IE pers.pron. (dual forms)

[Jens Elmegaard Rasmussen]

On Wed, 12 May 1999, Patrick C. Ryan wrote:

> [ moderator re-formatted ]

> Dear Jens and IEists:

[ moderator snip ]

> Of all the arguments employed to explain divergent forms, analogy is surely
> the weakest because it implies a *mistake* on  the part of native speakers
> of the language. Were Sanskrit speakers all **childs**?

I'm sure all speakers of Sanskrit were children at one time. Aren't you
in effect dismissing the existence of analogy as a factor in language
change? In the case of the non-neuter dual, the Sanskrit form -a:(v)
corresponds fully with the evidence from the other languages, only
elsewhere the other stem-classes use a different morpheme that can
everywhere continue an IE *-e. So, the other languages distinguish
o-stems stems from non-o-stems in this point, Sanskrit does not. English
has -s in the genitive sg. and pl., most related languages only in the sg.
The two problems are quite parallel, and analogy is known to be the answer
in the latter case, what's wrong with suspecting in the former?

> Now, I have two questions:

> 1) If no IE syllable may begin with a vowel in a root, and affixes derive
> from grammaticalized morphemes, why should we expect any affix to originally
> be simply -V?

Oh boy. I'm not saying the non-neuter was _originally_ *-e, I'm saying it
was (o things look as if it was) in the IE protolanguage. I am not sure
there were no vowel-initial roots, it is mostly very hard to prove that
something as hazy as *H1 was not present. On the other hand, there is
absolute certainty that IE did have vowel-initial affixes. You may take
the gen.sg. ending, whether you want to posit as *-os or as *-es, there is
no place for "-Hos" or "-Hes"; the 3pl active ends in *-ent, certainly not
"-Hent" (I am speaking of the full forms, I know there as zero-grade
variants, but they are plainly derived from the fuller variants by rule).
As for the ultimate origin of the dual *-e, I have given that question a
good deal of thought, and perhaps we are not so far apart in this. Form a
purely IE point of view, the form *-e is odd in a point which seems ti
have caused no concern to anybody else, namely its being a strong case.
For strong and weak inflectional forms are normally (and I believe,
ultimately completely) distributed by a phonetic principle: The stress
simply shift one syllable towards the end of an inflected stem if the
added flexive has an underlying vowel. Thus, there is not stress shift
before *-m, *-s, *-t of the sg. active, for there is no syllable to go to,
and hense these are strong forms; by contrast before *-me, *-te, *-ent,
and those of the dual and the whole of the middle voice (e.g. *-H2e) there
is a stress shift, so these are weak forms. Likewise in declension, where
*-s, *-m, *-H2 and zero leave the accent where it was (strong forms),
while *-os, *-ey, *-VH1, *-oom, *-bhyos, *-bhis, *-su do cause the accent
to move (weak forms). Only the dual in *-e and the nom.pl. in *-es seem
reluctant to fall into line. I have managed to explain the nom.pl. *-es
from an earlier vowel-less sequence *-z-c (i.e., two different syllables,
one marking the nom., the other the pl., in that order, structurally
parallel with the acc.pl. in orig. *-m- + sibilant), the calculation
giving at the same time phonetic explanation of all the many other
oddities of the nom. pl. forms, esp. the type in *-'-or-es, *-'-on-es: Why
is the -o- not lost? Why is it o, not e? Why is it not long? Why has the e
not been lost? And, of course, why is the e not accented? All of this is
explained by **-z-c where there was no vowel to shift to: An unaccented
-e- is first reduced to -o-; then the nom. sibilant ("-z-") lengthens
(result now *-'-o:r-zc); then short unstressed vowel are lost (but this
form does not have any, so the rule operates vacuously here); the a long
vowel is shorted before word-final triconsonantal clusters containing the
nom. sibilant (result now *-'-or-zc, much like nom.sg. of prs.ptc.
**-ent-z, through **-ont-z and *-o:nt-z with length at the critical time
when short unaccented vowels were lost, is shortened to *-ont-z, PIE
*-onts); from *-'-or-zc to the actual *-'-or-es the road is short: it
takes a vowel insertion, and since we have no contrasting evidence to stop
us we may just postulate a change to *-'-or-ezc before the last step, the
well-known change of all sibilant clusters to plain /s/, creates the
IE output *-'-ores which everybody posits on different grounds already.
Thus encouraged, one would like to derive the dual *-e from something
ultimately in equal fashion. Now, since the pronouns make one posit /H3/
as the morpheme of the dual, one may simply toy with an older form in
*-H3. From a consonant stem like *H2ner- 'man', the form PIE *H2ne'r-e
would then have to be derived from an older form *H2ne'r-H3. This entails
the postulate that word-final *-H3 developed an auxiliary vowel after a
preceding consonant, i.e. went to *-eH3. I know of no IE words in
unaccented *-eH3, nor any words in *-C-H3, so the postulate that this gave
*-C-eH3 and then lost the H3 to yield IE *-e would contradict nothing we
know, and so could be correct. - Note, however, that even without this
possibility of a principled understanding I would still have to posit *-e,
for that is what the existing evidence gives you.

