arXiv:2512.04765v1 Announce Type: new 
Abstract: Large language models and multilingual machine translation (MT) systems increasingly drive access to information, yet many languages of the tribal communities remain effectively invisible in these technologies. This invisibility exacerbates existing structural inequities in education, governance, and digital participation. We present AdiBhashaa, a community-driven initiative that constructs the first open parallel corpora and baseline MT systems for four major Indian tribal languages-Bhili, Mundari, Gondi, and Santali. This work combines participatory data creation with native speakers, human-in-the-loop validation, and systematic evaluation of both encoder-decoder MT models and large language models. In addition to reporting technical findings, we articulate how AdiBhashaa illustrates a possible model for more equitable AI research: it centers local expertise, builds capacity among early-career researchers from marginalized communities, and foregrounds human validation in the development of language technologies.

AdiBhashaa هي مبادرة يقودها المجتمع وقد طورت أول مجموعات بيانات متوازية مفتوحة وأنظمة ترجمة آلية مرجعية لأربع لغات قبلية هندية رئيسية: بهيلي، مونداري، غوندي، وسنتالي. يهدف هذا المشروع إلى معالجة عدم وضوح اللغات القبلية في نماذج اللغة الكبيرة وأنظمة الترجمة الآلية متعددة اللغات، مما ساهم في تفاقم عدم المساواة الهيكلية في التعليم والمشاركة الرقمية.

AdiBhashaa es una iniciativa impulsada por la comunidad que ha desarrollado los primeros corpus paralelos abiertos y sistemas de traducción automática de referencia para cuatro importantes lenguas tribales indias: Bhili, Mundari, Gondi y Santali. Este proyecto aborda la invisibilidad de las lenguas tribales en los modelos de lenguaje y sistemas de traducción automática multilingüe, lo que ha contribuido a desigualdades estructurales en la educación y la participación digital.

AdiBhashaa est une initiative communautaire qui a développé les premiers corpus parallèles ouverts et systèmes de traduction automatique de référence pour quatre grandes langues tribales indiennes : Bhili, Mundari, Gondi et Santali. Ce projet vise à remédier à l'invisibilité des langues tribales dans les modèles de langage et les systèmes de traduction automatique multilingues, contribuant ainsi à des inégalités structurelles dans l'éducation et la participation numérique.

AdiBhashaa is a community-driven initiative that has developed the first open parallel corpora and baseline machine translation systems for four major Indian tribal languages: Bhili, Mundari, Gondi, and Santali. This project addresses the invisibility of tribal languages in large language models and multilingual machine translation systems, which has contributed to structural inequities in education and digital participation.

AdiBhashaa: A Community-Curated Benchmark for Machine Translation into Indian Tribal Languages

arXiv:2510.06249v4 Announce Type: replace 
Abstract: The 2025 Multimodal Models for Low-Resource Contexts and Social Impact (MMLoSo) Language Challenge addresses one of India's most pressing linguistic gaps: the lack of resources for its diverse low-resource languages (LRLs). In this study, we investigate whether enforcing cross-lingual similarity in specific internal layers of a decoder-only multilingual large language model (LLM) can improve translation quality from LRL to high-resource language (HRL). Specifically, we combine Centered Kernel Alignment (CKA), a similarity metric that encourages representations of different languages to align, with REPINA, a regularization method that constrains parameter updates to remain close to the pretrained model, into a joint method we call TRepLiNa. In this research project, we experiment with zero-shot, few-shot, and fine-tuning settings using Aya-23 8B with QLoRA across MMLoSo shared task language pairs (Mundari, Santali, Bhili) with Hindi/English pivots. Our results show that aligning mid-level layers using TRepLiNa (CKA+REPINA) is a low-cost, practical approach to improving LRL translation, especially in data-scarce settings.

تم تقديم طريقة TRepLiNa، التي تجمع بين محاذاة النواة المركزية (CKA) وREPINA، لتحسين الترجمة الآلية للغات ذات الموارد المنخفضة، وخاصة للغات هندية مثل مونداري، سانتالي، وبهيلي، باستخدام نموذج Aya-23 8B. يهدف هذا النهج إلى تحسين جودة الترجمة من اللغات ذات الموارد المنخفضة إلى اللغات ذات الموارد العالية مثل الهندية والإنجليزية.

Se ha introducido el método TRepLiNa, que combina la Alineación de Núcleo Centrado (CKA) y REPINA, para mejorar la traducción automática de idiomas de bajos recursos, especialmente para idiomas indios como Mundari, Santali y Bhili, utilizando el modelo Aya-23 8B. Este enfoque busca mejorar la calidad de la traducción de idiomas de bajos recursos a idiomas de altos recursos como el hindi y el inglés.

La méthode TRepLiNa, qui combine l'Alignement de Noyau Centré (CKA) et REPINA, a été introduite pour améliorer la traduction automatique des langues à faibles ressources, en particulier pour des langues indiennes comme le Mundari, le Santali et le Bhili, en utilisant le modèle Aya-23 8B. Cette approche vise à améliorer la qualité de la traduction des langues à faibles ressources vers des langues à fortes ressources comme l'hindi et l'anglais.

The TRepLiNa method, which combines Centered Kernel Alignment (CKA) and REPINA, has been introduced to enhance low-resource machine translation, particularly for Indian languages like Mundari, Santali, and Bhili, using the Aya-23 8B model. This approach aims to improve translation quality from low-resource languages to high-resource languages such as Hindi and English.

AdiBhashaa: A Community-Curated Benchmark for Machine Translation into Indian Tribal Languages

Was this article worth reading? Share it

Translate a Book with AI

Zemith-3bda3b

MarsHub