GENBOX-L Archives

Archiver > GENBOX > 2003-03 > 1048197203


From: "Tom Morris" <>
Subject: [GENBOX] Place hierarchies - too powerful or not powerful enough?
Date: Thu, 20 Mar 2003 16:53:48 -0500
In-Reply-To: <003901c2e673$970cc820$6401a8c0@micronxp>


One of the things that appears most promising about Genbox is its rich
data model (leaving aside for the moment the question of how one would
ever transfer this information *out* of Genbox).

A key facility in this data model is the structured naming of places.
This is a powerful facility which organizes the comma separated pieces
of a place name into a hierarchical tree of place name elements.

On the surface this seems like a great idea, but there are a couple of
things that worry me about it. For one thing, it would appear to
encourage early binding of evidence to conclusions. The sequence of
characters representing the place name as it appeared in the document is
never stored anywhere, but is rather split into a linked hierarchy of
names. A later operation, far from the time and context that the
original hierarchy was created can rearrange this hierarchy losing all
trace of the original place name. This seems dangerous to me. Another
thing that concerns me is the complexity of the scheme, both as a source
of bugs during development/maintenance and as a source of user
confusion. I think the complexity is probably manageable, but I'd like
to see a straightforward way to record the place name as it originally
appeared.

On the other hand, I've been thinking for a while of a similar system
for dealing with some of the complexities of New England boundary
changes, but I think even more power/complexity might be needed. For
example, Dunstable, Massachusetts was established in 1673. The King
(yes America had kings back then) split it into Dunstable, New Hampshire
and Dunstable, Massachusetts in 1741 and various bits and pieces were
split of and renamed, some several times, over the years. Depending on
where in Dunstable you lived, you might still live in Dunstable or you
might have changed from Dunstable, MA to Dunstable, NH to Nottingham
West, NH to Hudson, NH -- all without moving.

Talking about "Nottingham West, present day Hudson" is fine because it
was never subdivided, just renamed. However, there's no way to
automatically say the same thing about "Dunstable, MA" or even
"Dunstable, NH" because you'd need to know where in the town the event
occurred.

Depending on my goals, I may the database to find all events that
occurred anywhere in the areas that we part of the original Dunstable,
MA or want to know something about the governmental structure that a
place was part of at a particular time (county, state).

Regions which grouped states together, either as part of the hierarchy
or separately, would be another useful addition and would support things
like meeting the NEHGS Register style of not using counties in place
names within New England, but using them every where else.

As an aside, if you're interested in what can be done by combining a
geographic database with powerful contextual natural language
capabilities, have a look at
http://www.metacarta.com/geotagger-technology.html. How'd you like to
be able to search the web for "Paris, Texas" and have it return not only
hits on that exact string, but also the hits for "Paris" that it could
tell were in Texas from context as well as "Paris, Lamar, Texas" and
"Paris, Lamar Co., TX" and, if you wished, hits on the neighboring
towns. Pretty cool!

So what do others think of Genbox's placename facility? Too powerful,
not powerful enough, or just right?

Tom


This thread: