27-04-2004 15:06, elmeras2000 wrote:
> The point about /setstos/ is this: If PIE has /s/ and /t/ as
> phonemes, then there is no reason not to interpret a sequence
> [setstos] as a concatenation of precisely the phonemes /setstos/. It
> is only on a more abstract level that it is discovered that *[tt]
> and *[dt] do not exist, so that [tst] can just as well be
> interpreted as /tt/ or as /dt/ (and as /dst/ too for that matter).
> But since phonemic analysis acts on a level preceding identification
> of morphological connections, the knowledge that the root of (at
> least some realisations of) [setstos] is [sed-] in other contexts
> has no part in it.
How about the surface-true interpretation /sectos/? (I'll transcribe the
affricates [c] and  with single letters to emphasise their unitary
status.) I think the idea of a phonemic */c/ could be defended for PIE,
even if it was a marginal member of the inventory, and if its
distribution was defective. It's arguable that what we find finally in
*po:d-s and *h1dont-s was a single affricate segment rather than a
cluster, and */c/ is an attractive analysis of the "thorn" in the "core"
IE groups. I'd suggest that the reason why we find *gWHð, *g^Hð, *kWþ
and *k^þ, but no *gWð or *g^ð, is that */c/ was the only phonemic
affricate. It could become allophonically breathy-voiced by progressive
assimilation when preceded by a breathy-voiced stop (like *t by
Bartholomae's Law), but ordinary modal-voice assimilation was
regressive, so any underlying media was automatically devoiced before
*/c/. But if */c/ was the first segment in a cluster, its phonation type
was automatically assimilated to that of a following stop (*//-d-d-,
-t-d-// --> */-cd-/ [3d], etc.). In fact, it behaved exactly like */s/
in these respects.