China

Despite growing interest in Theory of Mind (ToM) tasks for evaluating language models (LMs), little is known about how LMs internally represent mental states. Understanding these internal mechanisms is critical - not only to move beyond surface-level performance, but also for model alignment and safety, where subtle misattributions of mental states may go undetected in generated outputs. In this work, we present the first systematic investigation of belief representations in LMs by probing models across different scales, training regimens, and prompts - using control tasks to rule out confounds. Our experiments provide evidence that both model size and fine‑tuning substantially improve belief representations, which are structured - not mere by-products of higher dimensionality - yet brittle: even semantically neutral prompt variations can impair them. Crucially, we show that these representations can be strengthened: targeted edits to model activations can correct wrong ToM inferences.

EMNLP 2025

Brittle Minds, Fixable Activations: Understanding Belief Representations in Language Models

activation editing

theory of mind

large language models

probing

## Welcome to EMNLP 2024! 
We are excited to welcome you to one of the most prominent conferences in the field of Natural Language Processing. This year, EMNLP 2024 is being held in a hybrid format,
offering both virtual and in-person participation in beautiful Miami. Due to a record-breaking number of submissions, we've expanded the total number of accepted papers to accommodate more cutting-edge research from around the globe.
### [Conference Handbook](https://drive.google.com/file/d/1WPROgxjLAC96AJL7Ugy0tEnYm7dkrbHt/view?usp=sharing)

You are required to register for this event. **Please register [here](https://2024.emnlp.org/registration/).** The EMNLP 2024 event page on Underline will be open to public one week prior to the event.

Please register!

EMNLP 2024

EMNLP 2024 will take place in Miami, Florida from Nov 12th to Nov 16th, 2024, at the Hyatt Regency Miami Hote and on Underline for remote participants.

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.

Combining pre-trained expert models offers substantial potential for scalable multimodal reasoning, but building a unified framework remains challenging due to the increasing diversity of input modalities and task complexity. For instance, medical diagnosis requires precise reasoning over structured clinical tables, while financial forecasting depends on interpreting plot-based data to make informed predictions. To tackle this challenge, we introduce MEXA, a training-free framework that performs modality- and task-aware aggregation of multiple expert models to enable effective multimodal reasoning across diverse and distinct domains. MEXA dynamically selects expert models based on the input modality and the task-specific reasoning demands (i.e., skills). Each expert model, specialized in a modality task pair, generates interpretable textual reasoning outputs. MEXA then aggregates and reasons over these outputs using a Large Reasoning Model (LRM) to produce the final answer. This modular design allows flexible and transparent multimodal reasoning across diverse domains without additional training overhead. We extensively evaluate our approach on diverse multimodal benchmarks, including Video Reasoning, Audio Reasoning, 3D Understanding, and Medical QA. MEXA consistently delivers performance improvements over strong multimodal baselines, highlighting the effectiveness and broad applicability of our expert-driven selection and aggregation in diverse multimodal reasoning tasks.

MEXA: Towards General Multimodal Reasoning with Dynamic Multi-Expert Aggregation

Vision-language models (VLMs) often process visual inputs through a pretrained vision encoder, followed by a projection into the language model's embedding space via a connector component. While crucial for modality fusion, the potential information loss induced by this projection step and its direct impact on model capabilities remain understudied. We propose two novel approaches to quantify such visual information loss in the projection by analyzing the latent representation space. First, we evaluate semantic information preservation by analyzing changes in k-nearest neighbor relationships between image representations, before and after projection. Second, we directly measure information loss by reconstructing visual embeddings from the projected representation, localizing loss at an image patch level. Our experiments reveal that connectors fundamentally alter visual semantic relationships---k-nearest neighbors of the visual embeddings diverge by 40-60% post-projection, correlating highly with degradation in retrieval performance. The patch-level embedding reconstruction provides interpretable insights for model behavior on visual question-answering tasks, finding that areas of high information loss reliably predict instances where models struggle.

