China

Answering complex real-world questions requires step-by-step retrieval and integration of relevant information to generate well-grounded responses. However, existing knowledge distillation methods struggle to effectively transfer step-specific reasoning abilities, as they fail to account for the variation in accessible information and learning difficulty across reasoning steps. To address this, we propose Step-wise Knowledge Distillation for Enhancing Reasoning Ability in Multi-Step Retrieval-Augmented Language Models (StepER). StepER employs step-wise supervision to align with evolving information and reasoning demands across stages. Additionally, it incorporates difficulty-aware training to progressively optimize learning by prioritizing suitable steps. Our method is highly adaptable across various frameworks of multi-step retrieval-augmented language models, including those based on reasoning paths or question decomposition. Extensive experiments show that StepER outperforms prior methods on multi-hop QA benchmarks, with an 8B model achieving performance comparable to a 70B teacher model.

EMNLP 2025

StepER: Step-wise Knowledge Distillation for Enhancing Reasoning Ability in Multi-Step Retrieval-Augmented Language Models

retrieval-augmented generation

multi-hop qa

knowledge distillation

### 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**. 

ACL 2025

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

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.

It has been demonstrated that incorporating external information as textual modality can effectively improve time series forecasting accuracy. However, current multi-modal models ignore the dynamic and different relations between time series patterns and textual features, which leads to poor performance in temporal-textual feature fusion. In this paper, we propose a lightweight and model-agnostic temporal-textual fusion framework named Cross-MoE. It replaces Cross Attention with Cross-Ranker to reduce computational complexity, and enhances modality-aware correlation memorization with Mixture-of-Experts (MoE) networks to tolerate the distributional shifts in time series. The experimental results demonstrate a 8.5% average reduction in Mean Squared Error (MSE) compared to the SOTA multi-modal time series framework. Notably, our method requires only 75% of computational overhead and 12.5% of memory usage compared with Cross Attention mechanism.

Cross-MoE: An Efficient Temporal Prediction Framework Integrating Textual Modality

Taxonomies play a crucial role in helping researchers structure and navigate knowledge in a hierarchical manner. They also form an important part in the creation of comprehensive literature surveys. None of the existing approaches to automatic survey generation compare the structure of the generated surveys with those written by human experts. To address this gap, we present our own method for automated taxonomy creation that can bridge the gap between human-generated and automatically-created taxonomies. For this purpose, we create the CS-TaxoBench benchmark which consists of 460 taxonomies that have been extracted from human-written survey papers. We propose TaxoAlign, a three-phase topic-based instruction-guided method for scholarly taxonomy generation. Additionally, we propose a stringent automated evaluation framework that measures the structural alignment and semantic coherence of automatically generated taxonomies in comparison to those created by human experts. We evaluate our method and various baselines on CS-TaxoBench, using both automated evaluation metrics and human evaluation studies. The results show that TaxoAlign consistently surpasses the baselines on nearly all metrics.

TaxoAlign: Scholarly Taxonomy Generation Using Language Models

Unified vision large language models (VLLMs) have recently achieved impressive advancements in both multimodal understanding and generation, powering applications such as visual question answering and text-guided image synthesis. However, progress in unified VLLMs remains constrained by the lack of datasets that fully exploit the synergistic potential between these two core abilities. Existing datasets typically address understanding and generation in isolation, thereby limiting the performance of unified VLLMs. To bridge this critical gap, we introduce a novel dataset construction framework, \textbf{UnifiedVisual}, and present \textbf{UnifiedVisual-240K}, a high-quality dataset meticulously designed to facilitate mutual enhancement between multimodal understanding and generation. UnifiedVisual-240K seamlessly integrates diverse visual and textual inputs and outputs, enabling comprehensive cross-modal reasoning and precise text-to-image alignment. Our dataset encompasses a wide spectrum of tasks and data sources, ensuring rich diversity and addressing key shortcomings of prior resources. Extensive experiments demonstrate that models trained on UnifiedVisual-240K consistently achieve strong performance across a wide range of tasks. Notably, these models exhibit significant mutual reinforcement between multimodal understanding and generation, further validating the effectiveness of our framework and dataset. We believe UnifiedVisual represents a new growth point for advancing unified VLLMs and unlocking their full potential.

