Graded inheritance nets for knowledge representation

Abstract

Inheritance nets are used for the representation of hierarchical knowledge. If we allow the representation of positive and negative information, conflicts can occur. They are resolved according to the deduction strategy used (cp. [T 86], [HTT 87], [THT 87]) that is based on two fundamental principles: 1. A link is better than a compound path. 2. More specific information is better than less specific information. We doubt these principles and give an alternative approach instead. It is more general in that it allows to label the links in an inheritance net. The decision which path to believe depends on a partial order given on the labels of the net.

We will show that we can model preemption as defined by [T 86] and get the set of sceptical valid paths as the intersection of all possible sets of credulous valid paths.