Volume 85 1976 > Volume 85, No. 3 > A third palatal reflex in Polynesian languages, by Robert Blust, p 339-358
A THIRD PALATAL REFLEX IN POLYNESIAN LANGUAGES
1. The historical development of the Proto-Austronesian palatal obstruents (*(n)s, (n)z/(n)Z, (n)c, (n)j) 1 in Polynesian languages was first studied systematically by Otto Dempwolff. 2 In his earliest (partial) analysis of palatal reflexes Dempwolff noted that *z/Z, *j yielded Samoan, Maori, Marquesan, Easter ø, while *nz/nZ gave rise to Mqa, Eas h and *nj to Sm s, Mao, Mqa, Eas h: 3
A similar pattern of mergers in Fijian and certain other Oceanic languages (as Marshallese) was left unexplained.- 340
Somewhat later the basis of this bifurcate development was made more definite, and at the same time its details were subjected to considerable modification. Based on an examination of the entire palatal series, Dempwolff concluded that PAN *s, *z/Z, *c, *j > s in Futunan and Samoan, but h (X following i, o, u in Tahitian) in other Polynesian languages. 4 He suggested that the corresponding nasal grades (*ns, etc.) merged with this series in Tongan and Uvean, but disappeared in the remaining members of the Polynesian group. To explain the observed correspondences, two Proto-Polynesian phonemes *s and *h were postulated as continu- - 341 ations of Proto-Oceanic (‘Urmelanesisch’) *z, *nz respectively:
The essential elements of this modified viewpoint were retained by Dempwolff in his comprehensive description of the phonological derivation of the Polynesian languages from Proto-Austronesian. 5 Following a suggestion made earlier, 6 it was argued further that Fi s in directly inherited words corresponds to PPN *h and Fi to PPN *s. A common feature of Dempwolff's claims as expressed in three publications extending over a 14-year period is thus that the PAN palatals exhibit two reflexes in PN languages, traceable ultimately to a difference of consonant grades.
2. The first serious attempt to determine the internal subgrouping of the Polynesian languages, and to reconstruct the Proto-Polynesian sound system “from the bottom up” was made by Elbert. 7 Like Dempwolff, Elbert reconstructed PPN *s as the indifferent reflex of the entire simple palatal series, and PPN *h as the corresponding nasal grade reflex. However, for three of his Polynesian Outlier witnesses (Sikaiyana, Ontong Java and Nukuoro) it was necessary to recognise a three-way split: PPN *s > Sik, OJ, Nuk s, h, PPN *h > Si, OJ ø, Nuk s ∽ ø ∽ h. There is no accompanying discussion, but borrowing between contiguous dialects or closely related languages was presumably thought to lie behind this development, since no additional phonemic distinctions were posited to account for the additional correspondences.
3. Grace followed Elbert in his interpretation of palatal reflexes in Polynesian languages. 8 For Rotuman, however, he noted (p. 30) that PAN *s, *Z, *j and seemingly also *ns, *nz, *nZ and *nj > s, but that the palatals sometimes inexplicably yield j (described as ‘apparently a voiceless alveolar or palatal affricate’): PAN *ŋu(cs)u ‘lips’ > Rot nuju ‘mouth’.
4. In an important study of the evolution of the Proto-Austronesian palatals in Oceanic languages generally, Wilhelm Milke pointed to a correlation between Sm s: Fi s, Sm ø: Fi (sample size 72 initial, 57 medial), and Sm s: Gd s, Sm ø: Gd ø (43 initial, 28 medial) which he found by the Chi-square test to be statistically significant at or above the 95% level of confidence. 9 To account for this agreement he proposed that the - 342 palatal reflexes in Fijian be reassigned to their alternative prototypes. Milke further departed from Dempwolff's treatment by writing POC *s, *z, thereby implicitly rejecting an appeal to consonant grade as an explanation of the relevant correspondences. While this decision created no problems for purely internal Oceanic comparison which did not already exist, it effectively left the relationship of the Oceanic correspondences to the PAN palatal series unexplained, as seen in the following graphic representation:
In addition to these overt innovations in method and interpretation, Milke's analysis also differed from that of Dempwolff in a less conspicuous, though perhaps ultimately more significant, way. Several of Dempwolff's influential predecessors and contemporaries 10 had held that the comparative method as developed in Indo-European studies could not be applied to the Austronesian languages, in particular to the languages of Melanesia. Their argument was based on the belief that the phonological correspondences holding between the better-known languages of Indonesia with those of Melanesia (and of the languages of Melanesia with each other) were so complex and numerous that they could not plausibly be explained as divergent developments of a single original system in directly inherited forms. Dempwolff saw the appeal to consonant grades as the definitive answer to such criticism: 11 by permitting consonant grade cross-over, a large number of correspondences could be subsumed under a fairly small number of reconstructed distinctions that were at least partially supported by the nasal clusters independently posited for Proto-Austronesian.
