PIE e/o Ablaut

Jens Elmegaard Rasmussen jer at cphling.dk
Sat Apr 22 14:59:07 UTC 2000


Dear Pat and anyone,

I am truly grateful for the very explicitly critical reply to my mail, for
it gives me occasion to comment on some points that appear to have become
common heritage in the field of IE, even though the basis for them appears
slender or non-existing. I have a problem, however, with dragging named
authorities into this; do we have a right to bother third party just
because _we_ cannot come to an agreement? Still, the literature is there,
and for this very purpose. So, if our moderator permits, I'll react to
your posting in full.


On Sun, 9 Apr 2000, proto-language wrote:

> Dear Jens and IEists:

>  ----- Original Message -----
> From: "Jens Elmegaard Rasmussen" <jer at cphling.dk>
> Sent: Tuesday, April 04, 2000 5:54 PM

>>>> On Mon, 27 Mar 2000, Pat Ryan (proto-language at email.msn.com) wrote:

> <snip>

> [PR]

> To clarify what my understanding, wrong though it may be, of the purported
> change from *e to *o is, I will quote Lehmann's description of the alleged
> phenomenon, from page 110 of _Proto-Indo-European Phonology_, which I
> support with some reservations:

> "After various studies the conditions of change have been defined: /e'/
> /e':/ [e' e': a' a':], with phonemic pitch accent, became [o' o':] when the
> chief accent was shifted to another syllable, and the syllable accented
> formerly received a secondary pitch accent."

That is not what we find. It may be a popular guess about the unknowable,
but, as far as observations _can_ be made, they are very potently against
it. This is not ad hoc, it's rather contra hoc. I could understand that a
de-accented /e/ turned into /o/ if all unaccented /e/'s became /o/ (as I
think they did at one point), but the next step for all would be to go on
to zero (as I believe they in fact did). Surely, this cannot explain IE
"o-grade", except for the plain cases where lengthening has occurred, so
that we get, e.g., *-e':n vs. *'-o:n from accented *-e'n-s and unaccented
*'-en-s respectively. The rule is completely inadequate to account for an
alternation between accented o and zero (as in the perfect), nor does it
tell us why the pretonic -o- of the causative has not been lost.

> Now I feel, in view of the fact that this idea was originated and defended
> by an Indo-Europeanist of undoubted competence, that a dismissive question
> like "When will you ever learn?" is wholly unjustified. I, like some others,
> may well have incorrect ideas about some (or many) things but, as I
> understand it, one of the purposes of this list is to get constructive
> feedback on ideas so corrections, where appropriate, may be made.

Agreed, and now it's happening, in both directions. Too bad that
suprasegmentals are not being conveyed on the list, but my "When will you
ever learn?" was meant with a ring of sarcasm, potentially against myself
- meaning "How long can I go on disagreeing with everybody?" As opinions
stand, it seems to be the facts that ought to give in: I am still waiting
for solid evidence favoring the most popular views about ablaut; when will
the IE languages ever learn how they are supposed to be?

> Lehmann's position is maintained more recently (1993) in his _Theoretical
> Bases of Indo-European Linguistics_, where he writes on page 131:
> "Deflected grade is explained by loss of primary accent on a vowel and
> replacement by a secondary accent. If in derivation the accent fell on an
> affix rather than on the root, the root vowel under such secondary accent
> changed to o, as in Greek nomo's "pasture," nomeu's "shepherd" in contrast
> with the vowel of the accented root in the verb ne'mo: "I pasture".

If there are two types to'mos and tomo's, it cannot be the accent that
caused any of the syllables to assume the vocalism /o/. If the cited rule
were correct, there should be -o- and not zero grade in the to-participle,
and an u-stem like *pe'r-tu-s (ON fjordr) should alternate with
**por-te'w- and not with *pr.-te'w- (Eng. ford, Welsh rhyd, Avest.
p at r@tu-; ambiguous only Lat. portus). I find the explanation by a
"secondary accent" wholly circular; there is no other evidence for such a
secondary accent than the o-vocalism it is supposed to explain. And
zero-grade appears under the very same circumstances - so the -o-
looks like something that has a reason _different_ from the changes that
led to zero.

> Though Jens may assert correctly that I personally am not as familiar with
> the literature as he is, I sincerely doubt whether Jens would be justified
> in asserting the same for Professor Lehmann.

Hey, I never expressed an opinion about your familiarity with the
literature, nor would I find it pertinent to the matter: Mine isn't
perfect, and where I know it, it doesn't always help much.

