1. Introduction
Syntactic ergativity is usually defined in terms of interclausal syntax – indeed Dixon (Reference Dixon1994) equates ‘syntactic ergativity’ with ‘interclausal ergativity’ in the chapter title ‘Inter-clausal or syntactic ergativity’. The notion of a syntactic pivot was developed to deal with the fact that languages differ in the set of arguments they take as privileged for the purposes of syntactic processes like clause combining and extraction. The terms S, A, and O are used to refer to the single argument of a one-participant verb, the more agentive argument of a two-participant verb and the non-agentive argument of a two-participant verb, respectively.Footnote 2 In some languages only S and A are accessible to the type of syntactic processes mentioned above, while in others only S and O are. It is the latter pattern which is usually described as syntactic ergativity. Not all languages have pivots, and pivotless languages clearly cannot be said to show either ergative or accusative interclausal patterning. Since syntactic ergativity is generally equated with interclausal ergativity, it follows that pivotless languages should, by their very nature, show no signs of syntactic ergativity.
Models of syntax differ in how easily they are able to accommodate ergativity at the level of intraclausal syntax – i.e. in the structure of the verb phrase.Footnote 3 Some formal syntactic theories unambiguously assume a clause structure where external arguments are ‘subjects’ in the sense of A and S, and so cannot easily incorporate a notion of syntactic ergativity at the intraclausal level. At the same time, even theories which allow for intraclausal ergativity in the sense of an ergatively structured verb phrase – i.e. the A argument of a transitive clause being internal to the verb phrase while the O is external to it – assume the VP-external argument position to be linked to pivothood; that is, external position is associated with the syntactically most privileged argument, whether this argument is A or O in a transitive construction.
This paper argues that the Oceanic language Äiwoo shows an ergatively structured verb phrase, while lacking a syntactic pivot. In other words, the fact that the VP-external argument position in Äiwoo is filled by S and O, at least in some clause types, has no discernible consequences for the structural organisation of the language as a whole. This raises fundamental questions concerning the theoretical significance of the external argument position: external arguments are associated with an especially prominent syntactic status (typically equated with ‘subjecthood’) in most theories of syntax, but no such status is apparent for external arguments in Äiwoo.
The paper is structured as follows: Section 2 discusses the notions of ‘pivot’ and ‘syntactic ergativity’. Section 3 describes the basic properties of Western Austronesian voice systems which are relevant to the analysis of Äiwoo. Section 4 presents the evidence for a verb phrase in Äiwoo which includes V and A, but not O or S, while Section 5 shows that Äiwoo lacks a syntactic pivot. Section 6 proposes a historical explanation for this situation, and Section 7 discusses the implications of the Äiwoo data for syntactic theories in which the external position is assumed to have a privileged status. Section 8 offers concluding remarks concerning the demonstrated lack of a universal association between VP-external position and a privileged status in syntactic structure.
2. Subject, pivot, and syntactic ergativity
The notion of a syntactic pivot was developed as a result of the recognition that the grammatical properties associated with subjects in traditional grammar do not pick out a single category in all languages (Keenan Reference Keenan and Li1976; Schachter Reference Schachter and Li1976, Reference Schachter, Cole and Sadock1977). The pivot of a clause is the set of arguments that are accessible to syntactic processes of clause combining and extraction, e.g. the arguments which may be omitted under coreference in coordination constructions or extracted in constructions such as relative clauses or question formation (e.g. Dixon Reference Dixon1994: 11). A language which restricts the set of arguments that may be involved in such constructions to S and A is said to have an S/A pivot, while one that restricts it to S and O is said to have an S/O pivot. The pivot of a language may or may not match the argument singled out by other properties often thought to be characteristic of subjects, such as being the agent argument if there is one, being the addressee of imperatives, or controlling reflexives (Keenan Reference Keenan and Li1976).
The idea that there are languages where traditional ‘subject properties’ do not all pick out the same argument was first introduced by Schachter (Reference Schachter and Li1976), who showed that in Tagalog and other Philippine languages, some subject properties are characteristic of the actor argument, whereas others are characteristic of the ‘topic’ or ‘pivot’, the syntactically most prominent argument which is picked out by these languages’ characteristic voice morphology (see Section 3 below). Schachter suggests that there are in fact two types of ‘subject’ properties, which he calls ‘reference-related’ and ‘role-related’, respectively. Reference-related properties derive from the fact that subjects are typically referentially prominent, and include, on Schachter's account, the ability to relativise, the ability to launch floating quantifiers, and the inability to occur in existential sentences. Role-related properties derive from prominence on a hierarchy of semantic roles, and include being the addressee of imperatives, the antecedent of reflexives, and the controlled argument in control constructions (see Section 5.4 below for more discussion of control), as well as typically being ordered before any other nominal in a clause.
Various terms have been employed for the two categories defined by the two sets of properties. Dixon (Reference Dixon1994) uses the term ‘subject’ for the set of S and A, essentially a semantic grouping in that A is defined as encoding ‘that role which is most likely to be relevant to the success of the activity’ (Dixon Reference Dixon1994: 8), whereas he uses ‘pivot’ for the set of arguments picked out by rules of clause combining. Manning (Reference Manning1996) distinguishes between ‘a-subject’, the most prominent argument at the level of argument structure, and ‘gr-subject’ or pivot, the most prominent argument at the level of grammatical relations. Falk (Reference Falk2006), who, like Manning, works within the framework of Lexical-Functional Grammar (LFG), uses the terms 
for the function of encoding the argument which is most prominent on the thematic hierarchy, and pivot or piv for the element with the function of connecting its clause to other clauses in the sentence.Footnote 4 The discussion in this paper takes Falk's analysis as its starting point, for two reasons. Firstly, the LFG framework on which it builds allows for crosslinguistic variation in verb phrase structure, i.e. it incorporates the notion of an ‘ergative verb phrase’, which is at the centre of my discussion. Secondly, it proposes a principled explanation for this crosslinguistic variation – though, as will be shown below, an explanation which does not apply to the facts of Äiwoo. My purpose, however, is not to propose changes to the LFG framework on the basis of the Äiwoo data, but to show that the data poses a challenge to any theory which assumes a universal function of particular argument positions within a clause, even when that theory is flexible enough to allow for crosslinguistic variation in which arguments can be found in particular positions.
Dixon (Reference Dixon1994: 175) and Van Valin & LaPolla (Reference Van Valin and LaPolla1997: 275–278) take pivot to be a construction-specific property, in order to account for cases where a single language appears to exhibit different pivots in different construction types. Falk (Reference Falk2006), on the other hand, considers pivot to be a language-specific property, that is, each language has only one type of pivot. However, for each of the complex constructions typically taken to involve a pivot, there are alternative strategies a language may use which do not involve a pivot, and which therefore may not pattern in the same way as those which do involve a pivot. Of these strategies, the most directly relevant for the present paper is that of the use of anaphoric devices such as null pronouns or agreement to indicate the identity of arguments across constructions, which will be discussed further in Section 5 below. Falk's claim is that in so-called mixed-pivot languages, the constructions in question use different strategies, giving the appearance of different pivots in a single language, where in fact some of the constructions do not involve a pivot at all.
The term ‘syntactically ergative’ is largely equated in the literature with ‘having an S/O pivot’ (see also Levinson, in press). Dixon (Reference Dixon1994: 144) states that ‘[o]nly some languages work in terms of a pivot, and it is only these that can be characterised as “accusative” or “ergative” … at the level of inter-clausal syntax’. Falk (Reference Falk2006: 79) similarly defines a syntactically ergative language as one where ‘the piv is the obj if there is one’. A language which lacks a pivot, then, should not be characterisable as syntactically ergative – nor, for that matter, as syntactically accusative.
However, there are a number of languages attested which show what appears to be an ergative intraclausal structure, in the sense of having constructions where the argument internal to the verb phrase in transitive clauses is the A rather than the O. This is, however, most commonly seen in languages which show complex morphosyntactic mechanisms for altering the pivot of a clause, while having no obviously ‘basic’ construction from which the others are derived – a language type often referred to in the literature as ‘Philippine-type’ because it was first described for languages of the Philippines. It should be noted, however, that in the Austronesianist literature, the term ‘Philippine-type’ is often used for a subset of languages of this overall ‘symmetrical voice’ type, contrasted e.g. with ‘Indonesian-type’ languages showing a somewhat different set of morphological markers, but retaining the basic distinction between an ‘actor-pivot’ and an ‘undergoer-pivot’ construction, where neither is clearly ‘basic’ or unmarked (Ross Reference Ross, Wouk and Ross2002, Arka & Ross Reference Arka, Ross, Arka and Ross2005, Himmelmann Reference Himmelmann, Adelaar and Himmelmann2005). In the following, I will therefore simply refer to ‘(Western) Austronesian voice systems’ when referring to languages showing marking of this type.
3. Austronesian voice systems
Western AustronesianFootnote 5 languages are known for their complex voice systems, sometimes also referred to as ‘focus systems’. The main characteristic of such systems is that they mark the verb for the role of the syntactically most prominent argument, distinguishing minimally between an ‘actor voice’, which has the actor as the most prominent argument, and an ‘undergoer voice’ or ‘object voice’, which has an undergoer as the most prominent argument. Some languages distinguish further between several non-actor voices, such as e.g. location voice or circumstantial voice. Typically, all voices show an overt morphological marker on the verb; that is, none of the voices can be identified as ‘basic’ with the others being derived from it. The examples in (1) below illustrate this for Tagalog, which has an actor voice (AV) and three undergoer voices: a patient voice (PV), a location voice (LV), and a circumstantial voice (CV). The prenominal marker ang, glossed as ‘specific’ in the source because one of the properties of the ang-marked argument is that it is necessarily interpreted as specific, marks the argument which is singled out as prominent by the voice morphology of the verb, and it can be shown that the ang-marked argument functions as the pivot of its clause (Schachter Reference Schachter and Li1976, Reference Schachter, Cole and Sadock1977).
(1) Tagalog (Schachter Reference Schachter and Li1976: 494–495; glosses from Ross Reference Ross, Wouk and Ross2002: 26)
(a) Mag-alis ang babae ng bigas sa sako para sa bata.
av-take.out spec woman npiv rice loc sack for loc child
‘The woman will take some rice out of a/the sack for a/the child.’
(b) A-alis-in ng babae ang bigas sa sako para sa bata.
dur-take.out-pv gen woman spec rice loc sack for loc child
‘A/the woman will take the rice out of a/the sack for a/the child.’
(c) A-alis-an ng babae ng bigas ang sako para sa bata.
dur-take.out-lv gen woman npiv rice spec sack for loc child
‘A/the woman will take some rice out of the sack for a/the child.’
(d) Ipag-alis ng babae ng bigas sa sako ang bata.
cv-take.out gen woman npiv rice loc sack spec child
‘A/the woman will take some rice out of a/the sack for the child.’
There is considerable crosslinguistic variation in the exact formal properties of such voice systems. For some languages, the actor-voice construction has been argued to be formally intransitive (see Ross Reference Ross, Wouk and Ross2002: 26–31 for a discussion of the issues involved in determining the transitivity of a clause in Tagalog), while in others it is clearly transitive. In Tagalog, Kroeger (Reference Kroeger1993) argues that there is no verb phrase, at least not in all constructions; a number of other languages do show a verb phrase, the postverbal argument of a transitive construction forming a constituent with the verb (Ross Reference Ross, Wouk and Ross2002: 54; Himmelmann Reference Himmelmann, Adelaar and Himmelmann2005: 142–143). Crucially, which argument enters into this constituent depends on the voice marker on the verb: the ‘focused’ argument, i.e. the one picked out by the voice morphology, tends to be either clause-initial or clause-final, and so the postverbal argument will be a non-focused core argument, i.e. the undergoer in the actor voice and the actor in the undergoer voice.
The pattern can be illustrated by example (2) from Balinese, which has two voices, an actor voice and an undergoer voice (typically described in the literature on Balinese as ‘object voice’); note that the undergoer voice in this language is morphologically unmarked, unusually for a language of this type.
(2) Balinese (Wechsler & Arka Reference Wechsler and Arka1998: 388)
(a) Tiang [numbas bawi-ne punika].
1 av.buy pig-def that
‘I bought the pig.’(actor voice)
(b) Bawi-ne punika [tumbas tiang].
pig-def that ov.buy 1
‘I bought the pig.’(object voice)
Wechsler & Arka (Reference Wechsler and Arka1998) apply a number of tests to show that in Balinese, the preverbal argument, i.e. the one picked out by the voice marking, is the pivot, the privileged argument for the purposes of constructions such as relativisation, raising, and control, while the verb forms a constituent with the non-pivot argument, as indicated by brackets in (2a–b) (see also Manning Reference Manning1996: 27–31). A system like this, where the NP picked out by the voice marking is the pivot and the non-pivot NP is part of the verb phrase, effectively alternates between syntactic ergativity and accusativity in the undergoer vs. the actor voice. The structure where the non-pivot NP is part of the verb phrase while the pivot NP is external to it is in line with the general assumption in formal syntactic theories that the position external to the verb's domain is associated with ‘subjecthood’ in the sense of a privileged status in the syntactic makeup of a language. In what follows, however, I will show that there is no necessary correlation between external structural position of an argument and a syntactically privileged status.
4. The verb phrase in Äiwoo
4.1 The Äiwoo languageFootnote 6
The Äiwoo or Reefs language is spoken in the Main Reef Islands in Temotu Province, the easternmost province of Solomon Islands in the southwest Pacific Ocean. It belongs to the so-called Reefs-Santa Cruz language group, which for a long time was cited as an example of a group of languages of mixed origin, having a largely Austronesian vocabulary but a non-Austronesian (Papuan) grammatical core (e.g. Wurm Reference Wurm and Wurm1976, Reference Wurm, Wurm and Carrington1978, Reference Wurm, Gonzalez and Thomas1981, Reference Wurm, Dutton, Ross and Tryon1992). However, recent research has shown that the evidence for a Papuan substrate is flawed, and has conclusively placed Äiwoo and the other Reefs-Santa Cruz languages in the Oceanic subgroup of Austronesian, as originally argued by Lincoln (Reference Lincoln, Wurm and Carrington1976) (Næss Reference Næss2006, Ross & Næss Reference Ross and Næss2007, Næss & Boerger Reference Næss and Boerger2008). Current evidence points to the Reefs-Santa Cruz languages, together with the languages of nearby Utupua and Vanikoro, forming a previously unidentified first-order subgroup of Oceanic, which has no members outside Temotu Province (Ross & Næss Reference Ross and Næss2007: 471). Taken together with the archaeological evidence (Spriggs Reference Spriggs1997, Sheppard & Walter Reference Sheppard and Walter2006), this suggests that these islands were settled in a very early population movement straight from the Bismarck Archipelago, the presumed homeland of Proto Oceanic, and that the population has lived in its present location in relative isolation from the rest of the region for well over 3,000 years. It is therefore not surprising if we find features in these languages that are unknown in other Oceanic languages, or that have been retained in the languages of Temotu Province while largely being lost in the rest of Oceanic.
The main grammatical properties of Äiwoo which are of relevance to the account developed in this paper are outlined below.
Äiwoo is a head-marking language where the person/numberFootnote 7 of intransitive subjects is marked by prefixes on the verb, as shown in (3) below, while transitive subjects – and, in some cases, objects – are marked by suffixes, as shown in (4). Äiwoo person marking must consequently be considered to show tripartite alignment, though it might be argued that S and A to some extent have a privileged position in this system, in the sense that not all objects get marked on the verb. There is no object suffix for 1min, 3min, or for augmented-number objects in combination with non-3min subjects; these may instead be referred to by independent pronouns.
(3) (a) I-ku-wä. (b) Mi-ku-wä.
1min.S-ipfv-go 2min.S-ipfv -go
‘I go.’ ‘You go.’
(c) Ku-wä. (d) Ku-lu-pwä.
ipfv-go ipfv-3aug.S-go
‘S/he goes.’ ‘They go.’
(4) (a) I-togulo-no. (b) I-togulo-mu.
pfv-hit-1min.A pfv-hit-2min.A
‘I hit him/her.’ ‘You hit him/her.’
(c) I-togulo-nee-mu. (d) I-togulo-gu-i.
pfv-hit-1min.A-2min.O pfv-hit-3min.A-3aug.O
‘I hit you.’ ‘S/he hit them.’
Person marking on verbs is the main means of distinguishing between arguments, in combination with word order (see below); there is no case marking of nouns or independent pronouns.
Basic word order with intransitive verbs is SV, shown in (5a). With the verbs described above as transitive, i.e. those taking subject suffixes, the order is OVA when the object is a noun, as in (5b); pronominal objects are postverbal, as in (5c).
- (5)
(a) Temaale lâ ku-basiki=to=wâ.
needlefish deic:dist ipfv-run=cs=deic:dist
‘The needlefish ran off.’
(b) Lâto nuwo i-luwa-kä toponu
thus bottom pfv-take-dir:3 turtle
eä nupaa i-luwa-kä lâpu.
conj top pfv-take-dir:3 rat
‘The turtle took the bottom half, and the rat took the top.’
(c) I-lowâle-mä-mu iu.
pfv-help-dir:1-2min.A 1min
‘You helped me.’
This paper is mainly concerned with the properties of OVA-type clauses. Äiwoo also has a distinct set of two-argument verbs which show AVO order; these are typically used when the emphasis is on the performing of an action rather than its effect on an object. The object noun in this construction cannot be indexed on the verb; instead, the verb takes intransitive subject marking (recall example (3) above). The verbs occurring in OVA and AVO clauses are formally related; no productive derivational process exists which derives one from the other, but there are recurring patterns of alternation, e.g. vei (AVO) – vili (OVA) ‘weave’, kei (AVO) – kili (OVA) ‘dig’; läke (AVO) – läki (OVA) ‘chop’, benge (AVO) – bengi (OVA) ‘block’; tou (AVO) – tu (AVS) ‘carry, bring’, gou (AVO) – gu (OVA) ‘husk’; dâwââ (AVO) – dââ (OVA) ‘tie up’, eâwââ (AVO) – eââ (OVA) ‘pull’. The AVO construction will be briefly discussed in Section 6 below.
4.2 Tests for VP
The aim of this section is to show that in OVA clauses in Äiwoo, the verb and the A argument form a constituent, i.e. a verb phrase. However, a number of commonly used tests for identifying verb phrases are difficult to apply to Äiwoo. Coordination is not a useful test because, as is common in Oceanic languages, any argument NP can be omitted if it is retrievable from context. It is certainly possible to have structures like (6), where two V+A constituents are coordinated and the shared O argument omitted:
(6) I-loteläi-kä isä=naa, mo lâ
pfv-prepare-dir:3 mother.3min=fut conj deic:dist
ku-tu-kä tumwä=nâ.
pfv-bring-dir:3 father.3min=deic:dist
‘Her mother prepared (her food), and her father brought (it) to her.’
But the shared argument is omitted from both clauses, showing that this is not a property of the VP coordination but a general possibility for identifiable arguments.
Other common tests require highly specific data which must normally be elicited for the purpose, and such elicitation has not been carried out for Äiwoo. This concerns, for example, the use of pro-verbs equivalent to English ‘do so’. Äiwoo has a form doo ‘(do) like that’, which can be used as a pro-verb, as in (7), where it refers back to the intransitive verb complex kuluweewaabulagokä ‘jump around’, and takes le ine ‘(just) he’ as its subject:
(7) Lâ ku-luwee-waabulago-kä=to=wâ.
deic:dist ipfv-jump.up-all.around-dir:3=cs=deic:dist
Mo isä-pelivano=nâ ba.
conj mother.3min-child.3min=deic:dist neg
Doo=to=we le i=ne.
like.that=cs=deic:prox deic:prox 3min=deic:prox
‘He started jumping up and down. But not his wife. Only he did this.’
The distribution of doo is fairly flexible, and there is no reason to believe that it could not be used to replace a V+A constituent; however, this cannot be confirmed from the available material.
The criteria that do pick out the VP constituent in Äiwoo concern the distribution of markers of negation and tense/aspect, as well as a quantifier which, when floated, must remain within the VP; these phenomena all independently define a constituent which includes the verb and an A argument, but not any other arguments.
4.3 Negation
Negation in Äiwoo is bipartite, consisting of a preverbal particle and an enclitic. The unit enclosed by these two morphemes includes the verb plus inflectional and derivational morphology, serialised verb roots, and bound morphemes such as manner adverbs, directional markers, and markers of quantificational aspect (‘again’, ‘still’, ‘yet’, ‘always’), but excludes prepositional phrases or clausal adverbs:
- (8)
(a) Bulaape nuwo ba ku-polââ-päko-ngä=gu…
next.day surroundings neg ipfv-clear-good-yet=neg
‘Tomorrow (when) it is not yet fully light …’
(b) Ba i-kää-de=gu kâlâ pelivali-besi-de
neg pfv-know-1 + 2aug=neg there pl-friend.1min-1 + 2aug
kele ngä temotu mi-nenge.
here loc island bn:genr-dem:prox
‘We don't know about our friends in the other islands.’
The morphemes within the negation structure are those which modify the verb directly, specifying manner, direction, repetition, etc. of the verbal action. In an analysis of a very similar structure in Mwotlap, an Oceanic language of the Banks Islands directly southeast of the Reefs-Santa Cruz area, François (Reference François, Bril and Ozanne-Rivierre2004: 112) notes that ‘[s]emantically speaking, [the negation-internal morphemes] have the function of a head-modifier, in very much the same way as adjectives modify the nominal head of an NP’, and, consequently, labels the unit enclosed by the bipartite negation in Mwotlap a verb phrase. The same argument applies to the negation structure in Äiwoo.
In Äiwoo, S and O arguments never occur within the negation structure. Example (8a) shows that a preverbal S (nuwo ‘surroundings’) precedes the negation particle; postverbal S arguments are rare, for reasons to be elaborated on below. The contrast between O and A arguments is illustrated in (9):
- (9)
(a) Ba i-te-kä sime=gu.
neg pfv-see-dir:3 person=neg
‘No one saw (him).’
(b) Sime ba i-te-kâ-no=dä=gu.
person neg ipfv-see-dir:3-1min.A=some=neg
‘I didn't see anyone.’
In example (9a), the noun sime ‘person’, occurring inside the negation structure preceding the negation clitic, can only be interpreted as the subject. In (9b), on the other hand, where it is outside the negation structure, it must be interpreted as the object.
The example in (9b) shows the usual word order for nominal objects, which, as noted in Section 4.1 above, is OVA. Pronominal objects, on the other hand, are postverbal; however, they still occur outside the negation structure, and cannot precede the negation clitic:
(10) Ba i-te-kâ-mu=gu ijii?
neg pfv-see-dir:3-2min.A=neg 3aug
‘Haven't you seen them?’
In other words, the verb phrase as delimited by the negation construction in Äiwoo includes A arguments, but excludes S or O.
4.4 Tense and aspect clitics
Äiwoo has two phasal aspect clitics, =to ‘change of state’ and =jo ’progressive’, as well as a future clitic =Caa which, like many future tense markers crosslinguistically, can also be used to indicate habituality. These clitics pick out the same unit as the negation construction: they attach either directly to the verb complex, or to the verb plus postverbal A argument ((a) examples), while postverbal O arguments consistently follow the clitics ((b) examples):
- (11)
(a) Ki-vääpo-kä tumwä=jo=wâ.
ipfv-ask-dir:3 father.3min=prg=deic:dist
‘Her father asked her.’
(b) Ku-wo-bii=jo=wâ nuwotaa.
ipfv-go-follow=prg=deic:dist head.3min
‘He would follow his own mind.’
- (12)
(a) I-luwa-kä tumwä=naa.
pfv-take-dir:3 father.3min=fut
‘Her father would take it.’
(b) I=nâ ku-pu-kä=naa
3min=deic:dist ipfv-go-dir:3=fut
ki-amoli-kä=naa=kâ talâu na.
ipfv-see-dir:3=fut=deic:dist meal poss:food.3min
‘He would come and find his food there.’
Since Äiwoo, as noted above, often omits overt NP arguments, and since the position of S is overwhelmingly preverbal, it is difficult to find examples where an overt S NP occurs postverbally and where the relevant clitics are present. Those which do occur are typically cases of heavy NPs, either nominalised verbs or NPs with multiple modifiers, which show a tendency to be postverbal. In all such cases, the clitics precede the NP; that is, S patterns like O rather than like A in this regard:
- (13)
(a) Wâ=nâ lâ botou-woli=to=wâ
go=deic:dist deic:dist pour-go.down=cs=deic:dist
nyibengä teuwâ mi-lâdo.
huge rain bn:genr-nasty
‘Then a horrendous rain started pouring down.’
(b) Bââ=kaa de-na-potaa-mu.
not.be=fut bn:thing-irr-search-2min.A
‘You will not lack anything
(lit. there will not be things you look for).’
It is important to note that while S arguments are placed in postverbal position largely if they are heavy, this is not the case for O arguments. As noted above, pronominal Os, which are anything but heavy, are postverbal by default; additional examples are given in (14), showing that postverbal pronominal Os follow the tense/aspect clitics:
- (14)
(a) Ki-do-kâ-mu=waa ijii go doo?
ipfv-pay-dir:3-2min.A=fut 3aug with what
‘What will you pay them with?’
(b) Le i-vetängä-ive-wâ-i-le=to
deic:prox pfv-destroy-appl-dir:2-3aug-ua=cs
iu=nge.
1min=deic:prox
‘They have spoiled me.’
Note also nuwotaa ‘his head’ in (11b), which cannot be characterised as particularly heavy. The clause-final position is contrastive in Äiwoo, and it is possible that this is the intended reading here (he would follow his own mind as opposed to his parents’ insistent advice); as there are no other postverbal elements in this clause, it is impossible to distinguish between the O being immediately postverbal and being in the clause-final, contrastive position. It is clear, however, that such a distinction exists, as can be seen from (15a–b); the O argument in (15a) is contrastive, following the prepositional phrase and occurring as the final element of the clause, whereas in (15b) it is immediately postverbal and has no contrastive reading:
- (15)
(a) Ngaa i-wâ-ilâ-usi-kä Paelat=to
so pfv-caus-go.out-again-dir:3 Pilate=cs
ngâ lokâpu=kâ Barabas.
loc prison=deic:dist Barabbas
‘So Pilate released from prison Barabbas (instead of Jesus).’(Mark15:15)
(b) I=nâ ku-pu-kä=nâ ki-amoli-kä=naa
3min=deic:dist ipfv-go-dir:3=deic:dist ipfv-see-dir:3=fut
talâu na ngä nye-ki-vitelie
meal poss:food.3min loc bn:loc-ipfv-hang
tumwä=nä mo isä.
father:3min=cappl conj mother.3min
‘He would come and find his food in the place where his father and mother had hanged it.’
A arguments can occur in clause-final position – and hence outside the VP – only if they are contrastive; compare (16a) with its non-contrastive counterpart in (16b):
- (16)
(a) Mo i-kää=jo=wâ iumu.
conjpfv-know=prg=deic:dist 2min
‘But it's up to you
(lit. you yourself know; i.e. it is your decision, not mine).’
(b) Mo i-kää-mu=jo=wâ.
conj pfv-know-2min.A=prg=deic:dist
‘But you know.’
A postverbal O, then, is not necessarily contrastive, whereas an A occurring outside the tense/aspect clitics always is.
Like the negation construction, the tense/aspect clitics treat V and A as a unit, which excludes S or O. Since clitics are often described as morphemes which operate at the level of the phrase rather than at the level of the word (e.g. Anderson Reference Anderson2005), this is a further reason for labeling the unit in question a verb phrase.
4.5 The quantifier dä
The VP as defined so far is also the domain of a third construction, namely quantifier float. The quantifier dä ‘some, a’ can either occur in the NP preceding or following the noun it quantifies,Footnote 8 as in (17a), or it can follow the verb, as in (17b), where it quantifies sii milaki ‘small fish’:
- (17)
(a) Dä sime ki-ngâ=to=wâ.
some person ipfv-call=cs=deic:dist
‘Someone is calling.’
(b) Sii mi-laki le ki-e-mo-lämä dä
fish bn:genr-small deic:prox pfv-vpl-stay-inside some
ngâ nupanubulâ mi-laki.
loc pool bn:genr-small
‘A small fish was living in a small pool.’
When floated, dä must remain inside the VP, preceding the negation enclitic =gu and the tense/aspect clitics as well as any VP-external adjuncts:
- (18)
(a) Nupää ba ki-ko-li-mä dä=gu ngämi.
cloth neg ipfv-lie-go.down-dir:1 some=neg obl.pro
‘There is no cloth lying on it.’
(b) Pleti i-bangii-to-epu-kä dä=to ngä corner eä tebol.
plate pfv-smash-go.in-also-dir:3 some=cs loc corner of table
‘She broke another plate on the corner of the table.’
However, dä follows a postverbal A argument, further indicating that such arguments are inside the VP:
(19) Däjelâ le i-wâlu-woli-wâ ibesi-ji
something deic:prox ipfv-put-go.down-dir:2 friend-1 + 2min
dä=ne.
some=deic:prox
‘Our friend here put something down.’
Compare (19) to (20), which has a postverbal pronominal O following dä:
(20) Kä=nä maa sibiliwâlili nä-ekäle-mu dä ijii.
say=cappl if young.woman irr-ask.for-2min.A some 3aug
‘He said, Ask for some young women (to work for you).’
4.6 Summary of verb phrase properties
A number of constructions pick out the verb plus transitive subject argument as a constituent in Äiwoo – the bipartite negation construction, the tense/aspect clitics and the quantifier dä all take the unit consisting of V and A as their domain in a transitive clause. By contrast, none of these constructions treat V and S or V and O as a unit.
It is clear that what is involved is a phrasal unit rather than some form of incorporation construction, as even A arguments which are clearly phrasal – i.e. consist of more than just a bare noun – follow the same pattern. That is, this pattern in Äiwoo is not a case of ‘agent incorporation’, as suggested by Myhill (Reference Myhill1988) for Indonesian; it is simply the default position for A NPs:
- (21)
(a) Lâto i-kää [ilâ sikimapolâ eângâ]=to=wâ…
thus pfv-know deic:dist giant dem:dist=cs=deic:dist
‘Then the giant realised…’
(b) Ngaa Pale lâto i-meli-usi-kä
so Pale thus pfv-release-again.tr-dir:3
[eâumobä lilu eângâ]=to=wâ.
wood.sprite two dem:dist=cs=deic:dist
‘So those two wood-sprites left Pale alone again’
The fact that negation and quantifier float are among the properties defining the VP in Äiwoo may have a broader crosslinguistic correlate. Van Valin & LaPolla (Reference Van Valin and LaPolla1997: 218) propose that ‘the universal basis for the language-specific phrasal category VP is focus structure’, as the focus domain of predicate-focus constructions coincides with the VP in the languages they discuss. They further note that ‘[f]ocus structure is crucially involved in the interpretation of negation and quantification’ (Van Valin & LaPolla Reference Van Valin and LaPolla1997: 219). That is, if VP structure is linked to focus structure, one might expect negation and quantification constructions to make reference to the VP; see Section 7 below.
The above discussion shows, then, that OVA clauses in Äiwoo show an ergative verb phrase structure: the VP includes A, but not S or O. That is, Äiwoo exhibits an ergative pattern at the level of intraclausal syntax. In the next section, I will demonstrate that the language does not show any signs of ergativity at the level of interclausal syntax; in fact, it appears to lack a pivot altogether.
5. Äiwoo as a pivotless language
5.1 Alternatives to pivots
Dixon (Reference Dixon1994: 145) notes that head-marking languages often seem to lack a pivot. Falk (Reference Falk2006) provides a principled explanation for this: Head marking involves what he refers to as incorporated pronouns, i.e. marking of arguments directly on the verb. In such languages, then, each clause in a complex construction has a pronoun which is coreferential with an element in the first conjunct; this is then a case of semantic sharing of arguments rather than a syntactic constraint on coordination. For such constructions, ‘[p]ivothood is completely irrelevant’ (Falk Reference Falk2006: 88).
Given that Äiwoo shows extensive marking of arguments on the verb, then, one might expect it to lack a pivot. There is evidence that, in Äiwoo, the subject and object affixes on the verb function as the syntactic arguments of the clause rather than merely agreeing with the arguments; when an overt argument NP is present, the affixes are usually left out:
- (22)
(a) go ku-mo ngâ nuumä=ke iumu.
because ipfv-stay loc village=deic:prox 2min
‘… because you are the one who stays at home.’
(b) Dee sii=ee ku-wâ-nubo=kâ iu.
this fish=dem:prox ipfv-caus-die=deic:dist 1min
‘I killed this fish myself.’
That is, the default pattern is to have only one representation of the subject in any clause – either as an affix or as a full NP. This is clearly not an agreement pattern, because when there is an overt argument to agree with, the ‘agreement’ disappears. Rather, affixes and independent NP arguments are two alternative ways of coding syntactic arguments in Äiwoo clauses.
With information about the arguments coded directly on the verb, then, the notion of pivot as a single syntactic element shared by multiple clauses may be expected to be irrelevant. In the following I will examine the types of construction to which the notion of pivot are typically taken to apply, and show that there is no evidence of a pivot in these constructions in Äiwoo.
5.2 Constructions involving pivots
The following discussion will take Falk's (Reference Falk2006) LFG-based analysis as its starting point, as this account is explicitly designed to accommodate the possibility of an ergatively structured verb phrase. Accordingly, in order to test for the presence of a pivot in Äiwoo, it must be determined which ‘subject’ properties are characteristic of pivots as opposed to 
. As noted above, Schachter (Reference Schachter, Cole and Sadock1977) designates relativisation, the ability to launch floating quantifiers, and the inability to occur in existential constructions as reference-related properties, while being addressee of imperatives, antecedent of reflexives, and controllee in control constructions are seen as role-related properties.
Falk's (Reference Falk2006) conception of pivot differs from Schachter's account in that he sees referential prominence not as a defining characteristic of pivots but as a secondary property deriving from their prominent status in the clause structure. Furthermore, he notes that control, considered a role-related property by Schachter, actually does not pattern uniformly across languages: in some languages the controllee in control constructions is the , in others it is the pivot.
According to Falk, this is in fact the only ‘subject’ property which does not show a consistent distribution crosslinguistically. Within the LFG framework, this falls out of the distinction between two types of control: anaphoric control and functional control (Bresnan Reference Bresnan1982). Functional control is a relationship between the functional structures (f-structures) of the main clause and the controlled clause; in other words, it involves the sharing of a single argument across clauses, and is therefore a pivot property. By contrast, anaphoric control involves a null pronoun in the controlled clause which is coreferent with an argument in the main clause; anaphoric coreference does not involve pivots, but tends instead to be restricted by the relational hierarchy, of which the most prominent member is .
In Falk's framework, then, essential pivot properties are those which relate to the sharing of a single element by more than one clause, as the pivot is defined as ‘the element with the function of connecting its clause to other clauses in the sentence’ (Falk Reference Falk2006: 74). These include being the shared argument in coordinated clauses, the ability to participate in raising and extraction, and – in some languages – being the controlled argument in control constructions. Other properties typically taken to be characteristic of pivots are seen as secondary, because they are not related to the cross-clausal function of pivots. Instead, Falk suggests, they stem from the notion of pivot as being a distinguished element of the clause; they include obligatoriness, definiteness or wide scope, and external structural position.
In what follows, I will show that none of the constructions taken by Falk to exhibit essential pivot properties show any evidence of a pivot in Äiwoo. In addition to the properties discussed by Falk, I will also discuss the ability to launch floating quantifiers, which is included in Schachter's list of reference-related properties and argued by Wechsler & Arka (Reference Wechsler and Arka1998) to provide evidence for S/O being the ‘subject’ (pivot) in Balinese undergoer-voice clauses. As may be seen from example (2), discussed in Section 3 above, such clauses are structurally parallel to OVA clauses in Äiwoo in that both show an ergative verb phrase structure; however, as I will show below, there is no evidence of an S/O pivot in Äiwoo. Ross (Reference Ross2004: 533) notes that ‘canonical Oceanic languages’ typically lack a pivot, and links this to the appearance of subject marking on the verb; so Äiwoo is in fact a typical Oceanic language in this respect.
5.3 Coordination
Coordination in Äiwoo may involve the sharing of any two arguments across the coordinated clauses. Some examples are given in (23):
- (23)
(a) S–O: Na-pu-ute-mä=kaa mo wâ-nubo-wâ-i-de.
irr-come-again-dir:1=fut conj caus-die-tr-suff-1 + 2aug.A
‘If he comes back, we will kill him
(lit. he comes back and we kill him).’
(b) O–S: I-to-guli-kä=nâ, mo lâ
pfv-strike-poke-dir:3=deic:dist conj deic:dist
i-luwee-kä=to=wâ.
pfv-jump.up-dir:3=cs=deic:dist
‘She poked at it, and it jumped up.’
(c) S–A: Usaliki lâ i-pu-kä=to=wâ mo kä=nä …
Usaliki deic:dist pfv-go-dir:3=cs=deic:dist conj say=cappl
‘Usaliki went to them and said …’
(d) A–S: Mo pesikimâpolâ eângâ ilâ sii eângâ
conj giant dem:dist deic:dist fish dem:dist
i-ngä i-du=kâ, mo lâ ku-bakisi=kâ.
pfv-eat pfv-finish=deic:dist conj deic:dist ipfv-run=deic:dist
‘And when the giant had finished eating the fish, he ran on
(lit. and the giant ate the fish finished, and he ran on).’
Note that some of these examples translate most idiomatically into English as subordinate rather than coordinate constructions, but this is an artefact of the translation; all examples involve the same conjunction mo and the same syntactic structure.
5.4 Control
As with coordination, constructions with typical ‘control verbs’, such as kA Footnote 9 ‘say, want’, involve anaphoric coreference rather than syntactic control:
- (24)
(a) Kâ-no=ngä nyä-eamole ute.
say-1min.A=cappl 1min.S:irr-see again
‘I want to see again.’
(b) Kâ-no=ngä nä-lowâle-mä-mu iu.
say-1min.A=cappl irr-help-dir:1-2min.A 1min
‘I want you to help me.’
In (24a), the 1min subject marking on both verbs indicates that the A argument of kA ‘say, want’ is the same as the S argument of eamole ‘see’. In (24b) the subjects of the two verbs are different, and each verb is marked for its subject, kA for 1min and lowâle ‘help’ for 2min.
The fact that such constructions involve anaphoric coreference is shown clearly by examples like (25), where the controller is included in, but not identical with, the controlled argument.
(25) Kâ-no=ngä ji-nâ-motävilooli=ta ji-nä-täve.
say-1min.A=cappl 1 + 2min-irr-swing.down=just 1 + 2min-irr-hang
‘I want us to swing down and hang (from the branch).’
Such patterns are a characteristic of anaphoric constructions, because pronominal coreference may involve inclusion (e.g. the reference of I is included in the reference of we) as well as identity. By contrast, such patterns are not expected to occur in a pivot-based construction where two clauses share a single syntactic element, as such a shared element would necessarily have to be identical in both clauses (Falk Reference Falk2006: 90–91, 142).
5.5 Raising
Falk (Reference Falk2006: 155) defines raising as ‘a control construction in which the controller is not a thematic argument of its verb’. In such a construction, ‘[a]n anaphoric analysis is not possible, because that would leave the controller without a thematic role’. In other words, if Äiwoo is a pivotless language, this would appear to imply that it should lack raising constructions.
There are some constructions in Äiwoo which share central properties with raising constructions; notably, the fact that the clause-initial noun phrase is not assigned a thematic role by the following verb. The first involves the verb umo ‘to be difficult’, a stative verb which may take a subject noun phrase denoting the thing that is difficult, as in (26a), or a complement clause introduced by go, in which case the construction translates into English as ‘it is difficult to X’ (26b). The latter is by far the most frequent construction in my data, while (26a) is elicited:
- (26)
(a) Maths enge umo.
maths dem:prox difficult
‘Maths is difficult.’
(b) Umo go nâ-lâgovapä-kä go sime.
difficult conj irr-hide-dir:3 prep person
‘It was difficult to hide it from people.’
In the latter construction, an argument of the go-marked clause may be fronted to the position preceding umo:
- (27)
(a) Ee, mo temaale=kâ umo go nâ-wâ-nubo-wâ.
yes conj longtom=deic:dist difficult conj irr-caus-die-tr
‘Yes, but longtom was hard for him to kill.’
(b) Äiwoo enge umo go nä-kää-eke-nyi-no.
Äiwoo dem:prox difficult conj irr-know-quick-tr-1min.A
‘Äiwoo is hard for me to understand quickly.’
However, the frequent appearance of a demonstrative or deictic clitic at the end of the clause-initial noun phrase in this construction suggests that this is a case of topicalisation rather than raising. What has sometimes been called the ‘double subject’ construction (especially in descriptions of Japanese, e.g. Kuno Reference Kuno1973), where a clause has both a grammatical subject and a clause-initial topic distinct from the subject, is very common in Äiwoo:
- (28)
(a) Go iu=nge nyisi biou mana.
because 1min=deic:prox body.1min heavy very
‘Because my body is very heavy
(lit. because me here, my body is very heavy).’
(b) Iu=nge dee ku-wââ-gu=ngä tumo.
1min=deic:prox this ipfv-forbid-3min.A=cappl father.1min
‘My father has forbidden it to me
(lit. me here, my father has forbidden this).’
Note that the initial topicalised NP bears no grammatical relation to the verb of the clause, and accordingly is not assigned a thematic role by the verb either, analogous to the definition of raising offered by Falk. The two constructions are thus superficially similar, but syntactically different: since there is no requirement in Äiwoo that a topicalised NP bear any syntactic relation to the verb of the clause in which it is a topic, no pivot condition applies to such constructions.
A similar analysis is applicable to the other verb in my data which shows syntactic patterns reminiscent of raising, namely käsä ‘resemble, (be) like, seem’, as in (29):
(29) Sime enge käsä=nä opulo.
person dem:prox like=cappl red
‘That person seems to be red.’
Käsä has a highly complex distribution and may function both as an independent predicate, as the second predicate in a serial verb construction, as a preposition, or as a subjunction. Most importantly for the present purposes, when used as a predicate, it has both personal uses (‘X resembles Y’) and impersonal uses, that is, lacking a referential subject (‘it looks like/seems that X’).
- (30)
(a) Mo nulä i-ea mana ngâgo-i, go käsä sipsip
conj neck.3min pfv-bad very to-3aug conj like sheep
mi-bââ me-ki-vevaale ngâgo-i.
bn:genr-not.be bn:person-ipfv-wait to-3aug
‘He had compassion on them, because they were like (resembled) sheep without a shepherd.’(Mark 6:33)
(b) I-mo=kâ, mo käsä=nä lâ ki-to-vesi
pfv-stay=deic:dist conj like=cappl deic:dist ipfv-exist-cont
ngä nulä=nâ.
loc neck=deic:dist
‘He stayed at home, but he kept thinking about it
(lit. it was like it stayed in his mind/neck).’
The most plausible analysis of (29) is that it shows the impersonal use of käsä with the subject of opulo ‘be red’ topicalised, i.e. ‘that person, it seems she is red’. The presence of the clitic =nä on käsä is characteristic of the impersonal as opposed to the personal predicate; in my data, only one of 17 clear examples of the use of käsä as an impersonal predicate lacks =nä, whereas only one of 16 uses as a personal predicate shows it without any other triggering factors being present. This pattern is likely to be a result of the origins of the impersonal predicate in the use of käsä as the second verb in ambient serial verb constructions (Crowley Reference Crowley2002: 41–42, 75–76). Note also the demonstrative in the initial NP, which as noted above is typical of topicalised phrases.
5.6 Extraction
Äiwoo is a wh-in situ language: question words are not fronted, but occur in the same syntactic position as the questioned element would in a declarative clause. Note that, since the default position of both S and O is preverbal, clause-initial question words are common, but need not involve extraction; the clause-initial position of doo ‘what’ in (31a) is simply the default position for the object of a transitive verb. Compare (31d), where the questioned argument is A, and ie ‘who’ occurs in the normal position of A arguments: following the verb, but preceding the future clitic.
- (31)
(a) Doo=lâ kiki-mu=wâ?
what=deic:dist cry.for-2min.A=deic:dist
‘What are you crying for?’
(b) Imu=wâ mi-ku-wo doo?
2min=deic:dist 2min.S-ipfv-go what
‘Why have you come?’
(c) Sime mi-doo lâ i-amoli-kâ-no=ngâ?
person bn:genr-what deic:dist pfv-see-dir:3-1min.A=deic:dist
‘What kind of person was that I saw?
(d) Sii le ki-täveli-dä=ne mo
fish deic:prox ipfv-hang-some=deic:prox conj
i-tu-mä ie=kaa?
pfv-bring-dir:1 who=fut
‘There is fish hanging here, but who brought it?’
There is no relative marker in Äiwoo. Relative clauses are either unmarked or formed by nominalisation with the prefix mi-. In the latter case, the construction as a whole is equivalent to an apposition of two nouns (Næss 2006: 277–278). Relative clauses in Äiwoo may be formed on any argument, thus showing no evidence of a pivot:
- (32)
(a) S: Ku-mo=kâ Nookâ, Nede, ngä naa bäli
ipfv-live=deic:dist Nookâ Santa.Cruz loc end side
ki-päie-kä ngä nyibä näle.
ipfv-face-dir:3 loc eye sun
‘He lived at Nookâ, on Santa Cruz, on the side which faces the sunrise.’
(b) A: Ku-wä ki-ea nulo ilâ sime
ipfv-go ipfv-bad neck.1min deic:dist person
ki-singä-ive-gu=ngâ.
ipfv-lie-appl-3min.A=deic:dist
‘I hate the man who told lies about me’
(c) O: Ngaa i-lotolâ-kä=nâ de-na
so pfv-prepare-dir.3=deic:dist bn:thing-poss:food.3min
ki-pi-kä tumwä=jo.
ipfv-bring-dir.3 father.3min=prg
‘So she prepared some food which her father had brought.’
5.7 Floating quantifiers
As noted in Section 4.5 above, Äiwoo has a quantifier dä, which can occur either as a modifier in the NP or following the verb within the VP. The data shows no restrictions on the type of argument that can occur with a floated dä, as the examples in (33) show. In (33a) dä refers to the S argument, in (33b) to the A and in (33c) to the O.
- (33)
(a) S: Sii mi-laki le ki-e-mo-lämä dä
fish bn:genr-small deic:prox ipfv-vpl-stay-inside some
ngä nupanubulâ mi-laki.
loc pool bn:genr-small
‘A small fish was living in a small pool.’
(b) A: Ki-li-mo-le=nâ ilâkâ
ipfv-3aug.S-live-ua=deic:dist there
ba i-kää-i dä=gu ngä nuumä=kâ.
neg pfv-know-3aug.A some=neg loc village=deic:dist
‘No one in the village knew that they were living there.’
(c) O: Lâto ki-li-amole-ngengäle=nâ, kä-i-le=nä
thus ipfv-3aug-look-around=deic:dist say-3aug.A-ua=cappl
ngamaa nyie nä-te-kä-i-le dä.
if fire irr-see-dir:3-3aug.A-ua some
‘So they looked around, hoping to see a fire.’
A second quantifying morpheme shows a somewhat different distribution, but, like dä, can occur either as a modifier in the NP or within the verb phrase. Du ‘finish’ is basically a verb which, when used to quantify a noun, may be translated with ‘all’. It may either function as a modifier in the noun phrase (34a) or be serialised to the main verb of the clause (34b):
- (34)
(a) nyenaa-du ngä paveli eângâ
tree-finish loc garden dem:dist
‘all the trees in the garden’
(b) Ngaa tepekoulâ ku-luwa-du-kä=jo.
conj things ipfv-take-finish-dir:3=prg
‘She did all the chores.’
Du differs from dä in its position when ‘floated’; while dä follows the verb within the VP, du occurs inside the verb complex, serialised to the main verb. However, like dä, du shows no restrictions on the type of argument it can quantify when occurring in the verb complex:
- (35)
(a) S: Sime lu-po-du=to ngââgu.
person 3aug.S-go-finish=cs bush
‘All the people had gone to the bush.’
(b) A: I-amoli-du-mä-i iu.
pfv-see-finish-dir:1-3aug.A 1min
‘They all saw me.’
(c) O: I-wâ-nubo-wâ-du-no ijii.
pfv-caus-die-tr-finish-1min.A 3aug
‘I killed them all.’
To summarise, none of the constructions taken to involve pivots crosslinguistically show any evidence of a pivot in Äiwoo: there is no raising, no wh-fronting, relativisation and quantifier float show no restrictions on the type of core argument they may involve, and coordination and control make use of anaphoric coreference rather than pivot restrictions. It seems fair to conclude, then, that Äiwoo is in fact a pivotless language.
Indeed, with the exception of the properties taken to define the VP in Section 4, no processes, interclausal or intraclausal, appear to single out S and O as a privileged category in Äiwoo. That is, Äiwoo exhibits yet another type of syntactic ergativity from that discussed by Levinson (in press) in the Papuan language Yélî Dnye, where intraclausal processes such as focus and quantifier floating follow an ergative pattern, but there is no evidence of a verb phrase. Like Äiwoo, Yélî Dnye lacks a pivot; but unlike Äiwoo, it shows ergative case marking. The Äiwoo data, then, serves to reinforce Levinson's point that syntactic ergativity is a more diverse phenomenon than has hitherto been recognised, and adds another element to his expanded typology of ergativity: a language type with a VA verb phrase, but no S/O pivot.
6. The origins of the ergative verb phrase
The presence of an ergative verb phrase structure – i.e. a structure where the A is internal to the verb phrase while S and O are external to it – in a pivotless language seems highly unusual. As noted in Section 5.2 above, Falk (Reference Falk2006: 74, 76, 97) considers ‘external structural position’ to be a secondary, though not a necessary property of pivots, derived from its status as a ‘sentence-internal topic’. In Falk's analysis, a pivot is
intermediate in its scope between the local argument and adjunct functions on the one hand and the discourse-related functions on the other. The structural position of piv reflects this intermediate status … The structural position typically associated with piv, [SPEC, IP], is closer to the lexical head than the place of discourse functions but farther than most adjuncts. The general picture that emerges is that configurational languages represent grammatical functions iconically in the c[onstituent]-structure, and the external position of the piv is part of this iconicity. (Falk Reference Falk2006: 97)
However, this explanation does not readily apply to Äiwoo, which lacks the grammatical function – pivot – that this structure supposedly iconically represents. Rather, the explanation for the ergative VP structure in Äiwoo is historical. The alternation between the OVA and the AVO clause patterns in Äiwoo derives fairly transparently from an original actor-voice/undergoer-voice system like that described in Section 3 above, and some of the morphological differences between the verbs in such clauses can be brought back to what was originally voice morphology. In particular, the final vowel -i in many OVA verbs, which alternates with -e or Ø in the corresponding AVO verb, is transparently cognate with the common Oceanic transitive suffix -i, which in turn reflects an earlier undergoer-voice suffix *-i (Lynch, Ross & Crowley Reference Lynch, Ross and Crowley2002, Pawley & Reid Reference Pawley and Reid2011 [1979], Ross Reference Ross2012). The distinction between subject prefixes for intransitive verbs and subject suffixes for transitive verbs, highly unusual for an Oceanic language, can be taken to originate in actor pronouns which have become bound to the stem, reflecting the position of the actor argument in actor-voice and undergoer-voice clauses, respectively. The likely origin of Äiwoo person affixes in earlier independent pronouns is noted by Ross & Næss (2007: 479), who also note that ‘[t]here has always been a tendency [in Oceanic languages] to innovate new subject prefixes from reduced forms of the free pronouns’.
This analysis explains both the presence of an ergative verb phrase in the OVA clause and the absence of a pivot. The fact that V and A form a constituent in the original undergoer-voice construction parallels the situation in languages like Balinese discussed in Section 3 above. The development of head marking through the accretion of subject/actor pronouns to the verb led to the loss of the pivot through the increased use of anaphoric coreference in multiclausal constructions. The properties discussed in Section 4 as picking out a VP constituent consisting of V and A in OVA clauses do not similarly treat V and O in AVO clauses as a constituent: the O argument is here outside the negation structure and the tense/aspect clitics, and follows a floated dä:
- (36)
(a) Ile ngä nye-polââ enge
deic:prox loc bn:loc-clear dem:prox
sime ba ku-wokene dä=gu nupwää
person neg pfv-wash some=neg cloth
ilâ ku-wopa doo=wâ.
deic:dist ipfv-white like.that=deic:dist
‘No one in this world could wash a cloth that white.’(Mark 9:3)
(b) Me-ku-nu=to nenu=kâ.
1aug.S-ipfv-drink=cs coconut=deic:dist
‘We drank coconuts.’
(c) Gi-laki ki-lää-kä dä nupwää
bn:male-small ipfv-give-dir:3 some cloth
ngâgo si-laki dä.
to.3min bn:female-small some
‘A boy gave a cloth to a girl.’
(The postverbal dä quantifies the initial noun gilaki ‘boy’; the speaker offered the alternative structure gilaki dä kiläakä nupwää ngâgo silaki dä.)
This, too, is a reflex of the historical situation: Proto Malayo-Polynesian (PMP) actor-voice clauses were formally intransitive, with an oblique O argument. In other words, the O in the actor voice was less tightly bound to the verb than the A in the undergoer voice, and it is this situation that is reflected in the clause patterns of present-day Äiwoo. The fully symmetrical clause patterns of ‘Indonesian-type’ languages like Balinese is clearly a separate development (Ross Reference Ross, Wouk and Ross2002, Reference Ross2012; Næss Reference Næss2013).Footnote 10
7. Theoretical implications
If a language can have an ergatively structured verb phrase – i.e. one where the A argument is internal to the VP while the S and O are external to it – without this being matched by an S/O pivot, this has implications for approaches to syntax which define grammatical relations in terms of structural configurations. Most formal models of syntax accord the structural position external to the verb's domain – whatever the exact label given to that domain in the theoretical framework in question – a unique status in the overall clause structure of a language.
For some approaches, accommodating an ergative verb phrase is inherently difficult. In Minimalist theory, the external argument is the argument of a separate predicate in the syntactic tree, and this position is associated with the semantic role of Agent (Harley Reference Harley and Boeckx2011). This is clearly incompatible with S/O arguments which do not necessarily – in the case of O, cannot – show an Agent semantic role.
In a recent paper on syntactic ergativity, Polinsky (Reference Polinsky2013) proposes a distinction between ‘PP-ergative’ languages, where the ergative argument is a PP, and ‘DP-ergative’ languages, where it is a DP, and suggests that syntactic ergativity is ‘a side-effect of the PP-ergative type’: it arises when ‘a. the ergative is a prepositional phrase, and b. extraction is limited to DPs’ (Polinsky Reference Polinsky2013: 4). However, here, too, syntactic ergativity is equated with an S/O pivot, defined as meaning that ‘the ergative DP cannot be extracted under relativization, wh-question formation, or focusing’ (Polinsky Reference Polinsky2013: 1). Both PP and DP ergative arguments are generated in [Spec,vP] under this analysis, and so there does not appear to be any way of accommodating an ergative verb phrase structure. Äiwoo, being a DP-ergative language, may not directly contradict the prediction that only PP-ergative languages show syntactic ergativity on the definition of syntactic ergativity as meaning inter-clausal ergativity; but it does present a different type of syntactic ergativity which is not accommodated by this analysis.
Approaches which do seek to incorporate the possibility of an S/O external argument generally do so by associating it with the function of pivot. In the Government-and-Binding–based analysis of Bittner & Hale (Reference Bittner and Hale1996), syntactic ergativity involves raising of S and O to [Spec,IP]. This position is then associated with syntactic properties essentially equivalent to what in this paper has been called pivot properties (Ā-dependencies in the terminology of Bittner & Hale). Similarly, as discussed in Section 6 above, the LFG-based analysis of Falk (Reference Falk2006) takes VP-external position to be associated with pivothood. Indeed, the S/O pivot structure is here assumed to be the explanation for the ergative verb phrase, in that the verb phrase structure is taken to iconically reflect the syntactic status of the pivot as being ‘intermediate in its scope between the local argument and adjunct functions on the one hand and the discourse-related functions on the other’ (Falk Reference Falk2006: 97).
The Äiwoo data shows that it is possible for a language to have an ergatively structured verb phrase with no matching ergative alignment in other areas of syntax. The verb phrase structure in Äiwoo OVA clauses is a historical accident, a remnant of a previous actor-voice/undergoer-voice system; and while it was presumably at an earlier stage matched by an S/O pivot, this is not the case in the present-day language, which lacks an identifiable pivot.
This situation is obviously problematic for any approach that seeks to assign universal linguistic functions to aspects of constituent structure. Falk argues for a competing-motivations explanation for the crosslinguistic variation between ergative and accusative verb phrase structures:
Constituent structure is the overt expression of syntactic elements, and thus can be expected to reflect information about other dimensions. This leads to a problem, since different dimensions have different kinds of information and relations between elements, and it is impossible for constituent structure to express all of them simultaneously. The existence of different constituent structure patterns in different languages … is the consequence of differences in which dimensions of linguistic information are reflected in constituent structure. (Falk Reference Falk2006: 97)
We have seen, however, that it is equally feasible for differences in basic clause structure to be a consequence of purely historical processes, and in fact have no obvious information value at all.
There is nothing new in the claim that explanations for synchronic structural properties must be sought in diachronic developments – this assumption is the basis of the strand of linguistic research known as grammaticalisation theory. However, such diachronic explanations are usually formulated in terms of some form of functional optimisation, e.g. the tendency for frequent forms to undergo phonological reduction (see Bybee & Hopper Reference Bybee, Hopper, Bybee and Hopper2001 and references therein), the notion of competing motivations as providing explanations for alternative ‘solutions’ to the same ‘grammar problem’ but ruling out certain possibilities as not satisfying either of the motivations in competition (see e.g. Du Bois Reference Du Bois and Haiman1985, Croft Reference Croft2003), etc. The ergative verb phrase structure in Äiwoo, on the other hand, is simply a by-product of other diachronic processes, as outlined above. That is, the phenomenon of syntactic ergativity, which has long been perceived as reflecting a fundamentally different basis for grammatical organisation from that found in accusative languages (e.g. DuBois Reference Du Bois1987, Dixon Reference Dixon1994, Manning Reference Manning1996), neither necessarily reflects essential differences in the principles that determine grammatical organisation nor necessarily implies wide-ranging structural differences between languages with ergative intraclausal syntax and languages with accusative intraclausal syntax.
The structural facts of Äiwoo strongly suggest that assigning universal functions to syntactic configurations is untenable. Other authors have suggested that the VP-external position may not be universally associated with ‘subjects’ or pivots. For example, Diesing (Reference Diesing1990) argues that [Spec,IP] in Yiddish functions as a topic position, while Kroeger (Reference Kroeger1993) claims that this position in Tagalog is associated with the pragmatic functions of topic and focus. This echoes the proposal, cited in Section 4.6 above, that VP structure is motivated by focus structure (Van Valin & LaPolla Reference Van Valin and LaPolla1997: 218; Van Valin Reference Van Valin2005: 80–81). Synchronically, in Äiwoo, there is no evidence that the VP-external position is a ‘topic’ position. For example, the O argument in an OVA clause can be questioned, as in (31a), and new participants can be introduced in this position, as in (37), where the A is the topic and O+V constitute the focus domain:
(37) Lu-po-kä-le, nyâ-nou nyigi i-te-kä-i-le
3aug.S-go-dir:3-ua bn:tree-banana one pfv-see-dir:3-3aug.A-ua
ki-ko.
ipfv-lie
‘They went along, and they saw a banana tree lying there.’
That is, Äiwoo VP structure cannot synchronically be explained either in terms of the syntactic property of pivothood or of the pragmatic property of focus structure. Diachronically, on the other hand, the origins of the ergative VP in a symmetrical voice system speaks in favour of a link to focus structure, given that the privileged nominal in a symmetrical voice system is essentially a topic (Schachter Reference Schachter and Li1976, Reference Schachter, Cole and Sadock1977). This could further explain why the VP in Äiwoo is defined partly by grammatical features such as negation and quantification, which are closely linked to focus structure (Van Valin & LaPolla (Reference Van Valin and LaPolla1997: 291). However, in present-day Äiwoo, the position external to the verb phrase is not associated with any particular syntactic or pragmatic function, except in the trivial sense that it is the position of S and O arguments.
8. Conclusion
This paper has shown that, in languages which have a verb phrase, there is no universal association between VP-external position and a privileged status in syntactic structure. This constitutes a problem for models of syntax which define grammatical relations in terms of structural configurations, as these generally take VP-external status to be associated either with agentivity or with some version of the syntactic concept which has here been called pivot.
Nonconfigurational models which associate VP structure with focus structure fare somewhat better, in that focus structure is a plausible basis for the system from which the Äiwoo clause patterns historically derive. In terms of synchronic grammatical systems, on the other hand, the paper has shown that what has traditionally been seen as fundamental differences in grammatical organisation – the difference between an accusative and an ergative pattern of VP structure – need not in fact be associated with any broader differences in syntactic or pragmatic organisation.
What the VP structure of Äiwoo demonstrates is, rather, that any linguistic system at any point in time is the outcome of a number of interacting factors which do not add up to either a universal structural template or a set of functionally-based competing motivations. This is essentially the point made by Evans & Levinson (Reference Evans and Levinson2009: 444), who state that ‘most linguistic diversity is the product of historical cultural evolution operating on relatively independent traits’. Rather than language universals, they argue, there are
evolutionarily stable strategies, local minima as it were, that are recurrent solutions across time and space … they result from myriad interactions between communicative, cognitive, and processing constraints which reshape existing structures through use. (Evans & Levinson Reference Evans and Levinson2009: 444)
In other words, languages tend to converge on similar structural patterns – evidenced in Äiwoo through properties such as the existence of a verb phrase and the correlation between head marking and pivotlessness – but there is more variability inherent in these patterns than rigidly universalist models allow for.
- =
clitic boundary
- 1, 2, 3
first, second, third person
- 1 + 2
1st + 2nd person
- A
agentive argument of two-argument verb
- appl
applicative
- aug
augmented number
- av
actor voice
- bn
bound noun
- cappl
clausal applicative
- caus
causative
- conj
conjunction
- cont
continuous
- cs
change of state
- cv
circumstantial voice
- def
definite
- deic
deictic (particle/clitic)
- dem
demonstrative
- dir
directional
- dist
distal
- dur
durative
- fut
future
- gen
genitive
- genr
general
- ipfv
imperfective
- irr
irrealis
- loc
locative
- lv
location voice
- min
minimal number
- neg
negation
- npiv
non-pivot
- O
second argument of two-argument verb
- obl.pro
oblique pro-form
- ov
object voice
- pfv
perfective
- pl
plural
- poss
possessive
- prep
preposition
- prg
progressive
- prox
proximal
- pv
patient voice
- S
single argument of one-argument verb
- spec
specific
- suff
suffix (precise function unclear)
- tr
transitive suffix
- ua
unit-augmented number
- vpl
verbal plural
