PIE
Satem Kentum
palatovelar *K^ >
*$ *K
plain velar *K >
*K *K
labiovelar *KW >
*K *KW
... where *$ is a cover symbol for the various Satem reflexes of
*K^.
According to one dissident view, defended by Antoine Meillet, PIE had
only *K and *KW, and the apparent contrasts on which the Brugmannian
reconstruction of *K^ is based resulted from the split of *K into palatal and
non-palatal variants in different contexts in the Satem languages. The main
weakness of this approach is that those ‘different contexts’ are difficult to
define. It seems as though Meillet’s *K series had both palatal and non-palatal
Satem reflexes in certain environments, in particular before *r.
Another unorthodox theory (Møller and Kuryłowicz), claims that the
original series were *K^ and *K, whereas *KW developed from *K -- mainly before
front vowels, through some kind of dissimilatory process -- as a Kentum
innovation. There are acute problems with this scheme, too, e.g. the occurrence
of labiovelars before consonants and the lack of typological precedent for the
proposed developments. It is extremely unlikely that such an odd change should
have occurred independently in the various Kentum branches. However, in order to
posit it as a common innovation one would be forced to make the indefensible
assumption that the Kentum languages are more closely related to one another
than any of them is to any Satem language, or that they used to form a cohesive
convergence area.
I don’t think this theory can be maintained. The real choice is between
Brugmann and Meillet. Let me begin by listing the most powerful arguments
against the classic Brugmannian aproach.
(1) No IE language preserves contrastive reflexes of all the three
series. This suggests that the ‘K-overlap’ (Satem *K and Kentum *K in the same
root) may represent a neutralisation zone between two original series.
(2) In a system with three dorsal articulations the ‘neutral’
(unmarked) one (*K) should be at least as common as any of the complex (marked)
ones. This is a general property of natural classes of phonemes, and any
violation of it should be commented on and, if possible, accounted for. But in
PIE the *K series (across phonation types) is definitely the rarest, and many of
the examples of *k, *g and *gH cited in the literature are too doubtful for
profitable discussion. In the basic vocabulary *K^ and *KW are overwhelmingly
more common. For example, *k^ occurs in the words for 6, 8 and 10 (not to
mention the decads and 100), *kW in 4 and 5, whereas no number words contain
*k.
(3) In the Satem languages there is a good deal of variation between *K
and *$. Even in Indo-Iranian *K for expected *$ is sometimes found, and in
Balto-Slavic non-satemised *K is very common. The opposite (*$ for expected *K)
is virtually impossible for purely technical reasons. There is a bias in the
Brugmannian reconstruction: anything that has a *$ reflex in any Satem language
is automatically assigned to the *K^ set. One should note, however, that the
contrast between *K^ and *KW in Satem is disturbed unidirectionally: some
instances of PIE *K^ are reflected as *K, but PIE *KW never emerges as *$. To
sum up, the boundary between *K and *K^ is fuzzy, while the *KW lexical set
preserves its integrity in th Satem languages.
Possible
counterarguments:
(ad 1) It has been claimed that Albanian and Luwian retain the
three-way contrast in some positions. The Albanian evidence was first presented
by Holger Pedersen early in this century. Now that Albanian historical phonology
has made substantial progress, Pedersen’s argument has lost its force --
notwithstanding which, Albanian is regularly mentioned by Ivanov as a language
reflecting ‘three rows of velars’, also in his UCLA lectures on Advanced IE
Phonology this very quarter:
The attention of linguists has shifted to Luwian, which has the
advantage of being a dead and imperfectly known language, so any claim about it,
no matter how odd, may be hard to refute. There are just a few instances of
Luwian <z> [ts] that might correspond to PIE *k^. In most cases <z>
is followed by <i>, which suggests a trivial palatalisation like
Proto-Anatolian *ki > tsi (as in the putative Luwian and Lycian reflexes of
*k^(e)i- ‘lie, rest’). One remarkable exception is Luwian za- ‘this’ vs. Hittite
and Palaic ka:-, but even this may be secondary (analogical generalisation of
the regular [ts] that arose in zi-). Anyway, Hieroglyphic Luwian luha- ‘light’
and dakam- ‘ground’ show velar reflexes (also in Cuneiform Luwian tijamni- we
have lenition but no evidence of affrication).
One ought to remember that velars are particularly susceptible to
palatalisation and may go through several cycles of ‘softening’ in the history
of a single linguistic lineage (Slavic and Romance provide striking examples).
Brugmannian *k^ often ends up as French [s] (as in cent) before front vowels,
while *k yields <ch> [S] when palatalised before Proto-Romance *a (as in
chevre or chien), and *kW is not palatalised (qui, quatre). This might easily
result in a mirage of a three-way contrast if French were a little-known and
poorly documented language.
(ad 2) I’ve never seen a convincing counterargument.
(ad 3) If we lump *K^ and *K together, how shall we account for the
stable *K that occurs in a number of roots (e.g. *kr(e)uh2- ‘blood; raw meat’:
Latin cruor, Old English hre:aw, Greek kreas, Sanskrit kravis.-, Old Polish kry,
etc.)?
There are some loose regularities here. *K tends to occur before
non-syllabic *r (in Baltic also before *l) both word-initially and medially.
Indo-Iranian and Armenian introduce $ where there is analogical support for it,
but retain *K otherwise (cf. Middle English kerven ‘carve’ < OE c^eorfan
[tSe@...] influenced by pret.pl. curfon [kurvon], pp. corfen [korven]).
Balt-Slavic languages have uniform *Kr. I’d like to suggest that non-palatal *K
is also the normal Satem reflex before *h2 [x] as in *dHugh2ter-, and probably
after *s.
Only *K is found in suffixes such as *-(i)k- and *-(i)sk- presumably
because only the ‘basic’ velar was allowed in word-forming suffixes already in
PIE and suffixal *-k- did not contrast with *-kW-. In many words (especially in
Balto-Slavic) apparent ‘dispalatalisation’ must be due to lexical borrowing from
Centum languages. My favourite example of a similar phenomenon is the
‘restoration’ of formerly palatalised velars in English words like egg, get,
give, score, etc., thanks to Old Norse influence. This would easily account for
a number of cases like Slavic *goNsI- (*gansi-) ‘goose’ vs. Baltic
*z^ansi-.
What can’t be defended is *ok^to: < **okito: as proposed by
Szemerényi. There is no independent evidence of syncope in such words; besides,
Satem *$ is too frequent in environments where palatalisation caused by front
vowels can be ruled out. It must be concluded that the Satem palatalisation was
not restricted to a particular context (although it was blocked in some
environments). Such a shift is hardly the usual thing that happens to velars,
but fortunately the Satem grouping (as opposed to Centum) has always been
geographically cohesive (I’d also argue that it’s possibly monophyletic) and
it’s quite possible that ‘Satemisation’ was a single event rather than a number
of independent developments in several branches.
My preferred scenario is therefore as follows:
THE SATEM SHIFT
PIE *K > Satem *$ (except in blocking environments, where *K
is retained)
PIE *KW > Satem *K
... while Centum languages simply preserve the PIE state of
things.
It’s getting so late that if I started taking issue with G & I’s
root-structure argument I wouldn’t be able to finish tonight. I’ll return to it
later, but -- in case you wonder -- my opinion is that the examples cited in
support of it are for the most part artificial and that the whole argument is
entirely specious.
Piotr