China

Image recaptioning is widely used to generate training datasets with enhanced quality for various multimodal tasks. Existing recaptioning methods typically rely on powerful multimodal large language models (MLLMs) to enhance textual descriptions, but often suffer from inaccuracies due to hallucinations and incompleteness caused by missing fine-grained details. To address these limitations, we propose RICO, a novel framework that refines captions through visual reconstruction. Specifically, we leverage a text-to-image model to reconstruct a caption into a reference image, and prompt an MLLM to identify discrepancies between the original and reconstructed images to refine the caption. This process is performed iteratively, further progressively promoting the generation of more faithful and comprehensive descriptions. To mitigate the additional computational cost induced by the iterative process, we introduce RICO-Flash, which learns to generate captions like RICO using DPO. Extensive experiments demonstrate that our approach significantly improves caption accuracy and completeness, outperforms most baselines by approximately 10% on both CapsBench and CompreCap.

EMNLP 2025

RICO: Improving Accuracy and Completeness in Image Recaptioning via Visual Reconstruction

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."<br>

*Message from the General Chair, Dirk Hovy*

[**Link to Conference Handbook**](https://drive.google.com/file/d/1johU5QqVVYO4RfH7QcIORr7qrVBdzdwC/view?usp=sharing)





<br>

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

Understanding human intents from multimodal signals is critical for analyzing human behaviors and enhancing human-machine interactions in real-world scenarios. However, existing methods exhibit limitations in their modality-level reliance, constraining relational reasoning over fine-grained semantics for complex intent understanding. This paper proposes a novel LLM-Guided Semantic Relational Reasoning (LGSRR) method, which harnesses the expansive knowledge of large language models (LLMs) to establish semantic foundations that boost smaller models' relational reasoning performance. Specifically, an LLM-based strategy is proposed to extract fine-grained semantics as guidance for subsequent reasoning, driven by a shallow-to-deep Chain-of-Thought (CoT) that autonomously uncovers, describes, and ranks semantic cues by their importance without relying on manually defined priors. Besides, we formally model three fundamental types of semantic relations grounded in logical principles and analyze their nuanced interplay to enable more effective relational reasoning. Extensive experiments on multimodal intent and dialogue act recognition tasks demonstrate LGSRR's superiority over state-of-the-art methods, with consistent performance gains across diverse semantic understanding scenarios.

LLM-Guided Semantic Relational Reasoning for Multimodal Intent Recognition

Large Language Models have advanced significantly in complex reasoning, often leveraging external reward model to improve the reliability of their multi-step processes. However, existing process verification methods struggle with reliably assessing incomplete reasoning traces and are limited by the cost of high-quality human annotations or the inherent noise in automatically generated labels. Therefore, we present Dyve, a dynamic process verifier that enhances reasoning error detection in large language models by integrating fast and slow thinking, inspired by Kahneman's Systems Theory. Dyve adaptively applies immediate token-level confirmation (System 1) for straightforward steps and comprehensive analysis (System 2) for complex ones. Unlike traditional verifiers that only evaluate final outputs, Dyve employs a {step-wise consensus-filtered supervision} strategy, leveraging Monte Carlo estimation, LLM-as-a-Judge, and specialized reasoning models to extract high-quality training signals from noisy rollouts. Experimental results on ProcessBench and the MATH dataset confirm that Dyve significantly outperforms existing process-based verifiers and boosts performance in Best-of-N settings while maintaining computational efficiency by strategically allocating verification resources.

Dyve: Thinking Fast and Slow for Dynamic Process Verification

To alleviate the hallucination problem of large language model (LLM), retrieval-augmented generation (RAG) has been proposed and widely adopted. Due to the limitations in cross-chunk summarization task of naive RAG, graph-based RAG has emerged as a promising solution. However, a close study reveals several flaws in these works. First, most graph-based RAGs suffer from less efficient indexing process, which leads to information loss and expensive costs. Second, they heavily rely on LLM for retrieval thus inference slowly, which hinders their application in industry. To build a more efficient and effective RAG, we propose the multi-semantic RAG (MS-RAG). In this work, we combine knowledge graphs with dense vector to build a multi-semantic RAG. To be specific, (i) at indexing stage, we create multiple semantic-level indexes, including chunk-level, relation-level, and entity-level, to leverage the merits of dense vector and knowledge graph. (ii) at retrieval stage, unlike the previous LLM-empowered entity extraction, we propose a novel mix recall algorithm. Finally, we employ a multi-semantic rerank module to purify the results. Extensive experiments show that MS-RAG achieves superior performance. In terms of retrieval effect, MS-RAG achieves state-of-the-art performance, which is about 10%-30% improvement than the existing methods. In terms of question-answering effect, MS-RAG still achieves promising results with faster inference speed. More analysis and experiments are provided in Appendix.

