¹ So far as I am able to determine, Czekanowski first applied the technique to ethnology in his Objektive kriterien in der ethnologie, Korrespondenzblatt der deutschen gesellschaft für anthropologie, ethnologie, und urgeschichte 42.71–5 (1911), reprinted in Sitzungsberichte der anthropologischen gesellschaft in Wien 1911–1912.17–21 (1912); and to linguistics in his Wstęp do historij Slowian 50–8 (Lwow, 1927). S. Klimek, Culture element distributions: I. The structure of California Indian culture, UCPAAE 37.1–70 (1935), introduced the method into ethnology in this country, and A. L. Kroeber and C. D. Chrétien, Quantitative classification of Indo-European languages, Lg. 13.83–103 (1937), into linguistics.

² A. Capell, The linguistic position of south-eastern Papua (Sydney, 1943).

³ For our purpose, poor quality is not, in general, a hindrance : though the spelling may be only crudely approximate to phonemic and even partly inaccurate, a word can be recognized as cognate with an equally badly spelled word in another language; M. Swadesh, Salish internal relationships, IJAL 16.157–67 (1950), faced the same problem in Salishan, and has an appropriate remark on this point : ‘In many cases the phonetics are inaccurate but it is usually possible to judge whether forms are cognate or not’ (157). But meager quantity of material is a handicap which cannot be overcome.

⁴ The languages and their groupings are given by Capell 6–7, and in the separate treatments of areas, 9–107; the names have been checked against S. H. Ray, The languages of the Central Division of Papua, JRAI 59.65–96 (1929), and The languages of the Eastern and South-Eastern Divisions of Papua, JRAI 68.153–208 (1938), from which also some group names were taken. The list follows. Area (1) : Western mainland group (Redscar Bay area): Mekeo, Rapeka, Kuni, Roro, Pokau, Kabadi, Doura. Area (2) : Central mainland group (Port Moresby area): Motu, Sinaugoro, Ikoro, Hura, Keapara, Rubi, Aroma, Keakaro. Area (3): Southern Massim group (Suau group): Mugura, Daui, Suau, Rogea, Sariba, Bonarua, Wari, Tubetube. Area (4) : Southern Duau group: Guregureu, 'Urada, Bunama, Sawabwara, Nuakata. Area (5) : D'Entrecasteaux group: Roboda, Dobu, Kukuya, Bwaidoga. Area (6): Milne Bay group: Keherara, Tavara, Wagawaga, Basilaki (Bohila'i). Area (7) : Wedau group: Awalama, Taupota, Wedau, Galavi, Boniki. Area (8) : Collingwood Bay group: Paiwa, Anuki, Mukawa, Kwagila, Ubir, Oyan, Miaitopa. Area (9) : Northern Massim group: Kiriwina, Gumasi, Gawa, Murua, Nada. Area (10) : Southeastern Massim group: Tokunu, Panayati, Misima, Du Chateau, Panakrusima, Sabari, Nimoa, Panatinani. Area (11): Sud Est group: Tagula, Rewa, Juru, Pantava, Griffon Point, Piron Island, North Coast.

⁵ In New Guinea linguistics Papuan is a cover term for all languages not demonstrably Melanesian; it implies no genetic relationship among the languages to which it is applied.

⁶ A bibliography of these papers may be found in D. W. Reed and J. L. Spicer, Correlation methods of comparing idiolects in a transition area, Lg. 28.348–59, footnotes 3, 4, 5, 7 (1952).

⁷ Swadesh, op.cit. 157, seeks the same end: ‘For the purposes of comparative linguistics and for the reconstruction of the prehistory of peoples, it is not sufficient merely to know that a given group of languages are or are not related to each other ... For the Salishan stock, the fact of relationship was determined decades back and has become widely known, but very few scholars have any conception whatever of the degrees of relationship or the stretch of time needed to account for the formation of this stock.‘ Swadesh does not use the Czekanowski-Klimek method, but his procedure is nevertheless based on the theory of mathematical probability. The present paper, unlike Swadesh's, does not seek to determine time-depth.

⁸ The earlier papers hardly went beyond the groupings. It is to be observed that groupings arrived at by nonstatistical means are also used as a basis for historical interpretation; compare the now generally discarded centum-satem grouping of Indo-European languages, or, better, such groups as the Balto-Slavic or Indo-Iranian. In this connection it is to be stated that Kroeber and Chrétien arrived at the usual groupings of Indo-European because the source of their element list, A. Meillet, Les dialectes indo-européens (Paris, 1922), was a collection of data designed to establish just such groups; the fact that it also suggested a Germanic-Baltic-Slavic group gives reason to believe that a statistical technique is useful in pointing out what a nonstatistical technique may overlook. I have elsewhere criticized Kroeber and Chrétien on other grounds. See Chrétien, The quantitative method for determining linguistic relationships, UCPL 1.11–20 (1943).

⁹ This assumption involves denial of Dempwolff's Proto-Melanesian hypothesis and acceptance, in principle, of Capell's position. See O. Dempwolff, Das austronesische sprachgut in den melanesischen sprachen, Folia ethnoglossica 3.42 (1927); O. Dempwolff, Vergleichende Lautlehre des austronesischen Wortschatzes 2.165–6 (1937); Capell 269–70. I use the term ‘Indonesian’ rather than ‘Austronesian’, ‘Malayo-Polynesian’, or Oceanic' because I believe that Capell has established the specifically Indonesian source of the Oceanic strain in Melanesian, at least for southeastern Papua. This position may be taken whether or not we accept Capell's three waves of contact.

