TBSM : a task-based specification methodology for expert systems

Vita.There has been an increasing interest in improving the quality and reliability of artificial intelligence systems. As a result, several methodologies for specifying expert systems have been proposed. However, these approaches are limited in formally verifying and validating the intended functionality and behavior of an expert system. In this dissertation, a novel methodology is proposed, the task-based specification methodology, for specifying the model and the process knowledge of an expert system at different abstraction levels. Specifications are acquired and organized around the system's functional units called tasks. To capture a specification at its appropriate abstraction levels, we use the task structure to achieve a functional decomposition that supports polymorphism. A formal foundation for the proposed methodology is presented to formulate two major components: functional and behavioral specification. The notion of state transition model is adopted for formulating the functional specification. The notion of task structure is formalized for the analysis of task process refinement. Task state expressions are used to describe expected behavior specifications. Progression operators and frame axioms are adopted for constructing the state description. The formalism provided by the framework serves as a basis for the verification of specifications. The proposed methodology and its benefits are demonstrated using a specification of R1/SOAR constructed in a reverse engineering fashion. Based on the proposed methodology, a knowledge engineering tool is developed to facilitate acquiring and organizing the specification and the prototype. In the proposed framework, the prototype complements the specification throughout the expert systems life cycle. The traceability between them is facilitated by organizing both types of artifacts using a common functional decomposition structure. In a preliminary study of the role of fuzzy logic in specifying imprecise requirements, the proposed methodology is further extended by formulating soft functional requirements using test-score semantics, and by using the criticality qualifier to represent the degree of importance of soft requirements. The proposed specification methodology not only enables the verification of an expert system's knowledge in an earlier phase of the software life cycle, but can be used to facilitate the validation and the maintenance of large expert systems