¹ The writing of this article is due mainly to the encouragement of Professor Leonard Bloomfield, under whose direction I worked on problems of general linguistics and Indo-European comparative grammar at Yale University during the academic year 1945–6. To his careful reading of the manuscript I owe not a few improvements. For all this, as well as for advice given during my research work, I wish to express my cordial appreciation. I am likewise grateful to Professors Franklin Edgerton and E. H. Sturtevant for having read the first draft of the article and given me their criticism. It will be obvious to readers of this article that I am indebted to them on several points. Thanks are due also to Professor Bernard Bloch for advice and assistance on technical matters, and to my good friend Garner Hoyt for help in preparing the final version of the article.

^1a All references are to the 7th edition unless otherwise stated.

² The superior o represents a reduced vowel (> ә secundum).

³ For a similar reason I do not take into account the so-called ә secundum, which is subject to a good deal of discussion and doubt.

⁴ It should be remarked that the rules about zero grade really apply only to roots containing a sonant after the alternating vowel. Otherwise the full grade is generally found instead, e.g. Sk. sannáḥ and sattáḥ from the root sad-, IE *sed- ‘to sit’, Gk. λεκ-τός, root λεγ-, IE *leê- ‘to gather, etc.‘.

⁵ With a long ī, frequent in Sanskrit for an old ә. Unexplained.

⁶ Edgar H. Sturtevant, The Indo-Hittite Laryngeals 16 (Baltimore, 1942).

⁷ Sturtevant, op.cit. 17 f.

⁸ See footnote 6. A very interesting hint of the existence of an a-colored laryngeal, I believe, is the following fact. While the alternation e/zero etc. is, so to say, constant in the short-vowel roots, there is one case where a number of roots have a instead of e as full-grade vowel. This happens when the vowel is an absolute initial. Now, in this position we may suppose there was earlier an initial laryngeal, as was pointed out by Hermann Möller and by Sturtevant (op.cit. 55 f.). Some examples follow. Root *aydh- ‘to burn’: full grade in Lat. aedēs, Skt. édhas- ‘wood’, Gk. αθος, zero grade in Skt. indhé ‘he kindles’ (with infix -n-); root *aws- ‘to shine’ (as noun 'dawn'): full grade in Lat. aur-ōra, Lith. aušra, aũsta 'it dawns', zero grade in Skt. uṣáḥ, verb uccháti (*us-ske-ti) ‘it dawns’; root *ayzd- ‘to feel ashamed, to venerate’: full grade in Gk. αἰδὀς, verb αἴδομαɩ, Gth. áistan, zero grade in Skt. īdé (*izdáy) ‘he supplicates’. The frequency of the a in this position, while it is apparently absent between consonants (except, of course, in the monosyllabic long-vowel roots, where certainly a laryngeal was present) points to a special phonetic condition. This cannot be simply the initial position, for there are too many exceptions. The supposition that the disturbing factor here too is a laryngeal seems to be the most natural explanation of this peculiarity, and at the same time is indirectly confirmed by it. (See also §11.)

⁹ The symbols are Sturtevant's (op. cit. 23). I take sub-type (c) as very probably a case of (a) and (b), in which no form with the full vowel e survived. See on this point Sturtevant, Hittite Evidence Against Full-Grade o, Lang. 14.104–11.

¹⁰ See §8 and §9 on this point.

¹¹ I omit here the thematic vowel, which belongs only to the present.

¹² Sturtevant, IH Lar. 16.

¹³ See especially p. 118, where the references to DeSaussure are given.

¹⁴ For a statement of the rule see Meillet, Introduction 212.

¹⁵ I use ә̯ to represent any laryngeal (i.e. ә with consonantal value), following Kurylowicz.

¹⁶ See Edgerton, The Indo-European Semivowels, Lang. 19.86 ff., whose symbols I use: t and k are any consonants, y is any sonant, a is any vowel, and so on.

¹⁷ Of course, if we admit that ә was a consonant (laryngeal).

¹⁸ See Meillet, Introduction 129 (6th edition), and A. Cuny's article.

¹⁹ Cf. J. Kurylowicz, Études indoeuropéennes 1.46 ff., and Sturtevant, IH Lar. 83, who gives this innovation as Indo-European (op.cit. 90). The first suggestion had been made by DeSaussure in 1891.

²⁰ One e-colored and one a-colored. That would give us four primitive laryngeals. See Sturtevant, IH Lar.

²¹ Sturtevant, A Comparative Grammar of the Hittite Language 130, 133; IH Lar. 37.

²² The spelling indicates the absence of a vowel between consonants. On the meaning of this spelling consult Sturtevant, Hitt. Grammar 47 f.

²³ represents an a-colored laryngeal which disappeared in Hittite as well as in Indo-European. On the neuter plural ending see IH Lar. 44.

²⁴ This can perfectly well be granted without diminishing the strength of my contention. My point is only that we cannot distinguish between a case in which there was originally a full vowel and a case where there was no vowel at all.

²⁵ Edgerton, Sievers's Law and IE Weak Grade Vocalism, Lang. 10.235–65 (see especially 263).

²⁶ See op.cit. 55.

²⁷ On the other hand, Sanskrit i (< ә) spread often to monosyllabic roots.

²⁸ In Revue de Phonétique 2.106.

²⁹ These represent IE *-tay (*-tay), *-nós (*-nós), and *ṛ-sk̑eti respectively.

³⁰ Ernout and Meillet, Dict. étym. s.v. tero.

³¹ In IH Lar. 52.

³² It is, perhaps, not irrelevant to remark that both present classes agree in having generally a sonant before the final consonant of the root. Cf. Lat. yungō, rumpō, tundō, -cumbō, linquō, findō, scindō; and see the Sanskrit nasal-class presents in William D. Whitney, Roots, Verb-Forms, and Primary Derivatives of the Sanskrit Language 213.

^32a The dissyllabic forms like *perek̑- and so on are the traditional ones and purely theoretical. I keep them in this list, in order to show that instead of these hypothetical forms what really occurs is either *perk̑- or *prek̑-, etc. But, in fact, I think they must be historically untrue.

³³ Ạ represents an a-colored laryngeal (Sturtevant).

³⁴ With ā or ē depending on the laryngeal present in the root.

³⁵ ә only before consonants.

³⁶ See Meillet, Introduction 161.

³⁷ Including here the sonants.

³⁸ Meillet already avoids the forms with two full vowels, using simply the double forms proposed in this article (op.cit. 162 ff.).