ABSTRACT
Valuable information that is useful or necessary to certain design activities is often available only in natural language text. To be useful, the information must be extracted manually and formal design models must be generated. These extracted models are . used for simulation studies, product evaluation and patent analysis. The approach considered here to automating model extraction from natural language documents includes parsing, semantic analysis, coreference detection and model generation. This paper focuses on the semantic analysis step and the semantic domain of digital computer systems. Conceptual graphs, a form of semantic networks, have been selected as the knowledge representation. With word meanings represented by conceptual graphs, semantic analysis proceeds by unification of these graphs, as guided by the parse trees developed in the parsing step.
