¹ See e.g. Brugmann, Vergl. Gr.² 1.256 f.—Leonard Bloomfield and Edgar H. Sturtevant have added to my indebtedness to them by reading the first draft of this paper and making suggestions which have led to notable improvements in its form and substance. It is, of course, not implied that they indorse all, or necessarily any, of its contents.—I also acknowledge with deep gratitude a number of valuable suggestions received from Paul Tedesco.

(The author has a plentiful supply of reprints of this article, and will gladly distribute them gratis on request; failing such request, he will assume that persons who receive Language would not care for a separate reprint.)

² PBB 3.52 ff.

³ Curtius Stud. 9.287 ff. On the parallelism with i (y), u (w), see e.g. Wackernagel, AIGr. 1.73; Hirt, Ablaut 8; Brugmann, KVGr. 1.121.

⁴ Although it was elaborately attacked in 1891 by Bechtel, Hauptprobleme 114–153 (see especially 151–3). Cf. also Johannes Schmidt, Pluralb. 218 (1889): ‘die sonantentheorie, welche n und r zu nahe an i und u rücken will.‘ These were probably the last two distinguished Indo-Europeanists who tried to stem the tide.

⁵ This term, as now commonly used, seems appropriate, since each was a ‘minimum unit of distinctive sound-feature’ (Leonard Bloomfield, Language 79), and the choice between the several forms (or ‘allophones‘) manifested by each was determined automatically by the phonetic surroundings. Phonemically, of course, one symbol (say /y w r l m n/) should be used in all cases. As a matter of practical convenience I have thought best to vary the spelling according to the actual phonetic values, as a rule.

⁶ I shall henceforth use the term ‘semivowel’. It is less ambiguous than ‘sonant’, a term which has been used in linguistics with other meanings. As applied to them, the terms ‘syllabic’ and ‘vocalic’ will be used interchangeably; likewise ‘non-syllabic’ and ‘consonantal’. I shall refer by ‘WP’ to the Vergleichendes Wörterbuch der indogermanischen Sprachen, by Alois Walde, ed. Julius Pokorny. The major and minor ‘St. Petersburg Lexicons’ of Sanskrit, respectively by Boehtlingk and Roth and by Boehtlingk alone, are referred to as ‘BR’ or ‘PW’, and as ‘pw’ respectively. I think other abbreviations in references will be self-explanatory. A raised dot before an r or l (e.g. pṛá) will be used to denote a vowel derived from IE ṛ or ḷ, see footnote 27.

⁷ IE ə may be ignored at this point; see §§60–63.

⁸ It seems fairly clear (Brugmann, VGr.² 2.2.145) that neuter noun stems in IE -men-, -wen- had n. acc. -mṇ, -wṇ (Skt. -ma, -va). Here the preceding sound was a semivowel, so that strictly they should not be quoted here, though we shall see that the expected result was the same (§73). Other stems in IE -n- seem to have been heteroclitic, or at least cannot be clearly shown to have had n. acc. forms in -ṇ.

⁹ See Lang. 10.244–5. In urú- and most other words beginning in ur-, an initial v- has been lost by Wackernagel's law (above); urú- is for pre-Vedic *vurú- for IE (and Indo-Iranian) *wṛrú-; the initial was w, not (ṛ)r (§36).

¹⁰ These are the cases designated by the term Sievers's Law, as extended and corrected by various other scholars, last of all by the present writer (Lang. 10.235 ff.). Note here that IE ṛr, ḷl become Indic ir or ur, rarely il or ul; and, of course, IE ṃm, ṇn become Indic am, an.—A brief history of what was previously done by others introduces my former article and will not be repeated here, except skeleton-fashion (with some additions), as follows. Sie vers (1878) said ‘unaccented i and u (sc. after a consonant and) before a vowel were consonantal after a short syllable, vocalic after a long,’ on the basis of Vedic and Germanic evidence. Others showed that Iranian, Greek, Latin, and Balto-Slavic largely support the law, in certain categories of forms. That iy, uw must be assumed instead of Sievers's ‘vocalic i, u’ was clearly stated by Osthoff, MU 4.399 (1881), with some evidence from Indic and some from Greek (the latter partly, though not wholly, convincing today); it was more impressively demonstrated by Wackernagel, AIGr. 1.203. That nasals and liquids behave like the i-y-iy and u-w-uw phonemes was clearly demonstrated on Vedic evidence by Wackernagel, AIGr. 1.11, 29; confirmatory evidence from Greek in Hirt, IGr. 2.198 f. But in some almost forgotten writings by Osthoff, MU 2.14 ff., 4.398, the essential facts are really presented. The first of these (1879) seems to have been written before Sievers's publication (1878), and would have profited by knowledge of this, but on the other hand brings in the nasals and liquids which Sievers ignored.

