You are on page 1of 23

THE SANSKRIT NOM. SING.

VIT by
F. B. J. KUIPER

Leiden
In memory of Jakob Pieter Smit t 28.8.1965

1. The origin of the final -.t in the nominatives of noun stems in ~r j, and h (representing proto-Indo-European k, g, and ~h) and of the corresponding .d before the case endings -bhih. and -bhyah.1 is not merely one of the minor problems of the historical grammar of Indo-Aryan. 2 Since it has often been assumed that the retroflex stop indirectly reflects a palatal stop [t'] or an affricate [tg] in proto-Indo-Iranian, 3 it is also of some importance for the reconstruction of the prehistoric development of Iranian, and especially for the interpretation of Old Persian 0, 4 for the
1 For the Vedic forms in -.t see Lanman, Noun-Inflection in the Veda ( = JAOS, 10) (1878), pp. 463, 490, 499; Macdonell, Vedic Grammar (1910), pp. 183, 237, 239. 2 See Wackernagel, Altindische Gramrnatik, I (1896), pp. 173ff., 176 and Nachtriige, pp. 92-94; III (1930), pp. 232f., 253f. 3 Cf., e.g., F. MOiler, Wiener Sitzungsberichte, 89 (1878), p. 5: "ConsonantenDiphthonge" (ts'etc., different from affricates, see WZKM, 9, p. 138 and Bartholomae's comment, ZDMG, 50 [1896], p. 705 n. 1), "im Indischen richtiger ~ " ; Meillet, MSL, 8 (1894), p. 284, J. Bloch, L'Indo-Aryen (1934), p. 53 (tA~ .t.y~ .t), Kurytowicz, Actes du 4e Congrks des linguistes (Copenhagen, 1938), p. 63 (ks ~ .t~~ t.), Comptes rendus de la Socidtd des Sciences et des Lettres de Wroctaw, III/1 (1948), p. 5, L'Apophonie (1956), p. 373f. (ks>t's.~t..s). 4 But OP. Oadayati can only be explained on the assumption of proto-Iranian *g or or *s, see Hfibschmann, KZ, 23 (1877), p. 397, Nyberg, Studia Indo-Iranica, Ehrengabe fiir W. Geiger (1931), pp. 213-216, Leumann, IF, 58 (1941), p. 3 ( = Kleine Schriften [1959], p. 3291".). It is hard, in the light of this evidence, to accept the view that Kafiri [ts] corresponding to Av. s and Skt. L is decisive proof for the existence of affricates in proto-Iranian. See for the evidence of Kafiri Morgenstierne, NTS, 13 (1945), p. 225f. and the theories based on it by Pisani, Archivio Glottologico Italiano, 33 (1941), p. 71, Rysiewicz, Studia Jfzykoznawcze (1956), p. 290, V. V. Ivanov, Voprosy Jazykoznanija, VII/4 (1958), p. 14. As for Oss.faerxt "axe", which Hfibschmann was the first to draw attention to, this may be a loan-word; see Abaev, Osetinskijjazyk ifol'klor (1949), p. 139f. As for *~h the possibility that uaganna "race-track" in the Kikkuli-text stands for prehistoric (Indo-)Aryan *v~hana- should be noted (see Benveniste, Hittite et IndoEuropden, p. 9 and Mayrhofer, llJ, VII [1964], p. 210, Die Sprache, 10 [1964J, p. 175), whereas for *z"the rendering by z (as in -~aza = Ved. vdja-) is instructive. The evidence of Kafiri (which may have split off in the common Indo-Iranian period, see Morgenstierne, p. 235) has no clear support in either Iranian or Indo-Aryan.

104

r . B . J . KUIPER

PIE. reconstruction of the phonemes traditionally denoted by k/5 or k 8, etc., 5 and for the relations between centum and satem languages. ~ In view of the far-reaching theories which, partly at least, have been based upon the final -.t of Sanskrit it may be useful to reconsider the problem. Before entering into a discussion of the general problem itself, however, I should like to make some remarks about a point of detail. Many of the explanations proposed start without further discussion from the assumption that a (proto-Indo-Iranian) affricate [t~] had become a retroflex cluster [t..s] in some prehistoric period (which most authors omit to specify sufficiently). Thus Jules Bloch writes that in [t~] "le premier 616ment tendait ~t s'assimiler au second, donc ~ prendre la forme c6r6brale". 7 If, however, the proto-Indo-Iranian representative of PIE. ~: still was ITS], the question necessarily arises whether this has actually become [.t.s] in any position, no matter at what time this must be supposed to have taken place. If so, the prototype of Skt. gatdm "hundred" must have been *t.s.atdm, with an initial .t which even in classical Sanskrit was extremely rare. A second objection that must be raised concerns the cluster [t..s], which is, in some way or other, the basis on which most of the more recent explanations have been founded. This cluster, however, is entirely non-existent in the phonemic system of Sanskrit in historical times, the only comparable clusters that actually occur being ts and .ts. A third objection concerns the obvious fact that the assumption of retroflex phonemes for proto-Indo-Iranian is fully unwarranted. In old Iranian such retroflexes are entirely lacking, and in Indo-Aryan the genesis of these phonemes is in all likelihood a comparatively late process which must have taken place in the separate branch owing to foreign influence in the Indian linguistic area. 8 A theory, therefore, which presupposes the existence of such retroflex phonemes at an early date, either in proto-Indo-Iranian or in proto-IndoAryan, would for this very reason seem a priori implausible. It is curious that so little attention has been paid to these quite obvious difficulties by the advocates of the theories in question. 2. To get a better insight into the prehistory of the enigmatic final -.t it may be convenient first to consider more closely the problem of the long
5 T. Burrow, JAOS, 79 (1959), pp. 85-90; cf. pp. 255-262. 6 V.V. Ivanov, Voprosy Jazykoznanija, VII/4 (1958), pp. 12-23. 7 L'lndo-Aryen (1934), p. 53f. He here follows Meillet, MSL, 8 (1894), p. 284. Cf. also F. Mtiller, Wiener Sitzungsber., 89 (1878), p. 5 (see above, note 3). Wackernagel, Altincl. Gramm., I (1896), p. 176, accepted .t.s as the previous stage of .t, Bartholomae, ZDMG, 50 (1896), p. 704, rejected it. 8 See above, "The Genesis of a Linguistic Area" (p. 81 ft.).

THE SANSKRIT NOM. SING. "r

105

sibilant in Sanskrit. Here the long ("geminate") s occurs only in cases where one morpheme ends in -s and the following begins with it. It is a well-known fact that in these cases we find three different representations, viz. : a) In the phonemic system of proto-Indo-European a long s seems always to have been shortened: 9 cf. PIE. *dsi for *dssi "thou art", as reflected by Skt. dsi, Av. ahi, OP. ahiy, Gr. g[, etc. Some last traces of this shortening are found in the noun-inflection of the oldest Indo-Iranian texts: Skt. apdsu, loc. pl. of apds- "active" (RS. VIII. 4.14), drhhasu, id. of drhhas- "narrowness" (AS. VI.35.2), Avestan qzahu, id. (Y. 10.7, etc.). The same shortening is found in word-final position (in accordance with the general rule) in such Rigvedic forms as dvive.h, vivdh. (for *vi-ve.s-s), etc) ~ In view of the explanation of vi.dbhih that will be put forward below, the OP. instr, pl. raucabiJ for *raucazbi~ deserves mention in this connection. It is true that a similar anomaly is found in such an apparently recent compound as OP. zftrakara- "evil-doer" for *zftras-kara-, where the Iranian stem form zarah- seems to function as the first member of the compound. The contrast with Vahyaz-data- should be noted. In the case of raucabiY, however, it is possible that the analogy of the corresponding case forms of the n-stems, e.g., *nama-hu : *nama-bi~, has given rise to a new form rauca-biYby the side of the loc. pl. *raucahu (analysed as *rauca+hu). In that case, a parallel instance in Sanskrit would be the instr, pl. dho-bhi.h which, owing to a different analogy, has been created by the side of the loc. pl. 6hasu.11 However, this explanation, first proposed by Bartholomae, 12 is far from certain. The chief difficulty is that the corresponding Avestan forms point to [-ah-big], which can be due either to a proto-Iranian innovation or to one of Avestan alone. As far as I can see, the most important evidence in this respect is the Gathic form huddbyO Y. 34.13 (also occurring in Later Avestan). Owing to an incorrect interpretation of its stem form Bartholomae was led to explain this as an analogical formation. 13 Since, however, d in the vulgate text is a graphic representation of [ah], and since the stem must have been [huda'ah-] in Zarathustra's dialect, 1~ it can hardly be doubted that the words hya.t cavi~ta huddby6 in Y. 34.13c stand for [yat cai~ta huda'ahbyah]
Cf., e.g., Wackernagel, Altindisehe Grammatik, I, p. 111, III, p. 289f. lo See Bartholomae, KZ, 29 (1888), p. 575. 11 Wackemagel-Debrunner, Altind. Gramm., III, pp. 290, 310. 12 Grundriss der iran. Philologie, I, 1 (1895), p. 215. 18 Altiranisehes W6rterbueh, col. 1824 n. 1 (not mentioned by Reichelt, Awestisches Elementarbuch [1909], p. 180). x4 See IIJ, I (1957), p. 89.

