TOWARDS PROBABILISTIC KNOWLEDGE BASES

We define a new fixpoint semantics for rule-based reasoning in the presence of vague or imprecise information. Such a semantics should fulfill two requirements. First, it must coincide with the intuitive understanding of the given information, and second, the semantics must be computationally tractable. We show that our semantics fulfills the first requirement and we formally verify that our fixpoint semantics.reduces to the usual fixpoint semantics of Datalog programs if all the given information is certain or non-vague. Moreover, we rigorously prove that this new semantics also satisfies the second requirement. At the end of this study we emphasize the strong similarity between our semantics and basic probability theory.