Version of 2008-06-18
Wersja polska • Bilanguage version • Wersja dwujęzyczna
The following article refers arguments of the followers of the glottalic theory as well as critique made by defenders of the classic views. Where [GJ] is given, the author of this article is the creator of the described hypothesis.
Professor Charles Barrack will be so kind as to accept my acknowledgements for sending me his works on the glottalic theory.
Classical historical grammar reconstructs three stop series for the PIE language:
These three standard series can be marked with symbols T, D, Dʱ. Such a reconstruction is based partially on Greek and partially on Sanskrit. In older works the fourth series Tʰ of voiceless aspirates (pʰ, tʰ, ḱʰ, kʰ, kʷʰ) was also considered but finally that idea was given up.
However, there also exist other reconstructions of the PIE consonantism. The most known of them is so called glottalic theory. Some elements of it were already postulated by Holge Pedersen, André Martinet and Morris Swadesh but the theory in its mature form was formulated in '70s. Its creators have been, independently, the Soviet scholars Thomas Gamkrelidze and Vyacheslav Ivanov in 1972, as well as the American linguist Paul Hopper in 1973.
| s. | An | To | Gr | It | Ce | Ge | Ar | Al | BS | Ir | In | PIE | ||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| cl | g1 | g2 | ||||||||||||
| 1 | T: | T | T | T | T | Θ | Tʰ | T | T | T | T | T | Tʰ | T: |
| 2 | T | T | D | D | D | T | T | D | D | D | D | D | Tˀ | Tˀ |
| 3 | T | T | Tʰ | Tʰ | D | Ð | D | D | D | D | Dʱ | Dʱ | Dʱ | Tʰ |
Abbreviations and symbols:
s – series;
language groups: An – Anatolian, To – Tocharian, Gr – Greek, It – Italic, Ce – Celtic, Ge – Germanic, Ar – Armenian, Al – Albanian, BS – Balto-Slavic, Ir – Iranian (with Kafir), In – Indic (with Dardic);
PIE – reconstruction for the Proto-Indo-European language: cl – classical (standard) model, g1 – the glottalic theory in the version of Gamkrelidze and Ivanov, g2 – version of Kortland.
T: – strong (fortes) voiceless stops in opposition to weak (lenes) ones,
T – voiceless stops,
D – voiced stops,
Tʰ – voiceless aspirates,
Dʱ – voiced aspirates,
Tˀ – voiced ejectives (abruptives),
Θ – voiceless spirants,
Ð – voiced spirants.
A similar chart can be found ex. in Barrack's works. The opposition of force in Anatolian (T: – T) has accepted here after classical reconstructions (Barrack accepts Tʰ – T). The symbol T need not mark weak consonants: it is so in Anatolian and Armenian, but not in ex. Greek (where T is phonologically strong).
In Gamkrelidze and Ivanov's version the reconstruction of the PIE system of stops was as follows:
It was also assumed that sometimes plain voiceless stops p, t, ḱ, k, kʷ used to happen as allophones of voiceless aspirates, and similarly plain voiced stops b, d, ǵ, g, gʷ used to happen together with aspirated voiced stops as their allophonic variants. In other words, according to the glottalic theory the 3 series of stops were Tʰ ~ T, Tˀ, Dʱ ~ D.
The non-aspirated allophones occurred in roots where another aspirate was present because of a rule that prohibited more than one aspirate in the same root. It depended on a dialect which stop lose aspiration: the first one in Indic and Greek, and the second one in Italic.
The proposed system of stops did not result from methods of reconstuction which are used in historical linguistics but from dependencies which can be observed in languages of the world, called typological universals. Actually it was based on Armenian consonantism. The differences when comparing to the classical view where that it contained ejective stops (just like some Armenian dialects).
For some time the glottalic theory was enough popular, and even it was treated as the next step in the development of the historical phonetics of Indo-European languages which follows the laryngeal theory. However it appeared with time that many arguments of glottalicists are easy to refute, and the theory itself must have been modified under pressure of facts.
