Dempwolff Reinvented: A Review of Wolff (2010)
This ambitious study, the result of over forty years of intermittent labor, essentially attempts to restart Austronesian comparative linguistics from the ground up. It rejects several key distinctions made by Dempwolff and accepted by virtually all subsequent scholars, and proposes a number of modifications to the phoneme inventory and word structure that reportedly are motivated at least in part by the author’s belief that Austronesian and Sino-Tibetan are branches of a larger language phylum. The body of the work contains sketches of the historical phonology of 37 languages reaching from Taiwan to Polynesia. These vary greatly in quality, accuracy, and relevance. The book concludes with a glossary of around 1,760 reconstructions, about 95 percent of which are drawn from Dempwolff or from Blust and Trussel’s more recent Austronesian Comparative Dictionary.
The foundations of Austronesian (An) comparative linguistics were laid by the German scholar Otto Dempwolff (1871–1938) in a three-volume work that appeared just before his death (Dempwolff 1934–38). Volume 1 laid out the goals of the work, sources of material, and the methodological approach adopted. It then systematically compared the lexicons and phonology of three “Indonesian” (In) languages (Tagalog, Toba Batak, and Javanese), and on the basis of this comparison proposed an “inductive reconstruction” of Urindonesisch (Proto-Indonesian; PIn). Volume 2 applied the PIn sound system “deductively” to eight other languages that were labeled “Indonesian” (Malay, Ngaju Dayak, and Hova/Malagasy), “Melanesian” (Fijian and Sa ‘a), or Polynesian (Tongan, Futunan, and Samoan). Since the data of all eleven languages could be explained fully from the PIn sound system, Dempwolff concluded that PIn was equivalent to Uraustronesisch (Proto-Austronesian; PAn), and he added that the languages classified as “Melanesian” and Polynesian are branches of a single group called Melanesisch (known today as Oceanic). Volume 3 was a lexicon of slightly over 2,200 reconstructions, with supporting data from two or more languages, at least one of which was In.
Dempwolff’s work was foundational, but not flawless. Most critically, it omitted the Formosan languages, which are crucial to understanding the history of An, and so lacked important data needed to reconstruct the phonemes most Austronesianists write as *q, [End Page 560] *S, and *h, and to distinguish the pairs written *t/*C and *n/*N. From about the mid- 1940s until the early 1970s, the American linguist Isidore Dyen, who was Wolff’s teacher, was the major figure involved in refining Dempwolff’s treatment of sound correspondences, and in providing a comprehensive, but highly problematic, subgrouping of An as a whole. As is well known, Dyen’s approach to phonological reconstruction was controversial, and can be characterized globally as one in which observed irregularities, sometimes in a single language or word, were projected onto PAn in a seemingly endless string of potential new protophonemes that awaited confirmation, but over time came to be treated as though they were an established part of the reconstructed language (Blust 2009a:536–46, 2009b).
The preceding paragraphs are a necessary prelude to any review of this book, as what Wolff (W) sets out to accomplish is nothing less than to reinvent Dempwolff from the ground up with an approach that arguably is guided by a deep-seated reaction to the work of Dyen. Proto-Austronesian phonology with glossary (hereafter PANPWG) is a massive compendium of data and analysis, much of it idiosyncratic, but nonetheless displaying W’s obviously wide knowledge of the languages and command of detail. Following a list of maps and abbreviations, three pages of acknowledgements, a three-page foreword by Laurent Sagart (to which I return below), and a preface, Part A, Introduction (pp. 3–65) is devoted to background on the language family, subgrouping, and homeland (Ch. 1), “Considerations of Theory and Methodology” (Ch. 2), and an “Inventory of PAn Phonemes and Other Issues of Phonology” (Ch. 3). The book then moves on to Part B, Development of the Formosan Languages (Pazih, Saisiyat, Thao, Atayalic, Saaroa, Kanakanabu, Rukai, Bunun, Amis, Kavalan, Puyuma, Paiwan); Part C, Development of the Philippine Languages (Tagalog, Chamorro, Ratahan, Tondano, Pamona, Buginese, Salayarese, Muna); Part D, Development of the Languages of Kalimantan, Malagasy, and Malay (Kelabit, Ngaju Dayak, Malagasy, Malay); Part E, Development of Old Javanese, Toba Batak, and Moken; Part F, Development of the Languages of Eastern Indonesia (Manggarai, Buru, Leti, Kei); Part G, Development of the Oceanic Languages (Tolai, Motu, Sa‘a, Fijian, Tongan, Samoan); Part H, Glossary (a nearly 300-page listing of about 1,760 reconstructions and supporting evidence); Part I, Indices of Citations; Part J, References; and Part K, Topical Index. With only minor deviations, the treatment of these 37 languages follows a modular format, first placing the language in space and scholarly context, then treating general weakening processes and deletions in the antepenultimate syllable, followed by similar processes in the penult, consonant cluster simplification, what is called “disyllabization of monosyllabic roots,” metathesis, intercalated consonants, and finally a step-by-step account of how W’s version of the PAn phonemes developed, with comments on both regular and irregular changes, speculations about borrowing, and so on. These sketches range from eleven pages (Rukai, Paiwan) to 26 (Muna), averaging about 17 pages per language.
3. The Proto-Austronesian Phoneme System
By any account, PANPWG is a major endeavor that clearly represents a lifetime of scholarship and intensive labor in bringing it to completion. In view of the tremendous effort that was obviously [End Page 561] expended to produce this book, one could only wish the result might be more congenial to the majority of scholars working in the Austronesian field. The problem is not with W’s command of the data, as it is abundantly evident that he is familiar with the material; the problem rather is with his approach to method, which is often so quirky as to be distracting. A good place to begin is with a critique of the sound system posited for PAn.
In a number of publications spanning three decades, Wolff (1974, 1982, 1988, 1991, 1993, 1997, 2003) argued (i) that certain of the phonological distinctions recognized by Dempwolff and accepted by virtually all subsequent scholars are unfounded, and (ii) that the PAN *t/*C and *n/*N distinctions proposed by Ogawa and Asai (1935) and later revived by Dyen (1965) are products of suprasegmentally conditioned sound change. These claims are repeated here as a fait accompli, despite serious criticisms raised by Ross (1992) and Blust (2009a:546–85) that are ignored in this work. I will consider only the stops, where the principal points of disagreement between what can be called the “standard view” (SV) and that of Wolff are shown in (1):2
labial alveolar palatal palato-velar velar uvular SV: p t, C c k q Wolff: p t (none) k q SV: b d z j g Wolff: b (none) j g [g] (none)
A comparison of W’s Chart Two with the material in his glossary shows that the chart contains several errors in the area of the voiced coronal and velar stops. First, SV *d in syllable-initial position is equivalent to W’s *d, as in *duSa (SV): *dusa (W) ‘two’, or *qudaŋ (SV) : *qudáŋ (W) ‘shrimp, lobster’, not to the absence of a protophoneme as he states. Second, W’s *j is equivalent to SV *z in syllable-initial position, as in *jalán (W) : *zalan (SV) ‘path, road’, or *qujáɬ : *quzaN ‘rain’, but to SV *d syllable-finally, as in *lahuj (W) : *lahud (SV) ‘seaward’, *likuj (W) : *likud (SV) ‘back’, or *kuj ‘lower shank’ : *kudkud ‘hoof’. Third, W’s *g, which is reconstructed in all positions, is said to be equivalent to SV *j, which has been reconstructed only in medial and final positions. In fact, W’s *g in initial position is equivalent to two SV phonemes, neither of which is *j: first to *g, as in *gatél (W) : *gaCel (SV) ‘itch’, *gemegem : *gemgem ‘hold in fist’, or *gusam : *guSam ‘thrush’, and second to *k, as in *galih (W) : *kalih (SV) ‘dig’, *getil : *ketil ‘pinch and break off’, and *gíta : *kita ‘see’. For convenience of reference, the standard view of the PAn voiced coronal and velar stops is shown in table 1, together with the correspondences they represent in Paiwan of southeast Taiwan, the three languages of Dempwolff (1934), and one of the languages of Dempwolff (1937).3
Perhaps the most conspicuous feature of table 1 is that *j and *z have defective distributions that are partly complementary: *j is not reconstructed word-initially and *z is not [End Page 562] reconstructed syllable-finally (but the two contrast medially as syllable onsets). Just as Nature abhors a vacuum, so W evidently abhors a gap, and one of the most problematic features of his reconstruction is that he has filled these gaps so that every consonant occurs initially, medially, and finally. He has done this in a series of steps that many will find not only arbitrary, but also in conflict with distributional properties of the consonants in the modern languages.
Wolff (1982) argued that the consonants *c, *r, and *g (voiceless palatal affricate, apical tap, and voiced velar stop, respectively), which were inherited from Dempwolff, are not well supported and should therefore be dropped, leaving the stop system as: /p, t, C, k, q; b, D, Z, j/.
Click for larger view
View full resolution
His position remained unchanged in 1988, but in 1991 he suggested that *C, proposed by Ogawa and Asai (1935), and again by Dyen (1965), be considered a stress-conditioned allophone of *t, leaving the stop system as /p, t, k, q; b, D, Z, j/. In 1997, he introduced a further restructuring of this system, relabeling *D as *d, *Z as *j, and *j as *g, now admitting some instances of *g- that were rejected in 1982. In other words, after dismissing all examples of *g in 1982, W partially retracted this position in 1997 by admitting word-initial SV *g-, which he combined with SV *j, labeling both as a new *g, but one with awkwardly different patterns of reflexes in initial and noninitial position, as with Paiwan g-, -d-, -d; Tagalog g-, -l-, -d; or Proto-Oceanic *k-, -s-, -s. In addition, W also now called *ɣ (SV *R, usually seen as an alveolar or uvular trill) a “voiced stop,” leaving the stop system as /p, t, k, q; b, d, j, g, ɣ/. These proposals are presented again in basically the same form in Wolff (2003) and the work under review. [End Page 563]
In his discussion of method, W reassures the reader (19) that “the new phonology enables us to account for the innovations that have taken place in a natural way—that is, the changes can be presented as the result of plausible articulatory processes of the sort that are paralleled in documented changes in languages belonging to any number of language families located in all parts of the world: assimilation, weakening, simplification, and so forth.” In the SV, the diachronic picture for *g and *j differs strikingly: *g has essentially two reflexes, /g/ and /k/, while *j is reflected as /d/, /ð/, /g/, /j/, /ʔ/, /s/, /x/, /y/, /n/, zero, and in other ways (/l/, /r/, /t/, /k/) that are positionally conditioned (Blust 2009a:572). Given the diversity of its reflexes, including historically independent instances of both /d/ and /g/, as well as continuants such as /s/ and /y/, it has long appeared most reasonable to interpret *j as an island in the phonological system, probably a palatalized velar stop (palatalization pulling it forward in many languages, and sometimes providing it with a continuant feature). By reinterpreting *j as [g] rather than the SV [gy], W is claiming instead that changes such as *j > /d/ or *j > /s/, which are recurrent in An languages, were actually *g > /d/ and *g > /s/, but only noninitially, and that such changes exemplify “plausible articulatory processes.” It is not surprising that few scholars agree with his position on this issue, as it is clearly at odds both with the comparative evidence for a contrasting *g in medial and final positions (SV *g) and with our general knowledge of likely types of sound change.4
Having replaced SV *j ([gy]) with *g to fill the first gap in table 1, W then relabeled SV *z as *j ([J̶]), and reassigned the correspondences for SV *-d to his *j, thus filling the second gap. While the use of *j to represent the correspondence under *z in table 1 is phonetically more appropriate, it claims in effect that palatal consonants could occur word-finally, yet one of the most striking facts about languages with palatal stops and /ñ/ (Ngaju Dayak, Malay, Old and Modern Javanese, Chamorro, and so on) is that these segments may not occur syllable-finally, and there is no compelling reason to believe that a distributional limitation of this type is a late development rather than a feature inherited from PAn. In addition, the correspondence assigned to SV *-d was moved to *j, leaving no *d in final position. To fill the gap produced by this game of phonemic musical chairs, W moved the correspondence that Dempwolff assigned to *-D to *d, and dismissed the etymologies upon which Dempwolff reconstructed *D in nonfinal position. Since *-D was relatively uncommon, *-d is now oddly rare.
Even the uninitiated reader will no doubt begin to sense how potentially confusing these “hygienic” alterations to the SV are, even without considering the substantive problems they introduce. Some Dempwolff etymologies that W dismisses in order to achieve his consolidated consonant system no doubt are invalid, but many others appear to be victims of an ideological position. A clear methodological weakness in much of W’s comparative work when he is confronted with a puzzle is his facile appeal to “analogy,” with no reference to a documented proportion that could account for the data in terms of pattern recognition and transfer on the basis of noncontroversial parallelism. As for *g/*k vacillation, which W takes to vitiate many instances of SV *g (in all positions), this is not unexpected, given the shorter duration of voicing for velar, as opposed to prevelar, stops, but it rarely obfuscates the original voicing properties (Blust 1996). [End Page 564]
The next major issue in (1) concerns the*t/*C (and by implication *n/*N) distinction. W argues that PAn had only *t and *n, and that the correspondences assigned to *C (probably [ts]) and *N (possibly [ɬ]) arose by stress-conditioned allophony. As noted by Blust (1997a, 2009a:553ff), although Proto-Philippine stress remains a thorn in the side of Austronesian comparativists, no reliable argument has yet been presented for attributing contrastive stress to PAn. W acknowledges this repeatedly throughout Part A of this book, yet insists on appealing to contrastive stress to explain the correspondences that have been assigned to *C vs. *t and *N vs. *n. In descriptions of individual languages such as Pazih (76), Saisiyat (90), and Thao (105), to name just a few, he then states matter- of-factly that *t has two outcomes that happen to agree cross-linguistically in cognate morphemes, but he is unable to state conditions for the divergent development.
Other issues that I will not deal with at length here are the reality of *c and *r, both of which are well supported by a substantial number of published etymologies. As noted in Blust (2009a:556–60), *c is distinguished from *s in only 20–25 languages of western Indonesia and mainland Southeast Asia, but these languages do not belong to a single subgroup, are geographically discontinuous, and *c is consistently distinguished from *s in monosyllabic roots that form part of different morphemes, thus eliminating borrowing as a possible explanation. Without access to *c, over 90 current reconstructions will either have to be discarded, or will show reflexes with unconditioned phonemic splits, a pattern that would suggest to most comparativists that a generalization has been missed. Much the same statement can be made for *r which, however, is distinguished by a more widely scattered set of languages (Blust 2009a:581–84).5
4. Treatment of Variation
While these distortions of the reconstructed consonant system are unsettling enough by themselves, they are far from being the only matter that most readers of this book will find disturbing. A second way in which Wolff’s comparative approach departs sharply from that of Dempwolff and most subsequent scholars is in his treatment of variation. Since at least the work of Brandstetter nearly a century ago, it has been recognized that many etyma differ only slightly in shape from others of similar meaning, as with *bañaw ‘wash, bathe’, *ñawñaw ‘rinse, wash’, *Señaw ‘wash’, *Siñaw ‘wash’ (Blust 1988:127); *asa, *esa, *isa ‘one’; or *qali-maŋaw, *qali-maŋu ‘mangrove crab’. As noted in Blust (2011), lexical variants in An languages fall into two formally distinct classes: those that have a recurrent submorphemic ‘root’, as *-ñaw ‘rinse, wash’, and those that do not. Submorphemic roots almost always have the shape -CVC, and only onomatopes are freestanding: Proto–Malayo-Polynesian (PMP) *balatuk ‘woodpecker’, *getuk ‘knock, pound’, *katuk ‘knock’, *ketuk ‘knock, pound’, *retuk ‘sound of chopping’, *tuktuk ‘knock, pound, beat’, but *tuk ‘the sound tuk’. [End Page 565]
W recognizes the reality of roots, but despite the total absence of comparative evidence to support this procedure he reconstructs them as free-standing morphemes, whether they are onomatopes or not, as with his *but ‘pluck out’ (cp. the bound element *-buC ‘to weed, pluck, pull out’ in Blust 1988:86), or *jak ‘step on’ (cp. the bound element *-zak ‘step, tread’ in Blust 1988:163). The result is a thorough obfuscation of PAn canonical shape. Chrétien (1965) determined that over 90 percent of the lexical bases in Dempwolff (1938) were disyllables, and Blust (2007a), following up insights first expressed by Brandstetter nearly a century earlier, has shown how an inherited disyllabic canonical target has powered a number of structurally distinct drifts in widely separated An languages. Moreover, in the largest collection of comparative An lexical data available (Blust and Trussel, in progress; hereafter “ACD”) there are no monosyllabic content morphemes at all. W nonetheless posits many monosyllabic bases as free morphemes in PAn, and in his descriptions of the phonological history of individual languages he claims that these were doubled or united with affixes that fossilized in order to satisfy the predominant disyllabic canonical target.
W’s belief in a disyllabic canonical target is certainly correct, but his ideas about how it works are contrary to what most other scholars assume, namely that monosyllabism in content morphemes was an epiphenomenon of regular sound change, not an original feature of PAn content morphemes. To cite three of many possible examples, W posits *kan ‘eat’, explaining the last vowel of Bunun kaun, Amis kaen, Tagalog kaín, Cebuano kaún, Sangir kaeŋ, and so on, as due to historically secondary “stretching of the root” that happened to result in regular sound correspondences.6 In *ɬuɣ (SV *niuR) ‘coconut’, the “stretching” that produced Tagalog niyóg, Chamorro niyok ([nidzok]), Palauan líus, Loniu niw, Mono-Alu niunu, or Bugotu niu is different from that posited for *kan, and cannot be predicted from context. Moreover, there is abundant evidence (some of it discussed in Blust 2007a) that in content morphemes that have become monosyllabic through sound change, a lost disyllabism is reconstituted by schwa epenthesis, not “stretching” of a nucleus. This is seen, for example, in Ngaju Dayak eñuh ‘coconut’, where initial schwa was added after the base reduced to a monosyllable through palatalization and absorption of the high front vowel. The wrongheadedness of W’s approach becomes even more obvious with his *tan ‘set trap’, which must somehow account for both the medial consonant and the final vowel of forms such as Cebuano táʔun ‘set a trap or fishline in place’, Malay tahan ‘to set (snares and traps)’, Muna tagho ‘set a trap’, Palauan d<el>áchel ‘good place for setting fish traps’, me-láchel ‘set a trap’, or Lakalai (ta)taho ‘set a snare’, which together point unambiguously to *taqen.
The other side of this coin is seen with reduplicated monosyllables such as *tuktuk, cited above. W acknowledges no consonant clusters in PAn, and so must intercalate a schwa in such forms, altering all of them to trisyllables (*tuketuk, etc.). To derive the attested forms he claims that medial schwa syncope—a real and widespread change in An languages—reduced such forms to disyllables (*tuketuk > *tuktuk, etc.). What he does not mention is that such “reduction” is seen even in languages that did not syncopate [End Page 566] medial schwa in nonreduplicated bases, as Moken, Nias, Muna, or Proto-Oceanic, thus implying that this vowel was never there. The simplest hypothesis needed to account for the comparative data seems clearly to be that some languages that allowed no other types of medial consonant clusters independently separated the abutting consonants in reduplicated monosyllables by schwa epenthesis.7
W is clearly aware of the importance of canonical form, since he states (53) that “canonical form—the canonical shape of the root and syllable, play an important role in the development of PAn phonology in the daughter languages. Not only is a community constrained to make forms conform to existing phonological patterns, but there is a tendency for a community to change infrequent but permissible patterns or shapes to conform to ones that are more frequent.” W’s statement is completely reasonable. However, one wonders how to reconcile it with the practice in PANPWG of recognizing monosyllabic content morphemes, reduplicated monosyllables of the shape C1V1C2eC1V1C2, and many longer words made by appending a redundant third syllable to a valid disyllabic base, as with *basequ ‘smell’ (SV *bahu ‘stench’), *dásuwen (*dahun) ‘leaf’, *jáqewis (*zauq) ‘far’, *layehu (*layu) ‘wither’, *miniyak (*meñak, miñak) ‘fat’, *peniyu (*peñu) ‘green turtle’, *puɬetiq (*putiq) ‘white’, *qisuwab (*Suab) ‘yawn’, *sehapúy (*Sapuy) ‘fire’, *seqeɣét (*SeReC) ‘tight’, *taquwéɬ (*taqun) ‘year’, or *wasiyeɣ (*wahiR) ‘fresh water’. In some cases, the added syllable is based on jaw-dropping speculation, as with *basequ ‘smell’, where reflexes of PMP *bahu are combined with Bunun saq-un ‘stinking’, a form that has the wrong stem vowel plus a final consonant that agrees with Tagalog báhoʔ (erroneously cited as bahóʔ) ‘stink’, but conflicts with every other witness for *-q that is known to reflect this base (Karo Batak, Malay, Sundanese bau, Old Javanese abo, Chamorro pao, Palauan báu, etc.), and moreover lacks the first syllable (said to be a PMP “prefix”). In other cases, as *dásuwen ‘leaf’ or *salac ‘forest’, the presence of *s implies that the reconstruction has Formosan reflexes, but no such evidence is presented in the glossary.
W’s way of handling variation with bases that do not contain a submorphemic root is even more troubling, as he essentially tries to eliminate doubleting from the lexicon through stratagems that have no determinable motivation other than to ensure that each lexical entry is phonologically and semantically uniform. For example, it is well known that both *esa and *isa are widely reflected in the meaning ‘one’. W’s solution to this problem (840) is to reconstruct just *ica (SV *isa), and to argue that the initial vowel dropped “in atonic position,” after which the lost disyllabism of the base was restored by schwa epenthesis: *ica > ca > əsa. As noted above, researchers on An languages have long recognized the importance of canonical disyllabism. However, W’s argument is suspect. In many languages a proclitic form sa- occurs on the words for ‘ten’ (PAn *sapuluq, PMP *sa-ŋa-puluq) and ‘hundred’ (PAn *sa-Ratus, PMP sa-ŋa-Ratus), but the full disyllable occurs as the numeral ‘one’. If PAn had *isa, but *isa-puluq > *sa-puluq the reduced form of the numeral would be found only as a proclitic (hence part of the same phonological word as the base), and neither it nor the full base isa would be subject to schwa epenthesis. To make matters worse, a third variant, *asa ‘one’, is reflected in a [End Page 567] smaller set of languages, but one that includes Saisiyat and Bunun in Taiwan, the Bashiic languages of the northern Philippines, and several languages of eastern Indonesia.
Examples such as *asa, *esa, *isa ‘one’ are just the tip of the proverbial iceberg, since a comprehensive denial of doubleting in An languages would require hundreds of distinct explanations (406 different patterns of variation are documented in Blust 2011, and there is no guarantee that this count is complete). How do we account for attested forms that point to both PAn *baNaR and *baNaw ‘a thorny vine: Smilax spp.’ by positing a single protoform? And likewise with PMP *tuqelan, *tuqelaŋ ‘bone’; *buRuk, *busuk, *buyuk ‘rotten’; *gilak, *gilap, *kilap, *kilat ‘shine, flash; lightning’; *lipen, *nipen, *ŋipen, *qipen ‘tooth’; and hundreds of other sets of variants? W’s solution for some sets of doublets (he simply ignores many others) is reminiscent of what Matisoff (1990:116) called “proto-form stuffing,” namely, add enough extra material to a protoform to allow it to span two or more incommensurate comparata through ad hoc changes. A somewhat tongue-in-cheek example might be *dogasu, ancestral both to English dog by loss of the last two syllables, and Tagalog asó, by loss of the first. This procedure can be illustrated by *aqetíh ‘water recede’, an etymon that is not reflected as a trisyllable in any daughter language, and is given this shape solely to unite *keti, *qati, *qeti ‘ebb; low tide’ under a single reconstructed shape (in fact, it still fails to account for the first variant, for which a morphological solution is also problematic). Other examples are *iseci ‘contents; flesh’, proposed to unite *asi, *isi, *Sesi by an assumed metathesis of the first two vowels in some languages followed by aphesis, *isepí ‘dream’, proposed to unite *Sepi and *Sipi in the same way, and *bacuheq ‘wash’, posited to unite *baseq, *basuq, *biseq, *isuq ‘wet; wash’ (said to consist of a base *cuheq that occurs with “prefixes” *ba- and *si/ise-, but is unattested as a free form). In a few cases, W simply runs out of tricks and is forced to acknowledge the reality of doublets, as with *tiduR and *tuduR ‘sleep’ (27), although he claims incorrectly that “reflexes of *tiduɣ are found only in the Philippines and Indonesia” (in fact, reflexes of *tiduR are almost universal in languages of the Admiralty islands, and are found in several languages of Paama and the Shepherd Islands in Vanuatu, as shown in Tryon 1976:473–74).8
One of the patently undesirable consequences of this reconstructional procedure is that W is forced to assume a metathesis of initial and medial vowels that targeted the same forms independently in numerous languages. Thus *isepi is posited to reconcile Pazih sipi, Proto-Rukai *sipi, Toba Batak, Old Javanese, Pamona ipi, Chamorro gwifi ‘dream’ with Proto-Atayalic *sepi, Paiwan mi-sepi, Tetun m-ehi ‘dream’ through metathesis in all languages that show /i/ as the penultimate vowel; *iseɬaw ‘wash’ (inconsistently written *isenaw on p. 72) is posited to reconcile Pazih sinaw, Thao shinaw, Truku Seediq sinaw, Kanakanabu m-ari-sináw, Proto-Rukai *sinaw, and Bunun ma-sinaw with Pazih me-senaw ‘wash utensils’, Bikol hanáw ‘wash the hands or feet’, Cebuano hunáw ‘wash the hands’, and so avoid the reconstruction of doublets *Señaw, Siñaw; and *iseyup is said to produce Saisiyat hiop, Thao iup, and Kavalan m-siup ‘to blow’ through [End Page 568] separate metatheses intended to reconcile these forms with apparent Philippine cognates such as Cebuano huyúp ‘blow air’, and so on.
The ad hoc nature of this treatment of variation is clear, and like most ad hoc proposals, it leads to other problems. As just noted, for what others would write as *baseq, *basuq, *biseq, and *isuq W cites a single form *bacuheq, with a proposed base *cuheq prefixed by *ba- in the first two forms and *si/*ise- in the last. The problem is that such “prefixes” are manufactured out of convenience: there is no comparative evidence at all to support either one.9 More generally, it is notable that PANPWG, like Dempwolff (1938), limits its scope to lexical bases with little attention to the reconstruction of affixed forms, unlike the ACD, where lexical bases are accompanied by all affixed forms, reduplications, and compounds that can be justified by comparative data. Where W does talk about well-established affixes (51), it is to reiterate the worn-out claim that the PAn infixes *-um- and *-in- were actually prefixes that metathesized into the stem, increasing the markedness of the affixed word, and negatively affecting its learnability by children (for an overview of the many problems with this interpretation cf. Blust 2009a:382–88).
The problems identified so far with PANPWG have concerned phonology, canonical shape, and morphology. But semantic reconstruction is also an important part of any fully articulated etymology, as shown by such classic works as Benveniste (1973). Dempwolff (1934–38) showed little interest in semantic detail, and W is no different, often giving glosses that are vague or impoverished. Examples of the first type include *bánuwa ‘land, place where there is s.t.’ (*taneq was ‘land, earth, soil’, but as argued in Blust 1987, *banua referred specifically to inhabited territory, the territory that provided the life support of the community in both material and spiritual terms); *gumis ‘down, body hair’ (*bulu meant ‘body hair’; the doublets *gumis, *kumis referred exclusively to facial hair, hence ‘beard, moustache’); *lutuŋ ‘ape’ (178), ‘monkey’ (900), (given the Formosan homeland that W assumes, the referent could only be the Formosan rock monkey); and *qudáŋ ‘crustacean’ (reflexes of this term refer almost exclusively to shrimp and lobsters, never to crabs, barnacles, or other types of crustaceans). An example of the second type is *bataŋ ‘main part of tree’, where the gloss for the PMP form is far richer, including ‘tree trunk, fallen tree, log’ ‘stem of a plant’, ‘body’, ‘log bridge’, and ‘numeral classifier for cylindrical objects’; on the Proto-Western Malayo-Polynesian level we can further add ‘corpse’, ‘self’, ‘bridge of the nose’, ‘most important or pre-eminent thing’, and ‘main course of a river’. Where greater attention is given to semantic detail it is almost entirely by noting glosses given in the ACD.
6. Speculative Etymologies
Entirely apart from the unexpected shapes found in protoforms such as *aqetíh, *isepí, or *bacuheq (a problem of reconstruction), a number of proposed etymologies are forced or surprisingly farfetched, given W’s [End Page 569] clear understanding of the sound correspondences (a problem of identifying relationship). An example of the first type is *baɣeqaŋ > Saisiyat rælæʔæm ‘molar tooth’, where we read (93) that “*b- is reflected with /r/.” In fact, both *bageqaŋ, and *baReqaŋ (ACD) must be posited for ‘molar’, and other forms in individual languages cannot be reconciled with these, as with Malay geraham ‘molar tooth’, so the lack of fit between *baReqaŋ and the Saisiyat form is hardly surprising. In other cases, the critical reader is left wondering why a relationship is assumed at all, as with *biɣeni (ACD: *beRŋi) > Pazih xinian (expected **bixiŋi) ‘night’ (72), *naɬaq > Pazih laŋa (expected **nala) ‘pus’ (72), *taŋíla, *talíŋa > Saisiyat saɬiʔil (expected **saŋira, sariŋa) ‘ear’ (94), *iceɣáb (ACD: *neRab, *niRab) > Thao tiɬa (expected **ɬaf, niɬaf) ‘yesterday’ (98), *bacuheq (ACD *baseq, *basuq) > Thao fɬuq (expected **fatiq, fatuq) ‘wash’ (98), *ɣuqaɬay > Thao ayuði (expected **ɬqaðay) ‘male’ (99), *baqesin > Thao ɬqauʃin (expected **faqʃin) ‘sneeze’ (100), *sulaɣ > Thao qɬuran (expected **turaɬ) ‘snake’ (100), *puɬetiq > Thao puði (expected **puðtiq) ‘white’ (105), *wasiyeɣ > Atayal qosiyaʔ (expected *wsiəg) ‘water’ (117), *ɣuqaɬay > Bunun ba-nanað (expected **luqanay) ‘male’ (174), *b<in>a-bahi > Bunun binanauʔað (expected **binabað ‘woman’ (176), *tambuɣi ‘triton shell’ > Kelabit buriʔ (expected **temburi) ‘someone’s words’ (417), *baheɣat > Tongan mama-fa (expected **foa) ‘heavy’ (702; < Proto-Oceanic *mapat), *kanuhec > Tongan ŋuu-feke (expected **kanuo) ‘squid’ (703), *buɣesu (SV *buRu) > fuaʔa (expected **fū) ‘jealous’ (703), and many others that require special pleading, given multiple problems with sound correspondences, and in some cases with the semantics as well.
The foregoing shortcomings are products of a questionable analysis with regard to the reconstructed sound system, of supernumerary syllables in what most scholars regard as disyllabic bases, of imaginary morpheme boundaries and manufactured affixes, and of forced comparisons. The last of these issues borders on what might be called “errors of fact.” In a work as large as this one, it would not be surprising to find an occasional factual error. I do not want to suggest that there is an inordinate number of factual errors in this book, but there are a number that could easily have been avoided, some of which are little short of mind-boggling. Here are some of the more prominent of these:
• With reference to Bunun, W states (167) that “the language has several hundred thousand speakers.” However, the largest Formosan aboriginal language is Amis, with a speaker population probably numbering below 180,000 in 2010 (Blust 2009a:49), and statistics provided by The Council of Indigenous Peoples, Executive Yuan (Taiwan) show 45,796 ethnic Bunun in 2004, and about 34,000 speakers in 1993.
• Amis is said to lack a phonemic stop in the glottal/pharyngeal region, but to have both glottal and pharyngeal fricatives (181). In fact, the central dialect of Amis, which W has chosen to represent the language, has an epiglottal/pharyngeal stop that the writer recorded during work on the language in a Field Methods course taught at National Chengchi University in Taipei in the Spring of 2002, and which has been described in detail by Edmondson, Esling, Harris, and Huang (2005).
• In Kelabit, stress is said to be “predictably on the penult of the word unless the vowel of the penult is *e, in which case the stress is on the final syllable.” In Blust (2006:314), a publication that W (423) cites in his text, but ignores in his discussion, [End Page 570] stress is clearly transcribed on the penult even when the nucleus is schwa. This is a key to understanding the conditions for gemination that gave rise to the typologically rare and theoretically intriguing phonemic voiced aspirates in this language and the synchronic alternations between them and their plain voiced counterparts. W’s entire account of this process is garbled in ways that are difficult to comprehend.
• Separately from the series of errors found in his description of the origin of the Kelabit voiced aspirates, W claims (415) that “final vowels became voiceless.” This is untrue: voiceless final vowels and full vowels followed by -h (terminal devoicing) are clearly distinguishable, and Kelabit has the latter, not the former.
• The claim (421, footnote 21) that *t > s /__i in Kelabit is “exemplified here but is not discussed” is surprising in view of the detailed discussion in Blust (1974), and the mention of this Kelabit change in a broader context in Blust (2009a:609).
• Oceanic (Oc) specialists will be startled to read (625) that most innovations in Proto- Oceanic (POc) “are in the realm of morphology and syntax, but a couple of phonological innovations are shared by all the Oc languages: (1) the development of three labio-velar consonants, (2) the development of prenasalized consonants in certain forms.” We then read that “the merger of *p and *b and … of the voiceless stops *c, *t, and *k with their voiced counterparts *j, *d, *g” were shared by “almost all of the Oc languages.” W seems unaware that the merger of *p and *b is one of the central pieces of evidence for an Oceanic subgroup (and hence is attested in all Oceanic languages). He also seems unaware that *t and *d did not merge in POc, either in their oral grade or their nasal grade reflexes.
• Those familiar with the history of Motu will be surprised to learn (651) that etymologies such as *aku > lau ‘1sg’ and *sáŋin (SV *haŋin) > lai (incorrectly written l-aai) ‘wind’ contain a “widespread prefix … l-, which occurs only before /a/.” W again seems unaware of the relevant literature (Blust 1990), and even though he mentions parallel changes in Kei (617) and Sa ‘a (670), he fails to connect the Motu data to other cases in which a palatal glide has been added before word-initial low vowels (followed by *y > l in Motu).
• Finally, Polynesianists will surely feel a jolt when they read (701) that Tongan “is closely related to the languages of Niue and Niue-fo’ou [sic], and together with them constitutes the Eastern Polynesian subgroup of the Polynesian languages. All other Polynesian languages are in the western subgroup.” Even a cursory glance at Marck (2000), or a basic familiarity with papers written forty years ago, as Pawley (1966, 1967) would have sufficed to correct such fundamental errors.
In other cases, an error appears in one part of the book but the facts are stated correctly elsewhere, as where we are first told that Chamorro normally reflects *-aw as -o rather than the actual -aw (47), but the facts are later presented accurately (272). More could be mentioned, but I will stop with the statement (471) that “the Malay homeland is most likely eastern Kalimantan,” attributed to an unreferenced publication by Collins and Sariyan. It has been known at least since Hudson (1970) that the greatest concentration of diverse, indigenous, longhouse-dwelling animist populations with languages that are closely related to Malay is in southwest Kalimantan, an area from which southern [End Page 571] Sumatra and the Malay peninsula could have easily been settled by crossing a fairly narrow waterway.
Somewhat different from outright errors of fact are peculiarities of usage or judgment, as where W refers to the island of Borneo repeatedly as “Kalimantan,” when it is well known that this term applies only to the Indonesian part of the island, or where he describes the Thao–English dictionary of Blust (2003), which is 1,106 7″ ×10″ pages of single-spaced text in double columns, with over 4,000 sample sentences, as “a short, but very reliable dictionary” (67), and the far shorter and less detailed works by Egerod (1980) on Atayal, and Pecoraro (1980) on the Truku dialect of Seediq, as “substantial.”
8. Attention to the Literature
In more than one place in this book, W seems unaware of the relevant literature, as in the claim that Amis lacks a phonemic glottal stop, or that the historical and synchronic process of *t assibilation in Kelabit has not been described in print. In a similar vein, he seems to puzzle over the irregular transition from PAn *siwa to PMP *ciwa ‘nine’ (983), with no mention of the detailed description of this problem in Blust (1995), and he observes (418) that in Kelabit “in some cases /e/ was changed to /i/ as part of the morphological process of forming the past tense,” evidently with no inkling of the importance of ablaut in the languages of central and northern Sarawak (Blust 1997b). Likewise, in discussing the historical phonology of Tongan and Samoan, W shows no knowledge of the third palatal reflex that characterizes all Polynesian languages (Blust 1976). In other cases, one wonders whether the problem is ignorance or a pervasive underacknowledgement of primary sources. In his description of Malay, for example, W notes that “Malay’s closest relative is Cham. The Chamic languages, spoken mainly in Vietnam and Cambodia, derive from Old Cham, which was brought to mainland Southeast Asia from the Malay homeland in the first millennium bc.” Much of the content of this statement derives from Blust (1981, 1994), but this can hardly be inferred from the references cited. To make matters murkier, W states elsewhere (37) that “TB [Toba-Batak] and Malay are clearly in a subgroup as opposed to other MP [Malayo-Polynesian] languages,” and then adds alternatively that Toba Batak may be next of kin to the “Malayo-Javanic” languages, reviving a proposed genetic group that has been thoroughly discredited (Adelaar 1992).
