Meta wants to create real-time AI translation software for ‘hundreds of languages’

Meta is working on real-time translation software that the company says should work for “everyone in the world.” Meta CEO Mark Zuckerberg reports this during a presentation in which the company explains its AI plans. It is unknown when that software will be released.

Meta reports that the company is working on two translation projects. The company cites a “No Language Left Behind” project, through which Meta aims to develop an AI model that can learn languages ​​that have few written examples, such as languages ​​that are primarily spoken. The company mentions examples such as Asturian, Luganda and Urdu. Translation software generally uses written resources when learning a language.

In addition, Meta says it is working on ‘universal translation software’, which should be able to translate ‘hundreds of languages’ mutually in real time, without having to be transcribed first. The company says that many current translation software converts speech translations into text before they are actually translated.

Meta writes, among other things, that the translation software would benefit its own products. The company mentions in its blog post that the software can be used by AR glasses to translate real-time conversations, for example in a market abroad. The company also talks about real-time translations in virtual spaces in the metaverse. However, Meta does not report when this software should be ready for use, nor does it share a roadmap. The company just says it’s “making progress.”

Meta has been working on AI models and translations for some time. For example, the company uses Laser, an open source language processing toolkit that currently includes “more than 125 languages.” Meta says it has expanded this toolkit with speech capabilities to create the software. The company also talks about CoVoST 2, a dataset with 22 languages ​​that are mainly spoken. The company also recently built a supercomputer, which will eventually contain 16,000 Nvidia A100 GPUs and will be used for AI work. The company previously indicated that it would use the supercomputer for speech recognition and analysis of speech and text.

Meta made the announcement during an online Meta event focused on AI. There, the company also showed a prototype that allows users to create VR worlds based on descriptions.