Lexical Replacement and Cognate

This article was downloaded by: [University of Auckland Library]On: 05 January 2014, At: 20:54Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Australian Journal of LinguisticsPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/cajl20

Lexical replacement and cognateequilibrium in AustraliaBarry Alpher a & David Nash ba American University , 3218 Wisconsin Ave NW, Apt B2,Washington DC, 20016, USA E-mail:b ANU, AIATSIS E-mail:Published online: 14 Aug 2008.

To cite this article: Barry Alpher & David Nash (1999) Lexical replacement and cognateequilibrium in Australia, Australian Journal of Linguistics, 19:1, 5-56

To link to this article: http://dx.doi.org/10.1080/07268609908599573

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the“Content”) contained in the publications on our platform. However, Taylor & Francis,our agents, and our licensors make no representations or warranties whatsoeveras to the accuracy, completeness, or suitability for any purpose of the Content. Anyopinions and views expressed in this publication are the opinions and views of theauthors, and are not the views of or endorsed by Taylor & Francis. The accuracy of theContent should not be relied upon and should be independently verified with primarysources of information. Taylor and Francis shall not be liable for any losses, actions,claims, proceedings, demands, costs, expenses, damages, and other liabilitieswhatsoever or howsoever caused arising directly or indirectly in connection with, inrelation to or arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden. Terms& Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

http://www.tandfonline.com/loi/cajl20

http://dx.doi.org/10.1080/07268609908599573

http://www.tandfonline.com/page/terms-and-conditions

http://www.tandfonline.com/page/terms-and-conditions

Australian Journal of Linguistics, Vol. 19, No. 1, 1999

Lexical Replacement and CognateEquilibrium in AustraliaBARRY ALPHER AND DAVID NASH

We estimate the degree to which languages resort to borrowing as a means of lexicalreplacement, within a group of neighbouring languages of southwestern Cape YorkPeninsula, using several methods: (1) sound-correspondences and correspondence-mimicry;(2) the proportion of 'local' words in single-language lists; and (3) the creation of thevocabulary of special registers. We find that borrowing accounts for at most half of lexicalreplacement in these languages, and most usually is well below half. We demonstrate thatthis rate is crucial in the prediction of what fraction of vocabulary might in the long termbe common to two neighbouring languages (the 'equilibrium percentage') in a model oflexical similarity that does not distinguish borrowings from common retentions. We thenapply these findings to the case study, and compare determinations by lexicostatisticalsubgrouping (with and without recognition of loans), with results from classification byclassical means. We find substantial agreement, and that the effect of 'borrowing toequilibrium' on lexicostatistical subgrouping is tolerably small.

Introduction1

Loanwords, both detected and undetected, pose problems for the geneticclassification and subclassification of languages. In some cases, notably languageswith longstanding traditions of writing, like those of the Germanic family, borrow-ings over a period of a thousand years and more can be identified with precision and

1 Our collaboration on this paper began upon Alpher's presentation 'Borrowing and non-borrowing'to the Seminar on Australian Aboriginal Languages at the Massachusetts Institute of Technology,8 January 1979, which included the key ideas of Sections 2.2 and 3.1.2. We are grateful to Paul Blackfor providing us with a copy of Black (1997) in preliminary version (and his 1979 paper referred totherein) and for other useful comments, most recently in his capacity as a referee. We thank GraemeWilliams and Adolfo Constenla for kindly running J. Guy's SIMULA program and the SPSS clusteranalysis program, respectively, on our data sheets. Numerous others helped us with one or anotheraspect of the work; we thank among them Marcia and Robert Ascher, Gavan Breen, Sheila Embleton,Philip Hamilton, John Haviland, Kenneth Hale, Harold Koch, Johanna Nichols, and Bruce Rigsby.We are grateful for extensive comments from an anonymous referee. The 1982 version of this paperwas pre-circulated and presented at an annual meeting of the Australian Linguistics Society (AliceSprings, August 1984) and was further revised and expanded in December 1985 (Alpher) andFebruary 1986. Nash's work was funded in part by National Science Foundation grant numberBNS-7913950 and by grants from the Australian Institute of Aboriginal Studies and the AustralianResearch Council (No. A58932251).

0726-8602/99/010005-52 © 1999 The Australian Linguistic Society

5

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14

6 B. Alpher and D. Nash

dated. From cases like these, it is demonstrable that languages have incorporatedloanwords into their vocabularies at different rates (the rates differ from onelanguage to another over a given period and within the same language at differentperiods). Whatever the case may be as to the limits and distribution of variation ofrates of lexical replacement in a given list of 'meanings', it is clear that the fractionof lexical replacement that is accomplished by borrowing from other languages hasa frequency distribution. Among the Germanic languages, this rate, calculated withrespect to a standard 200-word list, has varied from 2% (Transylvanian Saxon, sincethe time of Old High German) to 48% (Faroese, since the time of Old Icelandic),and the rate for English (45%, since the time of Old English) during the last 1,200years or so is among the highest of those surveyed (Embleton 1986: 99; see ourTable 2).

Where there is no time-depth to the written corpus and where the normalmethods of identifying borrowings fail to give revealing results (a problem longrecognized; see for example Hale 1962; Dyen 1963: 61; Embleton 1986: 126), therate at which the lexicon is replaced by borrowing seems currently to be left out ofconsideration or to be a matter of guesswork. These conditions are notoriously thecase with the languages of the Aboriginal peoples of Australia. Athough related toone another in various degrees, many of these languages are phonologically con-servative, making the use of shared phonological innovations as a tool in languageclassification and the identification of loanwords highly problematic. The totalpopulations of speakers of these languages are rarely over 1,000 and frequentlyunder 500; marriage with a speaker of another language, and hence bi- andmultilingualism, is in some areas the norm (Sutton 1978: 106-113; Heath 1978a:14-21; to our knowledge, the first report in the literature of an extensive system ofthis kind is that of Sorensen (1967, with regard to peoples of the Amazonianrainforest)), and extensive borrowing of bound forms as well as free-form lexicalitems has been shown to occur (in northern Australia, for example, by Heath(1978a)). Nowhere to our knowledge has a quantitative characterization of borrowingunder such circumstances appeared, but the general impression given by writings onthe subject is that it is considerably more frequent than in other parts of the world.As well, the practice, common among speakers of Australian languages, of proscrib-ing the use of the name of a recently deceased person and words that resemble it hasbeen advanced as a cause of an unusually high rate of lexical replacement (whetherby borrowing or other means), again without quantitative data but with presumablydrastic effects on the usefulness of quantitative methods and indeed of the compara-tive method itself insofar as it relies on the comparison of 'cognate' forms.

The problem is of interest in areas of study other than Australia, notably withregard to the assessment of language relationships of such great time-depth that themethod of comparative reconstruction becomes difficult because of the degree ofattrition of common vocabulary. Such is the case with the indigenous languages ofthe Americas, where the time-depths at which the comparative method becomesunworkable are of the same order as the time-depth of the beginnings of agriculture.Before agriculture, the sizes of populations of speakers, the affinal relationshipsamong them, and the extent of bi- and multilingualism might very well have

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14

Lexical Replacement 7

resembled those observed in parts of Australia at present. There is another area oflong-range comparison to which prior multilingualism is considered to have been ofrelevance, in the production of unidentifiable borrowing in presumably massiveamounts (Austerlitz 1991: 363): this is the internal and external relations of theUralic languages (whose common ancestor is presumed to have been the languageof people with a hunting-and-gathering economy (Austerlitz 1991: 355)). Anothercontemporary region for which a situation of this kind has been observed isPapua New Guinea (Ross 1996: 202; the groups in question in this studypractise agriculture). Even the wildest estimate of the rate of lexical replacement byborrowing in a case like the Australian one might prove of some use in such cases.

An estimate of the degree to which lexical replacement is not totally accomplishedby borrowing in areas like Aboriginal Australia is also of relevance for general modelsof vocabulary distribution over space. If, as runs a tacit assumption of one of thequantitative models (Dixon's—1970 and 1972) of the distribution of sets of lexical'cognates' in neighbouring Australian languages, all lexical replacement is by bor-rowing and the origin of a word as a retention or a loan is always opaque,vocabularies of adjoining groups (at least, where each has just two neighbours) willresemble each other to an extent that approximates 50% over the long run. We showbelow that it is unlikely that more than half, and likely that far less than half, oflexical replacement in these languages is accomplished by borrowing and thus thatthe 50% equilibrium rate is probably never a concern in language classification. Weemphasize here that the assumption that all lexical replacement is by borrowing is animplicit premise of Dixon's model and is not one subscribed to by its proponent; itis in fact one Dixon (1980: 28, 99) has explicitly denied. Our interest here is inobtaining quantitative data and feeding it into an improved model.

How often do languages resort to borrowing as a means of lexical replacement?We attempt below, by various direct and indirect methods, to estimate this rate fora group of neighbouring languages of southwestern Cape York Peninsula. Weevaluate one of these methods, that of 'local words' (Section 2.2.2), against theotherwise inexplicably high2 lexicostatistical count of 'sames' between two languagesof this group, Yir-Yoront and Kuuk-Thaayorre. These two languages we believe tobe in a Sprachbund relationship but not to be especially closely related genetically:the genetic interrelationships of these languages are highly problematic. Using thedata obtained through the local-words method, we compare the lexicostatisticallygenerated subgrouping with subgrouping inferred on other grounds. There aregeneral implications for the use of lexicostatistical methods in genetic subgrouping,and in addition, we believe that the methodology of estimating rates of borrowinghas an intrinsic interest.

Among other conclusions, we find that the expected 'equilibrium' figures are ingeneral low enough that language classification can proceed using lexicostatistics asa pointer to a first approximation, and with one eye kept on relevant geographical

2 In hypothesizing genetic subgroups, we attach much more importance to shared morphologicalparadigms, particularly irregular ones, than to information of other types (shared phonologicalinnovations, shared vocabulary retentions).

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


facts, without undue worry about undetected borrowings. We consider the types ofsituations where lexicostatistics gives anomalous results in language subgrouping,examining in detail a case from Cape York Peninsula, and in so doing unearth somecurious and at present mysterious facts with regard to unexpectedly high rates oflexical sameness in neighbouring languages. We believe that our findings apply tothe question of geographical boundaries across which languages are related onlydistantly, with particular regard to the reality of the proposed 'Pama-Nyungan'genetic subgroup, but we do not pursue this matter in this paper.

1. Replacement Strategies

Australian Aboriginal languages, like all languages, replace vocabulary items with newwords, either shifting the meaning of the old form, limiting it to specialized uses, orforgetting it. An example from English is animal, replacing the older and nowspecialized form deer (as retained in modern Dutch dier, German Tier 'animal'). Thereasons for such replacement are numerous; one suspects that semantic specializationof the old form often triggers it. With Aboriginal languages, however, anotherphenomenon is of relevance: the death-tabooing of words. The Australian Aboriginesavoid saying the name of a recently deceased person, and avoid other words thatsound similar or otherwise recall a proscribed name. This custom has been remarkedupon by many observers for over a century (see Black 1997 for references). It hasrepeatedly been alleged as the driving force behind lexical replacement in Australianlanguages, often to the exclusion of any other mechanisms. We are aware at present,however, of only a few recorded examples of the name of a deceased together withthe words avoided after this person's death, and, in the nature of the case, eventhough a particular lexical item in a language may be identifiable as a loan, it isimpossible to tell after three or four generations have elapsed whether the motive forthe lexical replacement was death-taboo or something else.

The usual technique for avoiding a tabooed word is to say another word in itsstead. The expressions chosen to replace words homophonous with the name(or totem) of a recently deceased person are, as far as we can tell, of the followingtypes: (a) the temporary use of a synonym, from any of the languages in therepertoire of the particular community (thus perhaps from the same language orperhaps from a neighbouring language or English), (b) compounds or other newformations, (c) use of widened or metonymically or metaphorically shifted meaningsof existing (and untabooed) words (such as widening the term for 'blood' to include'water', 'liquid'), (d) use of the corresponding term from the lexicon of the auxiliary'respect language' (which, of course, has the requisite index of politeness), (e) theuse of a hand-sign, or gesture, to serve for the proscribed term, (f) the use of ahypernym (a more general term), with the extreme being a 'whatsaname' word (suchas Warlpiri nganayi), (g) use of a particular word reserved just for the purpose ofsubstituting for any tabooed name,3 and (h) borrowing.

3 Strategies (a)-(d) are mentioned by Dixon (1980: 99), (f) is discussed by Nash and Simpson(1981), and (g) by Alpher (1991: and individual entries).

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


Methods (b) and (c) are of course normal in all languages; that they (as well asborrowing) are well attested in Australian Aboriginal languages as means of lexicalreplacement is not disputed. For exemplification of the others, see Black (1997:56-58), who shows in some detail how tabooed words often find their way back intothe vocabulary after a suitable period of mourning has passed: instances of tabooingof which we are in a position to check the history have proved to be temporary. Weare therefore skeptical that death-taboo rather than other motivations has beenresponsible for lexical replacement in any large proportion of the cases.

2. Borrowing and Replacement Rates

Dixon (1972: 331) hypothesized that the increased pressure to borrow a word froma neighbouring language in the context of the Australian practice of name-avoidance(under strategy l(a)) causes the vocabulary replacement rate for the typicalAustralian language to be higher, ceteris paribus, than that of other languages of theworld. In his 1980 book surveying the Australian language family, he does not makeas strong a claim, but still sees the death-taboo as the most significant force in lexicalreplacement:

The social custom of name taboo, and the associated proscription onlexical words that have similar form, is of the utmost significance forunderstanding one of the ways in which Australian vocabularies change.Sometimes a proscribed noun or verb may come back into use within amatter of months, but it is more normal for it to be some years before thetaboo is relaxed. Often it may be replaced by a synonym from within thelanguage ... or by a newly coined compound; but more often a new wordwill be borrowed from the language of a neighbouring tribe. Even when thishas happened, the original lexeme may return to everyday use ... but inother cases it will have been completely eliminated from the language, itsplace being taken by the borrowed item. (Dixon 1980: 28)

The crucial question is: just how common is this last possibility? This is the onlypoint in this otherwise unexceptionable summary which does not receiveconfirmation in the writings of other researchers.

There are three respects in which Dixon's hypothesis can be examined:

(i) observation of particular instances of replacement (Section 2.1);(ii) estimation of replacement rates in Australia (Section 2.2);

(iii) estimation of the proportion of shared vocabulary between neighbouringlanguages presumed to be 'in equilibrium' (Section 3).

2.1. Estimates of Replacement Rate

Dixon's (1972) hypothesis is, of course, difficult to test directly, for to do so wouldrequire observing a group of languages over a long period of time, of the order ofcenturies. To test it indirectly, we need an estimate of the time-depth of the

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


break-up of a genetically related group of languages independent of glottochronol-ogy. Neither of these tests has been carried out in Australia, and it is inappropriateto put the hypothesis in such definite terms as the following:

Frequently, when an ordinary word must be dropped from speech, peopleborrow the equivalent word from a neighbouring dialect or language—apractice that accounts in part for the relatively large amount of sharedvocabulary between Australian languages. (Haviland 1979a: 210)

[...] it is certainly the case that sporadic tabooing will have been a factor inthe development and reorientation of pronominal systems in a number ofAustralian languages. (Dixon 1980: 351)

What actual evidence is there on replacement rates in Australia? Consider thefollowing commentary in O'Grady et al. (1966: 26):

Many investigators have felt that the lexical retention rate in Australia isvery low—lower than that of languages in the world generally. But Haleand O'Grady obtained a test list in Parnkalla in 1960 and checked itagainst a vocabulary published by Schurmann in 1844. The two listsshowed almost total agreement (and the few disagreements may well bedue to the fact that the Parnkalla language recorded in 1844 was represen-tative of one dialect while the Parnkalla language recorded in 1960 wasrepresentative of a slightly different dialect).

Of course, the time allowed for change in the Parnkalla vocabulary in the cited testprobably amounts to a century at most, since the language and culture were notfunctioning in the traditional manner for very long this century. Furthermore,glottochronologists hypothesized a retention rate of around 81% per millennium forthe Swadesh 200-word list (see Swadesh (1950, 1952, 1955) and Lees (1953)), andthus almost 98% per century. Note that findings of retention rates differ accordingto variations in the sizes of lists and by variations in the composition of same-sizedlists. For the Swadesh (1955) 100-word list, for example, findings of retention ratescluster around 86% per millennium. On the assumption that the list used forParnkalla in 1960 was the O'Grady 100-word list (used in this paper and reprintedin the Appendix)—and the published source does not tell us—then it should benoted that the O'Grady list and the Hale 100-word list (also in the Appendix),although modelled on the Swadesh list, differ from it in various ways that improveits usefulness in Australia ('spear [N]' is in, 'ice' is out, for example) but introduceunknown deviations in the retention rate. If, however, the predicted retention ratefor the Swadesh 100-word list of 98.5% per 100 years is anywhere near the mark forthe O'Grady and Hale lists as used in Australia, it predicts a change in aboutone or two words in the 100-word list in a century,4 and the error term in

4 An interval of 10 millennia at this rate corresponds to a shared vocabulary of about 10% (i.e. aboutthe amount shared by two nearby non-Pama-Nyungan languages), and 50 millennia to a negligiblefraction of a per cent, so there is no need to postulate a lower retention rate just to account for thesort of diversity observed in Australia.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


the prediction would usually make observations over a period as small as a centuryfairly inconclusive.

On the other hand, if the retention rate were much lower, say 50% per millen-nium, there would on average be a change in seven words5 in the 100-word listobserved at an interval of a century, or 13 words after an interval of 200 years, thelongest period we can hope to investigate in Australia.

The maximal period of 200 years is available to us in just one language—GuuguYimidhirr. Haviland gives a detailed comparison of Cook's and Banks' 1770 listswith the modern language, and finds that '[m]ost of Cook's words are completelyrecognisable today' (1974: 231), including the one personal name Cook recorded(in use in 1901, and remembered in Haviland's time—1974: 229). Alpher (1997)has compared the 77 most reliable of Cook's and Banks' items with their modernGYim equivalents and found a retention rate of 67/77 (87%) for the 200-yearperiod. The corresponding items in the English of 1770 are retained at the rate of72/77, or 94%, over the same period.6 These translate to 50% and 73% permillennium. To these figures and to the difference between them we can attachvirtually no statistical significance, since they are based on a single 77-word sampleover a single 200-year period.7 Note that the drastically lower figures derived fromthe Cook-Banks list for both languages likely result from differences from theSwades'h lists in both size and composition. The lack of certain stable items likepronouns, among other things, is bound to make them less conservative. The sameis doubtless true of the standard lists we cite in the Appendix, although for these welack an absolute time scale.

If borrowing is high in Australia, we should recognize items in Cook's list as beingsubsequently replaced with loans from the neighbouring languages. Of the 10 itemschanged in the version of the Cook-Banks list used above, a maximum of three canbe construed as loans (all from Kuku-Yalanji).

2.2. Estimate of Replacement due to Borrowing

2.2.1. Estimation of Rate of Borrowing using Sound-correspondences as Tracers.2.2.1.1. Anomalous Sound-correspondences in Two-language Lists. Investigators nor-

5 antilog (log 0.5/10) = 0.93, where log is the natural logarithm.6 The English items that we count as different between the Cook-Banks list and current usage are

beard for contemporary barb, brow for forehead, ham for thigh, lance for spear, and throttle for throat.In making these determinations, we have deliberately turned our back on philological knowledge thatwe possess for English, for which we have more than a thousand years of written texts and sourceslike the Oxford English Dictionary, on the grounds of the shallowness of our knowledge of the historyof Guugu-Yimidhirr. We grant that forehead is attested much earlier than brow as a term for 'forehead'and that spearhas a far longer history in English than lance (possibly the prototypical military analoguein Banks' time), and we concede that Banks' and Cook's speech varieties were not necessarily directlyancestral to modern colloquial Australian. But we lack knowledge of Guugu-Yimidhirr at this depth.Bergsland and Vogt (1962: 116) noted similar problems with other languages with long literaryhistories as a serious obstacle.

7 We acknowledge, however, the possibility that these figures reflect a genuine imbalance in theretention rates for the two languages.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


mally identify as loanwords those apparent cognates that have failed to undergoexpected sound-changes or which attest the sound-changes proper to anotherlanguage's history. Among the languages for which we have reasonably rich data,Yir-Yoront (YY) and Kuuk-Thaayorre (KTh)8 are an appropriate pair for usingsound-correspondences to test for loanwords in this way, in order to study the extentof borrowing. A comparison of verb paradigms argues against an especially closegenetic relationship, but their rate of vocabulary-sharing is unexpectedly high (seeSection 3.2) and they have in common certain innovations in morphophonemicalternations (one, for example, apparently due to borrowing of a stress pattern andreduction of unstressed vowels). The speakers of these languages are documented ashaving been in intense contact before the days of continuous European presence(Sharp 1958). Yet a number of sound-changes, particularly those in Yir-Yoront,have caused these languages to sound conspicuously different, even to unpractisedears, and provide a means to identify loans.

A detailed comparison of Yir-Yoront and Kuuk-Thaayorre phonological processesand vocabulary is in process (Alpher 1998). Here are the results for one set ofsound-correspondences. YY has lost nasals that preceded a homorganic stop; KThhas retained them. Twenty-six forms exhibit this correspondence. An example is YYputh, KTh punhth 'arm' (PP *punyja). YY unexpectedly attests a nasal in sixknown forms of this type; an example is YY pornt, KTh punt 'elbow' (PP *punti).We accordingly identify the latter as a likely loanword in YY. The other fivemembers of our putative list of loanwords contain an nhth cluster in YY, like wanhth'sickness', KTh wanych (PP *wanyji).9 The six anomalous pairs out of a total of 32in this set (a ratio of 19%) are by our hypothesis loanwords.

We can treat a list like this as a random collection of 'sames' from a two-languagelexical sample in which the direction of the borrowing and something about theprovenance of all the words can be known. Taking, as an example, the group of 26

8 Kuuk-Thaayorre (KTh) data are mostly from Hall (1968, 1972); Yir-Yoront (YY) data are fromAlpher (1991 and fieldnotes). Transcription of examples in all languages is in a practical orthography,with lamino-palatal stop j, trilled or tapped liquid rr, continuant r, glottal stop '. For Yir-Yoront, chand g represent a lamino-alveopalatal stop and a glottal stop, respectively, and v represents amid-central vowel, shwa. Transcriptions of the form 'karr, V for YY verbs display the imperative formof the verb followed by the ablaut-vowel that replaces the original vowel in certain tense-forms (inthis example, karr 'look (imperative)', kin 'looked'). YY, YTh, and KTh transcriptions of the form'kolhth, i' for nouns display the citation (absolutive) form followed by the thematic vowel whichfollows it in the oblique cases, and which, if it is i or u, conditions umlaut in YY (in this example,kolhth 'tail', dative kilhthi). Other abbreviations are L, local (in 'local word'); PP, Proto-Paman;PPNy, Proto-Pama-Nyungan; and as on the map (Figure 1).9 For some of these forms, there is at least a suspicion that the nasal in YY escaped reduction because

it was not originally homorganic: compare YY kunhthn, KTh kunychn 'pandanus' (Local *kunyjan)with Djambarrpuyngu gundjak, Gupapuyngu gundjalk 'pandanus' (Zorc 1986). Both of theseNortheast Arnhem Land languages contrast ndj with nydj. The possibility cannot be discounted thatNE Arnhem gundjak was borrowed from a language with the commonly attested class prefix *gun:*gun-djak; but compare Ngandi ma-gunjak 'pandanus', Nunggubuyu maguj (Heath 1978b, 1982).If the Arnhem Land forms are true cognates, the other four forms fall under suspicion, leaving uswith just one undoubted loanword out of 32.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


words discussed above, we identify in it some eight words that are 'Old' under thecriteria set out in Section 2.2.3. The remaining 18 can be taken to be 'sames' eitheras a result of retention from as yet unsubstantiated 'Old' forms, or as borrowingsthat preceded the sound-shift, or as common innovations that are no longer attestedin the other (southern) members of the subgroup (see Section 3.2). We have, then,24 new 'sames' (these 18 plus the six known loanwords), of which six, or 25%, canbe unequivocally shown by their sounds to have become 'same' by borrowing fromKuuk-Thaayorre to Yir-Yoront. From such a list (that is, without looking at otherlanguages) nothing can be learned about the sources of words that are 'New' in bothlanguages but are not 'sames'.

In lists of this type (counts of items shared in two languages, with the ratio ofnumber same by borrowing to total number of sames in the lists; see for exampleBlake (1979: 117-131)) where no judgement is given or possible concerning thedirections of borrowings and the sources of other 'sames', still less can be learnedof the degree to which innovation is accomplished by borrowing in any singlelanguage. To see why this is so, consider, for example, Heath's (1978a: 29-30; seealso Heath 1979: 405) finding, in a list of 70 nouns in Ngandi and Ritharrngu withmeanings in the 'body-part and -product' domain, of 18 'sames'. Heath claimsthat all (or at the very least all but one) of these are borrowings. The time-depthof the genetic split between Ngandi (non-Pama-Nyungan) and Ritharrngu (Pama-Nyungan) is very great; hence it can be postulated that of the 52 'unlikes', at mosta very small number are retentions from the protolanguage in either Ngandi orRitharrngu. Either language can have in this domain, then, as many as 52 (fromthe 'unlikes') plus 18 (from the 'sames', if one language was universally the donorand the other universally the borrower), or in other words the entire 70, asinnovations by means other than borrowing. In this limiting case, the otherlanguage of the pair has produced 18/70, or 26%, of its lexical replacements byborrowing. Split the difference, on the assumption that borrowing went in bothdirections, and the result is 13%. However, an unknown number of the 52'unlikes' in either language are themselves the results of borrowing from still otherlanguages (at most only a very few, by supposition, are retentions from theprotolanguage; in the limiting case of two unrelated or very distantly relatedlanguages, all the 'sames' will be loans); the effect of these, if known, would be toraise the 13% estimate by an unknown amount. On the other hand, with languagesmore closely related, such as the Cape York languages considered below, thenumber of 'sames' will include a significantly large fraction that are 'same' not byvirtue of borrowing but by common retention; the grand total of innovationsbecomes proportionately smaller, with unpredictable effects on the ratio of innova-tions by borrowing to all innovations. As can be seen from the above, if all that isgiven is a pair of word lists of equal size and of the same 'meanings' in twolanguages, together with the number of 'sames' and the number judged 'same' byvirtue of borrowing, it is not possible to infer within any narrow limits the fractionof lexical replacement that is accomplished by borrowing. We may take our figureof 13% (above), however, as not incompatible with the estimated maximum of50% obtained by other means below.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


Kuuk-Thaayorre Yir-Yoront verbsverbs

CV:thCVmhth

CVthCVnhthCVy

Figure 1. "Some possible many-to-one regular sound correspondences".

2.2.1.2. Correspondence-mimicry. We have, in the reckoning above, counted asloanwords those forms that have not undergone the sound change in question. Thepossibility exists, however, that borrowed forms are reshaped to conform withknown correspondences between the two languages. We (e.g. Alpher 1992; Koch1997; Nash 1997) call this phenomenon CORRESPONDENCE-MIMICRY, or MIMICRY.As a process it is distinct from the normal assimilation of loanwords—avoidance ofimpossible sound-sequences and sound-types.

An example from the area under discussion is Uw-Oykangand adjun 'tail',evidently borrowed from a language that continued *tjuni (cf. Proto Central NewSouth Wales *dhun 'tail' (Austin 1997: #40, 29); Black (1980: 233) gives*tju(:)n(V)). UO has lost all protolanguage initial consonants, as for example atjin'yamstick' < P P *katjin. Familiar with a-initial words as a salient feature thatdistinguishes their language from those of their neighbours, the UO have assimilated*tjuni as a loan by adding initial a (rather than by the equally plausible, but as yetunattested, method of dropping the initial consonant to produce *uri). Because—and only because—in opting for initial a they have chosen the 'wrong' form, we areable to recognize the result as a loan.

Although UO is forced to perform one of the above options to assimilate aconsonant-initial loan to its vowel-initial canon, pronounceability is not in generalthe major motivation for correspondence-mimicry. Consider the form akwertengerle(Alpher's Eastern Arrernte notes) 'a group which holds 'managerial' responsibilitiesfor country' in some Eastern Arrernte dialects (pronounced (a'kurtV|ngurlv);kwertengerle in other dialects). The form is clearly a loan from Warlpiri kurdungurlu(Nash 1982). Since Arrernte in general permits consonant-initial words, pronounce-ability cannot be the motivation for prefixing the a-. For a full treatment ofcorrespondence-mimicry, see Alpher (1992).

The frequently many-to-one nature of KTh to YY sound-correspondences sug-gests another way to estimate the contribution of borrowing to lexical sameness,should the loan have gone from YY to KTh. If KTh borrows a YY form and subjectsits shape to mimicry, there are frequently several ways (depending on the phonolog-ical category) in which the 'wrong' form will result. Because, for example, YY haslost nasals homorganic with following stops, lenited stops in the second syllables

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


of unstressed forms (as, routinely, in verbs, see Alpher 1988), and lost distinctivevowel length, correspondences like those diagrammed in Figure 1 can result(C = any consonant, V = any vowel):

To take a KTh word and mimic a YY word, it is a simple enough matter toperform the necessary changes—the word, especially if a verb or a generic noun, willsound foreign in YY without them—and the result will usually be indistinguishablefrom a common retention. But to move in the reverse direction is quite anothermatter.

Where correspondences of this type exist, and where there is independentcomparative evidence against which to check the 'correctness' of the KTh forms,one could proceed as follows: (i) identify 'bad' forms by checking a KTh formagainst the protoform and noting whether it is a regular development; (ii) count the'bad' correspondences and ascertain that they are evenly distributed amongthe possible types.

One might assume for a rough calculation that Kuuk-Thaayorre in borrowingYir-Yoront forms mimics them always and at random. The frequency of such formsis then a basis for an estimate of how much of KTh's vocabulary is borrowed fromYY. Suppose, for example, that there are 100 YY-KTh pairs of words of thesound-class illustrated above (the YY form terminating in a vowel followed by y) forwhich the Proto-Paman etymology is known. Of these, suppose that 25 of the KThforms exhibit the 'wrong' correspondence. But one in six of the YY forms in thissound-class that are borrowed by KTh will, on the assumption of correspondence-mimicry at random, exhibit the 'right' correspondence (for the wrong reason), andour figure of 25 would then represent only 5/6 of the actual number of borrowings,which will have been closer to 30 (out of 100). So this hypothetical case leads to anestimate that 30% of the 'sames' in a YY-KTh list would be the result of borrowingfrom Yir-Yoront into Kuuk-Thaayorre. The estimate could be checked byexamining correspondences in other sound-classes.

In the real world, however, the question is academic. We must suppose thatcorrespondence-mimicry takes place in only a minority of instances, and in fact wehave to date found no instance of a 'mistake' generated by correspondence-mimicryin a KTh form known to be borrowed from YY. We conclude on this criterion thatif there has been massive borrowing between these two languages, it has beenunidirectional, from KTh to YY, where its traces are largely invisible.

2.2.1.3. Phonological Conservatism. There remains the very real possibility that agreat amount of borrowing took place before, not after, the diagnostic sound-changes. This is, of course, the situation that the investigator must suspect in manyother regions of Australia, where the languages are phonologically conservative. Inorder to estimate borrowing rates, if possible, despite this handicap, we use theconcept of the LOCAL WORD, as discussed below (Section 2.2.2).

2.2.2. Estimate Based on 'Local' Words and Single-language Lists.Frequently the researcher encounters a group of cognates that are attested in a groupof more-or-less contiguous languages which do not form a subgroup, such that theattestation of these cognates is not widespread enough to suggest that the words

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


continue a protoform of the next-larger subgrouping. We designate such formsLOCAL WORDS. The pseudo-protoform reconstructable from them we write prefixedwith 'Local *'. Cognate sets with a significantly larger distribution, and theirassociated protoforms, we designate OLD WORDS. If a word's distribution is 'local',we suspect that it has been borrowed into one or more of the languages that attestit, although it is not clear which, or which language is the source. We designate aword (etymon) whose attestation (as currently known) is limited to a single languageas UNIQUE.

We are conscious that these definitions are dependent on a genetic tree model oflanguage differentiation, and also are relative to the level of (sub-)grouping underconsideration.

The notion generalizes the situation considered by Guy (1980a: 26). Guy consid-ers four languages that share forms for two different meanings in the followingpattern:

languageALPHABRAVOCHARLIEDELTA

meaning 1XXYY

meaning 2Z

wzw

On the assumption that a daughter language makes lexical innovations indepen-dently, 'and that no case of an innovation resembling an already existing form mayever occur', then in a situation such as the above (where four forms X, Y, Z, W areshared in the given pattern), there must have been two instances of borrowing(though of which form by which language we cannot immediately tell).

The number of 'local words' shared by two languages appears to decrease as thetwo languages are separated by one or more intervening languages.

It is possible to produce an estimate of how much lexical replacement is due toborrowing by counting the numbers of words in the local, old, and other categoriesin a list. Such an estimate involves the heuristic assumptions that 'local' words arein fact borrowed into at least some of the languages where they occur and that 'old'words are present in their languages as retentions rather than borrowings. Weemphasize that these and other assumptions made below are heuristic. We areinterested in the amount of the contribution of borrowing to lexical replacement,and we prefer to overestimate rather than underestimate. Of course, there are waysin which a word may be falsely identified as 'old' or 'local'. A word may owe itsgroup-wide distribution to recent diffusion and look like an 'old' word, or converselyan 'old word' may have been retained only in one area and thus be a 'local word'.Alternatively, a word may appear to be 'local' simply because its cognates in distantplaces have not come to our attention. Because we have had in the course of ourinvestigation to revise a number of 'local' words into the 'old' category, we suspectthat the latter is the case far more frequently.

Applying the notions 'local' and 'old' to a group of languages centred approxi-mately around the mouths of the Mitchell River in southwestern Cape York

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


NORTHERNTERRITORY

SOUTHAUSTRALIA

Adnyamathanha

NEW SOUTHWALES Gumbaynggir

Figure 2. Approximate relative location of languages mentioned. Abbreviations: A, Ayapathu; AL,Aghu-Laya (Kuku-Thaypan) Aim, Almura; BP, Barrow Paint; Djab, Djabugay; Djam, Djambar-rpuyngu; FI, Flinders Island; Gup, Gupapuyngu; GYim, Guugu-Yimidhirr; HR, Hann RiverAghu-Tharnggala; Kaa, Kaantju; KN, Kok-Narr; KB, Koko-Bera (Koko-Pera, Kok-Kaper); KTh,Kuuk-Thaayorre; KYak, Kuuk-Yak; KYa'u, Kuuku-Ya'u; MKul, Mayi-Kulan; MKut, Mayi-Kutuna; MTh, Mayi-Thakurti; MY, Mayi-Yapi; NP, Northern Paman; Nung, Nunggubuyu; Oik,Olkola (Olkol, Olgol); ON, Ogo-Nyjan (Ogunjan); Pak, Pakanh (Bakanh); R, Rimanggudinhma;Ump, Umpila; UO, Uw-Oykangand; WM, Wik-Mungknh (Wik-Mungkanh); Wme, Wik-Me'nh;WNg, Wik-Nganyjirr; YTh, Yirrk-Thangalkl (Yirr-Thangell); YY, Yir-Yoront (YirrqYorront).Note: Muluridji and China Camp (CC) are co-dialectal with KYal and south of it. Koko-Babongk

(KBab) is on the coast between KB and KN.

