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

10. PIE ‘Initial y’ in Greek

10.1. Development of ζ from PIE ‘initial y-’

In the IE dialects except Greek, PIE initial y undergoes a consistent development to a palatal resonant. In Greek there is a sharp divergence of development; some y's develop to an initial aspirate, e.g. ὅς, Skt. yás ‘who’; ἥβη ‘youth’, Lith. -jėgà ‘strength’; others to a dental affricate, e.g. ζωστός, Avestan yāsta- ‘yoked’; ζειαί ‘spelt’, Lith. javaĩ ‘grain’.

Many explanations have been suggested for the development in Greek. Until Sapir discussed the problem, however, no convincing evidence to support an explanation was found in any IE dialect other than Greek, and none of the explanations was generally accepted. Indo-Europeanists who explained the divergence in Greek as a purely Greek development looked for support for their theories in linguistic methodology rather than in supporting evidence in other IE dialects. That this was of little help in providing a solution is demonstrated by two explanations, each of which assumes the alternate development as the regular one. Pedersen assumed that all IE y- > gy- > ζ except in certain phonetic environments, Schwyzer that y- > ʽ  as the normal development.1 The various theories have been discussed in so many places that I shall sketch briefly only the important theories or those that still seem to be current. They may be listed in two groups: theories that consider the development of /y-/ to ζ and ʽ  purely Greek developments, in section 2; theories that try to explain the Greek developments from phonetic differences in PIE, in section 3.

10.2. Explanations of ζ < y- as a Gk. development

10.2a. Sommer succeeded in showing how little evidence there was for parallel developments in other IE dialects; but in suggesting an explanation he was able to present only a series of formulae, rather than a comprehensive theory.2 In one of these he, as did Sapir later, drew a parallel between the development in Greek of PIE [y-] and [w-]. Sommer assumed a parallel between [w-] and [y-] in words with intervocalic -s-; when -s- was lost, [w-] and [y-] became aspirated, e.g.

ἑαρόν from *Ϝεσαρόν   ζέω from *ι ̯έσω
    ζώννῡμι from *ι ̯ώσνυμι.
Here, however, the parallel ceases; Sommer's proposed Ϝh became h, but i ̯h hi ̯ became ζ whereas i ̯ became h. Another formula of Sommer's was that i ̯ became ζ before υ as in ζυγόν and ζύμη

10.2b. Pedersen has discussed the subject in a number of articles.3 In IF 22 he objected to Sommer's explanation for various reasons: 1. if ζ developed from hi ̯, we shouldn't expect it to be voiced; 2. Sommer's analogies are not convincing; 3. ὑσμῑ́νη does not conform with Sommer's rules.

He himself has proposed various explanations. In KZ 36 he considered ζ a purely Greek development. In IF 22 he proposed an explanation by means of lost pre-IE consonants, mentioning that Möller had considered the y of Skt. yaúti, cf. Gk. ζεύγνῡμι, Skt. yámati to be , an IES spirant. To my knowledge this is the first attempt at explanation by means of the lost pre-IE consonants. Pedersen did not expand his suggestion, and since he later proposed a different explanation apparently abandoned it.

In his last treatment of the problem, Symb. Phil., Pedersen assumes that every PIE y- became gy- became ζ except under certain circumstances. Pedersen tries unconvincingly to define these; even after he has made his explanation, he finds three difficult words. Besides, he gives no convincing argument for assuming a development of y- to gy-.

10.2c. Linguists who continue to suggest that ζ developed in neuter nouns because of wrong syllabic division4 have not met the arguments that 1. the article το(δ) would have been only one of the forms of the article found before *ιυγον and 2. that the masculine and feminine nouns as well as the verbs would be unaccounted for.

10.2d. Schwyzer, Gr. Gr. 331, suggested that y- normally developed to ʽ ; in some words, however, it was doubled in an effort to maintain it; double y-may have developed to dy-, this to ζ. To uphold Schwyzer's view one would have to establish that y- normally became ʽ . Even if this were granted, the doubling of y- in random words seems unlikely.

