Meta-UNESCO partnership will bring AI translation to underserved languages worldwide

Meta has launched a new Language Technology Partner Programme in collaboration with UNESCO to enhance AI translation and speech recognition capabilities for underserved and indigenous languages worldwide.

Program Overview; The initiative, spearheaded by Meta’s Fundamental AI Research (FAIR) team, combines open-source AI models with research to expand linguistic diversity in digital spaces.

• The program seeks partners to contribute speech recordings, transcriptions, and translated text to improve AI language models
• The Government of Nunavut, Canada has already joined the initiative, providing data for Inuktitut and Inuinnaqtun languages
• Partners must contribute at least 10 hours of speech recordings with transcriptions and 200+ sentences of written text

Technical Implementation; Meta is introducing new tools and benchmarks to evaluate and improve AI language capabilities.

• An Open Source Translation Benchmark has been launched to assess machine translation model performance across seven languages
• All translations and resources will be made open source and freely available to the community
• Partners will receive access to technical workshops led by Meta researchers

Building on Previous Success; Meta’s new program extends its existing commitment to language technology development.

• The 2022 No Language Left Behind (NLLB) project created an open-source machine translation engine
• The Meta Massively Multilingual Speech (MMS) project expanded speech recognition to over 1,100 languages
• Recent updates include zero-shot transcription capabilities, allowing AI to transcribe previously unencountered languages

Future Impact and Implications; The success of this initiative could significantly affect digital inclusion and cultural preservation efforts, though challenges remain in gathering sufficient data for truly comprehensive language coverage and ensuring accurate translations across diverse linguistic contexts.