Proto-Indo-European Phonology

3. The Laryngeal Theory

3.1. Definition of the laryngeal theory and a sketch of its history

The laryngeal theory is the name commonly given to an assumption made about the phonological system of an early stage of Indo-European. It is assumed that this system included a number of phonemes, usually called laryngeals, of which the various IE dialects other than the Anatolian languages show no direct reflexes. Evidence for laryngeals in the IE dialects can be deduced only on the basis of phonemes that are also reflexes of other PIE phonemes. In Gk. δω̑ρον ‘gift’, long o had developed in pre-IE from a combination of vowel and laryngeal; in φώρ ‘thief’, from lengthening of short vowel upon loss of the following vowel. Because the reflexes of laryngeal phonemes merged with reflexes of other PIE phonemes, the possibility that laryngeals had once been IE phonemes was recognized relatively late in the study of IE historical phonology; even then it was almost completely disregarded for fifty years.¹ Only after Kurylowicz pointed out reflexes of laryngeals in Hittite did Indo-Europeanists generally begin to accept the laryngeal theory.

In the twenty years that have elapsed since Kurylowicz suggested that the ḫ in Hittite iš-ḫa-a-i ‘binds’ and ḫ, ḫḫ in other words corresponds to an assumed IE laryngeal,² Indo-Europeanists have published a great number of works in which the theory has been examined, some seeking further evidence in Hittite, others finding evidence in the IE dialects. Such evidence was sometimes presented together with inadequately supported explanations for previously unsolved phonological developments; in some such studies the chronology of the loss of laryngeals has been inadequately treated; as a result such explanations and laryngeal theory have often been rejected.³ With the evidence that has been assembled in favor of the theory we can hardly reject it.⁴ If it is valid, we should expect it to clarify some of the unsolved phonological and morphological problems of IE linguistics. We might also expect that our reconstructions of PIE will have to be revised.

With such a revision we may be able to distinguish various stages of IE. Some of the confusion caused by Hirt's presentation of IE results from his failure to make this distinction. A more accurate statement of the development of IE is probably the greatest current need in IE phonology. Such a statement can be made on the basis of the analysis of the changes in the IE vowel system, especially with the clarified form of that system which the laryngeal theory provides.

3.2. Arguments adduced by Saussure and Möller for the theory

Two different arguments were advanced for the laryngeal theory when it was first proposed. Saussure arrived at it by comparison of the PIE ablaut classes. Since the vowel interchange in Gk. ἵστᾱμι : στατός is parallel to that of πείθω : ἔπιθον, he assumed an earlier similarity in root structure and deduced that στᾱ- was a reflex of /steA-/. Saussure's basic assumption was that a similarity—such as the similarity between (σ)ἔχω, λείπω, ἵστᾱμι and ἔσχον, ἔλιπον, στατός of phonemic variation in morphemes of a seemingly different structure pointed to anterior forms of a similar structure. Saussure's assumption has been upheld by the discovery of confirming evidence in Hittite.

The second argument was suggested by linguists who assumed a common origin for IE and Hamito-Semitic. The first of these was Möller. Shortly after Saussure proposed his theory of two lost sonants, Möller suggested various correspondences between PIE and Proto-Hamito-Semitic forms. Because the PHS phonological system contained laryngeals, Möller assumed them for Proto-Euro-Hamito-Semitic and identified them with the lost pre-IE phonemes that Saussure had postulated. The advantage of Möller's argument lay in his ability to cite earlier, though reconstructed, forms with laryngeals; the disadvantage, that the forms he cited were so far removed from the various IE developments that the interrelationships seemed nebulous.

PHS need no longer be used as a support for the laryngeal theory. Hittite, in which were found forms with orthographical evidence for reflexes of laryngeals, has replaced it as a source of forms to provide actual evidence for laryngeals.

From the first there have been points of difference between various forms of the laryngeal theory. When Saussure's basic assumption of a similarity in ablaut was rigorously applied to the various PIE developments, it was suggested, by Möller and later Cuny, that three laryngeals rather than two had to be assumed. For the ablaut relationships of ἔχω and λείπω are parallel to those of τίθημι as well as those of ἵστᾱμι and δίδωμι. There is still wide disagreement about the assumption of three laryngeals because of three different vowels; some Indo-Europeanists ascribe the ο of δοτός to coloring by a neighboring laryngeal, others explain it as an analogical vowel. (SSP 168-78.) But since the time of Möller's suggestion most Indo-Europeanists who accept the laryngeal theory find in the IE dialects evidence for the assumption of three laryngeals A, ʔ, γ, although some assume from IE evidence only two, E and A.

3.3. Hittite evidence in favor of these arguments

The great importance of Hittite for the laryngeal theory is that it contains forms in which distinct reflexes of the laryngeals are recorded. These are found where we should expect to find them in accordance with theoretical analysis of IE phonological sequences and form categories.

Saussure had assumed earlier laryngeals from the PIE patterns given below; Hittite forms which have been cited in support of his assumptions are listed after these.

A. Long vowels that had not arisen in ablaut he assumed were due to lengthening of short vowels upon loss of laryngeals, thus stā- < /steA-/. (Mém. 135.)

Hittite preserves some of the uncontracted forms, in which the laryngeal had not yet been lost. Corresponding to long a in Latin pāscō is -aḫ- in Hittite pa-aḫ-ša-an-zi ‘protects’; compare also Gothic mēl ‘time’, Hittite me-e-ḫu-e-ni ‘time’.

