言 語研 究(Gengo Kenkyu)88(1985),68~85

Problems in the Mediopassive Endings

of Indo-European

Kazuhiko YOSHIDA

1. One of the morphological categories most resistant to his- torical analysis in IE is mediopassive endings." The correspondence of present mediopassive endings inherited into IE daughter languages is given below (the proto-forms for the 1 and 2 p1. endings are hardly to be determined because of the variety in forms in the daughter languages, and therefore I limit my discussion below to the 1,2 and 3sg. and 3 p1.):

Ved. Gk. Lat. OIr. Toch. A[B] Hitt. Goth. (Arcado-Cypriote) sg.1. -e -ƒÊƒ¿ƒÇ -or -ur -mar [-mar] -ha(ri) -da

2. -se -ƒÐƒÍƒÇ -re/-ris -ther -tar [-tar] -ta(ri) -za

3. -(t)e -ƒÑƒÍƒÇ -tur -thir -tar [-tar] -(t)a(ri) -da pl. 3. -nte -ƒËƒÑƒÍƒÇ -ntur -tir [-ntar] -nta(ri)] -nda

As is noted, everywhere present mediopassive forms are attested, simple endings are extended by the hic et nunc particle -i or by the r-element, both of which were optionally used in PIE. In Indo- Iranian, Greek and Germanic, -i was generalized, while in Ana- tolian, Tocharian, Italic and Celtic it yielded to the r-element. The function of -r(i) of the Anatolian, Tocharian, Italic and Celtic mediopassive is quite parallel to that of the -i which charac-

1) I am happy to express my gratitude to Selina Stewart, Fred Biggs and Gudrun Th6rhallsdottir who read an earlier version of this paper and gave me helpful comments and suggestions.

68 Problems in the Mediopassive Endings of Indo-European 69 terizes the primary endings of Indo-Iranian, Greek and Germanic, although in Italic and Celtic the r-endings function as both prima- ry and secondary, which is obviously an innovation. At first glance, it is extremely difficult to establish the original mediopassive endings, especially those of the 1 and 2sg. Meillet's reconstruction is 1sg. *-ai, 2sg. *-sai, 3sg. *-ai or *-tai and 3pl. *-ntai, which can hardly be reconciled with, among others, Hittite and Tocharian forms.2) It is not my intention here to give a perfectly satisfying expla- nation to this problem. Readers may find more questions than answers in this paper. But I will be immensely gratified if I suc- ceed in tempting readers to re-examine their own assumptions on the IE verbal system. Below I will attempt to place my discussion in the widest possible context of IE comparative grammar.

2. One of the most significant contributions recently made to our understanding of the IE verbal system is C. Watkins' study of the origin of the IE thematic active." He argues that Proto-Indo- European had a second series of endings characterized by 1sg. *-h2o ,2sg.*-th2o,3sg. *-e and that the resegmentation of 3sg. *-e as a suffix *-e+zero-ending brought the generalization of the thematic vowel *-e- to the 1 and 2sg. (*-e-h2o, *-e-th2o) with the subsequent change *-e->*-0- before laryngeals and resonants: sg.1.*-h2o *-h2o *-e-h2o *-o-h2o

2. *-th2o•¨*-th2o•¨*-e-th2o•¨*-e-th2o 3. *-e *-e-0 *-e-0 *-e-0

The reconstructed *-o-h2o(>*-oo>*-o) for the 1sg. explains

2) Introduction a l'etude comparative des langues indo-europeennes,232f. 3) C. Watkins. Indogermanische Grammatik III/1(1969),chapter VIII. Of course, his view has not been accepted by everybody. A strong objection to it was given by Warren Cowgill in his lecture before the Cornell Linguistics Circle on November 12, 1981 ("Thematic active endings of Indo-Hittite"). 70 Kazuhiko YOSHIDA

Gk.-ƒÖ,Lat.-o and OIr.-u.Later in most languages *-o-h2o,

*-e-th2o ,*-e were remodeled by the influence from the corres- ponding athematic active endings -m,-s,-t: sg.1.*-o-h2o *-o-m

2.*-e-th2o•¨*-e-s

3.*-e *-e-t

The endings *-o-m,*-e-s,*-e-t are well represented by daughter languages (e.g. Skt. abharam, abharah, abharat). The vowel color of -o of the 2sg.-th2o in Watkins' reconstruction is totally based on the parallelism to the 1sg. -h2o, which is, in turn, posited solely to account for *-o(<*-oh2o). The reconstruction of *-h2 for the 1sg. is, however, more preferable to *-h2o, because *-o-h2 beautifully explains the bimoric *-o(<*-oh2) of Germanic (e.g.

Goth. baira<*-5<*-oh2)4) and the acute intonation of Baltic (e. g. Lith vedil<*veduo<*vedo<*-oh2).5) Outside the third person, there is no substantial evidence for an apophonic contrast among

the thematic active, mediopassive and perfect endings, so the reconstruction of *-th2e for the 2sg. ending is supported unam-

biguously by the Greek perfect ending -ƒÆƒ¿ and the Tocharian medio-

4) Proto-Germanic final syllables preserve a contrast of acute and circumflex intonations on etymologically the same endings as Baltic and Greek, although the intonations are not anywhere directly attested. The circumflex prevents shortening of final syllables in Gothic. If the 1sg. ending were *-o-h2o, the resultant form would be Goth.-o(< trimoric *-5<*-oh2o).For details, see P. Hollifield's "The Phono- logical Development of Final Syllables in Germanic (part I and II)" in Die Sprache 26 (1980). 5) Jay Jasanoff called my attention to this Baltic evidence for the 1sg. -h2 in his Balto-Slavic class (Fall 1983, Cornell). PIE *-o goes to Lith. uo. The change, vedu<*veduo, is accounted for by two rules: "Leskien's Law" (acute vowels in final syllables are shortened) and "de Saussure's Law" (acute final syllables attract the ictus from the preceding mora if it is there). Cf. Chr. S. Stang, Vergleichende Gram- matik der Baltischen Sprachen (1966: 126f.). Problems in the Mediopassive Endings of Indo-European 7 1 passive ending A. -tar, B. -tar(

3. In Watkins' system, the mediopassive endings for the singu- lar are reconstructed as 1 sg. *-h2o, 2 sg. *-th2o and 3 sg. *-o, which differ from the active counterparts only in the vowel quality of the 3 sg. *-o. Cowgill proposes the same reconstruction in his On the Prehistory of Celtic Passive and Deponent Inflection" in Eriu 34 (1983: 75)." On the other hand, F. Kortlandt believes that the 1 sg. mediopassive ending underwent shortening from *-h2o to *-h2 in the PIE period , so he reconstructs 1 sg. *-h2, 2 sg. *-th2o

6) A possibility that the Hittite hi-conjugation preserves intact the

active characterized by the endings *-h2e, *-th2e, *-e has been proposed by J. Jasanoff, "The Position of the hi-Conjugation" in Hethitisch und Indogermanisch (1979) and "s-subjunctive, s-aorist, and the Preterite of the hi-conjugation" presented at the First East Coast Indo-European Conference held at Yale University in June, 1982. Since this is not a

place to discuss the hi-conjugation, I cannot give Jasanoff's hypothesis the sufficient discussion it deserves. I will merely say that I am not convinced at all by the so-called standard view that the hi-conjugation

is a descendant of the IE perfect. Its recent exponents are H. Eichner, Die Vorgeschichte des hethitischen Verbalsystems" in Flexion und

Wortbildung (1975), E. Risch, "Zur Entstehung des hethitischen Verbal- paradigmas" in the same volume and H. Izui, "Indo-European Perfect and Hittite Verbal System" Gengo Vol.12, No.8 (1983). The corre- spondence between Hittite hi-conjugation and IE perfect is in ending shapes only. The differences in function and lexical distribution of the two formations are such that it is practically impossible to plausibly derive the hi-conjugation with the same function as mi-conjugation from the IE perfect with the stative value.

7) In his earlier article, "More Evidence for Indo-Hittite: The Tense- Aspect System" in Proceedings of the 11th International Congress of Linguists (1975: 559), Cowgill reconstructed the singular endings as 1 sg. *-h2o or *-h2e, 2 sg. *-th2o or *-th2e and 3 sg. *-o, but the condition for the alternation *o•`*e in the 1 and 2 sg. was not specified. 7 2 Kazuhiko YOSHIDA and 3 sg. *-0.8) As mentioned above, I think that there was origi- nally no apophonic relationship among the thematic active, middle and perfect for the 1 and 2 sg. endings. Accordingly, *-h2e and

*-th2e may be reconstructed for the 1 and 2 sg ., respectively. In sum, the candidates are *-1120, *-h2e and *-h2 for the 1 sg., *-th2o and *-th2e for the 2 sg. and *-o for the 3 sg.

Scholars find agreement in the reconstruction *-o (3 sg.), which is everywhere altered to *-to under the influence of the corre- sponding active ending *-t. The original *-o is well preserved in the Hittite a-class mediopassive (e. g. da(ri) 'sits'=Skt. aste=

Gk. ƒÅƒÐƒÑƒ¿ƒÇ) and indicated by Vedic present forms of the type adye

lies'<*kei-o-i (cf. Classical Skt. sete<*kei-to-i=Hitt. kitta (ri)

=Gk. ƒÈƒÃƒÇƒÑƒ¿ƒÇ), perfect middle, e.g. jujuse (

There are not many difficulties in reconstructing the 3 pl.

*-nto on the basis of forms in daughter languages . But the more archaic 3 pl. *-ro is attested in the Vedic perfect middle, e.g. duduhre(

-at (in e .g. aduhat) is by *-o+active *-t. This *-ro itself may have been produced by attachment of the mediopassive maker *-o in the 3 sg. to the earliest 3 pl. ending in *-r, which reminds us of the 3 pl. perfect ending, e.g. Ved. -uh and the 3 pl. preter- ite ending of the Hittite hi-conjugation, -er.

In the following sections, I will discuss these reconstructed

8) Cf. "1st sg. middle *-H2" in IF 86 (1981). 9) Some scholars argue that there was an original functional difference between the o-mediopassive and the to-mediopassive (e.g. N. Oettinger, Der indogermanische Stativ." MSS 34 (1976), H. Rix, "Das keltische Verbalsytem auf dem Hintergrund des Indo-iranisch-griechischen Rekon- struktionsmodells." in Indogermanisch und Keltisch (1977), F. Kort- landt, "Toward a Reconstruction of the Balto-Slavic Verbal System" Lingua 49 (1979), but I suspect that their difference is simply of chronological nature. Problems in the Mediopassive Endings of Indo-European 7 3

endings in terms of the IE languages with inherited mediopassives.

4. Vedic must have undergone some remodeling of mediopassive endings in its prehistory. The attested forms are:

athematic thematic

primary sg. 1. -e -e 2. -se -se 3. -(t)e -(t)e

pl. 3. -nte -nte secondary sg. 1. -e 2. -thah -thah 3. -ta -ta

pl. 3. -nta -nta

Since Indo-Iranian changed PIE *o and *e into the same *a, two types of reconstruction, *-h2o, *-th2o and *h2e, *-th2e, would

cause the same outcome. The disappearance of the laryngeals must have been posterior to the change *o and *e•¨a, according to the date of "Brugmann's Law" (*o•¨a before resonants in open syl- lables. e.g. pf. 1 sg. cakara <*kuekuorh2e, 3 sg. cakara<*kuekuore).

Starting from the stage *-h2a, *-th2a, *-a, Vedic Sanskrit had the

following historical change with the influence from the active ending taken into consideration:

primary sg. 1. *-h2ai•¨*-ai•¨-e (athematic)

1. *-a-h2ai•¨*-a-ai•¨*-ai (thematic)

2. *-th2ai•¨*-thai•¨*-the•¨-se

3. *-ai•¨*-e•¨-(t)e

pl. 3. *-ntai•¨-nte

secondary sg. 1. *-h2a•¨*-a (athematic)

1. *-a-h2a•¨*-a-a•¨*-a (thematic)

2. *-th2a•¨*-tha

3. *-a•¨-ta

pl. 3. *-nta•¨-nta 7 4 Kazuhiko YOSHI DA

The 2 and 3 sg. and 3 pl. endings present no problems except the 2 sg. secondary ending -thah which is puzzling to me. As P.

Hollifield has observed,10) the formation may be interpreted as the expected ending -tha+active thematic ending -as in the same way as the anomalous 3 sg. a-duh-at and 3 pl. imperfect a-duh-ran

(old mediopassive ending *-o, *-ro+active athematic ending -t,

nt). The condition for the insertion of a thematic vowel, however, remains to be explained.11)

The situation is complicated in the case of 1 sg. Only the primary athematic ending -e is phonological. The thematic ending expected for the 1 sg. primary, *-ai, takes the same form as the subjunctive -ai(<*-a-a-h2ai). For this reason Cowgill proposes that the thematic verbs use the pattern of the athematic ones to

differentiate the indicative and subjunctive moods.12) He also argues that the secondary endings -i (athematic) and -e (thematic) must

both be analogical for *-a (<*-h2-a) and *-d(<*-a-h2-a). The source for -i is a pattern -mahe (1 pl. primary): -e (1 sg. primary)

=-mahi (1 pl . secondary): X (1 sg. secondary), X=(cf. W.

Peterson, Lg. 12 (1936: 162)). The thematic secondary ending -e

can be explained as the addition of the athematic secondary ending,

*-i , to the expected *-a (*-a+athematic *-i•¨-e).

What picture of the development is given if we start from *-h2

(1 sg.)?:

primary *-h2-i (athematic) *-a-h2-i•¨*-a-i•¨-e (thematic)

secondary *-h2•¨-i (athematic)

*-a-h2•¨*-d (thematic)

10) "The Personal Endings of the Celtic Imperfect" KZ 92 (1978: 219). 11) H. Rix gives another alleged explanation in his Historische Grammatik des Griechischen (1976: 247), which I find ad hoc. 12) W. Cowgill, "The First Person Singular Medio-Passive of Indo- Iranian" in Pratidanam (Studies Presented to F.B.J. Kuiper) 1968. Problems in the Mediopassive Endings of Indo-European 7 5

This time the thematic primary and athematic secondary endings, e and -i, are phonologically regular, but the athematic primary *-i and the thematic secondary *-a underwent innovations. The remodeling of the athematic primary *-i to -e of the thematic primary is motivated by homonymy between the athematic primary and secondary endings. As for the thematic secondary ending -e, the same explanation as above may be given (*-a+athematic *-i -e) . True, we sometimes need a number of analogical changes to account for the facts of various IE languages. But we must not forget that when analogy is postulated it is preferable to find not only the proportion by which it worked, but also some motivation for it. In this sense Cowgill's proportion to derive the athematic secondary ending -i is not well motivated in my judgement. The same -i can be derived phonologically by positing *-hs. In addition, the assumption of the 1 sg. *-h2 necessitates less analogy than the other two views, *-h2o and *-h2e, so that it is not unreasonable to reconstruct *-h2 for the 1 sg. rather than *-h2o or *-h2e in the case of Vedic Sanskrit.

5. The reconstructed Greek mediopassive endings were *-ƒÊƒ¿ƒÇ,

*-ƒÐƒ¿ƒÇ, *-ƒÑƒ¿ƒÇ , *-ƒËƒÑƒ¿ƒÇ until 1952, when M.S. Ruiperez proposed *-ƒÊƒ¿ƒÇ, *-ƒÐƒÍƒÇ , *-ƒÑƒÍƒÇ,*-ƒÑƒÍƒÇ on the evidence of Arcado-Cypriote -ƒÊƒ¿ƒÇ, -ƒÐƒÍƒÇ,-ƒÑƒÍ,

. The reconstruction by Ruiperez is also compatible with the secondary endings -ƒÊƒ¿ƒË, -ƒÐƒÍ, -ƒÑƒÍ, -ƒËƒÑƒÍ except for the 1 sg. -ƒÊƒ¿ƒË.

Considering the influence from the corresponding active endings, the reconstruction of *-h2o, *-thso, *-o, *-nto explains Gk. *-ƒÊƒ¿ƒÇ,

*-ƒÐƒÍƒÇ , *-ƒÑƒÍƒÇ, *-ƒËƒÑƒÍƒÇ without difficulties except the 1 sg. *-ƒÊƒ¿ƒÇ, where

cannot have affected the timbre of a neighboring *o. It is*h2 also impossible to derive *ƒÊƒ¿ƒÇ, *ƒÐƒÍƒÇ, *ƒÑƒÍƒÇ, *ƒËƒÑƒÍƒÇ directly from *-h2e

13) M.S. Ruiperez, "Desinendias Medias Primarias Indoeuropeas sg. 1. *-(m)ai , 2. *-sol, 3. *-(t)oi, pl. 3. *—ntoi" in Emerita 26 (1952). 7 6 Kazuhiko YOSHIDA

*-th2e , *-o, *-nto, but if we suppose that the vowel color of the

3 sg. *-o spread to the 1 and 2 sg. endings, replacing *e of *-h2e and *-th2e, the situation becomes the same as the above case with the proto-forms *-h2o, *-th2o, *-o, *-nto. The analogical spread from the 3 sg. to other positions of the paradigm is typologically quite unremarkable.

G. Klingenschmitt, via Hollifield (1978 : 219), pointed out that

Gk. secondary 1 sg. -ƒÊƒ¿ƒË can have arisen, in a parallel way to Skt.

2 sg. secondary -thah (<*-th2e+es), from *-h2e+om with secon- dary insertion of m. But, since Greek historical phonology regularly produces -ƒÊƒ¿ƒË, not -ƒÊƒ¿ƒË from the contraction of ƒÊƒ¿-ƒÍƒË, his expla- nation is inadequate.

In my opinion, the only way to explain the difference in the vowel length of the primary -ƒÊƒ¿ƒË and the secondary -ƒÊƒ¿ƒË phono- logically is to posit *-mh2 for the 1 sg., which yields *-ma after a vowel and *-ma after a consonant. The primary -ƒÊƒ¿ƒÇ arose with attachment of the hic et nunc *-i to *-ma and the secondary -ƒÊƒ¿ƒË with addition of the 1 sg. athematic active ending *-m to *-ma.

The dialects other than Arcado-Cypriote have -ƒÊƒ¿ƒÇ, -ƒÂƒ¿ƒÇ, -ƒÑƒ¿ and -ƒËƒÑƒ¿ƒÇ. The persistent a found there is usually taken as ex- tension of a in the 1 sg. -ƒÊƒ¿ƒÇ. But as Cowgill and Neu have pointed out,14) it is not very likely that the 1 sg. *-ƒÊƒ¿ƒÇ alone would have been able to influence analogically the 2 sg. *-ƒÂƒÍƒÇ, 3 sg. *-ƒÑƒÍƒÇ and

3 pl. *-ƒËƒÑƒÍƒÇ.15) There may be some additional motivation which has

14) W. Cowgill, " Dialectology in the Light of Mycenaean" in Ancient Indo-European Dialects (1966) and E. Neu, "Die indo- germanischen primAren Medialendungen *-(m)ai, *-sai, *-(t)ai,*-ntai" in IF 73 (1968).

15) This analogical extension violates Kurylowicz' second analogical law: les actions dites "analogiques" suivent la direction: formes de fondation•¨formes fondees, dont le rapport decoule de leurs spheres d'emploi ("La nature des proces dits 'analogiques'" in Esquisses linguistiques 1960). The critical position of the third singular in the paradigm as a foundation for analogical spread is cogently discussed in Watkins' Indo-European Origins of the Celtic Verb (1962: 99ff.). Problems in the Mediopassive Endings of Indo-European 77 not been discovered yet.

6. Latin mediopassive endings characterized by the attachment of the r-element except in the second person, -or, -re/-ris, -tur, -ntur can be explained by any type of reconstructed endings as is illustrated below:

sg. 1. * -o-h2or•¨* -or•¨-or

2. * -th2o•¨* -to•¨* -so•¨* -ro•¨-re

3. * -or•¨* -tor•¨-tur

pl. 3. * -ntor•¨-ntur

sg. 1. * -o-h2er•¨* -o-h2or•¨* -or•¨-or

2. * -th2e•¨* -th2o•¨* -so•¨ -re

3. * -or•¨* -or •¨* -tor •¨ -tur

sg. 1. * -o-h2r•¨* -or•¨ or

Note that in the case of the reconstructed endings *-h2e and *-th2e

the vowel timbre of *-h2er and *-th2e changed into *o under the

influence of the 3 sg. *-or and that the r of the 2 sg. *-re is a

result of rhotacism (*s•¨*z•¨r/V-V). The 2 sg. -ris side by side

with -re is a new form enlarged by the 2 sg. athematic active

ending -s.16)

7. The morphology of Old Irish is extremely complicated.

Playing an important role are, among others, palatalization and

lenition of consonants, which are basically conditioned by the follow-

ing front vowels and the preceding vowels, respectively. The

deponent and passive endings of the absolute inflection are:17)

16) Cf. A. Meillet et J. Vendryes, Traite de grammaire comparde des langues classiques (1968: 351). 17) Traditionally, absolute endings are supposed to come from PIE prima- ry endings and conjunct from secondary. But a very convincing alterna- tive theory has been proposed by Cowgill, who takes both from primary, with an added particle in the absolute forms ("The Origins of the Insular Celtic Conjunct and Absolute Verbal Endings" in Flexion und Wortbildung 1975). 7 8 Kazuhiko YOSHIDA

sg. 1. -ur 2. -t(h)er 3. -thir pass. -th(a)ir pl. 3. -tir pass. -t(a)ir The palatalization of final -r in the 3 sg. and 3 pl. deponent and passive endings suggests that the most likely reconstructed endings are in -ri: -thir<*-Vtri, -th(a)ir<*-Vtori, -tir<*-ntri, t(a)ir<*-ntori. The alternation between the passive *-(n)tori and the third person deponent *-(n)tri is fully discussed by Jasanoff.18) The 2 sg. has an enigmatic ending -t(h)er. There is a fundamental difficulty in the vowel timbre. PIE *-th2o and *-thle ought to develop into *-t(h)o and *-t(h)a respectively in Old Irish. Thur- neysen suggests that the unextended -the connected with the 2 sg. imperative may go back to *-thes and thus correspond to the Skt. secondary -thah,19) which itself is also hard to understand as was mentioned above. The 1 sg. ending is explained in the same manner as in Latin, starting from any type of reconstruction (i.e. *-h2o, *-h2e, *-h2).

18) "The r-endings of the IE Middle" Die Sprache 23 (1977: 162ff.). The ending *-(n)tori is, in all probability, original. G. Schmidt de- rived Celtic *-tri and *-ntri from PIE *-tr and *-ntr, which he thinks are correct endings reconstructed for PIE, instead of usual *-tor and *-ntor ("Das Medium im vorhistorischen Keltisch" in Indogermanisch and Keltisch (1977: 105ff.)). But PIE *-(n)tr would have given *-(n) tur rather than actually attested -(n)ter in Osco-Umbrian. Some Irish verbs take a passive endings, -air (e.g. berair 'be carried'). Kuryiowicz ("La desinence verbale -r en indo-europeen et en celtique" Etudes celtiques 12 (1968-69)) and Cowgill (1983: 102) assume that this is not a lineal descendant of PIE *-ori, but a creation of Celtic itself. I have a hunch that this problem should remain unsolved rather than be considered to have been explained. 19) R. Thurneysen, A Grammar of Old Irish (1946: 366). Problems in the Mediopassive Endings of Indo-European 79

8. So far, no one has offered an exhaustive description of

Tocharian phonology. The discussion in this section is based on

Jasanoff's view presented to me in his Tocharian class (Spring 1984, Cornell), which seems to best account for the diachronic and synchronic facts of Tocharian. It is suggested that vowel length was not phonologically distinctive in Toch. A and B and that the

"Fremdvokal" a is treated as the central high vowel [i] . Pre-

Tocharian had the following vowel system (with the conditions for the secondary changes *u•¨*a and *o•¨*a unspecified here for simplicity of argument):

*i(<*1) *a(<*e; *u) *u(<*u) *e(<*e) *o(<*o) *a(<*a , *a(<*a, *e), *o;*o) Pre-Toch. *e and *o basically fall together as Common Toch. [e], but *e as well as *a goes to o under the influence of a neighboring rounded vowel or labialized consonant and pre-Toch. *o survived in Common Tocharian under the same condition. Accordingly, the Common Tocharian vowel system is: *i [i] *a [i] *u [u]

*a[e] *o [o] *a[a]

The Common Toch. vowels are retained in Toch. B, where the only subsequent change was the introduction of a[^] as the stressed variant of a and the unstressed variant of a. On the other hand, Common Toch. *-a is centralized to a [^] in Toch. A (a new e being brought by the monophthongization of *ai). Toch. A also underwent secondary rounding and unrounding assimilation (Com- mon Toch. *a>A o, Common Toch. *o>A a). So Toch. A and B have synchronically the same vowel system (Common Toch. a is designated by e in Toch. B): 80 Kazuhiko YOSHIDA

i [i] a [i] u [u] e [e] a [^] o [o] a [a]

In the light of the above brief description of Tocharian pho- nology we may analyze the mediopassive endings of Toch. A [B] -mar [-mar] , -tar [-tar], -tar [-tad -ntar [-ntar] as follows. The original forms in Common Tocharian are obviously *-mar, *-tar, *-tar , *-ntar respectively, which, in turn, directly go back to pre-Tocharian. The pre-Tocharian 3 sg. *-tar and 3 pl. *-ntar should be interpreted as coming from PIE *-tri and *-ntri with the epenthetic a,20) both of which are inherited in Old Irish as the 3 sg. and 3 pl. deponent endings, *-thir and -tir, respectively. The vowel a in pre-Toch. 1 sg. *-mar and 2 sg. *-tar can be a reflex of PIE *a, *a, *a, *o or *o, as is shown above. Among these five possi- bilities, *o must be excluded because lowering of PIE *o to pre- Toch. *a occurs before a in a following syllable. This means that the reconstruction of *-h2o, *-th2o as PIE 1 and 2 sg. mediopassive endings is inadequate because *-h2o (>*-o) and *-th2o (>*-to) are not realized as Pre-Toch. -mar and -tar by any means. The reconstruction of 1 sg. *-h2(e) and 2 sg. *-th2e, however, explains pre-Toch. *-mar and *-tar without any difficulties (PIE *-h2e, *-h2, *-th2e>post-IE *-a, *-a, *-ta). Post-IE *a regularly yielded *a in pre-Tocharian (of course , m of *-mar was transferred from the active). Tocharian clearly demonstrates that the reconstruction *-h2o, *-th2o, *-o can only be treated as inadequate.

9. The essential features of Proto-Indo-European mediopassive endings are retained in Hittite:

20) The only final vowels that disappeared in Common Tocharian are *-u and *-a (<*-e) . Of these, only *-i is compatible with Old*-i Irish r-endings. Problems in the Mediopassive Endings of Indo-European 81

sg. 1. -ha(ri) 2. -ta(ri) (-ti) 3. -(t)a(ri) pl. 3. -nta (ri)

Synchronic and diachronic patterns of the distribution of the seemingly capricious element -ri will be exhaustively discussed in my forthcoming Cornell University dissertation, where I will show that earlier speculations on the genesis of -ri made by Neu, Watkins, Cowgill, Eichner and Oettinger are wrong in assuming that ri- and ri-less forms are Hittite innovations.21) Instead, I will argue that *-ri, not *-r reconstructed by most scholars, was al- ready optionally attached to unextended mediopassive forms in the Proto-Indo-European period and that Old Hittite essentially retained this PIE situation in the form of -(t)a and -(t)ari. Since IE *o and *a merged into a in Hittite, any type of reconstruction can perfectly account for the Hittite mediopassive endings. The only modification inside Hittite itself is the a of the 1sg. -ha, which is analogically transferred from the other positions of the paradigm.

21) Cf. E. Neu, Das hethitische Mediopassiv and seine indogermanischen Grundlagen (=StBoT 6) (1968: 140 ff.) and "Hethitisch /r/ im Wortauslaut" in Festschrift fur G. Neumann (1982: 222ff.), C. Wat- kins, Indogermanische Grammatik III/1 (1969: 78, 174), W. Cowgill, "More Evidence for Indo -Hittite" in Proc . XI Int. Cong. Ling. (1975: 561), H. Eichner, "Die Vorgeschichte des hethitischen Verbalsystems" in Flexion and Wortbildung (1975: 76), N. Oettinger, Die Stammbildung des hethitischen Verbums (1979: 520ff.). All these scholars ignore the various periods at which our Hittite documents are attested. Without the distinction of successive synchronic stages of the the true development of any morphological category cannot be under- stood. As H. Craig Melchert says in his recent book, Studies in Hittite Historical Phonology (Vandenhoeck & Ruprecht 1984: 79), "the crucial role of the relative chronology of sources in determining the features of Hittite grammar should no longer be open to question." Dating criteria are fully discussed in chapter I of the same author's Harvard University dissertation, Ablative and Instrumental in Hittite (1977). 8 2 Kazuhiko YOSHIDA

This modification is necessary only if we reconstruct *-h2(>-h) for the 1sg.

10. Germanic is not an Indo-European dialect group ex- ceedingly notable for archaism in its verbal system. Yet surprisingly archaic Indo-European aspects are still retained there. Below is given the paradigm of the Gothic inflectional passive. As a citation form, haitan 'call' is used here:

sg. 1. haitada

2. haitaza

3. haitada

pl. 1, 2, 3. haitanda

Although Gothic lacks the original 1sg. form, it is unmistaka- bly attested in runic inscriptions: haite 'I am called' (=OIc. heite,

OE. hatte), e.g. in the Kragehul spearshaft (cf. W. Krause, Die

Sprache der urnordischen Runeninschriften 1971: 152). Judging from the Gothic forms haitaza, haitada and haitanda, runic haite obviously includes the thematic vowel, PIE *-o-, but unfortunately our data of runic inscriptions are too fragmentary to determine the original form of its ending.

The 2 and 3sg. and 3pl. forms in Gothic developed regularly

from the reconstruction of *-th2oi or *-th2ei, *-oi, *-ntoi (>*-tai,

*-ai , *-ntai>*-sai, *-tai, *-ntai) through "Grimm's Law" and

Verner's Law" (s, t•¨s, p•¨z, d) and Germanic final syllable

reduction rule: Gmc. bimoric final *-ai•¨ Goth. -a. The origin of

the 1 and 2pl. forms identical to the 3pl. form is not very clear,

although the 1pl. ending might be explained by the vowel re-

duction caused by syncope (*-omedhai>*-amida>-anda).

Germanic handbooks show general agreement on two points:

the thematic vowel -a- in the singular (2sg. -a-za, 1 and 3sg.

a-da) was transferred from the 3pl. -a-nda (cf. Gk. 2sg. Problems in the Mediopassive Endings of Indo-European 8 3

*φ ερ-ε-σαι>*φ ερ-ε-α ι>φ ερ-η , 3sg. φερ-ε-ται, 3pl. φερ-o-ν ται)22) and the 3sg. form, haitada, was extended to the 1sg.23) Of these traditionally admitted analogical explanations, however, the former seems to me now less convincing than it used to be, in the light of the recent progress of Indo-European studies. The second point is probably explained by the homophony of the 1sg. and 3sg. in the preterite (-da), which spread from there to the passive para- digm, although an earlier identity of 1sg. *-ai(<*-h2ei) and 3sg. *-ai(<*-oi) could be possibly relevant here . The first point, the analogical spread of the thematic vowel -a-(<*-o-) from the 3pl. to the 1, 2, 3sg., is highly dubious. This hypothesis fails to explain why the corresponding active haitis (2sg.) and haitip (3sg.) were not remodeled under the influence of haita (1sg.) and haitand (3 pl.). Liihr's idea, 'Vokalharmonisierung' of *-i-(<*-e-) with the following a of the ending, is hardly persuasive, either.24) The per- sistent thematic vowel *-o- parallel to the Gothic passive is also found in Hittite thematic mediopassives and the Tocharian presents of Krause-Thomas' class III and class IV (Toch. A -a- [B-e-]< Common Toch. <-a-

22) H. Krahe, Historische Laut- und Formenlehre des Gotischen (1967: 131) and Germanische Sprachwissenschaft II (1969:115), W. Krause, Handbuch des Gotischen (1968: 262), E. Prokosch, A Comparative Germanic Grammar (1939: 219). 23) W. Streitberg, Urgermanische Grammatik (1896: 262), W. Krause. op. cit. p.262, J. Wright, Grammar of the Gothic Language (1954: 140). 24) R. Luhr, "Die Kontinuante der urindogermanischen Medialflexion im Germanischen." MSS 37 (1978). 8 4 Kazuhiko YOSHIDA passives for languages like Greek (*-o-ƒÊƒ¿ƒÇ, *-ƒÃ-ƒÐƒ¿ƒÇ, *-ƒÃ-ƒÑƒ¿ƒÇ) is, in all probability, due to the pattern in the active (*-o-h2, *-e-th2e,

*-e) . This remodeling did not occur in Gothic passives, Hittite thematic mediopassives and Tocharian class III and IV verbs. It

is suggestive that Tocharian class III and IV lack the active counterparts.25)

11. Although the discussion in this paper has been rather speculative, it will certainly shed a new light on our understanding of the IE verbal system.

As already discussed above, the reconstruction *-h2o for the 1 sg. brings difficulties to Gk. -ƒÊƒ¿ƒÇ,-ƒÊƒ¿ƒË and Toch. A[B]-mar[-mar].

Gk. -ƒÊƒ¿ƒË cannot be accounted for by *-h2e, either. On the other hand, the 1sg. *-h2, with *-m from the athematic active, can derive Gk. -ƒÊƒ¿ƒÇ and -ƒÊƒ¿ƒÇ phonologically from -Vmh2-i and -Cmh2

m, respectively. Furthermore, in the case of Vedic Sanskrit, *-h2o and *-h2e have to assume a great deal of remodeling in cases where

*-h2 does not need any analogy. Only Hittite causes a problem with positing the 1sg. -h2. Of course, the Hittite 1sg. -ha(ri) can be derived from -h2 with the spread of a from the other endings as Ruiperez (op. cit.) did, but I would like to think otherwise. Old

Hittite has an archaic 1sg. ending of the hi-conjugation, -he,

which is from *-h2e-i (>*-hai>-he). This ending unmistakably

shows that Hittite has not undergone the reduction of *-h2e to

*-h2 , which universally occurred to the 1sg. thematic active ending

25) Another possibility that Anatolian, Germanic and Tocharian extended o-timbre in the post-IE period, but at a time when the 3sg. ending was still *-o has also been pointed out by Jasanoff in chapter II of his Stative and Middle in Indo-European (1978). He also suggests that the IE mediopassive is retained in the Germanic third weak class disguised by the 'activization' ("The Germanic Third Weak Class" Lg. 49 (1973) and chapter III of his book referred to above.). Problems in the Mediopassive Endings of Indo-European 85 in the non-Anatolian languages. An attractive inference to be obviously drawn from these facts is that the vowel reduction of *-h2e to *-h2 occurred both to the thematic active and mediopassive throughout the non-Anatolian languages, while Hittite preserved the original shape of the 1sg. ending both in the hi-conjugation and mediopassive. I hardly believe that this assumption alone will contribute to Cowgill's famous 'Indo-Hittite hypothesis'. For the 2 sg. mediopassive ending, *-th2e ought to be reconstructed rather than *-th2o because *-th2o cannot possibly be reconciled with the Toch. A[B]-tar[-tar]. The overall discussion in this paper leads us to assume that the mediopassive endings most adequately reconstructed for the earliest stage of PIE are 1sg. *-h2e, 2sg. *-th2e and 3sg. *-o, which pattern with the corresponding earliest active endings before the reduction of the 1sg. *-h2e to *-h2, 1sg. *-h2e, 2sg. *-thse and 3sg. *-e. An apophonic contrast between the thematic active and mediopassive endings is found only in the 3sg.

(原稿受 理 日 昭 和59年12月1日)