Re: need to determine what RDF is

From: Dan Connolly <connolly@w3.org>
Subject: Re: need to determine what RDF is
Date: 30 May 2002 12:32:33 -0500

> On Thu, 2002-05-30 at 11:06, Peter F. Patel-Schneider wrote:
> > From: Dan Connolly <connolly@w3.org>
> > Subject: Re: need to determine what RDF is
> > Date: 30 May 2002 10:31:43 -0500
> > 
> > > On Thu, 2002-05-30 at 10:26, Peter F. Patel-Schneider wrote:
> > > [...]
> > > > I'm only interested in relationships between RDF graphs.  Which such
> > > > relationships are RDF relationships?  
> > > > 
> > > > My view is that the only such relationships are RDF entailment and RDFS
> > > > entailment.  Any agent that computes any other relationship between RDF
> > > > graphs is not doing RDF.
> > > 
> > > Why is RDFS special? It's just the first of many RDF vocabularies,
> > > no?
> > 
> > No.
> > 
> > RDFS is not an RDF vocabulary at all.  It is an extension to RDF, as
> > witness its treatment in the RDF Model Theory document.
> 
> Sigh... terminology again... in what way is it not
> an RDF vocabulary? You probably don't even know what I meant
> by that term. What distinction do you mean to draw
> between 'RDF vocabulary' and 'extension to RDF'?

I know what that term should mean, based on the meaning of the terms it
contains.  I don't know what you meant by that term, as evidenced by your
comments above.  All I know is that you don't mean what the term should mean.


> > RDFS is special because RDFS entailment is defined in the same document as
> > RDF entailment
> 
> Ah... then that's misleading; I worried about that.
> I'll see that we split them into separate documents.
> 
> > and the RDF Schema document is being produced by the RDF
> > Core Working Group.  
> 
> > > I could understand a definition that said 'anybody doing more than
> > > RDF simple entailments isn't doing RDF', but I don't understand
> > > a definition of RDF that includes RDFS but not dublin core,
> > > RSS, DAML+OIL/OWL, etc.
> > 
> > Then why is RDFS entailment in the RDF Model Theory document?
> 
> Editorial convenience.
> 
> >  Why is the
> > RDF Schema document being produced by the RDF Core Working Group? 
> 
> For the same reason that the group that produced the C
> standard also produced the C standard library: end-users
> could write strstr(), but it's a pain; standardized
> idioms promote interoperability.
> 
> And because we tried having them developed in separate
> groups the first time around, and the communications
> costs were very high.

OK.

I'm willing to remove RDFS from any special position, by the way.  I don't
even have a preference.  You may get resistance from other quarters.


> > Why should RDF not include Dublin Core, RSS, DAML+OIL/OWL, etc?  Well,
> > RDF can include any portions of these efforts that are expressible in RDF
> > (by the simple expedient of including the RDF documents produced by these
> > efforts).  However, RDF should not include any other portion of these
> > efforts.  Why?  Well simply because these efforts are not part of RDF, as
> > witness the fact that the RDF documents do not give them any special
> > status.
> 
> Again, by analogy, the ANSI C WG doesn't have to write all
> the C programs in the world.

Precisely, all it has to do is to determine what these C programs are
supposed to do (or mean).  However, you appear to believe that RDF should
encompass the entire meaning of DAML+OIL documents.  This is just as silly
as saying that the C programming language standard should encompass the
entire meaning of C++ programs.

> You might say that the semantics of all those C progams
> is specified by the C standard, while dublin core semantics
> aren't actually expressed in RDF.
> But consider C programs that do I/O withe external
> devices and such.

Yes, and if you want to be very precise, you don't think of the standard
library as part of the C language.  Programs that use functions defined in
the standard library, and expect to get the behaviour of the standard
library, often have to take explicit steps to ensure that they are loaded
with the standard library, in ways that go outside of the C language.  I
know that I have had to do this with my C programs.  (I also know that it
was extraordinarily difficult to get C++ programs to use the C++ ``standard
library'', which was not standard on several systems that I used.)


> I'm willing to stop saying that 'doing OWL is doing RDF',
> but I need some term for the relationship between OWL
> and RDF that is the same as the relationsihp between RDFS
> and RDF, and the same as the relationship between
> dublin core and RDF; as a placeholder, let's
> call it an RDF watermelon.
> 
> An RDF watermelon is
>   * a list of absolute URI references
> 	(usually, they share a prefix, for convenient use
> 	with XML namespace syntax)
>   * an agreement that these terms may be used in RDF/xml
>     syntax
>   * an agreement that when these terms are used in RDF/xml
>     syntax, assertion of such a document licenses
>     all [RDF or RDFS] entailments
>   * a specification of further constraints on the
>     meaning of these terms; i.e. more constraints
>     on which interpretations are models.
>     This specification may end up licensing
>     further formal inferences, or it might just
>     relate the terms to existing conventions and
>     practices, in such a way that humans are
>     expected to be able to judge which interpretations
>     are models, but a machine's understanding
>     will be incomplete.
> 
> RDFS, dublin core, and DAML+OIL look like RDF watermelons
> to me.

Why not call the notion a same-syntax extension of RDF?  That seems to
cover all the bases, and, moreover, makes it clear that RDFS/OWL/... are
*extensions* of RDF.


> > How could the situation be any different?  It seems that you are asking for
> > W3C to bless any effort (e.g., DAML+OIL or KIF) that has any relationship
> > to RDF, even if the only relationship is the effort uses URIrefs to
> > identify its tokens.
> 
> Well, yes; that's pretty much what the Resource Description Framework
> is, to me.

Well, then I think that you need to embark on a major education effort.  I
know that your view is certainly not universal.  I think that your view is
sufficiently different from the normal interpretation of a standard that it
would have to be written in big, bold, flashing, red type at the top of
every part of the RDF specification documents.

> There are a few things beyond using URIrefs: monotonicity,
> completeness (but not necessarily soundness) w.r.t.
> simple entialment, the use of unicode strings (and XML
> infosets) as literals. And for at least some period of time,
> a willingness to use RDF/XML syntax for exchange.



> -- 
> Dan Connolly, W3C http://www.w3.org/People/Connolly/

peter

Received on Thursday, 30 May 2002 13:55:43 UTC