Deep Relationships Between Languages

NEWS & VIEWS RESEARCH

The authors’ preferred model, in which the Colm Ó Cofaigh is in the Department of icebergs remained intact and upright after Geography, Durham University, Durham initial breakout, also contrasts with the DH1 3LE, UK. modern Larsen B ice shelf, where icebergs e-mail: [email protected] capsized during collapse8. 1. Vaughan, D. G. & Doake, C. S. M. Nature 379, It is particularly noteworthy that, as 328–331 (1996). Jakobsson et al. emphasize, the corrugated 2. Domack, E. et al. Nature 436, 681–685 (2005). ridges and furrows on the floor of Pine Island 3. De Angelis, H. & Skvarca, P. Science 299, Bay occur widely on the Antarctic continen- 1560–1562 (2003). 50 Years Ago 4. Jakobsson, M. et al. Geology 39, 691–694 (2011). tal shelf, including in the Ross and Weddell 5. Wingham, D. J., Wallis, D. W. & Shepherd, A. seas. This suggests that tides may be a more Geophys. Res. Lett. 36, L17501; http://dx.doi.org/ On the whole the average alcohol pervasive influence on the record of Antarctic 10.1029/2009GL039126 (2009). consumption within a species ice-sheet deglaciation than has hitherto been 6. Lowe, A. L. & Anderson, J. B. Quat. Sci. Rev. 21, of animal is constant. It can be 1879–1897 (2002). recognized, and it raises the interesting ques- 7. Graham, A. G. C. et al. J. Geophys. Res. 115, F03025; said that each species has its own tion of whether spatial and temporal variability http://dx.doi.org/10.1029/2009JF001482 (2010). specific liking for alcohol … In in tidal range could act as a control on ice-sheet 8. MacAyeal, D. R., Scambos, T. A., Hulbe, C. L. & this work the consumption of retreat9. Clearly, much remains to be discov- Fahnestock, M. A. J. Glaciol. 49, 22–36 (2003). 10 per cent alcohol and water in 9. Uehara, K., Scourse, J. D., Horsburgh, K. J., ered about the complexity of the geological Lambeck, K. & Purcell, A. P. J. Geophys. free-choice experiments in six record of West Antarctic Ice Sheet deglaciation Res. 111, CO9025; http://dx.doi. animal species has been studied and the associated controls. ■ org/10.1029/2006JC003531 (2006). … The hedgehogs preferred water to alcohol solution. Sometimes, however, the animals drank LINGUISTICS considerable amounts of alcohol, and it could at times be judged from their movements that they were slightly drunk. The hamsters Deep relationships preferred alcohol. They seldom tasted the water and, when they did so, only until they found out between languages that they were drinking from the wrong bottle. Tracing a common ancestry between languages becomes harder as the connection From Nature 19 August 1961 goes further back in time. A new test has revealed a surprisingly ancient relationship between a central Siberian and a North American language family. 100 Years Ago JARED DIAMOND for two unrelated languages. For instance, the word for ‘eel’ is almost the same in Japanese On the publication of the final umans crossed from Siberia into and in New Guinea’s Haruai language, despite report of the Royal Commission Alaska some 14,000 years ago, long a lack of other similarities between the two on Tuberculosis the view was before the 5,000–10,000-year barrier languages and the impossibility that one could frequently expressed that those Hbeyond which most linguists think that lan- have borrowed the word from the other1. in authority ought, long ago, to guage evolution erases traces of shared ances- To address this problem, linguists scrutinize have taken precautions against the try. Efforts to break this barrier by establishing word lists for putative cognates — phonetically dangers arising out of the use of a relationship between Old World and New similar words with the same meaning in two milk containing tubercle bacilli World languages have been unconvincing until related languages, and differing by regular … Now that the commissioners now. The change is due mainly to the efforts of sound correspondences — such as English have reported, and in no uncertain Edward Vajda, who offers evidence in Anthro- and Spanish cognates beginning with ‘f’ and voice, that tuberculosis, especially pological Papers of the University of Alaska1 for ‘p’, respectively (father, fish and foot versus in the child, may be the result of the first convincing link, over an astonishingly padre, pez and pie). The late Joseph Green- infection with tubercle bacilli large distance, between Old World and New berg emphasized the value of multilateral conveyed in the cow’s milk, it World languages. comparisons, in which many related languages is essential that the question Most of the world’s 7,000 present-day lan- are simultaneously compared to identify of regulations relating to milk guages fall into language families — such ancestral roots preserved in only some of the and meat supply should be as the Indo-European family, to which this daughter languages. He concluded that most carefully reconsidered, and journal’s English, along with 140 other lan- Native American languages belong to one large that, as the commissioners put guages, belongs. Languages of the same family ancient family, termed Amerind2, and that it, “Government should cause resemble each other because they arose from a Indo-European and other established families to be enforced throughout the common ancestor and then diverged. form a superfamily, termed Nostratic. Twenty- kingdom food regulations, The difficulty in detecting ancient linguis- seven shared worldwide etymologies suggested planned to afford better security tic relationships — those beyond the 5,000– even more ancient shared ancestry among all against the infection of human 10,000-year barrier — is that all languages the world’s languages3. However, most linguists beings through the medium of share the same pool of consonants, vowels reject these conclusions, in part because of a diet derived from tuberculous and grammatical structures. Therefore, some lack of statistical tests to prove that the claimed animals.” resemblances are expected to occur by chance similarities are not due to chance. From Nature 17 August 1911 if one compares lists of thousands of words There are two methods for testing the role

he identifies include a dozen grammatical prefixes and about 100 cognate words with sound correspondences. Because the prefixes appear in the same sequence between Yeniseian and Yeniseian Na-Dene verbs, and because Vajda analysed all prefixes and didn’t just fish for one matching pair, both his prefix comparisons and his word lists comfortably pass Nichols’ significance Arctic tests. Vajda’s evidence is judged to be plausi- Ocean SIBERIA ble by linguists previously sceptical of claimed relationships between language families1. The reported Yeniseian–Na-Dene link raises CANADA many questions. Where was the common

ancestral language spoken (Siberia or North

t Na-Dene America), and did Yeniseians move west or

i a

r Na-Denes move east across the Bering Strait?

