Re: [tied] The phonetic value of PIE *h3 and the 'drink' root.

From: Jens Elmegaard Rasmussen
Message: 14179
Date: 2002-08-01

On Fri, 26 Jul 2002, Glen Gordon wrote:

> Miguel:
> >A Bizkaian causative XXX-e-ra-gin (XXX-to be-make) or
> >XXX-e-ra-zo (XXX-to be-make(?)) is then not entirely unlike my analysis
> of
> >the PIE causative as *R-XXX-?e-
> >(be(?)-XXX-make-).
> And where is this **R-? Last time I checked, IndoEuropean
> tended towards suffixation, not prefixation. Plus,
> the **R- is all the more unlikely considering that PIE
> looks to have been originally SOV. Why can't an o-grade
> just be an o-grade??

Pardon me for barging in, but this is something I am resposible for, so
let me face the music.

The traditional term "o-grade" of IE linguistics covers quite a few
different things that cannot all have the same causal background. Some
have a very clear distribution, others are a bit more debatable. We have
to distinguish at least the following types:

1. Alternation ó / zero depending on accent. This is practically only seen
in reduplicated verbal categories, thus (1) perfect *kWé-kWór-e, 1pl
*kWe-kWr.-mé, (2) intensive *wr.-wórt-mi, 3pl *wér-wr.t-n.ti, (3)
causative aorist, athematic kind, 3sg *H1gi-H1gór-t > Ved. áji:gar
'awakened', as opposed to the thematic kind (old middle?) *wé-wkW-e-t
> Ved. ávocat 'spoke'. I fail to see solid ground to assume ó/zero for the
reduplicated presents, which I rather posit as *dhi-dhéH1-ti, 3pl
*dhé-dhH1-n.ti; but I know that some see this as the explanation of the
o-vocalism of Germanic *do:- 'do'.

2. Alternation o / zero, both unaccented, further alternating with é; in
case of lengthening we have unaccented /o:/ and accented /é:/. That is the
basic ablaut. We have lengthened /o:/ in nom. *H2ék^-mo:n 'stone' vs. /é:/
in *wr.H1-é:n 'lamb', both animate, i.e. based on //-men-s// and
//-én-s//; in both subtypes the nominative sibilant has produced
lengthening of the vowel preceding the stem-final consonantism, an old
rule that was revived by Szemerényi in a time when nobody else apparently
believed it; before the lengthening, unaccented /e/ must have become /o/
so that it yields /o:/ when lengthened. The zero alternant is seen in
neuters in *-mn. which of course had no lengthening sibilant. This means
that the lengthening produced by the nominative marker is older than the
complete loss of anaccented short vowel; that makes the /o/ it did work on
in the animate nominatives an intermediate stage between /e/ and zero.
There are some cases of phonetically regular shortening of e:/o: to e/o
again (as, after clusters in nominatives), and also some cases of
analogical /o/ (as, in *-mon-m., which is based on the proportion
*-mé:n : *-mén-m. = *-mo:n : x).

3. A related regularity has produced with /o:/ from stems with
underlyingly long /e:/, as *pe:d- 'foot' => nom. *pó:d-s. All such cases
are animate and contain the nominative marker (or its effects), thus
*pe:d-s also underwent lengthening, and the superlong /e::/ developed into
a vowel with o-timbre. Other examples are *de:m- house' which gave a nom.
with o-timbre (Arm. NAsg tun) and the odd *dém-s, and *ne:kWt-
'night' which formed nom. *nókWt-s (with shortening because of the
cluster) and the gen. *nékWt-s (Hitt. nekuz), both processed the usual
way, starting with gen. *de:m-ós, *ne:kWt-ós which underwent shortening to
*demós, *nekWtós and then fell under the INITIAL ACCENT RULE shifting the
accent to *démos, *nékWtos out of which the main ablaut rule formed the
output *dém-s, *nékWt-s. The initial accent rule is also responsible for
the ablaut é:/é in Narten presents. If the basic difference between /e/
and /o/ was once one of pitch accent (that is a property of the PIE
accent), then it would be *very* natural that lengthening of a high-pitch
vowel would produce a falling tone (it is not natural to keep a high tone
in speech for very long, nor of course to raise a high tone even

