Almost a decade after the launch of Google Translate, computer translations may finally stop being a punchline.
On Tuesday, Google unveiled the Google Neural Machine Translation system, a system by which it hopes to improve the machine learning that informs its linguistic tools. And it’s kicking things off with one of the most difficult language pairs around: Chinese and English.
Older translation technology, called Phrase-Based Machine Translation, broke down any input into its component words and phrases and dealt with translating those independently. This led to skipped words, trouble with rare words, and really clunky output of long sentences. By contrast, Neural Machine Translation “considers the entire input sentence as a unit for translation.” It’s proven that this reduces the number of awkwardly translated colloquialisms and complex sentences significantly.
Google’s latest contribution to this technology is to make it both faster and more accurate—in fact, approaching the speed of an average human translator.
Google concedes that GNMT may still struggle with proper nouns and rare vocabulary, but it’s enough of an improvement that your intro linguistics homework might have just gotten much easier.
A much more technical breakdown of the announcement can be read here.
