[sc34wg3] revised draft Reference Model document N0298

[Steven R. Newcomb]

"H. Holger Rath" <holger.rath@empolis.com> writes:

> Are all node types of the same 'level' (I don't find
> a better word here)? I mean from the programming point
> of view it is obvious that X-, R-, A-, and P-nodes 
> are first class objects with an id. But I am uncertain
> if this is the case with the C-node? I assume C-node
> were introduced to avoid hyper-edges connecting three 
> nodes at once in the model (at least I can remember
> a discussion about this). It is probably just a minor
> point but worth to discuss.

I'm not sure what you mean by "level".  I think a node
doesn't have any "position", really, except within the
context of an assertion in which it participates.

Every C-node must serve as the C end of exactly one Cx,
one CR, and one AC.  But instead of saying it that way,
the dRM says that if a node serves as the C end of one of
these arcs, it must serve as the C end of exactly one
of them, and it must also serve as the C end of exactly
one of each of the other two kinds of C-ended arcs.

Two reasons that there are C nodes are:

(1) As you suggest, to avoid the need for a
    hyper-edges, so as to allow the model to be
    described in terms of two-ended arcs.  Some pretty
    important graph paradigms offer only two-ended
    arcs, and we thought the dRM should be
    representable in their terms.  (The earliest
    version of PMTM4 was rightly criticized, by
    SteveP and others, for being unrealistic about
    this.)

(2) To preemptively reify (as nodes) the playings of
    roles, so the graph will remain stable even though
    somebody decides to make an assertion about the
    playing of a role.

> And finally, a more academic question: 
> 
> Michel, SteveN  have you had HyTime or groves in mind and 
> how the RM could be mapped to them? 

We've been thinking about it, but we haven't tackled
the problem of creating a property set for the RM yet.
The highest priority is to get consensus on some kind
of Reference Model.  With that in hand, we can
translate it into RDF, a property set, and whatever
else.

One more comment.  The magic property of nodes whose
subjects are addressable, that make the addressable
subjects of such nodes direct properties of them, is
strongly reminiscent of the way that GROVEs work.
There's nothing really new or alarming about it; the
DOM, for example, also smells strongly of its roots in
the GROVE paradigm.

Oh, yes, one more thing: in a property set for the RM,
I suspect that the C, A, P, and R node types (that we
don't officially admit the existence of) will likely
become explicit node classes.  


SteveN