Rob:
> So in Vedic pa:t, the long /a/ is not from compensatory lengthening
> due to loss of word-final /s/?
No. We reconstruct IE *po:ds with *s and with _long_ vowel. The *s
only disappears after resonants. Since *d isn't a resonant, it
doesn't disappear. Rather it disappeared sometime during Vedic's
own development. We may observe Latin /pe:s/ and Gr /pous/ which
keep *-s.
> I had hypothesized that the o-vocalism came from the vowel
> preceding a nasal, but now I'm not so sure. However, I'm not
> sure what else could have caused it.
Nothing caused it. It was simply the way it was. Rather you
should simply be concerned about why it went zero in *kun-os.
> Yet why would a word for 'dog' be a collective?
(Good question. I don't this is a tenable statement.)
PS Szemerenyi Lengthening is the lengthening of the preceding
vowel of athematic stems when the nominative is marked with
*-s. This doesn't operate on thematic stems however which
still show *-o-s, not **-o:-s. Sometimes this lengthening is
present but the *-s isn't because 'Nominative Loss', the loss
of *-s after resonants, is necessarily a seperate event.
In other words, Nominative Loss and this lengthening cannot
logically correlate because of examples like *po:ds.
= gLeN