Google talks up its ‘neural machine’ translator

SAN FRANCISCO — Google is promising that its widely used translation service is now even more fluent, thanks to an advance that is enabling its computers to interpret complete sentences.

That may sound simple, but it took years of engineering to pull it off.

Until now, Google’s technology analysed phrases in pieces and then cobbled together a sometimes stilted translation.

Now that Google’s machines can interpret entire sentences, the translations of extended passages of text should read and sound much more like a native speaker of the language.

Google described its “neural machine” tool as the biggest leap for its translation service in a decade.

The company expects to eventually expand it to all 103 Google Translate supports.

“In 10 years, Google Translate has gone from supporting just a few languages to 103, connecting strangers, reaching across language barriers and even helping people find love. At the start, we pioneered large-scale statistical machine translation, which uses statistical models to translate text. Today, we’re introducing the next step in making Google Translate even better: Neural Machine Translation,” Google said in a blog post.

Starting on Tuesday, the technology will be used to translate phrases to and from English and eight other languages — French, German, Spanish, Portuguese, Chinese, Japanese, Korean and Turkish.

These represent the native languages of about one-third of the world’s population, covering more than 35 per cent of all Google Translate queries, the company said.

Mr Barak Turovsky, the product lead for Google Translate, was cited on CNET as saying that the new method cuts down on errors by 55 to 85 per cent.

Google Translate has more than 500 million users, and translates more than 100 billion words a day. Agencies