The University of Texas at Austin; College of Liberal Arts
Hans C. Boas, Director :: PCL 5.556, 1 University Station S5490 :: Austin, TX 78712 :: 512-471-4566
LRC Links: Home | About | Books Online | EIEOL | IE Doc. Center | IE Lexicon | IE Maps | IE Texts | Pub. Indices | SiteMap

Proto-Indo-European Phonology

Winfred P. Lehmann

12. The Laryngeals in PIE

12.1. Laryngeals survived into PIE; their distribution and allophones unclear

From the analyses of the phonological developments given above it is clear that in some phonetic surroundings laryngeals survived into PIE as independent phonemes. Only on such an assumption can we explain the differing reflexes of laryngeals which are found in various dialects; PGmc. has reflexes of laryngeals in the neighborhood of /y w r l m n/ but none in the neighborhood of /p t k kʷ/. On the other hand Plnd.-Ir. shows only slight traces of reflexes of laryngeals in the neighborhood of resonants, but has aspirated stop phonemes that developed from stop and laryngeal. One cannot account simply for this variety of development by assuming that laryngeals were lost in pre-IE but left as reflexes compound phonemes in PIE; by such an assumption PIE ph th kh would have survived only in a few dialects, PIE hw and wh in another. If we were to assume for PIE compound reflexes of laryngeals we would have to draw up complicated formulae of their development in the various dialects. It is more credible to assume laryngeals as independent PIE phonemes. In some phonetic surroundings they were already lost in PIE. In the phonetic environments in which they survived into the dialects their loss or survival varies from dialect to dialect.

Although the developments investigated above have established the necessity of assuming laryngeal phonemes for PIE, they do not enable us to determine their number, whether two or more, or their allophones, whether these had laryngeal articulation, or the occurrences of these in PIE. From the contrast in development of resonants in Gk. we need only assume two pre-Gk. phonemes, one voiceless, the other voiced; the Ind.-Ir. and Gmc. developments likewise require the assumption of no more than two phonemes in any of the environments examined. Whether more than two such phonemes are to be assumed for PIE can be determined only after examination of these and whatever other developments are ascribed to laryngeals.

Examination of the reflexes of laryngeals in PIE as well as in the dialects is necessary to determine their positions of occurrence in PIE, and their allophones in PIE and pre-IE. By PIE they had been lost in various environments, e.g. /teʔt-/ had become /te·t-/; whether their allophones had been modified in the environments in which they had survived must also be determined.

12.2. Reflexes of short vowel plus laryngeal in PIE

The most general reflexes of laryngeals in PIE are the ‘original long vowels’ and their unstressed forms, ‘schwa indogermanicum.’ Most of the commonly accepted evidence in favor of the laryngeal theory is based on these reflexes, as was noted in chapter 3. All long vowels which did not arise as a result of compensatory lengthening upon loss of a following vowel, that is, Dehnstufe,1 or possibly sporadic rhythmic laws,2 developed from short vowels lengthened upon loss of laryngeals.

The reconstructions of PIE long vowels and resonants have been based on various criteria: A. on the reflexes in the dialects; B. on the ablaut relationships; C. on the parallelism in PIE morphological classes. A. Since virtually all dialects have a long vowel in an adjective for ‘alive’, Skt. jīvá, Lat. vīvus, OCS živъ, a long ī has been reconstructed for PIE. B. Since the ablaut relationships of Gk. ἰτός : εἰ̑μι ‘I go’ parallel those of στατός : Dor. Gk. ἵστᾱμι ‘I stand’, for which an ‘original long vowel’ is assured by comparison with Skt. sthā-, etc., we assume an ‘original long vowel’ also in the PIE etymon of Gk. φατός : φᾱμί ‘I say’. C. Since the PIE normal grade vowel is preserved in the present sing. of athematic verbs, e.g. Gk. εἰμί ‘I am’, an ‘original long vowel’ is assumed for τίθημι ‘I place’, etc.

Brugmann and Indo-Europeanists who follow him have reconstructed for PIE nine long vowels: ā ē ō ī ū ṝ ḹ ṃ̄ ṇ̄. Evidence of all three types may be adduced for the first five of these; for ṝ ḹ ṃ̄ ṇ̄ there are no similar reflexes in the dialects. Other Indo-Europeanists assume clusters rather than PIE ṝ ḹ ṃ̄ ṇ̄; thus Hirt reconstructs ırə, ılə, ımə, ınə. In seeking an answer to this and other problems of the PIE long vowels one must rely on ablaut theory as well as on the laryngeal theory; both Brugmann's (Saussure's) reconstruction ṝ and Hirt's ırə were arrived at by analysis of ablaut relationships. The laryngeal theory provides us with new interpretations of pre-IE phonology; it does not modify the theory of ablaut. The theory of ablaut is an attempt to state the pre-IE phonological relations for all PIE phonemes other than the obstruents. Because the laryngeal theory provides new conjectures about pre-IE vowels and continuants, some provisions of the ablaut theory will be modified. Among these are the relationships between the stressed vowels [a· e· o·], and the vowels [i· u· r̥· l̥· m̥· n̥·] and ə which developed in unstressed syllables. Since most of these unstressed vowels have their origin in a short vowel plus laryngeal, we must further attempt to find any possible traces of a diversity of laryngeals in their reflexes. These vowels are examined below in three groups: A. the combinations of [r̥ l̥ m̥ n̥] with laryngeals; B. [i u] plus laryngeal; C. the weakened forms of PIE /e· a· o·/.

12.3. Laryngeals preserved in the dialects after [r̥ l̥ m̥ n̥]

There is general agreement on the reflexes of the vocalic allophones of the PIE resonants; the following chart is a composite one based on various handbooks. It gives the chief developments of resonants, when vocalic, in the various dialects. The IE reconstructions are those of Brugmann.

IE Skt. Gk. Ital. Arm. Celt. PGmc. OCS Lith.
i i i i i i, e i, e ĭ(ι) i
ī ī ī ī i í ī i y
u u u u u u,o u,o ŭ(ъ) u
u ū ū ū u ú ū y ū
ṛ + C ar, ra or, ur ar ri ur, ru rĭ, rŭ ir̃, ur̃
+ V ir, ur ar ar ar ar ur ĭr, ŭr ir, ur
ṝ īr, ūr rā, rō, ara rā, ār, ara ar ar/rā? ur rĭ, rŭ ìr, ùr
ḷ + C   al, la ol, ul al li ul, lu lĭ, lŭ il̃, ul̃
+ V see r al al al al ul ĭl, ŭl il, ul
ḹ lā, lō, ala lā, āl, ala al al/lā? ul lĭ, lŭ ìl, ùl
          Irish      
ṃ + C a a em, um am em um im̃, um̃
+ V am am am, em am am um ĭm, ŭm im, um
ṃ̄ ā, ām mā, ama ām, mā, ama am am/mā? um ìm, ùm
          Irish      
ṇ + C a a en, an an en un iñ, uñ
+ V an an en, an an an un ĭn, ŭn in, un
ṇ̄ ā nā, ana ān, nā, ana an an/nā? un ìn, ùn

Examination of the reflexes shows that the evidence for assuming PIE /i· u·/ differs from that for assuming r̥̄ l̥̄ m̥̄ n̥̄. The parallelism between the reflexes of PIE /i· u·/ in the dialects leaves without question the assumption of these phonemes for PIE. The four other long vocalic resonants are posited by assumption of a complete parallelism between the vocalic resonants. Ablaut relationships supply the chief evidence. The PIE etymon of Skt. jātá, Gk. γνητός ‘born’ stands in the same ablaut relationship to /genX-/ as does the etymon of Skt. bhūtá, Gk. φῡτός to /bhewX-/. Since the accented bases are parallel in structure, it was assumed that the unaccented forms maintain the parallelism. Because there is pretty conclusive evidence for PIE //, r̥̄ was also reconstructed as a PIE phoneme. In adopting the four phonemes, PIE r̥̄ l̥̄ m̥̄ n̥̄, Brugmann, though somewhat tentatively, followed Saussure, Gdr. I.417-8; it should be noted that Brugmann assumed different reflexes of these phonemes from those assumed today.

Hirt preferred reconstructions which indicated the morphological relationships of the ‘long resonants.’ ‘Long resonants’ are found primarily in the unaccented forms of laryngeal bases, e.g. Skt. jātá beside jánita, pūṛṇá beside prāti. Since for Hirt the laryngeal bases were characterized by a final long vowel, not a laryngeal consonant, he reconstructed PIE gınə, as the unstressed form of genē-, and as the etymon of jātá. (IG 2.124-48.) For without the laryngeal theory a was the only possible weakened form of long vowels.

The weakness of Hirt's reconstructions lies in the difficulty of accounting for their developments. By the usual laws Hirt's ırə, for example, should have developed to Skt. ari. Since instead it developed to īr and ūr, Hirt had to set up an additional phonetic law for the treatment of ə in combination with ır and other resonants; the aberrant developments here Hirt ascribed to accent. (See BHL 14-5.)

Both Hirt's and Brugmann's reconstructions are unsatisfactory, Hirt's for its disregard of phonological developments, Brugmann's for its disregard of morphological relationships. With Brugmann's formulation we cannot account for the prevocalic forms of the supposed ‘long resonants,’ e.g. Gk. aor. inf. θαν-ει̑ν ‘die’ beside θνητός; in both forms the root vocalism is the same. With the assumption of PIE /nX/, however, or even Hirt's ınə, one can state that the /X/ was lost before vowels. (See IG 2.144-7.)

12.3a. With the laryngeal theory we analyze the seṭ-roots of the Skt. grammarians as laryngeal bases, not like Hirt, as heavy bases; we write them with final laryngeal rather than with final long vowel; instead of Hirt's genē- we reconstruct PIE /genʔ-/. The unstressed form of this base, the form from which Skt. jātá, Gk. γνητός developed, was /gnʔ-/ [gn̥ʔ-]. We cannot determine from the laryngeal theory, however, whether for PIE we should assume [n̥ʔ] or [n̥·]. Our decision must be based on the reflexes in the dialects.

Only a few dialects have reflexes pointing to PIE long [r l· m· n·]. In Armenian, Gmc., Baltic, Slavic, and possibly Celtic3 the laryngeals were lost with no effect on the quantity of the preceding resonant. The difference in intonation in some Baltic and Slavic dialects is sometimes ascribed to a quantitative loss. The quantity is ascribed to an earlier syllabic loss. By the laryngeal theory we need only posit one stage instead of two, and this is a syllabic loss affecting intonation. The alternative theory, the assumption of PIE r̥̄ l̥̄ m̥̄ n̥̄, must hold that the laryngeals were lost in PIE with increase in quantity, that is, that pre-IE /rX/ [r̥X] became r̥̄, and that this quantity was in turn lost, that is, that r̥̄ became ; this theory is needlessly complex. Only if the evidence from the other dialects were overwhelmingly in favor of long vocalic r̥̄ l̥̄ m̥̄ n̥̄ would we need to assume two series of phonemic changes rather than one.

