Re: question on domain

David Powell wrote:
> The domain of property doesn't restrict the property to that set of
> classes. Instead, the domain is a set of classes that the resource is
> implied to be an instance of.

I do not fully agree with that opposition...
Assume

  :p rdfs:domain :C .
  :a :p :b .

By forcing inference engines to infer that :a is an instance of :C, the
domain assertion *does* restrict, in a way, the property.

The opposition between inference and validation, as stated also by
Richard Newman below, is in the eye of the beholder: it depends how you
are using it.

Richard Newman wrote:
> Matthew,
>   Yes, you are. This is a very common misconception about RDFS.
>
>   RDF/RDFS is a property-centric model. Domains and ranges are used for
> *inference*, not validation. E.g., saying that
>
>      color:blue foaf:mbox <mailto:blue@example.com>
>
>   is not an error — it means that an inference engine can look at that
> triple, and the triple from the FOAF ontology:
>
>     foaf:mbox rdfs:domain foaf:Agent
>
>   and deduce that
>
>     color:blue rdf:type foaf:Person
>
>   *not* that there is a conflict, because blue hasn't been defined as a
> Person.

Again, the statement about color:blue is still correct, *not* IMO
because rdfs:domain would not be about validation, but because of the
open-world assumption, which leaves the possibility for color:blue to be
a Person after all, unless the opposite is *explictly* stated.

If you state that color:blue is not a person, and that it has an mbox,
you will get an inconsistency, due to the domain of foaf:mbox. This is
inference *and* validation.

  pa

Received on Friday, 21 April 2006 10:22:32 UTC