Facebook claims that it is now providing 2 billion text translations per day thanks to its automated system. Additionally, close to half of Facebook’s massive user base, 800 million people in total, now see translations each month.
Those insane statistics come courtesy of Alan Packer, the man in charge of Facebook’s language technology. Packer revealed the figures during his talk at MIT Technology Review’s Emtech Digital conference in San Francisco, which itself centers on artificial intelligence and machine learning. The context is important as Packer took the opportunity to announce Facebook’s planned move to a new translation system based on artificial neural networks.
Facebook’s present system took over translation tasks from the social network’s crowdsourcing tool, which relied on its users in various countries to help with the process, in 2011. The automated software can detect 40 different languages in 1,800 directions, such as French to English, reports TechCrunch.
Facebook language stats. In case you're curious: emojis get no translation. #universallanguage? #EmTechDigital pic.twitter.com/q5rFWAAY0k
— leahthehunter@gmail.com (@leahthehunter) May 23, 2016
The move to AI is expected to be completed later this year. Packer claims that compared to the statistical machine translation currently being utilized, neural networks will produce more natural-sounding translations.
The large stores of data that Facebook has access to, which includes an archive of roughly 2 trillion posts and comments, are being used to teach its AI more accurate interpretations of global idioms and metaphors. One particular case referenced by Packer in his talk saw the AI spot a French variation on the word “wow,” which was being changed to “uau” by teenagers in the country. As a result, the machine-learning system now translates that term appropriately.
Packer also stated that its popular translation software is leading users to “have more friends, more friends of friends, and get exposed to more concepts and cultures.”
