Meta Levels in Conditional Logic

Wolfgang Nejdl and Markus Banagl

Abstract

First, we present a formal account of splitting a knowledge base into two parts, expanding our previous results motivated by avoiding the Gärdenfors triviality results for revision semantics for conditionals. Second, we propose a specific revision function and show that this function is optimal in the sense, that after revision as few conditional consequences as possible are changed. Third, we provide an algorithm, which for certain propositional sentences $\phi$ and $\psi$, computes a belief revision ordering, which has the property that after revision by $\phi$ all conditional consequences of $\psi$ remain unchanged. A complexity result (upper bound for execution time) is presented as well. Last, we compare our results to those of recent proposals by Morreau and Geffner/Pearl and present some ideas for extending our ideas for default reasoning.

Keywords: Belief revision, iterated revisions

The full paper is available as a postscript file .