We address the problem of representing common sense knowledge about action domains in the formalisms of logic programming and default logic. We employ a methodology proposed by Gelfond and Lifschitz which involves first defining a high-level language for representing knowledge about actions, and then specifying a translation from the high-level action language into a general-purpose formalism, such as logic programming. Accordingly, we define a high-level action language AC, and specify sound and complete translations of portions of AC into logic programming and default logic. The language AC includes propositions that represent 'static causal laws' of the following kind: a fluent formula ψ can be made true by making a fluent formula φ true (or, more precisely, ψ is caused whenever φ is caused). Such propositions are more expressive than the state constraints traditionally used to represent background knowledge. Our translations of AC domain descriptions into logic programming and default logic are simple, in part because the noncontrapositive nature of causal laws is easily reflected in such rule-based formalisms.
Bibliographical noteFunding Information:
Proof in the other direction is similar. Thus we have shown that the two domains agree on Res, which is sufficient to establish the fact that they have the same models, given the earlier observation that they have equivalent sets of value propositions (Section 6). Moreover, since they also agree, for each action A and state S, on the question of whether or not A is prohibited in S, we can conclude that either both domain descriptions are qualification-free or neither is. \[\] Many thanks to Vladimir Lifschitz and Norman McCain. Thanks also to Chitta Baral, Michael Gelfond, Enrico Giunchiglia, G. Neelakantan Kartha, Teodor Przymusinski and Michael Thielscher for helpful discussions. This work is partially supported by National Science Foundation grant #IRI-9306751.