The chief evidence for the assumption of long vocalic resonants is found in Skt. and Gk. The development of these reflexes becomes clear when we compare them with reflexes of the short vocalic resonants. The reflexes of the short vocalic allophone of the resonants, e.g. [], are twofold in many dialects; we find one reflex before consonants, e.g. Skt. , another before vowels, e.g. Skt. ir, ur. Other dialects, notably those with only one development of vocalic resonant plus laryngeal, have, only one development for short vocalic resonants, e.g. Arm. and Lith. I assume that such dialects preserved the original allophonic pattern; originally the vocalic allophone of resonants was limited to preconsonantal position. After this distribution was disturbed, further allophones developed in some dialects depending on the following phonemes, e.g. in Skt. I assume that one such reflex became generalized to the exclusion of others. The way to such generalization may be indicated by Gk.; here one reflex of [] before consonant is the same as that before vowel, another different. One such development may have been generalized to the complete exclusion of the other.

Before laryngeals the reflexes of PIE [r̥ l̥ m̥ n̥] parallel those elsewhere. I assume that by interchange resulting from ablaut a weakened vowel was preserved before the laryngeal in some forms, e.g. /treX-to-/, lost in others, e.g. /trX-tó-/. We should expect the Skt. reflexes of these to be lengthened forms of tir- and tr̥; since we find only the lengthened form of tir-, I assume that in Skt. the pre-vocalic form was generalized before laryngeal. In OCS, on the other hand, we find before laryngeal trĭ-, the pre-consonantal reflex.

12.3b. Although only one reflex of vocalic resonant plus laryngeal is usually found, I assume that we still have evidence for a twofold development of [r̥ l̥ m̥ n̥] before laryngeals, e.g. in the reflexes in Skt. of [m̥X]. Two reflexes are found in Skt. for [m̥X], the pre-consonantal reflex in Skt. dāsá ‘slave’, the pre-vocalic reflex in dāntá ‘tamed’. Since the minority of forms has ā, I assume that the prevocalic forms here too were being generalized. For /nX/ the preconsonantal reflex has been completely generalized.

In Gk. too we find a twofold development of [r̥ l̥ m̥ n̥] before laryngeal, to ρᾱ, αρα λᾱ, αλα μᾱ, αμα νᾱ, ανα; here both reflexes are maintained for all resonants. ρω, λω correspond to the pattern of ρᾱ, λᾱ; the difference in vowel-color will be discussed below. The reflexes αρα, αλα, αμα, ανα have been the subject of much discussion, see Gdr. I.418-9, IG 2.136-8; Buck, CGLG 113-7; the reason for the twofold development has apparently been obscured by later changes. For we find such twofold developments in words similar in morphological structure, and presumably similar too in phonological structure at an earlier time: ανα in θάνατος ‘death’, νᾱ in θνητός ‘dead’. Although the original distribution can only be suggested, I assume from comparison with Skt. that Gk. ρᾱ etc. are the lengthened reflexes of [] to be expected before consonant, αρα etc. the lengthened reflexes of [] to be expected before vowel. I assume that /e/ was found between resonant and laryngeal in forms where the immediately following syllable was not accented, e.g. Aeol. εστοροται, which would have had the accent on the augment, and *ἄθανατος, *ἄδαματος, which would have had the accent on the negative prefix, see Whitney 1283.a. /e/ was lost, however, when the following syllable was accented, e.g. θνητός, στρωτός.

After the Gk. accent replaced the PIE accent the pattern was broken; from compounds with an accent pattern like that of *ἄθανατος were made forms like θάνατος ‘death’. We may see in their accentual patterns an indication that θάνατος, δάμασις were secondary. θάνατος like πότος ‘drink’, OHG mord ‘murder’ has root accent; θνητός ‘mortal’, like ποτός ‘drunk’, Skt. mr̥tá ‘dead’ has suffix accent; see IG 5.220-3 for further examples. Although we-do not have such doublets for all αρα : ρᾱ words, I conclude from the conformity of θνητός, δμητός with the expected accentual and semantic pattern that they developed from the PIE -form with suffix accent and consequent loss of /e/ before laryngeal.

With the assumption that laryngeals were preserved into PIE after [r̥ l̥ m̥ n̥] and that [r̥ l̥ m̥ n̥] developed regularly before them, we can account for most of the developments in the dialects, e.g. Skt. īr, ūr beside ir, ur, ā beside a, ā, ām beside a, am. Gk. μᾱ and νᾱ do not show the expected developments, that is, the lengthened form of the reflex of [] and []; I assume that they continue lengthened forms of [] and [] older than is represented by α, the usual Gk. reflex of these. In Italic too some of the developments are lengthened forms of the short resonant; others are reflexes of forms that elsewhere were lost.

The wide variety of development of [r̥X l̥X m̥X n̥X] in the various dialects may be ascribed to the loss of /X/ in this environment only in the individual dialects.

The diversity of development in the individual dialects such as Gk., Ital., Celt., and even Skt. may be ascribed to the small number of words with vocalic allophone of resonant before laryngeal, and the lack of morphological patterning between them. Analogical regularization is hardly to be expected between words of such different morphological categories as παλάμη, σφαραγέομαι, and θνητός.

