United States

For Mapudungun arn→es translation, morphology-aware tokenization can substitute for a 5× increase in model parameters. We fine-tune three sizes of Meta&#39;s NLLB-200 on Mapudungun–Spanish translation across eight tokenization strategies, including our novel Morfessor-VC method, which
constrains Morfessor morpheme segmentation to tokens already present in NLLB&#39;s pretrained
vocabulary. Our 600M Morfessor-VC model is competitive with our own fine-tuned 3.3B Standard BPE model on arn→es (43.2 vs. 42.9 chrF++, ∆ = +0.3, p = 0.039, 95% CI [0.02, 0.60]) while using five times fewer parameters, and all fine-tuned conditions surpass frontier LLMs by over 27 chrF++. Mapudungun is an indigenous polysynthetic language spoken by 200,000+ Mapuche people in Chile and Argentina, absent from NLLB-200 and not supported by major commercial MT providers; prior work predates large-scale multilingual models and does not address the tokenization challenges posed by its agglutinative
morphology. These results establish new state-of-the-art baselines for Mapudungun MT and provide a practical foundation for community language tools in pedagogy, social media, and language revitalization.

ACL 2026

Bringing Mapudungun into the Modern MT Ecosystem: Morphology-Aware Tokenization for NLLB-200 Fine-Tuning

mapudungun

morphology-aware tokenization

nllb-200

polysynthetic languages

low-resource machine translation

fine-tuning

### Welcome!
The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. Our Virtual Poster Sessions will take place online Thursday, August 22, 2024.

You are required to register for this event. **Please register [here](https://2024.aclweb.org/registration). **

If you have already registered, please check your inbox for an email from Underline granting you access to ACL 2024 content.

Please register!

ACL 2024

The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. More information will be announced soon.

For Mapudungun arn→es translation, morphology-aware tokenization can substitute for a 5× increase in model parameters. We fine-tune three sizes of Meta's NLLB-200 on Mapudungun–Spanish translation across eight tokenization strategies, including our novel Morfessor-VC method, which
constrains Morfessor morpheme segmentation to tokens already present in NLLB's pretrained
vocabulary. Our 600M Morfessor-VC model is competitive with our own fine-tuned 3.3B Standard BPE model on arn→es (43.2 vs. 42.9 chrF++, ∆ = +0.3, p = 0.039, 95% CI [0.02, 0.60]) while using five times fewer parameters, and all fine-tuned conditions surpass frontier LLMs by over 27 chrF++. Mapudungun is an indigenous polysynthetic language spoken by 200,000+ Mapuche people in Chile and Argentina, absent from NLLB-200 and not supported by major commercial MT providers; prior work predates large-scale multilingual models and does not address the tokenization challenges posed by its agglutinative
morphology. These results establish new state-of-the-art baselines for Mapudungun MT and provide a practical foundation for community language tools in pedagogy, social media, and language revitalization.

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.

Qom, a language of the Guaycuruan family, is a low-resource language for NLP and speech processing. We present the first parallel Qom–Spanish corpus in a computationally usable format, comprising 33,392 parallel segments, totaling 1,469,905 Qom tokens and 891,344 Spanish tokens. A subset of 2,943 segments excludes Bible-derived content. It includes alignments at different levels: sentences, sentence fragments, and paragraphs, and is compiled from multiple sources, both previously available and newly collected. 
We also present bidirectional neural machine translation baselines based on NLLB-200, achieving competitive performance in both translation directions on the full dataset, and lower performance on the non-Bible subset. An ablation study shows that training exclusively on biblical data reduces performance on non-biblical text, highlighting the importance of domain diversity in low-resource machine translation.

QomL'aqtaqa: A Qom–Spanish Parallel Corpus for Natural Language Processing with Machine Translation Evaluation

Publicly available spoken language identification (LID) systems provide sparse and inconsistent coverage of indigenous languages of the Americas and languages of the Pacific Islands. No system on HuggingFace covers Central Alaskan Yup'ik except the largest variant of Meta's MMS-LID family, and only three MMS-LID variants cover Samoan, while Whisper and VoxLingua107-based models lack both despite including other Polynesian languages. We describe an ongoing effort to build a coarse-labeled LID dataset for Yup'ik and Samoan from US public broadcast archives, benchmark publicly available LID systems on it, and train a simple MLP classifier on frozen wav2vec~2.0 representations as a prototype. We report preliminary corpus statistics, off-the-shelf model performance, and prototype results. Guided by the distinctive phonological typology of the target languages, we outline a phonologically-informed fine-tuning direction as future work.

