Abstract

In this dissertation, we investigate the role of causal knowledge in commonsense reasoning about action and change. We define a language in which a relatively simple form of causal knowledge is expressed. Using this language, we describe a novel approach to formalizing action domains as "causal theories"--including domains that involve concurrency, nondeterminism, and things that change by themselves. We show that a subclass of causal theories can be translated into propositional logic by a generalization of Clark's completion procedure for logic programs. Finally, we describe an implemented approach to automated query answering and "satisfiability planning" which is based on this translation