Discourse structures in medical reports - Watch out! The generation of referentially coherent and valid text knowledge bases in the MEDSYNDIKATE system

The automatic analysis of medical narratives currently suffers from neglect ing text structure phenomena such as referential relations between discours e units. This has unwarranted effects on the descriptional adequacy of medi cal knowledge bases automatically generated from texts. The resulting repre sentation bias can be characterized in terms of incomplete, artificially fr agmented and referentially invalid knowledge structures. We focus here on f our basic types of textual reference relations, viz. pronominal and nominal anaphora, textual ellipsis and metonymy and show how to deal with them in an adequate text parsing device. Since the types of reference relations we discuss show an increasing dependence on conceptual background knowledge. w e stress the need for formally grounded, expressive conceptual representati on systems for medical knowledge. Our suggestions are based on experience w ith MEDSYNDIKATE, a medical text knowledge acquisition system designed to p roperly deal with various sorts of discourse structure phenomena. (C) 1999 Elsevier Science Ireland Ltd. All rights reserved.