1 Introduction

This paper focuses on subject extraction in Māori, the indigenous Polynesian language of New Zealand. Māori is generally described as having accusative alignment with various traces of ergativity (Bauer 1993; 1997; though see Sinclair 1976; Pucilowski 2006 for (split-)ergative treatments), and only subjects are generally accessible for the purposes of relativisation, topicalisation, focus and questioning (Keenan & Comrie 1977; Bauer 1993; 1997). The combination of accusative alignment and subject-only accessibility is important since many of the languages identified by Keenan & Comrie as having subject-only accessibility exhibit ergative alignment, suggesting that these languages may really have absolutive-only accessibility. Māori thus shows that subject-only accessibility is a genuine phenomenon. Nonetheless, as in many other languages, Māori subjects are not equally accessible in all contexts. More specifically, subject questioning/focus is more restricted than subject topicalisation. This paper is primarily concerned with the nature of this restriction.¹

Māori has basic VSO order in verbal constructions and Predicate-Subject order in non-verbal constructions, i.e. it is predicate-initial. Non-verbal constructions can be subdivided into two types based on the category of the predicate phrase (prepositional or nominal) following standard practice in the description of Māori (Reedy 1979; Bauer 1993; 1997; de Lacy 1999) and related Polynesian languages (see e.g. Seiter 1980 on Niuean; Mosel & Hovdhaugen 1992 on Samoan; Otsuka 2005 on Tongan). I argue that, whilst subject topicalisation is generally permitted in all types of construction, subject questioning/focus is permitted in verbal and prepositional predicate constructions, but prohibited in nominal predicate constructions (see also de Lacy 1999).

In this paper I argue that de Lacy’s (1999) descriptive generalisation is essentially correct and can successfully describe a wider range of Māori constructions. I combine this generalisation with Bauer’s (1991; 1993; 1997) analysis of Māori focus/question constructions as clefts in which the questioned constituent is the matrix predicate phrase and the matrix subject phrase is a headless relative clause. I propose an intervention account based on featural Relativised Minimality (Starke 2001; Rizzi 2013) whereby the C head of this headless relative clause probes for a feature shared by both nominal arguments and nominal predicates, which I call [D]. Consequently, a nominal predicate phrase will block the creation of a headless subject relative clause by intervention, whilst a verbal or prepositional predicate phrase will not. The schematic structure of the headless relative clause CP is illustrated below:

This analysis also accounts for why only subjects are accessible for focus/questioning: if the C head of the headless relative clause probes for a [D] feature, the subject phrase will act as an intervener for the relativisation of any lower nominal (see also Aldridge 2004; 2008a; b).

As for topicalisation, I propose that the Topic head probes for a feature found only on nominal arguments, which I call [X]. Consequently, no predicate phrase acts as an intervener for subject topicalisation. However, the subject phrase will still act as an intervener for the topicalisation of any lower nominal argument, thus accounting for why only subjects are accessible for topicalisation.

I am thus proposing that C and Topic in Māori probe for [D] and [X] features respectively, i.e. features associated with nominal syntax, rather than [REL] and [TOP] features more familiar from analyses of European languages. The choice of feature in triggering Aʹ-movement defines in large part the extraction profile of a given language/construction, i.e. the set of elements that is accessible to Aʹ-movement and that intervenes with Aʹ-movement. I discuss this in greater detail in Section 7.

The structure of the paper is as follows: in Section 2, I lay out my assumptions concerning Māori clause structure in neutral declaratives. In Section 3, I describe verbal predicate constructions and introduce the strategies used to question subjects and predicates and to topicalise subjects. In Section 4, I describe various non-verbal predicate constructions in Māori, considering whether it is possible to question the subject, question the predicate phrase, and topicalise the subject. I also consider the syntactic category of the predicate phrase, i.e. whether it is prepositional or nominal. In Section 5, I discuss how the Actor Emphatic construction, which has received a lot of attention in the Māori and Polynesian literature, fits into the proposed generalisations. In Section 6, I propose that intervention plays a central role in the proposed generalisations, and discuss and develop a formal analysis. In Section 7, I discuss the conceptual underpinnings of the formal analysis, arguing that using nominal features to trigger Aʹ-movement in Māori is consistent with an emergentist approach to formal features (Biberauer 2011; 2017; Wiltschko 2014; Biberauer & Roberts 2015a; b; 2017). Finally, Section 8 concludes.

