¹ For a comprehensive review of lexicostatistical research through 1959, see Dell H. Hymes, ‘Lexicostatistics so far’, Current anthropology 1.3–44 (1960) and ‘More on lexicostatistics’, id. 338–45.

² In their present form, my arguments owe much to a series of discussions I have been privileged to have with Isidore Dyen of Yale University, one of the most careful among the scholars who have interested themselves in lexicostatistical techniques.

³ C. F. Voegelin, Current anthropology 3.147 (1962).

⁴ Such a strong claim as this for lexicostatistics is seldom made explicitly, but one must agree with Voegelin (see fn. 3) that is often implied. Various weaker claims are evaluated as well in the sequel.

⁵ For example, Harry Hoijer, ‘Lexicostatistics : A critique’, Lg. 32.49–60 (1956), and more recently, in great detail, I. Fodor, ‘The validity of glottochronology on the basis of the Slavonic languages’, Studia slavica 7.295–346 (1961).

⁶ C. Douglas Chrétien, ‘The mathematical models of glottochronology’, Lg. 38.11–37 (1962).

⁷ Leonard Bloomfield, Language 78 (New York, 1933).

⁸ It is not, of course, meant that all linguistic change is extralinguistically caused (the Saussurian overstatement, e.g. Cours ⁵ 124), but simply that the weakening of communication necessary for differential linguistic changes to take hold is external to the language system.

⁹ Language 314.

¹⁰ Morris Swadesh, ‘Towards greater accuracy in lexicostatistic dating’, HAL 21.121–37 (1955), esp. 123.

¹¹ Swadesh, loc.cit.

¹² George Peter Murdock, ‘Cross-language parallels in parental kin terms’, Anthropological linguistics 1:9.1–5 (1959). For a discussion of the underlying factors here, see Roman Jacobson, 'Why “mama” and “papa”?', Perspectives in psychological theory 124–34 (New York, 1960), reprinted in Jakobson's Selected writings 1.538–45 ('s-Gravenhage, 1962).

¹³ Jules Gilliéron, Pathologie et thérapeutique verbales (Paris, 1921).

¹⁴ A method for extricating certain such nonrelevant items is proposed by Isidore Dyen, ‘Lexicostatistically determined borrowing and taboo’, Lg. 39.60–6 (1963). As his title indicates, Dyen's method does use statistical inference. I believe, however, that it also requires an assumption as to the regularity of lexical change (this is denied by Dyen), and I argue the point in §6. Incidentally, he does not claim that his method can be used except for the determination of subgrouping among languages for which the fact of genetic relationship has already been established.

¹⁵ Gordon H. Fairbanks, ‘A note on glottochronology’, IJAL 21.116–20 (1955), is an interesting discussion of this problem.

¹⁶ La méthode comparative en linguistique historique 33 (Oslo, 1925).

¹⁷ Much such discussion and references to much more are to be found in the references cited in fn. 1.

¹⁸ See §8.

¹⁹ Some recent work by Isidore Dyen is explicitly limited to this aim. See ‘The lexicostatistical classification of the Malayopolynesian languages’, Lg. 38.38–46 (1962), and ‘The lexicostatistically determined relationship of a language group’, IJAL 28.153–61 (1962).

²⁰ An example is the work of Dyen cited in fn. 19. ^20aDell H. Hymes, Current anthropology 3.138 (1962).

²¹ Walter W. Arndt, ‘The performance of glottochronology in Germanic’, Lg. 35.180–92 (1959).

²² Knut Bergsland and Hans Vogt, 'On the validity of glottochronology^, , Current anthropology 3.115–53 (1962). This article presents evidence for difficulties in the assumption of uniform retention rates, along with considerable discussion of both this and more general problems, by the authors and others.

²³ Dell H. Hymes, loc.cit.

²⁴ I am no mathematician, but the use of statistics in some lexicostatistical studies does not seem faultless to me. Thus, in the first paper of Dyen's cited in fn. 19, the application of a chi-square test requires the assumption that cognation is equiprobable for each pair of items, as Dyen himself points out (41, fn. 5). The method he employs in a more recent paper (that cited in fn. 26), however, makes precisely the opposite assumption, that each gloss has its own probability of retention. The assumptions contradict each other, and, if my nonmathematical mind does not mislead me, the latter one puts the problem beyond the range of statistics entirely, since individual events do not have probabilities.

²⁵ C. F. Voegelin, Current anthropology 3.147 (1962).

²⁶ H. A. Gleason Jr. as quoted by Hymes, Current anthropology 1.11 (1960); Isidore Dyen, ‘On the validity of comparative lexicostatistics’, paper read to the IX International Congress of Linguists in Cambridge, Mass., 1962.

²⁷ See §5.

²⁸ Edward Sapir, Language 234 (New York, 1921).

²⁹ As Bergsland and Vogt put it, ‘It is difficult to see … how the various social, cultural, and linguistically patterned factors of lexical change could be determined from short word lists designed … to eliminate those very factors‘—op.cit. (fn. 18) 151.

³⁰ David G. Mandelbaum (ed.), Selected writings of Edward Sapir 9 (Berkeley and Los Angeles, 1951). The use of the word ‘intuitively’ perhaps requires a brief comment, since I think Sapir's intentions in such usage have been misunderstood. Intuition of the native speaker does not function directly in explanations (this is a circle), but provides data for the linguist, and conditions which his analyses and explanations must meet where relevant.

³¹ Further discussion and examples of such rules are given in my paper, ‘Algonquian languages and genetic relationship’, read to the IX International Congress of Linguists and to appear in the Proceedings of that Congress.

³² Bergsland and Vogt 151.

³³ ‘Pour chaque individu, le langage est … une récréation totale faite sous l'influence du milieu qui l'entoure‘—Antoine Meillet, Linguistique historique et linguistique générale 2.74 (Paris, 1952).

³⁴ Confusion over this point has, I believe, led to much misunderstanding in questions of ‘structural borrowing’ and ‘language mixture’. Grammars cannot be borrowed, nor can they mix, since they cannot come in mutual contact. Thus the evidence presented in M. B. Emeneau, ‘Bilingualism and structural borrowing’, Proceedings of the American Philosophical Society 106.430–42 (1962), is important in exploring how far it is possible for the members of a speech community to ‘reinvent’ a foreign language within their own grammars. (The answer Emeneau gives for the case he studies—Brahui—is reassuring to the comparative linguist, for what it amounts to is ‘surprisingly far, but not so far that it is not possible to detect borrowings and to reconstruct unambigously’.) The article does not, however, as its author considers, prove ‘structural borrowing’.

³⁵ Meillet, Méthode (cited in fn. 16) 38.

³⁶ Brilliant light has been thrown on this process by Calvert Watkins, in his book Indo-European origins of the Celtic verb (Dublin, 1962).

³⁷ Isidore Dyen, Lg. 29.580–3 (1953); M. B. Emeneau, Kolamι, a Dravidian language, Chapter X (Berkeley and Los Angeles, 1955). Consideration of shared innovations, however, is not enough. Since we are dealing with systems and not lists of items, retentions and divergence must also be brought into the picture.

³⁸ Methods in structural linguistics 3 (Chicago, 1951).

³⁹ Readings in linguistics 96 (Washington, 1957). This view is ascribed to Boas, I think unfairly, as argued in my paper, ‘Descriptive linguistics in America: Triviality vs. irrelevance’, to appear.

⁴⁰ A well-written and entertaining exposition of the nature of scientific discovery is found in Norwood Russell Hanson, Patterns of discovery (Cambridge, England, 1958). On scientific theory see e.g. Ernest Nagel, The structure of science (New York, 1961),