weiler at ERS.UNR.EDU
Wed Oct 2 10:01:15 CDT 1996
Tom DiBenedetto wrote
> well fine, but your initial comments were directed at the
here are my original comments in response to Alexey:
On Wed, 2 Oct 1996, Alexey V. Kuprijanov wrote:
> Linnaeus said: 'Remember: not the character constitutes genus
> but genus constitutes its character'. It should mean the following.
> You should first recognise in some way
This is all that matters here. Current taxonomic science occurs in a
variety of ways, and, in fact, a great deal of research has gone into
comparing the diversity of ways.
> ...units of diversity,
> then > find the characters to make these units recogniseable for other
> taxonomists, and then protect your units against the mad cladists
> using cladistic analysis. Such units will be immunised against
> varying of datasets. The distribution of characters will be highly
> consistent with the distribution of objects in units.
This is a caricature of modern taxonomic science, and ignores entirely
the century - old goal of having natural classifications. So far as
cladists being mad, they use an explicit algorithm that can be repeated
and understood (note the algorithm need not exist as software, but it is
an algorithm nevertheless. I submit that taxonomy by authority also
uses algorithms, the authors just don't bother to write them down and
test their generality. The human brain is like a computing machine
itself; examine its beahavior and we see that it needs algorithms to run
(language structure), that it performs heuristic approximations all the
time, and in doing so, is entry-order sensitive (that's in part why
jokes seem funny to it). Cladistics alone is not the full scope of
Systematic Biology, but we have learned a great deal from it.
> As to discordances:
> It is obvious that 1) the apo- plesiomorph polarity of character-
> states depends on the outgroup chosen by the author and 2) varying the
> character-sets you could get various results. The latter is particularly
> true for such cases when anybody having no idea about the structure of
> diversity in the group under study simply put in the computer a bag of
> characters and strikes the button.
But it is also true when people who think they do have an idea about
the structure of diversity of the group they study.
> Of course several such persons
> can obtain the very different results. I am not sure that comparison
> of these results will lead us to important conclusions.
This is also quite true when people use their authority to justify
a classification; hence the development of taxonomy as a science.
Discordance arising by any means implies that
1) the data being used (molecular or morphological) are either
noninformative or are in some way misleading, or
2) the authority has changed his/her mind in light of new data shedding
light on old trees, or has decided to weight (in their mind) different
sets of characters than they did before, or
3) there may have been more than one genealogical history
(e.g., reticulation, introgression, coalescence, etc.), or
4) To quote John Trueman, your alignment is wrong.
R. Zander wrote:
> Response to W. Wuster:
> 1. Decide whatever the Hennigian algorithm, a truly simplistic model of
> evolution, gives you must be the right answer.
> 2. Decide that supportive evidence must come from this source,
> 3. and only this source.
The Hennigian algorithm, maximum parsimony, clique analyses, compatibility
trees, and maximumlikelihood will give the wrong tree under some
situations. Discordance (as above) can arise because the same methods are
mislead by the data in different ways. Oddly enough, CONGRUENCE can
result when a method is mislead by two or more sources of data in a
similar manner (i.e., you can get the same wrong max pars tree for two
molecules, and both can be wrong). I'm hoping that we can move on from
relying on congruence alone, and test our data for symptoms of
uninformativeness (i.e., no or low signal) and for symptoms of misleading
information (e.g., long branches). generalizations about methods just
don't matter, unless they just plain don't work well at all. What matters
is what the data are likely to do to the algorithm, not vice versa.
We clearly need to ask critical questions of each data set, which, after
all, most of the person-hours goes ito collecting, be it morphological or
molecular. I'll have more to offer on this later.
More information about the Taxacom