1. Pearson, Karl, The Grammar of Science (London, 1892), 92–93.CrossRefGoogle Scholar

2. Medawar, P.B., “Hypothesis and Imagination” in Schilpp, P.A. ed., The Philosophy of Karl Popper (2 vols.: La Salle, 1974), 1:279.Google Scholar

3. See for example Oliver, R., “The Beginnings of Bantu History” in Perspectives nouvelles sur le passé de l'Afrique noire et de Madagascar: mélanges offerts à Hubert Deschamps (Paris, 1974), 159–69.Google Scholar In this article (which, from internal evidence, was probably written in 1971) the contradiction is still described as “apparent” on the first page and on the last. It is clear from later publications, however, that Oliver's views have changed.

4. Some of this ground has already been covered in Vansina, Jan, “Bantu in the Crystal Ball, II,” HA, 7(1980).Google Scholar

5. Guthrie, Malcolm, Comparative Bantu: An Introduction to the Comparative Linguistics and Prehistory of the Bantu Languages (4 vols: Farnborough, 1967–1971).Google Scholar

6. For biographical information I have used the obituaries by Arnott, D.W., BSOAS, 36(1973), 629–37CrossRefGoogle Scholar, and by Carter, H., Proceedings of the British Academy, 59(1973), 473–98.Google Scholar For a bibliography see Atkins, G., “Writings of Malcolm Guthrie,” African Language Studies, 11(1970), 2–4.Google Scholar

7. Guthrie, , “Comparative Bantu: a Preview,” Journal of African Languages, 4(1965), 40–45.Google Scholar As its title implies, this article was written as a trailer for his definitive publication. It gives a relatively full but not always reliable account of the way in which his investigations developed.

8. For a review of this earlier work see Doke, C.M., “The Growth of Comparative Bantu Philology” in Doke, C.M. and Cole, D.T. eds., Contributions to the History of Bantu Linguistics (Johannesburg, 1961), 54–79.Google Scholar

9. Guthrie, , “Comparative Bantu: a Preview,” 40.Google Scholar

10. Guthrie, , “Some Developments in the Prehistory of the Bantu Languages,” JAH, 3(1962), 273.CrossRefGoogle Scholar

11. “Archetype” is my word for it, not Guthrie's.

12. Guthrie, , “A Two-Stage Method of Comparative Bantu Study,” African Language Studies, 3(1962), 6.Google Scholar

13. In fact the regularity did not usually show up in individual consonants and vowels, but rather in consonant-vowel or vowel-consonant combinations which Guthrie called “patterns.” The “starred forms” were then built up by allowing such “patterns” to overlap.

14. Guthrie, , “Two-Stage Method,” 21–22Google Scholar; but for the full complexity of the situation see Comparative Bantu, 1:124–25.Google Scholar

15. In Guthrie's usage it was the “comparative series” which were “osculant,” not the “starred forms.”

16. Guthrie, , “Comparative Bantu: a Preview,” 40.Google Scholar

17. Guthrie, , The Classification of the Bantu Languages (London, 1948).Google Scholar A draft of this book had been completed by May 1946: Africa, 16(1946), 261.Google Scholar In the published version I find 24 “starred forms,” only two of which were not included later in Comparative Bantu. There are differences of presentation -- capital letters, *L instead of *d -- but few of substance. Tones were not marked.

18. Guthrie, , Classification, 11.Google Scholar See also Guthrie, “La classification des langues bantu: approche synchronique, méthodes et résultats,” Travaux de l'Institut de Linguistique, 4(1961), 78.Google Scholar

19. Guthrie, , Classification, 18–19.Google Scholar

20. Ibid. 27.

21. Ibid. 59. Zone D, for instance, was a residual category, “of little linguistic significance:” in Guthrie's view there were “reasons for not placing any of these groups in the neighboring zones, but few, apart from geographical contiguity, for making a zone out of them.” (p. 40). See also Guthrie, , “Problèmes de génétique linguistique: la question du bantu commun,” Travaux de l'Institut de Linguistique, 4(1961), 89Google Scholar, where he stated that “la validité des zones est géographique plutôt que linguistique.”