The appeal of Dempwolff's consonant-grade hypothesis depended crucially on a critical limitation: that in directly inherited words no more than two unconditioned reflexes of any given PAN obstruent appear in a single Oceanic language. Indeed, it was an implicit prediction of Dempwolff's theory—and one for which he himself never found falsifying evidence—that no OC language would exceed this limitation. It is thus a matter of some interest that Milke was unable to accommodate all Gedaged reflexes of the PAN palatals under his *s, *z, but noted d as a third (unexplained) reflex in two items. 12 Apparently because of the - 343 small number of forms reported, little importance was attached to this observation, and it passed entirely unnoticed in a subsequent paper. 13
At about the same time, Goodenough, in a preliminary comparison of Nakanai with several other languages of New Britain, 14 reconstructed Proto-Nakanai *d as the outcome of (among other segments) PAN *ns, *nz/nZ, *nc, *nj, and PNk *s and *j as the indifferent reflex of PAN *s, *z/Z, *c and *j. Milke later referred to the “remarkable convergence” of his results with those of Goodenough (viz. POC *s, *z: PNk *s, *j), 15 but failed to mention that Goodenough's Proto-Nakanai shows three developments from the PAN palatals, and that *d corresponds to Gd d in the one cognate available (PNk *dala, Gd dal ‘path, road, way’). Multiple reflexes of the palatals were observed in the language of the Tami islands, and a third palatal reflex was cited in various other languages of the New Guinea area. Despite his general thoroughness, Milke overlooked a similar development in some other languages, as Manam, where POC *s, *z are said to yield Mn s, r, yet such basic non-conforming items as dzala < PAN *Zalan ‘way, path, road’ are not noted. For all reported cases of supernumreary reflexes, borrowing is suggested as the source of the observed discrepancies.
In a final, posthumously published paper that makes use of data from some 87 speech communities, 16 it is clear that Milke had become aware of the seriously problematic nature of the third palatal reflex in certain OC languages. To account for a largely overlapping correspondence that none the less appears to be distinct from the reflexes both of *s and of *z, he posited a new segment *nj in some seven items:
In so doing Milke introduced a curious asymmetry into his theory of Proto-Oceanic, since *j does not form part of his reconstructed system of phonological contrasts. Unfortunately, because there is no discussion of method the reconstruction of *nj is not explicitly justified. A tabulation of Milke's material, however, shows that the crucial evidence for *nj in the data considered is the appearance of Gedaged d, Nggela nd or Yabem ns in conjunction with the normal reflex of *s or *z in most other OC languages, as in:
Where a cognate is known from both languages, Gd d, Ngg nd are mutually corroboratory (2, 6); Yabem, on the other hand, shows ns for Ngg nd and ns for Gd d, Ngg nd in items 3 and 6, but s for Gd d in items 1 and 4. Despite this vacillation *nj is apparently reconstructed entirely on the testimony of Yabem in item 7.
Milke evidently believed that *nj represents the “true nasal grade” of *s, 18 an interpretation which clearly leaves the derivation of POC *s, *z from the PAN palatals unexplained. Grace, on the other hand, writes Milke's *s as *s and *z as *ns, thereby leaving *nj unexplained. 19 Since none of the seven items reconstructed with *nj derives from a known PAN etymon, Grace's decision would appear to represent the safer course: it can still be assumed that *nj represents a segment which occurred exclusively in loanwords borrowed by speakers of Proto-Oceanic, or in certain POC lexical innovations. Closer inspection of the lexical materials available for OC languages, however, reveals that even this interpretation is untenable.
Thus, by Milke's implicit criteria, the nd of Ngg ndoko (PAN *cekel) ‘choke, strangle’, Ngg nduli ‘sucker of banana’ (PAN *suliq ‘plant shoot, sucker’), Ngg gundi (PAN *gusgus) ‘rub’, 20 Ngg lundi ‘slough the skin’ (PAN *luslus ‘slip off, slip away’), Ngg pondi (PAN *pespes) ‘squeeze’, and possibly Ngg ndulu ‘go down, set, of the sun’ (PAN *surup ‘penetrate, enter’) must point to POC *nj. The reconstruction of a POC segment distinct from Milke's *s, *z in one of these forms is corroborated by Gedaged: dilu-n (met.) ‘sprout, shoot, sucker; offspring, progeny’. Through the evidence of at least Gedaged, Nggela and possibly Yabem, then, we are confronted with a third reflex of the PAN palatal stops in some OC languages. 21 While this fact in itself is surprising, the phonological history of Nggela, Gedaged and Yabem is generally less well known than that of any of the languages described by Dempwolff in the Vergleichende Lautlehre. For this reason it would be considerably more surprising to discover a third palatal reflex in any of the languages whose external histories Dempwolff studied in greater detail.- 345
5. The reflexes of the PAN palatals in Fijian and Samoan were subjected to a critical re-examination by Milner, who sought to account for the relatively large number of palatal-derived correspondences in these languages (s:s, :s, s:ø, :ø) as a product of borrowing. 22 A number of previously unnoticed doublet forms (one showing the oral grade, the other the nasal grade reflex of *s) were identified, but no more than two palatal reflexes were found for any given language.
6. Although a few PPN lexical reconstructions were advanced by Elbert, 23 the first large-scale attempt at reconstruction of the Proto-Polynesian lexicon was undertaken by Walsh and Biggs. 24 Like Dempwolff and Elbert before them, 25 Walsh and Biggs posit PPN *s, *h as reflexes of the PAN palatal stops; these reconstructed segments correspond to Milke's POC *s, *z (Grace's *s, *ns). One comparison is especially noteworthy. To account for the agreement of Tagalog ŋôsoq ‘upper lip’, Sa'a ŋudu ‘lip’, Fi ŋusu ‘mouth’ Dempwolff posited PAN *ŋu(cs)u ‘lips’. 26 He cited no Polynesian reflexes. Walsh and Biggs, on the other hand, reconstruct PPN *ŋutu ‘beak, orifice, mouth’, and note Fi ŋusu ‘mouth’ as a probable external cognate. The unelaborated implication is that PPN *ŋutu derives from PAN *ŋu(cs)u by some secondary change. 27 The relevance of this form to the problem at hand will soon become apparent.
