Re: rdf inclusions (was Re: DAML Level of Effort for FY03-FY05)

   Drew McDermott writes:
   > Anyway, the following strikes me as the only reasonable point of view:
   > There is simply no way to import part of an ontology, as Sandro and
   > Frank both seem to want to do. 

   [Sandro]
   I'm not sure how I was misunderstood.  I think the only reasonable
   thing to import is another RDF graph (serialized as some
   web-accessible document and named by the web address).

Hey, I can misunderstand anyone.  I was probably accepting something
Frank implied.

   [Sandro]
   So are you committing to some particular web content being served at
   the namespace address?  That's a difficult issue to get consensus on.
   People want to be able to use XML instance data with schemas not on
   the web and not necessarily shared.  That seems to me like a bad idea
   with ontologies, but it's a big problem any XML-namespaces proposal
   will run into.

Unless I'm misunderstanding again, I think we're on the same
wavelength here.  There should be a difference between grabbing XML
instance data on an ad hoc basis, and saying, I'm committing to the
claims of yonder ontology.

   [me]
   [example of typical RDF using stuff without importing it]
   > But how can a computer see all this?  Should it just assume that any
   > namespace that expands out to something ending in ".daml" somehow is
   > an ontology?  This strikes me as a silly tactic for avoiding an
   > "imports" declaration.

   [Sandro]
   Is it important to you to keep the instance data and the ontology in
   separate graphs?  I think the Semantic Web is based on merging all the
   graphs one believes and can access, so the instances and their
   ontologies should not be considered separable graphs.

No, the graphs aren't separate.  But some subgraphs are more equal
than others.  For instance, suppose I write a DAML type checker.
Given some instance data, and the ontologies they use, I should be able
to check that the data don't contradict the ontology.  (E.g., no one
has two mothers if the ontology says they shouldn't.)  Now suppose I
suck in some instance data, and there happen to be some daml
assertions mixed in.  It seems to me that without an explict "imports"
statement I would be justified in just taking the instance data that I
wanted (say, with a particular syntactic form) and ignoring the
extraneous daml assertions.

Hmmm.  I seem to contradict myself, given that I want to forbid
picking and choosing pieces of an ontology, but allow exactly that in
other cases.  The resolution is that ontology importing seems to be a
distinguished case, which should have a clear syntactic tag.  I
understand what it means to import an ontology: take it all.  I am
willing to allow for all sorts of other ad hoc cases, but we don't
have to legislate those up front.  

                                             -- Drew McDermott

Received on Monday, 22 April 2002 08:17:52 UTC