Robert Robbins rrobbins at GDB.ORG
Fri May 6 17:37:11 CDT 1994

On Fri, 6 May 1994, Julian Humphries wrote:

> I have a different take on this entire problem.  First of all, we need to
> admit several things:
> The problem, as I see it, is that people are assuming that the barcodes (or
> specimen data) in question need to reflect current institutional/discipline
> based abbreviations.  Lets admit that that goal is impossible and move on
> to a solution.

Let me second Julian's point as strongly as possible.  Any attempt to
produce stable identifiers with historical or present semantic content
will fail.  Even if everybody were to agree, the existence of codens that
looked familiar but in fact contained changed values would be a source of
endless confusion.

>  If our primary purpose is to have labels on specimens that
> are uniquely identifiable to institution (and collection) all we need is a
> list of the players and a indentically long list of numbers.  Match 'em up
> and there you go.

It's not likely to be quite that simple, but close.  If, as Julian noted,
there may be duplicate accession numbers within different components of
the same institution, then there is a need for a hierarchy of arbitrary
numbers.  My phone number is 903 0041 within parts of Maryland and 301 903
0041 in North America.  Internationally, it gets a bit longer.  The idea
is the same.

> Leave it to computers to tell you what institution that
> bar code (data record) is associated with.  Nothing will prevent
> collections from adding on associated labels additional ways of notating
> their collection.

Exactly!  The trick is to delegate responsibility to the lowest
APPROPRIATE level.  Some central authority is responsible for translating
arbitrary identifiers of sites; then the identified site is responsible
for managing the next level (and on down) identifiers at that site.  When
I dial an international telephone number, the US phone system parses the
number enough to figure out what country and what phone district should
receive the call, then the resposibility for resolving the local number
devolves to that district.

> Note that this makes the job of creating unique identifiers much simpler
> (and cheaper).  All that is required is the assembly of lists of
> collections (of which there are lots of such lists) and the agreement that
> *somebody* is the source for the numbering scheme.  Whether this is ICZN or
> IUBS or IOPI or ASC doesn't really matter if we can agree that agency is
> the custodian.

Note that the agency must be singular.  If multiple organizations claim to
be the official keeper of the list, trouble arises.

>  I bet that someone with access to the net, a scanner and a
> good editor could create a 98% complete list in less than a month.  If I am
> right, money shouldn't be a factor.  Such a simple 1 to 1 correspondence
> could then easily be incorporated into any software that needed the data.

This estimate is probably pretty good.  This is not a massive project.
The problems associated with following it out to a logical conclusion are
sociological, not technical.  What is needed is the clear realiztion that
acronyms are always interpreted in a context.  When I lived in California
I was surrounded by people who knew as surely as they knew anything that
USC meant southern cal.  Here in the mid atlantic region, USC just as
unambiguously means University of South Carolina.

As Julian noted, there are things that a computer system does very well,
and that is keep track of lots of information without requiring that any
mnemonics be involved.  What a computer does extremely poorly is resolve
information in which context is a crucial clue.  Few biologists would
confuse the identical acronyms of a major botanical collection and that of
a zoological collection, if the subject matter were herbarium sheets.
Programing a computer to even that level of context resolution is not easy.

Also, bear in mind that using aribtrary identifiers does not rule out the
use of familiar acronyms in the software interface.  Type in USC, the
system recognizes that an ambiguity exists (there are two different USCs
in the system) and asks the users to clarify his/her intent.  Users could
even store their own preferences, if the system were appropriately
configured (when Jones is logged in, the default for USC is Southern Cal;
for Smith, the default is South Carolina, etc...)

The sine qua non of a working database is absolutely unambiguous, stable
identifiers.  Absolutely unambiguous means ABSOLUTELY UNAMBIGUOUS, it does
not mean mostly unambiguous, especially if you are a true scholar and
understand the context.

I recently read about a database designed for, and successfully used by,
practicing physicians that suddenly failed hard when it was used for a
different medical discipline.  The problem was that it had been designed
to avoid those nasty, dehumanizing numbers so it identified patients by a
combination of last-name, address, and birth-date.  This worked fine for
its original area of application, which I think was geriatrics, but when
it was first used in pediatrics it failed and failed hard.  Twins tend to
have the same last-name, address, and birth-date.  And, they tend to see
the same pediatrician.

More information about the Taxacom mailing list