We will present various arguments of followers of the glottalic theory below, as well as their detailed analysis.
It was stated that the putative inventory of stops in PIE, consisting of series of voiceless, voiced and voiced aspirated (breathy voiced) stops, is not plausible as linguistic typology does not know examples of similar systems. On the other hand the system of stops proposed by the glottalic theory is as if common among the languages of the world.
In fact Roman Jakobson was wrong when asserting that no language is known which would have a system of stops like the one postulated for PIE. Actually there are few languages which have voiced aspirates but no voiceless aspirates. Robert Blust even showed that a system of voiceless, plain voiced and voiced aspirated stops exists in Kelabit, a language of Borneo. Thus Jakobson's argument is a simple urban legend, even if it is true that the standard reconstruction contains a typological rarity.
Nevertheless one should have in mind that the system proposed by the glottalic theory is not any better in this respect. One can even say that it remains even more implausible. Namely, in order to explain known facts from various languages, Gamkrelidze and Ivanov asserted that the 3rd series of stops contained plain voiced and aspirated voiced stops as allophonic variants: D ~ Dʱ. Even this fact would be a typological oddity.
Which is more, there is not a single instance of existing ejective consonants together with aspirated voiced stops in languages of the world.
Early reconstructions of the Proto-Indo-European language slavishly copied structures known in Sanskrit, which was acknowledged to be the most conservative and the less distant from the protolanguage. Among others, existence of 4 series of stops: T, Tʰ, D, Dʱ was assumed in PIE, just like in Sanskrit. Existence of the Tʰ series has not been confirmed ultimately: Sanskrit voiceless aspirates come mainly from PIE sequences of a voiceless and a laryngeal, and their counterparts in other IE languages are, as a rule, plain voiceless stops. The voiceless aspirate series was removed from the reconstruction of the protolanguage consonantism, but the voiced aspirated (breathy voiced) consonants remained in it, causing typological problems described above.
However, the objection of copying Sanskrit structures slavishly was recognized as still valid. Particularly it concerned distinguishing of series of voiced and voiced aspirates (D : Dʱ). Actually this kind of opposition occurs only in Indic, from among all IE languages.
This argument does not seem to be too strong. The following argumentation is totally wrong: if some elements of the reconstruction that were based on Sanskrit appeared inapt, then basing on this language is a mistake in general. It is not so, and thus we should not reject Sanskrit data.
The next argument of glottalicists is again of a typological nature. Namely, it is pointed out that if a voiced stop is missing from a phoneme inventory, it would normally be the velar g (or uvular ɢ) that is missing. On the other hand, if a voiceless stop is missing in the system, the labial p is the most likely candidate. And similarly pˀ is cross-linguistically a much rarer sound than the other ejectives.
A labial consonant of the 2nd series, reconstructed as b in the standard theory, and as pˀ in the glottalic theory, was probably totally absent in PIE. Thus the glottalic proposal would be close to the typological universal.
The next argument of glottalicists is the fact that no IE root in the standard reconstruction contains two plain voiced stops. The lack of the roots of the type *deg- cannot be explained in any way because it is not known what factor would prevent coexistence of two voiced stops.
On the ground of the glottalic theory these roots would have the structure of the type *tˀekˀ-, thus they would have two ejective consonants. The lack of them in PIE would have typological explanation again: roots with two ejectives do not occur in languages in the world.
Not only roots of the type *deg- but also roots which contain both voiceless and voiced aspirated stops, i.e. roots of the types *tegʱ- i *dʱek-, are absent in the classic reconstruction. If the plain voiced stops were not voiced but abruptive, the voiced aspirated stops were the only voiced stops. In order to explain the absence of the types *tegʱ- i *dʱek-, Hopper formulated a constraint that two nonglottalic stops in one roots must agree in voicing. And because the ejective stops are outside the opposition of voicing, they were excluded from the constraint.
