Re: Axiom annotations

On 19 Apr 2007, at 17:51, Paul Gearon wrote:

> Hi Bijan,
>
> On 4/18/07, Bijan Parsia <bparsia@cs.man.ac.uk> wrote:
> > Just a few days ago, we had an analog discussion for rdf:List and
> > friends here in the "Restrictions on Bags and Seqs content" thread,
> > remember?
>
> I've not finished my message for that, but short answer from me is
> that I think, even for RDF, Bags and Seqs are known to be broken, and
> rdf:List is not a good idea. Historically, I believe rdf:List was
> introduced in order to encode DAML+OIL syntax in the first place.
>
> I have my own ideas about problems with Containers, but do you have  
> some references for what you mean by "known" to be broken please?

I was writing an exposition of this but I can't find it at the  
moment, so I'll toss out a short answer:

The proliferation of properties in Containers is noxious on several  
levels and achieves no good. It has broken systems that presume a  
sane number of properties and can overwhelm others, if not strictly  
breaking them.

(Found this out working with the RDF archives of #rdfig.)

The semantics have always been nonexistent and to get sensible  
permutations of e.g., bags probably would require comprehension  
principles such as those that appear in OWL Full.

Finally, if we stayed first order, they would have the usual  
unintended models (heck, they can't be closed as is; one of the  
reasons lists came about in the first place; probably the most  
standard complaint about them).

Finally, they are redundant. Given that there is no built in theory  
that enforces their intuitive semantic, you can get much the same  
effect by rolling your own without the baggage and a good deal more  
control.

For example, I'd rather model sequences as linked elements with prior/ 
next properties. Then I at least have a *shot*.

>   Similarly, why is rdf:List not a good idea?

Lists similarly are suggestive, but lack semantics and you can't give  
the intuitive semantics in owl (since you can't rule out models with  
rogue elements and branches). People get all excited by them because  
they are used to modeling with data structures so lists seem  
friendly, but they 1) don't give any *real* joy and 2) they are  
incredibly hard to work with in OWL anyway. So people put out a lot  
of effort and don't get anything useful.

Then there's the fact that they are used in the syntax of OWL which  
raises its own issues.

So awkward, underspecified, overload, and poorly supported.

> Of course, the question then becomes, what do you think *should* be  
> used?

It depends on what you are trying to model, obviously, and to what end.

For example, in Lisp systems and Prolog systems, lists are sometimes  
uses to model XML structures. I would not use rdf:List for this (and  
I would be very careful about using OWL at all for this).

>   I have a couple of practical approaches for specific  
> circumstances, but the semantics are lacking.

Cheers,
Bijan.

Received on Thursday, 19 April 2007 17:21:22 UTC