Austria

The past years have witnessed a proliferation of large language models (LLMs). Yet, reliable evaluation of LLMs is challenging due to the inaccuracy of standard metrics in human perception of text quality and the inefficiency in sampling informative test examples for human evaluation. This paper presents a sample-efficient human evaluation method for LLMs based on the principle of MAximum Discrepancy (MAD) competition. MAD automatically selects a small set of informative input instructions, each of which maximizes the discrepancy of two LLMs’ reponses, which are subsequently subject to three-alternative forced choice by human subjects. The pairwise comparison results of multiple LLMs are then aggregated into a global ranking using the Elo rating system. We compare eight representative LLMs in terms of four skills: knowledge understanding, mathematical reasoning, writing, and coding. Experimental results show that the proposed method reliably achieves the &quot;golden&quot; ranking of LLMs with a minimum set of input instructions, which in turn reveal their relative strengths and weaknesses, and offers valuable insights for further LLM advancement.

ACL 2025

Sample-Efficient Human Evaluation of Large Language Models via Maximum Discrepancy Competition

automatic evaluation of datasets

automatic creation and evaluation of language resources

evaluation methodologies

### Welcome to ACL 2023, the 61st Annual Meeting of the Association for Computational Linguistics! 
 The conference will be held in Toronto, Canada, July 9-14, 2023. 
Following the succession of the recent conferences in our field, ACL 2023 will adopt a hybrid format.
While the impact of Covid has considerably diminished in terms of traveling, obtaining visas to Canada
entails a very long process. Moreover, the global economic conditions pose challenges for many individuals to travel to conferences. Recognizing these circumstances, we know many participants may not be
able to attend the conference in person. Therefore, we are committed to providing a great virtual platform
so everyone has the opportunity to interact with other participants and enjoy the conference. Based on the
current registered participants, approxiately 30% have chosen to attend the conference virtually. Whether
you join us in person or virtually, we sincerely hope everyone has a remarkable conference experience. 
This General Chair’s message is where I express my gratitude to the many individuals who have made
enormous contributions to the conference over the past year.

Read [**ACL 2023 General Chair's message**](https://docs.google.com/document/d/1WobYM7norbG4dI48s75HfJoD89qgX5a_F-6U8AteLSA/edit?usp=sharing/) in full.

##### **[Conference Handbook](https://2023.aclweb.org/downloads/acl2023-handbook.pdf)**

ACL 2023

The Association for Computational Linguistics (ACL) is the premier international scientific and professional society for people working on computational problems involving human language, a field often referred to as either computational linguistics or natural language processing.

The past years have witnessed a proliferation of large language models (LLMs). Yet, reliable evaluation of LLMs is challenging due to the inaccuracy of standard metrics in human perception of text quality and the inefficiency in sampling informative test examples for human evaluation. This paper presents a sample-efficient human evaluation method for LLMs based on the principle of MAximum Discrepancy (MAD) competition. MAD automatically selects a small set of informative input instructions, each of which maximizes the discrepancy of two LLMs’ reponses, which are subsequently subject to three-alternative forced choice by human subjects. The pairwise comparison results of multiple LLMs are then aggregated into a global ranking using the Elo rating system. We compare eight representative LLMs in terms of four skills: knowledge understanding, mathematical reasoning, writing, and coding. Experimental results show that the proposed method reliably achieves the "golden" ranking of LLMs with a minimum set of input instructions, which in turn reveal their relative strengths and weaknesses, and offers valuable insights for further LLM advancement.

poster

### Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Message from the General Chair: 
*It is my great pleasure and honor to welcome you to the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025), held in beautiful Vienna, Austria, from July 27 to August 1, 2025. ACL2025continues our field’s tradition of excellence in scholarship, innovation, and inclusivity, and I am deeply grateful to the many volunteers who have worked tirelessly to bring this event to life.* 
[Read more](https://drive.google.com/file/d/1GI_hvOpjswAuYdUTromfeDiPpCcqidwg/view?usp=sharing)

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to ACL 2025". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you need to log in with the **email address you registered with**. 

Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Text-to-image (T2I) models excel at generating high-quality images from text via powerful text encoders but training these encoders demands substantial computational resources. Consequently, many users seek pre-trained text encoders from model plugin-sharing platforms like Civitai and Hugging Face, which introduces an underexplored threat: the potential for adversaries to embed Trojans within these plugins. Existing Trojan attacks often require extensive training data and suffer from poor generalization across different triggers, limiting their effectiveness and scalability. To the best of our knowledge, this paper introduces the first **T**ext-encoder **W**eight-editing method for **I**nserting **S**ecret **T**rojans (**TWIST**). By identifying the *bottleneck MLP layer*—the critical point where minimal edits can dominantly control cross-modal alignment—TWIST achieves training-free and data-free Trojan insertion, which makes it highly efficient and practical. The experimental results across various triggers demonstrate that TWIST attains an average attack success rate of 91%, a 78% improvement over the state-of-the-art (SOTA) method proposed in 2024 and highlights the excellent generalization capability. Moreover, TWIST reduces modified parameters by 8-fold and cuts injection time to 25 seconds. Our findings underscore the security risks associated with text encoders in real-world applications and emphasize the need for more robust defense mechanisms.

