Explaining Global Patterns of Language Diversity

Daniel Nettle

Merton College, Oxford OX1 4JD, United Kingdom

Received July 2, 1997; revision received March 23, 1998; accepted April 8, 1998

The six and a half thousand languages spoken by humankind are very unevenly distributedacross the globe. Language diversity generally increases as one moves from the poles toward theequator and is very low in arid environments. Two belts of extremely high language diversitycan be identified. One runs through West and Central Africa, while the other covers South andSouth-East Asia and the Pacific. Most of the world’s languages are found in these two areas. Thispaper attempts to explain aspects of the global distribution of language diversity. It is proposedthat a key factor influencing it has been climatic variability. Where the climate allows contin-uous food production throughout the year, small groups of people can be reliably self-sufficientand so populations fragment into many small languages. Where the variability of the climate isgreater, the size of social network necessary for reliable subsistence is larger, and so languagestend to be more widespread. A regression analysis relating the number of languages spoken inthe major tropical countries to the variability of their climates is performed and the resultssupport the hypothesis. The geographical patterning of languages has, however, begun to bedestroyed by the spread of Eurasian diseases, Eurasian people, and the world economy. © 1998

Academic Press

INTRODUCTION

The diversity of human language is oneof its most intriguing features, and overthe last decade scholars from several dis-ciplines have become interested in whatpatterns of linguistic diversity might tellus about the human past (Cavalli-Sforzaet al 1988; Nichols 1990, 1992; Renfrew1991; Mace and Pagel 1995; Dixon 1997). Atleast three senses of the term linguisticdiversity need to be distinguished. Firstly,there are regions of the world, such asCameroon or Papua New Guinea, wherethere are very large numbers of differentlanguages, and others where there arevery few. Such diversity is the topic of thispaper, and I will refer to regions withmany languages as being high in languagediversity, just as biologists refer to regionswith many species as high in species di-versity.

Language diversity needs to be clearlydistinguished from phylogenetic diversity of

languages, which is a matter of how manydifferent language families or branches oflanguage families are present.1,* Some-times the two types of diversity go to-gether, as in Papua New Guinea and thePacific. However, there is no necessaryconnection between them, and there aremany discrepancies. Central Africa, forexample, has hundreds of different lan-guages and so is high in language diver-sity. However, almost all of those lan-guages are very closely related, belongingto the Bantu family, and so the region islow in phylogenetic diversity. I will arguein this paper that the language diversity ofLatin America is quite low relative to com-parable parts of Africa and Asia. However,it is generally agreed (the claims of Green-berg [1987] notwithstanding) that the lan-guages of Latin America belong to dozensof different families, and so the phyloge-

* See Notes section at end of paper for all foot-notes.

journal of anthropological archaeology 17, 354–374 (1998)article no. AA980328

3540278-4165/98 $25.00Copyright © 1998 by Academic PressAll rights of reproduction in any form reserved.

netic diversity of the continent is high,something which has implications for ourunderstanding of the Latin American past(Nichols 1990).

The third sense of linguistic diversitywhich must be distinguished is that ofstructural diversity on some linguistic pa-rameter. For example, the languages ofthe world order their basic constituents ina number of different ways. There areSOV languages (Subject-Verb-Object),SVO languages (Subject-Object-Verb),VSO languages and so on. A set of lan-guages exhibits high structural diversityin word order if many different orders arerepresented among them. Structural di-versity will tend to be correlated with phy-logenetic diversity, since where there aremany different families there will often bemany different structural types of lan-guage. However, the connection is not anecessary one, since there can be severalstructural types within one family, or, ob-viously, several families with the samestructural type. Nonetheless Nichols(1992:250) found that high structural di-versity on the basic syntactic and morpho-logical parameters she looked at was in-deed coincident with high phylogeneticdiversity, both being typical of the Pacificand New World.

This paper, then, is an investigation oflanguage diversity, which is not closelyrelated to the other two types of linguisticdiversity. Structural diversity and particu-larly phylogenetic diversity give us infor-mation about the evolution of humanpopulations along the axis of time. Forexample, it is the lack of phylogenetic di-versity amongst the languages of Centraland Southern Africa which tells us thatthe spread of Bantu-speaking farmers totheir present locations was comparativelyrecent.

I will argue that language diversity alsocarries information of interest to the an-thropologist, archaeologist or historian.However, this information is not primarily

about events in deep time but about theorganisation of people in space. That is,where a region is dotted with many smalllanguages, we can infer that the popula-tion has arrayed itself, socially and eco-nomically, in many small (though not nec-essarily isolated) communities. Where asingle language is spread over the wholeregion, we can infer that powerful mech-anisms of regional integration have beenat work, and we may wish to ask whatthose are. This paper, then, asks the ques-tion of, what, in general, determines thesize of language communities found in ahuman population. In Section 1, I discussthe main geographical patterns in the dis-tribution of languages across the globe. InSection 2, I discuss the principal vectorswhich spread languages, and suggest thatthe theory developed in Nettle (1996) toexplain patterns of language diversity inWest Africa might explain patterns inother parts of the world too. That theorylinked the size of language groups, viasubsistence patterns, to climate. In Sec-tions 3 and 4, I test the theory for all themajor tropical countries by means of aregression analysis relating the number oflanguages spoken in each country to cli-matic variables. The data presentedstrongly support the theory.

Before turning to Section 1 and theglobal distribution of languages, however,a brief discussion of how languages areidentified is in order. Ascertaining the sizeof community speaking a particular lan-guage presupposes being able to delin-eate it, and throughout this paper I willrefer to counts of languages in particularcountries. I have been able to do this be-cause linguists routinely list and refer toentities called languages. However, theconcept is not unproblematic.

There is variation in speech norms bothwithin and between all known communi-ties. The problem for an analysis of lan-guage diversity is distinguishing languageboundaries from dialectal or idiolectal

355LANGUAGE DIVERSITY

variation. The criterion usually proposedto do this is that of mutual intelligibility,and formal techniques for measuring thishave been devised (Casad 1974). How-ever, not only are such techniques rarelyused in the field, but their very basis isproblematic. First, intelligibility is a gradi-ent phenomenon, and there can be vari-ous degrees of partial understanding. Onemay find chains of dialects in which adja-cent varieties are perfectly intelligible, butthose at opposite ends of the chain aredefinitely not. It is unclear how best todraw boundaries in such cases. Second,mutual intelligibility can be asymmetrical.Third, intelligibility varies a great deal ac-cording to the context and the particularspeakers involved, and fourth, intelligibil-ity depends upon much the parties in-volved want to understand each other,which is a product not of linguistic struc-ture but of social factors (Wolff 1959; Hud-son 1996:35-36).

Such arguments have lead many au-thors to stress the problematic nature ofthe term “language” (Hudson 1996:36; Ro-maine 1994: Chapter 1). Whilst acceptingits difficulties, I would argue that we neednot disregard as worthless data from fieldlinguists on the number of languages spo-ken in different areas. To see why, theparallel with the problem of individuatingbiological species can be considered.

