China

Despite significant advancements in Large Vision Language Models (LVLMs), a gap remains, particularly regarding their interpretability and how they locate and interpret textual information within images. In this paper, we explore various LVLMs to identify the specific heads responsible for recognizing text from images, which we term the Optical Character Recognition Head (\textbf{OCR Head}). Our findings regarding these heads are as follows: (1) Less Sparse: Unlike previous retrieval heads, a large number of heads are activated to extract textual information from images. (2) Qualitatively Distinct: OCR heads possess properties that differ significantly from general retrieval heads, exhibiting low similarity in their characteristics. (3) Statically Activated: The frequency of activation for these heads closely aligns with their OCR scores. We validate our findings in downstream tasks by applying Chain-of-Thought (CoT) to both OCR and conventional retrieval heads and by masking these heads. We also demonstrate that redistributing sink-token values within the OCR heads improves performance. These insights provide a deeper understanding of the internal mechanisms LVLMs employ in processing embedded textual information in images.

EMNLP 2025

How Do Large Vision-Language Models See Text in Image? Unveiling the Distinctive Role of OCR Heads

multimodal

interpretability

It is my great pleasure and honor to welcome you to the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026), held in San Diego, California, USA, July 2–7, 2026. ACL 2026 continues our fields 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.

It is exciting to see the many developments in the field reflected at our conference, such as the advancement of large language models (born out of our community) into many application directions. This is not without challenges, partly due to unsolved problems, and partly due to misuse of the technology. All this is reflected well at ACL 2026. The conference also reflects that the field is not only about advancing technology, but also the science of fundamentally understanding language and how it is use by many different speaker communities. 
[Full Message from General Chair Philipp Koehn](https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing) 
<html><button style="display: inline-flex; align-items: center; justify-content: center; white-space: nowrap; border-radius: 9999px; font-weight: bold; background: #7c3aed; color: white; font-family: 'Space Grotesk', sans-serif; height: 40px; font-size: 16px; padding: 0 20px; border: none; cursor: pointer" onclick="window.open('https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing')">Conference Handbook</button></html>

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 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://2026.aclweb.org/registration/) first.

**Online Registration Form**: https://acl.swoogo.com/acl2026

Registration Required

ACL 2026

The 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026) will take place in San Diego, California, from July 2 - 7, 2026.

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." 

*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.

Word-level quality estimation (WQE) aims to automatically identify fine-grained error spans in machine-translated outputs and has found many uses, including assisting translators during post-editing. Modern WQE techniques are often expensive, involving prompting of large language models or ad-hoc training on large amounts of human-labeled data. In this work, we investigate efficient alternatives exploiting recent advances in language model interpretability and uncertainty quantification to identify translation errors from the inner workings of translation models. In our evaluation spanning 14 metrics across 12 translation directions, we quantify the impact of human label variation on metric performance by using multiple sets of human labels. Our results highlight the untapped potential of unsupervised metrics, the shortcomings of supervised methods when faced with label uncertainty, and the brittleness of single-annotator evaluation practices.

Unsupervised Word-level Quality Estimation for Machine Translation Through the Lens of Annotators (Dis)agreement

We report on investigations into the characteristics of outgoing connections in feedforward layers of large language models. Our findings show that inner neurons with diverse outgoing connection strengths contribute more significantly to model performance than those with uniform connections. We propose a new fine-tuning loss that takes advantage of this observation by decreasing the outgoing connection entropy in feedforward layers. Using this loss yields gains over standard fine-tuning across two different model families (PaLM-2 and Gemma-2) for downstream tasks in math, coding, and language understanding. To further elucidate the role of outgoing connection heterogeneity, we develop a data-free structured pruning method, which uses entropy to identify and remove neurons. This method significantly surpasses the effectiveness of random and even magnitude-based neuron removal.