22. Guthrie, , Classification, 28.Google Scholar

23. Ibid. 20. It happened that this book was reviewed by Greenberg, who confessed his inability to understand what Guthrie was trying to accomplish: Greenberg, J.H., “Review of Guthrie, Classification,” Word, 5(1949), 81–83.Google Scholar

24. In 1961 Guthrie wrote that he had been working on Bantu “during the last fourteen years or so:” “Some Developments in the Prehistory of the Bantu Languages” (a mimeographed paper for the Third Conference on African History and Archaeology, SOAS, July 1961), 1. In 1962 he said explicitly that the “two-stage method” had been adopted “some fifteen years ago:” Guthrie, , “Some Developments,” 273Google Scholar, or “Two-Stage Method,” 1.

25. Guthrie, , “Comparative Bantu: a Preview,” 41.Google Scholar

26. Among many statements to the same effect the three that follow are all from Guthrie, “Two-Stage Method:” “it is evident that the ultimate purpose of a comparative study… lies in the realm of history or, where there are no written records, of prehistory” (p. 1); “no comparative study that does not lead to conclusions in prehistory is really worth undertaking” (p. 1); “the ultimate aim of the whole procedure is to reach some conclusions about Bantu prehistory” (p. 4).

27. Guthrie, , “A Two-Stage Method,” 1.Google Scholar Note that there is no sense of a distinction between positive and negative feedback.

28. Guthrie, , “Comparative Bantu: a Preview,” 41.Google Scholar

29. Part of this trip was spent supervising the work of the two linguists who had been assigned to survey the western half of the “Northern Bantu Borderland” under a scheme organized by the International African Institute. See especially the report by Richardson, I., Linguistic Survey of the Northern Bantu Borderland, vol. 2 (London, 1957).Google Scholar

30. Guthrie, , The Bantu Languages of Western Equatorial Africa (London, 1953).Google Scholar Checking with Comparative Bantu, the only discrepancies I can find are: three cases of a change in the final vowel; three cases of a change in tone; and *k instead of *g in *-ganda “cloth.” On the other hand, out of 80 items which are cited in relation to these “starred forms,” only 53 occur in Comparative Bantu. A number of Guthrie's “starred forms” were also quoted by Richardson, , Linguistic Survey, vol. 2.Google Scholar I count 23, of which 16 had not been cited previously by Guthrie himself.

31. Carter, , “Obituary,” 487.Google Scholar She was a student of Guthrie's in 1951/52. A standard word-list which was “circulated in cyclostyle by Guthrie from 1949/50 onward” is mentioned by Sharman, J.C., “Some Uses of Common Bantu” in Whiteley, W.H. ed., Language in Kenya (Nairobi, 1974), 117Google Scholar; cf. Guthrie, , Comparative Bantu, 2:15.Google Scholar

32. Guthrie, , “Comparative Bantu: a Preview,” 41.Google Scholar

33. Guthrie, , Comparative Bantu, 1:17.Google Scholar

34. Guthrie, , “Some Developments,” 273.Google Scholar

35. Guthrie, , “Contributions From Comparative Bantu Studies to the Prehistory of Africa” in Dalby, David ed., Language and history in Africa (London, 1970), 21.Google Scholar This is explicitly a statement of his reaction to Greenberg's theory.

