China

Aligning Large Language Models (LLMs) with human preferences is crucial in ensuring desirable and controllable model behaviors. Current methods, such as Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO), rely on the Bradley-Terry (B-T) model to maximize the likelihood of pairwise choices. However, when multiple responses are available, the B-T model fails to guarantee an accurate list ranking of the responses. To address this issue, we propose Permutative Preference Alignment (PPA), a novel offline listwise approach that incorporates the Normalized Discounted Cumulative Gain (NDCG)—a widely-used ranking metric—as an alternative training objective for LLM alignment. We develop an end-to-end alignment algorithm by approximating NDCG with a differentiable surrogate loss. Experiments demonstrate that PPA outperforms existing pairwise and listwise methods on evaluation sets and general benchmarks such as AlpacaEval. Furthermore, we show that NDCG-based approaches improve ranking accuracy more effectively than B-T-based methods and provide a theoretical explanation for this improvement.

EMNLP 2025

Permutative Preference Alignment from Listwise Ranking of Human Judgments

listwise

alignment

learning to rank

### Welcome to 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics

Welcome to the 2025 meeting of the Nations of the Americas Chapter of the Association for Computational Linguistics! I am proud to help organize the first NAACL conference to carry the new name of our organization, one that emphasizes inclusion for all of the Americas. I am also pleased to welcome you to Albuquerque, New Mexico, a state whose unique blend of cultural influences will make for an excellent backdrop for NAACL 2025, especially with this year’s special theme on NLP in a Multicultural World. 
**[Continue reading...](https://drive.google.com/file/d/1jX-qGhqVSZZCIrAnJaz798pflu5Irrdn/view?usp=sharing)**

*- Colin Cherry, Google, NAACL 2025 General Chair* 

[![](https://assets.underline.io/markdown_image/1/image/b087f8a4dc5816d6e1a6514e59c59ac3.png)](https://drive.google.com/file/d/1T96GzPqObXrMln2BMByCSXTSizjTg69P/view?usp=sharing)

You need to log in with the email address you registered with. Access credentials have been sent to your email. 

Please be sure to check your spam and other email folders if you do not see an email confirmation right away.

Please log in to explore this event.

To access NAACL 2025 event page you are required to register. Please follow [**this link**](https://2025.naacl.org/registration/registration/) to register. Access will depend on your registration type.

Please register!

NAACL 2025

Welcome to 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics

poster

## Welcome!
"I am excited to welcome you to this year’s edition of the Conference on Empirical Methods in Natural Language Processing! Importantly, it marks the 30th edition of EMNLP. With over 8,000 submissions, more than 3,000 accepted papers, and thousands of attendees, we have come a long way from that first
workshop, which had 14 accepted papers. As the field looks ahead, Suzhou is the fitting location for celebrating this milestone: rooted in a long literary tradition, yet modern and forward-looking, and home to a large share of the NLP community." 

*Message from the General Chair, Dirk Hovy*

Celebrate 30 Years of EMNLP! 
EMNLP 2025 will be held in Suzhou, China from November 5th to November 9th, 2025.

Recent advances in text-to-video (T2V) generation highlight the critical role of high-quality video-text pairs in training models capable of producing coherent and instruction-aligned videos. However, strategies for optimizing video captions specifically for T2V training remain underexplored. In this paper, we introduce VC4VG (Video Captioning for Video Generation), a comprehensive caption optimization framework tailored to the needs of T2V models.We begin by analyzing caption content from a T2V perspective, decomposing the essential elements required for video reconstruction into multiple dimensions, and proposing a principled caption design methodology. To support evaluation, we construct VC4VG-Bench, a new benchmark featuring fine-grained, multi-dimensional, and necessity-graded metrics aligned with T2V-specific requirements.Extensive T2V fine-tuning experiments demonstrate a strong correlation between improved caption quality and video generation performance, validating the effectiveness of our approach. All benchmark tools and code will be released to support further research.

VC4VG: Optimizing Video Captions for Text-to-Video Generation

To meet the increasing need for timely and accurate evaluation of large language model (LLM) responses, training LLM-as-judges to evaluate and critique other model responses has emerged as a popular paradigm. However, existing judge models are largely trained with supervised finetuning (SFT) on small data scales to perform limited types of evaluation tasks, fundamentally limiting generalization. To meet the need for strong, generalized judge models, we explore training foundational judge models at large data scales (680K) with direct preference optimization (DPO). Using four training tasks, we form three types of DPO preference pairs targeting different aspects of evaluation: Generating meaningful critiques, making accurate judgements, and understanding what comprises good and bad responses. To demonstrate the effectiveness of our method, we train judge models of three sizes: 8B parameters, 12B, and 70B, and evaluate on a comprehensive suite of 13 benchmarks (7 pairwise, 4 single rating, and 2 classification). Our models achieve the best aggregate performance, with even our 8B model outperforming GPT-4o in pairwise benchmarks. Further analysis shows that our judge models produce factual and actionable critiques and serve as strong foundational judges for continued finetuning.