B. The long resonants, ī ū ṝ ḹ ṃ̄ ṇ̄, which Saussure posited, developed according to him from short resonants with loss of laryngeals.

We find the reflex of a laryngeal in Hittite pal-ḫi-i-iš ‘broad’, compare Latin plānus ‘flat’.

C. Such long resonants are found in the zero grade forms made from Skt. seṭ-roots (Mém. 248ff.), e.g. bhávi-tum ‘be’, bhū-tá; vámi-ti, van-tá; váni-tā, vā-tá. On the basis of such forms Saussure assumed that seṭ-roots developed from PIE roots which ended in sonants, that is, laryngeals; Skt. bhavi- would point to PIE /bhewX-/, vami- to /wemX-/, vani- to /wenX-/.

In Hittite we find cognates of seṭ-verbs written with laryngeals. Compare with Skt. sanóti ‘gains’, pret. ptc. sātá, Hittite ša-an-ḫa-an-zi ‘conquers’; with Gk. περάω ‘traverse’, Hittite pár-ḫa-an-zi ‘drives’.

It might be added here that this analysis of seṭ-verbs showed a similarity of formation in three Skt. verb classes: (Mém. 257; BHL 17.)

After the ninth class was analyzed on the basis of the laryngeal theory, it became clear that these three classes developed from one earlier class in which n was infixed. That a phonological analysis clarified a morphological relation helped to support the laryngeal theory.

D. Saussure also assumed that all IE roots had a basic vowel e. (Mém. 133.) Other root vowels in words such as Gk. ἄγω ‘I lead’, in which PIE a seemed to occur in a form where a basic vowel is expected, Saussure explained as derived from a laryngeal; derivation from laryngeal plus vowel later seemed more plausible, i.e. /Aeg-/ > /Aag-/ > /ag-/.

In Hittite we find numerous examples showing stage two, whereas the IE dialects have stage three; Hittite ḫar-ki-iš ‘white’, Gk. ἀργής ‘white’; Hittite ḫa-an-ti ‘front’, OHG enti ‘front’.

After the Hittite cognates were found, the postulations of Saussure and Möller were raised to the status of explanations; for now, although the older forms were few, IE reconstructions could be supported with earlier forms rather than with phonological formulae or remote reconstructions.

3.4. The Hittite evidence in favor of the laryngeal theory

Even though Hittite has supplied the clinching evidence for the laryngeal theory, the Hittite evidence is not without difficulty, and almost disappointing as a support for the theory. Hendriksen has listed the Hittite words in which ḫ, ḫḫ gives evidence of laryngeal, (BHL 27-33) and their number is relatively small; I have added to his list the additional words cited in IHL. The combined list is given below in five groups.

