Automated analysis of natural language properties for UML models

It is well known that errors introduced early in the development process are commonly the most expensive to correct. The increasingly popular model-driven architecture (MDA) exacerbates this problem by propagating these errors automatically to design and code. This paper describes a round trip engineering process that supports the specification of a UML model using CASE tools, the analysis of specified natural language properties, and the subsequent model refinement to eliminate errors uncovered during the analysis. This process has been implemented in SPIDER, a tool suite that enables developers to specify and analyze a UML model with respect to behavioral properties specified in terms of natural language.