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.