In 2001, Google started providing a service that could interpret viii languages to as well as from English. It used what was as well as thence state-of-the-art commercial auto translation (MT), but the translation character wasn’t really good, as well as it didn’t improve much inwards those starting fourth dimension few years. In 2003, a few Google engineers decided to ramp upward the translation character as well as tackle to a greater extent than languages. That's when I got involved. I was working equally a researcher on DARPA projects looking at a novel approach to auto translation—learning from data—which held the hope of much improve translation quality. I got a telephone telephone band from those Googlers who convinced me (I was skeptical!) that this data-driven approach mightiness piece of occupation at Google scale.
I joined Google, as well as nosotros started to retool our translation organisation toward competing inwards the NIST Machine Translation Evaluation, a “bake-off” amid enquiry institutions as well as companies to construct improve auto translation. Google’s massive computing infrastructure as well as powerfulness to squash vast sets of spider web information gave us potent results. This was a major turning point: it underscored how effective the data-driven approach could be.
But at that fourth dimension our organisation was equally good irksome to run equally a practical service—it took us twoscore hours as well as 1,000 machines to interpret 1,000 sentences. So nosotros focused on speed, as well as a twelvemonth afterward our organisation could interpret a judgement inwards nether a second, as well as amongst improve quality. In early on 2006, nosotros rolled out our starting fourth dimension languages: Chinese, as well as thence Arabic.
We announced our statistical MT approach on Apr 28, 2006, as well as inwards the half-dozen years since as well as thence we’ve focused primarily on essence translation character as well as linguistic communication coverage. We tin orbit notice straightaway interpret amid whatsoever of 64 unlike languages, including many amongst a pocket-sized spider web presence, such equally Bengali, Basque, Swahili, Yiddish, fifty-fifty Esperanto.
Today nosotros pick out to a greater extent than than 200 1 chiliad one thousand monthly active users on translate.google.com (and fifty-fifty to a greater extent than inwards other places where y'all tin orbit notice operate Translate, such equally Chrome, mobile apps, YouTube, etc.). People also seem eager to access Google Translate on the become (the linguistic communication barrier is never to a greater extent than astute than when you’re traveling)—we’ve seen our mobile traffic to a greater extent than than quadruple twelvemonth over year. And our users are really global: to a greater extent than than 92 percentage of our traffic comes from exterior the United States.
In a given twenty-four hours nosotros interpret roughly equally much text equally you’d honour inwards 1 1 chiliad one thousand books. To seat it some other way: what all the professional person human translators inwards the footing hit inwards a year, our organisation translates inwards roughly a unmarried day. By this estimate, almost of the translation on the planet is straightaway done yesteryear Google Translate. (We can’t utter for the galaxy; Douglas Adams’s “Babel fish” likely has us vanquish there.) Of course, for nuanced or mission-critical translations, aught beats a human translator—and nosotros believe that equally auto translation encourages people to utter their ain languages to a greater extent than as well as behave on to a greater extent than global conversations, translation experts volition endure to a greater extent than crucial than ever.
We imagine a futurity where anyone inwards the footing tin orbit notice eat as well as portion whatsoever information, no thing what linguistic communication it’s in, as well as no thing where it pops up. We already render translation for webpages on the wing equally y'all browse inwards Chrome, text inwards mobile photos, YouTube video captions, as well as speech-to-speech “conversation mode” on smartphones. We desire to knock downward the linguistic communication barrier wherever it trips people up, as well as nosotros can’t hold off to come across what the adjacent half-dozen years volition bring.