The fate of pre-PIE **i and **u.

From: Miguel Carrasquer
Message: 14120
Date: 2002-07-23

Even though *i and *u are reconstructed for PIE, the usual
interpretation is that they are vocalic allophones of the resonants *y
and *w. This interpretation is something of an overgeneralization,
because of a small but irreducible number of instances where *i and *u
do behave as vowels in their own right. In most cases, however, *i
and *u are indeed zero grade variants of *ey and *ew (sometimes *ye
and *we), and they behave as sonorants (like *l, *r, *m or *n) rather
than as vowels.

This leaves the vocalic system of PIE as consisting merely of *e and
*o, plus their long (or lengthened) counterparts *e: and *o:. There
was also an *a, but its frequency is minimal. From a typological point
of view, this is a strange system indeed. It gets even stranger if we
take into account the fact that *e and *o can be merely Ablaut
variants of the same underlying phoneme, as the evidence allows.
Despite what has been said about some NW Caucasian languages,
one-vowel languages do not exist, or are at least extremely rare.

If we accept that *e, *o (and *a) ultimately derive from a single
underlying vowel (let's call it **a), the most elegant way out of the
dilemma is to reinstate **i and **u as vowels in their own right for
the period in question ("pre-Ablaut PIE"). This would give us a
typologically normal three-vowel system {a, i, u}, or, more likely, a
six-vowel system with short {a, i, u} and long {a:, i:, u:}.

Now the merger of short /i/, /u/ and /a/ into a single phoneme (e.g.
/&/) is a not uncommon phenomenon. It's exactly what must have
happened in the pre-history of the two-vowel NW Caucasian languages
alluded to above. We know that something similar happened in
pre-Tocharian, where PIE *i, *u and *e merge as *ä. The vocalic
merger is accompanied by a feature transfer to the neighbouring
consonants. That is to say, former **i causes palatalization and
former **u causes labialization, while former **a is accompanied by
neutral quality. The NW Caucasian languages indeed show a plethora of
labialized and palatalized consonants besides the plain ones.
Tocharian too has an almost full set of palatalized consonants, but
the labialized and plain consonants have largely merged, except
sometimes in the velar series (*kuC > *kwäC). In Old Irish too, the
loss of final vowels and syncope of certain medial vowels led to a
three-way quality distinction in the consonants: palatalized
("slender"), plain ("broad") and u-quality. The u-quality consonants
had largely merged with the plain ones by the time of attested Old
Irish, but some instances remain, especially in morphology (e.g. biur
/b^irW/ "I carry" < *beru).

The existence in PIE of a full labio-velar consonant series (*kW,
*ghW, *gW) is strongly suggestive of something similar having happened
in pre-PIE times (cf. the erratic retention of u-quality in the
Tocharian velar series). A careful analysis of other phonological
anomalies in the reconstructed material produces some other possible
(but erratic) instances of palatalization and labialization. Besides
*kW, *ghW and *gW, we have possible cases of labialization in the
variation between Germanic *f, *b vs. *kW, *ghW elsewhere (e.g. the
"wolf" word *wl.kWos ~ *wl.pos and the "bid" word *ghWeid- ~ *bheid-,
suggesting original labialized labials **pW and **bhW); the
alternation *t ~ *s in the "month" word *méh1not- ~ *meh1nés- and the
ptc.pf.act. *-wot- ~ *-ús-, as well as in the verbal second person
endings sg. *-s ~ pl. *-té[C], suggesting a development **tW > *s
under certain circumstances (cf. a similar development in pre-Greek).
Further, *h3 may be interpreted as the labialized variant of *h2 (and
*h1?), and the alternation *m ~ *w (in e.g. the first verbal person
sg./du./pl.) suggests a pre-form **mW. As to possible cases of
palatalization, we have Greek pt- and phth- corresponding to *p and
*bh elsewhere (< **p^ and **bh^ ?), the alternation *t ~ *y in the
"three" word (*trei- ~ *tr.ti- < *tret^-) and *s(w) ~ *y (< *tW ~ *t^)
in the pronominal and thematic plural (*mes ~ *wei- "we", plural *-es
~ oblique (o-stem) *-oi-). A similar case of labialized ~ palatalized
alternation as in the plural may underlie the dual (*h3 ~ *h1 < **xW ~
*x^ ?). In the sonorants, we have sporadic alternations *n ~ *y
(*nem- ~ *(y)em-) and *l ~ *y (*ye:kWr(t) ~ *le:pWr.(t) > Skt. yakrt ~
Arm. leard), while Balto-Slavic shows an opposition between two types
of syllabic sonorants, one set giving PBS *im, *in, *il, *ir
(palatalized/plain), the other giving *um, *un, *ul, *ur (labialized).

In the past I have suggested the following simple scenario: the PIE
vowel system was originally a six-vowel system

i u i: u:
a a:

In normal stressed position, this gave:

**i é, with palatalization of the preceding consonant.
**u é, with labialization of the preceding consonant.
**a é
**i: éi
**u: éu
**a: ó

When unstressed:

**i 0, with palatalization of the preceding consonant.
**u 0, with labialization of the preceding consonant.
**a 0
**i: i
**u: u
**a: e

As to the development of pre-PIE *a and *a:, I'm still convinced that
this is the right solution. A development /a:/ > /o(:)/ is so common
as not to require further comment. That PIE *o was formerly a long or
"strong" vowel is demonstrated by the fact that it is still long in
open syllables in Sanskrit, that it does not get coloured by
laryngeals (as *e: and other long vowels don't), that it was not
reduced to *ä in Tocharian (it eventually merged [but for
palatalization] with *e: instead). The qualitative Ablaut *e ~ *o
(which evidently has nothing to do with front/back vowel quality) can
now be explained as a simple quantitative Ablaut (lengthening and/or
shortnening) of *a ~ *a:, while cases of non-apophonic *o can
unproblematically be taken represent an original long vowel **a:.

Unfortunately, the above simple scenario fails for the alleged
reflexes of pre-PIE **i and **u. Firstly, *éi and *éu cannot continue
the pre-PIE simple vowels **i: and **u:. The phonetic development
itself (**i: > **ay > *ei) is sound, a simple diphthongation of long
vowels. But in the sequences *ey and *ew, the elements *y and *w
really do behave as resonants phonotactically: there are no (or not
enough) roots of the structure CéiRC ~ CiRC (CéuRC ~ CuRC)
[R=resonant, C=obstruent], as we would expect to see if *ei and *eu
originally were single (long) vowels (cf. the extremely common root
structure *CoRC ~ *CeRC). A second objection is that the simple
scenario of palatalization / labialization of the preceding consonant
is also inadequate: sometimes it's the _following_ consonant which is
affected, and sometimes traces of palatalization or labialization fail
to appear at all.

It is clear that the scenario above is not adequate. I would like to
suggest an alternative one, which is more complex, but hopefully has
greater explanatory power. The point of departure is the same:

*a *a:
*i *i:
*u *u:

All short vowels merged in principle as *a (= /&/), which later gave
*e. More exactly, however, *i gave *ya and *u gave *wa under most
circumstances, where the *y and *w could subsequently be lost (with or
without changing a preceding consonant). When they were no lost,
stressed *yé and *wé eventually gave *í and *ú (but remained as *yé,
*wé after a vowel), while unstressed *ye and *we resulted in
zero-grade *y (*i) and *w (*u). In the long vowels, **a: and **u:
merged as **a:, which later gave *o. Long *i:, however, resulted in
*e: (in svarita position: *e). As in the case of the short vowels,
the reflexes *u: > *wo and *i: > *ye(:), with glide maintained, can
also occur under certain circumstances. Besides regressive
palatalization and labialization (*Ci > *Cye > *C^e), **i and **u
sometimes also caused palatalization and labialization of a following
consonant (*iC > *iC^> *(y)eC^). Unstressed long vowels were
shortened to **a, **i and **u, and subsequently evolved to *e, *(y)
and *(w), respectively. Schematically:

stressed unstressed lengthened svarita
**a *é 0 *o
**a: *ó *e --
**i *é 0 *e
*yé / *í *i *ye
**i: *e: 0 --
*yé: *i --
**u *é 0 *o
*wé / *ú *u *wo
**u: *ó 0 --
*wó *u --
**ay *éi *i *oi
**a:y *ói *ei --
**aw *éu *u *ou
**a:w *óu *eu --

To test this scheme in practice, we have to see how well it performs
in explaining the attested PIE forms. Firstly, we turn to the i- and
u-stems, which I reconstruct as stems ending in Pre-PIE **i and **u.
I will distinguish between initial-stressed (proterodynamic) words
with light root syllable (PD-L), proterodynamic words with heavy root
syllable (PD-H) and end-stressed (hysterodynamic) words (HD). At the
first stage, we have:

PD-L PD-H HD
N '-î-z '-i-z -í-z
V '-îy '-iy -íy
G -íyâs -íyas -iyás

The circumflexed vowels have been lengthened by the svarita-rule ("in
final syllables, the vowel is lengthened if the preceding syllable is
stressed and short"). In the vocative and the oblique cases (but not
in the nominative and accusative), the *i is followed by a glide
(linking the *i with the next vowel, which in the vocative had already
been lost). Notice that at the stage of svarita-lengthening, the
glide did not count for purposes of syllabification (otherwise we
would have had *-í-yâs in the PD-H form, as in the PD-L declension).

For the next stage, we apply the developments **i > ye, **i: > **ye:,
except before *y, where we have **iy > *ey, **i:y > *e:y. Further we
have the usual developments **a > *e, **a: > *o(:). In the oblique
cases of the PD-L declension, the accent unexpectedly shifted to the
ending (which may already have happened in the previous stage),
producing:

N. '-ye:s '-yes -yés
V. '-e:y '-ey -éy
G. -eyós -éyes -eyés

After zero grade (*e > 0, *e: > e), we get:

N. '-yes '-ys -yés
V. '-ey '-y -éy
G. -yós -éys -yés

In the next stage, *ye is simplified to *i:

N. '-is '-ys -ís
V. '-ey '-y -éi
G. -yós -éys -ís


Finally, to get to the attested situation, there is the analogical
transfer of the ending -ei- to the oblique cases of the HD stems (to
distinguish the G. from the N), while the vocative -ei was generalized
(analogical forms in brackets):

N. '-is '-is -ís
V. '-ei ('-ei) -éi
G. -yós -éis (-éis)

For the u-stems, the developments are largely similar:

N '-ûz '-uz -úz
V. '-ûw '-uw -úw
G. -úwâs -úwas -uwás

[**u > we, **i: > **wo, but **uw > *ew; **a > *e, **a: > *o, accent
shift in HD-L oblique]

N. '-wos '-wes -wés
V. '-ow '-ew -éw
G. -wós -éwes -ewés

[zero grade: *e > 0, *e: > e]

N. '-wos '-ws -wés
V. '-ow '-w -éw
G. -wós -éws -wés

[*we > *u]

N. '-wos '-ws -ús
V. '-ow '-w -éw
G. -wós -éws -ús

Finally, Ausgleich of the nom. and acc. to *-us, *-um (for **-wos,
**-wom in the HD-L declension). Further developments as in the
i-stems:

N. ('-us) '-us -ús
V. '-ou '-u -éu
G. -wós -éus (-éus)

Another nominal class with contains original **i in the suffix is that
of the ih2-stems:

PD-L PD-H HD
N. '-îx '-ix -íxz
G. -íxâs -íxâs -ixás

After **i > ye, **i: > **ye: and **a > *e, **a: > *o, we get:

N. '-ye:x '-yex -yéxs
G. -yéxos -yéxos -yexés

After zero grade:

N. '-yex '-yx -yéxs
G. -yéxos -yéxos -yxés

Next is the development *ye > *i, which here occurred in
unstressed/closed syllables only. In stressed/open syllables [i.e. in
the oblique forms], it was perhaps the back quality of *h2 wich
prevented the closing of *ye to *(y)i.

N. '-ih2 '-yh2 -íh2s
G. -yéh2os -yéh2os -yh2és

Next we have laryngeal colouring:
N. '-ih2 '-ih2 -íh2s
G. -yáh2os -yáh2os -ih2ás

And after laryngeal loss:
N. '-i: '-i: -í:s
G. -yã:s -yã:s -(y)yás

Some other relevant paradigms:

The word *mélit- "honey":

**mál.it > **mélye:t > **mélyet > *mélit
**mal.ít.âs > **melyétos > **mlyétos > *mlítos

The word *meh1not- "month, moon":

**mát.nût.s > **métnots > *méh1nots > *méh1no:ts
**mat.nút.âs > **metn(w)étwos > **m(e)tnésos > *meh1nésos

Here we see that stressed *ú caused labialization of a following *t,
while the labial element was lost after *n (or perhaps after the
complex cluster *mtn-?).

In the ptc.pf.act., **u is suffix-initial:

**'-.ûts > **-wots > *-wo:ts
**-út.âs > **-wétwos > **-úsos

Again, *ú causes *t > *s(w), but in morpheme-initial position, the
*w-element is maintained (so it's not a case of the labialization
being only progressive, as it might the case of *meh1not- might
suggest).

The cases where we have Ablaut *o ~ 0 can be explained as stressed vs.
unstressed *<u:>:

*pú:ntxz > *p(w)ont(e)hs > Skt. pánt(h)a:s
*pu:ntxás > *p(w)n.thés > Skt. pathás

Note the loss of labialization after *p.

For short *u, we have:

**músz > **mwéss > *múss > *mú:s
**musás > **mwesés > *mwsés > *musés

Ablaut with *e: can be explained as involving **i:

**kí:rd > *ke:r
**ki:rdás > *kr.dés

**lí:punt > *lyé:pwert > *yé:kWr.(t)
**li:púntas > *lyepwéntes > *y(e)kWén(t)s

Note, unlike in the case of *pont(e)H-, the labialization of *pw (>
*kW).

Further details are provided by the reconstruction of the pre-PIE
personal pronouns. The basic forms would have been: 1. **mu, 2. **tu,
3. **su. The accusative/oblique was made by (late) suffixation of a
stressed element *-má, the plural by suffixation of **-átu or **-áti,
the dual by suffixation of **-íku or **-íki. This gives the following
forms:

N **mú **tú **sú
Ob **mu + má **tu + má **su + má

Npl **muátu **tuátu **suátu
**muáti **tuáti **suáti
Ob **muatu + má **tuatu + má **suatu + má

Ndu **muíku **tuíku **suíku
**muíki **tuíki **suíki
Ob **muiku + má **tuiku + má **suiku + má

In the absolute Auslaut *ú was maintained, whence 2sg. *tú. We
expect:

N *mú *tú --
Ob *mumé *tumé *sumé

Npl *mWétW *tWétW --
*mWét^ *tWét^ --
Ob *mWetWmé *tWetWmé *sWetWmé

Ndu *mW(y)éxW *tW(y)éxW --
*mW(y)éx^ *tW(y)éx^ --
Ob *mW(y)exWmé *tW(y)exWmé *sW(y)exWmé

After zero grade:

N *mú *tú --
Ob *mwmé *twmé *swmé

Npl *m(W)ésw *(s)wésW --
*mWéy *sWéy --
Ob *m(W)sWmé *(s)wsWmé *s(W)smé

Ndu *m(W)éxW/*m(W)íxW *(s)wéxW --
*mWéx^/*mWíx^ *sWéx^/*sWíx^ --
Ob *m(W)xWmé *(s)wxWmé *s(W)xWmé

Sequences of two consecutive labialized consonants were dissimilated,
in the case of the 2pl./du. this led to dissimilatory loss of s- (as
in the numeral *sWesWek^s > *usWek^s).

Further developments led to attested:

N -- *tú --
Ob **mwé > *mé *twé *swé

Npl *mésW *(y)úsW --
*wéy *sWéy --
Ob *n.smé *(y)usmé *(s)usmé
*nosW *wosW --

Ndu (*míh3) *(y)úh3 --
*wéh1/*wíh1 *sWéh1/*sWíh1 --
Ob *n.h3wé *(y)uh3wé *sh3wé
*noh3 *woh3 --


Clearly, not all the details have been worked out (especially where it
concerns labialization and palatalization), but I'm not pessimistic
about finding a solution for the remaining problems.

=======================
Miguel Carrasquer Vidal
mcv@...