W3C home > Mailing lists > Public > public-owl-wg@w3.org > November 2007

Re: UFDTF Metamodeling Document

From: Peter F. Patel-Schneider <pfps@research.bell-labs.com>
Date: Fri, 30 Nov 2007 13:15:03 -0500 (EST)
Message-Id: <20071130.131503.195143113.pfps@research.bell-labs.com>
To: conrad.bock@nist.gov
Cc: ekendall@sandsoft.com, public-owl-wg@w3.org, cawelty@gmail.com

From: "Conrad Bock" <conrad.bock@nist.gov>
Subject: RE: UFDTF Metamodeling Document
Date: Fri, 30 Nov 2007 12:44:03 -0500

> Peter,
> 
>  >  > For example, graphs can span multiple documents, multiple graphs
>  >  > can occur in the same document,
> 
>  >  Umm, where is the relationship between graphs and documents
>  >  represented in the metamodel?  I don't see it anywhere.
> 
> I thought it was the relation between Document and RDFStatement in
> Figure 10.7, but not sure.

Nope, that has relationships between documents and lots of other things,
but no relationship between documents and rdf graphs.

>  >  It is true that RDF Concepts does conflate statement and triple in a
>  >  couple of places, but it also make the distinction clear in several
>  >  other places.  RDF Semantics is quite clear that the term is triple
>  >  and does not use statement in this context at all, reserving it
>  >  almost exclusively for rdf:Statement.
> 
> It doesn't conflate these, as far as I can tell.  In Figure 10.2,
> Statement has a property called "reified" with enumerated values for
> unrefified statements (which I gather you are calling "triples"),
> statements that are refified but not asserted as true, and statements
> that are both refied and asserted.

Well, an RDF graph contains triples, not statements, so the metamodel
category RDFStatement has to be the category of triples.  However,
RDFStatment is also standing in for rdf:Statement, the class that has as
its instances tokens that correspond to triples.  (These tokens are
often called reified triples.)  Triples and reified triples are
different in the RDF semantics.  See http://www.w3.org/TR/rdf-mt/#Reif
for more information.  As RDFStatement is used for both triples and
reified triples in the metamodel on page 35, the metamodel conflates
them.

>  >  By the way, it appears to me that the RDF metamodel allows regular
>  >  (unreified) triples (RDFStatement) to be missing a subject,
>  >  predicate, or object, which is not allowed in RDF.
> 
> If your interpretation of the RDF spec is generally agreed (which
> it isn't clear it is), 

Huh?  Who does not agree with the statement that RDF triples have
exactly one subject, one predicate, and one object?

> then this would be an additional constraint on
> Statement on page 40.  BTW, I notice the second paragraph above Figure
> 10.2 says asserted statements must have subject, object, and predicate,
> but this doesn't appear in the constraints on page 40.

>  >  > multiplicities (which may appear to be unintuitive to someone less
>  >  > familiar with MOF as you mention),
> 
>  >  Umm, it appears to me that the metamodel states that the
>  >  CommentForResource and LabelForResource relationships are 1 to many
>  >  (in which direction I'm not sure).  I believe that they should both
>  >  be many to many.
> 
> Just to clarify reading UML, the figure 10.3 says that RDFResource has a
> property RDFScomment with cardinality zero to many, PlainLiteral a
> property commentedResource with cardinality zero to one.  Whether the
> place of the cardinalities is intuitive seems to depend on whether you
> think in objects or relations.

Ok, but this is still wrong.  Any literal, not just plain literals, can
be the comment for an arbitrary number of resources.

>  >  Agreed.  You already know my views on the best way to proceed.
> 
> Yes, just not why.  :)  Even the short alignment discussion we've been
> having has turned out areas you thought were agreed among RDFers but
> weren't.  

Umm, which areas?  RDF triples *have* three components, for example - no
ifs, ands, or buts about it.

> Multiple structural models for OWL would severely limite
> interoperability and consequencely adoption.

I'm not convinced about this.  It may be that tools like Jena need a
different structural model for OWL than do tools like Protege, Pellet,
Racer, and FaCT++.  It may even be that tools that process OWL in yet
other, different, ways, like KAON2 and Hoolet, may need yet another
structural specification.

> Conrad

Peter F. Patel-Schneider
Bell Labs Research
Received on Friday, 30 November 2007 18:32:10 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Tuesday, 8 January 2008 14:13:27 GMT