Zoubin Ghahramani, Vice President of Google Research, speaks about Google’s “1000 Languages” initiative at a Google AI event in New York. Image: Sabrina Ortiz/ZDNET
At a conference on artificial intelligence held this week in New York, Google announced its intention to multiply by ten its already very extensive language portfolio, thanks to artificial intelligence.
Google is committed to building an AI model that can support the world’s 1,000 most spoken languages to make information more accessible.
“Language is a fundamental part of communicating and understanding the world,” said Jeff Dean, senior researcher at Google. “But more than 7,000 languages are spoken around the world, and only a few are well represented online today.”
Towards a universal voice model
As the undertaking is extremely ambitious, the project will likely take many years to bear fruit. However, Google is already working towards its goal.
The tech giant has developed a Universal Speech Model (USM) that has been trained on more than 400 languages, making it the most extensive speech model to date, according to a blog post. Google has also partnered with communities around the world to obtain voice data.
Google’s focus on expanding its language capabilities is not new. Recently, Google added 24 more languages to its Google Translate platform and enabled voice typing for nine more African languages on Gboard.
Meta is also working on its translation AI model
Google is also working with local governments, NGOs and academic institutions in South Asia to collect audio samples of the region’s various dialects.
Other big tech companies are also building big language models. In July, Meta announced an AI model called No Language Left Behind that can translate into 200 languages.
Meta’s efforts have also been made to bring content to communities that are not represented on the web. Meta’s AI model includes translations for 55 African languages, which is a significant step forward, since less than 25 African languages are supported by widely used translation tools.
Source: ZDNet.com