The prehistory of the h2e-conjugation

The prehistory of the h2e-conjugation Miguel Carrasquer Vidal, January 2016 In this paper, I will present a number of hypotheses about the prehistory of the Proto-Indo-European verb, mostly dealing with the origins and development of the PIE h2e-conjugation. Hypothesis 0: PIE had a mi-conjugation and a h2e-conjugation Although the Hittite verbal system is in general much simpler than the system that had been reconstructed for PIE largely on the basis of Greek and Vedic, one unexpected complication was the existence in Hittite (and Anatolian in general) of two separate conjugations: the mi-conjugation, featuring a set of endings largely compatible with the extra-Anatolian material, and the hiconjugation, with a different set of endings (at least in the singular), clearly related to the endings of the perfect and/or the middle, but making active forms, in both present and preterit. There are two possibilities to explain the Hittite facts: (1) the hi-conjugation was an Anatolian innovation that developed out of the PIE perfect, or perhaps the PIE middle, or (2), the opposition between the two conjugations, the mi- and the hi-conjugation, was a feature of PIE itself, with the latter subsequently lost in the extra-Anatolian branches, leaving only relics such as the perfect. As argued by Jasanoff 2003, option (1) poses more problems than it solves. We cannot derive the hiconjugation from the middle, because Hittite already has a middle, which can be connected to the extra-Anatolian middle forms by a number of exact etymologies. If we try to derive the hiconjugation from the perfect, we encounter the opposite problem: there are not enough convincing etymologies linking the Anatolian hi-conjugation verbs to the extra-Anatolian perfects. For instance, if we take a look at LIV2, the e a e o l A atolia fo s listed u de the atego Pe fekt 1. Of these, only 8 correspond to extra-Anatolian perfects: *dwei-, *h1er-, *h2edhgh-, *men-, *neiH-, *spend-, *steh2- and *wers-, and most of these equations are doubtful for several reasons. The best cases can in my opinion be made for *h1/3er- and *neiH-, but it should be clear that such a distribution is more in line with the hypothesis that the perfect is one of the remnants of an earlier hi-conjugation, than that it fits the hypothesis that the Anatolian hi-conjugation sprang from the perfect. In that case, we would expect a much more solid core of convincing and basic equations between the Hittite hi-verbs and the extra-Anatolian perfects. In my opinion, then, we need to assume two different active conjugations in PIE, for which I will use Jasanoff s terms: mi-conjugation vs. h2e-conjugation2. 1 Interestingly, more than half of these (12) are from roots beginning with a laryngeal, and more than half of the rest (6) are from roots that end in a laryngeal. 2 Rather than the Hittite-centric mi- and hi-conjugations. I will not bother to introduce new terminology, say, mconjugation vs. χ-conjugation. Hypothesis 1: The root aorist was originally h2e-conjugated Although the h2e-conjugation was still very much alive in Hittite, I think there are reasons to believe that it had already lost ground to the mi-conjugation even in Anatolian. There is a significant body of evidence suggesting that the PIE root aorist (as exemplified by Hitt. tē u I spoke , a d Ved. ádhā I put , oth f o the ao isti oot *dheh1-), was originally h2e-conjugated. 1sg We have evidence for 1sg *-h2a in Greek ἔθηκα I put *dheh1-h2a), ἔ ωκα (*deh3-h2a I ga e , ἧκα (*Hieh1-h2a), Latin fē ī (*dheh1-h2a-i), iē ī (*Hieh1-h2a-i), Phrygian addaket (ultimately from *dheh1h2a), and Tocharian tākā (*(s)teh2-h2a). The origin of these k-aorists has long been a source of puzzlement, but becomes understandable if the 1sg ending of the root aorist was once *-h2(e), which was almost universally replaced by *-m (both in and outside of Anatolian), except in the special case of roots ending in a laryngeal, where, in some dialects, *-Hh2- > *-k-. It is evident that the k-aorist is not copied from the k-perfect: where the possibility exists (in Greek), it makes no sense at all within the verbal system, and where it would make some kind of sense, as in e.g. Latin, the k-perfect does not exist. What the k-perfect and the k-aorist have in common is that they both have their origin in a sound law affecting the h2e-conjugation 1sg ending *-h2 when directly preceded by a laryngeal, as originally proposed (for the k-perfect) by Sturtevant 1940. 3sg As recently demonstrated in Pooth 2015 and Kümmel 2015, we have considerable evidence for a 3sg ending *-s in the root aorist. The case is clearest in the aorist optative, where the 3sg ending in the Rigveda is exclusively -yās (*-jeh1-s . This sg fo late e a e the asis o hi h the p e ati e was built, reinterpreting the ending -yās as an aorist optative suffix. But even outside of the optative, there are clear traces of 3sg *-s in the root aorist in Vedic, Avestan and Old Persian, as for instance the 3sg injunctive dhās in HirGS 1,13,15, ĀpMB 2,10,17 tan ma ūrjaṃ dhāḥ Das e s haffe i K aft! . 