Factors in Causal Explanation

The construction of explanations is a key process in human understanding, having many applications for internal tasks such as learning through to external tasks such as diagnosis and image interpretation. In this paper we introduce ve orthogonal dimensions along which any explanation task can be characterised and show how their identiication gives rise to a new model of explanation construction and evaluation. The identiication of these dimensions is important because any system that attempts to model the explanation capability of humans must incorporate some notion of each dimension. In addition the dimensions help to identify the type of knowledge human reasoners and hence any com-monsense reasoning system must possess. We argue that the exibility of the explanation process introduced above means that backward chaining algorithms from the event to be explained should still be an important topic of research in AI.