Re: Media types for the Semantic Web

From: Mark Baker <distobj@acm.org>
Subject: Media types for the Semantic Web
Date: Wed, 31 Mar 2004 12:13:19 -0500

> BTW, is this on topic for this list?  I can't tell any more. 8-/
> 
> On Fri, Mar 26, 2004 at 01:28:02PM -0500, Peter F. Patel-Schneider wrote:
> > An OWL ontology is much, much more than an RDF graph, unless you believe in
> > the strongest of the various one-meaning proposals that have been bandied
> > about.
> > 
> > For example, if I ask you whether the RDF graph
> > 
> > 	ex:foo owl:sameAs ex:bar .
> > 
> > entails
> > 
> > 	ex:foo rdfs:subClassOf ex:bar .
> > 
> > you *should* say no, because I have only licensed you to treat this graph
> > as an RDF graph, not an OWL ontology.
> 
> Sounds reasonable.  I like the use of the term "licensed" there too, as
> it jives with my view of the role of media types.
> 
> I'm still a novice on "model theory" stuff, so please help me understand
> where I might be wrong, but here's how I currently see the different
> meanings that one might have when transferring a graph;
> 
> 1.  I'm communicating just this graph
> 2.  I'm communicating this graph and some axiomatic triples
> 3.  I'm communicating this graph and every inferred triple
> 4.  I'm communicating this graph, these axiomatic triples, and all
> inferred triples from the resulting merged graph

I don't think that *any* of these are appropriate.  What is being
communicated should be information, not syntax.  (Sometimes the information
is carried by the particular syntax, but, again, it is the information, not
the syntax that is being communicated.)

> I believe that using application/rdf+xml handles #1 easily, and #2
> when the axiomatic triples are the RDF axiomatic triples (since those
> are equivalent in that case, it seems).  When the axiomatic triples are
> not (just) RDF's, then a new media type seems necessary.  

I do agree that attaching different meanings to the syntax should require
some sort of difference in the communication, be it either a different
media type or some information carried in the headers, or even a magic
number at the beginning of the contents (as in Unix).

> FWIW, what I
> believe this suggests is that axiomatic triples should be avoided where
> possible, precisely because they lead to an explosion of media types
> (or an explosion of non-self-descriptive messages if new media types
> aren't minted 8-).  

Why should there not be many different kinds of meaning associated with
communications?  There are lots of different kinds of meaning associated
with human communications, for example.  (Think chat (English) vs chat
(French).) 

> This seemed to be Dan's concern;
> 
> ""
> Now further up the spectrum, we might consider application/owl+xml.
> 
> I find that objectionable because it suggests that
> dublin core and adobe XMP and RSS and so on
> need their own media types, and it leaves me
> wondering what media type to use if for
> a document that mixes all these vocabularies
> together.
> ""
>   -- http://lists.w3.org/Archives/Public/www-webont-wg/2002Oct/0162.html

Well, what is wrong with this?  Well maybe it is an indictment of using
media types instead of some other mechanism, but why otherwise?

> AFAIK though, RSS and XMP don't have their own axiomatic triples, nor
> require inferencing, so Dan's argument seems to be a strawman;
> application/rdf+xml should suffice for them.
> 
> For #3 and #4, I think new media types are also required.
> 
> Thoughts?
> 
> Mark.
> -- 
> Mark Baker.   Ottawa, Ontario, CANADA.        http://www.markbaker.ca

Peter F. Patel-Schneider
Bell Labs Research

Received on Wednesday, 31 March 2004 12:35:32 UTC