Updates regarding UPenn tree

Sean Crist kurisuto at unagi.cis.upenn.edu
Tue Sep 21 00:48:46 UTC 1999


I've finally had a chance to talk to Don Ringe a few times recently, and
I've got some answers to questions which previously came up on this list,
and some other new information.

Regarding the grouping of Italo-Celtic, here's the scoop.  Going back to
early versions of the work, the optative has always been included as a
character.  Most branches of Indo-European have a reflex of *-oy-; Italic
and Celtic have *-a:-, and others (Hittite, Armenian, Albanian, and
arguably Tocharian) don't have it, either because they never had it or
because they lost it.  If you go back to the tree I gave, you'll see that
this situation _allows_ an Italo-Celtic grouping, but does not _force_ it.
(Either *-a:- is an Italo-Celtic innovation, or else the ancestor of
Greco-Armenian-Balto-Slavic-Germanic-Indo-Iranian used to have *-a:- too
and then replaced it with *-oy-.  If the latter is true, then we need
not group Italic and Celtic together).

Later, the team added _two_ new characters which forced the Italo-Celtic
sub-branch (not just one character as I previously incorrectly reported).
One was the *p..kw > *kw..kw character (a sound change shared by Italic
and Celtic), and the other is the morphological construction underlying
Latin -tio (Don says that a lookalike in Armenian can be shown to be an
independent innovation, but I don't know the details).  So much for
Italo-Celtic.

A bit of very recent news regarding the tree is that in the most recent
runs, the position of Germanic has turned out to be even more
indeterminate that the team had originally thought.  As I noted earlier,
the team had claimed that Germanic patterns with Italo-Celtic with regard
to lexical characters, but with Balto-Slavic with regard to morphological,
etc. characters.  What the team had already found is that if they take
Germanic out and run the algorithm again, they robustly get the same tree
over and over.  When they put Germanic in, they've now found that it keeps
showing up in all sorts of different places in the tree.

There were several side points to this which were mentioned in passing.
One is this: if the tree is a perfect phylogeny (roughly, one where the
character values are distributed so that there are no parallel innovations
involved), the algorithm will give the right result.  However, if the set
of characters is such that there is no perfect phylogeny, the algorithm is
not guaranteed to give the best possible tree; it will give a pretty-good
one, but not necessarily the best one. As best as I understood, different
runs of the program will give different results in this case (Don used the
word "randomizing" in this context; this is getting into the gory
internals of the algorithm which I don't understand).

What this means is that things which stay the same run after run are the
things you can take with a greater level of confidence (e.g., the grouping
of the satem core; the early branching of Anatolian, etc.).  Places where
you get more variation between runs are places where the data are more
indeterminate; Germanic is a case in point, and Albanian had already
turned out to be so indeterminate that there was no point in including
it.  Given the current, mature set of characters, and leaving Germanic
out, the same tree keeps coming up, which is a very good sign.

I asked Don in passing if the team had considered the position of Phrygian
in the tree.  I already knew that the team could not include Phrygian
because the attestion is so very poor that they wouldn't be able to assign
values for most of the characters; but once the tree had already been
drawn, they might be able to tell from what little data we have where
Phrygian might fit in.  Don said something about the mediopassive which I
didn't understand (Don is known for lecturing at a very rapid pace), but
he said "One way you could work it is this," and drew a tree on the
blackboard which had Phrygian branching off after Italo-Celtic, but before
Greco-Armenian.  What I understood was that Phrygian seems to pattern with
the "core" (Greco-Armenian + satem core) in some respects but not others.

One other comment in passing led me to believe that I might have been
wrong when I earlier said that the characters had been equally weighted,
but I didn't get any details.

That's the latest from UPenn; coming up next, we have the weather and the
sports scores.

  \/ __ __    _\_     --Sean Crist  (kurisuto at unagi.cis.upenn.edu)
 ---  |  |    \ /     http://www.ling.upenn.edu/~kurisuto/
  _| ,| ,|   -----
  _| ,| ,|    [_]
   |  |  |    [_]



More information about the Indo-european mailing list