The sigmatic forms of the Hittite verb

Central to the problem of the Hittite verbal system is the status of the ḫi- conjugation 3 sg. pret. in - š and its relationship to other sigmatic morphemes—the partly overlapping 2–3 sg. ending - šta , the 2 pl. endings - šten(i) and - šdumat , and the synchronically unanalyzable *- s - of ganeš - ‘find, recognize’ and other s -extended verbal roots. The account of these endings given in Jasanoff 2003 is reviewed and, where necessary, revised. splits, as seen were characteristic of certain types of protomiddle aorists. PIE neoactives of the type typologically to the deponents of up the


Introduction
If the title of this article reminds the reader of Holger Pedersen's monograph Les formes sigmatiques du verbe latin et le problème du futur indo-européen (Pedersen 1921) the déjà vu is intentional. Nearly a century ago, Pedersen tried to disentangle the confusing array of Latin verbal forms in *-s-, *-is-, and *-(i)ssin order to find Latin evidence for his theory of a PIE s-future. In the present paper the sigmatic suffixes and endings to be disentangled come from Hittite, and the IE category hovering in the background is the s-aorist. In Jasanoff 2003 (henceforth HIEV ) I set forth a theory of the PIE verbal system that placed particular emphasis on the evidence of Hittite and Anatolian. The essence of this theory was that the Hittite ḫi-conjugation (i.e., the conjugational class consisting of verbs with a 1 sg. pres. act. in -(ḫ)ḫi) was the continuant of a late PIE formal category that I called the "h2e-conjugation." PIE h2e-conjugation presents and aorists, I said, were grammatically active, but characterized by the endings, and in some cases also the ablaut, traditionally associated with the perfect (1 sg. *-h2e, 2 sg. *-th2e, 3 sg. *-e, … 3 pl. *-(é)rs). Since these endings were formally related to those of the middle (1 sg. *-h2e, 2 sg. *-th2e, 3 sg. *-(t)o, … 3 pl. *-ro/*-nto), I posited a pre-IE "protomiddle" diathesis from which the perfect, the h2e-conjugation, and the middle were all descended. At a certain point in the prehistory of the family, I argued, protomiddle forms were either formally renewed as true middles (e.g., provided in typical cases with *-o (later *-to) for *-e in the 3 sg., deprived of paradigmatic ablaut, given "primary" variants with hic et nunc *-r, etc.) or reinterpreted as h2econjugation actives. Only one such class of actives-the formation we know as the perfect-was robustly preserved in the IE languages that remained after the separation of Anatolian from the rest of the family. Although I avoided speculation on the precise value of the protomiddle, there is good reason to think that it was originally patient-oriented and intransitive.1 On the whole, the response to the "h2e-conjugation theory" over the past fifteen years has been gratifying. Inevitably, some particular analyses have found greater favor than others. Thus, "molō-presents" (the ideal type 3 sg. *mólh2e : 3 pl. *mélh2-r̥ s (> *-n̥ ti) 'grind') have found considerable acceptance, while "i-presents" (3 sg. *dhéh1-i̯ -e : pl. *dhh1-i̯ -érs (> *-énti) 'suck') have called forth 1 My most recent views on the functions of the protomiddle and the prehistory of the perfect, which have developed considerably since HIEV, are presented in Jasanoff 2018a. As described there, the pre-PIE protomiddle, marked by endings of the "h2e-series," included the full range of functions associated with the classical middle, including the formal subtype that some writers (e.g., the editors of LIV ) call "stative." Opposed to the protomiddle, both in present and aorist stems (the perfect had not yet emerged as an autonomous category), was the active, with its familiar endings of the "m-series" (1 sg. *-m(i), etc.). The major functional difference between the protomiddle at this stage and the later classical middle (sensu lato) was that, for reasons possibly rooted in the deeper prehistory of the protomiddle as an intransitive "patientive," the protomiddle endings often also occured a) with forms that had come to function as ordinary actives (e.g., 3 sg. pres. *mólh2-e 'grinds'), and b) with forms that functioned both as transitive actives and intransitive middles depending on syntactic context (e.g., 3 sg. pres. *ḱónk-e 'hangs (up)' (trans.) and also 'dangles, is suspended' (intrans.)). The transition from this older system to that of late PIE was marked by the formal differentiation of the true middle, with renewed, but still recognizably h2e-aligned endings, from the protomiddles that retained their unrenewed h2e-series endings and henceforth patterned as actives.
In the aftermath of this step, one and the same original paradigm could be represented by both a renewed middle (e.g., 3 sg. *ḱónk-or or *ḱn̥ k-ór 'dangles, is suspended'; cf. Hitt. kangattari) and an unrenewed neoactive (e.g., *ḱónk-e 'hangs (up)'; cf. Hitt. kānki). Such paradigm splits, as will be seen below, were particularly characteristic of certain types of protomiddle aorists. Late PIE neoactives of the type 3 sg. *mólh2-e, *ḱónk-e (1 sg. *-h2e(i), 2 sg. *-th2e(i), etc.), loosely comparable typologically to the deponents of later IE traditions, made up the h2e-conjugation. a variety of alternative proposals.2 But disagreements over the correct reconstruction of this or that putative h2e-conjugation present type mainly concern differences over ablaut grade, relative date of thematization, and other questions of detail. h2e-conjugation aorists, or at least some of them, raise questions of a more fundamental nature. Aorists with h2e-inflection were the focus of chs. 6 and 7 of HIEV. In ch. 6 I discussed the relatively straightforward "stativeintransitive" subtype, characterized by *-o-: *-e-/zero ablaut (e.g., *lógh-e : *légh-r̥ s 'lay down') and derivationally associated with other protomiddle-based formations, notably including the perfect, in what I called "stative-intransitive systems."3 Stative-intransitive aorists, with paradigmatic ablaut still preserved in the Indo-Iranian "passive" aorist (Ved. 3 sg. ádarśi : pl. adr̥ śran 'appeared') and the Tocharian class V subjunctive (A 3 sg. wekaṣ : pl. *wikeñc 'will disappear'), are in fact as well as or better grounded than molō-presents; they will be taken for granted in what follows. In ch. 7 I argued for a second and more controversial h2e-conjugation subtype, likewise with *-o-: *-e-/zero ablaut, but with different semantics and a different derivational profile. In this second subtype, I said, the paradigm was partly infiltrated by sigmatic forms in the parent language, so that an aorist like *nóiH-/*néiH-'lead' had a 1 sg. *nóiH-h2e (cf. Hitt. nēḫḫun) and a 3 pl. *néiH-r̥ s (: Hitt. naier), but also an unexpected sigmatic 3 sg. *nḗiH-s-t (: Hitt. naiš), seemingly based on a suppletive stem in *-s-. The resulting "presigmatic" aorist, I argued, was later fully sigmatized in the classical IE languages, becoming the familiar s-aorist. Full sigmatization did not take place in Tocharian, showing this branch to have been the second to leave the IE family.
The idea that the familiar s-aorist, with *-s-running through all its forms, was a post-Anatolian and post-Tocharian innovation, and that its PIE ancestor was a h2e-conjugation root aorist with extraneous s-forms, has been sharply contested.4 It is easy to see why: the traditional reconstruction, with *-s-everywhere, is supported by the evidence of many branches; the Hittite facts are genuinely complicated; and the intricacies of the highly structured Tocharian verbal system remain a closed book to many Indo-Europeanists. Central to the debate is the ḫi-conjugation 3 sg. pret. in -š and its relationship to other sigmatic morphemes in Hittite-the partly overlapping 2-3 sg. ending -šta, the 2 pl. endings -šten(i) and -šdumat, and the synchronically unanalyzable *-s-of ganeš-'find, recognize' and other s-extended verbal roots. The full gamut of sigmatic forms of the verb has been intensively studied and debated since 2003, settling some questions and raising new ones. The time has come for a fresh look at the evidence.

11
To which the Latin future type faxō 'I will do' was, according to the most common view, the subjunctive. 12 That the *-s-of *-sḱe/o-is the "same" as the *-s-that stands alone is an obvious default assumption, although the identity of the second element is obscure. See Oettinger 2013 for a recent proposal.
Of these, karš-is an active mi-verb in Old and Middle Hittite; ḫann(a)-has an active 3 sg. ḫannāi that is at least as old as the "stative" 3 sg. ḫannari;21 and iškall(a)-has no attested indicative before Neo-Hittite, where it is mostly active. There remain only paḫš-, with its robustly attested imperative paḫši, and šalig-, with its imperative šaliki, found only on a single tablet.22 Everything points to paḫši itself, haplologically reduced from a present subjunctive *peh2-s-esi and synchronically reanalyzed as paḫš-i, as the locus of the paḫši group as a whole. The discovery that paḫši was a si-imperative has the important corollary, which will be exploited below, that the PIE subjunctive must have existed prior to the separation of Anatolian from the rest of the family.
Indo-European Linguistics 7 (2019) 13-71 2014) with Ved. pres. náyati 'leads' (= YAv. naiieiti 'id.'), pointing to a root *neih1or *neih3-(we will write *neiH-). nai-is one of the relatively small company of Hittite verbs-the rest are mostly old root presents-whose morphological profile actively confirms part of what we know about its PIE averbo from the extra-Anatolian comparative evidence. Like *u̯ eǵh-'move by vehicle' , *u̯ edh-'move/lead' , *nem-'bend' , *pekw-'ripen' , *dhegwh-'burn' , *gwes-'extinguish/go out' , and others, *neiH-belonged to a specific family of roots that denoted a change of position or physical condition in the parent language. ) and in the associated class VIII (s-) presents, which correspond formally to s-aorist subjunctives in the other languages (e.g., 3 sg. namṣäṃ, pakṣäṃ, tsakṣäṃ, keṣäṃ; cf. § 11). In Hittite, a continuant of the 3 sg. s-aorist *nḗiH-s-t is traditionally seen in the regular 3 sg. pret. naiš. This identification has latterly been contested (wrongly, in my opinion; see below). But even without naiš there is unmistakable evidence for an aorist stem *neīH-s-in the unique 2 sg. mid. impv. nešḫut 'turn (intr.)!' . The form nešḫut (MH/MS; later naišḫut) provides a bridge between the familiar terrain of the Indo-Iranian verbal system and the more exotic morphological landscape of Hittite. Like most Hittite ḫi-verbs with stems ending in a diphthong, nai-takes an unexpected -s-before the -t-of the second person plural: cf. act. naišteni, naištani (pres.), naišten (impv.), mid. 2 pl. naišdumat 23 I use "Inner IE" to refer to the phylogenetic clade consisting of all the IE branches other than Anatolian and Tocharian. The term incorporates my already-mentioned view that Tocharian was the second branch to leave the family, after Anatolian.
(impv.).24 The origin of the "intrusive" -s-in these forms, on which opinions are divided, will be discussed in the following section. But whatever the prehistory of the sigmatized endings -šten(i), -šduma(t), etc., a special explanation is needed for the -s-of nešḫut. The non-dental middle imperative ending -(ḫ)ḫut is never otherwise sigmatized.25 Since HIEV 182-184 I have compared nešḫut with the Vedic middle imperatives rasva, mátsva, yákṣva. The latter are not root aorist forms-the roots in question do not make root aorists-but medializations of the attested si-imperatives rasi, mátsi, yákṣi, made by replacing what appeared to speakers as an anomalous active imperative ending -si by the normal middle ending -sva.26 nešḫut is such a form in Hittite, medialized from an inferrable si-imperative *nēši (< *néiHsi < *néiH-(se)si). From an inner-Hittite point of view, *nēši would have been an "i-imperative" like paḫši, ḫuitti, or šaliki; its medialization consisted in replacing the perceived active ending *-i by the middle ending -(ḫ)ḫut. It is true that there is no actual *nēši in Hittite; the attested active imperative of nai-is the obviously analogical naī (like dai 'put!' , išpāi 'be sated!' , etc.). But the Vedic si-imperative néṣi is found ten times in the Rigveda, making it more common than all the other s-aorist forms of the root nī-combined. There is also a well-attested subjunctive néṣa-(cf. OAv. naē-šat̰ ). A PIE si-imperative *néiHsi is thus very nearly guaranteed. nešḫut can be seen as its indirect reflex in Hittite.