MS-RAG: Simple and Effective Multi-Semantic Retrieval-Augmented Generation

Recently, large vision–language models (LVLMs) have emerged as the preferred tools for judging text–image alignment, yet their robustness along the visual modality remains underexplored. This work is the first study to address a key research question: Can adversarial visual manipulations systematically fool LVLM judges into assigning unfairly inflated scores? We define potential image-induced biases within the context of T2I evaluation and examine how these biases affect the evaluations of LVLM judges. Moreover, we introduce a novel, fine-grained, multi-domain meta-evaluation benchmark named FRAME, which is deliberately constructed to exhibit diverse score distributions. By introducing the defined biases into the benchmark, we reveal that all tested LVLM judges exhibit vulnerability across all domains, consistently inflating scores for manipulated images. Further analysis reveals that combining multiple biases amplifies their effects, and pairwise evaluations are similarly susceptible. Moreover, we observe that visual biases persist despite prompt-based mitigation strategies, highlighting the vulnerability of current LVLM evaluation systems and underscoring the urgent need for more robust LVLM judges.

Fooling the LVLM Judges: Visual Biases in LVLM-Based Evaluation

Large models achieve strong performance on Vision-and-Language Navigation (VLN) tasks but are costly to run in resource-limited environments. Token pruning, by reducing model input size, offers appealing tradeoffs for efficiency with minimal performance loss. However, existing pruning strategies, designed for general Vision-Language tasks, overlook VLN-specific challenges: input information loss makes navigation agents uncertain about their decisions and thereby their walks longer, increasing computational cost. Non-VLN-specific strategies often fail to recognize uninformative instruction tokens, thus undermining efficiency gains. We propose NAP, a navigation-aware pruning framework that addresses these issues from three angles: pruning only background view tokens while preserving action-relevant ones, removing low-importance nodes from the navigation map to discourage backtracking and thus shorten navigation length, and leveraging a Large Language Model to pre-identify instruction words irrelevant to navigation, which helps the agent prioritize pruning them during navigation. Experiments on standard VLN benchmarks show that NAP significantly outperforms prior pruning strategies, preserving higher success rates while saving more than 50% FLOPS.

Walk and Read Less: Improving the Efficiency of Vision-and-Language Navigation via Tuning-Free Multimodal Token Pruning

Ambiguity resolution is key to effective communication. While humans effortlessly address ambiguity through conversational grounding strategies, the extent to which current language models can emulate these strategies remains unclear. In this work, we examine referential ambiguity in image-based question answering by introducing RAcQUEt, a carefully curated dataset targeting distinct aspects of ambiguity. Through a series of evaluations, we reveal significant limitations and problems of overconfidence of state-of-the-art large multimodal language models in addressing ambiguity in their responses. The overconfidence issue becomes particularly relevant for RAcQUEt-bias, a subset designed to analyze a critical yet underexplored problem: failing to address ambiguity leads to stereotypical, socially biased responses. Our results underscore the urgency of equipping models with robust strategies to deal with uncertainty without resorting to undesirable stereotypes.

