Re: Comments on informal meaning of the RDFS vocabulary

From: Brian McBride <bwm@hplb.hpl.hp.com>
Subject: Re: Comments on informal meaning of the RDFS vocabulary
Date: Wed, 29 Jan 2003 13:20:39 +0000

> At 07:01 29/01/2003 -0500, Peter F. Patel-Schneider wrote:
> 
> I really should have left Danbri to respond to this, (my fingers engaged 
> before my brain), but having started it makes sense for me to keep going.
> 
> [... deleted part where Peter explains  ...]
> 
> Trying to reflect back the issue in my words, your concern is that schema 
> document does not distinguish between parts of the specification that are 
> formally captured in the model theory, and parts that are not, e.g.
> 
>    http://www.w3.org/TR/rdf-schema/#ch_type
> 
> is supported by the model theory and
> 
>    http://www.w3.org/TR/rdf-schema/#ch_comment
> 
> is not.
> 
> I agree that the schema doc does not make that distinction.
> 
> > >
> > > Ah, right, I think I see the trend of what you are getting at here, but I
> > > still need a clearer statement of what the issues are.
> >
> >The lack of a distinction in the wording describing rdf:type and rdfs:label
> >takes part of the social meaning that is being promoted for RDF (i.e., the
> >social meaning of rdfs:label) and tries to give it the same status as the
> >model theory meaning.
> 
> What I'm not clear about is why it matters.
> 
> Can you offer a test case to illustrate a problem.  

Well, if one really believed RDF Schema, then the model-theoretic behaviour
of RDF should abide by whatever is said in rdfs:comment value.  For
example,

	ex:Cretan rdf:subClassOf ex:Person .
	ex:Cretan rdfs:comment "All Cretans are liars" .

would mean that the model theoretic consequences of 

	ex:John rdf:type ex:Cretan .

would include that John is a liar.  I leave it up to the RDF Core WG to
turn this into a test case.

> What difficulty is an 
> implementor or a user going to have?

Well, none, provided that the implementor only uses Semantics as a
specification, and the user doesn't expect RDFS systems to actually do what
Schema says.  The text in Schema is, of course, only text, and only
informative text at that.  As far as actual implementations go, reading
Schema is completely unnecessary.

However, given that Schema is part of the output of the RDF Core Working
Group, I do expect it to reflect the meaning of RDF as given in Semantics.
Right now it doesn't.

So, an implementor who looks to Schema for guidance on how to build an RDF
system is going to get the impression that there is no difference in import
between the meanings given to rdf:type and rdfs:label.  The implementor may
decide that the only suitable way of presenting resources to users is via
values of their rdfs:label properties.  The implementor may decide that RDF
lists have unique firsts and rests and write an RDF system accordingly.  The
user may believe that RDF lists must always have unique firsts and rests.

Social meaning adds yet another complication to the above situation.
Because the meaning of rdfs:comment is given in the same way as the meaning
of rdf:type, users may believe that rdfs:comment values actually (must)
describe the model-theoretic behaviour of resources.

> Brian

Peter F. Patel-Schneider
Bell Labs Research
Lucent Technologies

Received on Wednesday, 29 January 2003 08:42:38 UTC