We cannot explain why in some dialects laryngeals were lost everywhere without compensatory lengthening, why in others they were lost without such lengthening only before vowels. Presumably in Skt., Gk., Lat., and Celt. the laryngeals were lost before consonants at a time when they still caused compensatory lengthening, as they had after [i] and [u]. Whether this lengthening was ever found in Armenian is unknown. In Baltic and Slavic the loss of laryngeals produced a modification in intonation. On the basis of the evidence of forms cited in chapter 7 I assume that in Gmc. the loss occurred at a period when there was no longer such compensatory lengthening.

12.4. [i· u·] from [i u] plus laryngeal already in PIE

The reflexes in the dialects lead us to conclude that [iX] [uX] before consonants had contracted to [] [] already in PIE. [] [] became phonemic in PIE, presumably upon loss of the following laryngeal. I assume further that this change in phonemic status was aided by coalescence with [], and possibly [], from other sources; thus Skt. bhū-tá developed from [bhuX-to-], Skt. mū́ṣ from [mŭs] by lengthening in monosyllables.

Before vowels, however, the laryngeals were lost after [i u] without lengthening. Reflexes of such prevocalic uncontracted forms are Skt. ábhuvat ‘became’ beside bhūtá, gen. sg. dhiyás or -dhyás beside nom. sg. dhī́. (See Whitney 352.b.) The allophones iy, uv developed originally in accordance with the provisions of Sievers' Law, as formulated by Edgerton; later they were generalized, with y as hiatus breaker.

Besides these uncontracted forms we find in Gk. nom. sg. forms of -stems ending in -ια after consonants, but in all the other dialects in , e.g. Gk. πότνια ‘lady’ (*πόσις ‘master’) but Skt. pátnī ‘mistress’ (páti ‘master’); Gk. πίειρα < *πιϜερyα ‘fat’ but Skt. pī́varī; Gk. ἰδυι̑α < *Ϝιδυσ-yα but Skt. viduṣī. For others see Risch 124-9, and Buck, CGLG 178-9.

We find such uncontracted forms also medially in reflexes of a few laryngeal bases: πρίαμαι ‘buy’ but Skt. krīṇā́ti, μετε-κῑ́αθον ‘marched’, διερός ‘active’, κύαμος ‘bean’, and possibly κύανος ‘steel’. Hirt, IG 2.139, lists a few more doubtful words. Derivation from the contracted forms [] and [] is highly unlikely. The Gk. forms can be more credibly explained as reflexes of PIE forms in which vocalic resonants did not contract with a following laryngeal.

Such forms then are parallel to the αρα- forms, as Pedersen, Cinq. 33, pointed out. As in the αρα-forms a vowel was preserved between the resonant and the laryngeal, presumably when the stress did not fall on the syllable immediately preceding the laryngeal. The form of the resonant in Gk. supports the assumption of such vowel survival. In πίειρα, for example, the allophone of /y/ must have been consonantal, an allophone which was possible here only if a vowel followed /y/. The uncontracted form of the nom. survived only in Gk.; in the other dialects long was generalized by analogy with the oblique cases. Gk. thus illustrated the pattern of -stems before contraction. We have ample evidence for assuming a feminine/collective noun ending -h, IHL 44. I assume that the stem is preserved in a few Homeric vocative forms, e.g. πότνι, which are often assumed to have an elided final vowel.

The presence in Gk. of such uncontracted forms does not contradict the assumption that // and // were phonemic in PIE. Gk. merely preserved and generalized some infrequent case forms.

12.5. [e· a· o·] also PIE. Their unaccented forms

For /e· a· o·/ there is no evidence of uncontracted forms. None of the dialects have reflexes of /eX/ corresponding to those in Gk. of /yeX/ or those of /reX/. I assume therefore that //, /eA/, and // had already contracted in PIE; the resulting /e· a· o·/ fell together with /e· a· o·/ which had been lengthened in Dehnstufe. Hirt gives numerous examples, IG 2.36-51. The Lithuanian reflexes /ů/ and /o/ of PIE // are dialect developments and do not give us evidence for assuming two long // phonemes; see Meillet, Introduction 103-4.

The long PIE vowels then point to a gradual loss of laryngeals, varying according to phonetic environment. They were lost earliest between /e a o/ and consonant. The resulting long vowels became phonemes, falling together with the /e· a· o·/ that had developed in Dehnstufe.4 The raising of [] and [] to phonemic status was apparently also aided by the prior presence of long [] and []. But there were no previous long [r̥ l̥ m̥ n̥] in PIE with which the short vocalic resonants might have fallen together; after [r̥ l̥ m̥ n̥] laryngeals were lost in the individual dialects leaving the reflexes listed in the chart given 12.3.

12.5a. Although there is general acceptance of /e· a· o·/ as PIE phonemes, the unaccented form of these is disputed. Most Indo-Europeanists assume one unaccented form, ə. By a less widely held theory there were three unaccented vowels corresponding to the three long vowels. Proponents of the schwa theory, ə, base their arguments on the presence in most dialects, and in most unstressed forms, of one reflex, i in the Ind-Ir. dialects, a elsewhere. Proponents of three unaccented vowels point to e and o which are found as reflexes of PIE unstressed vowels, especially in Gk. Proponents of the schwa theory have been unable to explain these e and o vowels except by recourse to analogy, a method which is not particularly credible for words for which no source of the analogical vowel can be discovered. On the other hand the assumption of three vowels seems unnecessarily complicated because most dialects have a uniform reflex of the three hypothetical vowels.