2 pl. -šten(i), -šduma(t), etc. (part I)
The intrusive -s-of 2 pl. naišten(i), etc. was briefly discussed in HIEV (119-120, 184). The explanation offered there was that with the establishment of the siimperative pair *nēši : nešḫut, or the forms ancestral to these, the sequence *nēš-(*neiH-s-) was interpreted as the "imperative stem" of the verb that later became nai-. As such, *nēš-became the basis for the creation of 2 pl. imperatives *nēšten (act.) and *nēštuma (mid.).27 But since the imperative and preterite were normally identical in the 2 pl., I argued, these forms came also to serve as preterites. In the latter role, they analogically took on the regular "strong" 24 There is also a 2 sg. pres. naištari (NH). 25 In nai-, on the other hand, it is never not sigmatized: a 2 sg. mid. impv. *nēḫḫut would have been perfectly well-formed, but does not occur. 26 So too váṃsva (: subj. váṃsa-'grant') and sakṣva (: subj. sāḱṣa-'conquer'), though the corresponding forms in -si are not attested. 27 To which was later added the final particle -t, presumably identical with the -t (< *-dhi) of the active imperatives in -nut (cf. arnut 'bring' , etc.).
Kloekhorst has no doubt what conclusion should be drawn from these forms: When we look at the 21 verbs for which an ending -šten(i) is attested … we immediately see that they are all ḫi-inflected verbs. There is not a single mi-inflected verb that shows the ending -šten(i). This cannot be coincidental: statistics show that the chance that a random collection of 21 Hittite verbs consists of ḫi-verbs only, is 1 in 1.2 trillion … Our conclusion therefore should be that the ending -šten(i) is the original ḫi-ending that contrasts with the mi-ending -tten(i).30 A footnote explains the mathematics: "To my knowledge, we find about 210 ḫi-verbs and 580 mi-verbs in Hittite. The ratio ḫi-verbs : total number of verbs therefore is 210 : 790 = 1 : 3.76. The chance that a random list of 21 Hittite verbs consists of ḫi-verbs only then is 1 : 3.7621 = 1 : 1,212,170,547,718." It takes no great sophistication with numbers to verify that Kloekhorst's list is, as he says, not random. The 21 verbs on the list are indeed all ḫi-verbs, which cannot be accidental. But, more to the point, and not emphasized by Kloekhorst, they are ḫi-verbs of a very particular type. Eight of the 21 are monosyllabic stems in which -ai-alternates with -i- (išḫai-, išpai-, mai-, pai-, parai-, šai-, dai-, zai-). Three others are ḫalzai-, a disyllabic stem which inflects the same way as the shorter forms (specifically, pai-); nai-, an alternating monosyllabic stem that differs from the others only in that the weak stem is nērather than ni-; and au-, a monosyllabic stem with the mirror-image alternation of -au-and -u-. Of the remaining ten items, two (penna(i)-and ūnna(i)-) are compounds of nai-, and a third (nanna(i)-) is somehow related to these.31 uppa(i)-has been influenced by pai-.32 This leaves only mema(i)-and dāla(i)-, which have the same inflection in the present as penna(i)-, ūnna(i)-, etc.; and ḫanna-, tarna-, šunna-, and peda-, where the full sequence -išten, with -i-reflecting an i-diphthong that is not etymological in these stems, was mechanically taken over from penništen, etc.33 The association of the endings -šten and -šteni with ablauting diphthongal ḫi-verbs is orders of magnitude more significant 30 ibid.

31
I leave open whether nanna(i)-is a reduplicated form of nai-, as claimed by Kloekhorst (2008b: 600), or an imperfective in -anna/i-, as maintained by Melchert (1998: 416). In the latter case it would have to have been influenced by penna(i)-and ūnna(i)-. 32 But uppa(i)-is not a compound of pai-; cf. Melchert forthcoming. 33 If these forms were archaisms we should have expected *-ašten. It is clear from Kloekhorst's own discussion that tarništen is more recent than tarnatten. Apart from the isolated Old Hittite hapax petišten (OS), peda-has only pedatten(i); the parallel compound uda-'bring hither' has only udatten(i) (MS+).
Indo-European Linguistics 7 (2019) 13-71 than the correlation of these endings with the ḫi-conjugation as a whole. It is therefore incumbent on Kloekhorst, who maintains that -šten and -šteni were originally the general 2 pl. endings of the ḫi-conjugation, to explain how and why these endings came to be restricted to diphthongal stems. It is not an easy task. Kloekhorst's sketchy remarks on this subject are put into clearer form by Melchert (2015: 129-130), who cautiously aligns himself with Kloekhorst's position. Melchert provisionally posits a pre-Hittite loss of *-s-when preceded by a stop or *h2 and followed by a stop (i.e., in sequences of the form *-TsT-and *-h2sT-). Forms like akten(i) (: ak(k)-'die') and watarnaḫten (wātarnaḫ-'order') are thus phonologically regular, according to Melchert, while, e.g., ārten (: ar-'arrive') and dattēni, dātten (: da-'take') are analogical for *āršten and *daštēni, *dāšten, respectively. In the end, the only verbs where *-s-was not replaced by sound change or analogy, under this scenario, were those with stems ending in a diphthong. None of this, I confess, seems in the least plausible to me. The correlation of -šten(i) with stem-final -ai-is a stunning fact-too stunning to be explained by ad hoc sound laws and random analogies. Yet this is basically what Kloekhorst and Melchert have to offer. The supposed loss of *-s-in *-TsTand *-h2sT-sequences is not an independently grounded sound change; the strongest thing that can be said for it is that apparent counterexamples like 2 pl. pres. paḫ(ḫa)šduma (: paḫš-) and 3 sg. pres. tak(ki)šzi (: takš-'inflict') contain a morpheme boundary and thus could, if necessary, be explained away as analogical. It is entirely unobvious, in the last analysis, why the steady analogical advance of s-less -(t)ten(i) at the expense of -šten(i) should have stopped precisely at the diphthongal stems.34 Particularly striking is the absence of any sign of the theoretically expected sigmatic forms *daštēni and *dāšten in the otherwise morphologically archaic verb da-.
From a more general point of view too, it would be quite surprising to find etymologically distinct mi-and ḫi-endings in the 2 pl. No one would dispute the possibility-and indeed, the likelihood-that the PIE active (> miconjugation) and perfect/h2e-conjugation (> ḫi-conjugation) endings were at one point distinct in all paradigmatic positions. Hittite, however, otherwise merged the two conjugations in the plural. Both mi-and ḫi-verbs have -wen(i) (-men(i) after -u-) in the 1 pl. and -(t)ten(i), with or without intrusive -s-, in the 2 pl. In the 3 pl., both mi-and ḫi-verbs have the original mi-conjugation ending (-anzi) in the present and the original ḫi-conjugation ending (-er) in the 34 Only to spread from these later, fitfully and uncertainly, to the non-diphthongal stems šunna-, ḫanna-, and peda-.
preterite. Against this background, it is a priori unlikely that the presence or absence of intrusive -s-would have a deep IE-level explanation. Nevertheless, Kloekhorst boldly compares (2008a: 498f.) the -s-of -šten(i) with the hitherto unexplained Toch. B 2 pl. pret. ending -s, -so (cf., e.g., takās 'you (pl.) were') and its Toch. A counterpart -s (tākas 'id.'). According to Kloekhorst, the underlying PToch. *-sǝ (vel sim.; see below) was the continuant of a PIE perfect-related sending whose exact form and relationship to the "real" 2 pl. perfect ending *-e (cf. Ved. vidá 'you (pl.) know') he is unable to specify. This PIE s-ending, he says, gave a pre-Hittite ḫi-conjugation ending that was recharacterized by adding the -(t)ten(i) of the mi-conjugation to yield the attested -šten(i).
Kloekhorst's explanation of the ending -šten(i), though a classic case of obscurum per obscurius, figures crucially, as we shall see, in a proposal of Melchert's regarding the prehistory of the s-aorist (Melchert 2015: 129 ff.; cf. § 10 below). Before going further, therefore, it will be worthwhile to take a closer look at the supposedly relevant Tocharian 2 pl. in *-sə and what it actually tells us.

Excursus: Toch. B 2 pl. -s, -so
The 2 pl. ending that Kloekhorst sets up as *-sǝ is discussed by Malzahn (2010: 42-43, 47, 514-515 and2011: 48-49). As her presentation shows, not one but two morphemes are actually subsumed under this formula, one the true 2 pl. pret. ending and the other the ending of the 2 pl. imperative. In the preterite, Toch. A has invariant -s, while Toch. B has both -s and -so. The position of the accent (B takas, klyauṣaso 'you heard') points to an underlyingly final vowel in both cases. As a rule, the -o that participates in this type of variation in Toch. B is the so-called mobile or "bewegliches" -o, which surfaces in metrical texts as the realization of an underlying final *-ǝ. In this case, however, as Malzahn points out, the longer variant in the 2 pl. pret. is disproportionately common, being found also in prose passages where simple -s would have been expected. In the imperative the preference for the longer form is even stronger-so much so that in one imperative class (cl. III) -so is found four times and bare -s is not attested at all. There is reason to think, therefore, that the -o of -so, at least in the imperative, might originally not have been *-ǝ, but some more substantial syllabic sequence that yielded normal (i.e., non-mobile) -o by sound change.
Strengthening this impression is the remarkable Toch. A 2 pl. imperative form päklyossū 'hear!' beside regular päklyoṣäs, with -sū for normal -s. The patterning of these forms suggests a single historical ending with secondarily differentiated preterite and imperative variants. Let us assume that the starting point was a 2 pl. pret./impv. of the form *-sX. In the preterite, this ending would have been subject to pressure from paradigmatically related forms, particularly the 1 pl., which ended in *-mǝ (< *-mes?). We can speculate, therefore, that *-sX was remade to *-sǝ in the preterite, giving Toch. A 2 pl. pret. -s and Toch. B alternating -s, -so; in the imperative, where there was no such remodeling, *-sX survived and gave Toch. A -su35 and Toch. B invariant -so. Inevitably, the now distinct preterite and imperative endings would have tended to be confused. This is why -so gives the appearance of being overused in the preterite in Toch. B, while -s figures as a variant of "correct" -so in the imperative. In Toch. A, the -s of the preterite (< *-sǝ) was introduced into the imperative, where it all but replaced the phonologically regular *-su that survives in the unique päklyossū.
Our ending(s) are thus most plausibly traced back to a sequence of the type *-s(u)u̯ ō. Here, however, we reach an apparent dead end. No PIE desinence of this shape is known.38 The closest formal lookalikes are the 2 pl. middle ending *-dh(u)u̯ e/o (Ved. -dhvam, Gk. -sthe, etc.) and the Indo-Iranian 2 sg. middle imperative in *-su̯ a (Ved. -sva), which, as I have suggested elsewhere (Jasanoff 2006(Jasanoff [2008), may go back to a preform *-sh2(u)u̯ o. All these contain an underlying *-Cu̯ -cluster and end in a vowel; whether there is any historical significance to this general resemblance is unclear. What is minimally clear is that, whatever its antiquity and internal structure, the pre-Tocharian ancestor of Toch. B -s, -so and Toch. A -s (+ impv. -su) was not a bare *-s or some other undercharacterized sequence (e.g., *-su) that would have invited "clarification" to -šten(i) in Hittite, as suggested by Kloekhorst. Nor is there any reason to think that it stood in some primal relationship to the simple sigmatic ending of the 3 sg. s-preterite in Tocharian (e.g., Toch. B preksa 'asked') or the 3 sg. pret. of the ḫi-conjugation in Hittite (e.g., dāš 'took'), as argued by Melchert (cf. below).