1/2pl The a d pl of the oot ao ist, as dis ussed i Malzah *dheh1-) and G eek I e t < *gweh2-): a-dhā-m a-dhā-s a-dhā-t a-dhā-ma a-dhā-ta a-dh-ur ἔ-βηἔ-βη-ς ἔ-βη ἔ-βηἔ-βη-τ ἔ-β-α , sho s : A laut i Vedi I put < The Ablaut cannot derive from anything in the mi-conjugation, where the endings are always asyllabic in the singular (*-m, *-s, *-t), and syllabic in the plural (*-més/*-wén, *-té(n), *-ént), abstraction made of the particle *-i, which has no effect on the accentuation or the Ablaut. Malzahn suggests influence from the 2pl imperative, which is possible, but it should be noted that the se o da h2e-conjugation endings, on which later, are precisely characterized by being asyllabic in the singular and the 1/2pl (*-h2, *-th2, *-s, *-m, *-(s)), and syllabic only in the 3pl (*-ér-s). 3pl Finally, the root aorist in Vedic, as can be seen from the 3pl form adhur above, is also characterized by the 3pl ending -úr (from *-r̥ s, with the accent of *-ér-s > *-́r), an ending that surely belongs to the h2e-conjugation: the 3pl ending of the perfect is, depending on dialect, *-r̥ , *-r̥ s or *-́r, and the iddle stati e has *-r-o-. In summary, everything seems to indicate that the root aorist was formerly conjugated with *h2econjugation endings, and that it acquired the familiar mi-conjugation endings later, albeit before the split-off of Anatolian. If so, the diffe e e et ee ao isti a d p ese ti e al oots i.e. oots ha i g a u a ked root aorist and roots having an unmarked root present) was not based on imprecise semantic factors su h as fo i sta e teli o atelic oots. Instead, there was an exact morphological distinction: aoristic roots were aoristic because they belonged to the h2e-conjugation, whereas presentic roots belonged to the mi-conjugation. Hypothesis 2: Verbal categories with Vedic 3pl -úr were originally h2e-conjugated Apart from the root aorist, there are a number of other verbal categories that in Vedic have -úr as the 3pl preterit ending. We will assume that Vedic -úr is the a k of the h2e-conjugation, and that all of these forms once belonged in that conjugation. Apart from, obviously, the perfect, this includes the optative, the edupli ated p ese t , the reduplicated aorist, and the s-aorist. The optative need not concern us here. That the sigmatic aorist is connected to the h2e-conjugation should come as no surprise. As argued at length in Jasanoff 2003, the origin of the whole formation is likely to be the 3sg h2e- o jugatio se o da e di g *-s, reinterpreted in a way similar to the way in which, as sketched above, the 3sg optative -yās led to the atego of the p e ati e i Vedi . I a ase, the s-ao ist e e tuall p o ided a ea s fo p ese ti oots to e p ess a ao ist, i su h languages as had introduced the aspectual difference between the present and the aorist (as well as the perfect). The reduplicated aorist (preserved as a separate verbal category mainly in Vedic) normally has thematic endings, but there is a small number of athematic forms. These have a 3pl in -úr, which ould suggest that the edupli ated o ausati e ) aorist also once was h2e-conjugated. As we will see, the presence of reduplication itself, and the thematization of the verbal category are also arguments in favour of this. Perhaps the most surprising item yielded by setting up -úr as a litmus-test for membership of the h2econjugation is the so- alled edupli ated p ese t . Although it is e tai l ade al ost e lusi el f o ao isti oots, i I do-Iranian (where the category is most extended) it routinely serves to provide imperfective forms (present and imperfect) to root aorists. For instance, the root aorist adhā makes a present didhā i. As su h, it fu tio s i a a si ila to othe ha a te ized p ese ts , su h as *-nú, *-jé or plain thematics such as bhárati. In fact, both bhárati and bibhárti o u side side i Vedi as p ese ts ith the ea i g a ies, i gs . Outside of I do-Iranian, the category of reduplicated presents is not common, and only a few verbs show reflexes in more than two branches of Indo-European: *dhi-dhéh1- to put, ake , *di-déh3- to gi e , *́i-́é h1- to gi e i th, e o (middle), *sti-stéh2- to sta d up , a d the the ati s *pi-ph3-é- to d i k a d *si-sd-é- to sit do . It is st iki g that ost of these fo s e d i a la geal. If e assu e that outside of Indo-Iranian, the reduplicated presents all became simple thematics of the bhárati-type based on their h2e- o jugatio sg p i a e di g *-e, the fact that this failed to happen in roots of the structure *CEH- is obviously due to the contraction of 3sg *CEH-e to *C̃ e.g. -dhéh1-e > -dh̃). This no longer sounded like a thematic form lacking *-t(i), and consequently, the form was never thematized. Note that this implies a pre-existing category of thematic verbs to which the thematized bibharti-verbs were assimilated. This