• A. Words with ḫ initially:
  1. ḫa-an-na-a-i ‘judge’ : Lat. ā ‘from’ (example of γe prefix; not in BHL)
  2. ḫa-an-na-aš ‘grandmother’ : Lat. anus ‘old woman’
  3. ḫa-an-ti ‘front’ : OHG enti ‘front’
  4. ḫia-a-ra-aš ‘eagle’ : OHG arn ‘eagle’
  5. ḫa-me-eš-ḫa-an ‘spring and summer’ : Gk. ἀμάω ‘reap’ (not in BHL)
  6. ḫar-ak-zi ‘be destroyed’ : OIr. orgaim ‘destroy’
  7. ḫar-ki-iš ‘white’ : Gk. ἀργής ‘white’
  8. ḫar-kán-zi ‘have’ : Lat. arceō ‘enclose’ (questioned in BHL)
  9. ḫar-ra-an-zi ‘break’ : Gk. ἀρόω ‘I plough’ (doubtful in IHL, not in BHL)
  10. ḫa-aš-ši-i ‘hearth’ : Lat. āra ‘hearth’ (not in IHL)
  11. ḫa-aš-ta-i ‘bone’ : Skt. ásthi ‘bone’
  12. ḫa-at-ta-an-zi ‘pierces’ : Arm. hatanem ‘I cut’ (not in BHL)
  13. ḫa-at-ki ‘close’ : Skt. átka ‘mantle’ (not in BHL)
      (ḫatti : Arm. hay ‘Armenian’; loan-word in both languages; not in BHL)
  14. ḫé-kur ‘peak’ : Skt. ágra ‘peak’
  15. ḫi-in-ik-zi ‘bows’ : Gk. ὄγκος ‘curve’ (IHL connects the Hittite words in 15 and 16; BHL proposes the connection of Hittite 15 with the Gk. word)
  16. ḫi-en-kán ‘death’ : OIrish écen ‘necessity’
  17. ḫu-uḫ-ḫa-aš ‘grandfather’ : Lat. avus ‘grandfather’ (doubtful in BHL)
  18. ḫu-uk-zi ‘charms’ : Gk. ἄγνῡμι ‘I break’ (not in BHL)
  19. ḫu-la-a-li-az-zi ‘wraps’ : Skt. vr̥ṇóti ‘covers’ (not in BHL)
  20. ḫu-ul-la-an-zi ‘destroys’ : Gk. ἁλίσκομαι ‘am captured’
  21. ḫur-na-an-zi ‘sprinkles’ : Skt. vāri ‘water’ (not in BHL)
  22. ḫu-wa-ar-ta-aš ‘curse’, ú-e-ri-ya-zi ‘names’ : Gk. εἴρω ‘I say’
  23. ḫu-u-uš-ki-ši ‘awaits’ : Lat. aveō ‘I desire’ (not in BHL)
  24. ḫu-u-wa-an-te-eš ‘winds’ : Lat. ventus ‘wind’
  25. ḫu-i-iš-zi ‘lives’ : Skt. vásati ‘dwells’
  26. ḫu-u-i-tar ‘animals’ : Goth. wáit ‘know’ (not in BHL)
• B. Words with ḫ, ḫḫ medially:
  • a. Between consonant and vowel:
    1. e-eš-ḫar ‘blood’ : Skt. ásr̥k ‘blood’
    2. iš-ḫa-a-aš ‘master’ : Lat. erus ‘master’
    3. iš-ḫa-a-i ‘binds’ : Skt. syáti ‘binds’
    4. dam-me-eš-ḫa-an-zi ‘injures’ : Gk. δαμάσσω ‘subdue’ (not in BHL)
    5. pal-ḫi-i-iš ‘broad’ : Lat. plānus ‘flat’
    6. ša-an-ḫa-an-zi ‘strives’ : Skt. sanóti ‘gains’
    7. par-ḫa-an-zi ‘hunts’ : Gk. περάω ‘traverse’
    8. tar-ḫu-un ‘conquers’ : Skt. trāti ‘protects’ (ti-eš-ḫa-aš ‘kept away from’ : Skt. dásyati ‘lacks’; not in IHL, doubtful in BHL)
  • b. Between consonant and consonant:
    • Only inflected forms, such as the verb forms ša-an-aḫ-mi, and ša-an-aḫ-zi, and the noun form e-eš-ḫa-ni. Possibly iš-ḫu-u-wa-i ‘throws’ : Skt. iṣṇā́ti ‘sends’; not in IHL, doubtful in BHL
  • c. Between vowel and consonant:
    1. e-ḫu ‘come, go’ : Skt. yā́ti ‘goes’ (not in BHL)
    2. e-ḫu-ra-da-iz-zi ‘stop one's ears’ : Gk. οὖς ‘ear’ (not in BHL; Sommer, HH 43, rejects connection with οὖς)
    3. ir-ḫa-a-aš, ar-ḫa-an ‘boundary’ : Lat. ōra ‘boundary’ (not in BHL)
    4. pa-aḫ-ša-an-zi ‘protects’ : Lat. pāscō ‘I protect’
    5. na-aḫ-za-ri-ya-az-zi ‘is afraid’ : Irish nár ‘modest’
    6. la-a-ḫu-i ‘pours’ : Lat. lavō ‘wash’
    7. me-e-ḫu-e-ni ‘time’ : Goth. mēl ‘time’
    8. pa-aḫ-ḫu-e-ni ‘fire’ : Gk. πυ̑ρ ‘fire’
    9. tuḫ-ḫu-iš ‘smoke’ : Gk. θύος ‘incense’ (doubtful in BHL)
    10. tuḫ-ḫu-uš-zi ‘takes off’ : Gk. δαιδύσσασθαι ‘draw away’ (doubtful in IHL; not in BHL)
  • d. Between vowel and vowel:
    1. la-aḫ-ḫa ‘war’ : Gk. λᾱός ‘army, people’ (not in BHL)
    2. ma-aḫ-ḫa-an ‘when, as, if’ : Dor. Gk. μᾱ́ν ‘truly, surely’ (not in BHL)
    3. pí-e-ḫu-te-iz-zi ‘leads away’ : Lith. vedù ‘lead’
    4. še-e-ḫur ‘urine’ : OIcel. súrr, saurr ‘semen’
    5. wa-ḫa-an-zi ‘turns’ : Skt. vāya ‘weaving’
       Also the forms: pa-aḫ-ḫu-ur ‘fire’, me-ḫur ‘time’ of words already cited, and na-aḫ-ḫa-an ‘obeisance’, cf. naḫ- ‘fear’.
    6. Furthermore the verbal endings: -ḫḫari, -ḫḫi, -ḫḫuen are used with a great number of verbs, e.g. da-aḫ-ḫi ‘I take’, a-ar-ḫu-un ‘I arrived’, etc. The -ḫi ending may be compared with the -α of Gk. οἰ̑δα ‘I know’.
    7. -ḫḫ- in factitive verbs and nouns, e.g. ne-wa-aḫ-ḫu-un ‘renewed’; Lat. novāre ‘renew’; al-wa-an-za-aḫ-ḫa-az ‘witchcraft’.

This evidence is small, but when one compares it with the total amount of lexical evidence for a Hittite-IE relationship, it is quite remarkable.

3.4a. But the small number of words with reflexes of laryngeals is not the greatest shortcoming of our Hittite material. Hittite is written in an orthographic system which itself presents many problems. Couvreur assumed different sounds from the writing of single and double ḫ. (Hett. ḫ. 188, 193, 194.) But when Hendriksen investigated the occurrences of ḫ : ḫḫ, he noted that the single writings of ḫ, like those of n, m, l, r, and š are found after e. (BHL 38-42.) Because of this patterning he ascribes the occurrences of ḫ or ḫḫ to the presence of the preceding vowel, rather than to a difference in laryngeals. Speiser on the other hand assumes that the voiced : voiceless distinction indicated in Hurrian is also true of Hittite; Sturtevant, following Speiser, concludes that ḫ is a voiced velar or post-velar spirant, the voiced counterpart of ḫḫ. (Speiser, Lang. 16. 319—40; see IHL 34 for Sturtevant's views, and further bibliography.) It is difficult to decide in favor of either assumption. For a decision depends on phonological criteria which have not been independently assembled. If the Hittite scribes grouped ḫ with p t k, as one of the Hittite obstruents, then they presumably were distinguishing between a voiced and a voiceless counterpart. If they grouped ḫ with l r m n š, as one of the Hittite resonants, they were scarcely making a distinction in voicing. Before we can use Hittite variation in orthography with full certainty in support of one or other forms of the laryngeal theory, this and other orthographical problems must be solved from Hittite phonological evidence. Since Hittite provides no eḫḫ : eḫ, or aḫḫ : aḫ contrast, but merely an aḫḫ : eḫ contrast, it is not wholly certain that we can distinguish different laryngeals from this orthographical variation.

In reading Hittite one is struck by the variation in orthography. [ezzi] ‘he eats’ may be written e-iz-zi-az-zi, e-za-az-zi, e-iz-az-zi, [uk] ‘I’ may be written ú-uk, ú-ga, ú-ug-ga, and so on. Moreover we find variation in usage of forms. The lack of standard spellings and fluctuation in use of forms may reflect variations in usage during the period of time in which Hittite texts were written. It may also indicate dialect variation; the Hittite texts contain many Luwian forms which are marked. But apparently some dialect (or Luwian) forms were not especially marked. Sommer has pointed out that more and more Luwian forms that are not marked as Luwian are being found in the Hittite texts. (HH 18.)

Whatever may be the causes of variation, it is clear that the description of Hittite is left with many problems; before these are solved Indo-Europeanists must use Hittite material with caution.

3.4b. It has also been pointed out that Hittite has preserved reflexes of laryngeals only in certain phonetic environments. Hendriksen found, when examining the Hittite material where we should expect laryngeals internally in words, that they have been represented by ḫ : ḫḫ only before and after r l m n y w and š. (BHL 86.) We may infer that in other positions laryngeals were lost. This inference is supported by the observation of Pedersen that Hittite t did not become z before i if it occurred in a form that previously contained a laryngeal between the t and i, such as ti-it-ti-an-za < *ti-tA-iya-, compare Gk. ἵστᾱμι, as against tu-uz-zi-iš ‘army’, compare Gothic þiuda ‘people’. (Hitt. 87, 175.) And the Hittite forms cognate with the forms made from the root dhē- ‘place’ and dō- ‘give’ give us no evidence for assuming reflex of a laryngeal. (But cf. IHL 55.) In Hittite, vowel plus laryngeal plus consonant may have yielded long vowel plus consonant, unless the laryngeal was restored by analogy. Since quantity of vowels is not marked in Hittite texts, cognates of dō- and dhē- would presumably not give us written evidence for laryngeals.

For these reasons with our present store of Hittite materials we are left with deductions from other IE dialects as our most secure evidence for the laryngeal theory. Hittite has served to put the theory on a firm basis; the theory must be refined from analysis of phonological developments such as ablaut in the various IE dialects.

3.5. Various approaches to the laryngeal theory

Various linguists differ markedly in their evaluation of the laryngeal theory. We may divide the approaches to the theory into three groups:

• A. those that reject it;
• B. those that superimpose it on the phonological system set up for PIE by Brugmann;
• C. those that assume the laryngeal theory for a stage of IE which antedates that reconstructed by Brugmann.

A. Linguists in this group, which is continually decreasing, dismiss the Hittite evidence as inadequate and prefer to assume no laryngeals for PIE.⁵ Until 1920 or even 1930 the evidence published may have been insufficient. But with the gradual increase in the number of available Hittite texts the evidence is almost incontestable. The material assembled from Hittite may not equal in sheer quantity that for the Gmc. consonant shift, but in proportion to the Hittite vocabulary with cognates in IE dialects it is quite convincing.

One may find as counter-arguments statements that the laryngeal theory has been used to explain all the difficulties left in IE phonology. Even if this were so, it would not disqualify the theory. False extensions of a theory do not alter its validity. Moreover, if the laryngeal theory explained nothing, there would be little point in holding it. It is quite obvious that the laryngeal theory demands a change of analysis of some of the most important IE form classes, such as the set-roots. Or stated differently, laryngeals were once present in a high proportion of IE words. We therefore expect some change in explanations of IE difficulties.

B. At present the greatest number of Indo-Europeanists, Benveniste, Couvreur, Cuny, Hendriksen, Kurylowicz, Messing, Pedersen, and Sommer, fall into this group;⁶ they assume that Hirt's and Brugmann's phonological systems are to be retained in general, but modified with respect to data that the laryngeal theory clarifies. Thus, long vocalic ṝ is not analyzed ṝ or ırə, but rather r̥X; original ā is analyzed aA, a is analyzed Aa, and ēi is analyzed eʔi. The old established orthography for reconstructions, ē for eʔ, etc. may, however, be maintained. (BHL 95.)

There has been much discussion between linguists of this and the next group, with forceful arguments presented on either side. (See IHL 23–6 and references there.) Most of the discussion, however, centers about the Indo-Hittite hypothesis. This hypothesis is beyond the scope of my work. For in discussing the laryngeal theory it is unnecessary to take into account all of the morphological and phonological arguments advanced in favor of the IH hypothesis. The present work is concerned only with considering those parts of the discussion which apply to the laryngeal theory.