Toward a Coarse-Labeled Spoken Language Identification Dataset for Central Alaskan Yup'ik and Samoan from US Broadcast Archives

This system description presents a Wixarika image captioning system for the AmericasNLP 2026 Shared Task on Cultural Image Captioning for Indigenous Languages. The system uses Spanish as a pivot language, combining CLIP-based image retrieval, Qwen3-VL Spanish caption generation, the official Sheffield-compatible Spanish-to-Wixarika MT model, and character n-gram language-model reranking. We report local 5-fold development results, official test results, error analysis, and implementation details for reproducibility.

6fanle Submission to the AmericasNLP 2026 Shared Task on Wixarika Image Captioning

Culturally grounded image captioning for under-resourced Indigenous languages is challenging due to severe data scarcity and the need to describe culturally specific visual content. This paper describes our submission to the AmericasNLP 2026 shared task, where we evaluate two architectural paradigms for caption generation across Bribri, Guaraní, Yucatec Maya, Wixárika, and Orizaba Nahuatl. First, we implement a cascaded system that combines a large vision-language model with a machine translation pipeline, showing that culturally contextualized, persona-based prompting improves over the official baseline in most comparable settings. Second, we develop a direct, end-to-end Single-stage approach by adapting PaliGemma 2 using LoRA fine-tuning, continued pre-training, and multilingual joint training. Our single-stage experiments show that, despite severe domain mismatch and reliance on synthetic training data, multilingual training and continued pre-training improve automatic chrF++ relative to single-language LoRA fine-tuning in some settings. Overall, cascaded pipelines remain the strongest among the evaluated approaches under current data constraints, while single-stage models remain a promising but currently data-limited path toward direct Indigenous-language image captioning.

Culturally Grounded Image Captioning in Indigenous Languages with Vision-Language Models: Cascaded and Single-Stage Approaches

We describe the USP system for the AmericasNLP 2026 Shared Task on Culturally Relevant Image Captioning for Indigenous Languages, covering Guaraní (grn), Maya Yucateco (yua), Nahuatl (nah), Wixárika (hch), and Bribri (bzd). We propose a two-stage cascade: Qwen3-VL-8B-Instruct (Bai et al., 2025) generates Spanish captions via language-specific cultural prompts; language-specific fine-tuned NLLB-200-distilled-600M (NLLB Team et al., 2022) models then translate them into each target language. We train on AmericasNLP 2023 data (Ebrahimi et al., 2023) augmented with public parallel corpora. Our system achieves competitive results, including 3rd place in Guaraní human evaluation (2.41/5.0) and 5th in Bribri (1.09/5.0) among 8 teams. We also report that NLLB-200-distilled-600M silently lacks vocabulary entries for Bribri and Maya Yucateco, producing English output without error.

USP at AmericasNLP 2026 Shared Task: Culturally-Aware Image Captioning for Indigenous Languages via Vision-Language Models and Fine-Tuned Neural Machine Translation

Multilingual large language models (LLMs) are increasingly used to interpret mental health narratives in settings such as support chatbots, screening, and content moderation. A basic reliability question remains underexplored: when the same narrative is presented in different languages, does a model apply comparable evaluative standards? We study cross-linguistic variation in mental health evaluation between English and Chinese using two widely deployed LLMs (GPT-4o and Qwen3-32B). We separate (i) construct-level evaluative orientation from (ii) decision-level behavior on downstream tasks. At the construct level, we adapt validated psychometric stigma instruments spanning social, self, and health-professional stigma and quantify language-conditioned shifts in model responses. At the decision level, we evaluate models on binary stigma detection and four-way depression severity classification using semantically aligned English--Chinese datasets. Across instruments and models, Chinese prompts elicit systematically higher stigma scores than English prompts. These differences propagate to downstream decisions: Chinese prompts reduce sensitivity in stigma detection and induce a downward shift in predicted depression severity, producing substantially more under-estimation errors. Our findings show that prompt language can shift both evaluative orientation and effective decision thresholds, underscoring the need for multilingual evaluation and calibration before deploying LLMs in socially sensitive domains.

Language Shapes Mental Health Evaluations in Large Language Models

The evaluation of societal biases in NLP models is critically hindered by a geo-cultural gap. This leaves regions such as Latin America severely underserved, making it impossible to adequately assess or mitigate the perpetuation of harmful regional stereotypes in language technologies.

This paper presents LACES, a stereotype association dataset, for 15 Latin American countries. This dataset includes 4,789 stereotype associations\footnote{The de-identified dataset can be accessed via \href{https://github.com/fvialibre/LACES}{GitHub}}, manually created and annotated by 83 participants. The dataset was developed through targeted community partnerships across Latin America.