2 Māori clause structure

Māori has basic VSO order in verbal constructions and basic Predicate-Subject order in non-verbal constructions. Following much work on the syntax of Austronesian and Polynesian languages (see especially Massam 2000; Aldridge 2004; Collins 2017), I assume that the predicate phrase undergoes predicate fronting² to a position higher than the external argument, regardless of whether the predicate phrase is verbal, prepositional or nominal (I will return to the derivation of VSO order shortly).³ This is schematised in (4) (the heads T, F and R are explained below).

Tense-Aspect-Mood markers are generally merged in T. They potentially raise to a C-domain position (see Massam 2010 on Niuean; Collins 2017 on Samoan), but as this is unimportant for the present paper, I ignore it here. R stands for Relator, i.e. whichever head mediates the predication relation between the subject (in its specifier) and the predicate (in its complement) (den Dikken 2006; see also Bowers 1993). The predicate, which I have labelled XP in (4), can be one of three categories: VP, PP or DP, yielding verbal, prepositional or nominal predicate constructions respectively. For example, in verbal predicate constructions, the predicate would be a VP and R would be a v head (I remain agnostic concerning the exact category of R in non-verbal predicate constructions). Following Collins (2017), F is a functional head between the subject and T, and bears a [PRED] feature which triggers predicate fronting to SpecFP (see Massam 2000; 2010; Aldridge 2004; 2006 for different implementations of the same basic idea). Assuming that the DP subject remains low in SpecRP, this straightforwardly captures the basic predicate-initial order of all Māori constructions.⁴

Further following the work of Massam (2000), Aldridge (2004) and Collins (2017), I assume that Māori’s basic VSO order is derived by extracting the object from the VP to a position below the subject prior to (remnant) VP predicate fronting. Assuming that subjects do not move, this position would be a SpecvP or adjoined position lower than that occupied by the subject (see Collins 2017).

Something analogous may also be observed with complex non-verbal predicate constructions, i.e. the complement of the head of the non-verbal predicate may appear following the subject. Such splitting is not generally obligatory though it is preferred in many cases (see Bauer 1997: 31, 33, 63–64 for examples and discussion). I illustrate here using a locational construction (see Section 4.2 below) – the subject phrase is bracketed and the (discontinuous) predicate phrase is in bold.⁵

I assume that the complement of the head of the predicate phrase moves out of the predicate phrase to a position below the subject prior to predicate fronting, analogous to the derivation of VSO. This is schematically illustrated for (6) in (7).

3 Verbal predicate constructions

In this section, I describe the verbal predicate construction and introduce the strategies for subject questioning, predicate questioning and subject topicalisation.

4 Non-verbal predicate constructions

Māori has a number of non-verbal predicate constructions, which I introduce in their own subsections. For each construction I consider the following questions: (i) Can the subject be questioned? (ii) Can the predicate phrase be questioned? (iii) Can the subject be topicalised?¹⁰ I will also consider whether the category of the predicate phrase is nominal or prepositional. As above, the predicate phrase will be in bold, the subject in brackets.

5 Actor Emphatic constructions

The Actor Emphatic (AE) construction, which is well-known from the literature on Māori and Polynesian languages more generally, is treated separately because the empirical description of subject extraction in this construction is more complicated. Furthermore, whilst the constructions in Sections 3 and 4 are typically thought to be monoclausal in neutral declarative contexts, there is some debate about whether the AE is monoclausal (Waite 1990; Pearce 1999) or biclausal (Bauer 1993; 1997; Potsdam & Polinsky 2012) among other issues (see Waite 1990; Bauer 2004; Potsdam & Polinsky 2012 for overviews).

Descriptively, the AE construction, as the name suggests, emphasises the actor/agent argument. The emphasised agent is expressed in a prepositional phrase introduced by a possessive preposition: nā for actual/realised actions, as in (46), and mā for future/intended actions, as in (47). These prepositions are the same as those found in P-POSS constructions, though the -ō form possessive prepositions which are also found in P-POSS constructions, nō and mō, are impossible in the AE construction for reasons that need not concern us here – the O-class possessors are also incompatible with the AE construction in Tahitian (Potsdam & Polinsky 2012). The TAM marker co-varies with the tense of the preposition: i with nā, e with mā. The internal argument of the transitive predicate is grammatically a subject – it is unmarked and can be topicalised (see below) – but the verb is in active form, i.e. it is not passivised.

The AE construction is usually only possible with transitive predicates (Waite 1990: 400). Examples with intransitive predicates are attested in corpora but constructed examples are often rejected by native speakers (Bauer 1997: 506).

Following Bauer (1993; 1997) and Potsdam & Polinsky (2012), I adopt a biclausal analysis of the AE construction. Potsdam & Polinsky (2012) show that Māori has two variant orders in the AE construction, which they call AE2 and AE3 (AE1 is a variant attested in Tahitian but not in Māori). In AE2 the subject (the internal argument) follows the verbal predicate, whilst in AE3 it precedes the verbal predicate but follows the prepositional predicate phrase.

Potsdam & Polinsky (2012) argue that the prepositional predicate phrase, nā Pita, is the matrix predicate phrase; that the internal argument, te hipi, is grammatically a subject in both AE2 and AE3; and that the TAM marker and verbal predicate, i tīhore, belongs to an embedded clause (following Chung 1978; Bauer 2004). Evidence comes from negation, relativisation and the distribution of TAM markers, among other things (see Potsdam & Polinsky 2012 for details and references).

Potsdam & Polinsky point out that there is still the issue of whether the grammatical subject te hipi is part of the matrix clause or the embedded clause. On the basis of a comparison with the Tahitian AE construction, they propose that the subject is in the embedded clause in AE2, but is raised in AE3 (they do not commit themselves to the exact position of raising). Their analysis as applied to (48) is given in (49); they do not specify how the subject (the internal argument) patterns in terms of constituency in AE3 (see Potsdam & Polinsky 2012: 84) – ec stands for ‘empty category’.

It goes far beyond the scope of this paper to attempt to resolve the outstanding constituency issues or how the overt agent and theme relate to the embedded clause. The schematic structures in (49) will suffice for present purposes.

Returning to our description, the AE construction fits the generalisation from above concerning predicate questioning and subject topicalisation, i.e. both of these are permitted. Predicate questioning has in fact already been seen in (14) above, but the examples are repeated in (50) for convenience, now shown with the matrix predicate phrase in bold and the (embedded) subject in brackets.

(51) shows that subject topicalisation is permitted in AE constructions (recall that the grammatical subject corresponds to the internal argument).

This is one of the most common ways to topicalise internal arguments of transitive predicates in Māori, direct objects typically not being very accessible (see also Section 6.1). The AE construction can also be used to relativise internal arguments (Bauer 1997: 570).

Given that the matrix predicate phrase is prepositional, our generalisation would lead us to expect that ko-fronting of the subject for questioning and/or focus would be permitted. However, the empirical facts are somewhat complicated. On the one hand, it is reported that the subject of AE constructions can be focused using ko-fronting, as predicted by our generalisation. Two examples are given below (small capitalisation indicates strong stress):

On the other hand, it is also reported that subject questioning in AE constructions is prohibited, as in (54), which is not predicted by our generalisation.

Such examples can only receive an echo interpretation, a point I return to below.

Pearce (1999) proposes that subject questioning in AE constructions is prohibited because the PP containing the emphasised agent occupies SpecFocP, the dedicated focus position in the left periphery (see also Waite 1990). This provides a straightforward explanation for why the agent in the prepositional predicate phrase can be questioned, and would correctly rule out examples like (54), though something more would have to be said about (52) and (53), e.g. perhaps these examples involve a different type of focus which targets a distinct and higher focus position. However, there are problems for this type of analysis.

First, Potsdam & Polinsky (2012) note that, if the emphasised agent in the focused PP is Aʹ-moved to SpecFocP or some left peripheral position, it will presumably move across the grammatical subject (the internal argument). Therefore, in cases where the agent is an R-expression and co-indexed with a theme subject, we would expect a Strong Crossover violation. However, Potsdam & Polinsky observe that such examples are grammatical, regardless of the position of the theme subject.

Potsdam & Polinsky note a second prediction, namely the theme subject is predicted to be able to bind the representation of the agent in the agent’s supposed base position. However, this is not possible.

