contextual significance. We also introduce a new metric GREAT (Geospatial, Reasoning and Event Accuracy with Temporal alignment) for a reasoning capturing evaluation. Our layered multi-agentic approach, evaluated using a reasoning-weighted metric, demonstrates that meaningful information can be inferred from images, allowing them to be effectively linked to their corresponding events and broader contextual background.

Generative methods significantly advance event argument extraction by probabilistically generating event argument sequences in a structured format. However, existing approaches primarily rely on a single prompt to generate event arguments in a fixed, predetermined order. Such a rigid approach overlooks the complex structural and dynamic interdependencies among event arguments. In this work, we present GEMS, a multi-prompt learning framework that Generates Event arguments via Multi-perspective prompts and ontology Steering. Specifically, GEMS utilizes multiple unfilled prompts for each sentence, predicting event arguments in varying sequences to explicitly capture the interrelationships between arguments. These predictions are subsequently aggregated using a voting mechanism. Furthermore, an ontology-driven steering mechanism is proposed to ensure that the generated arguments are contextually appropriate and consistent with event-specific knowledge. Extensive experiments on two benchmark datasets demonstrate that GEMS achieves state-of-the-art performance, particularly in low-resource settings. The source code is available at: https://github.com/AONE-NLP/EAE-GEMS

Recent advances in large language models have highlighted the critical need for precise control over model outputs through predefined constraints. While existing methods attempt to achieve this through either direct instruction-response synthesis or preferential response optimization, they often struggle with constraint understanding and adaptation. This limitation becomes particularly evident when handling fine-grained constraints, leading to either hallucination or brittle performance. We introduce Generative Adversarial Policy Optimization (GAPO), a novel framework that combines GAN-based training dynamics with an encoder-only reward model to progressively learn and adapt to increasingly complex constraints. GAPO leverages adversarial training to automatically generate training samples of varying difficulty while utilizing the encoder-only architecture to better capture prompt-response relationships. Extensive experiments demonstrate GAPO’s superior performance across multiple benchmarks, particularly in scenarios requiring fine-grained constraint handling, where it significantly outperforms existing methods like PPO, DPO, and KTO. Our results suggest that GAPO’s unique approach to preferential prompt learning offers a more robust and effective solution for controlling LLM outputs.

Large Language Models (LLMs) are increasingly deployed in real-world applications that demand complex reasoning. To track progress, robust benchmarks are required to evaluate their capabilities beyond superficial pattern recognition. However, current LLM reasoning benchmarks often face challenges such as insufficient interpretability, performance saturation or data contamination. To address these challenges, we introduce GAMEBoT, a gaming arena designed for rigorous and transparent assessment of LLM reasoning capabilities. GAMEBoT decompose complex reasoning in games into predefined modular subproblems. This decomposition allows us to design a suite of Chain-of-Thought (CoT) prompts infused with domain knowledge to guide LLMs in addressing these subproblems before action selection. Furthermore, we develop a suite of rule-based algorithms to generate ground truth for these subproblems, enabling rigorous validation of the LLMs’ intermediate reasoning steps. This approach facilitates evaluation of both the quality of final actions and the accuracy of the underlying reasoning process. GAMEBoT also naturally alleviates the risk of data contamination through dynamic games and head-to-head LLM competitions. We benchmark 17 prominent LLMs across eight games, encompassing various strategic abilities and game characteristics. Our results suggest that GAMEBoT presents a significant challenge, even when LLMs are provided with detailed CoT prompts.

Programming languages possess rich semantic information - such as data flow - that is represented by graphs and not available from the surface form of source code. Recent code language models have scaled to billions of parameters, but model source code solely as text tokens while ignoring any other structural information. Conversely, models that do encode structural information of code make modifications to the Transformer architecture, limiting their scale and compatibility with pretrained LLMs. In this work, we take the best of both worlds with GALLa - Graph Aligned Large Language Models. GALLa utilizes graph neural networks and cross-modal alignment technologies to inject the structural information of code into LLMs as an auxiliary task during finetuning. This framework is both model-agnostic and task-agnostic, as it can be applied to any code LLM for any code downstream task, and requires the structural graph data only at training time from a corpus unrelated to the finetuning data, while incurring no cost at inference time over the baseline LLM. Experiments on five code tasks with six different baseline LLMs ranging in size from 350M to 14B validate the effectiveness of GALLa, demonstrating consistent improvement over the baseline, even for powerful models such as LLaMA3 and Qwen2.5-Coder.

