¹ I am grateful to Y. Bastin, A. Coupez, M. Mann, J. Miller and T. Spear, who have commented on an earlier draft of this paper. Current views of the dynamics involved are still based on the scenario developed by Oliver, R., ‘The problem of the Bantu expansion’, J. Afr. Hist., VII (1966), 361–76.CrossRefGoogle Scholar See also Curtin, P. D. et al. , African History (Boston, 1978), 25–30.Google Scholar

² The linguists concerned are preparing their own publications to present and discuss the new evidence. The computer inputs and outputs can be consulted at the Royal Museum for Central Africa in Tervuren. This article discusses merely its historical implications. Lehmann, W., Historical Linguistics: An Introduction (New York, 1962)Google Scholar, is still excellent for beginners. Among the more advanced texts, Hock, H. H., Principles of Historical Linguistics (Berlin, 1986)Google Scholar, and Anttila, Raimo, An Introduction to Historical and Comparative Linguistics (New York, 1972)Google Scholar, who relies less on Indo-European examples than other textbooks do, are recommended.

³ On sociolinguists, see Hock, , Principles, 627–61.Google Scholar Once a written form of a language is created from one or more of its dialects, this becomes the standard and is often called the language, while all the oral forms are then dubbed dialects.

⁴ Hence historical linguistics are often called comparative linguistics.

⁵ Nurse, D. and Hinnebusch, T., Swahili and Sabaki: A Linguistic History (Berkeley, 1993).Google Scholar is an excellent example of such an endeavor.

⁶ Swadesh, M., ‘Towards greater accuracy in lexicostatistic dating’, International Journal of American Linguistics, XXI (1955), 121–37CrossRefGoogle Scholar, and earlier references there.

⁷ A full discussion of lexicostatistical procedures is provided by Embleton, S. M., Statistics in Historical Linguistics (Bochum, 1986).Google Scholar

⁸ Other factors involved are the exact list of rubrics used, the sources of the data and the counting procedures employed. The two published earlier attempts still in use are Heine, B., Hoff, H. and Vossen, R., in ‘Neuere Ergebnisse zur Territorialgeschichte der Bantu’, Zur Sprachgeschichte und Ethnohistorie in Afrika (Berlin, 1977), 57–70Google Scholar (on 147 languages and many lists drawn from dictionaries), and Bastin, Y., Coupez, A. and de Halleux, B., ‘Classification lexico-statistique des langues bantoues (214 relevés)’. Bulletin de l'Académie Royale des Sciences d'Outre-Mer, XXVII (1981), 173–99.Google Scholar The Tervuren study produced an unpublished tree for western bantu in 1983. The reason for the sensitivity of the results to the total number of languages included derives from the fact that the Bantu languages form a huge single pool of dialects. When only a fraction of the total number of languages is tested many places in the dialect network between the languages are unrepresented and the nature of the exact links between the languages tested may remain obscured.

⁹ On the procedures to assess retention and the eight items on the Swadesh list that had to be discarded, Bastin, Coupez and de Halleux, ‘Classification lexico-statistique des langues bantoues’.

¹⁰ Most of the missing languages are spoken in southern Tanzania, south-eastern Angola and an adjacent area in Zambia between the Zambezi and Cuando Rivers. For southern Tanzania the comparative study of Derek Nurse allows one to foresee that their addition to this universe would probably introduce only minor and local adjustments. Cf. Nurse, D., ‘The diachronic background to the language communities in southwestern Tanzania’, Sprache und Geschichte in Afrika, IX (1988), 15–115.Google Scholar The situation in south-western Zambia, known from unpublished data, is similar. More worrisome are the ommissions of presumably Northern Bantu languages such as Homa (Sudan) and Kari, Boguru and Bali (north-east Zaire), for which only scanty and insufficient data are available.

¹¹ This in turn yields data about the relative stability of each of the 92 items compared to all the others.

