Proto-Indo-European Phonology

14. The Allophones of the Laryngeals

14.1. Necessity of determining allophones

In the preceding chapters reflexes of laryngeals have been examined and material assembled in favor of the laryngeal theory. The investigation has been primarily on the phonemic level. It was directed at the role of the laryngeals in the phonological structure of PIE and of their reflexes in the phonological structures of the various dialects. I shall now examine the material pertinent for determining the allophones of the laryngeals.

As Sturtevant has stated, IHL 15, the term ‘laryngeal’ is a traditional term; it is now used with a meaning as far removed from its etymological meaning as is the term ‘guttural,’ which is used by some linguists in the sense ‘palatals and velars.’ In studies of PIE the term laryngeal is used to refer to phonemes which have no direct reflexes in the dialects. These phonemes may have had glottal, or laryngal, articulation, but few Indo-Europeanists have assumed such articulation for all laryngeals.

Some proponents of the laryngeal theory have consistently attempted to identify the laryngeals phonetically. Möller was the first of these, Eng. St. 3.151 (1880). Of contemporary Indo-Europeanists holding the laryngeal theory Sturtevant indicated by the symbols he chose the primary allophones of each laryngeal phoneme. Others, following Saussure, prefer to deal with laryngeals only on the phonemic level; Kurylowicz draws his only inference about allophones of laryngeals from assimilation phenomena, and catalogues the laryngeals with non-phonetic symbols.

It is of course important to deal not only with phonemic but also with phonetic units. Our only justifiable reason for restricting discussion of laryngeals to the phonemic level would be lack of information. Such information has gradually been assembled and should give us a basis for at least a general description of the laryngeals.

14.2. Evidence in Hittite

The potentially most promising evidence, the evidence from orthographical representation in Hittite, can at present be only imperfectly utilized. Cuneiform symbols transcribed with ḫ are found in some Hittite cognates of IE forms containing laryngeals. While Hittite orthography was extremely important in establishing the laryngeal theory, it has not provided much information about the phonetic units symbolized. For the description of Hittite is by no means complete;¹ until it is, our use of Hittite for historical purposes will be severely limited.

With current theories about Hittite phonology we can find in Hittite evidence for at best two laryngeals. In some words we usually find ḫ written singly, in other words it is usually repeated. Comparing this orthographical pattern with that of the stops Sturtevant concluded that single ḫ represents a voiced sound, ḫḫ a voiceless sound. Since the ḫ symbols represent velar spirants in the languages with which Hittites were in contact, Sturtevant assumes that ḫḫ represents a voiceless velar spirant, ḫ a voiced velar spirant, supporting this conclusion with the ḫ : r confusion in Muršilis Sprachlähmung. This conclusion seems valid but is not wholly certain. For the orthographic pattern may be correlated with vowel patterning. Hendriksen, BHL 38-42, found ḫ chiefly after repeated vowels, ḫḫ after vowels written singly, and concluded that the repetition of ḫ, like that of l m n r, gives us no information about the consonant. Unless we find further evidence in support of Hendriksen's thesis, Sturtevant's theory is more convincing than is Hendriksen's. Sturtevant's theory is supported by Hurrian orthographical patterns; to judge by Hurrian orthography, which follows the same patterns as the Hittite orthography, ḫ : ḫḫ should be correlated with obstruents, not with the resonants.

Much work, however, needs to be done on the Hittite system of orthography. The Hittite orthographic patterns may not be clarified until we obtain further knowledge of Hurrian, a language which is not as well known as is Hittite, and until the history of the spread of the cuneiform symbols is clarified. Until such information is obtained I shall accept Sturtevant's interpretation of the ḫ : ḫḫ orthography. Moreover until such time we can find no further evidence for laryngeals in Hittite orthography.

The suggestion has twice been made that Hittite orthography marks another laryngeal; spellings like e-eš- have been interpreted as indicating initial glottal stop, i.e. [ʔeš-]. One suggestion was made in a posthumous paper of Sapir's, Lang. 15.180 fn.1, another in Sturtevant's report of a conversation with Götze, Lang. 18.181 fn.2. Neither suggestion has been amplified; Sapir indicated that he planned to develop in a later paper his reasons for this interpretation, but no collection of material has been published for, or against, the suggestion.

14.3. Attempts to determine allophones in IE : HS studies

Möller, the first Indo-Europeanist to try to identify the laryngeals phonetically, based his identification on the relationship between the Indo-European and Hamito-Semitic languages. Since one of the chief differences between the consonant systems of PIE and PHS was the presence of laryngal phonemes in PHS, and since these subsequently were lost in some Semitic languages with compensatory vowel lengthening, Möller suggested that Saussure's hypothecated phonemes were glottal sounds. He hesitantly defined A as a voiced glottal stop, E (/ʔ/) as a voiceless glottal stop, O (/γ/) as a laryngal r, Eng. St. 3.151. In the same year Sweet suggested the following articulation:² A, ‘the glottal r, or voiced glottal trill’; O (/γ/), ‘the same labialized, the Danish r’; E (/ʔ/), ‘A palatalized.’ Möller modified his description in PBB 7.492, defining A as similar to the Semitic alef, a ‘voiceless guttural stop,’ and E as probably the voiced counterpart.

Möller later assumed two more laryngeals, see IHL 16; he changed both his symbols and phonetic descriptions, assuming in 1917 that all of the laryngeals had glottal articulation. His earlier phonetic descriptions, however, are those which have been maintained by Indo-Europeanists who identify the laryngeals phonetically. Sapir, for example, defines the four laryngeals as follows, Lang. 14.269: “ʔ (a glottal stop followed by e-timbre of full grade vowel in its primary form), ʔ̣ (another glottal phoneme followed by a-timbre of full grade vowel in its primary form), x (presumably a voiceless velar spirant = ḫ-, -ḫḫ- of Hittite), and γ (presumably a voiced velar spirant, Arabic ‘ghain’, = ḫ-, -ḫ- of Hittite).” In a posthumous article, Lang. 15.181, Sapir's ʔ̣ is more narrowly defined as a ‘glottal stop with velar timbre’; the other descriptions are unchanged. These descriptions are those given by Sturtevant in IHL; later, Lang. 24.259-61, he uses the symbol h instead of ʔ̣, presumably to indicate a glottal fricative rather than a stop.

The foundation for description of laryngeals from IE-HS relationships is weak, for we are not yet able to reconstruct the early stages of IE and HS, to say nothing of the common source. To illustrate the possible variation in interpretation of our data we may note that Cuny defined the three laryngeals he assumed as palatal (ç), velar (A̱), and labio-velar (γ) resonants, RHA 5.38-9. I conclude that we will again arrive at a more useful formulation by analysis of definable reflexes in the IE dialects than from tenuous material drawn from hypotheses of prehistoric relationships about which we have no accurate descriptive analyses. When we begin from attested forms we of course run the risk of reconstructing reflexes of laryngeals, not the laryngeals; for example, the ‘laryngeals’ which survived into Ind.-Ir. after voiceless stops may be different from their etyma which produced phonological changes in PIE and pre-IE. In view of the problematic laryngeals arrived at by comparison of IE with HS a reconstruction of such reflexes from attested forms is preferable to derivation from somewhat dubious reconstructions.

In drawing general conclusions about the articulation of the laryngeals I assume that there was continuity in the structural bases of the IE phonological system. The phonological structure itself was changed. But the patterns of change were similar in pre-IE, PIE, and the early dialects, more similar than those between the early dialects and PIE, and those of some non-IE languages like Chinese or Nootka. Thus I would not expect to find characteristic pitch patterns such as those of Chinese for each word in PIE, but rather a use of pitch like that in early Greek. And though a glottal stop is assumed for PIE and pre-IE I should not expect glottalized stops for pre-IE like those of Nootka. The bases for such assumptions have been fully discussed by Sapir in his book, Language 157-82 (New York, 1921). Such assumptions are legitimate when used as a guide in detecting developments, rather than as explanations for them.

14.4. Evidence in the IE dialects for continuant articulation

A phonetic process attributed to all laryngeals is lengthening of preceding vowels that are not homorganic, in the pattern -eXC. Such lengthening occurs in IE dialects upon loss of continuants.

In Skt. such lengthening is found when the continuant is a voiced sibilant, e.g. nīḍá < *nizdo- ‘nest’, rūḍhá < *ruẓ̌ḍha- < *rugdha- ‘climbed’, voḍhum < *vaẓ̌ḍha- < *vagdha- ‘drive’.³

In Gk. too such continuants are voiced sibilants, e.g. Ion.-Att. εἰμί from *ἐσμι ‘I am’, τρήρων from *τρασρων ‘cowardly’.

In Ital. such continuants are voiced sibilants, r, and nasals, e.g. Lat. nīdus ‘nest’ < *nizdos, svāsum < *svarssom ‘black’, equōs < *equons ‘horses’. In late Old French voiceless spirants are lost with compensatory lengthening, e.g. testa > Mod. Fr. tête ‘head’. Cuny cited this development, RP 2.103, to support the plausibility of the pre-IE changes. Since it is generally assumed that testa > *tehte > tête this development seems very like that found for laryngeals. Cuny also cited similar developments in Semitic: Cl. Arab. ra's > Mod. Arab. rās.

In Gmc. such continuants are voiced sibilants, e.g. OS mēda, cf. Goth. mizdō ‘reward’, velar spirants, OHG mālōn < mahlōn ‘summon before a court’, and nasals, PGmc. aηχ, iηχ, uηχ > āχ, īx, ūχ, e.g. Goth. fāhan, OE fōn ‘capture’, Goth þeihan ‘thrive’, Goth. þūhta ‘seemed’. It may be noted that the nasalized a in OE was raised as well as lengthened, and fell together with ō rather than with ā. We do not. find such changes after lengthening of short vowel upon loss of laryngeal in pre-IE. In British English such continuants are voiced spirants, r, e.g. [ha·d] hard, [hɔ·d] hoard, [hə·d] heard. See Gdr. I.804-7 for further examples of such developments, and references.

These developments parallel those that we hypothecate for pre-IE where /e· a· o·/ resulted from a sequence of /eXC/. As a result I assume that the allophones of pre-IE X had continuant articulation.

14.5. Evidence for fricative articulation

Another phonological feature common to all laryngeals is their role in syllables. From the patterns which Edgerton found we may draw two conclusions: 1. from patterns like tekt-iy-, tēt-iy- < teXt-iy- I conclude that laryngeals had the same effect as obstruents in determining the following allophone of resonants; 2. the patterns also show that laryngeals did not undergo the allophonic variation of resonants. See 2.2.

IE root structure gives us further evidence that laryngeals pattern differently from resonants, and from obstruents. We find bases of the following structure: (the patterns are arranged in order of frequency; the initial is not significant)

• CeRC e.g. /bʰeyd-/ Skt. bhédati ‘splits’
• CeRX e.g. /pewX-/ Skt. puṇā́ti ‘purifies’, Inf. pavitum
• CeXR e.g. /leXw-/ Lat. lāvit ‘washed’
• CeXC e.g. /peXs-/ Lat. pāscō ‘pasture’
• CeCX e.g. /metX-/ Skt. mathnā́ti ‘shakes’, pptc. mathitá.

Since we do not find bases of the structure CeCC, and only rarely CeCR,⁴ the laryngeals show a greater variety of patterning than either obstruents or resonants, and must be classed in a separate structural set. IE root structure shows progression from most open sound at the peak of the syllable to sounds of greater closure, e.g. CeRC, CReC, except for the one fricative, /s/. According to these sequences the laryngeals patterned like fricatives.⁵

From their role in IE root structure and in the patterns determined by Edgerton I assume that the allophones of laryngeals were neither oral stops nor resonants, but fricatives.

14.6. Further evidence for relatively open articulation

Continuant articulation may be supported by the role of laryngeals in Homeric meter. Laryngeals plus ρ, λ make position in Homer. The same patterning is found for σ and Ϝ before ρ, not so consistently for voiceless stops and ρ, cf. Munro, Hom. Gram. 342-5. We may note further that such lengthening is more frequent before velar voiceless stop plus ρ, than before dental and labial stops plus ρ. Although this evidence by itself is not weighty, it supports the assumption of continuant articulation for the laryngeals. A similar support for this assumption is the lengthening of w and j before laryngeals in Gmc.; such lengthening is found elsewhere in Gmc., especially in the WGmc. dialects, before resonants.

These are the general criteria which may be used in broadly defining the allophones of laryngeals. They point to a relatively open consonantal articulation.

14.7. Evidence for voice

We may conclude from their reflexes in combination with initial ρ, λ, μ, ν in Gk. that three of the laryngeals were voiceless. The reflex of σ plus ρ, λ, μ, ν is written ῥ-, λ-, μ-, ν-, but early inscriptions give evidence that the resulting resonant was unvoiced. (Pron. of Gk. and Lat. 61-4.) Moreover the reflex of /sw-/ and /sy-/ was ʽ . Sapir pointed out that the Gk. reflexes of PIE resonants after three of the laryngeals, /h x ʔ/, were the same as those after PIE /s/. Since the change in the resonants was voicelessness, and since this was produced by one voiceless continuant, σ, it may be assumed that the other continuants which produced the change were also voiceless.

The assumption of voiceless continuant articulation for one of these three laryngeals, /x/ is supported by Hittite orthographic evidence. Sturtevant has demonstrated that the Hittite reflexes of /x/ were voiceless, those of /γ/ voiced velars of similar articulation; the use of the ḫ-signs points to a velar fricative articulation. The orthographical evidence in Hittite for voiceless and voiced counterparts is supported by a phonological observation made by Kurylowicz. In some forms of the root /po·-/ < /peγ-/ ‘drink’, the p is voiced, e.g. Skt. píbati, presumably by assimilation to the neighboring laryngeal. Such voicing is found in PIE when the phoneme in question is a member of a structural set that contrasts with another structural set without voice; thus the resonants, though voiced, do not produce such voicing. We may therefore assume that the allophones of /x/ were voiceless velar spirants, those of /γ/ voiced velar spirants.

The identification of the allophones of /x/ and /γ/ as velar fricatives is supported by developments in Gmc.

This description is further supported by reflexes in Gk. In Gk. the reflex of PIE /γy-/ is ζ, of /xy-/ ʽ . Since other PIE etyma of ζ belong to a structural set contrasting with voiceless phonemes, the same assumption may be made for /γ/.

We find that vowels contiguous with /x/ have an open, relatively back articulation. Vowels coalescing with /γ/ on the other hand have a relatively high back articulation, presumably with lip rounding. Because of this vowel articulation I assume that allophones of /γ/ had labial articulation, in addition to velar fricative articulation, and I describe them more fully as labio-velar voiced fricatives. (Evidence in morphological categories in favor of this description has not yet been explored.)

The chief allophone of /x/ then is [χ], of /γ/ [γʷ].

The chief allophone of /h/ may be most readily determined from its reflexes in Skt.; /p t k/ plus /h/ became Skt. /ph th kh/. Since these reflexes fell together with unvoiced reflexes of PIE /bʰ dʰ gʰ/, the reflex of /h/ must have been similar to the unvoiced aspiration of these phonemes. I conclude that the chief allophone of /h/ was [h]; the friction was presumably produced in the larynx, but it may also have been produced in the pharynx.

14.8. Summary of the evidence

I am unable to find any further information from which to determine more closely the allophones of /ʔ/. In Gk. a sequence of /ʔe-/ becomes ἐ. Although the Gk. smooth breathing contrasts with rough breathing, which was aspiration, we cannot conclude that it represented a glottal stop phoneme. (The Pron. of Gk. and Lat. 70-1.) /ʔ/ did, however, unvoice a following resonant, e.g. Gk. ὅς beside Lat. ea, Skt. ayám; we may conclude that its allophones were voiceless. If it had been a stop we might have expected it to be marked by a symbol when the Greeks adopted a set of symbols which included one for the glottal stop. I therefore assume from negative evidence that the chief allophone of /ʔ/ was a sub-oral spirant with little friction. If its place of articulation had been oral, I should expect its reflexes to be represented by Hitt. ḫ. The allophone of /h/ was apparently a glottal aspirated fricative. I conclude that the allophone of /ʔ/ was either a weakly aspirated glottal fricative, or a pharyngeal fricative. I can find no evidence in favor of either description. This is the laryngeal for which we have the smallest amount of phonetic information.

The allophone of /ʔ/ is often described as a ‘glottal stop of palatal color,’ possibly because /ʔ/ had no effect on neighboring vowels. I can find no support for assuming ‘palatal color.’ If the chief articulation of /ʔ/ was sub-oral, I should expect no effect on the articulation of neighboring vowels.

I thus assume the following allophones for the laryngeals: [χ] for /x/, [γʷ] for /γ/, [h] for /h/, [h̺] or [ḥ] for /ʔ/. At present we have only general criteria for such conclusions, and these have little more certainty than our conclusions about the allophones of /bʰ dʰ gʰ/. But only if we make such assumptions will we be able to extend our information about the laryngeals and reflexes which as yet are unknown.⁶

Footnotes

¹ Sturtevant gives a convenient list in transcription of the cuneiform symbols used in Hittite texts for ḫ, Comp. Gram. 43-6, 72-3. For subsequent discussions of his on orthographical conventions, see Lang. 16.82-3 with the references there, and IHL 34.

² See the Collected Papers of H. Sweet, Arr. by H. C. Wyld, 147 (Oxford, 1913).

³ A. Thumb, Handbuch des Sanskrit 107 and 118-9 (Heidelberg, 1905).

⁴ Some of the Skt. 5th class verbs, e.g. dabhnoti ‘harms’, seem to point to a structure CeCR. I assume that the pattern here is dialectal, not PIE.

⁵ Some Indo-Europeanists define the laryngeals as ‘weaker phonologically’ than the resonants because laryngeals are lost with compensatory lengthening of resonants, e.g. [bhuXt-] > [bhūt-]. The non-significance of such a statement is clear when one compares the relation of Modern British English r to vowels, e.g. board [bɔ·d], bore [bɔ·], rod [rɔd]. In only one environment, the post-vocalic position, is r ‘phonologically weaker’ than the vowel. Unless such observations are restricted to descriptions of phonemes in definite environments, they are pointless. From them we can draw no general conclusions about the allophones of phonemes.

⁶ Reflexes of the laryngeals in PIE were presented in Chapter 3. A summary of the Gmc reflexes of laryngeals in the neighborhood of resonants was given in Chapter 8, a further Gmc. reflex was discussed in Chapter 9, a further Gk. reflex in Chapter 10, a further Ind.-Ir. reflex in Chapter 11. Studies of reflexes of laryngeals in other dialects and in other environments of the dialects already investigated remain to be done.

I do not assume vocalic allophones for the laryngeals. In Chapter 13 I have listed the PIE vocalic phonemes; in Chapters 12 and 15 I discuss the relationships between laryngeals and vowels and resonants, and the development of the PIE vowel system. The influence of laryngeals on the vowel system, is discussed there.