7. Pawley follows Elbert, and later follows Walsh and Biggs in recognising just two PPN reflexes of the PAN palatals: PPN *s and *h. 28 The imperfect correspondence of these reconstructed segments with Ngg s, h is noted, 29 but the third Nggela reflex of the palatals (nd) is not mentioned.
8. Basing himself largely on a comparison of the Bauan Fijian dictionary of Capell 30 with the expanded PPN lexicon of Biggs, Walsh and Waqa, Hockett has reconstructed some 555 roots for the immediate common ancestor of Fijian and the Polynesian languages (“proto-Fijian-Polynesian”). 31 In addition to the well-known Fi-PPN correspondences s:s (< POC *s), :s, s:h, :h (< POC *ns), t:t (< *t) and d:t (< *nt)' Hockett recognises a correspondence s:t in two comparisons (Fi ŋusu ‘mouth’: PPN *ŋutu ‘beak, orifice, mouth’; Fi usa ‘carry cargo by boat’: PPN *uta ‘cargo, freight (of canoe)’), 32 and a correspondence :t in two - 346 others (Fi bua ‘valley’: PPN *puta ‘hole’, Fi vue ‘swelling’ ∽ bue(kovu) ‘with abdomen distended by hookworm’: Proto-Eastern Polynesian *puutee ‘bag, sack’). He points out that the sequence *-us- followed by any vowel other than *i is—with one exception to be noted—attested only in PPN reconstructions that lack a Bauan cognate (i.e. only in possible PPN innovations):
Based on this observation it is proposed that PFP *ns > Fi s, PPN *t following *u and preceding any vowel other than *i, but otherwise yielded Fi s, PPN *s. In a reversion to Dempwolff's treatment, PFP *s is posited to account for the correspondence Fi , PPN *h. That is:
In sum, then, apart from the marginal cases of Sikaiyana, Ongtong Java and Nukuoro, where the appearance of three palatal reflexes seems to be readily explainable by an appeal to dialect borrowing, the consensus of scholarly opinion spanning nearly five decades in one of the most intensively studied branches of the Austronesian family, is that the Polynesian languages exhibit just two reflexes of the PAN palatal stops.
9. Of all studies done to date, Hockett's comes nearest to the discovery of a third palatal reflex in Polynesian languages. The reasons seem clear: direct comparison of the PN languages with Proto-Austronesian as reconstructed by Dempwolff provides few cognate identifications that suggest a third development of the palatals; purely internal Polynesian comparison provides none, since the third palatal reflex fell together with PAN *t, *T as PPN *t. The comparison of Fijian (where all palatal stops fell together as s or ) and the Polynesian languages, with the relatively large number of cognate identifications that this makes possible, thus provides ideal conditions for the recognition of a previously unnoticed feature of the phonological history of the Fijian-Polynesian group.
The derivation of PPN *t from an earlier palatal in the first two comparisons suggested by Hockett seems undeniable. Thus, in addition to the comparison on which Dempwolff based *ŋu(cs)u ‘lips’ we find such western Austronesian forms as Saaroa ŋuso: ‘lips, mouth’, Yami ŋusuq ‘mouth’, - 347 and such additional OC forms as
(1) POC *ŋusu(q)
Lou ŋusu-, Leipon ŋucu-, Sori usu- ‘lips’, Mussau ŋusu, Takia ŋudu ‘nose, snout, beak’, Motu udu ‘mouth; nose, beak’ (alternatively derivable from PAN *ujuŋ ‘nose’), Mendak ŋusu-, Roviana ŋuzu-na, 'Āre 'Āre nusu-na ‘mouth’, Mota ŋusiu ∽ ŋusui ‘snout; a man's two lips projecting together’, Rotuman nuju ‘mouth, beak, spout’, Fi ŋusu ‘mouth’, PPN *ŋutu ‘beak, orifice, mouth’.
Similarly, although western AN cognates of the second item are unknown, 33 widespread OC forms point clearly to POC *Rusan ‘cargo, load a canoe’:
Seimat uxan-i, Nauna us ‘to load, as cargo in a boat’, Puluwat wôtán ‘cargo, load; be loaded’, Dobuan qe-usana 34 (*R > ø unexpl.) ‘load’, Motu udauda (*R > ø unexpl.) ‘load (of pots in lagatoi)’, 35 Bogotu lu-luja ‘cargo’, Nggela lunda ‘load a canoe or ship with cargo; cargo’, Sa'a luda ‘carry cargo, load a canoe’, Fi usa ‘carry a cargo by boat’, PPN *uta ‘cargo, freight (of canoe)’.