36. Westphal, E.O., “Analysing, Describing, and Teaching Bantu Languages,” African Language Studies, 11(1970), 383.Google Scholar

37. Corresponding in fact to two of the “groups” defined in 1948.

38. Guthrie, , “Teke Radical Structure and Common Bantu,” African Language Studies, 1(1960), 1–15.Google Scholar For an American reaction see Gleason, H.A., “Review of African Language Studies, 1,” Language, 37(1961), 306–08.CrossRefGoogle Scholar

39. On this point see Mann, M., “Sound-Correspondences and Sound-Shifts,” African Language Studies, 14(1973), 26–35.Google Scholar

40. Guthrie, , “Comparative Bantu: a Preview,” 41.Google Scholar In the previous paragraph a total of 26,000 items was mentioned. Elsewhere he wrote that the output from “stage 1,” before the inclusion of exhaustive data for the 28 “test languages,” consisted of 15,000 items: Guthrie, “Some Developments,” (mimeo) 2. I conclude that 11,000 items were eliminated at the end of “stage 1.” As to the dating see below, n. 64.

41. This innocuous assumption was justified at great length in terms of a “hypothetico-deductive method;” see Guthrie, , Comparative Bantu, 1:19–20.Google Scholar

42. Guthrie's term was “topological.”

43. Over the next few years Guthrie's progress can be monitored fairly closely, thanks to the following sources: (a) reports of meetings of the African History Seminar during January-March 1959; (b) two lectures given in Paris at Easter 1960 and published in Travaux de l'Institut de Linguistique (see notes 18 and 21); (c) a paper presented at the Third Conference on History and Archaeology in Africa in July 1961 (see note 24); and (d) three papers published in 1962 (see notes 10, 12, and 68).

44. Zones S and T were amalgamated to form a new zone S. It is not clear exactly when this was done, or why.

45. The partition is ABCHKLR:DEFGMNPS. To avoid ambiguity I use capital letters (as in “Western”) when speaking of sets of zones and lower case (as in “western”) when speaking of sets of roots.

46. The sample in question consists of every twentieth “comparative series,” beginning with C.S. 14 and ending with C.S. 2174: it thus comprises about 5% of the total data. In some cases, two or three subsets of data belonging to the same C.S. were treated separately by Guthrie, and that brings the number of roots in the sample to 123. A listing is available on request.

47. In defining the “regions,” “a certain amount of experimenting” is said to have been involved: Guthrie, , “Some Developments,” 275.Google Scholar From this, and from the emphasis placed on the size of the residual set, it appears that Guthrie had tried splitting the zones into “regions” along a number of different boundaries, and chosen the partition which left the smallest number of roots unclassified. Suppose we require -- as he seems to have done -- that the “regions” should both be connected and should also contain (as nearly as possible) an equal number of zones: then, on a strict definition of “connectedness” (two zones are connected if they share an edge), there are 78 possible partitions. Again, a listing is available on request. For Guthrie, in fact, two zonss were considered to be connected even if they only met at a vertex, like P and S. The number of possibilities would then be greater still. As an extreme example, the partition ABEGKLMN:CDFHPRS would be admissible by his definition, but not by mine. Guthrie cannot have tried more than a few alternatives, and there is no guarantee that he had found the partition which was optimal -- optimal in his terms -- as a basis for classification. See below, n. 51.

48. On a two-dimensional grid, the only roots that could possibly fail to classify would be those that occur in fewer than five zones. On a three-dimensional grid, by contrast, a root that occurs in six zones could still fail to classify for some partitions. The “belts” can thus be seen to have suffered from a built-in disadvantage; and here again, of course, there is no guarantee that Guthrie had found the best three-way partition.

49. The partition into “areas” is ABC:DEF:HL:GMN:KR:PS.

50. Guthrie, , Comparative Bantu, 1:100.Google Scholar The “modulus of dispersion” is a strange statistic which Guthrie devised for himself: even the instructions he gave for its calculation are mutually inconsistent. On this evidence, therefore, we cannot say that there is no bimodality, only that Guthrie failed to demonstrate one.

