Re: Media types for the Semantic Web

From: Mark Baker <distobj@acm.org>
Subject: Re: Media types for the Semantic Web
Date: Wed, 7 Apr 2004 12:51:30 -0400

> On Wed, Mar 31, 2004 at 03:25:51PM -0500, Peter F. Patel-Schneider wrote:
> > > So by "what if you don't", do you mean that there's no equivalent
> > > licensed-as-RDF graph which can communicate that same information?
> > 
> > Precisely.  There are far more things under the sun than RDF can imagine.
> > (I wish I knew the source of this ``quote'' - it doesn't appear to be on
> > the web.)
> 
> Understood.  I suppose the difference between something like owl:sameAs
> and your run-of-the-mill foo:bar predicate is that the former, when
> added, changes the existing graph rather than just extending it,
> wouldn't you say?  

No, I would't say.  The graph (by which I take to mean the RDF graph) is
just a set of triples [RDF Semantics, Section 0.3).  In the graph there is
no distinction between owl:sameAs, foo:bar, and rdf:type - they are all
just URI references.

> (Aside; is there a name for these sorts of
> predicates?).  I'm wondering whether a mandatory extension mechanism
> in RDF wouldn't be appropriate when such a predicate was used, in order
> to permit the sender to say "If you don't understand this predicate,
> then you can't build the graph that I'm trying to communicate"?  That
> seems sensible to me.

Well, some of these things might not be predicates.  Even requiring that
everything be filtered through the mechanism of an RDF graph might preclude
a particular kind of extension.

> DanC also said something to me at one point (can't find it), that he
> felt there was an argument to be made for assuming that any use of
> the OWL namespace (or maybe it was rdfs) necessarily imports OWL (rdfs)
> semantics.  

Hmm.  Well this might work, except that, as I indicated above, it may be
that a particular extension requires a different kind of logical formalism,
which cannot be couched as strictly an extension of the RDF machinery.
Further, this approach doesn't seem to fit into one of the preferred ways
of using URI references from the OWL namespace, namely just adding a means
of coidentification to RDF.

> While that would certainly be a simple solution, I don't
> believe it to be a workable (scalable) one, as given an arbitrary RDF
> document, one would not know whether or not some namespace required
> such importation, and therefore whether one was able to understand the
> meaning of the graph as intended by the sender.  But ... mandatory
> extensions might help there, I think, so that a sender could label any
> predicate requiring importation as "must understand"; if the recipient
> understood the predicate, then it would presumably understand the
> intended semantics and could do the right thing.  If it didn't, at
> least it would know it was missing out on some triples.

What does ``missing out on some triples'' mean?  

My preferred solution to all this, by the way, is to have different
languages, perhaps all being written down in XML, and perhaps not.  Each
language would have its own media type, at least.  (I'm not sure whether
media types are adequate for this.)  

Then there is no notion of missing out on triples.  A piece of software
gets to see out front what the language is, and can either handle it or not
(or, perhaps, partly handle it).

[...]

> Cheers,
> 
> Mark.
> -- 
> Mark Baker.   Ottawa, Ontario, CANADA.        http://www.markbaker.ca


peter

Received on Wednesday, 7 April 2004 14:29:32 UTC