China

Large Language Models (LLMs) can struggle to balance gullibility to misinformation and resistance to valid corrections in persuasive dialogues, a critical challenge for reliable deployment. We introduce **DuET-PD** (**Du**al **E**valuation for **T**rust in **P**ersuasive **D**ialogues), a framework evaluating multi-turn stance-change dynamics across dual dimensions: persuasion type (corrective/misleading) and domain (knowledge/safety), using MMLU-Pro and SALAD-Bench. With **DuET-PD**, we uncover a primacy effect in initial persuasion and a capability-robustness trade-off: capable models often resist valid corrections, especially in safety tasks, while open-source models show higher gullibility. To address this, we introduce Holistic DPO, a training approach balancing positive and negative persuasion examples. Unlike prompting or resist-only training, Holistic DPO enhances both robustness to misinformation and receptiveness to corrections. Our framework and quantitative insights, coupled with the Holistic DPO method, enable LLMs to better navigate persuasive dialogues, improving reliability in knowledge- and safety-critical contexts.

EMNLP 2025

Persuasion Dynamics in LLMs: Investigating Robustness and Adaptability in Knowledge and Safety with DuET-PD

technical paper

## 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.

Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities by integrating visual and textual inputs, yet modality alignment remains one of the most challenging aspects. Current MLLMs typically rely on simple adapter architectures and pretraining approaches to bridge vision encoders with large language models (LLM), guided by image-level supervision. We identify this paradigm often leads to suboptimal alignment between modalities, significantly constraining the LLM's ability to properly interpret and reason with visual features particularly for smaller language models. To address this fundamental limitation, we propose Supervised Embedding Alignment (SEA), a token-level supervision alignment method that enables more precise visual-text alignment during pretraining. SEA introduces minimal computational overhead while preserving language capabilities and substantially improving cross-modal understanding. Our comprehensive analyses reveal critical insights into the adapter's role in multimodal integration, and extensive experiments demonstrate that SEA consistently improves performance across various model sizes, with smaller models benefiting the most (average performance gain of 7.61\% for Gemma-2B). This work establishes a foundation for developing more effective alignment strategies for future multimodal systems.

SEA: Supervised Embedding Alignment for Token-Level Visual-Textual Integration in MLLMs

To serve global users safely and productively, LLMs need culture-specific knowledge that might not be learned during pre-training. How do we find such knowledge that is (1) salient to in-group users, but (2) unknown to LLMs? The most common solutions are single-initiative: either researchers define challenging questions that users passively answer (traditional annotation), or users actively produce data that researchers structure as benchmarks (knowledge extraction). The process would benefit from mixed-initiative collaboration, where users guide the process to meaningfully reflect their cultures, and LLMs steer the process towards more challenging questions that meet the researcher's goals. We propose a mixed-initiative methodology called CultureCartography. Here, an LLM initializes annotation with questions for which it has low-confidence answers, making explicit both its prior knowledge and the gaps therein. This allows a human respondent to fill these gaps and steer the model towards salient topics through direct edits. We implement this methodology as a tool called CultureExplorer. Compared to a baseline where humans answer LLM-proposed questions, we find that CultureExplorer more effectively produces knowledge that leading models like DeepSeek R1 and GPT-4o are missing, even with web search. Fine-tuning on this data boosts the accuracy of Llama-3.1-8B by up to 19.2% on related culture benchmarks.

Culture Cartography: Mapping the Landscape of Cultural Knowledge

Reasoning language models (RLMs) excel at complex tasks by leveraging a chain-of-thought process to generate structured intermediate steps. However, language mixing, i.e., reasoning steps containing tokens from languages other than the prompt, has been observed in their outputs and shown to affect performance, though its impact remains debated. We present the first systematic study of language mixing in RLMs, examining its patterns, impact, and internal causes across 15 languages, 7 task difficulty levels, and 18 subject areas, and show how all three factors influence language mixing. Moreover, we demonstrate that the choice of reasoning language significantly affects performance: forcing models to reason in Latin or Han scripts via constrained decoding notably improves accuracy. Finally, we show that the script composition of reasoning traces closely aligns with that of the model’s internal representations, indicating that language mixing reflects latent processing preferences in RLMs. Our findings provide actionable insights for optimizing multilingual reasoning and open new directions for reasoning language control to build more interpretable and adaptable RLMs.

Language Mixing in Reasoning Language Models: Patterns, Impact, and Internal Causes

Large Reasoning Models (LRMs) generate explicit reasoning traces alongside final answers, yet the extent to which these traces influence answer generation remains unclear. In this work, we conduct a three-stage investigation into the interplay between reasoning and answer generation in three distilled DeepSeek R1 models. First, through empirical evaluation, we demonstrate that including explicit reasoning consistently improves answer quality across diverse domains. Second, attention analysis reveals that answer tokens attend substantially to reasoning tokens, with certain mid-layer attention heads closely tracking the reasoning trajectory, including self-reflective cues. Third, we apply mechanistic interventions using activation patching to assess the dependence of answer tokens on reasoning activations. Our results show that perturbations to key reasoning tokens can reliably alter the final answers, confirming a directional and functional flow of information from reasoning to answer. These findings deepen our understanding of how LRMs leverage reasoning tokens for answer generation, highlighting the functional role of intermediate reasoning in shaping model outputs.

From Reasoning to Answer: Empirical, Attention-Based and Mechanistic Insights into Distilled DeepSeek R1 Models

With thousand years of history, calligraphy serve as one of the representative symbols of Chinese culture. Increasing works try to digitize calligraphy by recognizing the context of calligraphy for better preservation and propagation. However, previous works stick to isolated single character recognition, not only requires unpractical manual splitting into characters, but also abandon the enriched context information that could be supplementary. To this end, we construct the pioneering end-to-end calligraphy recognition benchmark dataset, this dataset is challenging due to both the visual variations such as different writing styles and the textual understanding such as the domain shift in semantics. We further propose CalligraphicOCR (COCR) equipped with calligraphic image augmentation and action-based corrector targeted at the challenging root of this setting. Experiments demonstrate the advantage of our proposed model over cutting-edge baselines, underscoring the necessity of introducing this new setting, thereby facilitating a solid precondition for protecting and propagating the already scarce resources.

CalligraphicOCR for Chinese Calligraphy Recognition

Parameter-Efficient Fine-Tuning (PEFT) methods, such as LoRA, significantly reduce the number of trainable parameters by introducing low-rank decomposition matrices. However, existing methods perform extensive matrix multiplications in domain specialization tasks, resulting in computational inefficiency and sub-optimal fine-tuning performance. Hence, we propose LoSiA (**Lo**w-Resources **S**ubnet **I**ntegration **A**daptation), an innovative method that dynamically localizes and optimizes critical parameters during the training process. Specifically, it identifies a sub-network using gradient sparsity analysis and optimizes it as the trainable target. This design enables effective high-rank adaptation by updating only the sub-network parameters, reducing the additional matrix multiplication. We also present LoSiA-Pro, a faster implementation of LoSiA, which reduces the training latency by about 27% compared to LoRA. Extensive evaluations show that our method achieves minimal performance drop compared to full fine-tuning, while requiring the least training time across domain specialization and common-sense reasoning tasks. Further analysis shows that LoSiA also reduces forgetting during continued training. The source code will be publicly available.

