"centum"/"satem" "exceptions" [was Re: Northwest IE attributes]

[Miguel Carrasquer Vidal]

Jens Elmegaard Rasmussen <jer at cphling.dk> wrote:

>On Wed, 1 Mar 2000, Miguel Carrasquer Vidal wrote:

>> [...]
>> The accent pattern had certainly something to do with zero grade
>> vs. normal grade, no doubt about that.  The influence of accent
>> on e-grade vs. o-grade is less transparent [quite apart from the
>> fact that it makes little sense phonetically].  There are of
>> course obvious cases like the pattern [stressed] -e'(:)R vs.
>> [unstressed] -o(:)R in the resonant stems.

>> The e/o alternation in the thematic vowel is, one had to agree
>> with Jens Rasmussen, caused by the quality of the following
>> consonants (*e before voiceless/silence, *o before voiced).

>Thanks! we are still a minority I believe, but a viable alternative does
>not seem to have been proposed.

I apologize for the typo, I meant "one haS to agree".

>> The solution, I think, is to derive qualitative *e/*o-Ablaut from
>> an earlier quantitative **a/**a:-Ablaut, with developments /a/ >
>> /&/ > /e/ and /a:/ > /A:/ > /o(:)/.  Lengthening caused by
>> ensuing voiced/lenis consonants is well-known (e.g. English).
>> The transition from quantitative to qualitative distinction in
>> vowels is also commonplace, in the case of /a/ with languages
>> generally equally divided between long-backers (a: > o:) and
>> short-backers (a > o).  Pre-PIE was a long-backer.  I don't have
>> a good explanation for the poim'e:n ~ d'aimo:n phenomenon
>> (stressed vowels resisted lengthening by following resonant?).
>> Not all cases of e/o alternation seem to be due to secondary
>> lengthening of **a, there were probably primary **a:'s as well.
>> The backed *o: resulting from **a: generally lost its length
>> (i.e. at a time when length was no longer phonemic), so it must
>> predate "lengthened grade" and the laryngeal lengthenings.
>> Brugmann's Law shows that the length was still allophonic in PII.

>I don't get the part on Brugmann's Law.

In my scheme, after **a had become [E] and **a: had become [O:]
(and something else had happened to **i and **i:, *u and *u:),
vowel length momentarily ceased to be phonemic in PIE, much as is
the case in modern Dutch, or was the case in proto-Western
Romance (quantitative distinctions had given way to qualitative
ones).  The back vowel can thus be phonologically represented as
*/o/, even if at least in the PIE dialect ancestral to
Indo-Iranian it still had an allophonic vaiant *[O:] in at least
the open syllables (with or without the Kleinhans restriction,
etc., see Collinge's "The Laws of Indo-European" if you have the
stomach).  This allophone later merged with the new long vowels
resulting from nominative lengthening, sigmatic aorist
lengthening, derivative vrddhi, laryngeal lengthening etc., all
resulting in Proto-Indo-Iranian *a:.

In this context, it is perhaps interesting that in Tocharian the
reflexes of *o and lengthened *e: also merge (as Toch. <e> ~
<(y)e>).

>But stressed -e:'n vs. unstressed
>'-o:n (and -e:'r vs. '-o:r and -e:s vs. '-o:s in parallel
>fashion) indicates that the expected loss of the unstressed vowel of the
>suffix (/-en-/, /-er-/, /-es-/) had not progressed further than to a stage
>of weakening when the "nominative lengthening" set in: The nominative
>marker (some variety of /s/) caused lengthening of the nearest preceding
>vowel in the environment VC(C)_# Thereby, underlying /e/ appears as /e:/,
>and the weakened counterpart is found, at the end of the day, to surface
>as /o:/. Phonetically, this could perhaps be seen as something like the
>French e muet which is indeed a weakened vowel and does exhibit a marked
>rounding (at least in some varieties of the language). Supposing the
>relevant prestage of PIE to have been comparable to this, one gets a "long
>rounded schwa" from where the /o:/ of the PIE forms could well have
>developed. It takes a rule saying the lengthened vowels were not deleted
>by the working of the accent-governed ablaut, and in fact I see no
>counterexamples.

As I said, my **a: > *o rule does not give a totally satisfactory
account of the d'aimo:n ~ poim'e:n distribution.  It is tempting
to include it anyway, but it can only be done by the
counterintuitive step of assuming that stressed vowels resisted
lengthening, while unstressed ones did not (and that is
contradicted by the existence of stressed thematic vowels).

The main advantage of assuming lengthening of the unstressed
vowels in this position is that it explains why they did not
disappear as they should have after the working of "zero grade".

Similarly, an unstressed (pretonic) *primary* long vowel was
shortened, but did not disappear (and in fact attracted the
accent secondarily) in the (acro)static nominal and verbal
paradigms, some of which show o-grade (*wodr, *wedn- "water"),
others lengthened grade (*ye:kwr, *yekwn- "liver").  Of course,
in the case of *ye:kwr the preceding palatal consonant (*y, maybe
originally *l^) may easily have prevented the backing of *aa to
*o(o), and the original long vowel merged with new *e: instead.
Perhaps other cases of acrostatic *e: instead of *o can also be
explained by my theory that *all* (pre-)PIE consonants had
labialized and palatalized versions.  I'd have to investigate
that.

>So, I agree, IE /o/ has a multitude of sources.

Indeed.  Besides *h3e, I believe another source of *o can be
labialized consonant + *e (if we assume *h3 itself was
labialized, as it most probably was, we need no special rule for
the laryngeal here).  For instance, the Italo-Celtic(-Slavic)
1pl. verbal ending *-mos can be derived from *-mwes (cf. Hittite
-wen but Luwian -wan), with labialized *mw (from **mu "I") +
plural *-es.  Another example may be the apparent o-grade *mol-
"to grind" (Lat. mole:re), which appears where we would
morphologically expect e-grade, and which in zero-grade appears
as e.g. Greek mul-os "mill".  If we assume a single labialized
phoneme */mw/, the situation is normalized: e-grade *mwel- (>
Lat. mol-), zero-grade *mwl- (> Grk. mul-).  With loss of the
labialization we have *mel- and *ml- elsewhere.

=======================
Miguel Carrasquer Vidal
mcv at wxs.nl