Morocco

Fine-tuning LLMs introduces many important behaviors, such as instruction following, safety alignment. This makes it crucial to study how fine-tuning changes models&#39; internal mechanisms. Sparse autoencoders (SAEs) offer a powerful tool for interpreting neural networks by extracting concepts (features) represented in their activations. Previous work observed that SAEs trained on base models transfer effectively to instruction-tuned (chat) models, attributed to activation similarity. In this work, we propose *feature drift* as an alternative explanation: the feature space remains relevant, but the distribution of feature activations changes. In other words, fine-tuning recombines existing concepts rather than learning new ones. We validate this by showing base SAEs reconstruct both base and chat activations comparably despite systematic differences, with individual features exhibiting clear drift patterns. In a refusal behavior case study, we identify base SAE features that drift to activate on harmful instructions in chat models. Causal interventions using these features confirm that they mediate refusal. Our findings suggest that monitoring how existing features drift, rather than searching for entirely new features, may provide a more complete explanation of how fine-tuning changes model capabilities.

EACL 2026

Feature Drift: How Fine-Tuning Repurposes Representations in LLMs

sparse autoencoders

interpretability

### Welcome to 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics

Welcome to the 2025 meeting of the Nations of the Americas Chapter of the Association for Computational Linguistics! I am proud to help organize the first NAACL conference to carry the new name of our organization, one that emphasizes inclusion for all of the Americas. I am also pleased to welcome you to Albuquerque, New Mexico, a state whose unique blend of cultural influences will make for an excellent backdrop for NAACL 2025, especially with this year’s special theme on NLP in a Multicultural World. 
**[Continue reading...](https://drive.google.com/file/d/1jX-qGhqVSZZCIrAnJaz798pflu5Irrdn/view?usp=sharing)**

*- Colin Cherry, Google, NAACL 2025 General Chair* 

[![](https://assets.underline.io/markdown_image/1/image/b087f8a4dc5816d6e1a6514e59c59ac3.png)](https://drive.google.com/file/d/1T96GzPqObXrMln2BMByCSXTSizjTg69P/view?usp=sharing)

You need to log in with the email address you registered with. Access credentials have been sent to your email. 

Please be sure to check your spam and other email folders if you do not see an email confirmation right away.

Please log in to explore this event.

To access NAACL 2025 event page you are required to register. Please follow [**this link**](https://2025.naacl.org/registration/registration/) to register. Access will depend on your registration type.

Please register!

NAACL 2025

Welcome to 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics

Fine-tuning LLMs introduces many important behaviors, such as instruction following, safety alignment. This makes it crucial to study how fine-tuning changes models' internal mechanisms. Sparse autoencoders (SAEs) offer a powerful tool for interpreting neural networks by extracting concepts (features) represented in their activations. Previous work observed that SAEs trained on base models transfer effectively to instruction-tuned (chat) models, attributed to activation similarity. In this work, we propose *feature drift* as an alternative explanation: the feature space remains relevant, but the distribution of feature activations changes. In other words, fine-tuning recombines existing concepts rather than learning new ones. We validate this by showing base SAEs reconstruct both base and chat activations comparably despite systematic differences, with individual features exhibiting clear drift patterns. In a refusal behavior case study, we identify base SAE features that drift to activate on harmful instructions in chat models. Causal interventions using these features confirm that they mediate refusal. Our findings suggest that monitoring how existing features drift, rather than searching for entirely new features, may provide a more complete explanation of how fine-tuning changes model capabilities.

poster

#### *Message from the General Chair, Aline Villavicencio*
I’m delighted and honoured to welcome you to the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), taking place in the beautiful city of Rabat, in Morocco, in March 24-29, 2026. EACL is the flagship European conference of the Association and EACL 2026 proudly continues our field’s tradition of excellence in scholarship, innovation, and inclusivity. I am deeply grateful to the many volunteers whose dedication, generosity, and tireless efforts have made this conference possible.
For the first time EACL is being hosted in the African continent. This is an important milestone for our community, and we are grateful to our Moroccan hosts for enabling this historic moment by bringing this edition of EACL to Rabat. We are also delighted that the Second Arabic NLP School is co-located with EACL. We hope attendees enjoy this wonderful opportunity to strengthen ties with the Computational Linguistics communities across the African continent. *[Read full message](https://drive.google.com/file/d/14NlmHvwM6fPJuMmOvVh7K0vtQbEyv3SZ/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://underline.io/events/522/reception','_blank')">Go to Workshops and Tutorials Program</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 EACL 2026". 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 are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://2026.eacl.org/registration/) first.

**Online Registration Form**: https://acl.swoogo.com/eacl2026

Registration Required

Welcome to the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL). EACL 2026 will be held in Rabat, Morocco, from March 24–29, 2026. 

