[sc34wg3] Illustrating SIDPs

[Robert Barta]

On Tue, May 04, 2004 at 02:30:03PM -0400, Patrick Durusau wrote:
> "Oh, I can distinguish between people who have the same name by means
> of their spouses.  If they have different spouses, even if they have
> the same name, they must be different persons."
...
>     * Property name:              "personID"
> 
>     * Value type:                 complex:
>                                      "name"   : string
>                                      "spouse" : topic
> 
>     * SIDP or OP?:                SIDP

> There is only one SIDP per topic (per TMA). (note that personName was
> replaced by personalID)
> 
> SIDPs (and, for that matter, OPs) can be arbitrarily complex.

Patrick,

I agree with (almost) everything you say here, but I think my
conclusions are quite different from yours.

As I read the TMRM document(s), I understand the objective to provide
a framework to define these TMAs. What I am missing, though, is a
notation to formalize this properly.

I think that the whole concept of an 'identifier' is rather misleading
(not only in TM universe, I mean in general): as long as you have such
an identifier at hand for your topic maps everything is neat and nice.
In general, though, the concept of identity is an inferred one, coming
- as you also say - from a combination of values, or - more
abstractedly - being the result of an application-specific rule.

If we assume a rule like "two persons should be regarded the same if
they have the same email address", then this rule is actually nothing
else as a function which computes an "identity value" out of the email
address and the fact whether the object is of class 'person' or not.
If two (or more) objects have then same function result, then they are
identical _for this very application_. Technically speaking, the
objects are all in the same equivalence class as induced by the
function.

[ This is how deductive databases work in general: They have explicit
knowledge (facts) together with identity and they have implicit
knowledge as provided by the inference rules. ]

My conclusion now for TMs is that

  - such rules are simply part of the ontology which characterizes an
    application. One (or more) of such identity-inducing rules might
    exist in a single application.

  - it would be an overkill to put such a formalism into TMRM to
    express this.  Why not burden TMCL with the ability to express
    such rules?

  - TMRM could then be simplified in that it merely captures the "nature
    of TMs", making the association concept explicity as it already does.

This is roughly my argumentation line I had in Amsterdam.

\rho