Nonmonotonic Reasoning in Ldl ++

Deductive database systems have made major advances on eecient support for nonmonotonic reasoning. A rst generation of deductive database systems supported the notion of stratiication for programs with negation and set aggregates. Stratiication is simple to understand and eecient to implement but it is too restrictive; therefore, a second generation of systems seeks eecient support for more powerful semantics based on notions such as well-founded models and stable models. In this respect, a particularly powerful set of constructs is provided by the recently enhanced LDL++ system that supports (i) monotonic user-deened aggregates, (ii) XY-stratiied programs, and (iii) the nondeterministic choice constructs under stable model semantics. This integrated set of primitives supports a terse formulation and eecient implementation for complex computations, such as greedy algorithms and data mining functions, yielding levels of expressive power unmatched by other deductive database systems. 1 INTRODUCTION The LDL system (Chimenti et al., 1990) developed at MCC in the 80's is representative of a generation of Deductive Database Systems (Minker 1996) that sought to integrate logic-based rule languages into relational databases, in order to provide eecient and scalable support for sophisticated queries on large data sets and knowledge bases. Therefore, most of these systems 525