Google began their foray into the AI and machine intelligence sectors thanks to the computing power they’re able to leverage in the company’s data centers. This worked out great as it allowed people all around the world to see how beneficial it was without needing to invest in new hardware. This was when they started to develop their own hardware to handle these computational cycles, and that saved the company a ton of money as this was better designed to handle the tasks required for quick and efficient machine learning algorithms. The latest extension of their progress has come in the form of on-device machine learning hardware. Today, the company showed how Google Translate benefits from using on-device machine learning technology.
We’ve seen a number of silicon vendors begin adding special processors to their SoCs that allow mobile apps and services to benefit from these new neural network algorithms without needing to be connected to the internet. These special CPUs aren’t needed in order to handle the machine learning tasks that applications implement, but if they have them then the device will save a lot of time and battery life. Google even went as far as to add their own custom silicon in the Pixel 2 and Pixel 2 XL to handle their photography processing algorithms.
Neural machine translation (NMT) was introduced to Google Translate close to two years ago and since then people have seen significantly improved accuracy in their translations online. The application utilized this feature if it was online, but when there wasn’t a connection it used on-device phrase-based machine translation (PBMT) instead. So while you could usually get a general idea across, it wasn’t as accurate as it could be. Today’s announcement is the next step up from that with on-device NMT. The image above shows how it is better than on-device PBMT, but still not as accurate as online NMT.
Offline translation can be very useful when you’re traveling, so getting things as accurate as possible can really improve the experience. To use this newly announced on-device NMT in Google Translate, you will need to go into the application and download the language pack you’re interested in (which can be between 35MB to 45MB in size). This new feature is being rolled out to 59 total languages over the next few days.
