Meta’s No Language Left Behind

Apparently there are over 1700 languages spoken in the world today. Wow. Given that there are only 195 recognised countries, that’s a lot of languages. So being able to translate is important to any company wanting to address a worldwide customer base.

Meta, the company formerly known as Facebook, has released an open source AI language translation model called No Language Left Behind (NLLB). The latest release, NLLB-200, supports 200 languages. That’s a long way from 1700, but about double what was available in any single model before. NLLB includes 55 languages from Africa alone as well as other lesser known languages such as Asturian and Lao.

Facebook claims the NLLB-200 model to be on average 44% more reliable than previous automatic translation systems – though, to be honest, that’s starting from a low baseline. As well as NLLB, Meta is also releasing its FLORES-200 dataset for evaluating translation effectiveness.

Meta says:

We’re committed to bringing people together. That’s why we’re using modeling techniques and learnings from our NLLB research to improve translations of low-resource languages on Facebook and Instagram. By applying these techniques and learnings to our production translation systems, people will be able to make more authentic, more meaningful connections in their preferred or native languages. In the future, we hope to extend our learnings from NLLB to more Meta apps.

One of these apps will, of course, be the metaverse.