UnifiedVisual: A Framework for Constructing Unified Vision-Language Datasets

Translating natural language into formal language such as Lean 4 has gained attention for its potential to automate formal proof development. Automated methods provide a scalable and cost-effective alternative to manual formalization, driving increasing interest in this task. However, existing LLMs mainly rely on instruction tuning and lack fine-grained structural and semantic alignment, making it difficult to generate syntactically and logically sound formal proofs.To address this, we propose a reinforcement learning framework that enhances the ability of LLMs to generate high-quality Lean 4 statement from natural language. We first fine-tune a LLaMA3-8B model on NL–Lean 4 data to obtain a base translator with basic translating ability. Then, we design a multi-aspect dense reward mechanism covering four key dimensions: semantic alignment, tactic-level alignment, global-level alignment, and compile-checking. Separate reward models are trained via preference modeling, and their normalized outputs are combined to guide optimization via PPO. Finally, a curriculum learning strategy based on multi-dimensional difficulty allows the model to learn progressively from simple to complex cases. Experiments on NL-to-Lean 4 tasks show that our method consistently outperforms baseline models. Further analysis on reward model and curriculum learning confirms their effectiveness in enhancing model performance. The code will be released upon publication.

NL2Lean: Translating Natural Language into Lean 4 through Multi-Aspect Reinforcement Learning

Large Vision-Language Models (L-VLMs) have demonstrated remarkable performance in various vision and language tasks, including Visual Question Answering (VQA). However, their high computational cost makes them impractical for resource-constrained settings and inference-heavy applications. In contrast, Small Vision-Language Models (S-VLMs) offer efficiency but suffer from a significant performance gap compared to their larger counterparts. In this work, we introduce the Model Parity Aligner (MPA), a novel framework designed to systematically improve S-VLMs by leveraging unlabeled images and effective knowledge transfer from L-VLMs. Instead of traditional knowledge distillation methods that rely on labeled training data, MPA employs a strategic parity-based approach that precisely identifies the knowledge disparities between S-VLMs and L-VLMs, and optimizes training by targeting only these disparities. We conduct extensive experiments on four diverse VQA benchmarks, namely TextVQA, ST-VQA, ChartQA, and OKVQA, each of which required specialized reasoning capabilities such as text recognition, chart interpretation, and commonsense and factual understanding. Our results demonstrate that MPA consistently enhances the performance of S-VLM on all benchmarks, reducing the performance gap while maintaining computational efficiency. We shall make our code and MPA-aligned models publicly available upon acceptance of this work.

When Big Models Train Small Ones: Label-Free Model Parity Alignment for Efficient Visual Question Answering using Small VLMs

Large vision-language models (LVLMs) have witnessed significant progress on visual understanding tasks. However, they often prioritize language knowledge over image information on visual reasoning tasks, incurring performance degradation. To tackle this issue, we first identify the drawbacks of existing solutions (i.e., limited multi-modal reasoning capacities, and insufficient and irrelevant visual descriptions). We then decompose visual reasoning process into two stages: proactive visual perception (i.e., eyesight) and textual reasoning (i.e., wisdom), and introduce a novel visual reasoning framework named ProReason. This framework features decoupled vision-reasoning capabilities and multi-run proactive perception. Briefly, given a multi-modal question, ProReason iterates proactive information collection and reasoning until the answer can be concluded with necessary and sufficient visual descriptions. Notably, the disassociation of capabilities allows seamless integration of existing large language models (LLMs) to compensate for the reasoning deficits of LVLMs. Our extensive experiments demonstrate that ProReason outperforms existing multi-step reasoning frameworks on various benchmarks for both open-source and closed-source models, with the average performance gain reaching 13.2%. Besides, the integration of LLMs allows ProReason to produce high-quality visual reasoning data, which empowers ProReason-distilled models (i.e., ProReason-VL and ProReason-Q3) to achieve superior performance in downstream tasks. Our insights into existing solutions and the decoupled perspective for feasible integration of LLMs illuminate future research on visual reasoning techniques, especially LLM-assisted ones.