4. There is the "thematic vowel", i.e. vowels in stem-final position which
regularly behave in a way of their own. Being independent of the accent,
they have no zero-grade alternants, but show up as /o/ before voiced
segments (and surprisingly also before the nominative marker which I then
take to have been earlier voiced), and as /e/ before voiceless segments
(and zero). The dependence upon the environment is generally acknowledged,
but for reasons beyond my comprehension common opinion has apparently only
got as far as "o before sonant", covering cases of following m, n(t), w, y
and r, while disregarding *-od, *-odhi, *-obhi and *-o- + vowel, and of
course the difference between 2sg *-es and *-os. The phonetic
rationale is probably again one of tone: voiced sounds are spoken on a
lower tone than voiceless segments.

5. Then there is the funny -o- of the causative *mon-éye-ti. That /o/ is
unaccented, and there is nothing present that could be suspected to cause
lengthening; and even if there were, the accent should have been drawn
back to it, which is also not the case. On top of this the causative
structure comes in two flavors, one with -o- and one with zero: Ved.
dha:ráyati 'holds', but s'ubháyate 'shines pretty'. As seen, this is
especially the case when the causative function has been lost; that must
mean that the constant presence of -o- represents an innovation, i.e. a
standardization of the form with causative meaning preserved, while the
old *forms* are those that survive in the type that has lost the causative
meaning (which they would do quite naturally in the middle voice). So we
had the middle-voice forms *dhor-éye-tor and *k^ubh-éye-tor meaning
originally 'is made to hold, i.e. holds' and 'is made to shine pretty,
i.e. shines pretty'. The very same double-headed appearance is presented
by the verbal nouns made from root + -aH2, as Lat. toga vs. fuga, or Gk.
phorá: vs. phugé:. Why would we have IE *dhor-, *tog-, *bhor-, but
*k^ubh-, *bhug-? I have investigated the matter at length and found that a
given root-structure type consistently uses either the form with -o- or
the form with zero in such formations, thus also *tomH1-áH2 (Gk. tomé:),
*tois-áH2 (Lith, tiesà) like *g^onH1-éye- (OE cennan, Ved. janáyati) and
*tois-éye- (Lith. taisýti). There is thus complementary distribution, but
why is that?
The evasive -o- of *tog-áH2/*bhug-áH2 cannot be the root vowel pure and
simple, for that should have been lost consistently; especially, there is
no phonetic reason why *tomH1-áH2 does not have zero-grade which would
have been entirely possible. The vowel cannot have been originally long
and subsequently shortened in pretonic position, for then it should have
retracted the accent (and of course have been present in the fuga type
also). The only algebraically correct analysis is one that makes the
o/zero element an ORIGINALLY ASYLLABIC ELEMENT. Then *all* examples have
lost the root vowel proper, for the root was unaccented, and thus in the
examples quoted the root was in fact reduced to (1) *dh-r-, *t-g-, *bh-r-,
*t-ys-, *t-mH1-, *g^-nH1- as against (2) *k^-wg-, *bh-wg- (the first -
marking the original place of the root vowel). It will seem that some
roots took the -o- while others did not, and that it was the complexity of
the root that governed the choice.
Since the two shapes make up only a single morphological type, they
must have the same original structure, and since the /o/ cannot come from
zero, zero must come from the prestage of /o/, i.e. a consonant whose
regular development it was to end up as either /o/ or zero. The zero
variant will then reflect loss of the consonantal predecessor of the /o/,
the deletion having occurred in a time when the element was still a
consonant. That fully explains that, in *bhOwg-éH2 where the element was
in a heavy cluster, the -O- was pushed out, while in lighter sequences
such as *dhOr-éye-, *tOg-éH2, and even *tOys-éH2 and *tOmH1-éH2, the -O-
was preserved and so, at a later date, changed into the vowel /o/. Both
events, loss and retention, were processed in a time posterior to the
ablaut that produced zero-grade, for it was the loss of the root
vowel ("zero-grade") that caused the clustering that was treated
differently according to its exact degree of complexity.

