Meta’s Fundamental AI Research (FAIR) team is taking significant strides toward achieving Advanced Machine Intelligence (AMI), aiming to develop AI that mirrors human reasoning in complex tasks such as translation. As part of this mission, Meta has partnered with UNESCO to enhance AI support for underserved languages, fostering linguistic diversity and digital inclusivity.
Advancing AI for Multilingual Accessibility
Recognizing the importance of preserving and integrating indigenous and lesser-known languages into the digital landscape, Meta’s latest initiatives seek to expand AI-driven translation and speech recognition capabilities. These efforts not only enrich global communication but also strengthen AI’s adaptability to new linguistic and cultural contexts.
New Initiatives for AI-Powered Translation
Meta has introduced two key programs aimed at fostering collaboration and innovation in AI translation technologies:
Language Technology Partner Program
Meta is calling on partners to help broaden its open-source language technologies, particularly in underserved languages. This initiative aligns with UNESCO’s International Decade of Indigenous Languages, aiming to provide crucial linguistic data for AI model training.
Potential collaborators are invited to contribute speech recordings (10+ hours), written text (200+ sentences), and translated sentence sets. In return, partners will gain access to Meta’s technical workshops and work closely with AI researchers to integrate their languages into machine learning models.
The Government of Nunavut, Canada, has already joined the initiative, providing data in the Inuit languages Inuktitut and Inuinnaqtun, marking a significant step in preserving indigenous languages through AI.
Open Source Translation Benchmark
Meta is also launching an open-source machine translation benchmark—a standardized test designed to evaluate AI translation models. This benchmark includes linguistically diverse sentences curated by experts and is currently available in seven languages.
The initiative invites global contributors to submit translations, helping to build an unparalleled multilingual translation benchmark that will be freely accessible to researchers and developers.
A Continued Commitment to Language Diversity
Meta’s recent announcements build on its longstanding efforts to democratize AI-powered language technologies:
- No Language Left Behind (NLLB): In 2022, Meta launched the first neural machine translation model for numerous languages, setting the foundation for future research.
- Partnership with UNESCO & Hugging Face: A language translation tool was introduced during the 2023 UN General Assembly, leveraging NLLB.
- Meta Massively Multilingual Speech (MMS): This project, scaling AI audio transcription to over 1,100 languages, now includes zero-shot speech recognition, enabling transcription in previously unseen languages.
Shaping the Future of AI Translation
Meta’s vision is to develop intelligent systems capable of understanding and responding to human needs, regardless of language or cultural background. Through open-source collaboration and cutting-edge research, the company continues to push the boundaries of AI-driven linguistic accessibility.
For those interested in joining Meta’s Language Technology Partner Program or contributing to the open-source translation benchmark, participation details are available on Meta’s official platforms.