With the laryngeal theory this complication can be avoided. It is a statement of the ablaut theory that the unstressed vowels assumed for PIE developed from full vowels. Although formerly, original long vowels of various colors were assumed, it is now clear that this diversity of ‘original long vowels’ is a result of contraction with laryngeals. Since we do not assume three different ‘original long vowel phonemes,’ but rather a diversity of laryngeals, we no longer assume three different unaccented vowels, but rather one /e/. Laryngeals were found in the neighborhood of /e/ as well as in the neighborhood of /e/. Unless we favor the schwa theory, in place of the unstressed e a o we assume /e/ preceded or followed by laryngeals. Under the hypothesis of pre-IE laryngeals the problem of the unstressed variants of /e· a· o·/ is resolved to the question whether or not /e/ had contracted with neighboring laryngeals in PIE.

This problem can be solved only from examination of the reflexes in the dialects. The threefold reflexes in Gk. are inexplicable by the schwa theory. Yet the schwa theory has not been discarded, for the weighty evidence against it is taken primarily from one dialect. If, however, material from other dialects could be assembled that would point to PIE /eX/ rather than PIE ə, the schwa theory would lack conviction. I shall therefore list the various forms and morphological categories in which Gk. shows vowels other than a for the unstressed forms of PIE /e· a· o·/ and then attempt to find similar distinct reflexes in other dialects.

12.5b. The usual forms in Gk. that are cited in favor of three unstressed vowels are στατός, cf. ἵστᾱμι ‘I stand’, θετός, cf. τίθημι ‘I put’, δοτός, cf. δίδωμι ‘I give’; for others see IG 2.34-5, 119-21, Gdr. I.174-5, Introduction 154ff. Proponents of the schwa theory explain the ε, α, ο of the unaccented syllables as analogical vowels from those of the accented syllables. Such an explanation can apply only for words beside which are found such accented forms. These are not always attested, e.g. no long e is found beside ἄνεμος ‘wind’, no long ο beside ἀρόω ‘I plough’.

Other forms with e where an unaccented vowel is expected are so wide-spread that the e has been explained as an aberrant PIE /e/ : Gk. γενετήρ ‘progenitor’, Skt. jánitṛ, Osc. Genetaí, Welsh cenedl; Gk. φέρετρον ‘barrow’; Lat. tenebrae ‘darkness’. Brugmann even suggested that the e in the second syllable of ἄνεμος ‘wind’ is a reflex of PIE /e/, Gdr. I.486. Such an explanation of the e in these forms is possible, but not plausible; for beside all of these forms with aberrant e, we find forms attesting an e-colored laryngeal. And when an o rather than an e is found one cannot assume that this is a reflex of an original accented vowel. Thus a series of Gk. aorist forms, e.g. ἔθορον, cf. θρώσκω, is wholly inexplicable to proponents of PIE ə. If with the laryngeal theory these forms, as well as the ‘regular’ forms could be explained, the schwa theory might be abandoned. For the laryngeal theory offers a simpler explanation than the alternative possible before the development of the laryngeal theory, that is, the assumption of three unaccented vowels. This assumption, as Lommel pointed out KZ 59.195, would merely push back the problem from the dialects into PIE.

12.5c. In chapter 2 I assumed that by ablaut there was only one unaccented variant of /e/; this I have written /e/. The ə of Indo-Europeanists who hold the schwa theory is reconstructed where /e/ stood before laryngeals. According to the laryngeal theory this unaccented vowel may have been influenced or colored by a neighboring laryngeal. All proponents of the laryngeal theory admit the possibility of such coloring for /h/ and /x/; the evidence is so extensive that further proof is unnecessary. Examples are: Lat. stāre from /stah-/ from /steh-/ and Lat. status from /steh/; Lat. pāscō ‘feed’ from /pexs-/, cf. Hitt. pa-aḫ-ḫa-aš-mi, and Gk. ἄπαστος ‘without provision’ from /pexs-/. On the other hand no such change is found in vowels standing beside a third laryngeal, e.g. OLat. coēpī ‘I began’ from /eʔp-/, perf. ptc. coeptus from /eʔp-/. The effect of a fourth laryngeal on contiguous vowels is disputed. Some proponents of the laryngeal theory have assumed that this laryngeal, /γ/, too, changed the timbre of neighboring vowels, so that [γe] became [γo], [] became [ō]. Others, e.g. Sturtevant, have assumed that /γ/ did not change the color of a neighboring /e/ or /e/. Sturtevant must therefore assume that many forms of the root dō-, from /deγ-/ ‘give’, e.g. Lat. dōnum, Gk. δίδωμι, have o-vowels by analogy, and that the expected vowel is found in Lat. 2d sg. pres. dās, IHL 43. Since δίδωμι is parallel in formation to εἰ̑μι, τίθημι, ἵστᾱμι, assumption of analogy is somewhat difficult to uphold. Analogical ā in the 2d sg., however, is readily understandable, as are the a-vowels in other forms of the root. Indo-Europeanists who hold that /γ/ changed neighboring vowels from e to o, on the other hand, must explain /γ/ found contiguous with e in words like Skt. mā́ti ‘measures’, Hitt. me-ḫur ‘occasion’, and cognates. In sum, we find forms which show coloring of contiguous e-vowels to a-vowels; other forms show no change of color of a neighboring laryngeal. A third group have o-vowels, but beside these also e and a-vowels. In all three groups there were later analogical changes. The essential point of dispute among Indo-Europeanists centers about the question of deciding which forms are regular developments, which are analogical, and of developing the simplest theory to explain the relationships.

