Volume 86 1977 > Volume 86, No. 3 > Maori as an accusative language, by Sandra Chung, p 355-370
MAORI AS AN ACCUSATIVE LANGUAGE
In a recent article published in this Journal, 1 M. B. W. Sinclair has departed from previous treatments of Maori syntax to claim that Maori is an ergative language. According to him, this proposal provides a superior account of several long-standing problems of Maori syntax; notably, the high frequency of the sentence type referred to as “passive” in the traditional grammars. The purpose of this brief note is to give Sinclair's proposal the close scrutiny which it deserves. I will show, first, that the Maori facts discussed by Sinclair are not accounted for by his proposal at all, so that there is no reason to prefer his analysis to the traditional one of Maori as an accusative language. Second, I will argue that other facts of Maori—not discussed fully by Sinclair—cannot be handled naturally or adequately within his proposal. Hence his proposal must be rejected in favour of the traditional analysis, which is able to account for all of the facts. Finally, I will close with some general remarks about the Maori passive and the relationship of frequency to syntactic structure.
Before turning to the details of Sinclair's proposal, it may be helpful to place it in perspective.
In Maori, corresponding to the intransitive sentence type (1):
there are two sentence types involving underlying transitive verbs. These are the ‘active’, exemplified by (2):
and the ‘passive’, exemplified by (3):
In the active, the logical subject is unmarked and the logical object is marked with the particle i; while in the passive, the logical subject is marked with e and the logical object is unmarked. In the passive, further, the verb does not occur in its stem form but selects an alternant of the passive suffix—(C) (i)a, where C is h, k, m, n, ng, r, t, or wh. 5
Syntactically, the outstanding characteristic of these constructions is that the passive is more frequent than the active. The passive occurs as often as the active for middle verbs (e.g. ‘see’, ‘love’) and considerably more often than the active for true transitive verbs (e.g. ‘build’, ‘kill’, ‘gather’). This of necessity means that it is found in contexts which would be odd, unexpected, or unnatural for the English passive. 6
Despite this, early scholars of Maori identified the active as the basic transitive construction, mentioning in passing that “the Passive Voice . . . is more frequently required than the Active Voice”. 7 This view has been upheld by more recent investigators, although there are several notable exceptions. 8 Hale, Biggs, and Chung and Timberlake, 9 for instance, all maintain that the active is the basic transitive construction from which the passive is derived. On this view, transitive clauses may undergo a Passive rule which promotes the logical object to subject, removes the logical subject to an oblique prepositional phrase (marked with e), and attaches a passive suffix to the verb. Transitive clauses which have not been passivised undergo an Accusative Case Marking rule which marks their objects with i. Because subjects of intransitive clauses, on this analysis, have the same lack of marking as subjects of basic transitive clauses (2), Maori satisfies the definition of an accusative language.
Sinclair departs radically from this view by claiming that the passive is the basic transitive construction from which the active is derived. His proposal can be outlined as follows. Clauses of type (3) would be basic transitive clauses containing a subject and an object. Optionally, such clauses would be allowed to undergo a rule of Antipassive which would demote the logical object to an oblique prepositional phrase (marked with i). Transitive clauses which have not undergone Antipassive would undergo an Ergative Case Marking rule which would mark their subjects with e. Because subjects of intransitive clauses, under this proposal, would have the same lack of marking as objects of basic transitive clauses (3), Maori would satisfy the definition of an ergative language.
Sinclair's proposal is contrasted with the traditional view in (4-5):
(4) The Traditional View: Maori as an Accusative Language
(5) Sinclair's Proposal: Maori as an Ergative Language
Given that the syntactic identity of (2-3) themselves is in question, any terminology for these clause types is likely to reveal some bias. I will continue to refer to (2) and (3) as “active” and “passive”, respectively, since I will argue ultimately that the traditional view is superior to Sinclair's. I turn now to the details of Sinclair's proposal and the reasons for rejecting it.
SUPPOSED ADVANTAGES OF SINCLAIR'S PROPOSAL
Sinclair claims three principal advantages for his proposal over the traditional view that Maori is accusative. In this section I show that these advantages are illusory, and that Sinclair's proposal accounts for the relevant facts no better or worse than the traditional hypothesis.
The first supposed advantage of Sinclair's proposal is that it accounts for the form of transitive imperatives. Imperatives in Maori are either marked with the tense-aspect particle e (if the verb is two morae or less) or else unmarked. Intransitive imperatives employ the verb in its stem form. 10
Transitive imperatives are always passive; they use clause type (3) rather than (2): 11
In order to account for this under the view that Maori is accusative, it is necessary to place a condition on Passive making it obligatory in imperatives. Such a condition would not appear in Sinclair's proposal, since for him (7) would be a basic transitive clause. In this respect, he suggests, his proposal is superior to the traditional view.
It should be obvious, however, that Sinclair's proposal would require an equally complex condition on Antipassive making it impossible for this rule to apply in imperatives. For without such a condition, nothing would prevent Antipassive from applying to (7) and producing the ungrammatical (8). Since this condition does not appear in the traditional hypothesis, it follows that Sinclair's proposal and the traditional one are - 358 equally complex as far as imperatives are concerned. Imperatives, therefore, can hardly be said to constitute an advantage for his proposal. 12
The second supposed advantage of Sinclair's proposal is that it reduces the “ambiguity” of the particle i. As Sinclair notes, i is used in Maori to mark locations, sources, causes, time phrases, objects of comparison, and various other oblique NPs. In the traditional view of Maori as accusative, i is assigned not only to these NPs but also to objects of active transitive clauses (2). In Sinclair's proposal, however, clauses of type (2) would not contain objects but rather oblique prepositional phrases created by Antipassive. This enables him to eliminate transitive objects from the list of NPs to which i is assigned—that is, it enables him to claim that i is assigned to oblique NPs alone. The resulting decrease in ambiguity, according to him, argues in favour of his proposal.
This argument is based on two assumptions, both of which are dubious. First, it assumes that ambiguity is so uncommon in language that any reduction of it in linguistic description is to be viewed as a positive accomplishment. This assumption is misguided; as is well-known, every language tolerates a certain amount of ambiguity. Second, the argument assumes that the traditional hypothesis must mention by name every separate type of NP to which i is assigned, while Sinclair's proposal can state the distribution of i in terms of an “ablative-general oblique case”. 13 This assumption is false. Under the traditional view, the distribution of i can be stated as follows:
(9) Nonsubjects which are not already marked with ki or some other overt case particle are marked with i.
This statement is neither more nor less complex than the statement that would be required in Sinclair's proposal:
(10) Oblique NPs which are not already marked with ki or some other overt case particle are marked with i.
Again, the treatment of i in Sinclair's proposal is comparable in complexity to that within the traditional view. Hence this particle does not provide an argument for or against his proposal.
The third and final advantage claimed by Sinclair for his proposal is that it accounts for the unusually high frequency of the passive (3). It does this, according to him, by claiming that the passive is the basic transitive construction from which the active (2) is derived. But this argument, like the one involving i, is based on hidden assumptions. Logically, there is no reason to suppose that the choice of active or passive as basic should - 359 account for the frequencies of these constructions at all. Frequency is a question of usage; as such, it is independent of the identity of the basic transitive construction and its relationship to derived constructions, which are questions of structure. This means that, logically speaking, Sinclair's proposal explains the frequency of the passive no better or worse than does the traditional view.
