Re: LANG: syntactic version for imports (and other things)

From: Jeff Heflin <heflin@cse.lehigh.edu>
Subject: Re: LANG: syntactic version for imports (and other things)
Date: Tue, 17 Sep 2002 13:33:41 -0400

> "Peter F. Patel-Schneider" wrote:
> > 
> > From: Jeff Heflin <heflin@cse.lehigh.edu>
> > Subject: Re: LANG: syntactic version for imports (and other things)
> > Date: Mon, 16 Sep 2002 17:59:02 -0400
> > 
> > > "Peter F. Patel-Schneider" wrote:
> > > >
> > > > From: Jeff Heflin <heflin@cse.lehigh.edu>
> > > > Subject: Re: LANG: syntactic version for imports (and other things)
> > > > Date: Mon, 16 Sep 2002 10:06:27 -0400
> > 
> > [...]
> > 
> > > > > Also, I expected to see some account of the meaning of this construct.
> > > > > You didn't like my entailment-based version, what would you suggest
> > > > > instead?
> > > >
> > > > The meaning is the obvious one.  The contents of the imported ontology
> > > > are considered to be part of the meaning of this ontology.
> > >
> > > I have to admit I am a bit suprised at this answer, especially since it
> > > is coming from someone who normally holds semantic precision as
> > > something of uptmost importance.
> > 
> > The meaning is quite clear, and quite precise.
> > 
> > > What are the "contents" of an ontology?
> > 
> > An ontology is a document.  Its contents are the contents of the document.
> > 
> > > Is it the RDF syntax? The triples? The abstract syntax?  The conditions
> > > imposed on interpretations by the syntax?
> > 
> > At the level of *syntax* it doesn't matter.  All you need to do is to take
> > the document, in whatever form it is, and add it to the importing ontology,
> > in whatever form *it* is.  Then the syntax-to-semantics mapping takes
> > over.
> 
> But that's where it's NOT clear. What does it mean to "add a document to
> an ontology?" If the documents are RDF/XML, you can't simply insert one
> document into the next and expect it to work (as we've seen with the
> iterations we've done on Mike Smith's approach). Clearly, the syntax
> does matter. We can't just say "ignore the XML junk" because the WG has
> made the decision that we must work in the RDF/XML world.

Well, there is a ``mapping'' from syntax (both RDF/XML documents and abstract
syntax) to semantics.  Just use that.    (The relationship is not really a
mapping, but this is an easy way of glossing over what is really going on.
In more detail, the meaning of an imports directive consists of treating
the referent of the directive as another source of constraints on models.)

The semantics mapping from RDF/XML requires RDF/XML parsing into triples,
so do that first, as a separate document.  Don't bother to create a single
document, because there is no need for it.  Simple.  (If you really want
to, you could create a single RDF graph, I suppose, but what would be the
purpose?) 

> > > Do you agree with Mike Smith's
> > > syntactic approach [1]?
> > 
> > Sure, except that it runs afowl of XML junk.
> > 
> > > If so, what about the issues I raised [2]?
> > 
> > By treating the imported ontologies as separate documents, and doing the
> > syntax-to-semantics translation in that context.  Actually only the XML/RDF
> > to RDF graph needs to be done in context.  You could also duplicate the NS
> > stuff on every top-level element in the included ontology.
> 
> Now this seems like we are getting somewhere.  Although it is still a
> little vague (it sounds like two or three alternatives are described), I
> think I hear you saying that one solution is to do inclusion at the
> abstract syntax level. 

In the abstract syntax, you can do inclusion, because there is no context
to get in the way (yet).  If we add extra-logical features to ontologies,
then there may be a reason to not think of imports in this fashion.

> Note, that this is very different from Mike's
> approach because he does the inclusion at the XML syntax level. 

This could work also, but it requires fiddling with RDF/XML.  But if you
don't bother to create one RDF/XML document then there are no issues.

> It needs
> to be fleshed out, but I think I see how you could go from an OWL RDF
> document into a semantics that takes imports into account and does what
> its supposed to. But don't get me wrong, I'm still not wild about it
> because it seems that we [a]re specifying a procedure for computing meaning
> as opposed to saying what the meaning is. It seems contrary to the whole
> notion of model theoretic semantics.

Nope, this is how semantics works, model-theoretic or otherwise.  You take
some syntax and you define its relationship to semantic constructs, be they
interpretations, rules of inference, or axioms.  The only extra bit here is
to follow a pointer.

> > > And
> > > finally, how is this "better" than my entailments based approach?
> > 
> > Your approach requires a treatment of ontologies and imports in the
> > semantics, which raises big issues that I don't want to have to handle in
> > the semantics.
> 
> I have to admit, I still don't understand what the big issues in the
> semantics are. 

Well, if you have ontologies in the domain of discourse, what can you do
with them?  Can you point to a blank node?  Can you deny importing?  Etc.
Etc.

[...]

> > > Jeff
> > >
> > > [1] http://lists.w3.org/Archives/Public/www-webont-wg/2002Sep/0167.html
> > > [2] http://lists.w3.org/Archives/Public/www-webont-wg/2002Sep/0197.html

peter

Received on Tuesday, 17 September 2002 14:58:34 UTC