Meta's Fundamental AI Research (FAIR) team is working towards advanced machine intelligence that can perform complex tasks such as translation. This initiative aims to support underserved languages and promote linguistic diversity in the digital world.
The FAIR team, in collaboration with UNESCO, is developing AI models that work on multilingual problems and adapt to new situations. They have announced recent programs, research, and models that support this goal and are inviting collaborators to contribute to AI translation technologies covering a wide range of global languages.
The Language Technology Partner Program seeks partners to help advance Meta’s open-source language technologies. The focus is on underserved languages as part of UNESCO's International Decade of Indigenous Languages. Partners are expected to provide speech recordings with transcriptions, written texts, and translated sentences in diverse languages. These contributions will aid in integrating these languages into AI-driven speech recognition and machine translation models, which will be made freely available upon release.
Partners will also access technical workshops led by Meta's research teams to learn how to use open-source models for building language technologies. The Government of Nunavut in Canada has agreed to collaborate by sharing data in the Inuit languages Inuktitut and Inuinnaqtun.
An open-source machine translation benchmark has been launched alongside the partner program. This standard test evaluates AI model performance in translation using sentences crafted by linguistic experts across seven languages. Contributors can access the benchmark and add translations that will be shared openly.
These initiatives align with Meta's long-term commitment to supporting underserved languages. In 2022, they released the No Language Left Behind project, an open-source machine translation engine for many languages developed with UNESCO and Hugging Face. The recent Meta Massively Multilingual Speech project scales audio transcription across over 1,100 languages and includes zero-shot speech recognition capabilities introduced in 2024.
Meta aims to create systems that understand complex human needs regardless of language or cultural background through collaborative enhancement of machine translation and other language technologies.