¹⁰ See A. L. Kroeber, Some relations of linguistics and ethnology, Lg. 17.290 (1941); S. H. Ray, A comparative study of the Melanesian island languages 597–8 (Melbourne, 1926).

¹¹ Kroeber, Lg. 17.290.

¹² Ethnology would rule out the alternative possibility that originally Oceanic-speaking Melanesiane came later under far-reaching cultural influences that caused them to replace the bulk of their lexicon with foreign elements.

¹³ Throughout this paper I use the numbers for the areas as given in footnote 4.

¹⁴ Capell appears to list 291 words, but he omits Nos. 83 and 84. This leaves 289. I omit 25 words which are indicated for one region only, leaving 264. I have, of course, made the table as described here, and have used it as the source of my numerical data. The problem of providing the raw data so that others can check statistical results is serious in view of printing costs, but fortunately can be met in the present paper with a little labor on the part of the checker. None of the various papers dealing with the glottochronologic method has provided the raw data.

¹⁵ See Chrétien, Culture element distributions: XXV. Reliability of statistical procedures and results, UC-AR 8.5.483–5 (1945).

¹⁶ I borrow the terms ‘subject’ and ‘relative’ from F. Galton, Co-relations and their measurement, chiefly from anthropometric data, Proceedings of the Royal Society of London 45.140 (1889). It does not matter which member of the pair is chosen as subject and which as relative; the result will be the same. Interchanging subject and relative interchanges b and c, but since the formula used below is symmetrical with respect to b and c, the value found is unchanged.

¹⁷ I have given the pairs in this table in order of decreasing values of V, since such an arrangement is advantageous in the later discussion.

¹⁸ For earlier references on association-coefficients see Chrétien, UCPL 1.13 and footnote 4. The statement made there (14) that tetrachoric r gives the true correlation is wrong; it gives the true correlation only under certain very restrictive assumptions which are not made here and which cannot justifiably be made. For a more modern treatment of association see M. G. Kendall, The advanced theory of statistics 1.308–13 (London, 1948).

¹⁹ I have discussed the null-hypothesis and the x ²-test in UCPL 1.17–9 and UC-AR 8.473–5. See also J. Neyman, First course in probability and statistics 259 (New York, 1950), and L. J. Savage, The foundations of statistics 247 (New York, 1954).

²⁰ Since there are 264 words in the non-Indonesian list, and we know their distribution over each area, N is the sum of the marginal values : (a + b) + (c + d) = N, and (a + c) + (b + d) - N.

²¹ The interval, in the terminology of linear point sets, is a closed interval, i.e. both +.121 and —.121 are included in the range of nonsignificant values, as well as the values lying between them. Ordinarily it is almost more than sufficient to compute V to two decimal places. Near the critical values, however, it is necessary to go to four decimal places in order to be sure whether the V is significant or not. This has been done in Tables 3 and 6, the results being afterwards rounded off to two places. This will explain why of certain pairs of areas with the same V some are significant and some not.

²² This remark is especially pertinent in the present paper. In UC-AR 8.482–3 I sought to show that the Czekanowski-Klimek method should not be used when N is less than 500. I have since found support for this view in a source unknown to me at the time: W. E. Deming and R. T. Birge, On the statistical theory of errors, Review of modern physics 6.124 (1934). Discussion with my colleagues A. L. Kroeber and D.W. Reed has persuaded me that the N = 500 rule may be relaxed somewhat provided that we use coefficients only to make broad groupings; this is what I do here.

²³ Chrétien, UCPL 1.15–6, gives a general statement on this interpretation.

²⁴ It is to be observed that Figure 2 represents a further reduction of data and a further loss of particularity. But this is justified in view of the fact that we are dealing with only 264 words.

²⁵ I borrow the term from A. L. Kroeber, Culture element distributions: III. Area and climax, UCPAAE 37.101–2 (1936), where it has geographical significance. Kroeber, Cultural and natural areas of native North America, UCPAAE 38.5 (1939), applies the term to time-scale.

²⁶ The (10)-(11) climax is not obvious in Figure 2(a), but appears in the rearrangement of figure 2(c).

²⁷ This will illustrate the importance of examining the data underlying the coefficients. Without so doing, we might have concluded that (1) and (2) possessed a common substratum larger than is actually the fact, simply on the basis of the high coefficient. Earlier studies have not been guiltless in this respect.

²⁸ The number of words of the master list found in either or both of two areas is a + b + c.

^28a In Table 6 we find a gap of .05 between (7)-(8) and (4)-(5), and again between (10)-(11) and (6)-(8). No other gaps this large or larger are found.

²⁹ This is Capell's view; see his map, p. 10. Investigation of the languages of the Admiralty Islands, New Britain, and New Ireland, together with the other Melanesian languages of the north coast of New Guinea, would undoubtedly throw light on the details of this movement.

³⁰ In the plus and minus table referred to at the beginning of this section, and not provided here.

³¹ The range 34% to 23% applies only to the members of the groups of the eastern branch. Outside these groups, but still within the branch, the figures are different.

³² IJAL 16.159, Table I (1950). Swadesh considers only what he regards as basic vocabulary.

³³ See footnote 9 above.

³⁴ If we had chosen a point between .34 and .35 as the level of discrimination whereby to include or to reject pairs for representation in figure 3, area (1) would have been marginal to group (3)-(4)-(5).

³⁵ I include the nonsignificant (2)-(10).

³⁶ Herein my procedure differs from Swadesh's glottochronology, if I understand him correctly : he seeks to establish time-depths which are to be taken as facts, apparently in much the way that physical measurements are taken as facts.

³⁷ See Capell, especially 266–76.