| Abstract
|
Knowledge-based systems, like other softwa … Knowledge-based systems, like other software systems, need to be debugged while being developed. In addition, systems providing "expert advice" need to be able to justify their conclusions. Traditionally, developers have been supported during debugging by tools which offer a trace of the operations performed by the system (e.g., a sequence of rule firings in a rule-based expert system) or, more generally by an explanation facility for the reasoner. Description Logics, formal systems developed to reason with taxonomies or classification hierarchies, form the basis of several recent knowledge-based systems but do not currently offer such facilities. In this thesis, we explore four major issues in explaining the conclusions of procedurally implemented deductive systems, concentrating on a specific solution for a class of description logics. First, we consider how to explain a highly optimized procedural implementation in a declarative manner. We begin with a formal proof-theoretic foundation for explanation and we illustrate our approach using examples from our implementation in the CLASSIC knowledge representation system. Next, we consider the issue of handling long, complicated deduction chains. We introduce methods designed to break up description logic queries and answers into small, manageable pieces, and we show how these are used in our approach and how they support automatically generated explanations of followup questions. Next, we consider the problem of explaining negative deductions. We provide a constructive method for explanation based on generating counter-examples. Finally, we address the issue of limiting both object presentation and explanation. We offer a meta-language for describing interesting aspects of complicated objects and use this language to limit the amount of information that should be presented or explained. The work in this thesis has been motivated by design and application work on a description logic-based system and a significant portion of our work has been implemented for CLASSIC and is in use. een implemented for CLASSIC and is in use.
|