¹¹ The only common present of this type in RV is tiráti (IE tṛréti). It occurs 65 times, and is extremely interesting for us. If I have counted right, there are 35 cases after a consonant or long vowel, or initial; here pronunciation tira- is to be expected, and always occurs. I have found 30 cases where it occurs after a short vowel, still written tira-. But in 18 of these the meter demands, or at least encourages, pronunciation tra-. Thus 8.59.7d prá tirataṃ na cannot possibly be read as a decent triṣṭubh line except by reading trataṃ, as our law demands. Hardly less compelling are a considerable number of cases where forms of prá tira- are thus written under conditions which, if historically regular, would compel pṛrá (purá?) for prá, and also tra- for tira- (that prá, is often dissyllabic, after a heavy syllable and initially, has long been recognized; Lang. 10.259). That is, after a heavy syllable or when initial, read pṛrá (or purá) tra- for written prá, tira-.

¹² But I was wrong in saying l.c. that ‘the meter requires’ it.

¹³ Lang. 10.260. A census of forms in Lanman, NInfl. 525. An analysis of his figures, with restudy of doubtful cases, yielded the following results for all weak cases of n-stems before vocalic endings in RV. After a light syllable the meter demands -n- about 200 times, -an- some four or five times. After a heavy syllable (whether due to consonant cluster or long vowel before single consonant), the meter demands -an- over 100 times, -n- apparently about 55 times. After a heavy syllable, even the traditional text writes -an- frequently, and that too in cases where Classical Skt. would not do so, namely, after a single consonant preceded by a long vowel. But most of the cases of -n- for expected -an- also follow a single consonant preceded by a long vowel; the language of the authors of the hymns had started towards the Classical Skt. condition. However, the aberrant cases belong largely to heteroclitic or defective stems, and their number might be reduced by speculative but not unreasonable interpretations. Thus e.g., the stem āsán- ‘mouth’ exists by the side of synonymous ; neither has any n.-acc. (for which āsíyam is used); , āsnás, dissyllabic, may be errors of tradition for , both of which exist in the RV.

¹⁴ Lang. 10.242 ff. These forms are specially important because they effectively dispose of the now fashionable theory that instead of iy, uw, ṛr etc. we should assume ъy, ъw, ъr etc., with ‘shwa secundum’ instead of the syllabic semivowel before homorganic non-syllabic semivowel. To my mind, all the cases of the converse of Sievers's Law tend strongly against that theory. But some of its defenders have suggested to me that such forms as antaryāt for theoretical antar-iyāt etc. (§12) could be analogical creations on such patterns as syāt (after short vowel): siyāt (after consonant or long vowel, or initial), the latter being supposed to be for a prehistoric sъyāt. This or any such theory, implausible even for (i)y and (u)w (relic forms, not new creations), cannot conceivably account for cakre, dadhre, since, according to the ‘shwa secundum’ school, there never existed any -ṛr- in the situations of Sievers's Law, but only -ъr- until this developed into Indic -ir- (or -ur-). At no point in the past, therefore, could Sievers's Law have provided any pair of alternants in the r series other than ъr: r, or ir (ur): r. Obviously neither of these could furnish the model for a substitution of -r- for -ṛr-. And if it were a possible assumption that cakre and dadhre were analogical, the older forms which they would have replaced could only be forms containing -ṛr; not, for instance, either -ir- or -ъr-. See Appendix E for further remarks on this topic.—The 3 pl. perf. mid. ending is -re after a light syllable, -ire after a heavy (IE -ray, -ṛray). It corresponded to a 3 sg. -e (IE -ay). Both these forms were at least Indo-Iranian (Reichelt, Avest. Elementarbuch 144), and both were very common in the RV (Macdonell, VGr. 359 f.). It seems to me probable that they, at least, were IE, whether the rest of the paradigm (very much rarer in RV) was or not. Reichelt cites, to be sure, only cāxrare as a 3 pl. in the Avesta; this is an analogical creation like Indic cakrire.

¹⁵ Classical Skt. has also, but much more rarely, the analogical sunu-vas and the like. The 1 plur. also has (allowably and commonly) sunmas, which must be analogical to the dual (so Wackernagel, AIGr. 1.59). The RV happens to have no 1 dual forms from verbs of this type, but does have the 1 plur. forms kurmás, kṛṇmahe, which clearly show that the duals existed in their regular forms as in Sanskrit. See Lang. 10.238. The Hindu grammarians also authorize ‘syncope’ of root-final u before 1 dual -vas in 3d class roots; thus juhvas for juhuvás (Whitney 647c). No such forms are known to occur. The RV has no 1 duals from any verb of this type.