106

F. B. J. KUIPER

in Z a r a t h u s t r a ' s dialect. 15 It is probable, therefore, that the Gathic instr. pl. raoc~b~ Y. 30.1c, 31.7a (cf. 58.6 a n d raoc3byd Y. 19.6) stands for [raucahbi@ As such it can s u p p o r t the theory that final -~ in the Gathic dialect is merely a graphic representation of [-ah]. 16 Since the occasional spelling with -6big (thus p r o b a b l y in gar6bf~ Y. 34.2c) is due to the secondary i n t r u s i o n of LAv. 6 for ~ into the G a t h i c orthography, 17 it is impossible to assume any historical c o n n e c t i o n between Avestan -6bio~ a n d Vedic -obhi.h, is which reflects the characteristic I n d o - A r y a n developm e n t of -az-? 9 I n the light of this Avestan evidence it is quite possible 15 For eavigtd (Geldner civigtd) = *cOigtd, see Darmesteter, Le Zend-Avesta, I (1892), p. 336, Tedesco, ZII, I (1923), p. 48 n. 2, Kurytowicz, Traces de laplace du ton en gathique ( = Biblioth~que de l'l~cole des Hautes t~tudes, fasc. 244) (Paris, 1925), p. 21, L'Apophonic en indo-europden, p. 164, Humbach, MSS, 2 (1952), p. 11. Bartholomae's reading cavi~t5 (Altiran. Wb., col. 442) has been corrected in col. 1889.-The reading huddby6, found in the best manuscripts, has rightly been accepted by most authorities ever since Westergaard, Zendavesta, I(1852-54), p. 67, in spite of Geldner, who reads

hudaby6.
18 See H. F. J. Junker, Das Awestaatphabet ( = Caucasica, II) (1925), p. 46ff., Nyberg, JAs., 1931, II, p. 43f. n. 1. p. 202 n. 1 (fracara apausa form of *fracarah; but in Irans Forntida Religioner [1937], p. 486, he takes a different view), Karl Hoffmann, Iranistik, I, p. 8: "-# (ftir auslautendes -ah, ved. -aO aus -as)". Wikander, Vayu, p. 70, took -~ as representing a visarga (= -ah, -ih). However, even if intervocalic h was voiced in Old Iranian, there would not seem to be any need for a particular term for a voiceless h in final position (Reichelt, Awestisches Elementarbuch, p. 83), which was, in any case, merely an allophone of the phoneme/h/. See below, note 20. For different opinions on a see Andreas, G6tt. Nachr., 1911, p. 2 (d = w), H. W. Bailey, Zoroastrian Problems in ninth-century Books (1943), p. 192f. (a = y), Jos. Markwart, Das erste Kapitel der GSp5 ugtavatf ( = Orientalia, No. 50) (1930), p. 15ft. 17 K. Hoffmann, l.c., who thereby corrects the older theory of GAv. ~ standing for 5 (e.g., Bartholomae, GrlrPhiL, I, 1, p. 154, Reichelt, Aw. Elementarb., pp. 33, 180). The explanation of LAv. 5 for ah in final position and for a in compounds remains a crux (see also Markwart, I.e.). The explanations proposed by Gauthiot, Latin de mot, p. 117 (ah >qh > oh >5) and W. S. Allen, Sandhi (1962), p. 107, fail to convince. See Hoffmann, p. 9, and note LAv. qzahu, u~ahva beside LAv. raucdhva [= raueahva], ravShu, tdmShva, like LAv. aoja~hvat, beside GAv. aojSnghva_t,aojSnghvantdm (Bartholomae, GrlrPhiI., I, 1, p. 174), which point to the conclusionthat a is merely graphical. The traditional explanation (e.g., Reichelt, op.c., p. 83) is unacceptable. ~s Bartholomae, GrlrPhil., I, 1, pp. 173f., 215, 225, ZDMG, 50 (1896), p. 708f. ("vorindisch"), Reiehelt, Aw. Elementarb., p. 180, Kent, OM Persian, pp. 41, 63, Thumb-Hauschild,Handbuch des Sanskrit, I/I, p. 333 ("-obhis vielleicht schon urarisch", like Wackernagel, Altind. Gramm., III, p. 289; similarly Scheftelowitz, IF, 33 [1913-14], p. 153: "Bereits in der arischen Grundsprache scheint inlautendes z vor bh zu h geworden zu sein, welches den vorhergehenden a-Vokal zu o triibte und dann in der Aussprache mit dem nachfolgenden bh zusammenfiel... Im Awesta ist dieses sektmd~irentstandene o durch #, 5 wiedergegeben ..." Against this view Meillet, Les dialeetes indoeuropdens (1908), p. 30, stressed the fact that the o in Skt. -obhiO and that in Av. -6big "sont donc deux ph6nomgnes radicalement ind6pendents Fun de l'autre'. ~9 Wackenaagel-Debrunner, Altind. Grammatik, III (1930), p. 289, rightly reject the older theory of Benfey, Bartholomae and Brugmann, who explained -adbhib and -adbhya.h as the regular phonetic developments of -azbhil.~, etc. In 1913 Scheftelowitz

THE SANSKRITNOM. SING. V[t.

107

that OP. raucabi5 also stands for *raucahbiL ~~ In that case the substitution of-ahbiY for *-azbiY might have been an innovation o f proto-Iranian. However that m a y be, this innovation is mostly explained as an intrusion of the nominative f o r m (or a sandhi variant o f it) into other case forms, which would be parallel to, e.g., v ~ i b y 6 [ = vd#byah] for *v~gby6; cf. the nora. sing. v~xL In view of the mutual influence exercized by the singular and plural fornls of the instrumental (see below, Ved. *i.dbhi.h idd ~ f.ddbhi.h) the stem-form of [raucahbiY] might as well be explained f r o m the instr, sing. *raucahd. b) The n o r m a l representation in Sanskrit, however, is the long s, which has analogically been restored. ~1 The loc. pl. o f the nouns in -asis, f r o m the Rigveda downwards, in -assu (or -a.hsu), 22 e.g., r6jassu, ~rdvassu, and similarly hav(s.su of a stem in -i.s-. 2s The explanation by analogy implies a (more or less conscious) effort on the speakers' part to m a r k off the m o r p h e m e -su as a separate element of the word. In the same way the loc. pl. of rods- " m o n t h " , viz. mdsfi TS. VII.5.2.2 was replaced by m~ssu KS. 33.1 (p. 27, 8), PB. IV.4.1, TB., etc. A parallel p h e n o m e n o n in the verbal inflection is the long s in ~dssi SB. (etc.), ~sse ChU. 24 N o instance o f a monosyllabic root m o r p h e m e with a short vowel is recorded. See below, p. 109. c) The third representation, viz. ts for ss, occurs only in some verbal stems of the roots ghas- "to eat", vas- "to dawn", vas- "to dwell", and vas- "to clothe". It is restricted to the stems of the future and the s-aorist had already combated it (IF, 33, p. 153). In this connection the evidence of Ved. mayobh~- andpurobh~-, which are fully parallel to raks.oh6n-, manoj~-, etc, would seem of particular importance. 20 With loss of the voiced h according to Kent, OM Persian, pp. 41, 63, or merely graphic for [raucahbid] according to Brandenstein-Mayrhofer, Handbuch des Altpersischen, p. 61. For a different explanation of OP. raucabi~, which seems equally possible, see Wackernagel-Debrtmner, Altind. Grammatik, III, p. 289. They assume that the nora. sing. had already become *rauca in OP. and that in raucabig this nominative form had been introduced. In this connection it may be observed that Meillet's assumption (apud Meillet-Benveniste, Grammaire du vieux-perse, p. 261) of an implosive s in the OP. endings -as and _ffsis certainly incorrect. As the genitive mazdah-a (with a new genitive ending added) shows, the ancient Iranian genitive [mazda'ah] of the Gathic dialect must have survived in Old Persian as *mazdah (with a final h) up to a certain time; no matter how these -ah and -ah had come to be pronounced in the period of the Achaemenids in the S.W. dialect. As far as I can see, final -as and -as must have become -ah and -ah in the common proto-Iranian language. ~ See, e.g., Wackernagel(-Debrunner), Altind. Gramm., I (1896), p. 111, III (1930), p. 2891'. 22 Cf. Lanman, Noun-Inflection in the Veda, p. 567, Macdonell, Vedic Grammar, p. 233. 2a CT. Lanman, op. c., p. 573. 24 See Nachtriige ad Altind. Gramm., I, pp. lll, 38 and 178, 39.

