Tools for Rapid Automated Development of Expert Models (TRADEM)

We will investigate, test and evaluate a new method for machine generation of domain and expert models for intelligent tutoring systems. This method uses a corpus of instructional, training, and procedural content that is semantically related to a domain of instruction. Semantic analysis, text analysis, and machine learning are applied to the domain corpus to identify competencies, map relationships and extract constraint statements that represent expert actions and behaviors. This method results in data that can be used by model-tracing, example-tracing, and constraint-based intelligent tutoring systems. It can incorporate information generated in a Virtual World or game-based training environment and, as an automated process, can continually update its underlying models. We will test the proposed method using concept maps and content sources for Elementary Mathematics and Cognitive Psychology and with competencies and data used to develop simulation-based training for combat medics. We will evaluate technical feasibility by using our method to program variants of AutoTutor. Finally, we will investigate compatibility with the event data model architecture defined in the Simulations for Integrated Learning Environments (SIMILE) project and will provide a high level system design for a prototype implementation compatible with the Generalized Intelligent Framework for Tutoring (GIFT).