Re: imports in S&AS

From: "Jeremy Carroll" <jjc@hplb.hpl.hp.com>
Subject: imports in S&AS
Date: Tue, 25 Mar 2003 13:43:42 +0100

> 
> 
> I am reviewing some of my earlier issues with S&AS.
> The good news is that nearly all of them are now fixed.
> 
> On imports, I have one minor but critical issue for Peter, and some
> discussion primarily intended for others in the WG (Peter and I had a brief
> discussion at the editors meeting - and I am not *too* unhappy with the
> current state).
> 
> Critical issue:
> ===============
> 
> The link
> http://www-db.research.bell-labs.com/user/pfps/owl/semantics/rdfs.html#RDF_g
> raph_imports_closure
> does not lead to a definition of imports closure.
> 
> I use this link in
> http://www.w3.org/2002/03owlt/editors-draft/snapshot#syntaxConformance
> 
> Suggest add new definition:
> [[
> Definition: The imports closure of an RDF graph K is the union of the
> smallest imports closed set of RDF graphs containing K.
> ]]
> (it seems odd to me to keep talking about sets of graphs rather than just
> one)
> 
> I could put this definition into test cases, but this would seem odd to me.

I've put this (slightly edited) just after the imports closed definition.

> More General Discussion
> =======================
> In S&AS the underlying model of imports is driven from the abstract syntax.
> In particular the model suggests that the documents imported are themselves
> abstract syntax documents - this is in tension with the WG decision about
> imports which is essentially triple based.
> In other words, S&AS (OWL DL) assumes that imported ontologies are in OWL
> DL, whereas the WG decision is that an OWL DL ontology can import documents
> that are *not* well-formed OWL DL, but taken together form an OWL DL graph.
> I am not too unhappy because, while I think the text is confusing on this
> point, I see no operational impact.

I don't agree with this.  In Sections 2 and 3, of course, ontologies are
abstract syntax ontologies but I don't see any other possibility.  In
Section 5 ontologies are RDF graphs.  The connection comes from the
mapping, and the mapping talks about collections of ontologies being in OWL
DL.  This allows for a particular ontology (RDF/XML document) to not have
all the typing required for individuals to make it alone be valid OWL DL,
but for the missing typing to come from another ontology in the collection.

> Critical phrases:
> section 2.1
> "idealizations of this operational meaning for imports are used"
> 2.3
> "If an ontology imports another ontology, the axioms in the imported
> ontology (and any ontologies it imports, and so on) can be used for these
> purposes"
> (assuming that the imported document *has* axioms)

Well this is Section 2, after all, so what else could there be?

> 3.4
> in definition of satisifies
> "I satisfies each ontology mentioned in an owl:imports annotation directive
> of O."
> (again assuming that the imported ontology is in OWL DL)

Again, what other possiblity could there be here?

> 4 mapping rules
> no mention of imports
> This is the key disconnect between imports in the abstract syntax, and
> imports in the concrete syntax - which is why Peter and I are not arguing
> this point!

Actually Section 4 does provide a mapping for imports in the abstract
syntax, as follows:

	Ontology(O
	  Annotation(owl:imports U) ...
	  )

maps to 

     O rdf:type owl:Ontology .
     owl:imports rdf:type owl:OntologyProperty .
     O owl:imports U .
     U rdf:type owl:Ontology .
     ...

> 5.4 OWL DL
> 
> "Definition: Let T be the mapping from OWL ontologies in the abstract syntax
> to RDF graphs from Section 4.1. Let O be a collection of OWL DL ontologies
> in abstract syntax form. If for any URI, u, in an imports directive in any
> ontology in O the RDF parsing of the document accessible on the Web at u
> results in T(K), where K is the ontology in O with name u. Then O is said to
> be consistent with the Web. "
> 
> Again this disconnects with the definition of an OWL DL document which is
> one whose import closure is an RDF graph that corresponds to a collection of
> OWL DL ontologies.

There is a slight disconnect between Sections 4 and 5.

Section 4 defines the notion of a set of OWL abstract-syntax ontologies
that have a *separated vocabulary*.  Such sets are the ones that can be
transformed into sets of RDF graphs in a way that entailment is preserved.
Section 4 then goes on to define an *OWL DL ontology in RDF graph form* as
the transformation of such a set of OWL abstract-syntax ontologies.
Section 5 doesn't use this latter notion, instead building on the former
notion. Section 5 does end up with a more-restrictive notion.

> Test Case:
> =========
> 
> http://www.w3.org/TR/owl-test/#imports-008
> 
> OWL Full <imports/imports008>:
> 
> first:sub rdf:type rdfs:Class .
> first:sub rdfs:subClassOf first:super .
> 
> and
> 
> OWL Lite <imports/main008>:
> 
> <imports/main008> rdf:type owl:Ontology .
> <imports/main008> owl:imports <imports/imports008> .
> first:sub rdf:type owl:Class .
> first:super rdf:type owl:Class .

This does indicate the difference.

> The definition of "consistent with the Web" is not applicable to this test
> case - this does not bother me too much since I do not see the concept as
> necessary for OWL, but it does strike me as somewhat broken. My preferred
> treatment would be to have imports as an operation on RDF graphs and leave
> the abstract ontologies out of it all together.

So what is the way to proceed?

> Jeremy

peter

Received on Tuesday, 25 March 2003 13:00:05 UTC