The science of ecology makes and testsquantitative predictions about the num-ber of species which will be found in dif-ferent environments (see, e.g., Huston1994), yet the concept of species is scarcelyless problematic than that of language.For sexual organisms, the criterion cited isthat of (in)ability to interbreed, yet thiscriterion has all the difficulties of the(un)intelligibility criterion for languages.There are chain species, examples of par-tial ability to interbreed, and even asym-metrical interfertility (Mayr 1963; Sokaland Crovello 1970; Abruzzi 1982). The jus-tification for holding on to the concept of

species is that, although there are prob-lem cases, it is a convenient way of cap-turing real facts about biological diversity.I would make the same case for linguisticdiversity and thus justify holding on to thelanguage concept in the following ways.

First, though there are cases of blurredboundaries, there are many others wherethe communicative discontinuities arequite obvious, and indeed highly salientto the people concerned. Indeed, Dixon(1997:8) argues that the clear cases are typ-ical and the blurred ones rather rare. Sec-ond, the linguistic arbitrariness of the lan-guage/dialect distinction is relativelyunimportant, if linguists employ it ap-proximately self-consistently. As long asthis is the case, then counts of what fieldlinguists have deemed to be languageswill give a reasonable relative measure oflanguage diversity. For example, Romaine(1994: Chapter 1) points out the arbitrari-ness of the language boundaries usuallyidentified in Melanesia, which do not cap-ture the range of communicative practicesof this highly multilingual population.However, she is quite clear that there ismore diversity of something to do with vari-eties of language in Melanesia than in otherareas, and the conventional count of lan-guages does capture this fact, even if itfails to capture other aspects of the lin-guistic geography of the area. Linguistsseem to agree often enough in practiceabout what languages are for the assump-tion to be made that their results fromdifferent continents are roughly compara-ble. Third, although error is introducedinto the data by the indeterminacy of lan-guage boundaries, this error is effectivelyrandom. It is thus more likely to obscurereal trends than create apparent ones, andany clear patterns which are still evidentare likely to represent genuine effects. Fi-nally, the magnitude of variation in lan-guage diversity between different regionsis very large. If there was a difference inthe average number of square kilometers

356 DANIEL NETTLE

per language in, say, Papua New Guineaand Namibia, of two- or threefold, itwould be tempting to conclude that therewas nothing more going on than noise inthe data. However, the real difference is80-fold (about 540 km2 for Papua NewGuinea, and about 40,000 km2 for Nami-bia), and so it seems reasonable to look forunderlying causes.

In what follows, then, I shall proceed touse counts of languages without furthercomment, though it should of course beborne in mind that these carry a largemargin of error around them.

1. GLOBAL PATTERNS OFLANGUAGE DIVERSITY

There are many more languages spokenin the world than is commonly realised.The best available catalogue is the Ethno-logue, compiled by the Summer Instituteof Linguists, of which Grimes (1993) is arecent edition. Grimes (1993) lists 6528 liv-ing languages, as well as a number of ex-tinct ones.

What is even more striking than thesheer number of languages is the uneven-ness of their distribution. Grimes (1993)lists languages by country, so, althoughthe country is not an ideal unit of compar-ison, I will use it here. Eight hundred andsixty-two languages are spoken in PapuaNew Guinea. This is 13.2% of the globaltotal, yet Papua New Guinea representsonly 0.4% of the world’s land area, andonly 0.1% of the world’s population livethere.2 At the other extreme, Russia has1.5% of the languages and 15.3% of theland area. We can gain an idea of theglobal distribution by plotting the relativelanguage diversity of each country, as inMap 1.

The simplest way to do this would be todivide the area of countries by the numberof languages spoken and shade them ac-cording to this value. However, doing soartificially inflates the apparent diversity

of smaller countries, for the followingreason. Many languages cross nationalboundaries, and the smaller the country,the more of its languages tend to do so.For example, 21 languages are spoken inthe Gambia, which is a very narrow stripof land, but all of them are also spoken inSenegal or elsewhere. Dividing the area ofthe Gambia by 21 would thus give an ex-tremely misleading figure for the averagearea occupied by a Gambian language. Ihave taken two measures to minimizethe small country effect. First, countriessmaller than 50,000 km2 have been ex-cluded from the analysis. Second, ratherthan dividing the number of languages bythe area of the country, I have performeda regression of the logarithm of the num-ber of languages on the logarithm of thearea of the country.3 The number of lan-guages does increase with increasing area(ln[LANGS] 5 0.49 ln[AREA] 1 0.53; r 5.45, df 5 124, p , .001), but with consider-able scatter and hence a relatively low rvalue. The relative language diversity ofeach country has thus been determinedby the value of the standardized residualfrom this regression line.

The relative language diversity of all thecountries of the world larger than 50,000km2 is shown on Map 1. The darker theshading, the more relatively diverse thecountry. Note that the resolution of themap is only at the level of the country—countries like India and Mexico whichhave very diverse and less diverse regionsare shaded uniformly using the mean.

Several clear patterns can be observedin Map 1. First, language diversity tends tobe greatest near the equator and decreaseas one moves North or South away fromit, as several authors have noted (Breton1991; Nichols 1990, 1992; Mace and Pagel1995). Species diversity decreases in asimilar way as one moves away from theequator (Stevens 1989; Mace and Pagel1995).

Second, there are more specific associ-

357LANGUAGE DIVERSITY

ations between language and biologicaldiversity. In the Old World, there are twogreat belts of very high language diver-sity, which correspond almost exactly tothe two great belts of equatorial forestwhich harbour so many of the OldWorld’s species. One runs from IvoryCoast across West Africa into Zaire. Theother runs from South India and peninsu-lar Southeast Asia down through the In-donesian islands into New Guinea andthe islands of the Western Pacific. Total-ling up the languages listed in Grimes(1993) for the 17 main countries in thesetwo belts4 gives 3929, which is 60% of allthose in the world. As well as being plen-tiful where species are numerous, lan-guages are few where species are few.Large areas of white on Map 1 correspond

to the Sahara and Arabian deserts, whichthough in the tropics are arid and poor inspecies.

Third, the New World distribution isslightly different from that of the OldWorld. There is a latitudinal pattern, but thehighest diversity is not in Amazonia, as wemight expect, but in Mexico. Furthermore,the overall level of diversity is lower than inthe Old World. I will discuss possible rea-sons for this in Section 4 below.

None of these three patterns is a simpleproduct of there being more people in themore diverse areas, as I shall show in Sec-tion 4. The orders of magnitude of thedifferences are also very large. We have,therefore, some strong trends to explain:language diversity is inversely related tolatitude, is low in deserts and arid places,

MAP 1. Map of the world showing the relative language diversity of the major countries. This iscalculated by regressing the logarithm of the number of languages spoken in the country (Source:Grimes 1993) against the logarithm of the area of the country, and shading each country accordingto the standardised residual. The shading sheme is as follows: White (least diversity), zres , 20.5;Light dooting, 20.5 , zres , 0; Heavy hathcing: 0 , zres , 0.75: Black (most diversity): zres . 0.75.

