Volume 83 1974 > Volume 83, No. 3 > Proto-Polynesian 'ai, by Paul G. Chapin, p 259-307
“[adverb] or verbal directive, . . . sometimes it answers to will or shall . . . It is often connected with a query, and the reason demanded or given” 1.
“. . . marque le futur: . . . marque une conséquence ou une interrogation: . . .” 2
“. . . a post-verbal particle which indicates that the clause in which it occurs is dependent upon some preceding phrase or clause” 3
“particle indicating movement” 4
“particle used to direct one's attention to the place of action rather than to the action itself: . . . translates ‘while’ or ‘as’ when it follows a verb which is preceded by the particle a: . . . a particle used in conjunction with the interrogatives: . . .' 5.
Despite appearances, these varied characterisations are all dictionary entries for a single particle, ai, in a single language, Tahitian. Their diversity is symptomatic of the confusion that ai, a pan-Polynesian form, has caused Polynesian grammarians and lexicographers. My purpose in the following pages is to survey the grammar of ai in 19 Polynesian languages 6 and to make some generalisations which may be taken as hypotheses regarding the syntax of *ai in Proto-Polynesian (PPN).
In brief, my conclusions will be as follows: PPN had a single anaphoric particle *ai and possibly one or more non-anaphoric particles with the same form, which may or may not have been related to it. Anaphoric *ai was not lexical but grammatical: generally speaking, it was a substitute for a noun phrase which was in an oblique case (or an adverbial prepositional phrase, if a distinction is to be drawn) and which was identical to and - 260 coreferential with some other noun phrase in the same sentence or a preceding sentence. The noun phrase repetition could arise either in the ordinary way or as a result of transformational copying rules. The cases where anaphoric *ai did not arise by actual substitution were those cases, as yet poorly understood in general linguistic theory, akin to the use of so-called “lexical” pronouns and pronominal demonstratives in other languages, where the anaphoric reference is not to some specifically identifiable noun phrase but to conceptual elements of the discourse as a whole, or to elements present in the real world context of the discourse and perceived by both speaker and addressee.
A word about logic and methodology is in order. It has sometimes been suggested that syntactic reconstruction is impossible, because nothing as tangible as cognate forms is available for comparison and manipulation. As Ross Clark has rightly pointed out, 7 such a suggestion confounds a basic methodological principle with a particular application of it. The logical basis of comparative reconstruction, which makes it acceptable as a scientifically valid procedure, is a sort of Occam's razor principle that that theory is best which invokes coincidence as an explanation the least often. Thus cognacy is explained by descent from a common form, rather than by chance convergence; just so, common syntactic patterning suggests reconstruction of that pattern in the proto-language.
There is even a sense in which syntax may be a more reliable indicator of genetic affiliation, and syntactic reconstruction a more certain procedure, than vocabulary comparison and reconstruction. This is so to the extent that the syntax of a language is more impervious to influence by neighbouring languages than its vocabulary. Separating loans from true genetic cognates is a notorious problem of traditional comparative reconstruction. In so far as syntactic patterns are not borrowed, syntactic reconstruction avoids this problem. On the other hand, syntactic comparison faces a special difficulty of its own in separating out the language-particular from the universal. One could scarcely postulate a genetic relationship between two languages on the grounds that both had relative clauses, or imperatives without overt subjects, or both were SVO. Such descriptors are either inherent characteristics of any human language or members of a small stock of variant possibilities from which each language “chooses”. The situation is clear in these cases, but less so in others. Affix types, case systems, pronoun paradigms—how reliable is correspondence in such matters as an index of genetic affiliation and a basis for comparative reconstruction? The answer, I think, must be one of degree: the greater the number of possibilities for organisation of a language with respect to an aspect under investigation, the less likely chance becomes as an explanation of commonality.
The organisation of the paper will be as follows. I shall offer first the comparative evidence for the various grammatical/semantic functions I reconstruct for PPN anaphoric *ai—that is, the determinant 8 roles it - 261 assumes (anaphorically) within the clauses in which it appears, such as Locative, Temporal, Instrumental, etc. For each sentence cited in evidence, I shall indicate the nature of the anaphoric reference—the antecedent of ai and the syntactic pattern which effects the anaphora. The summary of this first section will include some discussion of probable cognates of PPN *ai in non-Polynesian Austronesian languages. I shall then survey the non-anaphoric uses of ai in the various languages, and make some remarks about non-anaphoric *ai in PPN. Following that, I shall examine the individual languages in terms of their divergences from PPN in the use of ai. I shall conclude with some comments about the implications of the study for the general theory of syntax and of syntactic change.
ai functions as a Locative determinant of its clause in EUV, HAW, KAP, MAE, MAO, MOR, NIU, NUK, PIL, PUK, RAR, REN, SAM, TAH, TIK, TON, and VAI. The following are examples from each of these languages. In these and all following examples, the antecedent of ai is italicised. The analysis of the syntactic patterning will be given collectively at the end of each set of examples. Abbreviated citations identifying the source of the example are given in parentheses immediately following the example; a key to the abbreviations may be found at the end of the paper. Each example is given in the orthography of its source. Translations are from the source, except those enclosed in double parentheses, which I have provided.
Sentence (1) contains an ordinary anaphoric reference between co-ordinate clauses. Sentences (2) and (14) have undergone Adverb-fronting of the antecedent of ai. Examples (3), (6), (10), (15), and (17) contain restrictive relative clauses, and (13) a non-restrictive relative, of which the antecedent of ai is the head. Sentences (4), (9), and (12) exhibit interrogative pronoun fronting, sentence (4) in an indirect question, with ai referring to the interrogative pronoun. In sentences (5), (7), and (16), the antecedent of ai has been topicalised. Sentence (8) has undergone emphatic right dislocation of the Locative adverbial phrase, with the result that ai refers forward to its antecedent. 11 Sentence (11) contains an - 263 infinitive complement, with apparently ordinary anaphoric reference of ai to the Goal determinant of the verb in the main clause.
ai takes the place of the Temporal determinant in EUV, HAW, KAP, MAO, MOR, NIU, NUK, PUK, RAR, REN, SAM, TAH, TON, and VAI.
Sentences (18), (22), and (27) have undergone adverb fronting of the antecedent of ai. They show an interesting range of possibilities: in (18) an adverbial phrase has been fronted, in (22) a single adverb, and in (27) an adverbial subordinate clause. Examples (19), (20), (24), (25), and (30) all have relative clauses containing ai, with the antecedent of ai as the head of the clause. (21) and (26) have undergone topicalisation of the antecedent of ai. (23) and (31) exhibit ordinary anaphora, in (23) from a main clause antecedent to ai in an infinitive complement, in (31) from one clause to another conjoined to it. In (28) and (29) the antecedent of ai is the interrogative pronoun which has been moved to the front of the clause.
ai functions as the determinant specifying the Goal of motion in EUV, MAO, MOR, NIU, NUK, PUK, REN, SAM, TAH, TIK, TON, VAI, and possibly MQA.
Sentences (32), (36), (37), (40), and (41) display ordinary anaphora in various patterns of clausal relationships. (33), (35), (38), (39), and (42) contain relatives whose heads serve as antecedents for ai. (34) is a simple sentence in which ai (i.e. ei) refers forward to its antecedent, and in (43) the antecedent of ai (ei) has been topicalised.
It is possible but doubtful that MQA uses ai for a Goal anaphor. In the Handy texts the following sentence appears:
It may be that te vahi “the place” is the antecedent of ai, which would then be functioning as Goal determinant of the verb tihe “arrive” in the relative clause on te vahi. However, this is the only example of its kind, and in other sentences where a Goal anaphor is called for, i eia is used:
Thus it may be necessary to find a different analysis of (44).
ai functions as Accusative determinant of its clause in EUV, HAW, MAE, MAO, MOR, NIU, PIL, RAR, REN, SAM, TIK, TON, and VAI, and possibly in KAP, MQA, and NUK. It also has this function in modern PUK, which has come under heavy influence from RAR, but not in earlier stages of PUK.
Examples (46), (49), and (52) exhibit ordinary anaphora between sentences in juxtaposition or conjunction, and (56) and (58) ordinary anaphora into an infinitive complement. (47) is an example of a peculiarly Polynesian construction, most prevalent in HAW but found in many other PN languages as well, which I term the “genitive relative”. The genitive relative is a relative clause whose subject has been raised to become a genitive modifier of the head of the clause. Thus the subject of the relative clause in the underlying structure of (47) is ia “she”, which in the surface structure of (47) has been raised and put into the genitive form kana “her”, as genitive modifier of mea “thing”, the head of the relative clause and the antecedent of ai. (48), (50), (51), (53), and (57) contain standard relative clauses with the antecedent of ai being the head of the relative. Topicalisation has moved the antecedent of ai to the front of (54), and in (55) the antecedent of ai has undergone interrogative pronoun fronting.- 267
In the texts of KAP, MQA, and NUK there are a few examples suggesting the use of ai as an Accusative anaphor in those languages, which because of their relative infrequency and/or their equivocal nature do not establish the point with certainty.
The reference of ai in (59) might be to the “him” of the first clause, in which it would be Accusative; but it might as well be a Locative reference to the house.
A possible analysis of (61) is that dana hai e hai ai is a genitive relative clause on the order of “the deed that he would do”, with ai functioning as the Accusative object of hai “do” and referring anaphorically back to dana hai “his deed”. However, this example and one other are the only potential evidence for Accusative ai in the NUK text, and one would expect a function as frequently exercised as the Accusative to be more broadly represented in a run of text. Also, the English translations of the NUK text tend not to expedite analysis.
PUK, the easternmost Samoic language, has been extensively influenced in the last several decades by the dominant language of its geographical area, RAR, which is Central Eastern. 12 This has resulted in a number of changes in both vocabulary and syntax which may be discerned by contrasting PMT, collected in 1934-35, with Sandra Chung's field notes of 1971. One change that concerns us here is the use of ai as Accusative anaphor. No instance of this has been found in PMT, but a certain class of verbs of Chung's data permits it—namely, those transitive verbs which cannot be used ergatively in colloquial PUK. It seems likely that the newfound possibility of Accusative ai in PUK is due to the influence of RAR, which, as we have seen, permits Accusative ai.
In (62) the Accusative object laatou “they (Pl.)” has been topicalised, and serves as an antecedent of the ai.- 268
ai can function as Instrumental anaphor in KAP, MAO, MOR, NUK, PIL, PUK, RAR, REN, SAM, TAH, and possibly EUV, HAW, MAE, and VAI.
Examples (63) through (68) and (72) contain restrictive relative clauses in which ai holds the place of the Instrumental determinant, and refers anaphorically to the head of the clause. In (69) and (71) the antecedent of ai has been topicalised. Ordinary anaphora between conjoined clauses relates ai and its antecedent in (70).
The data from EUV, HAW, MAE, and VAI which support the postulation of ai as Instrumental anaphor in those languages are equivocal.- 269
The ai of (73) is probably Instrumental. However, EUV is basically an Ergative language, with verbs taking Accusative objects being those verbs of perception and emotion which Clark has labelled “Type B” verbs. 13 Furthermore, the verb taomia is in “passive” form. Thus it may be that the ai is replacing an Agentive determinant, which in the surface form of EUV is marked with the preposition e, rather than an Instrumental, which is marked with aki. The use of ai as Agentive anaphor will be discussed below.
(74) is typical of numerous examples in HAW and many of the other languages studied which are very difficult to sort out in a principled, non-arbitrary way into clearly distinct categories of Instrument, Cause, Means, Force, Indirect Agent, or whatever. In analysing the Polynesian languages, I have found it useful to distinguish Instrumental and Causal as major determinant types. I believe there are good semantic and formal grounds for this. Semantically, the Causal determinant specifies an event or a state of affairs which is logically and perhaps chronologically prior to and in some sense responsible for the action or state designated by the predicate of which it is a determinant, while the Instrumental determinant denotes something involved in the action itself, something not itself acting or being acted upon but rather being used by the actor to effect the action. Other semantic distinctions characterise the split, also, though they may be entailed by the basic contrast just defined. Instrumental determinants are typically concrete, though not always—note examples (68), (72)—while Causal determinants are typically abstract, though not always—see example (92) below. The predicates to which Instrumental determinants attach normally designate an action, while Causal determinants may attach to predicates designating actions or states, though here again the distinction is not always clearcut - is ora ‘survive’ in (65) an action or a state ? Formally, some of the languages (e.g. TON, EUV) use distinct overt prepositions for the two determinants, and the semantic distinction between the two prepositions corresponds to that just outlined.- 270
Assuming these distinguishing criteria to have some linguistic validity, it still isn't clear how they are to be applied in cases like (74). Is the authority by which one performs an action an instrument of the action ? I don't know. Since all of the examples that I have found in HAW which might support the establishment of Instrumental use of ai are of the same problematic nature as (74), it must remain an open question for the present whether ai functions as Instrumental anaphor in HAW.
The ai in (75) is most probably functioning as the Accusative object of a transitive verb bisiŋi “play with”. However, an Instrumental interpretation is possible. A fuller grammatical description of MAE than is presently available would be required to settle the question.
Example (76) is from a song. The use of ei as Instrumental anaphor, with toku fangongo “my coconut-shell” as antecedent, seems fairly clear from the translation. There is a problem, though, in that in VAI and every other Polynesian language, anaphoric ai/ei not preceded by a preposition always appears as a post-verbal clitic, preceding all determinants which follow the verbal complex. In (76) the ei follows the noun phrase toku kili “my skin”. For this reason, and because songs often display special syntactic patterns not part of the grammar of the non-poetic language, (76) cannot be taken as a reliable indication of the use of ai (ei) as Instrumental anaphor in VAI. Unfortunately, no other relevant examples appear in the CV texts.
There is in TON and in SAM a different form, TON 'aki, SAM a'i, which acts as Instrumental anaphor:
The antecedent of 'aki has been topicalised in (77); (78) is an example of ordinary anaphora into an infinitive complement.
Because of the appearance of cognate forms in distinct branches of Polynesian (Tongic and Nuclear Polynesian), PPN *'aki has been reconstructed as an alternate of *ai. 14 I believe, however, that although PPN did have a form *'aki, the use of 'aki as Instrumental anaphor is a specifically TON innovation, which was borrowed by SAM. I believe this on the following grounds:
(a) TON and SAM are the only languages (of the 19 investigated for this study) in which this form can function anaphorically, although other languages have cognates which serve to introduce Instrumental determinants (EUV, NIU). One would expect a PPN pattern to have more contemporary reflexes.
(b) SAM can also use ai as Instrumental anaphor—see (71) above—although TON cannot. This kind of free variation is exactly what one would expect if a form were borrowed into a language that already had another form for the same function.
(c) The use of 'aki to substitute for Instrumental determinants makes - 271 some kind of synchronic sense in TON, in which 'aki is also the Instrumental preposition. In SAM, it is an isolated form.
SAM a'i, while a regular cognate of TON 'aki in terms of the sound correspondences, is also a plausible loan form. The Samoans had extensive contacts with the Tongans in pre-European-contact times, including occupation by them. It is possible that the Samoans were aware of the sound correspondences between the Tongans' language and their own, and when they borrowed a word, that they made the appropriate adjustments in it. Even if this is not the case, the SAM of that period had no /k/, and /?/ would be a phonetically likely replacement, and initial /?/, particularly in grammatical formatives, is notoriously unstable. 15
ai can function as Casual anaphor in EUV, HAW, KAP, MAO, MOR, MQA, NIU, NUK, PIL, PUK, RAR, REN, SAM, TAH, TON and VAI. The definition of this determinant type has been discussed above, in the discussion of example (74).
RAR (89) No te nanaoanga a Pu-kuru mo Atoto i nga puku-a-mata o Apopo, i tupu ai te manga tuatua ra, (HTR29:329) It was from the circumstance of Pu-Kuru and Atoto's scooping out of the eyes of Apopo, arose this saying,
REN (90) “Po ni aa te ngea ai koe kia Sina, manga i maangama?” (FTC139)“Why are you talking about Sina who is in the world of light?”
SAM (91) se ā 'ea le mea 'ua 'ē faanoanoa ai? (SG60; Psalm 42:11) why art thou cast down?
TAH (92) na te varua o Ta'aroa i ha'api'i i te ha'a, i riro ai Tanemata-morari ei tamaiti hi'ora'a maita'i. (AT368) it was the spirit of Ta'aroa that directed the work, and caused Tane-mata-morari to become a good-looking boy.
TON (93) Na'e tō 'a e 'uha ia, pea ta'ofi ai 'a e kātoanga. (TG158) The rain fell, and so the festival was not held. ((lit., was cancelled))
VAI (94) tela te mea e pule ei mata o te palu. (CV173) That is the reason that the palu has large eyes.
In these examples, we see again a wide range of constructions that can effect the presence of anaphoric ai. In (79) and (93) the ai arises by ordinary anaphora, with the antecedent the entire preceding clause. In (80), (83), (89), and (92) Adverb-fronting has moved the antecedent of ai to the beginning of the clause. As the examples show, Adverb-fronting results in anaphoric post-verbal ai whether the nominal component of the fronted determinant is a pronoun (83), a noun phrase (92), or a nominalised clause (89). (92) is interesting in that the fronted noun phrase has a semantic function in both of the co-ordinated clauses that follow it, but only in the second clause is it anaphorically represented by ai. We shall discuss this point further in the general remarks to follow.
The antecedent of ai (ei) in (81) is the head of the genitive relative clause, and in (86), (88), (91), and (93) the head of the ordinary restrictive relative clause in which the ai appears. Note that the heads of these relative clauses are often reflexes of PPN *me'a “thing”, which serve the function in many Polynesian languages of a very generalised noun phrase, used where a lexical noun phrase is required but one essentially without any semantic content of its own. It appears to act as what Chomsky has called the “designated representative” 16 of the category Noun Phrase, and, consistently with Chomsky's theory, appears frequently to be deletable—consider, for example, the following:
HAW (95) Aia haawiia 'ku oukou, mai manao nui i ka oukou e olelo aku ai; no ka mea, e haawiia aku ia oukou i kela manawa, ka oukou mea e olelo aku ai. (Matt. 10:19)- 273
But when they deliver you up, take no thought how or what ye shall speak: for it shall be given you in that same hour what ye shall speak.
It is in this way, through deletion of an antecedent Art + *me'a, that an ai which is actually anaphoric may sometimes appear to have no antecedent, but simply to mark a subordinate clause, as for instance in
The analysis of (96) is that, despite the translation, the portion of the example following the comma is not a direct quotation, but rather a relative clause on te mea, which in the underlying structure of (96) is the accusative object of fai atu ‘ask’, but which has been deleted in the derivation of (96).
Returning to the analysis of the examples of ai as Causal anaphor, (82) and (85) have undergone topicalisation of the antecedent of ai, and in (84), (87), and (90) the antecedent of ai is an interrogative pronoun which has been moved to the front of the clause. (87) is interesting in that all three of the ais in it have the same antecedent.
There is a class of constructions containing ai found in many of the languages which may be called “Purpose clauses”, which I have subsumed under the Causal category. The following examples do not represent every language in which Purpose clauses containing ai are found, but only a sampling to demonstrate the construction.
While the interpretation of these examples is somewhat different from that of the examples given to illustrate Causal ai, there is a clear relationship between the two. The first clause describes a situation which is a necessary and sufficient condition for the situation of the second clause to come about. Thus in (98) the result of Ataranga's raising up Maui was that all the men could look at him. In (97) and (99), the situation of the first clause has not been realised at the time of the utterance, as the clause states a command. However, when it is realised, when the command is carried out, then the second clause situation is expected to follow as a consequence. There is, in other words, a cause-and-effect relationship in - 274 these examples just as there is in the Causal examples, and in all cases the ai appears in the clause stating the effect. Thus it seems justifiable to categorise them together.
These, then—Locative, Temporal, Goal, Accusative, Instrumental, and Causal—are the determinant categories which ai fills anaphorically in a sufficient number of Polynesian languages to justify confidence in reconstructing them as functions of *ai in PPN. Two other important determinant types are possible functions of ai in a smaller number of languages, Dative and Agentive. Possible explanations for the restricted distribution of ai in these categories will be discussed in the summary section of this part of the paper.
ai can function as Dative anaphor in EUV, NIU, NUK, PUK, REN, SAM, TIK, and VAI, and possibly in MAO.
Examples (100), (101), (102), (104), and (107) all derive ai by ordinary anaphora between co-ordinate clauses or separate sentences. (103) and (105) exhibit topicalisation of the antecedent of ai. (106) is an instance of emphatic right dislocation, with the antecedent following the ai. In (100), (102), and (104) I have inserted quotation marks surrounding dots to - 275 indicate that a quotation follows which functions as the Accusative object of the verb of speaking which the ai serves as Dative determinant.
In all the MAO texts examined (around 150 pages), only four or five possible examples of ai being used as a Dative anaphor were found, and all of these were susceptible to other interpretations. Perhaps the most likely example was:
It is possible that the two ais in (108) refer anaphorically to oo koutou maahita “your schoolteachers” and function as Dative determinants of the verbs koorero “talk” and paatai “ask”. However, it is also possible that their presence is due to the status of the infinitive complements in which they appear as purpose clauses, which take ai in MAO as discussed above under the category Causal. SRM does not provide translations, and in the absence of an informant I cannot be sure of the correct interpretation.
Another possible example is:
The presence of “to her” in the translation suggests a Dative interpretation of ai. However, the ai appears in a purpose clause introduced by kia, which typically demand ai in MAO, as in (97) above. Yet a third possibility is that ai is a Locative anaphor in (109), with antecedent te whare.
There is evidence for ai as Agentive anaphor in MAO, RAR, and TAH, and possibly EUV and PIL.
In all three sentences the ai appears in a relative clause whose head is the antecedent of ai. It is evident in the MAO and RAR examples that the relative clause has been passivised, as indicated in the literal translations. - 276 In both cases, the full determinant appearing overtly in the relative clause is the Nominative, and the function of ai (ei) is Agentive. The TAH example is more difficult to analyse, as mae'e does not appear in any reference on TAH that I have been able to find. Tryon, however, lists mā'e “to lift”, 17 and I suggest that what actually precedes the ai in (112) is mā'ea a'e “be lifted up”. If this is so, then a literal translation of the relative clause containing the ai would be “by whom the king, Atea, is lifted up”, and the analysis proceeds as for the other examples. 18
It is possible that the ai of the last sentence refers back to te tai “a man” as antecedent, in which case it is functioning as the Agentive determinant of the passive verb tukua “be lowered”. However, it is at least equally likely that the reference of ai is to the whole set of arrangements made by the man, which is recapitulated by the ai as the Instrument or the Means of the action of lowering the basket. Elsewhere in PIL, we find eia functioning as Agentive anaphor:
The possibility of ai as Agentive anaphor in EUV was discussed above with reference to example (73).
Evidence of anaphoric ai performing various other functions is scattered through the languages examined. For instance, in some languages we find ai taking the place of an adverbial of source:
The ai in (115) is in a relative clause whose head is the antecedent of ai and specifies the source of the motion referred to by the verb of the - 277 relative clause. Adverb-fronting has moved a source adverbial to the front of (116), which there functions as the antecedent of the post-verbal ai.
In HAW there is a set of forms with the prefix pe- which serve as promanner adverbials, and which are frequently fronted. When a sentence has one of these at the front, an anaphoric ai referring to it is found post-verbally:
Other such uses may well exist, whose relative infrequency of occurrence kept them from appearing in the texts examined. It is clear that they all conform to a general pattern, the outlines of which I shall now discuss.
Summary—PPN anaphoric *ai
The most striking fact about the list of determinant categories which ai can fill anaphorically is the absence of the Nominative. Every Polynesian language has this category, but in no Polynesian language do we find ai substituting anaphorically for Nominative determinants. 19 That this is not simply an accidental gap in the data can be seen from examples like (92) which offers something like a minimal syntactic pair.
The noun phrase te varua o Ta'aroa “the spirit of Ta'aroa” is in the underlying structure of (92) a part of both of the clauses which follow it. It is the Nominative determinant of the first clause, as may be seen from the fact that te ha'a “the work” is in the Accusative; and it is the Causal determinant of the second clause, of which Tane-mata-morari is the Nominative. In the surface form of (92), where te varua o Ta'aroa has been moved to the front of the sentence, ai appears as its Causal anaphor, and there is no Nominative anaphor.
The Nominative, of course, is distinct from the other determinant types both formally and semantically. Formally, the Nominative determinant is the unmarked determinant, the determinant without a characteristic - 278 introductory preposition (or equivalently whose characteristic preposition is ø). Semantically, the Nominative is the direct, or the neutral determinant, the determinant without semantic characteristics of its own; the semantic function of the Nominative determinant in a given clause is established by the predicate of that clause. The other determinants may collectively be termed “oblique”, in accordance with a long-standing linguistic tradition, and the rule governing the distribution of ai in every Polynesian language, and of PPN *ai, may be stated in terms of this notion.
I have been using the term “anaphoric” in description of the instances of ai adduced throughout this first section of the paper, and anaphora is indeed crucial to understanding the grammar of ai in these examples. The notion of ai as an anaphoric element is not original with me—Elbert and Pukui credit the first use of the term to Milner, 20 and we can see the basic insight prefigured in Churchward, who calls (TON) ai a “Pronominal Adverb”, 21 and even in Pratt, who writes of (SAM) ai, “This extremely important word performs the function of a relative and always implies an antecedent, . . .” 22—but it seems never to have gained wide acceptance among Polynesian linguists, as witness the quotations from various Tahitian grammars and dictionaries which opened this paper.
I believe there are significant reasons for this state of affairs. In only a minority of cases does ai appear by ordinary anaphora; that is, with an antecedent in an earlier clause or sentence which is understood to be functioning semantically as a determinant of the clause in which the ai appears, and to be represented in that clause by the ai. Examples of this type have appeared among the rest of the examples roughly in proportion to their overall numbers. A larger number of cases involve relative clauses, in which the antecedent of the ai is the head of the relative. That there is an anaphoric aspect to the structure of relative clauses is recognised by traditional grammarians in the use of the term Relative Pronoun, but it is primarily transformational grammarians who have argued that relative clauses are derived from underlying forms in which a full noun phrase identical to and co-referential with the head is present in the relative, and that the derivation of the surface form of the relative clause involves the substitution of a pro-form for a noun phrase under conditions of identity, a process which is essentially the same as ordinary pronominalisation. If this hypothesis is applied to Polynesian, it provides a partial unification of the description of ai.
For another large class of cases, however, the anaphoric nature of ai is not at all clear in traditional terms. These are the various instances of topicalisation, adverb-fronting, interrogative pronoun fronting, right dislocation, and the like which appear throughout the examples. It is largely because of these, I think, that so much confusion has attended the description of ai—understandably so, because within the framework of traditional or of structuralist grammar there is no way to see any unity - 279 among these superficially disparate patterns. Their commonality is not structural, but derivational. The property they share is that they have all been derived by some transformational disarrangement of an underlying clause. To justify this claim fully and explicitly is quite beyond the scope of the present paper, but the basic form the justification takes is simple enough: in its underlying structure, a simplex clause in a Polynesian language contains a predicate followed by a small number of determinants, no more than one determinant of each category. 23 Deviations from this structure, as for instance with some determinant preceding the predicate, are introduced transformationally.
We must be more precise. Transformational derivation alone does not establish the basis for anaphora. A special subclass of transformational rules is involved in the derivation of the examples we are discussing. These are the copying transformations, first discussed by Ross. 24 The effect of a copying transformation, as the name implies, is to create a copy of some constituent at some specified location in the clause or sentence. If the constituent copied is a determinant, then the result of the transformation is to create two identical determinants, and we have again the basic situation for anaphoric substitution for one of them. 25
By way of illustration, consider the derivation of (14)
Before Adverb-fronting applies, the structure is
After applying Adverb-fronting as a copying rule:
Then anaphoric substitution of ai for the second ?ii te farre, on the grounds of identity, results in (14).
There is one remaining class of cases of anaphoric ai which is not, I believe, handled satisfactorily by the mechanisms discussed so far. These are the Purpose clauses, discussed above under the Causal category, such as
According to some currently argued theories of generative grammar (generally those known under the rubric of “generative semantics”), at some level of derivation of (99) a copy of the structure underlying the first clause would be embedded in the second clause, to be replaced at the surface level by the ai. I do not subscribe to such a theory, however. - 280 Rather, I think, the status of the ai in (99) and like examples is close to that of the so-called “lexical” pronouns, pronouns which do not arise by substitution for some fuller noun phrase. Such pronouns are not truly lexical in the sense that their semantic contribution to the sentence is self-contained. Rather they serve to capture some conceived or perceived entity present in the situational awareness of both speaker and listener and thrust it into the stream of the sentence. We do not yet have a fully adequate theory of these, or indeed any initially plausible theory, but I am persuaded that the ai of (99) acts in essentially the same way. It recapitulates the proposition stated by the first clause and asserts its (Causal) relevance to the action of the second clause. It is anaphoric in the sense that it takes its meaning in the clause from its relationship to an antecedent, but it is not a substituent.
In a transformationalist framework, then, one can see that it is meaningful to speak of all the instances of ai adduced so far as anaphoric, and that in doing so one is capturing a linguistically significant generalisation, which of course applies to PPN as well.
A subsidiary generalisation which is worth noting is that PPN *ai could substitute only for third person noun phrases. Only two examples of second person antecedents of ai came to light in all the texts examined (one, REN (54), was cited above; the other is from RAR), and there were no examples of first person.
An important problem in reconstructing the syntax of PPN *ai is the status of the Agentive and Dative determinants. These were certainly a part of PPN, 27 and not Nominative; it is a fair question, therefore, in light of the discussion so far, why the distribution of ai in these categories is more limited than in the others discussed, especially for Agentive, which is represented by only a handful of examples. Perhaps some alternative hypothesis is required to account for this.
Since Dative and Agentive are the determinants which in general are restricted to referring to humans, and furthermore ai cannot substitute for first or second person, which are necessarily human, one attractive possibility would be that PPN *ai could not refer to humans. There were, after all, personal pronouns for this purpose, which are what we in fact find alternating with ai as Dative and Agentive anaphors. In a few of the languages—MAE, MQA, possibly KAP and PIL—there does appear to - 281 be such a restriction (although caution is required in interpreting an omission in textual data as reflecting the grammar of the language, rather than being simply accidental. However, in most of the languages the Accusative, which is the one oblique determinant which tolerates human and non-human noun phrases equally well, seems to allow ai to refer to humans as readily as to non-humans. See for example REN (54), TON (57), PUK (62), and MAO (Footnote 27); TON and MAO are languages which apparently do not allow ai as Dative anaphor. The possibility cannot be ruled out that *ai did not refer to humans in PPN but has come gradually to do so in limited ways in most Polynesian languages of the present; positive support of such a hypothesis, however, would require more convincing evidence than is presently available. It does seem probable, though, that the personal pronouns alternated freely with *ai in pronominalising human noun phrases in PPN.
An alternative hypothesis which bears considering is that the distinction between those determinant functions which PPN *ai could fulfil and those which it could not was not Nominative vs. oblique, as claimed above, but rather between a core set of determinants—Agentive, Nominative, Dative—which specify the basic participants in the action of a clause, and the rest of the determinants which specify the time, place, purpose, etc. of the action. 28 The historical development of contemporary Polynesian languages on this view would have to involve a process of seepage of ai into the Agentive and Dative functions in some languages.
The difficulties with this proposal are primarily theoretical. Why should the Accusative not belong to the core set? Why should ai-seepage not occur at all into the Nominative, only slightly into the Agentive, and to a much greater extent into the Dative? For that matter, why should there be ai-seepage at all? That is, if the kind of distinction just described exists at all, it seems like one which would be quite fundamental, and scarcely subject to historical weakening. It is hard to conceive of principled answers to these questions.
It seems to me that the proposal outlined cannot stand, but that a scaled-down version of it may well be an important part of the story. Ignore for the moment the problem of the Dative (I shall return to it below) and consider just the Agentive. The instances of Agentive ai are fewer by far than for any of the other oblique determinants—indeed they are only a few more than the unexplained apparent examples of Nominative ai discussed in Footnote 19. This is interesting, because the Agentive is clearly more closely related to the Nominative than any other determinant, both semantically and syntactically. Sandra Chung has argued extensively in favour of retaining the notion of “logical subject” in the description of Polynesian languages, even those languages which are primarily Ergative. 29 Her arguments show that in a number of respects, for example clitic pronoun fronting, the Agentive and the Nominative act in concert. - 282 The evidence assembled here suggests that general non-susceptibility to anaphoric replacement by ai may be another property they have in common.
But then how are the instances of Agentive ai that we do find to be accounted for? The answer, I think, is that the Agentive alone among the determinants has a dual personality. It has both a direct function, as logical subject, and an oblique function, as passive agent. Sometimes one function predominates, sometimes the other, and the variation may be by language—it is surely a significant aspect of the Accusative-Ergative split—perhaps by idiolect within a language, and by situation of a particular utterance. ai substitution is possible only when the Agentive is in its oblique function. Strong evidence for this contention is that in every putative instance of Agentive ai discovered, the verb of the clause has been in the passive form. No instances of “Ergative” ai, i.e. Agentive ai in an active clause, have been found.
A third hypothesis to account for the distribution of *ai in PPN relates it to the prepositional system of the language. 30 On this hypothesis, the prepositions *i and *ki and possibly *mai and *aki required *ai as anaphor for their (third person) objects. In so far as these prepositions acquired grammatical functions such as marking Accusative and Dative, the requirement carried over to those determinants as well. Unlike the other two hypotheses just discussed, the assumption here is that *ai did function as Dative anaphor in PPN (though not as Agentive anaphor), and that some Polynesian languages have lost this property.
Because of the extreme polyfunctionality of Polynesian prepositions, and of i in particular, this hypothesis is almost empirically indistinguishable from the oblique determinant hypothesis which is the central thesis of this paper. The one significant advantage I can see to the latter is the possibility of a principled account of the instances of Agentive ai, such as was offered above. It is hard to see how a non-arbitrary account of these exceptional cases could be constructed under the prepositional hypothesis. Another area where further research (with informants) might yield evidence to distinguish the two is in the compound locative prepositions which translate “under”, “over”, etc. In some languages these begin and end with i, e.g. MAO i raro i (te raakau) “under (the tree)”, while in other languages they begin with i but end with Genitive o, e.g. SAM i lalo o (le la'au) “under the tree”. Example (5) shows that at least in some circumstances the determinant introduced by one of these compound prepositions can be replaced by ai:
Since (5) is from MAO, the proponent of the prepositional hypothesis could say that the ai has appeared under the influence of the second i, and that the entire compound preposition has disappeared like all Locative prepositions before ai in MAO (this point will be discussed further below). If ai were to appear in the analogous position in SAM, however, this - 283 would be more difficult for the prepositional hypothesis to explain, while the oblique determinant hypothesis would hold both to be Locative determinants and hence both amenable to substitution by ai. 31
As implied above, the prepositional hypothesis and the oblique determinant hypothesis both make the claim that *ai functioned as Dative anaphor in PPN. This leaves the problem of accounting for the comparatively restricted incidence of Dative ai in present Polynesian. Here the best generalisation available seems to be in terms of the historical development of the prepositional systems of the various languages. This will be discussed in more detail below, but for the moment suffice it to say that some languages have deleted the PPN preposition *ki “to”, which marks Dative and Goal determinants, before ai, some have retained it, and in TON it may or may not appear. If we assume that MAO does not have Dative ai, that is, that the putative examples of it adduced for that language are not valid, then the generalisation is that just those languages which have obligatory *ki before ai retain ai as Dative anaphor. These languages are EUV, NIU, SAM, TIK, and VAI. There are two exceptions, NUK and REN, and one special case, PUK. NUK and REN appear to have Dative ai without ki—see (102) and (104)—and are unexplained. PUK has Dative ai without ki ((103)) which can be explained in terms of the known linguistic history of PUK.
PUK is a Samoic language which has come under heavy influence from RAR, an Eastern language. In earlier PUK, as represented in the Beagle-hole texts (PMT), *ki was regularly retained before ai, as in
Accordingly, consistently with the generalisation, PUK retained Dative ai. RAR is a language which has dropped all prepositions before ai. The result of the contact between the two languages has been that modern PUK now generally drops its prepositions before ai, but it has retained its Dative ai (which RAR does not have). If the association between preposition drop and loss of Dative ai is a causal one, then it may be that within a few generations of speakers PUK will lose its Dative ai. This will be interesting for some future linguist to check on (provided of course, that PUK is not totally supplanted by RAR in the interim).
The Polynesian languages vary quite a bit as to whether ai is introduced by a preposition. The extreme cases are TIK on the one hand, in which every ai must be preceded by a preposition, and on the other hand a number of languages in which ai never has a preposition. A number of - 284 intermediate cases also exist. I will limit my discussion of prepositions to *i and *ki, as instances of the other prepositions occur too infrequently in the texts to make reliable generalisations about them.
The languages categorise quite readily in terms of their treatment of prepositions before ai into Pawley's three major subgroups (excluding the Outlier languages, which reach both extremes of the range of possibilities and form no discernible pattern). The Eastern languages—HAW, MAO, MQA, MOR, RAR, TAH (EAS is a special case and will be discussed separately)—never have any preposition before (anaphoric) ai. The same is true of MAE and PIL, and generally of modern PUK for reasons already mentioned. The Samoic languages—EUV, early PUK, SAM, and VAI—all retain *ki before ai, and in all except SAM *i may optionally be retained before ai in Locative determinants only. In SAM Locative ai stands alone. Also, VAI obligatorily retains i before Accusative ai. 32 The Tongic languages—TON and NIU—retain *ki, obligatorily in NIU but optionally in TON, and retain *i before ai when they jointly form a Locative predicate, but not otherwise. For example, contrast TON (16) with TON (119):
As for the Outliers, MAE and PIL drop all prepositions before ai and TIK retains all, as already mentioned. In KAP i is optional before Locative ai, and in NUK and REN i appears with ai in Locative predicates but not otherwise.
On straightforward comparative grounds, assuming Pawley's sub-grouping, we can reconstruct the following situation in PPN: *ki was retained before *ai everywhere, and *i before Locative *ai either obligatorily, or else obligatorily in Locative predicates and optionally otherwise. *ai in other determinant categories had no preceding preposition. The history of developments into the modern languages has been one of gradual weakening of the requirements for prepositions before ai, to the point of dispensing with them entirely in the Eastern languages. The only innovation contrary to this general trend has been the VAI requirement of i as well as ki before Accusative ai, an innovation clearly motivated by the internal structural dynamics of VAI.
To summarise the summary, then, the evidence suggests the following scenario. PPN oblique determinants in the third person, generally marked by *i, *ki, or Agentive *e, were pronominalised by *ai under normal conditions of identity and coreference. These conditions could arise either in the underlying structure or as a result of the application of copying transformations. The Agentive could function as either an oblique or a direct determinant, and could be pronominalised by *ai only in its oblique function. Where the reference of the determinant was human, personal - 285 pronouns could optionally be used in place of *ai. *ai could also arise as a “lexical” anaphor, serving to recapitulate some concept present in the discourse and inject it into a Purpose clause as a kind of Causal determinant. If the characteristic preposition of the determinant was *ki, it was retained before the pronominal *ai, and the *i of at least some Locative determinants was retained, but other prepositions were deleted under *ai pronominalisation.
In subsequent developments, some of the daughter languages dropped some or all of the prepositions remaining before ai. Those languages which droppped the *ki, or made its retention optional, lost ai as Dative anaphor, except for NUK and REN. Various languages lost ai as anaphor in one or more other determinant categories, for reasons which are largely unexplained. TON adopted its Instrumental preposition 'aki as Instrumental anaphor in place of ai, and at some later time SAM borrowed this usage as an optional alternative to Instrumental ai.
ai is distinctively Polynesian, but it is not uniquely Polynesian. A number of languages of the South-east Solomons appear to have cognate forms. Oroha, for example, has
Arosi has ai, iei, glossed by Fox (AED5) as ‘there’, appearing in compound forms. Capell (LI126) cites Inakona ai “there (also relative, exactly like Samoan ai)”. Rotuman e is also cognate, though possibly borrowed. 33
Bauan Fijian has particles with closely analogous if not identical function to ai which do not, however, agree in form. These are kina for non-human and vuā for human antecedents. Milner (FG49, 68) gives examples showing the use of these particles as Temporal, Causal, Locative and Dative anaphors (my characterisation, not Milner's). To illustrate:
At least two uses can be reconstructed for a PPN form *ai which are not anaphoric. It is possible that the forms serving these functions were distinct from anaphoric *ai, merely homophones of it. However, there is some reason to believe that there may have been some relationship between these forms and anaphoric *ai—this will be discussed in the concluding remarks of this section. In addition, there are a few scattered non-anaphoric uses of ai manifested in individual Polynesian languages of the present, including some use in idioms.- 286
ai, normally but not universally bound to a preceding i, serves as the existential predicate ‘there is, there are’ in EAS, EUV, KAP, MOR, NIU (modern) PUK, REN, SAM, TON, and VAI, and possibly MAO, MQA NUK, and RAR.
That existential (i)ai is functioning as a predicate is seen from examples such as (125), (131), (132), in which it is marked with a tense/aspec marker, (124) where it is followed by the adverb hakaou ‘again’, and (132) where it is followed by anaphoric ai. 35 (124) and (132) also exemplify existential ai without preceding i. i never precedes existential ai in EAS, though it may in TON.
In modern PUK existential i ai always specifies existence AT SOME - 287 LOCATION, not merely existence in the abstract as in, for instance, (130), (131). Earlier PUK did not have existential ai.
(134) is a clear example, taken from the standard MAO-English reference dictionary. The only reason MAO is on the “possibly” list is because no examples of existential ai turned up in a fairly extensive search of texts. I have no reason to doubt that (134) is valid. (136) contains the clear negative existential e deai “there were not”; however, the NUK text contains no clear examples of affirmative existential ai. (135) and (137) share the properties of being difficult to parse comprehensibly and of being unique of their kind in their respective languages.
ai can serve as a particle of narrative continuity, perhaps best glossed “and then”, in EAS, NIU, PUK, RAR, SAM, TAH, TON, and VAI, and possibly TIK.
Then the time came for the food to be put in the oven of the man and woman, and they needed the big kape plant, and they wanted very much to get it. Therefore they asked Loau to go into the house.
As (144) shows, this ai can become a stylistic device which is used repeatedly, perhaps several times in a single sentence.
TIK (146) Poi fakasekeseke, fakasekeseke, fetoki nga tamariki, kae rere rei, tu rei i nga uta, forua. Fakapuaka ai foki na nana. (HTT188) He went and surf-rode, surf-rode and once more the children fell off, but he flew along, stood on the beach and whooped. Again his mother warned him to have respect.
The ai in the second sentence of (146) may possibly be the Continuity particle; at any rate, it is difficult to see what other function to assign to it. However, if there were a Continuity particle ai in TIK, one would expect it to be found more frequently, for instance in some of the clauses of the first sentence of (146).
Other uses, idioms
In HAW, MAO, and NIU evidence has appeared that ai is used as a marker of adverbial subordinate clauses of time, clauses introduced in English by words like “when”, “while”, “until”. This use is distinct from the use as Temporal anaphor already discussed; Temporal anaphor ai may take an adverbial subordinate clause of time as its ANTECEDENT, but when it does so it appears in the clause dominating the subordinate clause. In the use we are discussing here, the ai appears in the subordinate clause itself, and has no apparent antecedent. It is interesting that when ai does appear in the subordinate clause in this fashion, no anaphoric ai appears in the main clause. This complementary distribution suggests a relationship between the two at some level of analysis. Also noteworthy is the fact that in MAO and NIU, although not in HAW, ai marks time clauses only when they have been nominalised. I can offer no explanation for this.
HAW uses ai to mark clauses introduced by me “like, as”, where there is no clear candidate for an antecedent. Semantically there seems to be some connection between this usage and the use of anaphoric ai to refer to fronted forms in pe-, exemplified above by HAW (117).
In MAO we find ai after verbs of motion expressing the idea of indefinite extension of the motion referred to by the verb. This usage is reminiscent of the Continuity particle, but the latter does not appear in MAO.
In SAM ai can appear at the beginning of a clause with the interpretation ‘probably’.
ai has been frozen into a number of idiomatic phrases in various languages: HAW e pono ai “to need”, MAE na 'tafito ai “because”, MAO e ai “according to”, MQA e aha ai “why?”, PUK oti ai loa “that is all (of the story)”, TON talu ai “then, since then”, VAI fua ai “just, only”.
I have included these various other uses of ai not simply for completeness, but because one gets from them tantalising feelings of deep relationships which future research may elucidate. Some of these have already been mentioned, but there is one further point which I should like to make here concerning the relationships between the major non-anaphoric uses of ai presented here and the anaphoric uses. There is a fairly clear connection between existential (i)ai and Locative anaphor ai. Formally, of course, the existential is equivalent to Locative preposition i plus ai. In general the existential ai becomes ei by the same rules in the same languages as the anaphoric ai. We have observed that the Locative preposition i (unlike the other prepositions, apparently) can function predicatively, and it can function this way with nothing following it but anaphoric ai. Furthermore, in some languages (NIU, NUK, REN, TON) this Locative predicate i ai has a special status in terms of its historical preservability. In PUK existential i ai can only assert existence at some location. Thus although existential and anaphoric ai are synchronically distinct forms, as pointed out in Footnote 35, there is a clear continuum of grammatical patterns linking them.
The continuity particle ai is linked in similar fashion to ai as Temporal and Causal anaphor. In all of the examples cited of ai as continuity particle, although ai does not refer to any specific antecedent, it carries a clear sense of sequence and of consequence, of connectedness in time and in natural logic between the clause it adorns and the preceding discourse. - 290 In a sense like that discussed earlier for Purpose clauses, it recapitulates what has gone before and proclaims its relevance as temporal or causal precedent. The difference between the ai of continuity and the Causal or Temporal anaphor ai is just the difference between a precedent and an antecedent. It is a real difference, but not a sharp and distinct one. Compare English then, which can serve as a Temporal anaphor, but also as a marker of temporal sequence (first we wash, then we dry) and as a marker of logical consequence (if butter costs more than eggs, then eggs cost less than butter).
One more form should be mentioned here. A number of Polynesian languages, including EAS, KAP, MOR, NUK, SAM, and VAI, have a form ai as substituent for human noun phrases in interrogative pronouns. The ai is generally preceded by a characteristic preposition, depending on which determinant is being queried, or by the topic preposition *ko. Because of its shape, and because of the fact that it is used as a pronominal substituent, it is a priori possible that it derives from anaphoric ai. However, there are grave difficulties with this view. In the first place, interrogative pronoun ai varies in shape from language to language in ways that anaphoric ai does not. In HAW and MAO it is wai, in TON hai, in RAR ?ai. Furthermore, in RAR anaphoric ai becomes ei everywhere except after words ending in short -a, while interrogative ?ai never does. 36 Anaphoric ai does not appear in the Nominative, while interrogative ai may substitute for a Nominative determinant as readily as any other. Anaphoric ai never appears with the preposition *ko, while interrogative ai regularly does, and anaphoric ai never appears in the Genitive, while interrogative ai does (at least in SAM and EUV). Finally, the weight of authority is against identification of the two. I am not aware of a single Polynesian grammarian or lexicographer who has dealt with the two as different manifestations of a single entity. On the contrary, grammarians like Spencer Churchward are at considerable pains to distinguish them (SG53-55).
SUMMARY BY LANGUAGE
Anaphoric ai in EUV can be Locative (1), Temporal (18), Goal (32), Accusative (46), Causal (79), and Dative (100). It can probably be Instrumental (73), but (73) may be an example of Agentive use. The only non-anaphoric use of ai is Existential (125).
EUV retains ki before ai obligatorily, and i optionally in Locative determinants.
Anaphoric ai: Locative (2), Temporal (19), Accusative (47), Causal (80), Manner (117), and possibly Instrumental (74). Non-anaphoric ai: marker of adverbial subordinate clauses of time (147), of clauses introduced by me “like, as” (150), and in the idiom e pono ai “to need”.
HAW drops all prepositions before ai.- 291
Anaphoric ai: Locative (3), Temporal (20), Instrumental (63), Causal (81), and possibly Accusative (59). Non-anaphoric ai: Existential (126).
KAP optionally retains the Locative preposition i, obligatorily drops all other prepositions before ai.
ei and ai are free variants in KAP.
Anaphoric ai: Locative (4), Accusative (48), and possibly Instrumental (75). Non-anaphoric ai: the idiom na 'tafito ai “because”. The relative impoverishment by this account may well be a reflection of the scantiness of the data, rather than of the language itself. Only nine pages of text were available, with a total of 17 occurrences of ai.
MAE drops all prepositions before ai.
Anaphoric ai: Locative (5), Temporal (21), Goal (33), Accusative (49), Instrumental (64), Causal (82), Agentive (110), Source (115), and possibly Dative (108), (109). Non-anaphoric ai: marker of nominalised adverbial subordinate clauses of time (148), particle after verbs of motion indicating indefinite extension of the motion (151), in the idiom e ai “according to”, and possibly as Existential (134).
MAO drops all prepositions before ai.
Anaphoric ai: Causal (83), and possibly Goal (44) and Accusative (60). Non-anaphoric: in the idiom e aha ai “why?”, and possibly Existential (135). MQA ai has undergone a significant change: it can have only abstract antecedents. This is true of every example found in the text, with the single possible exception of the dubious Goal case (44). This change suffices to explain the limitation of the categories ai can fill; in general it is only the Causal category which can be filled by an abstract determinant (often a clause). The only two putative examples found of Accusative ai were as objects of the verb vivo “think”, which of course takes only abstract Accusative objects.
MQA drops all prepositions before ai.
Anaphoric ai: Locative (6), Temporal (22), Goal (34), Accusative (50), Instrumental (65), and Causal (84). Non-anaphoric ai: Existential (127).
MOR drops all prepositions before ai.
ai becomes ei in MOR after words ending in /i/ or /e/, and sometimes after words ending in /o/ or /u/.
Anaphoric ai: Locative (7), Temporal (23), Goal (35), Accusative (51), Causal (85), Dative (101), and Source (116). Non-anaphoric ai: Existential (128), Continuity particle (139), and marker of nominalised adverbial subordinate clauses of time (149).- 292
NIU retains ki before ai, 37 and Locative i when it is functioning predicatively.
Anaphoric ai: Locative (8), Temporal (24), Goal (36), Instrumental (66), Causal (86), Dative (102), and possibly Accusative (61), Non-anaphoric ai: possibly Existential (136).
NUK retains Locative preposition i before ai when it is functioning as a predicate. Otherwise all prepositions are dropped before ai.
Anaphoric ai: Locative (9), Accusative (52), Instrumental (67), Causal (87), and possibly Agentive (113). Non-anaphoric ai: none.
PIL drops all prepositions before ai.
Anaphoric ai: Locative (10), Temporal (25), Goal (37), Instrumental (68), Causal (88), and Dative (103). Modern PUK, but not earlier PUK, also has Accusative (62). Non-anaphoric ai: Continuity particle (140), and in the idiom oti ai loa “that is all (of the story)”. Modern PUK also has ai in sentences asserting existence at some location; earlier PUK did not have Existential ai.
Early PUK retained ki before ai, and could optionally retain Locative i. Modern PUK generally drops prepositions before ai except in the Locative existential, but ki may be retained before Dative ai; this appears to be regarded as a more colloquial style. ki is dropped before Goal ai.
Anaphoric ai: Locative (11), Temporal (26), Accusative (53), Instrumental (69), Causal (89), and Agentive (111). Non-anaphoric ai: Continuity particle (141), possibly Existential (137).
RAR drops all prepositions before ai.
In contemporary RAR ai becomes ei except after words ending in short /a/. In earlier texts (HTR) this rule holds in general, but there are some exceptions.
Anaphoric ai: Locative (12), Temporal (27), Goal (38), Accusative (54), Instrumental (70), Causal (90), and Dative (104). Non-anaphoric ai: Existential (130).
REN retains the Locative preposition i before ai when it is functioning predicatively. Otherwise prepositions are always dropped before ai.
Anaphoric ai: Locative (13), Temporal (28), Goal (39), Accusative (55), Instrumental (71), Causal (91), and Dative (105). a'i alternates with ai as Instrumental anaphor in SAM: see (78) and accompanying discussion. Non-anaphoric ai: Existential (131), Continuity particle (142), “probably” (152).
SAM retains *ki (SAM 'i) before ai; all other prepositions are dropped.- 293
Anaphoric ai: Locative (14), Temporal (29), Goal (40), Instrumental (72), Causal (92) and Agentive (112). Non-anaphoric ai: Continuity particle (143).
TAH drops all prepositions before ai.
Anaphoric ai: Locative (15), Goal (41), Accusative (56), and Dative (106). Non-anaphoric ai: possibly Continuity particle (146).
TIK retains prepositions before all instances of ai analysed here as anaphoric, though not before the ai analysed as a possible Continuity particle. Locative is the only category in TIK which takes the preposition i; all others take ki. Thus the difference between TIK and other languages in terms of preposition retention is that TIK is the only language in which Locative i is obligatorily retained.
ai becomes ei after the prepositions i and ki, which is to say in all anaphoric cases. It remains ai in examples like (146), where it follows a verb ending in /a/. On a comparative Polynesian basis, it is fair to guess that the raising is phonologically conditioned, i.e. ai goes to ei after non-low vowels. There are not enough data to be sure of this, however.
Anaphoric ai: Locative (16), Temporal (30), Goal (42), Accusative (57), and Causal (93). TON uses 'aki as Instrumental anaphor; see (77) and following discussion. Non-anaphoric ai: Existential (132), Continuity particle (144), and in the idiom talu ai ‘then, since then’.
TON optionally retains ki before ai, and retains Locative *i (TON 'i) before ai when it is functioning as a predicate.
Anaphoric ai: Locative (17), Temporal (31), Goal (43), Accusative (58), Causal (94), Dative (107), and possibly Instrumental (76). None-anaphoric ai: Existential (133), Continuity particle (145), and in the idiom fua ai “just, only”.
VAI retains ki before ai, and optionally Locative i. VAI uses both ki and i as Accusative prepositions, the choice depending on the verb, and both are obligatorily retained before ai.
ai generally becomes ei after non-low vowels, with some exceptions which may be due at least in part to SAM influence.
The reader may have noted that EAS was not represented at all in the discussion of anaphoric ai, and has been pulled to the end of this section, out of its alphabetic order. This is because EAS is the one language studied which has radically restructured the syntax of ai. Accordingly, it deserves separate and special attention.
In a word, ai in EAS is no longer anaphoric. The regular non-anaphoric uses of *ai reconstructed for PPN persist in EAS, and have already been reported. All of the remaining ais in EAS are tokens of one general - 294 functional type; ai is one of four members of a class of post-verbal demonstratives. The other members are era, nei, and ena. I shall describe here the syntax of this class of particles, and of ai within it. From the description it will become clear how the grammar of *ai in PPN has shaped the situation in EAS, even through such an extensive change as EAS has undergone.
The use of the post-verbal demonstratives (pvd) interacts strongly with the choice of tense marker. Of the six tense markers in EAS, most instances of pvd are associated with just one, i “Past perfect” (equivalent to i in other Eastern Polynesian languages). A few instances are found with e “Non-past”. There are no clear data showing pvd with verbs in any other tense, or with verbs unmarked for tense (in infinitive complements).
The converse relationship also holds: verbs in i tense generally take a pvd. The exceptions to this generalisation comprise about 15 percent of the cases, and can for the most part be explained in terms of the historical development of EAS.
A pvd, when it appears, is the final element in the surface order of the verbal complex. Following the verbal stem and preceding the pvd may appear an adverb like hakaou “again”, a particle ró whose function is unclear from the materials available to me, and a directional particle, mai or atu.
For verbs in i tense, it is possible to a considerable extent to predict which pvd will appear. I shall consider era, nei, ena, and ai, in that order.
era almost always appears in an adverbial subordinate clause of time which precedes the clause it modifies.
One example occurs in the texts in which the clause containing era does not actually begin the sentence:
As the translation indicates, the time reference of the era clause is ahead to the clause following it, rather than back to the main clause. Thus (154) is consistent with the generalisation. On the other hand, examples appear in which the when-clause follows the clause it modifies; in this case the pvd is ai rather than era:
nei follows a verb in i tense under the following conditions:
(A) the verb is a verb of motion toward a goal, and the goal is understood to be ‘here’, i.e. the vicinity of the speaker; AND- 295
(B) an adverbial phrase or clause specifying the time or the cause of motion described by the verb precedes the verbal complex.
It might be questioned whether condition B is actually necessary, or whether nei might not appear any time after a verb of motion to the goal “here”. That this is not the case is shown in part by example (154), where the motion is arrival in the vicinity of the speaker, but the pvd is era because the verb is in a when-clause. The validity of condition B is further supported by the following interesting example:
Sentence (158) is a quotation of a woman living in Mangareva. From her point of view nei would be semantically appropriate (that is, condition A is met). The failure of nei to appear can be explained by the fact that (158) does not meet condition B.
Conditions A and B are necessary, not sufficient conditions; even when they are met, the appearance of nei is not guaranteed:
Note that in the absence of nei, ai appears, rather than one of the other pvd or nothing.
It is difficult to generalise about ena, as only three examples containing it appeared in all of the texts examined. One was
EAS (160) “. . . oíra au i-oho-atu-ena i te rongo ki a koe; . . .” (THM384) ((“. . . therefore I went to inform you . . .”))
It is clear that the conditions for era and nei are not met by (160), and the same is true of the other examples discovered. Why ena was selected in these three cases, however, is not so clear. One clue may reside in the fact that all three examples refer the action of the clause in some way to the second person. Englert writes that the suffix ena following a verb in e tense refers the action of the verb to the second person (THM362). He does not make the same claim for verbs in i tense, but these examples would support such a claim. Fuentes calls ena a relative pronoun used when the person or thing spoken of is present (EIL630, 713). However, it is unclear how (160) would be analysed to include a relative pronoun.
The commonest pvd by far, and the one least constrained in its distribution, is ai. In a sense, ai appears whenever a pvd is called for and no other one will do. It is the “elsewhere” case. Thus the syntax of ai is really the syntax of the pvd as a class, minus the relatively well-defined conditions under which era and nei appear and the less well-defined conditions for ena. The presumption is that if era, nei and ena did not - 296 appear in their respective positions, ai would, as for instance in (159). This is important, because the syntax of the class as a whole can be understood in large part in terms of the reconstructible syntax of PPN *ai.
In what follows, I shall discuss first the cases where EAS ai appears precisely where one would expect it on comparative/historical grounds; then cases where ai could reasonably be accounted for by some natural extension or development of the PPN situation, or else by some special assumption (possibly incorrect) about other facets of EAS syntax; then finally cases where, in comparative terms, ai should not appear. Recall that in all cases reference for the time being is only to verbs in i tense. Verbs in other tenses will be discussed subsequently.
ai appears as per comparative expectations when an oblique determinant has been removed transformationally, by deletion or dislocation, from its normal position following the verbal complex. The oblique determinants represented in the EAS data are Locative, Causal, Source, Instrumental, Dative and Temporal.
In this set of examples we see the same range of transformational operations observed above in the comparative survey of the other languages. (161) and (165) have undergone relativisation, (163) and (164) adverb-fronting, (162) interrogative pronoun fronting, and (166) topicalisation.
If it were only for these examples, of course, ai in EAS would require no special treatment. However, there are numerous further examples which do not fit so neatly into the comparative pattern. For instance, ai appears in quite a number of sentences which specify a goal of motion with a prepositional phrase beginning ki “to”. Such phrases are a common referent for anaphoric ai in other Polynesian languages. The problem of EAS is that not a single one of the examples gives clear evidence of transformational disarrangement of the ki-phrase; in every instance the ki- - 297 phrase follows the verbal complex.
Examples of forward reference by anaphoric ai to its antecedent have appeared earlier in the paper (NUK (8), MOR (38), where they have been analysed as resulting from application of the emphatic right dislocation transformation. One might consider invoking this rule to explain (167), but it would not account for (168), since the effect of emphatic right dislocation is to copy the emphasised determinant at the end of its clause. What is more likely, I think, is that in an earlier stage of the language, a right dislocation rule was active in sentences with ki-phrases, with ai remaining as the pronominal remnant, but in the present stage this pattern has metamorphosed into a generalisation to the effect that ai appears post-verbally (after verbs in i-tense) in clauses containing a goal phrase. This generalisation covers all but two of the examples discovered, in which ai is lacking. Residual cases like these are of course not unusual in a process of historical change like the one described.
A difficult class of cases to handle are those in which the ai is apparently the anaphor of the Agentive or the Accusative determinant of a transitive verb. In other Polynesian languages, ai can function in these determinant roles, as shown earlier. In EAS the situation is complicated by the fact that the rules governing the Accusative and Agentive determinants are exceedingly hard to discern, if indeed there are any. I have encountered in the texts sentences with the subject marked with the Agentive e and the object unmarked, sentences with the subject marked with e and the object marked with Accusative i, sentences with both subject and object unmarked, and intransitive sentences with subject marked with e. EAS does not have passivisation as a productive rule, and all of this variation appears entirely arbitrary. Pending a thorough analysis of this problem, it is hard to have a great deal of confidence in assigning the responsibility for the appearance of ai to a transformationally affected Accusative or Agentive determinant. However, it is possible that something of the sort is involved in examples like the following:
In (169) the subject, Ko Cirino “Cyrenius”, is not marked for case. Thus it is possible that at one level of derivation the object te rerorero “taxing, enrolment” is Accusative, and that when it is fronted it gives rise to the ai - 298 following the verb. In (170) the object te Atua “the Lord” is not marked, so it may be that the underlying subject of the relative clause te vī'e “the woman”, which has been deleted by relativisation, is Agentive, and its deletion produces the ai.
Even such speculative suggestions as these have no force in a number of cases where on comparative grounds one would expect no ai at all. Consider for example:
In (171) the object toou ohio “your money” of the relative clause is marked with the Accusative i; in (172) the verb is intransitive. Furthermore, in (172) the presence of the ai appears to be insensitive to whether the subject precedes or follows the verbal complex. None of these facts offers any comfort to a hypothesis relating the appearance of ai to the transformational derangement of oblique determinants. As mentioned above, e-marked subjects do appear with intransitive verbs and with i-marked objects, but an ad hoc postulation of such markings in the underlying forms of (171) and (172) would clearly be circular. Thus we are led to the conclusion that ai in EAS no longer is bound by the constraints on PPN *ai, and in fact, as (172) shows, is no longer anaphoric.
It was mentioned earlier that around 15 percent of the tokens of verbal complexes with i tense contain no pvd at all. It is interesting that all but a few of these are patterns in which, on comparative grounds, anaphoric ai would NOT be expected. Some examples:
The verb noho “stay” of (173) is intransitive, and matou “we (excl. pl.)”, which has been topicalised, is its presumably Nominative subject. Topicalisation of a Nominative should not occasion ai. (174) is to be contrasted with (164); in (164), where the Instrumental phrase has been fronted, ai appears, while in (174) the fronted phrase is the Nominative subject, and there is no ai. (175) shows that the ai of (171), which would not be predicted on comparative grounds, is optional. 38- 299
One example in which ai would be expected to appear, but it does not, is
The fronted Temporal determinant would lead one to expect an ai after the verb haro “weigh (anchor)”.
Comparison of the class of cases in which ai does not appear with the class in which it does reveals the following pattern. Of the 26 cases discovered of verbs in i tense with no pvd, all but three or possibly four contain patterns which would lead one on comparative grounds not to expect ai. Of the nearly 100 cases of post-verbal ai, all but about a dozen appear according to comparative expectations. Assuming that our “comparative expectations” accurately reflect the situation in PPN, we can understand these imbalances as indicating that EAS is in the midst of a syntactic change, a generalisation of the use of ai (or another pvd if appropriate) to all verbs in i tense. The historical source of the usage was the use of ai as an anaphoric replacement for transformationally dislocated oblique determinants. The differentiation between use and non-use to pvd in particular cases can be predicted on this basis with an accuracy of about 85 percent. The number of counter-examples, though, shows that such a rule is not a living part of EAS grammar, but rather the origin of a pattern of ai usage which is gradually spreading beyond the contexts specified by the rule.
So far the discussion has been limited to verbs in i tense. The pvd also appear with verbs in other tenses, though to a much more limited extent. The only tense which is found an appreciable number of times is e, and that appears in only a dozen to 15 examples. A few cases with he and with no tense marker are found, 39 and two cases with subordinating ki. Since the tenses other than i, particularly he, comprise the majority of tense tokens in the texts, this represents a real prejudice in EAS grammar against pvd with those tenses. This restriction as applied to ai is unique to EAS among Polynesian languages, and shows a clear historical process of restriction of applicable context for ai in the development of EAS grammar, concomitant with the generalisation of ai usage already described.
Although this study has been sharply focussed as to topic and as to language family, a number of points have emerged from it which pertain significantly to general questions in the theory of syntax and of syntactic change.
We find first of all some important considerations for the general theory of case. Primary among these is the centrality of the distinction between the oblique, “marked” cases 40 as a class and the direct, unmarked, Nominative. The basic generalisation governing the appearance of - 300 anaphoric ai turns on this distinction, and the linguistic significance of the distinction is thereby demonstrated. That is, it is not the case that a case system consists of six or eight or however many cases all equally opposed to each other; rather, the Nominative is in a class by itself, in opposition to the oblique cases collectively. There may be further hierarchical structuring within the set of oblique cases, for example into “actants” (Agentive, Dative, Accusative) and “determinants” (Locative, Temporal, Instrumental, etc.). The evidence here is not decisive on the latter point, but it is a possibility which case grammarians should keep in mind.
Also of interest in this connection is the special status of the Agentive. Agentive sometimes has a direct function, like the Nominative, and sometimes an oblique function. This variation is intimately connected with the variation between Accusative and Ergative systems: an Ergative system is one in which the Agentive is functioning as a direct case, while in an Accusative system it is an oblique case.
The historical developments observed also provide evidence for a linguistically real distinction between Locative and Temporal, as distinct case types. It is not at all clear a priori that this should be so. There are strong similarities between the two both conceptually and formally. Conceptually, it is simple and in fact normal, at least for relativistically oriented Westerners, to regard spatial and temporal location as being in the same “space”. Formally, most languages, and certainly the Polynesian languages, use the same set of prepositions or equivalent formatives to mark Locative and Temporal cases. None the less, we see a clear distinction between the two being made in the historical development of MAE, PIL, and TIK, in which ai continues to serve as Locative anaphor, but no longer functions as Temporal anaphor. Furthermore, its use in VAI as Temporal anaphor is optional, whereas its use as Locative anaphor is obligatory.
The absence of ai as Temporal anaphor might be an accidental omission in MAE, where available data are scanty. In PIL, TIK, and VAI, however, where more extensive texts are available, we can observe examples in which ai should appear as Temporal anaphor and does not.
All three examples have undergone fronting of a time adverbial, a situation in which ai would be expected to follow the verb. Compare (179) with (31).
A further distinction between the two is observed in their different status with respect to retention of their characteristic prepositions under anaphoric substitution, as reconstructed for PPN. The characteristic preposition of both is *i. The *i is retained (optionally or obligatorily) before *ai as Locative anaphor, but dropped before *ai as Temporal anaphor.- 301
The conditions on dropping the preposition *i before Locative anaphor *ai are interesting in their own right. In several contemporary Polynesian languages — NIU, NUK, REN, and TON — i is retained before Locative ai just in case the Locative is functioning as a predicate, and dropped otherwise (as well as elsewhere before ai). It is as though the i becomes invested with extra responsibility when it serves to distinguish a determinant functioning predicatively, and thereby becomes indispensable, while the same preposition when not part of a predicate is redundant and subject to deletion.
A surprising result of the investigation is the necessity of postulating, at least for Polynesian, more copying transformations than have heretofore been thought necessary. Virtually any transformation which results in some determinant appearing to the left of the predicate in Polynesian languages must be regarded as a copying transformation, to sustain the present account of anaphoric ai. This includes rules such as adverb-fronting, topicalisation, and interrogative pronoun fronting, whose counterparts in other languages have generally been regarded as chopping transformations. 41 From the point of view of universal grammar, this poses a problem. Assuming that rules like the ones mentioned are part of the universal inventory of transformations, are they to be regarded as copying transformations universally? If so, how are we to account for their behaviour in languages where they act like chopping transformations? If not, are we to assume that the categorisation of a given universal transformation as a copying rule or a chopping rule is a language-particular decision? I cannot resolve this issue here, but only raise it.
An interesting sidelight to emerge from the study is the relationship between the anaphoric and non-anaphoric roles of ai. A variety of evidence shows that these functions are grammatically distinct. It is noteworthy, however, that the two non-anaphoric functions for ai found generally in Polynesian are the existential predicate (generally with a preceding i) and the continuity particle. What is striking about this is the analogy to the English situation, where the distinctive element of the existential predicate has the same shape as the Locative anaphor, there, and the continuity particle the same shape as the Temporal anaphor, then. Such convergence can scarcely be accidental; some deep-seated universal relation between these grammatical elements is revealed, which is eminently worthy of further investigation.
The development of modern EAS from PPN offers a valuable case study in syntactic change. The syntax of ai in EAS has been radically restructured: ai is no longer an anaphoric particle, but rather one of a class of four post-verbal demonstratives (pvd), of which the other three also have counterparts in other Eastern Polynesian languages. At the same time, the past still has a profound effect upon the present: the syntax of this whole class of pvd can be defined largely in terms of patterns which in PPN led to the presence of *ai. This represents as much of a change in the grammar - 302 of the other three particles, as compared with the other languages in which they appear, as it does for ai.
The mechanism by which the change in EAS took place is worth considering. The pvd with which ai merged in EAS are represented in PPN, but perhaps by only a single form, *ni “local, now”. 42 By the time of Proto-Nuclear Polynesian the class had two members, *nei “near in time-space” and *laa “distant in time-space”. These alternated with each other as one of several classes of post-verbal particles, also including Directionals (*mai “hither”, *atu “hence”, etc.), Manner particles (*noa “aimlessly”, *lava “definite, intensive”), and General Qualifiers (*foki “also”). The order of these classes was not rigidly fixed, but was commonly Directional, Manner, pvd, General Qualifier. When *ai was generated anaphorically, it was placed into this sequence following the Directional class.
When Proto-Nuclear Polynesian broke up into Proto-Samoic-Outlier and Proto-Eastern Polynesian, the post-verbal particles remained fairly much as before in PSO, but a significant change occurred in PEP: the favoured position of the Manner particle class changed to the beginning of the post-verbal sequence. The result of this was that *ai, still being inserted into the sequence after the Directional class, was now immediately followed by the pvd. Since the occasions on which both *ai and one of the pvd would be used in the same post-verbal sequence would be relatively few, the surface effect of this contiguity would be one of complementary distribution. A hypothesis worth considering is that this purely locational mechanism provided the dynamic of the restructuring in EAS by which ai and the pvd (increased to three in EAS and other Eastern languages by the addition of naa, EAS ena with a locational reference in between the other two, “near distance, vicinity of addressee”) came to constitute a single class. It is difficult to see any other semantic or syntactic connection between *ai as it was used in PPN and the pvd.
Support for this hypothesis is found in the fact that ai and these demonstrative particles have certain interactions in some other Eastern Polynesian languages, although no other language has carried the connection as far as EAS. In HAW, for example, la is the Continuity particle, appearing with great frequency, and it also substitutes for ai as Temporal anaphor in some cases, for instance in the first clause of HAW (47) above. nei may in some cases be used in place of ai as anaphor for Accusative, Locative, and Causal determinants, as in
In (180) the Causal determinant, containing an interrogative pronoun, has been moved to the front of the clause. Normally in such a situation one expects an ai in HAW following the verb, as for example in HAW (80). In (180) we find nei instead. (180) also exemplifies the use of la as Continuity particle, immediately following the aha “what?, why?”.- 303
Scattered comments in various sources on TAH, too diffuse and vague to cite here, lead me to believe that a similar situation may exist in TAH. I know of nothing comparable in any Samoic-Outlier language, where the normal position for the Manner class of post-verbal particles intervenes between ai and the pvd.
If the collocational hypothesis is correct, it shows that fully formal factors, divorced from any semantic considerations, can play a role in effecting syntactic change, a point of considerable importance for the theory of such change.
It has become something of a cliché to call Polynesia the world's greatest laboratory for anthropological and linguistic study. Actually, like most clichés, this is a partial truth, at least as regards linguistics. What we are able to observe in certain parts of Polynesia to a degree probably unequalled elsewhere in the world is spontaneous linguistic change, language change unmediated by any significant external influences for periods of centuries. One predictable result of such isolation is that change is comparatively slow. The PPN that we are gradually reconstructing is rather more similar to the Polynesian languages of the present day than the Romance languages are to their Latin parent, despite the fact that the time depth separating PPN from the present is greater by several centuries at least. 43
Beyond our observation of general conservatism, however, we can seek in the history of Polynesian answers to the qualitative question, what kinds of language changes are likely to occur spontaneously? This study offers us a view of some of these. The search for others will continue to be the theoretically most rewarding aspect of research in comparative Polynesian syntax.
This is a slightly revised version of a paper presented to the First International Conference on Comparative Austronesian Linguistics, Honolulu, Hawaii, January 2-7, 1974. Research and preparation were supported by the National Science Foundation, Grant Number GS-38997.
The paper could not have been completed in the time available without the invaluable research assistance of Sandra Chung and Bill Seiter. Chung did the primary research for NIU and VAI, and Seiter for EUV, KAP, MAE, MQA, MOR, NUK, PIL, PUK, RAR, REN, and TIK.
I am also indebted to J. E. Buse, Vern Carroll, Ross Clark, Ken Hale, Yves Lemaitre, Andrew Pawley, and Ralph White for valuable comments and additional examples. None of these people, assistants or correspondents, is responsible for any faulty interpretations I may have made of the information they have provided me.
Becky Carey did her usual superb and speedy job of typing the manuscript, for which I am very grateful.
Special love and thanks go to my wife and children, who rearranged their lives in some unpleasant ways so I could have the opportunity to write. This paper is dedicated to them.- 304
KEY TO CITATIONS
1 Davies 1851.
2 Jaussen 1898:87.
3 White 1948:18.
4 Tryon 1970:148.
5 Graham 1973:310-11.
6 Easter Island (EAS), East Uvean (EUV), Hawai'ian (HAW), Kapingamarangi (KAP), Mae (MAE), New Zealand Maori (MAO), Marquesan (MQA), Moriori (MOR), Niuean (NIU), Nukuoro (NUK), Pileni (PIL), Pukupukan (PUK), Rarotongan (RAR), Rennellese (REN), Samoan (SAM), Tahitian (TAH), Tikopia (TIK), Tongan (TON), and Vaitupu Ellice (VAI). These are all of the languages for which sufficient textual material was available to carry on the research.
7 Clark 1973:17-18.
8 I adopt here the direct English analogue of Martinet's term, from Martinet 1965, as reported by Milner (1974). It seems to me that it is time for Polynesian grammarians to settle on a common term, and I offer this as the best of the available alternatives. “Case-marked NP” is too cumbersome, and prejudges the issue as to whether prepositional phrases are to be counted as cases. “Actant”, which I myself have previously used, now seems to me to carry too strong a connotation of activity and agentiveness to be sufficiently general, although some considerations developed later in this paper may suggest a need to reintroduce this term to make a necessary distinction (see Footnote 28). “Complement” and modifier” have already been preempted for other uses. “Argument” has too logico-mathematical a flavour. The international parentage of “determinant” should augur well for its international adoptabliity.
9 ei is an alternate of ai found in some languages, sometimes predictably, sometimes not (e.g. in KAP). For languages which have an orthographic ei, the rules governing its appearance will be given under that language in the summary of individual languages. In some languages without orthographic ei, e.g. MAO, unstressed ai is frequently pronounced as ei in rapid speech.
10 The use of i (and other prepositions) with ai (or ei) is found under varying conditions in many Polynesian languages. It appears to be obligatory in TIK. This will be discussed in some detail below.
11 See Chapin 1970 for discussion of this rule and justification of its application in a Polynesian language.
12 See Pawley 1966, 1967 for evidence for the subclassification of Polynesian languages.
13 Clark 1973:88.
14 E.g. by Clark 1973:74.
15 See Clark 1973:14-15.
16 Chomsky 1964:71.
17 Tryon 1970:153.
18 Another problem in the analysis of (112) is the Accusative ia marking the appositive Atea, while the noun phrase to which it is in apposition is in the Nominative. It may be that an Accusative i has erroneously been dropped between ai and te ari'i, and that the verb is actually the active form mā'e a'e “lift up”. Under this analysis the ai would be unexplained.
19 There are a very few scattered apparent counter-examples to this generalisation in the texts examined, of which the following are the most serious: MAO: Koia nei te poaka, i puuhia ai e taku matua. (LLM122) This is the pig which was shot by my father. HAW: A i aloha aku oukou i ka poe i aloha mai ia oukou, heaha la auanei ka uku e loaa mai ai ia oukou? (Matt. 5:46) For if ye love them which love you, what reward have ye? Although I have no satisfactory explanation for these, it is worth pointing out that the antecedent of the ai in the MAO example is the logical object of the verb, and only derivatively (by the Passive) Nominative. This is not very significant, though, since in every other case in MAO and all other languages studied, it is the derived status of the determinant which determines whether ai may be used or not. See Footnote 27 below for an example. The HAW example involves the verb loa'a “accrue” which has rather peculiar properties in any case; see Elbert and Pukui forthcoming.
20 Elbert and Pukui forthcoming; Milner 1966:9, where at the head of the entry for (SAM) ai we find “Anaphoric and relative particle.”
21 Churchward 1953:147.
22 Pratt 1911:21.
23 At least this is true of the majority of Polynesian languages, including all of the Triangle languages. It is reported that some Outlier languages are of SVO form.
24 Ross 1967:426ff.
25 I have argued elsewhere that in the case of anaphoric substitution resulting from a copying transformation, it is always the original constituent which is replaced, never the copy. See Chapin 1970:376-7.
26 Ignoring the problem of proper derivation of the clitic pronoun taaua “we (Incl. Du.)”.
27 Whether Agentive was an underlying category of PPN or a derived category (introduced by passivisation) is still a live issue in Polynesian linguistics, to which I have no contribution to make here. The resolution of the issue is not material to the question at hand, however, since ai can (and in PPN presumably could) substitute for determinants which have become oblique by derivation. The following example illustrates: MAO: Ko Waikohika te tamahine i whakaaro ai a Manukipureora raua ko tana wahine, ko Tukitaua, me moe i a Maniauruahu; (NM294) It was Waikohika who was the daughter which Manukipureora and his wife, Tukitaua, wished to marry Maniauruahu, In the underlying structure of this sentence, te tamahine “the daughter” is the Nominative subject of the verb moe “marry”, whose Accusative object is Maniauruahu. te tamahine has been transformationally raised to the position of Accusative object of the verb of the next higher clause, whakaaro “wish”, and subsequent to the raising has been pronominalised by ai in the usual way.
28 If this were the case, a change of terminology reflecting the distinction might be in order; the term “actants” would be an appropriate one for the core set, leaving “determinants” for the rest.
29 Chung 1972.
30 A version of this hypothesis has been advanced by Clark (1973:73-4 and personal communication).
31 I would not, however, want to be committed to the position that the impossibility of ai in the SAM case would count against the oblique determinant hypothesis. The compound locative construction in SAM and similar languages is partially Genitive, and Genitive phrases do not occasion ai since they are not determinants but proper parts of determinants. Thus the absence of ai in the SAM analogue of (5) would be susceptible to the explanation on the oblique determinant hypothesis that the Genitive construction overrides the Locative. Since the prepositional hypothesis makes a stronger claim than the oblique determinant hypothesis, it is to be expected that more kinds of evidence should count against the former than against the latter.
32 VAI marks some Accusatives with i, some with ki, depending on the verb. Both are retained.
33 I am indebted to Andrew Pawley for the information in this paragraph.
34 Englert's translations in THM are extremely accurate. Since they are in Spanish, however, I have given English renditions here. Hence the double parentheses.
35 (132) also shows conclusively that in present TON, at least, anaphoric ai and Existential ai are distinct forms. A general constraint on anaphoric ai throughout Polynesian is that it can occur no more than once per clause.
36 I am indebted to J. E. Buse for this information.
37 ki in NIU becomes ke except before pronouns (including ai) or proper names.
38 Example (175) is from the text, while the contrasting (171), apparently drawn from it, is adduced by Englert as a grammatical example (to illustrate one use of possessive pronouns such as oou “your”). Thus the ai in (171) is probably Englert's own interpolation. Since Englert knew the language intimately, I think this pair of examples is a valid illustration of optionality of ai.
39 All in Métraux 1940, which is a very poor source from a linguist's point of view.
40 For the purposes of this discussion, I shall shift from “determinant” back to the term “case”, which is more generally familiar in general linguistic theory.
41 A “chopping transformation” is a movement rule which moves a constituent from one position to another without leaving any residual traces in the original position of the constituent. The distinction between copying and chopping transformations was first formulated in this way by Ross 1967.
42 See Pawley 1970:349-50 for this and the other reconstructions in this paragraph.
43 See Pawley and Green 1974.