51. There is a specific sense in which the partition adopted by Guthrie might be described as optimal -- not in terms of its performance as a basis for classification, but rather in terms of its intrinsic properties. Given any partition, we can measure the length of the boundary between the “regions” by counting the number of edges of which it is composed. In the case of Guthrie's partition, for instance, the edges that form the boundary are C/D, D/L, L/M, K/M and K/S: we therefore say that it has a length of 5. This happens to be the shortest boundary we can find for any admissible partition (see n. 47).

52. Thus the term ‘Proto-Bantu’ had no sooner been introduced than its meaning was re-defined: it now covered three “source-languages” rather than one, and a period of time besides.

53. For Guthrie's use of this term see the passage already cited in n. 41.

54. A ‘dendrogram’, as this term is generally used in the field of numerical taxonomy, can be defined as a rooted tree satisfying one special condition: it has no vertices of degree 2, with the possible exception of the vertex which is designated as the ‘root’. I regret the necessity of using the term ‘root’ in its graph-theoretic as well as its linguistic sense; but the meanings attached to the word are so very different that confusion ought not to arise.

55. Henrici did most to precipitate the crisis by reworking Guthrie's own data for the 28 “test languages” and obtaining a dendrogram which is very highly skewed in the direction of zone A: Henrici, A., “Numerical Classification of Bantu Languages,” African Language Studies, 14(1973), 82–104.Google Scholar For a reaction, see Dalby, D., “The Prehistorical Implications of Guthrie's Comparative Bantu, I: Problems of Internal Relationship,” JAH, 16(1975), 481–501.CrossRefGoogle Scholar An obscure remark contrasting “Eastern unity” with “a relative Western disunity” (493n30) is, I suppose, a polite way of saying that “Eastern” is at least roughly valid, but that “Western” is not. In fact the boundary between the “regions” does not seem to coincide at all closely with the boundary of Eastern Bantu.

56. This is the sub-branch called “Central” by Henrici, and covering the whole Bantu field except zones A-C and possibly part of D.

57. In the sample mentioned above (n. 46) there are 69 roots which can be referred to Proto-Bantu, granted the assumption that the languages are to be classified as in Fig. 4, No. 1. This assumption is simplistic, but not unfair to Guthrie. In the same sample, there are 33 roots identified as “general” (GG, G, or G'). The number of roots which are “generally” distributed and also referable to Proto-Bantu is 27: thus there are 42 roots which go back to Proto-Bantu without qualifying as “general,” and 6 roots which qualify as “general” without going back to Proto-Bantu. That makes 48 roots which Guthrie either missed or identified wrongly, as against 27 which he identified correctly. Proto-Bantu roots which happen to have been lost in all the northwestern zones cannot be identified as such -- unless perhaps with the help of evidence from related but non-Bantu languages.

58. See, for example, Guthrie, , “A Two-Stage Method,” 18–19Google Scholar (on “complex radicals”), 21-23 (on “osculant series”). Guthrie claimed that the distributional evidence was consistent with this conclusion (Comparative Bantu, 1:100Google Scholar), but the reasoning escapes me.

59. Ibid., 2:27.

60. Guthrie, , “Some Developments,” 280.Google Scholar

61. As recorded in the mimeographed report of a meeting of the African History Seminar, 28 January 1959.

62. The “test languages” contributed 13,000 items out of a total which was variously given as 21,000 or 22,000 (Guthrie, , “A Two-Stage Method,” 11Google Scholar, for example). On the other hand we know that the output from “stage 1” originally consisted of 15,000 items (see above, n. 40). It follows from this that 6000-7000 items must already have been included from the “test languages” before they were singled out for special treatment. Concerning Bemba see below, n. 64.

63. Guthrie, , “Comparative Bantu: A Preview,” 41.Google Scholar The actual wording is that “in principle no major additions were made after 1960” -- from which it seems fair to infer that “major additions” were still being made “in principle” till then, and that minor additions continued to be made in practice even later. After the “comparative series” had been numbered (by 1958) roughly 40 more were interpolated into the list; and these mostly seem to have been promoted from “partial series” on the evidence of new items from the “test languages. One other addition, made at an even later stage (not before 1965) was the incorporation of about 250 items from “zone Z.”

