An architecture and a knowledge representation model for expert critiquing systems

Abstract: The aim of an expert critiquing system (ECS) is to act as an assistant in a problem-solving situation. Once a user has presented a tentative solution to a given problem, the ECS reviews the proposed solution and then provides a critique — an evaluation of the solution. In this thesis we provide an architecture for an expert critiquing shell and present an implementation of such a shell, the AREST system. Experience from using this shell in two experimental implementations is reported. Further we show how the critique generation process can be supported by extending standard knowledge representation structures by explicitly providing conceptual relationships between items in the knowledge base in the form of a simple extension of a rule-based schema. Current results in text generation have opened the way for dynamically producing multi-paragraph text. Our work is based on a theory for text organization, Rhetorical Structure Theory (RST). To remedy certain shortcomings in RST we proposed improvements, e.g. rhetorical aggregates for guiding content selection and organization. In this thesis we discuss how the AREST system has been used to construct two ECSs with text generation capabilities.