Rational closure for all description logics

Author:  

Publication:  Artificial Intelligence

CITE.CC academic search helps you expand the influence of your papers.

Tags:     AI

Abstract

Rational closure is one of the most extensively studied nonmonotonic extensions of description logics. Nonetheless, so far it has been investigated only for description logics that satisfy the disjoint model union property, or limited fragments that support nominals. In this paper we show that for sufficiently expressive description logics, the traditional correspondence between rational closure and ranked interpretations does not hold. Therefore, in order to extend rational closure to a wider class of description logics it is necessary to change the definition of rational closure, or alternatively abandon its standard semantics. Here we pursue the former approach, and introduce stable rational closure, based on stable rankings. The resulting nonmonotonic logic is a natural extension of the standard rational closure: First, its refined exceptionality criterion yields a closure that satisfies the KLM postulates. Second, when a knowledge base enjoys the disjoint model union property, then stable rational closure equals the old notion. In the other cases, stable rankings may raise the exceptionality level of some concepts. Stable rational closure has a model-theoretic semantics based on upward-closed models, that relax the canonical models adopted in the past, in order to deal with logics that do not satisfy the disjoint union model property. Unfortunately, stable rankings do not always exist, and are not necessarily unique. However, they can be effectively enumerated for all defeasible knowledge bases in SROIQ, using any algorithm for reasoning with ranked models.

Cite this article

https://doi.org/10.1016/j.artint.2019.04.001

View full text

>> Full Text:   Rational closure for all description logics

Artificial immune systems can find arbitrarily good approximations for the NP-hard number partitioni

The 2016 and 2017 QBF solvers evaluations (QBFEVAL'16 and QBFEVAL'17)