6. Further Support. There are independent confirmations of the one-time
asyllabic status of the -o- of such formations:
One is the loss of prothetic vowels, as when Greek forms loigós from
the same root as olígos, i.e. *H3lig-, but *loyg-ó-; or Gk. moikhós
'adulterer' from the root of omeíkho: 'urinate', i.e. *H3méyg^h-, but
*moyg^h-ó-; also Hitt. warsa- 'rain' and Gk. ouréo: 'urinate' from the
same root as Gk. eérse:/aérse: 'dew' (one of which has assimilation or
dissimilation from *awers:a/ewersa:), i.e. *H1/2wers-, but *wors-éye-.
Such forms (of which there are a few other examples) are best explained on
the basis of the "consonantal o" theory, in that *H3mOyg^h-ó- was a
monosyllable in which the presence of the *consonant* -O- meant extra
crowding in the very heavy initial cluster resulting in loss of the
A second confirmation is the comparable loss of a laryngeal in the
environment oC_C (perhaps only oR_C) in the position after the root, a
rule that works only if the -o- is the one we're talking about: showpieces
are Gk. pórne: 'harlot' and tólme: (mostly, but secondarily, tólma)
'patience' from the roots *perH2- 'trade' and *telH2- 'endure, tolerate'.
These are derived from verbal nouns in *-men-, which would be IE
*pér&2-mn. and *tél&2-mn.; Greek has the compounded form with zero-grade
apó-pra:ma 'subletting' (*-pr.H2-mn.) and the animate telamó:n 'carrying
strap' which both demand these forms to have existed. From these lexemes,
IE *pórnaH2, *tólmaH2 are derived just like *tog-áH2 from the root (root
noun) *teg- 'cover'. The underlying forms are *perH2-men-é-H2 and
*telH2-men-é-H2 plus the addition of the -O-. By the zero-grade ablaut
this gives *pOrH2-mn-é-H2, *tOlH2-mn-é-H2, i.e. monosyllables in which,
again, the *consonantal* -O- caused so much crowding that the laryngeals
of the root segments were lost.
A third support is offered by the same forms in the reduction of the
suffixal *-mn- to *-n- (where labiality precedes somewhere in the word
already) or *-m- (where there is no other labialized consonant), an event
otherwise known only from cases where the suffix *-m(e)n- is itself in the
zero-grade and is on top of this preceded by zero-grade segments only (as
must have been the case in the old instrumental of man-stems in Sanskrit
which are found to end in -ma: or -na: with only one nasal).
A fourth support is also offered by these forms, viz. the accentuation
of the -ó- in derivatives from heavy roots like these (CeRH-) as opposed
to unaccented -o- in parallel derivatives from light roots (CeC-), as,
e.g. Gk. kormós 'stump vs. tórmos 'hole' from kérma and trêma
respectively. The underlying forms are here *kér-mn => *ker-men-ó-s and
*terH1-men-ó-s plus the -O-, i.e., after ablaut, monosyllabic *kOrmnós and
*tOrH1mnós. Again the -O- caused the laryngeal to vanish, but its earlier
presence has triggered the -O- to be syllabified so early that the result
was a vowel in this type when the initial accent rule operated, producing
*tórmos, but still for a while monosyllabic *kOrmós which only became *kormós
at a time when the initial accent rule has ceased to operate. I have
collected some 75 examples of these derivatives, some forty or so from
languages that can show the IE accent, and there is a complete match with
unaccented -o- with light roots as against accented -ó- with heavy roots.
Not all of these derivatives end in *-mo- or *-no-; some end in *-so-,
*-wo- or *-to-. Examples are Lith. bal~sas 'voice' (with laryngeal loss as
opposed to the root of bìlti 'start talking'), Skt. sárva-/Gk. hól(w)os
'all' from the root of OIr. slán 'healthy', Lith. vietà, acc. vie~taN
[acc.-class 2] = Germ. Weide (PGmc. *waitho:) from the root of Lith. výti
Then, all in all, there is no doubt that the -o- of these formations
was once consonantal. Other cases of IE o-vocalism do not delete
laryngeals, cf., e.g., the acute accent of Lith. kálti, bárti, málti,
kárti (with accent retraction reflected by the tones of Latvian and
Slavic, indirectly revealing the one-time presence of reduplication, this
being thus the IE intensive with ó/zero alternation), or the retained
schwa af the thematic -o- of the mid.ptc. *Hág^-o-m&1no-s > Gk. agómenos,
Toch.B akemane, or the optative type *bhér-o-yH1-te (2pl) which has acute
tone in Balto-Slavic. If there is such a thing a "lexical o", i.e. cases
of roots having /o/ as their unmarked vocalism, then that /o/ also leaves
laryngeals unharmed, as in *lów&1-tro-m 'washing tool' > Gk. loetrón, ON
laudhr (MBret. louazr).

7. The Function. What the function of the funny o was is another matter
that can only be addressed once the existence of the material is accepted.
On that point our knowledge is limited, but it is not zero:
In the causative, the problem is insoluble in its narrowest form: What
does one part of a causative marking mean? That's a silly question. In
derivatives from pre-existing stems the function appears to be
adjectivizing, most clearly in Gk. oûlos < *wól-no-s 'woolly' based on
*H2wl.'H1-na-H2 'wool' (with laryngeal loss fore and aft). Other examples
have specialized their meaning to such an extent as to make the original
semantic intention intransparent. The thematic vowel, which is always
present in the examples, ought to suffice to make the derivatory basis an
adjective, so this looks like a special kind of adjective. In fact the -o-
*never* occurs on its own, so it ought perhaps rather to be labelled a
circumfix, i.e. an elaborate morpheme consisting, in this case, most often
of an infix and a suffix which display a relation of complete solidarity.
Now, the infix part was apparently earlier a prefix, for roots beginning
with the consonant /r-/ do not infix the -o- but instead have it located
in initial position. Thus, while the lexeme *lówksnaH2 'moon' based on the
s-stem *léwk-os 'light' expanded by the verbal noun suffix *-men- (added
to an unmarked denominative verbal stem, one may assume), meaning
something like "pertaining to the making of light", has its -o- in the
usual infix position, its antonym based on *régW-os 'darkness' forms
*ó-rgWsnaH2 with prefixed /o-/; the result of the latter is Gk. órphne:
which also means 'darkness' (collective of the verbal noun from a
denominative verb made from a situational adjectival abstract, i.e.
"everything pertaining to the making of darkness"). I have found five more
examples of this kind and am now pretty sure they reflect a sound law.
Brent Vine of Los Angeles has called my attention to the Greek type
ago:gé: which he takes to have replaced earlier *o:gá: which he would find
attested in Mycenaean <o-ka> 'leadership'; accepting this, I would then
derive this type also from a prefix-type variant of the toga formation,
i.e. *H2eg^- => *o-H2g^-é-H2 on a par with *teg- => *tog-é-H2, only with
the o-part prefixed rather than infixed.
The adjectivizing meaning cannot be the whole story. The *bhor-ó-s type
forms agent nouns, which would qualify well as the derivational basis of
the causative. If *mon-ó-s means 'a thinker', then *mon-éye-ti may be
properly a denominative with factitive meaning 'makes a thinker', i.e.
'makes think'. The productive denominative type, however, has *-e-yé- with
accent on the present-forming *-yé/ó-, but that may be a younger
standardization. An older type with the same accent as the base-word may
be seen in Ved. pátyate 'governs, is master' from páti-; and also
*(s)kowH1-éy- 'wizard' (Av. kauua:, acc. kauuae:m 'priest, prophet' from
*-ay-am, not *-a:y-am) tallies remarkably well with *(s)kowH1-éye-ti (Gk.
koéo:, Lat. caveo: 'watch out'), the original meaning of the verb being
then 'make attentive', middle voice 'be attentive, pay attention, watch
out' (whence active form). The semantic relation of *bhor-ó-s to the root
*bher- would be 'one associated with carrying, i.e. a carrying one'; the
collective *bhor-é-H2 could then be 'a lot associated with carrying, i.e.
a burden for carrying'; it could also be a feminine adjective, if a word
meaning something like 'thing' can be understood and assumed to have been
feminine, in which case *bhor-é-H2 and *tog-é-H2 are 'thing associated
with carrying, thing associated with covering'. Then the underlying double
marking *O-...-e/o- would mean '(someone/something) associated with ...',
as *O-perH2-men-é-H2 would be 'female associated with trading'. The
causative *mon-éye-ti (probably from *mon-éy-ye-ti with one /y/ in the
verbal derivative suffix and another in the durative aspect marker) would
then be 'makes (the object) someone associated with thinking, i.e. makes
(the object) think'.
Under this refined analysis I would now be hesitant to accept my own
guess at a combining article in the old initial *O-, since that would make
very poor sense in the causative. If the causative is not secondarily
based on derivatives in which the *O- had already lost its original
meaning, it would be much more sensible to think in terms of
individualization, 'one of -'. In that case the initial *O- would mean
'one' and the thematic vowel, as so often, 'of'.

8. The Opposition. "Last time [Glen] checked, IndoEuropean tended towards
suffixation, not prefixation." Sure, so does Modern German, but still
there are such well-known formations as geliebt, gefunden, where the
participial suffix is obligatorily combined with a prefix. German is an
Indo-European language, and I am not aware that anybody has ever held this
morphological type against classifying it as such. In PIE, there is the
present-forming infix -ne-/-n- located before the final consonant of the
root. That looks like an older suffix which, by a process of metathesis
which is relatively easily formulated, moved into the position as close to
the end as possible, but not *at* the end, of the root. The o-infix is
mostly located in the place of the root vowel, i.e. immediately after the
initial consonantism (or, before the root-final coda, if you insist). In
derivations made from suffixed lexemes it is invariably located in (or
before) the root, never after the root. Then this is not a suffix, but
much rather an old prefix which, again by a process of metathesis, has
moved into the interior of the root.

9. Phonetics. The infix is then metathesized when clustering initially
with almost anything, being then moved to the position at the end of the
word-initial cluster. In many languages this is a normal fate of r-sounds:
English, for one, has words beginning with pr-, tr-, spr-, str-, br-, dr-
etc., but none beginning with rp-, rt-, rsp-, rps- etc.; in initial
clusters, /r/ always comes last in English. That makes the element
suspicious of being an r-sound of some kind. That could also explain that
there is no metathesis of *Or- (which even in metathesized form would
begin with a cluster ending in an r-sound) or *OHVCV- (but perhaps not in
*OHVCCV-, if Gk. ógmos tells us anything). That may suggest a back (or
spirant as opposed to sonant) r of some kind, i.e. [R] as in French or
Northern German. Vocalization of a uvular /R/ to an open variety of [o] is
very common in high-standard Danish (I myself rather vocalize to [a], but
I accept the existence of the other fashion as being a fact of human

10. Facing the Music. In a mail dated July 23rd, our friend gLeN reacted
to Miguel's report of the theory with the words: "That is by far the most
inane solution yet." I had to look up the word inane which I did not know.
I found it glossed in Webster by "lacking sense or meaning". I believe I
have shown that such is not quite the case.