The Role of Outgoing Connection Heterogeneity in Feedforward Layers of Large Language Models

Understanding natural language commands in situated Human-Robot Interaction (HRI) requires linking linguistic input to perceptual context. Traditional symbolic parsers lack the flexibility to operate in complex, dynamic environments. We introduce a novel Multimodal Grounded Semantic Role Labelling (G-SRL) framework that combines frame semantics with perceptual grounding, enabling robots to interpret commands via multimodal logical forms. Our approach leverages modern Visual Language Models (VLLMs), which jointly process text and images, and is supported by an automated pipeline that generates high-quality training data. Structured command annotations are converted into photorealistic scenes via LLM-guided prompt engineering and diffusion models, then rigorously validated through object detection and visual question answering. The pipeline produces over 11,000 image-command pairs (3,500+ manually validated), while approaching the quality of manually curated datasets at significantly lower cost.

Grounded Semantic Role Labelling from Synthetic Multimodal Data for Situated Robot Commands

What can we learn from network data representing concepts and their relations as perceived by students who reflect on their learning from a textbook via think-aloud data about students’ learning processes, knowledge acquisition, and learning outcomes? This study brings methods from entity and relation extraction using classic and LLM-based methods to the application domain of educational psychology. We built a ground-truth baseline of relational data that represent relevant (to educational science), textbook-based information as a semantic network. We identified SPN4RE and LUKE as the most accurate method to extracting semantic networks capturing the same types of information from transcriptions of verbal student data. Correlating the students' semantic networks with learning outcomes showed that students’ verbalizations varied in structure, reflecting different learning processes. Denser and more interconnected semantic networks indicated more elaborated knowledge acquisition. Structural features such as the number of edges and surface overlap with textbook networks significantly correlated with students’ posttest performance.

Semantic Networks Extracted from Students' Think-Aloud Data are Correlated with Students' Learning Performance

Pretrained language models (LMs) are prone to arithmetic errors. Existing work showed limited success in probing numeric values from models’ representations, indicating that these errors can be attributed to the inherent unreliability of distributionally learned embeddings in representing exact quantities. However, we observe that previous probing methods are inadequate for the emergent structure of learned number embeddings with sinusoidal patterns. In response, we propose a novel probing technique that decodes numeric values from input embeddings with near-perfect accuracy across a range of open-source LMs. This proves that after the sole pre-training, LMs represent numbers with remarkable precision. Finally, we find that the embeddings’ preciseness judged by our probe’s accuracy explains a large portion of LM’s errors in elementary arithmetic, and show that aligning the embeddings with the pattern discovered by our probe can mitigate these errors.

Pre-trained Language Models Learn Remarkably Accurate Representations of Numbers

This research investigates how to improve machine translation systems for low-resource languages by integrating loanword constraints as external linguistic knowledge. Focusing on the Portuguese-Emakhuwa language pair, which exhibits significant lexical borrowing, we address the challenge of effectively adapting loanwords during the translation process. To tackle this, we propose a novel approach that augments source sentences with loanword constraints, explicitly linking source-language loanwords to their target-language equivalents. Then, we perform supervised fine-tuning on multilingual neural machine translation models and multiple Large Language Models of different sizes. Our results demonstrate that incorporating loanword constraints leads to significant improvements in translation quality as well as in handling loanword adaptation correctly in target languages, as measured by different machine translation metrics. This approach offers a promising direction for improving machine translation performance in low-resource settings characterized by frequent lexical borrowing.

 Leveraging Loanword Constraints for Improving Machine Translation in a Low-Resource Multilingual Context

Large language models (LLMs) are increasingly used in the creation of online content, creating feedback loops as subsequent generations of models will be trained on this synthetic data. Such loops were shown to lead to \textit{distribution shifts} - models misrepresenting the true underlying distributions of human data (also called \textit{model collapse}). However, how human data properties affect such shifts remains poorly understood. In this paper, we provide the first empirical examination of the effect of such properties on the outcome of recursive training. We first confirm that using different human datasets leads to distribution shifts of different magnitudes. Through exhaustive manipulation of dataset properties combined with regression analyses, we then identify a set of properties predicting distribution shift magnitudes. Lexical diversity is found to amplify these shifts, while semantic diversity and data quality mitigate them. Furthermore, we find that these influences are highly modular: data scrapped from a given internet domain has little influence on the content generated for another domain. Finally, experiments on political bias reveal that human data properties affect whether the initial bias will be amplified or reduced. Overall, our results portray a novel view, where different parts of internet may undergo different types of distribution shift.

