Panel Discussion on Ontologies

Abstract

A discussion on the topic of ontologies in agent-based systems had Drs. Sidney Bailin, Gary Berg-Cross, and Tim Finin as panel members. Bailin, who chaired the session, led off with a series of questions for consideration: 1. Can we properly speak of an ontology (an artifact, as opposed to just ontology as a field of study)? Can we properly speak of multiple ontologies? 2. Is an ontology different from a dictionary? A taxonomy? A class hierarchy? A domain model? 3. Will the semantic web be an ontology? 4. Are ontologies needed now more than ever? Bailin answered his first question by noting that his Microsoft Word grammar checker continually objects to the phrase an ontology as well as to the plural ontologies. Nevertheless, he finds it impossible to do without these constructions given current conventions. He noted that ontologies have characteristics in common with dictionaries, taxonomies, class hierarchies, and domain models, and that the term is sometimes used synonymously with one or another of these alternates. It tends to connote things that the alternates do not imply, however. Ontologies typically have more structure than dictionaries, e.g., in their arrangement of concepts into an is-a hierarchy (that is, into a taxonomy). They may have more precise semantics than taxonomies, e.g., by identifying attributes associated with a given concept, and possibly rules governing the values that the attributes assume. They are, in this respect, similar to class hierarchies, but they tend to be used at the problem-domain level while classes are often viewed as solution domain (i.e., design or implementation) constructs. In this respect they are like domain models; but a domain model may contain, in addition, a variety of other kinds of information. One participant pointed out that a domain model may use an ontology.