Pushing Goal Derivation in DLP Computations

dlv is a knowledge representation system, based on disjunctive logic programming, which offers front-ends to several advanced KR formalisms. This paper describes new techniques for the computation of answer sets of disjunctive logic programs, that have been developed and implemented in the dlv system. These techniques try to "push" the query goals in the process of model generation (query goals are often present either explicitly, like in planning and diagnosis, or implicitly in the form of integrity constraints). This way, a lot of useless models are discarded "a priori" and the computation converges rapidly toward the generation of the "right" answer set. A few preliminary benchmarks show dramatic efficiency gains due to the new techniques.