Google has switched its machine translation service Google Translate to a new system called Google Neural Machine Translation (GNMT), which is an end-to-end learning framework that learns from millions of examples, providing a better translation quality (did you notice that?), the company has announced on its blog.
Besides the challenges of supporting all 103 languages Google Translate has previously supported, the switch also allowed Google to enable what they call “Zero-Shot Translation” translation between language pairs never seen by the system.
In a blog post shared by Google Research last week, Mike Schuster (Google Brain Team), Melvin Johnson (Google Translate) and Nikhil Thorat (Google Brain Team) explain that the step was triggered by their aim to answer a simple question: “Can we translate between a language pair which the system has never seen before?”
The answer was yes. An example language pair used in the blog post is Korean and Japanese. The improved system generates reasonable translations between Korean and Japanese, even though it has never been taught to do so. Google claims this is the first time this type of transfer learning has worked in Machine Translation.
During the past 10 years, Google Translate has grown to support 103 languages, and the service translates more than 140 billion words per day. If you are interested in more information about Google’s linguistic experiment, head over to Google’s dedicated research blog.