Google Inc. announced in November that it would be using an artificial intelligence technique called neural machine translation to completely change the way Google Translate understands languages, with the ultimate goal of creating a universal language translator.

Today the company came a little closer to that vision by rolling its system out to three more languages: Hindi, Russian and Vietnamese.

Translating languages is hard, even for a computer. Google Translate has always been an impressive piece of technology, but its translations rarely match the fluency of a natural speaker, especially for larger blocks of text. Google says that neural machine translation is changing that, allowing Google Translate not only to understand languages better, but also to create more natural-sounding translations.

Phrase-Based Machine Translation, Google Translate’s old way of doing things, would break sentences up into phrases and translate each piece separately. While the parts themselves might be accurately translated, when they were put back together in a whole sentence, they did not always fit the way they should. The phrasing and sentence structure would sound off to a native speaker, even if they could still understand the general meaning of the sentence.

“Neural translation is a lot better than our previous technology, because we translate whole sentences at a time, instead of pieces of a sentence,” explained Barak Turovsky, product lead for Google Translate. “This makes for translations that are usually more accurate and sound closer to the way people speak the language.”

Neural machine translation tries to understand sentences much like humans do, by looking for context clues that give a better idea of a sentence’s overall meaning. Google noted on its research blog that its test with Chinese-to-English neural translation showed a reduction in translation errors by 55 to 85 percent compared with phrase-based translation of the same text. Google admitted that neural translation “can still make significant errors that a human translator would never make,” but the company called its method “a significant milestone” in machine translation.

Turovsky said that Google will be rolling out its improved AI translations to several other languages over the next few weeks, and Google hopes eventually to roll it out to all 103 languages supported by Google Translate.

Image: Google