Re: 2 questions

Barry Meyers-Rice (barry@as.arizona.edu)
Mon, 29 Nov 93 13:16:01 MST

>I prefer (like Peter G. Taylor, S. Jost Casper, Elza Fromm-Trinta, Ludwig
>Diels, John M. Macfarlane, and Benedictus Hubertus Danser) the
>nineteenth-century method of studying the plants rather than studying
>character tables. (expert- based system vs. expert-system base). The

I will rant and rave for just a moment. I get so irritated when, in the
field, I come across a genus for the first time and am trying to figure
out what species it is, and the key says

18 Leaves somewhat larger, etc etc.....

18 Leaves somewhat smaller, etc etc....

This irritates me to no end. It's like a cookbook saying, "cook until done".
The only people who can use this kind of key are those people who already
know what species the plant is! At least give me some measurements! Sure,
this info is often given in the species descriptions, but still this kind
of key irritates the willies out of me.

>However, any binary (observation yes/no) classification schema like
>the proposed decision tables (Barry) suffer that attributes either
>vary widely or that observation are difficult to obtain, imprecise
>or subjective. It also does take not into account that different species
>are sometimes so closely related that it is hard to make distinction
>among them.

Not being familiar with expert systems, etc., I still wonder if the problems
you listed about could be accounted for quite nicely by simply ordering
the criteria in a table appropriately. For example, in _Sarracenia_ there
are only three species which form phyllodia. No variation here. So put
that in column 1 of the table. Then as you move to more subtle or unreliable
characters, put those in later columns. As you proceed from most significant
bit to least significant bit, you produce a number with progressing
accuracy, that narrows the number of candidate species. So, if the table
would result in (for example)

00001=S.flava 00001=decimal 1
01000=S.leucophylla 01000=decimal 8
10000=S.minor 10000=decimal 16

So numerically, you could say S.leucophylla=decimal 8 +/- 4, and classify
accordingly.

I was at a recent symposium on star classification. In this field you
look at a stellar spectrum, and kind of get the gestalt of the spectrum.
(Come on, LkHa 225, speak to me!) If this kind of classification can be
done automatically with neural networks etc as various contributors
suggested, computer botany shouldn't be too far behind, at least in
theory.

>Yay! One of our aquatic Utrics (U. gibba? - any luck with those seeds I sent
>you last year, Barry?) is in flower again this year. Sure they're only small
>but at least they're not in the pink/purple color scheme... We've also got

Yeah, the seeds you sent me germinated quickly and turned out to be
_U.gibba_. Very easy to key, that one.

>The tuberous Drosera (auriculata) on the block are just starting to
>enter their summer dormancy period, so they're seeding (but the wind
>gets them before me... and interestingly the upright stems turn a deep
>red - some plants are centirely red whilst still in growth, and I'm
>going to try and cultivate some of these in the coming years...

I've noticed in my plants, from the seed you've sent, some are completely
red as you mentioned, while others branch wildly. I've separated some
branching plants and am curious to see if this trait breeds true.

Barry