2 pl. -šten(i), -šduma(t), etc. (part II)
The Tocharian 2 pl. pret. in -s, -so is thus a red herring; it offers no support for Kloekhorst's claim that the -s-of -šten(i) comes from an ending that originally characterized the ḫi-conjugation as a whole. The salient distributional treatment are always actually reconstructable with a u-diphthong or a *-VwV(-) sequence, as in the examples above. 37 With generalized lengthened grade from the Narten s-present; cf. note 20. 38 Unless, as Michael Weiss suggests to me (p.c.), a connection can be shown to exist with the Umbrian future perfects benuso 'you will have come' and couortuso 'you will have turned back' , with -uso < *-us-so < *-swā. The possible relevance of these forms will be discussed elsewhere.
Indo-European Linguistics 7 (2019) 13-71 fact about intrusive -s-in Hittite is its intimate association with ḫi-conjugation diphthongal stems, and especially with nai-, the only verb where it also appears in the middle. Other things being equal, therefore, Kloekhorst's theory is at a disadvantage vis-à-vis the HIEV theory (cf. § 5), which places the locus of the -s-squarely in the diphthongal stems-specifically, in the si-imperative of nai-(*nēši) and its analogically extended "imperative system" (nešḫut, *nēšten, *nēštuma, later naišḫut, naišten, naišdumat). The nai-/*nēši/nešḫut scenario, however, has been faulted on other grounds. Melchert (loc. cit.) offers the following critique: Such a scenario is not impossible, but it rests on a series of unverifiable steps. Most problematic for such an account are the older zero-grade forms pišten(i) and uštēni, for which the paradigm of nai-(with an allomorphy of strong stem nai-versus weak stem nē-) provides no model (we expect either some trace of *pe/ēšten(i) and *u-uš-te-n(i) = /o:sten(i)/ or only renewed paišten and aušten). One should note that attested 2Pl pešten is only late and obviously secondary due to the confusion of e/i before š in New Hittite. Jasanoff's derivation also requires that a feature of an entire class be based on the alleged pattern of a single verb, one which furthermore did not originally belong to the class (hence precisely the difference of weak stem nē-< *néiH-versus stems in -Cī-/-Cy-< *-Cih1-C˚-/ -Ch1y-V˚-).
Before taking up these points individually, let us pause to recall that the monosyllabic diphthongal stems in -ai-, which form the nucleus of the verbs with intrusive -s-, are a heterogeneous group. Four of the monosyllabic examples on Kloekhorst's list (mai-, p(a)rai-, šai-, dai-) are standard i-presents with, as I have argued, e : zero ablaut (3 sg. *dhéh1-i̯ -e : pl. *dhh1-i̯ -énti).39 Another, išpai-, as discussed in HIEV 107-109, was probably originally an i-present with Narten ablaut (*spḗh2-i̯ -e : pl. *spéh2-i̯ -n̥ ti). The remaining two monosyllabic 39 The ablaut pattern of the main group of i-presents, discussed in detail in HIEV 98 ff., remains a contested issue. The case for an e : zero alternation, maintained in HIEV, rests on the evident etymological identity of these forms with the "verba pura" of the other IE languages, especially Germanic and Balto-Slavic (cf., e.g., PGmc. *sē( j)an = Lith. seju = OCS sějǫ 'sow': Hitt. šai-'press'). Since o-grade is conspicuously lacking in the verbs of this type outside Anatolian, it would be hard to justify a reconstruction with o : zero ablaut at the PIE level. The *-oi-: *-i-suffix posited by Kloekhorst (2008b: 808), which would be the only ablauting tense-aspect suffix in the PIE verbal system, is even more problematic. For the non-ablaut of the *-i-in i-presents compare the standardly assumed non-ablaut of the *-s-in s-presents (cf. LIV class Va: *u̯ éid-s-ti : *u̯ id-s-énti).

43
I know of no evidence that would decisively rule out the possibility that the sequence *-eh1i-retained its laryngeal long enough to escape the regular monophthongization of *-ei-to Hitt. -e-(cf. HIEV 99). 44 See HIEV 99. If, as generally assumed, PIE *-ei was the source of the normal ḫi-conjugation ending -i, either via a special close mid vowel *-ẹ or in some other way, then any of the sequences listed-*-eh1-i̯ -ei, *-oi̯ -ei, *-oiH-ei, and *-ēh2-i̯ -ei-could in principle have given Hitt. -āi. 45 "Normal," at least, in the sense that it was a general feature of disyllabic stems; cf. 3 pl. pret. uppier, mēmier, ḫalzier, pennier, etc. The unexpected scriptio plena of the -i-in the forms pīšten (OH/OS), pīweni (MH/MS), and pīwen (OH/NS), drawn to my attention by Melchert (p.c.), is perhaps to be interpreted as the phonological reflex of the weak stem Against this background, it is entirely natural that intrusive -s-should have spread from one member of our class-the extremely common verb nai--to the others. The form pišteni, singled out as problematic by Melchert, is notable only because, following the creation of pret./impv. *paišten on the model of naišten (or perhaps of earlier *pešten on the model of *nešten) and the transfer of the -s-from the preterite to the present ("*pittēni" → *pištēni → pišteni), the vocalism of the present (pi-) was generalized to the preterite, as just discussed. The case of uštēni, also cited by Melchert, is entirely straightforward:
*uttēni *uttēni uštēni The spread of intrusive -s-, then, must be envisaged as a two-stage process. First, -s-was extended from the preterite and imperative of nai-to the preterite and imperative of the other monosyllabic diphthongal stems; then -(t)teni was replaced by -šteni in the corresponding present forms. in the Neo-Hittite period (cf. Hoffner-Melchert 2008: 181, n. 6). The real locus of -šta must have been aušta, where the corresponding 3 sg. pres. is the unexpectedly sigmatic mi-conjugation form aušzi.46 The present in -šzi and preterite in -šta of this verb are clearly related. The standard view, obviously correct in principle, is that aušzi was a back-formation from aušta or some predecessor of aušta, and that it was created to replace the synchronically anomalous form *āwi (< *h1óu̯ -ei) that would have been inherited as the 3 sg. pres. of au-(contrast 1 sg. ūḫḫi < *h1óu-h2ei, 2 sg. autti < *h1óu-th2ei).47 The interest of aušzi and aušta lies in what they tell us about the prehistory of the regular ḫi-conjugation 3 sg. pret. in -š (naiš, dāš, etc.). As will be discussed below, there are two widely held positions on the formal history of this ending. According to one view, the ḫi-conj. 3 sg. pret. ended in simple *-s from the beginning, either because *-s was a PIE desinence in its own right or because it was originally some kind of PIE stem formative or enlargement followed by a zero ending. According to the other, more widespread view, the 3 sg. in -š was the continuant of earlier *-s-t, most often interpreted as the *-s-of the s-aorist followed by the normal secondary ending *-t. In the latter case, the treatment of word-final *-st would have been parallel to that of word-final *-nt. Inherited final *-nt gave both -n and -nta in Hittite: -n was the phonologically regular reflex, seen in the nom.-acc. nt. sg. of nt-stems (e.g., ptcp. appan 'taken' vs. animate nom. sg. appanza); -nta was the morphological (= analogical) reflex, seen in the 3 sg. pret. of mi-verbs, where -t was maintained or restored under the influence of primary -zi < *-ti (cf. 3 sg. pret. kuenta 'slew' (ultimately < *gwhén-t) beside pres. kuenzi < *gwhén-ti). There is no reason why there might not similarly have been two treatments for *-st-phonologically regular -š in the ḫi-conjugation, and "morphological" -šta in the 3 sg. pret. of mi-verbs with stem-final -š-(e.g., ganešta : pres. ganešzi, ēšta 'was': pres. ēšzi).48 A decision between original *-s and *-st is not possible on a priori grounds. 46 Possibly just as old, but not so well attested in older texts, is maušzi : maušta 'fall' . 47 A later development was the spread of aušta to the 2 sg., where it replaced "correct" *autta. A possible trigger would have been the identity, for independent historical reasons, of the 2 sg. and 3 sg. in the semantically related verb ištamaš-'hear' (pret. ištamaššun, -ašta, -ašta; cf. HIEV 121). 48 I take no firm position on the phonetics of the final vowel in these cases. The -a was almost certainly real, and can be assumed in the absence of evidence to the contrary to have been identical with the final vowel of the 2 sg. in -(t)ta (see the preceding note). Forms like Hitt. kuenta recall the Slavic 3 sg. aor. in -tъ of forms like OCS pitъ 'drank' < pre-Slavic *pīt. In Slavic the retention of the *-t in sentence sandhi before vowels (*pīt#V …) led speakers to infer an elided schwa (*pīt(ъ)# V …). This was then introduced into other environments, where it had the effect of protecting the otherwise vulnerable final stop.
In the case of aušzi and aušta, however, it is easy to see that the backformation of a present *austi (> aušzi) from a preterite *aust (> aušta) would have been a much more straightforward affair than the creation of a new *austi on the basis of a t-less preterite *aus. If the original ending of the ḫiconjugation 3 sg. pret. had been *-st, there would have been literally dozens of mi-conjugation verbs to provide a model for a present in *-sti-not only discrete lexical items like ganeš-and ēš-, but also the productive class of fientives in pres. -ešzi : pret. -ešta.49 If, on the other hand, the preterite of au-had been *aus from the beginning, it is unclear by what pathway a new *austi > aušzi would have arisen. A conceivable channel might have been the imperative, where the influence of 2 pl. *austen (aušten) could have led, via a nonproportional analogy, to the replacement of theoretically expected 3 sg. *awu (vel sim.) by *austu (cf. attested aušdu). The new 3 sg. impv. *austu could then have given rise to 3 sg. indic. *austi. But this is a much less intuitive scenario than simple back-formation of *austi from *aust. The aušzi : aušta pair thus constitutes meaningful evidence for the view that the 3 sg. pret. of the ḫiconjugation once ended in *-st.