12.5d. The commonly assumed reflex of PIE ə in Gk. is α, as in the first syllable of πατήρ ‘father’, cf. Skt. pitā́. Reflexes that differ from this are found in the following categories:

In addition we find unexpected vowel color o in some laryngeal bases. Such are the o in aorists, beside ρω, λω in other forms of the verbs. Examples are:

A. For most laryngeal bases we find in the s-aorist a weakened form of the base ending in α, e.g. ἐδάμασα ‘tamed’. (Schwyzer, Gr. Gr. 752.) For many of these there are cognates giving evidence for an a-colored laryngeal; cognate with ἐδάμασα are Lat. domāre and OHG zamōn.

Beside these, however, are found seven aorists with ε: ἐκάλεσα, ἐκόρεσα, ἐλόεσα, ἐστόρεσα, ἤμεσα, ἤνεσα, ὤλεσα. For some of these, we have evidence of an e-colored laryngeal in the base: Gk. καλήτωρ beside ἐκάλεσα, Lat. crēscō ‘grow’ beside ἐκόρεσα. WP find evidence for such a base also beside ἤμεσα 1.262-3, and ἐστόρεσα, 2.638-40. Cognates of the three other aorists give us no evidence to assume an e-colored laryngeal.

There are two aorists with o-vowel; for one of these, ὤμοσα, WP 1.178-9 assume a root omō; for the other, ἤροσα, they assume a root arā-, 1.78-9. Assumption of a final ā, however, is not without dispute; Persson, Beitr. 669, had assumed Gk. ἀρο- to be a reflex of the PIE form, but WP consider it analogical.

B. We also find such threefold development of vowels in the form of laryngeal bases found in word compounds. Beside the usual α found in γέλασμα ‘laughter’, παν-δαμάτωρ ‘all-conquering one’ we find ε in ἄνεμος, ὠλε-σίκαρπος, ἀρετή, ναέτωρ, γενέτειρα, βέλεμνον, and o in ὄνομα and ἐνοσίχθων, see Specht, KZ 59.83-9. Again we find beside these, cognates pointing to a laryngeal in the base; Lat. reor, rērī ‘consider’ point to an e-colored laryngeal beside ἀρετή, Lat. nō-men to an o-colored laryngeal beside ὄνομα.

C. For the two compounds with unaccented ε, δίεμαι and διερός, WP 1.775 assume spread of ε by analogy; but again this assumption is apparently based on the supposition that only α should be found in such an environment.

D. Aorists made from laryngeal bases generally have α in the root, which here was unaccented, e.g. ἔβαλον, cf. βέλεμνα, βλη̑το; ἔθανον, cf. τέθνηκα, θάνατος, etc. Some aorist forms, however, have ο; these have been listed above. Beside most of these are found forms with long ō. Such long ō vowels, which are reflexes of resonant plus laryngeal, are also found in:

Such reflexes of resonant plus laryngeal, in which α plus resonant would be the normal development, are found also in a number of Gk. substantives: (Schwyzer 363)

I assume that when vowels other than a are found in the neighborhood of unstressed resonants that the vowel color is a result of assimilation to the articulation of a neighboring phoneme. In some forms, e.g. μολει̑ν, the laryngeal was lost before lengthening the reflex of the resonant; in others e.g. βλώσκω, it changed both the color and the quantity of the vowel resulting from the preceding resonant.

12.5e. Assumption of such an influence of neighboring laryngeals may be supported by the Skt. development of [] in the neighborhood of laryngeals. [r̥x] may develop either to ūr or īr. Only ūr is found after labials, including u, cf. Wackernagel, Aind. Gr. I.28. After other consonants and initially īr is usually found. In some forms, however, only ūr is attested after consonants other than labials; Wackernagel, Aind. Gr. I.28 lists thirteen such forms. Most of these have a labial elsewhere in the base, e.g. kūrpara, kūrmá. In kūrdati ‘leaps’ which is cognate with Gk. κόρδαξ ‘dance’ there is a development of back vowels in both languages, but we have no evidence for a labial in the base. A similar correlation of back vowels is found in Gk. ὀρθός, Skt. ūrdhvá (with labial, to be sure), ὀργή, ūrja, οὐ̑λος, ū́rnā. Since the vowel color in Skt. of [] plus [X] was influenced by surrounding phonemes, I assume that here too the color was so determined. The only phoneme with such possible effect was the laryngeal.

I assume that in forms with preceding consonant other than labial the possible development was not fixed; [r̥X] developed either to īr or ūr in the same root, e.g. śīrtá and śūrtá from śr̥- ‘break’. This pattern had an analogical effect also on reflexes of [r̥γ]; cognate with Gk. κόρση we find Skt. śīrṣá. But because of the presence of back vowel in Gk. and Skt. in most words with a sequence of [] plus /γ/ I assume that the resulting vowel is a back vowel because of the laryngeal.

Such influence of a neighboring laryngeal was not restricted to reflexes of PIE []. A relatively great number of Gk. forms have been cited above to show such influence also on reflexes of /e/. I thus assume for PIE rather than ə a sequence of unaccented vowel plus laryngeal, /ex eh eʔ eγ/. We have further evidence to support this assumption.

