To find out whichlinguistic features travel together, and might thus be parts of Chomskyan models,means drawing up a reliable linguistic family tree. That is tricky. Unlikebiologists, linguists do not have fossils to guide them through the past (apartfrom a few thousand years of records from the few tongues spoken by literatesocieties). Also, languages can crossbreed in a way that spcies do not.English, for example, is famously a muddle of German, Norse and medievalFrench. As a result, linguists often disagree about which tongues belong to aparticular family. To leap this hurdle, Dr Dunn began by collecting basicvocabulary terms - words for body parts, kinship, simple verbs and the like -for four large language families that all linguists agree are real. These areIndo-European, Bantu, Austronesian(from South-East Asia and the Pacific) andUto-Aztecan (the native vernaculars of the Americas). These four groups accountfor more than a third of the 7,000 or so tong
ues spoken around the world today.
For each family, Dr Dunn and his team identified sets of cognates. These areetymologically related words that pop up in different languages. One set, forexample, contains words like ゛night゛.゛Nacht゛and ゛nuit゛. Another includes ゛milk゛and ゛Milch゛, but not ゛lait゛. That result is a multidimensional Venn diagram thatrecords the overlaps between languages.
Which is fine for the present, but not much use for the past. To substitute forfossils, and thus reconstruct the ancient branches of the tree as well as themodern-day leaves, Dr Dunn used mathematically informed guesswork. The maths inquestion is called the Markov chain Monte Carlo (MCMC) method. As its namesuggests, this spins the software equivalent of a roulette wheel to generate arandom tree, then examines how snugly the branches of that tree fit the modernfoliage. It then spins the wheel again, to tweak the first tree ever soslightly, at random. If the new tree is a better fit for the leaves, it istaken as the starting point for the next spin. If not, the process takes a stepback to the previous best fit. The wheel whirrs millions of times until suchrandom tweaking has no discernible effect on the outcome.
When Dr Dunn fed the languages he had chosen into the MCMC casino, the resultwas several hundreds equally probable family trees. Next, he threw eightgrammatical features, all related to word order, into the mix, and ran the gameagain.
The results were unexpected. Not one correlation persisted across all languagefamilies, and only two were found in more than one family. It looks, then, asif the correlations between grammatical features noticed by previousresearchers are actually fossilised coincidences passed down the generations aspart of linguistic culture. Nurture, in other words, rather than nature. If DrDunn is correct, that leaves Dr Chomsky's ideas in tatters, and raisesquestions about the very existence of a language organ. You may be sure,though, that the Chomskyan heavy artillery will be making its first rangingshots in reply, even as you read this article. Watch this space for furtherdevelopments.
