Re: Key questions about common upper ontologies
Danny,
I mostly agree with what you say. In particular,
the following observtion is fundamental:
DA> I'm not sure the commonest use case would be
> the large organization, rather the smaller, more
> domain-specific enterprise (which would in effect
> correspond to the departments in the big org).
Having spent 30 years in a large org (IBM), I'm well
aware of the fact that one department rarely knows what
another department is doing -- and when they do, they
jealously guard their turf. As we have heard from
the 9/11 hearings on TV, the same is true of the US
Government. On a finer scale, the branches of the
White House itself don't know what other branches are
doing, and the guy at the top is clueless.
DA> The scenario I envisage would be not unlike that
> which has been growing in the RDF world with the
> WordNet schema (and others similar). When defining
> a term in a domain-specific schema it's possible to
> point to the corresponding term in the dictionary
> (e.g. in the Pet Profile schema I've been working on,
> mypet:Pet is a subclass of wordnet:Pet). At present
> the usage is something of a blunt instrument, but
> ongoing work on the RDF Thesauri [1] should allow
> more fine-grain qualified application of broad-range
> vocabularies like WordNet.
I agree that WordNet and related tools are extremely
important, but if you talk to many people working on
ontology (e.g., Jim Hendler and Pat Hayes, among others),
they will say that WordNet is definitely not an ontology.
They will admit that it might be somehow relevant to
the way that ontologies are used, but that it's outside
their main field of interest.
After seeing what has happened to Cyc, which has spent
20 years developing a tightly-organized ontology from
top to bottom without achieving a single money-making
application, I have come to believe that the looser
coupling of WordNet is more likely to succeed than
any tight coupling.
An important point is that all the branches of IBM
or the US Government use the same words, even when
they (as usual) have different definitions for them.
The different definitions may create problems, but
the words provide a starting point for indexing,
finding, and comparing the definitions. When the
inevitable discrepancies are found, it is then
possible to begin negotiating.
DA> I believe there's a not insignificant benefit lurking
> in this selective, per-domain approach. Users of a common
> upper ontology don't have to subscribe to the whole ontology.
> If I didn't like WordNet's definition of 'Cat' then I can
> simply use a definition from elsewhere without there being
> a conflict with its definition in the WN vocabulary. Likewise,
> interoperability will be enhanced where terms are shared,
> grounded (if that's the correct word) in the SUO, without
> a need for adoption of the whole system. Aside from the
> practical implications this could be politically expedient,
> as I know there are plenty of folks that aren't comfortable
> with the notion of a single global Uber-ontology (self
> included). In other words, SUO can be useful without
> having to please all the people all the time.
I agree. And this is why I believe that Philippe Martin's
Multi-Source Ontology (MSO) is important. It indexes and
stores everybody's definitions in a convenient repository,
where you can find the different definitions and decide
which, if any, is best suited to your application.
John