The Construction of Explanations

Abstract

What sorts of processes are needed to construct explanations? My aim is to describe a computational model of explanatory reasoning. It implements a theory of the representations and processes that people rely on to construct explanations. I will report two experiments on the explanations people produce for unusual events. The first experiment shows that they explain a single aspect of an event even when multiple aspects are unusual, and they explain the actions carried out by an actor rather than the objects possessed by an actor, even when both are unusual. The second experiment shows that their communication goals determine what aspect of an event they explain. The results have implications for theories of explanation-based learning, and for intelligent tutoring systems.

I am grateful to Mark Keane for helpful comments on an earlier version of this article.