¹⁶ The one alleged case after a long vowel which I was unable to explain l.c. is now cleared up by Barret's edition of AVPpp. 20.28.10 (strangely cited in pw. as 20.5.5), where the meter proves that the forms there spelled (in the virtually unique and very corrupt ms.) cārvāk and cārvadanaḥ were pronounced cāruvāk and cāruvadanah (they are cpds. of cāru- and vāc-, vadana-).

¹⁷ It is worth noting that the only optative forms of i ‘go’ in the RV are three in which a pre verb ending in a vowel precedes (eyāt, preyāt, atīyāma), and one simple form which seems to have been monosyllabic: in 7.86.4, at the end of a triṣṭubh line, turá (p.p. turáḥ) iyām ‘quick(ly) may I go’ is usually supposed to be read turéyām with ‘secondary sandln’, but turó yām would do just as well.

¹⁸ ghṛtásya víbhrāṣṭim án(u) vaṣṭi śocíṣā; precisely like above, except that here the written tradition is wrong. Even Oldenberg, a stubborn opponent of Sievers's Law, speaks of this as ‘denkbar’ (RV Noten, ad loc.). On this method of dealing with the traditional RV writing see §§20–22 below.

¹⁹ Required are situations where an affix beginning -y- or -v- or -r- or -m- or -n-, plus a vowel, was affixed to a ‘stem’ ending respectively in i-, u-, ṛ-, a- (for ṃ, ṇ), preceded by a consonant which was preceded by a short vowel.

²⁰ Except Iranian in compound words, and possibly Homeric Greek, which has been held to show traces of it. Homer has e.g. a form written ϵἰν, for Attic ἐν, before vowels; it has been taken to represent *eny-, which would be perfectly consistent with Rigvedic sandhi. Most recent Hellenists, however, assume metrical lengthening; see e.g. W. Schultze, Qu.E. 216 ff.

²¹ The other languages do, indeed, sometimes show the one-time existence of such doublet forms; a welcome confirmation of our evidence. Some examples from Iranian are cited at second-hand in Lang. 10.236. Greek has e.g. δύω, Lat. duō, from IE duwō(ω); but δώ(δϵκα), δίς, Lat. bis, from IE dwō(w), dwis. But, so far as is known, the other languages show no trace of the original distribution. Only the RV shows that originally duw- occurred initially and after heavy syllables, dw- after light.

²² Why, it is not easy to say. Sometimes, one may guess, the disfavored forms may have been known to the poets, but associated with types of dialect which they despised or disliked. Sometimes, possibly, these forms would merely have sounded ‘queer’ to them: perhaps the forms were foreign to the poets' vernacular, but lacked the aura of respected archaisms, it may be for no better reason than that they happened not to occur in particular collections of poetry which were favorite models of composition.

²³ The magic-mongers of the Atharva Veda, on the other hand, mishandled the delicacies of the language as they did the meters, being comparable perhaps to a Tennessee mountain revivalist in his efforts to imitate Biblical language. I am sure that no one would ever have discovered Sievers's Law from a study of the AV, though its authors were obviously trying to compose in very nearly the same dialect as that of the RV. They belonged socially and intellectually to a very different class.

²⁴ For a few examples see Wackernagel AIGr. 1.XII. The Hindus themselves, in very ancient times, recognized that meter requires departures from Vedic tradition, e.g. that to make ‘correct’ meter, traditional y, v must often be pronounced syllabically: ibid. 202; RV Prātiśākhya 8.22 (527), 17.14 (974). At the time of these statements there was quite certainly no written tradition of the RV; and equally certainly the texts were then recited, at least in the ritual and presumably always, as they are now written, that is, in a manner inconsistent with the meter. The reconciliation with metrical requirements was apparently, to Hindus of classical times, nothing but theory. But the theory was very old; metrics is one of the traditional Vedāṅgas or branches of Vedic learning.

²⁵ One example: several scores of times the RV records as pāvaká- an adjective stem which can be proved to have been pavāká-, or at least to have had a short first syllable and a long second.

²⁶ Thus I write and siyoná-, §17, and waste no words over the fact that the texts write only and syoná-. All Vedists would agree, except that many write (intending two syllables) and sioná- (three syllables). I follow Wackernagel (see footnote 10) and some others. This whole paper, and especially §§9–13, should indicate why, in addition to Wackernagel's reasons, among which may be noted the regular and extremely common vṛddhi formations like sauvará- from (written) svará-, which imply by-forms like suvará-(not *suará-). RV Prātiśākhya 17.14 (974) definitely supports this view; it prescribes that one ‘separate consonant clusters containing semivowels by vowels homorganic’ (to the semivowels, i.e. by inserting them, in metrical analysis). Cf. footnote 24.

²⁷ See Lang. 10.262 f. Let me cite one sample of a word in which our law certainly works in part, but which is very largely recalcitrant, so far as we can yet see. It is recognized (l.c. 259) that the word always written prá was often pronounced as a dissyllable in the Rigveda. Oldenberg writes in such cases p·rá; the pronunciation doubtless lay between pṛrá and purá, more likely close to the latter. I have made a count (somewhat hurriedly) of all the cases cited by Grassmann of this word as a preverb in association with either finite or non-finite forms of verbal roots. They number about 1260. More than half (about 665) of these occur at the beginning of a line, where p·rá should be pronounced. And indeed nearly all the recognizable cases of p·rá do occur there; the others occur internally after a heavy syllable. It is hard to state their number precisely, because of alternative possibilities of reading the texts; but there are nearly a hundred cases (chiefly initial) where p·rá seems certain and necessary, and at least 75 more where, without real emendation, it may be read (generally by assuming ‘abhinihita sandln’, or ‘elision’ of a- after -e, -o). Internally after a light syllable, prá occurs, always (I think) with the proper monosyllabic pronunciation, about 400 times. On the other hand, without assuming alterations in the traditional text which to me were not immediately obvious, or defective meter (which may sometimes be assumed), we should have to admit close to seven hundred cases of monosyllabic prá out of place (over 500 initially, less than 200 medially after a heavy syllable). A more careful study than I gave to the problem, in my hasty count, would undoubtedly reduce, but could hardly be expected to eliminate, these cases. It would appear that the Vedic poets showed imperfect control of the priestly dialect in using this word, while yet showing at least some recollection of the positions proper to each of its forms.

²⁸ Curiously, the written tradition often bungles this in a strange way; it writes e.g. tanvàm, though the ‘svarita’ accent mark on à which we represent by the grave means that the preceding u (not written!) was accented (not the a of the first syllable, which is marked as accentless). This is probably mere adaptation to what actually happened in later Sanskrit (Lang. 10.239, §9).

²⁹ See on the whole subject Lang. 10.237–9, §§5, 7, 9.

³⁰ Arnold, Vedic Metre 83. It is customary (with Arnold) to assume that the accent is wrongly marked; that is, to emend the text to asuryá-, vasavyá-, wherever the words occur.

³¹ Thus, compounds of ví with a noun stem beginning in a vowel (such as víyaṃsa-) were often accented on the í of the adverb: there are 8 such stems, occurring 69 times, according to Grassmann. There are 8 stems with accent elsewhere, such as viyáñjana-, with only 18 occurrences. But such forms in víy-, as also in unaccented viy-, are used exclusively after heavy syllables, or initially; only once, after a light syllable, is there vy-, in 9.65.7 p·rá sómāya vyaśvavát. Similarly in the very common case-forms of the stem etc., with unaccented i, occur as dissyllables 82 times after heavy syllable or initial. They hardly occur after a light syllable; only 8 times in all, and in three of these monosyllabic pronunciation is certain, e.g. 9.15.1 eṣá (written ) yātiy áṇviyā. But dhíyam, dhíyas, with accented i, also occur 110 times after a heavy syllable or initially, only twice after a light syllable. But the compound always accents and pronounces í (sudhíyas, written sudhyàs). Other monosyllabic stems in radical ī, ū present a similar picture. Whether the i, u is accented or not, the syllable containing it is regularly used after heavy syllables or initially, except in compounds where this is impossible, in which case the í, ú is accented and pronounced.—The like is true of some forms which seem to go back to accented syllabic liquids and nasals: śíras- (as if from ) occurs 24 times, always after a heavy syllable or initially. Even if the accent were not on the first syllable, no *śras could occur in any of these positions. Oblique cases of pur-, IE pḷl-, with accent on the ending ( etc.), occur 12 times initially and after heavy syllables; there is one doubtful case after a light. With accent on the first syllable, púram etc., also almost always initially or after a heavy syllable (according to an incomplete count). I suspect that the paroxytone aorist-present stem gáma-, like the oxytone gamá-, goes back to g^wṃme- rather than to a full-grade form of the root, which would not be usual in an aorist (Brugmann, VGr. ²2.3.114). It also occurs virtually only initially or after heavy syllables. See §8 above, and Lang. 10.261.

³² Vedic Variants 2.414; Wackernagel 1.323, 326. Sanskrit retains -āv; for -av it usually has -a.

³³ Brugmann ²1.203 ff.

³⁴ See e.g. Wackernagel-Debrunner 3.45, 152.

