9.1 Uhlenbeck’s Ergative Hypothesis

In 1901, the Dutch comparative and anthropological linguist Christianus Cornelius Uhlenbeck published a short note entitled “Agens und Patiens im Kasussystem der indogermanischen Sprachen”, which opened a new chapter in Indo-European philology. Highlighting the identity of the nominative and accusative cases in all neuter paradigms and the identity of the neuter nominative-accusative ending with the non-neuter accusative ending in the o-stem inflection, Uhlenbeck concluded

Languages that feature the alignment system Uhlenbeck describes are classified as ‘ergative’ in more recent typological literature. Figure 9.1 visualises how they differ from ‘accusative’ languages, the type to which most ancient and modern Indo-European languages belong.Footnote ² As before (4.35, 4.41), the subject of intransitive sentences is labelled as S, the (logical) subject of transitive sentences as A, and the (logical) direct object of transitive sentences as O. Instead of Uhlenbeck’s activus and passivus, the terms ‘ergative’ and ‘absolutive’ are nowadays used for the relevant cases in an ergative system.Footnote ³

Fig. 9.1. ‘Accusative’ vs. ‘ergative’ alignment systems

On the assumption that neuter nouns, which prototypically refer to inanimate things, will readily occur in the S or O slots, but less readily in the A slot,Footnote ⁴ and that non-neuter nouns, which prototypically refer to animates, will commonly occupy the A slot, the distribution of the basic PIE case endings referred to by Uhlenbeck indeed provides a better fit with an ‘ergative’ than with an ‘accusative’ system (Fig. 9.2).

9.2 Chapter Outline

Figure 9.2 represents an idealised state of affairs. It ‘forgets’, for example, that non-neuter nouns may also occur in the S slot. Nevertheless, the general distribution is as noteworthy as it seemed to Uhlenbeck and its implications have been duly discussed by subsequent scholars. The present chapter will retrace this debate and show how what started as an attempt to elucidate the prehistory of the PIE nominal system (9.4–9.11) soon acquired a verbal dimension as well (9.12–9.14).Footnote ⁵ In our context, this latter aspect will be of particular importance. To sketch the history of the problem is just a way of preparing the ground before taking up again the loose ends left in the preceding chapters. We shall do this by first returning to the s-aorist (9.28–9.31); but afterwards, in Chapter 10, we shall also revisit all the other categories we have looked at throughout this study.

Fig. 9.2. PIE case endings and accusative vs. ergative alignment

9.3 On Studying ‘Pre-Proto-Indo-European’

From now on, our primary focus will thus be on a remote stage of Proto-Indo-European. For convenience, this stage will be called ‘Pre-Proto-Indo-European’ (‘Pre-PIE’). The defining difference between our ‘Proto-Indo-European’ and this ‘Pre-Proto-Indo-European’ will be the alignment structure, for an updated version of Uhlenbeck’s ergative theory will eventually be adopted here. It hardly needs stressing that some conclusions must remain tentative when we operate at such depths of time. This does not, however, make every exploration of Pre-Proto-Indo-European pointless. As long as the undertaking promises to refine our understanding of structural oddities in the later proto-language and its historical descendants, like the morphological point noticed by Uhlenbeck, it would be wrong to abandon it merely because it has not always been pursued with the necessary methodological rigour.

9.4 Van Wijk on Genitives, Impersonal Verbs, and Pronouns

Soon after the publication of Uhlenbeck’s piece, one of his pupils, Nicolaas van Wijk, went a step further. Van Wijk argued that at one point the ending *-s of the non-neuter nominative singular must have been identical, apart from the accent, with the ending *-es/-os of the genitive singular.Footnote ⁶ The near-homonymy is revealing because in ergative systems the ergative case often has functions other than referring to A, just as the accusative in accusative systems need not only refer to O (but for instance also acts as a directional case in Indo-European: 9.24). More specifically, ergatives tend to coincide with instrumental or similar cases, as may be expected of a case category which expresses that an eventuality occurs ‘by someone’.Footnote ⁷ Accordingly, it is also of interest when the PIE genitive and ablative singular appear to have been identical (except in the o-stems), since the ablative can have agentive functions too (an eventuality occurs ‘issuing from someone/something’). Moreover, van Wijk found support for his views in a number of verbal classes governing the genitive, in which the genitive can be interpreted in an agentive manner. For example, in impersonal verbs of the Latin type pudet me alicuius rei ‘I am ashamed of something’, the thing causing (A) the shame stands in an ergative-like genitive and the person affected (O) in an absolutive-like accusative.Footnote ⁸

More as a side remark, van Wijk finally added that Uhlenbeck’s theory also explains an inflectional peculiarity of the Proto-Indo-European demonstrative pronouns. Here, a stem *so in the non-neuter nominative (masc. *so, fem. *sā/*seh₂ > Gr. ὁ, ἡ, etc.) contrasts with a stem *to- in the other paradigmatic slots (e.g., nom.-acc. sg. ntr. *to-d, acc. sg. masc. *to-m > Gr. τό, τόν, etc.). This distribution resembles the one observed by Uhlenbeck for the case endings: ‘The distribution of the stems so and to suggests that in older Indo-European so was a pronoun in the activus, to a pronoun in the passivus’ (i.e., ‘ergative’ *s(o) vs. ‘absolutive’ *t(o)-).Footnote ⁹

9.5 Pedersen on Ergative → Accusative Alignment Change

Similar views were next expressed by Pedersen.Footnote ¹⁰ Like van Wijk, Pedersen believed in the original identity of nominatives in *-s, qua old ergatives, and genitives in *-es/os, and for him too synchronically irregular impersonal constructions were telling relics. However, by comparing sentences without a surface subject, like Russ. otca derevom ubilo ‘a tree killed the father [lit., it killed the father by a tree]’ with instr. derevom ‘by a tree’, he advocated a slightly more complex system for Pre-Proto-Indo-European. In his opinion, the ergative was used only when the A of a transitive sentence was animate, whereas inanimate A’s stood in the instrumental.Footnote ¹¹

More importantly, though, Pedersen also sketched for the first time a possible mechanism for the change from a Pre-PIE ergative system to the unquestionably accusative system of (later) Proto-Indo-European. At first confined to the A function in transitive sentences, the ergative ending would later have been used also to mark the S’s of intransitive sentences (cf. 9.18, 9.21). Among the animate o-stems, whose absolutive ended in *-om, this eventually led to the restriction of *-om to the O function of transitive sentences. Hence, *-om was reanalysed as an ‘accusative’ ending and its case-marking element *-m was transferred to other (animate) stem classes.

As we shall see instantly (9.6–9.7), one problematic aspect of this scenario is the assumption that *-om was the old absolutive ending of the o-stems; but in this as in most other respects, Pedersen was still very much in agreement with what Uhlenbeck had laid out.

9.6 Vaillant on Nominal Gender and Directional *-m

What had thus become the ‘classical’ theory of Pre-PIE ergativity was consolidated furtherFootnote ¹² by two French scholars, Vaillant and Martinet. Although Vaillant’s main contribution concerned the verbal domain (9.12), he also offered a brief comparison with a modern ergative language, Basque, in which inanimate nouns may occur in the ergative, but more commonly stand in the instrumental when referring to the A of sentences like Bread nourishes the man.Footnote ¹³

In one sense, this was merely an endorsement of Pedersen’s ideas, but Vaillant also spelled out an important corollary. The nominal gender distinction between neuter (= prototypically inanimate) and non-neuter (= prototypically animate) nouns may have come into being only as a consequence of the transition from ergative Pre-Proto-Indo-European to accusative Proto-Indo-European. When the formal expression of A and S was homologised – presumably due to the A → S extension envisaged by Pedersen – , all those (prototypically animateFootnote ¹⁴) nouns that frequently occurred in the A slot retained the form of the former A-only case (ergative in *-s) for the new A/S case (nominative), whereas those (prototypically inanimate) nouns that did not normally occupy the A slot generalised the form of the former S case (absolutive in *-Ø) for the new A/S case. Before this had happened, however, there was no morphological gender differentiation, just as the neuter and non-neuter genders later continue to inflect identically outside the nominative and accusative cases.Footnote ¹⁵ As for the accusative in *-m, this would initially have been not an alternative absolutive ending (next to *-Ø), but the ending of a directional case (‘towards X’) which secondarily superseded the old absolutive in the new non-neuter gender.Footnote ¹⁶

9.7 Martinet on Markedness

While the old absolutive was said to survive best in the neuter nominative-accusative, a likely trace of it was also left among the non-neuters. Vaillant pointed out that the vocative singular is regularly expressed by the nominal stem with a zero-ending (cf. athematic Gr. πάτερ, Skt. pítar ‘father!’ < *ph̥₂ter-Ø; thematic Gr. λύκε, OCS vlьče ‘wolf!’ < *u̯l̥k^u̯e-Ø, etc.). But it was only Martinet who realised that this difference between a marked nominative and an unmarked vocative in the same declension class constitutes a cornerstone of the ergative theory:

So, whereas it makes sense to have a formally marked ergative (next to an unmarked absolutive) in an ergative language, it is unexpected in an accusative language to have a formally marked nominative (next to a marked or unmarked accusative) – as in later PIE *h₂ner-s_nom. [sow] *seh₁mn̥-Ø_acc. ‘the man [sows] the seed’. And this oddity is compounded if, with the vocative, there is in fact an unmarked case somewhere in the system; for this shows that the system is perfectly able to cope with a bare stem designating an animate entity.

