[SKOS] Re: SKOS Concept schemes and OWL Ontologies (related to ISSUE-36)

Hi Simon,

It's a good thing that you sent this mail, because it's about a pattern that has been not commented upon a lot outside of the WG.

If we consider your example, in the light what the Primer and Reference say in [1,2]
ONT1 defines two concept schemes CSA and CSB
ONT2 owl:imports ONT1
so ONT2 includes CSA, CSB (as well as CSC of course) that's your solution b)
Notice that this import does not make the concepts of CSA (and CSB) member of CSC: appropriate skos:inScheme statements have still to be asserted!

In this respect the comment you make is quite true:
- we have nothing to import concept schemes on an individual basis
- and (my making it explicit) we have nothing to make the concepts in the imported concept schemes members of the importing concept scheme on a group basis.
The current solution is purely about containment of datasets (nformation sources? I admit that I'm not a real expert in this part of the RDF world): one information source imports another one, but this does not create extra knowledge that really glues the elements described in these two sources.

One first way to obtain a better result in your example is to put the 2 concept schemes in different ontologies. The pattern in [1] would indeed expect users that want to exploit owl:import not to mess the concept schemes together.
Note indeed that this is actually more-or-less required by OWL management of ontology containment, as far as I understand it from [3]. In OWL there is no RDF link between the description of the owl:Ontology instance and the classes and properties that are assumed to belong to it. Everything relies on the containment at the information-source level. I say that my file describes an ontology, and my file happens to contain some class and property descriptions. Remember that actually ontologies do not contain classes and properties, they contain statements/axioms about these.

So now, why not make concept schemes also always instances of owl:Ontology? Well I guess for advanced users aware of the problems I've just refered to, that could make sense. After all, we really want to say that a concept scheme includes a number of concepts (by means of inScheme) and contains some statements about them (by means of OWL ontology-based containment)
But I guess many other people just don't care about these subtleties, and would still like to put several concept schemes in a same dataset.

Considering this, it could make sense to include a skos:imports, as you suggest. This would be a combination of owl:imports and "creation" of skos:inScheme statements between the concepts from the imported CS and the importing one.
But in the end I think that could be counter-productive, precisely because of the mixture between OWL containment and SKOS membership aspects. I'm afraid users would just interpret it in a natural SKOS way without being really aware of the OWL machinery that is behind and requires extra care with the datasets handled.
Or we would have to define our own SKOS containment mechanism, very similar to the OWL one by the names that are used, but quite different with respect to what the methods indeed do. Again I'm afraid it would not be really productive, with respect to the compatibility between SKOS and other Semantic Web stuff.

I was not really enthusiastic about it at the beginning, but now I tend to adhere to the existing solution. It has at least the positive aspect of not fooling the users into thinking it's easy to import concept schemes :-(

Best,

Antoine

PS: by the way why do you say we can interpret Concept Scheme as OWL classes?

[1] http://www.w3.org/TR/skos-primer/#secextension
[2] http://www.w3.org/TR/skos-reference/#L1170
[3] http://www.w3.org/TR/owl-ref/#Header

> De: public-esw-thes-request@w3.org de la part de sjupp@cs.man.ac.uk
> Date: mer. 05/03/2008 16:36
> À: public-esw-thes@w3.org
> Objet : SKOS Concept schemes and OWL Ontologies
>
>
> Hi all,
>
> I am after some clarity on using owl:imports with respect to concept
> schemes, hopefully my problem is clear from the examples. I should point
> out that I create my SKOS vocabularies as OWL ontologies using Protege.
>
> So consider an OWL ontology called ONT1, it contains the following 
> instances.
>
> ex1:csA rdf:type skos:ConceptScheme
>
> ex1:csB rdf:type skos:ConceptScheme
>
> I also have an Ontology called ONT2 that conatins:
>
> ex2:csC rdf:type skos:ConceptScheme
>
> ex2:csC owl:imports ex1:csA
>
> What should I expect to get imported into ONT2. I get two different 
> results depending on how I interpret skos:ConceptSchema (i.e as a 
> owl:Class or an owl:Ontology - If I am correct it is consistent in 
> SKOS to do both)
>
> 1) ONT2 imports only the concepts that are in concept scheme csA? 
> (which is what I want to happen)
>
> 2) ONT2 imports all of ONT1? (which is how it could be interpreted 
> because of the semantics of owl:imports)
>
> If owl:imports is to be used with concept schemes then why not always 
> treat concept schemes to be the same as owl:Ontologies?  Was the 
> possibility of a skos:imports ever discussed by the working group?
>
> Cheers,
> Simon
>
> Simon Jupp
> simon.j...@manchester.ac.uk
> http://www.cs.man.ac.uk/~sjupp/
>
>
>