Natural Language Processing in Medicine. Design, Implementation and Evaluation of an Analyser for Dutch

This book describes a complex computer program (implemented in Prolog by BIM) that is, thanks to the application of Natural Language Processing and Artificial Intelligence techniques, able to analyse and process the content of a (limited) set of Dutch patient discharge summaries. Such a program achieves that not only the medium but also the message of medical documents can be treated computationally, which implies the possible automation of several medico-administrative tasks. An introduction to medical language processing as well as an overview of similar international projects is included. The bulk of the effort is dedicated to modules that are specific for the Dutch language. To this aim, existing resources (for Dutch) are re-used and integrated together with original developments. As a result, an electronic dictionary (containing some 8.000 lemmas) and a morphological recogniser (and a tagger/lemmatiser as a derived product) are realised. Another important achievement is a computer grammar (for sentence analysis) with a parser.
For other modules (sublanguage and discourse analysis), software that has been developed elsewhere (Linguistic String Project team of the New York University and the Service d'Informatique Medicale, Assistance Publique - Hopitaux de Paris resp.) is adapted and extended for use with Dutch. In this way, intelligent information processing becomes feasible. Two information extraction applications illustrate the potentialities of the program. The application domain is cardiology and cardiac surgery. All components of the system are situated in their theoretical framework, extensively discussed and illustrated by means of excerpts of Prolog code and figures. A through evaluation has been performed as well, and the results, if possible, are compared with other results of similar tests.