The argument of glottalicists seems to be strong apparently but a deeper analysis bears doubts again. Namely, first of all, the glottalic theory attempts to explain the absence of roots of the type *deg- (basing on universals), it is not engaged completely in an explanation of the absence of roots of the types *tegʱ- i *dʱek-. The statement that voiced stops do not occur with voiceless stops is not an explanation but a description of an observation rather. In the same time several explanations of these observations were proposed on the base of the classical theory.
First of all, it was noticed [Erhart 1982] that the total frequency of plain voiced and aspirated voiced stops is almost the same as the frequency of voiceless stops. Thus it was assumed that only two series of stops existed in the earliest phase of the protolanguage, and only next the 2nd series split into plain voiced and aspirated voiced. Then the four possible original types of roots which contained two stops developed as follows:
| phase I | phase II |
|---|---|
| TAK- | *tek- |
| TAG- | *teg- |
| DAK- | *dek- |
| DAG- | *dʱegʱ- |
In the reconstruction of the phase I, the symbols T, K denote stops of the first series (voiceless), the symbols D, G – stops of the second series (voiced), and A is a symbol of a vowel.
This hypothesis explains the absence of roots of the types *deg- (instead of them we have *dʱegʱ-), *tegʱ- (instead of them: *teg-) and *dʱek- (instead of them: *dek-), as well as the rarity of the types *degʱ- and *dʱeg- (which should not be present at all). However, it has two crucial faults which vote for rejecting it:
Absence of roots of the types *tegʱ- and *dʱek- can also be explained on the ground of the classical theory – paradoxically – when applying Gamkrelidze and Ivanov's supposition on the prohibition of existence of two aspirates in the same root.
In order to do it, we should suppose that the 1st series of stops really contained voiceless aspirates originally, while rejecting all suppositions on ejective articulation of the 2nd series naturally (thus instead of the classical system *t – *d – *dʱ we should reconstruct *tʰ – *d – *dʱ at least for the older phase). On a certain stage of the development of the IE protolanguage a loss of aspiration of the voiced stop took place in original roots of the types *tʰegʱ- i *dʱekʰ-, and as a result the voiced aspirated stops went from the 3rd series to the 2nd series (naturally under the condition that it contained voiced stops of the type *d and not ejectives of the type *tˀ). In the stage which is presented in standard reconstructions of PIE, the aspiration by voiceless stops had declined in all positions, at least in most IE dialects (ex. not necessarily in Proto-Germanic). And finally in particular dialects the aspiration used to decline in roots with two voiced aspirates:
| phase I | phase II | phase III | phase IV | |||
|---|---|---|---|---|---|---|
| Indic | Italic | Greek | ||||
| I | II | |||||
| *tʰegʱ- | *tʰeg- | *teg- | *teg- | *teg- | *teg- | *teg- |
| *dʱekʰ- | *dekʰ- | *dek- | *dek- | *dek- | *dek- | *dek- |
| *dʱegʱ- | *dʱegʱ- | *dʱegʱ- | *degʱ- | *dʱeg- | *tʰekʰ- | *tekʰ- |
We are not sure whether stops of the 1st series were aspirated for real, but we know for sure that the presence of two aspirates were not allowed actually. In addition, in Indic and Italic what declined was aspiration of voiced (see the chart) whereas in Greek it was aspiration of voiceless (already after devoicing of old voiced aspirates).
The next hypothesis calls observations of glottalicists in question.
As it can be seen from the above analysis, glottalicists' arguments are not serious. Some of them are of seeming character, others point at problems which can be explained on the basis of the classical theory, often better than when accepting assumptions of the glottalic theory. Moreover there exist facts which cannot be explained by the glottalic theory at all.
As it was mentioned above, the glottalic theory has not been formulated on the base of comparative methods of historical linguistics but of typological, and even statistic considerations. Direct evidence is absent, and even if it is not required unconditionally (for instance the classical theory has no direct evidence for gʷʱ – a labiovelar voiced aspirated stop), one cannot say that direct indications occur either. These are not typological universals, which are seldom absolute. Such an approach is methodologically suspicious, and it may even be considered to be a methodological error.