C. Present-day linguists who retain for PIE almost exactly the phonology of Brugmann are Sturtevant, Austin, Kerns, Schwartz, and Smith. (IHL 20–1.) According to them the laryngeals disappeared in pre-IE.

They are faced with numerous problems. Some troublesome phenomena which have been explained with the help of the laryngeal theory are found restricted to one, or possibly two, dialects, for example, the Ind.-Ir. voiceless aspirated stops. These Indo-Europeanists must explain such phenomena by assuming that the change was made in PIE and the effects were lost in virtually all dialects. Thus all PIE p t k before voiceless laryngeals became ph th kh, even though we have extensive evidence for ph th kh only in Ind.-Ir. Although Sturtevant explains voiceless aspirated stops in this way, he considers other theories weakened if they assume a PIE development which was lost separately in each branch of PIE. (IHL 25.)

A similar explanation adds six phonemes, hy, hw, hr, hl, hn, hm, to PIE, although the evidence is taken primarily from one dialect. (IHL 76–8; 90.) Again one questions the advisability of assuming for PIE six phonemes which are lost in all dialects but one, while rejecting the assumption that four or fewer phonemes were preserved in PIE and lost in the dialects.

In the course of this work an attempt will be made to settle this problem. But even on the basis of evidence presented in this chapter, outright rejection of the laryngeal theory seems untenable.

3.6. Evidence in the IE dialects in favor of the laryngeal theory

Before examining phonological problems with the help of the laryngeal theory I shall sum up the evidence for assuming laryngeals in various patterns. I list first those forms on which linguists agree, and then the more doubtful forms; I also give the reasons for such assumptions.⁷

3.6A. IE ə is the reflex of a laryngeal.

There are various reasons for this statement.

• a. In some Hittite cognates ḫ : ḫḫ corresponds to reconstructed IE ə.
• b. With it one can explain the long suspected relation between the Skt. fifth, seventh, and ninth verb classes. For the -i- of seṭ-roots, reconstructed in IE as ə, developed from a laryngeal; the PIE form of seṭ-roots is then to be reconstructed with a final laryngeal, e.g. Skt. pavi- from PIE /pewX-/, not pewā-, pewə-. The Skt. fifth, seventh, and ninth classes have an n-infix. The seventh class is composed of verbs with an obstruent as last phoneme of the root, e.g. rinákti < /ly-n-ekʷ-ty/; the fifth class with w, e.g. śr̥nóti < /kl-n-ew-ty/; and the ninth class with a laryngeal, e.g. punā́ti < /pw-n-eX-ty/.
• c. When one assumes that a consonantal laryngeal is to be reconstructed where formerly a vocalic schwa was assumed, some of the exceptions to an Ind.-Ir. phonological formula known as Brugmann's Law are removed. Much has been published about this formula.⁸ In general Indo-Europeanists no longer accept it. Pedersen, however, has published an acceptable formulation: ‘IE short o before r l m n became Ind.-Ir. ā in open syllables.’
  The formulation may be expanded to: ‘PIE short o before the consonantal allophones of the PIE resonants became in open syllables Ind.-Ir. ā.’ Thus is defined the relation between Gk. δόρυ, Skt. dā́ru, Gk. φέρομες, Skt. bhárāmas, but Gk. πόσις, Skt. páti, Lat. rota, Skt. rátha.
  The forms which were left unexplained even by Pedersen's formulation are clear with the laryngeal theory : such are causative stems like janáya-, and a in the perf. 1st sg. (there is fluctuation between a and ā in the later language but a predominates in the old language) but consistently long ā in the perf. 3d sg., e.g. 1st sg. cakára, 3d sg. cakā́ra. Forms like janáya- are from laryngeal bases; therefore pre-Ind.-Ir. o in these words stood in a closed syllable. The contrast in the perfect forms can be explained by comparison with Hittite. It is assumed that the Hittite ḫi-conjugation is related to the IE perfect. The 1st and 3d singular endings of the ḫi-conjugation are -ḫi and -i; by comparison with these the pre-IE endings may be reconstructed: 1st -Xe and 3d as -e. Consequently in the 1st sg. of perfects like cakára o stood in a closed syllable and regularly became a; (later ā was introduced in many forms by analogy with the third singular). In the 3d sg. o stood in an open syllable and regularly became ā. The forms 1st sg. cakára, 3d sg. cakā́ra thus represent the expected development. With the laryngeal theory we can therefore explain the chief difficulties and apparent exceptions to Pedersen's revision of Brugmann's Law.
• d. We may conclude from this evidence that any well-attested seṭ-root gives evidence of IE laryngeal. Any wide-spread formation, however, like the seṭ-roots may spread by analogy; i spread to aniṭ-roots especially in certain form categories. For example, in the sya-future, Indic shows i for both seṭ and aniṭ-roots; on the other hand, Avestan has no i in these forms. Persson suggests that both are analogical developments. (Beitr. 350–3.) Hence not all Skt. seṭ-roots may give us incontrovertible evidence for laryngeals; on the other hand some original seṭ-roots may have lost their laryngeals. But in general, Skt. roots are used in such a variety of forms that we can determine which of them are seṭ.
  Seṭ-roots are among the soundest evidence for laryngeals; but they give us information only about the last phoneme of the root. We can most easily define seṭ-roots from our evidence in Skt.; i is clearly preserved in many formations, such as the infinitive bhávitum, the s-aorist ásāniṣam, and other forms with suffix beginning in s or other consonants.
• e. We may also determine the presence of PIE laryngeals from the intonation in Lithuanian, Lettish, and Serbo-Croatian. Lithuanian syllables with acute accent (´ on long syllables, ` on short) point to PIE syllables with laryngeal; Lithuanian syllables with circumflex accent (˜) developed from PIE syllables without laryngeal. To Lithuanian ´ and ` correspond Lettish ˜ or ˆ, Serbo-Croatian  ̑; to Lithuanian ˜ correspond Lettish ` and Serbo-Croatian  ̑. Examples are:
  • Lith. búti, Lett. bût, Serbo-Cr. bȉti ‘to be’; PIE /bhewX-/ cf. Skt. bhávitum.
  • Lith. pìlnas, Lett. pil̃ns, Serbo-Cr. pȕn ‘full’; PIE /pelX-/ cf. Lat. plēnus.
  • Lith. acc. sg. šìrdį, Lett. acc.sg. sir̂di ‘heart’; PIE /keXrd-/, Gk. κη̑ρ.
  By contrast the following are examples of intonation patterns of words whose etyma did not contain laryngeals:
  • Lith. par̃šas ‘boar’, Serbo-Cr. prâse ‘shoat’; Lat. porcus.
  • Lith. añtras ‘the other’; Goth. anþar.
  • Lith. liẽšti ‘lick’; Skt. leḍhi. (See Hirt, IG 5.151–3.)
  Such intonation patterns do not give us evidence for the exact position of the laryngeal. Acute accent may result from the pattern CeRX- or CeXR-, from the so-called heavy bases or the so-called long vowels and diphthongs, for example, Lith. sáulė ‘sun’; Gk. ἠέλιος. Only the presence of a laryngeal may be determined from the intonation patterns; its position must be noted from evidence in other dialects. Moreover the acute intonation spread to verbs with circumflex intonation; this analogical spread is especially characteristic of intransitive verbs. (See Sandbach 11.) Thus Lith. svìlti, Lett. svil̃t ‘smoulder’ developed from etyma with no evidence of laryngeal, cf. OE swelan ‘glow’. In terms of the laryngeal theory Hirt's heavy bases are bases ending in laryngeal. The term ‘laryngeal base’ is therefore preferable both to Hirt's term ‘heavy base’ and to the term ‘seṭ-root’ which is taken over from Skt. grammar.

3.6B. The so-called ‘original long IE vowels’ are reflexes of short vowel followed by laryngeal; for all such long vowels we may assume earlier short vowels and laryngeal. The following evidence may be cited.

• a. Cognates with IE words having long vowels are found in Hittite with vowel followed by ḫ : ḫḫ.
• b. The ablaut relationships found in forms like Gk. τίθημι, ἵστᾱμι, δίδωμι is clarified. If one compares these with verbs like Skt. ás-mi, Gk. δέρω, one can analyze them as follows:

  PIE root			PIE root
  ás-mi	es-		τίθημι	dhē- < dheʔ-
  δέρ-ω	der-		ἵστᾱμι	stā- < steA-
  ‘flay’
  μέv-ω	men-		δίδωμι	dō- < deγ-
  ‘stay’

  With the assumption of root forms /dheʔ- dhʔ-/ some of the problems in the formation of Skt. reduplicated verbs become clear. We find, for example, 1st sg. bíbharmi, 1st pl. bibhr̥más ‘bring’, but dádhāmi, dadhmás ‘set’ and dádāmi, dadmás ‘give’. Although apparently different the plural forms are quite parallel. In bibhr̥más, -r- became syllabic between two consonants; in dadhmás and dadmás, -X- was lost, as regularly between consonants.
• c. Most of the problems connected with IE ‘long semi-vowels’ are clarified. When laryngeal bases with resonant before laryngeal are found in formations with accent on the suffix, the base shows a ‘long semi-vowel.’ Skt. gūrtá, Lat. grātus, formerly reconstructed gʷṝ-tó-, are from /gʷrX-tó-/, compare Skt. gṛṇāti; Skt. pūrná, Lat. plēnus are from /plX-nó-/; Skt. bhūtá is from /bhwX-tó-/. IE ‘long semi-vowels’ then are similar in origin to the original long vowels. ‘Long semi-vowels’ may also have developed from laryngeal plus /y w r l m n/ . Skt. sthāv-ará is an extension of /steX-/; we find alongside it the form sthūrá; compare also sphāy-ate, sphīta, Lat. fēmina, fīlius (from /dheXy-/), etc. This assumption is not made by some Indo-Europeanists; but those linguists who derive ‘long semi-vowels’ only from resonant plus laryngeal are led to assume earlier forms for which we have no evidence, such as *bhwıX-w-′.
• d. The treatment of ‘original long diphthongs’ points to anterior presence of a laryngeal. In Cretan Gk. ᾱ̓Ϝέλιος ‘sun’, short vowel plus laryngeal have contracted to a long vowel; presence of laryngeal is supported by the intonation in Lithuanian sáulė; in Skt. sū́ra ‘sun’ we find the expected zero grade.
  From long vowels and diphthongs we can derive evidence for laryngeals medially in roots, as well as finally. Skt. naú ‘ship’ gives us evidence for /neXw-/, dadhā́mi for /dheX-/.

3.6C. IE a developed from Aa which in turn came from Ae.

The following evidence may be cited.