¹² Assumptions, or the ‘weights’ given each possibility, involve priorities given either to the primacy of internal cohesion (connectivity) within identified language clusters (exclusivity) or to the primacy of connections between individual languages belonging to different clusters. Three figures are involved: (a) the lowest percentage of agreement within a cluster; (b) the percentage by which the cluster is linked to the most closely related cluster; and (c) the highest percentage at which any language within the cluster links up with a language outside of it. Figure (b) must always be lower than (a) and ideally higher than (c). In practice, borrowing from neighbors can render (c) higher than (a) or even (b), thus compromising the exclusivity from the cluster. In drawing a tree one must choose an intermediate between two extreme situations. These are the choice of not linking any cluster to any other before one has complete linkage, or total exclusivity (the statistical index of connectivity, VN = 100%) and the choice to link any language directly to its ‘nearest neighbor’ even if this neighbor does not belong to the same cluster (VN = 0%). The latter situation is called total connectivity.

¹³ Guthrie, M., Comparative Bantu (4 vols.) (Farnborough, 1967–1971).Google Scholar For the emendations, see Bastin, Y., Bibliographie bantoue sélective (Tervuren, 1975)Google Scholar (Musée Royal de l'Afrique Centrale: Archives d'Anthropologie 24), 1, and in the references there. For the linguistic criteria used, see Guthrie, M., The Classification of the Bantu Languages (London, 1948)Google Scholar, and Meeussen, E., ‘De talen van Maniema’, Kongo Overzee, XIX (1953), 385–90.Google Scholar

¹⁴ M. Mann, personal information.

¹⁵ In a situation where parents are dialects of the same language, mixed languages (Mischsprachen) are perfectly possible. To insist on a single ancestor becomes unrealistic when two neighboring related ancestral languages strongly influence the emerging new language. Such cases have often occurred in the Bantu linguistic continuum. One can come close to a mixed language situation even when one of the parent languages is very different from the other. The case of Mbugu in Tanzania is notorious. Cf. Whiteley, W. H., ‘Linguistic hybris’, Afr. Studies, XIX (1960), 95–7.Google Scholar

¹⁶ This section is based on a detailed study of the massive new data available at Tervuren. The situation of Bubi, a language of the northern division of North-west Bantu on the island of Bioko, makes this clear. The position of Seki (Muni estuary) on the various trees can also be explained only by seaborne travel.

¹⁷ Some trees include only Babole and Aka and even have Babole and Aka split before West Bantu splits off from East Bantu. Others include only Ngondi-Pande and place Babole-Aka in a Northern Zaire cluster.

¹⁸ Stahl, A. B., ‘Intensification in the west African Late Stone Age: a view from central Ghana’, in Shaw, T. et al. (eds.), The Archaeology of Africa: Food, Metals and Towns (New York, 1993), 261–73Google Scholar; Holl, A., ‘Cameroun’, in Lanfranchi, R. and Clist, B. (eds.), Aux origines de l'Afrique centrale (Libreville, 1991), 150–2.Google Scholar

¹⁹ As in the syntheses of Phillipson, David W.. Cf. his The Later Prehistory of Eastern and Southern Africa (New York, 1977), 210–30Google Scholar, and African Archaeology (Cambridge, 1983), 179–81.Google Scholar

²⁰ Migrations from the Sahara into the Niger bend area and westwards are very well documented, but despite archaeological research no such evidence turns up in northern or central Nigeria.

²¹ Vansina, Jan, Paths in the Rainforests (Madison, 1990), 71–100.Google Scholar The proto vocabulary which documents this (pp. 267–301) is not based just on what is now known to be West Bantu but includes the north-eastern and the north-western clusters of North Bantu. It therefore refers to the period when North Bantu was spreading and before narrow West Bantu emerged. While the assumption that agriculture leads to spurts in population increase seems to be supported in some cases (e.g. in the lower Nile valley) it is not inevitably so and should be proven rather than assumed.

