Czech scientists teach computers to "speak" Czech


Would you like your computer to understand you when you shout at it or when you are begging it to unfreeze? Well, that might soon be possible thanks to the work of Czech scientists from the Centre for Computational Linguistics at the Faculty of Mathematics in Prague. Among the projects they are currently working on is for example machine translation. But what is much more exciting is the one in which they teach computers to understand spoken language. In the case of Czech, it is especially difficult, as the spoken and written forms of the language differ significantly. I spoke to the head of the centre, Professor Eva Hajicova, and asked her first whether that was a major challenge.

"That's exactly one of the problems which are very specific about Czech because the written form is very different from how people speak. So this is one of the problems we have to deal with."

Is it possible to explain to the listeners how this works, for example what the machines look like?

"Well, the machines are normal computers and what is specific about them is that they must have a speech recogniser, which means that the machine should understand spoken language, or in the case of written language, there must be rules in the machine - just like at school. If people just think of their school years when they had lessons in Czech grammar, they were supposed to analyse the Czech sentence and say what the subject is, what the object is and to understand the sentence through some rules like that. So that's exactly what we had to teach the machines."

What are the potential uses for these machines that can recognise and transcribe speech?

"I can imagine many applications. Let's say there might be a travel agency which can answer the questions of people for train connections or plane connections. People could just phone the travel agent but the travel agent will not be sitting there the machine will do the work which a machine can do. I mean, look at the timetable and tell you when your train leaves and what you want and what kind of reservations you want to make. So that's one of the evident possibilities but there are many other things like that where the spoken signal should be understood. Maybe situations which require immediate voice signal from a human to do something, to stop something to rearrange something. So I think this is a very important application area."