Meta's Most Recent AI Is Capable of Translating 200 Languages In Real Time

There are now about 7,000 languages spoken in our world, and Meta appears to desire to comprehend them all. Six months ago, the company started its ambitious No Language Left Behind (NLLB) initiative, which aims to train AI to translate between several languages without first going by English. The company unveiled its first great success, named NLLB-200. It’s an AI model that can communicate in 200 languages, including a number of lesser-known languages from Asia and Africa, such as Lao and Kamba.

The NLLB-200 can translate 55 African languages with “high-quality results,” according to a company blog post. Meta claims that the model outperformed existing state-of-the-art models on the FLORES-101 benchmark by 44 percent on average, and by up to 70 percent for specific African and Indian dialects.

Translating between any two languages, particularly if neither is English, has proven to be a big difficulty for AI language models, in part since many of these translation systems depend on written data scraped from the internet to train on. It’s really simple if you speak the language in which this line is written; it’s far more difficult if you’re seeking for quality content in Fan or Kikuyu.

Meta has opted to open-source NLLB-200, as it has done with the majority of its other publicly promoted AI initiatives, and also award $200,000 in grants to organizations to create real-world applications for the technology. For example, consider Facebook News Feed or Instagram. “Imagine visiting a favorite Facebook group, coming across a post in Igbo or Luganda, and being able to understand it in your own language with just a click of a button,” the Meta post hypothesized. On Meta’s example site, you can see how the new model works.