12.5f. Before i, the Ind.-Ir. reflex of these sequences, there was secondary palatalization of velars. Wackernagel, Aind. Gr. I.1.42, cites the isolated word duhitr̥ ‘daughter’ in support of the statement that such palatalization did not spread analogically from related forms in which the velar consonant stood before /e e· y i·/. Two Vedic words, however, have an unpalatalized velar before i, okivas from uc(i)- ‘be pleased’ and tigitá ‘sharp’, cf. téjate ‘is sharp’. On the basis of these forms Wackernagel states that i, when a reflex of PIE /e/ plus laryngeal, cannot have had a consistent i-color. With the laryngeal theory we can explain the reason for such lack of consistency. We have evidence in Gk. concerning the color of the laryngeal in okivas, but not for that in tigitá. A cognate of uc(i)- is Gk. ἕκηλος, Pind. ἕκᾱλος and εὔκηλος ‘in undisturbed comfort’. We may therefore assume a laryngeal of a-color, /wek-A-/. Thus these two forms, as well as forms like -tiṣṭhighiṣan ‘wishing to climb’ and daghiṣyante ‘will reach’ are further evidence against the assumption of a PIE phoneme ə. /eX/ were maintained as separate phonemes into Ind.-Ir. For Ind.-Ir. i from /eX/ coalesced with the Ind.-Ir. i phoneme after the time of secondary palatalization, but before the change of s to . Ind.-Ir. s became after this i as well as after i from PIE /y/, Gdr. I.728-30.

12.6. The influence of laryngeals on neighboring accented vowels

If we assume on the basis of unaccented vowels PIE laryngeals of three colors, we must also describe their development in the neighborhood of stressed vowels. There is little doubt about /ʔ/ and /A/. /ʔ/ does not affect the timbre of a neighboring vowel: /e/ plus /ʔ/ yields PIE //, e.g. Gk. τίθημι, /e/ plus /ʔ/ yields /e/, e.g. θετός. PIE /e/ is changed to /a/ both when standing before or after the a-colored laryngeals: /Ae/ becomes /a/, e.g. ἀντί, and /e/ was lengthened as // when the following /A/ stood before an obstruent, /eA/ became //, Dor. ἵστᾱσι, /eA/ became /a/, e.g. στατός. The effect of the remaining laryngeal is disputed; in some forms is found evidence for //, e.g. δίδωμι, in others the e-vowel remains unchanged, e.g. Arm. sair ‘edge’ from *ke-ri, which is from the root /keγ-/.

There are very few roots on which to base a description of the development of PIE /e/ followed by /γ/; and cognates have been found for only /do·-/ in Hittite. Such roots are:

The reflexes of these roots are not very wide-spread in the dialects. Only Baltic and Slavic dialects give us evidence for the accented vowel of /o·g-/: e.g. Lith. úoga ‘berry’, OCS agoda ‘fruit’; evidence for the unaccented form is found in Gmc. and Celtic, and possibly Arm.: Goth. akarn ‘wild fruit’ and Irish áirne ‘gland’ from *agrīni ̯a ‘sloe’.

The roots /do· ko· po·/ give unambiguous evidence for assuming a development of to ; various forms from each have been cited above. The second root /po·-/, however, is not without difficulty. Reflexes are: Skt. pāyú ‘protecting’, Gk. πω̑υ ‘herd’, ποιμήν (< *πωι-) ‘shepherd’, Lith. piemuõ ‘shepherd’, Arm. hauran ‘herd’, Goth. fōdr ‘sheath’. Beside these there are attested forms of a root /pa·-/ ‘pasture cattle; nourish’, e.g. Lat. pāstor ‘shepherd’, OCS pastyrь. Indo-Europeanists have discussed variously the relationship of the two roots, cf. WP 2.73. Although the meanings are quite similar, they cannot be reconciled with the difference in phonology; it would be difficult to find a source from which o might have spread to Gk. πω̑υ by analogy. On the other hand the Lat. forms, e.g. pāscō, seem to be regular reflexes of a PIE root /pa·-/. Whether like Sturtevant, IHL 36, we derive Skt. pā́ti ‘protects’ from the root /pa·-/, we must assume a PIE root /po·-/ and one /pa·-/. The Hittite forms pa-aḫ-ḫa-aš-mi, etc. presumably are cognate with the root /pa.-/.

Although the evidence is small I assume that /γ/ changed the timbre of a contiguous vowel, but only when it coalesced with that vowel. This laryngeal, for example, is found in Hitt. ḫi-in-ik-zi ‘assigns’; the following vowel, however, is unmodified, cf. OIr. écen ‘necessity’. Examples given above illustrate the reflexes of /γ/ when it contracts with preceding /e/ or /e/.

There is thus a marked difference between the effect of /γ/ and that of /x/ and /h/ on a neighboring vowel. /x/ and /h/ change the timbre of any neighboring vowel, /γ/ changes the timbre of a neighboring vowel only when it contracts with that vowel.

12.7. The development of PIE /e/ in the dialects

With such an analysis of the effect of laryngeals, the development of the PIE unaccented vowel, /e/, becomes clear. Indo-Europeanists have attempted to give an explanation for the development of /eX/ (ə) to i in Ind.-Ir., to a elsewhere. We cannot of course predict the development of any phoneme, nor can we explain why it developed in any particular way. But we can attempt to understand its course of development, and this we must do in terms of the structure of the phonological system concerned.

In PIE the contrast between /e/ and /e/ was a contrast of accented versus unaccented vowel, not one of timbre. A similar contrast existed between // and /eʔ/, /eA/ and /eA/, // and /eγ/.

None of the dialects has a similar contrast. In all dialects the short vowels may occur in unaccented as well as accented syllables. None of the dialects has a vowel which is restricted to unaccented syllables as was /e/ in PIE. Theoretically then the reflex of /e/ could coalesce with those of /e a o/ in any dialect. If it did, and no distinction was preserved between the reflexes of unstressed and stressed vowels, important distinctions would of course have been lost.