²² B. Clist in Lanfranchi and Clist (eds.), Aux origines, 181–3, 225, 226Google Scholar, shows that new technologies, including metallurgy, diffused only very slowly. In Gabon, foragers, ‘neolithic’ food producers and iron users lived side by side for over four hundred years, which shows how relatively unimportant the innovations were. The presence of ceramics on Late Stone Age sites in Gabon, and especially on a site near Libreville dated to c. 3000 B.C. or even 4500 B.C. suggests an independent diffusion of ceramics among foragers. Clist also found that by 4200 B.C. foragers crossed the 25-km.-wide estuary of the Gabon river to fetch basalt stone (ibid. 166–7), which presupposes navigation equipment. Elsewhere in the Bantu-speaking area there are other cases of separate diffusions for pottery, herding, the farming of grain crops and metallurgy.

²³ In theory, careful archaeological excavation of both farming and contemporary foraging settlements should show the relevant differences. Even so, one may not retrieve many of the relevant but small differences between foragers and food producers.

²⁴ The scarcity of genetic data coupled to an a priori acceptance of the massive Bantu migration hypothesis renders the biological evidence inconclusive on this point. Cf. Excoffier, L. et al. , ‘Genetics and history of sub-Saharan Africa’, Yearbook of Physical Anthropology, XXX (1987), 151–94CrossRefGoogle Scholar, and especially 181–5; Irish, J. D., ‘Biological affinities of Late Pleistocene through modern African aboriginal populations: the dental evidence’ (Ph.D. thesis, Arizona State University, 1993).Google Scholar The scanty data would just as easily support the tentative scenario sketched here.

²⁵ Vansina, , Paths, 83–94Google Scholar, for the evidence which underlies the scenario sketched here.

²⁶ Several well known cases of language shift occurred in the last millennium, e.g. the spreading of Fang or Ntumu, in Gabon. But there are indications that north-western languages lost ground to those of the narrow West Bantu core as early as the first centuries before our era in the coastal areas of southern Gabon and Congo, and it is evident that even earlier language shifts must have occurred from North to West Bantu in the region of the great bend of the Zaire river.

²⁷ In most of Angola neither domesticated yams nor oil palm trees will grow. Trapping is much less efficient there than hunting, and foragers relied heavily on the latter. Immigrant farmers from further north could settle in only a few localities very far apart from each other and had to be taught new hunting techniques by the autochthons.

²⁸ C. Ehret, in numerous publications since 1967. For a recent synthesis, see Schoenbrun, D., ‘We are what we eat: ancient agriculture between the Great Lakes’, J. Afr. Hist., XXXIV (1993), 1–31CrossRefGoogle Scholar, and references to C. Ehret, 9, n. 30.

²⁹ Mangbele is an example of a Bantu language that died out as late as the nineteenth century through language shift in favor of a Mangbetu language. The Bergdamara and Coroka of Namibia and Angola may have abandoned a Bantu language for Khoi and San languages over a millennium ago.

³⁰ Proto East Bantu does not have a terminology for iron working. The distribution of the various forms for ‘iron’, ‘iron ore’, ‘to forge’, ‘smith’ and ‘hammer’ in the area support various interpretations but not a single origin. The question cries out for further study.

³¹ Collett, D. P., ‘The spread of early ironproducing communities in eastern and southern Africa’ (Ph.D. thesis, University of Cambridge, 1988)Google Scholar, concluded that the various ceramic styles cannot be grouped together into a single stylistic tradition.

³² Despite the claims made, especially by T. Huffman, that language, group identity and ceramic style in small-scale societies correspond (cf. Huffman, T., Iron Age Migrations: The Ceramic Sequence in Southern Zambia [Johannesburg, 1989], 5–9)Google Scholar, ceramic style and language are not always correlated. For instance, although all Bushong speak the same language the northern Bushong and their northern neighbors (who speak other languages) used household pottery made on the banks of the Sankuru river, whilst southern Bushong and their southern neighbors (who speak other languages) used household pottery of a different style made on the banks of the Lulua river.