> [JR]

>> I don't think the facts are anywhere near this way: In the perfect, the
>> /o/ is accented, its unaccented variant being zero;

> [PR]

> Perfect

> As Lehmann sees it, *o' is the result of a secondary tone-accent of a
> stress- and tone-accent stage that was preceded by stress-accent stage.
> During the combined stress- and tone-accent stage, a hypothetical perfect 1.
> p. s. *we'id-eH(2) would have become, in the plural, 1. p. p. *wid-me', with
> the full- and zero-grades being the result of the stress-accent while the
> tone-accents (marked by ') shifted from the root-syllable to the affix.

> As is well known, the perfect "often, but not always, had reduplication".
> Therefore, the easiest explanation for the *o of the attested *wo'id-eH(2)
> is to assume that it is the simplification of an originally reduplicated
> form: *we'-woid-eH(2) with the reduplicating syllable receiving the primary
> tone-accent and root-syllable receiving a secondary tone-accent, analogous
> with *de'-dork-eH(2) [Greek de'dorka].

Greek verbal accent is non-original, all finite verb forms being accented
as early as the laws of limitation permit. In languages reflecting a free
accent in continuations of the perfect (Vedic, Germanic, Hittite) the -o-
is accented. Note also that corrected *de-do'rk^-H2e cannot have its -o-
explained from an earlier form with accent on the reduplication, for in
such cases the language itself tells us that the rest of the verbal body
is reduced to zero-grade, cf. Vedic 3pl prs. da'dhati from
*dhe'-dhH1-n.ti. Theoretically, one could imagine that the -o- of the
perfect was earlier unaccented, and only _got_ accented in a relatively
late period, namely after the reduction of e to o, but before the further
reduction to zero; in that case the intermediate stage o would be
preserved. I believe that actually happened in a few cases, but the
scenario become unnecessarily complicated with the perfect, and it
divorces the explanation of the o/zero ablaut of the perfect from that of
other reduplicated categories.

> [JR]

>> the same goes for the
>> intensive and the reduplicated aorist; and if the reduplicated present has
>> o-vocalism (always or sometimes), for that as well (when applicable).

> [PR]

> In Beekes, I see no *o-vocalism in intensive reduplication (*we'r-w(e)rt-,
> 'to turn'; and I am not familiar with the IE reduplicated aorist (Beekes
> lists only three types: stem, thematic, and sigmatic) --- could you give an
> example?

I don't know why Beekes writes "-(e)-" here; I wouldn't (Beekes also
writes v instead of his usual w in this example, so maybe we should not
overinterpret every fine point here). The expected *-o- is posited e.g. in
the book on the subject by Christiane Schaefer, Das Intensivum im
Vedischen, and in LIV. The assumption of *-o- is based on the lack of
palatalization in Indo-Iranian (Ved. carkarmi, janghanti) and the
existence of an apparent "primary present with o-grade" elsewhere which
can hardly be anything but a dereduplicated continuation of this category,
given the special semantic specification spelled out by Hiersche as
"Bezeichnungen wiederholter und angestrengter physischer Taetigkeit" (IF
68, 1963, 157). As I claim to have shown myself, the Balto-Slavic examples
treat root-final laryngeals in a way that reveals the one-time presence of
reduplication in what is now Lith. ba'rti 'scold', ka'lti 'forge', ma'lti
'grind' and their Latvian and Slavic counterparts. I also believe I have
shown the Hitt. verb asa:si, 3pl asesanzi 'colonize' to be an old
intensive, levelled from *asa:si, *e:sanzi (by generalization of /as-/)
from *H1s-H1o's-ti/*H1e's-H1s-n.ti (passed to the hi-conjugation because
of the *-o-). This is precious in showing vowel gradation in the
reduplication also, a structure reappearing in the Arm. nominal derivative
karkut 'hail' from *gr.-gro'Hd- (via *karkrut) from the root of Slav.
gradU 'hail', Lith. gru'odas. - The reduplicated aorist is mostly thematic
and so has zero-grade, as in Ved. avocat, Gk. ei^pon (IE *we'-w{kw}-e/o-),
but there are a number of athematic forms in Vedic, e.g. aji:gar 1.
'awoke', 2. 'devoured' from two different IE verbs, 1 *H1gi-H1go'r-t, 2
*{gw}i-{gw}o'rH3-t, again with lack of palatalization before the *-o-.

> As for reduplicated presents, I cannot put my finger on an example without a
> final root laryngeal, which complicates the pictures. But if you have an
> example of root *Ce'C- and reduplicated present: *Ce'CoC-, the same
> explanation as above for the perfect could be applied.

