Meta's Open-Source AI Translation Initiative - Open Source For You
Meta with its latest efforts to support underserved languages in AI and, in collaboration with UNESCO, inviting partners to help enhance machine translation and language technologies to promote linguistic diversity and digital inclusivity.
Meta’s Fundamental AI Research (FAIR) team is dedicated to advancing machine intelligence (AMI) capable of performing complex cognitive tasks, including translation. The goal is leveraging AI for inclusivity and innovation, ensuring intelligent systems can adapt and learn across diverse languages and contexts.
Collaboration with UNESCO for Linguistic Diversity
A key focus of this effort is their collaboration with UNESCO to support underserved languages in AI models. By developing multilingual AI systems, they promote linguistic diversity and digital inclusivity while enhancing AI’s ability to adapt to new linguistic environments. They introduce new research programs and invite collaboration on AI translation technologies that incorporate a broad spectrum of global languages and dialects.
Meta is seeking partners to help expand its open-source language technologies, with a special focus on underserved languages. This initiative aligns with UNESCO’s International Decade of Indigenous Languages, aiming to preserve and strengthen linguistic heritage.
Partners can contribute by providing:
- Data sets for underserved languages
- Linguistic expertise
- Feedback on AI translation models
These contributions will help refine AI-driven speech recognition and machine translation models, which will be open-sourced for public use. Participants will also gain access to exclusive technical workshops led by Meta researchers, offering guidance on integrating open-source models into language technologies.
Open-Source Machine Translation Benchmark
To enhance AI translation accuracy, they are launching an open-source machine translation benchmark. This dataset, crafted by linguistic experts, will serve as a standardized test for evaluating AI translation models. The benchmark, currently available in seven languages, invites contributors to submit translations that will be open-sourced. Our aim is to create an expansive multilingual translation benchmark for global AI research.
Meta remains committed to advancing AI-powered language tools. In 2022, they launched No Language Left Behind (NLLB), an open-source translation model covering many languages, setting a foundation for future research. They also partnered with UNESCO and Hugging Face to develop an AI translator based on NLLB.
Most recently, they introduced the Meta Massively Multilingual Speech (MMS) project, expanding AI-driven audio transcription to over 1,100 languages. In 2024, we enhanced MMS with zero-shot speech recognition, enabling AI to transcribe languages it has never seen before without prior training.
Building Inclusive AI Communication
By continuing to innovate in machine translation and speech recognition, they strive to build AI that understands and responds to complex human communication across all languages and cultures.
For more information, click here.
