[sc34wg3] More comments on the Tau model

[Robert Barta]

On Thu, Aug 26, 2004 at 11:46:07AM +0200, Lars Marius Garshol wrote:
> | The other problem is that it doesn't appear that members are members
> | of /N/ either, which means that members cannot be reified, either.
>  
> * Robert Barta
> |
> | That's a feature, not a bug. I have no problem with others putting
> | the obscenity 'reifying members' into a model, but for me this is
> | clearly an abomination.
....
> ....., you can treat association roles as assertions in \tau,
> and the problem will go away. You'll get a much bulkier TMDM
> representation, which may make other things harder, but it's one
> possible route.

Yes, this is exactly the escape hatch we have built in. (I learned
having always an escape plan from James Bond, of course.) So it can
be emulated, no good reason to have it first-class in the model.

Yes, the mapping to TMDM will be bulkier. But, hey we already knew
that. :-)

> | But that is equivalent to addressing the assertion itself. And that
> | we have.
> 
> No. Reifying a role and reifying and association is not the same, and
> does not mean the same.
> 
> Let's take a parenthood association example. LMG is the child, Bj?rg
> is the mother, and Knut is the father. The association represents the
> whole fact that I was born of those two particular parents, whereas
> the role my father plays represents his involvement in the event,
> which is different from the event itself.

I know it has been there, but I would seriously doubt that you
can sensibly do that. Lets look at

 a = { <mother, Bj?rg>, <father, Knut>, <child, LMG> }

According to TMRM-thinking the roles /mother/, /father/, /child/ very
much characterize this assertion. Actually to a degree that you could
say that { mother, father, child } is the _TYPE_ of this assertion.
TMRM (at least the last time I read it) even goes so far as to say
"you cannot use the role /mother/, /father/ or /child/ anywhere else
in other assertions like

  a' = { <mother, Bj?rg>, <father, Knut> }

because it is the relative positioning of the things which makes up
the nature of this statement". And indeed, /mother/ and /father/ in
/a'/ have a subtly different meaning than in /a/. [ I have not built
in the latter (yet) into \tau, but I find this rather compelling and
precise thinking. ]

Now we want to make a statement about Knut's involvement in /a/, like
"everyone said he was a wonderful father". All of a sudden we would
have to tear <father, Knut> out of its context to make it "indirectly
reifyable". This contradicts - for me - the idea that /father/ cannot
exist independently from /mother/ and /child/.

This whole 'indirect reification' issue is pretty messy. You can do it
with whole associations, with topic characteristics and then also with
members. Why these? Why not do it with "substrings within basenames",
or reify "pairs of members" or "triples of basenames" or "I put this
scope and this type to the occurrence"? It is so completely arbitrary.
It is ok for a model like TMDM which is close to the application and
provides a fair bit of guidance for programmers. But in a fundamental
model, I don't know...

> The exercise I was referring to was representing TMDM in \tau, and you
> certainly haven't spelled out how to do that yet.

Pah, this is an exercise left to the reader. :-) Seriously, I do not
see this as a big problem, just work.

\rho