3 sg. pret. -š: the Tocharian connection
The question of whether the -š of the ḫi-conjugation 3 sg. pret. goes back to simple *-s or *-st is separate from the question of its morphological identity. Many scholars (e.g., Eichner 1975: 79f., Oettinger 1979: 71, 405 f., Kloekhorst 2008b identify the ending -š with the *-s-of the PIE s-aorist and set up *-s-t; others (e.g., Pedersen 1938: 97, Watkins 1969 maintain the connection with the s-aorist but reconstruct bare *-s.50 Melchert (2015) chooses an unusual third option, denying a connection with the s-aorist but still setting up *-s-t, which he sees as having been renewed from earlier desinential *-s by adding the productive 3 sg. ending *-t. We will take it for granted here here, largely on the strength of aušzi, -šta, that the immediate source of Hitt. -š, however it may have come into being, was *-st. We must now revisit the question of whether the widespread, but not universal identification of this *-st with the 3 sg. of the s-aorist is justified. The case for equating Hitt. -š with the canonical *-s-t of the PIE 3 sg. s-aorist receives support from what is usually taken to be the reflex of the s-aorist in Tocharian. The active of the much-discussed Tocharian class III (s-) preterite is actually a root formation outside the 3 sg.:
taught' (: kärs-'know')) and in the clearly innovative preterite classes IV, V, and VII, which are based on inner-Tocharian imperfects. This means that with the exception of the irregular "class VI" preterites of the roots käm-'come' and lät-'go out' ,53 all Tocharian preterites other than those of class III follow the class I pattern, with no *-sa-. The situation can be summed up as follows: 1) all active preterites in Tocharian have the same endings in the first and second persons; 2) apart from the variable difference between *-rə (< *-rəs < *-r̥ s), which predominates in class III, and *-rae (< *-raent < *-ront), which predominates in the other classes, all active preterites likewise have the same ending in the 3 pl.;54 3) all preterites have the same ending-zero-in the 3 sg. In class III, however, an element *-sa-is interposed between the root and the zero ending. While the relevance of the *-sa-of class III to our Hittite discussion is obvious, the endings of the 1-2 sg. and 1-3 pl. also have a story to tell. The 1 sg. in PToch. *-wa, the 2 sg. in PToch. *-sta, and the 3 pl. in PToch. *-rə/*-rae, are clearly connected in some way with the endings of the PIE perfect; some older works even take them directly from the perfect.55 Apart from reduplicated participles of the type B peparku/A papräku 'asked' and kakautau/kākotu 'split' , however, the perfect has left no convincing traces in Tocharian. The h2e-conjugation theory, which posits the existence of presents and aorists with the perfect endings, opens the door to a more attractive possibility. Tocharian preterites continue a number of different formations: "normal" root aorists with *e : zero ablaut (e.g., B 3 sg. śarsa 'knew' , mid. kärsāte < *kersH-/*kr̥ sH-); reduplicated aorists (e.g., A śaśärs), originally thematic; and h2e-conjugation aorists of various types, including the relatively uncontroversial stative-intransitive type with *o : *e/zero ablaut (e.g., A 3 sg. lip, pl. lepar 'remained' < *loip-/*l(e)ip-).56 The endings *-wa, *-sta, etc. must have been generalized from the h2e-conjugation aorist component of the composite category that we know as the "preterite," in 53 Both are significant relics. Cf. Malzahn (2010: 43-44). *-rǝ (< *-r̥ s) is the more archaic variant; *-rae (< *-ront), like the Latin 3 pl. perf. in -runt, is a blend of *-r̥ (s) and the formally unrelated 3 pl. ending *-ont. A possible historical reason for the association of *-rǝ with class III and *-rae with the other classes is offered in note 83. 55 So, e.g., Krause-Thomas 1960: 247, Adams 1988.; but see HIEV 175 ff. 56 With, among other apophonic peculiarities, o-grade (lep-< *loip-) in the plural. See § 13. much the same way that the endings of the PIE perfect were generalized in the historical mélange that we know as the "perfect" in Latin.
As seen over a half-century ago by Ivanov (1959: 29-31) and Watkins (1962: 61ff., 99ff.), the class III preterite bears a striking resemblance to the preterite of the ḫi-conjugation in Hittite. Compare, e.g., the preterite of da-: sg. 1 dāḫḫun pl. dāwen 2 dātta dātten 3 dāš dāir Ivanov and Watkins were impressed by the isolation of the sigmatic 3 sg. in both languages, but had no useful framework for dealing with the numerically predominant s-less forms. These, however, are equatable as well. Outside the 3 sg., the endings of Hittite dāḫḫun, etc. are the secondary endings of the ḫi-conjugation/h2e-conjugation, matching the endings of Toch. *praek-(ə)wa, *-(ə)sta, etc., which, as just discussed, are the endings of a h2e-conjugation aorist. It is hardly likely that this combination-*-s-in the 3 sg., h2e-endings elsewhere-could have developed independently in Anatolian and Tocharian. We therefore have no choice but to assume, whatever we make of it, that a type of aorist with a 3 sg. in *-s(t) and h2e-inflection elsewhere already existed in PIE. In my 2003 discussion of the problem (HIEV ch. 7), I characterized this formation as a suppletive "presigmatic aorist," from which later emerged, by generalization of the *-s-throughout the active, the fully sigmatic aorist of the Inner IE languages.57 I did not at the time seriously consider the logical possibility, recently reintroduced into the discussion by Melchert 2015, that the *-s(-) of the 3 sg. was a desinence.

Desinential *-s?
As an alternative to assuming a presigmatic aorist for PIE, which he emphatically rejects, Melchert proposes (2015: 129) that "the original h2e-aorist third singular ending was simply *-s (thus already Watkins 1969: 54 and Yoshida 1993: 33-34), probably renewed already in PIE (though parallel independent renewal cannot be entirely excluded): *-s → *-s-t." He denies any connection between this desinential *-s and the *-s-of the s-aorist, which he assumes to have been a wholly separate formation, fully developed in PIE. In this respect he differs 57 So already in nuce Jasanoff 1988b. from at least Watkins, for whom the radiation of the classical s-aorist from the 3 sg. was a cardinal point of doctrine. From a purely Hittite point of view, -š is indeed the synchronic 3 sg. secondary ending corresponding to the 1 sg. in -(ḫ)ḫun, the 2 sg. in -(t)ta, etc. Melchert's distinctive move is to project this state of affairs back to PIE, thus eliminating the need for a separate explanation of the Hittite ending on the basis of the s-aorist. But the price of assuming both a desinence *-s and a formally unrelated "classical" s-aorist is very high. Since the paradigms dāḫḫun, dātta, dāš, etc. and prekwa, prekasta, preksa, etc. are "cognate," Melchert has to analyze the -s-of 3 sg. preksa as historically desinential as well.58 This forces him to separate the -s-of preksa from the -s-of a series of other, fully sigmatic paradigms in Tocharian that one might have thought were related. Thus, e.g., the productive form of the middle of the class III preterite, which has -s-throughout, is obviously inseparable from the s-aorist middle in Greek and Indo-Iranian; cf. Toch. B parksamai, -satai, -sate, 3 pl. parksante (A präkse, -sāte, -sāt, 3 pl. präksānt), exactly like Gk. elusámēn, -sa[s]o, -sato, Ved. aneṣi, -ṣṭhāḥ, -ṣṭa, and (NB!) OAv. frašī (< *préḱ-s-) 'I take counsel' . Melchert cautiously accepts the historical identity of these latter forms. He makes a distinction, however, between 3 sg. act. preksa and 3 sg. mid. parksate: preksa, he says, is a h2e-conjugation aorist with originally desinential -s-; parksate is the formally unrelated 3 sg. middle of a bona fide s-aorist with suffixal -s-. No respecter of Occam's Razor can be comfortable with this position. Also associated with the class III preterite is the fully sigmatic present type in *-se/o-(class VIII): cf. Toch. B 1 sg. preksau 'I ask' , 3 sg. prekṣäṃ, 3 pl. prekseṃ (A praksam, prakäṣ, prakseñc) beside pret. prekwa; further pres. naksau 'I destroy' beside pret. nekwa 'I destroyed' , etc. In HIEV 180-182 and earlier publications I identified these as historical s-aorist subjunctives-an analysis to which Melchert is apparently well-disposed. Here too, however, he has to deny any hint of a connection with the -s-of preksa.
At this point the reader may be tempted to ask whether a simpler, less stipulative version of Melchert's desinential *-s scenario might not be more satisfactory. Could the classical s-aorist have come into being, for example, through a post-IE Watkins-style reanalysis in which the desinential *-s of h2econjugation forms like *proḱ-s 'asked' , *noḱ-s 'destroyed' , etc. was reanalyzed as a tense sign? This would have the advantage of eliminating the full-blown s-aorist altogether in PIE proper (i.e., the common ancestor of Anatolian, Tocharian 58 It is clear from his account that he envisages a PIE 3 sg. *próḱ-s(-t), though the form is not explicitly spelled out.
Indo-European Linguistics 7 (2019) 13-71 and the "Inner" languages), and with it the awkward redundancies just noted. But the facts are not so simple. The classical s-aorist is standardly, and with good reason, reconstructed with Narten ablaut, with ē-grade in the strong forms of the active. Lengthened grade is also unambiguously present in the Tocharian class III preterite, where it is reflected by PToch. *-ae-with preceding palatalization (cf. Toch. A 3 sg. ñakäs, pl. ñakär 'destroyed' < *ñaek-(s-) < *nēḱ-(s-) (not < *noḱ-(s-)); Toch. B 3 sg. lyauksa, pl. lyaukar 'illuminated' < *lyaewk-(s-) < *lēuk-(s-) (not *louk-(s-); cf. Malzahn 2010: 200 ff.).59 If the class III preterite had simply been a h2e-conjugation aorist with o-grade in the strong forms and a 3 sg. in *-s, there would be no way to explain the ē-vocalism of these forms. This is why Melchert needs the s-aorist: the similar-looking reflexes of *noḱ-, *louk-(> PToch. *naek-, *laewk-), proper to the h2e-conjugation aorist, and *nēḱ-, *lēuk-(> PToch. *ñaek-, *lyaewk-), proper to the s-aorist, were confused, he says, through "the fatal merger of *o and *ē into … *ae." The result was the intrusion, in effect, of *nēḱ-/*ñaek-and *lēuḱ-/*lyaewk-, "borrowed" from the s-aorist, into the h2e-conjugation paradigm. To the extent the desinential theory is viable at all, it requires an external (i.e., non-h2e-conjugation) source for the lengthened grade of forms like Toch. A ñakäs. All this is a major reason why, in my view, the desinential theory is not viable. Melchert is correct that since the Tocharian endings *-wa, *-sta, etc. were generalized from a nucleus of h2e-conjugation aorists ( § 9), one might have expected the 3 sg. in *-s(t) also to have some kind of deep historical connection to the h2econjugation. But whatever the nature of this connection was, it cannot simply have been that *-s (→ *-st) was the regular Tocharian continuant of the 3 sg. h2econjugation secondary ending, parallel to 1 sg. *-wa and 2 sg. *-sta. If *-s had been the normal 3 sg. ending of a h2e-conjugation aorist, it would surely have been generalized, like *-wa and *-sta, throughout the Tocharian preterite, and not just to class III. The common-sense inference, rather, is that a subset of the numerous pre-Tocharian preterites that rested on h2e-conjugation aorists for some reason substituted *-s(t) for *-e in the 3 sg. active. h2e-conjugation aorists of this special "presigmatic" type were inherited into Anatolian as well. Within Anatolian, *-s(t) was eventually generalized from the presigmatic nucleus to all h2e-conjugation/ḫi-conjugation preterites in Hittite. It was lost in Luvian.
In support of his case for a desinential *-s in the 3 sg., Melchert cites the supposed identity of this ending with the *-s that he sets up in the 2 pl.: "The 59 As first pointed out by Ringe (1990: 185-186), palatalization is general in Toch. A but confined to the active. In Toch. B palatalization is lexically restricted but found in both voices. On B lyauksa vs. lauksāte see note 42.
presence of an ending *-s just in the third singular and second plural of the h2eaorist can hardly be a coincidence, since it matches the same peculiar distribution of the ending *-e in the h2e-present and is equally unmotivated" (2015: 129). But it is not at all clear that *-e, rather than some other sequence of the form *-(H)e/o, was the source of the supposed h2e-conjugation ending underlying the Vedic and Avestan 2 pl. perfect in -a (cf., e.g., Ved. vidá 'you (pl.) know').60 And in any case, whether or not the 3 sg. and 2 pl. perfect/h2e-conjugation present endings were segmentally identical, the purported 2 pl. in *-s, as we have seen, is a fiction, based on an incorrect interpretation of Hitt. -šten ( § 5) and Toch B -s, -so ( §6). From a wider theoretical point of view, a h2e-conjugation 3 sg. desinence *-s makes little sense. The endings of the perfect, h2e-conjugation, and middle are formally related; all go back to a pre-PIE "H-series," from which the familiar variants (e.g., 1 sg. perf. *-h2e, 1 sg. h2e-conj. pres. *-h2ei (pret. *-h2e), 1 sg. mid. pres. *-h2er (pret. *-h2e), etc.) emerged via a process of differentiation. The basic form of the 3 sg. ending is reconstructable as *-e or *-e/o. If, as claimed in effect by Melchert, this *-e or *-e/o had, so to speak, a "companion form" *-s that took its place in the h2e-conjugation aorist, the *-s would have to have been the secondary ending corresponding to primary *-e. But this is unlikely for several reasons: 1) nowhere else in the system of H-series endings is the primary : secondary distinction marked by anything more deeply embedded than a hic et nunc particle (*-i or *-r);61 2) the perfect, a category with only unrenewed and hence "secondary" endings (cf. 1 sg. *-h2e, not *-h2ei or *-h2er), has *-e, not *-s in the 3 sg. (*u̯ óide); 3) h2e-conjugation aorists of the stative-intransitive type likewise had *-e, not *-s in the 3 sg. This is why the reflexes of this type-the class V subjunctive (type A wekaṣ : *wikeñc; cf. §1) and class I preterite (A lip : lepar; §9) in Tocharian, and the passive aorist in Indo-Iranian (Ved. ádarśi : adr̥ śran; §1)-show no sign of sigmatic morphology (see below). In short, it is hard to avoid the conclusion that the hypothesis of a h2e-conjugation desinence *-s creates more problems than it solves.62 60 As authority for the reconstruction *-e in the 2 pl. Melchert cites HIEV 32, where *-(H)e is set up on the strength of Paelign. lexe 'you (pl.) have read' . But as Michael Weiss points out to me (p.c.), it is not out of the question that this form was simplified from *leg-s-te, with ending *-te. 61 In some cases the endings of the mi-series are brought in to help, as, e.g., in the pluperfect (Ved. 3 sg. perf. dīdaȳa 'shines' , plpf. ádīdet; etc.). 62 In two recent publications, Kümmel (2016Kümmel ( : 83-86, 2018, partly building on Indo-European Linguistics 7 (2019) 13-71