358 DANIEL NETTLE

is high in the environment that producesequatorial forest, and is relatively low andanomalously distributed in the NewWorld. In the next section, I briefly discussthe vectors by which languages are spreadand outline the ecological theory fromNettle (1996) which, I will argue, may ac-count for the observed patterns.

2. VECTORS OF LANGUAGESPREAD

As we have seen, the vast majority of alllanguages are spoken by small communi-ties in the tropics. Most of these commu-nities are primarily dependent on non-industrial ways of making their living, orhave been until very recently. Few lan-guages are routinely written down, stillless used in television or radio. Very feware the official vehicle of any nation-stateor other large-scale political elite. In thissection, then, I will suggest and brieflyjustify a number of generalizations whichcan be drawn about the “typical” case ofhow a language develops. These are, first,that it is transmitted orally; second, that itpasses through primary social bonds;third, that state policies are not usuallyrelevant; and fourth, that it is encapsu-lated in the wider system of economic life.I will now explain each of these general-isations in turn.

First, language transmission is basicallyoral and informal. This means that lin-guistic norms pass between individualswho interact face-to-face. It seems to bepart of the nature of linguistic transmis-sion that, in the absence of positive inter-action, people’s languages diverge. It fol-lows that people who maintain the samespeech norms must be connected by socialor economic ties, and that lack of face-to-face interaction between groups will tendto mean that their languages diverge.

Second, however, the mere presence ofinteraction between people is no guaran-tee of their linguistic homogeneity. Socio-

linguistic differences can persist despiteroutine interaction between the people in-volved. Trade between groups can takeplace via lingua francas, or pidgins, or bi-lingual individuals, while the underlyingdifferences between their first languagesare undiminished.

What determines whether groups willconverge or remain uniform linguisticallyis a matter of social identification (LePage1968). Where individuals are closely in-volved enough to want to be identifiedwith each other, they will mutually ac-commodate their speech behavior. How-ever, they may also interact while activelymaintaining ethnolinguistic distance, andthe question of what determines whetherthey do so or not is an important one.

The answer seems to lie in the strengthand nature of the social bonds involved(Milroy 1980). It is convenient in this con-text to distinguish between primary andsecondary social bonds. Primary socialbonds are relatively enduring, are oftenformed early in life, and are multivalent.This means that they are not formed forany one specific purpose. Rather they aregeneralized social linkages (Sahlins 1972)which may bring the actors together inmany different contexts and at differenttimes. Social bonds within ethnolinguisticgroups in non-industrial societies typi-cally appear to have this character; theyform a dense web of relationships, whichare often reinforced by biological and cul-tural kinship, and which may form thebasis for common ritual activities and cel-ebrations, common defence, commonfarm work or hunting, gifts, and foodsharing, as well as trade narrowly defined.In most societies, these primary bondshave also been those on which peoplehave depended for their basic livelihood.

Secondary social bonds, by contrast, arebased on specific functions, such as atrade in a specialized good like salt ormetal, or a specialized service. Purelycommercial trade creates only a second-

359LANGUAGE DIVERSITY

ary bond, as reciprocation tends to be im-mediate, and no future moral responsibil-ity for the other party is entailed (this isSahlins’ [1972] “balanced reciprocity”).Secondary bonds are associated withgreater social distance than primary onesand are more typical of the relationshipsbetween ethnolinguistic groups thanthose within them. I would argue that sec-ondary bonds are not themselves suffi-cient for sociolinguistic identification.Thus, whether social interaction will leadto the adoption of a common mothertongue is determined whether the socialbonds created are primary or secondary.

Like almost all dichotomies, the distinc-tion between primary and secondarybonds simplifies an underlying contin-uum. Pairings of farming and hunting orherding peoples give a good example ofthis. The pygmies of central Africa, forexample, are specialized hunters, whononetheless consume a great deal of theirdiet in cereals grown by neighboringfarmers. It is unlikely that they could sur-vive reliably by hunting and gatheringalone (Bailey et al. 1989), and so this rela-tionship is of vital importance to them,although it is in origin a single-purposerather a multivalent one. Each pygmygroup is paired with a particular group offarmers. The interdependence is so greatthat, in each case, the pygmies no haveaccommodated to the farmers and nolonger have a distinctive language.

Further south in Africa, in the Kalahari,San hunter-gatherers also depend uponexchange with neighboring farmers, yethave remained linguistically distinct(Headland and Reid 1989). In the WestAfrican savanna, one finds widespreadsymbiosis between Fulani pastoralists andcereal farmers, particularly the Hausa.Even the most purely pastoral Fulani con-sume a high proportion of their calories ascereals which they have traded for milkand other products (Swift 1986:180). How-ever, the distinction between Hausa and

Fulani remains, and a relatively small pro-portion of those Fulani who are still herd-ers have adopted Hausa as a mothertongue (though most speak it as a linguafranca). The difference in outcome fromthe pygmy case may reflect a difference inthe extent of investment in the particularexchange relationship in the differentcases. Fulani pastoral networks are ex-tremely extensive, and go well beyondHausaland, and the San seem to haveboth a wide choice of trading partners andgreat flexibility in their own subsistencearrangements. Neither San nor Fulanihouseholds are, then, as irreversibly at-tached to their particular agriculturalistpartners than the pygmies, and so theyhave retained their distinct mothertongues.

Accepting that the distinction is some-what simplistic, then, we can nonethelessassert that, in general, people come tohave the same first language as those towhom they are linked by primary bondsin the sense I have described.

The third generalization about lan-guage transmission is that the influence ofspecialized political and governmentalstructures is of limited importance, atleast with regard to mother tongues.There are obviously some cases of the do-main of a language being determined bythe extent of a particular political forma-tion, such as the spread of vernacularLatin through Europe as part of Romanexpansion. However, there are manymore cases where great empires have con-trolled areas for long periods of time with-out ordinary country people coming tospeak the language of the elite at all, oronly learning it as a secondary tongue foruse in dealing with government. Theoverwhelming majority of languages eventoday are unofficial, minority tongues, andmost persist despite the incorporation oftheir speakers in wider state, religious andregional systems. Even in densely popu-lated, early industrialized Europe, “the

360 DANIEL NETTLE

virtual identify of language, state and na-tion was approximated only in the nine-teenth century” (Coulmas 1992:33), and inmany European countries minority lan-guages are still spoken. Overall, it is rea-sonable to conclude that the determinantsof which first language one learns are usu-ally based in very quotidian and local so-cial situations, rather than in courts orparliaments.

The final generalization about languagetransmission is that the spread of a lan-guage is rooted in an economic system. Itmight appear that the choice of primarysocial associates is a purely social or cul-tural matter. However, in non-industrialsocieties where relatively little of the cir-culation of food and labor passes throughthe wages and money, there is no realdistinction between economy and society.An individual’s social associates also tendto be those with whom he labors in thefields; or from whom he borrows land,livestock, or seeds; or with whom he com-bines to appropriate and defend land.Bonds which are social in character arecemented by real exchanges of food andservices, and it is to social associates thatpeople turn in time of shortage (Colson1979; for ethnographic examples, see e.g.Watts 1983; Cashdan 1985; Legge 1989). Itseems, in general, that languages arerooted in networks of social bonds whichhave a real economic importance. This is akey assumption of the ecological theorywhich I will advance below.

