On 05-01-19 10:13, petegray wrote:
>>This of course doesn't account
>>for the retention the velar in _all_ Indo-Iranian forms (except in
>>reduplication syllables, as in 3sg. perf. jagau < *g(W)e-g(W)oH-e).
> Pardon what must be a stupid question. Why should the velar not survive?
Well, it should have been palatalised before *e. The generalisation of
<g-> at the expense of <j-> would be understandable if the o-grade
*g(W)oh1- had been frequent enough. Who knows, perhaps <ga:tHá->
reflects o-grade (O-infixed?) *g(W)oh1táh2, and <gá:yati> comes from
*g(W)ó:h1-(e)je/o- (iterative, Narten?), but where are all the e-grades
gone? One would expect *ja:ti, *jija:ti rather than the attested
alternative presents <ga:ti, jiga:ti>, but these are late forms,
possibly back-derived from <gá:yati>. I can't say I understand this
>>something like *g(W)ah2(i)- could be posited instead, but how does one
>>derive *gi:tá- from it?
> There are other long vowels Sanskrit stems with the form -i:ta, e.g. "drink"
> 1 pa: / pi:ta and "suck" 2dha: / dhi:ta. I guess we need a solution for
> them all, not just an ad hoc one for each verb.
> (Most of the other Sanskrit long -a: roots have forms in either -a:ta
> or -ita (short vowel) or both), e.g. cha: ja: jna: ta: sa: stha: sna:
My own proposal, published some time ago, was that the "long-diphthong
roots" had the structure CeiH-, with monophthongisation assumed in some
full-grade forms (to avoid superheavy syllable rhymes), e.g. *poih3-C- >
*poh3C- (or perhaps even *peih3-C- > *peh3C- > *poh3C-, though this is
something I invented a moment ago), *trei-h2 > *trih2, and *dHeih1-C- >
*dHeh1C-. That's not very different from the generally accepted
diphthong smoothing in *djeum > *dje:m, etc. Of course in the nil grade
things like *pih3-tó- > pi:tá- and *dHih1-tó- > *dHi:tá- fall out
naturally without any extra assumptions. This solution seems to work in
many cases, but apparently not for <gi:tá->, since *g(W)iH-tó- would
have given *ji:tá-.