Re: problem introduced by recent change to RDF MT

From: pat hayes <phayes@ihmc.us>
Subject: Re: problem introduced by recent change to RDF MT
Date: Thu, 12 Jun 2003 10:55:53 -0500

> >From: pat hayes <phayes@ihmc.us>
> >Subject: Re: problem introduced by recent change to RDF MT
> >Date: Wed, 11 Jun 2003 15:50:52 -0500
> >
> >>  >The recent changes in the RDF MT (10a June 2003) have resulted in the
> >>  >following problem:
> >>  >
> >>  >An rdf interpretation I is now of an arbitrary vocabulary V. 
> >>This means that
> >>  >the domain of IS is V.  However, V does not necessarily include the members
> >>  >of the RDF vocabulary.
> >>
> >>  True; this was deliberate.
> >>
> >>  >  This results in the potential breakdown of the RDF
> >>  >semantic conditions.  For example, there might not be a domain element
> >>  >corresponding to rdf:type.
> >>
> >>  There might not, indeed, in a simple interpretation of an RDF graph
> >>  which did not use the URIref rdf:type; that was true previously, of
> >>  course. However, the RDF semantic conditions require that
> >>  IEXT(I(rdf:type)) contain at least infinitely many pairs of the form
> >>  <x, I(rdf:XMLLiteral)>, so require that I(rdf:type) be in IP; and the
> >>  first semantic condition requires that IP be a subset of IR, in every
> >>  rdf-interpretation. So the conditions do not break down for rdf- (or
> >>  rdfs-) interpretations.
> >
> >I guess I didn't make my point correctly.
> >
> >If rdf:type is not in V then I(rdf:type) is not defined so there is no way
> >that the semantic conditions can say that I(rdf:type) is in IP.
> 
> Ah, I take your point.  I think intuitively of all mappings as 
> partial on the entire universe, so automatically extendable to a 
> larger domain where required by context, but I understand this is not 
> widely regarded as kosher.

Well, I don't have any particular beef with having I work on all names, but
this is not how the RDF model theory works, at least as I read it.

> I have modified the text to refer to (V union crdfV), where crdfV is 
> a 'core' vocabulary containing rdf:type, rdf:Property and 
> rdf:XMLLiteral. Similarly the RDFS conditions refer to V union crdfV 
> union rdfsV. This seems to cover the necessary ground while keeping 
> the closures finite.

I am worried about this change.  It has significant consequences to the
observable entailments.

(I just noticed a typo in Section 1.4.  There is a ``than'' in the third
semantic condition.)

> >>  Pat

peter

Received on Friday, 13 June 2003 07:28:05 UTC