108

F. B. J. KUIPER

and some desiderative formations like upa-vivatsu- "wishing to fast", but does not occur before the personal endings -si and -se. The most complete list of forms attested is to be found in Debrunner's Naehtriige, p. 96 (ad Altind. Gr., I, p. 178, 39; add vatsyan JB. I, 1681~ The explanations proposed for this particular development are essentially of two types :55 The first group of explanations is based on the supposition that ts is the result of an analogical innovation, which may date from either the protoIndo-Iranian or the Indo-Aryan period. Bartholomae explained the t in the aorist a-vat-s-am AB. as due to the 3 sing. a-vat (from vas- "to dwell"). It has, however, long since been recognized that such forms as AS. vy-avat (from vi-vas-), Br. a-gat (root gas-) are themselves comparatively recent new formations, due to the fact that in the corresponding 2 sing. the s in -as (e.g., aghas AS. XX. 129.6, for *a-ghas.s-s) could easily be re-interpreted as the personal ending (like a-pa-s : a-pa-t). 26 Besides, this theory (which Bartholomae later put forward with the greatest reserve, ZDMG, 50 [1896], p. 711), can hardly account for such forms as jighatsat KS., vivatsydti MS. A different theory was developed by L. Bloomfield and M. Leumann. 27 They assumed that (apparently in protoIndo-iranian) ts had become ss, e.g., in *sarvdtat-s>*sarvdtas, fut. *pat-syati > *passyati, patsti > *passfi. This is, indeed, the development that is found in Iranian. In Indo-Aryan, however, ts is supposed to have been restored "aus Grtinden etymologischer Durchsichtigkeit", which accounts for Ved. sarvdtat (for *sarvdtat-s), datsva, patsyati and pats(t, contrasting with Av. haurvatas, dasva, etc. Now, it is on the analogy of this restoration of ts in patsyati (supposed to stand for *passyati) that vatsyati for *vas-syati (future of vas-"to dwell") can be explained. However, as Leumann rightly observes, this analogical explanation fails to account for Ved. mdtsya- : Av. masya- "fish", where such an analogy cannot have worked. 2s Leumann's main reason for maintaining the theory, in spite of the numerous analogical new formations which it presupposes, is Ved. dehi from *dazdhi, corresponding to GAy. dazd~. In this isolated case the group d-dh must have become (via regular *dzdh) zdh, before dh developed to h in Vedic. Here, accordingly, *dz seemed to have become z, which is still extant in Avestan. However, this isolated instance ofzdh at a prehistoric stage of Indo-Aryan has since satisfactorily been ex25 See Wackernagel,Altind. Gramm., I, p. 179 with Nachtriige. 26 For exact references see Wackernagel,I.c. 27 See resp. AmJPhil., 32 (1911), p. 55 and IF, 58 (1941), pp. 12-14 (= Kleine Schriften [1959], pp. 336-338). 28 No inference, it seems, can be drawn from Rigvedic tsdru-"a crawling animal".

THE SANSKRIT NOM. SING. Pit,

109

plained as the result of dissimilation (either in PIL. or in PIA. *dadzdhi). 29 Hence the only support for the supposed phonetic development in Indo-Iranian has been eliminated. The second group of theories tries to explain ts as the result of a phonetic development of ss. The possibility that this has taken place in p r o t o - I n d o - E u r o p e a n ( J o h . Schmidt, Pisani) must be ruled out because ss was itself an innovation of lndo-Aryan. 3~ Indeed, not a single instance of ts is found in the Rigveda, the earliest occurrences apparently being avdtsi.h 2 and jlghatsati 2 (with jighatsf~- 2) in the Atharva- Veda. Therefore, the development ofss to ts, far from dating back to PIE., must rather be a (comparatively late) secondary development of the separate Indo-Aryan branch. Nor can it be a regular development, such as can be formulated in a phonetic law. Attempts to formulate such a law have failed because vatsyati (for *vas-syati) cannot possibly be explained from such an ausnahmsloses Lautgesetz. The ten Rigvedic instances of a l o c . pl. in -assu (as against one instance in -asu) show that as a rule the long ss, when it was analogically restored, remained intact. There was, however, one difference between the loc. pl. of disyllabic stems in -as-, such as dmhasu, rcijassu, and the future stem of such a monosyllabic verbal root as vas- "to dwell". While the first category remained sufficiently identifiable as case forms of words in -as-, even if the long ss was not restored (as in dmhasu from drhhas-), the normal development of *vas-sydmi to *vasydmi would have resulted in a lack of morphological clearness as it would have suggested an analysis va-sydmi. The conscious will of the speakers to counteract the normal phonetic tendency of their language, and their effort to over-emphasize the s of the tense morpheme -syaapparently led to a realization of the second s as the affricate [ts], to mark off the beginning of the morpheme from the preceding s. Afterwards the cluster [sts] in [vas-tsy~mi] must have undergone the same dissimilation that is found in the future of vraAc- "to hew down, fell, cut to pieces", viz. *vrask-.sy~mi ~ vrak.sydmi. 31 Something of this kind Wackernagel must have had in mind when he wrote that ts is due to the fact "dass man den zugeh6rigen Formen zu lieb dem s ein zweites s vorschob; dann wurde s-s zu ts. ''82 Bartholomae, it is true, rightly objected that this did
29 K. Hoffmann, MSS, 8 (1956), p. 21. s0 E.g., Wackernagel(-Debrunner), Altind. Gramm., I, p. 179, III, p. 251 (on the loe. pl. in -tsu; cf. p. 289 on -dbhil~, etc. : "nur jungvedisch land ganz selten"); Bartholomae, ZDMG, 50, p. 711. 31 See Wackernagel, Altind. Gramm., I, p. 270. Cf. also the external sandhi of ddha k sa'rantf.h VII.34.2 and ~ for s in ~atdkratuO tscirat VIII. 1.11 (Macdonell, Ved. Grammar, p. 71). ~2 Altind. Gramm., I, p. 178f.

110

r . B . J . KtJn'ER

not happen in Ved. ~assi, massu, sddassu, etc., 33 but this objection is only valid in so far as it proves that ts from ss is not the result of a regular phonetic law. A more exact definition of the process is found in Brugmann's Grundriss, F (1897), p. 735 in the words: "Vermutlich entwickelten sich in den Geminatae -ss-, -~-, die auf zwei Silben verteilt waren, t und k in der Silbenfuge ... I m Loc. PI. ist diese Entwicklung durch Systemzwang gestSrt worden." However, apart from the fact that .s+s > k s (dveks.i) is the result of a fundamentally different process (see below, p. 112), Brugmann, too, was still thinking of a normal phonetic law. On the other hand, he was right in stressing the importance of the syllabic trench. A confirmation of the view that the development of s to the affricate Its] was determined by the monosyllabic character of the preceding root morpheme can be found in rnddbhih. RS. 1, madbhyd.h AS. 2, the instr. and dat.-abl, pl. of regis- "month". z~ I think Leumann is right in explaining these forms as new formations based upon a loc. pl. *rnats~. 35 Just as in the parallel cases discussed below, the form of the loc. pl., which alone can account for those of the instr, and dat.-abl, pl., is not attested. The lower frequency of the loc. forms in the Rigveda (1546 occurrences) as compared with those of the instr, and dat.-abl, pl. together (3847 occurrences in total) 86 sufficiently explains the non-occurrence of *matsu in our text. Thus, of the words of kindred meaning div- "day" and gardd- "year" there are 19, resp. 2 occurrences of dy~bbih, and ~arddbhih but not a single of their loc. pl. O f the disyllabic stems there is only one isolated instance of u.sddbhih. in one of the last parts of the Rigveda (I.6.3), and svdtavadbhyah, in Yajurvedic texts. As for u.sddbhih. (also AS. 3 in bk. XX), it is fully obscure why *u.s6bhi.h has gone out of use. The animate nouns in -ashave the normal ending (dhgirobhi.h, dhgirobhya.h). The Vajasaneyi Sarhhita, which uses the normal form sa-jd.sobhyam, has usdbhyam, on the analogy of the acc. sing. u.sdrn VS. 2 AS. 2. See Altind. Gr., III, p. 283. In the light of these facts it can hardly be questioned that the few

38 ZDMG, 50, p. 710. 8~ Macdonell, op.c., p. 221 n. 9, Wackernagel-Debrunner, op.c., III, pp. 250, 289. 85 IF, 58 (1941), p. 13 (= Kleine Schriften, p. 337). Taken by itself, madbhib RS. II. 24.5 could have been provoked by karddbhih.,which precedes it immediately (Scheftelowitz, IF, 33 (1913-14), p. 154), but it is somewhat hard to believe that madbhyd.hin the AS. could have been created on the analogy of the Rigvedic hapax and have nothing to do with the parallel forms in -d.bhih.(see below). 86 The exact figures are (according to Lanman's statistical account in Noun-Inflection in the Veda, Table I, opposite p. 582) 3360 for the instr, pl. and 487 for the dat.-abl. pl.

