Non-monotonic reasoning in a semantic network

Abstract

We define semantic networks for representing knowledge as being multilabeled semiordered directed graphs. Node labels allow us to treat the problem of universal and existential quantification. Arc labels are used for treating the "defined classes" and negations.

Question-answering process amounnts to searching for graph morphisms called "enlargements". Such morphisms are also used for making networks consistent.

Precedence rules are given for non-monotonic inferences, i.e. for selecting between enlargements or between "enlargement paths". These rules are suitable for simple hierarchical taxonomies, and also when predicates occur.