Peninsula (see Figure 2 for locations and Figure 3 for putative subgrouping), wehave used the following working criteria:

(i) A form is 'local' if (a) it is exclusively a west-coast word (Northern Pama toKukatj, or roughly from the tip of Cape York to Normanton, and extending nofurther inland than the first tier of inland languages, e.g. Ogo-Ndjan); (b) its

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


WM Pak ON AL Kurrtj KYak KTh UO Oik YY YTh KP KBab KN

Figure 3. Putative subgrouping of languages of southwestern Cape York Peninsula on the basis ofshared innovations.

distribution is west-to-east only, in a fairly narrow band; or (c) its distributionextends up the Mitchell River but not over the Dividing Range,

(ii) A form is 'old' if (a) it is attested beyond the Mitchell headwaters (Djabugay,Yidiny, and beyond); (b) its distribution extends both north (where northincludes Northern Pama—Ura and Ngkoth on the map) and east; (c) itsdistribution extends both well to the east and well to the south; (d) it is attestedin the Marie languages (a subgroup of Pama-Maric extending south to NSW);(e) it is attested in other groups south and west from Normanton, like Mayi(MKul, MKut, KTh, and MY on the map); or (f) it occurs well outside theregion: Centralia, NSW, WA, etc. These criteria should be considered rules ofthumb, and they are not mutually exclusive.

We are, in adopting these criteria, deliberately casting the net wide, in order notto underestimate borrowing by denning too few words as 'local'. Generalizing thesecriteria to something that is replicable in other regions, while bearing in mind therudimentary nature of our knowledge of Cape York subgrouping, our workingprocedure amounts to the following: 'old' forms are those which (1) occur outsidethe larger subgroup; (2) occur in widely separate regions of the larger subgroup'sarea; or (3) if they occur in a contiguous but restricted area, are attested in morethan four subgroups (out of some 10-20 in this case—see Section 3.2) of the largersubgroup, which do not lie along some known or plausible route of communication.

Examples of the application of these criteria to deduce both 'old' and 'local' (withYY at the notional centre) forms follow:

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


LOCAL

Criterion (a)—west coast only:

*nyapin 'egg' in Wik languages, KYak, KTh, YY, YTh;*punti 'elbow' in Wik-Muminh, KTh, KYak; YY, YTh (in this case phonological

criteria identify the YY and YTh forms as loans, confirming the inference madeon the basis of local-ness);

Criterion (b)—west-to-east only:

*ngata 'fish' in Wik languages, KTh, KYak, YY, YTh, Flinders Island, and BarrowPoint (replacing *kuyu);

*pirra 'leaf in KB, GYim, Mul, CC, and KYal (Oates & Oates 1964);*walmpi 'possum' in YTh, KB, UO, and Flinders Island;*patin(a) 'skin' in Wik languages, KYak, KTh, YY, YTh, UO, Oik, ON, HR, Laya;*jamal 'foot' in YY, YTh, KB, UO, Oik, ON, HR, Laya, GYim, Rimanggudinhma,

Almura;*yirrka 'speak' in YY, YTh, UO, Oik, ON, HR, GYim, CC; 'speech, language' in

YY, YTh.

Criterion (c)—west-to-east (criterion b) together with north, but not includingNorthern Pama:

*kalu 'take, carry' in Wik languages, Umpila (Harris & Ogrady 1976), KYak, KTh;*punyja 'arm' in Wik languages, Kaantju-Umpila, KYak, KTh, YY, YTh; 'elbow' in

KB;*kulan 'possum' in KYak, KTh, GYim, HR, ON, Wik-Me'nh, KYa'u (Thompson

1976);*yangan 'hair' in Wik languages, Kaantju-Umpila, KYak, KTh, Ayapathu;

Criterion (d)—up the Mitchell River and contiguous languages in a generallyeastwards direction, but not to the east coast or southeast over the Dividing Range:

*warri 'where' in YY, YTh, KB, UO, Oik, ON, HR.

Forms with very widespread cognates sometimes occur in a 'local' region with aspecialized sense. We recognize here the borrowing of meanings and categorize theform as LOCAL SEMANTIC-SHIFTED, although it is in most cases premature to judgewhich is the original and which the modified sense. Examples:

*yampa 'leaf locally in YY yap, YTh yap, Ngkoth ambamb, Uradhi yamba; 'ear' inUmpila; 'place, camp' in Bidjara and other Marie yampa. Compare *pina whosereflex pin in YY has the senses 'ear', 'leaf, and 'site'.

*kamu 'blood' locally in YY, YTh, KTh, KYak kam; Wik-Muminh, Kaantju-Umpila kamu; 'nectar' in Yidiny gamu; 'water' in Bidjara, other Marie languages,and the Mayi languages Ngawun, Mayi-Kulan, and Mayi-Yapi (Breen 1981b:190) kamu; 'tear' in Wik-Munkan mee'-kam. For an illustration of parallel

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


semantic shifting, compare YY and YTh kawn 'water'; KNarr kawung, Yugarabal(basin of Brisbane and Caboolture Rivers; Watson 1944) 'kaoun' 'blood',Nyungar (of Narrogin, WA; Douglas 1976) kawun 'wine'.

OLD:

*pungku 'knee' in Northern Pama, Wik languages, Kaantju-Umpila, KTh, KYak,YY, YTh, UO, KNarr, GYim, Warungu; possibly in Mayi punkul (Breen 1981b:108); possibly in Maranungku pingkarra (Tryon 1970);

*jipa 'liver' in Northern Pama (except Ngkoth), Kaantju-Umpila (O'Grady 1976),Almura, KTh, KYak, YY, YTh, KNarr, UO, Oik, Kurrtjar, GYim, FlindersIsland, Gunya (Marie); 'stomach' in Ngkoth and Mayi-Kutuna; 'heart' inDjabugay (Hale 1961 and nd b). We assume that the 'liver' sense is primary;

*yaka~yaki 'cut' in KTh, KYak, YY, YTh, KB, UO, Oik, ON, HR, AL, Mu, CC,KYal (Hershberger 1964); 'split' in Djirbal; poss. ake- 'cut' in Arrernte.

We consider reflexes of *jangkar 'to laugh', which would otherwise appear to belocal to Northern Pama, the Wik languages, KTh, KYak, YY, YTh, and UO, as'old' on the strength of Djaru (WA) jingkiri 'laughing', a noun-like preverb. But wepersist in listing *yangkarV 'shin' as 'local' to Wik-Munkan, KTh, KYak, YY, andYTh, because of the uncertainty of the semantic connection of Western Desertyangkarl-pa in meanings like 'hip, hip bone, thigh, buttock area'. By the same token,YY ngelqer, YTh ngelker, and KYak ngalkar 'tongue' are 'local', because the semanticconnection of Proto-Ngayarda *ngalka.ri 'liver' remains unestablished.

We are aware that there are, on the other hand, a few extremely widespreadcognate-sets with meanings in the religious or abstract area, which for all theperfection or near-perfection in their sound-correspondences may well be recent(and ubiquitous) loans. A potential case is *pijarr(a), widely continued in Cape Yorkas 'dream' (N) and as far south as Gumbaynggir, with bijaarr 'name, language'(Eades 1979: 349). We are confident, however, that such instances are far outnum-bered by the unrecognized cognates in distant places of 'local' and, most especially,'unique' forms.10

10 An example of the kind of reshuffling that follows from the recognition of previously unrecognizedcognates is that of the words for 'fire' in UO, Olgol, KTh, and KYak. Here, the pseudo-reconstruction*paathu accounts for KTh paath (ergative paathu + n) and KYak paath, and, with an otherwiseunattested sound-correspondence, for UO alh (ergative alhu +I), Olgol alh. The form *paathu is(wrongly, in light of data presented below) classed as 'subgroup' in the counts that produced thefigures cited below, because (by assumption) these four languages constitute a subgroup (C.2, aslisted in Section 3.2.1; for readers who do not accept C.2 as a subgroup, this form is—wrongly, inlight of data presented below—classed 'local'). But, subsequently to making these counts, theexistence of the Adnyamathanha (South Australia) form yalhu 'flame' (McEntee & McKenzie 1992:100; O'Grady 1979: 120) came to the authors' attention. This form strongly suggests that UO andOlgol alh continue a PPNy form *yalhu (and therefore belong in the 'old' category) and throws theKTh and KYak forms into the 'subgroup' list (i.e. 'unique' to subgroup C.2.a.i, a putative dialect-set,as listed in Section 3.2.1; for readers who do not accept this as a subgroup this cognate set becomes'local'). Parenthetically, with regard to *yalhu, the existence of a laminal lateral (lh) in an eastern

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


The category UNIQUE can at present be said merely to reflect the ignorance of theresearchers; the less well known a language is to us, the more inflated are the figuresfor this category. It is likely, of course, that further research will shift a number ofour 'unique' forms into the 'local' category rather than the 'old', although we cannotat present predict how many will go each way. And it is of course possible that someof our 'old' forms are in fact 'pseudo-old', since 'old' words are as borrowable as anyothers.

In the face of such uncertainties, we have striven to use methods that will notproduce underestimates of the rate of borrowing.

The eight languages in the area in question for which we have anything approach-ing adequate data in this regard are Wik-Munkan (WM), Kuuk-Thaayorre (KTh),Kuuk-Yak (KYak), Yir-Yoront (YY), Yirrk-Thangedl (YTh), Koko-Bera (KB),Uw-Oykangand (UO), and Ogo-Nyjan (ON). For each of these languages we haveassembled lexicostatistical lists of 100, 120, and 151 words, each inclusive of the last(see Appendix for the lists and sources).11 We have counted the forms in each list ineach of the categories in the left margin of Table 1 and calculated the fraction ineach category.

The weighted figures in Table 1 are calculated by totalling up the number oflanguages in which each 'local' word is attested, subtracting 1, dividing the result bythe original number: (n— l)/n; then summing these figures to give a total for the'local' category (compare in this regard Gleason's weighting system for numericalcomparison, the 'characteristic vocabulary index' (Gleason 1962: 28-29)). We areinterested in these weighted figures not with regard to the individual languages, butonly with regard to producing an average for all the languages. The reasoning is that,if a certain number of languages share a form, at least one of them must haveoriginated it and not borrowed it. If we knew more about the relevant subgroupingsand sound-correspondences, we could weight the figures by subgroup ratherthan individual languages; but rating by individual languages errs in the preferreddirection of overestimation.

The subgroups recognized are KTh-KYak, YY-YTh-KB-Koko-Babongk, theWik group, UO-Olk, and ON-Hann River Aghu Tharrnggala-Aghu Laya. We havenot included a Koko-Babongk (sister-dialect to KB) list in our count; hence the'unique' figures are greatly inflated (at the expense of the 'subgroup' figures) for KB,which is only much more distantly related to YY and YTh. They are similarlyinflated for UO, because we have had access only to a 100-word list of itssister-dialect Oik; similarly again for Aghu Tharrnggala and Aghu Laya, which aresister-dialects to each other and rather more distantly related to Ogo-Nyjan (ON).

footnote 10—continuedlanguage (UO) with a phonologically and semantically transparently cognate form in a south-centrallanguage (Adnyathamathanha) virtually settles the issue of the Proto-Pama-Nyungan depth ofthe apical-laminal contrast for laterals (see Dixon (1980: 485) for a discussion of the imperfect setof putative cognates for 'earth, ground, soil' in forms like walya, yalya, and, it might be added, YYyulh, a; cf. also the discussion in Alpher (1991: 301) of the YY form melh 'lily seed')-11 Copies of the lists, which incorporate the judgements summarized here with respect to cognationand 'old'-nesSj are available from the authors at http://www.anu.edu.au/linguistics/nash/aust/

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


Table 1. Composition of word lists in eight languages (all entries are percentages)

Retentions

l.OLD

Innovations2. SUBGROUP

3. LOCAL

4. IDENTIFIABLEBORROWING

5. UNIQUE

6. Innovations(SUBGROUP+LOCAL+ BORROWING+ UNIQUE)

7. Fraction ofinnovationsthat are dueto borrowing(LOCAL+ BORROWING)

8. Weightedfraction ofinnovationsthat aredue toborrowing

Sizeof list

100120151

100120151

100120151

100120151

100120151

100120151

100120151

100120151

WM

424136

201917

272728

000

111319

585964

474544

353534

KTh

444238

89

13

403840

000

81110

565862

716663

534948

KY

403936

81014

413939

221

9109

606164

726764

555149

YY

423935

212223

312928

114

4811

586165

565049

443940

YTh

434237

212223

323133

112

245

575863

595456

464243

KP

413834

333

282528

221

253334

596266

524344

332828

UO

383533

151411

252627

111

222623

646763

414045

282629

ON

313127

10108

282727

000

323238

696973

403936

262524

Mean

403835

131414

323031

111

141719

606265

555150

403737

Notes:SUBGROUP subsumes forms found exclusively in a single subgroup, whether they are singlemorphemes, new-formations, caiques with parts not cognate, or semantic-shifted. LOCALcorrespondingly subsumes morphemes, new-formations, caiques of all kinds, and semantic-shiftedforms; and UNIQUE subsumes single morphemes, new-formations, and semantic-shifted forms.

The figures in rows 1-5 were obtained by totalling up for each list the number of forms in eachcategory and dividing by the total number of 'meanings' for which there were entries. Row 6 wasobtained by summing the figures as indicated in the margin and dividing as above. Row 7 wasobtained by summing in the categories indicated and dividing by the figure used as the numeratorto obtain row 6; row 8 was obtained as row 7 was but with figures weighted as explained above.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


We defend these subgroupings below (Section 3.2), but we are confident thatdifferent subgroupings would give figures that are in general comparable to those inthe table. Where subgrouping does make a difference is in the grouping of YY andYTh with KB rather than with KTh and KYak. To reclassify in this way has theresult of shifting a large number of words common to YY, YTh, KTh, and KYakfrom the 'local' class to the 'subgroup' class and shifting the small number of wordswhich KB shares with YY and YTh from the 'subgroup' to the 'local' class. In otherwords, re-subgrouping in this case would lower the estimate of the rate of replace-ment by borrowing. Questions of the correctness of subgroupings aside, we prefernot to underestimate this figure. Estimates of the fraction of lexical replacementaccomplished by means of borrowing (Table 1, row 8) range from 55% to 26% forthe 100-word list, 51% to 25% for the 120-word list, and 49% to 24% for the151-word list. We note again that these figures overestimate the fraction due toborrowing. It is extremely unlikely that this quantity exceeds 50% in actuality in anyof these cases, and it is quite possible that the actual fractions have been in theneighbourhood of, or below, the lowest figures given.

2.2.3. Estimate Based on Respect Vocabulary.There is a part of the lexicons of Australian Aboriginal languages that providesanother means to count instances of lexical addition and their types. In the respect(or 'avoidance') vocabulary and speech-style, a special word replaces one or moreordinary words in conversation concerning (or in the presence of, or with) certainrelatives. Examination of such items yields information on the types and frequencyof processes of vocabulary-creation. Consider, for example, the respect vocabularyof Yir-Yoront and its etymological origins (this material, together with relevantforms from neighbouring languages, is listed in full as Table 12, pp. 105-107, ofAlpher (1991)). Among the 45 respect vocabulary items recorded for Yir-Yoront, wefind some 29 forms for which the Kuuk-Thaayorre respect vocabulary equivalent isrecorded; some 14-16 of these, or about 50%, are 'local' (where 'local' includeswords recorded only in YY and KTh); by hypothesis, these are loanwords in one orthe other or both languages, and for the purposes of this particular reckoning, all arecounted as loans in YY. Examples include YY lirrch, KTh rich 'spear' ( < L *tirrcha;the ordinary terms are YY kalq, KTh kirk, both < *kalka); YY nhag, KTh nhangk,WMng nhengk 'meat, animal' ( < L *nhangki; the ordinary terms are YY minh,KTh minh, WMng minh, all<*minya). One term clearly identifies itself as a loanfrom YY respect vocabulary to KTh respect vocabulary: YY yapm, KTh yapim,Bakanh yaampany12 'eye' (L *yaampany). There is one correspondence of a YYrespect term with a KTh ordinary one: YY (respect) wal 'ear' (ordinary term pin),KTh (ordinary) waal 'ear'. The rest are constructed of YY resources on one oranother metaphor: larr = ma 'go' (ordinary terms larr 'ground', ma 'tread'),latr=olhth 'lie, sit; fall' (ordinary tholhth 'fall'), thorrchonh 'dog', 'yam' (ordinarythorrchn 'hair'), nhin 'smell (N)' (ordinary motr-nhin 'sweat', ngulhthrr 'smell'); orthey are apparently reversals: yiwn 'speak' (ordinary 'tell a lie'); or they include

12 Pakanh yaampany 'eye' is erroneously spelled 'yaapany" in Alpher (1991: 107).

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


reversals within their semantic range: wallon 'put, leave it', 'hit', 'carry' (ordinarywa\l + on 'cause to go'); or are apparently unique (mur 'house', ngerrcher 'hair,head'). Other instances that we are aware of in this region of a respect-vocabularyitem that corresponds to another language's ordinary-vocabulary item are KTh(respect) ngoolpng 'blood' (Bakanh ngolpong, UO olbong 'blood', both ordinary), UO(respect) ormyjogh 'hair' (YY ordinary thorrchn 'hair'), UO (respect) owilam 'blood'(HR ordinary vghwilvni).

Comparable data exist for another part of northeast Australia, in Dixon's (1990:•50) study of the origins of items of respect vocabulary in Yidiny and in the Dyirballanguages. Like us (see below in this section), Dixon has concluded that elaboratedrespected vocabularies are quite recent in the area, and like us he has usedGuugu-Yimidhirr as a more distantly related point of comparison (1990: 8-9, 18,21), finding that its respect vocabulary is related not at all to those of Yidiny andDyirbal.13 For Yidiny respect-vocabulary roots, Dixon finds that those with cognatesin other registers and/or in other languages constitute 112 of a total of 191, or59%.14 The comparable figure for Dyirbal is 245 of 622, or 39%. These figuresaverage to 49%, remarkably close to our estimate of 50% for southwestern CapeYork Peninsula (and based on a much larger corpus), and we will accept them atthat nominal value for the sake of our argument.15

It is of course to be expected that cognates will turn up for a number of words nowconsidered unique. And, since in this area no respect-vocabulary items have beenidentified to date for which a reconstruction is possible for any subgroup of greatertime-depth than the dialect-group, newly-identified 'sames' will in all probablyincrease the numbers of 'local' words, i.e. of putative loanwords. Clearly, however,borrowing from other languages' ordinary vocabulary is not a statistically majorsource of respect vocabulary in southwestern Cape York Peninsula. What is neededif it is to be shown for other areas (the Wik area to the north is a possibility; seeJohnson 1990: 429 [1991: 213]) is a demonstration not just that it has occurred ina number of words but also that these words constitute a large proportion of somesuch fixed and short list of 'meanings' as those used here. No quantitative data areas yet available, however.

We note in passing that the only language of this area for which anythingapproaching the replacement of every ordinary lexical item takes place is UO. In YY,by contrast, many words have no separate respect equivalent: waql 'crayfish', pothn'prawn', thortm 'sand', kun-kolhth 'tail', kaqar 'moon', thaw 'mouth', etc. Theseforms are used in the respect register as is, either preceded by the generic wangal('hand [respect]' in isolation; wangal-pothn would be 'prawn [respect]') or insentences containing wangal as a particle. KTh apparently does the same.

13 A number of GYim respect-vocabulary items correspond to ordinary vocabulary items in Yidiny.14 All categories except 'derived from the everyday style of the same dialect by phonologicaldeformation' (three of 191 in Yidiny; 48 of 622 in Dyirbal) and 'no cognate known' (78 of 191 inYidiny, 329 of 622 in Dyirbal).15 Overestimating, if anything, the contribution of borrowing to vocabulary elaboration andreplacement. Note that extending Dixon's counts to include polymorphemic items would probablyincrease the ratio of items derived language internally to those not so derived.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


YY in this matter as in others apparently participates in a respect-vocabulary'pool' which centres to the north—Sharp's (1939: 256) area V. We know of just onerespect item (possibly) shared with a language to the south or east of this area: yulh'tree' ( < *yuunyja), UO onjchar. In the larger area that includes both YY and GYim,there is apparently no sharing of respect items: GYim—a language for whichHaviland (1979a and b) has provided an extensive respect lexicon—shares 23% ofordinary vocabulary16 with YY but shares not one respect item with it. It is apparentfrom this and other facts that respect vocabularies are a relatively recent creation inthis part of Australia. It is, then, proper to treat the creation of respect vocabulariesas a task of wholesale generation of new lexical items.

2.2.4. Conclusion: the contribution of borrowing to lexical replacement in these languagesis probably less than 50% at a maximum and most usually well below 50%.Estimates of the extent of the contribution of borrowing to lexical replacement rangefrom 0% (based on lack of correspondence-mimicry between Yir-Yoront andKuuk-Thaayorre; Section 2.2.1.2) to 25% (based on anomalous sound-correspon-dences between Yir-Yoront and Kuuk-Thaayorre; Section 2.2.1.1) to the somewhathigher figures 50% (based on counts of local words in respect vocabulary, and likelyan anomalously high case; Section 2.2.3) and the following (based on counts of localwords in standard lexicostatistical lists of ordinary vocabulary; Table 1):

Estimates of fraction of replacement due to borrowing (percentages)

Size100120151

of listwordswordswords

High555149

Low262524

Mean403737

We conclude that a maximum of about 50% is a fair guess for lists of about thisorder of magnitude. We can use this estimate in the equations for equilibriumpercentage presented in the next section. For Sankoffs (1973: 103) general ex-pression y + O-yO, when 0^0.5 then the equilibrium is at most 0.5 -y /2 (where yis the 'probability of chance recurrent cognation'), i.e. less than 50%. In ourrefinement of Dixon's particular model of languages along a strip (under theassumption that loans are never distinguished analytically from common retentions;see Section 3.1 and especially Section 3.1.2) the equilibrium figure is, in this range,approximately one-half of the fraction of replacement due to borrowing, that is, amaximum of about 25%. We stress that these figures are maximum estimates. Withthese compare the figures for Germanic languages in Table 2. Here the highestfigures are 48% for Faroese and 45% for English, very much comparable with ourpresumably overestimated maximum of 50%. The mean for the Germanic figures is19%, lower than all of our positive estimates for Australian Aboriginal languages,but not much lower. Note that the extreme estimate of 0% is based on lack of

15 Of the items in Hale's 100-word list, 23% are shared; 20% are shared and 'old'. The comparablefigures for O'Grady and Klokeid's list are 22% and 19%, respectively, and for our 120-word list, 21%and 19%, respectively.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


Table 2. Fraction of lexical replacements (Swadesh 200-word list; see Lees 1953) that are loans inGermanic languages (abstracted from Embleton 1986: 79, 99-101)

Language1

EnglishGermanLSTrSYiddishPennGIcelandicNorwegianDanishSwedishFrisianFaroese

Mean

Ancestor2

OEOHGOSaxOHGOHGOHGOlcelOlcelOlCelOlcelOFrisOlcel

Time3

0.981.131.131.131.131.131.081.081.081.080.631.08

N4

200200177200200200200200200200163200

C5

131159140148138144167145142146106144

Rp6

694137526256335558545756

L7

31331

1241

1114101727

RpB8

0.44930.07320.08110.01920.19350.07140.03030.20000.24140.18520.29820.48210.1937

1. Modern languages: LS, Hamburg Low Saxon; TrS, Transylvanian Saxon; PennG, PennsylvaniaGerman.

2. OE, Old English; OHG, Old High German; OSax, Old Saxon; Olcel, Old Icelandic; OFris, OldFrisian.

3. In millennia.4. N, size of list (attestations from the Swadesh 200-word list).5. C, number of true cognates (loanwords excluded).6. Rp, number of replacements: N — C.7. L, number of loans.8. RpB, fraction of replacements attributable to borrowing: L/Rp.

attestation from correspondence mimicry, and is not to be trusted because of thespeculative nature of the assumptions made and because of the preliminary nature ofthe data.

3. Equilibrium Percentages

The classifier of Australian languages often has to estimate the degree of relationshipof two neighbouring languages that have apparently undergone little or no differen-tiation by sound change. Suppose the two languages have a word in common,similar or identical in sound or meaning. It is impossible to tell whether thiscorrespondence results from common retention from a protolanguage or is aborrowing in one or both languages. Certain rough-and-ready methods of languageclassification, such as lexicostatistics as usually practised, become difficult to applyin such cases, because they are designed on the assumption that common retentionsare counted and borrowings are excluded from the count.

This problem prompted the innovation (Dixon 1970, 1972: 330-337) of alexicostatistical model with a new assumption: for a given pair of languages, allwords that are similar in sound and meaning are counted in a single category('cognate'), and no attempt is made to weed out borrowings. The interesting resultfollows that two languages in contact over a long period will approach, in their

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


vocabulary lists (whether 100-word or the entire vocabulary), an equilibrium fre-quency of 'cognate' words, in which losses of vocabulary items from one or the otherlanguage are balanced by items that become similar because one language hasborrowed them from the other. Sankoff (1972, 1973: 103) and Embleton (1986: 66)touch on the same long-run equilibrium situation. Certain consequences of ethno-historical interest obtain: if two languages have matching vocabulary items atsignificantly higher than the equilibrium rate, they have not been separate languagesfor very long; if they have matching vocabulary items at significantly lower rate thanthe equilibrium, they have not been in contact for long. Dixon (1972: 331-336) isthe only author to have nominated a specific percentage for the equilibrium, andputs it this way:

If the two dialects have been contiguous for a long enough time, they willhave about 50% vocabulary in common ... Fifty percent is an 'ideal'equilibrium figure. We would expect in practice in two contiguous dialects,that had been borrowing back and forth for sufficient time, to have between40% and 60% common vocabulary ... Considering the amount of time thataboriginal languages have been occupying Australia we should expect mostdialects to show common vocabulary percentages within these ranges; itseems that very many do so.

(It is clear that 'contiguous dialects' as used here does not mean that the two dialectsbelong to the one language.)

On the other hand, if the map of the lexicostatistical classification of O'Grady,Wurm and Hale shows nothing else, it provides numerous instances of adjacentlanguages which do share less than 40% or more than 60% common vocabulary. Forinstance, Hale calculated that the adjacent western Gulf languages Mara andYanyuwa share around 2%, and O'Grady and Klokeid (1969: 309) give 90% for theWestern Desert dialects Antikirrinya and Yankuntjatjarra.

Dixon's hypothesis has an answer for such cases, of course, namely that the timeof contiguity is too short for the equilibrium to have been established (though thereis little indication of what sort of times are required to reach equilibrium). Thus,because of 'the rather low lexical score between Yidiny and Ngadjan—29%', Dixonsuggests that these two north Queensland languages 'had been in contiguity for arelatively short time' (1977: 8), or that '[m]utual unintelligibility, coupled withtraditional pride and intertribal antagonism, is likely to produce a situation thattends to increase, rather than reduce, linguistic differences' (1976: 218).

To what extent, then, is the '40%-60% equilibrium' model applicable, given thehost of other factors acknowledged to be at work? What proportion of the contigu-ous languages in Australia with lexical scores outside the '40%-60%' range are to beexplained by factors other than length of time in contiguity, such as have beenadvanced for Yidiny and Ngadjan?

The possibility that we investigate below is that equilibrium (with a definition of'cognate' that includes loanwords) can occur, but that the equilibrium rate is notnearly as high as 50%. The consequence for ethnohistory is that neighbouringlanguages which share significantly less than 40-60% of their vocabulary can

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


nevertheless have been spoken in contiguity (in their present locations, say) duringthe greater part of the period of their linguistic separation. It is not necessarily thecase that two such languages have only recently become neighbours.

The estimate we reach, based on the data summarized in Section 2.2.4, is that theequilibrium rate is very unlikely to exceed 25%, and quite likely to have been lessthan this. Consequently, where Dixon's figure of 50% ( ± 10%) leads to the con-clusion that the per cent 'cognate' between geographically contiguous languagescannot fall below 40%, our figure of 25% (± 5%) indicates that this figure cannot fallbelow 20% (maximum estimate). Therefore, geographically contiguous languageswith 'cognation' percentages in the 20-40% range need not be assumed to havedeveloped apart and come into contact only recently, and claims as to the distanceof their relationship can be made (admittedly tentatively) on lexicostatistical groundsalone.

We are also of the belief, although we cannot demonstrate it from currently availablefigures, that language pairs like the geographically contiguous Mara and Yanyuwa, with2% common vocabulary, as mentioned above, can well have been neighbours for quitea long time and possibly for the entire period of their linguistic separation (and arerelated only very distantly, since the lexical retention rate, as discussed in Section 2.1,seems not to be to any great degree lower than that found elsewhere in the world).Such claims constitute a large part of the original (O'Grady et al. 1966) postulation ofthe 'Pama-Nyungan' and other subgroups in Australia. We believe this (i) because ourequilibrium figure, as we have been at pains to point out, is an overestimate, possiblyby a considerable amount, and (ii) because the wholesale migration of linguisticallydefined groups (as postulated in Dixon 1972: 331, 332, 1980: 239, 1991: 5-7 (massmigration of Yidiny)) is inconsistent with current anthropological understanding of thenormal Aboriginal relationship to land. Language shift by a group who maintain theirgeographical situation and land-tenure cannot of course be ruled out (and Dixon's(1991: 7) hypothesis of Yidiny intermarriage with 'Pygmoid tribes' and the adoption ofthe Yidiny language by these latter not only does not rule it out but postulates it). Butmass migration or language shift in situ is, we believe, by no means an assumption thatis necessary to account for very low cognate percentages, and language shift in situwould (all things being equal) be expected to produce an increase of shared vocabulary(persisting via a substrate effect) rather than a decrease.

3.1. Models of Borrowing and Lexical Change

Advances in lexicostatistical models have been concerned with refinements such asthe computation of error limits, allowing for variable retention rates (both with time,and across the vocabulary),17 and allowing for chance recurrent cognation—see, forinstance, Dyen et al. (1967), Sankoff (1973), and Embleton (1986).

17 Any respectable lexicostatistical method must be controlled for the interlocking effects of list sizeand 'basicness' of vocabulary. This paper (aside from the obviously less 'basic' nature of the itemsin our list items numbered 121 and higher; see the Appendix) does not deal with these effects. Weemphaize, however, that they are indeed important, and we recommend Breen (1990: 154-163) andBlack (1997: 62 et passim) for the Australian situation.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


As far as we know, Sankoff (1972) and Dixon (1970) were the first to model theeffect of borrowing on the proportion of shared vocabulary between languages in thesame region—other lexicostatistical models assume that loan words had beenidentified and 'weeded out' from calculation of cognate percentages. We know ofthese subsequent treatments: Guy (1980a: 26-27) has some advice on the treatmentof loanwords within lexicostatistical analysis; Embleton (1981, 1986) corrects,condenses, and runs some simulation tests of Sankoff s (1972) model of borrowing;and Embleton (1986) runs models incorporating a treatment of loanwords on theGermanic, Romance, and Wakashan families to compare the fidelity of lexicostatis-tically generated trees with trees generated on the basis of other evidence (forGermanic and Romance) and to explore a high-borrowing situation (Wakashan).Dixon's treatment remains the only one in which the equilibrium possibility isfocused on and a percentage proposed.

Sankoff (1973: 103), building on his 1969 dissertation, shows a general expressionfor the expected value of the proportion of unaffected meanings after time t, andremarks that 'As t gets very large, the lexicostatistic relationship approaches anequilibrium value y + 0 — yO, dependent only on borrowing and chance cognation.'(The two quantities are y the probability of chance recurrent cognation, and 0 the'borrowing probability' or the fraction of replacement attributable to borrowing(Sankoff 1973: 100), identical to our L below.) Sankoff s equilibrium expression isendorsed by Embleton (1986: 66).

We need then to explore where Dixon's reasoning differs from the generallexicostatistical model, beginning with the assumptions made in Dixon's (1972)model. First, we list assumptions that we too are prepared to entertain:

(i) the word lists used in the comparison are of the same size; in the limit, theassumption is that the two languages' lexicons are of the same size;

(ii) both languages replace the same fraction by borrowing, and the same fractionby other means;

(iii) items that become similar because both languages make identical new-formations with old cognate items are not represented in the model (ignoringthis probably has statistically minor consequences);

(iv) borrowings, even if identifiable as such by sound-changes, are counted as'cognates' along with retentions (note that the opposite has been explicitly theprocedure since the earliest literature on the subject; Lees (1953: 115) andBlack (1997: 62) for example count loans as noncognate; Guy (1980a: 27) feelsthey should be ignored; identifiable loans comprise at any rate a statisticallyminor part of our own data); and

(v) vocabulary is replaced at the same rate in both languages (the assumptioncommon to most lexicostatistical models).

These assumptions, however much they oversimplify, are consistent with the use ofthe model as a 'rough-and-ready' method of language subclassification.

Now, consider Dixon's model. We write p(i,j) for what Dixon writes as pij, viz.the proportion of vocabulary that languages I and J have in common, and reconsiderhis situation of four languages A, B, C, D in a coastal strip so that B has just A and

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


C as neighbours, and C has just B and D as neighbours. (Dixon also has an E butnever uses it. Note that a plausible geographic setting also is along a peninsula.)

3.1.1. Preferential Borrowing. The model assumes that languages B and C arereplacing their vocabulary over time, and are doing so by loans only from the twoadjacent languages (A and C for B, B and D for C). Dixon first assumes that alanguage will borrow with equal probability from both adjacent languages, and thendevotes a page to investigating that assumption.

Note that both sets of Dixon's calculations assume that p(b,d) = 0—since heassumes that the amount of vocabulary shared by C with either B or D is simply thesum of the amounts shared with B and D, respectively. This goes against hisobservation (1972: 335) that two languages with just one language separating themwould in 'equilibrium' share about 20% vocabulary. This figure of 20% is also alogically necessary minimum in the case where language C shares 70% with B and50% with D. Dixon's model does not specify for the minimum of 20% which allthree of B, C, D have in common how a form is to be replaced. By hypothesis, it isreplaced by borrowing from B or D, but this is not possible.

In general, a reasonable estimate for p(b,d) is the product of p(b,c) and p(c,d):

p(b,d) = p(b,c)-p(c,d)

In the 50% 'equilibrium' situation, then, p(b,d) would be 25%, and would range upto 35% and down to 10% in the two non-equilibrium cases considered by Dixon.Thus the expressions for borrowing trends need to be amended. Let C(b) [resp.C(d)] be the probability that C copies a replacement term from language B [resp.D]. First, consider how to calculate C(b). The replacement term must perforcecome from the vocabulary of B not shared with C—the absolute chance of this is[1 — p(b,c)]. The alternative to the replacement being from B is that it is from D(which has chance [1 - p(c,d)]), and to be clearly from B it cannot be in thecommon vocabulary of B and D (which has chance p(b,d)). Parallel reasoning forC(d) gives the pair of formulae:

C(b) = [1 -p(b,c)]/{[l -p(b,c)] + [1 -p(c,d)] -p(b,d)}

C(d) = [1 -p(c,d)]/{[l -p(b,c)] + [1 -p(c,d)] -p(b,d)}

Consider the examples investigated by Dixon, and the effect of assuming p(b,d) is20% rather than zero:

1p(b,c)

0.20.7

0.20.7

2p(c,d)

0.50.5

0.50.5

3p(b,d)

00

0.20.2

4C(b)

0.620.37

0.730.5

5C(d)

0.380.63

0.450.84

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


Note: Column 3 controls the two cases considered; columns 4 and 5 are calculatedfrom columns 1-3 by the above formulae.