Lost in Embeddings: Information Loss in Vision–Language Models

We investigate the use of LLM-generated data for continual pretraining of transformer encoder models in specialized domains with limited training data, using the scientific domain of invasion biology as a case study. To this end, we leverage domain-specific ontologies by enriching them with LLM-generated data and pretraining the encoder model as an ontology-informed embedding model for concept definitions. To evaluate the effectiveness of this method, we compile a benchmark specifically designed for assessing model performance in invasion biology. After demonstrating substantial improvements over standard MLM pretraining, we investigate the feasibility of applying the proposed approach to domains without comprehensive ontologies by substituting ontological concepts with concepts automatically extracted from a small corpus of scientific abstracts and establishing relationships between concepts through distributional statistics. Our results demonstrate that this automated approach achieves comparable performance using only a small set of scientific abstracts, resulting in a fully automated pipeline for enhancing domain-specific understanding of small encoder models that is especially suited for application in low-resource settings and achieves performance comparable to masked language modeling pretraining on much larger datasets.

Enhancing Domain-Specific Encoder Models with LLM-Generated Data: How to Leverage Ontologies, and How to Do Without Them

The rapid use of large language models (LLMs) has raised critical concerns regarding the factual reliability of their outputs, especially in low-resource languages such as Urdu. Existing automated fact-checking solutions overwhelmingly focus on English, leaving a significant gap for the 200+ million Urdu speakers worldwide. In this work, we introduce UrduFactCheck, the first comprehensive, modular fact-checking framework specifically tailored for Urdu. Our system features a dynamic, multi-strategy evidence retrieval pipeline that combines monolingual and translation-based approaches to address the scarcity of high-quality Urdu evidence. We curate and release two new hand-annotated benchmarks: UrduFactBench for claim verification and UrduFactQA for evaluating LLM factuality. Extensive experiments demonstrate that UrduFactCheck, particularly its translation-augmented variants, consistently outperforms baselines and open-source alternatives on multiple metrics. We further benchmark twelve state-of-the-art (SOTA) LLMs on factual question answering in Urdu, highlighting persistent gaps between proprietary and open-source models. UrduFactCheck's code and datasets are open-sourced and publicly available at [URL redacted].

UrduFactCheck: An Agentic Fact-Checking Framework for Urdu with Evidence Boosting and Benchmarking

Understanding human social behavior such as recognizing emotions and the social dynamics causing them is an important and challenging problem. While LLMs have made remarkable advances, they are limited to the textual domain and cannot account for the major role that non-verbal cues play in understanding social situations. Vision Language Models (VLMs) can potentially account for this gap, however their ability to make correct inferences over such social cues has received little attention. In this paper, we explore the capabilities of VLMs at social reasoning. We identify a previously overlooked limitation in VLMs: the Visual Social-Pragmatic Inference gap. To target this gap, we propose a new task for VLMs: Visual Social-Pragmatic Inference. We construct a high quality dataset to test the abilities of a VLM for this task and benchmark the performance of several VLMs on it.

VIBE: Can a VLM Read the Room?

Vision-Language Models (VLMs), such as CLIP, have exhibited significant advancements in recognizing visual concepts through natural language guidance. However, adapting these models to downstream tasks remains challenging. Existing adaptation methods either overlook the structural knowledge between the text and image modalities or create overly complex graphs containing redundant information for alignment, leading to suboptimal classification performance and increased computational overhead. This paper proposes a novel adapter-tuning methodology named Homogeneous Graph Adapter (HomoGraphAdapter), which transforms diverse textual and visual descriptions into a unified set of node representations and establishes edges between nodes for inter-modal and cross-modal semantic alignment. We leverage a straightforward homogeneous Graph Neural Network (GNN) to adapt positive and negative classifiers across text and image modalities. The classifiers comprehensively enhance the performance for few-shot classification and OOD generalization. Compared with the SOTA approach HeGraphAdapter, HomoGraphAdapter improves classification accuracy by an average of 1.51% for 1-shot and 0.74% for 16-shot on 11 datasets, while also reducing both precomputation time and training time. The code will be released upon acceptance.

