Indo-European Phonology

[Bomhard at aol.com]

Since there has been some discussion on this list regarding the reconstruction
of the Proto-Indo-European stop System and of the glottalic reinterpretation
of Indo-European consonantism, it might be helpful to review the history of
the reconstruction of that system and the reasons why the glottalic
reinterpretation was proposed in the first place.

1.  August Schleicher

Although the comparative-historical study of the Indo-European languages did
not begin with August Schleicher, he was the first to attempt, in the first
volume (1861 [4th edition 1876]) of his (in English translation) "Compendium
of the Comparative Grammar of the Indo-European Languages," to reconstruct the
phonological system of the Indo-European parent language.  Earlier scholars --
especially Rasmus Rask and Jacob Grimm -- had worked out the fundamental sound
correspondences between the various daughter languages, and the need to
reconstruct the phonological system of the parent language had been recognized
as early as 1837 by Theodor Benfey, but no one prior to Schleicher had
actually undertaken the task.

2.  The Neogrammarian Period

Schleicher's reconstruction remained the accepted standard until the late
1870's, when a series of brilliant discoveries were made in rapid succession:

1. First, there was the discovery of "The Law of Palatals" (Das
Palatalgesetz), which established the antiquity of the vowel systems found in
Greek and Latin and recognized, for the first time, that the Sanskrit vowel
system was an innovation.
2. The next major discovery was that Proto-Indo-European had syllabic nasals
and liquids.
3. Following these discoveries, the system of vowel gradation (Ablaut) became
clear, and the original patterning was worked out in precise detail.
4. Finally, Verner's Law explained several annoying exceptions to the expected
developments of the earlier voiceless stops in Proto-Germanic.  First, the
voiceless stops became voiceless fricatives in Proto-Germanic.  Then, at a
later date, these voiceless fricatives became voiced fricatives except (A)
initially and (B), in some cases, medially between vowels.  The problem was
that both voiceless and voiced fricatives appeared medially between vowels,
and the choice between voiceless fricatives, on the one hand, and voiced
fricatives, on the other hand, appeared to be entirely random.  What Verner
figured out was that the patterning was tied to the original position of the
accent -- the voiceless fricatives appeared medially between vowels when the
accent had originally fallen on the contiguous preceding syllable.  If the
accent had originally fallen on any other syllable, however, voiced fricatives
appeared.

By the end of the nineteenth century, the phonological system reconstructed by
the Neogrammarians was widely accepted as being a fairly accurate
representation of what had existed in Proto-Indo-European.  To this day, the
Neogrammarian system, or slightly modified versions thereof, commands a great
deal of respect and has many defenders.

The Neogrammarian reconstruction of the Proto-Indo-European phonological
system, which was arrived at through strict adherence to the principle that
sound laws admit no exceptions, was notable for its large inventory of stops
and its extremely small inventory of fricatives.  The stop system consists of
a four-way contrast of (A) plain voiceless stops, (B) voiceless aspirated
stops, (C) plain voiced stops, (D) voiced aspirated stops.  This system is
extremely close to the phonological system of Old Indic.  Actually, there were
two competing versions of the Proto-Indo-European phonological system at this
time:  (A) the German system (as exemplified in the works of Karl Brugmann,
for example), which was phonetically based, and (B) the French system (as
exemplified, in particular, in the works of Antoine Meillet), which was
phonologically based.  It must be pointed out that, in spite of its wide
acceptance, a small group of scholars has, from time to time, questioned the
validity of the Neogrammarian reconstruction, at least in part.

Brugmann, in particular, reconstructed five short vowels and five long vowels
plus a reduced vowel, the so-called "schwa indogermanicum", which was written
with an upside down e and which alternated with so-called "original" long
vowels.  A full set of diphthongs was posited as well.  Finally, the system
contained the semivowels *y and *w, a series of plain and aspirated spirants,
several nasals, and the liquids *l and *r.  The nasals and liquids were unique
in their ability to function as syllabics or nonsyllabics, depending upon
their environment.  They were nonsyllabic (A) when between vowels or initially
before vowels, (B) when preceded by a vowel and followed by a consonant, and
(C) when preceded by a consonant and followed by a vowel.  The syllabic forms
arose in early Indo-European when the stress-conditioned loss of former
contiguous vowels left them between two nonsyllabics.

It should be noted here that the Proto-Indo-European vowels were subject to
various alternations that were partially correlated with the positioning of
the accent within a word.  These vowel alternations served to indicate
different types of grammatical formations.  The most common alternation was
the interchange between the vowels *e and *o in a given syllable.  There was
also an alternation among lengthened-grade vowels, normal-grade vowels, and
reduced-grade and/or zero-grade vowels.

Meillet's reconstruction differs from that of Brugmann in several important
respects.  First, Meillet reconstructs only two tectal (guttural) series,
namely, palatals and labiovelars -- he does not recognize a separate pure
velar series.

Brugmann posited a separate series of voiceless aspirates for Proto-Indo-
European on the basis of an extremely small, and somewhat controversial, set
of correspondences from Indo-Iranian, Armenian, and Greek.  In the other
daughter languages, the voiceless aspirates and plain voiceless stops have the
same treatment, except that *kh appears to have became x in a small number of
examples in Slavic -- however, these examples are better explained as
borrowings from Iranian rather than as due to regular developments in Slavic.
As early as 1891, in a paper read before the Societe de Linguistique de Paris,
the Swiss scholar Ferdinand de Saussure suggested that the voiceless aspirates
might have had a secondary origin, arising from earlier clusters of plain
voiceless stop plus a following "coefficient sonantique".  This idea was taken
up by Meillet, who pointed out the great rarity of the voiceless aspirates,
noting in particular that the dental voiceless aspirate *th often appears to
be the result of aspiration of a plain voiceless dental by a following *H:  *t
+ *H > *th, at least in Sanskrit.  Current thinking on the part of a great
many linguists is that the series of voiceless aspirates reconstructed by
Brugmann and other Neogrammarians for the Indo-European parent language should
be removed, being secondarily derived in the individual daughter languages.
The main opponent of this view has been Oswald Szemerényi, who has argued for
the reinstatement of the voiceless aspirates and, consequently, for a return
to the four-stop system (plain voiceless, voiceless aspirated, plain voiced,
voiced aspirated) of the Neogrammarians.

Particularly noteworthy is Meillet's treatment of the resonants.  Here, he
considers *i and *u to be the syllabic allophones of *y and *w respectively
and classes them with the resonants, thus:  *i/*y, *u/*w, *m/*m, *n/*n, *r/*r,
*l/*l (the first member is syllabic, the second non-syllabic), that is to say
that he does not consider *i and *u to be independent phonemic entities.  The
diphthongs are analyzed by Meillet as clusters of (A) vowel plus nonsyllabic
resonant and (B) nonsyllabic resonant plus vowel.

3.  The Twentieth Century to 1970

In 1878, the young Ferdinand de Saussure attempted to show that so-called
"original" long vowels were to be derived from earlier sequences of short
vowel plus a following "coefficient sonantique".  In 1927, Jerzy Kurylowicz
demonstrated that reflexes of de Saussure's "coefficients sonantiques" were
preserved in Hittite.  On this basis, a series of consonantal phonemes,
commonly called "laryngeals", was then posited for Proto-Indo-European.  Jerzy
Kurylowicz, in particular, set up four laryngeals.  The overwhelming majority
of scholars currently accept some form of this theory, though there is still
no general agreement on the number of laryngeals to be reconstructed for
Proto-Indo-European or on their probable phonetic values.

With the reduction of the gutturals to two series, the removal of the
traditional voiceless aspirates, the reanalysis of the diphthongs as clusters
of vowel plus nonsyllabic resonant and nonsyllabic resonant plus vowel, and
the addition of laryngeals, we arrive at the system of Winfred P. Lehmann,
which consists of the contrast (A) plain voiceless stops, (B) plain voiced
stops, and (C) voiced aspirates.

Now, the removal of the traditional voiceless aspirates creates a problem
from a typological point of view.  Data collected from the study of a great
number of the world's languages have failed to turn up any systems in which
voiced aspirates are added to the pair plain voiceless stop / plain voiced
stop unless there are also corresponding voiceless aspirated stops in the
system.  This is an important point, affecting the entire structure of the
traditional reconstruction.  In order to rectify this imbalance, several
scholars have sought typological parallels with systems such as those found,
for example, in Javanese.  In these rare systems, there is a three-way
contrast, sometimes described as (A) plain (unaspirated) voiceless, (B)
voiced, (C) "voiced aspirated":  /T/, /D/, /Dh/.  However, this interpretation
is based upon a lack of understanding of the phonetics involved.  Series (C)
in such systems is, in reality, voiceless with breathy release and not "voiced
aspirated".