S ALASKA Are there archaeological correlates of such a g

n i movement? (The only plausible candidates are r e B archaeological horizons at around 12,000 and 5,000 years ago.) Do Yeniseian and Na-Dene form a superfamily by themselves, or do they P a cific belong to a larger superfamily that includes Ocean Sino-Tibetan and Basque? Are Yeniseian and Na-Dene people more closely related geneti- cally to each other than to other Siberians and Native Americans? (Apparently not, possibly Figure 1 | Circumpolar map of the Arctic showing the locations of Yeniseian and Na-Dene languages. because of millennia of intermarriage with There are two small groups of outlying Na-Dene populations in North America, far south of those 1 surrounding peoples.) Why did Yeniseian depicted, in the United States southwest and on the California and Oregon coasts. Vajda has found speakers have such a tiny geographical range evidence of a link between the Yeniseian and Na-Dene languages. (Figure based on ref. 1.) in modern times? But there are two questions that most trou- of chance in assessing deep relationships1,4–7. for assessing the role of chance depends on ble linguists. Why do Yeniseian and Na-Dene One is also used by biologists to test evolution- ‘individual-identifying thresholds’1,5,6. With languages still show such a strong relationship ary trees: the Monte Carlo simulation. This about 7,000 languages in the world, one could if they diverged 12,000 years ago, when other involves taking word lists for two or more estimate P = 1/7,000 = 0.000143 as a level of languages diverge beyond recognition after languages, repeatedly and randomly scram- rarity to suggest a unique individual language, 5,000–10,000 years ago? Either Yeniseian and bling the meanings within each list, recounting or (1/7,000) (0.05) = 7 × 10−6 as a level of Na-Dene languages really diverged only 5,000 the number of cognates in each randomized rarity significant at P < 0.05. Johanna Nichols1 years ago, or they are unusually conservative pair of lists, and then calculating statistically uses this test to support Vajda’s remark- and evolve especially slowly. And how did whether the observed number of cognates able new evidence for a relationship between Yeniseians and Na-Denes get separated by exceeds that expected by chance, according to central Siberian Yeniseian languages and North 5,200 km in only 5,000 or 12,000 years, mak- the randomized lists. American Na-Dene languages. ing their overland migration the longest and Such tests have been extended by Kessler Yeniseian is a tiny family of less than a dozen fastest recorded by hunter-gatherers? In com- and Lehtonen4 to multilateral comparisons, languages that were formerly spoken along ing years, we can watch as these questions are after first confirming statistical significance Siberia’s Yenisei River, and now survive only debated — and perhaps as other links between among 11 languages already known to belong as the Ket language, which has fewer than Old and New World language families are to the Indo-European family. This was true 200 speakers. Na-Dene languages (that is, the revealed. ■ even for Albanian, a language whose Indo– Tlingit, Eyak and Athabaskan languages) are European affinities had proved difficult to spoken across much of Alaska and northwest Jared Diamond is in the Geography establish by bilateral comparisons. This suc- Canada, with outliers in the southwestern Department, University of California, cess vindicates Greenberg’s view that mul- United States, California and Oregon. Four Los Angeles, California 90095-1524, USA. tilateral comparisons can uncover evidence intervening language families and 5,200 kilo- e-mail: [email protected] of a relationship that is obscure in bilateral metres separate the nearest Ket and Na-Dene 1. Kari, J. & Potter, B. (eds) Anthropological Papers of comparisons (because any single Indo-Euro- speakers today (Fig. 1). the University of Alaska Vol. 5 (issues 1, 2) (Univ. pean language alone happens to lack certain In 1923, one possible cognate was found Alaska Fairbanks/Alaska Native Language Center, 2010). Indo-European roots retained in Albanian). between Ket, Athabaskan and Tlingit, and 2. Greenberg, J. H. Language in the Americas (Stanford Unhappily for those multilateralists who other similar words were later identified (see Univ. Press, 1987). consider Indo-European to have an ancient ref. 3 for examples). Vajda has now placed this 3. Ruhlen, M. On the Origin of Languages (Stanford Univ. Press, 1994). relationship with Uralic languages (which postulated relationship on a rigorous footing. 4. Kessler, B. & Lehtonen, A. in Phylogenetic Methods include Finnish and Hungarian), the two fami- He was struck by how Ket’s system of tones and and the Prehistory of Languages (eds Forster, P. & lies turned out to be related at a significance bewilderingly complex strings of eight verb Renfrew, C.) 33–42 (McDonald Institute, 2006). 5. Nichols, J. in The Handbook of Language Contact level of only P = 0.45, far from the P < 0.05 prefixes were utterly out of place in Siberia — (ed. Hickey, R.) 361–379 (Wiley-Blackwell, 2010). 4 required for significance . It seems likely that otherwise occupied by toneless suffixing lan- 6. Nichols, J. & Warnow, T. Lang. Ling. Compass 2, the few sharings between the two families are guages related to Turkish and Finnish — and 760–820 (2008). 7. Wichmann, S. & Holman, E. W. Assessing Temporal due to chance or to borrowing. by how those tones and prefixes corresponded Stability for Linguistic Typological Features (Lincom The other, more approximate, set of methods in detail to Na-Dene languages. The parallels Europa, 2009).