- From: Alan Ruttenberg <alanruttenberg@gmail.com>
- Date: Tue, 9 Jan 2007 16:45:35 -0500
- To: Michael Schneider <m_schnei@gmx.de>
- Cc: bernard.vatant@mondeca.com, der@hplb.hpl.hp.com, semantic-web@w3.org, timbl@w3.org, marc@geonames.org, public-owl-dev@w3.org
On Jan 9, 2007, at 4:21 PM, Michael Schneider wrote: > What I have found in OWL1.1 is this "punning" mechanism: When you > have some class C, you can then also give this same name, "C", to > some instance. So one could perhaps come to the conclusion, that > one can just attach the datatype property to the /instance/ C, > which is of course formally allowed in OWL/DL. But, as I understand > punning correctly, there is no semantical connection between Class > (C) and Individual(C) (see [1, §3.3]) - it just seems to be meant > as a form of name "overloading". And this would /not/ solve the > above problem! (It could instead easily lead to some hard to find > errors: One might intend to add some property to a class, but, in > fact, the property is added to an instance, which happens to have > the same name.) You are correct. This is true of the punning mechanism in general. It was believed that this mechanism was possible to implement in the short run, and seemed better than the alternative of having to use annotation properties. That said, there will likely be a working group on OWL 1.1 and hopefully feedback from people using the feature will indicated whether, on the whole, its benefits outweigh the problems generated by it. > >> In the spec its use is (halfway) constrained to be a datatype >> property because when specified via the Label() construct its >> value must be a "constant". > > Another, more exotic use of rdfs:label, which comes to my mind, > would be to label a resource with some /graphical object/ instead > of a text label. Not, that I would argue for doing so, but people > will perhaps come to this idea, and perhaps some people already did > so. In this case, the label would better be an URL to the graphical > resource instead of a text string. And then, a datatype property > would not be the right thing, again. Agreed. It would be useful to see how prevalent this usage is. There are a couple of alternatives that come to mind. 1) Allow subproperty statements relating annotation properties to each other, but forbid that they participate in restrictions. I think this would mean that there would no need to have the reasoner involved with these properties, but queries for property values would need to take the subproperty statements into account. Some DL person will surely explain to me why this can't work.... 2) Allow rdfs:label to be declared either an ObjectProperty or a DatatypeProperty and have this declaration trump any built in AnnotationProperty declaration. The downside of this is that an owl:imported ontology could declare rdfs:label to be the other type and render the ontology inconsistent. -Alan
Received on Tuesday, 9 January 2007 21:46:32 UTC