The presigmatic aorist: descriptive
In what follows we will use the term "presigmatic aorist" to refer to the subclass of PIE h2e-conjugation aorists in which the normal ending *-e was replaced by *-s(t) in the 3 sg. active. Aorists of this type, ex hypothesi, were the source of class III preterites in Tocharian and ḫi-conjugation preterites in Hittite. More can be inferred about them from other facts we know: 1) Neither Anatolian nor Tocharian has a classical s-aorist. Therefore, since a critical mass of Tocharian class III preterites correspond etymologically to s-aorists in the Inner IE languages (e.g., the preterites of näm-, päk-, Melchert's discussion of Hitt. -š, argues for a parallel desinence *-s in Indo-Iranian. The case is very tenuous. On the Indic side, the 3 sg. root aorist optatives in -yāḥ and *-īḥ (the latter presupposed by precatives of the type jéṣma 'we would win' and yeṣam 'I would travel') were originally proper to the s-aorist, and go back, as I have argued, to *-yāst and *-īšt, respectively; see HIEV 186-188 and the discussion of the presigmatic aorist optative at the end of this section. The Rigvedic hapax dhāyīḥ likewise looks very much like a precative. The isolated 3 sg. s-aorists ápāḥ (: pā-'drink') and aprāḥ (along with dhāḥ, atārīḥ, and a few other post-Rigvedic forms), so far as I can see, do nothing to strengthen the case for *-s as opposed to *-st. In Iranian, Avestan has the much-discussed optatives YAv. fratuiiāand *aiβituiiā(< *-āh) 'have power'; following Hoffmann 1967: 28-29, these are most often explained as 2 sg. forms. Avestan also has the t-less 3 sg. s-aorists āiš, xšnāuš, vąs, sąs (vs. dār ə št, with -t), but the value of these is undercut, as Kümmel recognizes, by the observable simplification of *-st to *-s in the imperfects as, cinas, and didąs. The situation is more puzzling in Old Persian. Here the normal secondary ending *-t has been uniformly replaced by -š in ruki environments; cf. 3 sg. impf. āiš, akunauš, adəršnauš and 3 sg. opt. kəriyaiš, vināθayaiš, fraθiyaiš. Given the crosslinguistic tendency of optatives to develop into iterative preterites, it is probably best to look for the formal origin of this -š in the optative of the s-aorist. As in Vedic, we can assume a remodeling of the 3 sg. s-aorist opt. in *-īt to *-īšt in the prehistory of Old Persian, with *-št subsequently spreading to the present system and taking on preterital value. On the basis of the 3 sg. in -š (< *-št) an analogical 3 pl. in -šan was created, seen, e.g., in mā yadiyaiša n 'they should not be sacrificed to' , akunavaša n beside akunava n 'they made' , etc. A 3 sg. *abara h is inferred by Kümmel from 3 pl. abara n , aba[ra]ha n 'they brought' , but the status of this form is uncertain. The possibility of an original 3 sg. in *-ast cannot be excluded. Kümmel further claims (2016Kümmel further claims ( : 87-88, 2018) that there was a formal relationship between the alleged Indo-Iranian 3 sg. in *-s/-š and the 3 pl. ending that he writes as *-r-s (so earlier Cowgill 1979: 39). It is true that the supposed 3 sg. in *-s is paired with a 3 pl. in *-rs in some Indo-Iranian optatives and in the PIE paradigm of the presigmatic aorist (see below), but the distribution of the two endings was not otherwise parallel. In my view, the *-s of *-rs was not a morpheme, but an organic part of the unanalyzable original 3 pl. ending of the H-series, like the *-t of the active 3 pl. in *-nt (cf. HIEV 32-34). tsäk-; also of pärk-, etc.; cf. §4), the natural assumption is that the presigmatic aorist, with its partially sigmatic paradigm, was the source of the fully developed classical s-aorist.
2) The Narten ablaut of the classical s-aorist and the ē-grade of Toch. A 3 sg. ñakäs, pl. ñakär, etc. show that lengthened grade must have been proper to some position or positions in the presigmatic paradigm. 3) Since the s-less, non-3 sg. h2e-conjugation component of the presigmatic paradigm had *o : *e/zero ablaut, the specific locus of the lengthened grade could only have been the sigmatic 3 sg. We are thus led to posit a PIE paradigm sg. 1 *próḱ-h2e pl. *próḱ-me-2 *próḱ-th2e *próḱ-(t)e-3 *prḗḱ-s-t *préḱ-r̥ s Compare the slightly different version in HIEV (178).63 It cannot be emphasized too strongly that this composite reconstruction, despite its aberrant and clearly heterogeneous 3 sg., is based on close-to-the-surface comparative evidence. Most of the details of the paradigm, including the *o : *e ablaut of the non-3 sg. forms, follow directly from the etymological equatability of the dāḫḫun and prekwa paradigms in Hittite and Tocharian, respectively. The lengthened grade of the 3 sg. is necessitated by the palatalization of ñakäs, -är in Tocharian and the ablaut of the s-aorist in the Inner IE languages (Ved. ávāṭ, Lat. uēxī, etc.). The basis for locating the "home" of ē-vocalism specifically in the 3 sg. will be taken up further in §14. Before considering the obviously looming question of how a paradigm of this type could have come into being, we must round out our synchronic picture with a survey of what we know about the distribution of sigmatic and non-sigmatic forms in the presigmatic aorist outside the active indicative. The overview below follows the general outlines of HIEV 179-188, with a few differences of emphasis and detail.
Two distinct middle paradigms are associated with the reflexes of the presigmatic aorist in our data. One is the fully sigmatic formation that we recognize 63 The paradigm in HIEV has e-grade rather than o-grade in the 1-2 -pl. But since the 1-2 pl. (and 1-3 du.) were strong positions in the "normal" root aorist (cf. Ved. ákarma, ákarta, etc.), it is more economical to assume they were strong in the h2e-conjugation root aorist as well (Jasanoff 2013: 108). The assumption of o-grade in the 1-2 pl. also comports better with the ablaut patterns of ḫi-verbs in Hittite (cf. Melchert 2013: 142 f.) and class I preterites and class V subjunctives in Tocharian (see §13).
Of the modal forms of the presigmatic aorist, the subjunctive is reflected, as I argued in HIEV, in the class VIII (*-se/o-) presents of Tocharian (type B preksau, A praksam). 68 Malzahn's discussion of the historical position of class VIII (2010: 429-432) shows (though this is not her intent) that there is no coherent alternative to this view. Since class VIII is fully sigmatic and the class III preterites are sigmatic only in the 3 sg., it is hardly credible that class VIII was created from class III within the internal history of Tocharian. Adams' objection (1994: 4f.) that PIE subjunctives are not normally the source of presents in Tocharian, though correct as far as it goes, is neutralized by the fact that the well-established class II (thematic) present of the root B klyaus-/A klyos-'hear' (B 3 sg. klyauṣäṃ, etc.) is historically the subjunctive (*ḱleu-s-e/o-) of a Narten s-present *ḱleū-s-(cf. §3).69 And while it is true that full-grade s-aorist subjunctives of the type *neḱ-s-e/o-, *dhegwh-s-e/o-, etc. would regularly have given palatalized *ñäk-s-, *śäk-s-, etc. rather than the attested näk-s-, tsäk-s-, etc., root-initial palatalization has been so heavily morphologized in Tocharian that its absence is never probative (pace Hackstein 1995: 159-165, andAdams loc. cit. and2012: 48f.).70 As mentioned earlier (cf. note 59), initial palatalization is also systematically absent in the Toch. A reflex of the s-preterite middle indicative (pattern 1), where the root certainly had e-grade.
It may seem a fool's errand to inquire into the fate of the presigmatic aorist subjunctive in Hittite, since the PIE subjunctive was lost in Anatolian. But its disappearance was not total. As seen above, the historical si-imperative paḫši was haplologized from a 2 sg. subj. *peh2sesi ( § 3), and the imperative nešḫut implies a si-imperative *nēši, haplologized from a 2 sg. subj. *neiHsesi 68 In this particular case, with analogical root vocalism (*-ae-) from the preterite and subjunctive. 69 It can hardly be accidental that the same inherited subjunctive that underlies the siimperative päklyauṣ and the thematic present klyauṣäṃ in Tocharian was also the source of a si-imperative and a thematic present in Vedic (si-impv. ( §4). It is no accident that the unique nešḫut is specifically a form of nai-. All ḫi-conjugation verbs belong to one of three types: 1) former h2e-conjugation presents, 2) former h2e-conjugation aorists of one of the non-presigmatic varieties (e.g., stative-intransitive aorists), and 3) former h2e-conjugation aorists of the presigmatic type. Owing to the productivity of the 3 sg. pret. ending -š, which spread from the presigmatic type to the ḫi-conjugation as a whole, it is easy to make the mistake of supposing that the three "input" classes fell together in every other respect as well. That this was not the case is shown by nešḫut, *nēši and their analogical spinoffs naišten, naišdumat, etc. ( § § 4-5). PIE *neiH-belonged to the small nucleus of roots that passed into Anatolian with a presigmatic aorist. It thus inherited, along with the indicative *nóiH-h2e, *nóiH-th2e, *nḗiH-s-t, etc., a subjunctive *néiH-s-e/o--the Hittite counterpart, so to speak, of a Tocharian class VIII present. Unlike the 3 sg. indicative in *-s-t, the subjunctive in *-se/o-never spread to other ḫi-verbs. But the haplologized 2 sg. subj. *neiH-si/*nēši survived long enough to engender nešḫut, *nešten (→ naišten), and, ultimately, the phenomenon of intrusive -s-. See further § 16. The subjunctive of the PIE presigmatic aorist, then, was built to the sigmatic "stem" of the indicative. It was otherwise with the optative. Hittite, where the optative has well and truly disappeared, has nothing to tell us here; our sources of information are Tocharian and, unexpectedly, Indo-Iranian. The formation of the Tocharian optative is usually passed over quickly because of the synchronic rule, repeated in all the major handbooks, that the optative is formally based on the subjunctive stem. Thus, e.g., the 3 sg. opt. of pärk-'ask' in Toch. B is parśi (< pre-Toch. *préḱ-ih1-t), descriptively built to the (non-sigmatic) class I subjunctive 3 sg. prekäṃ, pl. parkäṃ (< pre-Toch. *próḱ-/*préḱ-). But synchronic regularities have historical explanations, and in this case the explanation is of considerable interest. The class I subjunctive whose optative was *préḱ-ih1-was a reflex-a second reflex, as we shall see-of the very same h2econjugation aorist (*próḱ-/*préḱ-) whose indicative furnished the non-sigmatic h2e-conjugation stem of the class III preterite (prekwa, etc.). The process of paradigm split by which h2e-conjugation aorists yielded preterites and subjunctives in Tocharian will be discussed in §13.

