The UPenn IE Tree (the stem)

[X99Lynx at aol.com]

In a message dated 9/17/1999 2:09:06 AM, alderson at netcom.com writes:

<<The innovations in question are not single-point phenomena, nor does any
branch
on the tree represent a single change:  Germanic, for example, is defined
...by having PIE *m rather than PIE *bh in the plural oblique cases, and PIE
*dh as a past-tense formant unlike any other in all of IE, and PIE *e > PG.
*i, PIE *o > PG. *a, PIE *a: > PG. *o:, PIE *e: > PG. *a:, etc. ad nauseam.>>

In a message dated 9/03/1999 12:39:20 AM, kurisuto at unagi.cis.upenn.edu writes:

<<-Reconstructed forms were not included in the data.>>

Well, I don't know what * in front of all those vowels mean, but obviously
either you and Mr. Crist are talking about two different trees.  Or sometime
recently PIE was crossed off the list of reconstructed languages.

You write:
<<What is absurd is your insistence on seeing this tree not for what it is,
but
for what you think it ought to have been.>>

I have some reason to think that I'm not the one who is seeing this tree for
what it is.  The example above just being one instance.  It might be far more
productve to take a critical look at what you are defending for a moment.

Do you find nothing extraordinary about writing <<Germanic, for example, is
defined ...by having PIE *m rather than PIE *bh in the plural oblique cases>>
when you have an unqualified quote that no reconstructed data was used?
Could it be that you misunderstand what the tree is about?

<<Yes, and you have been told so, though perhaps not in so many words:  *Both*
sides of each fork represent innovations, as has been pointed out to you
several times.>>

Once again you are not aware of what was said about the tree.  The case where
both sides innovate was given as a possibility, but not necessary.

In a message dated 9/14/1999 2:40:06 PM, kurisuto at unagi.cis.upenn.edu wrote,
specifically in response to my bringing up the issue:

<<It's true that you can imagine cases where you can't tell which is the
innovation; there could even be cases where both branches innovate, and
the original form is retained in neither branch.  But that's OK, since
what we're doing is showing relatedness on the basis of shared
characteristics.

This won't always work, as in the case where the innovation is an
irreversible phonological merger; the branch with the merger has
innovated. But for lexical characters, for example, there can certainly be
cases where you can't tell who innovated.>>

<<The program was not handed a list of Sanskrit, Greek, Latin, and Germanic
words and expected to derive Grimm's Law therefrom, but was rather given the
information that Grimm's Law encodes...>>

And if I'm not mistaken it was also given a date for Grimm's Law and an
assumption that of course that Law is relatively unique to Germanic languages.

Now here's how Sean Crist (to the best of his understanding, of course)
described the basic "units" being measured by the algorithm and reflected in
the Stammbaum:

>From a message dated 8/24/99 10:02:51 PM from kurisuto at unagi.cis.upenn.edu:

<< as far as I'm aware, Ringe et. al. don't make any
statements regarding mutual intelligibility.  It's a matter of innovations
being shared or not shared.  Presumably, two dialects which had recently
branched (i.e., undergone innovations which are not shared) would remain
mutually intelligible for awhile, ...>>

I can only take that to mean that Grimm's Law (as some form of quantitive raw
data?) would mark the branching off of Germanic.  Nothing more and nothing
less.  Whenever the Law is deemed to have taken effect in relative chronology
would be the point of branching off of Germanic, UNLESS it branched off
before Grimm Law - in which case that law would not even show up in the
analysis.

I don't have a copy of the methods statement - I'd love to see it - but the
point of branching is supposed to be a real event.  In which case all later
innovations should be irrelevant.  Please follow me closely on this one.
Innovations that occur after branching should not be included in the data, or
they will falsify the relative or absolute time of occurence of the actual
branching.  So the algorithm must either assume Grimm's Law happened at the
point of branching (and include that data).  Or date it afterwards and
exclude all GL data.  Again I'd love to see a formal statement of the
methodology here.  But I've gone over it with some very good research people
and I'm pretty sure I'm right.

My basic question is how does this "algorithm" and its tree improve our
understanding in that case?  What could the algorithm applied to the data
tell us about Grimm's law that it hasn't been told to tell us?  Is it capable
of testing whether that law is not in fact a late innovation but an early
archaism?  Of course not.  I recall a different value was given to a
particular phoneme or lexical element and that somehow brought Italo-Celtic
into a single group.  The entirely pertinent question is why any of the
values were given and whether they simply reflect the same guess work that
would go into them without the trappings of a computer application.

My problem has always been with what this apparent use of technology adds.
That's why I asked questions about what was assumed in the data.  What does
it add about Grimm's Law?  I suspect nothing much.  Because, e.g., it must
already assumes Grimm's law happened at a specific relative time.

When I mentioned Miguel Carrasquer Vidal's tree you said it was based on
intuition.  If you really look deep into the methodology applied here, I
think you'll see that it is based on nothing more than that - but wears the
cloak of being a statistical analysis.  I suspect MCV's tree is superior for
that simple reason.  It has the courage to place branchings in a real time
and place, based on both history and linguistics that can be understood and
evaluated.  The way this approach has been described, it does none of that.

And finally if YOU do understand the way this algorithm arrives at any kind
of independently derived chronology (as opposed to an assumed one), please
explain how adding dates of attestation in any way would have any statistical
effect on the chronology of most of the branchings shown in this tree.  I've
thought about this and tried to give the benefit of the doubt.  And it simply
does not make sense.

There are many good uses that a computer with proper use of statistical
probability can add to our understanding of the "divergence" found in IE
languages.  But I have good reason to suspect that this application does
nothing more than feed back the results that it was given in the first place.

And I have been reminded that I should say again that I have not seen any
detailed statement of the methodology used and should qualify everything I've
said accordingly.

Regards,
Steve Long