We have now discussed in generalterms the vectors by which languages arespread. Clearly, there are a small numberof languages which present a very differ-ent profile to that outlined above. For ex-ample, the last 500 years has seen a smallnumber of Eurasian languages spreadwidely, first in Eurasia, and then acrossthe globe, by elite political, technological,and demographic dominance. I would ar-gue that these events are rather untypicalof language evolution as a whole. This is

not because there have not been local lan-guage spreads at other times and otherplaces; there always have, as politieswaxed and waned. The difference is thescale of the European spreads, whichwas quite unprecedented. These recentspreads have until very recently affectedonly a fairly small proportion of theworld’s languages, since in most areas,indigenous languages have persistedalongside the colonizers. The Americasare an exception, which we discuss furtherin Sections 4 and 5. Perhaps a few dozenof the six and a half thousand languageswere spread to their present range by re-cent elite dominance, with a few hundredalready obliterated by their expansion.For the great bulk of the remainder, myearlier assertion that more local factorsshould be considered seems justified.

Thus we can return to our basic task ofexplaining the global distribution shownin Map 1. It is clear from the foregoingdiscussion that the formation of any par-ticular ethnolinguistic group will be acomplex interplay of many locally specificfactors; formation of social bonds will de-pend upon precise topographical, mili-tary, epidemiological, demographic, andcultural situations, as well as more nebu-lous contingencies such as the rise and fallof local prestige and influence. However, Ibelieve general explanations are appro-priate for the global trends. The mainpatterns—the latitudinal gradient, theparallel with species diversity—patternstrongly with biological or ecological phe-nomena, which suggests that the appro-priate theory will be one that links humanagents to their ecological setting.

Individuals in any non-industrial soci-ety have to simultaneously solve manydifferent ecological problems, from cop-ing with disease, to providing fuel, fertil-izer, and drinking water, to optimizingintergroup relations and population pres-sure. All of these may influence the evo-lution of their social systems. In my West

361LANGUAGE DIVERSITY

African study (Nettle 1996), I made out acase that one environmental factor couldbe identified which had a preeminent in-fluence on many aspects of life, especiallythe formation of linguistic groups. I calledthat factor ecological risk.

Ecological risk is the amount of varia-tion which people face in their food sup-ply over time. Variation can occur bothseasonally (within a year) and inter-annu-ally, with periodic years of shortage.Households cannot operate for more thana few days without food and so must de-velop mechanisms to deal with temporaryshortfalls in its supply.

There is a large literature on the mech-anisms by which people cope with risk(e.g., Cashdan 1985; Minnis 1985; Huss-Ashmore, Curry and Hitchcock 1988; De-Garine and Harrison 1988; Halstead andO’Shea 1989; Cashdan 1990). Four keystrategies may be identified ([Halsteadand O’Shea 1989:3]; Colson [1979:21] di-vides them into five rather than four buther schema is basically the same). Theseare diversification of productive and in-come-generating activities, storage offood, mobility, and exchange.

In the current context, I will discussonly the mechanism most relevant to lan-guage diversity, that of exchange. In Net-tle (1996), I argued that patterns of socialexchange, which determined languagespread, were related to patterns of rain-fall. Where rainfall is continuous throughthe year, households can reliably producefood all year round and so are little de-pendent on exchange outside the imme-diate vicinity for subsistence purposes.Where small groups are basically self-sufficient, there is no incentive for them toform networks of primary bonds outsidethe immediate vicinity, and so many smalllanguages evolve. Where there is amarked dry season or probability of adrought year, on the other hand, house-holds form social ties over a wide area, togain access to resources elsewhere in time

of local shortage. Here, the languages aremuch more widespread, as the extensivesocial networks spread linguistic normsover a wider area. In the extreme case ofthe northern savanna of West Africa, pas-toralists cope with between the brief anderratic rainy seasons by moving, but alsoby exchange of stock and services through“wide networks of kinsmen throughoutWest Africa” (Berg 1976:24; cited in Legge1989). Correspondingly, languages likeFulfulde are spread over millions ofsquare kilometers.

The theory which emerged from thediscussion in Nettle (1996) was, then, thefollowing: the greater the ecological risk,the larger the social networks people mustform to ensure a reliable supply of basicsubsistence products. Since linguisticnorms are spread through those same so-cial networks, it follows that the averagesize of a language group should also in-crease as ecological risk increases. Thisprediction was tested using a regressionanalysis and was found to hold true bothwhen the size of a language was estimatedin terms of area occupied and when it wasestimated in terms of number of speakers.

In this paper, I test whether the predic-tions of the ecological risk theory hold forother parts of the world as well as WestAfrica. I use the country as the basic unitof analysis. This is convenient because thebest available dataset on global languagediversity (Grimes 1993) is organized bycountry. However, in many ways, coun-tries are troublesome units of comparison.First, they are not all the same size. Sec-ond, they are not ecologically homoge-neous, and, third, languages overlap theirboundaries. All of these problems have tobe taken into consideration in the statisti-cal analysis, as I will describe in Section 3below.

In Nettle (1996), the magnitude of eco-logical risk for each location was mea-sured by calculating the number ofmonths in the year in which enough rain

362 DANIEL NETTLE

falls for useful plant growth to occur usingthe Growing Season formula (LeHouerou1989), which is described in Section 3 be-low. The formula captures the extent towhich biological production is variablewithin a year. It may also be a reasonableproxy for how much it varies betweenyears, since locations with shorter rainyseasons tend to have more erratic rainfallthan those with longer ones. The formulais also used here, and an average value ofthe Growing Season (the Mean GrowingSeason or MGS) is obtained for eachcountry using climatological records. TheMGS can vary from 0 (no months in whichthere is adequate rainfall for food produc-tion, hence no possibility of self-suffi-ciency and extremely high risk) to 12months (continuous rainfall and hencefood production all year round, very lowrisk). The Growing Season formula takesaccount of rainfall and temperature. Theassumption implicit in its use is thus thatthese are the main limiting factors on foodproduction. This is certainly a simplifica-tion. It may, however, be reasonably accu-rate at least for the tropics. In the temper-ate latitudes, variation in day length andthe presence of frost become more impor-tant, and for this reason as well as forsome others given in Section 3, only trop-ical countries are considered in the re-gression analysis.

Extensive social networks are effectivein mitigating risk for two slightly differentreasons, and these give rise to two differ-ent hypotheses to be tested. First, increas-ing the spatial extent of a social networkgives individuals in it access to more dif-ferent microenvironments and types ofland, which may have food products avail-able at different times. This spatial averag-ing of ecological risk leads to the predic-tion that the spatial extent of languagegroups should increase as the degree ofecological risk they are exposed to in-creases. Converting this into the number

of languages per country gives the firsthypothesis to be tested:

Hypothesis 1. The longer the Mean GrowingSeason, the more languages there will be spokenin a country of a given area.

