Google's neural translation cut errors by up to 85 percent

When Google replaced the statistical engine behind Google Translate with the deep-learning Google Neural Machine Translation system in September 2016, it reported a large jump in quality. By human side-by-side evaluation, GNMT “reduces translation errors by more than 55%-85% on several major language pairs.” The system went live first for Chinese-to-English, which Google said handled about 18 million translations a day. It was one of the most visible examples of a deep-learning method making a mass-market product abruptly better.