Recursive Training Loops in LLMs: How training data properties modulate distribution shift in generated data?

Detecting human morals and values in language is essential for interpreting both human and machine communication and for developing AI systems aligned with human values. However, this task remains challenging due to the variety of theoretical frameworks and the resulting diversity in available datasets. This work contributes MoVa, a set of resources a set of resources that define and benchmark generalizable classification, consisting of: 1) 16 labelled datasets across 4 theoretically-grounded moral and value frameworks; 2) a lightweight LLM prompting strategy that outperforms fine-tuned models in multiple different domains; and 3) a new application that facilitates a detailed design of psychological questionnaires. One methodological insight from MoVa, defined as all@once, is to classify a set of related concepts simultaneously with LLMs, further explained via an evaluation resembling the well-known multi-label classifier chain. In sum, MoVa provides valuable resources for developing and evaluating generalizable text classification and aligning LLMs with human values.

MoVa: Towards Generalizable Classification of Human Morals and Values

Large language models (LLMs) are used in various applications. However, despite its wide capabilities, it is shown to lack cultural alignment \citep{ryan-etal-2024-unintended, alkhamissi-etal-2024-investigating} and produce biased generations \cite{naous-etal-2024-beer} due to a lack of cultural knowledge and competence. Evaluation of LLMs for cultural awareness and alignment is particularly challenging due to the lack of proper evaluation metrics and the unavailability of culturally grounded datasets representing the vast complexity of cultures at the regional and sub-regional levels. Existing datasets for culture-specific items (CSIs) focus primarily on concepts at regional levels and contain several inconsistencies regarding the cultural attribution of items. To address this issue, we created a novel CSI dataset for Indian culture, belonging to 17 cultural facets. The dataset\footnote{Our dataset, code, and LLMs generations will be made available upon acceptance of the work.} comprises sim8k cultural concepts from 36 sub-regions. To measure cultural competence, we evaluate the adaptation of LLMs to cultural text using the created CSIs, LLM-based, and human evaluations. Also, we perform quantitative analysis demonstrating selective sub-regional coverage and surface-level adaptations across all considered LLMs.

 DIWALI - Diversity and Inclusivity aWare cuLture specific Items for India: Dataset and Assessment of LLMs for Cultural Text Adaptation in Indian Context

Large language models (LLMs) are helping millions of users write texts about diverse issues, and in doing so expose users to different ideas and perspectives. This creates concerns about issue bias, where an LLM tends to present just one perspective on a given issue, which in turn may influence how users think about this issue. So far, it has not been possible to measure which issue biases LLMs actually manifest in real user interactions, making it difficult to address the risks from biased LLMs. Therefore, we create IssueBench: a set of 2.49m realistic prompts for measuring issue bias in LLM writing assistance, which we construct based on 3.9k templates (e.g. "write a blog about") and 212 political issues (e.g. "AI regulation") from real user interactions. Using IssueBench, we show that issue biases are common and persistent in state-of-the-art LLMs. We also show that biases are remarkably similar across models, and that all models align more with US Democrat than Republican voter opinion on a subset of issues. IssueBench can easily be adapted to include other issues, templates, or tasks. By enabling robust and realistic measurement, we hope that IssueBench can bring a new quality of evidence to ongoing discussions about LLM biases and how to address them.

Downloads

Next from EMNLP 2025

Unsupervised Word-level Quality Estimation for Machine Translation Through the Lens of Annotators (Dis)agreement

Similar lecture

Retrieval Heads are Dynamic