A semantics for open normal defaults via a modified preferential approach

Abstract

We present a new approach for handling open normal defaults that makes it possible 1. to derive existentially quantified formulae from other existentially quantified formulae by default, 2. to derive universally quantified formulae by default, and 3. to treat cardinality formulae analogously to other formulae. This was not the case for previous approaches. Reiter uses Skolemization in his treatment of open defaults to achieve the first goal, but this has the unpleasant side-effect that logically equivalent facts may lead to different default consequences. In addition, Reiter's approach does not comply with our second requirement. Lifschitz's main motivation for his approach was to satisfy this second demand. However, to achieve this goal he has to violate the third requirement, and the first condition is also not observed. Differing from these two previous approaches, we will not view open defaults as schemata for certain instantiated defaults. Instead they will be used to define a preference relation on models. But unlike the usual approaches to preferential semantics we shall not always take the minimal models to construct our semantics. Due to this new treatment of preference relations the resulting nonmonotonic consequence operator has nice proof-theoretic properties such as cumulativity.