10. Hockett's conditions for the development PFP *ns > PPN *t are fairly unusual. Moreover, Walsh and Biggs list only three instances of the PPN sequence *-usi (viz. *pusi ‘blow’, *tusi ‘point to, indicate’, *unu(si) ‘withdraw’) and two instances of *-us- followed by a vowel other than *i (*(su)sulu ‘illuminate’, *tusa ‘be equal, divide, share’). As the first of the latter two items is evidently a partially reduplicated form of PAN *suluq ‘torch’, Hockett's claim can be maintained only if the proposed change *ns > t took place before reduplication. Although there is no evidence requiring the assignment of reduplication in this root to a period antedating the emergence of Proto-Polynesian, neither is there evidence requiring the assignment of the change *ns > t to a period antedating the emergence of Proto-Polynesian. The appearance of *-s- in PPN *(su)sulu ‘illuminate’ thus gives reason to question Hockett's proposed phonological conditions. Moreover, Hockett himself cites Nukuoro usu (Fi uu) ‘nose’ as a prima facie violation of the *ns > t change. He attempts to account for this fact by positing PPN doublets *isu, *utu ‘nose’, “the second surviving in Nukuoro with its consonant reshaped under the influence of the first”. 36 But if both forms were found in Proto-Polynesian, and the medial consonant of Nukuoro usu results from contamination, it is surprising that contamination (in either direction) did not occur in any other PN language, or for that matter in Proto-Polynesian itself.
11. Given the phonetically unmotivated and phonologically unusual nature of his proposed conditions, together with the existence of at least - 348 one prima facie counterexample, one is tempted to regard Hockett's explanation as a skilful contrivance made possible by fortuitous agreements in a limited sample. Closer inspection, in fact, reveals other equally plausible Fijian-PN comparisons which involve the correspondence s:t or :t in an environment other than that stated by Hockett:
Given the imperfect semantic match, it is possible to quibble with cp. (3). Cognate forms meaning ‘yardarm’ or ‘yardarm and boom of the sail’ are widespread in the languages of the Admiralty Islands, however, and these clearly agree with Fijian in pointing to an original palatal: 38
12. If the external comparison of the PN languages is extended to include languages other than Fijian, the number of plausible comparisons involving the correspondence of /t/ in a PN language to a regular reflex of the PAN palatals in a non-PN language mounts rapidly. Limited searching already suggests the following:
Gedaged sol ‘become mixed or compounded’, Sm tolo ‘of certain foods, mix, stir with a stick’.
One of the most striking examples of PPN *t as the reflex of a palatal involves a well-known but highly contentious etymology first proposed by Dempwolff. 42 To account for the agreement of Tagalog kátig, Javanese, Malay katir ‘outrigger boom’, Toba Batak hatir ‘pinion’ with Tongan kiato, Futunan kiqato, Samoan qiato ‘outrigger boom’ Dempwolff posited PAN *katiR. However, the Polynesian forms could be compared with those in Indonesia only on the assumption that they derive from a suffixed form of the root (*katiR-en) with metathesis both of the vowels and of the medial and final root consonants: *katiR-en > kitaR-en > kiRat-en (> kiato by regular loss of final consonants and *R, and *e > o). Although a suffix *-en can be reconstructed for Proto-Austronesian, 43 its meaning cannot easily be reconciled with the assumption that it was earlier suffixed to the root *katiR.
But there is a more serious objection to Dempwolff's etymology. In all PN languages with the possible exception of Futunan, PAN *R has invariably disappeared. Futunan afaqaa 44 ‘break forth violently, as a storm’ < *SabaRat ‘north-west monsoon’ next to e.g. Fu fai < *paRi ‘stingray’ suggests that Futunan may occasionally reflect *R as q. The appearance of -q- in Fu kiqato thus seemed to lend some support to the claim that the original medial and final consonants had metathesised in the PN forms of this word. Milke compared related words in some 20 OC languages, 45 many of which normally reflect *R as a consonant, but found no support for *R in this root; he accordingly reconstructed POC *kianto ‘outrigger boom’, and rejected Dempwolff's etymology on the grounds that an implicit prediction of this claim is falsified by the New Guinea Austronesian evidence.
Curiously enough, Milke himself overlooked a recurrent discrepancy between his reconstruction and the evidence he cites in support of it. Thus, of the 20 languages for which reflexes of *kianto are quoted, we have sufficient information to determine the regular development of *nt, *s and *ns for at least five (Manam, Lou, Pak, Mouk, Gilbertese) that are diagnostic for the dental/palatal distinction. These languages appear to fall into three first-order subgroups of Oceanic, as follows: (1) Manam, (2) languages of the Admiralty Islands, (3) Gilbertese. Members of all three groups point unambiguously to an earlier palatal as the source of the medial consonant in the word for ‘outrigger boom’: 46- 350
Manam qiadzo, Nali, Lou kias, Loniu, Ahus kiec, Lindrou kies, Seimat ayas, Wuvulu-Aua ato 47 ‘outrigger boom’, Gilbertese kiaro ‘stick uniting the outrigger to a canoe’, PPN *kiato ‘outrigger boom’
next to Manam dzala < *Zalan ‘path, way, road’, but udi < *pun(tT)i ‘banana’, Nali say, bun, Lou sal, mun, Loniu can, pu, Ahus cal, buh, Lindrou san, bur, Seimat sal, hut < *bu(n)tun ‘a shore tree: Barringtonia asiatica’, 48 Wuvulu-Aua tala, puqu, Gilbertese rama < *(cs)a(R)man ‘outrigger float’, ati < *qa(tT)u(nN) ‘a fish: the bonito’.
In addition, Friederici notes Fijian (Suva and Levuka) nai kaso ‘Auslegerstange’, 49 Hornell cites kaso ‘accessory booms’ 50 and Capell lists i kaso ‘cross beam fastening together the outrigger and the hull of a canoe’, 51 a form which is irregular, but almost certainly cognate with reflexes of PPN *kiato ‘outrigger boom’. 52
13. Since only cp. (1) (*ŋu(cs)u) has known cognates in non-OC languages, it might be objected that the phrase “palatal reflex” is inappropriate in the present context. In all forms that have a known PAN prototype, however, POC *s/ns is derived from one or another of Dempwolff's palatal stops. It thus seems justifiable to regard the members of similar purely internal OC comparisons as “palatal reflexes”. Moreover, more extensive searching reveals further examples of /t/ in Polynesian languages corresponding to a palatal reflex in one or more non-OC languages: 53- 351
14. Alternatively, we might appeal to chance, borrowing or derivation from a doublet with *t/T to account for cps. (1)—(14). The importance of chance as an alternative explanation is illustrated most dramatically for certain arbitrary, or demonstrably untenable comparisons which show considerable initial plausibility, as
Comparisons (a)—(d) seem to offer a strong case for cognation. The meanings of the forms compared are similar, and the phonological correspondences are regular under our present assumptions. Nevertheless, the claim that the PN members of these comparisons show /t/ from an earlier palatal must be rejected or seriously questioned in each case: (a) next to Sa tahe we find ahe ‘to flow’ which, along with Arosi ahe ‘to flow, as a current, be in flood’, Lau afe ‘current, wave, tide; be in flood; freshet’ Fi dave ‘to flow, of liquids in a small stream’ would seem to provide a superior comparison (< earlier *tafe *ntafe), (b) once the full semantic range of Hawaiian kau (‘place, put, hang, suspend, gird on, set, settle, perch, rest . . .’) is considered, the comparability of meaning in these items is seen to be no more compelling than that between Hwn kau and e.g. Fijian tau ‘a verb with the general meaning of something falling quietly - 352 into its proper place’, (c) next to Tongan tā-tatau and cognate forms in other PN languages, we find To hau ‘instrument used in tattooing’, which offers a superior comparison, (d) an alternative comparison of the Samoan form with Tagalog tagós ‘reaching through whole width, thickness or length; reaching as far as the other side or end’ is possible. 54 In short, any one of cps. (1)-(14) could be a product of chance, but the probability is extremely small that this is the correct explanation in all cases. 55
Borrowing would seem to have little explanatory value in accounting for cps. (1)-(14), since no language that is likely to have served as a source of massive borrowing in Proto-Polynesian shows /t/ as a reflex of the PAN palatal series. It is none the less probable that early-Fijian // would have been borrowed by PPN speakers as /t/. 56 While this suggestion is perfectly plausible in itself, it is difficult to reconcile with two observations: 1) since there is no reason why a word could not be borrowed into a language in which the inherited cognate was also found, we would expect doublet forms for items (1)-(14) in at least some PN languages, but to date none has been detected, 2) in various cases (as 1, 2, 3, 10) the attested Fijian form contains /s/, not //. 57 Since PPN *s was presumably a sibilant, early-Fijian /s/ would almost certainly be borrowed as PPN *s, not *t.
Another potential explanation of the development in question is that some or all of cps. (1)-(14) derive from doublet forms with *t. Evidence for such a claim, however, is known in only one case. Thus, the comparison Bunun ŋutus ‘snout, nose’, Sa'a ŋudu ‘lip’ might be taken to justify a PAN root *ŋu(ŋ)(CtT)u(sS). But Sa'a d reflects a palatal in some other roots (luda < POC *Rusan ‘load; cargo’), and hence need not point to *t. In conjunction with the other material cited in cp. (1) it seems likely that this form too derives from earlier *ŋusu.
15. As seen in the preceding sections, there appears to be no way to avoid the conclusion that the PAN palatal stops show three PPN reflexes (*s, *h, *t). These developments do not correspond to reconstructed PAN distinctions, and cannot plausibly be explained as products of chance, borrowing or derivation from original doublets. With respect to the palatals, then, we have returned to the state of affairs for which Dempwolff proposed the OG/NG distinction as a remedy: there simply is no obvious explanation for these facts that is compatible with present reconstructional - 353 hypotheses at the Proto-Austronesian level.
Despite our inability to account for the three-way split of the PAN palatals in Polynesian languages, we might none the less hope to derive the third PPN palatal reflex (*t) from a POC segment or segment sequence that is independently supported by reflexes in other first-order OC sub-groups. The obvious candidate for such a prototype would be Milke's *nj. As noted earlier, Milke reconstructed *nj to explain the conjunction of Gd d, Yb ns, Ngg nd with the regular reflex of a palatal in other OC languages. 58 It would not be unreasonable to expect the third PPN palatal reflex to agree with this correspondence. But this expectation is not borne out by observation. Thus, if PPN *s is taken to reflect POC *s (PPN *samu: POC *samu(k) ‘eat scraps’, *sua:sua(ŋ) ‘turn over, raise up, dig roots, hoe’) and PPN *h is taken to reflect POC *ns (*hAkau: *nsakaRu ‘coral reef’, *hake:nsake ‘up, upwards’), there would seem to be no alternative but to derive the third palatal reflex from *nj. If for the sake of argument we posit POC *Runja (Ngg lunda) ‘load, cargo’, *kianjo (Gd aiad) ‘outrigger boom’, and possibly *ŋunju (Takia ŋudu- 59 ‘mouth; nose, snout, beak’, we are indeed able to point to instances of PPN *t (*uta ‘load, cargo’, *kiato ‘outrigger boom’, *eng;utu ‘nose, snout, beak’) from earlier *nj. In other examples, however, *nj> s, as with *lanje> *lase ‘coral limestone’ and *njuli(q)> *suli ‘shoot of a plant’. Not only is the third palatal reflex in PN languages therefore inexplicable from our theory of Proto-Austronesian, it is also inexplicable from Milke's theory of Proto-Oceanic. 60
In summary, cps. (1)—(14) present evidence for a previously unnoticed palatal reflex in PN languages which fell together with POC *t, *nt as PPN *t. Since a different pattern of merger (with the reflex of POC *s) is observed in Fijian, it follows that Proto-Fijian-Polynesian had a distinct consonant phoneme which gave rise to Fi s, PN t. Phonetically, I assume that this consonant was similar to the partially corresponding Rotuman /j/ (i.e. [ts]); 61
phonemically we can write *c. Hence the following PFP - 354 reconstructions:
16. It is often stated, though seldom in print, 62 that reconstruction on a given level is inherently unreliable if undertaken before systematic reconstruction has first been carried out on lower levels. The basis for this belief is not at all clear, but seems to be founded in the inductivist aversion to “inferential leaps”, that is, in the requirement for a demonstrable catenation leading from observation to inference. But the irrelevance to science of the conditions of discovery seems by now to be well established. 63 Moreover, carried to its logical conclusion this restriction would lead to an infinite regress, or at best would require us to complete reconstruction on the dialect level before the most closely related distinct languages could be compared with the eventual aim of reconstructing the family ancestor. Since the number of reconstructed languages would be equal to the number of identifiable subgroups on all levels and this could, dependent on the patterning of hierarchical inclusions, be nearly equal to the number of attested languages, the task of high-level reconstruction in a genetic grouping as large as the Austronesian family (est. 500 languages) would be made insuperably—and unnecessarily—difficult.
If there is a general lesson to be learned from the present case it is the unproductiveness—indeed pointlessness—of such a methodological stricture. 64 Although the inclusion of cps. (3), (4) and (10) is sufficient to demonstrate that the correspondence Fi s or : PPN *t cannot be explained as a conditioned reflex of PFP *ns, the need to reconstruct PFP *c is barely discernible from a comparison of Fijian and the Polynesian languages themselves. It is only through the expanded comparative base made accessible by appeal to external witnesses, and hence through reconstruction “from the top down”, that this inference can be seen as thoroughly justified.- 355
Since reaching approximately its present form, the argument that the Polynesian languages contain a third palatal reflex has been bolstered through several additional examples supplied by Andrew Pawley. These items and Pawley's commentary, where appropriate, are as follows:
In addition, the following comparison has been brought to my attention by Ross Clark, whose comments are noted:
I am grateful for comments on an earlier version of this paper by Ross Clark, William A. Foley, Peter C. Lincoln, D. S. Walsh and especially Andrew Pawley, who pointed out several new examples of the third Polynesian palatal reflex (see Postscript). Any remaining errors of fact or interpretation are entirely my own responsibility.
1 Written *(ń)d′, *()k′, *()g′ by Dempwolff. The distinction of *z and *Z, originally proposed by van der Tuuk (1865), was ignored by Dempwolff (who wrote both *d′), but was reinstated by Dyen (1951). Since we are not concerned in the present paper with the reflex of PAN *ñ in Polynesian or other Oceanic languages (but see Blust to appear (a)), ‘Proto-Austronesian palatals’ will hereafter be taken to mean ‘Proto-Austronesian palatal obstruents’. Unless noted otherwise, the orthography of PAN reconstructions follows Dyen through (1965) as modified by Blust (1970); the orthography of Proto-Oceanic reconstructions similarly follows Grace (1969). Abbreviations and sources of material not indicated by context are:
Ahus (Blust n.d.)
'Āre 'Āre (Geerts 1970)
Arifamu (Capell 1943)
Arosi (Fox 1968)
Baluan (Blust n.d.)
Bugotu (Ivens 1940)
Cebuano Visayan (Wolff 1972)
Dobuan (Bromilow 1904)
Gapapaiwa (Blust n.d.)
Gedaged (Mager 1952)
Gilbertese (Bingham 1908)
Hwn:Hawaiian (Pukui and Elbert 1971)
Iban (Scott 1956)
Ilocano (Carro 1956)
Javanese (Pigeaud 1938)
Lau (Fox 1974)
Leipon (Blust n.d.)
Likum (Blust n.d.)
Lindrou (Blust n.d.)
Loniu (Blust n.d.)
Lou (Blust n.d.)
Mailu (Capell 1943)
Makasarese (Matthes 1859)
Malay (Wilkinson 1957)
Manam (Blust n.d.)
Mao:Maori (Williams 1971)
Maranao (McKaughan and Macaraya 1967)
Mendak (Blust n.d.)
Mota (Codrington 1896)
Motu (Lister-Turner and Clark?)
Mukawa (Capell 1943)
Mussau (Blust n.d.)
Nali (Blust n.d.)
Nauna (Blust n.d.)
Ngaju Dayak (Dempwolff 1934-38)
Nggela (Fox 1955)
Niue (McEwen 1970)
Numfor (van Hasselt and van Hasselt 1947)
Pak (Blust n.d.)
Pati (Leenhardt 1946)
Penchal (Blust n.d.)
Puluwat (Elbert 1972)
Rotuman (Churchward 1940)
Roviana (Waterhouse 1949)
Sa'a (Ivens 1929)
Saaroa (Ferrell 1969)
Sangir (Steller and Aebersold 1959)
Seimat (Blust n.d.)
Sm:Samoan (Milner 1966)
Sori (Blust n.d.)
Tagalog (Panganiban 1966)
Takia (Blust n.d.)
Titan (Blust n.d.)
To:Tongan (Churchward 1959)
Toba Batak (Dempwolff 1934-38)
Wuvulu-Aua (Blust n.d.)
Yami (Ferrell 1969)
2 Dempwolff 1924-25, 1928, 1934-38.
3 Dempwolff 1924-25.
4 Dempwolff 1928.
5 Dempwolff 1937.
6 Dempwolff 1927.
7 Elbert 1953.
8 Grace 1959:70
9 Milke 1961.
10 von der Gabelentz 1873; Codrington 1885; Friederici 1913; Ray 1926.
11 Dempwolff 1927; 3; 1937:121.
12 Milke 1961.
13 Milke 1965.
14 Goodenough 1961.
15 Milke 1965:334.
16 Milke 1968.
17 Also the doublet sika ‘hate, despise’, not cited by Milke.
18 See Grace (1969), where this view—otherwise unattested in print—is discussed.
19 Grace 1969.
20 For a discussion, and further examples of the PAN:OC correspondence C1V1C2C1 V1C2:CVCi, see Blust (to appear (b)).
21 In addition to nd, Nggela h or s may reflect a PAN palatal (hava < *sapa ‘what?’, hala < *Zalan ‘path, way, road’, uha < *quZaN ‘rain’, ihu < *ijuŋ ‘nose’; sala ‘break a custom or law’ < *salaq ‘wrong, in error’, suru ‘diarrhoea’ < *ZuRuq ‘liquid, gravy, juice’). Yabem s, ns (see previous examples) or ø (as in y-ab ‘high land, upland’ < POC *nsa(m)pe ‘up, high’, wá < *wa(n)se ‘divide’) corresponds to the regular reflex of a PAN palatal in most other OC languages, though examples of all three developments are difficult to find in items traceable to a PAN prototype.
22 Milner 1963.
23 Elbert 1953.
24 Walsh and Biggs 1966, Biggs, Walsh and Waqa 1970 and subsequent computer print-outs.
25 Dempwolff 1928, Elbert 1953.
26 Dempwolff 1938.
27 Biggs (1965:391) earlier wrote Proto-Eastern Oceanic *ŋu(ts)u, PAN *ŋu(ts)u ‘lips, mouth’, thus implying a separate derivation of the Fi and PN forms from PAN doublets. Since Dempwolff did not reconstruct *ŋutu, and comparative evidence requiring the reconstruction of such a form is not presented by Biggs, this argument remains unsupported.
28 Pawley 1966:40; Pawley 1972:40.
29 Pawley 1972:120.
30 Capell 1957.
31 Hockett 1973.
32 Although they cite Fi ŋusu ‘mouth’ as a probable cognate of PPN *ŋutu ‘beak, orifice, mouth’, Walsh and Biggs (1966) do not include a similar citation for Fi usa under PPN *uta.
33 The association of the PN forms with Iban mutan ‘cargo’, proposed in Blust (1972), now seems clearly to have been in error. Instead, the Iban item is probably cognate with, or a borrowing of Malay muat-an ‘cargo’, with irregular contraction of the earlier vowel sequence.
34 Dobuan qe- is assumed to be the predication marker frequently offered with the citation forms of verbs in languages of the Massim area (cf. Gapapaiwa e gibugibu-i ‘to roast’, e taratara-i ‘to hew, chop, fell trees’, e gamogamo ‘to quarrel’, etc.).
35 Listed sub ‘load’, but not sub uda.
36 Hockett 1973:11.
37 Capell's gloss for this item is potentially confusing, but apparently refers to the standard nautical usage (i.e. the ropes used to manipulate the sail). This interpretation is supported by Paul Maivusaroko and Joseph Sotutu, speakers from two distinct dialect areas in Vanua Levu. Hornell (1936:335), on the other hand, gives Fi karikari sila ‘boom of sail’ (supporting bamboo that runs the length of the bottom of the sail), karikari tu ‘yard of sail’ (supporting bamboo that runs the length of the top of the sail).
38 PAN *t and *T > t in most Admiralty languages (Seimat, Lou, Baluan, Penchal, Nauna tina-, Ahus tine- < *tina ‘mother’), but *s, *z/Z, *c > c or s (see Blust to appear (a)).
39 A doublet is evidently needed to account for Likum sah, Ahus car ‘sea urchin’, Mota sar ‘echinus’. Puluwat ccaaŕ ‘k.o. common edible sea urchin’ is possibly connected, but shows several unexplained irregularities.
40 PAN *s, *z/Z, *c and *j > t and *t, *T > ø before a non-high vowel in Lau and other languages of the Malaita-San Cristobal group, as in Lau qato < *kasaw ‘rafter’, tala < *Zalan ‘path, road’ fita < *pija ‘how much/how many?’, ano c *taneq ‘earth, land’, olu > *telu ‘three’.
41 Biggs (1965) cites Rotuman /h/ as the normal reflex of a palatal in indirectly inherited words. If borrowed, however, the source language for this item apparently could not have been Polynesian.