Optative
Indicative As in Narten presents; cf. 3 sg. opt. *u̯ élh1-ih1-t (> Lat. uelit 'wants' , Go. wili 'id.' , OCS -velitъ 'orders') : *u̯ elh1-'choose' . 72 Wrongly listed without the asterisk in HIEV. Depending on the interpretation of the Avestan and Old Persian optatives discussed in note 62, the Proto-Indo-Iranian preform corresponding to YAv. vainīt̰ would have been *u̯ ánīt (so HIEV ) or *u̯ ánīšt. 73 The classic discussion of these forms is by Hoffmann (1967). 74 That such a stage would have existed is not in doubt, since the sigmatization of the indicative was of Inner IE date and final *-st survived into Proto-Iranian. 75 With quasi-regular failure of Stang's Law, the IE-internal rule by which /-VHm/ "should" have developed via *-Vmm to *-Vm. Stang's Law never applies in the 1 sg. optative, suggesting the possibility that *h1 was not part of the environment for the rule.

12
The progress of sigmatization: overview The progressive sigmatization of the presigmatic aorist from the more to the less archaic branches of the family can be displayed as follows: jasanoff Indo-European Linguistics 7 (2019) 13-71

The presigmatic aorist in Tocharian
The presigmatic aorist split into two daughter paradigms in Tocharian, one a preterite (class III), and the other a subjunctive (class I). This double treatment was briefly noted above in connection with the optative *préḱ-ih1-(> Toch. B 3 sg. parśi), where the h2e-conjugation aorist stem *próḱ-/*préḱ-was the source of both the class I subjunctive (3 sg. prekäṃ, pl. parkäṃ) and the non-sigmatic component of the class III preterite (prekwa, prekasta, etc.). It is time now to examine this pattern more systematically. Tocharian subjunctives, which in function combine modal and future value, go back to former indicatives. Sometimes the underlying indicative is a present. When this is the case, the synchronic subjunctive and synchronic present can stand in various possible formal relationships to each other. Thus, e.g., the two stems can be identical, as in the case of many presents in -sk-, e.g., B anā-sk-(pres. and subj.) 'breathe'; or the present can be a recharacterized form of the stem that underlies the subjunctive, e.g., B kär-nā-sk-(pres.) vs. kär-nā-(subj.) 'buy' (PIE pres. *kwri-n(e)-h2-); or the present and the subjunctive can be based on completely separate stems, making for no systematic formal relationship at all, e.g., A ar-s-(pres.) vs. ar-ñ-(subj.) 'bring forth' . In other cases the subjunctive goes back to an aorist. Here the functional development was evidently by way of the "injunctive" uses of the aorist. Aorist injunctives, to the extent they were employed in a modal or future capacity, were brought into morphological alignment with present-based subjunctives in several ways-most obviously by substituting the primary (= present) for the secondary (= preterite/ injunctive) endings.79 Three major varieties of PIE aorist were "subjunctified" in Tocharian-normal root aorists with *e : zero ablaut, h2e-conjugation aorists of the stative-intransitive type, and presigmatic aorists.80 Although our focus below will be on the presigmatic aorist, it will be useful first to study the treatment of the other two types.
The history of the class V subjunctive and class I preterite is instructive because it shows how one and the same aorist paradigm could split into very different-looking subjunctive and preterite daughter paradigms. Other aorists had the same twofold treatment, both individual lexical items92 and, more to 88 ə-grade in the pret. sg. imported from the kärs-type. 89 o-grade in the pret. pl. imported from the wik-type. 90 With inherited strong vocalism in the 1-2 pl. of the aorist. 91 o-grade analogically extended to the 3 pl. from the 1-2 pl. 92 This was the case, in my view, with käm-'come' , which inherited an active root aorist from PIE (cf. Ved. 3 sg. ágan, pl. ágman). The 1 sg. of this stem developed a long *-ē-by sound change (*gwem-m > *gwēm), leading to a pre-Tocharian paradigm with three root the point, the core membership of the presigmatic type. The inner-Tocharian history of the presigmatic aorist (using *neḱ-instead of *preḱ-to show palatalization effects) began with the paradigm given in § 11: sg. 1 *nóḱ-h2e pl. *nóḱ-me-2 *nóḱ-th2e *nóḱ-(t)e-3 *nḗḱ-s-t *néḱ-r̥ s From this array emerged a nascent subjunctive and a nascent preterite. In the subjunctive, 1 sg. *nóḱ-h2e and 2 sg. *nóḱ-th2e were remade to *nok-mi and *noksi (vel sim.), respectively, with the primary endings; 1 pl. *nóḱ-me-and 2 pl. *nóḱ-(t)e-were remade to *nek-me-and *nek-te, with weak root vocalism; and 3 pl. *néḱ-r̥ s became *nek-n̥ ti, again with the primary ending. In the 3 sg., the theoretically expected form would have been *nēk-s-ti, based on inherited *nḗḱ-s-t. But in an emergent subjunctive paradigm that was in the process of becoming distinct from the emergent preterite, another possibility for the 3 sg. would have been *nok-ti, the analogical o-grade companion to 1 sg. *nok-mi and 2 sg. *nok-si. It was this *nok-ti that was selected, creating the canonical class I subjunctive alternation of strong o-grade (later ae-grade) and weak e/zero grade (later ə-grade), as in class V (cf. above). The preterite was likewise subject to morphological simplification. Unlike the Inner IE languages, which generalized the stem *nēḱ-s-throughout the active indicative, Tocharian generalized the lengthened grade throughout the preterite paradigm without generalizing the *-s-. It thus maximized the distinctiveness of the emergent preterite while maintaining its decidedly presigmatic character. The vowels *ē and *o subsequently fell together by sound change, but the generalization of *-ē-throughout the active is confirmed by the palatalization of forms like Toch. A 3 pl.  sg. 1 *naek-< *nok-< *noḱ-*ñaek-< *nēk-← *noḱ-2 *naek-< *nok-< *noḱ-*ñaek-< *nēk-← *noḱ-3 *naek-< *nok-← *nēk-s-*ñaek-s-< *nēk-s-< *nēḱ-spl. 1 *nǝk-93 ← *neḱ-← *noḱ-*ñaek-< *nēk-← *noḱ-2 *nǝk-← *neḱ-← *noḱ-*ñaek-< *nēk-← *noḱ-3 *nǝk-← *neḱ-< *neḱ-*ñaek-< *nēk-← *neḱ- The intimate relationship between the class III preterite and the class I subjunctive, like the relationship between the class I preterite and the class V subjunctive, is essential to understanding the historical position of these forms (cf. further §15). It is impossible to explain the shape of the class III preterite without reference to the shape of the class I subjunctive, and vice versa: the two categories, historically speaking, are one and the same. Failure to appreciate the position of the class I subjunctive has been one of the reasons for the surprising lack of focus or consensus in the literature on the class III preterite; see the summary in Malzahn 2010: 208-212. Much of this literature reflects an older stage in the "culture" of Tocharian studies, when Tocharian tended to be treated as a grabbag of exotic-looking forms, to be extracted from their morphological context and made sense of by mechanically projecting them back into the protolanguage. Within Tocharian grammar, this approach is particularly ill-suited to the verbal system, where inherited ablaut and palatalization patterns were systematically altered in the course of morphological restructuring.

The origin of the presigmatic aorist
The presigmatic aorist, as described above and in earlier publications, was transparently a suppletive category, made up of forms that originally belonged to two different paradigms. The core of the active indicative was a h2e-conjuga-

93
With non-palatalization before the weak ə-grade. It is hard to know whether this was simply analogical to the lack of palatalization in the singular, or whether a non-palatalizing *-ə-was actually substituted for pre-Toch. *-e-at a date prior to the phonological change of *e to PToch. *ə.
tion root aorist that also supplied the optative, while the 3 sg. active, along with the whole subjunctive, came from a sigmatic formation with Narten ablaut, strongly recalling the Hittite type ganešzi. Since the very idea of a suppletive s-aorist has struck some observers as a priori unlikely, it is important to bear in mind that suppletion has been part of the discourse surrounding the s-aorist for the better part of a century. Under the long-standard "perfect" theory of the ḫi-conjugation, ḫi-conjugation preterites like Hitt. dāḫḫun, dātta, dāš, etc. were taken to be a mixture of the classical s-aorist and the perfect. Exactly the same explanation was given for the Tocharian s-preterite: "in dem toch. s-Pt. sind der idg. s-Aorist … das idg. Perfekt mit Abtönung des Wurzelvokals … sowie nur im Otoch. der idg. athematische Wurzelaorist … kontaminiert" (Krause-Thomas 1960: 247); "those PIE perfects which did not take the preterital -ābecame amalgamated with the sigmatic aorist" (Adams 1988: 82). The novelty of the presigmatic aorist hypothesis is that instead of assuming suppletion twice, once in Anatolian and once in Tocharian, it assumes suppletion once, in PIE itself.
It goes without saying that, as a general principle, caution has to be exercised in setting up mixed paradigms in protolanguages; inner-paradigmatic suppletion is so potent a "trick" that it can be made to generate a pseudo-explanation for almost anything. Nevertheless, as an actual phenomenon that occurs in real languages, suppletion cannot in principle be disallowed for the languages or language fragments we recover by comparative reconstruction. The heteroclitic ("r/n-stem") declension (e.g., *u̯ ód-r̥ , gen. *u̯ éd-n̥ -s 'water') is a case in point. The alternation of r-and n-stem forms must already have been present in the parent language; it is unthinkable that precisely the same suppletive pattern could have arisen separately in the individual branches. The same is true of the presigmatic aorist. It is, of course, logically possible that the association of o/e-ablauting root-based forms and Narten-ablauting s-forms was due to parallel innovations in Anatolian, Tocharian, and (to a limited extent) Indo-Iranian. But anyone who makes this claim must explain a) why it was precisely these two formations, and not, e.g., the "normal" root aorist or the perfect, that independently combined into a single paradigm in both Anatolian and Tocharian;94 94 Melchert's approach, by positing a 3 sg. ending *-s and avoiding direct reference to suppletion as such, might seem to be exempt from this and the questions that follow. But both in his acceptance of Toch. parksate and Hitt. nešḫut as reflexes of the classical s-aorist and in his contamination-based account of the palatalization in the active of the Tocharian s-preterite (cf. §10), Melchert assumes a merger or intersection of two paradigms as well.
b) why it was specifically the 3 sg., and no other form in the active indicative paradigm, that was independently targeted for sigmatic morphology in both branches; c) why, outside the indicative, the subjunctive was sigmatized in both branches (cf. Hitt. nešḫut < *néiH-s-e/o-), not just in the 3 sg., but throughout the paradigm; and d) why the optative nevertheless remained non-sigmatic in both Tocharian and-notably-Indo-Iranian. The scenario that would have to be devised to explain these details as independent innovations would be like the independent creation scenario for r/nstems-too inefficient and repetitious to be viable.
Sometimes the historical background of a suppletive relationship-or of any synchronic anomaly-is known; sometimes it is not. In HIEV 179 I laid stress on the importance of distinguishing between the synchronic grammar of late PIE, which is partly accessible to us through the comparative method, and the historical rationale or 'explanation' for this synchronic grammar, which we can only recover by internal reconstruction … It would clearly be desirable, at a deeper level of analysis, to know why the expected 3 sg. *próḱ-e 'asked' and *nóiH-e 'led' were replaced by *prḗḱ-s-t and *nḗiH-s-t in pre-PIE … But the problem of motivating the replacement *prók-e, *nóiH-e ⇒ *prḗḱ-s-t, *nḗiH-s-t must be viewed in perspective from the outset. Even if it should prove impossible, six or seven thousand years after the fact, to discover the rationale for the peculiar PIE mixed inflection of the inaccurately named 's-aorist' , it is still far simpler to operate with a single unexplained suppletion in the parent language than to assume two separate and unexplained suppletions, one in Hittite and one in Tocharian.
The object of this frankly speculative account was not to convince anyone of its literal facticity, but to show how a mixed paradigm of the type seen in the presigmatic aorist could have come about through a succession of ordinary changes within the PIE system. Some parts of the proposed scenario were more compelling than others. I still think it likely that the specific root *preḱ-, which had its foot in both the sigmatic and h2e-aorist doors, so to speak, was prominently involved in the blending process that created the presigmatic paradigm. On the other hand, Melchert is entirely right to point out (2015: 128) that there would have been much simpler ways to clarify the active, transitive nature of forms like 3 sg. *nóḱ-e and *nóiH-e than by remaking them to *nḗḱ-s-t and *nḗiHs-t, respectively. Why, e.g., were the active forms not just renewed as *nóḱ-et, *nóiH-et or *nóḱ-t, *nóiH-t?95 And how sure can we be that pairs of the type *nóḱ-e : *nóḱ-o and *nóiH-e : *nóiH-o ever existed in the first place? As long as there are no satisfactory answers to these questions, Melchert's verdict on the HIEV scenario-"not remotely credible" (133)-is a legitimate opinion that others will no doubt share. What Melchert actually says, however, goes further: I contend that the attested facts of Tocharian, Hittite, and "Core Indo-European" can be accounted for without attributing to Proto-Indo-European a hybrid "pre-sigmatic" aorist created by an unmotivated and not remotely credible suppletion.
Here, I submit, is an instance of the kind of confusion I cautioned against above: blurring the distinction between a well-grounded, evidence-based reconstruction-the presigmatic aorist-and the scenario we speculatively invent to explain how it may have come about. To appreciate the methodological point, the reader should substitute the words "hybrid 'heteroclitic' declension" for "hybrid 'pre-sigmatic' aorist" in the lines just quoted.
As it happens, the HIEV scenario for the origin of the presigmatic aorist can be substantially improved with a few changes of causation and relative chronology. I will continue, in what follows, to assume a split of h2e-conjugation aorists of the type *nóiH-/*néiH-, *nóḱ-/*néḱ-, etc. into separate transitive and intransitive paradigms, with the inherited h2e-conjugation forms mostly being interpreted as active and transitive, and new middle forms mostly being created by adding the normal middle endings to the weak stem (1-2 sg. *néiH-h2e, *-th2e, pl. *néiH-medhh2, *-dh(u)u̯ e, *-ro). The crucial difference between the account in HIEV and what I would now propose lies in the treatment of the 3 sg. form. In the "old" scenario I envisaged a split of 3 sg. *nóiH-e into an active *nóiH-e 'led' and a middle *nóiH-o 'turned (intr.)' , with retained o-grade. Later, I said, the near-homophony of *nóiH-e and *nóiH-o made it necessary for one of the two to be replaced. A question left unanswered was why, if the near-homophony of 3 sg. act. *nóiH-e and 3 sg. mid. *nóiH-o was to prove so inconvenient in the not-too-distant future, the new 3 sg. middle form was not simply remade as e-grade *néiH-o from the beginning, with the same vocalism as the rest of the middle. Everything falls into place if we instead assume that pre-split h2e-conjugation forms of the type 3 sg. *nóiH-e, *nóḱ-e were themselves assigned to the emerging intransitive middle paradigm. As pointed out by Villanueva Svensson (2006: 297-299), word equations like 3 sg. OCS pade (< *pode) 'fell' = Ved. 3 sg. "passive" aorist padi 'id.' strongly suggest that the PIE form of the dentalless ("stative") 3 sg. middle ending in the root aorist was not *-o, as assumed in HIEV, but *-e.96 Hittite and, separately, Tocharian later modernized 96 The same idea was independently suggested to me at around the same time by Yaroslav this to *-o, which in turn became *-to in Tocharian (cf. Hitt. nē(y)at 'turned' , with back-formed present nē(y)a, nē(y)ari; Toch. A nakät 'perished' < PToch. *naektae < *noḱto). Under the new scenario, late PIE 3 sg. *nóiH-e and *nóḱ-e would have meant 'turned (intr.)' and 'perished' , respectively, and the replacement of intransitive *-e by *-o would have been an event of the post-PIE period.
If it is indeed the case that *-e was specialized with middle and/or intransitive value in the h2e-conjugation root aorist, an obvious difficulty would have arisen in the marking of transitivity in the 3 sg. While a distinction between transitive active and intransitive middle forms would have existed in most paradigmatic positions (1 sg. act. *nóiH-h2e ≠ mid. *néiH-h2e; 2 sg. act. *nóiH-th2e ≠ mid. *néiH-th2e; 1 pl. act. *nóiH-me-≠ mid. *néiH-medhh2; 2 pl. act. *nóiH-(t)e-≠ mid. *néiH-dh(u)u̯ e; 3 pl. act. *néiH-r̥ s ≠ mid. *néiH-ro), the 3 sg. had only *nóiH-e, historically bivalent but tending increasingly to become aligned morphologically with the intransitive middle. It was this problem that the introduction of the unequivocally transitive sigmatic form *nḗiH-s-t was intended to solve. The question is why, once the inherited 3 sg. *nóiH-e stopped meaning 'led' , the specific form that replaced it should have been taken from the imperfect of a Narten s-present.97 Gorbachov, who further compared Slavic -bъde 'awoke' with Ved. ablauting ábodhi, pl. ábudhran 'id.' (see now Gorbachov 2014: 52-53). My reluctance to entertain this idea in HIEV was rooted in my view that at the time of the breakup of PIE, *-o was consistently "middle" and *-e was consistently "active." If true, this would have implied that the ancestor of the passive aorist, which had to be middle on semantic grounds, also had to end in *-o. Villanueva Svensson, however, has made a strong case (op. cit. and 2007-2008 [2009], 2010-2011 [2012]) that the canonical alignment of *-e with the functions of the active and *-o with the functions of the middle, though valid for the present system, did not hold in the root aorist. Specifically, he argues that the 3 sg. aor. in *-e was always either specifically middle in the narrow sense (as, e.g., in *dóh3-e 'took' < *'gave to oneself') or, more typically, intransitive (as in the stative-intransitive aorist: *u̯ óh2g-e 'broke' , *pód-e 'fell' , *bhóudh-e 'awoke' , etc.). There is no evidence at all for a transitive active 3 sg. aor. in *-e apart from the purely hypothetical *nóiH-e 'led', *nóḱ-e 'destroyed' , etc. These, of course, are the very forms whose purported near-homophony with the supposed intransitive middles *nóiHo 'turned (intr.)' and *nóḱ-o 'perished' was supposedly responsible for their replacement by *nḗiH-s-t and *nḗḱ-s-t in the scenario presented in HIEV. 97 The problem becomes clear when the final paradigms are put side by side: act. sg. 1 *nóiH-h2e mid. sg. 1 *néiH-h2e 2 *nóiH-th2e 2 *néiH-th2e 3 *nḗiH-s-t 3 *nóiH-e ⋮ ⋮ pl. 3 *néiH-r̥ s 3 *néiH-ro In the 3 sg., the form that "should" have been assigned to the active paradigm (*nóiH-e) was Indo-European Linguistics 7 (2019) 13-71 The adoption of *nḗiH-s-t as the new transitive 3 sg. is best understood by considering the spread of the sigmatic 3 sg. and the spread of the sigmatic subjunctive (*néiH-s-e/o-, *néḱ-s-e/o-, *préḱ-s-e/o-, etc.) as parts of a single process. The -sḱe/o-presents *h2is-sḱé/ó-and *pr̥ ḱ-sḱé/ó-, as we have seen, probably replaced earlier Narten s-desideratives *h2eīs-s-'seek' and *preḱ-s-'ask' (cf. § 2 with note 13). In Greek, the desiderative (> future) marker -s-has the effect of neutralizing the present : aorist opposition: a form like grápsō 'I will write' can equally well be paraphrased as méllō gráphein 'I am about to write/be writing' (present infinitive) and as méllō grápsai 'I am about to write/get written' (aorist infinitive). On the assumption that the *-s-of *h2eīs-s-and *preḱ-s-had the same property, the pre-PIE subjunctives *h2éis-s-e/o-and *préḱ-s-e/o-would have had two readings, one more presential ('be inclined to (want to) be looking for' , 'be inclined to (want to) ask'), and the other more aoristic ('be inclined to (want to) find/obtain' , 'be inclined to (want to) get an answer'). In their "presential" value, these subjunctives would have retained their synchronic link to the underlying sigmatic presents *h2eīs-s-and *preḱ-s-, even as the latter forms were gradually restricted to "narrative" (i.e., less vivid, more aorist-like) imperfect functions. But the aoristic readings 'be inclined to (want to) find/obtain' and 'be inclined to (want to) get an answer' would effectively have merged with the non-sigmatic aorist subjunctives corresponding to the h2e-conjugation root aorists *h2óis-/*h2éis-and *próḱ-/*préḱ-.98 From the synchronic point of view of a speaker, then, the aorist inflection of the roots *h2éis-and *preḱ-would have included 1) an ordinary h2e-conjugation root aorist indicative (e.g., *próḱ-h2e, etc.);99 2) an ordinary h2e-conjugation root aorist subjunctive (*préḱ-e/o-); and 3) a sigmatic subjunctive (*préḱ-s-e/o-) which also-via the "presential" reading of the underlying desiderative-served as the subjunctive of the associated narrative imperfect (*prḗḱ-s-m̥ , etc.).
forced by the intransitive specialization of the ending *-e into the middle, and a Narten sigmatic form (*nḗiH-s-t) was generated in the 3 sg. active to take its place. 98 That is, the meanings "be inclined to want to get X done," proper to the aoristic reading of the sigmatic subjunctive of the desiderative, and "be inclined to get X done," proper to the non-sigmatic subjunctive of the h2e-conjugation aorist, would have been operationally indistinguishable. h2e-conjugation aorists can be assumed for *h2eis-and *preḱon the strength of the s-aorist profiles of these roots in the later languages (cf. OAv. s-aorist injunctive 3 sg. āiš 'wishes'; 1 sg. mid. frašī 'I take counsel' , Ved. aor. 3 sg. áprāṭ 'asked' , Toch. B prekwa, etc.). 99 Since the sequence *h2éis-s-would have fallen together with *h2éis-by the PIE degemination rule (cf. 2 sg. *h1ési 'you are' < *h1és-si), the discussion will henceforth focus on *preḱ-.
Indo-European Linguistics 7 (2019) 13-71 *-i. Elsewhere, including in Hittite and Tocharian, it was renewed as *-o (> *-to), the ending that played the same role in the present system. The middle of Hitt. nai-is largely built on a stem nē(y)a-, which was extracted from the 3 sg. pret. nē(y)at(i) and its back-formed present nē(y)a, nē(y)ari. Whether nē(y)at(i) and nē(y)a(ri) rest directly on a mechanically back-formed *nóiH-o(r), with o-grade, or on an e-grade preform *néiH-o(r), with vocalism leveled from the rest of the paradigm, is impossible to tell. What is certain is that despite the thematic appearance of Neo-Hittite forms of the type 1 sg. neyaḫḫari or 3 sg. pret. neyattat, the well-attested Old Hittite forms in -a(ri) and -at(i) are unambiguously athematic.101 There is thus no possibility of a close connection between Hitt. nē(y)a-and the true thematic present represented by Ved. náyati and YAv. naiieiti. As already noted, the absence of evidence for a thematic stem *néiH-e/o-in Hittite is of a piece with the absence of evidence for thematic presents beside class III preterites in Tocharian.
The treatment of the presigmatic aorist middle of bivalent verbs in Tocharian provides yet another example of the now-familiar split of an aorist into a preterite and a subjunctive. Just as the presigmatic active gave rise to both the class III preterite active and the class I subjunctive, and just as the other root aorist types gave class I preterites and class V subjunctives (cf. § 13), the middle of the presigmatic aorist was simultaneously parent to the "class 0" preterite and the class III subjunctive, both exclusively middle formations. The class 0 preterite is seen in Toch. A 3 sg. mid. nakät, pakät, tsakät, tamät, and a handful of other intransitive forms (cf. §7); here too belong 3 pl. tamänt and lyokänt (with secondary ly-for *l-). In Toch. B, as we have seen, the corresponding preterites were secondarily sigmatized, but are still recognizable by their o-grade: neksate (3 pl. -ante), temtsate (1 sg. temtsamai, 1 pl. -amte, 3 pl. -ante), lauksāte, etc. The main formal development in the history of these class 0 middles, apart from the replacement of *-e by *-(t)o and other desinential renewals, was the generalization of o-grade from the 3 sg. to the rest of the paradigm. The parallel with the corresponding preterite active (class III), where the ē-vocalism of the 3 sg. was generalized, could hardly be clearer.
Given the generalization of o-grade as the root vocalism in the class 0 preterite, the form of the subjunctive that would emerge from the presigmatic aorist middle might almost have been predicted on grounds of patterning alone. Here it was the e-vocalism of the non-3 sg. forms (1 sg. *néḱ-h2e, 3 pl. *néḱ-ro, etc.) that was extended to the historically o-grade 3 sg. In keeping with Indo-European Linguistics 7 (2019) 13-71 less type is lexically restricted and specifically valency-reducing, is no doubt an archaism vis-à-vis the situation in Indo-Iranian and Greek. Yet even in these branches we find echoes of "pattern 2." Such a case is the Vedic pair ahauṣīt (MS) 'poured ' (cf. si-impv. impv. hoṣi (RV)) vs. passive aorist áhāvi, matching Gk. act. ékheua 'id.' (< *ǵhéu-s-) : mid. khúto.104 The antiquity of the presigmatic aorist in this verb is underscored by the Tocharian class III pret. A 3 sg. śosā-ṃ 'poured' beside class I subj. B 1 sg. kewu 'I will pour' .105 Whether Anatolian also once had a fully sigmatic middle like Tocharian is unclear. Only the naiš : nē(y)at(i) = ñakäs : nakät pattern (pattern 2) is actually attested, but pattern 1 could theoretically once have been present in Hittite as well. It is suggestive that pattern 1 is associated with (inter alia) the root *preḱ-, a "donor" of *-s-in our framework, while pattern 2 is associated with the roots *neḱ-and *neiH-, which were *-s-"receivers." It is thus not out of the question that at the time in pre-PIE when the 3 sg. active of the narrative imperfect *prḗḱ-s-m̥ , *-s-s, *-s-t, etc. was revalorized as a transitive 3 sg. aorist (*prḗḱ-st), the middle of the narrative imperfect (*préḱ-s-h2e, *-s-th2e, *-s-to, etc.) was put to use as a specifically non-intransitive, self-benefactive aorist middle. In the absence of any actual sigmatic forms of this type in Hittite or elsewhere in Anatolian, however, it is difficult to be sure.

16
The root *neiH-in Hittite The HIEV picture of the presigmatic aorist, an updated version of which has been presented above, places a great deal of emphasis on the profile of Hitt. nai-. The question now arises: how reliable is the testimony of this verb? Is this one lexical item in fact a unique repository of archaic features, or is too much being made of a handful of scattered oddities? In approaching this question, the first point to understand is the sense in which nai-is special. nai-is the only verb in Hittite that is linked by word equation to an inherited s-aorist elsewhere in the family (Ved. naiṣ-/neṣ-, Av. naēš-). This does not mean (although it is a useful worst-case assumption) that nai-was the only verb from which the 3 sg. in *-s or *-st spread to the ḫiconjugation as a whole. At the time of the pre-Hittite propagation of *-s(t) it Indo-European Linguistics 7 (2019) 13-71 The odds that the uniquely s-rich morphological profile of nai-in Hittite is unrelated to its particular history and etymology are vanishingly small. Suppose, as a thought experiment along the lines of Kloekhorst's probabilistic speculations in §5, that the only facts we knew about PIE *neiH-were that it made a sigmatic or presigmatic aorist in PIE and that it had a lexical reflex in Hittite. Morphologically speaking, what would we expect a purely random Hittite continuant of *neiH-to look like? Would it have been a mi-verb or a ḫi-verb? The mi-conjugation, according to Kloekhorst's statistics, is almost three times as populous as the ḫi-conjugation; yet nai-is a ḫi-verb, with a 3 sg. pret. in -š. Is it credible that nai-, the only Hittite verb etymologically linked to a (pre)sigmatic aorist, would also have been-by pure chancethe only Hittite verb with intrusive -s-in the 2 sg. imperative and in the middle (*nēši, nešḫut, naišdumat)? Is it likely, a priori, that the general phenomenon of intrusive -s-would have been specifically associated with verbs of the same diphthongal shape as nai-? No actual calculations are needed to answer these questions. The special properties of nai-cannot be explained on the basis of a dubious 3 sg. s-ending that would have been common to all ḫiverbs. The inflection of this verb in Hittite only makes sense as a continuation of the special paradigm that also gave rise to the Tocharian s-preterite and the Inner IE s-aorist-a h2e-conjugation root aorist with suppletive sigmatic forms.