[topicmapmail] SIDPs and Identification Process [Was : Identities and names ... ]

[Bernard Vatant]

Thanks Daniel for the invitation to re-read our classics :)

I would tend to think that your distinction between (physical or abstract)
things, facts and processes, although a pertinent one, is somehow
orthogonal to the identification issue. They all are "subjects of
conversation", right? Anyway I will not wander on this debate any more here
...

We have agreed so far (at least a few people in this thread, that's a good
start for an universal agreement) that names are used as
(subject)identifiers, given a proper context and protocol, and that there
is no essential, absolute distinction between a name and an identifier.

But wait, there's more ... since we have opened this Pandore's box, I would
like to expand the reflection (before being striken by the week-end), to
the notion of SIDP introduced in the TMM (was PM4TM, RM4TM ...). I hope
Steve Newcomb will correct me here if I got it wrong.

Some identification process do not rely on the
name-as-identifier-in-a-context paradigm. Many other asserted properties
(or characteristics) can be used as identifiers.

The property "Queen or King of UK in 2000" identifies a single "subject"
(no offence to Her Royal Highness, hopefully), as surely as
"Elizabeth.II.DEI.GRA.FID.DEF" on the ?2 coin of the same date, although
it's not stricto sensu a name. In a Topic Map, it would probably be
asserted through an association linking a person playing the role of
"sovereign" and a country playing the role of "kingdom", in the time scope
"Year 200". Nevertheless, it would be good if an application could know
this is an identifying property, in a context of previous knowledge, namely
that at some point of time, a kingdom is ruled by a unique sovereign. That
could be expressed in OWL weird terminology through an
"inverseFunctionalProperty" (although taking time into account would be
perhaps an issue).

So I guess this is an example of what Steve Newcomnb is now about with
"identification of subjects through assertions". In fact, a name is just
one among so many properties that could be used for identification purpose.

Note that XTM distinction between <subjectIndicatorRef> and <baseName>, and
the long story of the TNC rise and fall, look now  quite short-sighted in
this perspective. So, if we want Topic Maps to keep on leading the path on
identification issue, we have to expand what has been said above about
names and identifiers, with related notions of context and protocol of
identification, to any possible identifying property
(inverseFunctionalProperty in OWL terminology) or characteristic (SIDP in
TM terminology), along the following lines:

- A subject is assigned characteristics or properties - including names -
through assertions.
- An identifying property (SIDP) is an assertion about a subject that, in a
given context, following a given protocol, makes for subject identification
(that is, ability to declare that two subjects are identical or not). SIDPs
can be assignment of names, or other assertions.
- A semantic application or language has to clearly declare what are the
properties it uses as SIDPs.
- Semantic applications or languages are interoperable if they are able to
match their respective identifying properties, so that two subjects
identical for one of them are also considered identical by the other(s).

Certainly vocabulary and details of all the above have to be clarified, but
I hope the principle is clear ...

Have a great week-end

Bernard

Bernard Vatant
Senior Consultant
Knowledge Engineering
Mondeca - www.mondeca.com
bernard.vatant@mondeca.com