Second, increasing the number of pro-ductive households in a social networkdecreases the variation experienced by allof them as a simple consequence of thelaw of large numbers. This is true as longas there is some degree of statistical inde-pendence between them. This numericalaveraging leads to the prediction that thenumber of individuals in a languagegroup should increase as the degree ofecological risk increases. Converting thisinto the relationship between MeanGrowing Season and languages per coun-try gives the second hypothesis.

Hypothesis 2. The longer the Mean GrowingSeason, the more languages there will be spokenin a country of a given population.

The two hypotheses would only beequivalent in a world where populationdensity was the same everywhere. I willtherefore test them separately. In Section3, I give the details of the methods used todo this. In Section 4, I give the results.

3. TESTING THE THEORY:METHODS

Ecological and linguistic data were col-lected for all the countries of the worldover 50,000 km2. Smaller countries wereexcluded because the problem of lan-guages crossing national boundaries in-flates their apparent diversity, as de-scribed in Section 1 above. Exceptionswere made for the Solomon Islands andVanuatu, since, though both these coun-tries are small, they are rich in languagesnone of which crosses their borders asthey are islands.

The analysis was restricted to countriesfalling wholly or mainly within the tropi-cal zone. This is strictly defined as the area

363LANGUAGE DIVERSITY

between the tropics of Capricorn andCancer, but for the present purposes, thebroader definition of the area between30°N and 30°S is preferable, as it capturesmore of the area of the climatic regimeswe know as “equatorial” and “tropical.”The analysis has not been extended to thetemperate latitudes, for several reasons.First, the available measure of ecologicalrisk (the Mean Growing Season) takes ac-count only of rainfall and average temper-ature, and thus cease to realistically reflectthe possibilities of food production as onemoves out of the tropical zone. Second,the economies of the tropical countries aremore rural and more dominated by sub-sistence activities than those of most ofthe temperate countries. The linguisticmap of, say, European countries, reflectsto a considerable extent the rise of stan-dard national languages since the indus-trial revolution, rather than longer termecological processes.

The Growing Season formula (Le-Houerou 1989) was employed as a mea-sure of ecological risk. A month is in-cluded in the growing season if theaverage daily temperature is more than6°C and the total precipitation in millime-ters is more than twice the average tem-perature in centigrade. The Growing Sea-son is an inverse measure of ecologicalrisk: the more growing months there are,the less the ecological risk.

The Mean Growing Season (henceforthMGS) was found for each country usingmeteorological records in Wernstadt(1972). The number of weather stationsper country varies from 100 to over 200,and the time period for which informationis available varies from a few years to 50 ormore. The stations are designed to be dis-tributed evenly around each country.Since countries are not ecological units,and are often large enough to span manyecological regimes, there is in some casesa danger of producing a meaningless “av-erage” climate which does not correspond

to that experienced by any of its commu-nities. As a safeguard against this, coun-tries where the standard deviation ofthe growing seasons from the differentweather stations was greater than 2months were excluded. This affected 19countries, which are marked with an as-terisk in the Appendix.

The number of living languages spokenby a resident community in each country(LANGS) was obtained from the Ethno-logue 12th edition (Grimes 1993). For thereasons given in Section 1, simply divid-ing the area or population of each countryby the number of languages exaggeratesthe diversity of smaller countries. I haveinstead taken the area of the country (forHypothesis 1) or its population (for Hy-pothesis 2) as one independent variable inthe multiple regression analysis, the MeanGrowing Season being the other.

A number of countries were included inthe sample used in Nettle (1996). As thisstudy is intended to independently testthe applicability of the theory to other ar-eas, those countries were also excludedfrom most of the analyses. The 13 affectedcountries are marked with a dagger (†) inthe Appendix.

There are two different linguistic vari-ables of interest, corresponding to the twodifferent hypotheses. One is the numberof languages spoken relative to the coun-try’s area, the other the number of lan-guages relative to the number of inhabit-ants. To calculate these variables, figuresfor both the area (AREA) and the popula-tion (POP) of each country were obtainedfrom the UN Demographic Yearbook. Thepopulation figures are mid-year estimatesfor 1991.

To reduce skewness and kurtosis. loga-rithms were taken of the LANGS, AREAand POP variables. The skewness andkurtosis of the MGS did not differ signif-icantly from 0, and so it was left untrans-formed.

Obviously, both the ecological and lin-

364 DANIEL NETTLE

guistic variables are highly approximate,the former because of the simplicity of theformula and the unevenness of the data,the latter because of the inherent difficul-ties involved in counting languages. Weshould thus expect at best an approximaterelationship.

4. TESTING THE THEORY: RESULTS

The appendix displays the number ofliving languages (LANGS), population(POP), area (AREA), number of completeweather station records, and Mean Grow-ing Season (MGS) for 74 countries, includ-ing, for the sake of completeness, the WestAfrican countries and those with variableclimates, which are excluded from the re-gressions unless otherwise stated.

Hypothesis 1. The longer the Mean GrowingSeason, the more languages there will be spokenin a country of a given area.

Multiple regression shows that this is in-deed the case. The equation produced is:

ln[LANGS]

5 0.58 ln[AREA] 1 0.23 MGS 2 5.32

~43 countries: r 5 .66, df 5 40, p , .001!

The relationship between Mean GrowingSeason and the number of languages oncethe area of the country has been con-trolled for is shown in Fig. 1. If the WestAfrican countries are included, it is obvi-ous that they sit in the same distribution.Indeed, the regression equation includingthem is virtually the same:

ln@LANGS#

5 0.53 ln@AREA# 1 0.24 MGS 2 4.77

~55 countries: r 5 .65, df 5 52, p , .001!

It is also immediately apparent that all theLatin American countries (including the Ca-

ribbean) fall well below the rest of the dis-tribution. There are a number of possiblereasons for this. One is simply sparsity ofpre-colonial population. Even now only asmall fraction of the land in tropical SouthAmerica is populated (less than 5% accord-ing to Partridge 1989:5). It is therefore nosurprise that there are fewer groups thanmight be expected. Mexico, where diversityis the highest in the Americas, was alsomore thoroughly peopled at the time of Eu-ropean expansion than areas further South.Perhaps a stronger reason for low diversityin the Americas is that so many indigenouslanguages have been lost. There are a num-ber whose demise is known about, butmany more who must have disappearedwithout record, principally through the in-fectious diseases which ravaged the conti-nent in the decades following Europeancontact (Crosby 1986). In Cuba, the mostextreme outlier in Fig. 1, the indigenouspopulation was decimated within a very fewyears of European contact (Niddrie 1971:82),and the precolonial linguistic situation islargely unknown.

If the Latin American countries are ex-cluded, the regression relationship ismuch improved (Fig. 2):

ln@LANGS#

5 0.42 ln[AREA] 1 0.30 MGS 2 3.48

~36 countries: r 5 .82, df 5 33, p , .001!

The relationship is not just a product ofcomparing different continents. Signifi-cant relationships are also found withinthe Asia/Pacific and African groups. Thedata from Latin America are too few toobtain an independent multiple regres-sion.

