Explanation requires a dialogue. Users must be allowed to ask questions about previously given explanations. However, building an interface that allows users to ask follow-up questions poses a difficult challenge for natural language understanding because such questions often intermix meta-level references to the discourse with object-level references to the domain. We propose a hypertext-like interface that allows users to point to the portion of the system's explanation they would like clarified. By allowing users to point, many of the difficult referential problems in natural language analysis can be avoided. However, the feasibility of such an interface rests on the system's ability to understand what the user is pointing at; i.e., the system must understand its own explanations. To solve this problem, we employ a planning approach to explanation generation which records the design process that produced an explanation so that it can be used in later reasoning. In this pa...
Johanna D. Moore, William R. Swartout