On Fri, 6 Jun 2003, Glen Gordon wrote:

> I fully accept that *e:/*e is parallel with *e/zero and that both
> are the result of accent alternations.

Fine, nice to be understood. Most colleagues south of where I am won't
have it that way, but want e:/e to be *analogical* on e/zero instead. I
find that misguided and so unrealistic that we can disregard it. If e:/e
were analogical it would be productive, but it is so rare that it was not
even observed until Narten did that. And if you have an old type with
*unchanging* accented /e/, as they say it was before the analogy gave it
grades, then a different levelling would be much, much more natural: I
mean, if there was unchanging /e/ and changing e/zero as two types, why
not instead change the latter into stable /e/ by generalizing the
e-allomorph? That would have reduced all types to a single one with stable
/e/ all through, the easiest language in the world. So I am very sure we
are right in agreeing that the gradation is an old thing here, and that
there is an underlying parallel between the two sets e:/e and e/zero due
to a shared causation which is the changing accent (which has been
secondarily stabilized in one of the types).

>
> What I don't accept is the analysis that *pod- has a long vowel
> in its default form. It's default form (the strong form) is *pod-,
> regularly lengthened in the nominative by the presence of *-s.
> The lenghtening of the weak case form *ped- (as seen in
> *ped-os) can likewise occur, producing *pe:d-. This doesn't
> mean that the root contains long vowel. In fact, it is the
> presence of the short vowel in cases where the root is clearly
> accented and clearly always accented that demonstrates that
> the root has _short_ vowel. That proof is the accusative
> *podm.

The situation is neutral on this point. If o/e can be shown to work better
and to produce more illumination of dark spots in the grammar, it should
be preferred. If it causes obscuration of things that are otherwise clear,
and no necessity can be shown for this, it should not. Everyone is free to
think as best he can.

>
> To support your theory, you must assume the existence of
> triple-morae and assume that the roots were reduced, even
> though we know full well that the nominative _lengthens_
> the preceding vowel. It's a multiplication of hypotheses.
> We either end up claiming that all stems showing length in
> the nominative are long-vowel stems (making for a severely
> lop-sided system), or we assume yet more nonsense as to
> why some stems are to be considered "long" and others not.

Not "even though", but "because". It's a repetition of the *same*
hypothesis as elsewhere in the algebraic system. You are not describing
what I have done, which surprises me after all this amount of discussion.
It is a remarkable fact noted by myself and also by Schindler that there
is a marked recurrence of *the same* roots in the lists of examples of
"Narten ablaut", even across paradigms. There are nominal and verbal
derivatives from *k^ey- 'lie', *plew- 'swim', *bher- 'carry' and quite
afew others with an addditional mora in the root segments when compared
with the same formations made from most other roots. I believe that should
be considered. I therefore also believe that a theory accomodating that
fact is superior to one that does not. - I used not to accept the
underlying length as a property of the lexical roots (although it was my
first guess), but wanted it to be paradigm-specific only, but later
discussion and analyses I have made myself have made the distinct
impression that the length sometimes *is* a trait of the root itself. It
is not the whole story, however, for there are also long-vowel paradigms
that can be formed by *all* roots, such as the *do'h3to:r agent-noun type
and the sigmatic aorist which are fine Narten examples.

The fact that I do not buy the theory of an IE "hi-conjugation at face
value" is decisive for my view on the status of the o-timbre. I can derive
it by rule in such a way that analogy stays well within the territory in
which it cavorts in other parts of the language. I therefore believe a
theory that dispenses with an o/e type as basically different from the
e:/e type is superior to a theory that must accept both a two enigmas
instead of one. If I am proved wrong in my presuppositions I'm willing to
accept anything that does work.

Jens