proposed response to Jeff Pan's response of 23 June concerning datatypes

Thanks for your continued comments.

> Peter,
> 
> Thanks for your reply to the last call comments. This email contains
> comments to your reply and the online editor's draft of the S&AS document
> [1].
> 
> In section 2.1 of [1],
> 
> > The other built-in OWL Schema datatypes are problematic for OWL, as
>                               ^^^^^
> discussed in the RDF Semantics [RDF MT].
> 
> Do you mean the other built-in "XML" schema (instead of OWL schema)
> datatypes? 

Yes, sorry for the typo.  This is correct in the editor's draft at
http://www-db.research.bell-labs.com/user/pfps/owl/

> And it is not clear to me where in RDF MT can I locate the
> discussion about why the datatypes are problematic. It might be easier for
> the readers to follow if some more details (e.g. section number in the RDF
> MT document) are provided.

A direct pointer will be included as soon as the RDF MT document is updated
to include one.  For now, I have included the section number (5).

> > Because there is no standard way to go from a URI reference to an
> > XML Schema datatype in an XML Schema, there is no standard way to
> > use user-defined XML Schema in OWL.
> 
> I am not sure about this. Usually a URI reference of this form
> http://any.domainname/anyxsdfile.xsd#sss will be understood to denote a
> user-defined XML Schema datatype named sss.  Even though it is not a
> standard way in XML Schema, there is no harm adding that in OWL (implicitly
> require that the datatype sss be derived from one of the built-in OWL
> datatypes). Or do we want to support more datatypes than XML Schema
> datatypes, so we don't like the file extension xsd?

Unfortunately, this would be a non-standard access mechanism.  The OWL
specifications should not depend on this mechanism.  Also, consider what
would happen if the XSD file had both a top-level datatype and a top-level
attribute with this name.

> > Sections 2 and 3 have been modified to make rdf:XMLLiteral a built-in OWL
> > datatype and rdfs:Literal not a built-in OWL datatype.  Data ranges now
> > include rdfs:Literal as a special case.  VD is then the built-in OWL
> > datatypes plus rdfs:Literal.  This has no real effect on the semantics.
> > You can see the revised document at
> > http://www-db.research.bell-labs.com/user/pfps/owl/
> 
> Yep, it is clearer now.

Thanks.

> > > (3) In the definition of datatype theory, it is not clear that what kinds
> > > of datatypes can be in the set D. Does it contain only the built-in OWL
> > > datatypes, or also their derived datatypes? If it can only consist of
> > > built-in OWL datatypes, the datatype theory is quite limited and seems to
> > > me not enough in many cases.
> >
> > Datatype theories could include more than the built-in datatypes. However,
> > there would have to be a private understanding as to the meaning of these
> > datatypes, as OWL has no mechanisms for providing such meaning.
> 
> It is good to have more than the built-in datatypes. However, it is not
> clear to me how this "private understand" approach works.
> 
> [1] http://www-db.research.bell-labs.com/user/pfps/owl/semantics/

One possibility would be to use the above non-standard mechanism for
user-defined XML Schema datatypes.  Communities could have a private
understanding to treat URI references into XML Schema documents in this
manner.

> Regards,
> Jeff

Please respond, copying to public-webont-comments@w3.org whether this
satisfactorily addresses your comment.

Peter F. Patel-Schneider
Bell Labs Research
Lucent Technologies

Received on Monday, 23 June 2003 13:11:28 UTC