LoSiA: Efficient High-Rank Fine-Tuning via Subnet Localization and Optimization

The reasoning capabilities of large language reasoning models (LRMs), such as OpenAI's o1 and DeepSeek-R1, have seen substantial advancements through deep thinking. However, these enhancements come with significant resource demands, underscoring the need for training effective small reasoning models. A critical challenge is that small models possess different reasoning capacities and cognitive trajectories compared with their larger counterparts. Hence, directly distilling chain-of-thought (CoT) results from large LRMs to smaller ones can sometimes be ineffective and often requires a substantial amount of annotated data. In this paper, we first introduce a novel Critique-Rethink-Verify (CRV) system, designed for training smaller yet powerful LRMs. Our CRV system consists of multiple LLM agents, each specializing in unique abilities: (i) critiquing the CoT qualities according to the cognitive capabilities of smaller models, (ii) rethinking and refining these CoTs based on the critiques, and (iii) verifying the correctness of the refined results. Based on the CRV system, we further propose the Cognitive Preference Optimization (CogPO) algorithm to continuously enhance the reasoning abilities of smaller models by aligning their reasoning processes with their cognitive capacities. Comprehensive evaluations on challenging reasoning benchmarks demonstrate the efficacy of our CRV+CogPO framework, which outperforms other methods by a large margin.

Enhancing Reasoning Abilities of Small LLMs with Cognitive Alignment

Intervening the internal activations of large language models (LLMs) provides an effective inference-time alignment approach to mitigate undesirable behaviors, such as generating erroneous or harmful content, thereby ensuring safe and reliable applications of LLMs. However, previous methods neglect the misalignment discrepancy among varied tokens, resulting in deviant alignment direction and inflexible editing strength. To address these issues, we propose a token-aware editing (TAE) approach to fully utilize token-level alignment information in the activation space, therefore realizing superior post-intervention performance. Specifically, a Mutual Information-guided Graph Aggregation (MIG) module first develops an MI-guided graph to exploit the tokens' informative interaction for activation enrichment, thus improving alignment probing and facilitating intervention. Subsequently, Misalignment-aware Adaptive Intervention (MAI) comprehensively perceives the token-level misalignment degree from token representation and prediction to guide the adaptive adjustment of editing strength, thereby enhancing final alignment performance. Extensive experiments on three alignment capabilities demonstrate the efficacy of TAE, notably surpassing baseline by 25.8% on the primary metric of truthfulness with minimal cost.

Token-Aware Editing of Internal Activations for Large Language Model Alignment

Large language models (LLMs) are increasingly integrated into users’ daily lives, leading to a growing demand for personalized outputs. Previous work focuses on leveraging a user's own history, overlooking inter-user differences that are crucial for effective personalization. While recent work has attempted to model such differences, the reliance on language-based prompts often hampers the effective extraction of meaningful distinctions. To address these issues, we propose Difference-aware Embedding-based Personalization (DEP), a framework that models inter-user differences in the latent space instead of relying on language prompts. DEP constructs soft prompts by contrasting a user’s embedding with those of peers who engaged with similar content, highlighting relative behavioral signals. A sparse autoencoder then filters and compresses both user-specific and difference-aware embeddings, preserving only task-relevant features before injecting them into a frozen LLM. Experiments on personalized review generation show that \ours consistently outperforms baseline methods across multiple metrics. Our code is available on an \href{https://anonymous.4open.science/r/DEP-A8A0}{Anonymous GitHub}.

Latent Inter-User Difference Modeling for LLM Personalization

Large language models (LLMs) have shown promise in clinical decision making, but current approaches struggle to localize and correct errors at specific steps of the reasoning process. This limitation is critical in medicine, where identifying and addressing reasoning errors is essential for accurate diagnosis and effective patient care. We introduce Med-PRM, a process reward modeling framework that leverages retrieval-augmented generation (RAG) to verify each reasoning step against established medical knowledge bases. By verifying intermediate reasoning steps with evidence retrieved from clinical guidelines and literature, our model can precisely assess the reasoning quality in a fine-grained manner. Evaluations on five medical QA benchmarks and two open-ended diagnostic tasks demonstrate that Med-PRM achieves state-of-the-art performance. We improve the performance of the base model by 6.95% using Med-PRM. Moreover, we demonstrate the generality of our reward model by integrating it in a plug-and-play fashion with strong policy models such as Meerkat, achieving over 80% accuracy on MedQA for the first time using small-scale models with fewer than 10 billion parameters.

Downloads

Next from EMNLP 2025

SEA: Supervised Embedding Alignment for Token-Level Visual-Textual Integration in MLLMs

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES