United States

We evaluate whether open-source LLMs can produce proficiency-graded English adaptations of entries from the \textit{Diccionario de colombianismos} (DiCol), a Colombian Spanish lexicographic resource used in language teaching. Three 7--8B instruction-tuned models---Llama~3.1, Qwen2.5, and Mistral---generate Beginner, Intermediate, and Advanced translations for all 8{,}252 definitions using structured zero-shot prompts identical across levels except for the target CEFR band. Automated metrics show that Intermediate targeting collapses (73--83\% classified as Advanced by vocabulary, $\chi^2 &gt; 705$, $p &lt; .001$) and that Advanced outputs expand 4.9--8.2$\times$ relative to the source. Expert annotation of a 360-entry stratified sample ($\kappa = 0.61$--$0.68$) identifies hallucination in 19\% of entries (Fleiss&#39; $\kappa = 0.77$ for cultural preservation categories, 97\% unanimity among flagged cases). Hallucination concentrates in the Advanced condition (81\%, $\chi^2 = 86.6$, $p &lt; .001$) and is associated with higher expansion ($U = 16{,}662$, $p &lt; .001$, $r = 0.68$), manifesting primarily as generic elaboration and, in a smaller proportion, as Colombia-stereotyping and pragmatic polarity inversion. We discuss these findings through the lens of \citeauthor{venuti1995invisibility}&#39;s domestication framework and describe the observed pattern as \emph{algorithmic domestication}.

ACL 2026

LLM-Adapted Colombian Spanish Lexicography: Proficiency Control, Hallucination, and Cultural Distortion

cefr-controlled text simplification

cross-cultural nlp bias

llm hallucination

## 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*

EMNLP 2025

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

We evaluate whether open-source LLMs can produce proficiency-graded English adaptations of entries from the \textit{Diccionario de colombianismos} (DiCol), a Colombian Spanish lexicographic resource used in language teaching. Three 7--8B instruction-tuned models---Llama~3.1, Qwen2.5, and Mistral---generate Beginner, Intermediate, and Advanced translations for all 8{,}252 definitions using structured zero-shot prompts identical across levels except for the target CEFR band. Automated metrics show that Intermediate targeting collapses (73--83\% classified as Advanced by vocabulary, $\chi^2 > 705$, $p < .001$) and that Advanced outputs expand 4.9--8.2$\times$ relative to the source. Expert annotation of a 360-entry stratified sample ($\kappa = 0.61$--$0.68$) identifies hallucination in 19\% of entries (Fleiss' $\kappa = 0.77$ for cultural preservation categories, 97\% unanimity among flagged cases). Hallucination concentrates in the Advanced condition (81\%, $\chi^2 = 86.6$, $p < .001$) and is associated with higher expansion ($U = 16{,}662$, $p < .001$, $r = 0.68$), manifesting primarily as generic elaboration and, in a smaller proportion, as Colombia-stereotyping and pragmatic polarity inversion. We discuss these findings through the lens of \citeauthor{venuti1995invisibility}'s domestication framework and describe the observed pattern as \emph{algorithmic domestication}.

workshop paper

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.

Registration Required

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

Large Language Models (LLMs) show unbalanced knowledge of cultures across the globe, favoring high-resource cultures over low-resource ones. A possible way to tackle this issue is to fine-tune LLMs on culturally specific data. However, fine-tuning recent LLMs requires high computational resources as well as memory storage, which triggered the development of parameter-efficient fine-tuning (PEFT) approaches, the most widespread being LoRA. In this article, we investigate the use of another class of PEFT approaches, namely soft prompt methods (prompt-tuning and prefix-tuning), to improve LLMs' cultural knowledge across diverse cultures. We focus on cultural alignment on Multiple-Choice Questions of cultural commonsense knowledge. On this task with limited fine-tuning data, we show that soft-prompt-based methods outperform LoRA in comparable settings. Moreover, the trained soft prompts are interpretable and capture similarities between cultures.

Soft Prompts for Adapting LLMs to Cultural Commonsense Knowledge

Culture shapes the objects people use and for what purposes, yet mainstream Vision-Language (VL) datasets frequently exhibit cultural biases, disproportionately favoring higher-income, Western contexts. This imbalance reduces model generalizability and perpetuates performance disparities, especially impacting lower-income and non-Western communities. To address these disparities, we propose a novel function-centric framework that categorizes objects by the functions they fulfill, across diverse cultural and economic contexts. We implement this framework by creating the Culture Affordance Atlas, a re-annotated and culturally grounded restructuring of the Dollar Street dataset spanning 46 functions and 288 objects. Through extensive empirical analyses using the CLIP model, we demonstrate that function-centric labels substantially reduce socioeconomic performance gaps between high- and low-income groups by a median of 6 pp (statistically significant), improving model effectiveness for lower-income contexts. Furthermore, our analyses reveals numerous culturally essential objects that are frequently overlooked in prominent VL datasets. Our contributions offer a scalable pathway toward building inclusive VL datasets and equitable AI systems.