As we have seen from the preceding discussion, Lehmann's reconstruction is
problematical from a typological point of view.  However, from a structural
point of view, it presents an accurate analysis of Proto-Indo-European
phonological patterning.
Several scholars have proposed various solutions in an attempt to eliminate
the problems caused by the removal of the traditional voiceless aspirates.
For example, in 1964, Kurylowicz tried to show that the voiced aspirates were
not phonemically voiced.  However, this interpretation seems unlikely in view
of the fact that the daughter languages are nearly unanimous in pointing to
some sort of voicing in this series in the Indo-European parent language.  The
main exceptions are Tocharian and possibly Hittite (at least according to some
scholars).  In each case, however, it is known that the voicing contrast was
eliminated and that the reflexes found in these daughter languages do not
represent the original state.  The Greek and Italic developments are a little
more complicated:  in these daughter languages, the traditional voiced
aspirates were devoiced, thus becoming voiceless aspirates.  Then, in Italic,
the resulting voiceless aspirates became voiceless fricatives.

According to Eduard Prokosch (in 1938), on the other hand, the voiced
aspirates of traditional grammar were really voiceless fricatives.  This
interpretation seems unlikely for two reasons:  (A) as noted above, the
daughter languages point to voicing in this series in Proto-Indo-European, and
(B) the daughter languages point to stops as the original mode of articulation
and not fricatives.  This latter objection may also be raised against the
theory -- advocated by Alois Walde (in 1897) and Johann Knobloch (in 1965) --
that the voiced aspirates may have been voiced fricatives.

Next, there is the theory put forth by Louis Hammerich (in 1967) that the
voiced aspirates may have been emphatics.  Hammerich does not define what he
means by the term "emphatics" but implies that they are to be equated with the
emphatics of Semitic grammar.  Now, in Arabic, the emphatics have been
described as either uvularized or pharyngealized.  Such sounds are always
accompanied by backing of adjacent vowels.  In Proto-Indo-European, all vowels
were found in the neighborhood of the voiced aspirates, and there is no
indication that any of these sounds had different allophones here than when
contiguous with other sounds.  Had the voiced aspirates been emphatics such as
those found in Arabic, they would have caused backing of contiguous vowels,
and this would be reflected in the daughter languages in some manner.
However, this is not the case.  If, on the other hand, the emphatics had been
ejectives such as those found in the Modern South Arabian languages, the
Semitic languages of Ethiopia, and several Eastern Neo-Aramaic dialects (such
as, for instance, Urmian Nestorian Neo-Aramaic and Kurdistani Jewish Neo-
Aramaic), the question arises as to how these sounds could have developed into
the voiced aspirates needed to explain the developments in Indo-Iranian,
Greek, Italic, and Armenian.

Oswald Szemerényi was one of the first (in 1967) to bring typological data to
bear on the problem of reconstructing the Proto-Indo-European phonological
system.  Taking note of Roman Jakobson's famous remark that "...no language
adds to the pair /t/ ~ /d/ a voiced aspirate /dh/ without having its voiceless
counterpart /th/," Szemerényi reasoned that since Proto-Indo-European had
voiced aspirates, it must also have had voiceless aspirates.  Though on the
surface this reasoning appears sound, it puts too much emphasis on the
typological data and too little on the data from the Indo-European daughter
languages.  As mentioned above, there are very cogent reasons for removing the
traditional voiceless aspirates from Proto-Indo-European, and these reasons
are not easily dismissed.  Szemerényi also tried to show that Proto-Indo-
European had only one laryngeal, namely, the voiceless glottal fricative /h/.
Szemerényi does not include diphthongs in his reconstruction since their
"phonemic status is disputed".

Szemerényi's reconstruction is in fact typologically natural, and he defended
it strongly right up to his dying day (1996).  His system -- as well as that
of the Neogrammarians, it may be added -- is merely a projection backward in
time of the Old Indic phonological system.  In certain dialects of
"Disintegrating Indo-European" (specifically, in the early development of Pre-
Indo-Iranian, Pre-Greek, and Pre-Italic), such a system no doubt existed in
point of fact.