• a. We find Hittite cognates written with ḫa-, for which the IE cognates do not have initial e, rather a.
• b. a- is found in ablaut relationship with zero, e.g. Gk. ἄ(Ϝ)ησι, Skt. vā́ti; the Hittite cognate ḫu-wa-an-te-eš gives us secure evidence for initial laryngeal. If we assume a pre-IE root /Aew-/ [Aaw-] to which suffixes were added, it is clear that the alternation was aw : w. a < /Ae/ was preserved before w; with loss of the vowel /e/, A too was lost. Forms with similar ablaut are Lat. aurōra, Skt. uṣás, Lat. aemulus, imitor. (See also Pedersen, VGK 107.) From original a we can derive evidence for laryngeals occurring initially or medially.
  Since e and o may be adjacent to two of the laryngeals, we must assume other laryngeals adjacent to full grade a. Furthermore, a number of IE words with initial a have Hittite cognates with ḫa, e.g. ḫa-an-ti, OHG enti, anti, while others have Hitt. cognates with a-, e.g. Lat. aurōra, Hitt. a-uš-zi. Consequently some Indo-Europeanists assume two a-colored laryngeals: 1. x, which became ḫ in Hittite; 2. h, which is not represented in the Hittite orthography. (This point has been the subject of much dispute; see SSP 206–22.)

3.6D. The aspiration of Skt. voiceless aspirated stops developed when laryngeals were lost after original voiceless stops. This clarifies the relation between tíṣṭhati, Gk. ἵσταμεν, Lat. sistit, Goth. standan, etc.; aspiration developed in the zero-grade forms of the root /steA-/, and spread to forms with other grades. (See EI 46–53.)

From words with voiceless aspirated stops in Skt. we can assume laryngeals medially in many words.

3.6E. The quantity of vowels in Vedic Sanskrit often demands that we assume a laryngeal. For example, Vedic abhí normally retains its final short vowel; in some compounds, however, we find long i. The second member of most such compounds is from a root beginning with laryngeal. (EI 30-3.) abhī́sat is composed of abhí and Xes-. Moreover, augmented forms of such roots with initial laryngeals often have a long augment, e.g. Skt. āyan, impf. 3d pl., PIE /Xey-/ ‘go’.

Greek metrics gives us further evidence that laryngeals formerly were found initially in certain words. Such words beginning with λ, μ, ν make position, e.g. νέφος. Assumption of an initial laryngeal is supported by cognates with /en-/, from /Xen-/, in other dialects; compare Skt. ámbhas, Gk. νέφος. (Lang. 17.91–2; LaRoche, HU 49–59 gives a complete list of such words.)

Vedic metrics also demands in some passages that we read two short vowels where one long vowel is written, e.g. yaánti where yā́nti is written. Kurylowicz, EI 33-40, cites many other examples; the superlatives in -iṣṭha- made from roots with long final vowel, e.g. déṣṭha from dā-, must often be read with a bisyllabic e, that is to say, with -aXi-. In such passages a laryngeal formerly stood between the two short vowels, then was lost, and the vowels coalesced to a long vowel.

From these irregularities we can determine laryngeals initially, medially, and finally in words. The irregularities are rare. We may assume that they were removed as the Vedic and Homeric poems were handed down. If we find them surviving, we can look upon them as good evidence for earlier laryngeals.

3.6F. Developments of PIE initial resonants in some Gk. words give evidence of former laryngeal.

Sapir demonstrated this by clarifying the treatment of initial r w and y on the basis of the laryngeal theory. (Language 14.269-74; also Language 17.88-91 for a recapitulation of this explanation by Austin.) Sapir suggests that:

With this assumption the complicated relationship between Hitt. hu-wa-an-da-aš ‘wind’, Skt. vā́ti ‘blows’, Gk. αἵνω ‘winnow’, and Gk. ἄησι ‘blows’ is explained. The pre-IE initial was /xewe-/; the Hitt. and Skt. forms, and Gk. αἵνω are from the variant /xwe-/, Gk. ἄησι from /x_ewe-/ > /we-/ > /α-/.

From such forms we can determine the presence of laryngeals initially; since in Gk., resonants in these words undergo unvoicing, Sapir ascribed the effect to voiceless laryngeals.

3.6G. Hittite apparently preserves an initial a in the reduced grade of roots beginning with a laryngeal; compare Lat. est, sunt; Hitt. e-eš-zi, a-ša-an-zi. Austin, following a suggestion of Möller's, assumed that Gk. and Armenian too preserved initial vowel in such words; he ascribes vocalic prothesis of many words to an earlier laryngeal, e.g. Hom. Gk. ἐών ‘being’. Compare also Gk. ἀνήρ and Skt. nára; this assumption is strengthened by the numerous Vedic compounds in which the final vowel of the first element has been lengthened, such as sū-nára. Austin assumes that prothesis is found where initial ʼ , γ, x were the laryngeals. (Lang. 17.85.)

These are various phenomena on the basis of which we can suggest laryngeals. Some of them may be found in different developments from the same word, possibly in different dialects; if so, the evidence is convincing.

Laryngeals have been assumed also for other patterns; (IHL 83 and 86-7) these, however, are found in a small number of words; moreover, since these patterns have been explained on the basis of the laryngeal theory, we must use them with caution as further evidence.

