Google says it is has vastly improved the accuracy of Google Translate through its new Neural Machine Translation (NMT) system. NMT utilizes neural networks to train machines how to produce more natural, grammatically correct translations. The new system improves Google Translate's capacity for contextual translation by processing whole sentences or paragraphs at a time, rather than analyzing individual words. Translation errors have been cut by 55 percent to 85 percent in several languages, but the system still makes mistakes, such as dropping words or misinterpreting a person's name.
Google announced its attempts to replicate human translations using neural networks in September, at which time NMT only supported translations between English and Chinese. NMT has since been rolled out for eight language pairs on the Google Translate website, mobile application, and Google Search. Translations to and from English and French, German, Spanish, Portuguese, Chinese, Japanese, Korean, and Turkish are enabled, covering 35 percent of Google Translate queries. Eventually, NMT will support translations for 103 languages.
"With this update, Google Translate is improving more in a single leap than we've seen in the past 10 years combined," says Google Translate's Barak Turovsky.
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA