Proto-Indo-European Phonology

13. The PIE Phonemic System

13.1. The PIE phonemes

For PIE I therefore assume a phonemic system like that given in 2.1a with slight modifications:

With these four classes of phonemes we can account for all IE developments. The phonemic structure of PIE is clear. The allophonic variation of the phonemes, however, is left with various problems. Two such problems will be sketched briefly, one for which a phonemic but no allophonic solution is offered in terms of this phonological system, another for which both a phonemic and an allophonic solution is possible. The material pertinent for describing the allophones of the laryngeals will be presented in the next chapter.

13.2. The problem of the IE ‘interdental spirants’

The first problem is the analysis of some clusters consisting of velar plus dental which Brugmann represented by velar plus interdental spirants, Gdr. I.790-3. Examples are found for the following velars; the dental element is indicated by D:

• k + D: Gk. κτάομαι ‘gain’, Skt. kṣáyati ‘rules’
  Gk. ἄρκτος, Skt. ŕ̥kṣa, Arm. arǰ, Alb. ari, Lat. ursus, Ir. art ‘bear’
• gh + D: Gk. χθών, Skt. kṣám, Lat. humus, Hitt. te-kán, Toch. tkam, kem ‘earth’
• gʷh + D: Gk. φθίνω, Skt. kṣiṇóti ‘destroys’, Toch. ktsai-. (MSL 18.24.)

Evidence in favor of velar-dental articulation may be drawn from the consistent representation in the various dialects by means of velar and dental reflexes; the dental reflexes are stops in Gk., Celt., Toch., and Hittite, spirants in the other dialects.

Proof that we are dealing with a cluster may be found in the consistent similarity of articulation of its members; the members are always alike in manner of articulation; we only find sequences like [kt], never [kd] or [kdh]. Moreover in some phonetic environments only one phoneme represents the cluster. Before a the dental element is eliminated in Gk.; compare Gk. κτείνω ‘kill’ and καίνω, χθών and χαμαί. Preconsonantally too the cluster is simplified in Gk.; compare τέκτων ‘builder’ and τέχνη ‘art’, τέκμαρ ‘goal’.

The allophones in these clusters cannot be determined more closely from unusual spellings in the dialects; reflexes of the cluster are everywhere written with symbols for reflexes of PIE dentals and velar phonemes, but the order of these varies. In Gk. the reflex of the velar usually precedes, in Hittite the reflex of the dental, in Tocharian we find both orders; in Gk. too a vase inscription gives evidence for the form Φιλοσκητ[ης] beside Φιλοκτήτης, cf. κτάομαι; and beside κτείνω we find Gort. κατα-σκευει.

Various attempts have been made to analyze the articulation of the cluster, and to find its origin. Apparently the dialect variation in representation of the dental element influenced the analysis of Brugmann; he decided that the dental member of the cluster was the aberrant one, and wrote it with symbols differing from those for the usual dentals.

Benveniste listed the older attempts at explanation, BSL 38. 139-47 (1937), and himself suggested that we should set up another series of PIE velars, comparable to the labio-velars; these he writes k^s, g^z, g^zh. Benveniste's argument is chiefly morphological. According to his theory of the root one should assume for τέκτων a root tek, and expect reflexes of a form tkeþ, but these are never attested. Likewise beside the form kþei-, cf. κτίζω, one should expect a root keþ-, but again this is not attested. Consequently Benveniste assumes unit phonemes. By the same argument one would have to assume a series of slit fricative-stop phonemes beside the simple stops; for beside the root /ste·-/ one finds no forms from set-. I consider both combinations, ‘movable s’ plus stop, and velar plus dental, to be clusters phonemically, but units morphologically.

Hammerich has attempted to establish that the plus element in the cluster resulted from the presence of a laryngeal, Laryngeal Before Sonant 15ff. Since he has merely listed the PIE forms and suggested his hypothesis, his attempt is unconvincing. To be sure, there is evidence of laryngeals in some forms with velar-dental cluster, but Hammerich has not made use of such evidence in suggesting his hypothesis.

I conclude that the allophones of the clusters have not been determined, nor their origin. Phonemically, however, the structure is clear. It never contrasts minimally with a succession of velar and dental obstruent phonemes. I therefore write the clusters attested: /kt/, /gdh/ from ghdh, /gʷdh/ from gʷhdh.

13.3. The problem of the IE palatals, velars, and labio-velars

A problem of PIE allophonic structure that has been treated variously is the relationship of the palatal and velar stops. Phonetic evidence is found for three positions of articulation in each of three manners of articulation, palatal, velar, and labio-velar; these may be written [k̑ k kʷ g̑ g gʷ g̑h gh gʷh].¹ From the PIE reconstructions of Brugmann and Walde one would assume three phonemes for each manner of articulation. In no IE dialect, however, do we find three contrasting phonemes; the greatest number in any dialect is two. The so-called centum dialects have a distinct reflex for [kʷ] and only one for [k̑] and [k]. On the other hand the satem dialects have distinct reflexes for [k̑], but [k] and [kʷ] have fallen together. Two different phonemic interpretations of the data have been made. Meillet, Introduction 91-5, assumes PIE phonemes k and kʷ, Kurylowicz, EI 1-26, the PIE phonemes k̑ and k.