Asia/Pacific:

ln@LANGS#

5 0.49 ln@AREA# 1 0.32 MGS 2 4.22

~18 countries: r 5 .87, df 5 15, p , .001!

365LANGUAGE DIVERSITY

FIG

.1.

Th

en

um

ber

ofla

ngu

ages

inea

chco

un

try

regr

esse

dag

ain

stth

em

ean

grow

ing

seas

onin

mon

ths

once

the

effe

cts

ofth

eco

un

try’

ssi

zeh

ave

been

con

trol

led

for.

366 DANIEL NETTLE

FIG

.2.

Th

en

um

ber

ofla

ngu

ages

inea

chco

un

try

regr

esse

dag

ain

stth

em

ean

grow

ing

seas

onin

mon

ths

once

the

effe

cts

ofth

eco

un

try’

ssi

zeh

ave

been

con

trol

led

for,

wit

hth

eL

atin

Am

eric

anco

un

trie

sex

clu

ded

.

367LANGUAGE DIVERSITY

Africa: (West African countries ex-cluded)

ln@LANGS#

5 0.50 ln@AREA# 1 0.23 MGS 2 4.37

~18 countries: r 5 .77, df 5 15, p , .01!

The remarkable similarity of the indepen-dent equations from the two continentalregions show what a powerful and univer-sal determinant ecological risk is.

Hypothesis 2. The longer the Mean GrowingSeason, the more languages there will be spokenin a country of a given population.

Multiple regression again confirms thatthis is the case. The equation is:

ln@LANGS#

5 0.33 ln@POP# 1 0.18 MGS 2 0.62

~43 countries: r 5 .63, df 5 40, p , .001!.

Again, the WestAfrican countries sit in thesame distribution. Including them givesthe equation:

ln@LANGS#

5 0.34 ln@POP# 1 0.18 MGS 2 0.65

~55 countries: r 5 .64, df 5 52, p , .001!

Once again, excluding the Latin Americancountries improves the relationship.

ln@LANGS#

5 0.25 ln@POP# 1 0.22MGS 2 0.32

~36 countries: r 5 .78, df 5 33, p , .001!

Once again, the relationship holds inde-pendently in the two major continents:

Asia/Pacific:

ln@LANGS#

5 0.24 ln@POP# 1 0.26 MGS 1 0.02

~18 countries: r 5 .82, df 5 15, p , .001!

Africa: ~West African countries ex-cluded!

ln@LANGS#

5 0.10 ln@POP# 1 0.20 MGS 1 1.55

~18 countries: r 5 .64, df 5 15, p , .05!

The predictions of both ecological riskhypotheses are clearly met for the wholeof the tropical world. If the previously ex-cluded countries with very variable grow-ing seasons are included, the significanceof the relationships is not changed, the rvalues being slightly lowered (Hypothesis1: r 5 0.61, df 5 58, p , .001; Hypothesis 2:r 5 .54, df 5 58, p , .001; Latin Americaincluded, West African countries ex-cluded in both cases).

The results of this analysis show a con-sistently strong relationship between lan-guage diversity and climatic patterns.They suggest that ecological risk has beenan extremely important factor—probablythe most important single factor—in peo-ple’s strategies of group formation andcommunication in the tropical world as awhole. The r values using the two differ-ent hypotheses do not differ significantly,so it is not possible to conclude that eitherthe spatial or the numerical process ismore important. Usually, they go hand inhand.

5. CONCLUSIONS

It seems, then, that a key determinant oflinguistic diversity is to be found in thebasic facts of ecology and subsistence.There are many other factors, too, whichwe have not considered here and nodoubt contribute greatly to the observed

368 DANIEL NETTLE

variance. A more sophisticated analysiswould incorporate such factors as relief,epidemiology and population density. Itwould also consider the influence of dif-ferent subsistence systems on the rela-tionship of social network and ecology.Access to marine resources, for example,allows communities to diversify whenfaced with seasonal shortages. Livestockherding also functions as a form of proteinstorage, and the available species of stockvaried widely between areas in traditionalsocieties (Diamond 1997a: Chapter 9).Nonetheless, given the unavoidably ap-proximate nature of the data used, thecorrelations are impressive and testify to astrong influence of ecology on the historyof human communities.

The assumption of this paper has beenthat the current pattern of language diver-sity represents some kind of ecologicalequilibrium. This assumption seemslargely justified by the correlations ob-tained. However, it should be remem-bered that the equilibrium observed is aproduct of a particular historical period,since almost all of the societies on whichthe data set is based are, or were untilrecently, subsistence farmers and herders.The pattern of language diversity was nodoubt very different when the continentswere covered with hunter-gatherers andwill again be very different as the changesassociated with the modern world econ-omy become more widespread.

Before the transition from hunting andgathering to farming, the amount of lan-guage diversity may have been evengreater than it is now, since hunter-gath-erer communities tend to be smaller thanfarming communities living in similarhabitats. In the Kalahari, San hunter-gath-erer groups number a few thousand whiletheir Tswana farmer neighbors are twomillion (Grimes 1993). In aboriginal Aus-tralia, communities never exceeded about500 individuals, though their territorysizes varied according to the ecological

productivity of the area (Birdsell 1953).Thus even in arid areas, very large com-munities did not evolve. This difference ofoutcome may reflect the different subsis-tence strategies of farmers and foragers;foragers respond to ecological variabilityprimarily by moving to alternative re-sources. Farmers, with their heavy invest-ment in land, cannot so easily move andso must use large social networks to bringthe resources to them.

As agriculture spread out from its cen-ters of development, it pushed waves oflinguistic and demographic homogeneitybefore it (Renfrew 1991). This process wasparticularly clear in Africa, with the Bantuexpansion, and Eurasia, with the Indo-Eu-ropean, Sino-Tibetan and other spreads(Diamond 1997a, 1997b). The languages ofthe spreading farmers eventually brokeup into daughters whose number was de-termined by the ecological regime of thearea, but it is likely that the number ofnew languages in many areas was lessthan the number of hunter-gatherer lan-guages which had been subsumed or dis-placed. The number of languages in theworld may thus have been lowered at theonset of the Neolithic, when the globalpattern we have detected in this paperbecame established.

The global pattern is currently under-going another transformation. This iscaused by the economic, technological,and demographic take-off of Eurasianpopulations and the consequent spread ofEurasian peoples, crops, diseases, andlanguages to the other continents. Thecauses and dynamics of this expansion arewell beyond the scope of this paper, but itis clear that their net effect will be togreatly reduce the world’s language diver-sity; one recent projection suggested that90% of living languages are threatenedwith extinction in the next century (Pinker1994:259).

In some continents, such as Australiaand the Americas, the Eurasian expansion

369LANGUAGE DIVERSITY

has involved a demographic and linguisticreplacement of indigenous peoples, whomainly succumbed to infectious disease,with Eurasian daughter populations(Crosby 1986). In South America, the in-digenous diversity has already been lost,which is why South America fails to pat-tern with the other continents in Fig. 1. InNorth America and Australia, most of thelanguages had, as of 1993, a few elderlyspeakers, so they still appear in the dataset of this paper, but they are not beingtransmitted to new generations and sowill disappear rapidly over the next 20years (Krauss 1992; Dixon 1997).

