October 2, 2022

Meta has taken another step towards creating a universal language translator.

The company has released an open source artificial intelligence model that translates more than 200 languages ​​- many of which are not supported by current systems.

The search is part of the Meta initiative launched earlier this year.

Greetings people

Subscribe to our newsletter now to get a weekly summary of our favorite AI stories right in your inbox.

We call this project There is no language behindThe AI ​​modeling techniques we used from NLLB help us make high-quality translations on Facebook and Instagram for languages ​​spoken by billions of people around the world, Mark Zuckerberg, CEO of Meta, said in a Facebook post.

NLLB focuses on low-resource languages, such as Maori or Maltese. Most people in the world speak these languages, but they lack the training data that AI translations usually require.

The new Meta model is designed to overcome this challenge.

To do this, researchers first interviewed speakers of disadvantaged languages ​​to understand their needs. Then they developed a new data mining technique that generates training sentences for Low resource languages.

Next, they trained their model on a mixture of mined data and human-compiled data.

The result is NLLB-200 – a massive multilingual translation system for 202 languages.

The team evaluated the performance of the model on the FLORES-101 dataset, which evaluates translations of low-resource languages.

“Despite doubling the number of languages, our final model performs 40% better than the most recent predecessor model on the Flores-101,” Study authors wrote.

SOTA مقارنة comparison