64. As appears to be claimed in Guthrie, , “Comparative Bantu: a Preview,” 41.Google Scholar A personal communication from Guthrie was cited by Tucker in an article which, from internal evidence, must have been written towards the end of 1957. Tucker, A.N., “Philology and Africa,” BSOAS, 20(1957), 541–54.CrossRefGoogle Scholar According to Tucker, (552) Guthrie had constructed “some 2,200 starred forms,” but “assures me that it is rare for more than 500 of these to be represented in any one Bantu language.” More specifically, “the greatest number he has recorded in any one language is 600” (552nl). The language in question must surely have been Bemba, from which 800 items were recorded in the end. It therefore seems clear that the output from “stage 1” had already been “pruned” by the end of 1957, but that the “test languages” had not yet been subjected to exhaustive analysis.

65. The “test languages” were used for only two purposes: for the calculation of various “coefficients,” and for an experiment in numerical taxonomy which we know to have been in progress in 1960: see Guthrie, , “Problèmes de génétique linguistique,” 92.Google Scholar

66. In a letter to Michael Mann, quoted by Bynon, T. and Mann, M., “Papers on Comparative Bantu: an Introduction,” African Language Studies, 14(1973), 2–3.Google Scholar

67. This is my reading of some obscure remarks in Guthrie, , Comparative Bantu, 1:97.Google Scholar

68. Guthrie, , “Bantu Origins: a Tentative New Hypothesis,” Journal of African Languages, 1(1962), 14.Google Scholar

69. Guthrie, , Comparative Bantu, 1:102Google Scholar, topogram T.1.

70. Ibid, 1:103, topogram T.5.

71. Guthrie, , “Some Developments,” 280.Google Scholar

72. This is admitted even by linguists closely associated with Guthrie, like Dalby and Mann: see Dalby, , “The Prehistorical Implications, I,” 495–96Google Scholar, or the paper by Mann cited in n. 74. Greenberg observes that some linguists have taken exactly the opposite view -- that peripheral languages tend to be more convervative -- but does not think it safe to generalize in either sense: Greenberg, J.H., “Linguistic Evidence Regarding Bantu Origins,” JAH, 13(1972), 193.CrossRefGoogle Scholar

73. For a brief statement of this principle, the first formulation of which is generally credited to Sapir, see Greenberg, , “Historical Linguistics and Unwritten Languages” in Kroeber, A.L. ed., Anthropology today: an Encyclopedic Inventory (Chicago, 1953), 284.Google Scholar

74. Mann seems to reconstruct the argument along similar lines, though he does not use the word Mischsprache. It appears that he had discussed the problem with Guthrie himself. See Mann, W.M., “Internal Relationships of the Bantu Languages: Prospects for Topological Research” in Dalby, , Language and History in Africa, 137–38.Google Scholar

75. See for example Guthrie, , “Bantu Origins,” 17Google Scholar, or “Contributions From Comparative Bantu Studies,” 26.

76. My colleague P.F. de Moraes Farias has drawn my attention to a passage in the “Dream of Maxen” where the Welsh invaders of Brittany are said to have killed the men and not only raped the women but also cut out their tongues, so that their own language should not be contaminated. Gantz, J., ed., The Mabinogion (Harmondsworth, 1976), 126.Google Scholar This is quite irrelevant, but too good to be forgotten.

