Jonah Peretti, the founder of BuzzFeed, disclosed in a recent letter to investors that its traffic tripled over the past year, hitting 85 million visitors in August. Soon—thanks to our collective inability to resist such sugary listicles as "36 Things You Never RealizedEveryone Else Does Too"—Mr. Peretti says, BuzzFeed will be one of the largest sites on the Web.
Until recently, though, BuzzFeed's towering traffic ambitions were held in check by a simple fact of global demographics. Everything BuzzFeed publishes is in English—and at the rate it's growing, BuzzFeed may be running out of new English speakers to colonize.
Mr. Peretti has long believed that BuzzFeed's appeal is universal, that people in Paris and Mexico City would be just as engrossed by "32 Cats Who Were Way Too Curious For Their Own Good" as are folks in New York and San Francisco. But going international presented logistical challenges. How do you translate dozens of BuzzFeed posts, many of them lousy with English idioms, into several languages every day, within a few hours of each story's publication?
Now BuzzFeed has found a method to do so—a system that's simultaneously brilliant, brazen, and more than a little nutty.
The site this month will launch versions in French, Spanish and Brazilian Portuguese. These international sites will be populated with BuzzFeed posts that originally appeared in English, but BuzzFeed won't be using professional translators to create them. Instead, BuzzFeed's posts will be translated by crowds of foreign-language speakers who are learning English using an app called Duolingo. In theory, as part of their coursework, these hordes will translate a BuzzFeed post in a matter of hours—at a quality that rivals that of professional translators, but at the speed, scale and price that you'd get from a machine.
If it works—and BuzzFeed's tests say it does—the effort could prove the utility of something known as human computation, a theory that argues that instead of rivaling one another, machines and humans can get more done by banding together.
Crowdsourced translation isn't new. Many sites, including, most famously, Facebook, have asked their loyal users to adapt their sites to different countries. But Facebook's was a one-time translation of its interface text. BuzzFeed and Duolingo's system will be larger, more routine and more Rube Goldbergian.
Duolingo is something like a videogame version of Rosetta Stone, and it's been found to be quite effective at teaching people new tongues. According to a study commissioned by the company, in about 34 hours with Duolingo, a person with no knowledge of Spanish will become as proficient as someone who's taken a first-semester college Spanish course. As of last month, the app, which is free to use, had garnered about 10 million users.
When you first begin using it, Duolingo teaches you the most basic concepts of a new language. As you become more adept, you're asked to translate texts as a test of what you've learned.
But why should people waste their energies translating dummy texts? Shouldn't their work amount to something?
That's the theory behind Duolingo's partnership with BuzzFeed. As part of their lessons, language learners will be presented with BuzzFeed posts that are trending in popularity.
For instance, people learning French might be asked to translate as much of "24 People Who Are Really Nailing This Parenting Thing" as they can. Duolingo will show each piece to dozens of people. Once enough French learners see and translate a piece, Duolingo's systems will shuttle a translated version to BuzzFeed, which will publish it internationally.
The system fits into a larger framework pioneered by Luis von Ahn, Duolingo's founder. Mr. Von Ahn says that in many human endeavors, people are spending a lot of time and energy on "wasted" work. Why not put their efforts to use?
Consider the Captcha, those squiggly letters that you're asked to decipher as a way to prove to computers that you're a human. Mr. Von Ahn is one of the engineers credited with creating that system.
But in 2007, he began to feel guilty about forcing people to spend so much time doing busy work. At the same time, big companies like Google and Amazon.com had begun projects to scan books into digital text, but computers are notoriously bad at reading old-timey typesetting.
Mr. Von Ahn had an epiphany: Instead of asking people to decipher dummy text as part of Captcha, why not give people images of words from scanned books? Thus was born reCaptcha, which Google purchased in 2009, and which now helps digitize more than two million books a year.
Mr. Von Ahn calls this human computation, and you can think of it as a specialized instance of crowdsourcing. People engaged in human computation aren't trying to accomplish work together. Instead, they're doing something useful incidentally, as a side-effect of some other endeavor.
At least for translation, the system could be very effective. A professional human translator costs about 20 cents a word. Mr. Von Ahn says Duolingo can translate dozens of articles into many different languages simultaneously at a small fraction of that price.
He says it can match the quality of humans, especially when it comes to colloquialisms. One example: English speakers who were learning German were asked to translate a document that included the German phrase "it is the drop that makes the barrel overflow." That's not an English idiom, but several people translated it into "the straw that broke the camel's back," which is essentially the same thing.
BuzzFeed's tests support this. The cryptic-sounding "24 people who are really nailing this parenting thing" was translated into "24 parents qui ont trouvé le truc pour éduquer leurs enfants," which French natives tell me is spot on.
Mr. Von Ahn says BuzzFeed will be the first of many paying clients to translate its work through Duolingo. With its millions of users, he says the app has enough capacity to earn "tens of millions of dollars" a year from its translations.
In a sense, when you click on a BuzzFeed link, you're subsidizing language education for millions of people around the world.
"It feels like something out of a William Gibson novel," says BuzzFeed's Mr. Peretti. "There's a certain cyberpunk logic to it that got us very excited."
—High Definition is a twice-weekly column about technology issues, people and companies.
A version of this article appeared October 13, 2013, on page B4 in the U.S. edition of The Wall Street Journal, with the headline: BuzzFeed's Brazen, Nutty, Growth Plan.
No comments:
Post a Comment