DOI: 10.1126/science.1078208 , 597 (2003); 300Science

et al.Jared Diamond,ExpansionsFarmers and Their Languages: The First

www.sciencemag.org (this information is current as of November 12, 2009 ):The following resources related to this article are available online at

http://www.sciencemag.org/cgi/content/full/300/5619/597version of this article at:

including high-resolution figures, can be found in the onlineUpdated information and services,

found at: can berelated to this articleA list of selected additional articles on the Science Web sites

http://www.sciencemag.org/cgi/content/full/300/5619/597#related-content

http://www.sciencemag.org/cgi/content/full/300/5619/597#otherarticles, 2 of which can be accessed for free: cites 21 articlesThis article

87 article(s) on the ISI Web of Science. cited byThis article has been

http://www.sciencemag.org/cgi/content/full/300/5619/597#otherarticles 20 articles hosted by HighWire Press; see: cited byThis article has been

http://www.sciencemag.org/cgi/collection/anthroAnthropology

: subject collectionsThis article appears in the following

http://www.sciencemag.org/about/permissions.dtl in whole or in part can be found at: this article

permission to reproduce of this article or about obtaining reprintsInformation about obtaining

registered trademark of AAAS. is aScience2003 by the American Association for the Advancement of Science; all rights reserved. The title

CopyrightAmerican Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by theScience

on

Nov

embe

r 12

, 200

9 w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

Farmers and Their Languages:The First Expansions

Jared Diamond1 and Peter Bellwood2

The largest movements and replacements of human populations since the end of the IceAges resulted from the geographically uneven rise of food production around the world.The first farming societies thereby gained great advantages over hunter-gatherersocieties. But most of those resulting shifts of populations and languages are complex,controversial, or both. We discuss the main complications and specific examplesinvolving 15 language families. Further progress will depend on interdisciplinary researchthat combines archaeology, crop and livestock studies, physical anthropology, genetics,and linguistics.

Until the end of the Pleistocene, allpeople on all continents lived as hunt-er-gatherers. Then, at different subse-

quent times between about 8500 and 2500B.C., food production based on domestica-tion of relatively few wild plant and animalspecies arose independently in at most ninehomelands of agriculture and herding, scat-tered over all inhabited continents except Austra-lia (Fig. 1) (1–11). Be-cause food productionconferred enormousadvantages to farmerscompared with hunter-gatherers living out-side those homelands,it triggered outwarddispersals of farmingpopulations, bearingtheir languages andlifestyles (12–14).Those dispersals con-stitute collectively themost important pro-cess in Holocenehuman history.

The agriculturalexpansions ultimate-ly resulted from threeadvantages that farm-ers gained over hunt-er-gatherers. First,because of far higherfood yields per areaof productive land, food production cansupport far higher population densities thancan the hunter-gatherer lifestyle. Second,

whereas most hunter-gatherer societies aremobile, most food-producing societies aresedentary and can thus accumulate storedfood surpluses, which were a prerequisitefor the development of complex technolo-gy, social stratification, centralized states,and professional armies. Third, epidemicinfectious diseases of social domestic ani-mals evolved into epidemic infectious dis-

eases of crowded farming populations, suchas smallpox and measles— diseases towhich the farmers evolved or acquiredsome resistance, but to which unexposedhunter-gatherers had none. These advantag-es enabled early farmers to replace lan-guages and societies of hunter-gatherersliving in their main paths of expansion.

Whereas recently expanding Europeans

described their conquests in writing, mostof the major pre-Columbian expansions ofagricultural populations occurred in pre-literate times. Hence the evidence for themcomes from five other independent sources:archaeology, records of plant and animaldomestication, human skeletal remains,modern human genes (and sometimes an-cient DNA), and dispersal histories of ex-isting or extinct but attested languages.Thus, study of the agricultural expansionsis preeminently interdisciplinary. To syn-thesize evidence from disparate fields isexciting but also challenging: Few scien-tists possess technical competence in all ofthese fields, and the different types ofevidence may seem to yield conflictingconclusions.

This review begins by introducing the basichypothesis and by explaining six complications

sometimes raised as objections. We then dis-cuss 2 general issues and 11 specific examplesinvolving linked spreads of prehistoric farm-ers and language families outward from ag-ricultural homelands, proceeding from rela-tively unequivocal examples to uncertainones. Finally, we call attention to new typesof evidence required to settle the many con-troversies in this field.

1Department of Geography, University of California,Los Angeles, CA 90095-1524, USA. 2School of Archae-ology and Anthropology and Research School of Pa-cific Studies, Australian National University, CanberraACT, 0200, Australia. E-mail: jdiamond@geog.ucla.edu( J.D.); peter.bellwood@anu.edu.au (P.B.)

Fig. 1. Archaeological map of agricultural homelands and spreads of Neolithic/Formative cultures, with approximateradiocarbon dates.

REVIEW

www.sciencemag.org SCIENCE VOL 300 25 APRIL 2003 597

on

Nov

embe

r 12

, 200

9 w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

The Basic Hypothesis and SixComplicationsThe simplest form of the basic hypothesis—thatprehistoric agriculture dispersed hand-in-handwith human genes and languages—is that farm-ers and their culture replace neighboring hunt-er-gatherers and the latter’s culture. This hy-pothesis would be supported if all five indepen-dent types of evidence coincided in attesting thereplacement of local hunter-gatherers by ex-panding farmers bearing their own archaeologi-cally visible culture, domesticates, skeletaltypes, genes, and languages, and if all thoseindicators were traceable back to the farmers’homeland of origin. Our two clearest examplesof such concordance of evidence are the colo-nizations of previously uninhabited Polynesiaand Micronesia by Neolithic populations speak-ing Austronesian languages (Fig. 2, no. 8), andthe expansion of farmers speaking Bantu lan-guages out of their tropical West African agri-cultural homeland after 1000 B.C. over mostsummer-rainfall regions of sub-equatorial Afri-ca (Fig. 2, no. 1).