These figures support our adoption of this particular simplification of Dixon's, asC(b) still exceeds C(d) when p(b,c) is low, and vice versa.

3.1.2. Reduced Borrowing. There is another effect on the equilibrium proportion ofcognates, to which Dixon's model is quite sensitive. Dixon's model assumes that aword is always replaced by borrowing (rather than by resources within the lan-guage—see the possibilities listed in Section 1), whereas it is more realistic to assumethat borrowing is resorted to in fraction L of cases; with L < 1, whereas Dixon'smodel has L = 1; our L is Sankoff s 0). Keeping Dixon's other assumptions, we candeduce that the equilibrium cognate proportion is L/2 (consistent with Dixon's50%). In other words, the equilibrium rate is half of the fraction of lexical replace-ment that is accomplished by borrowing.

We argue that we cannot apply the l40%-60% equilibrium' rule of thumb if L issignificantly less than 0.8, that is, if borrowing is not the highly dominant replace-ment strategy. (On the basis of the data from some Cape York languages consideredelsewhere in this paper, it appears that L may be somewhere between 0.25 and 0.5.Embleton (1981: 112, 1986: 79) modelled situations with Lr [her b] ranging from0 to 0.3.)

The model with L < 1 proceeds as follows: Let p'(b,c) be the fraction of B'svocabulary that is also in language C after the lapse of a fixed period of time in whicha fraction r of the vocabulary of each language has been replaced. Since the lexiconsare assumed to be equal in size (assumption (i)), p'(b,c) = p'(c,b), etc. Now,Dixon's (1972: 332, with p(b), etc. assumed to be 0.5) model may be recast as thestatement that:

p'(b,c) =p(b,c) -r-p(b,c) -r-p(b,c) +r/2 +r/2new = old - fraction lost - fraction lost + fraction + fractionshared shared by C by B borrowed by borrowed by

C from B B from C= p(b,c) + r ( l -2p(b,c))

Equilibrium, at which p'(b,c) = p(b,c), obtains when p(b,c) = 1/2, as Dixon shows.18

18 This calculation appears to assume that no word borrowed by C can be identical to a word of Bunless it was borrowed from B (and similarly for B)—in other words, that 50% of the borrowedvocabulary of each language happens to be identical to vocabulary in the other language whether ornot the other language is the proximate source of the borrowing. Further, as Paul Black (p.c., April1981) brought to our attention, the calculation also appears to assume that B and C do not lose thesame words. If word replacement is independent in the two languages, and also independent ofwhether or not the word is shared, the term 2rp (b,c) is too large, and should be reduced by theproportion of common replaced vocabulary, r2p(b,c). By the same token, the fractions B or C appearto have borrowed from each other are less than r/2, as some items borrowed by B from C (say) maybe lost by C before the end of the time period under consideration. Our tentative understanding isthat the amendments effectively cancel each other out and the equation above is not affected.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


Now, consider the modification of the above equation when a borrowing pro-portion19 L is incorporated:

p'(b,c) = p(b,c) - r-p(b,c) - r-p(b,c) + Lr/2 + Lr/2= p(b,c) - 2rp(b,c) + Lr= p(b,c)+r(L-2p(b,c))

Equilibrium, defined as before as the state in which p'(b,c) =p(b,c), occurs whenp(b,c) =172. And, just as in Dixon's model, we can see that if p(b,c) <I72, then p '(b,c)>p(b,c); conversely, if p(b,c)>L/2, p'(b,c) <p(b,c)—in other words, thepassage of time brings the cognate percentages towards the equilibrium figure of(172)* 100%, that is, an equilibrium in which the contiguous languages share thefraction L/2 of their vocabulary.

A similar result is obtained by using the somewhat more sophisticated model ofSankoff (1972), as refined by Embleton (1981, 1986). First, we need to clarify thedifferent usage of symbols for the parameters involved:

PARAMETER ALPHER and NASH SANKOFF, EMBLETONcognate fraction p(x,y) SXYreplacement rate r (including loans) r (excluding loans)borrowing rate Lr b, 0(r + b)

Note that we use r for the quantity that Sankoff and Embleton break up into their(r + b); i.e. our L is their b/(r + b).

Now the general equation (Embleton 1981: 107, [14], 1986: 73, equation 3-11),for a model incorporating borrowing between neighbouring languages, allows forany number of languages, each with a different number of neighbours. If we take thecase of their equation which applies to our example of the four languages A, B, C,D, and use our symbols for the quantities involved, the following expression resultsfor the time-derivative of p(b,c):

- 2(r - Lr)p(b,c) + (Lr/2) [2 - 4p(b,c) + p(a,c) + p(b,d)]= r(L - 2p(b,c)) + (Lr/2) [p(a,c) + p(b,d)]

At equilibrium, the derivative (rate of change of p(b,c) with time) is zero, and thus:

p(b,c) = 172(1 + [p(a,c) + p(b,d)]/2)

which is approximated by our result, viz. p(b,c) = 172, when the languages that donot border (A and C, B and D) have relatively low cognate percentages, but evenwere p(a,c) = p(b,d) = 1, at equilibrium p(b,c) would not exceed L. We can morerealistically assume that the equation for p(b,c) holds for the long-run equilibriump between any adjacent pair of languages along the modelled strip. We canreasonably assume independence between the languages and so two non-neighbour-ing languages separated by a third language have at equilibrium a common vocabu-lary fraction p2; i.e. p(a,c) =p(b,d) =p 2 at equilibrium. Thus at equilibrium the

19 That is, vocabulary replacements which result in shared forms between B and C, whether becauseof borrowing or because of identical new formations using available cognate forms.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


0.8-

0.6-

0.4-

0.2-

0.2 0.4 0.6 0.8 1.0

Figure 4. Graphs showing two models of dependency of equilibrium fraction (p) on fraction ofreplacement due to borrowing (L); p = (1 - V(l — L2))/L compared with p = L/2.

equation may be simplified to p = (L/2)(l + p2). When solved for p, this is equivalentto p = (1 — V( l — L2))/L. For L < 0.6, L/2 is a good approximation to equilibriump, as can be seen from the graph of the two quantities as in Figure 4. As the graphshows, when L = 1 (all replacement is by borrowing) this model (and Sankoffs1973:103) predicts p = 1 at equilibrium, namely merger of the erstwhile neighbour-ing languages across the entire lexicon.

As we stated, this model is of languages with two neighbours, as along a strip.When there are more than two neighbours of each language, the implication of theSankoff-Embleton general model is that equilibrium p between neighbours will beless than in the two-neighbour model. In other words, we take equilibrium p to beusually less than L/2, especially when L<0.6) .

We emphasize that this finding is more than a suggestion (contra Dixon (1997:27, fn. 14) in his comment on a draft of this paper). So long as it is recognized thatnot all lexical replacement is by borrowing (and Dixon has repeatedly concurred inthis judgement), the conclusion that the equilibrium rate must be less than 50% iscompelling. Furthermore, if it is accepted that estimates of the amount of resort toborrowing that are based upon consideration of the origins of respect vocabulary areapplicable to ordinary vocabulary as well (Dixon's and ours are about 50%; seeabove), then the conclusion is compelling that the equilibrium rate is not above25%. Moreover, the other methods of estimation that we have employed suggestthat the rate can be well under 25%.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


3.2. Lexicostatistical Subgrouping with Loans Taken into Account: the SouthwesternPaman Case

It is instructive to perform an exercise in lexicostatistical subgrouping with an eyetowards the effects of recognizing or not recognizing 'local' forms among the 'sames'.This we do below for the southwestern Cape York languages studied above (Section2.2.2), with the addition of Kok-Narr (Breen 1972; 1976 a,b). These are languagesfor which we possess the relevant data on word distribution, and whose subgroupingcan be hypothesized on independent grounds—the usual criteria of shared innova-tions in morphology and phonology. The lack of fit between any of the severalpossible lexicostatistically generated subgroupings of these languages with the sub-grouping(s) based on shared innovations constitutes a suitable 'worst-case' test forlexicostatistics. To the extent that subgroupings to be presented in this section, whichare arrived at by nonlexicostatistical means, are in future refined in the directionindicated by lexicostatistical results, our general point is strengthened.20

The classification of these languages on the basis of shared innovations is approx-imately as follows (based on Alpher 1972 and 1976, with certain modifications assuggested by Black 1980 and p .c , and with other modifications formulated for thefirst time here). A graphic version of this classification is given as Figure 3. We mustemphasize that there is very little about this classification that is not controversial tosome degree or another (the list of criterial features below is not, however, to betaken as exhaustive). But there are no variants except the lexicostatically generatedones themselves that do not come into more-or-less the same degree of conflict withthe lexicostatistically generated classifications. Furthermore, the claim most relevantto the present discussion seems fairly solid, that Kuuk-Thaayorre and Yir-Yoront donot, lexicostatistical evidence notwithstanding, constitute a genetic subgroup.

Pama-Maric subgroup (Hale 1964, 1976a,b,c nd a,b,c).A. Wik languages: Wik-Mungkanh, Pakanh, and others (Hale 1976c, nd a).B. Inland Pama: based on the imperative suffix -ng, certain vocabulary not recorded

elsewhere, like *ata- 'to look, see', and on highly complex vowel inventories(whose elaboration may or may not post-date the language split).21

1. Ogo-Nyjan—e.g. atvng 'look (imp)'.2. Aghu-Laya (Kuku-Thaypan, Aghu-Tharnggala)—e.g. tang 'look (imp)'.

2 0 Studies relating lexicostatistically determined subgroupings to those determined by other meanshave been done since the beginning of lexicostatistics in the 1950s. W e d o no t review these here ,bu t we call readers ' at tention to two recent studies of Indo-European language classification whichhave obtained excellent degrees of fit between subgroupings made according to the two differentmethods : Dyen et al. (1992) and Emble ton (1986) .21 G r o u p B (Inland Pama) is problematic because Olgol (of Chillagoe; Philip Hami l ton , p .c . , 1996),which is in the area bu t no t in this putative group, attests an imperative in -ng and because there isattestation elsewhere in this area (Yir-Yoront for one; see Alpher 1991: 17) of the addition ofword-final ng as a strictly phonological process. Th i s grouping does however receive somelexicostatistical support (from Hale 1961 and n d a); it is the same as the 'Southern P a m a ' of O 'Gradyetal. (1966:54).

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


C. Southwest Pama: based on the past perfective suffix -/ in the intransitiveconjugation(s) and in contrast with a past imperfective tense-aspect.22

1. Norman Pama (Kurrtjar, Kuthant, and Walangama; see Black 1980 and nd b).2. RR languages: based on a past perfective suffix -rr in the transitive conjugation(s)

and in contrast with a past imperfective tense-aspect (see notes 24 and 26 andremarks below on -rr).

a. M languages: past perfective tense-form of 'to see' a reflex of *nhaawa + l;common retention of a past imperfective suffix -(n)m in the transitive (L)conjugation in contrast with a past perfective tense-aspect.23

i. Kuuk-Thaayorre (past perfective of 'see' nhaaw + r, r in this context descendsregularly from *1) and Kuuk-Yak; both languages with past imperfective -m.M

ii. Uw-Oykangand (past perfective of 'see' ewa +1) and Olkola past imperfective in-nm (Alpher's notes, Sommer 1969 and 1972, and p.c. from P. Hamilton).

3. NT languages: past perfective suffix -nt, possibly an innovation consisting in theaddition of -t to the existing (conservative) *-n.25

a. Yir-Yoront and Yirrk-Thangalkl: loss of the nasal in a homorganic nasal-stopcluster.

b. Koko-Bera and Koko-Babongk (Alpher's notes, Black nd a).

22 G r o u p C (Southwestern Pama; proposed by Alpher 1972 and with the addi t ion of N o r m a n P a m aas suggested by Black 1980: 1930) is problematic . See also the remarks on Kok-Nar r in notes 2 3 ,24 , and 26 .23 Group C.2.a (Kuuk-Thaayorre and Uw-Oykangand, etc.) is among the least compelling of thesepostulated subgroups. Their common past imperfective in (n)m now appears (despite Alpher 1972:80; cf. Alpher 1990: 157) to be a common retention rather than an innovation—but in this regardnote that Kok-Narr imperfectives in -nh and -ny are also conservative, and hence their relationshipwith the putatively conservative *-(n)m is problematic.24 Grouping C.2.a.i (Kuuk-Thaayorre and Kuuk-Yak) is problemat ic because tense-paradigm datafor the latter are almost unavailable and what data there are do no t reflect some of the criterialinnovations. T h e past tense-forms of 'see ' , 'fall', 'grow' , and 'b i te ' in Kuuk-Yak are, respectively,nhakvn ( K T h nhaawr), wontvn ( K T h wontr or wantr, Koko-Bera wantal),piinhthvn ( K T h piinhthitr),andpathvn (KThpatharr) T h e complete absence from available Kuuk-Yak data (Alpher 's notes; Hall1968) of past perfectives in -rr (an innovation in this area) or -r ( < *-l; conservative in this area) infavour of -n (which is conservative in Pama-Nyungan , al though no t in the paradigms of ' see ' and'fall'—see Alpher 1990: 162 -164 ; it apparently does no t cont inue at all in K T h ) would be difficult,though not impossible, to explain away. Alternatively, it could force us to recognize a spectacularcase of massive borrowing, bringing a distantly-related language in to dialect-like similarity withKuuk-Thaayor re . In suppor t of the latter possibility are the inflated n u m b e r s of ' local ' words inKuuk-Thaayor re and Kuuk-Yak (see T a b l e 1).25 T h i s grouping (Yir-Yoront and Koko-Bera) is a mat te r of disagreement; cf. Alpher (1972: 76 -78 )on the one h a n d and Black (1980: 193) on the other. N o t e , however, that the a rguments given inAlpher (1972) include not just (i) the *-nrpas t ending bu t also (ii) the existence o f ' s t r o n g ' versus'weak ' verb conjugations (the former with disyllabic forms of the root al ternat ing with disyllabic ones)with corresponding cognate verbs and (iii) lowering of sequences of two high vowels to mid incorresponding contexts. See Alpher (1972) for details and more explication; the au thor stands byhis conclusions.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


4. Other: Kok-Narr: past perfective in -rr contrasting with conservative pastimperfective in -nh or -ny.26

The groupings that are nonproblematic here are C.I (Norman Pama; fide Black1980: 191-194), C.2.a.ii (Uw-Oykangand and Olgol; dialects of a single language),C.3.a (Yir-Yoront and Yirr-Thangalkl; dialects), and C.3.b (Koko-Bera and Koko-Babongk; dialects). With regard to problems with various less-compelling subgroup-ings, see the appropriate notes. Most problematic of all are groups C.2.a(Kuuk-Thaayorre and Uw-Oykangand; see note 23) and C.2 (see remarks on -rrbelow).

The Pama-Maric grouping itself has never received published justification. Wetake the composition of Pama-Maric, for the purposes of the argument, to be thesame as that listed in O'Grady et al. (1966: 51-54), with the exception (exclusion)of the 'Yara' subgroup (Dyirbal, Warrgamay, Nyawaygi, and other languages of thesouthern rainforest area), and with 'Gulf Pama' defined as 'Norman Pama' as above(group C.I). The positions of Kukatj and the Mayi languages, spoken to thesouthwest of Normanton, remain problematic but do not affect present conclusions.

Subgroups A-C of Pama-Maric as postulated here are three out of an estimated10-25 subgroups within this larger group. The exact number and composition ofthese subgroups have not been the subject of an exhaustive piece of research, but weare assuming that it will turn out to be of this general order of magnitude; thenumber is of relevance in our working definition of 'local words' (Section 2.2.2).

The *-l and *-rr past perfective endings on the basis of which groups C and C.2(respectively) are hypothesized have apparent cognates outside their respectivegroups. There is an -l imperfective in Kukatj contrasting with perfectives in -n and-nh (Breen 1976a: 159); there is a -la, designated pluperfect and in contrast withpasts in -na and -nya, in Umpila (O'Grady 1976: 193, 199); and there is a -la pastin the Marie languages Warungu (Tsunoda 1974), Bidjara (Breen 1973), andMargany-Gunya (Breen 1981a). By the same token, Umbuygamu (Sommer 1976:21) contrasts a tense in -rr designated pluperfect with a past in -n; note also theproblems with Kuk-Narr -rr mentioned above (and it is quite possible that *-rr is theconservative ending in languages of this area, in contrast to an innovative *nt). Butthe criterial value of *-l and *-rr for distinguishing C and C.2 as subgroups lies intheir being conjugation-specific and having past perfective value in contrast withpast imperfective.

Although it has been claimed (Merlan 1982: xii-xiii; Thomason & Kaufman1988/1991: 18 and Section 2.1 passim) that verb paradigms, and irregular ones atthat, are not immune from borrowing between fairly distantly-related languages, we

26 Logically, this belongs with the 'RR' languages, but there are problems with the interpretation ofthe -rr endings in the several languages in which they occur (see below), and for that matter thereis a problem in the apparent absence of a past perfective in *-l in Kok-Narr; the past perfective -ngis an apparent innovation which is assumed to have replaced it. Kok-Narr appears to group with theSouthwest Paman languages on other grounds, but it is here placed with them in a 'flat' groupingas a matter of caution.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


WM KTh KYakUO YY YTh KP KN ON

Figure 5. Subtree from Figure 3, with languages of Table 3.

believe that these particular suffix distributions are not susceptible of this interpret-ation. Take as an example the Yir-Yoront past perfect in the L conjugation, whichis zero accompanied by vowel ablaut in certain verbs, like puy 'bit' (nonpast pay +1).This can be shown (see Alpher 1972, 1989, 1990, 1991: 119) to have replaced anearlier *paya + r, itself cognate with Yirrk-Thangalkl payd +1 and Koko-Berapathent, both of which continue Proto-Pama-Nyungan *patja 'bite' and the pastperfective ending *-nt(V). But the modern reflexes of *-nt(V) are phoneticallysufficiently different to make borrowing a highly unlikely explanation. Note thatsocial closeness including a Sprachbund relationship, whose importance in theborrowing of morphological systems is stressed by Thomason and Kaufman (1988/1991: 13-34), is attested between Yir-Yoront and Kuuk-Thaayorre (Sharp 1934,1958) but not between Yir-Yoront and Koko-Bera.

We now pass to the classifications suggested by lexicostatistics. The material forlexicostatistical subgrouping is the numerical information contained in Table 3.

The 'correct' subgrouping of these nine languages, based on the 'Pama-Maricsubgroup' as detailed above, is as shown in Figure 5 (a subset of the tree shown inFigure 3).

We applied lexicostatistical methods—such as the n-way splitting algorithmapplied to unrecomputed linear-correlation coefficients, favoured by Guy (1980a:38)—to the data of Table 3 to produce subgrouping trees. A subgrouping producedby this algorithm and based on a count of all lexical 'sames' in two-language lists,regardless of source (i.e. derived from the figures in the right-hand column ofTable 3) yields the classification shown in Figure 6.

A subgrouping made by the same algorithm but based on a count which excludesputative borrowings (derived in this case from the figures in the middle column['old' + 'subgroup'] in Table 3) yields a slightly different grouping (Figure 7).

From both calculations there emerges a close relationship between Yir-Yorontand Yirrk-Thangalkl on the one hand and Kuuk-Thaayorre and Kuuk-Yak on the

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14

CO

Table 3. Fraction cognate. Left column: per cent of sames retained in the OLD category; middle column: per cent of sames retained in the OLD and

SUBGROUP categories; right column: per cent of sames in list regardless of source P

List KTh KY YY YTh KB UO ON KN 1?

100 30 30 40 27 27 37 27 27 34 28 28 33 25 25 25 25 25 27 19 19 22 19 19 19 §

120 W M151100120151100120151

100120151

100120151

100120151

100120151

100120151

30 30 4029 29 4125 25 38

KTh

272724

373634

272724

454546

KY

373836

807881

272522

333026

312926

272522

333026

312926

YY

343431

545046

545150

282624

323027

323028

444135

282624

323027

323028

636157

YTh

333331

514848

545251

969289

252422

252422

222221

262420

262521

252422

252422

222221

282522

272623

KB

252422

282725

252524

322926

333128

252121

302825

302826

292623

292724

252423

252121

302825

302826

292623

292724

252423

UO

272624

343129

333230

343127

353330

282729

191916

181816

181816

232118

222219

201918

191918

191916

181816

181816

232118

222219

201918

191918

ON

222219

212119

212220

272621

282823

242322

363433

192017

191918

171817

201916

191916

242321

171717

121313

192017

191918

171817

201916

191916

242321

171717

121313KN

192119

202120

182010

222218

212220

383534

171818

131516

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


WM KTh KYakYY YTh KP UO ON KN

Figure 6. Subgrouping based on a count of all lexical "sames" in two-language lists, regardless ofsource (i.e. derived from the figures in the right-hand column of Table 3).

other, a relationship which finds no support in the comparison of verb paradigms.From neither calculation does the close relationship between Yir-Yoront and Yirrk-Thangalkl on the one hand and Koko-Bera on the other emerge. This linkage ofYY-YTh with KTh-KYak (with the exclusion of KB) is in fact the subgrouping of

WM KTh Kyak YY YTh KP UO ON KN

Figure 7. Subgrouping based on a count that excludes putative borrowings.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


these three sets of languages that was postulated in Sommer (1969) and againin Wurm (1972). Exclusion of putative borrowings (the second calculation) hasa few desirable effects, however: it eliminates pseudo-subgroups linkingWik-Mungkan with Kuuk-Thaayorre and other languages to the south, andlinking Uw-Oy-ka-ngand with Ogo-Nyjan; and it (correctly, as we believe) suggeststhat Koko-Bera and Uw-Oykangand group with Kuuk-Thaayorre, Kuuk-Yak,Yir-Yoront, and Yirrk-Thangalkl but not with Wik-Mungkan.

Taken as a whole, none the less, neither of these lexicostatistically-derivedsubgroupings is conspicuously successful. With regard to inferences made from thedata in Table 3, it is of course possible to use algorithms other than the ones (Guy'sLXSTAT and SIMULA) used to produce the trees shown above. We were able toobtain cluster analyses of the nine lists of percentages shown in Table 3 (SPSS-XRelease 3.1, run 17 May 1989 on an IBM 3081KX computer); some of the resultingtrees are shown as Table 4.

With nine languages, there are 84 possible sets that consist of three languageseach, and within each of these sets the three languages have some Stammbaumrelationship. Each such group of three is susceptible of four different subgroupings:(AB)C, A(BC), B(AC), or the 'flat' (ABC). We compared the lexicostatisticallygenerated subgrouping (calculated here on the 120-word lists), for each of thesetriplets, with the subgrouping hypothesized on the basis of shared innovations (byassumption the 'correct' one), and scored the match as correct or incorrect. Theresulting 'fidelity' scores, expressed as the number 'correct' out of 84 (by definition,84 for the 'correct' subgrouping),27 are 32 and 45, respectively, for the 'all sames'and 'old + subgroup' groupings produced by Guy's algorithm (above) and are asgiven in Table 4 for those produced by cluster analysis from the 120-word lists. Thehighest score is the 45 (54%) for 'old + subgroup' (Table 4C) by cluster analysis,identical to the 45 (54%) obtained with Guy's algorithm for the same data.

If, however, Wik-Mungkan is removed from consideration, the apparent relativemerits of the methods change considerably. Out of 56 possible, the scores for thetwo subgroupings obtained by Guy's algorithm and the three in Table 4A-C are,respectively, 19, 29, 18, 31, and 40 (71%). This last is significantly better than 29(52%), the highest obtained with the other algorithm. And it is reasonable toexclude Wik-Mungkan, on the grounds that it is a dialectally complex language andits word list has been assembled from different sources, and that furthermore morelanguage territories intervene between it and its nearest neighbour in the sample(KTh) than between any other 'adjacent' members of the sample. The exclusion ofWik-Mungkan also results in an improvement of fidelity scores calculated by asecond method, which is described below.

For a statistical method applied to a known 'worst case', 71% is perhaps not toobad. Note however that this is by virtue of considerable manipulation; the methodis hardly a 'rough and ready' one. The scores obtained with both algorithms show

27 This is itself a rough-and-ready measure of the similarity of trees, highly impracticable beyondseven languages without a computer. For a more sophisticated measure of tree similarity, which takesinto account genetic distance as well as tree topology, see Embleton (1986: 80-93).

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


Table 4. Subgroupings generated by cluster analysis: dendrograms using averagelinkage between groups.a The 'fidelity' to the presumed 'correct' subgrouping is

expressed as the number of triads 'correct' (out of a possible 84)

A. 120-wordYYYThKThKYWMKBKNUOON

B. 120-wordYYYThKThKYUOWMKBONKN

C. 120-wordYYYThKThKYUOWMKBKNON

D. 151-wordYYYThKThKYUOWMKBKNON

list, all sames111112222

list, 'old' only111111112

list, 'old' and

2

list, 'old' and111111112

(right columns of Table 3); fidelity = 281

>

22

1122

(left columns); fidelity = 4111

2

'subgroup'11111122

'subgroup'11111122

2

(middle112222

(middle112223

111112

columns);

1112

columns)

112

11112

fidelity = 45

112

1122

a The program produced dendrograms of three other types as well: (i) average linkagewithin groups (generally less felicitous than the others); (ii) single linkage; and (iii)complete linkage. The representations here do not show an important part of thesedendrograms, the genetic distance (as shown by length of tree branches).

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


Table 5. 120-word list, per cent of 'sames' in the OLD and SUBGROUPcategories (subsets of Table 3—the figures in the centre of each cell)

a. YIR-YORONT 61 25YIRR-THANGALKL 26

KOKO-BERA'Correct' grouping: (YY YTh) KB.Score: Correct; statistics decisive.

b. KUUK-THAAYORRE 30 24YIR-YORONT 25

KOKO-BERA'Correct' grouping: KTh (YY KB).Score: Incorrect; statistics moderately decisive in their incorrectness.

c. WIK-MUNGKAN 19 20OGO-NYJAN 13

KUK-NARR'Correct' grouping: (WMng ON KN).Score: Correct; statistics given the benefit of the doubt.

significant improvement with an increase of stringency of conditions, and thereforeof labour-intensiveness, for counting a pair as 'cognate'.

But in this gradient lies another point of interest (and here, the discrepanciesbetween the lists appear more orderly under cluster analysis than they are underGuy's algorithms). The subgrouping based on data collected under the least strin-gent conditions (Table 4A) is patently 'areaP in nature: KB, KN, ON, and KN aresouthern languages and the rest are northern. Within the southern group, OU andON are inland and KB and KN are coastal. Within the northern group, the majorsplit is between WM, which is inland, and the rest, whose territories are coastal orestuarial.28 A point should be noted with regard to the 151-word subgrouping(Table 4D): although its placement of Wik-Mungkan (not as a sister of KTh-KYak-UO but as a sister of YY-YTh as well as KTh-KYak-UO in a three-way split) ismarginally better than the one based on the comparable 120-word list (is less'areal'), it is known in general (Black 1997; Embleton 1986: 53 and references) thatbeyond some (as-yet-undetermined) maximum longer word lists are more likely toshow an 'areal' effect than shorter ones (other things being equal).

One virtue of all the algorithmic methods whose results are shown above and inSection 3.2.2 is that they adjust statistical anomalies among various triads as judgedone-by-one for a 'best fit'. We did, however, attempt a test of fidelity to the 'correct'subgrouping by comparison of the 'correct' grouping of each triad with the groupingas read directly from the percentages in a matrix for each set of three. In judgements ofcorrectness (in the absence of a convincing test of statistical significance), thefollowing rules of thumb were taken as indicating a 'flat' grouping: (i) if the totalpoint spread was 10 or less and there was no pattern of two low (and similar) and

28 These areal dichotomies are in part, but only in part, artifacts of the criteria for local and old wordsset forth in Section 2.2—notably the west coastal versus inland criterion (a).

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


Table 6. 'Sames' among Yir-Yoront, Yidiny, andDyirbal. Hale's 100-word list; no weeding out ofputative borrowings; all figures are percentages

YIR-YORONT 17 15YIDINY 18

DYIRBAL

one high, and (ii) if two scores were high and one low (one language appears to bevery closely related to each of two others, but these two others do not appear tobe closely related between themselves). Although this procedure is in principleconstructible as an algorithm, in actual practice in the absence of a computerizedroutine there is a vulnerability to subjective judgements, and the data below shouldbe read with this problem in mind. Table 5 displays some examples of the individualjudgements.

The results of the survey of all 84 such triplets judged against the trees computedwith Guy's algorithm are as follows:

(i) Comparison counting all 'sames' regardless of origin (the right-hand column ofTable 3): 32 of 84 (38%; 20/56 or 36% if Wik-Mungkan is excluded) correct ifKuk-Narr is classed with Southwest Pama as above; 38/84 correct if Kuk-Narris assumed to comprise a separate subgroup within Pama-Maric; 38/84 correctif Kuk-Narr is assumed to comprise a separate subgroup within Pama-Maricand if Kuuk-Thaayorre and Uw-Oykangand are assumed not to constitute asubgroup.

(ii) Comparison counting 'sames' from which putative borrowings have been elim-inated (the centre column of Table 3): 40 of 84 (47%; 28/56 or 50% ifWik-Mungkan is excluded) correct if Kuk-Narr is classed with Southwest Pamaas above; 54/84 correct if Kuk-Narr is assumed to comprise a separate subgroupwithin Pama-Maric; 58/84 correct if Kuk-Narr is assumed to comprise aseparate subgroup within Pama-Maric and if Kuuk-Thayorre and Uw-Oykan-gand are assumed not to constitute a subgroup. The greatest success rate amongall these options is 58/84 or 69%, not an encouraging score in light of thenumber of 'correct' subgrouping assumptions that had to be dropped.

The corresponding calculations, run against the findings by cluster analysis shownin Table 4, are (i) for the 'all sames' column of Table 3, 31/84 (37%; 20/56 or 36%if Wik-Mungkan is excluded), and (ii) for the 'old' + 'subgroup' cloumn, 43/84(51%; 38/56 or 68% if Wik-Mungkan is excluded).

Let us have a brief look at where it succeeds and where it fails. It succeeds wheretwo of the three languages compared are related as close or distant dialects and thethird is not. It fails at the intermediate level where groups hypothesized on othergrounds tend to be controversial—to be sure, one of the reasons why these hypoth-esized groupings are controversial is that they are in conflict with lexicostatisticalresults. The method also fails in at least certain cases where 'sames' are countedacross major genetic boundaries; here too the possible statistical anomalies invited

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


by the use of individual triads of languages are a problem, but the comparisons cannone the less be of interest. Failures of this kind have been observed and commentedon for some time (for example, Embleton 1986: 94-95).

Consider the comparison displayed in Table 6 among Yir-Yoront, Yidiny, andDyirbal, for which grammatical considerations29 suggest that the first two are relatedto each other more closely than either is to Dyirbal, although the last two are spokenin neighbouring countries, and numerous intervening languages and great distanceseparate both from Yir-Yoront.

There is no way to read statistical significance into the difference between 17%(Yir-Yoront/Yidiny) and 15% (Yir-Yoront/Dyirbal). All of these shared items are(almost by definition) in the 'old' category.

Two facts of interest emerge here. The first is that what Yir-Yoront shares withYidiny does not significantly exceed what it shares with the putatively less closely-related Dyirbal. For this fact, no obvious interpretation suggests itself other thana reevaluation of the degree of relationship of Dyirbal to the Paman languages.The second fact of interest that emerges from this comparison is that 17 of the 18items shared by Yidiny and Dyirbal are in the 'old' category. They are reflexes of*tirra 'tooth', *jarra 'thigh', *pungku 'knee', *jina 'foot', *jana 'stand', *nyina 'sit',*paja 'bite', *pa(r)na 'water', *kuta(ka) 'dog', *yuku 'tree', *kungkarr 'north',*wanyu 'who', *wanyja 'where', *ngayu/*ngaju T , *nyuntu 'you (sg)', *ngali 'we(2 incl)', and *nyupul(a) 'you (2)'. To be sure, a few of these 'old' pairs, such asthe first and second person singular pronouns (Yidiny ngaya, Dyirbal ngaja T ;Yidiny nyuntu, Dyirbal nginda 'you') count as 'cognate' by virtue of certainarbitrary assumptions, although they in fact contain noncognate morphology andtheir sameness cannot in all probability be a matter of borrowing. But most ofthese 'old' pairs do not differ from each other enough to rule out a claim that theyare loanwords in one or both of the languages. Only one 'local' word is commonto both lists: kupu 'leaf.

The implications are either (i) that at least some of these 17 'sames' are only'pseudo-old' (see Section 2.2.2)—that they do in this case result from borrowing,evidence for their very wide distribution notwithstanding—or (ii) that there is aprocess at work in situations of prolonged language contact that works againstdifferential loss. This latter possibility is an attractive one in accounting for theanomalously high rate of sharing between Yir-Yoront and Kuuk-Thaayorre. Onemight hypothesize it to work as follows, for languages A and B which belong to

29 Here are a couple of examples. The Dyirbal dialects Jirrbal and Mamu have a future-tense ending-ny that has apparent cognates in coastal languages to the south but not in ('Paman') languages tothe north and west; and a third dialect of Dyirbal, Girramay, has a future ending -djay with noapparent cognates in languages to the north and west (Dixon 1972: 55, 1977: 207). The Dyirbaldialects have an elaborated set of demonstratives differentiated for four noun classes; the absolutivecase-forms of these have no apparent cognates in languages to the north and west (Dixon 1972:44-47, 1977: 186-194). Of course, more evidence is needed, especially of innovations in Paman,but we are operating here on the basis of Dixon's conclusion: 'Dyirbal and Yidiny are totallydifferent languages. Although both belong to the Pama-Nyungan subgroup of the Australian family,in their grammars they are as different as any two Pama-Nyungan languages.'

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


Table 7. 'Sames' among Yir-Yoront, Margany, and Wankumara(Galali): Hale's 100-word list: no weeding out of putative

borrowings] all figures are percentages

YIR-YORONT 16 8MARGANY 18

WANKUMARA (GALALI)

adjoining countries, whose speakers regularly intermarry, and for which there arenumerous bilinguals. Languages A and B share form X; an event, possibly a death,occurs in group A such that X will not be spoken. X is differentially avoided ingroups A and B (it is used more frequently in B, in certain social contexts). Speakersof A who are fully bilingual in A and B never forget X, think of it as a word that isthe same in both languages, and reinstate it at the end of the mourning period(something like this, with a standard language like High German or Latin as thebasis of continuity, is discussed under the term 'prevented loss' by Embleton (1986:102, 140, 146); Bergsland and Vogt's discussion (1962: 128-129) of the loweredretention rates for one of the various eastern dialects of Eskimo (all groups of whompractise death-taboo) that was spoken and transmitted in isolation from the otherswould appear to support this reasoning). Alternatively, X is replaced in bothlanguages with a similar caique, with cognate words semantic-shifted in the sameway, or with the same form borrowed from a particular third language—the lattercase giving rise to an inflated count of 'local' words.

It is clear that problems of this kind have been a serious and explicit concern ofglottochronologists from the beginning. Here is Swadesh (1950: 159-160):

... if two languages have been in contact for all or part of the time that haselapsed since their common period, they will have exerted an influence oneach other. If one of the two languages displaces a word of the originalstock, the second language may imitate the displacement or it may eventu-ally cause the first language to return to the original form. Since theseinfluences may be either in the direction of promoting or of retardingchange, the trends may cancel each other out. The total percentage ofchange may be the same as in the case of a single language out of contactwith related languages, but the two languages will tend to stay together. Inconsequence they will diverge from each other less than the maximumprobable amount under the formula as we have given it.

As a second case of three languages separated by either great geographicaldistance or a major genetic boundary, consider the relationship of Yir-Yoront, itssouthernmost relative in the Pama-Maric family Margany, and Margany's distantly-related southern neighbour Wangkumara (Garlali) (McDonald and Wurm 1979).The statistics are displayed in Table 7.

Here the relative amounts Yir-Yoront shares with the other two languages are asexpected, but the tally for the neighbouring Margany and Wangkumara is inflated.The figure of 18% represents 16 items out of the 87 in Hale's 100-item list for whichentries were recorded in both languages. Of these 16, some 10 are 'old': *jina 'foot',

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


*taaku 'ground, earth' (questionable), *mara 'hand', *nga- T , *jaa- 'mouth', *nyaa-'see', *jalany- 'tongue', *nyi- 'you (Sg)', *nyu-/*nyi- 'he', and *ngali 'we (Du Incl)'

Whatever the explanation for the inflated 'sames' counts, the user of lexicostatis-tical subgrouping as a rough-and-ready method needs to beware. Certain rules-of-thumb suggest themselves as ways to avoid some of the pitfalls. If for example (asin the cases illustrated in Tables 6 and 7) three languages A, B, and C all shareequally and in the low range, and if B adjoins C while A is geographically distantfrom both, then the 'sames' count for AB and for AC will be given greater weightthan that for BC. A more generalized use of this rule is the recognition of chains ofgeographically contiguous languages with cognate densities that gradually decreasewith the number of intervening languages; this procedure is stated explicitly inO'Grady's treatment of the Ngayarda subgroup (1966: 73-74); see also Swadesh'sdiscussion of the mesh principle (1962: 7-14). Recall in this regard that the figuresfor any given single triad of languages are apt to represent a statistical 'bump' andthat a sample of more languages, especially where gradients of cognate density withgeographical distance are encountered, is desirable. A second rule of thumb is, ofcourse, to weed out obvious borrowings. Another principle is to avoid the use ofdifferent lists and of different-sized lists, or to make allowances for the discrepanciesthey introduce. A cursory look at morphology and phonology can help to settle someof the remaining problems.

3.2.2. Past Practice of Classification by Lexicostatistics of Australian Languages. Theuse of lexicostatistics for subgrouping was at its peak in the 1960s, when it wasapplied to word lists of a few hundred words in various language families with manypresent-day languages in which very little historical-comparative work had beendone. These families included Austronesian and Australian. For many reasons, themethod in its rough-and-ready form proved to be unreliable in anything but its grosssubgrouping. Subsequently, Guy (1980a, b) proposed some improved methods ofsubgrouping, some which still use the table of cognate percentages as the basic data.Further, some attention has been given to the effect that borrowing has, with a viewto overcoming Dixon's (1972) dismissal, at least for Australian languages, of themethod, even at the coarser levels of subgrouping.

Dixon (1980: xiv) advances 'the beginnings of a proof that all the languages ofAustralia (except two or three northern tongues such as Tiwi and Djingili) aregenetically related', even though 'present knowledge of the relationships betweenlanguages is not sufficient to justify any sort of fully articulated 'family tree' model'(1980: 264-265). He proceeds by considering the systems of phonology, nominalcase inflection, pronouns and verbal conjugations, usefully complementing previousresults based on word lists. His claim that 'little should be inferred about thesubgrouping of languages in Australia from vocabulary comparisons' (1980: 254) is,we believe, overstated, and is subjected to empirical evaluation above.

It seems evident that a number of considerations in addition to uninterpreted rawlexicostatistical data went into the classification proposed by O'Grady et al. (1966;see especially pp. 22-25; see also O'Grady & Klokeid 1969: 301-302 and 308-309and O'Grady 1966: 73-74). Although it wrongly included Dyirbal and certain of its

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


neighbours in the Pama-Maric group, it did (correctly, we believe) group Koko-Berawith Yir-Yoront, as 'Western Pama'—a result not achievable with raw lexicostatis-tics. Note also the OGW's disregard of raw lexicostatistical data (1966: 114 and119) in their classing of N(h)anda with the Kardu languages (whatever the correctgrouping of N(h)anda may be—note that Blevins (forthcoming) groups it yet a thirdway—what is under consideration here is the methods actually used by O'Gradyet ah). And we must emphasize that the Pama-Maric case illustrated in Tables 3and 5 is indeed a 'worst case'. Numerous other cases give the 'correct' results in astraightforward manner, notably where the genetic boundaries in question are notshallow ones. Such a case is that of the Arandic languages (Hale 1962).

4. Conclusion

Languages in contact spoken by small populations for whom marriage with speakersof other languages has been frequent and even normal pose special problems forlanguage-historical inference, especially in the area of loanwords. With the aims ofquantifying the contribution of borrowing to lexical change in a situation of this kindand of estimating the significance of shared vocabulary in cases where loanwordscannot be identified with certainty, we have examined a group of AustralianAboriginal languages. We have estimated the contribution of borrowing to lexicalchange at a maximum of 50%, comparable to the rates found for languages knownto have borrowed heavily over a long period, for example English, with 45% oflexical replacement accomplished by borrowing largely from French and fromScandinavian languages. We infer that the problem of unidentified borrowings invocabulary shared by languages whose relationship, if a genetic one at all, is distantenough to preclude use of the normal comparative method and remote enough intime to belong to the period of hunting-and-gathering and the population sizes andrelationships postulated for this era is, while certainly a real one, not a crippling one,provided that certain safeguards are applied.

We have tested, by lexicostatistical means, our indirect method ('local words')for identification and quantification of borrowings, against a group of languages(southwestern Paman, in Australia) for which relevant facts about subgroupingcan be inferred (nonstatistically) from shared innovations in phonology and mor-phology. Finding that exclusion of loans 'identified' by this method does in factimprove the performance of lexicostatistical methods of subgrouping, we go on toa more general consideration of lexicostatistics in historical inference. There aregeneral reasons why glottochronology must be used with much circumspection,and which considerably limit its usefulness, but these apply the world over. Inparticular, it is known that the retention rate may be considerably higher than 80%per millennium—O'Neil (1964) shows a case where the 100-item list shows 94%common vocabulary after 1,000 years (and the 200-item list 93%). To the extentthat the retention rates vary between different languages and between differentgroups, the value of lexicostatistics (despite the fact that it differs from glot-tochronology in not concerning itself with computing time-depths in absoluteunits of time) in determining genetic subgrouping of languages is vitiated in an

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


absolute sense (a fact long explicitly recognized; see Bergsland & Vogt 1962: 126;Gudschinsky 1955: 149). However, even an imperfect lexicostatistics is a usefulsource of hypotheses about subgrouping of languages and dialects. And, given anassumption that the retention rate is fairly constant across the particular languagesunder study, whether this rate is relatively high or low on a world-wide scale isirrelevant to its use in subgrouping.

Although the evidence is slim, it does not support the particular inapplicability inAustralia of the retention rate minimum (roughly 80% per millennium) proposedfrom data elsewhere in the world. The main impetus for questioning its applicabilitycame from noticing the widespread word-tabooing practices in Australia on a scaleunknown in Indo-European. However, because a 'borrowed' word is usually 're-turned' (both in the observed cases and because of the recycling of names), andsince borrowing is just one of many replacement strategies, the taboo practice doesnot lead to a higher replacement rate.

The estimates of the fraction of lexical replacement that is attributable to borrow-ing (however crude), and the estimated equilibrium rate that follows from this, arelow enough to suggest that lexicostatistics, as a rough-and-ready method of languagesubgrouping, can proceed without undue concern for the effects of borrowing. Inparticular, there appears to be no reason from this quarter to question the lexico-statistical evidence for the 'Pama-Nyungan' subgroup suggested by O'Grady et al.This evidence amounts to a transcontinental linguistic boundary across which thepercentage of items shared in a 100 or 200 (or 500) word list hovers around 10%.This is well within the equilibrium range suggested above for languages that havebeen in long contact, and it is not possible to argue that unrelated or equallydistantly-related languages will acquire a much higher percentage of items sharedfrom borrowing over a long period of time. There is always the possibility, of course,that one or more languages may be misclassified by this rough-and-ready method.But the above evidence, together with evidence of a different kind from Merlan(1979), Blake (1988), Evans (1988), Alpher (1990), and Heath (1990) suggests thatthe general outlines of the Pama-Nyungan subgroup, as proposed in the literature,are tolerably accurate.

With regard to the possibity of fine-tuning lexicostatistical methods to move itbeyond the 'rough-and-ready' stage, we cite Embleton (1986: 92): 'lexicostatistics isbest used for the construction of provisional family trees only; the slightly improvedaccuracy for N = 500 over N = 200 is not worthwhile for what should be merelyprovisional results anyway'. Here she is speaking of list size. We extend this thinkingto problems of retention rates and borrowing rates. It seems likely that both of theserates follow a normal distribution. From estimates of these rates and their distributionscan come a lexicostatistics and glottochronology that is a blunt but useful instrument.

ReferencesAbbreviations

AIAS Australian Institute of Aboriginal Studies (now the Australian Institute of Aboriginaland Torres Straits Islander Studies, AIATSIS), Canberra and, for publicationsmid-1976 to circa 1984, Humanities Press, Atlantic Highlands, New Jersey.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14

Lexical Replacement 49ANU Australian National University.GCAL Grammatical Categories in Australian Languages, RMW Dixon (ed.), AIAS, 1976.HAL Handbook of Australian Languages, RMW Dixon & Barry J Blake (eds), Australian

National University Press, Canberra: Vol. I, 1979; Vol. II, 1981; Vol. III, 1983.LCY Languages of Cape York, Peter Sutton (ed.), AIAS, 1976 (Australian Aboriginal Studies

Research and Regional Studies No. 6).PAL 13 Papers in Australian Linguistics No. 13: Contributions to Australian Linguistics (Pacific

Linguistics Series A, No. 59), Research School of Pacific Studies, Australian NationalUniversity, Canberra, 1980.

Alpher B 1972 'On the genetic subgrouping of the languages of southwestern Cape YorkPeninsula, Australia' Oceanic Linguistics 11: 67-87.

Alpher B 1976 'Some linguistic innovations in Cape York and their sociocultural correlates' inLCY: 84-101.

Alpher B 1988 'Formalizing Yir-Yoront lenition' Aboriginal Linguistics 1: 188-197.Alpher B 1989 'The origin of ablaut as a grammatical process in Yir-Yoront' Unpublished MS.Alpher B 1990 'Some Proto-Pama-Nyungan verb paradigms: a verb in the hand is worth two in

the phylum' in GN O'Grady & D Tryon (eds).Alpher B 1991 Yir-Yoront Lexicon: sketch and dictionary of an Australian language Trends in

Linguistics Documentation 6 Mouton de Gruyter Berlin.Alpher B 1992 'Correspondence mimicry' Draft.Alpher B 1997 "The Banks-Cook wordlist of 1770: can it be adapted as a lexicostatistical tool?'

Studies in Comparative Pama-Nyungan Pacific Linguistics Series C No. 1ll Research Schoolof Pacific Studies ANU Canberra: 155-171.

Alpher B 1998 'Yir-Yoront and Kuuk-Thaayorre: a Sprachbund' Draft.Austerlitz R 1991 'Alternatives in long-range comparison' in Sydney M Lamb & E Douglas

Mitchell (eds) Sprung from some Common Source: investigations into the prehistory of languagesStanford University Press Stanford California: 353-364.

Austin P 1981a 'Proto-Kanyara and Proto-Mantharta historical phonology' Lingua 54: 295-333.Austin P 1981b A Grammar of Diyari, South Australia Cambridge University Press Cambridge.Austin P 1992a A Dictionary of Payungu, Western Australia Department of Linguistics La Trobe

University Melbourne.Austin P 1992b A Dictionary of Thalanyji Department of Linguistics La Trobe University

Melbourne.Austin P 1997 'Proto Central New South Wales phonology' in D Tryon & M Walsh (eds)

Boundary Rider: essays in honour of Geoffrey O'Grady Pacific Linguistics C-136 AustralianNational University Canberra: 21-49.

Bergsland K & H Vogt 1962 'On the validity of glottochronology' Current Anthropology 3(2):115-153.

Black P 1980 'Norman Pama historical phonology' in B Rigsby & P Sutton (eds) PAL 13:181-239.

Black P 1997 'Lexicostatistics and Australian languages: problems and prospects' in D Tryon &M Walsh (eds) Boundary Rider: essays in honour of Geoffrey O'Grady Pacific LinguisticsC-136 Australian National University Canberra: 51-69.

Black P nd a 'Lexicostatistical lists: Kokaper (with Kok-Babonk)' Unpublished MS.Black P nd b 'Kurtjar [on verb conjugations]' Unpublished MS.Blake BJ 1979 A Kalkatungu Grammar Pacific Linguistics Series B No. 57 Research School of

Pacific Studies ANU Canberra.Blake BJ 1988 'Redefining Pama-Nyungan: Towards the prehistory of Australian languages'

Aboriginal Linguistics I: 1-90.Blevins J forthcoming 'Nhanda and its position within Pama-Nyungan' in M Laughren &

P McConvell (eds) The Origins of the Western Desert Languages.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


Breen JG 1972 'Kok-Nar fieldnotes' MS.Breen JG 1973 Bidyara and Gungabula Grammar and Vocabulary Monash University Melbourne.Breen JG 1976a 'Gugadj and Gog-Nar. a contrastive sketch' in LCY: 151-164.Breen JG 1976b 'An introduction to Gog-Nar' in LCY: 243-259.Breen JG 1981a 'Margany and Gunya' in HAL II: 274-393.Breen JG 1981b The Mayi Languages of the Queensland Gulf Country AIAS Canberra.Breen JG 1990 Salvage Studies of Western Queensland Aboriginal Languages Pacific Linguistics

B-105 Australian National University Canberra.Dixon RMW 1970 'Languages of the Cairns rain forest region' in SA Wurm & DC Laycock (eds)

Pacific Linguistics Studies in Honour of Arthur Capell Pacific Linguistics Canberra: 651-687.Dixon RMW 1972 The Dyirbal Language of North Queensland Cambridge Studies in Linguistics 9

Cambridge University Press Cambridge.Dixon RMW 1976 'Tribes, languages and other boundaries in northeast Queensland' in Nicolas

Peterson (ed.) Tribes and Boundaries in Australia Social Anthropology Series No. 10 AIASCanberra: 207-238.

Dixon RMW 1977 A Grammar of Yidiny Cambridge Studies in Linguistics 19 CambridgeUniversity Press Cambridge.

Dixon RMW 1980 The Languages of Australia Cambridge Language Surveys CambridgeUniversity Press Cambridge.

Dixon RMW 1990 'The origin of 'mother-in-law vocabulary' in two Australian languages'Anthropological Linguistics 32: 1-56.

Dixon RMW 1991 Words of Our Country Queensland University Press St. Lucia.Dixon RMW 1997 The Rise and Fall of Languages Cambridge University Press Cambridge.Douglas WH 1976 The Aboriginal Languages of the South-west of Australia (2nd edn) Research and

Regional Studies 9 (1st edn 1968) AIAS Canberra.Dyen I 1963 'Lexicostatistically determined borrowing and taboo' Language 39: 60-66.Dyen I, AT James & JWL Cole 1967 'Language divergence and estimated word retention rate'

Language 43: 150-171.Dyen I, JB Rruskal & P Black 1992 'An Indoeuropean classification: a lexicostatistical experiment'

Transactions of the American Philosophical Society 82(5) American Philosophical SocietyPhiladelphia.

Eades D 1979 'Gumbaynggir' in HAL I: 245-361.Embleton SM 1981 'Lexicostatistical tree reconstruction incorporating borrowing' Toronto

Working Papers in Linguistics 2: 98-122 (republished 1982 Eighth Lacus Forum HornbeamColumbia SC: 265-272).

Embleton SM 1986 Statistics in Historical Linguistics Quantitative Linguistics vol. 30 StudienverlagBrockmeyer Bochum.

Evans N 1988 'Arguments for Pama-Nyungan as a genetic subgroup, with particular reference toinitial laminalization' Aboriginal Linguistics I: 91-110.

Gleason HA Jr 1962 'Counting and calculating for historical reconstruction' AnthropologicalLinguistics 1(2): 22-32.

Gudschinsky SC 1955 'Lexico-statistical skewing from dialect borrowing' International Journal ofAmerican Linguistics 21: 138-149.

Guy JBM 1980a Experimental Glottochronology: basic methods and results Pacific Linguistics SeriesB No. 75 Research School of Pacific Studies ANU Canberra.

Guy JBM 1980b Glottochronology without Cognate Recognition Pacific Linguistics Series B No. 79Research School of Pacific Studies ANU Canberra.

Hale KL 1961 'Attestations [of 100-word list in 33 Cape York languages; also exists under titleVocabularies and Cognation Judgments for 30 Cape York Peninsula Languages]'Unpublished MS dittoed.

Hale KL 1962 'Internal relationships in Arandic of Central Australia' in A Capell (ed.) SomeLinguistic Types in Australia Oceania Linguistic Monograph 7 Oceania (The University ofSydney) Sydney: 171-183.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


Hale KL 1964 'Classification of Northern Paman languages, Cape York Peninsula, Australia: aresearch report' Oceanic Linguistics 3: 248-265.

Hale KL 1976a 'Phonological developments in particular Northern Paman languages' in LCY:7-40.

Hale KL 1976b 'Phonological developments in a Northern Paman language: Uradhi' in LCY:41-49.

