Re: [topicmaps-comment] Re: RDF/Topic Maps: late/lazy reificationvs.early/preemptive reification

["Thomas B. Passin" <tpassin@home.com>]

[Piotr Kaminski]

>
> 2.  There is redundancy:  you're effectively recording the membership of
> the club in two places.  Once in the original association, and again in
> the sponsorship associations you're creating.
>

This is a very interesting matter.  Maybe you are duplicating it and maybe
you are not.  It depends a lot on exactly what these roles are supposed to
mean.  Also, don't forget, we had a ground rule of sorts- not to mess with
the original association too much.  Now about the sponsorship association -
you are making a statement about the ***membership*** of the person, not
about the club per se.  So the club actually does not appear twice in the
same way.  Instead, it plays a different role in the sponsorship association
than it did in the membership association.  In fact, you could have a
sponsorship that did not result in a membership - because the person was
rejected by the club.  So does the "sponsorship" association include the
success of the sponsored attempt to join, or not?  This could make a
difference in the degree of redundancy, too.

I don't think that having a separate association for sponsorship in itself
indicates a loss of normalization (in the relational database sense).  In a
relational system, one key test for correct normalization is whether any
items in a row (which represents an association between the items) actually
depend on each other (not counting the primary key, on which they must all
depend).  Whether they do in this case depends on the exact meaning of
"sponsorship", as we've just seen.

On the other hand, if you did not have to start with an existing
association, you might model this example with a single association (for
each person who is a member) called "membership".  Membership could have an
organization role, a person role, and a sponsor role.  This approach would
eliminate the original assocation altogether.

It's good that topic maps have the flexibility to model the situation either
way.  The real issue, at least for me, is establishing the intended meaning
of the relationships, especially if we are interested in machine processing
as well as human consumption.  This is an unsolved problem, so far as I am
concerned.  RDF Schema tries to address it, DAML+OIL tries to do it with
logic and ontologies, but I think there is a long way to go.

In Conceptual Graphs, you can define a relationship by means of a subgraph
that has formal parameters (lambda expressions or their graphical
equivalent).  At present, topic maps don't have that capability, but maybe
it will come out of activities to define templates.  Once you can do that,
you have a lot of freedom to represent a situation with relationships or
with concepts as the focus, as seems to fit the case or your particular view
of it.

Cheers,

Tom P