But the basic hypothesis is more oftencontroversial, because in most other cases thefive types of evidence are less concordant.Some critics believe that these discordancesrefute the hypothesis and that farming andlanguage families spread mainly by diffusionamongst existing populations of hunter-gath-erers (15). We conclude that reality is muchricher and more complex than the simpleversion of the hypothesis, for many obviousreasons. The main classes of discordance areas follows:

Clinal genetic admixture between hunter-gatherers and farmers. Usually, arrivingfarmers do not exterminate or drive out hunt-er-gatherers completely. Instead, there issome intermarriage, especially of hunter-gatherer women to farmer men, resulting indilution of farmer genes with hunter-gatherergenes. If the farmers’ expansion consisted ofhundreds of successive such steps of inter-marriage and gene dilution, the hybrid popu-lation at the most remote step would haveonly low frequencies of the original farmers’

genes, even though the hybrid population ateach step might have consisted of only 10%local hunter-gatherers at that step and 90%invading hybrid farmers from the previousstep. This is the wave-of-advance model bywhich Ammerman and Cavalli-Sforza [(16 ),see also (17 )] interpreted southeast-northwestgene gradients across Europe, which theyattributed to the northwestward expansion offarmers from Anatolia. As a result, genes ofthe modern population of western Ireland (thenorthwest terminus of the advance) are esti-mated to be derived 99% from Europe’s orig-inal hunter-gatherers and only 1% from Ana-tolian farmers, even though Ireland’s lan-guage, crops, livestock, religion, and writingsystem as of 1492 A.D. were derived almostentirely from eastern Mediterranean proto-types, with little or no contribution from theculture of Ireland’s original Mesolithic hunt-er-gatherers (18).

Adoption of farming by peripheral hunter-gatherers. Some hunter-gatherer populationsin the path of farming expansions succeeded

1

1

3a

3b

3c

9

8

3c3c

7

8

86

8

1

1

BB

A

B

10

10

11

11 11

11

11

11

??

To Hawaii

Australian

Austronesian

Nilo-Saharan

Afro-Asiatic

Dravidian

Sino-Tibetan

Austroasiatic

Tai

Papuan

Altaic

Chukchi-Kamchatkan

Khoisan

Niger-Congo

Caucasian

Indo-European

Uralic

Fig. 2. Language families of the Old World and their suggested expan-sions. Map based on information in (87) and other sources. Numberedexamples discussed in text are 1 (Bantu), 3a to 3c (Austro-Asiatic, Tai,

and Sino-Tibetan, respectively), 6 (Trans New Guinea), 7 (Japanese), 8(Austronesian), 9 (Dravidian), 10 (Afro-Asiatic), 11 (Indo-European). Oth-er possible examples mentioned only briefly: A (Turkic), B (Nilo-Saharan).

R E V I E W

25 APRIL 2003 VOL 300 SCIENCE www.sciencemag.org598

on

Nov

embe

r 12

, 200

9 w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

in acquiring livestock and/or crops and somematerial culture (such as pottery) from foodproducers, while retaining their own languag-es, genes, and skeletons. The clearest exam-ple is the development of herding amongsome of southern Africa’s indigenousKhoisan peoples (so-called Hottentots), whoacquired livestock and pottery from food pro-ducers expanding southward in Africa about2000 years ago (19, 20). Other cases involvethe Navajo in the southwestern United Statesadopting sheepherding between 1650 A.D.and 1700, and some Apache populationsadopting casual maize farming (21, 22).

Reversion of expanding farmers to thehunter-gatherer life-style. When expandingfarmers reach areas unsuitable for farmingwith the domesticates available to them, theymay survive by reverting to the hunter-gath-erer life-style. Undoubted examples are thederivation of Polynesian hunter-gatherers onthe Chatham Islands and New Zealand’sSouth Island from ancestral Polynesian farm-ers (23–25), and of Punan hunter-gatherers inBorneo rainforests from other Austronesianfarmers (26, 27 ).

Language shift by indigenous popula-tions. Discordance between languages andgenes may arise when an expanding languageis imposed on or adopted by a peripheralpopulation, with only a minor contribution ofexpanding genes. This situation differs fromthe situation of clinal gene dilution, in whichinvaders constitute a majority at every step. Aclear modern example is the increasing adop-tion of English as the language of govern-ment in Papua New Guinea, whose inhabit-ants nevertheless remain indigenous NewGuineans with negligible admixture of Euro-pean genes. Possible historical examples arethe imposition of the Magyar and Turkishlanguages on medieval Hungarians and Ana-tolians, whose genes today are estimated tobe derived only 10 and 30%, respectively,from their conquerors’ genes (28). Likelyprehistoric examples are the adoption of Aus-tronesian languages by some former speakersof Papuan languages in the western islands ofMelanesia and by Agta Negritos in the Phil-ippines (29, 30). However, though such re-placements can be attested, under pre-statepre-literate conditions they were over onlyshort distances compared with the spreads ofmany of the major agriculturalist languagefamilies over thousands of kilometers.

Replacement of the expanding farmers’language in the original homeland, after theexpansion began. If this happened, modernlanguage distributions might conflict with thecombined evidence from genes, archaeology,skeletons, and domesticates. The originalhomeland might now either lack the originalfarmers’ language family altogether, or elsemight support only one branch of the familycompared with many branches in the periph-

ery. Suggested examples of this tend to becontroversial because they involve eradica-tion of the original languages that can now, atbest, only be reconstructed. Nevertheless, thisseems to us the most plausible interpretationin some cases. For instance, one can suggestthat languages closely related to Austrone-sian, Indo-European, and Japanese are nolonger spoken in their putative ultimatehomelands in South China, Anatolia, and Ko-rea, respectively, because of the historicalexpansions of the Sinitic languages, Turkish,and Korean. The discovery of written docu-ments attesting to the former existence ofHittite and other now-extinct Indo-Europeanlanguages in Anatolia as well as the resultingbig changes in our understanding of that lan-guage family confirm the reality of languageloss in the potential homeland for that family(31, 32).

Hunter-gatherer expansions. Not onlyfarmers, but also sometimes hunter-gatherers,can expand at the expense of other hunter-gatherers, producing concordance of genes andlanguages without crops. Examples include theInuit expansion eastwards across the CanadianArctic and the Athabaskan expansion south-ward into the southwestern United States withinthe last millenium (33–35).

Language Family Origins and SpreadsAssociation of language family origins with ag-ricultural homelands. If our basic hypothesis iscorrect, then a single agricultural homelandmight have given rise to more than one languagefamily radiating from it (36). This suggests thathomelands will be areas where several majorlanguage families intersect geographically andwhere the methods of comparative linguisticssuggest that those families originated. In con-trast, most regions that lack independent agri-cultural origins should have a lesser diversity oflanguage families. But these are not intended asuniversal generalizations, and regions with highnumbers of language families need not alwaysbe regions of agricultural origin. For example,California in Native American times, certainlynot a homeland for agriculture, is neverthelessfamously diverse linguistically for other reasonsincluding a long-term existence of partly seden-tary hunter-gatherer populations. Each situationof linguistic diversity will need to be examinedon its own terms.

Thus, the Fertile Crescent, or else nearbyareas reached early by Fertile Crescent domes-ticates, is a zone of intersection for the Indo-European, Elamite (with Dravidian?), Afro-Asiatic, and Caucasian language families (13,37). In central and southern China and adjacentareas lies the intersection of the Sino-Tibetan,Tai-Kadai, Austro-Asiatic, and Hmong-Mienfamilies, with Austronesian as an offshoot fromTaiwan (Fig. 2) (38). Mesoamerica formed thehomeland for the Uto-Aztecan, Oto-Manguean,Mayan, Mixe-Zoquean, and several other minor

families. Highland New Guinea gave rise to theTrans-New-Guinea family, tropical West Africato the Bantu subfamily and indeed to the wholeNiger-Congo family of which Bantu languagesform a subgroup. Linguistic relationships in theeastern United States and in the Andes-Amazonregion are too uncertain to test this generaliza-tion, in part because of migrations and disease-caused language extinctions after European col-onization. However, agriculture may haveplayed a role in the spread of Iroquoian andSiouan languages in the eastern United Statesand of the Chibchan, Quechuan, and Aymaranlanguages in the Andes-Amazon (Fig. 3).

East-west versus north-south expansions.All other things being equal, crops and live-stock, and people and the technologies andlanguages associated with them have spreadmore rapidly along east-west axes than alongnorth-south axes (4 ). The reason is that day-length and seasonality (hence daylight-de-pendent plant germination schedules) dependonly on latitude, so that sites at the samelatitude but different longitudes are likely toshare, and sites at the same longitude butdifferent latitudes are likely to differ in, theirdomesticates, habitats, climates, diseases, andagricultural systems. As applied to linguis-tics, these differing average rates of spreadalong east-west and north-south axes mayprovide the underlying reason why there arethree well-established language families(Indo-European, Afro-Asiatic, and Austrone-sian, and possibly a fourth if one includes themore controversial Altaic) with geographicranges spanning 7000 to 14,000 km along theeast-west axis of the Old World, but no lan-guage families spanning more than a fewthousand kilometers along the north-southaxis of the New World. These generalizationsdo not deny that a few language familiesspread rapidly north-south over much moremodest distances (e.g., Bantu languages insubequatorial Africa, and Uto-Aztecan lan-guages from Mexico into the southwesternUnited States).

Examples of Specific LanguageFamiliesWe now examine how the basic hypothesisilluminates the dispersal histories of 15 lan-guage families, listed in an approximate se-quence from unequivocal to controversial.

1. Bantu (Niger-Congo family). Beginningaround 2000 B.C., farmers from the tropicalWest African agricultural homeland in east-ern Nigeria and western Cameroon speakingearly Bantu languages expanded east andthen south over most of subequatorial Africa,replacing or intermarrying with most of theoriginal inhabitants related to modern Pyg-mies and Khoisan people. The Bantu sub-group that covers most of this region is justone of the 177 subgroups of the whole Niger-Congo language family and comprises about

R E V I E W

www.sciencemag.org SCIENCE VOL 300 25 APRIL 2003 599

on

Nov

embe

r 12

, 200

9 w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

500 of the family’s recorded 1436 languages.Of particular interest, as a model for whatcould be achieved elsewhere, is the excep-tionally detailed integration of linguistic evi-dence with other types of evidence (fromgenetics, archaeology, and domesticatedplants and animals) in this case (20, 39–42).

2. Arawak (Taino). Around 400 BC, farm-ers from the Orinoco River of South Americacolonized the West Indies and eventually re-placed most of the islands’ earlier occupants,spreading up the Lesser Antillean chain to theGreater Antilles and Bahamas. They therebybecame ancestral to the modern Taino peoplespeaking Arawakan languages. The evidencefrom linguistics, pottery, and domesticates is

detailed, but genetic evidence is slight be-cause few Tainos survived European con-quest. Linguistic relationships suggest thatArawakan languages had previously originat-ed in and spread over much of the upperAmazon (43–45).

3. Austro-Asiatic, Tai (“Daic”), and Sino-Tibetan. Several independent sources of evi-dence suggest expansions of these three lan-guage families from agricultural homelandsin China, at different times and over differentgeographic ranges. Austro-Asiatic spreadwest and south from southern China into theIndian subcontinent and Malay Peninsula(46 ), Sino-Tibetan spread from the YellowRiver or Sichuan into Burma and the Hima-

layas (47, 48). Much of the southward expan-sion of the Tai languages, like that of theHmong-Mien (Miao-Yao) languages, hasbeen within historic times. Austro-Asiaticlanguages are especially diverse, suggestingthat they were the first of these three familiesto expand. They occur today almost entirelysouth of the political border of China, raisingthe possibility that they too originated insouthern China, but were then largely re-placed there by later expansions of Sino-Tibetan and Tai.

4. Uto-Aztecan. Maize, beans, and linguis-tic evidence suggest strongly that agriculturebased on Mexican domesticates reached thesouthwestern United States from northernMesoamerica with speakers of Uto-Aztecanlanguages (49–51). The principal discordanceis that the northernmost groups of Uto-Aztec-ans (Numic, Tubatulabal, and Takic) aredesert hunter-gatherers in the Great Basin andsouthern California, leading to the formerassumption that Uto-Aztecan languages orig-inated among northern desert hunter-gather-ers who spread south and became farmers.More likely, as Hill has recently argued (49),northward-expanding Uto-Aztecan farmersreverted to hunting and gathering when theyreached habitats either unsuitable for agricul-ture or rendered marginal by drought or ag-ricultural over-exploitation, just as Austrone-sian farmers did in southern New Zealandwhen they entered a climate zone unsuitablefor their main crop, the sweet potato.

5. Oto-Manguean, Mixe-Zoquean, Mayan.Oto-Manguean has the widest geographicrange of any language family within the Me-soamerican agricultural homeland, spanninga distance of 1300 km from Mesoamerica’snorthwest boundary to its southeast bound-ary, although that range is still small by OldWorld standards. The reconstructed Proto-Oto-Manguean language, as well as Proto-Mayan and Proto-Mixe-Zoquean, includesterms for the major Mesoamerican crops, es-pecially maize, supporting the hypothesis ofan agricultural expansion. So-called glot-tochronological calculations of languagefamily time depths (43, 52) suggest that allfour of these Mesoamerican families—Uto-Aztecan, Oto-Manguean, Mixe-Zoquean, andMayan—spread initially between roughly3000 and 1500 B.C. (53), a span that overlapswith the archaeological time span (�2500 to1500 B.C.) for the beginnings of agriculturalintensification and sedentary village life inMesoamerica (54 ).

6. New Guinea Highlands. By far thegreatest linguistic diversity in the modernworld occurs on the island of New Guinea,with 1000 of the modern world’s 6000 lan-guages, and with dozens of language isolatesor families that have no demonstrable rela-tionship to each other or to any languageoutside New Guinea. Recent linguistic stud-

Fig. 3. Language families of the New World and their suggested expansions. Maps based oninformation in (89) and other sources. Numbered examples discussed in text are 2 (Arawakan,Cariban, and Tupian), 4 (Uto-Aztecan), 5 (Oto-Manguean and Mayan). Other possible examplesmentioned only briefly: C (Iroquoian and Siouan, with maize after 500 A.D.), D (Chibchan), and E(Quechuan and Aymaran).

R E V I E W

25 APRIL 2003 VOL 300 SCIENCE www.sciencemag.org600

on

Nov

embe

r 12

, 200

9 w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

ies suggest that at least half of those 1000languages belong to a family (the Trans-New-Guinea family) whose spread may havebeen driven by agricultural origins in theNew Guinea Highlands (55). The principaluncertainties concern the age of agriculturalorigins in the Highlands (as early as 7000B.C. or as late as 4000 B.C., perhaps) (56–58), the identity of the first staple crops, andthe linguistic limits of the Trans-New-Guineafamily.

7. Japanese. Around 400 B.C., intensiverice agriculture, new pottery styles, and newtools, all based on Korean models, appearedon the southwestmost Japanese island of Ky-ushu near Korea and spread northeast up theJapanese archipelago. Genes and skeletons ofthe modern Japanese suggest that they aroseas a hybrid population between arriving Ko-rean rice farmers and a prior Japanese popu-lation similar to the modern Ainu and respon-sible for Japan’s earlier Jomon pottery. Mod-ern southwest-to-northeast gene clines in Ja-pan and DNA extracted from ancientskeletons support this interpretation (59, 60).Japanese origins would thus rival Bantu ori-gins as the most concordant and unequivocalexample of an agricultural expansion, were itnot for the flagrant discordance of the linguis-tic evidence. If Korean farmers really didbecome dominant in Japan as recently as 400B.C., one might have expected the modernJapanese and Korean languages to be asclosely similar as other languages that di-verged at such a recent date (e.g., Germanand Swedish), whereas their relationship is infact much more distant.

The likely explanation is language re-placement in the Korean homeland. EarlyKorea consisted of three kingdoms with dis-tinct languages. The modern Korean lan-guage is derived from that of the ancientKorean kingdom of Silla, the kingdom thatunified Korea. However, the now-extinct lan-guage of one of the two ancient Korean king-doms that Silla defeated, Koguryo, was muchmore similar to Old Japanese than is Sillan ormodern Korean (61). Thus, a Koguryo-likelanguage may have been spoken by the Ko-rean farmers arriving in Japan, may haveevolved into modern Japanese, and may havebeen replaced in Korea itself by Sillan thatevolved into modern Korean.

8. Austronesian. Detailed archaeologicalevidence demonstrates the colonization ofTaiwan by Neolithic pottery-making andrice-growing farmers from South China be-fore 3000 B.C., followed by the spread offarming, pottery, and Neolithic tools to thePhilippines (2000 to 1500 B.C.), then south-west to the Southeast Asian mainland and toMadagascar, and then east through Indonesiaout across the Pacific to the furthest islands ofPolynesia, eventually reaching New Zealandby about 1200 A.D. (23, 24, 27 ). In terms of

distance covered, this was the world’s largestpre-historic agricultural expansion, and it ri-vals the Bantu expansion in the degree ofdetail of its linguistic reconstruction (62, 63).Austronesian primary subgroups are concen-trated in the Taiwan homeland, just as Niger-Congo primary subgroups are concentrated inthe West African homeland. These facts sug-gest that colonists derived ultimately fromAustronesian-speaking farmers of coastalSouth China replaced or hybridized with theoriginal population of the Philippines andIndonesia (related to modern Philippine Neg-ritos, New Guineans, and Aboriginal Austra-lians), and that language shift and hybridiza-tion were especially complex when spreadingAustronesians encountered established densepopulations in the New Guinea area andNorthern Melanesia.

This interpretation faces three objections,at least with respect to Southeast Asia (theprogression of human settlement into RemoteOceania—Polynesia, Micronesia and easternMelanesia—is easiest to interpret becauseAustronesians were the first human popula-tion there). The most obvious objection is theabsence of languages related to Austronesianin coastal South China today; an equally ob-vious explanation may be replacement thereby expanding Sino-Tibetan speakers. A sec-ond issue is that rice of subtropical SouthChinese origin was abandoned and tropicalcrops were domesticated or adopted when theexpansion entered the equatorial tropics; thisshift strikes us as an entirely expected devel-opment. Not a problem. A third issue is ge-netic studies interpreted to suggest geneticdivergence between populations in what arenow southern China and Island Austronesialong before 4000 B.C. (64 ), but the molecularclock calculations underlying that interpreta-tion are controversial, and a Taiwan home-land for Austronesians is not ruled out byother genetic analyses (65–67 ).

9. Dravidian. Food production reachedSouth India at about 3000 BC, partly throughthe spread of Fertile Crescent and Sahel do-mesticates via the Indus Valley and the north-western Deccan, and partly through a simul-taneous spread of rice cultivation from South-east Asia with speakers of Austro-Asiatic(Mundaic) languages. In addition to theseundoubted spreads of crops into India fromelsewhere, Fuller (68) has recently argued forprimary (independent) origins of rice, millet,and gram domestication in the Ganges Valleyand South India.

The Dravidian language family is concen-trated in South India, with one distinctiveoutlier (Brahui) far to the northwest in Paki-stan, and perhaps an even more distinctiveextinct outlier (Elamite) much further to thenorthwest in southwest Iran (Elamite’s rela-tion to Dravidian languages is debated) (69).Either Dravidian languages are the original

languages of much of the Indian subcontinentor they arose to the west and spread at about3000 B.C. with Fertile Crescent domesticatesinto the Indian subcontinent, subsequentlybecoming extinct in their homeland. If thelatter interpretation were correct, then onewould have to assume either clinal gene di-lution or else language shift to explain whySouth Indians today are phenotypically andgenetically so unlike peoples of the FertileCrescent. Thus, for South Asian early agri-culture, both the archaeological and the lin-guistic records remain equivocal.

10. Afro-Asiatic. This language familyconsists of six branches, five (including An-cient Egyptian) confined to North Africa, one(Semitic) also extending in ancient times toSouthwest Asia. That distribution suggests anAfrican origin for the family, whose Semiticbranch might then have spread into South-west Asia. But the overwhelming archaeo-logically attested flow of domesticated cropsand animals from Neolithic times onward,into Egypt and through the Arabian Peninsulainto Ethiopia, is out of Southwest Asia ratherthan out of Africa. That would make it sur-prising for Semitic languages to have spreadagainst that stream.

There are two principal competing hy-potheses for the origin of Afro-Asiatic. One,based on reconstruction of early vocabularyfor cultural and environmental referents,places the homeland in the Levant during theearliest Neolithic (the late Natufian culture,9500 B.C.) (32, 70, 71), with a subsequenttwo-pronged spread by 5000 B.C. that is welldocumented archaeologically: mixed farmingacross the Nile into Egypt and North Africa,giving rise to the Egyptian and Berberbranches of Afro-Asiatic languages, andsheep- and goat-based pastoralism from west-ern Arabia across the Red Sea into Ethiopiaand Sudan, giving rise to the Cushitic,Omotic, and Chadic branches (Semitic spreadinto Ethiopia much later). That SouthwestAsian origin would now be masked by lan-guage replacement in the homeland, includ-ing the spread of the Semitic branch of Afro-Asiatic languages (including Akkadian orBabylonian, Aramaic, and Arabic) in historictimes. The other hypothesis, reflecting Afro-Asiatic language subgrouping but with noclear archaeological support, favors a home-land in northeastern Africa (72, 73). ThatAfrican origin would imply a preagriculturalspread for Afro-Asiatic, perhaps with popu-lation movement into a wetter early HoloceneSahara.

11. Indo-European. We have saved forlast the most intensively studied, yet still themost recalcitrant, problem of historical lin-guistics: the origin of the Indo-European lan-guage family, distributed before 1492 A.D.from Ireland east to the Indian subcontinentand (as represented by the extinct Tocharian

R E V I E W

www.sciencemag.org SCIENCE VOL 300 25 APRIL 2003 601

on

Nov

embe

r 12

, 200

9 w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

languages) western China. Unlike the Niger-Congo and Austronesian language families,each consisting of about a thousand languag-es that sometimes intergrade geographically,the Indo-European language family containsonly 144 languages divided among 11 mark-edly distinct branches. These and other factssuggest that the task of reconstructing Indo-European origins is complicated by massiveextinctions of Indo-European languages inthe past, resulting from the expansions of afew highly successful subgroups (Germanic,Romance, Slavic, and Indo-Iranian).

The two main competing hypotheses ofIndo-European origins both face severe diffi-culties. One hypothesis views Proto-Indo-Eu-ropean as having been spoken in the steppesnorth of the Black Sea by horse-riding no-madic pastoralists, whose supposed domesti-cation of the horse and invention of the wheelaround 4000 B.C. enabled them to expandmilitarily (74–76 ). But objections includethat horse domestication and riding may nothave begun until thousands of years later(77 ); that it is hard to understand (perhapseven inconceivable) how steppe pastoralistscould have imposed their language on somuch of Europe west of the steppes (78); andthat even linguists who reject glottochronol-ogy agree that Indo-European languages (in-cluding Anatolian) are so different from oneanother that their divergence probably beganbefore 4000 B.C. (31).

The other hypothesis, based on the recog-nition that the extinct Anatolian languages(including Hittite and Luvian, the probablelanguage of Troy) represent the most distinc-tive branch and hence the earliest document-ed branching in the family tree, views Proto-Indo-European (or, more strictly speaking,Proto-Indo-Hittite) as a language of NeolithicAnatolian farmers who carried Fertile Cres-cent domesticates west into Europe, east tothe Indus Valley, and north and then eastacross the Central Asian steppes beginningaround 7000 to 6000 B.C. (32, 78–80). Butobjections include that the reconstructedProto-Indo-European lexicon has a strongbias toward domesticated animals rather thancrops (81, 82) and that reconstructed Proto-Indo-European words relating to wheels andwheeled vehicles suggest (some would say“prove”) late Indo-European origins aroundthe time of the invention of the wheel (�4000B.C.) (76 ). Even we two authors of this paperhave differing views on this issue.

Future DirectionsWe mentioned at the outset that study of thefirst agricultural expansions requires interdisci-plinary research synthesizing five types of evi-dence. To resolve the many controversies con-cerning agricultural expansions, we need newevidence of all five types. Archaeological evi-dence of the first farmers is basic to many

questions, such as the distinction between pri-mary and secondary homelands of agriculture(83): Did agriculture really arise independentlyin Ethiopia, the Sahel, tropical West Africa,South India, and Amazonia? We need a morebalanced record of the earliest crops and live-stock, their wild relatives, and their dates andplaces of domestication, so that regions poorlyknown in this respect can take their place along-side the better understood Fertile Crescent (9).Detailed studies of gene gradients similar tothose available for Europe and the Pacific areneeded for other parts of the world; that infor-mation will be especially difficult to obtain inthe New World, because so many Native Amer-ican populations have disappeared or shifted inthe last 500 years. Studies of ancient skeletonsand of ancient DNA may give us snapshots ofhuman distributions at known times in the past.

Most importantly, given that genetic datacan, in theory, provide an ultimate test ofwhether farming systems and languagesspread with or without people, it is essentialto understand more about how genetic datashould be interpreted in terms of human his-tory. Problems with natural selection, differ-ing rates of mutation, the effects of popula-tion bottlenecks on stochastic loss of lineag-es, and the modeling of clinal variation acrossgeographical space are behind many of thecurrent debates between geneticists, particu-larly in Europe (17, 84, 85) and Austronesia(64–67 ). Conflict in these fields is the mainreason why this review is focused on lan-guages and archaeology, rather than genetics.Early farming systems and languages haveclearly spread, probably on many occasionsin historical association, and we believe thathuman populations have spread too, but de-tailed documentation of this inference stilllies in the future.

We also need more studies of languagesthemselves. Hundreds of historically impor-tant languages remain poorly described, ef-forts to trace so-called deep language rela-tionships (i.e., relations between languagesthat diverged long ago) remain highly contro-versial, and relationships of New World lan-guages are especially controversial (43, 86,87 ), due in part to the scale of loss andreplacement since 1500. Fortunately, lin-guists are today concerned with modeling theformation of linguistic diversity in time andthrough space (88), and this is a developmentto be applauded.

Environmental issues also arise, for not allearly farmers underwent territorial expan-sion. The Mixe-Zoqueans essentially stayedat home, as did speakers of languages of theCaucasus and also the Hmong-Mien, beforetheir recent movements into Southeast Asia.Evidently, the early spreads of farmers andtheir languages depended on being in theright place at the right time. Relevant factorswould have included availability of new

lands nearby into which to spread (pull fac-tors) and perhaps resource shortfalls at homedue to climate changes or prior resource mis-management (push factors).

To extract reliable conclusions from allthis evidence will require comparative re-search on a worldwide scale within multipledisciplines. It is quite a challenge, but auniquely fascinating one.

References and Notes1. B. Smith, The Emergence of Agriculture (Scientific

American, Washington D.C., 1995).2. D. Harris, Ed., The Origins and Spread of Agricultureand Pastoralism in Eurasia (Univ. College of LondonPress, London, 1996).

3. A. Damania et al., Eds, The Origins of Agriculture andCrop Domestication (International Center for Agricul-tural Research in the Dry Areas, Aleppo, Syria, 1998).

4. J. Diamond, Guns, Germs and Steel (Norton, NewYork, 1997).

5. The Transition to Agriculture in the Old World, O.Bar-Yosef, Ed., special issue of Rev. Archaeol., 19 (no.2), 1998).

6. D. Piperno, D. Pearsall, The Origins of Agriculture inthe Lowland Neotropics (Academic Press, San Diego,CA, 1998).

7. B. Smith, Proc. Natl. Acad. Sci. U.S.A. 98, 1324(1998).

8. M. van der Veen, Ed., The Exploitation of Plant Re-sources in Ancient Africa (Kluwer Academic, NewYork, 1999).

9. D. Zohary, M. Hopf, Domestication of Plants in theOld World (Clarendon, Oxford, ed. 3, 2000).

10. P. Bellwood, Ann. Rev. Anthropol. 30, 181 (2001).11. P. Bellwood, C. Renfrew, Eds., Examining the Lan-

guage/Farming Dispersal Hypothesis (McDonald Insti-tute for Archaeological Research, Cambridge, 2003).

12. P. Bellwood, in (2), pp. 465– 498.13. C. Renfrew, in (2), pp. 70 –92. Renfrew has been one

of the most active researchers in favor of stronglinkages between the spreading of early farmers andlanguage families, as discussed here, especially forIndo-European and other western Eurasian langagefamilies. See (11, 36, 78, 80) and further references toRenfrew’s work therein.

14. L. L. Cavalli-Sforza, F. Cavalli-Sforza, The Great Hu-man Diasporas (Addison-Wesley, Reading, PA, 1995).

15. A. Whittle, Europe in the Neolithic (Cambridge Univ.Press, Cambridge, 1996).

16. A. J. Ammerman, L. L. Cavalli-Sforza, The NeolithicTransition and the Genetics of Populations in Europe(Princeton Univ. Press, Princeton, NJ, 1984).

17. L. Chikhi et al., Proc. Natl. Acad. Sci. U.S.A. 99, 11008(2002).

18. E. W. Hill, M. A. Jobling, D. G. Bradley, Nature 404,351 (2000).

19. C. B. Bousman, African Archaeol. Rev. 15, 133 (1998).20. C. Ehret, An African Classical Age (Univ. Press of

Virginia, Charlottesville, VA, 1998).21. J. F. Downs, Animal Husbandry in Navajo Society and

Culture (Univ. of California Press, Berkeley, CA, 1964).22. D. H. Snow in Farmers, Hunters and Colonists, K. A.

Spielmann Ed. (Univ. of Arizona Press, Tucson, AZ,1991), pp. 71– 88.

23. P. V. Kirch, On the Road of the Winds (Univ. ofCalifornia Press, Berkeley, CA, 2000).

24. P. Bellwood The Polynesians (Thames and Hudson,London, 1987).

25. A. Anderson, Prodigious Birds (Cambridge Univ. Press,Cambridge, 1989).

26. V. T. King, The Peoples of Borneo (Blackwell, Oxford,1993).

27. P. Bellwood, Prehistory of the Indo-Malaysian Archi-pelago (Univ. of Hawaii Press, Honolulu, rev. ed.,1997).

28. L. L. Cavalli-Sforza, P. Menozzi, A. Piazza, The Historyand Geography of Human Genes, (Princeton Univ.Press, Princeton, NJ, 1994).

29. M. Ross, in Archaeology and Language I, R. Blench andM. Spriggs, Eds. (Routledge, London, 1997), pp. 209 –261.

R E V I E W

25 APRIL 2003 VOL 300 SCIENCE www.sciencemag.org602

on

Nov

embe

r 12

, 200

9 w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

30. L. Reid, in Language Contact and Change in theAustronesian World, T. Dutton, D. Tryon, Eds. (Mou-ton de Gruyter, Berlin, 1994), pp. 443–76.

31. R. Drews, Ed., J. Indo-European Studies Mono. 38,“Greater Anatolia and the Indo-Hittite LanguageFamily” (Institute for the Study of Man, Washington,DC, 2001).

32. A. Dolgopolsky, Mediterranean Lang. Rev. 6, 230(1993).

33. D. Dumond, The Eskimos and Aleuts (Thames andHudson, London, 1987).

34. M. Fortescue, Language Relations across Bering Strait(Cassell, London, 1998).

35. R. G. Matson, in (11), chap. 27.36. A. Sherratt, S. Sherratt, Antiquity 62, 584 (1988).37. C. Renfrew, Cambridge Archaeol. J. 10, 7 (2000).38. P. Bellwood, in Archaeology in Southeast Asia, Y.

Chun-tong, L. Wai-ling, Eds. (University Museum andArt Gallery, Hong Kong, 1995), pp. 11–22.

39. C. Ehret, M. Posnansky, Eds. The Archaeological andLinguistic Reconstruction of African History (Univ. ofCalifornia Press, Berkeley, CA, 1982).

40. D. Phillipson, in (11), chap. 15.41. K. Williamson, R. Blench in African Languages, B.

Heine, D. Nurse, Eds. (Cambridge Univ. Press, Cam-bridge, 2000), pp. 11– 42.