42 Dempwolff 1934-38.
43 Grace, n.d.
44 Listed with long last vowel by Grézel 1877: afàā (i.e. afaqaa) ‘tempête, ouragan’. Dempwolff writes afaqa.
45 Milke 1968.
46 To facilitate fuller comparison, forms for Manam and the Admiralties are cited from my own (1975) fieldnotes. Mailu (tail of New Guinea) iato, Mukawa, Arifamu kiato ‘outrigger boom’ show /t/ as against the usual Mailu /s/ (susu < *susu ‘breast’) or /r/ (are-na < *qa(zZ)ay ‘chin, jaw’, marai < *ma-sakit ‘sick, painful’), Mukawa /s/ (kusana < *quZaN ‘rain’, puso < *pusej ‘navel’) or zero kawa-kae-na < *qa(zZ)ay ‘chin, jaw’, kaoa < *qa(cs)awa ‘spouse’, nua < *nusu ‘island’) from an earlier palatal. Since no other example of /t/ as the reflex of a palatal is known in any of these languages, there is a distinct possibility that the word for ‘outrigger boom’ (and hence presumably other terms in the canoe-complex) is a Polynesian loan. The same may be true of Efate kiat ‘outrigger boom’ (cited by Milke) and Pati (New Caledonia) kioto ‘fork for the attachement of the outrigger boom’.
47 Together with Seimat ayas, Nali kayas, Pak kayat, Titan kacac, this form points to a doublet, Proto-Admiralty *kayaco. Gedaged aiad ‘outrigger boom’ suggests that the Proto-Admiralty form in turn derives from a POC doublet.
48 Reflexes of this word in OC languages that are diagnostic for *t/nt point to the simple stop (Fi vutu, Sa'a huu ‘a tree: Barringtonia spp.’). Since no instance of Seimat /s/ other than that in /ayas/ is known to correspond to the regular reflex of *t or *nt in any other language, however, it seems safe to conclude that /s/ in this word could not reflect *t/nt. A similar argument applies to the cognate forms in Wuvulu-Aua (where /t/ indicates an earlier palatal, but apparently never *t or *nt) and Gilbertese (where /r/ is the regular reflex of the palatal series, but seems never to derive from *t/nt).
49 Friederici 1912:310.
50 Hornell 1936:311.
51 Capell 1957.
52 If from earlier *kiaso this item shows unexpected loss of the first vowel; if from *kayaso it may show sporadic loss of earlier (< *y), with contraction of the resulting vowel sequence. I have been unable to determine whether Hazlewood (1872) lists kaso, but Dempwolff, Milke and Hockett do not refer to it even in passing.
53 In addition to cps. (1) and (11)-(14), it is tempting to regard such western AN forms as Ilocano lôgan ‘unborn child, fellow passenger; ride on or in a vehicle’, Tagalog lôlan ‘load, cargo; capacity or accommodation (of a vehicle)’, Cebuano Visayan lôlan ‘board, put on a vehicle’, Maranao roran ‘cargo, transport’, Sangir luraŋ ‘load, cargo’, Makasarese luraŋ ‘to load’ as cognate with members of cp. (2). All non-OC forms point clearly to *luja(nN), however, while OC forms (with the possible exception of Motu and Dobuan) point to a root with initial *R. Despite this discrepancy it could still be assumed that the medial consonant of the OC words derives from PAN *j, but in forms such as Nauna us ‘to load, as cargo in a boat’, this assumption appears to be in conflict with other observations (see Blust, to appear (a)).
54 Blust 1972:cp. 224.
55 A fairly thorough search has been made for evidence that would permit the PN members of cps. (1)-(14) to be assigned to alternative comparisons with Proto-Fijian Polynesian *t. Since this search eliminated cps. (a)-(d) from an earlier list that included them, the probability that the remaining comparisons are valid would seem to be strengthened. Among additional chance resemblances we can cite PPN *mutu ‘cut off, ended’, Fi musu ‘break or cut crosswise’ (but also Fi mudu ‘cut off, ceased, ended’, Mota mut ‘maimed in foot or hand’), PPN *sunu ‘cook on open fire’, POC *sunu ‘singe, burn, burnt land’ (but also PPN *sunu ‘singe’, POC *tunu ‘warm up food’) and PPN *pito, Fi viovio, POC *mpuso(s) ‘navel’ (but also Fi buto (dialect form cited by Walsh and Biggs), Mussau bito, POC *mputo ‘navel’).
56 I am indebted to Paul Geraghty for having brought this point to my attention.
57 Also cp. Bau lasa ‘easy, contented, tame, accustomed’, To lata ‘to feel at home or at ease, to be comfortable or happy and contented’, Sm lata ‘tame, used to, familiar with’, Mao rata ‘tame, quiet; familiar, friendly’, Hwn laka ‘tame, domesticated, gentle’ (comparison courtesy Paul Geraghty).
58 Milke 1968.
59 The Takia reflexes of the palatals appear to be essentially the same as those of Gedaged.
60 Milke's criteria for *nj are, in fact, partially incompatible even on the basis of the Gedaged and Nggela evidence, as with Gd dal, Ngg hala ‘path, way, road’. As shown in Blust (to appear (a)) it is apparently impossible to reconstruct either fixed (lexical) or free (morphological) consonant grade for POC *s without entailing major contradictions.
61 The statistical significance of this correspondence is evidently no greater than that for the correspondence of PFP *c with Milke's *nj. Rotuman /j/, on the other hand, generally corresponds to *nj in the small number of available examples:
62 However, see Haudricourt 1965:315.
63 Popper 1959.
64 This is not to deny, of course, that prior reconstruction on lower levels can sometimes benefit higher-level reconstruction. What is at issue is whteher such an ordering is logically necessary to the justification of reconstructional inferences.