Meta's New Open Source AI Can Translate 200 Languages

Meta's artificial intelligence-based 'No Language Left Behind' (NLLB) project can now reportedly translate 200 different languages. Meta founder and CEO Mark Zuckerberg first revealed the NLLB project during a livestream on Facebook earlier this year. Talking about the software, Zuckerberg said that the company is working on not one but two machine translation projects that will make it easier for people to communicate with others even if they speak a rare or low-resource language.

As for NLLB, Meta describes it as an "advanced AI model that can learn from languages with fewer examples to train from." The company also says that it will be used to enable high-quality translations to and from hundreds of languages, including many that are not yet adequately addressed by the current machine translation tools. Meta also touted a universal speech translation system that it said can translate speech from one language to another in real-time to support languages without a standard writing system.

SCREENRANT VIDEO OF THE DAY

Related: Should All AI Be Open-Sourced And What Would Actually Change?

According to Meta, its new No Language Left Behind project aims to build an AI model that can translate many more languages with precision and accuracy than any other available machine translation tool. The company on Wednesday announced that the system can now translate across 200 languages and said that it intends to use the technology to improve translations on Facebook and Instagram before eventually deploying it in the metaverse. The company also open-sourced the AI-based project and made it available on Github for anyone else to use. Meta also announced grants of up to $200,000 to nonprofit organizations for real-world applications of the new technology.

Not All Languages Have The Same Digital Resources

Meta logo on brain neural

While some languages like English, Spanish or Mandarin have enough resources that make it easier to offer machine translation solutions, others like Asturian, Luganda or Urdu are not so lucky. These languages and many others do not have the same level of digital resources to enable high-quality machine translation solutions. This is where Meta believes its new AI model comes in. According to the company, the NLLB project is an 'AI superpower' that can translate across 200 languages, including many that are not supported by current translation systems.

As an example of the problem faced by native speakers of obscure languages, Meta said that less than 25 African languages are currently supported by available translation tools. What's worse is that many of these are of relatively poor quality, thereby preventing native speakers from accessing the same online resources as everyone else. In contrast, NLLB supports 55 African languages "with high-quality results." The improved quality over current resources is reportedly especially noticeable with African and Indian languages. People can check out examples of NLLB's translations on Meta's demo site.

2022-07-06T19:03:32.000Z

Comment