According to reports from The Verge, Mark Zuckerberg’s Meta is developing an early-stage AI translation model that can work with over 200 languages from all over the world.
The report suggests that Meta will dig deep into this project and include a lot of languages that arent supported in many commercial tools. They are also open-sourcing the project so that others can chime in on their work. This AI model is supposed to be a dream R&D project and Meta is prioritizing creating their “Universal Speech Translator”. They also believe that this will help in the company’s growth across both Instagram and Facebook, it’ll also be beneficial in developing domains like AR and VR.
This in turn will also help them boost the advertisements they show to their users, which comprises 97% of their revenue.
The use of machine translation is going to come in very handy for Meta as the company will be able to understand the wants of their users better without the language barrier. This in turn will also help them boost the advertisements they show to their users, which comprises 97% of their revenue.
Machine translation experts told the website that although Meta’s latest research was ambitious and precise, the quality of translations done by the model would probably be below that of languages that are better-supported, such as Italian or German. Translation can be a touchy subject as it might rub some people the wrong way.
the lack of importance given to low-resource languages is what made them work on this project so dedicatedly.
Angela Fan, Meta Research Scientist.
Often times we see people posting something on Facebook with great intent only for it to be lost in translation. It gets pretty heated when the app translates something and gets a derogatory result as it can lead to some controversies. Meta has made sure that they have done thorough research over this as their researchers have run quite a few language tests using a benchmark common in machine translation known as Bilingual Evaluation Understudy (BLEU).
Angela Fan, Meta’s research scientist said that the lack of importance given to low-resource languages is what made them work on this project so dedicatedly. The features which are developed in this model will be put soon into Meta’s translation tools.