HomoGraphAdapter: A Homogeneous Graph Neural Network as an Effective Adapter for Vision-Language Models

The embodiment of emotional reactions from body parts contains rich information about our affective experiences. We propose a framework that utilizes state-of-the-art large vision language models (LVLMs) to generate Embodied LVLM Emotion Narratives (textbfELENA). These are well-defined, multi-layered text outputs, mainly comprising narrative-based descriptions focused on the salient body parts involved in emotional reaction. We also employ attention maps and observe that contemporary models exhibit a persistent bias towards the facial region. Despite this limitation, we observe that LVLMs can effectively recognize embodied emotions in masked images compared to baseline naive prompts. They achieve this without any fine-tuning when guided by the ELENA framework. ELENA charts a new trajectory for embodied-emotion analysis across the modality of vision and enriches modeling in an affect-aware setting.

Anatomy of a Feeling: Narrating Embodied Emotions via Large Vision-Language Models

Knowledge discovery from large-scale, heterogeneous textual corpora presents a significant challenge. Document clustering offers a practical solution by organizing unstructured texts into coherent groups based on content and thematic similarity. However, clustering does not inherently ensure thematic consistency. Here, we propose a novel framework that constructs a similarity graph over document embeddings and applies iterative graph-based clustering algorithms to partition the corpus into initial clusters. To overcome the limitations of conventional methods in producing semantically consistent clusters, we incorporate iterative feedback from a large language model (LLM) to guide the refinement process. The LLM is used to assess cluster quality and adjust edge weights within the graph, promoting better intra-cluster cohesion and inter-cluster separation. The LLM guidance is based on a set of success Rate metrics that we developed to measure the semantic coherence of clusters. Experimental results on multiple benchmark datasets demonstrate that the iterative process and additional user-supplied a priori edges improve the summaries' consistency and fluency, highlighting the importance of known connections among the documents. The removal of very rare or very frequent sentences has a mixed effect on the quality scores. Our full code is available here: \url{https://github.com/D2CS-sub/D2CS}

D2CS - Documents Graph Clustering using LLM supervision

The efficient implementation of large language models (LLMs) is crucial for deployment on resource-constrained devices. Low-rank tensor compression techniques, such as tensor-train (TT) networks, have been widely studied for over-parameterized neural networks. However, their applications to compress pre-trained LLMs for downstream tasks (post-training) remains challenging due to the high-rank nature of pre-trained LLMs and the lack of access to pretraining data. In this study, we investigate low-rank tensorized LLMs during fine-tuning and propose sparse augmented tensor networks (Saten) to enhance their performance. The proposed Saten framework enables full model compression. Experimental results demonstrate that Saten enhances both accuracy and compression efficiency in tensorized language models, achieving state-of-the-art performance.

Saten: Sparse Augmented Tensor Networks for Post-Training Compression of Large Language Models

LLMs have paved the way for truly simple document-level machine translation, but challenges such as omission errors remain. In this paper, we study a simple method for handling document-level machine translation, by leveraging previous contexts in a multi-turn conversational manner. Specifically, by decomposing documents into segments and iteratively translating them while maintaining previous turns, this method ensures coherent translations without additional training, and can fully re-use the KV cache of previous turns thus minimizing computational overhead. We empirically show this multi-turn method outperforms both translating entire documents in a single turn and translating each segment independently according to multiple automatic metrics in representative LLMs, establishing a strong baseline for document-level translation using LLMs.

Downloads

Next from EMNLP 2025

MEXA: Towards General Multimodal Reasoning with Dynamic Multi-Expert Aggregation

Similar lecture

Probing the Robustness of Theory of Mind in Large Language Models