RAcQUEt: Unveiling the Dangers of Overlooked Referential Ambiguity in Visual LLMs

The rise of multimodal data, integrating text, audio, and visuals, has created new opportunities for studying multimodal tasks such as intent detection. This work investigates the effectiveness of Large Language Models (LLMs) and non-LLMs, including text-only and multimodal models, in the multimodal intent detection task. Our study reveals that Mistral-7B, a text-only LLM, outperforms most competitive multimodal models by approximately 9% on MIntRec-1 and 4% on MIntRec2.0 datasets. This performance advantage comes from a strong textual bias in these datasets, where over 90% of the samples require textual input, either alone or in combination with other modalities, for correct classification. We confirm the modality bias of these datasets via human evaluation, too. Next, we propose a framework to debias the datasets, and upon debiasing, more than 70% of the samples in MIntRec-1 and more than 50% in MIntRec2.0 get removed, resulting in significant performance degradation across all models, with smaller multimodal fusion models being the most affected with an accuracy drop of over 50 - 60%. Further, we analyze the context-specific relevancy of different modalities through empirical analysis. Our findings highlight the challenges posed by modality bias in multimodal intent datasets and emphasize the need for unbiased datasets to evaluate multimodal models effectively.

Text Takes Over: A Study of Modality Bias in Multimodal Intent Detection

This paper examines two aspects of the isolated sign language recognition (ISLR) task. First, although a certain number of datasets is available, the data for individual sign languages is limited. It poses the challenge of cross-language ISLR model training, including transfer learning. Second, similar signs can have different semantic meanings. It leads to ambiguity in dataset labeling and raises the question of the best policy for annotating such signs. To address these issues, this study presents Logos, a novel Russian Sign Language (RSL) dataset, the most extensive available ISLR dataset by the number of signers, one of the most extensive datasets in size and vocabulary, and the largest RSL dataset. It is shown that a model, pre-trained on the Logos dataset can be used as a universal encoder for other language SLR tasks, including few-shot learning. We explore cross-language transfer learning approaches and find that joint training using multiple classification heads benefits accuracy for the target low-resource datasets the most. The key feature of the Logos dataset is explicitly annotated visually similar sign groups. We show that explicitly labeling visually similar signs improves trained model quality as a visual encoder for downstream tasks. Based on the proposed contributions, we outperform current state-of-the-art results for the WLASL dataset and get competitive results for the AUTSL dataset, with a single stream model processing solely RGB video. The source code, dataset, and pre-trained models are publicly available.

Logos as a Well-Tempered Pre-train for Sign Language Recognition

Multimodal large language models (MLLMs) have demonstrated strong performance across diverse multimodal tasks, achieving promising outcomes. However, their application to emotion recognition in natural images remains under-explored. MLLMs struggle to handle ambiguous emotional expressions and implicit affective cues, whose capability is crucial for affective understanding but largely overlooked. To address these challenges, we propose MERMAID, a novel multi-agent framework that integrates an emotion-guided visual augmentation module and a multi-perspective self-reflection module, enabling agents to interact across modalities and reinforce subtle emotional semantics, thereby enhancing emotion recognition and achieving autonomy. Extensive experiments demonstrate that MERMAID outperforms existing methods, achieving up to 8.70%-27.90% absolute accuracy gains across diverse benchmarks and demonstrating greater robustness in emotionally diverse scenarios.

MERMAID: Multi-perspective Self-reflective Agents with Generative Augmentation for Emotion Recognition

Multimodal large language models (MLLMs) are increasingly deployed in open-ended, real-world environments where inputs are messy, underspecified, and not always trustworthy. Unlike curated benchmarks, these settings frequently involve instructions that refer to missing objects or contradictory facts, rely on ambiguous references, or request infeasible actions. In such cases, success hinges not on task execution alone, but on a model’s ability to detect when something is silently wrong. This paper presents a systematic analysis of how current MLLMs handle such implicit reasoning scenarios: cases where the flaw is not explicitly stated but must be inferred from context. Using a curated diagnostic suite spanning four categories of real-world failure modes, we evaluate six MLLMs, including o3 and GPT-4o, and find that models frequently fail to surface hidden issues, even when they possess the necessary perceptual and reasoning skills. Explicit prompting reveals that the underlying capabilities exist but are often suppressed in favor of user compliance. We further show that simple inference-time interventions, such as cautious persona prompting and, in particular, requiring a clarifying question, can dramatically recover performance. Our findings highlight a persistent gap between reasoning competence and behavioral compliance in current MLLMs, and suggest practical strategies for making these models more trustworthy in underconstrained environments.

Downloads

Next from EMNLP 2025

LLM-Guided Semantic Relational Reasoning for Multimodal Intent Recognition

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES