3 The intransitive two-word stage: Absolutives, unaccusatives, and middles as precursors to transitivity

The previous chapter discussed (root) small clauses, which were argued to be evolutionary precursors to more complex (TP) counterparts, with both still in use to varied degrees in present-day languages. It is of note that all the small clause data included in Chapter 2 involved intransitive clauses, that is, clauses with a subject but no object. While small clauses in present-day languages can definitely be transitive, my argument is that the proto-grammars in the paratactic (non-hierarchical) stage were intransitive.¹

There are many reasons to postulate that proto-syntax started intransitive. First of all, children’s language acquisition proceeds through a two-word stage, as noted by many (e.g. Bloom 1970); a two-word stage can accommodate a predicate with only one argument, and thus cannot be transitive, at least not without positing various null categories in order to bridge the gap between adult grammars and early child grammars.² Similarly, early stages of sign languages constructed from scratch also seem to show a two-word, intransitive stage, as discussed below. Next, many fossils of proto-grammars are intransitive two-word structures, including certain compounds and unaccusative and passive-like small clauses (see also Chapter 2).

In addition, paratactic combinations of (small) clauses are almost always binary, merging only two clauses at a time (1–2). Thus, two-word grammars creating a (small) clause are paralleled by “two-clause” grammars:

(1)

(2)

Combining more than two paratactic phrases/clauses typically leads to a processing quagmire, as the following example helps illustrate:

(3)

This example is an expanded version of attested binary examples from pidgin languages, such as “No mani, no kom” from Hawaiian Pidgin English (Winford 2006). Unlike the relatively clear message behind “No money, no come,” it is hard to know how to interpret (3). Is it that if you do not come, then you cannot get paid, or get any shelter? Or does it mean that if you do not come with the money, then you will not get any shelter? Or is it a prediction or a threat that you won’t come, won’t get the money, and won’t get any shelter either? The grammar on its own cannot decide among these options.

This is not just an example involving the familiar kind of ambiguity, as found in e.g. “He saw the man with the binoculars,” where language users typically reach for one interpretation and do not even consider the other(s). With the one in (3), we are at a loss right away. It seems that our brains are just not prepared to readily assign meanings to paratactic ternary structures such as (3). But we can handle binary structures.

In Minimalism (e.g. Chomsky 1995) and predecessors, the central combinatorial operation Merge is widely considered to be binary, that is, it is considered that Merge can combine only two elements at a time (see e.g. Kayne 1984). The same assumption holds for the operation Adjoin, which is akin to paratactic attachment (see Chapter 4).³ As a consequence of binary Merge, it is considered in Minimalism and predecessors that binary branching is a syntactic universal, characteristic of all languages. To be more accurate here, because it was empirically determined/discovered that the vast majority of syntactic structures across languages can be analyzed as involving binary branching, the operation Merge was hypothesized to only be able to combine two words/phrases at a time.

If so, then the initial proto-syntax, characterized by parataxis, could not have been transitive in the modern sense of the word, given that transitivity involves three obligatory constituents (subject, verb, object), and accommodating three such constituents structurally would require hierarchical syntax.⁴ At least this is the claim in Minimalism: on top of the small clause (or VP layer) in transitive structures one must project another verbal layer, the layer of vP, as discussed later in the chapter. If, as I argue, proto-syntax did not have hierarchical capabilities, then it could not have had true transitivity.

But, can there be languages without transitivity? How would one express the basic notions such as “who does what to whom” in such languages? At first glance, such grammars might seem impossible to imagine. However, as will be shown in this chapter, there are many constructions in present-day languages that exhibit exactly such non-transitive properties.

A good initial illustration is provided by the emergence of Nicaraguan Sign Language (NSL) by deaf children in the 1970s and 1980s, to be discussed further in Section 3.5 (see also Aronoff et al. 2008 for Al-Sayyid Bedouin Sign Language, which exhibits similar properties). According to Kegl, Senghas, and Coppola (1999: 216–17), the earliest stages of NSL, with the first generation of speakers, do not utilize transitive N V N constructions, such as (4), at least not when both nouns are animate (Senghas et al. 1997). Instead, the speakers resort to a sequence of two intransitive clauses, an N V—N V sequence (5–6), clearly resembling the paratactic structures in (1–2):

(4)

(5)

(6)

Focusing on (6), one can say that the sign for WOMAN is the subject of PUSH, but the sign for MAN here is not the object of PUSH, but instead the subject of FALL. In this kind of grammar, there are no structural objects, as these structures are intransitive.⁵ Similar considerations hold for Homesign syntax, as reported in e.g. Goldin-Meadow (2005), to be discussed in Section 3.5. But, one can argue, this may just be a phenomenon of early stages of sign languages, and nothing like that is possible in spoken languages.⁶

However, this chapter will go over a variety of present-day structures which blur the subject/object distinction in this same way. One example of this kind of grammar is the so-called exocentric VN compounds, which necessarily consist only of two words (i.e. two free morphemes), a verb and a noun, resembling small clauses (see Section 3.3.2 for more details). If Givón’s (1971) well-known slogan “today’s morphology is yesterday’s syntax” has some truth to it, then it provides additional support for the claim that the mold these compounds are poured into may be just fossilized syntax of an earlier era. One more recent example of a compound which preserves a stage of English syntax is the name for the plant forget-me-not. While English speakers no longer use this kind of syntax in sentences (e.g. *You forgot me not), it is preserved in this particular compound.⁷

In the underlined compounds in (7) the noun is subject-like, while in the rest of the compounds it is object-like, as discussed below.

(7)

Even though these compounds contain a verb, and the verb takes one argument (the noun), which is typically object-like, it would be wrong to analyze such compounds as transitive. First of all, clearly, there is no second argument in these compounds, which would count as a subject. Also, the noun is not necessarily object-like, but can also be subject-like, as is the case with the underlined compounds. While a scarecrow is somebody who scares crows (crow is object-like), a rattlesnake is a snake that rattles (thus subject-like), and a cry-baby is a baby (or somebody) who cries (again subject-like). But the nouns in both of these cases appear in exactly the same position and the same form in the compound, following the verb, and thus there is no formal differentiation between object-like and subject-like arguments in this sense. This is quite comparable to the clauses characterizing early stages of Nicaraguan Sign Language, as illustrated above in (5–6). The VN compounds in other languages, including Serbian, show exactly the same properties in this respect, as discussed in detail in Chapter 6.

Furthermore, the intransitive constructions in some modern ergative languages share this property as well. In these languages, the subject of the intransitive clause is structurally not distinguishable from the object, both appearing in the so-called absolutive case, as illustrated in the following example from Tongan (Austronesian language spoken in Tonga; Tchekhoff 1979: 409):⁸

(8)

As the two distinct translations indicate, the only argument (the fish) can be interpreted here as either the subject or the object of eating, once again illustrating an intransitive grammar which does not make a formal distinction between subjects and objects. As pointed out in Tchekhoff (1973), as well as by other researchers (e.g. Authier and Haude 2012; Blake 1976; Mithun 1994: 247; Shibatani 1998: 120), the subject/object distinction does not play a role in such ergative/absolutive patterns, especially those which are both syntactically and morphologically ergative, as will be explained below. In addition to these, several other absolutive-like constructions found in present-day languages, in fact languages classified as nominative-accusative, will be considered in this chapter, including unaccusatives, nominals, and middles.

The main proposal in this chapter is that the initial paratactic (non-hierarchical) grammars were intransitive grammars, whose clauses consisted of just two (proto-)words. In this proposal, transitivity is seen as an innovation brought about by superimposing an additional layer of structure (perhaps the vP layer of Minimalism) upon the foundational (absolutive) layer, with some intermediate “middle” constructions paving the way toward transitivity. Not only can this approach shed light on the ergative-absolutive and nominative-accusative dichotomy found across today’s languages, but it can also explain the availability of the foundational absolutive-like patterns in various guises in primarily nominative-accusative languages. The recurring theme of this monograph is that each stage preserves, and builds upon, the achievements of the previous stage(s). In this case, the proposal is that transitive structures (vP shells), as well as middles, are built upon the foundation of intransitive (absolutive-like) VPs (or small clauses), shedding light on a host of quirky phenomena across languages.

As was the case with the small clause proposal in Chapter 2, this proposal also involves an internal reconstruction based on the theoretical postulates within Minimalism. Just as one can peel the TP layer off a modern sentence (Chapter 2), one can also peel off the vP layer, resulting in intransitive small clauses. Recall that the reconstruction method used in this book is based on the hierarchy of functional projections which allows a SC/VP to be composed without a TP or vP, but does not allow either a vP or a TP to be composed without a VP/SC. This renders the proposed progression of stages theoretically plausible.

In the process of evolving transitivity, i.e. grammaticalizing the syntactic positions of more than one argument, I propose that there are/were various types of intermediate steps, as discussed in Section 3.4. The evidence for these intermediate stages includes various “middle” constructions, which straddle the boundary between transitivity and intransitivity, passives and actives, as well as neutralize the distinction between subjects and objects. I exemplify this with se middle constructions to be introduced below, where se is analyzed as a meaningless proto-transitive marker.

As with the analysis of small clauses in Chapter 2, the argument for the proposed progression through stages (absolutive, to middle, to transitive) has three prongs to it: (i) evidence of “tinkering” with the language design, so that fossils of one stage provide foundation for the next, possibly through intermediate steps; (ii) identifying “living fossils” of each stage in modern languages; (iii) existing or potential corroborating evidence. Moreover, the goal is to show that each identified stage accrues concrete and tangible advantages over the previous stage(s), advantages that are significant enough to be targeted by natural/sexual selection.

In this respect, Section 3.2 shows how intransitive absolutive-like structures get built into the transitive (vP) structures, thus providing evidence of evolutionary tinkering with the language design. Section 3.3 introduces further living fossils of the postulated absolutive-like stage in the evolution of syntax. Section 3.4 considers middle constructions and serial verb constructions, both of which straddle the boundary between intransitivity and transitivity. There is also some corroborating evidence for an intransitive stage, as well as potential testing grounds, as discussed in Section 3.5.

This chapter postulates a stage in the evolution of syntax in which only intransitive absolutive-like patterns were available, i.e. patterns in which a verb takes only one argument, and in which the distinction between subjects and objects is neutralized, in fact, irrelevant. This is the sense in which I am using the term “absolutive-like” in this context, just to indicate that an intransitive structure does not distinguish subjects from objects grammatically. This is not to imply in any way that there was a special marking of an ergative argument, to contrast with the absolutive one. My proposal is that this intransitive proto stage could only have one argument per clause. From there, one can see how ergative and nominative languages would have diverged in the way they express additional arguments in sentences. Ergative languages would have kept the absolutive pattern for intransitive sentences, but added ergative arguments to this absolutive structure in order to express transitive patterns. On the other hand, nominative-accusative languages would have developed a special, accusative case only for the lower argument, establishing a category of the object. It could be that certain middle constructions in the latter languages paved the way toward developing the accusative case, as discussed in Section 3.4.2.

It should be pointed out that this is a very different view from the one that would advocate missing or null arguments. In this analysis, one is dealing with a two-slot grammar with only one argument slot, and there is nothing missing or null syntactically speaking.⁹ This is a perfectly coherent grammar, even if simpler than e.g. transitive grammars. Developing such a grammar would have constituted an enormous advantage over no grammar at all, but this kind of grammar has less expressive power than a fully transitive grammar, exactly the kind of scenario that would allow evolutionary forces such as natural selection to operate (see Chapters 2, 4, and 7). Pressure to accommodate additional arguments would have been a powerful driving force behind the evolution of more complex (transitive) patterns.

This proposal is entirely consistent with the analysis of transitivity in e.g. Minimalism, where transitivity is considered to involve an additional layer of verb structure, a vP shell (e.g. Chomsky 1995). In this analysis, the internal (lower argument) is generated in the VP (or SC), and the external argument (e.g. agent) in the vP (9–10), as discussed in Section 1.7.

(9)

(10)

In deriving the sentence in (9), one starts with the basic, small clause layer in (10a). Then, the agent (Maria) is merged in the higher vP layer (10b), which is responsible not only for accommodating this additional argument, but also for assigning (abstract) accusative case to the object (the ball). Finally, the TP layer is projected on top of the vP layer, and “Maria,” the highest argument, moves to become the subject of the TP (10c).

Thus, just as is the case with the small clause vs. TP distinction discussed in Chapter 2, here as well we have a layer of structure (vP) superimposed upon the foundational, absolutive (small clause) layer. In both cases, the small clause with one argument is the foundation. In more elaborate grammars, full transitive sentences will have all three layers, arranged in a hierarchy of projections (see e.g. Abney 1987):

(11)

Assuming this kind of structure building in Minimalism, my proposal in fact does an internal reconstruction to arrive at the intransitive small clause proto-syntax, as proposed in Chapters 1 and 2, and repeated below:

Internal reconstruction of clause structure (based on Minimalism)

A structure X is considered to be (evolutionary) primary relative to a structure Y if X can be composed independently of Y, but Y can only be built upon the foundation of X.

This hierarchy of functional projections is an influential theoretical construct, with good empirical foundation, and it is significant that it can be used to reconstruct proto-syntax.

While Chapter 2 provided evidence for living fossil structures without a TP, my focus in this chapter is on the fossil structures without vPs. In fact, intransitives, especially unaccusatives (see Section 3.3.1), can be accommodated without the vP layer (12–13), as discussed in Section 1.7. In other words, the vP layer is optional.¹⁰

(12)

(13)

Given that there is no agent, and no accusative case either, the vP shell need not project in (13a). In English, the object-like argument (the ball) has to move to the TP projection and become a structural subject (13b).¹¹

The verbs like roll, which participate in both transitive and intransitive patterns, clearly show fluidity in the expression of subjecthood (see also Sections 1.7 and 3.4). Observe that (9) and (12) start with exactly the same foundation, the small clauses in (10a) and (13a), respectively. Whether the ball will be the object or the subject of the sentence depends on whether or not there is an additional argument. What counts as a subject is thus relative to the number of arguments expressed.¹²

Recall from Chapter 2 that in the absence of the TP layer in unaccusative small clauses in Serbian of the kind in (14), only one layer of structure is available, the [SC/VP] layer:

(14)

In conjunction with the examples above, we see a gradual progression toward more syntactic complexity, from one single layer of structure in (14), to two layers of structure in English tensed unaccusatives (13), to three layers of structure with English tensed transitive clauses (10c), abstracting away from some other functional projections that may be there. Crucially, this gradual increase in complexity is arrived at not through impressionistic means, but by a precise method of internal reconstruction based on theoretical considerations.

Grammaticalizing transitivity in e.g. nominative-accusative languages, with a structural accusative case and the vP/VP shell, would not have precluded some other structures (e.g. unaccusatives, se clauses, nominals, compounds) from retaining the absolutive-like flavor. If these simpler grammars are easier to process, then their retention at least in some constructions is to be expected.

This section has shown that intransitive (absolutive-like) structures get built into the transitive vP shells, providing the necessary foundation for transitivity, thus offering evidence of evolutionary tinkering with the language design. The following section introduces further types of living fossils of the postulated absolutive-like stage in the evolution of syntax.

In this section I consider in more detail the following “living fossils” of the postulated absolutive-like stage in the evolution of human language: unaccusatives (Section 3.3.1), exocentric VN compounds (Section 3.3.2), absolutives in ergative-absolutive languages (Section 3.3.3), as well as (other) absolutive-like constructions found in nominative-accusative languages, including nominals and dative subject clauses (Section 3.3.4).

Unaccusative small clauses were introduced in Chapter 2, where the focus was to establish that such clauses are structures without a TP layer, showing neither Move nor subordination. What is relevant about them in this chapter is that they are intransitive structures which can be generated without projecting the vP layer either. This kind of grammar is a good approximation of the hypothetical two-word stage, as discussed in Section 3.1, as well as in Chapter 2. Moreover, this kind of grammar is reminiscent of the grammar found in exocentric VN compounds, as discussed further in Section 3.3.2, as well as in Chapter 6.

As pointed out in Section 3.2, unaccusatives can be accommodated without projecting the vP layer:

(15)

Recall that vP is projected primarily in order to accommodate an additional argument, typically the agent, as well as the accusative case, but unaccusative structures have only one argument and no accusative case (hence their name). Unaccusatives can be roughly characterized as intransitive structures whose sole argument is typically a theme, showing some object-like properties, including the postverbal position in some cases (see e.g. Perlmutter 1978; Burzio 1981; Levin and Rappaport Hovav 1995, for crosslinguistic manifestations and characterizations).

In Serbian, for example, there is a clear preference for unaccusative “subjects” to follow the verb, the position typically associated with objects. If these unaccusatives are at the same time TP-less small clauses (e.g. 16), this preference becomes more rigid, with strong preference for the otherwise non-canonical VS order (see Progovac 2008a,b for details):

(16)

This provides support for the unaccusative analysis under which the “subjects” of the unaccusative verbs (e.g. arrive, fall, come, appear) Merge as “objects” of the small clause (e.g. Burzio 1981).

Recall that unaccusative small clauses in Serbian are analyzed as involving one single layer of structure, (SC/VP) layer, and that their subjects thus have no syntactic position into which to move (Chapter 2; see also Section 3.3):

(17)

Without a vP or TP layer, these are just rigid and flat two-word structures, reasonably good approximations of the postulated two-word stage.¹³

Given this proposal, the unaccusativity phenomenon can be seen as an option to retain (elements of) absolutive-type grammars in constructions which can be supported by such grammars, e.g. intransitive constructions with a single (internal) argument (Casielles and Progovac 2010, 2012).¹⁴ To put it another way, if proto-syntax involves less grammatical burden, and is less costly to process, then one can expect to find it in constructions in which more complex grammars do not confer much advantage. Transitive constructions, as well as intransitive constructions involving agents in some languages, may need extra syntactic space, e.g. a vP shell, and thus cannot be expressed as readily with this type of grammar.

Intransitive absolutive constructions in ergative languages, as discussed in Section 3.3.3, as well as various absolutive-like constructions elsewhere, are again structures which blur the distinction between subjecthood and objecthood in the sense that their only argument has characteristics of both.¹⁵ The only difference seems to be that unaccusatives tend not to encompass agents, and this may be due to the special status agents have acquired in nominative-accusative languages, possibly by being associated with their own functional projection, such as vP in Minimalism. In other words, even if there is an association of the role of agent with the vP in some modern languages today (but see Progovac 2014b), this association was certainly not there in the two-word grammar stage. What the two phenomena have in common, absolutives and unaccusatives, is the unavailability of the accusative case, that is, the structural case which is reserved only for objects.¹⁶

Another phenomenon that is difficult to explain given the postulates of modern morpho-syntax are exocentric VN compounds of the kind illustrated in (18–19) below for English and Serbian. However, their shape makes sense if they are seen as fossils closely approximating a two-word absolutive-like (intransitive) stage in the evolution of human language (see Progovac and Locke 2009; Progovac 2009a, 2012).

(18)

(19)

The grammar behind these compounds is quite simple: it is a two-place mold that can fit exactly one verb and one noun, whether the verb is semantically monovalent or bivalent. Moreover the thematic role of the noun is underspecified. While the noun in these compounds is often interpreted as an internal argument, corresponding to an object in a sentence, it can also be external, corresponding to a subject in a sentence, as is the case with the underlined compounds in both languages. As pointed out in Section 3.1, the noun in crybaby is subject-like, while the noun in scarecrow is object-like.

Once again, the grammar behind these compounds provides no morpho-syntactic differentiation between subjecthood and objecthood, leaving room for vagueness. For example, a rattlesnake is conventionally interpreted as a snake that rattles, but one can imagine this word also used for somebody who routinely rattles snakes, on analogy with e.g. pick-pocket.¹⁷ Notice that a more complex compound, snake-rattler, which has a transitivity layer, is no longer vague in this way, and can only be interpreted as somebody who rattles snakes.

Perhaps a better way to make this point would be to consider a contrast between a turn-coat and a turn-table. Conventionally, a turncoat is somebody who turns his coat/skin inside out (traitor), with the coat being object-like. On the other hand, a turntable is a table that turns, where the table is subject-like. But if a turntable can be a table that turns, then, in principle, grammatically speaking, a turncoat could be a coat that turns, perhaps a coat that is reversible. Likewise, if a turncoat can be somebody who turns his coat, then, in principle, a turntable could be someone who (routinely) turns tables upside down, perhaps a rowdy regular in a bar. Again, this flexibility is not there with syntactically more elaborated compounds, such as table-turner, which cannot mean, not even in principle, a table that turns.

Exocentric VN compounds can thus be seen as absolutive-like constructions which blur the distinction between subjecthood and objecthood, and which also lack accusative case, the properties these compounds share with unaccusatives and (other) absolutives.

It is of interest that exocentric VN compounds across languages seem to specialize for derogatory reference when referring to humans, possibly implicating their original use in ritual insult. Chapter 6 explores the proposal that the ability to create such compounds in the distant evolutionary past may have been sexually selected, contributing to the consolidation of proto-syntax, as well as to vocabulary building.

As pointed out by a reviewer, there are many other compound types in English, combining other categories, such as an adjective and a noun (blackbird), a noun and a noun (snowman), a noun and an adjective (sky-blue). There are several reasons why they are not considered in this monograph, although future research might reveal relevance of some of these for evolutionary considerations, perhaps compounds of the egghead type. First of all, even though N-N compounds in English may seem simple and straightforward at first sight, they are typically not only headed (the second element is the syntactic and semantic head of the whole compound), but they are also recursive, producing: styrofoam snowman, or policy committee decision process. While it may seem that such combinations of nouns directly reflect our cognitive abilities for headed Merge and recursion, it is worth pointing out that not all languages in fact use such compounds, and especially not recursively (see Section 1.6; Chapter 6; also Snyder 2014).

In contrast, as will be discussed in Chapter 6, VN compounds are typically neither headed (hence the name exocentric) nor recursive. Moreover, they are relevant for the purposes of this book because they are combinations of a verb and a noun, typical building blocks of clauses/sentences, and the first categories to emerge and be differentiated (e.g. Heine and Kuteva 2007). Finally, these VN compounds reveal evidence of ritual insult, rendering them of particular interest for evolutionary considerations for that reason as well. Chapter 6 offers additional reasons for their evolutionary significance.

The vagueness attested in exocentric VN compounds is also characteristic of intransitive absolutives in some ergative-absolutive languages. Consider another example from Tongan featuring an intransitive sentence with the absolutive case (Tchekhoff 1973: 283):¹⁸

(20)

In this intransitive sentence, Mary can be interpreted either as the agent of the action, or the patient/theme. But, as pointed out in Tchekhoff (1973), this sentence means neither “Mary calls” nor “Mary is called” in Tongan, these being just two different translations of one single underdetermined/underspecified structure in Tongan. In other words, these translations reflect our nominative/accusative bias. Instead, all this sentence means is that there is calling, and that Mary is implied in the process (Tchekhoff 1973: 284). This is also the essence of Gil’s analysis of vague sentences in Riau (Footnote 18), as well as my proposal for middles in Serbian, and the idea of a proto-role (Section 3.4.2).

In ergative-absolutive patterns, the subject of an intransitive predicate is morpho-syntactically equivalent to the object, both characterized as absolutive arguments (e.g. Comrie 1978; Dixon 1994). Only agents of transitive verbs are marked distinctly, with ergative case.¹⁹ It is only after the addition of the external ergative argument (e.g. John below) that the role of Mary disambiguates and is necessarily patient/theme (Tchekhoff 1973: 283). In other words, the addition of the external argument forces the inner absolutive layer to distinguish itself from the external argument, resulting in more precision.²⁰

(21)

The examples in (20–21) illustrate quite clearly how the ergative argument (John) is inserted into the basic absolutive layer. They also illustrate something that has been noted repeatedly in the typological literature, that the ergative-absolutive structures resemble passive structures in nominative-accusative languages, in which the agent is introduced as an oblique argument, e.g. as a by-phrase in English passives, as discussed further below (see e.g. Hale 1970). These similarities extend to the nominal domain as well, as discussed in the following section.

Dyirbal (Australian language spoken in northeast Queensland) is another ergative language which, like Tongan (see also Dukes 1998), exhibits syntactic ergativity, in the sense that the absolutive role even in transitive constructions continues to behave in a subject-like fashion, as illustrated with a coordinated structure below (Dixon 1994: 155):

(22)

This clearly contrasts with English (23) below, in which a comparable coordinated structure yields the opposite result for the missing argument:

(23)

In fact, if we were to coordinate a passive sentence and an active sentence in English, we would get the pattern comparable to the one in (22) from Dyirbal:

(24)

This is the sense in which the ergative phrase can be likened to the passive by-phrase. The by-phrase here, just like the ergative phrase in (22), is not the true, structural subject, but only the “logical” subject, as will also be discussed with respect to the noun phrases in Section 3.3.4.1. This is also the reason behind the proposals in Nash (1996) and Alexiadou (2001) that ergative phrases may be attached by adjunction, in a way similar to the attachment of the passive by-phrase in English.

While Tongan and Dyirbal are analyzed as syntactically ergative languages, in the sense that they exhibit both morphological and syntactic ergativity, there are many ergative languages spoken today which exhibit only morphological ergativity, patterning with English with respect to e.g. coordination (see Aldridge 2008 for an overview and discussion; thanks also to Robert Henderson, p.c. 2013, for a discussion on this). Likewise, ergative-absolutive languages typically show the so-called split-ergativity, in the sense that they are ergative with some nouns/pronouns, but accusative with other nouns/prounouns, as discussed in Section 7.3.3. Tongan has also developed certain morphological constructions that can be analyzed as accusative patterns (see e.g. Tchekhoff 1973). It may well be that every language has some ergative and some accusative patterns, and it is only a matter of which patterns prevail.

Assuming that there was an intransitive absolutive-like (proto-syntactic) stage in the evolution of human language, one can envision the subsequent development of the two basic language types, primarily nominative-accusative and primarily ergative-absolutive. Lehman (1985: 245) points to the gradient nature of the distinction between the ergative and accusative types: “a language is never wholly and exclusively either ergative or active or accusative, in all its grammatical patterns.” As pointed out in this section and in the following sections, there are many absolutive-like constructions in nominative-accusative languages. Likewise, so-called ergative languages often develop nominative-accusative patterns in some domains, e.g. in the domain of personal pronouns (which are higher on the animacy hierarchy), resulting in so-called split ergativity (e.g. Trask 1979 and references there; see Chapter 7 for more discussion). This overlap is what one would expect under the evolutionary approach explored here.

Bringing unaccusativity and ergativity under the same umbrella, Bok-Bennema (1991: 169) points out that ergativity and unaccusativity are both characterized by the inability of transitive verbs to assign structural case to their deep objects. To put it another way, neither ergative nor unaccusative structures can have true (syntactic) objects, that is, objects distinguished from subjects by means of a specific structural case (see Footnote 16). According to e.g. Alexiadou (2001: 18; also Hale 1970; Nash 1995), ergative/absolutive patterns are reflexes of a passive/unaccusative system. Therefore, what all these phenomena have in common (absolutives, exocentric VN compounds, unaccusatives, and passives) is that the verb is unable to assign structural case to its deep object. Given that the object does not receive a distinct (accusative) marking, the distinction between subjecthood and objecthood is blurred.

These phenomena begin to make sense if they are seen as survivors from a two-word proto-syntax stage, which could only accommodate one argument per verb, and which did not have the means to distinguish between subjects and objects. As pointed out above, it is perfectly plausible to expect that the absolutive-like patterns will be preserved in some constructions, especially those in which subject/object differentiation is not important. It is also conceivable under this approach that the foundational absolutive-like patterns will be found in some guise or another in nominative-accusative languages as well, as explored further in the following section. Languages may vary considerably with respect to the degree to which they rely on the foundational absolutive-like patterns, but my argument is that every language has at least some constructions which are absolutive-like in nature, providing continuity and common ground between the two language types.

As noted in e.g. Authier and Haude (2012: 2) “some notoriously ‘accusative’ languages such as Latin, French, and in fact many Indo-European languages may have some hints of ergativity” (see also Bauman 1979: 430; Lehman 1985). Such hints of ergativity have already been introduced in this chapter for English and Serbian exocentric compounds, as well as for unaccusatives. This section considers additional constructions that can be seen in a similar light, including nominals (Section 3.3.4.1), dative subjects (Section 3.3.4.2), and clausal complements (Section 3.3.4.3).

This section is there to show that even in English one finds, in productive use, these absolutive-like structures which do not distinguish subjects from objects, resulting in vagueness. According to e.g. Alexiadou (2001), nominals across various languages are intransitive, as well as absolutive-like (passive-like). In other words, all nominals, whether passive or not, have an intransitive base (see also Picallo 1991; Bottari 1992; Alexiadou and Stavrou 1998). In passive nominals the agent appears as an adjunct, as in (25) from Alexiadou (2001: 78).²¹

(25)

In this analysis, by-phrases in derived nominals can only be interpreted as affectors (agents, instruments, creators), rendering examples such as (26) not fully grammatical.²²According to the authors, unlike with the verbal domain, there is no structural external argument in nominalizations, generated in a vP (or an nP, the nominal equivalent of a vP), and the presence of the by-phrase seems to be lexically licensed. In that sense, the external argument in the by-phrase resembles ergative case, which is also often analyzed as lexical/prepositional case, rather than structural case (see above).²³

(26)

This is consistent with the proposal in this chapter that the intransitive, absolutive-like/passive-like patterns provided a foundation for evolving transitive structures, with ergativity and accusativity being different solutions to the same problem of accommodating an additional, external argument.

Consider next dative “subjects” in Serbian, which co-occur with nominative “objects” in what certainly looks like an ergative/absolutive pattern:

(27)

Nominative on the “object” is like absolutive, being also the case of intransitive subjects, while dative adds an external argument, akin to an ergative (see e.g. Alexiadou 2001; Nash 1996, for an adjunction analysis of the ergative argument). As pointed out in e.g. Trask (1979: 398), the ergative case is often identical to the genitive, dative, or locative. According to Nash (1996: 171), ergative subjects, like dative subjects, cannot co-occur with structural accusative, but instead appear with absolutive/nominative “objects.” This is yet another construction in which the verb fails to assign structural (accusative case) to what would be its object.

It is also of significance here that dative subjects in Serbian typically co-occur with the (middle) pronoun se. As per the proposal in Section 3.4.2, se is associated with the ancient absolutive-like pattern.

The clausal complements of the so-called raising predicates, such as seem, appear, likely, as well as of predicates such as obvious, are also absolutive-like/unaccusative-like in nature. While they are generated as complements of the verb, they do not receive accusative case, and there is no external argument either, which is reminiscent of the unaccusative grammars.

(28)

(29)

For purely grammatical purposes, the subject position of these sentences hosts an expletive (meaningless) pronoun it, but this pronoun is certainly not an argument of the verb. In fact, what looks like an external argument can optionally be added, as in:

(30)

(31)

Intriguingly, when it comes to comparable predicates in Serbian, their external argument, if expressed, would appear as a dative subject:

(32)

Both Serbian mi and English to me can be seen as a type of ergative case, added to the otherwise absolutive foundation. This just shows that various quirky and exotic-looking phenomena across languages can be understood in this evolutionary framework.

As a reviewer rightly points out, also of relevance to this discussion are the so-called serial verb constructions, widespread in Creole languages, in the languages of West Africa, Southeast Asia, Amazonia, Oceania, and New Guinea. Serial verb constructions can be characterized as sequences of verbs “which act together as a single predicate, without any overt marker of coordination, subordination, or syntactic dependency of any other sort,” describing what is conceptualized to be a single event (e.g. Aikhenvald 2005: 1). What one observes in these examples again is that there is one argument per verb, and the relationship of that argument to the verb seems absolutive-like. According to Givón (1979: 220), serial verb constructions involve “a concatenation of small propositions in which, roughly, a one-to-one correlation is maintained between verbs and nominal arguments.”

Aikhenvald further states that these constructions are monoclausal, and that their intonational properties are the same as those of a monoverbal clause, sharing just one tense, aspect, and polarity value. Importantly, she also mentions that serial verbs do not necessarily have to be next to each other, as they are in (34), but can also be separated by other constituents, as in (33).

Anyi-Sanvi (Kwa family, Niger-Congo: Van Leynseele 1975: 191–2)

(33)

Igbo (Igboid, Benue-Congo, Niger-Congo; Lord 1975: 27)

(34)

It is of note here that (33) in essence has an N V—N V structure, comparable to the structures attested in Nicaraguan Sign Language (e.g. WOMAN PUSH – MAN FALL) (see e.g. Section 1.6). Intriguingly, Senghas et al. (1997: 558) claim that the N V – N V structures of the first generation of NSL signers tend to become N VV N (WOMAN PUSH FALL MAN), or N N VV (WOMAN MAN PUSH FALL) combinations, with the second generation. In other words, the second generation is grouping the two verbs so that they are adjacent to each other, as is typically the case with serial verb constructions. As Senghas et al. (560) conclude in the article, the next stage may be a stage where transitivity emerges, with only one verb remaining to support both nouns. This is essentially the path of grammaticalization envisioned in e.g. Givón (1979: 220).

While I would like to leave a more detailed investigation of serial verb constructions from this perspective for future research, it is worth pointing out that developing an intricate system of such constructions may have been yet another route toward transitivity, by embracing the dual-verb structures. If so, it is significant that the intransitive foundation (one absolutive-like argument per verb) provides the common ground for such a wide range of strategies for expressing transitivity.²⁴ In fact, this astonishing variety of strategies for expressing transitivity across languages and constructions seems to have nothing in common except for this paratactic, absolutive-like foundation. If true, this provides significant support for the reconstruction explored in this chapter, and for the gradualist approach to the evolution of syntax more generally.

According to e.g. Kemmer (1994: 181), “the reflexive and the middle can be situated as semantic categories intermediate in transitivity between one-participant and two-participant events.”²⁵ Here I consider just one representative example: se constructions in Serbian that can be characterized as middles as they straddle the boundary between the passive and active voice. In addition to dative subject clauses introduced in the previous sections (3.3.4.2 and 3.3.4.3), se is also used in a wide variety of other constructions in Serbian, and may well be one of the most frequently used words in the language.

Where pragmatics allows, se constructions in Serbian exhibit astounding vagueness of meaning, and se clearly cannot be analyzed as a reflexive pronoun, reflexivity being only one of the available interpretations, and not even a preferred one, as the following examples illustrate:

(35)

(36)

(37)

If (37) is uttered with a sense of urgency, the most probable interpretation will involve the most salient discourse participant, the speaker, even though there is no word or morpheme corresponding to the first person at all! Even though (38) below offers an unambiguous way of expressing the first reading of (37), (38) is much less likely to be used in the heat of the moment, suggesting that se constructions are easier to process than regular transitives:

(38)

It is significant that the vagueness in se clauses illustrated above is comparable to that found with Tongan absolutives (20) and Riau intransitives (Footnote 18), as well as with exocentric compounds. In (36), as apparent from the translations, Marko can be either the subject (agent), or the object (patient), or both at the same time, the latter option yielding the reflexive interpretation. This kind of ambivalence can only be a result of underspecification, that is, of simple, unarticulated syntax and semantics.²⁶ Given this, the meaning of (37) and (38) can be roughly characterized in the following way:

(37’)

(38’)

It is probably more accurate to characterize (38’) as (38’’) below, building directly on the middle pattern in (37’):

(38’’)

This would essentially mean, as discussed in this chapter, that the basic absolutive layer is still preserved even in (38), and that it is by virtue of superimposing a higher argument that the initial participant is now interpreted as a non-agent, in this case as patient/theme. This is exactly what we see with the Dyirbal data in Section 3.3.3.

Interestingly, Dowty (1991) also questions the rigidity and descreteness of theta roles, and proposes that they can instead be seen as prototypes, or proto-roles, such as proto-agent and proto-theme roles (thanks to a reviewer for leading me in this direction). The participant role that I am using here can then be seen as an even more underspecified role, just a proto-role. This is then how one can characterize proto-predication – as involving a verb (predicate) and just one argument, with a proto-role of a participant.

The presence of se simply implies that there is one more participant involved in the event, in addition to the one surfacing, and typically its role can be inferred from pragmatic context (e.g. 37). But the role of the expressed argument (e.g. Deca, Pas, or Marko above) still remains absolutive-like, not grammatically specified as either subject or object, giving rise to massive ambiguities (Progovac 2005a, 2013b, 2014a,b). The constructions in (35–36) once again illustrate a two-word grammar at work (this time enhanced by the particle se). Even though, pragmatically speaking, one is dealing here with an event with two participants, this kind of fossil syntax cannot express both arguments, nor can it specify whether the only expressed argument is subject or object.²⁷

Comparable vagueness may also be found with cognate se constructions in other Slavic languages, but also in Spanish (Arce-Arenales, Axelrod, and Fox 1994: 5), clearly indicating that the phenomenon illustrated above is not just a quirk of Serbian:²⁸

(39)

Serbian se is analyzed in Franks (1995) and Progovac (2005a) as an expletive (meaningless) pronoun, “absorbing” accusative case. Another way to look at it is to say that se in these constructions is a proto-transitive/proto-accusative marker imposed upon an ancient absolutive pattern, but being stuck in this intermediate stage between absolutivity and transitivity. As pointed out by Maggie Tallerman (p.c. 2014), these se constructions, as well as other constructions which I consider “transitional” in this framework, are not transitional in the sense that they are unstable or in the process of changing – they can only be transitional in the sense that they straddle the boundary between transitivity and intransitivity.

It is hard to be sure how to analyze these se constructions by using the tools of Minimalism, and the derivation in (40) is just a suggestion:

(40)

Again, the idea is that the noun and the verb are first Merged in a SC/VP (Section 1.7). Next, a proto-transitive functional word se is Merged with the SC/VP to create some kind of functional projection, whose head is se, and which can be labeled as FP.²⁹ Finally, assume that the TP is created, and the noun deca Moves to the specifier of the TP. FP is still not a vP, as it does not introduce an agent, nor does it disambiguate the role of the absolutive-type argument in the SC/VP, but it can be considered as a precursor to vP. The next step(s) in developing vP-type transitivity in accusative languages would be to associate this FP with an additional, external argument, such as agent, and to associate the internal argument with the special (accusative) case.

Interestingly, without se, the absolutive pattern vanishes, and the only argument has to be interpreted as subject/agent performing an action on an unspecified object, as is also the case with English translations in (41) and (42), a familiar consequence of accusative grammars:

(41)

(42)

This suggests that the fossil absolutive-like structures in Serbian are only preserved under the wing of se (as further explored in Progovac 2014b).

It seems, then, that the distinctions between subjecthood and objecthood, transitivity and intransitivity, passive and active, can be neutralized, and can have a middle ground. One way to make sense out of this is to postulate an intransitive absolutive-like stage in the evolution of human language, a stage which provides a foundation for any subsequent elaboration of argument structure.

Importantly, however, introducing transitivity with a structural accusative case (vP/VP shell) to a language does not preclude some other constructions (e.g. unaccusative small clauses, nominals, se constructions, compounds) from remaining absolutive-like. What is also important to emphasize is that many of these foundational structures still live inside/within the more complex structures. For example, absolutives generated in small clauses/VPs arguably live inside nominals, se constructions, and transitives, and small clauses in general live inside TPs, as commonly assumed in syntactic theory (Chapter 2). This reinforces the claim in this monograph that small clauses and intransitive absolutives constitute the foundation, the platform on top of which one can build (or not) more complex syntax, namely TPs and transitivity, perhaps in the form of vP shells.

Transitivity in syntax thus need not be seen as conceptual necessity, but rather as an evolutionary innovation; it can be seen as an additional layer of structure superimposed upon the foundational (absolutive) layer, leading to a variety of crosslinguistic strategies for marking case relations, and reflected in the postulation of two verbal layers in Minimalism (two vP shells). This renders syntax a quirky system, a product of tinkering, rather than a system optimally designed from scratch. As was the case with the small clause/TP distinction discussed in Chapter 2, the hypotheses explored in this chapter are testable/falsifiable, as well as corroborated by evidence from other fields, as discussed in the following section.

The strongest corroborating evidence for the proposal in this chapter comes from language acquisition, both involving sign languages and spoken languages. Neuroimaging would, once again, provide a good testing ground for the hypotheses proposed in this chapter.

As pointed out in Section 3.1, the emergence of NSL provides excellent corroboration for the proposal. According to Kegl, Senghas, and Coppola (1999: 216–17), the earliest stages of NSL, observed with the first generation of speakers, do not exhibit transitive N V N constructions, such as (43) below, at least not when two animate nouns are involved (Senghas et al. 1997). Instead, one finds what look like sequences of two clauses of the kind (N V—N V) (44–45):

(43)

(44)

(45)

Aronoff et al. (2008, and references there) found a similar pattern for another sign language that emerged spontaneously about 70 years ago, Al-Sayyid Bedouin Sign Language (ABSL). They also report that there is a tendency toward one argument per predicate, where e.g. transitive events involving two animate referents are rendered by two or even three clauses.

These sequences can be analyzed as paratactic/symmetric combinations of two intransitive (small) clauses, which are interpreted as the first one causing the second.³⁰ In this sense, this grammar is absolutive-like, and resembles the grammar behind serial verb constructions and other absolutive-like constructions discussed in this chapter, in that only intransitive structures are available, that is, each verb can have only one argument.

It is perhaps of interest to mention here that the overwhelming majority of the world languages are classified as either Subject-Verb-Object (SVO) or Subject-Object-Verb (SOV). Both types can be derived easily from a binary N V – N V pattern, comparable to the paratactic patterns in (44–45). If one starts with an N V – N V sequence, and assigns the role of S (subject) to the first noun, and the role of O (object) to the second noun, one can easily derive the two word orders above by dropping one of the verbs (the dropped verb could then be grammaticalized as a null light verb (v), as per the syntactic theory). There is another symmetric paratactic possibility: V N – V N, the verb initial order being attested in e.g. unaccusatives and VN compounds. If, again, the first noun is associated with S, and the second noun with O (as per the Cause First principle discussed in e.g. Section 1.6), this underlying pattern can easily yield the SVO order again, but also another possible word order across languages, VSO. The other logically possible word orders are extremely rare across languages. Needless to say, this is a rather speculative observation.

According to Goldin-Meadow (2005), the syntax of Homesign languages, self-styled gestural communication systems spontaneously developed by deaf children not exposed to sign language, also appears to be absolutive-like. In Homesign, both patients/themes and intransitive agents tend to precede verbs, once again neutralizing the distinction between subjects and objects. Also, patients are more likely to be expressed than agents, as is also the case with exocentric compounds and nominals discussed in the previous sections. As Goldin-Meadow notes, both American and Chinese deaf children are more likely to produce the sign for the eaten than for the eater. In Zheng and Goldin-Meadow’s (2002: 171–2) study, the Chinese children showed a bias to omit only the subjects of caused motions (agents), not the subjects of spontaneous motions. Subjects of spontaneous motions were produced as often as objects.

Considering that early stages of NSL, ABSL, and Homesign are languages arguably constructed from scratch, the patterns of intransitivity and ergativity observed in their creation are of evolutionary significance (see Section 2.5.1 for much more discussion regarding the reasons why language acquisition can be relevant for language evolution studies). At the very least, these considerations demonstrate that there is a simpler way to break into syntax, starting with intransitive clauses and blurring the distinction between subjecthood and objecthood.

Moreover, children acquiring spoken languages also go through a two-word stage (Bloom 1970) which seems to be characterized by similar proto-syntactic patterns. It is often claimed for child language acquisition that children “delete” arguments in their speech, that is, that they do not express all the arguments that would typically be required in the adult grammar.³¹ According to Zheng and Goldin-Meadow (2002: 171–2), such “deletions” are not random, but rather follow an ergative pattern. If children in these cases are using absolutive-type intransitive grammars, as per the proposal in this chapter, then they are not deleting anything, but rather just using the syntactic mold in which there is room for expressing only one single argument.

Similar patterns in language acquisition of spoken languages have been reported by other authors. For example, when hearing children are exposed to Korean (Clancy 1993) or Samoan (Ochs 1982), they too follow the deaf children’s pattern—they omit transitive subjects and produce intransitive subjects and objects, exhibiting essentially an absolutive pattern. Indeed, the same pattern has been observed for English language acquisition (Goldin-Meadow and Mylander 1983: 63). As Zheng and Goldin-Meadow (2002: 171–2) conclude, the ergative pattern is more robust, considering that the omission pattern found in all of these hearing children and the deaf children is reminiscent of the alignment found in ergative languages. This ties in well with the approach explored in this chapter.

As pointed out by a reviewer, the preferred discourse pattern in a variety of languages is the pattern in which only one argument is given in full, while the other arguments are either omitted altogether or occur in a reduced (affix) form (see e.g. Newmeyer 2005: 132–3, and references there). For example, Du Bois (1985: 347–9) found that in Sacapultec, a Mayan language of Guatemala, most clauses in the discourse contain only one full noun phrase, with zero noun phrases also very common. The full NP that commonly occurs is the absolutive, consistently following the verb, while the ergative full noun phrases are infrequent.

In addition, Du Bois (1987) has noted that the pattern in which the grown is expressed more readily than the grower is common in the adult languages of the world, as attested with the intransitive constructions in (b) from English:

(46)

(47)

While the transitive pattern in (a) necessarily takes John to be an agent, the intransitive counterparts in (b) favor the interpretation in which John is undergoing the action, as a theme/patient. In other words, there is avoidance of agents/external arguments in these cases (see also Casielles and Progovac (2010, 2012) for the significance of this phenomenon for language evolution, and in particular for the Agent-First hypothesis).

Interestingly, the bonobo Kanzi has been reported to have mastered a VS (two-word) syntax in his use of lexigrams and gestures, based on the description in Greenfield and Savage-Rumbaugh (1990: 161), as well as Heine and Kuteva (2007: 145–7). First of all, Kanzi only uses two-word combinations, including creations with one verb and just one argument, in a way that does not distinguish agents/subjects from patients/objects, with both following the verb. While Kanzi’s initial combinations (during the first month) show free word order (hide peanut, peanut hide), the later combinations seem to converge on the productive VS order, even when the noun is the agent, in the sense that the verb is followed by an agent gesture.³²

There is a lot of controversy surrounding the interpretation of these and other reports on primate communication, and it is not my intention to engage with these controversies in this book. For now, suffice it to say that, if Kanzi is in principle capable of (sporadic) two-word (intransitive) combinations, then it is conceivable that at least some individuals of our common ancestor with bonobos were too. This would have been enough to allow the process of natural selection for language.

Last but not least, as pointed out in Chapter 2, neuroimaging can provide a fertile testing ground for the hypotheses explored in this chapter. The suggestion is that one can use the subtraction and other neuro-linguistic methods to determine how proto-syntactic structures are processed in comparison to their more complex counterparts, in the hope of finding neuro-biological correlates of, for example, vP shells and transitivity (see Progovac 2010b).

For the reasons given in the Appendix, while the processing of transitives with vP shells is expected to show clear lateralization in the left hemisphere, with extensive activation of specific Broca’s areas, the proto-syntactic structures, such as absolutive-type constructions, as well as middle se constructions, are expected to show less lateralization, and less involvement of Broca’s area, but more reliance on both hemispheres, as well as, possibly, more reliance on the subcortical structures of the brain.To take just one concrete example (not discussed in the Appendix), it follows from the analysis presented in this chapter that se constructions (and middles in general) are easier to process than regular transitives, given that they involve simpler, less articulated syntax. This can be tested given the availability of minimally contrasting pairs in Serbian involving se constructions (48) vs. true transitive counterparts (49), as suggested in Progovac (2014a,b):

(48)

(49)

If syntax evolved gradually, through several stages, then it is plausible to expect that modern syntactic structures and operations decompose into evolutionary primitives. If so, this will not only be measurable in the activation of the brain, but without these evolutionary considerations it may not be possible to achieve a true breakthrough in the field of neuro-linguistics (see Section 2.5.3).

This chapter builds on the arguments of Chapter 2, and reconstructs a stage in the evolution of human language which is characterized by intransitive small clauses, lacking vP and TP structure, and allowing only one proto-argument per clause, that is, an argument whose thematic role is underspecified. This stage is arrived at by internal reconstruction based on the syntactic hierarchy of functional projections. Peeling off the outer clausal layers, TP and then vP, one arrives at the basic predication structure of an intransitive small clause. As with the proposal in Chapter 2, there are three prongs to this argument. First, the absolutive-like pattern is shown to provide a foundation upon which transitive structures are built. Second, there is a variety of absolutive-like foundational structures even in nominative-accusative languages. And, third, there is good corroborating evidence and promising testing grounds for this proposal. Furthermore, postulating an intransitive absolutive-like stage allows one to clearly identify the kinds of evolutionary pressures that would have led to the rise of transitivity, as explored in Chapter 7.