In other words, the ergative theory also manages to account for the typologically irregular distribution of case marking in Proto-Indo-European by tracing it back to a typologically regular state of affairs. By contrast, to postulate that Proto-Indo-European had always featured an accusative system with an (animate) nominative in *-s and a less marked vocative in *-Ø would violate the principle that “that case which covers S (i.e. absolutive or nominative) is generally the unmarked term – both formally and functionally – in its system”.Footnote ¹⁸ Of course, the very fact that precisely such a ‘wrong’ system does exist in later Proto-Indo-European and its daughter languages proves that here as always exceptions can occur; but not to use these as waymarks in the reconstructive process would be wrong if there is a straightforward way to explain them in diachronic terms.Footnote ¹⁹

9.8 Split Ergativity and the Animacy Hierarchy

Having said that, it would be equally wrong to defend the classical theory of Pre-PIE ergativity if it entailed contradictions with typologically ‘normal’ behaviour in other respects. Since the 1980s, pertinent objections have been voiced more than once, and this has led to a more or less universal abandonment of what had once been a popular hypothesis – though not necessarily to a return to a strictly accusative reconstruction (cf. 9.15–9.18). As we shall see, this is an unfortunate development because the objections themselves are based on a misunderstanding.

In a famous article of 1976, the typologist Michael Silverstein established what is today known as the ‘animacy hierarchy’, a scale of animacy reaching from a high-animate end at which first- and second-person pronouns are located down to a low-animate end at which inanimate nouns are placed (Fig. 9.3).Footnote ²⁰

Fig. 9.3. Silverstein’s animacy hierarchy

Many languages that feature ergative alignment structures do so only in part. So-called split ergative languages commonly use an ergative pattern only for certain kinds of A, but an accusative one for others. As shown by Silverstein, the split between the two patterns is not random in these cases. While it can occur at any point within the animacy hierarchy, the accusative-patterning A’s will always be located towards the left and the ergative-patterning A’s towards the right. Thus, there cannot be a language where, in a sentence such as the stone hits my friend, the noun stone (= A) stands in the nominative when in a sentence such as my friend throws the stone the noun friend (= A) would stand in the ergative. By contrast, it would be perfectly normal to encounter a language in which friend in my friend throws the stone stands in the nominative, but stone in the stone hits my friend in the ergative.

The underlying reason for this cross-linguistic rule is not very difficult to see. If we take the participants my friend and stone as well as the verb throw, our knowledge of the real world tells us that it is much more likely to find them together in a sentence meaning ‘my friend throws the stone’ than in one meaning ‘the stone throws my friend’. So, in combination with a verb like throw, the semantics of my friend and stone make these a natural A and a natural O respectively, and we can in principle do without any case marking. But what if we did want to say ‘the stone throws my friend’? In that case, we would need to signal the unusual distribution of A and O. One way of doing that would be to give stone a special A marking, as an ergative. Alternatively, my friend could receive a special O marking, as an accusative. Thus, we end up with a situation where the upper end of the scale, populated by prototypical A’s, tends towards an accusative pattern with an unmarked nominative and a marked accusative; whereas at the lower end, where the prototypical O’s cluster, the ergative pattern with an unmarked absolutive and a marked ergative is at home.

Within Indo-European, Hittite operates a split ergative system of this sort. In it, neuter nouns functioning as subjects of a transitive verb (A) take a suffix -anza (sg.), -anteš (pl.), which may therefore be regarded as an ‘ergative’ suffix. By contrast, pronouns and nouns of common gender used in A function stand in the nominative.Footnote ²¹

9.9 Pre-PIE ‘Split Ergativity’ or ‘Split Accusativity’?

Impressed by Silverstein’s findings, Villar and Rumsey independently concluded that the ‘classical’ theory of Pre-PIE ergativity must be dismissed.Footnote ²² As we have seen, this theory derives the later nominative ending *-s of prototypically animate nouns from an ergative, whereas the nominative-accusative ending *-Ø of prototypically inanimate nouns is supposed to continue an absolutive (9.1). But if the lower end of the animacy scale favours ergative alignment, whereas the upper end does not, should we not expect the opposite, with an old ergative surviving among the inanimate nouns (essentially the later neuters) rather than among the animate (non-neuter) ones?

Following this logic, both Villar and Rumsey advocate a (Pre-)PIE ‘split-accusative’ alignment system instead. The items at the top of the animacy hierarchy would have followed a normal nominative-accusative pattern, whereas those at the bottom would have featured ‘neutral’ case marking (without formal distinctions between A, S, and O).Footnote ²³ Broadly speaking, this is what historical languages like Ancient Greek still present with their indistinct nominative-accusative case in the neuter, but a clear distinction between nominatives and accusatives among non-neuters. The only substantial changes that might have happened on the way from earlier to later Proto-Indo-European would thus have been (a) the introduction of *-s in the non-neuter nominative, because a typologically ideal accusative system should have had an unmarked nominative (9.7), and (b) possibly a gradual downward shift on the animacy scale of the boundary between the accusative and the ‘neutral’ systems.

9.10 Shortcomings of the ‘Split-Accusative’ Hypothesis

On closer inspection, such a ‘split-accusative’ hypothesis turns out to be less helpful than it may look at first sight. To begin with, it should be noted that there is no real motivation for (a), the introduction of nominatival *-s.Footnote ²⁴ Moreover, even if one were prepared to disregard this, two other obstacles would remain.

Firstly, the very starting point for Uhlenbeck’s proposal, the formal identity of the o-stem neuter nominative-accusative with the non-neuter accusative of the same stem class, would have to be a mere coincidence. While *-m should be the accusative ending among the animates (non-neuters), the nominative-accusative case of the inanimates (neuters) should have ended in *-Ø (i.e., *-e/o-Ø in an o-stem class). Hence, Villar suggests that inanimate nouns like *i̯ugóm ‘yoke’ were initially athematic m-stems (with zero-ending: i.e., nom.-acc. *i̯ug-om-Ø), which were subsequently reinterpreted as ‘o-stems’ (*i̯ug-o-m) because of their similarity with the animate o-stem accusative where *-o-m was indeed a combination of stem vowel + ending.Footnote ²⁵ However, it is far from clear that a chance similarity like this could have triggered such a fundamental reanalysis when there were many other inanimate nouns which all ended in nom.-acc. *-Ø, not *-m, and when all the oblique cases of the nouns in question (gen. *i̯ug-(o)m-és, etc.) would have betrayed their m-stem nature plainly enough.

Secondly, there is the problem that some non-neuters are also inanimate. Whereas proponents of the ergative theory can postulate that some inanimate nouns (e.g., *ped- ‘foot’) occurred in the ergative frequently enough to make the ergative become their later nominative, such inanimates with a nominative in *-s should never have existed if the ultimate boundary between an accusative and a neutral system coincided with that between animate and inanimate nouns. Villar therefore has to assume that the animate accusative marker *-m proved so useful that it was gradually extended to inanimates, and that the nominative marker *-s followed suit; but what triggered these extensions were formal, not semantic criteria.Footnote ²⁶ Had there been no phonotactic constraints, the split-accusative system might have become a fully accusative one, but this did not happen because certain inanimates, like those in *-mn̥ (e.g., *sper-mn̥ ‘seed’ > Gr. σπέρμα), would not admit an ‘accusative’ in *-m/-m̥ (e.g., †-mn̥-m). Whether or not such blocking is conceivable in a few cases, Villar himself must admit that he is unable to explain why, for example, the many neuter s-stem nouns did not then acquire an accusative in *-(e/o)s-m̥ when *-Vsm̥ was unquestionably a possible sequence in Proto-Indo-European.

9.11 Countering Animacy-Based Objections to the Ergative Theory

At least from a morphologicalFootnote ²⁷ point of view, the classical ergative theory thus remains superior to the alternative envisaged by Villar and Rumsey. Yet, its legitimacy does depend on whether it can be defended against the typological objections raised (9.9). Luckily, to do so is not particularly arduous as long as synchronic and diachronic considerations are not mixed up.