In other continents where Eurasiansnever settled in numbers, such as tropicalAfrica, language diversity is still beinglost, though in this case it is due to lin-guistic replacement without correspond-ing demographic replacement. Englishand French, along with the larger and

more dominant indigenous languagessuch as Swahili, are spreading at the ex-pense of local languages as more peopleare sucked into the larger social networksassociated with the modern world econ-omy. Similar processes can be observed inthe Pacific, with the spread of Tok Pisin,Tagalog, and Bahasa alongside English.What the ultimate results of these changesin human ecology on the world’s languagediversity will be it is as yet impossibleto say.

ACKNOWLEDGMENTS

This research was carried out in the department ofAnthropology at University College London, andgenerously funded by the Medical Research Council.I am grateful to Leslie Aiello, Robin Dunbar, RuthMace, Daisy Williamson and many others for adviceand assistance. All remaining errors are of coursemy own.

APPENDIX

Country Languages Areaa Populationb Stationsc MGSd SD(GS)e

Algeria* 18 2381741 25660 102 6.60 2.29Angola 42 1246700 10303 50 6.22 1.87Australia* 234 7713364 17336 134 6.00 4.17Bangladesh 37 143998 118745 20 7.40 0.73Benin† 52 112622 4889 7 7.14 0.99Bolivia* 38 1098581 7612 48 6.92 2.50Botswana 27 581730 1348 10 4.60 1.69Brazil* 209 8511965 153322 245 9.71 5.87Burkina Faso† 75 274000 9242 6 5.17 1.07C.A.R. 94 622984 3127 13 8.08 1.21Cambodia 18 181035 8442 9 8.44 0.50Cameroon† 275 475422 12239 35 9.17 1.75Chad 126 1284000 5819 11 4.00 1.81Colombia 79 1138914 33613 35 11.37 1.37Congo 60 342000 2346 10 9.60 1.69Costa Rica 10 51100 3064 38 8.92 1.78Cote d’Ivoire† 75 322463 12464 9 8.67 1.25Cuba 1 110861 10736 13 7.46 1.55Ecuador* 22 283561 10851 44 8.14 3.47Egypt 11 1001449 54688 50 0.89 0.89Ethiopia* 112 1221900 53383 36 7.28 3.10French

Guiana 11 90000 102 5 10.40 0.80Gabon 40 267667 1212 14 8.79 0.77Ghana† 73 238553 15509 28 8.79 1.68

370 DANIEL NETTLE

APPENDIX–—Continued

Country Languages Areaa Populationb Stationsc MGSd SD(GS)e

Guatemala* 52 108889 9467 59 9.31 2.23Guinea† 29 245857 5931 8 7.38 1.22Guyana 14 214969 800 5 12.00 0.00Honduras* 9 112088 5265 13 8.54 2.53India 405 3287590 849638 218 5.32 1.92Indonesia 701 1904569 187765 58 10.67 1.82Kenya* 58 580367 25905 34 7.26 3.61Laos 93 236800 4262 7 7.14 0.35Liberia† 34 111369 2705 21 10.62 0.84Libya 13 1759540 4712 54 2.43 1.60Madagascar* 4 587041 11493 81 7.33 2.96Malawi 14 118484 8556 20 5.80 1.50Malaysia 140 329749 18333 63 11.92 0.37Mali† 31 1240192 9507 17 3.59 1.97Mauritania† 8 1025520 2036 8 0.75 0.83Mexico* 243 1958201 87836 272 5.84 2.69Mozambique 36 801590 16084 90 6.07 1.39Myanmar 105 676578 42561 30 6.93 0.81Namibia 21 824292 1837 6 2.50 1.89Nepal 102 140797 19605 16 6.39 1.98Nicaragua* 7 130000 3999 8 8.13 2.15Niger 21 1267000 7984 10 2.40 1.28Nigeria*† 427 923768 112163 24 7.00 2.16Oman 8 212457 1559 2 0.00 0.00Panama 13 75517 2466 5 9.20 0.75Papua N.G. 862 462840 3772 8 10.88 1.96Paraguay* 21 406752 4397 16 10.25 2.51Peru* 91 1285216 21998 40 2.65 4.22Phillipines 168 300000 62868 64 10.34 1.92Saudi Arabia 8 2149690 14691 10 0.40 0.92Senegal 42 196722 7533 12 3.58 1.11Sierra Leone† 23 71740 4260 23 8.22 0.59Solomon Is. 66 28896 3301 1 12.00 0.00Somalia 14 637657 7691 28 3.00 1.69South Africa* 32 1221037 36070 114 6.05 3.50Sri Lanka* 7 65610 17240 17 9.59 2.59Sudan* 134 2505813 25941 43 4.02 2.82Suriname 17 163265 429 2 12.00 0.00Tanzania 131 945087 28359 45 7.02 1.90Thailand 82 513115 56293 54 8.04 1.57Togo† 43 56785 3643 11 7.91 1.78UAE 9 83600 1629 6 0.83 0.69Uganda 43 235880 19517 21 10.14 1.17Vanuatu 111 12189 163 4 12.00 0.00Venezuela* 40 912050 20226 44 7.98 2.73Vietnam 88 331689 68183 40 8.80 1.59Yemen 6 527968 12302 2 0.00 0.00Zaire 219 2344858 36672 16 9.44 1.90Zambia 38 752618 8780 30 5.43 0.67Zimbabwe 18 390759 10019 52 5.29 1.43Nicaragua* 7 130000 3999 8 8.13 2.15Niger 21 1267000 7984 10 2.40 1.28

371LANGUAGE DIVERSITY

APPENDIX–—Continued

Country Languages Areaa Populationb Stationsc MGSd SD(GS)e

Nigeria*† 427 923768 112163 24 7.00 2.16Oman 8 212457 1559 2 0.00 0.00Panama 13 75517 2466 5 9.20 0.75Papua N.G. 862 462840 3772 8 10.88 1.96Paraguay* 21 406752 4397 16 10.25 2.51Peru* 91 1285216 21998 40 2.65 4.22Phillipines 168 300000 62868 64 10.34 1.92

* Denotes a country with variable Growing Season, excluded from most of the analyses.† Denotes a country used in Nettle (1996) and excluded from some of the analyses.a km2.b Thousands.c The number of weather stations used in calculating MGS.d The Mean Growing Season (months).e The standard deviation of the Growing Season values from the different weather stations in that country.

REFERENCES CITED

Abruzzi, W. S.1982 Ecological theory and ethnic differentiation

among human populations. Current Anthro-pology 23: 13–21.

Bailey, R. C., G. Head, M. Jenike, B. Owen, R. Recht-man, and E. Zechenter

1989 Hunting and gathering in a tropical rainforest:Is it possible? American Anthropologist 91: 59–82.

Barbujani, G.1991 What do languages tell us about human mi-

croevolution? Trends in Ecology and Evolution6: 151–155.