ProReason: Multi-Modal Proactive Reasoning with Decoupled Eyesight and Wisdom

Automatically assessing handwritten mathematical solutions is an important problem in educational technology with practical applications, but remains a significant challenge due to the diverse formats, unstructured layouts, and symbolic complexity of student work. To address this challenge, we introduce VEHME-a textbfVision-Language Model for textbfEvaluating textbfHandwritten textbfMathematics textbfExpressions—designed to assess open-form handwritten math responses with high accuracy and interpretable reasoning traces. VEHME integrates a two-phase training pipeline: (i) supervised fine-tuning using structured reasoning data, and (ii) reinforcement learning that aligns model outputs with multi-dimensional grading objectives, including correctness, reasoning depth, and error localization. To enhance spatial understanding, we propose an Expression-Aware Visual Prompting Module, trained on our synthesized multi-line math expressions dataset to robustly guide attention in visually heterogeneous inputs. Evaluated on AIHub and FERMAT datasets, VEHME achieves state-of-the-art performance among open-source models and approaches the accuracy of proprietary systems, demonstrating its potential as a scalable and accessible tool for automated math assessment.

VEHME: A Vision-Language Model For Evaluating Handwritten Mathematics Expressions

Video-language models (Video-LLMs) excel at understanding video content but struggle with spatial relationships, temporal ordering, and cross-frame continuity. To address these limitations, we introduce textbfVideoPASTA (textbfPreference textbfAlignment with textbfSpatio-textbfTemporal-Cross Frame textbfAdversaries), a framework that enhances Video-LLMs through targeted preference optimization. VideoPASTA trains models to distinguish accurate video representations from carefully crafted adversarial examples that deliberately violate spatial, temporal, or cross-frame relationships. With only 7,020 preference pairs and Direct Preference Optimization, VideoPASTA enables models to learn robust representations that capture fine-grained spatial details and long-range temporal dynamics. Experiments demonstrate that VideoPASTA is model agnostic and significantly improves performance, for example, achieving gains of up to 3.8\% on LongVideoBench, 4.1\% on VideoMME, and 4.0\% on MVBench, when applied to various state-of-the-art Video-LLMs. These results demonstrate that targeted alignment, rather than massive pretraining or architectural modifications, effectively addresses core video-language challenges. Notably, VideoPASTA achieves these improvements without any human annotation or captioning, relying solely on 32-frame sampling. This efficiency makes our approach a scalable plug-and-play solution that seamlessly integrates with existing models while preserving their original capabilities.

VideoPASTA: 7K Preference Pairs That Matter for Video-LLM Alignment

Large Language Models (LLMs) often suffer from mode collapse, repeatedly generating the same few completions even when many valid answers exist, limiting their diversity across a wide range of tasks. We introduce \textbf{Group-Aware Policy Optimization (GAPO)}, a simple extension of the recent and popular Group Relative Policy Optimization (GRPO) that computes rewards over the group as a whole. GAPO enables learning from the group-level properties such as diversity and coverage. We demonstrate GAPO using a frequency-aware reward function that encourages uniform sampling over valid LLM completions, and show that GAPO-trained models produce valid and more diverse model responses. Beyond this setup, GAPO generalizes to open-ended prompts and improves response diversity without compromising accuracy on standard LLM benchmarks (GSM8K, MATH, HumanEval, MMLU-Pro). Our code will be made publicly available.

Group-Aware Reinforcement Learning for Output Diversity in Large Language Models

Multi-modal entity alignment (MMEA) aims to identify equivalent entities between two multimodal knowledge graphs (MMKGs). However, the intrinsic noise within modalities, such as the inconsistency in visual modality and redundant attributes, has not been thoroughly investigated. Excessive noise not only weakens semantic representation but also increases the risk of overfitting in attention-based fusion methods. To address this, we propose LGEA, a novel LLMguided MMEA framework that prioritizes noise reduction before fusion. Specifically, LGEA introduces two key strategies: (1) fine-grained visual filtering to remove irrelevant images at the semantic level, and (2) contextual summarization of attribute information to enhance entity semantics. To our knowledge, we are the first work to apply LLMs for both visual filtering and attribute-level semantic enhancement in MMEA. Experiments on multiple benchmarks, including the noisy FB YG dataset, show that LGEA sets a new state-of-the-art (SOTA) in robust multi-modal alignment, highlighting the potential of noise-aware strategies as a promising direction for future MMEA research.

Downloads

Next from EMNLP 2025

Cross-MoE: An Efficient Temporal Prediction Framework Integrating Textual Modality

Similar lecture

LlamaV-o1: Rethinking Step-by-step Visual Reasoning in LLMs