3.7. Suggestions about the allophones of laryngeals

We have little evidence for the phonetic description of laryngeals. Some Indo-Europeanists have preferred to deal with them as phonemic units whose allophones we cannot define without further material, except by observing their effect on neighboring phonemes. But assumptions about the pronunciation have been made from the spelling for the Hittite reflexes ḫ, ḫḫ. In the cuneiform syllabary this spelling indicates velar spirants. Moreover, an instance of spelling substitution has been noted in a Hittite text : warnu- and waḫnu-. Since this spelling is limited to one text we can hardly draw far-reaching conclusions about a similarity of articulation between r and ḫ. At best we have Hittite information about two laryngeals, for that is the maximum number for which we have Hittite reflexes.

Further evidence has been drawn from loan-words; like other inferences drawn from loan-words such material must be used with great caution. Our lack of knowledge of the sources of the Hittite vocabulary makes this source doubly precarious.

Unfortunately no source yet found has given us satisfactory material on which we can base phonetic descriptions. But by combining the various bits of evidence we can begin to make tentative suggestions.

3.8. Laryngeals tentatively assumed for PIE

On the basis of the various studies in PIE phonology in connection with the laryngeal theory, it is clear that we must assume a phonemic system of PIE with laryngeals (see 3.5B) or reflexes of laryngeals (see 3.5C). In citing PIE reconstructions in this book I write laryngeals, before deciding from the evidence assembled in favor of either alternative.

Some phonological problems of Gmc. will now be investigated on the basis of a PIE phonemic system with laryngeals. An attempt will then be made to set up a more rigorous system, one that accounts for all of the phonological developments of PGmc.

This PIE phonemic system will also be tested in the investigation of problems from other IE dialects. After such investigations some of the following problems may be answered.

1. When were the laryngeals lost?
2. Can we set up various stages of PIE and pre-IE?
3. Do we have evidence to determine how many laryngeals were lost and what was their phonetic description?
4. What is the earliest phonemic system that we can assume for pre-IE?

Footnotes

¹ Consonants that had disappeared in the IE dialects, leaving reflexes in lengthened vowels, were first suggested by Saussure in 1878. Möller, Eng. St. 3.157fn. (1879), connected these with Semitic consonants, and called them laryngeals. Other linguists who published statements between 1878 and 1927 upholding the laryngeal theory are Pedersen and Cuny.

² J. Kurylowicz, ə indoeuropéen et ḫ hittite, Symb. Gram. I.95-104. This publication is found in but few American libraries; a readily accessible review by C. Marstrander may be found in NTS 3.290-5 (1929).

³ G. Bonfante, rev. of IHL, CP 39.51-7 (1944), rejects the theory and lists others who have.

⁴ Objections are sometimes raised to explanations of IE phonological developments by means of the laryngeal theory; the reasons given for such objections are that so few examples are adduced, cf. CP 39.51-7. Such objections are invalid if one examines realistically evidence for linguistic developments which is found in a stage of that language a thousand or more years after the developments occur. In modern English the past forms was : were furnish the only surviving evidence for assuming the origin of the English past tense verb forms from IE perfect and aorist forms with a similar interchange. Without a series of similar anterior forms in Old English we would be hard pressed to establish such a statement; yet evidence for such an interchange is very common in other old Gmc. dialects.

Our evidence for propositions of the laryngeal theory is often as meager as that of modern English for the origin of the English past tense forms of irregular (strong) verbs. Indo-Europeanists who hold to the laryngeal theory have therefore varied in their interpretation of the data, and have been criticized by others for building ancient phonological structures on scanty evidence. But the anomalous forms, that the phonological systems of Brugmann and Hirt do not account for, though few, are found in words attested in various IE dialects. The only explanation for such linguistic forms is an older form. If we are to explain the anomalous forms, we have to posit acceptable earlier forms. Proper procedure demands as complete as possible use of relevant forms found in the various IE dialects and analyzed in accordance with established linguistic principles.

⁵ Bonfante, CP 39.57fn., cites the Indo-Europeanists who have published statements in which they reject the theory. Allusion to linguists, however distinguished, who in the past rejected the theory hardly justifies continued rejection of it. As the evidence for it became greater, such linguists might have changed their minds, as indeed one of them did; see Language 6.149-58 and the later publications of E. H. Sturtevant, listed IHL 22.

⁶ See publications listed IHL 20-1, BHL 4-11, SSP, and Sommer, HH 79. No attempt is made to refer here to all articles in which mention is made of the laryngeals; a selection of them is listed in the text. Works which do not present important new arguments for or against the theory are omitted, e.g. A. B. Keith, The relation of Hittite, Tocharian and Indo-European, Indian Historical Quarterly 14.201-23 (1938); likewise those are omitted that assume the theory in explaining IE phonological difficulties, e.g. H. Velten, The Germanic Names of the Cardinal Points, JEGP 39.443-9 (1940).

⁷ Saussure already assumed laryngeals for A B C D; to my knowledge all linguists holding the laryngeal theory accept these propositions. Kurylowicz, RO 4.196-218, and EI 33-46 suggested (E) that evidence for laryngeals may be discovered by analysis of poetry and vocalic quantity; Austin has expanded his suggestion; Sturtevant accepts it, as does Messing with reservations. Sapir, Austin, and Sturtevant accept the statement (F) that laryngeals affected ‘initial’ resonants in Greek; Messing rejects it. Möller, Pedersen, and Austin suggest that prothesis may indicate the presence of a former laryngeal (G); Sturtevant accepts it, Messing rejects it.

⁸ Hirt, IG 2.20-1; H. Pedersen examined the evidence in the various dialects in Wie viel laute gab es im Indogermanischen?, KZ 36.74-110 (1900) and suggested the revision cited on page 87.