9. Issues of Classification
Undoubtedly the most startling feature of PANPWG is W’s position on phylogenetic relationships, which represents a complete turnaround from his earlier views. On the basis of phonological, lexical, and morphological evidence, Wolff (1995) argued that there must have been “a period of common development between the Philippines and Taiwan after the eastern languages split off.” By contrast, he now accepts a Malayo-Polynesian subgroup that includes all An languages outside Taiwan, thus sharply separating the linguistic history of the Philippines from that of Taiwan with never a look backward, even in a footnote. However, it does not take the alert reader long to see what has happened. In his acknowledgements, W notes (xii) that Laurent Sagart “read almost the entire manuscript,” and in a seminar on the Sino-Austronesian hypothesis that Sagart offered at Cornell University in Spring 2008, W evidently experienced an epiphany, in which his exposure to Sino-Tibetan comparative data was “sufficient to [End Page 572] inform my understanding of any number of reconstructions and in persuading me of the relationship of the Austronesian and the Sino-Tibetan languages, and thus to shape the assumptions that underlie the analysis here presented.” This, then, is the explanation for W’s turnaround, as he has adopted the phylogenetic views advocated by Sagart (2004, 2005) in their entirety, including the Sino-Austronesian hypothesis and the higher level phylogeny of An, which claims that the first split within An is between Luilang, Pazih, and Saisiyat vs. the rest, the second between Atayalic, Thao, Favorlang, Taokas, Papora, Hoanya and the rest, the third between Siraya and the rest, the fourth between Tsouic, Paiwan, Rukai, Puyuma, Amis, Bunun and the rest, and the fifth a division into three coordinate branches, one containing Kavalan and Ketagalan, another the Tai-Kadai family, and a third the Malayo-Polynesian division of An.
Once this is understood, several puzzling statements in PANPWG become more readily intelligible. One of these is W’s concern (38–39) with denying the validity of the East Formosan subgroup proposed in Blust (1999) to link Basay, Trobiawan, Kavalan, Amis, and Siraya, since if East Formosan is accepted it presents a serious conflict with Sagart’s “higher level phylogeny,” in which Siraya, Amis, and Kavalan are split between three different higher-order branches of the An family. Despite this apparent alliance, W’s practice is inconsistent, as a strict adherence to Sagart’s views would allow reconstructed forms to be assigned to PAn only if they are attested in Luilang (extinct, and known only from short wordlists), Pazih (recently extinct, but fairly well documented), or Saisiyat, and at least one other An (or Tai-Kadai!) language (Sagart 2004:431). Since this would have drastically curtailed the size of his reconstructed lexicon, it was ignored. Other statements about subgrouping that are not required by Sagart’s model are puzzling, as W’s denial of a Philippine subgroup with no argument of any kind, despite the extensive data for such a group presented by Zorc (1986), and Blust (1991, 2005).
In his foreword to this book, Laurent Sagart spares no words of praise. Among the many virtues that he finds in it (xvi–xvii) are:
i. The sketches of sound change are distinguished by “going into much finer detail than the simplified phoneme correspondence tables that can be found in the literature.”
ii. Wolff takes into account “certain general processes, morphological or phonological, that interfere with normal phonological development, generating lexical doublets. Recognition and characterization of these processes allows him to unify forms that would otherwise be treated as independent.”
iii. “The Glossary is a trove of new cognate sets …” that “will be a major source of information to those interested in the evolution of the An lexicon.”
iv. Because W accepts Sagart’s subgrouping, he “relies less on Malayo-Polynesian evidence than does Blust in reconstructing Proto-Austronesian.” To illustrate what he regards as the beneficial consequences of relying less on Malayo-Polynesian evidence, Sagart notes that Dempwolff reconstructed *taliŋa ‘ear’ (Blust *Caliŋa). Since some Formosan languages reflect *Caŋila, “Wolff reconstructs the Proto-Austronesian word for ‘ear’ as *taŋila. He does so because even though *taŋila is not reflected in any Malayo-Polynesian language, it is in several Formosan languages… .”
v. W’s treatment of submorphemes “will, I believe, move the field forward. He argues that, far from being sound-symbolic strings of the kind of gl- in gloom, glow, gleam, [End Page 573] etc., Austronesian ‘roots’ are the lost monosyllables of Proto-Austronesian: he identifies several processes which turned Proto-Austronesian monosyllables into disyllables: adding a prothetic vowel, ‘stretching’ the nucleus, … reduplicating the root, and petrifying an affixed form. … The recognition of An roots as monosyllabic morphemes will be of the highest significance in establishing the outer connexions of Austronesian.”
With regard to point (i), it is unclear what standard of comparison Sagart has in mind. Certainly studies such as Tsuchida (1976), Zorc (1977), Geraghty (1983), Sneddon (1984), van den Berg (1991), Adelaar (1992), Ross (1996), Mead (1998), Thurgood (1999), Lynch (2001), Blust (2002, 2007b), François (2005), or Clark (2009) present analyses of historical phonology that are a far cry from “simplified phoneme correspondence tables,” and moreover are far more firmly grounded in reliable comparisons than many of W’s surprisingly speculative etymologies.
With regard to (ii) and (v), I believe I have already said enough about the reservations many will feel toward W’s general rejection of doubleting via the syllabic extension of protoforms, and toward his theoretically vacuous appeal to the “stretching” of vocalic nuclei in purported PAn monosyllables for which no generally recognized comparative support exists. In short, it hardly seems appropriate to count either of these proposals as strengths or as ways in which W’s work is “pushing the field forward.”
Point (iii) is ironic. As W himself acknowledges, his glossary is massively derivative. A tally of the first 200 entries that appear to be independent forms (cross-references and subentries complicate the count) shows that over 95 percent are taken either from Dempwolff (1938), or the ACD, with slightly more from the former source. Of the eleven forms that W proposes de novo in this set of 200 comparisons, most are problematic. W’s glossary is, therefore, hardly “a trove of new cognate sets,” as Sagart asserts; the bulk of his data can more accurately be described as a reworking of material collected by other scholars, in some cases with virtually wholesale unacknowledged borrowing, as in the note to [waŋkaŋ] ‘boat’ (see ACD).