Hale KL 1976c 'Wik reflections of Middle Paman Phonology' in LCY: 50-60.Hale KL nd a 'CYP cognate density matrices [computed from Hale 1961]' Unpublished MS

dittoed.Hale KL nd b 'Other Paman languages [companion to 1976a and 1976c]' Unpublished MS

dittoed.Hale KL nd c 'Wik-Mungknh wordlist' Unpublished MS.Hall AH 1968 A depth study of the Thaayorre language of the Edward River tribe Cape York Peninsula

Unpublished MA thesis University of Queensland.Hall AH 1972 A study of the Thaayorre language of the Edward River tribe, Cape York Peninsula,

Queensland Unpublished PhD thesis University of Queensland.Hall AH 1976a 'Methods of negation in Kuuk-Thaayorre' in LCY: 299-307.Hall AH 1976b 'Morphological categories of nouns in Kuuk-Thaayorre' in LCY: 308-314.Harris BP & GN O'Grady 1976 'An analysis of the progressive morpheme in Umpila verbs: a

revision of a former attempt' in LCY: 165-212.Haviland JB 1974 'A last look at Cook's Guugu Yimidhirr word list' Oceania 44(3) (March):

216-232.Havilan4 JB 1979a 'How to talk to your brother-in-law in Guugu-Yimidhirr' in Tim Shopen (ed.)

Languages and Their Speakers Winthrop Cambridge Massachusetts: 160-239.Haviland JB 1979b 'Guugu Yimidhirr' in HAL I: 27-180.Heath J 1978a Linguistic Diffusion in Anthem Land AIAS Canberra.Heath J 1978b Ngandi Grammar, Texts, and Dictionary AIAS Canberra.Heath J 1979 'Diffusional linguistics in Australia: problems and prospects' in SA Wurm (ed.)

Australian Linguistic Studies Pacific Linguistics Series C No. 54 Research School of PacificStudies ANU Canberra: 395-418.

Heath J 1982 Nunggubuyu Dictionary AIAS Canberra.Headi J 1990 'Verbal inflection and macro-subgroupings of Australian languages: The search for

conjugation markers in non-Pama-Nyungan' in Philip Baldi (ed.) Linguistic Change andReconstruction Methodology Mouton de Gruyter Berlin: 403-407.

Hershberger R 1964 'Notes on Gugu-Yalanji verbs' in R Pittman & H Kerr (eds) Papers on theLanguages of the Australian Aborigines Occasional Papers in Aboriginal Studies No. 1 AIASCanberra: 35-54.

Johnson S 1990 [1991] 'Social parameters of linguistic change in an unstratified Aboriginal society'in P Baldi (ed.) Linguistic Change and Reconstruction Methodology Mouton de Gruyter Berlin:419-433; reprinted in Philip Baldi (ed.) Patterns of Change, Change of Patterns: linguistic changeand reconstruction methodology Mouton de Gruyter Berlin 1991: 203-217.

Koch H 1997 'Comparative linguistics and Australian prehistory' in Patrick McConvell &Nicholas Evans (eds) Archaeology and Linguistics: Aboriginal Australia in global perspectiveOxford University Press Melbourne: 27-43.

Lees RB 1953 'The basis of glottochronology' Language 29: 113-127.McDonald M & SA Wurm 1979 Basic Materials in Wankumara (Galali): grammar, sentences, and

vocabulary Pacific Linguistics Series B No. 65 Research School of Pacific Studies ANUCanberra.

McEntee J & P McKenzie 1992 Adynya-math-nha English Dictionary Revised May 1992 Adelaide(ISBN 0959664432).

Merlan F 1979 'On the prehistory of some Australian verbs' Oceanic Linguistics 18(1): 33-112.Merlan F 1982 Mangarayi (Lingua Descriptive Studies vol. 4) North-Holland Amsterdam.Nash D. 1982 'An etymological note on kurdungurtu' in J Heath et al. (eds) Languages of kinship in

Aboriginal Australia Oceanic Linguistic Monographs No. 24 Sydney: 141-159.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


Nash D 1997 'Comparative flora terminology of the central Northern Territory' in PatrickMcConvell & Nicholas Evans (eds) Archaeology and Linguistics: Aboriginal Australia in globalperspective Oxford University Press Melbourne: 187-206.

Nash D & J Simpson 1981 "No-name' in central Australia' in Carrie S Masek, Roberta AHendrick & Mary Frances Miller (eds) Proceedings from the Parasession on Language andBehavior Chicago Linguistics Society Chicago: 165-177.

Newton PJF 1980 'Lexicostatistics: a minor analytical tool for Australian historical linguisticstudies' Working Papers in Language and Linguistics II: 1-8.

Oates W & L Oates 1964 'Gugu-Yalanji vocabulary' in W Oates, L Oates, H Hershberger,R Hershberger, B Sayers & M Godfrey (eds) Gugu-Yalanji and Wik-Munkan LanguageStudies Occasional Papers in Aboriginal Studies No. 2 ALAS Canberra: 79-146.

O'Grady GN 1964 Nyangumata Grammar Oceanic Linguistic Monographs No. 9 University ofSydney Sydney.

O'Grady GN 1966 'Proto-Ngayarda phonology' Oceanic Linguistics 5(2): 71-130.O'Grady GN 1976 'Umpila historical phonology' in LCY: 61-67.O'Grady GN 1979 'Preliminaries to a Proto Nuclear Pama-Nyungan stem list' in S Wurm (ed.):

107-139.O'Grady GN & TJ Klokeid 1969 'Australian linguistic classification: A plea for coordination of

effort' Oceania 39: 298-311.O'Grady GN, CF Voegelin & FM Voegelin 1966 'Languages of the world: Indo-Pacific fascicle

six' Anthropological Linguistics 8(2) (February): 1-197.O'Neil WA 1964 'Problems in the lexicostatistic time depth of Modern Icelandic and Modern

Faroese' General Linguistics 6(1) (Spring): 27-37.Ross MD 1996 'Contact-induced change and the comparative method: cases from Papua New

Guinea' in M Durie & M Ross (eds) Comparative Method Reviewed: regularity and irregularityin language change Oxford University Press New York: 180-217.

Sankoff D 1972 'Reconstructing the history and geography of an evolutionary tree' AmericanMathematical Monthly 79: 596-603.

Sankoff D 1973 'Mathematical developments in lexicostatistic theory' in TA Sebeok (ed.) CurrentTrends in Linguistics vol. 11 Mouton The Hague: 93-113.

Sharp RL 1939 'Tribes and totemism in north-east Australia' Oceania 9.3: 254-275, 904: 19-42.Sharp RL 1934 'The social organization of the Yir-Yoront tribe, Cape York Peninsula' Oceania

4: 404-431.Sharp RL 1958 'People without politics' in VF Ray (ed.) Systems of Political Control and

Bureaucracy in Human Societies American Ethnological Society Seattle: 1-8.Sommer BA 1969 Kunjen Phonology: synchronic and diachronic Pacific Linguistics Series B No. 11

ANU Canberra.Sommer BA 1972 Kunjen Syntax: a generative view Australian Aboriginal Studies No. 5 AIAS

Canberra.Sommer BA 1976 'Umbuygamu: the classification of a Cape York Peninsular Language' in JF

Kirton et al. (eds) Papers in Australian Linguistics No. 10 Pacific Linguistics (Series A No.47 Research School of Pacific Studies ANU Canberra: 13-31).

Sutton PJ 1978 Wik: Aboriginal society, territory and language at Cape Keerweer, Cape YorkPeninsula, Australia Unpublished PhD thesis Department of Anthropology and SociologyUniversity of Queensland.

Swadesh M 1950 'Salish internal relationships' International Journal of American Linguistics 16:157-167.

Swadesh M 1952 'Lexico-statistic dating of prehistoric ethnic contacts' Proceedings of the AmericanPhilosophical Society 96(4): 452-463.

Swadesh M 1955 'Towards greater accuracy in lexicostatistic dating' International Journal ofAmerican Linguistics 21: 121-137.

Swadesh M 1962 'The mesh principle in comparative linguistics' Anthropological Linguistics 1(2):7-14.

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


Thomason SG & T Kaufman 1988 Language Contact, Creolization, and Genetic LinguisticsUniversity of California Press Berkeley; paperback ed. 1991.

Thompson DA 1976 'A phonology of Kuuku-Ya'u' in LCY: 213-235.Tryon DT 1970 An Introduction to Maranungku (Northern Australia) Pacific Linguistics Series B

No. 15 Linguistic Circle of Canberra Canberra.Tsunoda TG 1974 A grammar of the Warungu language, North Queensland Unpublished MA thesis

Monash University Melbourne.Watson FJ 1944 'Vocabularies of 4 representative tribes of south eastern Queensland' Supplement

to Journal of the Royal Geographical Society of Australasia (Queensland) No. 34 Vol. 48Session 1943-1944.

Wurm SA 1972 Languages of Australia and Tasmania Mouton The Hague.Zorc R. David (comp.) 1986 Yolngu-Matha Dictionary School of Australian Linguistics Darwin

Institute of Technology Batchelor NT.

Barr AlpherAmerican University< [email protected] >

Address for correspondence:3218 Wisconsin Ave NW, Apt B2

Washington DC 20016, USA

David NashANU, AIATSIS

< [email protected] >

Appendix Lexicostatical Wordlists Used with the Cape York MaterialThe first 100 words constitute the list of O'Grady & Klokeid (1969: 303-7); the next 20 are thewords in Hale's (1961) list that are not also in O'Grady and Klokeid's; the next 25 are words inBlack's (nd a.) list that are not in either of the above; the last 6 words (relevant for the most partin the monsoon tropics of Australia) are the authors' additions. 'H' cross-references to Hale'snumbering for the same word; 'B' does the same for Black's.

The list follows, in two forms. The first column gives the list in numerical order (by referencenumber). The second column gives the list in alphabetical order of the English gloss, so that thereference number for a particular gloss can be easily found. (Note with regard to certain items:(i) for 'stomach' see 'belly'; (ii) 'to get, pick up' (item 34 in the O'Grady-Klokeid list) and 'totake' (item 106, Hale's item 39, Black's 212) are listed as separate items here despite the fact thatthey appear to work as near-synonyms in the elicitation of words in Aboriginal languages.)

Numerical orderNo. Gloss

1. armpit (Hll , B70)2. ashes (H64, B109)3. belly (HI3, B47)4. big (H85, B138)5. bite (H44, B196)6. black (H31, B135)7. blood (H21, B78)8. bone (H23)9. breast (B43)

10. to burn (intr; H65, B108)

Alphabetical orderNo. Gloss

1. armpit (Hll , B70)2. ashes (H64, B109)

126. axe (B23)111. bad (H50, B170)

3. belly (H13, B47)4. big (H85, B138)5. bite (H443 B196)6. black (H31, B135)7. blood (H21, B78)8. bone (H23)

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14



11. by-and-by (H80, B235)12. chest (B42)13. to climb (H38, B177)14. to cry (H47, B191)15. to cut (H45, B204)16. dog (H69, B88)17. down, below (H78, B231)18. ear (H6, B30)19. east (H73, B225)20. to eat (H27, B197)21. egg (H68, B86)22. elbow (HI5, B68)23. excrement (B52)24. eye (H4, B27)25. to fall (H37, B178)26. far (H87, B228)27. fat, grease (H22, B79)28. fingernail (B63)29. fire (H62, B107)30. fly (N; B98)31. vegetable food (H72, B106)32. foot (H20, B62)33. forehead (H2, B26)34. to get, pick up35. to give (H42, B215)36. to go (H35, B172)37. ground (H55, B126)38. hand (H16, B65)39. head (HI, B24)40. head hair (H25, B25)41. hear (H30, B195)42. heart (B44)43. to hit (with hand; H43, B199)44. hungry (H26, B161)45. I (H96, B239)46. knee (H18, B59)47. leaf(BlOl)48. to leave it (H40, B216)49. liver (HI2, B49)50. long (H89, B140)51. to be lying down (B187)52. many (H84, B133)53. meat, animal (H66, B84)54. moon (H57, B122)55. mouth (H7, B32)56. name (B16)57. nape (H3, B38)58. north (H75, B223)59. nose (H5, B28)60. now, today (B234)


123. boomerang (B19)9. breast (B43)

11. by-and-by (H80, B235)12. chest (B42)

127. chin (B37)136. cloud (B119)113. creek (H61)

16. dog (H69, B88)17. down, below (H78, B231)

122. dream (N; B18)143. dry (N, Adj; B158)18. ear (H6, B30)19. east (H73, B225)21. egg (H68, B86)22. elbow (HI5, B68)23. excrement (B52)24. eye (H4, B27)26. far (H87, B228)27. fat, grease (H22, B79)28. fingernail (B63)29. fire (H62, B107)

132. fish (B97)30. fly (N; B98)32. foot (H20, B62)33. forehead (H2, B26)

149. goanna110. good (H49, B169)134. grass (B105)37. ground (H55, B126)38. hand (H16, B65)

117. hard (H91, B144)118. he (H98, B241)39. head (HI, B24)40. head hair (H25, B25)41. hear (H30, B195)42. heart (B44)

140. heavy (B144)128. hip (B57)139. hole (B129)44. hungry (H26, B161)45. I (H96, B239)46. knee (HI8, B59)

121. language (B17)47. leaf(BlOl)

141. light (B145)49. liver (HI2, B49)50. long (H89, B140)

151. mangrove (Avicennia sp.)52. many (H84, B133)53. meat, animal (H66, B84)

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14


61. old man (B4)62. one (H81, B130)63. person, Aborigine (H51, Bl)64. rib (B46)65. rotten (B153)66. to see (H29, B195)67. short (H90, B141)68. to sit (H34, B185)69. skin (H24, B71)70. sky (B119)71. small (H86, B139)72. to smell it (B195)73. smoke (H63, B i l l )74. snake (B96)75. south (H76, B224)76. to speak (H32, B189)77. spear (N; B20)78. spit (N; B35)79. to be standing (H33, B186)80. star (H58, B121)81. stone (H54, B127)82. sun (H563 B123)83. tail (H67, B85)84. thigh (HI7, B58)85. this (H92, B227)86. throat (B39)87. tongue (H9, B34)88. tooth (H8, B33)89. tree (H70, B100)90. two (H82, B131)91. up (H77, B230)92. urine (B51)93. water (H60, B112)94. west (H74, B226)95. what? (H93, B249)96. where? (H95, B251)97. who? (H94, B250)98. wind (H59, B120)99. woman (H52, B2)

100. you (sg.) (H97, B240)101. shoulder (H10, B40)102. upper arm (HI4, B69)103. shin (H19, B60)104. to die (H28, B198)105. to run (H36, B174)106. to take (H39, B212)107. to throw (H41, B217)108. to spear (H46, B201)109. to laugh (H48, B192)110. good (H49, B169)111. bad (H50, B170)112. to dig (H53, B202)113. creek (H61)

54. moon (H57, B122)133. mosquito (B99)55. mouth (H7, B32)

135. mud (B114)56. name (B16)57. nape (H3, B38)

116. near (H88, B229)58. north (H75, B223)59. nose (H5, B28)60. now, today (B234)61. old man (B4)62. one (H81, B130)

150. pandanus131. pelican (B93)63. person, Aborigine (H51, Bl)

130. possum (B90)145. rain (B112)64. rib (B46)65. rotten (B153)

138. sand (B128)137. shade (B125)124. shield (B20)103. shin (HI9, B60)67. short (H90, B141)

101. shoulder (H10, B40)144. sickness (B82)69. skin (H24, B71)70. sky (B119)71. small (H86, B139)73. smoke (H63, Bi l l )74. snake (B96)

142. soft (B145)75. south (H76, B224)77. spear (N; B20)

125. spearthrower (B21)78. spit (N; B35)80. star (H58, B121)81. stone (H54, B127)

147. string82. sun (H56, B123)83. tail (H67, B85)84. thigh (HI7, B58)85. this (H92, B227)

115. three (H83, B132)86. throat (B39)51. to be lying down (B187)79. to be standing (H33, B186)10. to burn (intr; H65, B108)13. to climb (H38, B177)14. to cry (H47, B191)15. to cut (H45, B204)

104. to die (H28, B198)112. to dig (H53, B202)

Dow

nloa

ded

by [

Uni

vers

ity o

f A

uckl

and

Lib

rary

] at

20:

54 0

5 Ja

nuar

y 20

14



114. tomorrow (H79, B238)115. three (H83, B132)116. near (H88, B229)117. hard (H91, B144)118. he (H98, B241)119. we two inclusive (H99, B242)120. you two (H100, B243)121. language (B17)122. dream (N; B18)123. boomerang (B19)124. shield (B20)125. spearthrower (B21)126. axe (B23)127. chin (B37)128. hip (B57)129. wing feather (B87)130. possum (B90)131. pelican (B93)132. fish (B97)133. mosquito (B99)134. grass (B105)135. mud (B114)136. cloud (B119)137. shade (B125)138. sand (B128)139. hole (B129)140. heavy (B144)141. light (B145)142. soft (B145)143. dry (N, Adj; B158)144. sickness (B82)145. rain (B112)146. yamstick147. string148. yam149. goanna150. pandanus151. mangrove (Avicennia sp.)


20. to eat (H27, B197)25. to fall (H37, B178)34. to get, pick up35. to give (H42, B215)36. to go (H35, B172)43. to hit (with hand; H43, B199)

109. to laugh (H48, B192)48. to leave it (H40, B216)

105. to run (H36, B174)66. to see (H29, B195)68. to sit (H34, B185)72. to smell it (B195)76. to speak (H32, B189)

108. to spear (H46, B201)106. to take (H39, B212)107. to throw (H41, B217)114. tomorrow (H79, B238)87. tongue (H9, B34)88. tooth (H8, B33)89. tree (H70, B100)90. two (H82, B131)91. up (H77, B230)

102. upper arm (HI4, B69)92. urine (B51)31. vegetable food (H72, B106)93. water (H60, B112)

119. we two inclusive (H99, B242)94. west (H74, B226)95. what? (H93, B249)96. where? (H95, B251)97. who? (H94, B250)98. wind (H59, B120)

129. wing feather (B87)99. woman (H52, B2)

148. yam146. yamstick100. you (sg.) (H97, B240)120. you two (HI00, B243)D

ownl

oade

d by

[U

nive

rsity

of

Auc

klan

d L

ibra

ry]

at 2

0:54

05

Janu

ary

2014

Documents

Lexical Replacement and Cognate