Soon Facebook Will Instantly Translate Your Posts Into 44 Languages
More than 1.5 billion people use Facebook. And only half speak English. The rest speak so many dozens of other languages, effectively silo’d off from the English speakers and, in many cases, from each other. It’s a case of social media being rather asocial.
But that’s changing. If you stumble onto a Facebook post in a foreign language, Facebook lets you instantly translate it—in a semi-effective way. And beginning today, millions of people will have the option of instantly translating their own posts into any one of 44 other languages, so that they will automatically show up in your News Feed in your native tongue. For the first time across the social network’s general population, Facebook is testing its “multilingual composer,” and though the initial test is limited, the aim is reach that far off point where everyone in the world can readily talk to everyone else. “That’s why I came to Facebook,” says Fazil Ayan, who oversees the company’s translation efforts and grew up in Turkey. “That’s my personal agenda.”
Businesses and celebrity types could already use this multilingual composer through Facebook’s Pages service. Each day, about 5,000 businesses and celebs publish nearly 10,000 posts in multiple languages. These are viewed about 70 million times a day, and more than a third of the time, they’re viewed in a foreign language. Ayan follows international footballers like Ronaldinho, a Brazilian star who uses composer to post not only in Portuguese, but Spanish and English. “I only see the English,” Ayan says. Now, millions of others can post in the same way.
Ayan and team designed the composer specifically for people with a multilingual audience. It also lets them edit the machine’s translation or even provide their own. But the ultimate goal is to automate the entire process, for everyone.
Machine translation is hardly perfect, but it’s improving. Today, Facebook will automatically translate among 45 languages, and it handles this task largely with traditional algorithmic models that rely on language statistics (essentially how often words and phrases appear in natural language). But when translating from English to German, the company is now leaning heavily on deep neural networks—networks of hardware and software inspired by the web of neurons in the brain—and this, Ayan says, provides a noticeable improvement. In recent years, deep neural nets have proven enormously adept at learning certain tasks—like recognizing faces in photos or identifying spoken words—by analyzing vast amounts digital data. Now, they’re also improving machine translation and natural language understanding, where a machine truly grasps the meaning of the words and sentences it translates. The plan is to push this tech across Facebook’s entire machine translation engine.
The same transformation is happening across the net. Microsoft’s Skype translation service leans on neural networks, and according Joseph Sirosh, who oversees Microsoft cloud computing services related to data and machine learning, the tech is moving into other Microsoft translation services as well. Certainly, neural nets are still a long way from mastering machine translation, even in tandem with other tech. But many researchers see a path toward that goal.
To reach this future, we need more and better data. That’s what neural nets thrive on. And Facebook’s multilingual composer plays a role here as well. Because people can edit translation and add their own, it generates additional data. This will be particularly helpful, Ayal says, with all the languages outside the 45 where the company already does translation. “From Catalan to Turkish? We don’t have enough data,” he says. “So this too will help with our overall mission: breaking language barriers.”