Google today announced that it’s switching over to neural machine translations, resulting in more natural suggestions, for two more languages: Arabic and Hebrew. Google first started using this approach in September and has been gradually rolling it out to more languages, with a goal of using it for all 103 supported languages.
In addition to transitioning away from a more traditional phrase-based machine translation system, Google is also enhancing translations in Arabic and Hebrew thanks to contributions from the Google Translate Community, which lets people contribute translations, Lee Reshef of Google’s Israel consumer and brand team wrote in a blog post.
Neural machine translations are also live in Google Translate for Chinese, English, French, German, Hindi, Japanese, Korean, Portuguese, Russian, Spanish, Turkish, and Vietnamese.
Microsoft has also been gradually moving to a neural machine translation, as opposed to more traditional statistical machine translation — 11 languages had made the switch as of March 14.
Google also uses deep learning — an approach that involves training artificial neural networks on a large data set and then getting them to make inferences on new data — to improve Android, Gmail, Google Maps, Google Play, Google Search, and YouTube, among other things.