SUO: Ontology Registry. Was Question about Example in KR Book
Speaking of Afghan bazaars, this topic was discussed while
you were away in Afghanistan and the conclusion was that a
registry of modules would be good. However, there was also
a lot of debate about the details.
> Monolithic standards are a problem. I agree. But usually
> what happens in the standards world is that one standard is used
> - it is stretched beyond its limits - and another replaces it.
> This is happening now with the ICD9 to ICD10 switch, for example.
One of the reasons for a modular system is the option of replacing
or updating modules while allowing interoperability with older
modules for older applications without causing total disruption.
My view is that the system should be more like Unix, which has
evolved in a modular way over the past 30 years. It is still
possible to run older applications that use older modules and
libraries even while installing new applications that use the
new modules. I think that the Windows model of total disruption
every couple of years is not something we should emulate.
> But does this argue for flock of competing upper ontologies?
> I'm not so sure. Yes, there is competition out there, but most
> of what we see as serious competition is just hype. There
> are in fact very few real upper level ontologies at use in
> any applications other than on academic toy problems.
We have to develop a migration path from where we are to where
we would like to be. Right now, people either have *no* upper
ontology or they have totally incompatible ones, such as Cyc,
SUMO, Dolce, or others. WordNet is perhaps the most widely used
starting point, but it has an upper ontology that is generally
rejected by all the others. Of those groups that do have upper
ontologies, there is probably only one statement that everyone
can agree on:
"Everybody else's upper ontology is totally unacceptable."
> Would it not be more fruitful to ask those who have "been there
> and done that" whether top levels are useful and which features
> they have found useful through experience?
You can ask, but not many people have given clear answers.
Since you have worked on the subject, what is your experience?
> We need to remember what has been forgotten on this list - that
> utility does not negate the validity of theory but does bound
> its scope, Theory that is not useful will be ignored, no matter
> how elegant or how true.
I certainly agree. That is why I have proposed that the registry
should contain annotations (or at least pointers) that record the
ways in which any particular module has been used, how well it has
performed, and how useful, practical, robust, or buggy it has been.