Zuckerberg announced: The new development that will translate content from 200 languages in high quality
The new model, which was first developed from a single language and later expanded to a variety of other languages, knows how to learn from languages with fewer examples of practice and produce excellent quality translations.
"The technological breakthrough will provide translation to more than 25 billion content on our platforms every day," Zuckerberg said.
Yinon Ben Shoshan
Thursday, 07 July 2022, 11:28 Updated: 11:37
Share on Facebook
Share on WhatsApp
Share on Twitter
Share on Email
Share on general
Mark Zuckerberg (Photo: GettyImages)
Mark Zuckerberg, founder and CEO of Meta (Facebook), yesterday (Wednesday) presented a breakthrough in the "No Language Left Behind" project, which is designed to build language tools and machine learning that will include most languages in the world. 200 different languages of the highest quality, including rare languages that have not been supported by existing translation tools to
date.The new model, first developed from a single language and later expanded to a variety of other languages, knows how to learn languages with fewer practice examples and produce excellent quality tests - tested over 10,000 Content has shown that the translations produced by the model provide an average of 44% more accurate and high-quality translation, compared to previous products.
Good to know (promoted)
7 risk factors for heart disease and one way to identify early and prevent heart damage
Served on behalf of Shahal
Content showed that the translations produced by the model provide a more accurate and high-quality translation by an average of 44% (Photo: GettyImages)
The breakthrough of Meta's AI teams will also be integrated into the company's platforms, and will help improve the quality of translation on Facebook and Instagram in a variety of languages.
The quality translations will allow, among other things, better control over the content that goes up to the platforms, locate offensive content and enforce publications that violate community rules if necessary.
In addition, Meta will also collaborate with Wikimedia, the company responsible for the Wikipedia site, in order to improve the quality of translations of various entries on the site - and help the site's editors publish in a wider variety of languages.
"It's amazing to see how much artificial intelligence enhances all of our services. Today we unveil a new AI model we've developed that is capable of translating more than 200 different languages and includes many languages that are not supported by existing translation services," Zuckerberg said.
He added: "We call this initiative 'No Language Left Behind' and we present our research breakthrough as an open source for industry use; the techniques and models we build as part of the project will help produce quality translations into languages spoken by billions of people around the world."
Zuckerberg added that "to put it bluntly - our model, which currently supports 200 languages, uses more than 50 billion different parameters. We trained it using our new supercomputer, one of the fastest and most advanced supercomputers in the world. The technological breakthrough will provide more translation "From 25 billion content on our platforms every day. Interpersonal interaction in different languages is a force that artificial intelligence will now allow."