A Skeptical Theory of Mixed Inheritance

Abstract

This paper is concerned with the problem of providing a semantic account for inheritance networks capable of representing both strict and defeasible information. The importance of representing defeasible information in a knowledge base—particularly, in a frame- or network-based inheritance reasoner — has been widely recognized ever since the publication of Minsky’s original paper on frames [11]. Although early systems designed to allow defeasible inheritance reasoning, such as FRL [12] and NETL [6], were subject to semantic difficulties in their treatment of cancellation, these problems by now are essentially solved. In fact, there exist today a number of well-defined and intuitively attractive theories of defeasible inheritance, including those of Touretzky [15], Sandewall [13], and Horty et al. [8]. The variety of these theories does not seem to indicate any kind of instability or chaos in our understanding, but instead, the presence of a range of options in the design space for defeasible inheritance reasoners; some of these options are surveyed in Touretzky et al. [16].