³⁵ And, since everything points to identity of treatment in ‘external’ and ‘internal sandhi in IE, the same would have to be assumed internally. The Rigveda has, to be sure’ quite a group (listed in Grassmann 1712) of noun stems in (written) -yya-, pronounced either -yíya- (with accented í) or -yiya- (after a long vowel, which always precedes when the i is not accented). Some are gerundives, some secondary formations from noun bases in -ya-. All seem to me evidently late and analogical. Certainly analogical is -yi in personal endings, like aor. 3 sg. ádhāyi. More puzzling is rayí-, which supplies the nom.-acc. sg. forms rayíḥ, rayím, and instr. pl. rayíbhiḥ (all those with endings beginning in consonants), to the steam rāy-, used before endings beginning in vowels (Wackernagel-Debrunner, 3.214 ff.). I believe that (though it was excluded from the RV, barring a single form in the late 10th book) is IE (= Lat. rem; to nom. probably , parallel to Brugmann ²2.2.131). There is no evidence whatever for rayí- outside of Indo-Iranian, and even the alleged Iranian evidence (Wackernagel-Debrunner l.c. 217) is far from convincing. I cannot explain the form, but suppose it to be an analogical creation, probably in pre-Indic or proto-Rigvedic. The only other such stem is yayí-, apparently connected with yā ‘go’, and certainly a late analogical creation. The two words, yayí- and rayí-, may even have been formed on the same (to us unknown) model.—There are finally some very obscure and difficult forms in Vedic -eya-, -eyā-, from roots in Indic -ā from IE -ā, -ē, -ō (heavy bases): gerundives like déya-, optatives like (párā)deyām. Sturtevant, CP 36.356 ff., has brought them into relation with Greek adjectives in (equally troublesome since y between vowels is normally lost in Greek), and explained both as from IE -əhyo- (-əhyē- etc.), the h representing an IH laryngeal. Such a theory would eliminate any suggestion that they come from forms containing double IE -yy-, which, without venturing to express an opinion of Sturtevant's reconstruction, I think cannot be maintained for reasons which this whole paper will, I trust, make clear.

³⁶ Cf. Wackernagel 1.73 ff., and Osthoff, MU 4.285 ff., which state this principle very clearly. Osthoff has some interesting confirmatory evidence, especially from Greek, along with some remarks, to be sure, which could hardly be accepted today. Apparent or alleged exceptions—barring forms with nasal infix, see Appendix A—are very scarce, and not to be admitted for IE. They are either foreign words, or late analogical formations. E.g. Indic caturthá- ‘fourth’ (not in RV) is clearly analogical to catur- before vowels; Lettish cętuĩtaĩs, connected with it by WP 1.512, is equally clearly analogical to the cardinal *ceturi, later četri (personal communication of Professor A. Senn; cf. Endzelin, Lett. Gr. 361, 368); most languages point clearly to k^wétwṛto- according to our rule (though this may be secondary in IE itself; the RV ordinal is , from (k^w)tur-). Another case is RV ‘firebrand, meteor’. Its connections are highly speculative, WP 1.321; it may well be a loanword; if it is really from an IE wḷl-, with w- lost by Wackernagel's law, see footnote 9, it must be analogical to forms in which a vowel followed, as in WP's alleged Greek cognates.—The forms with nasal infix after vocalic semivowel have not been explained previously and require more extended treatment; see Appendix A.

³⁷ Grassmann lists about 250 occurrences in the first 621 hymns (less than 2/3) of the RV, then stops. In the first 40 cases in Grassmann I have found only one after a short vowel. (In that position RV regularly uses dyós.)

³⁸ But forms of stuvá-, or stuv- before any other vowel, from root stu ‘praise’, occur more than half the time after a consonant or a long vowel, even counting as cases of final short vowel those which ended originally in short vowel plus s (which is lost in RV before s plus voiceless stop). This shows that it was not the initial consonant cluster which was responsible for the virtual limitation of bruv- to position after short vowels, but the fact that its second member was a semivowel.

³⁹ Here and elsewhere in this and the following sections, it is to be understood in default of contrary statements, that the pronunciation appears to be in accordance with my writing, i.e. in this case vr-, with v and r both non-syllabic. Not counted are the few cases where v·r- (i.e. presumably vur-, if not vṛr-) seems indicated, such as 2.24.3 ()v·radanta, and 10.69.10 (written ; see §41 and footnote 46).

⁴⁰ §12, and footnote 31. They occur almost always after heavy syllables or initially, and hence are pronounced viy-, niy-; in a few cases, after short vowels, vy- and ny- occur.

⁴¹ Av. vouru-; correctly taken as for *vurú-, IE wṛrú-, by Uhlenbeck, Leumann, and WP 1.285; Wackernagel's reasons (1.24) for preferring IE urú, have no weight.—That the ‘vowel’ after ur- in this word is a syllabic semivowel makes no difference (§50, end).

⁴² The loss of initial v- before u from IE ṛ, as well as IE u, shows that this was a relatively late change, at least later than Indo-Iranian. Iranian has no -ur- for -ṛr-.

⁴³ See §6 for iraj-, which certainly is for ṛreĝ-. Similarly iradh- and iras- are commonly and doubtless rightly derived from IE ṛr-, not yṛr- or ir-. The word írā probably does not contain original r or l; cf. íḍā (íḷā). There remain only íriṇa-, írin-, írya- (sometimes apparently íriya-, not always for clear reasons), and ilībíśa-; all of obscure origin; and none occurring often enough (the commonest six times), or under sufficiently regular conditions, to justify any inferences.

⁴⁴ There is one queer case: 1.177.4 ayáṃ yajñó ayáṃ miyédhaḥ, a triṣṭubh line with an extra syllable; ayámyédhaḥ would be metrically better (-my- for -mmy-? Whitney 232, Wackernagel 1.113), but still not ideal, since the eighth syllable should be long; and I could not approve it under our law, which, surely before the suggested simplification of the double m, would have had to yield ayáṃ miyé°. One does not like to admit , with ‘double crasis’, but this would give the best meter.

⁴⁵ As a matter of fact, there is evidence to show that even some isolated forms containing -am-, -an- from IE -ṃm-, -ṇn- continued to be used according to the original law. The RV has an enclitic stem sama- ‘some, any, each, all’, clearly from IE sṃmo- (Attic Gk. ‘in a certain way’, OS OE OHG sum, ON sumr). It occurs 13 times, invariably after a heavy syllable. Contrast samá- ‘same’, in RV a homonym of sama- but for the accent, from IE somó- (Gk. ὁμός, Gth. sama etc.), which has no such restriction in position (5 times after heavy syllable, 6 initial, 4 after short vowel). From the same base comes the particle sma ‘precisely’, originally ‘in the same way’, showing the form of the zero grade after short vowels (and so placed in RV 86 times, to 18 after heavy syllables).

⁴⁶ Note 10.69.10 utá avnor v·rádhataś cit, in which the augmented avanos (as traditionally written) shows the n, not an (ṇn), proper after short vowel. The stems tano-, sano- (the only stems of this type in RV not having initial semivowel) may all be interpreted as regular (-an- after heavy syllable or initial, -n- after light), whether augmented or not, except one (8.49.10 asanor), provided we read 6.60.1 śanáthad vṛtrám utá snoti (written śnáthad . . . sanoti), and 3.34.10 índ·ra . . . asnod (written asanod).

⁴⁷ Wackernagel-Debrunner 3.514 f.

⁴⁸ Brugmann, VGr. ²2.3.349.

⁴⁹ However, the acc. sg. ending -m after a syllabic semivowel (i or u) must, for the purposes of this study, be treated in RV as a ‘consonant’; it behaves precisely like the -s of the nom. sg., which also, of course, often appears as -r, changed from -s through -z, in sandhi; this r, too, is structurally as well as historically a ‘consonant’ for our purposes at this point. On the acc. sg. -m see the end of Appendix A, §§66–68, where I have tried to show that originally such combinations as IE -im, -um did not exist before consonants or in absolute final, though in the RV they have spread to such positions.

⁵⁰ Inconsistent are a group of words in -una- after a semivowel (cf. Whitney 1177c); aruṇá, karúṇa-, táruṇa-, dharúṇa, vayúna-, váruṇa-, yamúnā-, perhaps others. Note that there are also forms like mithuná-, píśuna-, in which -un- after a light syllable violates §31b. These latter, and, we may guess, those in [y] plus -un-, would be proper after heavy syllables, as in pracetúna-, árjuna-, nicumpuṇá- (§31a). Either they started in such positions and spread by suffix-adaptation, or some other analogical influences have been at work, too complicated to be unravelled now. The same is to be assumed of another little group of forms in -ina- (to some extent these two groups may be related, if the suffix was once -na- as some of them seem to suggest; and in some the -i- may derive from IE ə): aminá-, álina-, íriṇa-, dráviṇa-, vanína-, hariṇá-. Obscure are tílvila- (loanword), salilá-, and śakuni- (in RV only voc. śakune; certainly a loanword, cf. śakúnti-).

⁵¹ Doubtful are the presumably related words turvá-, turváṇi-, turváśa-, and one or two others. If connected with tūrv- and ultimately with IE , they would not contain IE u; their ur would be for ūr, by analogy of forms in tur- before vowels.

⁵² There is, to be sure, also niṇyá-, which appears to be trisyllabic 4 times (always after heavy syllable), dissyllabic 6 times (3 after light, 2 after heavy, once initial). But the ṇ shows that it is dialectic, and its etymology is obscure (it is generally supposed to be connected with ní).

