Abnormality Counts!
Aleks Knoks

Abstract:
Defeasible inheritance networks provide a fruitful environment for
modeling default reasoning. In this paper we aim to enhance their
expressive power with the idea of abnormality minimisation, which
plays a crucial role in circumscription. We implement it in two
alternative network-based frameworks. The first is in line with the
received way of modeling default reasoning by means of inheritance
nets, that is to say, it is path-based or direct. The second relies on
a set of non-monotonic inference rules. In both we are able to say
that an object is abnormal with respect to a certain default
statement, and, consequently, single out the conclusion sets that
imply the least number of abnormalities. We consider some cases
indicating that exactly these sets are the intuitively correct ones.