Volume 85 1976 > volume 85, No. 1 > Is Maori an ergative language, by M. B. W. Sinclair, p 9-26
IS MAORI AN ERGATIVE LANGUAGE?
The Maori language presents many problems to modern theories of grammar. Many of the anomalies in Maori—the so-called “passive-imperative” for example—have yet to be adequately explained and have given rise to a variety of ad hoc hypotheses. Hohepa 1 for example treats both “active” and “passive” as base forms, thus denying a transformational and (within the theory he used) semantic relation between them. 2 Further examination 3 has led to the hypothesis that Maori is in the midst of a drift to or from ergativity. The purpose of this paper is to examine seriously the hypothesis that Maori is an ergative language.
In nominative-accusative (henceforth abbreviated: N-A) languages the notional subject (or actor) of a transitive verb and the notional subject (actor) of an intransitive verb have the semantic and syntactic properties of the subject of the sentence. N-A languages are by far the most common; typical examples are English, French, German, etc. Ergative-absolute (henceforth: E-A) languages on the other hand identify the actor of an intransitive verb and the notional direct object (patient) of a transitive verb as subject of the sentence. The most notable example for which a detailed analysis is available is Dyirbal, a North Queensland, Australian language. 4 E-A constructions are also common in Caucasian languages 5 and are the basic forms of Basque and South Greenlandic Eskimo. Tongan is also generally considered to be E-A, although this has been disputed. 6
The theory in which we shall examine Maori is that of relational grammar as developed by Keenan, Comrie, Postal and Perlmutter, 7 and Johnson. 8 Unfortunately, little published work on relational grammar is available. However, a clear exposition can be found in Johnson, and the general ideas are used in Keenan and Comrie 9 and Keenan. 10 In - 10 “standard theory” (e.g. Chomsky 11 and “generative semantics” grammatical relations such as “subject of”, “direct object of” are defined configurationally, base structures being essentially ordered. This has proved both theoretically 12 and empirically inadequate, there being no conclusive way to demonstrate what the base order should be. Relational grammar treats grammatical relations as primitive and base structures as unordered.
Thus transformational rules refer to categories of grammatical relations, such as Subject, Direct Object, rather than to numbered elements as in standard theory. For the practical purposes of this paper however, the transformations are exactly as in standard theory, as they are described in introductory texts. 13
Transformations in relational grammar are divided into two classes: those which change grammatical relations (roughly, those which are cyclical in standard theory) and those which do not (post-, or non-cyclical in standard theory). Each language has a specific linearisation function which produces its characteristic surface order. Within this theory it seems that it may be possible to capture many linguistic universals and thus, ultimately, to provide a more accurate characterisation of human language. 14
In this paper we attempt to approach Maori as if it were a newly discovered language, and with no prejudice in favour of the N-A or E-A form. 15 Thus only relatively gross translations will be accepted.
The first problem is to determine the distribution of subject properties in Maori. This is done in Section 1 relying heavily on the work of Edward Keenan. 16 In Section 2 the traditional N-A hypothesis is examined briefly and some of its problems noted. The E-A hypothesis is discussed in Section 3. It is argued that the E-A hypothesis provides a better account of the Maori language than does the traditional N-A hypothesis, and resolves major anomalies arising from the latter. The case, however, is not absolutely conclusive and areas for further research will be suggested.
Subject Properties in Maori
Maori is basically VSO or VOS. The order of the actor NP (notional subject) and the patient NP (notional object) is interchangeable, not with complete freedom, but subject to minor constraints that have yet to be determined. Traditionally two forms of transitive sentence have been - 11 distinguished, the “active” and the “passive”, illustrated by (1) and (2) respectively.
Perf. hit John Mary
“John hit Mary.”
Perf. hit Mary John
“Mary was hit by John.”
The various NPs will be referred to as “a-phrase”, “e-phrase”, “i-phrase” and “ki-phrase” (see example 3). The subject of an intransitive verb is an a-phrase; indirect objects (datives) are ki-phrases.
go John to university (pres. continuous)
“John goes to university.”
The immediate problem is to determine the distribution of subject properties among these four most simple types of NP. The framework for this investigation is Keenan's work on the relational concept “subject of” as a linguistic universal. Of course not all of Keenan's tests apply to Maori; for example there is no quantifier float in Maori and thus a subject cannot be the NP most likely to launch floating quantifiers. Keenan's tests also presuppose an antecedently determined notion of basic sentence. As this is one of the things that are in dispute in this paper, for the present we shall simply apply the appropriate tests to all possible candidates. The following, although far from complete, provides a fairly clear picture of “subject of” in Maori.
Transformations provide a powerful test for subject:
“Subject NPs are the most potent, i.e., they get mentioned more often in transformations than do other NPs. More syntactic transformations will mention the NP which is in fact subject than any other NP.” 17
Subjects isolated by this test are syntactic subjects and may or may not coincide with the NPs picked by semantic tests. 18
In Maori, negation is an optional movement rule, fronting only the a-phrase. 19 Negatives in Maori are probably best treated as higher verbs and the fronting process as an instance of Raising. That only a-phrases can be thus raised is illustrated in the following examples. 20
To form the negation of (4):
Past go Pani to the house
“Pani went to the house.”
the negation word kaore is added and the a-phrase fronted:
Neg. Pani go to the house
“Pani didn't go to the house.”- 12
The indirect object ki-phrase cannot be thus raised under the higher predicate kaore, as is shown by the ungrammaticality of (6):
With transitive verbs also only the a-phrase can be raised under negation. To negate (7):
hit John Rewi
“Rewi hit John” or “John was hit by Rewi.”
It is possible, although slightly odd, to simply prefix kaore:
More usual is to raise the a-phrase, as in (9):
Semantically this indicates that it wasn't John who was hit by Rewi. The so-called “active” form of (7):
can be negated, again raising the a-phrase, to indicate that it wasn't Rewi who hit John:
That neither the e-phrase nor the i-phrase can be raised under negation is shown by the ungrammaticality of (12) and (13) respectively.
There is also in Maori a commonly used topicalisation rule which moves the a-phrase to preverbal position. Thus from (7) by topicalisation we get:
“It was John that Rewi hit.”
The e-phrase cannot be thus topicalised:
To topicalise the actor NP of (7) (Rewi) we have to use the “active” form, (10) in which the actor is in the a-phrase. Thus we get:
This is a very common form; the particle na emphasises agency, and the i is deleted from the i-phrase. The i-phrase of (10) cannot itself be topicalised, as is shown by the ungrammaticality of (17):
Equi-NP-deletion 22 (Equi) in Maori is constrained in much the same way as in English. With intransitive verbs the a-phrase must trigger deletion of the complement a-phrase if they are identical:- 13
With transitive matrix verbs the notional object can govern Equi; if the matrix is in the traditional “passive” this will be the a-phrase, if the “active” it will be the i-phrase. (19) and (20) respectively illustrate these:
“Tell/send Hata the shepherds to drive in the sheep to the woolshed.”
“Hata told the shepherds to drive the sheep in to the woolshed.”
The topicalised form of the matrix is used in (20) as the untopicalised form: “I tono a Hata i nga hepara ki te whiu mai nga hipi ki te wuruheti.” is not fully acceptable for reasons unrelated to Equi. nga hepara is the notional object of the matrix (the a-phrase in (19); the i-phrase in (20),) and the deleted a-phrase of the complement. No other NPs can govern Equi here. (21) and (22) show that the a-phrase of the traditional “active” cannot do so:
Neg. (a-ph.) Hata perf. tell (i-ph.) the children to help them.
“Hata didn't tell the children to help him.”
Hata perf. tell the children (i-ph.)
“Hata told the children to help him.”
Again forms using a fronting rule in the matrix have been used to avoid confusion of judgments. The correct form for, e.g., (21) is: “Kaore a Hata i tono i nga tamariki ki te awhina i a ia” with the i-phrase governing Equi. That a ki-phrase cannot govern Equi is shown by (23):
past ask I John comp. examine he the doctor
“I asked John to be examined by the doctor.”
That an e-phrase cannot govern Equi is shown in (24):
Incept. tell the children Pani comp. wash they her “Pani told the children to wash her.”
Thus Equi is inconclusive as a test for subject. Its restrictions seem to be semantic rather than syntactic.
Relative clause formation in Maori operates by simple juxtaposition of the relative clause and the head NP of the matrix, with the NP of the, relative clause identical with the head deleted. Relative pronouns are not used, but Williams 23 notes that in some cases one of the particles nei, na, ra, or ai is added following the verb of the relative clause. Relative clause formation provides a very good test for syntactic subject 24: in - 14 many languages only NPs in subject position can be relativised on, and in all languages yet investigated if NPs in non-subject position can be relativised on, then NPs in subject position can be also. 25 Williams records examples of relative clauses formed on a-phrases and i-phrases. 26 However, present speakers find relativising on the i-phrase unacceptable. 27 Relativising on an e-phrase or ki-phrase is ungrammatical.
The following examples illustrate relative clause formation in Maori. Suppose we want to insert (25) in (26) as a restrictive relative clause on te tangata (in answer to the anticipated question: “Which man did John see?”), the required sentence will be (27), which has the a-phrase (te tangata) deleted.
Past hit the man Sam
“Sam hit the man.”
“John saw the man.”
“John saw the man Sam hit.”
If we put (25) in the “active” form (28), relativisation on the i-phrase (te tangata) as in (29) is ungrammatical:
Past hit Sam the man
“Sam hit the man.”
The following three examples show that relativising on the e-phrase is ungrammatical. Example (30) is inserted in (31) as a relative clause to produce the unacceptable (32):
Past tell Tamahae the man comp. go to catch fish
“The man told Tamahae to go fishing.”
perf. go the man to the house
“The man has gone to the house.”
If (30) is put in the “active” form, as in (33), te tangata goes to a-phrase position and relativisation, as in (34) is acceptable.
“The man told Tamahae to go fishing.”- 15
“The man who told Tamahae to go fishing has gone to the house.”
The above tests show that the a-phrase has very strong (syntactic) subject properties. It could be raised under negation, topicalised, and be relativised on. Of the other types of NP, only the i-phrase could be used in any transformations: it could govern Equi.
The semantic subject properties discussed by Keenan 29 are distinct from the syntactic subject properties and the tests for them are of a less precise kind. Keenan claims that the semantic subject properties correlate with syntactic subject properties, but there seems to be good evidence that this is not correct. 30 In Maori the semantic subject seems to be in the e-phrase, and thus Maori provides counter evidence to Keenan's claim. We shall look at only two tests for semantic subject: the “Known Information Property” and “Definiteness”. 31 These are the most applicable to Maori, and can both be applied in a quite objective way.
The “Known Information Property” is described by Keenan:
“The referent(s) of a subject NP are usually known to the speaker and much more likely to be known to both speaker and hearer than the referents of non-subjects.” 32
We can test for this property by noting the position of the discourse topic in sentences within the discourse. The topic NP, if antecedently introduced, will tend to easily trigger anaphoric pronominalisation of a later NP with identical sense.
In Maori, the “Known Information Property” resides in the e-phrase which is readily and commonly pronominalised. The following example is of two sentences which begin a narrative in typical conversational Maori. 33
One Monday, incept. prepare Hata (& others) to shear
hipi. Ka tonoa e ia nga hepara ki te whiu mai i nga hipi
sheep. Incept. tell he the shepherds to drive in the sheep
ki te wuruheti.
to the woolshed.
“One Monday, Hata (& others) prepare for shearing. He tells the shepherds to drive the sheep to the woolshed.” 34
As noted earlier, this goes counter to Keenan's claims. Intuitively, however, it does not seem anomalous. If the syntactic subject of a sentence is the most prominent and noticeable position, then it would seem odd if what was already known had to occupy the most prominent position - 16 within the sentence, and correspondingly to be the least likely to be deleted. 35 The discourse topic, the old information, on the other hand should be less prominent and possibly deletable. Maori seems to follow this pattern, although English does not.
The same pattern occurs in Maori with the “Definiteness” property of semantic subjects. Semantic subjects tend to be more definite in reference than do non-subjects. By way of confirmation Keenan notes:
“In some languages subjects cannot be interpreted, generally. as ‘indefinite’ . . . Examples: Malagasy, Cebuano, Tagalog, Kinyarwanda.” 36
We would expect this property to correlate strongly with the “Known Information Property”, as information already introduced should be more definite than new information. It is so in Maori. It seems difficult to have an indefinite NP in an e-phrase. Conversely, as Chung and Timberlake demonstrate, the distribution of the indefinite article he is restricted to a-phrase with intransitive verbs and a-phrases which are patients of transitive verbs. 37 This is an important point to which we shall return in Section 3 below. Chung and Timberlake's 38 illustrative examples are:
burn a fire at middle of the house.
“A fire burned at the middle of the house.”
kill the man a octopus
“The man killed an octopus.”
kill a man the octopus.
“A man killed the octopus.”
“Thus he sought out incantations.”
The position now is fairly clear. The syntactic subject properties reside in the a-phrase, as is evidenced by the a-phrase's prominent behaviour in transformations. Discourse related semantic subject properties are most clearly to be found in the e-phrase. The particular sentence focus, the new - 17 information, is in the a-phrase. As was expected, i-phrases and ki-phrases do not show significant subjecty behaviour. 39
The Nominative-Accusative Hypothesis
The hypothesis that Maori is an N-A language is the traditional and standard one. On this hypothesis the a-phrase is identified as the syntactic subject, the i-phrase as direct object (patient), the e-phrase as the agent of the passive, and the ki-phrase as indirect object (dative). All translations to date have been on this basis although, because the frequency distribution of “active” and “passive” in Maori is the converse of English, the “passive” is usually translated into the English active.
The problems this hypothesis leads to are now well known and have been documented by Hohepa, Hale, Clark, and Chung and Timberlake. 40 Usually the “active” is treated as basic and the “passive” transformationally derived. The problems this brings led Hohepa 41 to treat both forms as basic, Clark 42 to suggest that this transformation is not really a true passive, and Chung and Timberlake to doubt that it is correct to think of one particular form as basic. 43 They also led to the hypothesis that Maori is in the midst of a “drift to ergativity”. 44
Even allowing for these various explanations there remain two important problems for which no satisfactory account has yet been given under the N-A hypothesis. These are: (a) the “passive imperative” and (b) the irresoluble ambiguity of the pre-nominal particle i.
(a) When a transitive verb occurs in an imperative in Maori the “passive” form of the verb is used:
Chase (pass.) the man bad
“Chase the bad man!”
The imperative transformation (whose justification is a paradigm of linguistic argumentation, both syntactic and semantic) deletes a second person pronoun subject. It is difficult to reproduce this argument in Maori because of the nature of the pronoun system and reflexivisation. The reflexive pronoun is formed by adding ano to the usual non-reflexive form. In first and second persons semantic ambiguity does not occur and thus it seems ano is used only with third person reflexives. However, ano can also have sentence-wide scope and mean “again”. Thus:- 18
Wash (pass.) him *reflexive/again
can be read: “Wash him again!” The semantic judgments needed to reproduce the normal imperative transformation argument for Maori have not, to my knowledge, yet been documented. Nevertheless there seems to be no reason to believe that the argument does not hold for Maori. It was, for example, accepted by Hohepa 45 and had not been seriously questioned.
It is interesting to note that Kendall 46 recorded examples leaving the second person singular pronoun koe in in “passive imperatives” (see also Section 3 below):
Speak pass. locative you
It is notable that koe is in the nominative case, not the agentive (e koe). However, native speakers now reject this form, as they do (43):
Wash (pass.) the dog by you
Example (43) may be acceptable if a strong break occurs after kuri. In this reading, e is the vocative particle, and the NP so marked can occur before or after the remainder of the sentence.
Wahanui! feed (pass) the children
Thus, under the N-A hypothesis the deletion in imperative formation is not of the syntactic subject but of the agent of the passive. The only reasonably plausible explanation 47 that I have heard is that there is a higher causative between the “passive” sentence and the topmost performative. Configurationally this would be: 48- 19
The e-phrase (agent) would then be raised to subject position in S2 and either delete under Equi with the DO (direct object) of S1, or the entire S1-S2 would delete immediately leaving S3 without agent. At first glance this seems reasonable, if, ad hoc, the extra higher predicate not being required in other languages. But in no other constructions in Maori does an e-phrase raise or delete under Equi. It is also counter-theoretical in that the e-phrase, being a demoted subject, should be a chômeur 49 and not take part in such promotional transformations. However, no NPs in Maori exhibit convincing chômeur behaviour and thus this objection is not especially telling.
Finally, one of the arguments for abstract higher predicates is that they can be realised in acceptable surface forms. Although causatives do occur in Maori “passive imperatives”, as is shown by (47):
Pronounce (pass) the word here
“Pronounce this word!”
(whaka is the causative prefix; literally, whakahuatia is built up from cause+bear fruit+pass.)
This requires there to be iterated causatives and these do not occur in Maori. Thus this tentative solution to the “passive imperative” problem does not, overall, look very promising.
(b) The prolific ambiguity of the prenominal particle i may be seen in the following sentence: 50
Perf. die the bird by Hata
“The bird has been killed by Hata.”
mate is intransitive and has the general meaning of “to lack”; in this example it means “to die”. In (48) te manu is an a-phrase. Williams 51 restricts these constructions to the special class of neuter verbs (called “inherently passive” in Hohepa 52; more usually they are called statives) of which some 18 to 22 are usually listed. 53 However, as mate is not usually included in this list, (48) indicates that this agentive use of i-phrase may be more widespread.
Thus, on the traditional N-A hypothesis, examples already given show i to mark either the accusative or agentive case, a critical ambiguity. Williams lists no fewer than 18 different meanings for the prenominal i. These include: accusative, from, beside, in comparison with, by reason of, agentive, at, upon, while, when, having, in possession of, in company with, in the opinion of, as affecting. It is also the standard marker of objects of prepositions. This seems more like a failure to translate than a solution to a translation problem. However, there does seem to be a pattern, - 20 namely, that i marks the ablative (by, with, from), general oblique, and agentive cases. But under the N-A hypothesis we have to include accusative case and with it the serious problem of ambiguity. The only solution has been to proliferate translations.
It has already been noted that the e-phrase does not trigger Equi, nor front under negation or topicalisation. However, it can trigger pronominalisation and reflexivisation:
past cut (pass.) H. himself
“John cut himself.”
Past. see (pass.) John Sam hit (pass) he he
“When John was seen by Sam he was hit by him.” 54
To account for some of these problems Chung and Timberlake have suggested that the passive is a derived stative verb in Maori. Apart from running counter to native speaker intuitions, this suggestion is not supported in Maori usage. An important characteristic of these verbs is that they take i-phrase agentive expressions, as in (51):
Perf. be caught the octopus by the man there (distant).
“That man caught the octopus.”/“The octopus was caught by that man (over there).”
But passives do not take i-phrase agents, as is illustrated by (52):
Perf. be dead be killed Tiger by the pig big
“Tiger was killed by a big pig.”
Thus the N-A hypothesis leads to some perplexing and still unsolved problems for grammatical theory. Most striking among these problems are the frequency and naturalness of the “passive”, the “passive imperative” and the multiple ambiguity of i.
The Ergative-Absolute Hypothesis
The E-A hypothesis being considered here is quite different from that considered by Hohepa, Chung and Timberlake. 55 In these works the traditional translation of prenominal particles was accepted as correct and a tendency toward ergativity posited as an explanation of the frequency of the “passive”, the “passive imperative”, and various other problems already noted. The hypothesis to be examined here is that Maori is an E-A language and is not in any state of change.
As noted in the introduction, E-A languages identify, for grammatical purposes, the actor NP in an intransitive construction and the patient - 21 NP in a transitive construction. These NPs occur in the basic or unmarked case and have all, or most, of the grammatical subject properties. They are said to be in the absolutive case. The actor NP in the transitive construction has a special case marking called ergative. 56 The most useful symptom of ergativity in a language is verb agreement: where the verb is marked for agreement with an NP in the sentence (as is the verb to be in English), in an N-A language it will agree with the actor of both a transitive and an intransitive construction, whereas in an E-A language the agreement will be with the actor of an intransitive and the patient of a transitive construction. Many combinations of these two basic types seem possible. 57
In terms of the most overt symptoms Maori is impossible to evaluate. There is no verb agreement so the most important test is inapplicable. There is a small difference in complexity of case markings, but this cannot serve to decide whether Maori is an E-A or N-A language.
Under the E-A hypothesis the translations of the prenominal particles will not be the same as the traditional ones. The hypothesised translations are: a: marks the syntactic subject, the absolutive case—in intransitive sentences this will be the actor, in transitive sentences the patient; e: marks the ergative case, the actor of a transitive sentence; i: marks the agentive, ablative and general oblique (object of preposition, etc.) cases; ki: keeps its traditional translation as dative (to, for) case marker. The more marked form of the transitive verb (the traditional “passive”) is considered basic and the traditional “active” is transformationally derived. Thus for example:
past wash the children Mary
“Mary washed the children.”
is read as active. nga tamariki is in the absolutive case and Mere is in the ergative case. A transformation—the “anti-passive”—can apply to this to produce (54):
where Mere has been promoted to absolutive and nga tamariki has been demoted to the general oblique case.
Many languages have constructions with dative or oblique patient NPs. Among N-A languages, examples from Latin of particular classes of verbs taking objects in genitive, dative, and ablative cases are familiar. Among languages having E-A constructions oblique patient NPs seem rather rare. However, it does occur in Burushaski 58 and the Lezgian language Udi at least. 59- 22
Thus (54) might be translated: “Mary washed the children.” but the E-A hypothesis predicts, correctly, that such a use of (54) would be unnatural. 60 The function of the anti-passive in Maori is to “feed” other transformations for various syntactic (e.g., Equi), semantic (e.g., negation), and pragmatic (e.g., topicalisation) purposes. This is not an ad hoc solution to a problem in Maori but is typical of E-A languages. 61 Thus, for example, if we wanted to say it wasn't Mary who washed the children, we would apply the anti-passive to (53) and then raise the new absolutive (Mere) under negation, yielding (55):
The feeding function of the anti-passive is clear. Similarly, if we wanted to emphasise that it was Mary who washed the children we would use the anti-passive on (53) to feed topicalisation, yielding:
This feeding function of the anti-passive parallels that of the passive on the N-A hypothesis. 62
The complexity of the simple case markings is consistent with both the N-A and E-A hypotheses. The a marking is simplest, never requiring an additional marker. In case a definite common noun occurs in the e-phrase the definite article te (which, like the indefinite he, substitutes for a when it occurs in an a-phrase) must remain:
pres. chase continuous Hata (abs.) the dog (erg.)
“The dog is chasing Hata.”
Thus the e-phrase is sometimes more complex than the a-phrase. The i-phrase always includes an a or te, etc., after the i and is thus more complex. This might be regarded as evidence that it is transformationally derived from an a-phrase, thus supporting the E-A hypothesis. However, as against such an argument we must note that the ki-phrase has the same requirements and that the indefinite article he 63 does not occur in i-phrase but does occur in a-phrases (absolutives).
The distribution of the indefinite article he does provide support for the E-A hypothesis although its failure to occur in transformationally derived i-phrases denies unanimity to that support. He can occur only in actor NPs of intransitives and patient NPs of transitives, i.e., only in absolutive NPs. Examples (33) to (36) above illustrate this. Under the N-A hypothesis this is anomalous. Under the E-A hypothesis it is quite normal: we expect to find properties shared only by absolutive NPs.
Under the E-A hypothesis the Maori transitive imperative (the traditional “passive imperative”) presents no difficulties. It is usual in E-A languages to delete a second person pronoun from the ergative position. An example from Dyirbal is:- 23
NP marker (abs.) woman (abs.) sit down/marry
“Marry (sit down) the woman!” 64
Similarly in Maori:
Kill the octopus (abs.) here!
“Kill this octopus!”
It is interesting to note that in Dyirbal the second person pronoun can optionally be left in in the imperative, but if it is then it is in the unmarked absolutive case, 65 not in the ergative case. This is exactly the form reported by Kendall in his 1820 grammar. Example (42) (Section 2 above) illustrates this:
“Speak thou!” 66
However, as this form is not found acceptable by present-day Maori speakers this should not be counted as of more than historical interest. That the form of the Maori transitive imperative is normal in E-A languages must count strongly in favour of the E-A hypothesis.
On the E-A hypothesis the ambiguity of the prenominal particle i can also be substantially reduced. However, there are superficial difficulties arising from the transformational derivation of i-phrases that are semantic patients and their governing of Equi. This is easily accounted for under the E-A hypothesis, as a matter of ordering of transformations. Thus Equi must apply to delete the absolutive (notional subject only) of the complement before the anti-passive applies in the matrix, demoting the (notional object, or patient) absolutive a-phrase to i-phrase. 67
The lack of chômeur properties of the i-phrase under the E-A hypothesis is similar to the same lack of chômeur properties of the e-phrase under the N-A hypothesis. That the e-phrase is sometimes deletable is not a problem, as ergatives are known to be deletable in other ergative languages. 68 Thus these facts do not count significantly for either hypothesis. They may, however, support an alternative hypothesis, viz., that Maori is a mixed E-A, N-A language (like many Caucasian languages, e.g., Avar, Karbadian 69) which was, in effect, Hohepa's position in making both “active” and “passive” forms basic. 70 Evaluating this will require considerably more research into subtle semantic features of Maori discourse than is presently available.
To return to the ambiguity of i: under the E-A hypothesis i will mark the ablative and general oblique cases. This fits well with the ergative - 24 accessibility hierarchy 71 (See Conclusion below) in which these cases are very low if not chômeurs. Furthermore, the conflation of these cases is not uncommon in E-A languages. 72
There remains the problem of the agentive use of i-phrases with intransitive verbs (see examples (48), (51) above). I suggest that the most accurate translation of these phrases into English is “at the hands of . . .”, or “by reason . . .”. In this way this use of the i-phrase could be subsumed under the ablative-general oblique case. But even if we list agentive as a separate translation of i the number of translations is reduced to only three.
An important point in favour of the E-A hypothesis is that it does account for the frequency of occurrence and the naturalness of the basic form (the traditional “passive”). The N-A hypothesis predicts that the derived (traditional “active”) form will be the most frequent, whereas in fact it is not fully acceptable.
A related fact that must be noted against the E-A hypothesis is that under it the more marked form of the verb is basic, the less marked form derived. Thus, for example, whangaia goes to whangai, titirotia (or tirohia) to titiro, etc. 73
Finally, the semantic subject properties (definiteness, discourse topic, old information) in Maori tend to reside in the ergative e-phrase, rather than in the absolutive a-phrase. To the extent that this might be considered anomalous 74 it is so for both E-A and N-A hypotheses. Under the N-A hypothesis these properties are in a phrase (the e-phrase) which should be a chômeur 75 under the E-A hypothesis semantic subject properties are located in a phrase (the e-phrase) which is at least basic and unequivocally a term. 76
If the above arguments are correct, Maori should be treated as an E-A language, and the translation of the prenominal particle i revised. The E-A hypothesis accounts for all the anomalies arising from the traditional N-A approach and itself gives rise to only one problem, viz., the morphological complexity of the basic verb.- 25
These arguments also demonstrate the interesting theoretical point that accessibility hierarchies 77 cannot be used to diagnose ergativity in this case at least. There are two accessibility hierarchies, the N-A hierarchy and the E-A hierarchy. The N-A hierarchy is:
Subject > Direct Object > Indirect Object > Oblique Object > Genetive > Object of Comparison.
The E-A hierarchy is:
Absolutive > Ergative > Indirect Object > Oblique Object > Genetive > Object of Comparison.
In brief, if a transformation can refer to an NP at a point in a hierarchy it can also refer to all those points above it. Thus, for example, if in a particular N-A language relative clause formation works on indirect objects it will also work on direct objects and subjects.
In Maori under the N-A hypothesis with the traditional “active” as basic, transformations obeyed the N-A hierarchy; under the E-A hypothesis with the traditional “passive” as basic, transformations obeyed the E-A hierarchy. Thus, with respect to the theory of relational grammar, which of the i-phase and e-phrase was a chômeur was the only test for which hierarchy was appropriate. As neither shows consistent chômeur behaviour, the hierarchies failed to distinguish. The argument thus turned on which hypothesis explained the greater range of data.
I am indebted to Mr Wiremu Tawhai for most of the Maori data in this paper, and for his enthusiasm and encouragement. Professor Ann Borkin, Messrs Russ Tomlin and Matt Dryer and the J.P.S. referee provided much valuable advice and criticism.
The standard notation for ungrammatical examples, (*), is used throughout; “?” is used to mark examples which are found odd by the native speaker, but which might be used in special circumstances, or with special intention.
1 Hohepa 1967.
2 cf. Hale 1968.
3 See e.g. Hale 1968, Hohepa 1969, Lynch 1972, Clark 1973, Chung and Timberlake 1974.
4 See Dixon 1972.
5 See Catford 1975.
6 Lynch 1972.
7 Lectures, L.S.A. Summer Institute, Amherst, Massachusetts, 1974.
12 See e.g. Johnson 1974: Chapter II.
13 See e.g. Grinder and Elgin 1973; the important rule of Equi-NP deletion (Equi) is introduced in Chapter 6.
14 See e.g. Keenan and Comrie 1972, Keenan 1974 and 1975, Johnson 1974.
15 This was not so with the pioneer translators of Maori who laid down the traditional translation pattern; see e.g. Williams 1862, Kendall 1820. As a matter of history E-A languages were not examined as such by linguists until after about 1860; see Catford 1975:2. However, the J.P.S. referee has pointed out that Norris (1846) compared the use of the passive in Maori with what would now be called an ergative construction in Hindi.
16 Especially Keenan 1975.
17 Keenan 1975.
18 See below, and Tomlin 1975.
19 See Hohepa 1967.
20 See also Chung and Timberlake 1974.
21 Question formation, also an optional fronting rule, behaves in the same way as the Negative formation and Topicalisation rules in Maori.
22 Equi deletes the subject of a complement sentence under identity with some NP of the matrix sentence called “the governor” of that deletion. In English the matrix NP governing the deletion can be in subject (A-Equi) or object (B-Equi) position. I have not yet been able to find an A-Equi verb in Maori. Thus all the discussion of transitive matrixes in the text is of B-Equi.
23 Williams 1950:49ff.
24 See Keenan and Comrie 1972.
25 Op. cit.
26 However, i-phrase relative clauses can mean “by”, “on”, “at”, “in”, “with”, “by means of”, “on account of”, “by reason of” . . . see Williams 1950:51.
27 It should be noted that a (unacceptable) relative clause formed on an i-phrase can always be phrased so that the relativisation is on an a-phrase. Note also that we are here considering relativisation only on “notional object” i-phrases and not on any of the other types of i-phrase.
28 As the J.P.S. referee has pointed out, examples such as these should be investigated using the pronominal copy ai.
29 Keenan 1975.
30 See Tomlin 1975.
31 See Keenan 1975.
32 Op. cit.
33 Source: Waititi 1964:16.
34 Traditional grammar tells us that the second of these two sentences could be in the “active” form:
? Ka tono a ia i nga hepara ki te whiu mai nga hipi ki te wuruheti.
Here the old information, the discourse topic, is in the a-phrase. However, native speakers find this form anomalous and would not use it.
35 Of course we should not expect this in the introductory sentence of a discourse. There we would expect either an intransitive or a topicalised construction. I believe this is so in Maori but do not have hard evidence to support that belief.
This argument, however, is somewhat speculative. As Professor Catford has pointed out (personal communication), some languages do allow deletion of discourse topics. It may even be more correct to regard the element of a sentence that is least accessible to transformations as the most highlighted, syntactically: cf. Tomlin 1975.
36 Keenan 1975.
37 This led Chung & Timberlake to suggest that the traditional “passive” is perhaps a derived intransitive verb, to account for the non-occurrence of he in a-phrases which are actors of transitives (in the traditional “active”). Intuitions and usage of native speakers run strongly counter to this suggestion.
38 Chung and Timberlake 1974.
39 Biggs (Biggs 1974) uses essentialness as a test for subject in Maori: the NP which is subject in a sentence is the most essential and thus the least easily deleted. On this test the a-phrase, as we would expect, is the subject. However, essentialness as a general test for subject is questionable; some languages, e.g., Avar (see Catford 1975:33) can freely delete either NP (notional subject or notional object) in a simple transitive construction.
40 See Hohepa 1967 and 1969, Hale 1968, Clark 1972, Chung and Timberlake 1974.
41 Hohepa 1967.
42 Clark 1972.
43 Chung and Timberlake 1974:fn. 3.
44 See Hohepa 1969; the argument there is based on the dubious identification of ergativity with (grammatical) passivity; see Catford 1975:33.
45 Hohepa 1967.
46 Kendall 1820.
47 Suggested in conversation by John Kimball. Hale (1968) suggested that this problem might be solved by an unusual ordering of rules in the transformation cycle, but Clark (1972) has shown this will not work.
48 The notation here is that of Generative Semantics: higher case letters indicate that these words represent abstract semantic notions; “CAUSE” is a general inchoative notion.
49 In relational grammar subject NPs that have been transformationally replaced—e.g., as in passive in English—become chômeurs (Johnson 1974, calls them x-rated) and only take part in a very restricted way in further transformations.
50 From Waititi 1964.
51 Williams 1971.
53 See e.g. Williams 1971:47.
54 This is an odd construction. Much preferred would be the participial construction: I te kitenga atu o Hemi ia a Hone ka patua e ia.
55 Hohepa 1969, Chung and Timberlake 1974.
56 Sometimes it has been called agentive; e.g. Lorimer 1935.
57 See Catford 1975.
58 See Lorimer 1935.
59 Of Udi Catford (1975:10) writes that it:
“. . . is unique among Caucasian languages in having an ergative-transitive construction with the object in an oblique case. This case form, like the South Caucasian one, functions also as a dative, and this point has been emphasized by those who wish to stress the ACCUSATIVELESS-NESS of the ergative construction. Whether one calls it dative or accusative is unimportant; the really interesting fact is that it is OBLIQUE.”
60 It also predicts that a strong contrastive stress would be needed to make (54) acceptable. Although preliminary investigation indicates that this is correct, further research is needed.
61 See, e.g., Dixon 1972.
62 See Chung and Timberlake 1974.
63 See Section 1 above, and Chung and Timberlake 1974.
64 Dixon 1972.
65 This is oversimplified. The pronoun system in Dyirbal works superficially on a N-A system. For an explanation of this phenomenon see Tomlin 1975.
66 Kendall 1820:47.
67 The apparent problem for the N-A hypothesis is solved similarly: only i-phrases will govern Equi and passive on the matrix will only apply after Equi deletes the subject of the complement.
68 E.g. Tongan; see Keenan 1975.
69 See Catford 1975.
70 Hohepa 1967.
71 See Johnson 1974.
72 See Catford 1975, Lorimer 1935. It has been correctly emphasised by the J.P.S. referee that the ambiguity of i is in fact reduced by only one degree. Nevertheless, the importance of eliminating that degree—i.e. accusative—should not go unremarked. Marking agentive, dative and oblique cases similarly is not uncommon, but adding accusative to that group is anomalous in the extreme.
73 A possible explanation of this oddity may be found in the apparent lack of distinction between intransitive verbs and common nouns in Maori (the semantic identity of these classes is generally accepted for English). Thus it may be that the traditional “active” form—the morphologically simpler—is a more “nouny” form; the more complex form would then be the “verby” one and the anti-passive transformation would change it to the “nouny” intransitive form with oblique object. However, much more research on this interesting phenomenon is required.
74 E.g. by Keenan 1975; but cf. Tomlin 1975.
75 The extent to which this is problematic for the N-A hypothesis is not very great. Many semantic properties may follow an NP in its transformational demotion (see Keenan 1975).
76 Cf. Johnson 1974.
77 See Keenan and Comrie 1972, Keenan 1975, Johnson 1974.