77. Tucker, , “Philology and Africa,” 550–51, 551n1.Google Scholar

78. As an example of what appears to be mumbo-jumbo, see the conclusions he reached from a study of Bantu-like features in Tiv: Guthrie, , “Bantu Origins,” 20.Google Scholar For some scathing comments by an American linguist see Welmers, W.E., “Review of Greenberg, Languages of Africa,” Word, 19(1963), 413–15.Google Scholar

79. Guthrie defined his “coefficient of generalness” as the number of “general” roots occurring in a given language divided by the total number of “general” roots. If we divide instead by the total number of roots recorded from the language in question, we can filter out any variation in ‘generalness’ which relates directly to variation in ‘commonness.’ Values for this coefficient were apparently calculated by Guthrie but not published in full: see Comparative Bantu, 2:14.Google Scholar As recalculated by Henrici, , ‘Numerical classification,’ p. 85Google Scholar, Table I, col. (d), they vary quite widely, between 30% and 50%, but do not show any obvious geographical pattern. In particular, the correlation between these values and those for Guthrie's “coefficient” is almost exactly zero (Spearman's R = −0.01). Conversely, if we define a Guthrie-like “coefficient of non-generalness,” we can confirm that “generalness” and “non-generalness” are correlated very highly (R = 0.82). The patterns of distribution differ slightly in detail, but by and large it is true that as the number of “general” roots decreases, so too does the number of “non-general” roots.

80. Guthrie, , Comparative Bantu, 1:102Google Scholar, topogram T.3.

81. Ehret, C., “Bantu Origins and History: Critique and Interpretation,” Transafrican Journal of History, 2(1972), 1–3.Google Scholar

82. Guthrie, , “Some Uses of Arithmetical Computation in Comparative Bantu Studies,” Transactions of the Philological Society (1964), 120–21.Google Scholar The multi-dimensional scaling performed by Henrici on Guthrie's data for the “test languages” would not have succeeded in the absence of a tendency towards connected distributions: Henrici, , “Numerical Classification,” 88–91.Google Scholar

83. This can be proved by using the same sample of roots already mentioned in n. 46. For any two zones the number of roots that can be expected to occur in both, under a null hypothesis of independence, is n₁ n₂/N, where n₁ and n₂ are the number of roots occurring in the first and second zones respectively, and where N = 123, the total number of roots in the sample. It turns out that for any two adjacent zones the number of roots which do in fact occur in both is greater than the expected number in every case apart from zones C and D.

84. Bynon, and Mann, , “Papers on Comparative Bantu,” 2Google Scholar, have made the point that heavy borrowing by one language from another is capable of generating regular correspondences. This, though true, is of doubtful relevance. In such a case, we would expect to find sets of conflicting correspondences -- one set for the items inherited by both languages from their latest common ancestor, and another set for the items acquired by borrowing. Guthrie was aware of this problem and dealt with it where it arose: Guthrie, , Comparative Bantu, 2:105–06.Google Scholar

85. Ehret, , “Bantu Origins and History,” 4–5.Google Scholar

86. The relevant sources are indicated in n. 43. For the discussion which followed Guthrie's paper at the SOAS conference in 1961 see Gray, R., “A Report on the Conference,” JAH, 3(1962), 184–87.CrossRefGoogle Scholar

87. Guthrie, , “Contributions from Comparative Bantu Studies,” 31–32.Google Scholar

88. Guthrie, , Comparative Bantu, 1:122.Google Scholar Because the root in question was missing from several “Eastern” zones, Guthrie thought it “emerged so late in the Proto-Bantu period that it did not spread to the whole of PB-B.”

89. For the “Pre-Bantu” theory see Guthrie, , “Some Developments,” 281–82Google Scholar, and “Bantu Origins,” 18-21; also Gray, , “A Report on the Conference,” 185–86.Google Scholar

90. Guthrie, , “Bantu Origins,” 21.Google Scholar

91. Gray, , “A Report on the Conference,” 186.Google Scholar

92. Guthrie, , Comparative Bantu, 2:111.Google Scholar

93. Gray, , “A Report on the Conference,” 186.Google Scholar

94. Guthrie, , “Problèmes de génétique linguistique,” 92.Google Scholar The “index of relationship” is another bizarre statistic (like the “modulus of dispersion”) that Guthrie devised for himself. It was defined for any given pair of languages as

where a, b, c are the numbers of roots that occur respectively in both languages, in the first but not the second, and in the second but not the first. This is excessively complicated, but it does measure what Guthrie was trying to measure. It is in fact related rather closely to the coefficient of “linguistic distance” used by Henrici, defined as

. Without much difficulty we can show that

. The difference in the resulting classifications, therefore, cannot be due -- except in some minor details -- to the difference in the coefficients employed. Note, however, that the values of D published by Henrici, “Numerical classification,” Table III, are based on the “general” roots alone, and so cannot be compared directly with the values given by Guthrie for the “index of relationship.”

