The social networking giant Meta has developed a single AI model that can translate between 200 distinct languages, many of which are not currently supported by commercial solutions. In an effort to encourage others to expand on its work, the business is open-sourcing the project.
The AI model is a component of Meta’s ambitious R&D initiative to produce a so-called “universal voice translator,” which the company views as crucial for development across its multiple platforms, including Facebook and Instagram as well as emerging fields like VR and AR.
Machine translation might be the basis of game-changing software for next projects like Meta’s augmented reality glasses. By better understanding its consumers, Meta is able to improve the advertising systems that account for 97 percent of its income.
The quality of some of the model’s translations would probably be well below that of better-supported languages like Italian or German, according to experts in machine translation, who also praised Meta’s most recent study for being ambitious and comprehensive.
Professor Alexander Fraser, a computational linguist at LMU Munich in Germany, said that the main contribution I, in this cases data. The fact that Meta’s model can translate 100 additional languages is noteworthy.
Contrary to popular belief, Meta’s successes are a result of both the breadth and the depth of its research. Most machine translation models can only handle a few languages, but Meta’s model is all-encompassing; it’s a single system that can translate between 200 different languages in more than 40,000 distinct ways.
The inclusion of “low-resource languages” in the model, or languages with less than 1 million publicly available translated sentence-pairs, is another goal of Meta. Many of the African and Indian languages spoken there are not often supported by commercial machine translation software.
To read our blog on “‘Metaverse Academy’ will be launched soon by Meta in France,” click here
