Extracting specific medical data using semantic structures

In this paper, we discuss the architecture, functionality and performance of a medical information extraction system. The system is based on an approach to automatic generation of semantic structures for free-text. Using a multiaxial nomenclature (Wingert Nomenclature) and existing language-engineering technologies, a conceptual graph-like representation is produced for each sentence of a text. These semantic structures are then exploited to extract information. The components that might be adopted for processing texts in another language than German are identified. Results of first evaluations of the system's performance in an information extraction (IE) subtask in the medical domain are presented: The filling of selected template slots obtained values of 81-95% precision and 83-97% recall.