So it looks like the consensus among those in this discussion is:

An ontology is a collection of

classes, each with possibly unique property values;

a few constant instances (e.g., equilateral triangle = special instance of generalized triangle, etc);

I’m not sure that I would see equilateral triangle as an instance. Surely there is more than one of them? On the other hand it could be a subtype of generalized triangle (some triangles are equilateral).

On the other hand there are plenty of individuals I might want in my ontology. If I want to define a class of “Ford Motor Car” in my ontology then it is useful to be able to have the individual “Ford Motor Company” in my ontology so that I can make the restriction class.

and

logical relationships among the classes and instances.

And nothing else. If that satisfies everyone, then any operational system would require more than just an ontology. It would also require that information nobody seems to want to call ontological, like the specific employees in the employee table.

If we accept this definition among the group of us, an ontology with a database to back it would be about the simplest semantic system I can imagine being useful. The database would store the instance data beyond the ontology, but the ontology would define the classes, properties and relationships among the entities.

MW: It will probably replicate much of the ontology too. It might be better to think of the ontology as an abstraction of the database, with some rules added, so that you can e.g. check the consistency of the database.

But then how do we account for the diverse viewpoints going into the system from multiple users? We all agree that each user has a unique ontology of her personal world. We know that subjectivity gets squeezed into the tightest databases with the strictest controls.

MW: You can’t. And in fact the problem really is just how do you impose sufficiently strict controls such that the range of meaning is sufficiently small that sufficiently accurate communication is possible.

So how do we account for personal ontologies in a semantic system?

MW: The alternative is to document all the individual ontologies and map between them. This is horrendously expensive, so a much cheaper alternative is to sit down and agree to use one in a particular way, which may not be the way that any of the individuals actually sees things, but at least is clear. This is at least analogous to the situation you find when an Italian, and Frenchman, a German and a Spaniard have a business meeting. They speak English, rather than each have 3 translators.

So I suggest we return to the real world.