doubts the relationship with the Germanic words. Quite possibly Greek has lost intervocalic s rather than į; and in any case we should probably recognize n as the vowel of the uncontracted theme, as in Sny 'live'. Hittite hameshanza, then, contains the PIE prefix **hoand the root **mē- with formative sh,22 which may have been present also in a prehistoric stage of Gk. åμáw.

The noun happar 'deal, trade' and the verb happara- 'sell'23 go with Gk. érépaσa 'I sold', eπopov 'I gave' and Lat. paro 'buy'; they contain the same prefix that we have been discussing.

Similarly we may connect hulaliya- 'wind, entwine', GIšhulali 'distaff', and hulali 'winding-sheet' (?)24 with IE *uel- 'wind' (Lat. volvo, Gk. eiλów, Skt. vrnoti, etc.), the second l belonging to a suffix. Just so hušk- 'await'25 contains the nil-grade of *ues- 'dwell' (Skt. vasati, OHG wisan, etc.), with iterative-intensive k (cf. halzišk- beside halziš-). Hittite harp- 'place, lay'26 must be connected with Skt. arpayati 'throws, places' and arpanam ‘a throwing, placing'. Either the Hittite verb contains the prefix ha- or Skt. arp- originally had no connection with r- 'go'; for the latter root appears in Hittite without initial h (ari, arai, artari, arnuzzi).27 In either case we have here another word in which Sanskrit causative p was inherited (see LANGUAGE 4. 4).

Hittite hapatiš 'servant, vassal'28 is identical with Homeric oηôós (Doric óradós) 'attendant', except for the stem vowel. The Hittite word disposes of Fick's2 connection of brŋdós with eroμaι, and suggests the PIE prefix **ho-. The Homeric synonym óráwv contains as its second member the participle of the verb which appears in Hittite as pa(i)- 'go'.30 Both the Hittite participle panza 'going' and the second member of oτάwv represent PIE **pāiont-. Homeric orŋdós probably, contains the root of ndáw 'leap', which therefore contains original ā, and must be separated from Tod-, etc. I do not know of any Hititte

22 Formative s is very common in the Hittite as in the IE verb. Friedrich, ZA NF 1. 16 f., called attention to the Hittite causative suffix ah(h). There is also a simpler form of the suffix (nil-grade?), whose meaning is not clear. 23 See Friedrich, Staatsvert. 92 ff.

24 See Sommer and Ehelolf, BoSt. 10. 72.

25 See Ungnad, Orientalistische Literaturzeitung 26. 572 (1923); Friedrich, Staatsvert. 170.

26 See Friedrich, ZA NF 1. 175, (1924).

27 See Friedrich, ZA NF 2. 41–5 (1925); Sturtevant, LANGUAGE 3. 165 f., 220 f.

28 See Götze, Madduwattaš 105 f.

" Vergleichendes Wörterbuch der Indogermanischen Sprachen 14. 141.

30 See most recently Götze, Madd. 597.

word, other than hapatis, which contains a dental extension of pa(i) 'go'.

Medial h appears in a number of personal endings, and here too it is impossible to find a corresponding sound in the IE languages. The endings with h belong to the second or hi-conjugation or to the mediopassive voice, and, with one exception, they designate the first person singular.

Since the hi-conjugation has several points of contact with the IE perfect and several others with the IE aorist, it is reasonable to search among the perfect and secondary endings for analogues to Hittite -hi, -ahhi (first person singular present) and -hun, -ahhun (first person singular preterit) If we carry -ahhi (e.g. šaggahhi 'I know': šakki 'he knows' back to PIE -ahi, IE ought to show -ai, and this is what we find in Skt. tutude 'I have thrust', Lat. tutudī 'I have beaten', Old Slavic vědě 'I know', etc. This is a middle ending in Skt., although Lat. and OSI., like Hittite, make it active. Since IE had another ending for the perfect active (e.g. Skt. veda, Gk. olda, Goth. wait), it probably agreed with Skt. rather than with Lat. and OSI. in this matter. But if Lat. and OSI. could transfer the ending from middle to active, a similar transfer was possible in Hittite or an opposite transfer in primitive IE; and so we need not hesitate to identify a Hittite active ending with an IE medio-passive ending.

In Skt. the primary as well as the perfect middle ending is -e, and there is a good chance that -ai was the primary middle ending in the IE period, at least for non-thematic verbs. Quite possibly, also the Skt. secondary middle ending -i (e.g. akri 'I made': akṛta 'he made') was used in the IE non-thematic verbs. With this we may identify the Hittite ending -hi, as in arhi 'I have arrived'. The preterit endings -hun and -ahhun are apparently the results of contamination of -hi and -ahhi with -un, the preterit ending of the mi-conjugation.

Since it is impossible to consider the medio-passive endings with h apart from the other medio-passive endings, we must examine the system as a whole. It is already well known that Hittite possesses mediopassive endings in r analogous to those of Italo-Celtic and Tocharian; but it is not so widely known that the language has an equivalent set of endings without r. The material is presented in parallel columns in Table I.