> 2) If view of what you have written below about length and its connection
> with open syllables, would you mind explaining why an open syllable seems
> sufficient grounds to you above to explain the length of Sanskrit <a:>?

Brugmann's law simply records the fact that IE short *-o- in an IE open
syllable turns up in Indo-Iranian with length, i.e. as IIr. /-a:-/. I see
no problem with such a phonetic rule pertaining to one of the IE braches.
In terms of phonetic naturalness, it is okay, for [o] is more sonorous tha
either [e] or [a], and open syllables do accord more space for lengthening
than closed ones, ergo, if only one of the three vowels e,a,o should come
out longer than the others, it would be o; and if it should be sensitive
to the syllable structure, it is expected to work better in open than in
closed syllables.

> >> He then proceeds to identify an inanimate (neuter) -*iH{1}.

> > He is right in that.

> Well, then it is incumbent upon you to provide the definitive argument for
> the existence of the "pure vowel" [i], which has eluded every IEist who has
> put his pen to it.

Neither Beekes nor I see the i of *-iH1 as underlyingly syllabic; in
phonemic terms it may just as well be given the notation *-yH1. And,
hurrah, it is a strong case, i.e. contains no underlying vowels.

> >> I would maintain that the great majority of the (animate and inanimate)
> >> forms can be more simply from *-y.

> > Not the ones we find, if they are to be treated by the phonetic rules we
> > normally accept.

> Well, generalizations are less than illuminating. Why not give a few
> examples if you believe this?

*H2ner-y cannot give Gk. ane'r-e; u-stem. *-u-y cannot give Skt. -u:. What
_is_  the basis of your decision in favour of *-y ??

[... (On "us two" in IE):]

> (JER:)

> > I'm not that much of an oracle, but my guess is *noH3 which stands to the
> > accented form *nH3we' as does *nos to *nsme', and in parallel fashion
> > *woH3 for 'you two' : accented form *uH3we' (apparently dissimilated to
> > *uH3e') which would match the 2pl *wos : *usme' - provided /m/ develops
> > into /w/ in the position after the dual marker /H3/.

> This is the least logical proposition that I have seen you advance. So
> enclitic *noH{3} just conveniently drops a *-we{'}? and *nos just drops an
> inconvenient *-me? and a stress-unaccented enclitic *woH{3} modifies its <w>
> to /u/ in *uH{3}we{'}? {3}.

> I think you are engaging in free association.

I'm trying to make sense of it all in a principled way that respects the
evidence of IE itself where there is some. It may be illogical to
abbreviate wordforms when forming enclitic variants of them, but many
languages plainly do that. It's like numerals and greetings, you get all
sorts of reduced shapes in allegro speech, since people already understand
the message at the beginning (sometimes even before). The Lithuanian
dative jam 'to him' is not from some enigmatic m-form in IE, its older
form was jamui, a perfectly regular dative, from which it has been
abbreviated by no known rule whatsoever.

Jens