A high-level categorization of explanations a case study with a tutoring system
Helmut Horacek · 2008
Giving explanations is an important functionality of most systems that perform some sort of problem-solving, including intelligent tutoring systems. In order to make explanation techniques transparent and better reusable across kinds of applications and domains, various categorizations have been proposed, but they appear to be on a comparably low level of abstraction, oriented on the surface form of explanatory requests. Inspired by our work on the tutoring system Dialog that teaches students how to prove mathematical theorems, we distinguish categories of explanations according to the depth of understanding that these explanations aim at, ranging from expositions of reasoning chains to ingredients of problem solving techniques. The modular design of our system does not only allow the identification of methods closely related to these categories, but it also enables us to clearly distinguish methods for problem solving from methods for proper explanation. We are convinced that the field as a whole can profit from factoring out explanation methods per se, which have much to do with presentation and tailoring issues, and methods for problem-solving that provide enhanced evidence about the results and support inquiries on top of them – the proper explanation-aware problem-solving.