IE Tree more revealing ?

[ECOLING at aol.com]

Whatever one thinks of Steve Long's points,
it is to me COMPLETELY WITHOUT QUESTION
that the form of our family trees could be vastly improved,
so that we can more easily take them as a framework and
follow the relation between hypothesized innovations
and branchings.  (There seems to be some agreement about this.
Please see below.)

And so that we can more easily check on the reasoning built into any
computer classifications and tree structures.

In a message dated 9/15/99 6:25:49 AM, alderson at netcom.com writes:

>The entirety of the [IE] tree is based on reconstructions

In that case, Steve Long is of course correct in highlighting
any statements that no reconstructions were used,
if there were indeed such statements (I'm not checking that just now).

And more important in the long run,
Steve Long is correct in pointing out that the results may IN EFFECT
be not very much more than the formalization of results which were
in fact arrived at by the 200 years of careful study by other methods.
Steve at least once appeared to me to equate it with "intuitive" etc.,
which I think is too strong,
because those 200 years of study have yielded a lot of solid analysis,
but it might be somewhat more supportable to use terms like
"not formalized" or "not explicit", since the result of 200 years
of work is partly rejected dead ends as well as results now considered
standard doctrine.

To the extent that the results of the computer work might be
easily changed by WHICH particular innovations are included,
even assuming a correct analysis of reconstructions and innovations
is used as data, this also is cause for concern.
For this reason, the tiny modification in the data used which caused
Italo-Celtic to be classified differently is an illustration of the
possible (!) non-robustness of the method.

We should try hard to make certain we understand the relation
between conclusions and the data on which they are based.

As to Steve Long's other point, about the ambiguity of
what is the "stem" vs. a "branch", I think people are still talking
past each other to at least some degree.

If an Italo-Celtic branch is defined by certain innovations,
not given in the visual diagram usually presented for the IE tree,
perhaps we would do better to have a diagram which could only
be printed on an 11 x 17 inch sheet of paper (which would not fit
on an 8.5 x 11 inch sheet), and which did include major innovations
on each of the branches.

In such a diagram, we would perhaps see an Italo-Celtic branch
with two or three innovations in common, not many, so what the
two families had in common was still pretty close to PIE,
while the other branch (what Steve Long regards as the stem)
would ALSO have a few innovations in common, which separated
them from the Italo-Celtic.  Is it the case that the remaining IE
languages do share any common innovations which exclude
Italo-Celtic?  If not, then the remainder could legitimately
be regarded as still derived from the stem itself, rather than
forming just another branch different from the Italo-Celtic one?

The second possibility, that there are no innovations shared by
all IE languages other than Anatolian, Tocharian, and Italo-Celtic
(I am not checking against Ringe's work just now)
brings us certainly much closer to the view of an
unrooted tree, in which Italo-Celtic branched off together at one point,
and other groupings or individual languages branched off at other points.

Rich Alderson writes to Steve Long (and the list):

>You have misunderstood the word "branch" in this context; a better
description
>of what is examined in the tree is that each *fork* represents a pair of sets
>of innovations, with a group of dialects on one side of the fork showing one
>set and dialects on the other side showing the other set.  It may be the case
>that one set in each pair is null, but it is not necessary that it always be
>the "left" set (in the representation of the tree presented in this list).

I believe Steve Long is focusing on the cases in which one of the branches
in fact contains no innovations, in which case we need to think about an
un-rooted PORTION of the tree.  We cannot tell such cases apart with
the currently most common notation for such trees, no innovations marked.

Brian M. Scott   BMScott at stratos.net writes:

>As I understand the algorithm, a binary character that has one value in
>Anatolian and the other in everything else would contribute to the
>branching shown above irrespective of whether both values, the Anatolian
>value, or the other value is an innovation with respect to PIE.  The
>tree does not directly show innovations at all.

To that degree, it should be represented as an unrooted tree?

It may be that Steve Long's points can assist us in
improving our family-tree diagrams, marking on
their branches the innovations, perhaps especially mergers,
which lead us to draw particular historical conclusions.

John McLaughlin has some very useful comments about the biases
of the kinds of family trees we often see in biology, with the "trunk"
leading to man by some kind of predeterminism.  He says:

>The family bush drawn on the endplate
>of the American Heritage Dictionary
>is how a relationship chart should look.

I agree heartily.

In the case we are mostly discussing,
the UPenn tree for IE makes stronger claims than the AHD bush,
because it groups together several of the families which are
shown in the American Heritage Dictionary as descending independently
directly from the root.
I believe Steve Long's points apply
more to the UPenn IE tree than they would to the bush in AHD.
Some of his points might
in effect be partly answered by the AHD version,
since the AHD version would grant more equal status to the descendants,
and could more easily accomodate
differences as to what was original vs. innovated
(I am NOT arguing any substance in that line just now).

I think listening to all of this discussion,
my own tentative conclusion is that the FORM (purely the form)
of the UPenn IE tree, like most family trees without indications
of innovations on ALL of the branches, leaves us with a highly
unnecessary gap between data and results, a gap in which
there may in fact be lots of structure which professionals
may be fully aware of, but which is not made available to anyone
in as educationally effective manner as is possible.

I think John McLaughlin's comments are on the right track here.

So what improvements would our list members want to make?

I strongly believe that searching for the most legitimate part of a
critique, and responding productively to that, is a much better way
of proceeding than trying to defend some abstract doctrine and
discredit the portions of a critique that may be wrong.
First, because it directs all of our limited amounts of attention into
more productive channels, and second, because it strengthens the
standard and the educational presentation of the standard,
to make those improvements which we can legitimately see.
Therefore, John McLaughlin's critique, and some elements of Steve
Long's critique, should be made, I believe.  We should make it clear
why family trees which do not distinguish between innovating and
non-innovating branches are rather dangerous both for our own
understandings and for the education of new entrants into the field.

Best wishes,
Lloyd Anderson