Barbujani, G., and R. R. Sokal1990 Zones of sharp genetic change in Europe are

also linguistic boundaries. Proceedings of theNational Academy of Sciences of the U.S.A. 87:1816–1819.

Berg, E.1976 The economic impact of drought and inflation in

the Sahel. Research for Economic Develop-ment, Ann Arbor.

Birdsell, J. B.1953 Some environmental and cultural factors in-

fluencing the structuring of Australian ab-original populations. American Naturalist 87:171–207.

Breton, R. J-L.1991 Geolinguistics: Language dynamics and ethno-

linguistic geography. University of OttawaPress, Ottawa.

Casad, E. H.1974 Dialect intelligibility testing. Summer Institute

of Linguistics, Norman, OK.

Cashdan, E.1985 Coping with risk: Reciprocity amongst the

Basarwa of Northern Botswana. Man (n.s.)20: 454–474.

1990 (Editor) Risk and uncertainty in tribal and peas-ant economies. Westview, Boulder.

Cavalli-Sforza, L. L., A. Piazza, L. Menozzi, and J.Mountain

1988 Reconstruction of human evolution: Bringingtogether genetic, archaeological, and linguisticdata. Proceedings of the National Academy of Sci-ences of the U.S.A. 85: 6002–6006.

Colson, E.1979 In good years and bad: Food strategies of

self-reliant societies. Journal of Anthropologi-cal Research 35: 18–29.

Coulmas, F.1992 Language and Economy. Blackwell, Oxford.

Crosby, A. W.1986 Ecological imperialism: The biological expan-

sion of Europe, 900–1900. Cambridge Univer-sity Press, Cambridge.

DeGarine, I., and G. A. Harrison (Editors)1988 Coping with uncertainty in the food supply. Ox-

ford University Press, Oxford.Diamond, J.

1997a Guns, germs and steel: The fates of human soci-eties. Jonathan Cape, London.

1997b The language steamrollers. Nature 389: 544–546.

Dixon, R. M. W.1997 The rise and fall of languages. Cambridge Uni-

versity Press, Cambridge.Greenberg, J. H.

1987 Language in the Americas. Stanford UniversityPress, Stanford.

372 DANIEL NETTLE

Grimes, B. F.1993 Ethnologue: The world’s languages. 12th edition.

Summer Institute of Linguistics, Dallas.

Halstead, P., and J. O’Shea1989 Introduction: cultural responses to risk and

uncertainty. In Bad year economics: Culturalresponses to risk and uncertainty, edited by P.Halstead and J. O’Shea, pp. 1–7. CambridgeUniversity Press, Cambridge.

Headland, T. N. and L. A. Reid1989 Hunter-gatherers and their neighbours

from prehistory to the present. Current An-thropology 30: 43–66.

Hudson, R. A.1996 Sociolinguistics. 2nd Edition. Cambridge Uni-

versity Press, Cambridge.

Huss-Ashmore, R., J. J. Curry, and R. K. Hitchcock(Editors)

1988 Coping with seasonal constraints. UniversityMuseum, University of Pennsylvania, Phil-adelphia.

Huston, M. A.1994 Biological diversity: The coexistence of species

on changing landscapes. Cambridge Univer-sity Press, Cambridge.

Krauss, M.1992 The world’s languages in crisis. Language 68:

4–10.

Legge, K.1989 Changing responses to drought amongst the

Wodaabe of Niger. In Bad year economics:Cultural responses to risk and uncertainty, ed-ited by P. Halstead and J. O’Shea, pp. 81–86.Cambridge University Press, Cambridge.

LeHouerou, H. N.1989 The grazing land ecosystems of the African Sa-

hel. Ecological Studies 75, Springer, Berlin.

LePage, R.1968 Problems of description in multilingual

communities. Transactions of the PhilologicalSociety 1968: 189–212.

Mace, R., and M. Pagel1995 A latitudinal gradient in the density of hu-

man languages in North America. Proceed-ings of the Royal Society of London, B 261:117–121.

Mayr, E.1963 Animal species and evolution. Belknap Press,

Cambridge.

Milroy, L.1980 Language and social networks. Blackwell, Ox-

ford.

Minnis, P. E.1985 Social adaptation to food stress: A prehistoric

south-western example. University of ChicagoPress, Chicago.

Nettle, D.1996 Language diversity in West Africa: An eco-

logical approach. Journal of AnthropologicalArchaeology 15: 403–438.

Nichols, J.1990 Linguistic diversity and the first settlement

of the New World. Language 66: 475–521.1992 Linguistic diversity in space and time. Chicago

University Press, Chicago.

Niddrie, D. L.1971 The Carribean. In Latin America: Geographi-

cal perspectives, edited by H. Blakemore andC. T. Smith, pp. 73–120. Methuen, London.

Partridge, W. L.1989 The human ecology of tropical land settle-

ment in Latin America: Overview. In TheHuman ecology of tropical land settlement inLatin America, edited by D. A. Schumannand W. L. Partridge, pp. 3–19. Westview,Boulder.

Pinker, S.1994 The language instinct. Penguin, Harmond-

sworth.

Renfrew, C.1991 Before Babel: Speculations on the origins of

linguistic diversity. Cambridge ArcheologicalJournal 1: 3–23.

Romaine, S.1994 Language in society: An introduction to Socio-

linguistics. Oxford University Press, Oxford.

Sokal, R. R., and T. J. Crovello1970 The biological species concept: A critical

evaluation. American Naturalist 104: 127–153.

Stevens, G. C.1989 The latitudinal gradient in species range:

how so many species coexist in the tropics.American Naturalist 133: 240–256.

Swift, J.1986 The economics of production and exchange

in West African pastoral societies. In Pasto-ralists of the West African Savanna, edited byM. J. Adamu and A. H. M. Kirk-Greene, pp..Manchester University Press, Manchester.

Watts, M.1983 Silent violence: Food, famine and the peasantry

in northern Nigeria. Berkeley: University of Cal-ifornia Press.

Wernstadt, F. L.1972 World climatic data. Climatic Data Press,

Lemont, PA.

373LANGUAGE DIVERSITY

Wolff, H.1959 Intelligibility and inter-ethnic attitudes. An-

thropological Linguistics 1: 34–41.

NOTES

1 Nichols (1992:232–237) calls this kind of diversitygenetic diversity. This designation is confusing giventhe recent interest in comparing the distribution oflanguages with that of human DNA (e.g., Cavalli-Sforza et al 1988; Barbujani and Sokal 1990; Barbujani1991), and so I avoid it here.

2 These percentages are calculated from the Worldbank’s World Development Report 1993.

3 The reason for taking logarithms of these vari-ables can be found in Section 3.

4 Ivory Coast, Ghana, Togo, Benin, Nigeria, Cam-eroon, Zaire, Tanzania, India, Vietnam, Laos, Phill-ipines, Malaysia, Indonesia, Papua New Guinea,Vanuatu, and the Solomon Islands. The numbergiven will be very slightly inaccurate, as languagesspoken in two neighboring countries will have beencounted twice. This probably involves no more than30 or 40 languages in a total of nearly 4000.

374 DANIEL NETTLE