Re: Weekly summary on predefined classes :-)

From: Peter F. Patel-Schneider (
Date: 02/07/01

Thanks, Frank, for an excellent summary.

From: Frank van Harmelen <>
Subject: Weekly summary on predefined classes :-)
Date: Wed, 07 Feb 2001 09:21:51 +0100

> Below is my weekly summary of the "datatypes debate" :-)
> (Purpose of writing this is to check if I still understand the issues, and if so, move them forward a bit again).
> [a] we speak about "predefined classes" and "user-defined classes"

I would prefer something like datatypes versus classes, although this is
not exactly correct.  Better names would be appreciated.  ``Predefined
classes'' is wrong, particularly if we allow XML-Schema datatype type

> [b] There seems to be agreement on a layered approach, as described in my earlier email (see quote [1] below)
> The main remaining complaints about the proposal from Ian and Peter (at
> seems to be the
> following two points: 
> [c] the syntax of how to actually include in DAML references to XML-Schema datatypes and their instances is unclear
> (apparently the examples in break RDF)
> Presumably, [c] should be fixable (I can't see a really deep issue here?)

To make the change that we have implicitly proposed in our proposal clear,
here is a modification to the RDF syntax that allows for datatypes in RDF
property elements.  As well as incorporating this into DAML+OIL, I strongly
believe that this should also be incorporated into base RDF.

I actually see this as the place of most disagreement.  The question is
whether we should augment RDF with a mechanism for including ``typed
literals'' or whether we should have some mechanism for incorporating
datatyping information in our language.  

I am strongly in favour of the former; I think that the latter is
completely misguided for a number of reasons, not the least of which is
that RDF(S) will almost certainly be extended to include typed literals at
some point in the future.

Proposal to allow datatypes into RDF:

Instead of having the following syntax rules

[6] propertyElt ::= '<' propName '>' value '</' propName '>'
		  | '<' propName resourceAttr '/>'
[8] value ::= description | string

we have 

[6] propertyElt ::= '<' propName '>' description '</' propName '>'
		  | '<' propName '>' string '</' propName '>'
		  | '<' propName dataType '>' string '</' propName '>'
		  | '<' propName resourceAttr '/>'

[x] dataType ::= 'xsi:type="' URI-reference '"'

> [d] the proposal relies on software that must be able to process in principle all of XML-Schema, which is unlikely to appear soon.
> A proposal to fix [d] is to redo the "easy part" of the XML-Schema datatypes (which seem to be the only parts we need) in DAML+OIL vocabulary. 
> Advantage: no reliance on XML Schema software, 
> disadvantage: duplication of effort between two standards. 

Instead of redoing the ``easy part'' of XML-Schema datatypes, I would much
prefer to define a subset of XML-Schema datatypes.  That is, instead of
having DAML+OIL+datatype names for the appropriate XML-Schema names, just
use the XML-Schema names directly.

I am rather unhappy that this seems to be required.  I think that it would
be better to just include XML-Schema datatypes in its full glory.  I don't
see that this requires us to include all of XML-Schema as I don't see where
XML-Schema datatypes uses anything in XML-Schema structures.  However, I do
realize that XML-Schema datatypes is extremely large, even without
XML-Schema structures.

> Frank.


This archive was generated by hypermail 2.1.4 : 04/02/02 EST