For bilingual and multilingual people trying to communicate with other people around the world, Google Translate has been a pretty great tool. However, admit it, you often laugh at some of the weird translations that they’ve given us. This happens because as much as they use machine learning, the phrase or sentence is translated piece by piece and not in a broader context. That’s why it comes out weird sometimes. But now Google has announced they are adapting the Neural Machine Translation technology to make translation better.
With this new system, whole sentences will be translated instead of the previous one where it is word by word. It then rearranges and adjusts the translation to become more like a person who’s actually speaking, with proper grammar and all. This way, the translation becomes more relevant instead of literal. The Neural Machine Translation system will make translations more natural-sounding over time, based on the end-to-end learning system that it employs.
With this initial rollout, eight languages will be part of this including to and from English and French, German, Spanish, Portuguese, Chinese, Japanese, Korean, and Turkish. They may just be a few languages but they already represent one-third of the world’s population. They already cover more than 35% of the queries being submitted to Google Translate.
Of course this is just the first phase as their goal is to eventually roll out the Neural Machine Translation to all of the 103 languages and and surfaces that use Google Translate. They are also making the system available for businesses that may be able to make use of it through the Google Cloud Transition API.