category was the subjunctive, which must be reconstructed for PIE i the se se of I do-Hittite . A atolia lost the su ju ti e as a atego , ut etai ed the *o ~ *e interplay of the thematic vowel that originated in the subjunctive. In any case, the t a sfe of the edupli ated p ese t from the realm of the perfective (i.e. the aorist) to the realm of the imperfective is a significant fact. We will see that the inverse occurred in the case of the perfect, which is likely to have started out as an imperfective or durative formation, made f o ao isti i.e. h2e-conjugation) verbs, but which ended up as a preterit category in e.g. Latin or Germanic (apart from the praeterito-praesentia). Hypothesis 3: In the h2e-conjugation, C1i- reduplication denoted the plural of the absolutive If the reduplicated present and the reduplicated aorist belong together with the unreduplicated root aorist, we may ask ourselves what the function of the reduplication was. In the case of the reduplicated present, it stands to reason to think that the reduplication had something to do with imperfectiveness (iterative, durative, etc.). As stated above, bibhárti-formations provide imperfectives to root aorists, so there should be little doubt that this is the case here. In the case of the reduplicated aorist, the formation usually has causative semantics, but that is hardly a problem given the connection between causatives and iteratives that we so often find in IE. We have the ausati e-ite ati es of the t pe *mon-éie (LIV2 4a), or the suffixes *-nu-, which is causative in Hittite, presentic elsewhere, and *-śe-, iterative in Hittite, causative in Tocharian. Nevertheless, I am going to deny that there was originally any connection between imperfectiveness and reduplication in the PIE h2e-conjugation. Instead, the hypothesis is that reduplication is related to the number of the intransitive subject and the transitive object. What this buys us , I will attempt to demonstrate in the following. In the first place, it yields three paradigms, to be filled in by the three categories mentioned above: the reduplicated present, the root aorist and the reduplicated aorist. We can reconstruct them as follows: sg. pl. intr. dhéh1dhí-dhh1- tr.sg. dhéh1dhéh1- tr.pl. dhí-dhh1dhí-dhh1- The endings were originally h2e-conjugation endings, which (except for the 3pl) were later replaced by mi-conjugation endings. Before we can elaborate further on the developments leading to the attested reduplicated present, root aorist, and reduplicated aorist, we need to take a closer look at the h2e-conjugation endings. Hypothesis 4: The h2e-co jugatio had pri ary a d seco dary e di gs The p i a 1 2 3 1 2 3 e di gs of the h2e-conjugation are basically the well-known endings of the perfect: *-h2e *-th2e *-e *-me *-(s)e *-r(e) The first four present no problems. In Vedic, the 2pl ending is -á, which must be original compared to the usual *-té. As argued by Kloekhorst 2007, the Hittite preterit ending in the hi-conjugation was originally -sten, possibly from *-s(V) + the usual mi-conjugation ending -ten. An ending *-sV is also suggested by the Tocharian 2pl preterit *-sä3. Based on Tocharian, Kloekhorst suggests a reconstruction *-su, but I think *-se (with depalatalization of regular *-s’ä) makes more sense. However, *-se (or *-su) is incompatible with Vedic -a. For a solution, we must turn to the middle 1/2pl endings. Prima facie, that does t sound like a good idea: the 1/2pl middle endings have always raised more questions than they have answered, but consider the following: the 1/2pl middle endings both seem to contain an element *dh(w): 1pl *-mVdh- and 2pl *-dhwV-. If we eliminate this element (a sound law can take care of that in the Auslaut), what we are left with are 1pl *-m and 2pl *-Ø, exactly the perfect endings (before the addition of *-e) as suggested by Vedic. The Tocharian and Hittite forms are then abstracted from roots ending in a dental, where *-Tdhw > *-Tsdhw. When *-dhw was lost, the synchronic ending *-s of the T-stems was analogically extended to all h2e-conjugation active forms. That this only happened in the h2e-conjugation, and not in the mi-conjugation, is of course due to the inherent awkwardness of a zero 2pl ending. The 3pl form is nowhere attested as expected *-re, except perhaps in Tocharian (*-rä). Avestan has *-r. Elsewhere, Vedic -úr and Latin -ēre have been introduced from the secondary endings. The element *-e, like the element *-i in the mi-conjugation, had no effect on the Ablaut or the stress. The h2e-conjugation primary endings therefore suggest an original 6:0 Ablaut. This was secondarily reshaped into 3:2:1 by a stress-shift to the 1/2pl endings *-mé, *-(t)é. The 3pl did not become *-ré, but, as was said above, the end-st essed e di g of the se o da e di gs as sometimes adopted, resulting in a mi-conjugation style 3:3 Ablaut. I e o st u t the se o da 1 2 3 1 2 3 e di gs as follo s: *-h2 *-th2 *-s *-m *-(s) *-ér-s But Latin -istis in e.g. 2pl a ā istis is rather a reduced periphrastic form *a āusi estes, with the 2pl of the copula added to the locative of the perfect participle. See Carrasquer Vidal 2013. 3 That the 1sg ending was *-h2 is primarily shown by the Hittite preterit ending -hhun, from *-h2 + - ̥ . We would have expected *-hhan if the ending had been *-h2a + -m. The third person ending *-s is also never followed by *-e (Hittite preterit -s, Indo-I a ia p e ati e -s, sigmatic aorist -s-), which indeed suggests that the se o da e di gs la ked *-e. The third person endings are characterized by an element *-s, added to the plural ending *-ér (< *-én) in the 3pl. The third person endings (*-s, *-ér-s) are then completely parallel to the third person endings of the mi-conjugation (*-t, *-én-t), except that the third person ending *-s was apparently agglutinated later than the third person ending *-t, namely after the working of the sound law *n > *r / _#. The A laut of the se o da e di gs ust ha e ee : , hi h is reflected, as we saw above, in the Ablaut of the root aorist in Vedic, Greek and Hittite (> 6:0). I ha e o s iousl used the te s p i a a d se o da i elatio to the diffe e t h2econjugation endings, because the exact distribution is far from clear. We do seem to find the secondary endings (*-h2, *-s and *-érs) primarily in past tense context: the root aorist, the reduplicated imperfect, probably the pluperfect. On the other hand, there is evidence for *-h2a in the k-aorists of Greek, Latin and Tocharian, and in the Luwian 1sg preterit ending -ha. And there is evidence for *-e in the reduplicated aorist, which has largely become thematized. Whether this reflects the original distribution, or whether this was a late development, I cannot say. It is clear that the secondary endings, especially in the 1sg (*-h2) and the 2pl (*-Ø) were apt to be replaced by miconjugation endings (*-m, *-(s)té) early on. An alternative strategy might have been to substitute them with h2e-conjugation primary endings (*-h2e, *-(s)e). This might then explain the further characterization of the primary endings with *-i in Anatolian (*-h2ai, *-th2ai, *-ei). We can now give the complete reconstructed paradigms of the reduplicated present, root aorist and reduplicated aorist: 1 2 3 1 2 3 intr. dhéh1-h2(e) dhéh1-th2(e) dhéh1-s dhéh1-e tr.sg. dhéh1-h2 dhéh1-th2 dhéh1-s tr.pl. dhí-dhh1-h2(e) dhí-dhh1-th2(e) dhí-dhh1-s dhí-dhh1-e dhí-dhh1-m(e) dhí-dhh1-(e) dhi-dhéh1-r̥ s dhí-dhh1-r̥ dhéh1-m(e) dhéh1-(e) dhh1-érs dhí-dhh1-m(e) dhí-dhh1-(e) dhi-dhéh1-r̥ s dhí-dhh1-r̥ To obtain the attested endings in Vedic, the following changes must have occurred: 1. replacement of the h2e-conjugation endings with mi-conjugation endings, except in the 3pl (and sporadically the 3sg and 1sg). 2. stress shift in the 1/2pl (except in the root aorist). 3. extension of the reduplication to the singular in the intransitive forms edupli ated p ese t . 1 2 3 intr. dhi-dhéh1-m(i) dhi-dhéh1-s(i) dhi-dhéh1-t dhéh1-e t tr.sg. dhéh1-m dhéh1-s dhéh1-t tr.pl. dhí-dhh1- ̥ dhí-dhh1-s dhí-dhh1-t dhí-dhh1-et 1 2 3 dhi-dhh1-mé dhi-dhh1-té dhi-dhéh1-r̥ s dhí-dhh1- ̥ t dhéh1-me dhéh1-te dhh1-́r dhi-dhh1-mé dhi-dhh1-té dhi-dhéh1-r̥ s dhí-dhh1- ̥ t Note that the thematization of the edupli ated p ese t ust ha e p edated the e te sio of the reduplication to the singular (*bhér-e > *bhér-et, not *bhi-bhér-et). This model correctly predicts the Ablaut and stress patterns of the reduplicated present and imperfect in Vedic (sg. bibhárti, later bíbharti, 3pl pres. bí-bhr-at, but impf. (a-)bi-bhár-ur4). The same goes for the root aorist (dhā́t, dhā́ a, dhúr). As to the reduplicated aorist, this model at least offers some plausible paths to arrive at the situation as attested in Vedic (mostly thematic forms, with a few athematics as well, root mostly in zero grade, but sometimes showing e-grade). Hypothesis 5: In the h2e-conjugation, V1-lengthening denoted imperfective aspect Aoristic roots could make presents and imperfects by means of different strategies. The use of the bibhárti and thematized bhárati forms was recent, because it postdates the split-off of Anatolian. The use of the suffixes *nú o a d *jé the ? with mi-conjugation endings is older: it is securely attested in Anatolian, and the nu-forms (LIV2 1k, 1l, 1m) have been affected by a metathetic sound law which led to the creation of n-infix forms, especially in complex roots (C(C)VCC-). An even older strategy was, in my opinion, the lengthening of the first vowel of the verbal root to denote an intensive, durative or iterative. To explain the resulting verbal forms, we need to invoke a couple of sound laws affecting the (Pre-)PIE long vowels. As proposed by Kümmel 2012b, and already Viredaz 1983, PIE stressed *ó is to be derived from a Pre-PIE long vowel *ā. The length of *ó has been partially preserved in Luwian (in open syllables: *doru = tāru ood , To ha ia *o merges with *ē to *æ, while short vowels go to *ä), and Indo-I a ia B ug a s la i ope s lla les . I further assume that Pre-PIE also had *ī and *ū, which under the stress give *́ and *ώ. The latter merged with *ó everywhere, except in Greek in a labial environment Co gill s la , see Carrasquer Vidal 2014. Mechanical substitution of the first vowel by its long counterpart yields the following paradigms: 1 2 3 1 2 3 intr. dhóh1-h2(e) dhóh1-th2(e) dhóh1-s dhóh1-e dh́-dhh1-m(e) dh́-dhh1-(e) dhi-dhéh1-r̥ s dh́-dhh1-r̥ tr.sg. dhóh1-h2(e) dhóh1-th2(e) dhóh1-s dhóh1-e dhóh1-m(e) dhóh1-(e) dhéh1-r̥ s dhóh1-r̥ tr.pl. dh́-dhh1-h2(e) dh́-dhh1-th2(e) dh́-dhh1-s dh́-dhh1-e dh́-dhh1-m(e) dh́-dhh1-(e) dhi-dhéh1-r̥ s dh́-dhh1-r̥ Applying more or less the same developments as before: 4 This is the regular form: as noted in the online discussion of this paper by Roland Pooth, the form actually attested earliest in ábhibhran, which conforms neither to the expected * hi- hér-r̥ s, nor is analogical after the present tense (in which case we would have expected * hí- hr- ̥ t = †bibhrat). The form looks like a root imperfect *bhr-ént (cf. 3sg. *bhér-ti = Ved. bhárti, Lat. fert), secondarily provided with reduplication. 1. replacement of the secondary h2e-conjugation endings with mi-conjugation endings, except in the 3pl. 2. stress shift in the 1/2 plural, with reduction of unstressed *ē to *e. 3. extension of the reduplication to the singular in the intransitive forms (i.e. the perfect), we arrive at the following paradigms: 1 intr. dhe-dhóh1-h2a 2 dhe-dhóh1-th2a 3 dhe-dhóh1-e 1 2 3 dhe-dhh1-mé dhe-dhh1-té dh́-dhh1-r̥ tr.sg. dhóh1-m dhóh1-h2a dhóh1-s dhóh1-th2a dhóh1-t dhóh1-s dhVh1-mé dhVh1-té dhéh1-r̥ s tr.pl. dh́-dhh1- ̥ dh́-dhh1-s dh́-dhh1-t dhe-dhh1-mé dhe-dhh1-té dh́-dhh1-r̥ The first (intransitive) paradigm is obviously that of the perfect. The only unexpected form is the 3pl, where application of the rules sketched above must have yielded a form with long reduplicated vowel. Such a form is indeed attested in Vedic. The material, collected in Krisch 1996, shows the following forms where the long vowel is apparently restricted to the perfect 3rd person plural (as well as the middle and the perfect participle), and where there is no independent evidence that the root started with a laryngeal: kalp- ( ākl̥p r, ākl̥pé), gardh- (jāghr̥ d r), tarp- (tātr̥p r), taṛ- (tātr̥̣ r), dhaṛ- (dādhr̥ ̣ r), randh- (rāradh r), vart- ( ā r̥ t r, ā r̥ té), aś- ( ā aś r, ā aśé), śad- (śāśad r, śāśadré), arś- ( ā r̥ ś r). We can add the following roots where the long vowel occurs only in the middle perfect or the perfect participle: vas- ( ā asé), maṁh- ( ā ahé), āś- ( ā aśé), darh(dādr̥ hạ̄ás), rabh- (rāra hé) and vañc- ( ā akré). I would conclude that the long vowel in the perfect 3pl is an archaism, and not an analogical extension triggered by roots starting with a laryngeal. That being said, the anomaly was usually corrected by aligning the 3pl form with the 1/2pl, resulting in the normalized 3pl form *dhe-dhh1-́r. Still, there seems to be additional evidence for a long reduplication vowel in two curious Hittite 3pl forms: hē er OH/NS, the d e li uids and hēs(s)er OH/OS, the ope ed , hi h a o l e de i ed f o *h2ē-h2n- and *h2ē-h2s-, with Ei h e s la . Using the secondary endings, a pluperfect was made. The paradigm, after replacement of the h2econjugation endings, and application of the relevant sound laws, must have been: dhe-dhóh1-m dhe-dhóh1-s dhe-dhóh1-t dhe-dhh1-mé dhe-dhh1-té dhi-dhéh1-r̥ s5 The anomalous reduplication vowel -i- and root Ablaut -e- of the 3pl are confirmed by Avestan ikōitərəš had appea ed < *ki-kéit-r̥s, with /k/ restored after sg. *(ca-)kait-. The second column in the table above corresponds to the category dubbed olō-verbs by Jasanoff. These had maintained h2e-conjugation endings in Anatolian (e.g. Hitt. malli he g i ds . The e e thematized or otherwise drawn into the mi-conjugation outside of Anatolian, except perhaps for the 5 Unstressed *ī is shortened to *i. form *woid- to k o , attested as a u edupli ated pe fe t i Germanic preterit-presents. ost IE la guages, as ell as the The singular of the molō-verbs shows o-grade and no reduplication. Reconstruction of the plural forms is more difficult. In Hittite, we mostly have 3:3 Ablaut (6:0 preterits, such as dās, dāer to take , from *déh3-, reflect old root aorists), and either zero grade or e-grade of the root. In Tocharian, the o-grade in the plural of preterit class I can only derive from o-grade plurals of olō-verbs. In the other branches, there are sufficient cases of alternation *e ~ *o to suggest that e-grade was common in the plural, and not just in the 3pl. We have, from the root *bhedhh2-, Hitt. paddai, Lat. fodiō and Slav. odǫ, but Lith. bedù. Other cases are: *bhlendh*́hengh*ghrebh*ghrem*leid*melh2*sten- Gmc. blandan Gmc. gaggan Gmc. graban Lith. gramù Lat. lūdō, Lith. láisti, Latv. laîst Hitt. malli, Gmc. malan, Lith. malù, Lat. olō (?) Slav. sto jǫ Lith. le dži Lith. že gi Latv. grebju, Slav. gre ǫ Latv. gremju Lith. léisti OIr. melid, Slav. eljǫ Slav. ste jǫ, Lith. stenù, Grk. sté ō, Gmc. stenan Especially the common pattern Baltic e-grade vs. o-grade elsewhere suggests a distribution sg. *-o-, pl. *-e-. This is similar to what we find in the nasal presents, where Baltic has n-infix forms, originally from the plural, while Slavic has generalized *-né-/-nó- from the singular. Likewise, i the essi efie ti es , Balti o ti ues the plu al fo *-h1i-, while Slavic has generalized the singular form *-éih1(see Carrasquer Vidal 2006). The plural forms would have been initially *mólh2-me, *mólh2-(t)e, *mélh2r̥ s. If the stress shift in the 1/2pl was early enough, that would have yielded *melh2-mé, *melh2-(t)é, *mélh2r̥s, but the Tocharian o-grade in the plural rather suggests *molh2-mé, *molh2-(t)é, *mélh2r̥ s. Analogy with the 3pl and the root aorist would have resulted in *melh2-mé, *melh2-(t)é, *mélh2r̥ s, while analogy with the perfect and the mi-conjugation would rather have resulted in * l̥h2-mé, * l̥h2-(t)é, and perhaps * l̥h2́r. Finally, the third column in the paradigm table above shows forms with long vowel reduplication (except in the stress-shifted 1/2pl) and zero grade of the root throughout. Such a category is not directly reflected by any known category in e.g. Vedic, Greek or Hittite. However, it is precisely the paradigm we might expect to underlie the Na te p ese ts i the analysis of Sandell 2014. In that paper, Sandell proposes that one category of long vowel Na te p ese ts , those of the shape *C(C)eu-, arose phonologically as a result of Sze e é i s la i the sg (*stéus > *st́us a d Sta g s law in the 1sg (*stéum > *st́ u ), which is an excellent suggestion6. The remaining Narten presents largely conform to the root shape *CeT, and Sandell proposes to derive those from formerly reduplicated forms *Ce-CT- > *CēT. In my reinterpretation, the original reduplicated forms would have shown the structure *Cē-CT- in the sg. and 3pl, which works just as well, if not better. At least 15 roots given in LIV2 as a ostati oot p ese ts conform to the pattern suggested by Sandell: *dek̂- a -, auf- eh e , ah eh e , *h3rê- ge ade i hte , ausst e ke , *k̂es- a s h eide , *tetk̂- e zeuge , he stelle , *̂e hesse , kaue , *g(w êh- i s Wasse ei d i ge , hi ei ate , *k(w)h2ed- ze uets he , *kwek̂sehe , e li ke , 1. *med- esse , fü Ei haltu g so ge , si h klü e , *rep- a si h eisse , 6 I would likewise explain the long ē-grade in the sigmatic aorist as the result of Szeme é i s la . upfe , *sekH- a t e e : . s h eide , . u te s heide , *teh2G- e üh e , fasse , *tresk̂aus uets he , d es he , * ek̂- ü s he , 3. *wes- g ase , eide , e zeh e , esse . To resume, I have tried to show that the h2e-conjugation yielded 7 distinct verbal categories in PIE: 1. 2. 3. 4. 5. 6. 7. the edupli ated p ese t the root aorist the reduplicated aorist the perfect the olō-verbs the Na te e s the sigmatic aorist Categories 1- e e o igi all pe fe ti es , espe ti el i t a siti e, t a siti e ith si gula o je t, and transitive with plural object. Categories 4- e e the o espo di g eo-i pe fe ti es , ith lengthened vowel, likewise split into three paradigms. The sigmatic aorist was an offshoot of the root aorist, by reinterpretation of the 3sg ending *-s as a perfectivizing marker. Hypothesis 6: In Anatolian and Germanic, reduplication was not extended to the singular Above, I hypothesized that edupli atio i the edupli ated p ese t a d i the pe fe t as originally confined to the plural forms, and that is was subsequently generalized to the singular, leading to the classic reconstructions *bhi-bhér-ti and *dhe-dhóh1-e. However, if we assume that this generalization only occurred after the split off of Anatolian, Tocharian and Germanic (and with the latter, perhaps Balto-Slavic), a number of phenomena in Germanic become easier to understand. In Germanic, we have direct evidence for a difference in syllable length between the singular and the plural of eithe the edupli ated i pe fe t o the pe fe t fo ou p ese t pu pose it does t atte which one), in the Gothic endings of the weak preterit: 1. 2. 3. 1. 2. 3. -da -des -da -dedum -deduþ -dedun The forms of the verb *dheh1- to do hi h e e used to apparently devoid of reduplication in the singular. ake the Ge a i eak preterit were As e al ead oted a o e, the ise of the the ati p ese ts out of the atego of edupli ated p ese ts is also better explained if we assume that the 3sg *bhére originally lacked reduplication. If we turn to the Germanic strong preterits, the apparent loss of reduplication in classes 1-6 contrasts with the retention of reduplication in class 7 (at least in Gothic, and sporadically in North-West Germanic as well). Evidently, reduplication was not lost in Germanic as the result of regular sound law. 1. steig- staig 2. beug- baug 3. bind- band stigun bugun bundun In classes 1-3, if the perfect was fully reduplicated, we have to explain the loss of reduplication in *ste-stoig- > *staig-, *ste-stig- > *stig-, etc. If we assume that reduplication was limited to the plural, we only have to explain reduplication loss in *ste-stig- > *stig-, which is at least twice as easy. 4. ber5. geb- bar gab ēru gē u In classes 4 and 5 (roots ending in -eC and -eR), there is no generally accepted explanation for the plural forms with long vowel. We again have to deal with irregular loss of reduplication in the singular (*bhe-bhor- > *bar-) as well as in the plural (*bhe-bhr- > * ēr-). If reduplication were confined to the plural, the singular form requires no explanation. The long vowel in the plural can be e plai ed as i the ase of the Sa dell e s a o e, edu tio of *bhe-bhr- > * ēr-. In other words, reduplication in the plural was not lost, merely reshaped. 6. far- fōr fōru Class 6 is the equivalent of classes 4 and 5 for roots with o-vocalism ( olō-verbs of the shape -oR, -oC). We can again hypothesize that the long vocalism started in the reduplicated plural forms: *fe-bar- > *fear- > *fār- > *fōr-. Note that this, like the Tocharian preterit class I, implies ograde in the plural. In class 6, the long vowel was transferred to the singular, where original *farwould have been ambiguous as to tense. The developments in Germanic classes 4, 5, and 6 find an exact parallel in East Baltic, where verbs of the structure CeR(H), CeC, CoR(H) and CoC make preterits with lengthened vowel: ber- > ėr̃ ė, lek- > lėk̃ ė, mal- > mólė, ag- > võgė. 7a. hait- Goth. haihaitNWG hē2t7b. staut- Goth. staistautNWG steut7c. hald- Goth. haihaldNWG hē2ld- ~ held- The verbs of class 7 with o-vocalism have reduplication in Gothic (singular and plural, both with ograde of the root). In the traditional framework, this has to have been analogically reshaped after original: *hegait- *hegit- In my alternative hypothesis, the forms become: *hait- *hegit-, and the analogy is somewhat easier to make. In Gothic, the singular *hait- acquired the reduplication syllable from the plural (> *hehait-)7, and the plural was replaced with forms analogous after the new singular. For North-West Germanic, we can assume a course of events roughly as postulated in Jasanoff 2007, although one can argue about some of the details. The plural, reduplicated, forms were in any case central to the reshaping of *hegit- > NWGmc *heit- > *hē2t-, and similarly *stezut- > *steut-, *hegald> *held-, etc. The unreduplicated singular forms resolved the ambiguity with the present by adopting the reshaped plural forms. The other class 7 verbs (7d.: lēt- : lelōt- / lē2t-; slēp-: seslēp- / slē2p-; 7e. flōk-: feflōk- / fleuk- ~ flē2k-) follow the same analysis as the a-grade verbs of classes 7a-c. In summary, although it is not impossible within the traditional framework to trace the attested Germanic strong preterits back to a perfect that reduplicated in all forms, a model with reduplication only in the plural offers a pathway that in my opinion is smoother in all respects. Similarly, the lack of reduplicated presents as a separate verbal category, and the almost complete absence of reduplicated perfects in Hittite is easier to explain if reduplication was limited to the plural in the proto-language. The reduplication was apparently dropped in Hittite (and presumably Tocharian) in the plural, leaving no reduplicated forms, except for those that were reduplicated in oth the si gula a d the plu al i.e. the edupli ated ao ist a d the p oto-Na te fo s). In Tocharian, these two categories gave rise to the Tocharian Kausativ, with reduplication, expressed as initial accent in TochB. Only where the reduplication was synchronically no longer recognizable as such, it was maintained, as in the case of Hittite hē er the d e li uids , hēs s er the ope ed , mentioned above, and presumably ār- ~ ēr- (*h1/3ór- ~ *h1/3é(:)h1/3r- to a i e . I To ha ia , Malzahn 2010 gives the following examples: otkasa-me sepa ated < *wäwætk-, orsa-c a a do ed < *h2eh2or(H)-, 3pl arar-c < *h2eh2rH-, with reduplication otherwise lost in the perfect (continued in the Tocharian preterit class III). In Hittite, a number of verbs do show synchronic reduplication. In Kümmel 2012a, the following items are found (not counting ite s ith full , i te si e, reduplication): mimma- a lehe e , pippa- u stü ze ? , lilhuwa- giesse , lilipa- le ke , sishae ts heide , titta- aufstelle = *si-sth2-), wiwa- s h eie , lilakk- fälle , wewakk- itte . This last form is sometimes considered to be a true perfect (with 3sg preterit wewakta its pluperfect), but in a form * e- ó́-ei we would have expected lenition to *wewāki, instead of attested wewakki. I would prefer to derive this form from a reduplicated aorist *wí-wk- ~ *wi-wék-, o a p oto-Na te form * ́-wk- ~ *we-wk- , ith <a> here being a prop-vowel. Most importantly, if the PIE perfect, as an intransitive form, only had reduplication in the plural, which was later eliminated in Anatolian, this at long last explains what happened to the perfect in Hittite. According to the traditional perfect-theory , the PIE perfect became the Hittite hiconjugation, and all hi-conjugation verbs ultimately derive directly or indirectly from the perfect, which is difficult to reconcile with the facts. By contrast, in Jasanoff s h2e-conjugation theory, the hiconjugation verbs are related to certain kinds of PIE presents and aorists, but the perfect itself would have left virtually no traces in Hittite. This is also hard to believe. In fact, Hittite hi-conjugation forms from aoristic roots do continue PIE perfects, with originally reduplication in the plural (lost except in cases like hē er and hēs s er), and o-grade and no reduplication in the (generalized) singular. 7 An interesting observation made by Hans-Werner Hatting in the online discussion of this paper is that the reduplication vowel is always written <ai> in Gothic, where one would expect <i>. By my hypothesis, the *e is here the result of shortened *ē in unstressed position. If Proto-Germanic *ē was [ɛː], as suggested by WGmc. *̄ ~ *ā, the shortening product may have been *ɛ rather than *e, which would explain the Gothic peculiarity better than the standard explanation: generalization of the vocalism of verbs starting with h- and ƕ-. One objection that can be made against the theory that Proto-Germanic did not have reduplication in the si gula of the pe fe t a d the edupli ated i pe fe t , a e the West-Germanic preterit forms of the e to do . The present is a olō-verb (*dō (i), *dōs(i), *dōþ(i), *dō (az), *dōþ(e), *dō þ(i)), providing the present tense to a preterit: OS deda dādi deda dādun OHG teta tāti teta tātu , -t, -n OE dyde dydest dyde dydon As e sa a o e, the e di gs of the eak p ete it appea to o ti ue the p ete it of the e to do , added to a gender- and number-neutral form of the past participle *-tó-. Compared with the WestGermanic preterit forms, the main difference is that there was no reduplication in the singular. But if so, where does the reduplication in West Germanic come from? With their reduced endings (1/3sg -da), these forms with reduplication in the singular appear to be old, although they cannot be as old as the creation of the weak preterit. The anomalous OE form dyde is in origin a subjunctive (*dudī-). The original indicative form must have been *dud-, which certainly had its origin in the plural (*didu- > *dudu-). Although it is tempting to see the form with reduplication *dēd- as a perfect with long reduplication vowel, no other examples of this variant are known to have survived in Germanic. The form is then better explained as resulting from an amalgamation of *dē- x *did- > *dēd-, in the manner of Balto-Slavic *dō- x *did- > *dōd- to gi e , an analogy which in fact also works better without reduplication in the singular. As to the OS/OHG singular *dedē, the only explanation I can think of is that the reduplication was introduced from the subjunctive, when that still had the shape *didī (later *dēdī). As the OS/OHG 2sg pret., and the whole of the OE preterit show, subjunctive forms repeatedly influenced or took over the indicative preterit pa adig of the e to do i West Ge a i , a d this ight e the o igi of the edupli atio i Ist aeo i / sg *deda. References Carrasquer Vidal, Miguel, 2006, Slavic verbal accentuation. Carrasquer Vidal, Miguel, 2013, The Latin perfect. Carrasquer Vidal, Miguel, 2014, Co gill’s la . Jasanoff, Jay H., 2003, Hittite and the Indo-European verb, Oxford. Jasanoff, Jay H., 2007, From Reduplication to Ablaut: The Class VII Strong Verbs of Northwest Ghermanic, in: Historische Sprachforschung 120: 241-284. Kloekhorst, Alwin, 2007, The Hittite inherited lexicon, Leiden. Krisch, Thomas, 1996, Zur Genese und Funktion der altindischen Perfekta mit langem Reduplikationsvokal; mit kommentierter Materialsammlung, Innsbruck. Kümmel, Martin Joachim, 2012a, Liste hethitischer Verben. Kümmel, Martin Joachim, 2012b, Typology and reconstruction: the consonants and vowels of ProtoIndo-European, in: Nielsen Whitehead, Benedicte, Thomas Olander, Birgit Anette Olsen and Jens Elmegård Rasmussen (eds.), The Sound of Indo-European: Phonetics, Phonemics, and Morphophonemics, Copenhagen. Kümmel, Martin Joachim, 2015, Anatolisches und indoiranisches Verbum: Erbe und Neuerung, Marburg. 2 LIV (Lexicon der indogermanischen Verben), 2001, Helmut Rix, Martin Kümmel, et al. (eds.), Wiesbaden. Malzahn, Melanie, 2004. 3:3, 5:1, or 4:2? On the ablaut of the root aorist in Greek and IndoEuropean, in: Historische Sprachforschung 117:1, 50-75 Malzahn, Melanie, 2010, The Tocharian verbal system, Leiden/Boston. Pooth, Roland, 2015, Is the "tēzzi pri iple" a plausi le i fere e?, Leiden. Sandell, Ryan, 2014, The phonological origins of Indo-Europea lo g o el Narte prese ts, Los Angeles. Sturtevant, Edgar H., 1940, The Greek κ-Perfect and Indo-European -k(o)-, in: Language 16-4, 273284. Viredaz, Rémy, 1983, The Proto-Indo-European and North-West Caucasian vowel systems.