Direct Judgement Preference Optimization

Advancements in Large Vision-Language Models (LVLMs) have demonstrated impressive performance in image-conditioned text generation; however, hallucinated outputs--text that misaligns with the visual input--pose a major barrier to their use in safety-critical applications. We introduce ConfLVLM, a conformal-prediction-based framework that achieves finite-sample distribution-free statistical guarantees to the factuality of LVLM output. Taking each generated detail as a hypothesis, ConfLVLM statistically tests factuality via efficient heuristic uncertainty measures to filter out unreliable claims. We conduct extensive experiments covering three representative application domains: general scene understanding, medical radiology report generation, and document understanding. Remarkably, ConfLVLM reduces the error rate of claims generated by LLaVa-1.5 for scene descriptions from 87.8% to 10.0% by filtering out erroneous claims with a 95.3% true positive rate. Our results further show that ConfLVLM is highly flexible, and can be applied to any black-box LVLMs paired with any uncertainty measure for any image-conditioned free-form text generation task while providing a rigorous guarantee on controlling hallucination risk.

Towards Statistical Factuality Guarantee for Large Vision-Language Models

Retrieval-augmented generation (RAG) addresses the limitation of large language models (LLMs) in achieving up-to-date information by integrating external knowledge sources, but it is hindered by noisy or irrelevant retrieved data, leading to reduced accuracy. Additionally, most RAG methods rely on task-specific supervision, reducing their adaptability across domains. To overcome these challenges, we propose WinnowRAG, a novel multi-agent debate-based RAG framework. WinnowRAG operates in two stages: in Stage I, query-aware clustering groups similar documents, with each cluster assigned to an LLM agent for generating personalized responses. A critic LLM then consolidates these answers, forming super-agents. In Stage II, the super-agents engage in a structured discussion to filter out incorrect or irrelevant information, ensuring only relevant knowledge is used for final response generation. Crucially, WinnowRAG is unsupervised and leverages pretrained LLMs without requiring fine-tuning, making it easily adaptable to various tasks. The experiments on various realistic datasets demonstrate the effectiveness of WinnowRAG over state-of-the-art baselines.

Separate the Wheat from the Chaff: Winnowing Down Divergent Views in Retrieval Augmented Generation

Detecting misogyny in multimodal content remains a notable challenge, particularly in culturally conservative and low-resource contexts like Bangladesh. While existing research has explored hate speech and general meme classification, the nuanced identification of misogyny in Bangla memes, rich in metaphor, humor, and visual-textual interplay, remains severely underexplored. To address this gap, we introduce BanMiMe, the first comprehensive Bangla misogynistic meme dataset comprising 2,000 culturally grounded samples where each meme includes misogyny labels, humor categories, metaphor localization, and detailed human-written explanations. We benchmark the various performance of open and closed-source vision-language models (VLMs) under zero-shot and prompt-based settings and evaluate their capacity for both classification and explanation generation. Furthermore, we systematically explore multiple fine-tuning strategies, including standard, data-augmented, and Chain-of-Thought (CoT) supervision. Our results demonstrate that CoT-based fine-tuning consistently enhances model performance, both in terms of accuracy and in generating meaningful explanations. We envision BanMiMe as a foundational resource for advancing explainable multimodal moderation systems in low-resource and culturally sensitive settings.

BANMIME : Misogyny Detection with Metaphor Explanation on Bangla Memes

Spreadsheets are critical to data-centric tasks, with rich, structured layouts that enable efficient information transmission. Given the time and expertise required for manual spreadsheet layout design, there is an urgent need for automated solutions. However, existing automated layout models are ill-suited to spreadsheets, as they often (1) treat components as axis-aligned rectangles with continuous coordinates, overlooking the inherently discrete, grid-based structure of spreadsheets; and (2) neglect interrelated semantics, such as data dependencies and contextual links, unique to spreadsheets. In this paper, we first formalize the spreadsheet layout generation task, supported by a seven-criterion evaluation protocol and a dataset of 3,326 spreadsheets. We then introduce \textbf{SheetDesigner}, a zero-shot and training-free framework using Multimodal Large Language Models (MLLMs) that combines rule and vision reflection for component placement and content population. SheetDesigner outperforms five baselines by at least 22.6%. We further find that through vision modality, MLLMs handle overlap and balance well but struggle with alignment, necessitates hybrid rule and visual reflection strategies.