Next, there are the proposals put forth by Joseph Emonds (in 1972).
According to Emonds, the plain voiced stops of traditional Proto-Indo-European
are to be reinterpreted as plain lax voiceless stops, while the traditional
plain voiceless stops are taken to have been tense and aspirated.  Emonds
regards the voicing of the lax stops as common to a Central innovating area
and the appearance of voiceless stops in Germanic, Armenian, and Hittite as
relics.
Similar proposals were put forth by Toby D. Griffen (in 1988).  According to
Griffen, Proto-Indo-European had a three-member stop system, which he
represents as (using the dentals for illustration) *[d], *[t], *[th] (media,
tenuis, aspirata).  While this system was maintained in Germanic with only
minor changes, a series of sound-shifts in the other Indo-European daughter
languages completely restructured the inherited system.  Thus, Germanic
emerges as the most conservative daughter language in its treatment of the
Indo-European stop system.

There are other problems with the traditional reconstruction besides the
typological difficulties caused by the removal of the voiceless aspirates.
Another problem, noted in most of the standard handbooks, is the statistically
low frequency of occurrence -- perhaps total absence -- of the traditional
voiced labial stop *b.

The marginal status of *b is difficult to understand from a typological
viewpoint and is totally unexplainable within the traditional framework.  This
problem was investigated by the Danish scholar Holger Pedersen (in 1951).
Pedersen noted that, in natural languages having a voicing contrast in stops,
if there is a missing member in the labial series, it is /p/ that is missing
and not /b/.  This observation led Pedersen to suggest that the traditional
plain voiced stops might originally have been plain voiceless stops, while the
traditional plain voiceless stops might have been plain voiced stops.  Later
shifts would have changed the earlier plain voiced stops into the traditional
plain voiceless stops and the earlier plain voiceless stops into the
traditional plain voiced stops.  In a footnote in his 1953 BSL article
entitled "Remarques sur le consonantisme semitique", Andre Martinet objected
to this "musical chairs" rearrangement.

"Since there are extremely few examples of the Common Indo-European phoneme
reconstructed 'analogically' as *b, it is tempting to diagnose a gap there as
well, as did the late Holger Pedersen...  But, instead of assuming, as did
Pedersen, the loss of a Pre-Indo-European *p followed by a musical-chairs
[rearrangement] of mediae and tenues, one should be able to see in the series
*d, *g, *gw the result of evolution from an earlier series of glottalics,
without labial representative."

This appears to be the first time that anyone had proposed reinterpreting the
plain voiced stops of traditional Proto-Indo-European as glottalics.
Martinet's observation, however, seems to have influenced neither Gamkrelidze
and Ivanov nor Hopper, each of whom arrived at the same conclusion
independently of Martinet as well as independently of each other.

In the preceding discussion, only the more well-known counterproposals were
mentioned, and only the briefest of explanations were given.  More details
could easily have been given.  Insights gained from typological studies, for
example, could have been used to strengthen the arguments:  no phoneme stands
alone; it is, rather, an integral part of the total system.  Each and every
phoneme is tied to the other phonemes in the system by discrete
interrelationships -- to disturb one phoneme is to disturb (at least
potentially) the entire system.  This is basically the message that Jakobson
and Martinet were trying to bring home.  All too often, this message is
ignored.  Moreover, the interrelationships are not only synchronic, they are
diachronic as well.

4.  The Glottalic Theory

Discovery -- perhaps "rediscovery" would be a better term since Martinet's
insightful remarks first appeared in 1953 -- of what has come to be known as
the "Glottalic Theory" came from two separate sources, each working
independently.  On the one-hand, the British-born American Germanist Paul J.
Hopper hit upon the notion that Proto-Indo-European may have had a series of
glottalized stops while he was a student at the University of Texas and taking
a course in Kabardian from Aert Kuipers.  Hopper went on about other business
after graduation, waiting five years before putting his ideas into writing.
On the other hand, the Georgian Indo-Europeanist Thomas V. Gamkrelidze, a
native speaker of a language containing glottalics (Georgian), had been
investigating the typological similarities between Proto-Kartvelian and Proto-
Indo-European.  It did not take Gamkrelidze long to realize the possibility
that Proto-Indo-European might also have had glottalized stops.  Gamkrelidze,
in a joint article with the now-immigrated Russian Indo-Europeanist Vjacheslav
V. Ivanov, was the first to make it into print (in 1972).  Hopper might have
beat them into print had his paper on the subject not been rejected by the
journal Language.  He was then obliged to search for another journal willing
to publish his views, which finally happened in 1973.  Then, in 1973,
Gamkrelidze and Ivanov published a German language version of their 1972
paper.

In his 1973 paper, Hopper proposed reinterpreting the plain voiced stops of
traditional Proto-Indo-European  (*b, *d, *g, *gw) as glottalized stops
(ejectives), that is, (*p'), *t', *k', *k'w respectively, because the
traditional plain voiced stops "show many of the typological characteristics
of glottalized stops (ejectives), e.g. they are excluded from inflectional
affixes, they may not cooccur with another in the same root, etc."  Hopper
also reinterpreted the traditional voiced aspirates as murmured stops.

Gamkrelidze and Ivanov also reinterpret the traditional plain voiced stops as
ejectives, but, unlike Hopper, they reinterpret the traditional plain
voiceless stops as voiceless aspirates.  They make no changes to the
traditional voiced aspirates.  They point out, however, that the feature of
aspiration is phonemically irrelevant in a system of this type.  In an article
published in 1981, Gamkrelidze claims that such a system exists in several
modern Eastern Armenian dialects (however, this is challenged by the Armenian
scholar Gevork B. Jahukyan in a 1990 rebuttal).

Many of the points discussed above by Gamkrelidze were also noted by Hopper,
in particular the root structure constraint laws.  Hopper also discusses
possible trajectories of the new system in various Indo-European daughter
languages.

The system of Gamkrelidze, Hopper, and Ivanov has several clear advantages
over the traditional reconstruction of the Proto-Indo-European stop system:

1. Their reinterpretation of the traditional plain voiced stops as glottalics
(ejectives) makes it easy to account for the fact that the phoneme
traditionally reconstructed as *b was highly marked in the system, being
characterized by an extremely low frequency of occurrence (if it even existed
at all).  Such a low frequency distribution is not characteristic of the
patterning of the voiced labial stop /b/ in natural languages having a voicing
contrast in stops, but it is fully characteristic of the patterning of the
labial ejective /p'/.
2. Not only does the reinterpretation of the traditional voiced stops as
ejectives easily account for the frequency distribution of these sounds, it
also explains the fact that they were used only very infrequently in
inflectional affixes and pronouns, since this type of patterning is
characteristic of the way ejectives behave in natural languages having such
sounds.
3. For the first time, the root structure constraint laws can be credibly
explained.  These constraints turn out to be a simple voicing agreement rule
with the corollary that two glottalics cannot cooccur in a root.  Hopper cites
Hausa, Yucatec Mayan, and Quechua as examples of natural languages exhibiting
a similar constraint against the cooccurrence of two glottalics.  Akkadian may
be added to this list as well if we take Geers' Law to be a manifestation of
such a constraint.
4. The so-called Germanic and Armenian "consonant shifts" (in German,
"Lautverschiebungen"), which can only be accounted for very awkwardly within
the traditional framework, turn out to be mirages.  Under the revised
reconstruction, these branches (along with the poorly-attested Phrygian as
well) turn out to be relic areas.

In 1984, Gamkrelidze and Ivanov published their monumental joint monograph (an
English translation of this work has since been published by Mouton de Gruyter
[in 1995]).  As is to be expected, this massive work (2 volumes, 1,328 pages)
contains the most detailed discussion of the Glottalic Theory that has yet
appeared.  Gamkrelidze and Ivanov's book also contains trajectories of the
revised Proto-Indo-European phonological system in the various Indo-European
daughter languages, original proposals concerning the morphological structure
of the Indo-European parent language, an exhaustive treatment of the Proto-
Indo-European lexicon, and a new theory about the homeland of the Indo-
Europeans (they argue that the Indo-European homeland was located in eastern
Anatolia in the vicinity of Lake Van).  One of the most novel proposals put
forth in the book is that Proto-Indo-European may have had labialized dentals
and a labialized sibilant.  Gamkrelidze and Ivanov also posit postvelars for
Proto-Indo-European.

The Glottalic Theory has attracted a good deal of attention over the past two
decades and has gained widespread -- though not universal -- acceptance.