Hi Jeff,
On Sun, Mar 04, 2012 at 11:55:49PM -0500, Jeff Young wrote:
> > That's precisely the question I was getting at earlier in this thread (on
> > Tuesday). Rephrasing...:
> >
> > The (current) DCAM Vocabulary Model [1] is grounded in RDFS. It has:
> >
> > -- Properties,
>
> Whereas RDFS allows properties to be either literals or entities, OWL
> makes a clear distinction:
>
> owl:DatatypeProperty (the range must be a literal)
> owl:ObjectProperty (the range must be an entity)
>
> I doubt that I would have ever grok'd RDF without this clear
> distinction.
The problem with this is that there is some principled resistance to the idea
that one _must_ neatly distinguish properties with literal ranges from
properties with object ranges. We see this is the preference by some projects
to continue using -- or even to revert to using -- the fifteen "free-range"
/elements/1.1/ properties in preference to the fifteen "ranged" /terms/
properties [1].
In the wild, I am told, well-meant range assignments often simply get ignored,
whether out of simple ignorance, willful pragmatism, or out of a principled
stance that the DatatypeProperty/ObjectProperty distinction forces "semantic
overcommitment". If were to happen on a wide scale (maybe someone here can
point to some numbers?), then it is consumers of Linked Data who would need to
adjust.
The Schema.org initiative, in my understanding, follows an (implicit?) policy
of defining ranges with precision, but in the expectation that those ranges
may, in practice, be ignored. If I have understood correctly, this seems wise.
Tom
[1] http://wiki.dublincore.org/index.php/FAQ/DC_and_DCTERMS_Namespaces
--
Tom Baker <[log in to unmask]>
|