Empirical Methods for “Ontological Engineering” — Case Study: Objects

Abstract

One of the main topics discussed at the LILOG workshop on ontology was methodology: how can we justify the claim that a certain realization of an AI ontology is a “good” way of representing some field of knowledge (or ‘knowledge cluster’)? A number of criteria are proposed for evaluating the ontology and inference rules that pertain to some cluster, in terms of the inferences about that cluster which they support: (partial) correctness and completeness, ex-tendibility, domain independence, and context neutrality; two criteria from software engineering, modularity and transparency, are also taken into consideration. It is argued that since we cannot presuppose formal definitions for these properties that are similar to the definitions of soundness and completeness for logic, we must apply the empirical methods of the humanities (e.g. linguistic semantics) to determine whether they are satisfied. This claim is then tested in a case study of possible ontologies of objects, in which I criticize two simple ways of representing the meanings of the German dimensional adjectives. It turns out that a knowledge representation pertaining to objects based on a theory from linguistic semantics (Lang 1987, 1989) about knowledge of spatial objects and the use of that knowledge in language stands up well under critical evaluation. This theory depends on making explicit distinctions between linguistic and extra-linguistic knowledge. The overall message is to support the popular view (promulgated especially by Hayes 1979, 1985a) that knowledge engineers are well-advised to seek out independent theories as a foundation of their work, since much of the empirical research relevant to the justification of their representation has been done for them in the formation of the theory.

Kai von Luck informs me that the term “ontological engineering” originates from Douglas Lenat.

Das Leben ist eher breit als lang und wir steh’n alle mittenmang