[topicmapmail] PSIs for years ?

[W. Eliot Kimber]

Jan Algermissen wrote:
> 
> Does anyone know an authority/registry/etc..
> that I could point at in order to establish
> subject identity for year-topics ?

This was discussed some time back (about a year ago, I think--I remember
discussing it at the Knowledge Technologies conference last march).

My suggestion is to use HyTime coordinate spaces as the targets of the
identity pointers. In HyTime, you can define a coordinate space that is
calibrated to some real measurement system, i.e., real time. You can
then use a finite-coordinate-space location address element to point to
any point or set of ranges of points on that space. These FCSLocs can
then be the targets of idenity pointers.

So, for example, you can define an FCS that represents some span of real
time at some convenient granularity (e.g., days) and then create a set
of FCSloc elements that represent either particular years as spans of
days or specific days (e.g., Jan 1 2002), depending on what you need.
See the archives from some examples I wrote up at the time. Any
arbitrary time span can be defined very simply (an FCSloc is nothing
more than a pointer to the FCS itself, represented by a single XML
element, and two numbers, start and length, representing the time span.
In addition HyTime provides a way to define any FCS addressing syntax
you want, so you could easily define a syntax that is more convenient
(e.g., pairs of Julian dates or something), as long as your implementing
software knows how to interpret the syntax in terms of the underlying
FCS abstraction).

A topic map engine that also understood the FCS stuff could, for
example, perform range comparisons to determine the time relationships
among two topics to see, for example, if they share any dates in common,
are completely disjoint, etc. You could set a date range as a "scope"
and then filter out any topics that didn't have that date as part of
their identity. You could have rules for precision of matching (e.g.,
any overlap, full overlap, within X units, etc.).

While the overall FCS facility in HyTime is somewhat complex (because of
its level of abstraction), the application of that facility to this
particular problem is pretty straightforward and shouldn't pose any
great implementation difficulty as long as you're not trying to
implement a completely general FCS engine.

Note that the same general approach could be expanded to define spatial
relationships among topics or even to model abstract n-dimensional data
spaces that topics are then bound to either for identity or for
occurrences. Not sure it would be the best way to do this, but it would
be possible.

Cheers,

Eliot Kimber
ISOGEN International, LLC