Learning non-monotonic causal theories from narratives of actions.
David Lorenzo · 2002
Non-monotonic formalisms for reasoning about actions and change have become a whole subfield of Artificial Intelligence. Current implementations allow to work with very expressive action theories involving ramifications, concurrent actions, complex qualifications and so on. A natural question that can be posed is whether this kind of declarative knowledge can be learned from observed time traces of property values from an existing dynamic system. For this task we consider a narrative-based logical theory of change in the form of Extended Logic Programs where logic-based learning methods can be applied effectively. The use of a narrative formalism provides more expressivity on the theories that are learned, for instance, to learn the effects of concurrent actions.