The glottalic theory has a problem to explain how the stop systems of particular IE dialects were derived from the system existing in the parent language. If PIE had a typologically unstable system, like the classical T – D – Dʱ, then it might be expected to evolve into different more typical systems in the various daughter languages, which is what one finds. But if PIE had a typologically common and thus stable system, which is postulated by glottalicists, it might be expected to have been preserved in at least some of the daughter languages, which is not the case: no IE language preserves putative abruptive consonants. Moreover it should be explained for what reason that stable system suffered multidirectional changes. The glottalic theory does not give the needed explanation.
In Tocharian all the old system collapsed and all three series merged into one. In Anatolian a binary opposition of strength (fortis – lenis, T: – T) arose in the place of the old ternary opposition. Similarly, a typologically frequent binary opposition of voicing (T – D) developed in Celtic, Albanian, Balto-Slavic and Iranian (and perhaps also in Ancient Macedonian). In Indic the old system preserved as a whole but the voiceless aspirate series was added for more symmetry. Greek and Italic devoiced the voiced aspirate series to a more common aspirate voiceless one. Finally, chain shifts of all three series happened in Germanic and Armenian.
There exist several trials to defend the glottal theory against this.
If PIE had not true ejective consonants in its inventory but ex. preglottalized ones, the system might be less stable and that is why it evolved in daughter languages.
Even stable systems change, and objections arising from theoretical considerations cannot overturn a reconstruction deduced by comparing the evidence of the daughter languages.
However it is problematic whether the glottalic theory delivers a reconstruction arising from comparing testified forms.
In five groups of Indo-European languages (Anatolian, Celtic, Albanian, Balto-Slavic, Iranian) identification of the series D with Dʱ took place, with the series T preserved distinct. In the same time only the series T with D or T with Dʱ did not mix together in any group (in Tocharian all three series identified). The facility of merging of the 2nd and the 3rd series may be an argument for that they differed with only one distinctive feature (aspiration).
It is true that stops of the 2nd and the 3rd series had lower frequency than those of the 1st series, and thanks to it their opposition was less loaded functionally. But one should doubt that it was a sufficient reason of the facility of their mixing.
According to the traditional conception, during their development some IE languages underwent the process called the consonantal shift. Germanic and Armenian belong to this group mainly. Among others, on the place of putative IE voiced stops, voiceless stops occur in both instances.
The glottalic theory inverts the situation totally. According to it the state in Germanic and Armenian is original (or close to original in any case), whereas the actual shift took place in those language groups where we have voiced stops on the place of ejective stops today, that is in ex. Indo-Iranian, Balto-Slavic, Italic or Greek.
In the classical theory all IE languages have been divided into 2 groups according to development of palatovelar consonants ḱ, ǵ, ǵʱ. In the first group plain velars k, g, gʱ occur today while affricates or spirants occur in the second group (ex. ¶, j, h in Sanskrit, s, z in Slavic). These groups have been termed kentum and satəm, after names of the numeral 100 in Old Latin and Avestan.
According to glottalicists the division of IE languages into two groups taihun and decem is important. These terms were created by Hopper in 1981 from names of the numeral 10 in Gothic and Latin. The old ejective tˀ in taihun languages (Anatolian, Tocharian, Germanic and Armenian) preserves as the voiceless t while as the voiced d in decem languages.
The cradle of Indo-Europeans should be searched for in areas where relatively least changed languages occur today. The other languages are said to have changed during migration from that area and taking new territories. Thus the differentiation model basing on the glottalic theory is completely different that the one elaborated for the classical theory. Glottalicists search for the IE homeland mostly in Armenia, whereas the traditional view points at different areas, most often at Ukrainian and Russian steppes.
A list of printed bibliography can be found here.