On Wed, Jun 02, 2010 at 07:36:18PM +0300, Antoine Isaac wrote:
> What I'd like to recall is that:
>
> - if something is the subject of a skos:inScheme statement it will be
> automatically classified as skos:Concept by virtue of SKOS semantics. And I
> think this fits quite well the idea of id.loc.gov providing a set of
> vocabularies, all of them could be exploited (among others) by
> SKOS-compatible tools.
>
> - if something is used in a predicate position for any statement, or is a
> sub-property of another property, then following RDFS semantics it is an
> instance of rdfs:Property. If one wants to keep the formal mappings between
> DC stuff and the MARC relators, I expect this will "contaminate" to a
> significant number of these MARC relators.
What this boils down to, I think, is the intended meaning of
the URIs. Since I have not seen any guidance materials, I
have no basis for assuming that these URIs would be recommended
for use for anything other than RDF properties.
> So unless LoC decides to drop one of the two features, the ambiguity will
> stay.
I wonder if we are assuming too much about the "typical"
uses of SKOS concepts -- e.g., as subject values. It is
not clear to me that an RDF property _cannot_ also be a SKOS
Concept (roughly, the concept of the relationship between two
resources). In the absence of incorrect usage examples I was
giving the maintainers benefit of the doubt and not assuming
that they were intended to be used as anything but predicates.
Ross has offered an example that clearly would be wrong,
but I'm not assuming that was the intent.
And if the intention were to use them only as properties,
it is still not clear to me that having the properties also
be SKOS concepts is _formally_ wrong.
It is perhaps interesting to note that when we assigned domains
and ranges to DCMI terms, we applied an informal rule _not_
to coin terms in the same namespace that differed only by
case, so given dct:language and dct:instructionalMethod,
instead of saying:
dct:language rdfs:range dct:Language (a hypothetical new class)
dct:instructionalMethod rdfs:range dct:InstructionalMethod (a hypothetical new class)
we coined dct:LinguisticSystem and dct:MethodOfInstruction:
dct:language rdfs:range dct:LinguisticSystem
dct:instructionalMethod rdfs:range dct:MethodOfInstruction
It was, however, always clear that a property was not a class (just
like nouns are not verbs).
> Now, you could say that LoC may decide to hide this issue under the carpet,
> and waits till inference engines or vicious minds (like us? ;-) ) spot the
> issue. But I'm not sure whether this is the perfect solution...
If the new relator URIs were used as anything but properties
it could get quite messy. I'm getting the picture that they
may indeed be intended for other uses, but I haven't heard
the views of the maintainers on that.
Tom
--
Tom Baker <[log in to unmask]>
|