The most frequent vowel in accented syllables was /e/ and its reflexes, in PIE as well as in the dialects. It is clear that a distinction between the reflexes of PIE /e/ and the reflexes of PIE /e/ was maintained in all dialects. The dialects which maintained a distinction of vowel timbre, the European dialects, also maintained a distinction of vowel timbre between reflexes of /e/ and /e/; /e/ generally became /a/. In the only IE dialect which failed to maintain at least partially the PIE threefold distinction of vowel timbre, Ind.-Ir., the contrast between /e/ and /e/ was preserved differently; the reflex of /e/ fell together with a phoneme which was characteristically found only in unstressed syllables, /i/.

12.7a. By a comparison of the structural systems of PIE and the dialects we can understand the sharply divergent developments of PIE /e/ to Ind.-Ir. i, to a in most forms in European dialects. It is only the piecemeal treatment of vowels, traditional in the study of historical linguistics, which has obscured the correlation between the development to Ind.-Ir. i and the shift in the structural basis of the vowel system. By comparing the structural systems we can also understand why /e/ may appear in the dialects with any vocalic timbre, e.g. Gmc. u in OHG -zug, Gmc. e/i in Goth. sitans beside the usual reflex a. In the European dialects the contrast of vowel timbre replaced that of accented versus unaccented vowel, and in the shift the particular vowel timbre of unaccented syllables was not significant. When viewed in this way the consistent development of /e/ to one vowel in any given dialect is more remarkable than are the particular reflexes in the individual dialects.

I assume that originally /e/ had various allophones, depending on the neighboring phonemes, especially the laryngeals. The allophones of the unstressed /e/ were probably more central than the allophones of the stressed vowels. In articulation, the allophones of /e/ were closest to those of /a/ among the accented vowels. Since two of the laryngeals were a-colored, a majority of the allophones of /e/ would have resembled /a/ in articulation. After the contrast of stressed : unstressed vowels was replaced by that of timbre even in unstressed syllables, the allophones of /e/ fell together with those of /a/. Only in Greek, where the threefold PIE contrast of vowel timbre was maintained, do we find some degree of retention of the varying articulation of the allophones of /e/. In those dialects in which the back vowels, /a/ and /o/, fell together, Gmc., Baltic, Slavic, the back, open allophones of /e/ exerted analogical influence on the higher allophones, and were generalized.

In Indo-Iranian, with its phonological shift in vowel system from a contrast in timbre to one of quantity, the allophones of /e/ fell together with the vocalic allophone of /y/, which occurred characteristically in unstressed syllables.

12.8. Four laryngeals assumed for PIE

Because of their differing effects on vowel timbre I conclude that we must posit at least three laryngeals for PIE:

This correlates with other evidence for laryngeals. In our Hittite records /ʔ/ is not represented by any cuneiform symbol, /γ/ is represented by ḫ, A may or may not be represented by ḫ(ḫ). To account most simply for such orthographic variation many Indo-Europeanists have assumed two laryngeals with a-color. It has been further noted that the a-colored laryngeal which is not recorded in Hittite combines in Ind.-Ir. with a preceding voiceless stop to yield an aspirated voiceless stop. See chapter 11.

I conclude that until we have further evidence from Hittite or other Anatolian languages we can most simply account for the various reflexes of laryngeals by assuming four laryngeal phonemes in PIE.5

Footnotes

1 See Hirt IG 2.51-76 for examples and bibliography.

2 No evidence for an origin in laryngeal, like that of Skt. bhū́ < /bhwX-/, has been found for the long ū in Gk. δρυ̑ς, μυ̑ς, Skt. srū́ ‘stream’. F. Specht, Die Flexion der n-Stämme im Baltisch-Slavischen und Verwandtes, KZ 59.213-98 (1931-2) ascribes the long ū to lengthening in monosyllables; see Excurs II Ein indogermanisches Dehnungsgesetz.

3 The reflexes of the resonants followed by laryngeal are disputed in Celtic; see IG 2.133 and Lewis-Pedersen 4-8.—A. Vaillant has discussed the Lithuanian intonation in Le probléme des intonations balto-slaves, BSL 37.112-4.

4 Kurylowicz suggested in 1937, Mélanges de Linguistique et de Philologie offerts à Jacques van Ginneken . . . 199-206 (Paris, 1937) that ā and ō cannot be assumed for PIE because laryngeals survived into the dialects. I cannot subscribe to this suggestion. I assume that laryngeals survived into PIE in some phonetic environments and were lost in others with compensatory lengthening. The status of laryngeals in PIE may be compared with that of r in modern British English. In pre-modern British English r is attested before and after vowels; in modern British English r has been lost after vowels with compensatory lengthening, but its status elsewhere is unchanged.

5 Compare Sapir, Language 14.270: ‘If, now, we posit an IE series ʼ y-, ʼ w-, ʼ m-, ʼ n-, ʼ 1-, ʼ r-, a series .̓y-, .̓w-, .̓m-, .̓n-, .̓l-, .̓r-, and a series xy-, xw-, xm-, xn-, xl-, xr-, we have to inquire what happened to these clusters (initially and in other positions) in each of the main branches of IE. In the end it will undoubtedly prove far more economical of effort to assume little or nothing in the way of sweeping reductions of these, to us, uncomfortable clusters and to keep our eyes open for distinctive reflexes of them in the IE dialects than to oversimplify our task by assuming radical reductions in the IE period.’