The classical model of Pre-PIE ergativity does not presuppose that inanimate nouns had no ergative case.Footnote ²⁸ If it did, it would indeed contradict Silverstein’s hierarchy (9.8). However, the hierarchy only refers to languages with animacy-based split-ergative systems. It has nothing to say about languages that are either fully ergative or show other types of split ergativity (cf. 9.21, 9.26–9.27). According to the classical model, Pre-Proto-Indo-European belonged to one of the latter types because it is assumed that ergative alignment occurred throughout the hierarchy, including the pronouns at the upper end of the animacy scale.Footnote ²⁹ The point is just that in this ‘fully’ ergative system, in line with every expectation, the more animate constituents of the hierarchy occurred more often in the ergative case (i.e., in A function) than the less animate ones. When the ergative alignment system of Pre-Proto-Indo-European was transformed into the accusative system of Proto-Indo-European, it was therefore natural that the new A/S case (= nominative) would be formally identical to, and originating from, the former A case (= ergative) in the paradigm of those (prototypically animateFootnote ³⁰) nouns that often occurred in A function; and similarly, it was natural that the new A/S case would be formally identical to, and originating from, the old S/O case (= absolutive) among those (prototypically inanimate) nouns that freely occurred in S function, but rarely if ever in A function (cf. 9.6).

Importantly, as long as we accept that the gradual spread of any A marker to the S function started at the upper end of the animacy hierarchy, among the pronouns, moving down the scale as time went by, there was not a single point in time at which Silverstein’s principle was violated. Schematically, this is shown in Fig. 9.4 (conflating, for the sake of simplicity, the different kinds of pronouns and the different groups of animate nouns).

Fig. 9.4. Ergative → accusative alignment change and the animacy hierarchy

9.12 Vaillant’s ‘Pseudo-Transitive Inflection’

The fact that the criticism of scholars like Villar and Rumsey is unjustified does not of course prove that the ergative theory is right. Ultimately, its value continues to rest on the number of apparent oddities it is able to explain.Footnote ³¹ From the preceding discussion, it will be clear that there are quite a few of them in the area of nominal morphology. But what about the verb?

In his article on Pre-PIE ergativity, Vaillant was one of the first scholars to ask how verbal inflection fits into the picture. Unaware that Pedersen had already expressed similar views a few years earlier (9.14), he suggested that the difference between the PIE ‘mi-conjugation’ on the one hand and the ‘h₂e-conjugation’ behind the perfect, the middle, and the Anatolian ḫi-conjugation on the other (2.9, 2.14, 5.3) might reflect two different underlying constructions, each requiring the use of different pronominal forms that secondarily became verbal endings. For ease of reference, and leaving aside the less distinctive plural, the two primitive ending systems – before the addition of ‘primary’ *-i (1.2, 7.30, 10.18) – are repeated here as conventionallyFootnote ³² reconstructed (cf. 1.3, 5.3):

Given its use in the prototypically intransitive perfect and middle, it is appropriate to associate the ‘h₂e-conjugation’ with intransitive environments; and in contrast with this, the ‘mi-conjugation’ may then be regarded as prototypically transitive. Moreover, we know that in several Indo-Iranian languages ergative constructions have arisen from syntactically reinterpreted patterns involving a past passive participle in *-to- accompanied by an agent in an oblique case (e.g., Old Persian X manā kr̥tam ‘X is done/has been done by me’ > Middle Persian man kart ‘I did’ with the oblique case man of the 1sg. personal pronoun).Footnote ³³ Vaillant therefore proposed to explain the ‘mi-conjugation’ as a similarly built ‘pseudo-transitive inflection’. In this, the 3sg. in *-t would reflect ‘a nominal form *g^when-t-, presumably a verbal abstract’, and the other persons would have arisen by a combination of the same nominal form with a personal pronoun in the ergative case:

9.13 Critique

Vaillant’s proposal is ingenious, but unsatisfactory in more than one respect. Firstly, much of the appeal of any such theory lies in the overlap between the 1sg. ending *-m and the oblique stem *m- of the 1sg. pronoun in Proto-Indo-European (cf. e.g. acc. sg. *me > Gr. με, *mē > Ved. mā(m), Lat. mē(d)). This stem *m- strikingly contrasts with whatever form without *m- one may have to reconstruct for the corresponding nominative (underlying Gr. ἐγώ, Lat. ego, Skt. ahám, etc.: e.g. *(h₁)egh₂-óm/-oH, cf. 9.28). It would thus seem simple enough to assume that the Pre-PIE ergative, being an oblique case, should have featured the stem *m-, exactly as required for Vaillant’s pattern. Surprisingly, however, only a few paragraphs earlier Vaillant had suggested that the historical nominative without the *m- continues the ancient ergative of the 1sg. pronoun.Footnote ³⁵

To remove this contradiction is easy. All one has to do is to correct Vaillant’s statements about the pronominal nominative continuing a pronominal ergative. In fact, there is a good reason why the historical nominative of the first-person (and second-person) pronoun should descend from an absolutive rather than an ergative, even if the relevant paradigms initially did contain an ergative. If, as is likely, later Proto-Indo-European behaved like Greek, Latin, or Sanskrit in its use of the personal pronouns, their nominative occurred mainly in emphatic, contrastive, or syntactically absolute contexts – but not as a rule in conjunction with first-person or second-person verb forms (where the endings alone sufficed to reference S/A). At least in its absolute function, the ‘nominative’ of the personal pronouns thus acts as a mere ‘name’ for the speaker (‘I’) and addressee (‘you’). As such it resembles a standard animate noun’s vocative (or naming form) more than such a noun’s S/A-referencing nominative. Hence, it is also as likely as a standard vocative to continue the unmarked absolutive of a Pre-PIE ergative system.Footnote ³⁶

Secondly, and more seriously, Vaillant’s idea depends on sound developments that are created entirely ad hoc. In the 2sg., for example, the starting point *g^u̯hent-t(i) is needed because the 2sg. pronoun features a stem *t(u̯)- corresponding to the stem *m- of the 1sg. pronoun (cf. e.g. acc. sg. *tu̯e > Gr. σε, *tē > Lat. tē(d), etc.); but for the associated change *-tt- > *-s-, one cannot very well postulate *-tt- > *-ss- > *-s- by arguing that ‘the group *tt changes in all the Indo-European languages, including Hittite (> zt …)’, that ‘the treatment which is closest to the primordial treament, without restitution of the dental stop, must be the most unusual one, Western Indo-European ss’, and that ‘a group ss had to be simplified into s (2sg. *ési …)’. If the stage *-s- had been reached in Proto-Indo-European before Anatolian split off (cf. 2sg. *-s(i) also in Hittite), it is chronologically impossible to relate this to the largely post-PIE changes of *-tt- > *-t^st- > Anatolian -zt-, Greek/Iranian/Balto-Slavic *-st-, Italic/Germanic -ss-. Similarly, one is left to wonder why in the 1sg. or 1pl. the internal *-t- was lost instead of the *-m of the ending becoming vocalic (e.g., 1sg. *g^u̯hent-m̥). Of course, nothing can prevent us from postulating Pre-PIE sound laws that produce the required outcomes, but a priori the phonological odds are stacked against what Vaillant needs to make his scenario work.

Furthermore, a general point is also worth making. While it is true that the ergatival constructions of ancient and modern Indo-Iranian have arisen in the way envisaged by Vaillant for Pre-Proto-Indo-European, the syntactic reanalysis of previous accusative-system passives is no necessary precondition for ergative systems to emerge.Footnote ³⁷ It is even somewhat paradoxical that Vaillant’s comparison with Indo-Iranian implies an even earlier (‘Pre-Pre-PIE’) accusative system when the conclusions that were drawn from the nominal data by Uhlenbeck and his successors were precisely what had pointed away from such accusatival beginnings in the first place.

9.14 Pedersen’s Alternative

A more promising way forward is therefore the one shown by Pedersen. While discussing the similarities, and possible genetic relationships, between the Indo-European and the Finno-Ugric personal pronouns, Pedersen stated that ‘the “perfect endings” [sc., 1–3sg. *-h₂e, *-th₂e, *-e] will thus originally have belonged to the intransitive verbs, the “normal” endings [sc., 1–3sg. *-m(i), *-s(i), *-t(i)] to the transitive ones’;Footnote ³⁸ and he observed that ‘the “intransitive” ending of the 1sg. is reminiscent of the initial sound of ἐγώ, the “transitive” ending of the non-nominative cases of the pronoun. … In the intransitive 3sg. the personal endings seem to be absent, for the -e of the perfect is hardly a person marker; in the transitive series there is an ending -t, which is reminiscent of the non-nominative pronominal stem *to- (Gr. τόν etc.).’Footnote ³⁹

What Pedersen proposes is thus a situation in which a verbal root could be combined, depending on syntactic context, with either the absolutive or the ergative case of a pronoun. For illustration, we may make up the following examples:

In these sentences, transitive *g^u̯hen- and intransitive *trep- represent the same kind of uninflected base form, and the ergative alignment is exclusively conveyed by the pronominal/nominal constituents. But because Pedersen outlines all this only summarily, it remains unclear whether he is assuming (a) that one and the same root could occur in both transitive and intransitive sentences or (b) that the selection of either (i) or (ii) was entirely determined by the root involved.Footnote ⁴⁰ If (a) were the case, we would end up with a nuclear system of verbal voice:Footnote ⁴¹

Although not every root might occur with equal ease in both ‘voices’, one advantage of positing such a system would be that the later active vs. middle voice contrast could naturally grow out of it. In fact, Pedersen does refer to Kuryłowicz and Stang, who at the time had only just discovered the common origin of the perfect and middle endings.Footnote ⁴² Yet, since he also treats the middle as a category that is both contemporary with and independent of the ‘active intransitive’ type, he may nevertheless have been inclined towards (b), the assumption that the intrinsic (in)transitivity of any given root ultimately decided which of the two constructions it had to take.Footnote ⁴³

9.15 Beginnings of the ‘Active Hypothesis’

If so, Pedersen’s ideas not just connect with the classical ergative hypothesis but foreshadow a rather different take on Pre-PIE alignment. Since this alternative approach, which flourished some decades later, is still fairly influential today, it must also be considered here, if only to show that its explanatory potential is inferior to that of its ergative counterpart.

Next to the ‘ergative’ and ‘accusative’ systems discussed so far, a third major alignment type is constituted by so-called active (or ‘split-S’) languages. In these, what matters is not the transitivity/intransitivity of any given verb but its belonging to either an ‘active’ or an ‘inactive’ class. With ‘active’ verbs, the functions of A and S are marked the same (so that S can be indexed as S_A), whereas with ‘inactive’ verbs the S formally matches an O (hence: S_O). Importantly, an intransitive verb should not sometimes mark its S like an O (e.g., when the eventuality is not volitional) and at other times like an A. If, say, a verb like fall were classified as ‘inactive’ because it usually refers to an involuntary and uncontrolled activity, its subject would still be marked as S_O if, for once, it voluntarily fell.Footnote ⁴⁴

That the existence of such active languages might impact on our understanding of Pre-Proto-Indo-European was first argued in 1947 by Iosif M. Tronskij; but a printed version of this paper was only published in the 1960s, in a collection that also contained one or two other pieces promoting similar views.Footnote ⁴⁵ According to Tronskij, Pre-Proto-Indo-European not only had two classes of verbs, activity verbs and verbs of state (i.e., the above ‘active’ and ‘inactive’ classes), but also two corresponding groups of nouns: an inflected active one with ergatives and absolutives, and an uninflected inactive one. As we shall see (9.16), the recognition that in some active languages the active/inactive dichotomy in this way extends to the nominal domain was to become quite important in later discussions of the supposedly ‘active’ character of Pre-Proto-Indo-European. In Tronskij’s scheme, however, it ultimately had little structural relevance. All it meant was that nouns of the inactive class were always uninflected, also when occurring with active verbs, whereas nouns of the active class would take the formally unmarked absolutive only when occurring as S_O’s of inactive verbs or as O’s of active verbs.Footnote ⁴⁶ In hindsight, we must say that this does violate Silverstein’s animacy hierarchy: for the absence of case marking should occupy the higher ranges of the animacy hierarchy when contrasting with an ergative (rather than an accusative) marking system.Footnote ⁴⁷

9.16 The Classical ‘Active Hypothesis’

Following these beginnings, the ‘active hypothesis’ gained in visibility thanks to Gamkrelidze and Ivanov’s monumental 1984 study on PIE and Pre-PIE language and culture as well as W. P. Lehmann’s work on the subject.Footnote ⁴⁸ These scholars, whose discussions are informed by the typological description of active languages by Klimov, agree that the following points deserve particular attention:Footnote ⁴⁹

1. 1. Active languages commonly divide their nouns into an active and an inactive class (cf. 9.15). These classes, within which there should be virtually no inflection apart from the opposition between an ‘active’ and an ‘inactive’ case, are said to be replicated by the Pre-PIE precursors of the later non-neuters and neuters respectively. In the wake of Meillet,Footnote ⁵⁰ reference is also made both to the alleged occurrence of PIE lexical pairs like ntr. *peh₂ur (vel sim.) as ‘inactive’ fire (> Gr. πῦρ, Umbr. pir, Hitt. paḫḫur, etc.) vs. non-ntr. *n̥gnis as ‘active’ fire (> Lat. ignis, Ved. agníḥ, OCS ognь), and to the parallelism of feminine nouns for trees vs. stem-formationally identical neuter nouns for fruits (e.g., Lat. pirus ‘pear tree’ vs. pirum ‘pear’; cf. Footnote 9.6, fn. 14).

2. 2. Next to the binary noun classification, active languages feature a corresponding dichotomy among the verbs. Whereas verbs with ‘active’ semantics (e.g., go, run, eat, live, kill, etc.) freely occur with subjects from the active noun class, a more restricted group of verbs with ‘inactive’ semantics (e.g., lie, fall, be heavy, etc.) typically require an inactive noun as their subject. If an active noun is to be predicated similarly, a separate active verb may be used, so that there will be synonym pairs for verbal concepts of the second type. In Indo-European, the erstwhile existence of such doublets is supposed to be reflected by pairs like *h₁es- vs. *b^huh₂- ‘be’, *ses- vs. *k̑ei̯- ‘lie’, and *steh₂- vs. *h₃er- ‘stand’ (with the first item in each pair representing the ‘active’ variant). Furthermore, the opposition between the PIE mi-conjugation and the h₂e-conjugation is taken to continue the same active vs. inactive divide.

3. 3. Active languages often possess a distinct class of verbs of affect and/or experiencing, whose experiencer subject does not stand in the active case. By contrast, passives are missing. Impersonal constructions in the ancient Indo-European languages, such as the Lat. pudet me ‘I am ashamed’ and Gr. δοκεῖ μοι ‘it seems to me’ types, might thus descend from such affective/experiencing verbs. And the variety of formations and constructions used to express the passive in the various Indo-European idioms would result from the lack of a single specific passive in the ancestor language.

4. 4. Active languages frequently distinguish alienable and inalienable possession, but have no separate verb for ‘have’. The latter point appears to have been shared by Proto-Indo-European since the daughter languages do not agree on what (if any) verb they use to express the concept of ‘having’. As for the former point, a pertinent phenomenon is observed in Hittite. There, the so-called ‘split-genitive’ construction (whereby a possessor in the genitive is ‘doubled’ by a possessive pronoun attached to the head noun: e.g., ša ^lúmašdá ēšḫar=šet ‘the blood of the poor man’, lit. ‘of the poor man his blood’) is by and large confined to situations of inalienable possession. Similarly, it has been suggested that the complementary use of the Latin dative vs. genitive of possession (domus patri/patris est ‘the house belongs to the father’) might originally have been determined by the (in)alienability of the thing possessed.Footnote ⁵¹

5. 5. Finally, the differentiation of inclusive vs. exclusive pronouns is claimed to be a standard characteristic of active languages. To match this, it is argued that the stem divergence between *u̯- and *m- in the Indo-European 1pl. pronoun could be due to a similar distinction. Inclusive *u̯- would be continued for instance in Hitt. u̯ēš, Ved. vayám, and Goth. weis ‘we’ – but because of its inclusivity also in 2pl. forms such as Lat. vōs, Gr. ὑμεῖς (with *us-), or OCS vy ‘you’ – , and exclusive *m- in OCS my, or also Lat. nōs and Gr. ἡμεῖς (with *n̥s-) whose *n- for *m- may be influenced by forms with regular *m̥s- > *n̥s-.

9.17 Assessment I: Minor Evidence

Not all of these arguments are equally weighty. The central point for us is (2), regarding the dichotomous verbal system. Before we discuss this (9.18), a few remarks on the others will suffice to show how inconclusive they are.Footnote ⁵²

1. ad 1. While it is not implausible to suspect an earlier animate/‘active’ vs. inanimate/‘inactive’ distinction behind the later non-neuter vs. neuter divide, we have already seen that the ergative theory can just as readily explain the genesis of nominal gender by reference to the change from an ergative to an accusative system (9.6). If there are pairs like *peh₂ur vs. *n̥gnis, these can only possibly tell us something about (Pre-)PIE world views and culture, not about grammar. In order to argue that the Pre-Proto-Indo-Europeans grammatically (i.e., not just semantically) distinguished an inactive *peh₂ur from an active *n̥gnis, one would have to produce evidence suggesting that combinatory restrictions applied to the selection and/or formation of predicates accompanying these ‘fire’ lexemes.

   As for the primitive declension system, again it is perfectly conceivable that there was a very early opposition between an ‘active’ case in *-s and an ‘inactive’ one in *-Ø (as would be required by the later neuter vs. non-neuter characteristicsFootnote ⁵³). However, there is no particular reason to regard, say, the ‘animate’ nouns’ case in *-m (> PIE acc. *-m) as more recent than the one in *-s. Moreover, the suggestive similarity of the later nominative in *-s to the later genitive in *-(e/o)s (9.4) would have to be accidental,Footnote ⁵⁴ and it might be asked why it was the ‘active’ rather than the ‘inactive’ case of the active/animate noun class that was marked.

2. ad 3. To have a special group of experiencing verbs is no prerogative of active languages. Several ergative languages of the Caucasus, for example, share this feature.Footnote ⁵⁵ Also, if the Indo-European coexistence of impersonal verbs with either a dative or an accusative experiencer (Lat. pudet me vs. Gr. δοκεῖ μοι) is anything to go by, the ergative hypothesis even has more explanatory power. The accusatival structure may then continue a normal transitive construction (with the experiencer in the absolutive → accusative) and the datival one a special experiencing type (with the experiencer in the dative qua experiencer case). In an active language with very limited inflection, one would not expect such a dual option.Footnote ⁵⁶

   Turning to the absence of a PIE passive, we should both note that the late-PIE middle may (also) have been used in passive contexts and stress that a passive transformation would have been at least as redundant in an ergative-type Pre-Proto-Indo-European as in a Pre-PIE system with active alignment. Just as an accusative language must not be expected to feature an ‘antipassive’ (9.22), so an ergative language can do without a passive.

3. ad 4. Similarly, it is unclear why the differentiation of alienable vs. inalienable possession should not also be possible in ergative (or accusative) languages. Even if it were true that “in active languages it is internally motivated and intertwined with the dominating pattern of the language” since “inalienable possession, which implies absence of change, is closely connected to stativity”,Footnote ⁵⁷ and even if there does seem to be some correlation with active-language structures in this case,Footnote ⁵⁸ the Hittite facts, which provide the most substantial – if not the only – support for the idea that the parent language also made such a distinction, could not possibly allow any inferences about the ‘active’ nature of Pre-Proto-Indo-European. Firstly, the phenomenon is in Hittite not backed by an active/inactive divide in the nominal and verbal systems. Secondly, it is most doubtful that the Hittite construction is inherited when an exact match is observed in neighbouring, but genetically unrelated, Hurrian, another ergative language.Footnote ⁵⁹

   The point about the missing ‘have’ verb, meanwhile, can be turned on its head. If later Proto-Indo-European, despite being an accusative language, did not develop such a verb although it might have done so just as easily as many of its descendants did, it can hardly be maintained that only active languages are able to go without for a prolonged stretch of time.

4. ad 5. Once again, it is simply wrong to assume that ergative or accusative languages are less inclined than active ones to distinguish inclusive from exclusive pronouns.Footnote ⁶⁰ Apart from that, the reconstruction of the PIE personal pronouns is so fraught with difficulties that any interpretation of the discernible patterns as reflecting an inclusive vs. exclusive divide is extremely uncertain at best.Footnote ⁶¹ If one concentrates on the 1pl., the two-stem situation with *u̯- vs. *m- is primarily reminiscent of the two-stem situation in the 1sg., not least because the oblique stem features *m- there too (9.13). Admittedly, the 1pl. with *u̯- does not in the same way resemble the 1sg. pronoun *(h₁)egh₂-óm/-oH (vel sim.), which is continued in most Indo-European languages, but its shape at least vaguely recalls the 1sg. nom. pronoun ūk ‘I’ of Hittite (itself of unclear provenance). This is not to say that there must be a connection, and that pluralic *u̯- vs. *m- therefore must have arisen from an absolutive/nominative vs. oblique difference rather than an inclusive vs. exclusive one; but the matter shows that alternative ways of looking at the same data do exist. Unless the appearance of *u̯- also in the 2pl. were a mere coincidence, this could incidentally still be explained by reference to inclusive usages of the 1pl. form, with *u̯- replacing whatever 2pl. pronoun might have been around before (cf. 10.6). However, the stem overlap no more proves that the (Pre-)PIE 1pl. pronoun contained [±inclusive] as an essential semantic feature than the confidential or condescending uses of 1pl. we in the sense of ‘you’ (in e.g. doctors’ speak) proves it for Modern English.

9.18 Assessment II: ‘Active’ vs. ‘Inactive’ Verbs?

So far then, there is little support for the active hypothesis. Unlike its ergative competitor, it also does not clarify any of the morphological patterns we have to reconstruct for Proto-Indo-European.Footnote ⁶² Whether it should nevertheless prevail therefore depends entirely on argument (2) in 9.16. Is this strong enough to override the advantages of the ergative view?

Effectively, the backing provided by (2) is very limited as well. It is true that intransitive mi-verbs like *h₁ei̯-m(i) ‘go’ pose a problem for a strictly ergative system where the mi-conjugation was confined to transitive structures (9.14). In order to account for them, one has to assume that the ‘ergatival’ endings were secondarily adopted by some intransitive verbs, notably ones with an agentive/controlling subject (‘unergatives’, as opposed to non-agentive/controlling ‘unaccusatives’ like lie or sleep). In the context of an ergative → accusative alignment change, this would be a natural development.Footnote ⁶³ Once the case marking of S’s and A’s coincided (cf. 9.5, 9.21), it was no longer necessary to operate with one verbal cross-referencer (‘ending’) for S and another for A. If both of them were to continue, their use had to become determined by semantic instead of the previous syntactic criteria.Footnote ⁶⁴ As we shall see later (9.20), the h₂e-conjugation could thus become specialised in a new ‘middle-voice’ category opposed to the characteristically ‘active’ m(i)-conjugation.

Within the framework of the active hypothesis, things are not equally smooth. First of all, it must be emphasised that a binary division into active vs. inactive verbs and nouns is no necessary characteristic of a language belonging to the ‘active’/‘split-S’ type. The fact that an ‘inactive’ verb like fall requires a subject marked as S_O need not mean that fall can only occur with subjects like stone, but not with others like man, and that a second quasi-synonymous ‘active’ verb fall₂ is therefore needed to formulate sentences like ‘the man falls’. But even supposing that Pre-Proto-Indo-European was one of those ‘active’ languages where such verbal pairs exist,Footnote ⁶⁵ the concrete examples given are questionable: *steh₂- and *h₃er- are no more synonymous than Engl. ‘(take a) stand’ and ‘rise’, and *ses- and *k̑ei̯- are normally taken to mean ‘sleep’ and ‘lie’ respectively. As for *h₁es- vs. *b^huh₂-, not only is it possible to diagnose a semantic divergence here too (‘be’ vs. ‘become’), but it is also odd to treat *b^huh₂- as the ‘inactive’ counterpart of *h₁es- when its mi-inflection is no less established than that of *h₁es- (cf. 1sg. aor. *(h₁e-)b^huh₂-m > Gr. ἔφῡν, Ved. ábhuvam; Table 6.1).Footnote ⁶⁶ Furthermore, there is no sign of any of the alleged ‘inactive’ verbal roots being restricted to ‘inactive’ nouns: *h₃er- frequently refers to people rising or getting up, *k̑ei̯- predicates lying persons as freely as lying things, and *b^huh₂- is the root used in various intrinsically animate, hence ‘active’, lexemes like Ved. bhūtá- ntr. ‘creature’ or Gr. φῦλον ‘tribe’. Related to this is the more general difficulty that one and the same root can readily be associated with both mi-endings and h₂e-endings in parallel formations, be they actives vs. middles or non-perfect (dynamic) vs. perfect (stative) ones. In order to defend the active hypothesis, one has to regard this as a secondary development. But the argumentation then becomes paradoxical. Because there are two conjugation ‘classes’, an active-language prehistory is postulated, but because the two ‘classes’ do not behave as they should in an active language, it is assumed that substantial changes obscured the original state of affairs.

Similar inconsistencies emerge when we look more carefully at one of the most fundamental traits in the two-class structure. Just like the mi-conjugation with 1sg./2sg. *-m(i) and *-s(i), the h₂e-conjugation possesses 1sg./2sg. forms, in *-h₂e and *-th₂e respectively (9.12). Yet, if the h₂e-conjugation were the original ‘inactive’ class, should we not expect it to combine only very reluctantly, if at all, with subjects from the top end of the animacy hierarchy (‘I’, ‘you’)? In order to address this problem, Gamkrelidze and Ivanov propose that *-h₂e initially indexed an inactive patient (O = ‘stone’) in a transitive sequence such as ‘I/you [move] stone’, following either a zero marker of the active agent (A) in the 1sg. (hence *-Ø-h₂e) or a *-t- marker of the active agent in the 2sg. (hence *-t-h₂e). By contrast, 1sg. *-m(i) and 2sg. *-s(i) would have indexed the active agent (A) only when the patient (O) also belonged to the active class (e.g. ‘I/you [kill] animal’).Footnote ⁶⁷ In other words, far from being restricted to intransitive contexts with inactive A’s, *-h₂e and *-th₂e would have belonged to a prototypically transitive type with active A’s, before being “extended to one-place verbs with an inactive argument” because of the “loss of the strict binary active/inactive classification of Indo-European nouns and reanalysis of the noun classes as grammatically agreeing nominal formations without regard to their semantic activity or inactivity”. That we are then not only dealing with an ‘extension’, but in many cases (with prototypical middles and perfects) also with an unmotivated restriction to intransitive uses (“one-place verbs”) is not commented upon, perhaps because the construct is so thoroughly arbitrary anyway. With it, the existence of two conjugation types is effectively no longer regarded as a telling relic of a Pre-PIE active alignment system. Instead, argument (2) comes to hinge exclusively on the semantic pairs discussed and dismissed above, together with the opposition of an allegedly ‘active’ 3sg. in *-t(i) vs. an allegedly ‘inactive’ 3sg. in *-e – with the proviso that the ‘inactivity’ of the latter is a pure petitio principii when this ending, at the stage we can really reconstruct, undoubtedly parallels the ‘active-class’ 1sg./2sg. endings *-h₂e and *-th₂e.Footnote ⁶⁸

9.19 ‘Split-S’ vs. ‘Fluid-S’

Just like the ‘split-accusative’ hypothesis (9.10), the active hypothesis thus fails to offer the degree of reconstructive economy that can be achieved with its ergative counterpart. Therefore to prefer the latter does not of course exclude the theoretical possibility that an even earlier active phase preceded the ergative one:Footnote ⁶⁹

However, care must be taken not to confuse such a conceivable – though exceedingly remote – ‘active’/‘split-S’ phase with an intermediate phase during the transition from the ergative to the accusative system. As argued in 9.18, this period too would have seen some S’s (already) marked as S_A’s while others were (still) marked as S_O’s. But this would not have been the same kind of S_A vs. S_O contrast as in an ‘active’ (‘split-S’) language. Instead, the transitional system would have belonged to what typologists call a ‘fluid-S’ type:

9.20 The Emergence of Verbal Voice

The emergence of verbal voice in Proto-Indo-European may be seen as a consequence of such a ‘fluid-S’ interlude between the Pre-PIE ergative and the later accusative systems. Although in the end the subjects of (prototypically, though not exclusively, transitive) active verbs and (prototypically intransitive) middle ones are identically marked as nominatives, no matter how ‘controlling’ or ‘agentive’ they are, the possibility of attaching either the active or the middle endings to one and the same verbal stem may still hark back to the characteristic flexibility of a ‘fluid-S’ type.

More concretely, as soon as the A-indexing verbal endings of transitive verbs (e.g., 1sg. *-m) started to intrude into the domain of intransitive verbs, a competition had to ensue between them and the old S-indexing endings of intransitive verbs (e.g., 1sg. *-h₂ as the likely precursor of *-h₂e: 9.28, 10.3–10.4). This was resolved not by getting totally rid of the S-indexing endings, as might have been possible, but by refunctionalising them. Not only were they kept wherever an S_O had to be indexed but, in line with this core function during the ‘fluid-S’ period, their application range was also somewhat extended into transitive environments. There they could now be used on condition that an A had some O-like properties, such as being the experiencer – and hence potentially the beneficiary – of an eventuality (‘A_O’). This explains why in Indo-European we observe not only intransitive middles like Gr. φέρεται ‘takes a course, moves, succeeds, (pass.) is carried’, but also transitive ones like Gr. φέρεται ‘carries off, wins sth. for him/herself’.Footnote ⁷¹

Figure 9.5 represents the development schematically, using the 1sg. endings as placeholders for the complete paradigms. For a more detailed account of the formal evolution of the middle voice, see 10.19.Footnote ⁷²

Fig. 9.5. The emergence of verbal voice in Proto-Indo-European

9.21 Extension of A Marking

The idea that a Pre-PIE ergative system could be transformed into the accusative system of later Proto-Indo-European by an extension of nominal A marking to S’s (9.5), accompanied by a reshuffling of verbal A vs. S indexing (9.18, 9.20), gains in plausibility when we observe that a change of this kind indisputably occurred in Mingrelian. Like its South Caucasian relatives, which include Georgian, Mingrelian originally used ergative alignment in the aspectually perfective aorist tense series, but accusative alignment in the imperfective present tense series (cf. 9.26). At some point, however, the original ergative marker was extended to cover S as well as A in the aorist.Footnote ⁷³ Since the alignment of both the present and the aorist tense series has thus become accusatival in Mingrelian, the old ergative marker (-k), still being confined to the aorist system, has effectively turned into a marker of aspectual perfectivity (though one that is attached not to the verb, but to its nominal subject).

9.22 Passives and Antipassives

Though simple and straightforward, the evolution witnessed by Mingrelian is not the only possible mechanism for an ergative → accusative alignment change. A second major pathway mirrors the accusative → ergative change brought about by the reinterpretation of a generalised passive in Indo-Iranian (9.12).Footnote ⁷⁴

Just as accusative languages often have a passive that can be used, for example,

1. (i) when the patient (underlying O) of an eventuality is located higher up on the animacy hierarchy than the agent (underlying A), so that the ‘normal’ relationship of agent and patient is inversed (e.g., John was hit by a car),

2. (ii) when the underlying O is considered to be of greater relevance than the underlying A, so that the latter need not even appear in the passive construction (e.g., The tree was felled (by the workmen)), or

3. (iii) when the primary focus is placed on the result of an eventuality rather than the eventuality as such (e.g., Mary was surprised by the news),Footnote ⁷⁵

so ergative languages commonly possess a so-called ‘antipassive’ transformation, featuring an obligatory S in the absolutive case and an optional Target in an oblique case (such as a dative or directional) (Fig. 9.6).

Fig. 9.6. Passive and antipassive transformations

Enabling semantic moves that match the ones just listed for the passive, such antipassives may for example allow speakers

1. (i) to remedy an atypical A–O relationship by demoting a highly animate underlying O into an indirect object or Target role, so that there is neither a non-prototypical (high-animate) O nor a non-prototypical (low-animate) A (e.g., The car is hitting at/towards John, rather than The car is hitting John),

2. (ii) to assign greater relevance to the underlying A than to the underlying O (especially if the latter is indefinite or generic), and possibly to omit the underlying O altogether (e.g., Smoking kills (people)),

3. (iii) to focus on certain modal features of an eventuality (rather than its actuation as such, let alone its result), such as the underlying O’s only partial involvement/affectedness (e.g., Jane is pulling at the table, as opposed to Jane is pulling the table) or the underlying A’s general inclination/habit to carry out the eventuality (e.g., Mary tells funny stories).Footnote ⁷⁶

Point (iii) in particular explains why “just as a passive is suited to perfective aspect or past tense – and is sometimes the genesis of an ergative system that is found only in perfect or past – so antipassive relates to tenses/aspects/moods in which the subject is likely to control the activity (e.g. purposive) or where the possibility of an activity happening depends on the propensity of the subject”.Footnote ⁷⁷ The use of the antipassive is therefore commonest in the non-perfective and non-past domain.

9.23 Reanalysis of Antipassives

Given the focus on the underlying A rather than the underlying O in the syntactically intransitive antipassive, an antipassive transformation constitutes an ideal locus for A marking to spread to S’s. Thus, far from being mutually exclusive with the evolution described in 9.21, the reanalysis of an ergative system’s ‘generalised antipassive’ as a transitive base construction in a newly accusative system should be able to go hand in hand with the spread of A marking to S roles. In fact, there are strong reasons to believe that the accusative system of the present tense series of South Caucasian (9.21) goes back precisely to such an antipassive transformation, since its functionally accusative-like case for O’s is also used more generally as a dative.Footnote ⁷⁸ If this is so, it means that Mingrelian has merely gone a little further along the road towards full accusativity than other South Caucasian languages. Its A-marker extension in the aorist tense series nothing but complements the antipassive-induced change from ergativity to accusativity in the present tense series which is shared by relatives like Georgian.

