Published online by Cambridge University Press: 22 April 2026
It is now nearly twenty years since my paper ‘The Indo-European semivowels’ was published in Lg. 19.83–124 (1943). Some recent publications, to be cited in §§15 ff. below, have suggested to me that a brief restatement of what I was trying to do there is needed.
1 In Lg. 10.235–65 (1934) I had published a preliminary study on the subject called ‘Sievers's law and IE weak-grade vocalism’, which is now largely superseded, though it contains materials dealing with ‘Sievers's law’, especially in non-Indic languages, which are not repeated in ‘Semivowels’ (as I shall henceforth call the 1943 article).
2 I refrain from using the customary slanting lines, as /y/ etc., because I found it convenient in my paper to use the line / for another purpose.
3 On Puhvel's different, and as it seems to me not very successful, method see §17 below.
4 See now my paper on 'Indo-European “s movable”', Lg. 34.445–53 (1958), which removes a whole large category of seeming 'exceptions' at one stroke, by another IE phonetic law which is independent of semivowel allophones but correlates with them interestingly.
5 University of California Publications in Linguistics, Volume 21 (Berkeley and Los Angeles, 1960).
6 Unless it happened to occur after a word ending in -a, so that saṃdhi would yield -āśiyām, because there would then be a heavy syllable before y, automatically producing iy. So e.g. the word āśva- ‘horse’, if the preceding word ends in -a, results in -
śuva-: Lg. 10.256 §44. This does not occur in aśyām.
7 Cowgill asserts (483–4) that the common Sanskrit root yā ‘go’ means ‘ride’, in a vehicle or on an animal, and that this word and its presumed cognates (chiefly Balto-Slavic) make ‘incontestably guaranteed’ a PIE root *yā with that meaning, ‘not “go” in general’. Any competent Sanskritist will be astonished by this allegation (which, by the way, Cowgill makes no attempt to prove). Whatever may have been the meaning of *yā in PIE, anyone with a minimal knowledge of Sanskrit should know (or should be able to find from any dictionary) that the Sanskrit root yā means simply ‘go’ in the broadest possible sense: on foot, on a vehicle or animal, by flying in the air, by swimming or floating in water or in any way whatever. It is an absolute synonym of i. Both these roots, by the way, often mean ‘attain, get into, arrive at’ (a state or condition), when no literal physical motion at all is involved, as in such expressions as vināśaṃ yāti ‘goes to destruction, is destroyed’, or punarjanmaiti ‘goes to rebirth, is reborn’. Grassmann's Wörterbuch zum Rig-Veda gives an identical list of meanings for i and yā. And all Sanskrit dictionaries, from Boehtlingk (and Roth) to V. S. Apte, confirm this. For instance, both Monier-Williams and Apte include ‘walk’ among their primary (first) definitions of yā. — In passing, Cowgill (484) need not have troubled himself about Epic Skt. iyāt, iyām used as preterites. They are real forms, used at least half a dozen times. But they have nothing to do with yā. They are normal optative forms of root i. The use of optatives with the meaning of preterites is a Middle Indic and Buddhist Hybrid Sanskrit idiom; many such Prakritisms, of course, appear sporadically in the Mahābhārata. See my BHS grammar §§32.85 ff.
8 He adds, in a footnote: 'If we use laryngeals freely enough, e.g. by writing *sweliHo- instead of *sweliyo-' [for ON. svilar]. As I said in §20, I think laryngeals are completely irrelevant to the semivowel phonemes. And as I said in §4, it is vain to even hope to find every isolated word in Germanic, such as svilar, preserving 'Sievers's law'. Even in the Rigveda (cf. §8 above) the allophonic alternations are not by any means perfectly preserved, though more nearly so than in any other IE language. In Germanic they are preserved only in a few morphological categories.
9 I believe that Skt. iy
ya is a late analogical formation, not an inheritance from PIE. (Cowgill apparently shares this view, for he refers, 488, to ‘the false assumption that *ey … formed perfect[s] already in the proto-language’.) I also happen to believe that PIE, like Greek, and according to Cowgill (487–8) like Germanic, had only e as vowel of the reduplicating syllable in the perfect; and (I would add; Cowgill does not) only i in that of reduplicating presents. Sanskrit has evidence which seems to support this. The -a- of babh
va and Vedic sas
va is, I think, a relic, pointing to the original condition; it is confirmed by the regular use of the reduplicating vowel -a- (IE -e-) with ṛ-roots (Whitney 782) in the perfect, but -i- in the present. I think that -u- originally did not occur at all in the reduplicating syllable of either present or perfect verb forms in IE. Verbs with radical u-vowels, just like other verbs, had i in the reduplicating syllable of presents, e of perfects. However, I am aware that this view of mine is rather heretical; Cowgill's is doubtless more orthodox on this point. Both views seem to me somewhat speculative.
10 Cf. Bloomfield, op.cit. 178: ‘only free forms can be isolated in actual speech’ — hence not morphs like y and i. A word, everyone will admit, may become lost from a language if it was ambiguous, so that it often led to misunderstandings (like English let ‘obstacle’, extant only in the game of tennis and in such a phrase as let or hindrance; and the cognate verb let, now probably completely obsolete). Shortness, however, can not be proved to have anything to do with this. And the ‘allomorphs’ y and i can hardly have resulted in ambiguity; nor does Cowgill seem to claim they could.
11 There was a time, some decades ago, when some linguists spoke of a phoneme as being ‘realized’ in one of its allophones. So far as I know, such expressions are not now used, at least by American linguists. Cf. my §26 above, where Cowgill is similarly dogmatic and cavalier in assuming this same change in Sanskrit (an unproved and unmotivated 'automatic y' stuck in 'between i and a following vowel').