Facebook announces an open source AI model that can translate directly between 100 languages

new.blicio.us Follow Oct 25, 2020 · 2 mins read
Facebook announces an open source AI model that can translate directly between 100 languages
Share this

Facebook has announced a major new development in its machine translation tools, a development that could help improve communications between users around the world on its platforms.

The company claims to offer 20 billion translations daily in their News Feed, and now thanks to their M2M-100 artificial intelligence they claim to be able to translate between any of the 100 languages without going through English. From Chinese to French without going through English.

This is an important change, and Facebook says that “breaking down language barriers is one of the most important ways to bring people together, providing authoritative information about COVID-19 and keeping them safe from harmful content.

Although translation tools have improved a lot in recent years (DeepL is a good example of this), we have all come across translations that are somewhat robotic or that don’t make much sense.

Facebook clarifies that when translating (for example) from Chinese to French, “most English-centric multilingual models train in Chinese to English and English to French, because English training data is the most available.

With M2M-100, Facebook trains directly on Chinese-to-French data “to better preserve meaning. They claim to be able to outperform English-centric systems by 10 points (in the BLEU metric, used to evaluate machine translations).

This is a really important change, considering that of the billions of daily postings in 160 languages on Facebook, two-thirds are in a language other than English. This is how Fact Check Explorer, Google’s anti-bullying search engine, works In Genbeta This is how Fact Check Explorer, Google’s anti-bullying search engine works

This is what Angela Fan, a research associate at Facebook AI, told Engadget. She says there are many regions in the world where people speak several languages, and none of them are English. It doesn’t seem too effective then to use translation systems that “rely heavily on English only data.

Although these advances are already significant in themselves, one of the most important points of this project is that it becomes open source for the research community and thus helps to drive new developments.

Facebook has made it clear that it also plans to continue developing this system independently and will eventually incorporate the resulting technology into its products.

Written by new.blicio.us Follow