Re: WSDL WG request for adding multiple version extensibility into Schema 1.1

On Feb 14, 2004, at 10:25 PM, Mark Baker wrote:

> On Sat, Feb 14, 2004 at 07:16:05PM -0500, Bijan Parsia wrote:
>>>   It gives you
>>> exactly the kind of extensibility you seem to require.
>>
>> Actually no. He requires "ignore unknowns" extensibilty *with*
>> validation.
>
> Right.  Granted, OWL can't express all the constraints of XML Schema,
> but many of them it can.

Technically, I believe that it can express almost none of them. It 
expresses *analogous* constraints, but one must be *very* careful not 
to be misled by the surface similarity. If nothing else, *what* they 
are constraining are radically different.

>   Actually, thinking more about it now, the
> "name" example is, I think, an example of the kind of constraint that
> OWL/RDFS can't express (limits on container structure).

No, the problem is that an OWL class definition can't (in most cases) 
mandate the presence of information in a knowledge base (kb) or 
document. But for data validation, that's exactly what you want to do.

For some discussion of the problmes see this thread:
	http://lists.w3.org/Archives/Public/public-sws-ig/2004Feb/0037.html
where I discuss the "Decker problem". I should be posting another more 
overviewy message soon. Perhaps I should put up a web page on it :)

Oh, more here:
	http://lists.w3.org/Archives/Public/public-sws-ig/2004Jan/0084.html

> If the example
> were Person/name/age then you could use OWL's cardinality mechanism.

?

I don't think so.

Suppose you had the class Name. This is a perfectly find OWL class. Add 
the restrictions that Name has exactly one firstName. Now consider 
these two RDF datasets:

	_:x rdf:type Name.

and
	_:x rdf:type Name;
	      firstName "Bijan".

Both are consistent with the class definition. OWL (and RDF) are 
*designed* to work with incomplete information. In the first kb (with 
the class definition) I know that the name *must* have a firstName (in 
*the world*), but not what it is.

> Ok, so maybe the fit isn't perfect, but it should work in many (most?)
> cases.

Hardly any, actually. And the pain would be high. You really need a 
different language of constraints on RDF, if this is what you want. But 
if that's true, you might as well go with a different language of 
constraints on XML. Actually, I believe schematron does the job.

>> If you try to validate a specific profile of RDF/XML, you
>> could have similar problems.
>
> What's a "profile" of RDF/XML, and what kind of validation do you mean?

For example, you might want to constrain your documents to socalled 
"nTriples in RDF/XML" (I.e., one top level rdf:Description per triple, 
no abbreviations except rdf:resource for the object) where you only 
have properties from the dublin core.

So,

<rdf:Description rdf:about="#article1">
	<dc:creator>Bijan Parsia</dc:creator>
</rdf:Description>
<rdf:Description rdf:about="#article1">
	<dc:title>A simple way to solve all problems</dc:title>
</rdf:Description>
<rdf:Description rdf:about="#article1">
	<dc:relation rdf:resource="#article2"/>
</rdf:Description>

You could fairly easily write a schema that rejected any document that 
used non dc properties in this format. So you could validate that the 
graph contained only assertions of dc properties, and even a certain 
number of ground ones.

Cheers,
Bijan Parsia.

Received on Sunday, 15 February 2004 09:04:08 UTC