A simple declarative language for describing narratives with actions

Abstract

We describe a simple declarative language ℰ for describing the effects of a series of action occurrences within a narrative. ℰ is analogous to Gelfond and Lifschitz's Language script A sign and its extensions, but is based on a different ontology. The semantics of ℰ is based on a simple characterisation of persistence which facilitates a modular approach to extending the expressivity of the language. Domain descriptions in script A sign can be translated to equivalent theories in ℰ. We show how, in the context of reasoning about actions, ℰ's narrative-based ontology may be exploited in order to characterise and synthesise two complementary notions of explanation. According to the first notion, explanation may be partly modelled as the process of suitably extending an apparently inconsistent theory written in ℰ so as to establish consistency, thus providing a natural method, in many cases, to account for conflicting sets of information about the domain. According to the second notion, observations made at later times can sometimes be explained in terms of what is true at earlier times. This enables domains to be given an alternative characterisation in which knowledge arising from observations is appropriately separated from other aspects of the domain. We also describe how ℰ domains may be implemented as Event Calculus style logic programs, which facilitate automated reasoning both backwards and forwards in time, and which behave correctly even when the knowledge entailed by the domain description is incomplete. © Elsevier Science Inc., 1997.