⁵³ RV 9.97.26d has a form written diviyájo, apparently too long by a syllable, for which Bartholomae, Stud. 1.101, suggested divyájo (cf. §12); but Oldenberg (see his Noten ad loc.) observed that dyuyájo would rather be expected. He did not concern himself with the preceding word, but it is in fact ná, providing a light syllable. Perhaps his observation was more profoundly right than he knew; it fits the above formula. Cf. §31a, b, and footnote 37.

⁵⁴ No inferences can be drawn from the group of words in ‘suffix -uri-‘ (Whitney 1191), as in , jáguri-, (sv)aṅgurí-, cúmuri-, jásuri-, sáhuri-. Obscure adaptations have clearly played a part here, and it is doubtful whether or to what extent IE u is really involved. In such forms as turí-, āduri- (and the irregular táturi-, pápuri-, which have been influenced by unreduplicated forms proper after heavy syllables), it is clear that ur goes back to IE ṛr or ḷl; and it may be that all the other -uri- forms started with imitations of these.

⁵⁵ With Saussure, Brugmann, and Wackernagel among others. Cf. Lang. 10.275 f. for special reasons supporting this formula. For Sturtevant's formula see below, footnote 58.

⁵⁶ Such a case is the weak-grade stem yūn-, to yuwen-, which is clearly IE (WP 1.200). No IE form of the strong stem before a consonant is preserved; but no other explanation of the ū in the weak stem is possible.

⁵⁷ See Lang. 10.257 f., which however is put too mildly and tentatively. There is no doubt of the parallelism IE ṛ: ṛr: r, like i: iy: y, and I should have been less hesitant about the conclusiveness of cakre, dadhre (§10 and footnote 14).

⁵⁸ Sturtevant, IH Laryngeals §69, prefers to assume a non-syllabic laryngeal in this position, which would be inconsistent with the behavior of the semivowels, as my §27 shows. But, as I suggest in this paragraph, this would be no argument against Sturtevant's view, even if one accepted IE shwa as a syllabic laryngeal (a view which, by the way, Sturtevant does not accept).

⁵⁹ It would probably necessitate changes in current fashions of designating some of the IH laryngeals; for I am informed that a syllabic glottal stop would be phonetically a questionable matter. But I believe no one claims that these fashions are more than arbitrary and speculative symbols, chosen for practical convenience and little else.

⁶⁰ See for example op.cit. §78, with §74. Cf. Appendix E, and footnote 70.

⁶¹ I regard the by-form as dialectic. Cf. Pkt. arihai (from arhati), usiṇa- (uṣṇa-), etc.; this is a very common case in Pkt. (Pischel, Pkt. Gr. §§131, 133 ff.). I find it hard to take seriously Brugmann's *pḷtəwo- as base for this form (²2.1.213), whatever evidence other languages may seem to offer for it. Why do we have , if is a regular form? I can conceive no circumstances which would justify both as IE. It is obvious that is regular. The two forms are synonyms, except that is restricted to the meaning ‘earth’, as a noun (which also has). It is quite natural that a dialectic form should be common in such a specialized meaning, and should not be used as fem. to the adjective. No finicky reconstructed difference can account for this.

⁶² But, of course, analogical influence of such forms as : āśú- is possible (§56). Probably not to be grouped here are (cf. sukratūyáyā also); the ū suggests different belongings, and there are parallel denominative verbs in -ūya- in each case, suggesting that the usual theory is right in taking the adverbs from stems in .

⁶³ The latter is found in the old śún-(as etc.) to , which occurs 7 times, always after heavy syllables. The strong stem is almost equally regular: 3 times after a light syllable (§8a), though apparently once initially; 4 times, always after heavy syllable or initially (including 2.39.4 in spite of Grassmann; §8b). We find theoretical -vn- (for IE -wn-) also, quite regularly (§26), after a vowel in maghón- (IE -e/own-) to maghávan-. On to yúvan- see footnote 56. If Classical Skt. adhunā is really connected with ádhvan- (Wackernagel-Debrunner 3.268 f.), it cannot be historic, since adhvanā would be necessary by §31b. It may however be dialectic, with the common Middle Indic change of (Vedic and) Sanskrit va to u (Pischel, Pkt. Gr. §152).

⁶⁴ In Upanisad and Mahābhārata language, tman appears a number of times for ātman, chiefly to be sure after a final -e or -o, where it is interpreted (e.g. by Whitney 135d) as an extension of ‘abhinihita’ sandhi of short a-. But other words in initial ā- are rarely treated thus, whereas this treatment of (ā)tman- is really less rare than would be suggested by the statements of Whitney and BR (s.v. ātman, end), as I know from personal experience (participation in the Critical Edition of the Mbh.).