I do not particularly advocate the reconstruction of -o- in the normal
reduplicated present, but I know some that do, and I have very little to
prove them wrong with. I rather believe the dissimilation in
reduplications that changed *wert-wert- into *wert-wort- and further into
either *we-wort- (pf.) or *wr.-wort- (intens.) was a spontaneous rather
than a regular event of phonetic change. But the main point of the issue
is that the -o- is here accented.

> [JR]

>> unaccented variant of /e/ is also zero,

> [PR]

> We are, if Lehmann is correct, dealing with *two* phenomena: 1) changes
> brought about by tone-accent shifts; and 2) changes brought about by
> stress-accent shifts.

My question will have to be: How can anyone know that? And how could old
forms avoid being hit by later changes? That appears to me to be such a
great obstacle to all theories I have seen that one will have to rank it
as lethal.

> Without specifying exactly which you have in mind, statements become
> problematical to interpret.

I avoid all such problems by not pronouncing a verdict about points for
which there is no evidence, and especially of course if there _is_
evidence to the contrary.

> [JR]

>> cf. Gk. ane'ra, andro's (acc.
>> *H2ne'r-m, gen. *H2nr-o's); a present like *H1e's-ti, 3pl *H1s-e'nti; an
>> optative like *H1s-ie'H1-t, 1pl *H1s-iH1-me'; or paradigmatic pieces like
>> *'-iH2, gen. *-ye'H2-s; acc. *'-im, gen. *-e'y-s; *'-um, gen. *-e'w-s;
>> ntr. *-mn, gen. *-me'n-s; aor. *dhe'H1-t, ppp *dh at 1-to'-s; 'sun' is
>> *se'H2-wl, gen. *sH2-ue'n-s. In all of this, and many, many other
>> examples, accented /e/ alternates with zero.

> [PR]

> That is exactly what we should expect as a result of the shift of
> stress-accent from *e.

Didn't somebody say this should end up being /o/?

> [JR]

>> However, lengthened /e:/ does
>> alternate with unaccented /o:/: nom.sg. *p at 2-te:'r as opposed to
>> *swe'-so:r; Gk. lime:'n as opposed to a'kmo:n; end-stressed s-stem
>> eugene:'s as opposed to root-stressed s-stem he'o:s /*a'uho:s/. Thus, if
>> the compounded form of Gk. pate:'r is as in eupa'to:r, the o-timbre is not
>> by virtue of the stem's being deaccented, but by its being simply
>> unaccented (for whatever reason), for words that never changed their
>> accent also show /o/ in case they have root-accent.

> [PR]

> I find the "contrast" between "deaccented" and "simply unaccented"
> unconvincing based on the examples given since the data could be explained
> as simply as due to the different times during which the compounds were
> formed: lime:'n at a time when the affix was stress-accented; a'kmo:n at a
> time when secondary tonal accent produced *o. What seems important from the
> examples is that the affix -*men at one time had both the stress- and
> tone-accents. Also, in the case of *swe'so:r, a component of *ser-,
> 'female', has been proposed (see Pokorny p. 911, under 4. *ser-).

Look good, it's not me that's making a distinction between deaccented and
unaccented. In my algebra, unaccented vowels are treated the same,
irrespective of the morphological status of the segments they are part of,
just as good phonetic rules ought to work.

> [JR]

>> The route to this /o:/
>> must go via a reduction of the underlying /e/ prior to the lengthening
>> induced by the nominative marker file://-s//, i.e. the /-o:-/ is nothing
>> but the lengthened variant of reduced /-e-/.

> [PR]

> Frankly, lengthened variants of reduced vowels need a swipe of Occam's
> razor.

The razor would either leave them unreduced or unlengthened; the former
option would yield unaccented /-e:-/, the latter zero; we find unaccented
/-o:-/. Occam shouldn't be allowed to produce wrong results.

> [JR]

>> In stems with underlying long
>> vocalism, lengthening of /-e:-/ yielded /-o:-/, thus *pe:d- => nom.
>> *po:'d-s; likewise *de:m- => *do:'m-s (exact form of nom.sg. insecure, but
>> acc. can only be *do:'m); I take this to indicate that the final part of
>> the superlong vowel was unaccented and so developed o-timbre, and the
>> /-o:(:)-/ is the product of contraction.

