Compiling Prioritized Default Rules into Ordinary Logic Programs Limited Distribution Notice

Prioritized default rules o er a conveniently higher level of speci cation that facilitates revision and modularity. They handle con icts, including arising during updating of rule sets, using partially-ordered prioritization information that is naturally available based on relative speci city, recency, and authority. Despite having received much study, however, they have had as yet little impact on on practical rule-based systems and software engineering generally, and had very few deployed serious applications. We give a new approach to the semantics and implementation of prioritized default rules: to compile them into ordinary logic programs (LP's). (By \logic program" and \prioritized default rules", we mean in the sense of declarative knowledge representation (KR), includingmodel-theoretic entailment and forward as well as backward inferencing. In particular, by \logic program", we do not simply mean Prolog.) We use the compilation approach both to expressively generalize and to implement courteous LP's, a previous formalism featuring classical negation and prioritized con ict handling. We show that we preserve courteous LP's' attractive reasoning behaviors and polynomial-time computational cost. Our expressive generalization enables recursion, and also reasoning about the prioritization. Our implementation enables courteous LP's functionality to be added modularly to ordinary LP rule engines, via a pre-processor, with moderate, tractable computational overhead. More generally, we show that the compilation approach is applicable to implementing, and to de ning, numerous variants of prioritized default rule KR's, beyond the particular courteous LP variant given here. This takes a long step towards actual deployment of prioritized default rules in commercially elded technology and applications. Copyright and Publication Information: The Limited Distribution Notice on the front page notwithstanding (it is standard boilerplate for all IBM Research Reports), copyright to this Research Report is not owned by anyone other than the author(s) and IBM. Related Papers and Material: can be found via the author's Web address and as IBM Research Reports at http://www.research.ibm.com .