The examples in (55) and (56) thus show that the agent (in the predicate phrase) does not undergo obligatory reconstruction and cannot even undergo optional reconstruction to a position below the theme subject. This strongly argues against Aʹ-movement of the predicate phrase, and suggests that the PP is base-generated in a position c-commanding the theme subject.

Could it be that the PP is base-generated in the left peripheral focus position? We have already seen that ko-fronting for focus is permitted in (52) and (53) above, although it was pointed out that these may potentially involve different types of focus. However, evidence from time adverbials suggests that the PP predicate phrase is not as high as a left peripheral position. In Māori, time questions require fronting of the question phrase (note that the subject also tends to be fronted in such cases). (57) is an example involving a verbal predicate construction.

Bauer (1997: 436) points out that time adverbials are commonly fronted in declarative clauses anyway, suggesting that this may be in-situ questioning of an already-fronted time phrase, rather than a fronting-for-questioning strategy.

Now, fronted time adverbials are possible in AE constructions in declarative contexts, as in (58a), but questioning the time phrase is not permitted, as in (58b).

These data suggest that the position required for questioning time adverbials is syntactically available in AE constructions, and thus that the ungrammaticality of (58b) results from an independent interpretive property of AE constructions. Supporting evidence for this idea comes from the observation that ko-fronting or Indefinite Subject Fronting of an interrogative subject is syntactically available, but the result can only be interpreted as an echo question (recall (54)).

Therefore, putting this independent interpretive property aside, I conclude that subject extraction in the AE construction is permitted, as expected for a prepositional predicate construction.

6 Analysis

In the previous sections, we concluded that subject topicalisation and predicate questioning are permitted in all constructions, whilst subject questioning is permitted in verbal and prepositional predicate constructions but prohibited in nominal predicate constructions (see also de Lacy 1999). In Section 6.1, I adopt a cleft analysis of questions whereby the questioned constituent is analysed as the matrix predicate, and the matrix subject is analysed as a headless relative clause (see Bauer 1991; 1993; 1997; Potsdam & Polinsky 2011). I will argue that subject questioning in nominal predicate constructions is ruled out by intervention of the nominal predicate phrase, as detailed below. In Section 6.2 I discuss an alternative type of proposal to the restrictions on subject questioning, which I dub the complementary distribution analysis, arguing that it is too restrictive, whilst in Section 6.3 I discuss some potential problems for a cleft analysis, ultimately concluding that they are not particularly severe.

7 Discussion

I have proposed that Māori triggers Aʹ-movement to SpecCP and SpecTopP using [D] and [X] features respectively, rather than more “traditional” and dedicated Aʹ-features such as [WH], [REL] and [TOP] familiar from analyses of European languages. I have argued that this captures the extraction profile of Māori in a straightforward fashion, utilising the familiar mechanisms of featural Relativised Minimality. I would also argue that there are conceptual benefits to the analysis proposed here. If Aʹ-movement is defined as movement into the C-domain and if the specific interpretations for different types of Aʹ-movement are determined by the field in which the moving element lands (as is standardly assumed in the cartographic literature following Rizzi 1997), then grammatical systems should in principle be free to trigger Aʹ-movement using features like [D] and [X], as in Māori, or features like [REL], [WH] and [TOP].²¹ Indeed, as Chomsky (1981: 7) recognised at the inception of the Principles and Parameters framework: “We need not expect, in general, to find a close correlation between the functional role of such general processes [i.e. the general processes underlying notions such as ‘passive’ and ‘relativisation’ – JD] and their formal properties, though there will naturally be some correlation”. In the present context, the Aʹ-movements underlying both Māori and English relativisation have the same ‘functional role’ and both share the formal property of involving movement into the C-domain, but they are triggered by distinct formal features and so the Aʹ-movements in both languages exhibit different formal properties in terms of which elements are accessible and which elements count as interveners.

Similar analyses have been independently proposed in the literature. Landau (2015) proposes that a [D] probe on a low C head derives the subjecthood of PRO; Belletti (2015) proposes that new information focus clefts in French are restricted to subjects because the relevant head in the C domain of the embedded clause probes for [D]; and Levin (2017) proposes that K’ichean C probes for [D], which results in movement of the absolutive or ergative argument depending on which argument is higher, which in turn depends on the internal structure of the vP.

Other authors have also questioned whether the A/Aʹ-distinction is universal. Davies (2003) and Davies & Kurniawan (2013) argue that Madurese and Sundanese respectively lack wh-movement, and that what has previously been analysed as Aʹ-movement is in fact A-movement, whilst van Urk (2015) and van Urk & Richards (2015) argue that Dinka has both A- and Aʹ-features but that these always form composite probes, meaning there is no formal distinction between A- and Aʹ-movement. Aldridge (2017a; b; 2018) proposes that in languages where there is no formal A/Aʹ-movement distinction, it is only a [uφ] feature which motivates movement to SpecCP. The A/Aʹ-distinction only arises when C has two sets of features which motivate two distinct movements. In such cases, there is C-T Inheritance (Richards 2007; Chomsky 2008), for example, in English, [wh] and [φ] originate on C: [φ] is inherited by T (hence A-movement targets SpecTP), whilst [wh] remains on C (hence Aʹ-movement targets SpecCP). Aldridge thus derives the A/Aʹ-distinction from C-T Inheritance – in the absence of C-T Inheritance, there is no A/Aʹ-distinction and only a [φ] feature is present, this being universally necessary for licensing DPs.

The present proposal is similar to Aldridge’s in that, if there is no formal evidence for Aʹ-features, I assume that they are not present in the grammatical system. However, rather than appeal to the mechanism of Feature Inheritance, I suggest that the formal feature used to trigger Aʹ-movement in a given language/construction results from the acquisition or development of the grammar itself. In this way, although the grammar itself places no constraints on the feature used to trigger Aʹ-movement, other factors will serve to limit the range of possible variation. Consider the question of why Māori triggers Aʹ-movement using nominal-features like [D] and [X], whilst English does so using features like [WH] and [REL].

I adopt the proposal that formal features are emergent (Biberauer 2011; 2017; Biberauer & Roberts 2015a; b; 2017), resulting from the interaction of the Three Factors in Language Design, namely Universal Grammar, the Primary Linguistic Data, and the Third Factor, i.e. principles of data processing and architectural/computational-developmental constraints (Chomsky 2005). Biberauer (2017) proposes as a Third Factor a domain-general cognitive bias called Maximise Minimal Means, which could be paraphrased informally as ‘do as much as possible with as little as possible’, and which in the linguistic domain has at least two language-specific manifestations that guide language acquisition, namely Feature Economy and Input Generalisation (see also Roberts & Roussou 2003; Roberts 2007; Biberauer & Roberts 2017).

According to this view, features are not innate or universal. If there is no evidence in the input – or more specifically the “intake” (see Evers & van Kampen 2008) – for the presence of a particular formal feature, then the language acquirer will not postulate it following Feature Economy. However, once a formal feature has been postulated, the acquirer will attempt to use it as much as possible following Input Generalisation. Crucially, this includes the possibility of recycling features in syntactic domains beyond the one for which a given feature was originally postulated. For example, Wiltschko (2014) argues that there is universally an Anchoring domain in the clausal spine, but that languages may encode Anchoring using Tense, Location or Person, at least some of which are plausibly recycled from prepositional or nominal domains.

Biberauer (2017) argues that the N/V distinction²² is plausibly made first in the acquisition process. Its acquisition is based largely on prosodic cues and accounts for the earliness of basic OV/VO directionality (see also Tsimpli 2014). It is thus plausible that nominal features are present in the early stages of the developing grammatical system, and are thus available for recycling in other domains if the intake provides adequate motivation. Now, an acquirer of Māori will be exposed to instances of subject questioning, subject relativisation and subject topicalisation in the input/intake. Subjects being nominal phrases, it seems plausible to suppose that the acquirer will encode the trigger for Aʹ-movement in Māori by recycling a nominal feature for use as a probe in the C domain. This makes maximal use of minimal means. The acquirer has not postulated any new formal features, and has maximised nominal features which already existed in the system. The result is a Māori-like extraction profile, i.e. one where only subjects are accessible to Aʹ-movement.

I assume that acquirers of languages like English also go through such a stage, i.e. recycling features to serve as triggers for Aʹ-movement is the default. This would account for the fact that questions and relative clauses emerge quite early during acquisition (at around age 3) and for the fact that subject questions and subject relative clauses typically emerge first (see Tsimpli 2014 and references therein).

In the absence of clear instances of Aʹ-movement of non-subjects, this system will persist, ultimately yielding the adult Māori grammar.²³ However, an acquirer of a language like English will be exposed to Aʹ-movement of non-subjects. Such positive evidence will (eventually) indicate that triggering Aʹ-movement using nominal features is insufficient to account for their intake, leading them to override Feature Economy (see e.g. Biberauer 2017; Biberauer & Roberts 2017) and postulate a new type of feature for triggering Aʹ-movement, namely (dedicated) Aʹ-features. Acquirers who have postulated Aʹ-features may still struggle with non-subject Aʹ-dependencies as a result of the computation required to evaluate featural subset and superset relations for the purposes of featural Relativised Minimality, but this arguably has more to do with processing (see Friedmann, Belletti & Rizzi 2009).

This sequence of development means that an acquirer of a Māori-type language will not fall into a superset trap. If they encoded Aʹ-movement triggers using Aʹ-features from the outset, one would either have to explain why Aʹ-movement of non-subjects is ungrammatical in the adult grammar given the standard assumption that negative evidence does not constitute part of the intake, or one would have to say that featural Relativised Minimality is computed differently in Māori-type and English-type languages, which is conceptually unappealing. On the view proposed here, featural Relativised Minimality is computed in exactly the same way in both types of language, the difference resulting instead from the features which enter into those computations.

I believe that this provides a new formal perspective on Keenan & Comrie’s (1977) Accessibility Hierarchy. As Keenan & Comrie and many investigations stemming from their proposal observe, the Accessibility Hierarchy describes both discrete differences between languages as well as parsing and processing preferences within languages (see also the discussions in Hawkins 2004; Guasti, Branchini & Arosio 2012). The Maximise Minimal Means model suggests a bridge between the two. The formal systems of languages with non-subject dependencies have to be motivated by robust cues from the intake. It seems reasonable to think there is a diachronic if not synchronic-acquisitional link between the robustness of cues and the ease with which such cues are parsed and processed. Robust cues will lead to the postulation of formal features during language acquisition, whilst non-robust or absent cues will not. In this way, parsing and processing preferences may be digitised and encoded in the formal system of a language during language acquisition, resulting in hierarchies, such as the Accessibility Hierarchy.

8 Conclusion

This paper has investigated subject extraction in Māori. I argued that, whilst subject topicalisation is generally permitted in all types of construction, subject questioning is restricted, being permitted in verbal and prepositional predicate constructions but prohibited in nominal predicate constructions. I argued that subject questions and subject focus take the form of clefts: the questioned/focused constituent is the matrix predicate phrase and the matrix subject phrase is a headless relative clause. I argued that the restriction on subject questioning reduces to an intervention effect in the CP of this headless relative clause. I proposed that the C head probes for a [D] feature ([D] being shared by both argument and predicate nominals). Consequently, nominal predicate phrases intervene with movement of the subject DP to SpecCP, whilst verbal and prepositional predicate phrases do not. In contrast, I proposed that the Topic head probes for an [X] feature ([X] being found on argument nominals only). Consequently, no predicate phrase intervenes with movement of the subject to SpecTopP.

I argued that by triggering Aʹ-movement using nominal features like [D] and [X] we can account for why generally only subjects are accessible to Aʹ-movement in Māori. I suggested that nominal features may be recycled as triggers of Aʹ-movement in line with recent emergentist ideas, according to which formal features emerge during language acquisition guided by the domain-general cognitive bias to Maximise Minimal Means.

Abbreviations

ACC = accusative, CLS = classifier, CLS(FUT) = future classifier, CONTR = contrastive particle, DIST = distal, DUP = reduplicated segment, EQ = equational ko, EXCL = exclusive, INCL = inclusive, KO = interrogative/focus ko, NEG = negator, PART = particle, PASS = passive, PERS = personal particle, PL = plural, PRES = present, PROX1 = proximal (near speaker), PT = past, Q = question word, TAM = tense/aspect/mood marker, TOP = topic ko, 3SG = third person singular

Funding Information

This research was carried out in part under the Rethinking Comparative Syntax project funded by the ERC (grant no. 269752) at the University of Cambridge. Publication of this research is supported by the Singapore Ministry of Education under WBS R-103-000-145-115 to guest editor Michael Yoshitaka Erlewine.

Competing Interests

The author has no competing interests to declare.