RE: comment: WD 10 May 2005

Hi all,

Thanks alot to John McClure for taking the time with this, very stimulating :)  Some specific comments:

1. Domain of labelling and documentation properties

The domain of the SKOS Core labelling properties (skos:prefLabel, skos:hiddenLabel, skos:altLabel, skos:prefSymbol, skos:altSymbol) has so far intentionally been left as rdf:Resource so that these properties may be used to describe anything at all.  

The domain of the SKOS Core documentation properties (skos:definition, skos:scopeNote, skos:example, skos:historyNote, skos:editorialNote ...) has also been left as rdf:Resource.  Again this is deliberate, as especially skos:definition and skos:example were envisaged as being useful to describe e.g. classes and properties in an ontology.  Commonly in current practice the rdfs:comment property is used to give documentation describing the meaning of e.g. a class or property in an ontology, but of course a class or property may have many 'comments', and rdfs:comment gets used in all sorts of ways - a more precise 'definition' property is undeniably useful.  

For individuals in an ontology the semantics aren't so obvious ... e.g. is it ok for a resource of type foaf:Person to have a 'definition' ... ?  

Also bear in mind that currently if you use any of the SKOS Core documentation properties you are  in OWL full, because the range of SKOS Core documentation properties is completely unspecified, and both literals and resources are allowed.  A hack around to keep you in OWL DL would be for you to declare locally that the documentation property you are using is either an object property or a datatype property, depending on how you are using it. 

2. SKOS and topic maps

I hope that as soon as the RDFTM task force [1] has published some RDF - TM mapping recommendations, we can go ahead and publish a natural mapping from the classes/properties of SKOS Core to some appropriate published topic map associations.  I spoke to Richard Light recently about this, who has worked on representing thesauri using topic maps.  Also Kal Ahmed lurks on this list, and I suspect he has already published a number of suitable associations from his 'topic map design patterns' work.  Hopefully Kal and Richard will be able to help out with that :)   

3. Can a resource be both a skos:Concept and an rdfs:Class or an rdf:Property?

This is something we've spent a lot of time discussing, but haven't got a strong position on yet.  There are some subtle issues, and a useful starting point is the short section in the guide on 'relationship to RDFS/OWL ontologies', see [2] - this requires careful reading.  

For recent discussion around proposed properties for mapping between RDFS/OWL ontologies and SKOS concept schemes see [3] and all subsequent in thread.  There's a lot of earlier stuff too, search for 'denotes' in the subject line of mails in the archive [4].

What I suggest we do is address this issue directly at the next (3rd) review in 2-3 months time (the second review is about to get underway).  I suggest that, for that review we ask someone from the Ontology Engineering Patterns (OEP) task force to be a reviewer, and we try to liase with OEP as much as possible.

Until then, the 'play it safe' position is: assume that skos:Concept is disjoint with rdfs:Class, owl:Class, and rdf:Property.

4. Range of skos:subject

The property skos:subject inherits its meaning from its super-property - dc:subject.  However, the range of the dc:subject property can (and does) take anything and everything.  The skos:subject property is primarily intended as a property with the same semantics as dc:subject, but with a more reliable range (i.e. you can expect to handle a resource of type skos:Concept).

As others have pointed out, you can of course use e.g. OWL classes in the range of skos:subject.  However, this will lead to the inference that a resource is both of type skos:Concept and owl:Class, and this leads you into the undecided territory of point 3 above.

To 'play it safe', if you want to state that a class is the subject of a document, use dc:subject, or your own refinement of dc:subject; only use skos:subject with resources of type skos:Concept, and assume that skos:Concept is disjoint with rdfs:Class, owl:Class and rdf:Property.

5. Broader/narrrower and class subsumption

I'm a bit confused with John's position here.  He states, 'I am unable to detect any
difference whatsoever between NT/BT and subclass/superclass relations'.  This directly contradicts a large body of consensus, and others have already disagreed with this position.  The OEP part-whole note provides a nice counter-example [5] in the section 'pattern 3'.

I'm going to leave it there for now, this email is long enough already, thanks everyone for chipping in.

Cheers,

Al.
 

[1] http://www.w3.org/2001/sw/BestPractices/RDFTM/
[2] http://www.w3.org/TR/2005/WD-swbp-skos-core-guide-20050510/#secmodellingrdf
[3] http://lists.w3.org/Archives/Public/public-esw-thes/2005Jun/0002.html
[4] http://www.w3.org/Search/Mail/Public/search?keywords=&hdr-1-name=subject&hdr-1-query=denotes&index-grp=Public__FULL&index-type=t&type-index=public-esw-thes
[5] http://www.cs.man.ac.uk/~rector/swbp/simple-part-whole/simple-part-whole-relations-v0-2.html

> -----Original Message-----
> From: public-esw-thes-request@w3.org
> [mailto:public-esw-thes-request@w3.org]On Behalf Of Mark van Assem
> Sent: 25 July 2005 10:55
> To: John McClure
> Cc: public-esw-thes@w3.org
> Subject: Re: comment: WD 10 May 2005
> 
> 
> 
> 
> Hi John,
> 
> Also thank you for your feedback, it is most instructive to see how 
> someone with another perspective looks at SKOS.
> 
> > to associate an rdfs:Class through skos:subject with a 
> resource. I am wondering,
> > if an instance of rdfs:Class has a <skos:prefLabel>, then 
> is the instance
> > implicitly typed as a skos:Concept? I am thinking of the 
> following triplets from
> 
> Nope, because <skos:prefLabel> has <rdfs:Resource> as domain 
> (inherited from <rdfs:label>). If the domain of prefLabel was 
> <skos:Concept> your statement is correct.
> 
> > Now, let me look for a moment at referencing rdfs:Class 
> instances during
> > construction of a ConceptSchema, the other side of the 
> coin. Are the following
> > triples recommended?
> > 
> > a:MyScheme	rdf:type		skos:ConceptScheme
> > a:Animal	skos:inScheme		a:MyScheme
> > a:Mammal	skos:inScheme 	a:MyScheme
> 
> Only if they are skos:Concepts.
> 
> > In any event, it seems to be a good idea to discuss in SKOS 
> Core Guide how to
> > reference instances of rdfs:Class as a resource subject and 
> as a node in a
> > concept schema.
> 
> I agree it would help for some people, but for others may be 
> confusing 
> and distracting from the main story. Am concerned that the Guide gets 
> "overloaded" with issues and becomes difficult to read.
> 
> Best,
> Mark.
> 
> -- 
>   Mark F.J. van Assem - Vrije Universiteit Amsterdam
>         mark@cs.vu.nl - http://www.cs.vu.nl/~mark
> 
>

Received on Monday, 25 July 2005 16:37:15 UTC