TWIST: Text-encoder Weight-editing for Inserting Secret Trojans in Text-to-Image Models

Role-playing enables large language models (LLMs) to engage users in immersive and personalized interactions, but it also introduces significant safety risks. Existing role-play fine-tuning techniques improve role adaptability but may degrade safety performance, particularly for villainous characters. In this work, we conduct the first comprehensive assessment of role-play fine-tuning risks by training 95 role-specific LLMs using RoleBench. Our experiments reveal that role-play fine-tuning leads to a noticeable decline in safety performance, with safety risks varying based on character traits. To tackle this challenge, we propose Safety-Aware Role-Play Fine-Tuning (SaRFT), a novel method designed to balance role-playing capabilities and safety. Extensive experiments on LLaMA-3-8B-Instruct, Gemma-2-9B-it, and Qwen2.5-7B-Instruct demonstrate that SaRFT consistently outperforms state-of-the-art baselines under both LoRA and full-parameter fine-tuning settings. Our findings highlight the necessity of role-adaptive safety measures and provide insights into mitigating role-specific safety risks in role-playing LLMs.

Beware of Your Po! Measuring and Mitigating AI Safety Risks in Role-Play Fine-Tuning of LLMs

Multimodal reference resolution, including phrase grounding, aims to understand the semantic relations between mentions and real-world objects. Phrase grounding between images and their captions is a well-established task. In contrast, for real-world applications, it is essential to integrate textual and multimodal reference resolution to unravel the reference relations within dialogue, especially in handling ambiguities caused by pronouns and ellipses. This paper presents a framework that unifies textual and multimodal reference resolution by mapping mention embeddings to object embeddings and selecting mentions or objects based on their similarity. Our experiments show that learning textual reference resolution, such as coreference resolution and predicate-argument structure analysis, positively affects performance in multimodal reference resolution. In particular, our model with coreference resolution performs better in pronoun phrase grounding than representative models for this task, MDETR and GLIP. Our qualitative analysis demonstrates that incorporating textual reference relations strengthens the confidence scores between mentions, including pronouns and predicates, and objects, which can reduce the ambiguities that arise in visually grounded dialogues.

Disambiguating Reference in Visually Grounded Dialogues through Joint Modeling of Textual and Multimodal Semantic Structures

Current vision-language models may grasp basic spatial cues and simple directions (e.g. left, right, front, back), but struggle with the multi-dimensional spatial reasoning necessary for human-like understanding and real-world applications. To address this gap, we develop SPHERE (Spatial Perception and Hierarchical Evaluation of REasoning), a hierarchical evaluation framework supported by a new human-annotated dataset. SPHERE systematically probes models across increasing levels of complexity, from fundamental skills to multi-skill integration and high-level reasoning that combines spatial, visual, and logical understanding. Benchmark evaluation of state-of-the-art models reveals significant deficiencies, especially in reasoning about distance and proximity, understanding both egocentric and allocentric perspectives, and applying spatial logic in physical contexts. These findings expose critical blind spots in existing models and underscore the need for more advanced spatial reasoning techniques, driving the development of vision-language models that align more closely with human spatial cognition.

SPHERE: Unveiling Spatial Blind Spots in Vision-Language Models Through Hierarchical Evaluation

Retrieval-Augmented Generation (RAG) offers an effective solution to the issues faced by Large Language Models (LLMs) in hallucination generation and knowledge obsolescence by incorporating externally retrieved knowledge. However, existing methods lack effective control mechanisms for integrating internal and external knowledge. Inspired by human cognitive processes, we propose Parenting, a novel framework that decouples, identifies, and purposefully optimizes parameter subspaces related to adherence and robustness. Specifically, Parenting utilizes a key parameter mining method that combines forward and backward propagation signals to localize subspaces representing different capabilities. Then, Parenting employs a type-tailored tuning strategy, applying specific and appropriate optimizations to different subspaces, aiming to achieve a balanced enhancement of both adherence and robustness. Extensive experiments on various datasets and models validate the effectiveness and generalizability of our method. Our code is available at https://github.com/Nostradamus4869/Parenting.