95. Guthrie, , “Some Uses of Arithmetical Computation,” 116–27.Google Scholar

96. For the term “ramification” see Guthrie, , Comparative Bantu, 2:13.Google Scholar

97. The key reference here is Sokal, R.R. and Sneath, P.H.A., Principles of Numerical Taxonomy (San Francisco, 1963)Google Scholar, later revised as Sneath, P.H.A. and Sokal, R.R., Numerical Taxonomy: the Principles and Practice of Numerical Classification (San Francisco, 1973).Google Scholar It may be mentioned that much of the pioneering work in this field was inspired by a conception of scientific method somewhat similar to Guthrie's especially in its abhorrence of feedback. On this point see Hull, D.L., “Certainty and Circularity in Evolutionary Taxonomy,” Evolution, 21(1967), 174–89.CrossRefGoogle ScholarPubMed

98. Henrici, , “Numerical Classification,” 82.Google Scholar For Guthrie's reaction see Bynon, and Mann, , “Papers on Comparative Bantu,” 5.Google Scholar

99. In general the evidence is not clearcut, but this is the impression I get from checking the distribution among the “test languages” of various “osculant” and synonymous roots: Guthrie, , Comparative Bantu, 1Google Scholar: ch. 7. For example, there exist two quite different roots which both have the meaning “python,” and the distribution of these is compatible with Guthrie's tree, but not with Henrici's.

100. Guthrie, , “A Two-Stage Method,” 4.Google Scholar

101. Guthrie, , “Comparative Bantu: a Preview,” 40.Google Scholar

102. Advertisement by Press, Gregg, Journal of African Languages, 6(1967), 293.Google Scholar

103. Carter, , “Obituary,” 491.Google Scholar

104. One reviewer had “paste-ups and page-proofs” of the last three volumes by 1970: Hamp, E.P., “On Bantu and Comparison [review of Guthrie, Comparative Bantu],” International Journal of American Linguistics, 36(1970), 274n.Google Scholar

105. Guthrie, , Comparative Bantu, 2:7Google Scholar, with emphasis added.

106. Bynon, and Mann, , “Papers on Comparative Bantu,” 2.Google Scholar

107. See above, n. 88.

108. Guthrie, , Comparative Bantu, 1:61–63.Google Scholar

109. Guthrie, , “Contributions From Comparative Bantu Studies,” 20–22, 33.Google Scholar This paper was published in 1970, but may have been written as much as three years earlier. In the following quotations emphasis is mine.

110. A classic reference here, for British linguists at least, is Allen, W.S., “Relationship in Comparative Linguistics,” Transactions of the Philological Society (1953), 52–108Google Scholar (where Guthrie, Classification, is cited several times). For an outsider's view of the problem see Ardener, E., “Social Anthropology and the Historicity of Historical Linguistics” in Ardener, E. ed., Social Anthropology and Language (London, 1971), 218–19.Google Scholar

111. See the collection of papers published in African Language Studies, 14(1973), 1–104Google Scholar: this includes the articles already cited by Bynon and Mann (n. 66), Mann (n. 39), and Henrici (n. 55), as well as an important contribution by Meeussen, A.E., “Comparative Bantu: Test Cases for Method,” 6–18.Google Scholar A long article by Greenberg (n. 72) appeared in 1972, as did Ehret's deadly critique of the “Bantu nucleus” (n. 81).