AI Watson Learns Finnish in Turku
For a decade, researchers at the University of Turku have been developing a technology that is able to interpret Finnish. Technology company IBM utilises the technology developed in Turku to teach Watson Finnish.
Filip Ginter and Tapio Salakoski find co-operation with IBM as a good example of the benefits of open science.
Around six years ago, IBM’s AI Watson defeated a human in the American television quiz show Jeopardy. At the moment, there are great expectations for Watson e.g. in the health care industry. Thanks to the technology developed by the researchers at the University of Turku, Watson now knows Finnish as well.
– In the spirit of open science, the technology we developed is available on the internet, and IBM happened to need exactly this kind of technology. They contacted us and we organised multiple seminars, in which we helped them utilise our language technology. As a result, IBM integrated the technology into Watson and also shared their updated version openly in the internet. This is a great example of the possibilities of open science that benefit everyone, say the Dean of the Faculty of Mathematics and Natural Sciences Tapio Salakoski and Assistant Professor of the Department of Future Technologies Filip Ginter.
– Especially in health care, there is a lot of valuable data, such as patient records, to analyse with the new technology. Until now, Finnish has prevented us from using the data to its full potential. The technology developed at the University of Turku has made the data better available, and now Finnish texts can also be mined. This opens new possibilities e.g. for combining patient care teams, observing trends, validating hypotheses, or enriching information pools, says Maarit Palo, IBM Executive of Staff Government Affairs IBM Finland, University Relations Nordic.
The researchers themselves have utilised the technology to process large amounts of text in health care. For example, language technology can help a doctor whose patient has very unusual symptoms. A computer can go through massive amounts of patient records and search for other patients with similar symptoms.
– The patient records in health care have great, unused potential. The law requires the collection of data, but the data is not yet utilised in the best way possible. Computers can read through such amounts of texts that humans do not have time for, says Salakoski.
– However, health care texts require understanding of hospital slang in addition to Finnish. Our multidisciplinary research group including researchers in language, AI and health care can help with this, says Professor of Clinical Nursing Science Sanna Salanterä from the University of Turku.
Language technology has been developed globally since the 1950s. However, the massive data that has only recently accumulated in the internet has enabled taking the technology a step further.
– Like a child, a machine is exposed to a great amount of examples, through which it will learn a language. In addition, an AI for learning the language needs to be coded. At this moment, there is so much data in the internet that this is possible, says Ginter.
Text and picture: Jenni Valta
Translation: Jenni Maja