On Mon, 14 Aug 2000, Anthony Finkelstein wrote:
> OK - I think my point has been misunderstood. A risk of writing a
> hasty response on a Friday afternoon. I am NOT arguing for the
> entirety of DC to be replaced by the use of UML. Rather, I believe
> that the relation element should be re-evaluated in the context of
> UML.
In a Web context I'd nod towards RDF rather than UML, but I agree with
your point either way. From an RDF angle, all 15 are relation
types, and the qualifiers for DC's Relation are just more of the same.
> PS If I was to be radical I would be suggesting that DC as a whole
> should be re-conceptualised with documents as active objects and
> metadata "elements" as methods that can be invoked by sending
> messages to those objects. But thats just my private eccentricity and
> you are entitled to ignore me.
(I'm also entitled to take your suggestion more seriously than it was
probably intended...)
What you sketch is not so far from the way the Web works right now,
HTTP being just an eccentric kind of distributed object system. I send
some URI-named object ('resource') a GET or POST message, passing in a
bunch of content-negotiation args, and I get back a mime-typed
rendering of the named resource. You might argue that the metadata-only
view of a resource is just one of its (possibly many) renderings.
Question then becomes, do we want one method for each and every
conceivable property of a Web object, as you sketch, or can we make do
with generic methods such as GET and POST (aka 'ASK' and 'TELL').
Since there'll be *way* more than the 15 DC properties, I'm inclined to
the latter. I want to be able to send a message to a Web object
saying: "tell me your DC::title and your XYZ::rating". If you map
document properties onto methods, that's two method calls by
necessity. If you take the ASK/TELL metaphor more seriously, you could
either have this as two calls to ask (passing in the property name,
ie. DC::title or XYZ::rating). Or you could have it as one call to ASK
passing in an SQLish expression such as:
"SELECT ?T, ?R FROM [?URI] WHERE DC::title(?URI,?T), XYZ::rating(?URI,?R) "
In which case, a generic method on the object wins over having 15
different per-property methods. Interesting to contrast this with the
Fedora model btw., eg: http://www.cs.cornell.edu/cdlrg/FEDORA.html
where they propose per-application methods, 'getDCField','getDCRecord'
etc.
Dan
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