⁶⁵ Commenting on this, Ludwig (Rigveda 5.352 f.) definitely stated the identity of the two forms, on purely philological grounds, without any phonetic or historic basis. Doubtless for this reason his brilliant perception has been unjustly ignored, but it is absolutely sound. The meanings of the two words are identical; both are very commonly used as mere adverbs of emphasis, as Grassmann's own statements under tmán- (meanings 4–7) suggest; Ludwig saw that the same meanings hold for tánā, táne (‘selbst, sogar‘).

⁶⁶ The strong stems are badly confused. Of course their nasalization is a late and purely Indic intrusion (Brugmann ²2.1.564), also the -i- often found before -v- (id. KVGr. 544 f.). Regular by §8 appear to be ; by §5 cf. Wackernagel 1.46, which does not quite content me); with intrusive i, granting which the form fits §26, (so always, 5 times, in RV; misquoted by both Grassmann and Macdonell!). Irregular are all other forms from roots in ṛ (violating §31b): , and all in nasals (apparently with strong grade of root by spreading, as also in indicative forms which should be weak): ; also (once). None of these forms are common. The ṛ-roots show the same form before consonantal endings in the indicative; one might suggest spread from them. But (since no 1 dual perfects occur) this means, in the active, only the 1 plural, cakṛmá etc., where the form is equally irregular before m; so that one needs to go to the middle (e.g. cakṛṣé) to find a historic source. Or—most likely—the analogical influence came from outside of the perfect; say from the p.p.p. .

⁶⁷ In the sequel, please understand this insertion after this phrase.

⁶⁸ In 1.34.1 read yant·ráṃ himyéva, not himiyéva with Grassmann.

⁶⁹ I cannot doubt that the obsession of ‘shwa secundum’ prevented such a scholar as Hirt from seeing the phonemic pattern of the IE semivowels. Hirt's conception of ‘shwa secundum’ as somehow related to the position of word accent also led him to connect Sievers's Law with the position of the accent (IGr. 2.198), which is utterly inconsistent with the obvious facts in the Rigveda (Lang. 10.246).

⁷⁰ Cuny, Revue de phonétique 2.101 ff., has some interesting observations bearing on this topic.—My point relates, inter alia, to Sturtevant's assertion that ‘in other cases an IH laryngeal has to be assumed within the group that later yielded the conditions requisite for Sievers's Law.’ I reply that Sturtevant would first have to prove that the presence of a laryngeal (supposing that it really ‘has to be assumed’) would make any difference in the application of Sievers's Law. His own remarks, op.cit. §74, tend to support the negative.

⁷¹ Only two common words in suffixal -ya- after heavy syllables are recognized by Grassmann and Arnold (85) as pronounced irregularly in a considerable minority of instances: dāíviya- and . The former is, according to Grassmann, regular 64 times, irregular (dāívya-) 14 times (Arnold wrongly adds another, 10.130.7). Possibly devá- or divyá- may have been the true reading in some of these. Grassmann recognizes 230 times, 119 (Arnold adds 3 more); besides fem., and compounds of one or the other, 25 times, sūry- 15 according to Grassmann. Here it is worth noting that RV has a by-form and synonym , fairly common even in the traditional text, but elsewhere recorded only in AV (according to BR; later lexicons cite it, because they knew their Veda). It may well be that sūry- of the text, when the stem is dissyllabic, is a mere error of tradition, which may easily have substituted the only form used later for RV etc. But also, Grassmann's count of dissyllabic is certainly too large; cf. §12.—A fair idea of how the suffix -ya- behaves may probably be got from a study of all such words with initial a- in Grassmann. According to my count, after heavy syllables, iy is both written and pronounced in 5 words, 12 occurrences; iy is pronounced though y is written in 42 words, 238 occurrences (in about three such words stray occurrences of pronounced y occur); y is pronounced in the single occurrence of abhyāyaṃsénya- (a highly anomalous form in other ways). After light syllables, y followed by svarita accent (§23) is written, pronounced, in 6 words, 19 occurrences (on asuryà- see footnote 30); y is written and pronounced in 26 words, 374 occurrences (179 of the stem anyá-; about nine or ten of these words show stray cases of apparent iy); iy is apparently pronounced in a majority of occurrences only of ápya- ‘watery’ (according to Grassmann 4 times with y, 11 with iy, but 3 of the latter must or may be counted as regular; should be read in the others, from the strong rather than weak form of the stem āp-, ap- ‘water‘? āpya- exists in the Upanisads and later) and in the single occurrence of avyathiyá-. Grassmann and Arnold, wrongly in my opinion, recognize more aberrant cases in some of these categories.

⁷² I treat as ‘unimportant’ isolated forms, and small minorities of irregular cases when a considerable majority is regular.