While the reasoning abilities of large language models (LLMs) continue to advance, it remains underexplored how such abilities vary across languages in multilingual LLMs and whether different languages generate distinct reasoning paths. In this work, we explore how reasoning knowledge in different languages complements one another. We propose a novel cross-lingual reward modeling framework that aggregates reasoning signals across languages to enhance mathematical reasoning. Our results show a substantial improvement (about 10 scores) in multilingual mathematical reasoning performance compared to using reward modeling within a single language, benefiting even high-resource languages. While English often exhibits the highest performance in multilingual models, we find that cross-lingual sampling benefits English, particularly under low sampling budgets. Our findings reveal new opportunities to improve multilingual reasoning by leveraging the complementary strengths of diverse languages.

Best-of-L: Cross-Lingual Reward Modeling for Mathematical Reasoning

Effectively identifying threats and mitigating their potential damage during crisis situations, such as natural disasters or violent attacks, is paramount for safeguarding endangered individuals. To tackle these challenges, AI has been used in assisting humans in emergency situations. Still, the use of NLP techniques remains limited, and mostly focuses on classification tasks. The significant potential of timely warning message generation using NLG architectures, however, has been largely overlooked. In this paper we present CrisiText, the first large-scale dataset for the generation of warning messages across 13 different types of crisis scenarios. The dataset contains more than 400,000 warning messages (spanning almost 18,000 crisis situations) aimed at assisting civilians during and after such events. To generate the dataset, we started from existing crisis descriptions, and created chains of events related to the scenarios. Each event was then paired with a warning message. The generations follow expert's written guidelines to ensure correct terminology and factuality of their suggestions. Additionally, each message is accompanied by three suboptimal warning types to allow for the study of different NLG approaches. To this end, we conducted a series of experiments comparing supervised fine-tuning setups with preference alignment, zero-shot, and few-shot approaches. We further assessed model performance in out-of-distribution scenarios and evaluated the effectiveness of an automatic post-editor.

CrisiText: A dataset of warning messages for LLM training in emergency communication

Sparse Autoencoders (SAEs) have been successfully used to probe Large Language Models (LLMs) and extract interpretable concepts from their internal representations. These concepts are linear combinations of neuron activations that correspond to human-interpretable features. In this paper, we investigate the effectiveness of SAE-based explainability approaches for sentence classification, a domain where such methods have not been extensively explored. We present a novel SAE-based model \verb|ClassifSAE| tailored for text classification, leveraging a specialized classifier head and incorporating an activation rate sparsity loss. We benchmark this architecture against established methods such as ConceptShap, Independent Component Analysis, HI-Concept and a standard TopK-SAE baseline. Our evaluation covers several classification benchmarks and backbone LLMs. We further enrich our analysis with two novel metrics for measuring the precision of concept-based explanations, using an external sentence encoder. Our empirical results show that \verb|ClassifSAE| improves both the causality and interpretability of the extracted features.

Unveiling Decision-Making in LLMs for Text Classification : Extraction of influential and interpretable concepts with Sparse Autoencoders

Despite significant progress in natural image editing with state-of-the-art MLLMs, compositional layout and content editing for structured visual domains (e.g., posters, websites) remains underexplored. In this work, we introduce SMART-EDITOR, a multi-agent framework for compositional editing for structured images like posters or websites. Unlike prior models that focus on isolated local edits, SMART-EDITOR maintains global coherence through two complementary strategies: Reward-Refine, an inference-time reward-guided refinement method, and RewardDPO, a training-time preference optimization approach leveraging reward-aligned layout pairs. To evaluate performance, we introduce SMARTEdit-Bench, a benchmark of cascading multi-step edit instructions that are implicit in nature yet require layout and semantic-consistency preserving reasoning about edit order to preserve spatial and semantic consistency. Both automatic and human evaluations confirm the value of reward-guided planning in producing semantically consistent and visually coherent edits, beyond what single-shot VLMs can generate.

SMART-Editor: A Multi-Agent Framework for Human-Like Design Editing with Structural Integrity

Temporal Knowledge Graph (TKG) reasoning seeks to predict future missing facts from historical evidence. While diffusion models (DM) have recently gained attention for their ability to capture complex predictive distributions, two gaps remain: (i) the generative path is conditioned only on positive evidence, overlooking informative negative context, and (ii) training objectives are dominated by cross-entropy ranking, which improves candidate ordering but provides little supervision over the calibration of the denoised embedding. To bridge this gap, we introduce **N**egative-**A**ware **D**iffusion model for TKG **Ex**trapolation (**NADEx**). Specifically, NADEx encodes subject-centric histories of entities, relations and temporal intervals into sequential embeddings. NADEx perturbs the query object in the forward process and reconstructs it in reverse with a Transformer denoiser conditioned on the temporal-relational context. We further derive a cosine-alignment regularizer derived from batch-wise negative prototypes, which tightens the decision boundary against implausible candidates. Comprehensive experiments on four public TKG benchmarks demonstrate that NADEx delivers state-of-the-art performance.