Parenting: Optimizing Knowledge Selection of Retrieval-Augmented Language Models with Parameter Decoupling and Tailored Tuning

LLMs have achieved remarkable fluency and coherence in text generation, yet their widespread adoption has raised concerns about content reliability and accountability. In high-stakes domains, it is crucial to understand where and how the content is created. To address this, we introduce the Text pROVEnance (TROVE) challenge, designed to trace each sentence of a target text back to specific source sentences within potentially lengthy or multi-document inputs. Beyond identifying sources, TROVE annotates the fine-grained relationships (quotation, compression, inference, and others), providing a deep understanding of how each target sentence is formed.
To benchmark TROVE, we construct our dataset by leveraging three public datasets covering 11 diverse scenarios (e.g., QA and summarization) in English and Chinese, spanning source texts of varying lengths (0–5k, 5–10k, 10k+), emphasizing the multi-document and long-document settings essential for provenance. To ensure high-quality data, we employ a three-stage annotation process: sentence retrieval, GPT-4o provenance, and human provenance. We evaluate 11 LLMs under direct prompting and retrieval-augmented paradigms, revealing that retrieval is essential for robust performance, larger models perform better in complex relationship classification, and closed-source models often lead, yet open-source models show significant promise, particularly with retrieval augmentation. We make our dataset available here: https://github.com/ZNLP/ZNLP-Dataset.

TROVE: A Challenge for Fine-Grained Text Provenance via Source Sentence Tracing and Relationship Classification

Despite rising global usage of large language models (LLMs), their ability to generate *long-form* answers to *culturally specific* questions remains unexplored in many languages. To fill this gap, we perform the first study of textual multilingual long-form QA by creating CaLMQA, a dataset of **51.7K** culturally specific questions across **23** different languages. We define culturally specific questions as those that refer to concepts unique to one or a few cultures, or have different answers depending on the cultural or regional context. We obtain these questions by crawling naturally-occurring questions from community web forums in high-resource languages, and by hiring native speakers to write questions in under-resourced, rarely-studied languages such as Fijian and Kirundi. Our data collection methodologies are translation-free, enabling the collection of culturally unique questions like "Kuber iki umwami wa mbere w'uburundi yitwa Ntare?" (Kirundi; English translation: "Why was the first king of Burundi called Ntare (Lion)?"). We evaluate factuality, relevance and surface-level quality of LLM-generated long-form answers, finding that (1) for many languages, even the best models make critical surface-level errors (e.g., answering in the wrong language, repetition), especially for low-resource languages; and (2) answers to culturally specific questions contain more factual errors than answers to culturally agnostic questions – questions that have consistent meaning and answer across many cultures. We release CaLMQA to facilitate future research in cultural and multilingual long-form QA.

CaLMQA: Exploring culturally specific long-form question answering across 23 languages

Large language models (LLMs) face significant challenges in handling long-context tasks because of their limited effective context window size during pretraining, which restricts their ability to generalize over extended sequences. Meanwhile, extending the context window in LLMs through post-pretraining is highly resource-intensive.
To address this, we introduce LongRecipe, an efficient training strategy for extending the context window of LLMs, including impactful token analysis, position index transformation, and training optimization strategies. It simulates long-sequence inputs while maintaining training efficiency and significantly improves the model's understanding of long-range dependencies. Experiments on three types of LLMs show that LongRecipe can utilize long sequences while requiring only 30\% of the target context window size, and reduces computational training resource over 85\% compared to full sequence training. Furthermore, LongRecipe also preserves the original LLM's capabilities in general tasks. Ultimately, \textit{we can extend effective context window of open-source LLMs from 8k to 128k, achieving performance close to GPT-4 with just one day of dedicated training using a single GPU with 80G memory.}
Our code is released at https://github.com/zhiyuanhubj/LongRecipe.

LongRecipe: Recipe for Efficient Long Context Generalization in Large Language Models

***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.

GODBench: A Benchmark for Multimodal Large Language Models in Video Comment Art

Despite the impressive capabilities of LLMs, they often generate content with factual inaccuracies in LegalAI, which may lead to serious legal consequences. Retrieval-Augmented Generation (RAG), a promising approach, can conveniently integrate specialized knowledge into LLMs. In practice, there are diverse legal knowledge retrieval demands (e.g. law articles and similar cases). However, existing retrieval methods are either designed for general domains, struggling with legal knowledge, or tailored for specific legal tasks, unable to handle diverse legal knowledge types. Therefore, we propose a novel **Uni**fied **L**egal **R**etriever (UniLR) capable of performing multiple legal retrieval tasks for LLMs. Specifically, we introduce attention supervision to guide the retriever in focusing on key elements during knowledge encoding. Next, we design a graph-based method to integrate meta information through a heterogeneous graph, further enriching the knowledge representation. These two components work together to enable UniLR to capture the essence of knowledge hidden beneath formats. Extensive experiments on multiple datasets of common legal tasks demonstrate that UniLR achieves the best retrieval performance and can significantly enhance the performance of LLM.

Downloads

Next from ACL 2025

TWIST: Text-encoder Weight-editing for Inserting Secret Trojans in Text-to-Image Models

Similar lecture

Can Large Language Models Be an Alternative to Human Evaluations?

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES