Trend towards machine translation

Computers can make translation easier, but they can't do everything (yet). An Interview with Philipp Köhn, pioneer of machine translation.

Alexander Heimann/Frankfurter Buchmesse

Philipp Köhn has revolutionized machine translation with his research. Born in Bavaria, he is Professor of Computer Science at the University of Edinburgh, Scotland, and at Johns Hopkins University in Baltimore, Maryland. An interview on progress and obstacles in machine translation.

Professor Köhn, most people who have a text automatically translated on the internet these days are using one of your inventions. What exactly lies behind your innovation?

We have fundamentally changed the model of machine translation. In the past, people tried to teach the computer single words and grammatical rules. That proved to be extremely difficult and was not really very successful. Our translation method, by contrast, is based on statistics and phrases.

What does that mean?

The computer doesn't translate individual words, but parts of sentences. To do this it accesses huge amounts of text that are already available in different languages and calculates probabilities for the meanings of words and phrases. In principle, we have also taught the computer to learn autonomously in this way. Our model is used by the most important companies in the field of internet translation, including Google Translator. By the way, one of the authors with whom I described the method in detail in 2003 is now in charge of the Machine Translation Group at Google.

Is machine translation also an option for literary texts?

Not at the present time. This is because there is an additional aspect in literary texts which is not an issue in an operator's manual: for example, linguistic aesthetics and hidden meanings or allusions between the lines. It's not our ambition to replace human translators, we just want to put a tool in their hands. This applies not only to translators, by the way, but also to everyone who would like to understand the gist of a text – whether professionally or privately. We can meet this ever-expanding need for information.

Your mother tongue is German. Today you live in Scotland and the USA. How do you rate the quality of machine translations into German?

Unfortunately, German is one of the few languages which still pose a challenge for machine translation. This had to do with its complex syntax, which differs fundamentally from languages like English or French. But of course we're always trying to get better – and are already working on a new translation model.

Frankfurt Book Fair from 8 to 12 October 2014 in Frankfurt-am-Main