Negative-Aware Diffusion Process for Temporal Knowledge Graph Extrapolation

Despite remarkable progress in multilingual machine translation (MT), the majority of African---especially East African---languages remain significantly underrepresented both in benchmark datasets and state-of-the-art (SOTA) MT models. This persistent exclusion from mainstream technologies not only limits equitable access, but constrains the development of tools that accurately reflect the region's linguistic and cultural diversity. Recent advances in open-source large language models have demonstrated strong multilingual MT capabilities through data-efficient adaptation strategies. However, little work has explored their potential for low-resource African languages. We introduce AfriMMT-EA, the first highly multilingual benchmark and MT dataset for East African languages. Our datasets comprise 54 local languages across five East African countries. We used these data to fine-tune two multilingual versions of Gemma-3. We compare models' performance on these languages with larger off-the-shelf baselines. We release our data and models, in the interest of advancing MT for these low-resource languages and their communities.

AfriMMT-EA: Multi-domain Machine Translation for Low-Resource East African Languages

Large language models (LLMs) are beginning to reshape how media professionals verify information, yet automated support for detecting check-worthy claimsвЂ”a key step in the fact-checking processвЂ”remains limited. We introduce the Multi-Check-Worthy (MultiCW) dataset, a balanced multilingual benchmark for check-worthy claim detection spanning 16 languages, six topical domains, and two writing styles. It consists of 123,722 samples, evenly distributed between noisy (informal) and structured (formal) texts, with balanced representation of check-worthy and non-check-worthy classes across all languages. To probe robustness, we also introduce an equally balanced out-of-distribution evaluation set of 27,761 samples in 4 additional languages. To provide baselines, we benchmark three common fine-tuned multilingual transformers against a diverse set of 15 commercial and open LLMs under zero-shot settings. Our findings show that fine-tuned models consistently outperform zero-shot LLMs on claim classification and show strong out-of-distribution generalization across languages, domains, and styles. MultiCW provides a rigorous multilingual resource for advancing automated fact-checking and enables systematic comparisons between fine-tuned models and cutting-edge LLMs on the check-worthy claim detection task.

MultiCW: A Large-Scale Balanced Benchmark Dataset for Training Robust Check-Worthiness Detection Models

We present a novel metric for the evaluation of morphological plausibility of subword segmentation. Unlike the typically used morpheme boundary or retrieval F-score, which requires gold segmentation data that is either unavailable or of inconsistent quality across many languages, our approach utilizes morpho-syntactic features. These are available in resources such as Universal Dependencies or UniMorph for a much wider range of languages. The metric works by probabilistically aligning subwords with morphological features through an IBM Model 1. Our experiments show that the metric correlates well with traditional morpheme boundary recall while being more broadly applicable across languages with different morphological systems.

Evaluating Morphological Plausibility of Subword Tokenization via Statistical Alignment with Morpho-Syntactic Features

We introduce MOSAIC (Masked Objective with Selective Adaptation for In-domain Contrastive learning), a multi-stage framework for domain adaptation of text embedding models that incorporates joint domain-specific masked supervision. Our approach addresses the challenges of adapting large-scale general-domain text embedding models to specialized domains. By jointly optimizing masked language modeling (MLM) and contrastive objectives within a unified training pipeline, our method enables effective learning of domain-relevant representations while preserving the robust semantic discrimination properties of the original model. We empirically validate our approach on both high-resource and low-resource domains, achieving improvements up to 13.4% in NDCG@10 (Normalized Discounted Cumulative Gain) over strong general-domain baselines. Comprehensive ablation studies further demonstrate the effectiveness of each component, highlighting the importance of balanced joint supervision and staged adaptation.

MOSAIC: Masked Objective with Selective Adaptation for In-domain Contrastive Learning

Large Language Models (LLMs) possess a remarkable capacity to generate persuasive and intelligible language. However, coherence does not equate to truthfulness, as the responses often contain subtle hallucinations. Existing benchmarks are limited by static and narrow questions, leading to limited coverage and misleading evaluations. We present **KGHaluBench**, a Knowledge Graph-based hallucination benchmark that assesses LLMs across the breadth and depth of their knowledge, providing a fairer and more comprehensive insight into LLM truthfulness. Our framework utilises the KG to dynamically construct challenging, multifaceted questions, whose difficulty is then statistically estimated to address popularity bias. Our automated verification pipeline detects abstentions and verifies the LLM's response at both conceptual and correctness levels to identify different types of hallucinations. We evaluate 25 frontier models, using novel accuracy and hallucination metrics. The results provide a more interpretable insight into the knowledge factors that cause hallucinations across different model sizes. KGHaluBench is publicly available to support future developments in hallucination mitigation.

Downloads

Next from EACL 2026

Best-of-L: Cross-Lingual Reward Modeling for Mathematical Reasoning

Similar lecture

Features that Make a Difference: Leveraging Gradients for Improved Dictionary Learning

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES