Meta introduced the NLLB-200 artificial intelligence model for online transmissions. The algorithm supports 200 languages, including less common ones.

🆕🆕🆕❗️NLLB-200 Model, Evaluation Dataset and Paper with improved translation quality for 200+ languages. Help advance AI translation for low-resource languages.
📃Paper: https://t.co/8TWbAbVOfu
⚙️Model: https://t.co/9m9m2VKizB
🛤️Comparison: https://t.co/mxSuPv2Ey2 pic.twitter.com/wgzQlOBerY

— MetaAI (@MetaAI) 6 July 2022

The company said the NLLB-200 provides “high-quality” translations in 55 African languages. The developers added that before the creation of the algorithm, less than 25 African languages ​​were covered by widely used tools.

Meta claims that the NLLB-200’s performance in the FLORES-101 test outperforms current current models by an average of 44%. For some African and Indian dialects, the advantage over analogues reached 70%.

Meta said its improvements will improve the quality of translations for text on Facebook and Instagram platforms, as well as third-party services like Wikipedia.

The company has made the NLLB-200 source code available on GitHub. The developers have created a website that showcases the work of the new model.

Meta also announced a $200,000 grant program open to nonprofits who want to explore ways to implement the NLLB-200 in the real world.

Recall that in June Meta announced a reorganization of its AI section.

In April, the company introduced the open source language model OPT-175B.