16 August 2018

Kees van Deemter appointed Professor of Natural Language Generation

Kees van Deemter

In February 2018, Kees van Deemter was appointed Professor of Natural Language Generation at the Department of Computer Science. “My research deals with the question: how can you convert data into text? The data can be anything: formulas, sensor readings, weather forecasts... In fact, anything that’s not text yet.” Does Van Deemter want to develop a digital meteorologist? “Not for the eight o’clock news. You would miss the personal aspect of a human meteorologist. But a system like that is very valuable for an offshore oil rig, which needs a very specific weather forecast.”

Van Deemter aims to achieve much more than just grammatically correct sentences. “We know pretty much how that works now. What’s more interesting to me, is how you can select information that is relevant to report from a huge quantity of data. And how do you categorise that information?” The more data, the more important that question becomes. “In medical applications, for example, you might deal with measurements of a patient in intensive care, who needs to be monitored using a lot of sensors. So when is something worth reporting, and how should it be reported? Is it better for the system to say: ‘the patient has a fever’, or ‘the body temperature is so many degrees’?”

Human language

Eventually, Van Deemter aims to understand human language behaviour: for example, how do doctors choose between using words like ‘fever’ and stating the actual number of degrees? Van Deemter therefore works closely together with linguists and experimental psychologists. His role as a computer science researcher is to express human thoughts in terms of algorithms, in order to understand how human thought works. “In order to capture human thought, we develop computer programs that behave in a human manner.”

Chinese weather forecast

One of Van Deemter’s latest projects focuses on the Mandarin language. “We suspect that there are some major differences between cultures. When we have weather data from China, along with the weather report based on those data, we can find out whether a Chinese weather forecast is structured the same way as a Dutch forecast. Once we understand that, we can learn something about the differences between our cultures. Plus, we can use the insights gained in order to develop practical applications using the Mandarin language.”