10.3. Explanation of ζ < y- on the basis of PIE

10.3a. Schulze was the first Indo-Europeanist to point out clearly the difficulty of the Greek developments.5 The theory he suggested has probably been accepted more widely than any since proposed. He assumed two different sounds for PIE, one a palatal semi-vowel that became Gk. ʽ , the other a palatal spirant that became Gk. ζ. Other Indo-Europeanists attempted to find evidence in other dialects in support of Schulze's assumption, but with such meager results that the evidence was otherwise explained and was thus unconvincing. In Skt. there is evidence for some kind of a difference; yastá, compare Gk. ζεστός, follows the pattern of paktá, compare Gk. πεπτός, differing from that of iṣṭá from yájati, compare Gk. ἅζεται, Gdr. I.793-4. In the Alemannic dialects of OHG are found the forms enēr and āmar which correspond to Gk. words with initial aspirates, while y- is preserved in words whose cognates have ζ as joch, Gk. ζυγόν, cf. 7.6. These supports for Schulze's theory have been rejected chiefly on a quantitative basis; the material in Skt. and Gmc., though meager, supports Schulze's assumption of different initial sounds; but the evidence cited from Armenian, Albanian, Celtic, and Slavic in favor of his assumption has been otherwise explained. (See Sommer, Gr. Lt. 139-45.)

10.3b. Sapir, whose theory is supported more soundly than any other both with linguistic evidence and methodological arguments, suggested a solution based on PIE phonology revised in accordance with the laryngeal theory.6 He accounts for the Gk. developments of PIE y- by assuming a persistence of a reflex of laryngeals before resonants in Gk. In contrast with Möller he assumes initial laryngeals for the words that show ʽ  in Gk. Like Sommer he draws parallels between the development of initial w and y, but ascribes the development to different reasons from those suggested by Sommer. If one of the voiceless laryngeals stood before w-, Sapir assumed that w- became aspirated. When w- was lost in Gk., words which had a ʽ w cluster before a vowel were left with initial ʽ  and vowel, e.g. *Aawelk- > *ʽ welk- > *ʽ elk-, cf. ἕλκω; words with simple initial w were left with initial vowel, e.g. *Ϝοιδα > οἰ̑δα. Sapir assumed a similar development of voiceless laryngeal and y-; voiceless laryngeal and y- too resulted in an aspirated cluster and ʽ  was left after the loss of y-, e.g. *ʔyós > *ʽ yós > ὅς.

Here, however, the parallelism stops. Where initial clusters of h plus w and h plus y both developed to ʽ , initial w became Ϝ and was lost, initial y according to Sapir became dz. If we are to rely as heavily on linguistic method as did Sapir and as seems necessary to suggest explanations for phenomena for which we can cite so little linguistic evidence, we should expect a more rigorous parallelism. If the simple resonant w- were lost initially, we should expect a similar development of simple initial y-. For medially y, like w, is lost in Gk.

By a convincing argument Sapir supported his assumption of an initial laryngeal before y- in the PIE form of ὅς, Skt. yás; connecting them with Skt. ayám and Lat. ea he reconstructed PIE *ʔeya- as the original form from which the various pronouns developed. If we accept parallelism between ὅς, yás and the other Gk. words which show ʽ  for PIE y-, it is difficult to reject Sapir's theory accounting for the development of such Gk. ʽ .

10.3c. Sapir assumed Gk. ζ e.g. ζωστός, as the normal development of PIE y-. He made this assumption because he found no cognates for words beginning with ζ like ea, ayám for yás.

He also assumed that PIE y- became ζ on the grounds that in Gk. ‘voiced continuants before vowels keep their voice.’ This, however, is not true of the PIE phoneme which is most similar in function and development to y, that is, w.

Moreover, when we examine the other PIE sources of Gk. we find that they are clusters: 1. dy- as in Ζεύς, cf. Skt. dyaús; 2. gy-, as in *gʷyē-, ζη̑ν; 3. zd medially, as in [ozdos], ὄζος. Except in a few proper names of uncertain origin, ζ makes position in Homer, as do similar clusters.

10.4. Words in which such ζ is found

Although Sapir has provided a convincing explanation for the development of PIE y- to Gk. ʽ , for the development of PIE y- to ζ he has simply assumed a phonetic development in Gk., supporting his suggestion by referring to a somewhat similar development in Italian. Neither the linguistic nor the methodological arguments of Sapir seem adequate for his suggestion of the origin of ζ. An examination of the Gk. words and their cognates in which ζ corresponds to y- of other dialects points to the necessity of assuming an initial PIE cluster, not simply y-.

10.4a. For Gk. ζωστός ‘girt’, ζώννῡμι ‘gird’, ζωστήρ, ζώνη ‘girdle’ we find cognates in Av. yāsta- ‘girt’, Lith. júostas ‘girt’, júosti ‘gird’, OCS pojasati ‘gird’, and Albanian n-ǵeš ‘gird’. Accepting Couvreur's suggestion, Hett. . 197-8, that Luwian ḫi-iš-ḫi-ya-an-ti is related to Hitt. iš-ḫi-ya-an-ti-iš Sturtevant has connected Gk. ζωστός with Hitt. iš-ḫa-a-i ‘bind’;7 he also cites the Hittite related words. Sturtevant reconstructs an IH root γyoγz- and explains from it the various Hittite developments; we find then for this group initial voiced laryngeal plus y. To be sure Sturtevant suggests a voiced laryngeal because the Gk. developments differ from those that Sapir assumed for voiceless laryngeal plus y; but he adds that the Hittite forms would not permit assumption of a voiceless laryngeal.

We thus have evidence from Hittite that one of the words in which PIE y-became Gk. ζ not only began with laryngeal but that the laryngeal was voiced. I assume for the ζ of ζωστός an origin parallel to that of Ζεύς and ζη̑ν; the ζ of ζωστός developed from γy-, that of Ζεύς from dy-, that of ζη̑ν from gʷy-.

10.4b. ζωστός is the only word with ζ for which a known native Hittite cognate has been found. ζυγόν seems to have a cognate in Hittite i-ú-ga-an ‘yoke’. But i-ú-ga-an has been assumed to be a loan-word from Indic; although other linguists consider it a native Hittite word, no evidence has been supplied to overthrow the thesis that it was borrowed.8 In view of the numerous borrowings from Indic of terms relating to horses and to horsemanship, such an explanation seems likely. Until decisive arguments to the contrary are given, i-ú-ga-an cannot be used to determine Hittite developments.

ζυγόν has many cognates in other IE dialects: Lat. iugum, Goth. juk, Lith jùngas, OCS igo, and verbs, Skt. yunákti, Lat. iungere, Gk. ζεύγνῡμι ‘yoke’. Evidence for assuming an initial laryngeal may be found in compounds or in sandhi phenomena, such as the lengthening of Vedic final short vowels in the first member of compounds when the second member begins with a laryngeal. We find such lengthening in compounds of Skt. yuj-: RV 3.11.6a abhīyújas ‘assailant’ is from abhí and yuj-; in RV 5.17.3b we find a long augment in ā́yukta ‘endowed with’.9 Since such lengthening does not enable us to determine the laryngeal in question, we cannot from the Vedic evidence reconstruct the initial cluster more precisely than Xy-.

10.4c. There is evidence for a similar compound of yáva ‘grain’, the Skt. cognate of ζειαί ‘spelt’. Related are ζείδωρος ‘producing pasture’, ψυσίζοος ‘producing grain’, Lith. javaĩ ‘grain’, and Irish eorna ‘barley’. In two poems of the RV the meter requires sūyavasa ‘abounding in grass’; the compound is from su- ‘good’ and yáva. I assume that in these forms retains the lengthening that resulted when it coalesced with initial laryngeal and I therefore reconstruct /Xyew-/.

Sommer, following Peppmüller, connected with ζειά a difficult Homeric word ἤϊα, ᾐ̑α ‘chaff’, the gen. pl. of which is found in ε368, ᾔων.10 Boisacq calls this connection improbable, but can cite in place of it only Thumb's connection of the word with the root ē̆s- ‘throw’. The semantic relation of Thumb's etymology is troublesome, especially when we find the meaning ‘flour’ for some occurrences of the word. One can assume that a word meaning ‘chaff’ may have developed from a meaning of ‘that which is thrown (away)’, but if ἤια also means ‘flour’ a connection with a word for ‘grain’ is much more plausible. The chief difficulty in relating the two words has been that of relating them phonologically. If we follow the procedure which Sapir used to indicate the relationship between Hitt. ḫu-u-wa-an-te-eš. ‘winds’, Gk. ἄησι ‘blows’, and αἵνω ‘I winnow’ we can arrive at a plausible explanation of the relationship between ζειά and ἤϊα. Instead of the conventional IE root wē- Sapir set up an IE base *xaweʼ - ‘to blow’. This underwent a series of ablaut modifications and changes caused by loss of phonemes which Sapir indicated as follows: (Lang. 14. 270)

1. *xaweʼ - or *xəweʼ - : *xweʼ -
2. *haweʼ - (or *həweʼ -) : *hweʼ -
3. *haweʼ - (or *həweʼ -) : *ʽ weʼ -
4. *hawē- (or *həwē-) : *ʽ wē-
5. *awē- (or *əwē-) : *ʽ wē-
6. *awē-   : *hē-.
For ζειά, ἤια I assume an initial γeye- : γye-. From the first of these developed ἤια; the η is a Homeric development, beside which occur also εἰαί and εἰ̑οι. From the second developed ζειά < *ζεεα, cf. ζεί-δωρος. This further evidence which we obtain from the clarification of the relation between ζειά and ἤια supports the conclusion that the root must be reconstructed with initial γ-.

