Tremble, Google Translate: Meta prepares real-time translator for 200 languages

It seems that Target wants to end the reign Google translator and developed an ambitious plan for this. Parent company Facebook, Instagram D whatsapp introduced “Not a single language left behind-200” (NLLB-200 or “No Language Shall Be Forgotten” for its Spanish translation), AI-based real-time translator that will support 200 languages.

As detailed in his official blogthe company wants to develop a platform that includes a large number of languages, including those that have not yet been taken into account by other tools. NLLB-200 will have support in 40,000 addresses different thanks to the combinations of 200 languages ​​that you can translate leaving behind 133 which supports Google translator.

This equation includes Languages which are not as common on well-known translation platforms such as “low resource languages”spoken by a small group of people in the world or who have less than a million available phrases to learn AI. For example, it is mentioned that some languages India or Africa they will also have translated phrases in the system.

Artificial intelligence as a means of translation

Scientific research carried out Target explains in more detail how this new platform was developed real time translation. According to the document, the company has merged its Target AIa human translation verification system called FLOWERS-200 and mechanisms for creating learning phrases in order to improve the translator and his results.

In order to get a decent or optimal translation, Meta created a test database with 3001 couple sentences in each of the languages ​​supported by your platform. Each of them was translated from English into the target language thanks to the team professional human translators who also speak the language by birth.

Based on these results, machine translation was compared to human links using Understudy Bilingual Assessment (BLEU)a reference point quite popular when it comes to translations. The test was very encouraging as the Meta model was considered 44% more accurate than the best automatic translation systems that currently exist.

However, this does not mean that all translations according to this system are error-free, since, for example, it is more difficult to identify errors in translation. languages ​​with few resources therefore, the results in such cases should be considered from a different point of view. Also, joining lists toxicity detect and avoid profanity or potentially offensive content also important.

A translator where everyone can contribute

Target pointed out that NLLB this is a project open source therefore, its code is open to the general public, and even the research tools used to develop it will be made public. This means that thanks to the work of the Internet community, more languages ​​could be integrated And up translations could be more accurate in future.

Finally, Target announced a partnership with Wikimedia Foundationthe foundation Wikipedia, to support the translation systems of the great Internet encyclopedia. Target NLLB is that more of the content available on the web can be understood by people even if they don’t know the language in which it is written.

There is currently a demonstration of how this system works and can be explored using following link. However, there is no exact release date yet.

We recommend you METADATA, an RPP tech podcast. News, analytics, reviews, recommendations and everything you need to know about the world of technology. To hear better, #StayHome.