Additionally, in this paper, we propose a novel adaptive data collection methodology that uniquely integrates the sourcing of new stereotype entries and the validation of existing data within a single, unified workflow. This approach results in a resource with more unique stereotypes than previous static collection methods, enabling a more efficient stereotype collection. The paper further supports the quality of LACES by demonstrating reduced efficacy of debiasing methods on this dataset in comparison to existing popular stereotype benchmarks.

Adaptive Data Collection for Latin-American Community-sourced Evaluation of Stereotypes (LACES)

Stereotype repositories are critical to assess generative AI model safety, but currently lack adequate global coverage. It is imperative to prioritize targeted expansion, strategically addressing existing deficits, over merely increasing data volume. This work introduces a multilingual stereotype resource covering four sub-Saharan African countries that are severely underrepresented in NLP resources: Ghana, Kenya, Nigeria, and South Africa. By utilizing socioculturally-situated, community-engaged methods, including telephonic surveys moderated in native languages, we establish a reproducible methodology that is sensitive to the region's complex linguistic diversity and traditional orality. By deliberately balancing the sample across diverse ethnic and demographic backgrounds, we ensure broad coverage, resulting in a dataset of 3,534 stereotypes in English and 3,206 stereotypes across 15 native languages.

SAFARI: A Community-Engaged Approach and Dataset of Stereotype Resources in the Sub-Saharan African Context

Large language models (LLMs) are increasingly used as evaluative tools across languages, yet bias research remains overwhelmingly Anglocentric, with most studies conducted in English using Latin-script names. It remains unclear whether bias patterns generalize across linguistic contexts. We investigate this question and introduce the stereotype perceptual map, a framework for analyzing how ethnic groups are positioned along evaluative dimensions. 
Using 900,000 model responses over 45,000 names spanning 9 ethnicities, we evaluate model behavior across prompt languages (English, Chinese, Thai), writing scripts (Latin, Chinese, Thai), evaluative domains (competence, warmth), and models (GPT, DeepSeek). We find that ethnic bias hierarchies are jointly shaped by local linguistic context and model origin and differ substantially between Western-centric and Sinocentric models. DeepSeek exhibits highly stable rankings across conditions in math competence judgments, consistently placing Chinese at the top, followed by Russian, and White, Hispanic, and Black names at the bottom. GPT, by contrast, shows strong script-dependent reordering: Latin-script conditions form one stable cluster, while native-script conditions form another, with substantially lower cross-cluster correlations. We term this script-gated bias: transliterating the same names into a non-Latin script can activate a different evaluative frame and produce rankings that are sometimes inversely correlated with Latin-script results. Warmth evaluations are less stable than competence across both models.
Our findings demonstrate that multilingual bias cannot be characterized through single-language, single-script audits. For multilingual users, code-switching between languages can toggle between different bias regimes. Fairness evaluations for multilingual LLMs must therefore account for deployment language, writing system, and model origin to capture the full range of potentially harmful stereotypes these systems exhibit.

Through the Looking Glass of Multilingual AI: Contrasting Language- and Name Script-Dependent Ethnic Hierarchies in GPT and DeepSeek

Bias benchmarks for LLMs largely focus on English, overlooking language- and culture-specific stereotypes. We introduce CrowS-Pairs-NL, a Dutch stereotype benchmark built by filtering, translating, and adapting the English CrowS-Pairs dataset to address known conceptual pitfalls, and extending it with newly crowdsourced Dutch sentence pairs. We evaluate six multilingual and Dutch-trained models using both a pseudo-log-likelihood metric adapted for autoregressive models and a prompt-based metric with three template variants. Models explicitly trained on Dutch data consistently exhibit higher stereotyping scores, suggesting that language-specific fine-tuning introduces language-specific bias. The two metrics broadly agree on model rankings but differ in sensitivity, with the prompt metric showing a narrower range of scores. Our benchmark and findings underscore the need for culturally grounded bias evaluation beyond English.

Content not yet available

Downloads

Next from ACL 2026

QomL'aqtaqa: A Qom–Spanish Parallel Corpus for Natural Language Processing with Machine Translation Evaluation

Similar lecture

Enhancing Turkish Word Segmentation: A Focus on Borrowed Words and Invalid Morpheme

Content not yet available

.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

QomL'aqtaqa: A Qom–Spanish Parallel Corpus for Natural Language Processing with Machine Translation Evaluation

Similar lecture

Enhancing Turkish Word Segmentation: A Focus on Borrowed Words and Invalid Morpheme

Downloads