Critiquing software specifications

The authors describe how the knowledge of expert-systems analysts is being input to use in an automated critic for specification debugging. They are working on a system, named Kate, that tries to acquire a problem specification that includes a description of the objects, transitions (actions), and constraints of the intended system. They have built a computer-based critic as part of the Kate environment that has its roots in a set of protocols they collected over several analysis sessions. The critic has three major components: a model of the domain, a matcher to connect parts of the model to the specification being analyzed, and a critique to supply the analysis. The model represents a set of policy issues for building systems in a particular domain and a set of relevant problem specification cases to consider for systems in the domain. The critic has been run on a library database problem. The results are presented and compared with a critique by an experienced library analyst.<>