> [PR]

> Well, this explanation does not explain Latin pe:s very well.  And the
> situation of *de/e:m-/*do/o:m- is so fluid that another example would surely
> be better.

pe:s does not match Gk. pous anyway, nor English foot. Since the weak
cases apparently have /-e-/, an alternation po:d-/pod-/ped- could easily
be levelled to pe:d-/ped-. But of course the explanation cannot be better
than the material it is based on, in this case Schindler's sifting of the
IE root-noun types, this being his "acrostatic" o/e type. Note that the
types were established without any theory about their phonological
prehistory, a part I have only added later. It should be assessed
as a very positive point that purely descriptive results like Schindler's
have proved open to a consistent phonological analysis without any
change.

> [JR]

>>    - There are special cases that demand special rules, thus the thematic
>> vowel (stem-final vowel of all kinds of stems) which is not reduced by the
>> accent, but alternates e/o depending on the phonetic nature of what
>> follows (the alternation is best preserved in pronouns and verbs, but
>> plainly applied originally also to nouns), actually in a very simple way:
>> /e/ is the form before voiceless segments and zero, while /o/ is the form
>> before old voiced segments, including the little surprise (or flaw, if you
>> look at it with a hostile mind) that the nominative *-s acts like a voiced
>> segment and produces *-o-s; thus, the nom. *-s is different from the *-s
>> of the 2sg of the verb which has *-e-s; note that the two also differ in
>> the detail that the 2sg marker does not cause lengthening and so must have
>> been originally phonetically different from the nom. morpheme.

> [PR]

> I would gladly grant the IE *-s (2. p. sing.), which I derive from earlier
> /s[h]o/ has a different origin from nominative *-s, which I derive from
> earlier /so/.

> But I cannot accept that voicing of a root-final obstruent determines the
> quality of the root-vowel --- at least, consistently, for we have *pe/e:d-
> and *de/e:m- alongside *po/o:d- and *do/o:m-.

It does not influence the root vowel, only the "thematic vowel", i.e. a
vowel in suffix-final position. This must be a junctural phenomenon, for
a suffix-final consonant has no such influence on a preceding vowel of the
suffix itself, cf. gen. in *-te'y-s or *-me'n-s. It only works _across_
the boundary between suffix and flexive, or (in the case of absence of a
flexive) across word-boundaries. That's not my discovery, Saussure saw it
already, I only added the rules.

> [JR]

>>    - Another special case is the "o-infix" I claim to have found in the
>> causative and in thematic derivatives like Gk. tome:', po'rne:. To my very
>> great surprise these forms only became amenable to normal algebra if the
>> /-o-/ segment was derived from an earlier consonantal added morpheme, i.e.
>> an infixed sonant which later, after the working of ablaut proper,
>> developed into /o/ (or was lost, the two results being in phonetic
>> complementary distribution).

> [PR]

> It is the firmest of my beliefs that IE had *no* infixes. An only apparent
> exception is the *metathesized* -n- in certain present stems.

We agree if I may derive my o-infix from an old prefix that got
metathesized (most of the time, i.e. whenever the root did not begin with
*r- in which case the o- appears prefixed). IE apparently had no prefixes
either, so it may in origin be a compositional member, a deictic stem
making an adjectival compound with the following stem. The phonetics would
be the same.

> [JR]

>> It is only in such forms that we find
>> "laryngeal loss in words with o-grade", often called Saussure's rule,
>> because Saussure collected a few examples of wanting laryngeal reflex and
>> found their common salient feature to be "o-grade". Saussure did not offer
>> any explanation of the strange fact, and as long as the o is taken to be a
>> phenotype of the old _vowel_, there can be none; however, if the o is seen
>> as an old consonant, the solution is obvious: laryngeals were lost where
>> there were many clustering consonants, and retained where there were
>> fewer. We also understand that the _unaccented_ -o- of, say, caus.
>> *mon-e'ye-ti 'causes to think' was not lost: it was a consonant when the
>> ablaut worked.

> [PR]

> A rather complicated solution when Lehmann's simple solution is at hand:
> *me'n- + *e'ye- -> *mone'ye-.

That is not simple, for it does not explain the difference it is supposed
to account for, viz. -o- vs. zero:  Why did this not give **mne'ye-, if
*me'n- + *-to'- gave *mn.to'-? Rules are supposed to work for all forms
present in a language at the time, not just for some.

> [JR]

>>    These facts are all well known - or based on the analysis of types of
>> examples that have been in the focus of attention for a century. Their
>> actual testimony is _very_ far from being "e goes to o when the accent is
>> shifted away from it". When will you ever learn?

> [PR]

> And when will you cease patronizing condescension?

Yeah, that was uncalled-for; I'm working on it.

Jens



More information about the Indo-european mailing list