42. P. Underhill, Ann. Hum. Genet. 65, 43 (2001).43. L. Campbell, American Indian Languages (Oxford

Univ. Press, New York, 1997).44. I. Rouse, The Tainos (Yale Univ. Press, New Haven, CT,

1992).45. A. Aikhenvald, in The Amazonian Languages, R. Dixon,

A. Aikhenvald, Eds. (Cambridge Univ. Press, Cam-bridge, 1999), p. 75.

46. C. Higham, The Bronze Age of Southeast Asia (Cam-bridge Univ. Press, Cambridge, 1996).

47. G. van Driem, Bull. Indo-Pacific Prehistory Assn. 18,43 (1999).

48. J. Janhunen,Manchuria: An Ethnic History (Suomalais-Ugrilainen Seura, Helsinki, Finland, 1996).

49. J. Hill, in (11), chap. 26.50. P. Bellwood, in Areal Diffusion and Genetic Inheri-

tance: Problems in Comparative Linguistics, A. Aikh-envald, R. Dixon, Eds. (Oxford Univ. Press, Oxford,2001), pp. 27– 43.

51. J. Hill, Am. Anthropol. 103, 913 (2001).52. P. Bellwood, in Time Depth in Historical Linguistics, C.

Renfrew, A. McMahon, L. Trask, Eds. (McDonald Insti-tute for Archaeological Research, Cambridge, 2000),vol. 1, pp. 109 – 40.

53. T. Kaufman, in Amazonian Linguistics, D. L. Payne, Ed.(Univ. of Texas Press, Austin, TX, 1990), pp. 13–73.

54. R. Adams, M. J. MacLeod, Eds, The Cambridge Historyof the Native Peoples of the Americas (CambridgeUniv. Press, Cambridge, 2000), vol. II, part I.

55. A. K. Pawley, in Perspectives on the Bird’s Head ofIrian Jaya, Indonesia, J. Miedema, C. Ode, R. Dam, Eds.(Rodopi, Amsterdam, 1998), p. 684.

56. J. Golson, in Sunda and Sahul, J. Allen, J. Golson, R.Jones, Eds. (Academic Press, London, 1977), pp. 601–638.

57. M. Spriggs, in (2), pp. 524 –537.58. S. G. Haberle, A. Chepstow-Lusty, Environ. Hist. 6,

349 (2000).59. S. Horai et al., Am. J. Hum. Genet. 59, 579 (1996).60. K. Omoto, N. Saito, Am. J. Phys. Anthropol. 102, 437

(1997).61. M. Hudson, Ruins of Identity (Univ. of Hawaii Press,

Honolulu, 1999).62. R. Blust, J. World Prehistory 9, 453 (1995).63. A. K. Pawley, in (11), chap. 20.64. S. Oppenheimer, M. Richards, Nature 410, 166

(2001).65. T. Melton et al. Am. J. Hum. Genet. 63, 1807 (1998).66. E. Hagelberg, in The Archaeology of Lapita Dispersal in

Oceania, G. R. Clark et al., Eds. (Pandanus, Canberra,Australia, 2001), pp. 167.

67. M. Hurles et al., Genetics 160, 289 (2002).68. D. Fuller, in (11), chap. 16.

69. V. Blazek, in Archaeology and Language IV, R. Blench,M. Spriggs, Eds. (Routledge, London, 1998), pp. 48 –78.

70. I. Diakonoff, J. Semitic Studies 43, 219 –217, (1998).71. A. Militarev, in (11), chap. 12.72. C. Ehret, J. African Hist. 20, 161 (1979).73. C. Ehret, Reconstructing Proto-Afro-Asiatic (Univ. of

California Press, Berkeley, CA, 1995).74. M. Gimbutas, J. Indo-European Studies 13, 185

(1985).75. J. Mallory, In Search of the Indo-Europeans (Thames

and Hudson, London, 1989).76. D. Anthony, Antiquity 69, 554 (1995).77. M. Levine, in Late Prehistoric Exploitation of the Eur-

asian Steppe (McDonald Institute for ArchaeologicalResearch, Cambridge, 1999), pp. 5–58.

78. C. Renfrew, Archaeology and Language ( JonathanCape, London, 1987).

79. G. Krantz, Geographical Development of EuropeanLanguages (Lang, New York, 1988).

80. C. Renfrew J. Indo-European Studies 27, 257 (1999).81. C. Watkins, The American Heritage Dictionary of

Indo-European Roots (Houghton Mifflin, Boston,1985).

82. J. Mallory, in Archaeology and Language I, R. Blench,M. Spriggs, Eds. (Routledge, London, 1997), pp. 93–121.

83. J. Diamond, Nature 418, 34 (2002).84. M. Richards et al., Am J. Hum. Genet. 67, 1251

(2000).85. L. Simoni et al., Am. J. Hum. Genet. 66, 262 (2000).86. J. Greenberg, Language in the Americas (Stanford

Univ. Press, Stanford, CA, 1987).87. M. Ruhlen, A Guide to the World’s Languages (Stan-

ford Univ. Press, Stanford, CA, 1987), vol. 1.88. D. Nettle, Linguistic Diversity (Oxford Univ. Press,

Oxford, 1999).89. Encyclopaedia Britannica (Encyclopedia Britannica,

Chicago, IL, ed. 15, 1987), vol. 2290. We thank C. Hilliker for drafting the illustrations.

R E V I E W

www.sciencemag.org SCIENCE VOL 300 25 APRIL 2003 603

on

Nov

embe

r 12

, 200

9 w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from