Such a variety of equivalent forms is somewhat bewildering; but a

31 For the former, see LANGUAGE 2. 33 f., 3. 161-8, 215-25; for the latter, see ib. 3. 223-5.

little study reveals the fact that the present forms without r form the basis of most of the others. The preterit consists largely of present forms with a suffixed t or ti. Thus kišat 'he became' is an extension of kiša 'he becomes', and the plural kišantat 'they became' adds t to the ending of iyanta 'they come'. There is the same relationship between the end

ings of the second pl., kišdumat 'you became' and pahšduma ‘you protect'. The second singular kištat 'you became' contains a third personal ending, a thing which is common in the preterit active; e.g. pešta 'you gave' and 'he gave'. If in the same way we remove the final t of the first singular, we arrive at basic endings -ha and -haha, which one might expect to find at the head of our first column.

The longer suffix -ti appears in išduwati 'he appeared' and nahšariyandati 'they feared'; but this, unlike t, sometimes appears in present forms. The second personal neyattati 'you turn, send' may or may not be an extension of a form *neyatta; but, at any rate, ešwaštati (e-šú-wa-áš-ta-ti) 'we sit' has a by-form eswasta (e-šú-wa-áš-ta).32

Similarly the present forms in the second column of Table I are clearly derived from those in the first column by the addition of a suffix -ri. Compare kiša and kišari, iyatta and iyattari, iyanta and iyantari. Consequently the first personal forms iyahhari 'I go' and arhahari 'I come' presume some such forms as *iyahha and *arhaha; that is, precisely such

present forms as should lie at the basis of the preterits iyahhat and iyahhahat. In view of these two lines of evidence I conjecture that Hittite once possessed a present middle first sing. ending in -ha or -haha. Probably we shall sooner or later find such forms in present value.

The imperative second pl. ardumat 'come ye' has the same ending as the preterit kišdumat, and the ending of kidumati 'lie ye' is related to this as is the ending of išduwati 'he appeared' to that of kišat he became'. The active voice also uses the preterit second pl. as imperative, e.g. iyatten 'ye made' and 'make ye'. Compare the 'injunctive' use of augmentless aorists in Skt. The r-forms of the imperative are obviously analogical; compare the active forms iyallu 'I will make', iyaddu ‘let him make', iyandu 'let them make'. I have no suggestion to offer as to the imperative second sing. ending -hhut.

The relationship of some of the Hittite medio-passive endings to IE 32 See Friedrich, AO 4. 95 (1927).

terminations is obvious. The -ta of the third sing. present and -nta of the corresponding pl. must be identified with the IE secondary endings -to and -nto. The only element in Table I that can contain the final -ai, which is characteristic of the primary tenses of the middle voice in Gk. and Skt., and to some extent also in IE, is the separable -ti of ešwaštati išduwati, etc. The particle or suffix **tai, which we must therefore assume for PIE, is the direct source of the IE primary ending -tai, and from there final -ai was no doubt carried by analogy to other IE primary middle endings, such as the plural *-ntai. After the close of the IE period the process was carried somewhat further in Greek and much further in Skt., as is evident from Table II (slightly altered from Brugmann).

The first pl. ending -wašta looks like a contamination of the IE dual active ending *yes, *-yos, which might appear in Hittite as *-was, with the dual middle ending *uedha (Hittite *-weta), which I infer for IE on the strength of Skt. -vahe, Gk. -μelov, (with -ov from the other persons), and the Gk. pl. ending -μea. Possibly Homeric Greek peрóμeσła 'we bring for ourselves' is a similar contamination of pepóuela with the active époμes, although the latter form survived into the historical period only in the Doric dialects. If so, the contamination probably occurred in primitive Greek, and this ending may have induced the σ of the second personal ending -σ0e (beside Skt. -dhve).33

The second pl. ending -duma must stand in some relation to the Skt. endings -dhve (primary and perfect) and -dhvam (secondary). It may contain the nil-grade of the secondary ending with a final a (originally o) from the third person.

Of peculiar significance is the correspondence of the r-endings of the third person with certain medio-passive endings of Italo-Celtic and Tocharian; -tari = Lat. -tur, and -ntari = Lat. -ntur, while -ari corresponds in form with the ending of OIr. berir, -berar 'he is carried', and of Osc. sakrafír 'let one sacrifice'. The evidence of Hittite does not favor the connection of these endings with the r-endings of the third pl. active, which has become almost a dogma of IE comparative grammar. Hittite employs the ending -er (often written -ir) to form preterit third plurals of either conjugation; e.g. kuenzi ‘he kills', kuenta 'he killed', kuennir 'they killed'; dai 'he takes', daš 'he took', dair 'they took'. While this active ending is appended directly to the verb stem to form

33 Brugmann, Grundriss der Vergleichenden Grammatik der Indogermanischen Sprachen 22. 3. 651, derives the σ of -μeo0a from the second personal endings -oe and -σov; but there is no evidence that the sound was original there either.