Enhancing Knowledge Representations by Ontological Relations

Several medical natural language processing (NLP) systems currently base on ontologies that provide the domain knowledge. But, relationships between concepts defined in ontologies as well as relations predefined in a semantic network are widely unused in this context. The objective of this paper is to analyse potentials of using ontological relations to produce correct semantic structures for a medical document automatically and to ameliorate and enrich these structures. Knowledge representations to unstructured medical narratives are generated by means of the method SeReMeD. This approach is based on semantic transformation rules for mapping syntactic information to semantic roles. Contextual relations expressed in natural language are automatically identified and represented in the generated structures. To achieve additional semantic relationships between concepts, the UMLS Medical Semantic Network and relationships between concepts predefined in the UMLS Metathesaurus are used to support the structuring process of SeReMeD. First results show that these relations can enhance and ameliorate the automatically generated semantic structures.