10.4d. ζέω ‘seethe’ has cognates in Skt. yásati, yásyati ‘seethes’, and OHG jesan ‘foam’. The pret. ptc. in Skt. is yastá, compare Gk. ζεστός; from yájati, compare Gk. ἅζεται, however, a Skt. pret. ptc. iṣṭá is found. When we compare Skt. paktá we infer that the initial of yaktá, like the initial p- of paktá could not become vocalized when the e of most to-forms was lost. I assume therefore that the initial was not simple y- but rather a cluster of laryngeal and y.

10.4e. Gk. ζῡ́μη ‘dough’ has cognates in Skt. yū́s, Lat. jūs ‘soup’, Lith. júšė ‘poor soup’ and the verbs Skt. yuváti yaúti ‘mixes’. The root from which these are derived is identical with that of ζεύγνῡμι; I therefore explain both initials in the same way.

10.4f. The other Gk. words with initial ζ- from y- have so few cognates or are so unclear in etymology that we can derive no evidence from them. ζόρξ ‘deer’ has a cognate only in Welsh iwrch. ζημίᾱ ‘difficulty’ and ζη̑λος have been 1. related to Skt. yavan ‘aggressor’, Slav. jarъ ‘violent’ and 2. derived from IE dy-, cf. δίζημαι. ζητέω ‘seek’ has been derived from *di ̯a-tei ̯o, and by other Indo-Europeanists connected with Skt. yátati ‘attack’. ζωρός, ζέφυρος, ζαρου̑ν, and ζάψ are obscure in origin.

10.5. Origin of such ζ in /γy-/

For the five groups of words with clearly identifiable cognates in other dialects I have found evidence for initial laryngeal: for γy- in ζωστός; for γy- or ʔy- in ζειαί. Cognates of ζυγόν, and ζέω give us no information to identify the laryngeal. But because Sapir has demonstrated that initial Hy- (voiceless laryngeal plus y) became Gk. ʽ , I assume that the X of ζειαί was γ, and that the unidentified laryngeals were also γ. The positive evidence from the Hittite cognates and the negative evidence from Sapir's formulation support each other.

I conclude that the ζ in these words developed from PIE γy-. If we assume such a cluster, we find that there was complete parallelism in the development of Gk. ζ. Initial ζ with the possible exception of the ζ in ζείναμεν, has its origin in three PIE clusters: dy, gy, γy, all of which are groups of voiced obstruent plus y.

This conclusion has been reached from an analysis of cognates of these words in Gk. and other dialects. It may be supported by metrical evidence. In verse ζ makes position in these and in most other words. Since clusters alone make position, analysis of ζ in ζυγóv etc. as a reflex of a cluster is much more probable than the assumption that it developed from a simple phoneme.

From the difference in development of PIE laryngeal plus y I assume that we have evidence for the survival into PGk. of voiced and voiceless (reflex of) laryngeals. Our Gk. evidence requires that we reconstruct with initial laryngeals PIE forms which formerly were reconstructed with initial y-.


1 H. Pedersen, Das Pronomen ΥΜΕΙΣ und das Idg. J. im Griechischen, Symbolae Philologicae. O. A. Danielsson. 262-8 (Uppsala, 1932). E. Schwyzer, Griechische Grammatik I.331. (München, 1939).

2 F. Sommer, Griechische Lautstudien 139-45. (Strassburg, 1905).

3 See fn. 1, and KZ 36.74-110 (1900), and Die idg. semitische Hypothese und die idg. Lautlehre, IF 22.341-65 (1907-8).

4 Proponents cited by Schwyzer, Gr. Gram. I.330. Finck, Initial Indo-European y in Greek, TAPA 68.120-2 (1937) suggests it again.

5 G. Schulze, Über das Verhältnis des Z zu den entsprechenden Lauten der verwandten Sprachen. (Göttingen 1867).

6 E. Sapir, Glottalized Continuants in Navaho, Nootka, and Kwakiutl, Language 14.248-78 (1938).

7 IHL 51. Hendriksen, who assumes a root seH(i)-, finds the of iš-ḫa-a-i and other forms difficult to explain.

8 A. Götze, IF 42.327ff. and Couvreur, Hett . 250 and 325. Additional references are cited in Edgar H. Sturtevant, A Hittite Glossary 184 2d ed. (Philadelphia, 1936).

9 Arnold, Vedic Meter 126ff.

10 Sommer, Gr. Lt. 154. Boisacq cites other references, Ety. Wb. 316 and 307.