For that matter, it is noteworthy that of the 37 languages for which sketches of the historical phonology are provided, 30 already have published descriptions from which W obviously could draw. It would be tedious to list these in full, but the only languages for which phonological histories had to be constructed more or less ex nihilo are Amis, Pamona, Salayarese, Old Javanese, Manggarai, Buru, and Kei. Nearly all of the languages used by Dempwolff (1934–38) were incorporated into W’s study, namely Tagalog, Toba Batak, Javanese, Malay, Ngaju Dayak, Malagasy, Fijian, Tongan, and Samoan, the only differences being that Dempwolff used Modern Javanese rather than Old Javanese, restricted his citation of Fijian data to Bauan (W cites both Bauan and Wayan, the latter more often), and included the Polynesian language Futunan, which W omits.
Point (iv) is equally ironic: *Caliŋa and *Caŋila are cross-referenced to one another in the ACD, and this pair is discussed in Blust (2009a:636), where it is noted that reflexes of *Caŋila (pace Sagart) are hardly restricted to Taiwan, but are also found in Inibaloi and Kallahan of the northern Philippines, in Maranao, Magindanao, and Iranun of the southern Philippines, in Irarutu/Irahutu of the Bird’s Head peninsula in Papua, in Kilivila of the Trobriand Islands, and in Mono-Alu, Banoni, and Kokota of the western Solomons. W’s [End Page 574] claim that forms like Malay or Samoan taliŋa arose because of a perception that body parts that come in pairs should be represented by lexical items that contain the plural infix *<al> is interesting, but would be far more convincing if it were supported by evidence that *<al> has been added to reflexes of *maCa ‘eye’, *qalima (but no **qima!) ‘hand’, *qaqay ‘foot, leg’, *siku ‘elbow’, *susu ‘breast’, and so on.
In summary, PANPWG is a highly personal view of the history of the An languages—one that combines a wide knowledge of the primary facts with an oddly idiosyncratic interpretation. The derivative nature of much of the material makes the book look at times like a quaint attempt to reinvent the wheel, and the treatment of the Oceanic data falls so far short of the standards of scholarship established in this area over the past 40 years (Clark ongoing; Ross, Pawley and Osmond 1998, 2003, 2008, to name just the two largest comparative studies) that one wonders why W would even consider publishing it. Moreover, it does not help matters that W’s analysis has been shaped by a conviction that An and Sino-Tibetan languages are part of a larger family—an intriguing proposal, but one that other specialists on both sides of this genetic divide have found unconvincing (Wang 1995, Blust 2009a:705–7).
John Wolff is a gifted descriptivist who has made lasting contributions to the field of An linguistics in the area of lexicography and pedagogical grammar. His Cebuano–English dictionary (Wolff 1972) is without question one of the finest examples of lexicography ever done for a Philippine language, and his multi-volume courses in Cebuano (Wolff 1966), Waray-Waray (Wolff and Wolff 1967), Pilipino (Wolff, Centeno, and Rau 1991), and Indonesian (Wolff, Oetomo, and Fietkiewicz 1992) have proven their value over the years. These are the contributions for which W undoubtedly will be remembered with gratitude by posterity, not for his historical work.
1. Many thanks to Sander Adelaar, Andy Pawley, and Malcolm Ross for feedback on an earlier version of this review. The usual disclaimers apply.
2. Wolff (2010:32) presents these purported equivalences as part of his Chart Two, in columns marked “Wolff” and “Blust.” The “standard view” is expressed, for example, by Dahl (1976:101), Zorc (1986), Ross (1992), Li (2004), and Blust (2009a). By “standard view,” I do not mean to imply complete uniformity, but all five of these scholars agree on certain key features in opposition to Wolff, including (i) the *t/*C distinction, (ii) the *j/*g distinction, and for some (Dahl, Blust, Ross with qualifications) (iii) the reality of a voiceless affricate *c. (Ross accepts this only for Proto–Malayo-Polynesian, while Dahl and Blust accept it for PAn).
3. X = not attested in this position. To accommodate W’s changes, I have added Dempwolff’s *ḍ (= *D in Dyen’s orthography, but rejected by most comparativists today). Blust (2009a) accepts *D only word-finally; other proposed voiced stops, such as *Z that are irrelevant to the point at issue here, are ignored in table 1.
5. To give a concrete idea of how facile W’s dismissal of contrary evidence often is, he claims that SV *c simply did not exist, and as a consequence he must derive Malay words like pacak ‘fixing pointed stakes in the ground’, pecah ‘breakage into bits’, or pucok ‘shoot, top branchlet’ from etyma with SV *s (his *c). The fact that this implies unconditioned phonemic splits in dozens of cognate forms across languages that distinguish SV *s and *c does not seem to weigh on his methodological conscience. Rather, in Malay (and by implication other witnesses for SV *c), “the replacement of /s/ by /c/ is due to sound-symbolism or contamination in most cases, and in a couple of cases the /c/ is in loan words from an unknown source” (484).
6. In many other languages, *kaen has reduced to kan, but this is hardly surprising, as the absorption of schwa by an adjacent vowel in bases of more than two syllables is well known, and *kaen is far more often found in affixed forms (*k<um>aen, *k<in>aen, *kaen-en, etc.) than as a bare stem.
7. Independent evidence for schwa epenthesis is seen in Kanakanabu and the Rukai languages, which added a schwa not only to break up medial clusters in reduplicated monosyllables, but also after word-final consonants to produce a uniform open-syllable canon, as in *beNbeN > Kanakanabu ta-bunəbunə, Saaroa ta-bəɬəbəɬə, or Proto-Rukai (Li 1977) *bələbələ ‘banana’.
8. It is worth observing that doubleting in An languages is not confined to lexical bases; as pointed out by Brandstetter (1916:28–32), it also affects submorphemic roots, and as noted by Blust (2001) it is a central feature of the exuberantly variable *qali/kali- prefix, which Wolff reconstructs in the single form *qaɬi-.
9. For another example of an equally arbitrary invention of morphology cf. *tageɣaŋ ‘ribs’, which W says “consists of *ta- ‘body part’ + *geɣaŋ.” There is no comparative support of any kind for this analysis, nor is there independent evidence for a body-part prefix *ta-. The true motive for this proposed morpheme division appears instead to be that *tageRaŋ presents unambiguous evidence for PAn *-g-, which W claims did not exist, and by the contrivance of making *g morpheme-initial he is able to account for it in terms of his unification of SV *g- and *-j(-).