Culture Affordance Atlas: Reconciling Object Diversity Through Functional Mapping

Large Language Models are widely used to generate and adapt cultural texts, yet the depth of their cultural representation remains poorly quantified. Intuitively, as a narrative text expands in length, the diversity of cultural words should scale proportionately. To formally test this, we evaluate the FairyTaleQA dataset, adapted by three models and introduce our primary contribution: the Contextual Stereotype Amplification Index (CSAI), an evaluation framework combining LLM-as-a-judge extraction, embedding-based cliché anchoring, and Natural Language Inference (NLI) congruence validation. By mapping the frequency of extracted Culture Specific Items (CSIs) against narrative length using Heaps' Law ($V = k \cdot T^\beta$), we present empirical evidence of a systematic limitation in current systems: they struggle to scale cultural diversity even under explicit cultural prompting. Models rapidly hit a "Cultural Vocabulary Ceiling," constrained to a fixed set of hyper-stereotypical terms. Furthermore, we demonstrate that merely optimizing for higher CSI frequency as done in prior works rewards logically broken tokenism. Our CSAI formulation actively penalizes such gratuitous stereotyping, offering a more principled approach to measuring and evaluating cultural homogenization in generative AI systems.

The Mirage of Diversity: Unmasking the Cultural Vocabulary Ceiling in LLMs

Large Language Models are increasingly deployed as writing assistants for users in the Global South, yet rewriting prompts can suppress institutionalized postcolonial varieties. We quantify South Asian English (SAsE) dialect erasure in a state-of-the-art open-weight model using a 500-sentence diagnostic benchmark (320 lexical and 180 syntactic markers). On Llama 3.3 70B, standard grammar correction retains only 26.0\% of markers (lexical 31.2\%; syntactic 16.7\%), while formalization is more destructive (14.0\% overall retention). For lexical items, we observe Americanization in 56.2\% (correction) and 59.4\% (formalization) of cases, typically via Standard American paraphrases. A simple dialect-aware prompt raises retention to 92.0\% and reduces lexical Americanization to 6.2\%, although some function-word phenomena remain resistant. A stress test shows even stronger suppression (6.7\% retention). We position dialect erasure within representational-harm and cultural-competence frameworks, and provide a replicable protocol for auditing writing-assistance systems.

The American Palimpsest: Quantifying South Asian English Dialect Erasure in LLMs

As Video Large Language Models (VideoLLMs) are deployed globally, understanding of the relevant cultural norms becomes essential for safety and accuracy. To advance cultural norm awareness evaluation in video models, we collect a dataset of over 3,000 annotations of cultural norms in video clips from popular US and Chinese TV shows. Each clip is labeled with cultural norms, adherence or violation labels, and verbal and non-verbal evidence supporting the label by at least three trained monocultural annotators with significant lived experience in the target culture. We conduct an analysis of annotator disagreements and construct a consensus-based benchmark for cultural norm understanding in videos, VideoNorms. Hierarchical linear modeling analysis of $7$ open-weight VideoLLMs' performance revealed that: 1) models exhibit cultural bias in norm adherence prediction; 2) models have more difficulty in providing non-verbal evidence compared to verbal for norm adherence or violation. Ablation studies confirm video modality is indeed necessary for high task performance, and scaling model size does not yield classification performance improvements. Our findings and data will contribute to culturally grounded video model training and evaluation.

VideoNorms: Benchmarking Cultural Norm Understanding of Video Language Models

LLMs are deployed globally, yet produce responses biased towards cultures with abundant training data. Existing cultural localization approaches such as prompting or post-training alignment are black-box, hard to control, and do not reveal whether failures reflect missing knowledge or poor elicitation. In this paper, we address these gaps using mechanistic interpretability to uncover and manipulate cultural representations in LLMs. Leveraging sparse autoencoders (SAEs), we identify interpretable features that encode culturally salient information and aggregate them into $\textbf{Cu}$ltural $\textbf{E}$mbeddings ($\textit{CuE}$). We use $\textit{CuE}$ both to analyze implicit cultural biases under underspecified prompts and to construct white-box steering interventions. Across multiple models, we show that $\textit{CuE}$ -based steering increases cultural faithfulness and elicits significantly rarer, long-tail cultural concepts than prompting alone. Notably, $\textit{CuE}$ -based steering is complementary to black-box localization methods, offering gains when applied on top of prompt-augmented inputs. This also suggests that models do benefit from better elicitation strategies, and don't necessarily lack long-tail knowledge representation, though this varies across cultures. Our results provide both diagnostic insight into cultural representations in LLMs and a controllable method to steer towards desired cultures.

Steering LLMs for Culturally Localized Generation

Large language models (LLMs) are increasingly deployed in multilingual settings, yet little is known about whether their moral and social judgments remain consistent across languages. In particular, when faced with moral and social dilemmas, LLMs must often implicitly or explicitly assign responsibility — to an individual, to broader social forces, or across multiple parties — a process known as responsibility attribution. This study investigates whether responsibility attributions vary across languages, whether any observed variation persists across thematic domains, and whether the degree of variation differs across LLMs. We evaluate three models (GPT-5.2, Gemini-2.5-Pro, and LLaMA-3.3-70B) across 12 scenarios spanning six thematic domains (marriage, career, authority, gender, elder care, and family). Each model was prompted to attribute responsibility for each scenario by selecting from four options: the primary individual, a secondary interpersonal actor, a broader societal factor, or distributed responsibility shared across multiple parties. Results reveal a significant overall association between language and responsibility attribution (Cramér's $V = 0.24$) that persists within every thematic domain ($V = 0.26$--$0.53$). The magnitude of cross-language variation is strongly model-dependent: GPT-5.2 and Gemini-2.5-Pro show modest shifts ($V \approx 0.19$), while LLaMA-3.3-70B exhibits substantially stronger divergence ($V = 0.52$). These findings suggest that normative consistency across languages cannot be assumed and should be treated as a distinct dimension of model evaluation.

Lost in Translation? How Language Shapes Responsibility Attribution in Large Language Models

This paper proposes a semi-automatic lexico-semantic modeling framework for Chinese chéngyǔ containing body-part and animal lexemes. The framework combines manual semantic annotation, lightweight RDF/OWL formalization and semantic classification in order to investigate whether lexical mediators such as 心 xīn “heart/mind”, 口 kǒu “mouth” or 马 mǎ “horse” are sufficient to predict idiomatic semantic interpretation. Based on 440 annotated chéngyǔ normalized into 18 semantic categories, we compare three classification approaches: a rule-based keyword baseline, character n-gram TF-IDF with logistic regression, and BERT-base-chinese. The results show that lexical mediators cannot be directly equated with semantic categories and that TF-IDF achieves the best overall performance, suggesting that lightweight character-level representations remain robust for very short idioms in low-resource settings. The study contributes an interpretable RDF/OWL-compatible resource for culture-aware modeling of Chinese idioms.

Ontology-oriented lexico-semantic modeling and neural classification of Chinese chéngyǔ: A culture-aware NLP approach

Cross-cultural psychology has shown that moral judgments about failures to help vary systematically across cultures. In a landmark study, Miller, Bersoff, and Harwood (1990) found that while Indian and American participants agreed that failures to help are undesirable, they differed in whether they considered helping a moral obligation subject to social sanction or a personal decision. We adapt Miller et al.'s paradigm---nine scenarios crossing need severity (life-threatening, moderate, minor) with role relationship (parent, friend, stranger) and their original probe questions---to a cross-lingual LLM setting, presenting them to four LLMs (GPT-5.4, Claude-Opus-4.6, DeepSeek-V3.1, Qwen3-235B) across ten languages. We find that language significantly shapes how LLMs categorize failures to help as moral violations, social conventions, personal-moral concerns, or personal decisions ($\chi^2(27) = 116.14$, $p < .001$, Cramer's $V = 0.147$). Models agree across languages that failures to help are undesirable, but diverge substantially in how they classify them, with the primary divergence falling between moral violations and personal decisions. The proportion of responses classifying failures as moral violations decreases as need severity decreases and the role relationship becomes more distant. Cross-lingual variation differs substantially across models, with open-weight models showing significantly stronger variation than closed-weight models. These findings indicate that users consulting LLMs in different languages may receive substantively different moral guidance, underscoring the need for cross-lingual normative auditing as a component of multilingual LLM evaluation.

Sorry, Can't Help You: How Large Language Models Judge Failures to Help Across Languages

Verbal humor involves reasoning through complex conversational contexts. Although LLMs have achieved strong performance on English humor datasets, their ability to interpret humor in Hindi remains unexplored. In this paper, we evaluate Hindi humor for which we extract dialogues from humorous video clips. We use a pipeline that transforms video content into detailed textual streams, including dialogue transcripts and scene descriptions, allowing reasoning over inputs exceeding 2,000 words. We test various LLMs, from efficient edge models (Qwen-2.5-7B, Qwen-3-7B, Gemma-3-27B) to Indic-focused models (Sarvam-M-24B) and large frontier models (Llama-3.1-70B, Gemini-2.0-Flash). Our findings show a concave performance pattern in long-context understanding, with reasoning quality peaking at moderate lengths (250–750 words) and declining at higher context lengths. We also show that standard metrics overstate pragmatic competence. While increasing model size generally improves performance, we also observe distinct failures in smaller LLMs due to instructional and linguistic issues, necessitating diversity metrics to capture hallucinations. Smaller, Hindi-focused models can compete with much larger generalist models. Importantly, our evaluation reveals that conversational humor is a challenge for even specialized models, making HinS a valuable benchmark for advancing research in Hindi Long-Context Humor Reasoning.

Downloads

Next from ACL 2026

Soft Prompts for Adapting LLMs to Cultural Commonsense Knowledge

Similar lecture

How Much Do LLMs Hallucinate across Languages? On Realistic Multilingual Estimation of LLM Hallucination

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from ACL 2026

Soft Prompts for Adapting LLMs to Cultural Commonsense Knowledge

Similar lecture

How Much Do LLMs Hallucinate across Languages? On Realistic Multilingual Estimation of LLM Hallucination

Downloads