Google Translate. Part one – look how far it’s come.
With over 500 million users, Google Translate has a very important role to play in breaking down language barriers and helping us all to stay better connected – not to mention pioneering some truly astonishing tech along the way.
So, who better to turn to for facts, figures and advice about using Google Translate than a (surprisingly unbiased) bunch of translation professionals…? Join us over the next couple of weeks as we explore the technology, benefits and limitations of Google Translate.
How does it work?
The basic premise behind Google Translate is the same as most machine translation software: it has a database of translated documents in over 100 languages from organisations, books and websites that have already been translated by human translators. It breaks down these texts into small chunks (phrases), which it analyses and quickly recycles to make a tailor-made translation for the user. This is called phrase-based machine translation (PBMT).
It’s no longer just a web-based service, either. Google Translate allows you to scan words and signs via your phone’s camera and provides automatic translations, turns speech in one language directly into text in another and even translates handwriting. And for those of us who still use internet browsers, Google Chrome’s in-built function can translate web content almost instantly and is used to translate more than 150 million webpages every day!
In the past, this so-called PBMT method made for some pretty clunky, ‘building-block’ translation results, but last year, ten years after Google Translate was launched, Google announced an exciting new development. It now uses neural machine translation (Google’s Neural Machine Translation, or GNMT) technology for a growing number of language combinations. As the name implies, neural networks are modelled on a biological brain, meaning, among other things, that they can learn and get better as they go along.
Google has run with this self-teaching technology to create a translation service that analyses source sentences as a whole, not just in broken-down parts, just like human translators. The company claims that Google Translate is producing higher-quality, more accurate translations than before, with 55-85% fewer errors. Research has shown that GNMT is producing translations that are creeping up towards the kind of quality we would expect from a human translator.
So, have the Google wizards cracked machine translation?
Although it would seem the brilliant scientists and mathematicians over in California are biting at the heels of human translators, Google Translate is by no means a fail-safe, clunk-free translation solution yet. It still struggles with rare words and proper nouns and just can’t take context into account like a good old-fashioned person. In part two, we’ll be looking at how to get the most of this immensely useful tool and when to trust your translation needs to a more human-based service.