Google Translate aims for perfection
by FFE EU News staff
Automated translation has been on the agenda of computer companies since IBM started the dream decades ago. Today, substantial progress has been achieved by Google’s Translate service which has 71 translatable languages in its system.
What else is in store for Google Translate? Franz Josef Och, project head of Google Translate, said they aim to perfect the service.
As a teenager, Och dreamt of intelligent machines that could simulate human attributes. Now at 41, he is focusing on a single goal: to build the perfect translation computer, so fast and subtle that “you hardly notice it all, except as a whisper in your ear.”
With the programme, texts or websites in one language can be translated into another in seconds with the touch of a button. An app version can also recognize and translate texts in an image. The latest in Translate’s list of services is an app that transforms smartphones into talking translation machines. So far, the app works well as long as the sentences are kept simple.
The fact that Google Translate has been accessed 200 million times last year shows just how valuable the service is today. Multinational companies who are interested in an automated translation service may be willing to add a price tag to get a similar tool in the future.
The Translate team in Google’s Silicon Valley headquarters has grown considerably, but Och reveals that not a single one of them is a linguist. He emphasises that this is the beauty of machine translation: that anyone can access it.
Och confesses that the team has encountered instances where the computer ends up “lost in translation.” To prevent this, they have bypassed rigid grammatical structures and instead fed the programme with a large set of linguistic data accessible through the internet so that the team merely “computes probabilities of translation.”
This way, the algorithms that govern the programme search for existing matching phrases to best translate a sentence.
While this strategy works well, Och clarifies that it is still far from perfect. Certain ambiguities still exist and the machine has no sense of creativity. Language pairs in the data so far have resulted in good combinations between English and Spanish. But other pairs like English and Japanese and English and German are still difficult to translate.
The Translate team’s quest for perfection has been motivated by the goal of making information accessible to many people and to keep up with competition. A few weeks ago, Facebook acquired a small company that dealt with automated translation. The European Union has its own translation programme designed to tackle bureaucratic language. Microsoft has also launched a translation project that can translate words using the speaker’s own voice.