In order to make his claims convincing, Sinclair would have to make the additional assumption that frequency and structure were correlated absolutely; in other words, that the more frequent of two related syntactic constructions was invariably the more basic of the two. But it is easy to show that this assumption is false. In Maori, for instance, negative sentences like (11-12) are related by a Raising rule:
Sentences like (12), in which the subject of the affirmative clause has been raised to become the subject of the negative, are considerably more frequent than sentences like (11), in which Raising has not applied—a fact borne out amply by Maori texts. Yet all recent investigations of negation in Maori agree in analysing (11) as more basic than (12). 14 The fact that Sinclair himself 15 treats negative sentences in this way is particularly telling evidence against any supposed correlation of frequency and “basicness”.
I conclude, then, that Sinclair's proposal does not account for the high frequency of the passive any better or worse than does the traditional view. Therefore frequency, like the other considerations discussed above, can hardly be counted as an advantage for his proposal.
DISADVANTAGES OF SINCLAIR'S PROPOSAL
Although Sinclair's proposal and the traditional view appear to be roughly comparable in handling the facts above, there are considerations which differentiate between the two. In this section I present three sets of facts which argue against his proposal.
These facts deal with the transitivity of the active (2) and the passive (3), which the reader will recall are treated differently under the two proposals. Within the traditional view, the active (2) is superficially transitive; the passive (3) is derived from it by a Passive rule which promotes the logical object to subject. Since the passive has no surface object once this rule has applied, it follows that it is superficially intransitive. Within Sinclair's proposal, in contrast, the passive (3) would be superficially transitive; the active (2) would be derived from it by an Antipassive rule which would demote the logical object to an oblique NP. - 360 Since the active would have no surface (direct) object once this rule had applied, it would follow that it would be superficially intransitive.
Given this, the choice between the two proposals reduces to the question of which of (2) or (3) is superficially transitive. The facts discussed below answer this question in favour of the active (2); hence they argue that the traditional view is superior to Sinclair's.
Having sketched the general strategy for this section, I now turn to the facts. These involve (a) the distribution of the nonspecific article he, (b) the process of ko Clefting, and (c) control of ki te Equi.
The Distribution of he
As was first noted, I believe, by Hale and Hohepa, 16 the nonspecific article he has an extremely restricted distribution. This article can modify subjects of intransitive clauses, as in (13-14):
It can also modify derived subjects (originally logical objects) of passive clauses:
But it does not appear with other types of NPs. In particular, it cannot modify the logical subject or object of an active clause:
The agent (originally the logical subject) of a passive clause:
Or any oblique NP:
Within the traditional view, the derived subject of the passive is a type of intransitive subject, so that the distribution of he can be stated as follows: He can modify intransitive subjects, but no other types of NPs. This restriction can be stated at surface structure or any level of structure after Passive has applied. Note that it automatically accounts for the inability of he to occur with the subject of the active (17), since the active is, within this view, syntactically transitive.
Things are not so easy within Sinclair's proposal, however. This can be seen by reviewing the facts under his assumption that the passive rather than the active is superficially transitive. Within Sinclair's proposal, he would be able to occur with subjects of intransitive clauses (13-14) and with logical objects of passive clauses as well (15-16). (Recall that these are derived subjects under the traditional view.) But it would not be able to occur with logical subjects of passive clauses ((19)—agents under the traditional view) or with any oblique NPs (20). Such a distribution would seem to suggest that he was restricted to absolutives, where an absolutive is an intransitive subject or a transitive direct object. Now recall that subjects of active clauses have undergone Antipassive within Sinclair's proposal and been turned into intransitive subjects; hence his proposal predicts that they should, like other absolutives, be able to take he. But they cannot (17). Their inability to do so argues that active clauses are not syntactically transitive, and so Sinclair's proposal is not correct.
For Sinclair, perhaps the only way to salvage this situation would be to claim that the restriction on he was stated before Antipassive had applied. At this point, subjects of active clauses would still be transitive subjects, and he would not be allowed to modify them. There are, however, other facts which cannot be described if this order of events is adopted. In particular, logical objects of active clauses (18) cannot be modified with he. Within Sinclair's proposal, these NPs are absolutives before Antipassive has applied, so they would be able to take he if the restriction on he preceded Antipassive. The only way to prevent this and block (18) would be to force this restriction to be stated after Antipassive had applied. But this, of course, contradicts the ordering required above.
We have, then, arrived at an ordering paradox of sorts. In order to describe (13-14) and (17) within Sinclair's proposal, the restriction on he cannot follow Antipassive; but in order to describe (16) and (18), the restriction on he cannot precede it. Put another way, there is no single level of structure at which the distribution of he can be determined. Perhaps the only way to resolve this paradox within Sinclair's proposal would be to require that the restriction on he refer simultaneously to two levels of structure, before and after Antipassive. It would then be stated as follows: He can modify NPs which are absolutives before Antipassive and still absolutives after Antipassive has applied. Significantly, no such appeal to dual levels of structure is required within the traditional view, which accounts automatically for the facts. It follows from this that the traditional view is superior to Sinclair's.- 362
Essentially the same sort of argument against Sinclair's proposal can be constructed from ko Clefting, a process which moves a focused NP to the left and marks it with the predicate/focus particle ko. The domain of this rule can be described as follows. Subjects of intransitive clauses can be clefted:
So can subjects of active transitive clauses:
And derived subjects (originally logical objects) of passive clauses:
Further, logical objects of active clauses can be clefted. These NPs leave behind a pronominal copy ai, which cliticises to the right of the verb:
NPs other than subjects or direct objects cannot be clefted, with or without the copy ai. Thus agents (originally logical subjects) of passive clauses cannot undergo this rule:
Neither can oblique prepositional phrases marked with i:
Within the accusative hypothesis, these facts can be accounted for by restricting ko Clefting to subjects and direct objects, and further requiring that clefted direct objects leave behind the copy ai. This simple statement generates (21-25) but excludes the ungrammatical (26-27).
But within Sinclair's proposal, the facts are more elusive. Recall that for Sinclair, the logical objects of active clauses have been turned into oblique NPs by Antipassive; hence his proposal predicts that they should be indistinguishable from true oblique NPs as far as ko Clefting is con- - 363 cerned. But this is not the case. The fact that (25) is grammatical, while (27) is not, argues against his treatment of active clauses.
For Sinclair, the only way to avoid this would be to claim that ko Clefting was ordered before Antipassive. Such an ordering would allow ko Clefting to apply at a point when logical objects of active clauses were still absolutives, and therefore distinct from oblique NPs. But this ordering cannot be correct, because there are other NPs which become eligible for Clefting only after Antipassive has applied. These are the agents of passive clauses (=transitive subjects for Sinclair), which can undergo Clefting after they have been turned into absolutives by Antipassive (23), but not before (26).
As before, in order to sidestep this paradox within Sinclair's proposal, we would have to allow ko Clefting to refer simultaneously to two levels of structure; one before, and the other after, Antipassive had applied. The restrictions on the rule would then be the following: Clefting applies to absolutives; 20 it also applies to NPs which were absolutives before Antipassive but not afterwards, provided that these leave behind the copy ai. But if Maori is accusative, it is possible to account for the total range of facts without having to resort to such an appeal. The greater naturalness and generality of this argues that the traditional view is superior to Sinclair's.
Control of ki te Equi
A final argument for the traditional view is provided by the process of ki te Equi, discussed in a number of recent works. 21 Here I will not attempt to provide a full description of this process, but will simply sketch those aspects of it which are relevant to the argument.
Ki te Equi applies to sentential complements embedded under verbs of volition or ability, verbs of motion, and verbs of sending or command. Complements of these verbs are typically introduced by the subjunctive particle kia:
When the subject of the complement clause is coreferential with some NP associated with the main verb, it can be deleted by ki te Equi. When this happens, kia is replaced by the complementiser ki te:
It is convenient to refer to the coreferential NP in the main clause, which can be said to trigger the deletion, as the controller of ki te Equi.
The choice of a controller is determined partly by the semantics of the main verb, in a way not well understood even for the English analogue of this rule. 23 But there are syntactic constraints on the controller as well, and it is these that we are concerned with. Ki te Equi can be controlled by subjects of intransitive (31) and active transitive (32) clauses:
It can also be controlled by derived subjects (originally logical objects) of passive clauses:
And logical objects of active clauses:
But it is never controlled by other types of NPs. In particular, it cannot be controlled by oblique prepositional phrases marked with i, or by passive agents: 26
Compare (35) with (36), where ki te Equi has not applied: 27
Within the accusative hypothesis, these facts can be dealt with by allowing ki te Equi to be controlled by subjects and direct objects, but no other types of NPs. When combined with the semantic restrictions alluded to above, this will correctly produce (30-34), but not (35).
Within Sinclair's proposal, however, a paradox arises like that discussed for ko Clefting above. Logical objects of active clauses (34) can serve as controllers; this would seem to indicate that controllers should be identified before Antipassive, when these NPs are still distinct from oblique prepositional phrases. 28 At the same time, logical subjects of active clauses can serve as controllers (32), but passive agents cannot (35); this would seem to suggest that controllers should be identified after Antipassive, when passive agents (=transitive subjects for Sinclair) have already been converted to absolutives. As before, there is no single level of structure at which the generalisation can be stated. The only way out of this paradox would be to allow the syntactic constraints on controllers to refer to two levels of structure. They would then be stated as follows: Ki te Equi can be controlled by NPs which are absolutives or were absolutives before Antipassive applied. Such a move, however, simply disguises the fact that logical objects of active clauses do not act like oblique NPs. This observation is captured naturally and automatically if Maori is accusative, since the essence of the accusative hypothesis is that logical objects of active clauses are syntactic direct objects. Therefore ki te Equi, like the other phenomena discussed above, argues for the traditional view and against Sinclair's. 29
In the above, I exhibited three sets of facts and showed how they can be accounted for within the traditional view that Maori is an accusative language. The fact that they cannot be described within Sinclair's proposal—at least not without appealing to ad hoc complications—argues that his proposal should be rejected. There is a sense, however, in which the similarities among these sets of facts provide the strongest argument that Maori is not an ergative language. There are two points to consider.
First, in order to account for each set of facts within Sinclair's proposal, it would be necessary to invoke the same type of complication in the - 366 grammar: 30 simultaneous reference to more than one level of structure (also known as global reference). Within Sinclair's proposal, the restriction on he would have to refer to NPs which were absolutives both before and after the application of Antipassive; the restriction on controllers of ki te Equi would have to refer to NPs which were absolutives either before or after Antipassive; the statement of ko Clefting would have to distinguish absolutives in derived structure from NPs which were absolutives before Antipassive, but not afterwards. The fact that global reference would be necessary in all three cases is revealing, for it points out that generalisations are being missed. Within Sinclair's proposal, nothing predicts or follows from the fact that ki te Equi, ko Clefting, and the nonspecific article all require global reference. It is simply a descriptive accident. Yet the three appeals are clearly connected, for all must mention the Antipassive rule, and all turn out to be unnecessary if his proposal is rejected in favour of the view that Maori is accusative. We would like a theory which recognises these underlying regularities and is able to account for them. The inability of Sinclair's proposal to do so argues strongly that it should be rejected.
Second, as was mentioned above, ki te Equi, ko Clefting, and the nonspecific article all provide evidence about transitivity in Maori. Ki te Equi and ko Clefting show that logical objects of active clauses do not have the syntactic character of oblique prepositional phrases, since they fail to pattern with the latter for the purposes of these rules. The nonspecific article he establishes that logical subjects of active clauses do not have the character of intransitive subjects, since—unlike other intransitive subjects—they cannot be modified with this article. These conclusions bear out the traditional view that the active is syntactically transitive. At the same time, they contradict Sinclair's proposal, which predicts that logical subjects and objects of active clauses should act, respectively, like intransitive subjects and oblique NPs. As such, they argue strongly that Maori is an accusative language.
It is, of course, always possible to produce an ad hoc description which will disguise these conclusions and mechanically grind out the facts; I have alluded to descriptions of this sort above. But no such description is strong enough to disguise the lack of positive evidence in Maori for Sinclair's proposal. If Maori were genuinely an ergative language, then we would expect to find many syntactic phenomena which treated logical subjects of active clauses like intransitive subjects, and logical objects of active clauses like oblique NPs. But Maori has nothing like this at all. In the complete absence of facts which clearly and directly support Sinclair's proposal, the claim that Maori is ergative reduces to a kind of game.
I conclude from this that Sinclair's proposal must be rejected in favour of the traditional view. That is, despite the high frequency of the passive, Maori has the accusative structure outlined in (4). This appears to be the only conclusion upheld by the facts.- 367
ON THE RELATIONSHIP OF FREQUENCY TO SYNTACTIC STRUCTURE
Although there are other debatable points in Sinclair's article, 31 the above should be sufficient to establish that his main thesis is untenable; Maori is not an ergative language. Rather, it is an accusative language, as was proposed first by the traditional grammarians and later by Hale, Biggs, and Chung and Timberlake, among others. The conclusion that Maori is accusative explains the facts of ki te Equi, ko Clefting, and the nonspecific article, as well as other syntactic phenomena not discussed here. 32 However, it leaves untouched the question of the high frequency of the passive, and it is to this issue that I would like now to turn.
It should be clear by now that the high frequency of the Maori passive does not truly conflict with the conclusion that it is derived from the active syntactically. As I pointed out above, the frequency of a construction is independent of its status as basic or derived; the former is a question of usage and the latter, one of syntactic structure. Logically, then, there is no reason to expect the two to be correlated absolutely. Further, there is abundant evidence from language indicating that frequency is not necessarily correlated with “basicness”. In Maori, as mentioned above, negative sentences which have been affected by Raising are far more frequent than ones which have not. Yet all recent investigators agree in deriving the more frequent, raised construction from the less frequent one, in which Raising has not applied. In English, verbs such as try do not allow their complement clauses to contain surface subjects (e.g. He tried to leave; but *He tried for him(self) to leave). But there are compelling reasons for believing that these complements are not subjectless underlyingly, but instead have - 368 subjects which are deleted by an Equi rule like that described above. Examples of this sort could be multiplied; they show that the most frequent surface manifestation of a construction is often not the same as its most basic or underlying form. Given this, any claim of an absolute correlation between frequency and “basicness” would be simplistic.
Still, there is a sense in which we feel the high frequency of the Maori passive to be unusual—and it is this intuitive feeling that lies behind much of the initial appeal of Sinclair's proposal. It is therefore incumbent on any alternative hypothesis to provide some explanation of it. That is, we should require the right analysis to account for the fact that the passive is derived from the active, as well as for the fact that its high frequency makes this derivation seem unusual or bizarre. In the hope of providing such an explanation, I offer the following remarks. 33
Despite the failure of frequency and syntactic status to coincide, I think it is clear that the two tend to be related as far as basic clause types are concerned. In English, for instance, the basic transitive construction (i.e., the active) is also statistically the most frequent. The same is true in many other languages. Although there are notable exceptions, such as Maori and Bahasa Indonesia, these should not obscure the fact that the most basic transitive construction is often the most common on the surface. The correlation is merely a tendency rather than an absolute requirement. Put another way, although the correlation does not always hold, it appears that grammars which do exhibit it are more highly valued, in some sense, than ones which do not. The idea that underlying and surface structure tend to reflect each other transparently is a close syntactic analogue of the opacity principle, proposed for phonology by Kiparsky. 34
Since this tendency is weak enough to allow exceptions like the Maori passive, one might ask exactly how it is relevant linguistically. The answer, I think, is to be found in the transmission of language from one generation to the next. In the process of constructing a grammar from the available surface data, new generations of speakers must themselves answer the question of whether the most frequent transitive construction is the most basic syntactically. The possibility then arises that these new speakers may reinterpret the facts to conform to the more highly valued situation. That is, confronted with a construction like the Maori passive, they may simply assume that it reflects the principle that underlying and surface structure tend to coincide. As a result, they may conclude either that the passive is not as frequent as it first appeared, or else that it must be (reinterpreted as) the basic transitive construction.
It appears, then, that the partial correlation of frequency and syntactic structure is relevant for linguistic change. While languages may have basic transitive constructions which are not common on the surface, or derived - 369 constructions which are, these situations may tend to be eliminated through change in the interests of greater transparency. This proposal explains how the Maori passive can be derived from the active synchronically, and yet still be considered unusual.
Finally, it is of some significance that the proposal just outlined also finds support within the larger Polynesian family. I have argued elsewhere 35 that Proto-Polynesian was an accusative language in which the passive was unusually frequent; and that a tendency towards transparency caused it to become less frequent in some daughter languages (e.g. Hawaiian, Tahitian) and was responsible for the rise of ergativity in others (e.g. Samoan, Tongan). If my analysis is correct, then the transparency principle provides a historical explanation for the diversity of case and voice systems in Polynesian — a diversity which has been the subject of much syntactic controversy.
But regardless of the outcome of this larger issue, the facts point clearly to the conclusion that Maori has not succumbed to this principle. It is still accusative in case marking and syntactic structure, and its passive — though more frequent than the active — is still derived from the active syntactically. To quote Hale: 36 “Maori is not yet an ergative language.”
1 Sinclair 1976.
2 Waititi 1969:35.
3 Biggs 1969:31.
4 Biggs 1969:33.
5 Williams 1971.
6 Clark 1973b, 1973c; Chung and Timberlake 1974; Chung 1976.
7 Williams and Williams 1965:42.
8 Hohepa (1967) treats both the active and passive as basic transitive constructions, neither being derived from the other. His proposal is discussed by Hale (1968). Clark (1973b) suggests that the active and passive are transitive constructions which differ principally in perfectivity, although he later (1973c:598) appears to retreat from this view. For a discussion of his (1973b) position, see Chung and Timberlake (1974).
9 Hale 1968; Biggs 1969; Chung and Timberlake 1974.
10 Maori sentences which are not accompanied by an explicit textual reference were provided by Patrick W. Hohepa in 1969-70. I would like to thank him and Kenneth Hale, with whom he taught a course in the structure of Maori in the fall of 1969, for their help and encouragement.
11 Exceptionally, transitive imperatives with a reflexive direct object may be active or passive.
12 Two objections might conceivably be raised to this line of reasoning. First, it might be objected that the condition on Antipassive required within Sinclair's proposal was characteristic of all ergative languages, and so less costly to a theory of grammar. But this objection is false. There are ergative languages, such as the Australian language Dyirbal, which do not exhibit any such condition and allow transitive imperatives to undergo Antipassive (Dixon 1972:110-11). Second, it might be objected that the traditional analysis of Maori imperatives was inferior because imperatives in natural language were never syntactically passive. This objection also is false. There are languages, such as the Western Austronesian language Malagasy, in which transitive imperatives are passive as often as, or more often than, they are active (Keenan 1976:259-60).
13 Sinclair 1976:24.
14 Hohepa 1967; Chung 1970; Clark 1973a; a number of arguments for this analysis are presented in Chung 1970, 1976.
15 Sinclair 1976:11.
16 In course lectures, 1969.
17 Orbell 1968:42.
18 Orbell 1968:74.
19 Orbell 1968:80.
20 These restrictions would have to be stated after Antipassive, just as the comparable restrictions within the accusative hypothesis must be stated after Passive has applied. The same is true for the other restrictions discussed in Sections 3.1 and 3.3.
21 Hale 1968; Hohepa 1970; Chung and Timberlake 1974; Chung 1976.
22 Biggs, Hohepa, and Mead 1967:83.
23 See, though, Postal 1970.
24 Orbell 1968:104.
25 Waititi 1969:16.
26 P. Hohepa's judgement on (35) was kindly confirmed by S. M. Mead and M. B. W. Sinclair. Note the ability of hiahia ‘want’ to undergo Passive in (36), which confirms that it is a transitive verb in (32); its logical subject in (32) is ngā tāngata ‘the men’ and its logical object, the entire complement. clause.
27 In (36a), from Biggs, Hohepa, and Mead (1967:46), the agent and derived subject of the complement clause have disappeared through pronominalisation by deletion.
The referee (who reviewed the manuscript) pointed out that (36a) may have an alternative reading in which the derived subject of hiahia-tia is ‘the woman’ rather than the complement clause (i.e. “When a man desired (her), so that (he) could sleep (with her) . . .”). In (36b), provided by the referee, this alternative reading is not possible and the complement clause must be the derived subject of hiahia-tia.
28 This is in fact suggested by Sinclair (1976:23).
29 It is possible to sidestep this conclusion by assuming that ki te Equi and Antipassive are freely ordered. This lack of ordering would produce (32) (Antipassive precedes Equi), (34) (Equi precedes Antipassive), and the other grammatical sentences without requiring reference to more than one level of structure. Significantly, though, it is not possible to get around the facts of ko Clefting and the nonspecific article in a comparable fashion. These phenomena thus provide the stronger arguments against Sinclair's proposal.
30 See, though, note 29.
31 For instance:
Sinclair claims that Maori has no rule of Quantifier Float (1976:11). But it is possible in Maori for the quantifier katoa ‘all’ to float from certain syntactic types of NPs to a position immediately following the main verb:
(i) ka rewa katoa ngā waka
aor start all the—pl canoe
‘All the canoes put out to sea.’ (Orbell 1968:90)
Sinclair claims further (1976:14) that it is impossible in Maori to relativise NPs other than subjects (his a-phrases). But in fact, many (perhaps all) nonsubjects can be relativised by pronominalising the relative noun to the copy ai.
Sinclair's discussion of the “semantic subject” properties of the passive agent (1976:15-16) ignores the fact that the properties which he discusses are not limited to subjects (or subjects and agents) in Maori, but extend to a wide range of NPs (see Chung and Timberlake 1974). Thus NPs of all syntactic types can be “readily and commonly” pronominalised; and all oblique NPs (as well as logical subjects and objects of active transitive clauses) are required to be marked formally as specific rather than nonspecific. Finally, Sinclair claims (1976:20) that the failure of the passive to co-occur with indirect clauses (marked with i) argues that it is not a derived stative construction. But within the traditional view, this failure follows from the fact that the passive is derived from the active, and active transitive clauses may not contain indirect causes. That is, the ungrammaticality of the passive (ii):
(ii) *i tuku-na mātou e te pirihimana i a Hōne
past release-Pass we Agt the policeman because pers John
(The policeman released us because of John.)
follows from the fact that it is derived from the active (iii):
(iii) *i tuku te pirihimana i a mātou i a Hōne
past release the policeman Acc pers us because pers John
(The policeman released us because of John.)
32 See Chung 1976.
33 As will become clear, these remarks constitute an explanation of the Maori passive only in a limited sense; they do not attempt to explain why the passive is so frequent, but only why its high frequency strikes us as unusual or remarkable. I have no synchronic explanation for why the passive should be so frequent in Maori. In general, it is difficult to give synchronic reasons for why any syntactic construction in a language should be more frequent than any other. A historical explanation is sometimes possible, though, as I have suggested below.
34 Kiparsky 1971.
35 Chung 1976.
36 Hale 1968:98.