This research is on automating the extraction of behavioral models from natural language descriptions of digital systems such as specifications, proposals and U.S. patents. The behavioral models are necessary for simulation in new system design and test generation on legacy system replacement. Often, only the natural language descriptions are available. An information extraction approach is being taken to develop an interactive system to help the modeler. The system will scan source documents, perform syntactic and semantic analysis to identify behavioral elements (actions, states, events) and their relationships, and instantiate conceptual templates from which augmented control dataflow graph models may be generated. A central research issue is the correct selection and filling of behavioral templates. Performance of the system is being evaluated on how much of the behavioral information is discovered automatically and how accurately the information is inserted into the templates.