SheetDesigner: MLLM-Powered Spreadsheet Layout Generation with Rule-Based and Vision-Based Reflection

Materials science research requires multi-step reasoning and precise material informatics retrieval, where minor errors can propagate into significant failures in downstream experiments. Despite their general success, Large Language Models (LLMs) often struggle with hallucinations, handling domain-specific data effectively (e.g., crystal structures), and integrating experimental workflows. To address these challenges, we introduce LLaMP, a hierarchical multi-agent framework designed to emulate the materials science research workflow. The high-level supervisor agent decomposes user requests into sub-tasks and coordinates with specialized assistant agents. These assistant agents handle domain-specific tasks, such as retrieving and processing data from the Materials Project (MP) or conducting simulations as needed. This pipeline facilitates iterative refinement of material property retrieval and enables the simulation of real-world research workflows. To ensure reliability, we propose a novel metric combining uncertainty and confidence estimates to evaluate the self-consistency of responses from LLaMP and other methods. Our experiments with real-world materials science tasks demonstrate LLaMP’s superior performance in material property retrieval, crystal structure editing, and annealing molecular dynamics simulations using pre-trained interatomic potentials. Unlike prior work focused solely on material property prediction or discovery, LLaMP serves as a foundation for autonomous materials research by combining grounded informatics and enabling iterative experimental processes.

LLaMP: Large Language Model Made Powerful for High-fidelity Materials Knowledge Retrieval

Large visual-language models (LVLMs) have demonstrated remarkable performance in visual-language tasks. However, object hallucination remains a significant challenge for LVLMs. Existing studies attribute object hallucinations in LVLMs mainly to linguistic priors and data biases. We further explore the causes of object hallucinations from the perspective of frequency domain and reveal that insufficient frequency information in images amplifies these linguistic priors, increasing the likelihood of hallucinations. To mitigate this issue, we propose the Multi-Frequency Contrastive Decoding (MFCD) method, a simple yet training-free approach that removes the hallucination distribution in the original output distribution, which arises from LVLMs neglecting the high-frequency information or low-frequency information in the image input. Without compromising the general capabilities of LVLMs, the proposed MFCD effectively mitigates the object hallucinations in LVLMs. Our experiments demonstrate that MFCD significantly mitigates object hallucination across diverse large-scale vision-language models, without requiring additional training or external tools. In addition, MFCD can be applied to various LVLMs without modifying model architecture or requiring additional training, demonstrating its generality and robustness. Codes are available at https://anonymous.4open.science/r/mfcd

Multi-Frequency Contrastive Decoding: Alleviating Hallucinations for Large Vision-Language Models

Social media platforms provide an ideal environment to spread misinformation, where social bots can accelerate the spread. This paper explores the interplay between social bots and misinformation on the Sina Weibo platform. We construct a large-scale dataset that includes annotations for both misinformation and social bots. From the misinformation perspective, the dataset is multimodal, containing 11,393 pieces of misinformation and 16,416 pieces of verified information. From the social bot perspective, this dataset contains 65,749 social bots and 345,886 genuine accounts, annotated using a weakly supervised annotator. Extensive experiments demonstrate the comprehensiveness of the dataset, the clear distinction between misinformation and real information, and the high quality of social bot annotations. Further analysis illustrates that: (i) social bots are deeply involved in information spread; (ii) misinformation with the same topics has similar content, providing the basis of echo chambers, and social bots would amplify this phenomenon; and (iii) social bots generate similar content aiming to manipulate public opinions.

How Do Social Bots Participate in Misinformation Spread? A Comprehensive Dataset and Analysis

Languages vary along a wide variety of dimensions. In Natural Language Processing (NLP), it is useful to know how ``distant'' languages are from each other, so that we can inform NLP models about these differences or predict good transfer languages. Furthermore, it can inform us about how diverse language samples are. However, there are many different perspectives on how distances across languages could be measured, and previous work has predominantly focused on either intuition or a single type of distance, like genealogical or typological distance. %In fact, often intuition, or an extremely generic method like number of language family is used. Therefore, we propose DistaLs, a toolkit that is designed to provide users with easy access to a wide variety of language distance measures. We also propose a filtered subset, which contains less redundant and more reliable features. DistaLs is designed to be accessible for a variety of use cases, and offers a Python, CLI, and web interface. It is easily updateable, and available as a pip package. Finally, we provide a case-study in which we use DistaLs to measure correlations of distance measures with performance on four different morphosyntactic tasks.

Downloads

Next from EMNLP 2025

VC4VG: Optimizing Video Captions for Text-to-Video Generation

Similar lecture

LiPO: Listwise Preference Optimization through Learning-to-Rank

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES