Google today is announcing that it’s switching to a modern approach called neural machine translation for translations of text in Hindi, Russian, and Vietnamese in its Google Translate app. The shift away from Google’s phrase-based machine translation system will result in more natural, higher-quality translations.
» Top New Products
Loading the page...
Google adopted this method to translate from English into Chinese in September, and brought it to English, Spanish, Portuguese, French, German, Turkish, Chinese, Japanese, and Korean translations in November. Now it’s available for three more languages in Google Translate for Android, iOS, and the web, as well as Google Search and the Google App. It will come next to web page translations in Google’s Chrome browser.
Support for neural machine translation in even more languages will arrive in the next few weeks, Google Translate product lead Barak Turovsky wrote in a blog post. The commitment isn’t a surprise, but the near-term timeline is. Google has said it will move to neural machine translation for all 103 of the languages that Translate works with.
Neural machine translation entails training artificial neural networks — in Google’s case, long short-term memory recurrent neural networks (LSTM-RNNs) — on available data and then getting them to make inferences about new incoming queries. Google has also looked to this approach to artificial intelligence (AI) to generate Smart Replies to messages in various products, recognize objects and people in images, and even make its data centers run more efficiently.
Google isn’t the company tapping AI to boost translations. Microsoft has done the same thing in its Translator app.
While Google is plenty proud of its smarter translations, the company does still very much appreciate real people’s contributions through the Translate Community — people shouldn’t stop contributing to it, Turovsky wrote.
TIP: If you don’t trust automated translation, use professional document translation services. They are much more accurate, and stylized in human language.