Meta is expanding its artificial intelligence translation efforts through a partnership with UNESCO, focusing on underserved languages. This move aligns with UNESCO’s International Decade of Indigenous Languages, which aims to preserve and revitalize languages that are at risk of disappearing. Meta’s AI-powered translation and speech recognition tools will support more diverse linguistic representation, with the first government partner being Nunavut, Canada, working to develop translations for Inuktitut and Inuinnaqtun.
Meta’s Language Technology Partner Program
As part of these efforts, Meta has launched the Language Technology Partner Program, inviting contributors to help improve AI translation by submitting speech recordings (10+ hours), large text samples (200+ sentences), and translated sentence sets. The goal is to build datasets that enhance machine learning models, making AI-powered translation more reliable for languages that lack extensive digital resources.
This strategy makes sense. AI translation flourishes with available training data, but many Indigenous and minority languages lack the massive text corpora that AI requires. By partnering with local governments and organizations, Meta is addressing a key obstacle—data scarcity. However, contribution requirements won’t be easy for smaller communities, so a question remains: will enough partners participate to make a difference?
BOUQuET: Meta’s New AI Translation Benchmark
To evaluate the accuracy of its AI translation models, Meta is launching BOUQuET, an open-source benchmark designed to test AI translation using linguistically diverse sentences. Unlike traditional benchmarks that rely on commonly used phrases, BOUQuET ensures that AI models are tested on more natural and culturally relevant language patterns.
Meta’s move toward creating better translation benchmarks is a step in the right direction. AI-powered translation often struggles with nuance, context, and dialectal differences—particularly in underrepresented languages. Hopefully, BOUQuET provides meaningful insights that improve AI accuracy beyond simple word-for-word translation.
AI-Powered Dubbing for Reels
In a separate effort to improve accessibility, Meta has introduced an AI tool that can automatically dub Reels into different languages. The system not only translates speech but also attempts to match lip movements with dubbed audio. Initially, this feature is rolling out for English and Spanish content creators in the U.S.
Automated dubbing tools have been improving, but they’re still far from perfect. Meta’s approach—incorporating lip sync—could set it apart from simpler translation overlays. If it works well, this could help social media videos reach more global audiences. However, AI-generated voices sometimes lack natural tone and expression, which may limit how immersive the experience feels.
Meta AI Expands to 43 Countries
Beyond translation efforts, Meta’s AI assistant is now available in 43 countries, supporting over a dozen languages. This expansion indicates Meta’s continued push to integrate AI across its ecosystem, competing with other digital assistants like Google Assistant and Apple’s Siri.
With generative AI advancing rapidly, voice assistants need to do more than just answer basic questions. If Meta can leverage its AI translation and dubbing tools into its assistant, it could stand out by offering real-time multilingual support. That said, AI translation still has inconsistencies, so it will take time before assistants truly handle multiple languages fluidly.
Final Thoughts
Meta’s partnership with UNESCO and its AI translation initiatives show a commitment to linguistic diversity, but execution will be key. AI-powered speech recognition and translation have great potential, but accuracy and adoption remain challenges—especially for languages with limited data. The success of Meta’s efforts will depend on whether enough contributors join the Language Technology Partner Program and how well the AI can handle complex linguistic structures.
For Indigenous communities and linguists, this could be a valuable resource in language preservation, but skepticism is understandable. If Meta approaches this with genuine collaboration and transparency, it could be a meaningful improvement in AI translation.