Social network simulation is developed to provide a comprehensive understanding of social networks in the real world, which can be leveraged for a wide range of applications such as group behavior emergence, policy optimization, and business strategy development. However, billions of individuals and their evolving interactions involved in social networks pose challenges in accurately reflecting real-world complexities. In this study, we propose a comprehensive Social network Simulation System (GA-S3) that leverages newly designed Group Agents to make intelligent decisions regarding various online events. Unlike other intelligent agents that represent an individual entity, our group agents model a collection of individuals exhibiting similar behaviors, facilitating the simulation of large-scale network phenomena with complex interactions at a manageable computational cost. Additionally, we have constructed a social network benchmark from 2024 popular online events that contains fine-grained information on Internet traffic variations. The experiment demonstrates that our approach is capable of achieving accurate and highly realistic prediction results.

Forecasting over Temporal Knowledge Graphs (TKGs) which predicts future facts based on historical ones has received much attention. Recent studies have introduced Large Language Models (LLMs) for this task to enhance the models’ generalization abilities. However, these models perform forecasting via simultaneously learning two kinds of entangled knowledge in the TKG: (1) general patterns, i.e., invariant temporal structures shared across different scenarios; and (2) scenario information, i.e., factual knowledge engaged in specific scenario, such as entities and relations. As a result, the learning processes of these two kinds of knowledge may interfere with each other, which potentially impact the generalization abilities of the models. To enhance the generalization ability of LLMs on this task, in this paper, we propose a General-to-Specific learning framework (G2S) that disentangles the learning processes of the above two kinds of knowledge. In the general learning stage, we mask the scenario information in different TKGs and convert it into anonymous temporal structures. After training on these structures, the model is able to capture the general patterns across different TKGs. In the specific learning stage, we inject the scenario information into the structures via either in-context learning or fine-tuning modes. Experimental results show that G2S effectively improves the generalization abilities of LLMs.

Large Language Model (LLM)-based Multi-agent Systems (MAS) have demonstrated remarkable capabilities in various complex tasks, ranging from collaborative problem-solving to autonomous decision-making. However, as these systems become increasingly integrated into critical applications, their vulnerability to adversarial attacks, misinformation propagation, and unintended behaviors have raised significant concerns. To address this challenge, we introduce G-Safeguard, a topology-guided security lens and treatment for robust LLM-MAS, which leverages graph neural networks to detect anomalies on the multi-agent utterance graph and employ topological intervention for attack remediation. Extensive experiments demonstrate that G-Safeguard: (I) exhibits significant effectiveness under various attack strategies, recovering over 40% of the performance for prompt injection; (II) is highly adaptable to diverse LLM backbones and large-scale MAS; (III) can seamlessly combine with mainstream MAS with security guarantees.

Speculative Decoding (SD) enforces strict distributional equivalence to the target model when accepting candidate tokens. While it maintains the target model’s generation quality, this strict equivalence limits the speedup achievable by SD and prevents users from trading deviations from the target distribution in exchange for further inference speed gains. To address these limitations, we introduce Fuzzy Speculative Decoding (FSD) - a decoding algorithm that generalizes SD by accepting candidate tokens based on the divergences between the target and draft model distributions. By allowing for controlled divergence from the target model, FSD enables users to flexibly trade generation quality for inference speed. Across several benchmarks, our method is able to achieve significant runtime improvements of over 5 tokens per second faster than SD at only an approximate 2% absolute reduction in benchmark accuracy. In many cases, FSD is even able to match SD benchmark accuracy at over 2 tokens per second faster, demonstrating that distributional equivalence is not necessary to maintain target model performance. Furthermore, FSD can be seamlessly integrated into existing SD extensions; we demonstrate this by applying FSD to EAGLE-2, greatly enhancing this existing extension’s efficiency while allowing it to leverage FSD’s tunable quality-speed trade-off.