Lists Home |
Date Index |
> > I personally think using an uber-framework about syntax and
> > business-specific models is the only way around this, and the RDF/OWL
> > approach seems a very good candidate.
> John Sowa just posted this on the Conceptual Graph list -
> In it he refers to another article, and says in part -
Thanks for the ref - I resubscribed to the CG list recently, and haven't yet
got around to seeing what's happening (267 messages unread - long weekend
> "I like his examples of companies that have no definition -- or
> sometimes, too many definitions -- of fundamental terms, such as
> "customer" and "product". Trying to define a "standard ontology"
> that would state official definitions of such terms would not help.
> On the contrary, it might make matters worse:
> 1. If there were an official definition, some programmer or
> database administrator might be tempted to implement it.
How would that make anything worse?
> 2. As a result, every application that assumed a different
> definition would fail.
Yes. They would also fail if they used non-standard definitions that were
> 3. In an ideal world, programs would stop with an error message
> that explained that there was an incompatible definition of
> some critical term.
What are errors doing in an ideal world? But in our subideal world,
fail/error message does seem reasonable.
> 4. In the real world, most programs would continue to run, but
> they would generate erroneous data that might cause important
> customers or products to be ignored or treated incorrectly.
What is there to support this statement? If the data is being validated,
then you'll get a message when it hits invalid data. In any case it's still
orthogonal to the idea of a standard. If you try and pass data between
different systems which use non-standard ontologies, then the same
(erroneous data or error messages) will happen.
> 5. Even if all offending definitions could be found and
> eliminated, programmers might be forced to implement
> "work-arounds" consisting of highly insecure and illegal
> code to handle customers and products that fall outside
> the official definitions. "
Again, what is there to back this up? Again, the problems people have when
they develop non-standard systems can hardly be blamed on the standards.
> Comments, Danny?
I don't know why Sowa passed on the material above, they look like arguments
but there isn't much substance when you look closely. But there certainly
are good arguments against standard ontologies. Standards can only work when
there is enough common ground between the users. Maybe there isn't enough
common ground for a single, global "customer" and "product" ontologies. So
different companies may well need different ontologies. But assuming there
is some level of compatibility between the systems (i.e. the companies can
trade) then there needs to be a way expressing the relationships between
those ontologies. If the ontologies themselves are expressed in a common
language, that becomes possible.
I really like Sowa's Analogical Reasoning ideas, and they could easily
become an important part of our systems a few years down the road. There are
difficult problems which need something like this. But I don't think he does
his case any good by pointing to problems that can be tackled directly using
a conventional approach to first-order logic.