We find evidence for reconstructing the following PIE phonological sequences of palatal and velar stops plus vowel; for sequences of such stops plus consonants; see Gdr. I.542–622.

• k̑e- Lat. cēnseō ‘rate’: Skt. śáṅsati ‘praises’
• k̑a- no evidence
• k̑o- rare. OHG bircha: Skt. bhūrja ‘birch’ (for g̑o-)
• ke- rare. Gk. κείρω ‘cut off’: Skt. kr̥ntáti ‘cuts’
• ka- Gk. καρκίνος ‘crab’: Skt. karkaṭa
• ko- Gk. μει̑ραξ ‘girl’: Skt. maryaká ‘young man’
• kʷe- Gk. τέσσαρες, Goth. fidwōr: Skt. catvā́ras ‘four’
• kʷa- OIcel. hualr ‘whale’: OPr. kalis ‘sheat-fish’
• kʷo- Gk. ποινή: Av. kaēna ‘punishment’.

We may conclude from the first six sequences that palatal or velar articulation varied with the articulation of the following vowel; before e the stop was fronted, before o it was back, before a it was neutral, but non-palatal.

Since e interchanged with o, there would have been an interchange of the velar allophones in various forms of one word; in verbal forms of the root /lewk-/ the velar allophone would have been found before the thematic vowel in the 1st sg., the palatal allophone before the thematic vowel in the 2d and 3d sg. A similar interchange would have been found in nominal forms, e.g. the -nt- participial forms from /lewk-/ with interchange of thematic vowels e and o.

Such a situation would remain stable as long as the conditioning features were not disturbed. When however the vowel system was disturbed, a phonemic contrast between the palatal and velar allophones would be established. Such changes in the vowel system occurred in most of the satem dialects. I assume that a contrast of palatal versus velar stops followed the vowel shift.

Because a consonant interchange was not permitted in the system of forms from one root in PIE while a vowel interchange was possible, either a palatal or a velar phoneme was generalized through a system of forms from one root. For some roots, however, we find some palatal, some velar forms, see Gdr. I.545–7. The velar allophone is found in most forms of the root /lewk-/, e.g. Skt. rócate ‘shines’, roká ‘light’, Lith. laũkas ‘pale’; but in a few forms the palatal allophone was generalized, e.g. Skt. rúśat ‘light’, OPrus. luysis ‘lynx’.

The assumption of a PIE system with only two velars is strongly supported by the distribution before /a/ and /r/. Before these phonemes we find primarily [k] and [kʷ]. Since we know that at one stage of PIE /e/ and /o/ were members of one phoneme, and have reasonable evidence that /a/ and /r/ never had one palatal, one velar allophone, we may conclude that this distribution before /a/ and /r/ represents a stage of the language older than that of [k̑ k kʷ] before /e/ and /o/.

In the centum dialects, on the other hand, where the three vowels /e a o/ did not fall together, the palatal allophones did not become phonemic.²

The relation between the development of [k̑] to a phoneme and the reduction of the PIE vowel system may be illustrated as follows:

The innovation presumably was initiated in one dialect, and spread throughout the satem dialects. After the vowel shift the palatal versus velar contrast was the fundamental opposition. The labial articulation of some of the velars became non-significant, and the labio-velars fell together with the velars.

In the centum dialects the vowels remained unshifted, and the original allophonic distribution was undisturbed. As Meillet has pointed out, we may support this conclusion about the conservatism of the centum dialects by adducing the observation from linguistic geography that an original distribution is often preserved in peripheral areas; although we cannot be certain about the distribution of the IE dialects, Tocharian, Celtic, and Gmc. seem to be languages of peripheral areas.

Except for restricted environments the allophones of the other PIE obstruents are clear, as are those of the resonants. Those of the vowels have been discussed above. In the next chapter I shall attempt to determine the allophones of the remaining class of phonemes, the laryngeals.

Footnotes

¹ Because of the parallelism in development between the voiceless, voiced, and the voiced aspirated stops, for simplification I give examples only of the voiceless stops.

² It seems at first glance as though the Gmc. development of the vowels is like that of Ind.-Ir., Baltic, and Slavic, to which I attribute the phonemicization of the palatal allophones. The threefold timbre pattern seems to have been lost too in the Gmc. dialects; for when our Gmc. materials were written down PIE /o/ and /a/, /o·/ and /a·/ had become single phonemes. The Gmc. development, however, may be dated after the operation of Grimm's Law, that is, after the PIE velar system had been rearranged. We have various kinds of evidence that the PIE vowel timbre distinction was preserved relatively late into Gmc.: the labio-velars lost their labial element before PIE /o/, maintained it before PIE /a/; the earliest Gmc. names given by Gk. and Lat. writers distinguish between /o/ and /a/ in unaccented syllables, cf. Streitberg, UG 45-7. The conditions to which I attribute the phonemicization of the palatals were not present in Gmc.