THE SANSKRIT NOM. SING. v[.t

111

isolated instances of-dbhi.h, -dbhya.h are due to a secondary development of Indo-Aryan. In 1930 Wackernagel and Debrunner (Altindische Grammatik, III, p. 289) rightly abandoned the older theory of zbh >dbh (vol. I [1896], p. 179) and explained dbh as analogical. The starting point probably was, in all cases, the loc. pl. 3. The idea that ts is the result of a conscious effort of some speakers in some cases to over-emphasize the second part of the long s after the syllabic trench, is no novel one. Ed. Schw3,zer has studied a similar phenomenon in modern Greek. ~7 On p. 234 of his article he makes the following remark: "Das neugriech. Nebeneinander von s, ss, und ts lfisst auch die Annahme unverf/inglich erscheinen, einerseits sei idg. ss zu s reduziert, anderseits (restituiertes) ss iibertreibend zu ts geworden." Apart from the fact that Schwyzer here seems to assume (or at least, not to exclude) a process in the Indo-European proto-language, his explanation is an improvement upon Brugmann's, who still operated with a regular phonetic law. The influence of such an emphasis on what would be the regular phonetic development is too well-known to require an ample discussion. Thus Berger has pointed out that the accent of dka- "one" and vi~va- "all" on the first syllable (instead of the older accentuation *ekd- and *vi~vdwhich is still normal in composition, e.g. ekd-manas-, vi~vd-rCtpa-) reflects the emphatic pronunciation which almost automatically results from the meanings "one" and "all". ~8 In this connection long consonants are particularly interesting in cases where shortening might cause a lack of morphological clearness. In more than one language they apparently constituted a specific problem. In Old English, for instance, this is manifest in the length of the preceding vowel. Long vowels in stressed syllables were indeed shortened before long consonants but this process did not happen if the long consonant belonged to two different morphemes. Thus OE. *mdbbum "treasure" (Goth. maibms) became mabbum, but the long vowels in the past forms rg~d-de "advised", m~t-te "met" remained long. As Kurath puts it, ~9 there is in this case a structural difference " b e t w e e n / % c - v / w i t h a contrastive long a~ "Dissimilatorische Geminataufl6sung als Folge yon l]bersteigerung', KZ, 61 (1934), pp. 222-252. 38 MSS, 3 (1953), p. lff., 32 (1958), p. 5ft., especiallyp. 20ff. 39 Language, 32 (1956), p. 436 n. 5. Cf. also the English pronunciation ofn in income, concord (in contrast with, e.g., conquer) which, as Kruisinga rightly observes, "is evidently due to the speaker's consciousness of the elements of the compound word" (A Handbook of Present-day English, 12, p. 143, par. 401).

112

F.B.J. KUIPER

consonant belonging to the same morpheme, and/~vc-cv/with a sequence of identical short consonants belonging to different morphemes ... where OE. retained long vowels". In terms of the classical historical grammar it might be stated that the long vowel was here analogically restored in accordance with the normal type of past forms like m#.lde. Either definition ultimately refers, not to a phonic difference between bb and dd, tt but to a specific attitude of the speakers. This may (but need not necessarily) have resulted in a particular articulation of the long consonants so as to bring out more clearly the trench between the root morpheme and that of the past tense. The introduction of external sandhi in Vedic mdnat.~-su for mdnassu (the manuscripts and the editions of the Rigveda vary in this respect) may be due to the same tendency. In Old Tamil a rule of external sandhi demanded that a sentence like mu[ t~tu "a thorn is (a) bad (thing)" should become muttrtu but an optional rule allowed this to be pronounced as muh t.~tu in case the first word was monosyllabic and had a short vowel. The latter restriction is irrelevant since after long vowels the long consonant was regularly shortened. The first condition, however, is essential, for the basic problem with which we are here concerned is that of "Sprachk/Srper". It is harder, indeed, to recognize monosyllables as individual words than to identify polysyllables. 4~ This is no doubt the reason why Vedic nd "like" is never contracted with any following vowel, nor dd "then" with a preceding -a. 41 The conclusion that the emphatic pronunciation of ss in *matsfi, vatsydmi was conditioned by the monosyllabic character of the noun stem and the verbal root is confirmed by the evidence of the stems and roots ending in -~. This will be discussed in the next section. 4. In cases where a final .s of a stem or root was followed by an s of a case or personal ending or a tense morpheme the result in Sanskrit was either .s~ (shortened to .s in final position and sometimes medially), e.g., :

jds.i (*jd.s-si), hav(s.su ; dviveh., viv~h., yauh., 4~


or it was k.s, e.g. :

*dve.s Wsi > dvek.si; *dve.s + s y d m i > dvek.syami; * advi.s 4-sam > ddvik.sam.
40 For this interpretation of the Old Tamil evidence see llJ, II (1958), p. 197. 4~ See Macdonell, Vedic Grammar, p. 61. 4~ Cf. bhaib in the later language, Bartholomae, KZ, 29 (1888), p. 574, Brugmann, Grundriss, I S(1897), pp. 725,812. In ddvf.s .salX. 128.5 there is a different sandhi.

THESANSKRITNOM.SING.V~t

113

Instances of this secondrepresentationare extremelyrare in the Rigveda,43 and ill the later languagethey are limited to the verbal inflection. Never, indeed, has -k.su for -ssu been introduced in the loc. pl. of the noun inflection. These facts point to the conclusionthat ks for sa is a comparativelylate, mostly post-Rigvedic,innovationof Sanskrit. It is, accordingly, neither an old,44 nor a regular 4~ development in Sanskrit, and it can hardly be doubted that Wackernagel's analogical explanation is correct. He assumedthat, as a result of the merger in the 3 sing. pres. of-.s+ti and -2+ti into -s[i (cL dvesti and vast.i from dvi.s-, vat-), the 2 sing. in -s+si (e.g., *dve.s+si) merged with that in -~+si (e.g., *va~+si), which was represented by -k.si. The 2 sing. dvek.si was, accordingly, created on the analogy of vaks.i. In the noun inflection the normal phonetic development in the paradigm of dvLs- "enemy" would consequently have been as follows: nom. sing. *dvis. +s > *dvi.s.s > *dv[h. (of. vivdh.). instr, pl. *dvi.s+bhih. > *dvi.zbhi.h > *dvirbhi.h (of. hav[rbhi.h). loc. pl. *dvi.s+su > *dvi.su ( ~ *dvi.s.su), with a shortening of the long .s.s which was properly a PIE., rather than an Indo-Aryan, development. It was assumed above that by the side of mds~t, massu, which represent the non-emphatic pronunciation of the case ending of the loc. pl., there has existed a variant *mas-tsu >*matsu, due to an optional emphatic articulation of-su, just as in modern Greek s, ss, and ts occur side by side as variants. To the same emphatic pronunciation must be due *dvi.s > *dvi.st.su >dvit.su, which is actually attested in the classical language. The contrast with the disyllabic form havLs.su RS. is the very thing that must be expected if-t.su owes its origin to an optional emphatic articulation after a monosyllabic stem. Since t after s always becomes .t in Sanskrit, whereas s after .t is never retroflex,46 this would seem the only explanation that is in full accord with the phonetic laws of the language. In the same manner the nora. sing. *dvLs+s must have become *dv(st.s > *dv#.s at a prehistoric time, before consonantal clusters lost their last components in final position. Afterwards *dv#.s regularly became dvit..

4a Wackernagel, Altind. Gramm., I, p. 137 with Nachtriige, Macdonell, Vedic Grammar, p. 25. 44 Thus Brugmann, Grundriss, 12, p. 735 ("Im Loc. P1. ist diese Entwieklung durch Systemzwang gestbrt"). 45 Burrow, JAOS, 79 (1959), p. 87. See below, p. 123. ae For the incorrect assumption of .t$ in proto-Indo-Iranian or proto-Indo-Aryan (e.g., J. Bloch, L'lndo-Aryen, p. 54, who followed Meillet's lead, Burrow, JAOS, 79, p. 87), see above, p. 104.

114

F.B.J.

KUIPER

The forms attested in the Veda are: -dv[t. RS. 1, viprfit. AS. 1, pravr" t. MS. 1,

KS. 1.
The forms *dvirbhih. and *dvirbhya.h are not attested although their former existence must be postulated on account of havtrbhih., f r o m *-i.zbhis (cf. Av. snaiOi~bya). Instead of such forms in -rbhi.h there are a few instances o f f o r m s w i t h -.d- before the endings -bhih and -bhyah.. The same analogy that gave rise to the instr, pl. madbhi.h RS. and the dat.-abl, pl. madbhyd.h AS. also accounts for vipr(t.dbhi.h VS. and vipr~.dbhya.h ~B. The correctness of the reconstruction *dvirbhi.h, etc., however, is guaranteed by the isolated f o r m dorbhydm VS. 1 of do.s- "fore-arm". 47 In this connection the root noun [.s- "prosperity, vigour, strength ''4s is particularly interesting. In p r o t o - I n d o - A r y a n the instr, sing. of root nouns was not formally distinguishable f r o m that of the stems in -a-. As a result, there occur some forms of the instr, pl. in -dbhih which are based on a reinterpretation of the instr, sing. in -d: cf. k.sap-dbhih. IV.53.7 (: k.sap-d IX.99.2), k.slp-abhi.h IX.97.57 (: *k.sip-d). 49 Inversely the instr, sing. has been remodelled after the instr, pl. in the inflection of i.s-, which can be reconstructed as follows: nom. sing. *(s+s>*i.st.s>*it.s>*#.. instr, pl. *i.z +bhi.h > *irbhi.h, instr, sing. *ir-d ~ nora.sing, ira, 6nira RS. etc. (iram AS. 2, etc.) loc. pl. *i.s+su > *i.st.su > *it.s~, instr, pl. *i.dbhih. ~ instr, sing. i.d-d In the RS. the last f o r m ild 1.128.7, III.24.2, 27.10 has not ousted i.sd RS. 26, AS. 6, but the new instr, pl. ildlbhih. 7 ~ instr, sing. il.aya 5 must soon have called into existence the noun il.a 16, ace. ilam 9, gen. ilaya.h 5, which continues to be used in the later Vedas (loc. i.dayam VS. 3). T h e word is equally frequent in the family books and in I and X (12 113 I l l 3 IV 1 V ~ V I P V I I P IX 1 X3)3 ~ The word ira 1 (iravant- 5) is m u c h less fre~7 Wackernagel-Debrunner, Altind. Gramm., III, p. 248. Wrist, Altindisch (1927), p. 79, still defended both .zbh> dbh (strongly questioned by Tedesco, Lg., 19 [1943], p. 18) and zbh > dbh. 48 For the inflection of is- see Wackernagel(-Debrtmner), Altind. Gramm., I, p. 212, III, p. 323. 49 See op.c., vol. III, p. 324, II/2, p. 260, resp. III, p. 241. 5o Wackernagel-Debrunner, Altind, Gramm., III, p. 323, questioned this origin of ida-, but see II/2 (1955), p. 259 and III, p. 248. As for ira- see L/iders, Philologica lndica, p. 552, Neisser, Zum W6rterbuch des 8gveda, I, p. 163, and Nachtriige ad Altind. Gramm., I, p. 24, 5. For the interplay of the declensions of is- and i.d- see Lanman, Noun-Inflection in the Veda, p. 493 ; for nouns in -a- replacing older stems in a consonant, cf. Altind. Gramm., HI, p. 323, II/2, p. 260, and llJ, I, p. 155, II, p. 307ff.

THE SANSKRIT NOM. SING.

V#.

115

quent, and it remains rare in the later texts (irWin AS. 2, irayai VS. 1). The differentiation between the three words (for their meanings see Neisser, Zum W6rterbuch des R.gveda, I, lI, 1924-1930) obviously dates from preRigvedic times. From the same time must date the shift of accent in k.sip~bhi.h (: k.sapdbhi.h), il.dbhi.h, Jr&. The aberrant accentuation ird in AS. XV.2.19 is hardly a trace of an older accentuation. The inference that may be drawn from these facts is that, in spite of the survival of the isolated instance dorbhydm in VS. 25.3 (with dot- in compounds, from the ~B. onwards), the outcome of the historical development can be stated in the morphophonemic rule that monosyllabic stems in -r.s-, -is-, and -us- change .s into t./.d before consonants and in final position. 5. The declension of the Sanskrit root nouns in -i-, -j-, and -h- (from PIE. k, ~2, ~h) was originally quite different from that of the stems in -.s-. It cannot be questioned that the regular representation of PIE. ks (for k +s, ~ +s, ~h +s) was ~ in Iranian and k.s in Indo-Aryan (as represented by Sanskrit). There is no reason, therefore, to question that the Vedic loc. pl. forms of vii-, f. "clan" and dii-, f. "direction", viz. vik.s~ (RS. 48 AS.2) and dik.sfl (AS. 9, VS. 1), represent the normal development of PIE. ks in Indo-Aryan. The Avestan form nddg "in straits" Y. 38.5, which apparently stands for PIE. *nefcs~, 51 furnishes additional corroboration for the conclusion that proto-Indo-Iranian had in the loc. pl. the normal representation of PIE. *-fcsu, no matter how this pre-stage of Indo-Aryan k.s and Iranian ~ is exactly reconstructed. 52 It follows that also in the nora. sing. the PII. forms must have been a direct continuation of PIE. *wlfcs and *dlfcs. Although the corresponding Avestan forms are not attested, 58 such parallel forms as spa~ "spy", par6.dard, name of a cock ("looking forward"), bard "mountain", -varg "working" (in du~var~tdvarg, gav~stravarM, hvard~, a~vara~ show the same representation of PIE. fcs as is found in the loc. pl. n~5~7. 5~ The regular Sanskrit form, accordingly, would have been *vlk.s > *vlk. Such nominative forms in -k actually occur in stems with r or .r immediately ~1 cf. Latin nex, necis. For Av. a in nd~, see Bartholomae, Altir. Wb., col. 1057n. 1. 5~ Cf., e.g., Bartholomae, GrIrPhil., I, 1, p. 12 (w n. 2), Kent, OM Persian, p. 34 (w92 with n.). Most important is the evidence of .so.d/~< *saz.d.hd< *s(v)ag(s)+dhd (el. Altind. Gramm., III, p. 354f.). 53 Av. vi5 Yt. 13.2, which according to Bartholomae, Altir. Wb., col. 1456, is a nominative of vis-, means rather "bird" (as in V. 2.42). See W. Henning, Asiatica, Fs. Weller (1954), p. 291. However, the existence of a nominative *rig is indirectly proved by the dat.-abl, pl. vi~ibyd, see p. 118. ~4 See Bartholomae, GrIrPhiL, I, 1, p. 216, ZDMG, 50 (1896), p. 705

116

v . B . J . KUIPER

preceding the final consonant (as in svar-d~k, etc., and upa-spr.'k from stems in -~-, vi~va-sr'k MS. from a stem in -j-, and perhaps gartd-rf& from one in -h-), ~5 or with the syllable beginning with a dental (as in dfk). The exact conditions under which -k occurs instead o f - f are hard to determine because there is some vacillation in Vedic. Hence the different formulations that have been proposed for the rule. 56 Such cases as the verbal forms asrgtk TS. : asr~t. MS. are characteristic of the uncertainty that existed among Vedic speakers regarding these forms. In later times -.t is, indeed, found even after .r (dh~n~bh.rt. "parching grain", mfda-v.rt. "tearing out roots" Pft.nini), where it stood traditionally in pravr.'t. "rainy season" MS, KS (stem prdvr.'s.-). It cannot be doubted, however, that the phonemic system of Indo-Aryan tended to avoid a retroflex stop in the immediate neighbourhood of a vocalic or consonantal r. 57 In the light of the preceding remarks on the restriction of ts (for ss) to the position after monosyllabic stems and roots, the fact that -k is also the only ending that occurs in disyllabic words like bhi.sdk "physician" RS. 358 and u.sn.fk "name of a metre" AS. 1 may be relevant. In rtv-ij- "sacrificing at the proper seasons, a priest" the nora. sing..rtvlk RS. II.5.7 may be due either to the fact that -/j- had early ceased to function as the root noun of yaj- "to sacrifice''59 so that rtvij- was no longer considered a compound, or to the preceding t. 6. Since there can be no doubt that -k was the regular representation, the problem is how the anomalous nominatives in -.t are to be accounted for. 5s For the stems in g see Lanman, Noun-Inflection in the Veda (= JAOS, 10) (1878),
p. 490; for those in ], see Lanman, p. 463, Wackernagel-Debrunner, III, p. 232f.; and for those in h, Lanman, p. 499, Wackernagel-Debrunner, p. 253f. The neuter form purusp.[k indirectly proves the existence of a nora. sing. masc. *purusp.~k "coveted by many" (root sprh-, cf. Av. sparaz-vant-); sraj- in hira.nya-srak X.6.4 "having a golden wreath" is ambiguous; for vi~va-s.[k see Nachtriige ad Altind. Gramm., I, p. 173, 40. 5e Cf., e.g., Meillet, IF, 18 (1905-06), p. 417ff.: nominative in -k after r/r (-dik, drk YV.), after dental (dik, .rtvik), and after a retroflex (bhis.6k, u&nik); Rysiewicz, Studia Jfzykoznawcze (1956), p. 290: -k also after i and u (like Ed. Hermann, KZ, 41 [1907], p. 43! accepted by V. V. Ivanov, Vopr. Jazykozn., 1958/4, p. 17, but the whole Rigvedic evidence is confined to the words dik, .rtvik, and vit.). Quite differently ThumbHauschild, Hb. d. Skt., I/l, p. 310. 57 Wackernagel, Altind. Gramm., I, pp. 166, 173; el. also ]invar av.rt TS., Sanskritization ofjinva rdvar MS., etc. (Vedic Variants, 2 [1932], p. 88). s8 bhir occurs IX.112.1 and 3, X.97.6, but bhisdktamarn II.33.4 shows that -k must have been generalized before consonants at an early date. Cf. bhi~dkti VIII.79.2 for *bhis.dat.i, and abhi.s.nak X.131.5: avoidance of .t after retroflex .s (MeiUet)? 89 Cf. deva-yd]- VS., which may have been inherited from PII. (: LAy. dagvayaz-), divi-yd]- RS., upa-ya]- ApS., and in the classical language -yd]- (Renou, Grarnmaire sanscrite, p. 323). There are no further traces of 4/-; Av. aibig is fully obscure (abi.iz-? but see Bartholomae, Zum altiranischen W6rterbuch, p. 106, Nyberg, MO., 35 [1931], p. 197 n. 2). [For rtv-i]- : ya]- see however Oldenberg, Noten ad 1.1.1.]

THE SANSKRITNOM. SING. v[.t The Rigvedic evidence comprises the following forms: stems in -g-: v#., spdf. Cf..sd.t from *swdks; stems in -j-: bhrdt. (vibhrdt.), rdt. (and compounds); stems in -h-: abhi-s.dt (and other words in -.sd.t), havya-vdt. (and other words in -vdt.).

117

Besides, the retroflex occurs before -bhi.h in pa.dbhi.h IV.2.12, 6~ .sa.dbhi.h II. 18.4 and, in the last books, in vi.dbhi.h X.28.8 and sard.dbhya.h I. 112.21 (with a disyllabic stem). The nom. sing. of PIE. stems in -k-, -~-, and -~h-, which ended in -ks, is found intact in Lat. (haru)-spex, Av. spaL In contrast with the latter form Ved. spdt. cannot directly represent PIE. *spdks} 1 However, in the light of the explanation proposed for dv#. the form spdt. is perfectly clear. In accordance with a well-known morphophonemic rule ~q-t is represented by .s.t. If, therefore, it can be made plausible that in proto-IndoAryan the inherited nominative form *spdk.s (from PIE. *specs) has been replaced by a new formation *spa~-s (with the normal antevocalic representation of k by d), the further development *spaits >*spd.st.s> *spdts > spdt. is much the same as in *dvi.s-s > dvit..6~ The motive for this substitution of the stem spdg- for *spak- may have been the tendency to treat the stems in the sibilants Aand .Sin an analogous way. In Avestan this parallelism was the natural result of the phonetic development of PII. *g-s and *g-s to 7; cf. spa~ and *i~ (of the stems spasand i~-) contrasting with vaxg, gen. v~c-~. In Indo-Aryan the normal development would have yielded *spdk, gen. spa~-d.h and vdk, gen. vac-d.h, contrasting with *it., gen. i.s-d.h. In the loc. pl. this contrast (-k.su vs. -fsu) was maintained, but in final position -k was replaced by -.t owing to the introduction of a new nominative form *spag-s. The real difficulty is not, however, this nominative but the instr, and dat.-abl, plural. For proto-Indo-Iranian these case forms of the stem *rig- must be reconstructed as *vil-bhi~ and *vi$-bhyas. They have, however, been replaced by new formations in all the older dialects: Old Persian has viObigfor *vidbiL Avestan (in both dialects: Y. 53.8, V. 13.10f., so But see Altind. Gramm., III (1930), p. 247, Nachtrlige ad I (1957), p. 172, 1-17. ~1 -t was explained as the result of a regular phonetic development (against Wackernagel, Altind. Grarnm., I, p. 173) by, e.g., Meillet, MSL, 8 (1894), p. 284, IF, 18, p. 417f., Ed. Hermann, KZ, 41 (1907), p. 40f., Bloch, L'Indo-Aryen, p. 53t". 63 Hermann Jacobsohn, Arier und Ugrofinnen (1922), p. 146, also assumed substitution of the antevocalic stem form (which he writes as *spat'-) for *spdk(~), but tried to explain sprit, as the normal development of a nominative *spdt" (without s), which is hardly possible. Similarly Meillet, MSL, 8 (1894), p. 284!

118

r . B . J . KUIPER

17f., 40) vT~iby6 for *vizbyah, and Ved. *vi.dbh[h. (see above for pad.bhih.). The latter form must have taken the place of the regular representation of PIE. *wi~,bhis (probably with elimination of the voiced hushed consonant and lengthening of the preceding vowel: *v~bhih cf. perhaps the nora. sing. puro.dd.h, stem puro.ddg-, from the instr, plur. *puro.dd[~]-bhi.h?). This statement calls for some comment. Wackernagel-Debrunner, although rightly rejecting the older theory of -zbh- becoming -dbh-, 6~ still hold on to the corresponding assumption that -z.bh- has become -.dbh-. 6~ However, Bartholomae's view that PII. and ~ had become ~ and ~ before stops 65 is not sufficiently supported by the evidence. As far as I can see, there is no reason to doubt that in Avestan PII. g and $ are also before stops regularly represented by s and z: LAy. v?spaiti~, corresponding to Ved. vi~pdti.h "chief of a community" points to the existence of a word *vi~pdti- in the common Indo-Iranian proto-language. It must, therefore, be an old inheritance in Avestan. As such it is very unlikely to have been transformed secondarily, with substitution of v~s- for *vi~-. The same is true of GAv. rnazibig [ = mazbi~, instr, pl. of maz- "great", which stands for PII. *ma~Hbhig. 66 On the other hand, for the only counter-instance, viz. vf~ibyO (mentioned above), an analogical explanation would seem quite natural. The form should obviously be paralleled with the corresponding forms of v~tk- "word", viz. va~ibig N. 41, 67, 103, and v ~ i b y 6 Y. 16.1, Yr. 10.88, Vr. 6.1, etc. In the light of vax~.barati- "offering of a prayer" and parallel formations Bartholomae was no doubt right in explaining v~?~- as the form of the nom. sing., which here has taken the place of the stemY In quite the same manner v~iby6 is likely to stand for *vizbyah (cf. v~spaitis3 and to contain the nora. sing. *viL It is impossible to determine what the regular development of *vi$bhig

68 Altind. Gramm., III (1930), p. 289. eL Altind. Gramm., III, p. 246 (*purod.a.zbh-> *purod.ad.bh-)with reference to vol. I, p. 173ff. Similarly Thumb-Hauschild, Handbuch des Sanskrit, I/1 (1957), p. 299, I/2 (1959), p. 92, where d.bh is explained from PII. ~bh. e~ See GrlrPhil., I, 1, pp. 12 and 18, Reichelt, Aw. Elernentarbuch, p. 48. Also Kurylowicz, L'Apophonie en indo-europden (1956), p. 372, assumes a proto-IndoIranian development "/~ devient .s devant occlusive". 66 The representation before continuants (e.g., Reichelt, op.c., p. 48) is of course a different problem: gn and sn reflect different dialectal developments, see MeilletBenveniste, Grammaire du vieux-perse, p. 72 (otherwise Hans Hartmann, OLZ, 1937, col. 156). For vi~pfti- see also Altind. Gramm., II/l, p. 242. 67 See Bartholomae, Altiran. Wb., col. 1335, and cf. druxLmanah-, bazuLaojah-, vigpaOa-, Bartholomae, GrlrPhil., I, 1, p. 150, Altiran. Wb., col. 781, etc., (but cf. Duchesne-Guillemin, Les cornpos~s de l'Avesta, pp. 15f., 186), Debrunner, Altind. Gramm., II/2, p. 724.

THE SANSKRITNOM. SING. V[.t

119

would have been in Indo-Aryan. The other voiced sibilant and the corresponding hushed consonant, viz. [z] and [.z], were mere allophones of the phoneme/s/. The first disappeared in Indo-Aryan in all positions with compensatory lengthening of the preceding vowel, the second disappeared before .d(h), .n and became r in other positions. The same is true of the representatives o f / ~ / a n d / 2 h / b e f o r e dentals and n, where they merged with .z, e.g. *mu~h+ta- ~ *mu~dha- ~ *muz. .dha- ~ rnft.dhd.. In all positions before vowels and continuants/~/became the palatal stop j, e.g., ajati, j~u-, jma.h, jraya.h, jvalati. Immediately before stops, however, j does not occur in Indo-Aryan. Since the evidence of Indo-Iranian points to the former existence of a phoneme/~/, parallel to/~/, *vi~-bhyas must have existed in proto-lndo-Iranian, as *viJpati- did. In Indo-Aryan voiced sibilants and hushed sounds were never preserved in this position. Perhaps, therefore,/~/was dropped in *vi~bhyas in proto-Indo-Aryan. However that may be, it is clear that in the stems in g, j, and h the stem was replaced by the nominative sing: s before the endings -bhi.h and -bhya.h. Hence it is that digbhi.h,69 digbhyd.h (AS. 3 VS. 6) are the forms corresponding to the nom. sing. dfk, whereas vi.dbh/.h RS. 1, vi.dbhyd.h AS. l (RS. KhiIa) are those corresponding to the nora. sing. v#.. In this way the parallelism between

vii, gen. vi~-dh. : vi.dbhih., vi.dbhyd.h


and

dvlt., gen. dvi.s-6.h : dvi.dbhi.h, dvi.dbhya.h


had become almost complete, the only exception being the loc. pl. vik.s(t as against dvit.su. The latter form, though not attested, it seems, before the classical language, probably dates back to a much earlier period. The fact that in the classical language also vik.su was replaced by vit.su shows that the tendency for parallelism continued to operate in later times. If praydtsu TS. VS. for praydk.su AS. indirectly proves the existence of *pray6t.su, this would be the earliest instance of this new formation. The use of this secondary stem in .t/.das the first member of compounds is (except for .s6.t-) not found before TB. v#.-pati-, X~v~. vit.-kula-"house of a Vai~ya". Some more instances are attested from Mauu onwards: prd.d.vivdka-"judge'" (for which the earlier language uses pra~navivdkdVS.), vit.-pa.nya-. The Mahdbhdrata has prd.d-vivdka-, the Rdm6ya.na 68 Either the original nominative form in -.ts, or the stem in -.t that was inferred from that nominative, or the later nora. in -.t. The relative chronologycannot be determined since this must have happened in the prehistoric period. ~9 Cf. RS. susarh-dfgbhi.h.

120

F. B. J. KUIPER

furnishes vil-gftdra-, 7~ and K~lid~sa has vi.d-ojas- (beside vid.aujas-, which is not morphologically clear). Similarly parivrd!- in compounds (Gan.apat.ha), from parivraj- which itself is not attested before the epics. 7. An analogous phenomenon is well known from the 2 and 3 sing. of the s-aorist and the root inflection. The forms attested in the Rigveda are the following: 2 sing.: s-aorist (for *-a~+s-s), from the root yaj- the injunctive form yat 1 (but indicative aya.h 2); from the root yah- the indicative form dvat 1. From the root aorist (for *-~-s) dnat (anal) 2 from the root nag-. 3 sing. : s-aorist (for *-a~+s-t) aprat. 1 (root p.rg-), like asral MS. from s.rj- (but RS. asrak 2, like adrak from d.rg- in the Br~hma.nas). 71 Root forms of bhraj-, raj- and nag-: 6bhral 2, rdt 1, dnal (anal) 24, abh[ nat. 1 (but pran. ak 4). In these cases the regular ending would have been -k (from -*ks[s]) in the 2 sing., but -~.s (for *-~sl) in the 3 sing. Of this original situation no traces are left. The ending of the 3 sing., with final -.s after a, could not remain intact since the rules of Sanskrit do not allow the occurrence of this final consonant in this position. In the 2 sing. -k might have been preserved but here a tendency for clarity caused the personal ending-s to be over-emphasized. The whole course of events can be reconstructed as follows: 1) Owing to the need of emphatic articulation of the personal ending -s in the 2 sing. the root could not be maintained in its inherited form with k before s, but was pronounced with voiceless palatal ~ before s. It is unlikely, however, that in the s-aorist such a pre-form like *ayak.s-s has at any time been replaced by *aya~-ks-s. Here the ending -I of ya.t is rather due to the analogy of the root forms. 2) J - s became -gts and further, in accordance with a morphophonemic rule which dates back to proto-Indo-Iranian, Jts. 3) A fundamental change in the consonantal system of Indo-Aryan, which is the earliest trace of the genesis of an Indian linguistic area, caused *-~ts to become *-s.ls. 4) As the result of dissimilation *-.sis became *-!s. 5) In this final consonant cluster the -s was dropped. This analysis of the various stages of this process gives us an insight into their relative chronology. At some time the old morphophonemic rule

To Ram. I. 6.21 in the Eastern and N.W. recensions. Cf. the critical ed., vol. I, p. 53 (248", 17ab). ~1 See Nachtrlige ad Altind. Gramm., I, p. 174, 37.

THE SANSKRIT NOM. SING. vii

121

"g > ~t" has been replaced in [ndo-Aryan by "d > .st". It cannot be determined whether or not the new formations of the type *abhrdd-s were anterior to the introduction of retroflex phonemes. Apart from this detail, the relative chronology of the various stages would seem unquestionable. This leads to the following conclusions. First, as Bartholomae has rightly pointed out, 72 the final -.t in the verbal and nominal forms must date from a period of Indo-Aryan when /~/ and/s representing PIE. ~ and ~h, had not yet merged w i t h / j / a n d /]h/, the results of a particular PII. palatalization of PIE. g, gU and gh, gUh. Second, the further development inevitably presupposes the introduction of a new series of retroflex phonemes, which again can only have taken place after proto-Indo-Aryan had come to be spoken in the Indian linguistic area. Third, all developments must have taken place at a time when final consonant clusters had not yet lost their last components. Not before the ending of the 2 sing. had become a single -.t could this ending also be adopted for the 3 sing., in accordance with the general formal identity of these two forms after the simplification of final clusters. The conclusion resulting from this analysis of the historical process is, accordingly, the very opposite of Wackernagel's. 7z

E X C U R S U S O N B U R R O W ' S T H E O R Y O F PIE. TK < Skt. K.S

In the introductory section it was pointed out that the problem under discussion, although of minor importance when taken by itself, has some consequences for problems of much wider interest. One of these is Burrow's theory of the origin of those Indo-European words for which the traditional Comparative Grammar used to posit such phonemes as *kb, *~hO, *gUhd, etc. Since his theory seems to have found more or less general acceptance, it may be useful to summarize the main objections to it, in so far as they are implied in the preceding article. An exhaustive critical analysis of this theory is not, however, aimed at. Burrow's interpretation of the evidence is to be found in two articles in JAOS, 79 (1959), pp. 85-90, resp. pp. 255-262. In the first article he studies the words commonly supposed to represent some modification of PIE. palatals (*~b and g,hd or */c8 and *~hz), in the second a few words with a PIE. labiovelar (*gUhd). Since the conclusions of my article have
72 ZDMG, 50 (1896), p. 709 73 See Altind. Gramm., I, (1896), p. 174. As for the 3 sing. Wackernagel's reserve
should be noted: " I n der 3. scheint -.t aus .s.t: ig.-~(s)t lautgesetzlich".

122

r. B,J. KUIPER

a bearing on the first of his articles only, I shall here mainly confine myself to this one. Burrow's argument is as follows: 1) It should not be taken for granted that the s. in Skt. r.'k.sa- "bear" stands for PIE. s or b, since in r e d . ks.umdnt-, Av. f~umant- "possessing cattle" Skt. ks.u- stands for PII. *p~u- (from PIE. *pku-). In other words, PII. d (from PIE. k) can be represented, under certain conditions, by ~. 2) Skt. k in ks. derives not only from the three PIE. guttural series but also from "the internal sandhi of .s-t-s". It follows that Burrow explains dvek.si as the result of a phonetic development. It is not clear, however, if he considers this the only regular development or rather a sporadic development, by the side of hav(s~u. This is a problem which he does not discuss. 3) On the other hand, in vit. and dvlt., which he takes to stand for *vlt.s. and *dv[t..s, the combinations d + s and s.-t-s are represented by t.s (>.t). From this fact he concludes that .ts. was the earlier stage of ks., dvek.si having passed through a stage *dvet..si. The development Ss >.t.s is said to be parallel to Skt. ss > ts, where "the first s is replaced by the corresponding (dental) occlusive". In quite the same way the word :ks.a- (Av. ar~a-) is explained from *rts.a-. In final position -.t.slost its last component before .t.sbecame k~. Only in cases like -d/k "the change took place earlier so as to affect the group even in final position". 4) "The ~, s. seen in Av. f~umant-, Skt. ks.umdnt- is an instance of a widespread tendency in Indo-Iranian to replace the palatal series (i, d, dh) by the cacuminal or cerebral sibilants when these palatals were in contact with other consonants. Similar examples are seen in Av. sa~kustoma"who understands best" (sak-), Av. vi~iby6, Skt. vi.dbhyds<*viz.bhyas (vii-), Skt. oak.s- "to see" (kad-), Av.fra~na- "question" (but Skt. pradna-)." 5) In this way Skt. r.'k.sa-"bear" can be explained from *.r.t.sa-< *.rtda- < PIE. *H.rtko-, and similarly Skt. ks.am- "earth" from *t..sam-< *tgam-< *ddham- < PIE. *dh~hom-. As far as I can see, the first four points do not justify the conclusion drawn sub 5. They will be discussed separately. As for the first point it should be observed that PIE. *ku~ regularly became *kd in proto-Indo-Iranian, as is shown by the root extension PIE. *kuk-eH -, which became xsgt- in Avestan, and kda- in Vedic. This is almost exclusively attested in the Maitrdyan.~ya Sarhhitd, viz. akAat I. 8.9, II. 1.4 (eL also M ~ . III. 3.6, VI. 1, 1; 15). In the other Vedas it is replaced by khy~-, but in KS. 7.10 (p. 72, 2) anukggttyai - anu vgt akhyan the older form is still used side by side with the later dialectal form. Cf. also 19,2 (p. 2, 20), Only in a few corrupt variant readings k.s is found for

THE SANSKRITNOM. SING. V[.t

123

khy (rather than for the original form with k~fi-), cf. KS. 4.2 (p. 30, 16) avak.sat for avakhyat, 16.10 (p. 231, 18) ak.sar for akhyad (but 16.9: p. 230, 20 akhyar). Inversely, the same text sometimes reads khy for ks, cf. 13.16 (p. 199, 8), akhyabhi.h, v. 1. for ak.sabhi.h. See also IIJ, IX
(1966), p. 222. The case of k.sumdnt- is different because the proto-Indo-Iranian form must still have been *pgu-mant-. In proto-Iranian *p5 seems to have merged with *ps into *fg, cf. PII. *nap(t)su > Av. nafju, and PII. *pgu- > Av.fgumant-, mod. Persian 5ubfin "herdsman". 74 In Indo-Aryan the following rounded vowel plus m caused the preceding labial p to be dissimilated, just as in *ploman- (Greek ~L~6gmv), which became kl6man"the right lung". Similarly after the labial phonemes pu- in *puru-pgu"possessing (or, consisting of) many cattle", which became puru-k.s{t-. In this comparatively late and specifically Indo-Aryan development *kSwas replaced by the current combination ks. Although it is impossible to give a fully satisfactory explanation of the .s in this isolated case of normalization, this much seems certain that it cannot be a basis on which to found such general statements as Burrow did. Skt..s does not stand for ~ except before t (where it reflects a PII. development). Burrow's second thesis to the effect that dvek.si represents a phonetic development .s-ks > k.s seems to disregard Ved. hav(s.su and dvive.h. If he has not actually overlooked these counter-instances he must have meant a "sporadic" phonetic law (see below, sub 4). Since he does not mention Wackernagel's analogical explanation, his reasons for his silent rejection of a generally accepted theory are not clear. In the absence of any argumentation it is difficult to see what may have induced Burrow to substitute for Wackernagel's theory a novel one, for which little support can be found in the evidence. From the third thesis it must be inferred, although this is not explicitly stated, that both Skt. k.s and Av. Y derive from *.t.s. If this is correct, Burrow attributes .t.s to the phonemic system of proto-Indo-Iranian. However, the assumption of a retroflex phoneme .t in PII. lacks any foundation, and the additional assumption of a retroflex .s, allophone of s after the retroflex .t, is not even supported by the evidence of Sanskrit. See above, p. 104. These assumptions, ill-founded in themselves, necessitate the further assumption of different chronological stages for the change *.t.s> k s in -d.rk on the one hand, and in vif on the other, which adds a new complication to the theory. As for the last thesis, Burrow's rather vague reference to "a widespread 74 Bartholomae, GrlrPhil., I, 1 (1895), pp. 11, 13,

124

F.B.I.

KUIPER

tendency" involuntarily evokes reminiscences of the long-forgotten notion of "sporadic sound laws". Burrow is, indeed, quite aware of the fact that he is not dealing here with a "normal" phonetic law. The tendency is supposed to have left just some occasional traces, sometimes only in Iranian but not in Indo-Aryan. In fact, none of the examples quoted can support the assumption of a change ~ > ~ before consonants in the widest sense of the word. Av. sa~ku~tama-, which contains the perf. ptc. of the root sak- (Skt. ~ak-) probably stands for *sa-sk-u~-. Unlike Bartholomae 75 1 think sagk- is rather due to the influence of the pres. ptc. saJqs (Geldner sax~qs). The present sa~aiti ( < *sacyati) may also account for the perfect preterite sa-~k-on. For Av. v~iby6 see above, p. 118. The brief reference to "Skt. cak.s- 'to see' (kai-)" seems to imply that Burrow accepts the theory which explains cak.s- as the reduplicated perfect stem ca-ks-. In spite of Leumann's tempting analysis of the noun cdk.su.s- "eye" as the stem of perf. pte., ~n this is, I am afraid, impossible because of Av. cajrnan- "eye", which shows the root caJ- to be an s-present of Av. kas- (dkasa.t "he noticed, perceived")/7 As for Av. fragna- "question" by the side of yasna- (with s, root yaz-), it has long been recognized that this is not an instance of a sporadic phonetic development but rather a dialectal (more particularly, S. W. Iranian) form; cf. OP. vaJna-. 78 It must be stated, accordingly, that none of the four theses upon which Burrow's explanation of Skt. r.'k~a- and ks.din- is based would seem to stand a critical examination. Therefore, it is impossible to derive these words from PIE. *I-I(tko- and *dh?,hdrn- in the way proposed. Since these critical remarks are confined to the Indo-Iranian evidence, I can only incidentally point out that also the thesis of a phonetic development kt > ks in Latin (Burrow, p. 88) is open to grave objections. So the explanation of Lat. ursus from PIE. *Hr.'tko-, which is based on this thesis, lacks a solid foundation. A single remark may be added on the second article, which gives an ingenious explanation of GAy. dgfi!.ardta- (LAy. jit.aga-) and GAy. d~jgtrn6spa- C LAy. jdrndspa-). However, Burrow omits to mention that this explanation isolates these two words from the other words where d and t are found 79 (a) before palatals: vi-J-c6i~ta-, superlative of vi-cira-,
75 ArischeForschungen, II(1886), p. 52f., GrIrPhil., I, 1, pp. 17, 92, Altiran. Wb., col.
1553.

~ Morphologische Neuerungen im altindischen Verbalsystem, p. 33 [ = Mededelingen Kon. Ned. Ak. v. Wetensch., Afd. Letterk., N.R., 15, 1952, p. 105]. ~7 See Acta Or., 12 (1934), p. 243. 78 See Meillet-Benveniste, Grammaire du vieux-perse, p. 71f. ~9 Bartholomae, GrlrPhil., I, 1, p. 158.

THE SANSKRITNOM. SING. V[.t

125

and urva-.t-ca6m s~ ardna-_t-ca6ga-81; (b) before b: fra-~-baoye, vi-~-baoye, GAv. ddbqz- (: LAv. bqz-, Skt. barhh-) ~2, (c) before k: _tka6da-, vi-.t-ka6~a-. F r o m the viewpoint o f Avestan text tradition there is, accordingly, no reason to p r o p o u n d a different explanation for the d in d~j?_t.arata- than for the other cases with ~ before palatals, like v~-~.c6i~ta-, and to reject silently the interpretation o f the spelling ddj- for j- given by Geldner, Bailey, and Karl H o f f m a n n ? ~ N o r would the traditional equation of Skt. k~ar-, Pkt. jhar- with Av. 7~ar- seem to be invalidated by the new theory which, t h o u g h testifying to its author's acumen, is nonetheless unacceptable.

so Geldner, Studien zum Avesta, I (1882), p. 55. 81 Cf. B. Geiger, Die Ama~a Spantas (1916), p. 217 n. 1, Kuiper, Aeta Or., 12 (1934), p. 280 n. 2, Humbach, MSS, 2 (1952), p. 10 n. 13, P. Thieme, Mitra and Aryaman (1957), p. 53 n. 39, Gershevitch, The Avestan Hymn to Mithra (1959), p. 186. 82 Spiegel, Commentar fiber das Avesta, I (1865), p. 300, Kuiper, Acta Or., 17 (1939), p. 51, Duchesne-Guillemin, BSOS, 9 (1937-39), p. 864f., Morgenstierne, NTS, 12 (1940), p. 70. For LAy. dbOidta- see Mayrhofer, Kurzgef. etym. Wb., II, p. 513 n. ss See Geldner, op.c., p. 54 ("ungenaue pleonastische Schreibtmg"), Bailey, Zoroastrian Problems in the Ninth-century Books(1943),p. 191 (] = d$ broken up by a vowel), K. Hoffmann, lranistik, I (1958), p. 11.

You might also like