9.24 Pre-PIE Antipassives and Nominal Inflection

Against this background, it will be useful to remember that several scholars already suspected an older ‘directional’ case behind the PIE accusative ending *-m (9.6).Footnote ⁷⁹ To support this, one may compare the directional accusative in various ancient Indo-European languages (e.g., Il. 1.322 ἔρχεσθον κλισίην Πηληϊάδεω Ἀχιλῆος ‘go to the tent of the Peleid Achilleus!’, Lat. Romam/rus ire ‘to go to Rome/to the countryside’, nuptum dare ‘give into marriage’, RV 10.14.13 yamáṃ ha yajñó gachati ‘to Yama goes the sacrifice’).Footnote ⁸⁰ Given such usages, it is unproblematic to assume that the non-directional accusative has its roots in a Target case in *-m representing an underlying O in the framework of an antipassive transformation.Footnote ⁸¹ And since we have seen in 9.22 that the use of such an antipassive would naturally have been prompted whenever a high-animate surface O was to be avoided, it also makes sense that the accusative in *-m is restricted – at least outside the thematic declensionFootnote ⁸² – to the higher end of the animacy scale.Footnote ⁸³ With inanimate O’s, the need to use the antipassive would have been less frequent, causing inanimate nouns to occur less often in the *-m case, just as they occurred less often in the ergative marked with *-s. When the old antipassive was reanalysed as a new transitive base construction in an accusative system (9.23), the generalisation of *-m as a surface O marker therefore did not affect these lexemes any more than did the extension of *-s from A to S function (Fig. 9.7).

Fig. 9.7. Antipassive case marking and (Pre-)PIE ergative → accusative alignment change

9.25 Pre-PIE Antipassives and Verbal Inflection

To keep things clear, our discussion of antipassives has so far focused only on the marking of the nominal constituents. However, if we still want to follow Pedersen in tying the PIE mi-conjugation to an earlier transitive syntactic pattern, and the PIE h₂e-conjugation to an intransitive one (9.14), the notion that an antipassive derivation may have been key to the Pre-PIE ergative → PIE accusative alignment change also has consequences for the verb.Footnote ⁸⁴

Specifically, we may consider two options:

1. (a) the Pre-PIE antipassive structure was exclusively marked on the nominal constituents, with the verb simply taking the required ‘intransitive’/S-indexing endings;Footnote ⁸⁵ or

2. (b) the Pre-PIE antipassive structure was (additionally) marked by some derivational affix attached to the verbal stem.

Taking again the example of 9.14 (though now writing *-h₂ for *-h₂e: cf. 9.20), we thus get either

1. (a)

   base construction	*gu̯hen-m̥	h2ner-Øabs.	=	‘I strike the man’
   antipassive	*gu̯hen-h2	h2ner-m̥dir.	=	‘I strike (at) the man’

or

1. (b)

   base construction	*gu̯hen-m̥	h2ner-Øabs.	=	‘I strike the man’
   antipassive	*gu̯hen-[suffix]-h2	h2ner-m̥dir.	=	‘I strike (at) the man’

In principle, the two options do not exclude each other. Compare the Greek passive, which is also partly marked by endings (act. πέμπ-ω ‘I send’ vs. pass. πέμπ-ομαι ‘I am sent’) and partly by a derivational suffix (act. ἔπεμψ-α ‘I sent’ vs. pass. ἐπέμφ-θη-ν ‘I was sent’). A priori, one might perhaps expect a preference for (a) with ‘labile’ roots, which freely occur in either transitive or intransitive contexts (e.g., *trep- ‘turn’), but for (b) with prototypically transitive roots whose use in an intransitive structure would require more signposting. In 10.35, we shall see that there is in fact a strong candidate for such a Pre-PIE antipassive suffix.Footnote ⁸⁶ But for the time being, we may content ourselves with stressing once more that the question of Pre-PIE alignment cannot adequately be discussed by reference to nominal inflection alone.

9.26 General Principles

In 9.21 it has been mentioned that most South Caucasian languages belong to a linguistic type with ‘split ergativity’, though one where it is not the animacy of the nominal constituents that determines the split (9.8). Instead, ergative syntax is found in the perfective aorist tense series, and accusative syntax in the imperfective present tense series. A similar restriction of the ergative pattern is observed among the modern Indo-Iranian languages where what used to be a passive periphrasis with a *-to- participle generallyFootnote ⁸⁷ replaces a transitive verb in the past tense only (cf. 9.12). This is not a chance coincidence. Ample cross-linguistic evidence shows that “if a split is conditioned by tense or aspect, the ergative marking is always found either in past tense or in perfective aspect”.Footnote ⁸⁸ The fact that antipassives tend towards non-past and non-perfective contexts (9.22) is also in line with this rule. Antipassives are, after all, ‘accusative-style’ transitive structures.

The reason for this phenomenon is easy to detect. As Plank observes,

Looking at the same matter from a different angle, one can then also add ‘ergativity’ to the correlates of high transitivity, and ‘accusativity’ to those of low transitivity (3.35). According to Hopper and Thompson, “[c]haracteristically, we find any of the following correspondences:

9.27 Implications for (Pre-)Proto-Indo-European

As we shall see later, the nexus between perfectivity and ergative syntax on the one hand, and between imperfectivity and accusative/antipassive syntax on the other, is of great importance for the evolution of the (Pre-)PIE verbal system. But already at this point, two things are worth stressing:

1. 1. The observation that few of the world’s languages display full (non-split) ergativityFootnote ⁹¹ cannot be used as an argument against an ergative reconstruction of Pre-Proto-Indo-European. Although Villar’s and Rumsey’s criticism of the ergative hypothesis is wrong-headed (9.11), we must concede that it would be unwise to posit a fully ergative Pre-PIE grammar without any split. However, given the widespread existence of partially ergative languages in which the split between ergatival and accusatival patterns relates to the verbal rather than the nominal domain, there is an obvious alternative.Footnote ⁹² We must envisage a system in which tense and/or aspect distinctions correlated with different ways of encoding eventualities that involve more than one participant.

2. 2. Once we do this, the ergative hypothesis offers a simple answer to the question how tense distinctions (present vs. past) could be expressed in the proto-language before ‘primary’ *-i was introduced (1.2, 10.18). In past-tense contexts, ergative structures may have been preferred, and in present-tense ones non-ergative structures (‘antipassives’ of whatever type: 9.25).Footnote ⁹³ Positively to prove this is of course difficult, but one or two morphological points can be cited in support.

   Firstly, the addition of ‘primary’ *-i seems more deeply rooted, and hence older, in the ‘ergatival’ m-series of PIE endings. If the h₂-series, where ‘primary’ *-i is still absent from the Greek and Indo-Iranian perfect (5.3), was naturally presentic anyway, it would stand to reason that it required a present-tense marker less urgently.

   Secondly, in the thematic conjugation the 1sg. pair (pres.) *-oH : (past) *-om is structurally less regular than 2sg. *-esi : *-es and 3sg. *-eti : *-et (1.5). How exactly 1sg. *-oH is related to the 1sg. in *-h₂(e) will occupy us later (10.4), but it is hard to deny that there is some link between the two, and that *-oH is therefore best rewritten as *-oh₂. So, the tense-conditioned *-oh₂ : *-om contrast in the thematic 1sg. concurs with the notion that, to begin with, m-series endings had a past-tense affinity and h₂-series endings a present-tense one.Footnote ⁹⁴

Meanwhile, Pre-PIE aspectual differences stricto sensu appear not to have been conveyed by the selection of ergatival vs. non-ergatival syntax/endings. As worked out earlier and last diagrammed in Fig. 8.1, perfectivity as such was signalled by reduplication in the earliest recoverable stages of the proto-language. The ergative hypothesis does not necessitate any modification of this theory. There is nothing to suggest that, for example, the use of a 1sg. in *-(o)m was ever unavailable in the imperfective past (imperfect) .

And yet, the subsequent evolution of aspect was not to remain unaffected by the aspectual dimension of ergativity. As will be argued in 10.41, it is precisely the change from Pre-PIE (split) ergativity to PIE accusativity that provides the key to all the aspectual restructurings we have dealt with in the preceding chapters. But before we can go there, we must first return to our unfinished discussion of the origin of the s-aorist (8.54). By now, we finally have all the theoretical equipment we need in order to bring it to a conclusion.Footnote ⁹⁵

9.28 From Pronouns to Personal Endings

In 9.12–9.14, we have encountered the idea that the personal endings of Proto-Indo-European go back to once independent personal pronouns. We have seen that there is a match at least between 1sg. *-m and the pronominal oblique stem *m-, which should also have been the stem used in the oblique ergative case. Whether there is a similar connection between the element *-h₂- in 1sg. *-h₂e and the laryngeal element *-h₂- in the pronominal nominative *(h₁)eg-h₂-oH/-om (> Gr. ἐγώ, Lat. ego, Skt. ahám) is much less certain; but given the tendency of emphatic pronouns to be renewed by accretion, it must not be excluded.Footnote ⁹⁶ In the 2sg., the pronominal t- stem is common to both the oblique and the nominative cases (cf. e.g. nom. Gr. (Dor.) τύ, Lat. tū, Skt. tvám < *tu(H)(-om), acc. Gr. σέ, Skt. tvā(m) < *tu̯-ē̆(-om), Lat. tē < *t-ē), but only replicated by the verbal ending *-th₂e next to 1sg. *-h₂e. The likeness between the 1sg. and 2sg. here suggests that a simpler 2sg. ending *-t (vel sim.) was remade into *-th₂e under the influence of the 1sg. For the m-series with 2sg. *-s, there is no such correspondence and one may therefore have to concede a complete renewal (cf. Footnote 9.29, fn. 100). In view of the *t- also in the oblique cases of the pronoun, one would rather expect a verbal ending 2sg. *-t next to 1sg. *-m.Footnote ⁹⁷

Turning to the 3sg. in *-t, a connection with the pronominal stem *to- has long been suspected.Footnote ⁹⁸ However, if one pays attention to the stem suppletion between *s(o) and *t(o)- (9.4) and adheres to an ergative framework, the stem *t(o)- should be represented primarily in the ‘non-ergatival’/‘absolutival’ h₂-series of personal endings, not in the ‘ergatival’ m-series. The distribution of *t(o)- vs. *s(o) requires that the latter, which eventually yields the non-neuter nominative, be the ergatival variant and the former the absolutival one.Footnote ⁹⁹ At the same time, the m-series should then feature a 3sg. in *-s (~ pronominal *so). In all these cases, an initially fuller-bodied pronoun may have been reduced to the shape we actually observe in the ending. If we allow for that, and therefore concentrate on the consonantism alone, the following rough-and-ready prediction of the ending patterns results:

9.29 Zero-Endings and Full Endings in the 3sg.

Importantly, though, the 3sg. must be viewed differently from the 1sg. and 2sg. In sentences like I strike the man or You strike the man, the logical subject (A = ‘I’/‘you’) is encoded in the personal pronoun. In God strikes the man, on the other hand, no pronoun is involved. So if we reconstruct a linguistic stage where the verbal idea was conveyed by an invariant form, to which the relevant participants were added as either pronouns or nouns, the basic 3sg. ending will be *-Ø, no matter if a transitive or an intransitive construction is at stake:

The 3sg. endings posited in 9.28 should only occur where there is no separate A/S. Under those circumstances, the demonstrative stem might step in anaphorically:

In other words, if a pronominal origin of the 3sg. endings is accepted, it entails an earlier situation where the 3sg. commonly bore no ending at all and where *-s and *-t were at best ‘optional’:

9.30 Spreading 3sg. *-t and the Genesis of Suffixal *-s-

The 3sg. ending *-Ø in the ‘absolutival’ h₂-series matches the absence of a 3sg. person marker in the PIE stative and perfect. As noted before (cf. 4.35, 5.27, 5.40–5.41), the final -e/o in 3sg. *CéC-e/o and *Ce-CóC-e, respectively, was originally part of the nominal stem behind these categories, not a true 3sg. ending. But what about the ‘ergatival’ m-series?

Due to the changes mentioned in 9.28 – the remake of the 1sg. pronoun, the remodelling of the 2sg. ‘absolutive’ ending *-t into *-th₂e, and the phonological attrition of the original pronouns when used as ‘endings’ – the connection between pronouns and endings was gradually obscured. The endings thus became independent grammatical morphemes, yielding a system in which the 3sg. in *-Ø, without a personal marker, contrasted with the fully marked 1sg. and 2sg. forms. To create distinct 3sg. endings out of the hitherto ‘optional’ 3sg. endings (*-s/*-t) must have been tempting at that point. However, this process may well have begun only when the accusativisation of the alignment system was already under way. If so, it would not have been clear that the ‘optional’ 3sg. in *-s should be confined to transitive sentences and the ‘optional’ 3sg. in *-t to intransitive ones. Rather, the ‘optional’ 3sg. in *-t would have had its natural locus both in intransitive contexts and in transitive contexts going back to antipassive structures (9.24–9.25). The ‘optional’ 3sg. in *-s, on the other hand, would have been confined to transitive contexts of non-antipassive origin – that is, the transitive contexts at the upper end of the semantic transitivity scale (cf. 9.26). With a functional distribution of this kind, the form in *-t had to be regarded as the ‘normal’, since least restricted, 3sg. marker. After all, the syntactic transitive/intransitive distinction had now lost its relevance to the selection of verbal endings. What newly counted was voice (9.20), but the 3sg. slot of the new early ‘middle’ voice could adequately be filled with the once nominal *CéC-e/o and *C(o)C-é forms instead (cf. 10.19).Footnote ¹⁰¹

Once *-t had thus become the default 3sg. person marker, it spread to all active 3sg. forms, including high-transitive (perfective) ones that hitherto ended in either *-Ø or ‘optional’ *-s. This could have spelled the end of ‘optional’ *-s. But what happened instead may have been the following. The competition between general (incl. transitive) 3sg. *CeC-t and high-transitive-only *CeC(-s) was resolved by adding the now universal person marker *-t to the form in *-s as well. Already before that, however, the restriction of *-s to the high-transitive domain meant that this element was no longer treated as an oddly deviant alternative person marker next to *-Ø and *-t, but as a signal of high transitivity – and hence, in the end, aspectual perfectivity (cf. 8.7, 8.55). Mutatis mutandis, this refunctionalisation of a previously ergatival element as a perfective marker is reminiscent of what happened to the ergative case ending -k in Mingrelian (9.21).

Figure 9.8 summarises the entire development that affected transitive structures, showing how the ultimate outcome was a split between a high-transitive s-paradigm and a root paradigm.

Fig. 9.8. A new model for the genesis of the s-aorist

9.31 Assessment

Unlike earlier explanations, the new scenario satisfies the requirements formulated in 8.55; namely, (a) that the 3sg. *CeC-s structure, which phonological considerations suggest to be the nucleus of the s-aorist paradigm, be characterised by high transitivity, and (b) that the same *CeC-s structure regularly stand next to a *CeC-t structure without *-s-, which will eclipse it whenever the s-less variant is prototypically telic too. Because we are dealing with a paradigm split, this virtual parallelism naturally follows from the fact that the later s-aorist and root aorist share a common starting point.

9.32 Conclusion

Although certainty cannot be reached in any such matter, the above derivation of the s-aorist is built around a widely accepted premise: that the PIE verbal endings are ultimately of pronominal origin. On this basis, it combines the results of our philological dissection of the s-aorist with an independently established and typologically viable model of Pre-PIE alignment (9.28–9.31). In the next chapter, we shall have to investigate whether the same model is also able to do justice to, and shed further light on, the other parts of the PIE verbal system we have been looking at.

To pave the way for this finale has been one aim of the present chapter. By retracing the history of research on Pre-PIE alignment since its beginnings (9.1–9.20), we have seen that what started off as an attempt to explain certain intriguing details of nominal inflection also has major implications for the verb, notably with regard to the two PIE conjugation ‘classes’ in *-m(i) and *-h₂e. While various scholars have tried to explain this dichotomy as a relic from an earlier ‘active’ (‘split-S’) stage of Pre-Proto-Indo-European, the same evidence can be integrated more easily into an ergative framework; and since the nominal data equally favour an ergative reconstruction, this is where we should look, no matter if such an ergatival system were in its turn the successor of an ‘active’ one (9.19).

For the classical ergative hypothesis to be rehabilitated, we must realise that typological objections raised against it are unfounded (9.8–9.11). These not only mix up diachronic and synchronic considerations, but also forget that, although few languages are purely ergatival, split ergativity is conditioned by (verbal) tense and aspect just as often as by (nominal) animacy. It is therefore proposed that Pre-Proto-Indo-European featured a split-ergative system in which a past/present distinction correlated with an opposition of ergative and antipassive structures (9.26–9.27). In combination with the simple spread of A marking to the S participant of intransitive sentences, the existence of such antipassives – reinterpreted as regular transitives – provides a plausible pathway for the transformation of the Pre-PIE ergative system into the accusative system of later Proto-Indo-European (9.21–9.25). The historical active vs. middle voice contrast may then be seen as the product of an intermediate ‘fluid-S’ phase (9.19–9.20).

The concrete paradigmatic changes involved in all this are still to be outlined. In order to do so, we must be prepared to continue operating on uncertain ground. As always, the ultimate measure of success can only be if, in the end, we manage to combine plausibility in the detail with coherence overall.