RE: semantics status of RDF(S)

From: "Danny Ayers" <danny@panlanka.net>
Subject: RE: semantics status of RDF(S)
Date: Sun, 1 Apr 2001 21:25:48 +0600

> While I can follow enough of your arguments to give me the impression there
> may well be some important flaws in the semantic model of RDF, I have
> trouble trying to fully pin down the problems identified. I'd like to act as
> Devil's Advocate for a moment :
> 
> The first point seems to be that some definitions are too loose be useful,
> which seems a bit like I can't walk down a street because it isn't on the
> map. Ok, solid formalisms should really underlie the model for
> interoperability (so two interpretations will be identical), but are they
> essential for the model to be useful? Isn't it likely that the assumptions
> made about the gaps in the foundations by different agencies are going to be
> close enough? (at least for the time being...).  I'm also having severe
> difficulty trying to figure out what you mean by the 'encoding' that
> provides the representational power - isn't a triple a perfectly good way of
> modelling a bit of semantic information? Also, why can't quantification be
> represented in triples?
>
> Personally I would appreciate one or two example cases where the model comes
> unstuck (ideally in informative prose), and a suggestion as to what
> formalisms & extensions are required to allow RDF(S) to transfer interesting
> semantic content. Also, a clearer idea of what you mean by 'interesting
> semantic content' would be most welcome.
> 
> Cheers,
> Danny.

Sure, anything can be represented in triples.  However, suppose that you
and I both want to represent negation in triples.  You use reified statements
and the resource http://foo.bar.net/negation#not.  I decide to use an
s-expression encoding (i.e., represent statements as lists) and
http://research.buu.com/sexpr#negation.  We both provide a decent semantics
for the triples we care about.  We both now have represented negation (or
quantification, or whatever).  

However, our meaning does not come from the triples, and certainly is not
related to the meaning that RDF gives to triples.  It instead comes from
the semantics that we have provided and the encoding of this meaning in the
triples.  Further, there is no way that we can directly use the triples to
pass meaningful information back and forth between us.   The presence
of RDF and RDF(S) does not provide any help either.  The only way out of
this impass is if we understand the other formalism's encoding and can undo
that encoding.

The situation can be even worse if we both actually meaningfully used RDF
and RDF(S) as part of our schemes due to the lack of a firm semantics for
RDF and RDF(S).  For example, suppose that I decided to use distributive 
referents and alternatives to represent disjunction.  I might feel so
justified because of the statement ``An application using a property whose
value is an Alternative collection is aware that it can choose any one of
the items in the list as appropriate.''  However, suppose that you decided
that distributive referents didn't work this way on Alternative collections
because of the statement ``Using a distributive referent on a container is
the same as making all the statements about each of the members
separately.''  

Now you might happily read RDF that I produced, and think that you
understood it.  However, you are turning all my disjunctions into
conjunctions, which will undoubtably result in problems.


Peter F. Patel-Schneider
Bell Labs Research

Received on Monday, 2 April 2001 09:10:11 UTC