Phonemic split

[Sean Crist]

In more than one post from Steven Long, there has been discussion of
phonemic split, altho Steven did not use this exact terminology in his
posts.  I think I can make myself clearer if I discuss this all at once
rather than responding piecemeal to each of his posts.  I'd like to
discuss phonemic split generally, and then discuss how it applies to the
cases Steven Long brings up.

When language change proceeds according to exceptionless sound law,
phonemic split occurs when the environment conditioning some allophonic
alternation is neutralized.  A textbook example occurred in the prehistory
of Indo-Iranian.  First, */k/ > */c/ before */i/, */y/, */e/
(palatalization). Then, */e/ and */o/ merged into */a/.  Notice that this
vowel merger destroys one of the conditioning environments for
palatalization; you can no longer predict where palatalization should be
found.

For example, the PIE 3 sg. pres. of the verb for 'to follow' would have
been something like *sekweti (I might not have the reconstruction quite
right, but it should be something like this; cf. Gk. _hepetai_).  In
Indo-Iranian, you'd first get the merger of the labiovelars into the
velers (*seketi), then palatalization (*seceti), and then the vowel merger
(*sacati), coming out in Skt. as _sacate_.  After the vowel merger, you
can no longer predict where you get */k/ and where you get */c/; at this
point, a new phoneme (= constrastive speech sound) has arisen.  (As an
aside, there were some forms of the verb where the velar was originally
followed by */o/ rather than */e/, as in the 1 sg. pres.; this should have
given */k/ as the output in Sanskrit, but analogy tends to destroy such
alternations within paradigms, and it did so here.)

As for phonemic splits which are _not_ the result of the neutralization of
environments conditioning an allophonic variation, we'd like to say that
such cases don't exist.  They can only arise by sporadic sound change,
which is contrary to the Neogrammarian Hypothesis.

The unfortunate reality is that we do run into such cases every once in a
while.  There is a case from Iranian which I'm not entirely able to
reproduce from memory, but the essence is that there was a single
prehistoric phoneme which does seem to sporadically split without any
obvious conditioning environment.  What we think probably happened is that
the dialect of the written records is one which had had massive borrowing
from some closely related sister dialect where this phoneme had undergone
some sound change.  Thus, the words where we get the unexpected segment
are actually borrowings.  There is external evidence which makes this
scenario plausible; if I'm recalling the story right, it's a matter of
recorded history that there had been a shift in political power from one
city to another, which is a situation where this sort of thing might be
expected to happen (i.e., prior to becoming the prestige center itself,
the new capitol had been borrowing words from the old prestige center).

I hesitate to bring this second sort of case up, because there will tend
to be an unbridled reaction to invoke this sort of solution willy-nilly.
I want to emphasize that a sporadic phonemic split is an analysis of last
resort; it's what you invoke when there is just no other plausible option
open to you.

To see how this looks on the ground, consider the following hypothetical
case, where Languages A and B are assumed to be related:

	Language A	Language B
	sabos		savas		'cucumber'
	zabate		savathe		'sun'
	
I'm looking at the initial consonant; note that Language A has an s/z
contrast, while Language B does not.  Notice also that the two are in the
same environment; there is nothing which could plausibly be conditioning
this as an allophonic alternation. There are two possible explanations:

1. The proto-language contrasted */s/ and */z/; Language A maintains the
contrast, but there was a merger of */s/ and */z/ in Language B.

2. The proto-language did not contrast */s/ and */z/.  Language B
maintains this state of affairs, but Language A underwent a sporadic
phonemic split.

What I'm saying is that we should pick #1 unless there are other,
extremely unusual considerations specific to this case which make this
explanation implausible. I'm sure that the overwhelming majority of
specialists in the field would agree, and with good methodological reason.

One other thing I'd note is this.  Even if we admit sporadic phonemic
splits as an analysis of last resort, one thing which we _cannot_ admit
_ever_ is that two daughter languages could both independently innovate by
undergoing the same sporadic split in exactly the same way, with the
outcome of the split being the same in both languages word for word.
Given that the outcome of a sporadic split in a given word is, by
definition, not predictable, the statistical probability of a sporadic
split repeating itself precisely is mind-bogglingly small.

With all of this in mind, let's return to the question whether
Proto-Celtic could have been the parent language giving rise to all of the
attested Indo-European languages.  In essence, what Steven Long is saying
is that for any set of attested, related languages, there can be multiple
prehistoric scenarios which could equally well give rise to those
languages, depending on what language you arbitrarily pick as the
proto-language for the entire family.

I pointed out that Proto-Celtic could not have given rise to all of the IE
languages, because the Celtic velar series would have to sporadically
split into a velar series and a palatal series in Anatolian, Indo-Iranian,
etc.  Steven Long's response was that this split could simply have been an
innovation in those branches.

So we've got two hypotheses:

1.  The proto-language for the IE family had a three-way contrast between
palatal, velar, and labiovelar consonants; Celtic merged the palatal and
velar series.  (Traditional reconstruction.)

2.  The proto-language for the IE family had a two-way contrast between
velar and labiovelar consonants.  There was a sporadic phonemic split
giving rise to the three-way distinction found in Luvian, Indo-Iranian
(with the later merger in Indo-Iranian between the velars and
labiovelars), etc. (Steve Long's Proto-Celtic hypothesis).

#1 involves only regular sound changes, while #2 involves a sporadic
change.  We therefore pick #1.

I'd like to note one other thing.  Under the most charitable version of
Steven Long's PIE = Proto-Celtic hypothesis, we'd have to say that the
sporadic phonemic split is a shared innovation of all the languages whose
data require it.  As I noted earlier, it is impossible that the various
branches independently underwent exactly the same sporadic split with
parallel outcomes in each word.  But note that doing this results in a
tree which is radically different in its structure from the one I
presented.  We'd have to group Indo-Iranian and Anatolian (and probably
others) under a single node; but Germanic, Greek, etc. would not be under
this node since their situation with the dorsals is like that of Celtic.

Steven Long's said that you could take the tree as I presented it and
substitute Proto-Celtic as the parent language for the entire family.  As
I hope I've made clear, even if we accept the account involving a sporadic
phonemic split (as we should not), it simply cannot be the case that the
tree could still look as it does.

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