WeepingElf wrote:Most instances of PIE *a are indeed due to *h2; most of the remaining have skewed distribution and may be loanwords. Some may have been borrowed into PIE about the time it broke up, and already had /a/ from *h2e. These possibilities probably cover all instances of PIE *a.
Though to what extent have actual sources been determined for these "possible borrowings"? There is apparently some evidence for roots that had not only *a, but also an
ablauting *a ~ *ā which would suggest a slightly older age.
KathAveara wrote:is it possible that the palato-velars were only phonemic in the satem dialects of PIE? That is, in the stages prior to what we can reconstruct, there were only two velar series - plain and labialised - and the plain velars had palatalised allophones in certain environments. Ablaut would cause alternations between the palatalised and plain allophones, so the satem dialects generalised the palatal allophone, while the centum dialects did not, leaving the distinct allophonic, or even dropping the distinction entirely. Does that make sense? Is it likely?
Well, since the satem dialects are in mostly good agreement on where to put palatals and where velars, and no language shows remains of this sort of an alternation (*Ḱe ~ *Ko ablaut), the levelling of this system would have to have happened quite early on. Which then makes it a bit difficult to understand why Armenian and Indo-Iranian share almost as many similarities with Greek as they do with Balto-Slavic (dunno how Albanian fits here).
I don't see the motivation to explain the uvulars (traditional plain velars) away entirely; they could simply have been a rare bunch of phonemes. The main problem seems to be the typologically odd K Kʷ Q system which doesn't have many parallels. Usually there would be labiouvulars around too in a system like that.
However, it's entirely possible to suggest some! The traditional *Ḱw clusters are a fairly strange contrast to phonemic labiovelars, and it would be even stranger if this was *[kw] versus [kʷ]. So, let's reinterpret things a little: the "labiovelars" were instead labiouvulars, while the "palatovelar+w clusters" were plain labiovelars.
This seems to work just fine in the descendants too: Satemization pushes both series forward, at what point the labiovelars break apart into *Čw clusters. Centumization merges the velars with uvulars, including the labiovelars with labiouvulars.