----------------------------------------------------------------------
2007-03-12 Subject: Range issue (Andy)

Feedback during the comment period about the DCAM and our proposed
assignment of ranges to properties in the DCTERMS namepsace has
highlighted the issue that it is problematic in RDF (most notably in
OWL DL) for properties to have a range that allows both literals and
non-literals as values.

In short, it is important in the context of RDF for us to be clear
about whether a property has a 'string' or a 'thing' as a value. 
(Note that the use of 'thing' here includes physical, digital and
conceptual things).

A case in point is dcterms:description, for which we've proposed a
range of rdf:Resource.  This range allows dcterms:description to be
used with a value that is a literal (a textual abstract for example)
and a non-literal (a thumbnail image for example).  Ditto for
dcterms:title.

In discussion between Andy, Mikael, Pete and Tom we agreed that it
would be sensible to revise our proposed assignments of ranges to make
it absolutely clear whether the value is a string or a thing.

Therefore, when we discuss this issue in Barcelona, we need to run
through each of the properties in the DCTERMS namespace (including the
new 15 properties) and decide:

1) whether the value is a thing or a string

2) whether we want to narrow the range of the property further by
using a subclass of 'thing' or 'string'

It may be there there are a few problematic cases (such as
dcterms:description above) where the wording of our definitions is
such that we really do allow both things and strings as values, in
which case we will need to adopt an alternative approach, such as
defining two properties, one with a string as the range and one with a
thing as the range.  I would hope that there will not be many such
cases!

In the context of this dicussion the rdf:Resource and owl:Thing classes
may be of use to us.

http://www.w3.org/2000/01/rdf-schema#Resource

http://www.w3.org/2002/07/owl#Thing

It is also worth noting that this discussion is premised on the notion
that OWL-DL compatability, i.e. clearly separating values as strings
from values as things, is a good idea.  However, recent discussion on
one of the semantic Web mailing lists seems to imply that this issue is
not clear cut, even within the SW community.

    So, this means that if one queries for the the
    skos:prefLabel of a concept, one either gets a literal
    or a resource with a label equal to this literal. This
    prevents the use of construction rules and keeps the SKOS
    vocabulary simple. The only extension to the current SKOS
    vocabulary would a a class skos:Term. [1]

[1] http://lists.w3.org/Archives/Public/public-swd-wg/2007Mar/0004

This is essentially an architectural issue.  We can usefully have a
discussion about semantics in Barcelona, and doing so will help clarify
our own understanding of how our properties should be used.  But there
may have to be a separate architectural discussion (on dc-architecture
or elsewhere) about whether DCMI should aim for OWL-DL compatability or
not.