Meta Launches 200 Language Translation Open Source AI Model
Meta AI has just announced a newly built single AI model called “No Language Left Behind” (NLLB), which currently translates 200 different languages, considered as a significant breakthrough in machine translation capabilities.
Meta AI researchers created NLLB as an effort to develop high-quality machine translation capabilities for most of the world’s languages. As described, the new NLLB-200 supports 55 African languages with high-quality results. In total, this single model’s BLEU scores improve on the previous model by an average of 44% across all 10k directions of the FLORES-101 benchmark. For some African and Indian languages, the increase is greater than 70% over recent translation systems.
As announced, the research advancements from NLLB will support more than 25 billion translations served daily on Facebook News Feed, Instagram, and Meta’s other platforms. Modeling techniques and learnings from the NLLB research are now also being applied to translation systems used by Wikipedia editors based on a partnership with the Wikimedia Foundation.
To train the NLLB-200 model, which has 54B parameters, Meta AI leveraged its newly built Research SuperCluster (RSC), an AI supercomputer. To evaluate and improve NLLB-200, the research team built FLORES-200, a many-to-many evaluation dataset that enables researchers to assess performance in 40,000 different language directions.
While it is more challenging to optimize a single AI model to work across hundreds of languages without compromising performance or translation quality in comparison to traditional systems which have separate models for each language direction, Meta AI states that its model is scalable and more maintainable.
Meta AI is also open-sourcing the NLLB-200 model and publishing some research tools to enable other researchers extend the latest solution to more languages and build more inclusive technologies. Furthermore, Meta AI is providing up to $200,000 of grants to nonprofit organizations for real world applications for NLLB-200.
The NLLB research paper can be found here.