Re: new version of S&AS

From: Jeremy Carroll <jjc@hpl.hp.com>
Subject: Re: new version of S&AS
Date: Wed, 19 Mar 2003 15:25:32 +0100

> I have been revising my version of the rules just enough to generate the 
> triple tables ...
> 
> I had a few points/questions - two of which Guus has already made
> 
> - 3.a EquivalentClass
> - 3.b Deprecation of datatypes?
> (I missed the datarange mapping rule qu)
> 
> Further ...
> 
> - the mapping rules don't yet seem to allow DAGs of descriptions; I was hoping 
> for changes in the introductory text

This was not approved last week.

> - the rules seem to allow an annotation with an object which is an arbitrary 
> uriref, which may not have a type triple - is that an oversight or 
> deliberate?

Hmm.  The answer is somewhere in between.   Annotations were initially
uninterpreted, which made URI reference the obvious target.  Recent changes
to annotations have changed the situation somewhat, but maybe not enough to
make a difference.

> - the following triple is incorrectly permitted:
> rdfs:Literal rdf:type rdfs:Datatype .

It may be that the generation of this triple is not a good idea, but how
can we prohibit it.  As well, the status of rdfs:Literal is dependant on
further fixes to the RDF model theory.  These changes will affect much of
the OWL treatment of datatypes.

Right now rdfs:Literal is a datatype name, i.e., its extension is a subset
of LV, not of R, but not necessarily the name of a built-in datatype, i.e.,
using it in a typed literal is probably not a good idea.  There is nothing
to prevent an OWL datatype theory from including rdfs:Literal as the name
of a datatype, just as there is nothing so preventing this being done in
RDF.  However, there would be certain breakage if one did so.

> - I had asked whether we could liberalise the rules concerning 
> InverseFunctionalProperty, TranstiveProperty, SymmetricProperty to avoid 
> requiring two type triples for these in OWL Lite and OWL DL. My suggestion 
> was:
>     + make the rdf:type owl:ObjectProperty triple optional
>     + add global constraint that every individual valued property had one of 
> these type triples.
>     + ensure that each of these classes is a subclass of owl:ObjectProperty in 
> the RDFS compatible semantics
> 
> What is the status of this suggestion?

I do not believe that these made it into the proposed resolution.

> - Further I had suggested that the optional triples permitting 
>       _:blank rdf:type rdfs:Class .
>   on description and restriction nodes were not useful and should be deleted. 
> Rationale: the optional RDFS triples are there to enhance interoperability 
> with RDFS, yet RDFS idiom does not use such triples with blank subjects. 
> Moreover the blank nodes cannot be shared between files, so occur only in OWL 
> files.

I don't see any issue with leaving these in, as they would allow certain
constructs that might be considered to be bening.

> Again, what is the status of this suggestion?

Also not part of the proposal, I believe.

> BTW the newish text on the side condition looks good - thanks for sorting 
> this.
> 
> Jeremy


peter

Received on Wednesday, 19 March 2003 10:55:49 UTC