The evolution of language

Languages all have their roots in the same part of the world. But they are not as similar to each other as was once thought

Apr 14th 2011 | from the print edition

WHERE do languages come from? That is a question as old as humanbeings’ ability to pose it. But it has two sorts of answer. The first isevolutionary: when and where human banter was first heard. The secondis ontological: how an individual human acquires the power of speech andunderstanding. This week, by a neat coincidence, has seen thepublication of papers addressing both of these conundrums.

Quentin Atkinson, of the University of Auckland, in New Zealand, hasbeen looking at the evolutionary issue, trying to locate the birthplaceof the first language. Michael Dunn, of the Max Planck Institute forPsycholinguistics in the Netherlands, has been examining ontology.Fittingly, they have published their results in the two greatest rivalsof scientific journalism. Dr Atkinson’s paper appears in Science, Dr Dunn’s in Nature.

The obvious place to look for the evolutionary origin of languageis the cradle of humanity, Africa. And, to cut a long story short, it isto Africa that Dr Atkinson does trace things. In doing so, he knocks onthe head any lingering suggestion that language originated more thanonce.

One of the lines of evidence which show humanity’s African origins isthat the farther you get from that continent, the less diverse,genetically speaking, people are. Being descended from small groups ofrelatively recent migrants, they are more inbred than their Africanforebears.

Dr Atkinson wondered whether the same might be true of languages. To find out, he looked not at genes but at phonemes.These are the smallest sounds which differentiate meaning (like the“th” in thin; replace it with “f” or “s” and the result is a differentword). It has been known for a while that the less widely spoken alanguage is, the fewer the phonemes it has. So, as groups of peopleventured ever farther from their African homeland, their phonemicrepertoires should have dwindled, just as their genetic ones did.

To check whether this is the case, Dr Atkinson took 504 languages andplotted the number of phonemes in each (corrected for recent populationgrowth, when significant) against the distance between the place wherethe language is spoken and 2,500 putative points of origin, scatteredacross the world. The relationship that emerges suggests the actualpoint of origin is in central or southern Africa (see chart), and thatall modern languages do, indeed, have a common root.

That fits nicely with the idea that being able to speak andbe spoken to is a specific adaptation—a virtual organ, if you like—thatis humanity’s killer app in the struggle for biological dominance. Onceit arose, Homo sapiens really could go forth and multiply and fill the Earth.

The details of this virtual organ are the subject of Dr Dunn’s paper.Confusingly, though, for this neat story of human imperialism, hisresult challenges the leading hypothesis about the nature of thelanguage organ itself.

The originator of that hypothesis is Noam Chomsky, a linguist at theMassachusetts Institute of Technology. Dr Chomsky argues that the humanbrain comes equipped with a hard-wired universal grammar—a languageinstinct, in the elegant phrase of his one-time colleague Steven Pinker.This would explain why children learn to speak almost effortlessly.

The problem with the idea of a language instinct is that languagesdiffer not just in their vocabularies, which are learned, but in theirgrammatical rules, which are the sort of thing that might be expected tobe instinctive. Dr Chomsky’s response is that this diversity, like thediversity of vocabulary, is superficial. In his opinion grammar is acollection of modules, each containing assorted features. Switching on amodule activates all these features at a stroke. You cannot pick andchoose within a module.

For instance, languages in which verbs precede objects will alwayshave relative clauses after nouns; a language cannot have one but notthe other. A lot of similar examples were collected by Joseph Greenberg,a linguist based at Stanford, who died in 2001. And, though Greenberghimself attributed his findings to general constraints on human thoughtrather than to language-specific switches in the brain, his findingsalso agree with the Chomskyan view of the world. Truly testing thatview, though, is hard. The human brain cannot easily handle theconnections that need to be made to do so. Dr Dunn therefore offered thetask to a computer. And what he found surprised him.

Place your bets

To find out which linguistic features travel together, and might thusbe parts of Chomskyan modules, means drawing up a reliable linguisticfamily tree. That is tricky. Unlike biologists, linguists do not havefossils to guide them through the past (apart from a few thousand yearsof records from the few tongues spoken by literate societies). Also,languages can crossbreed in a way that species do not. English, forexample, is famously a muddle of German, Norse and medieval French. As aresult, linguists often disagree about which tongues belong to aparticular family.

To leap this hurdle, Dr Dunn began by collecting basic vocabularyterms—words for body parts, kinship, simple verbs and the like—for fourlarge language families that all linguists agree are real. These areIndo-European, Bantu, Austronesian (from South-East Asia and thePacific) and Uto-Aztecan (the native vernaculars of the Americas). Thesefour groups account for more than a third of the 7,000 or so tonguesspoken around the world today.

For each family, Dr Dunn and his team identified sets of cognates.These are etymologically related words that pop up in differentlanguages. One set, for example, contains words like “night”, “Nacht”and “nuit”. Another includes “milk” and “Milch”, but not “lait”. Theresult is a multidimensional Venn diagram that records the overlapsbetween languages.

Which is fine for the present, but not much use for the past. Tosubstitute for fossils, and thus reconstruct the ancient branches of thetree as well as the modern-day leaves, Dr Dunn used mathematicallyinformed guesswork. The maths in question is called the Markov chainMonte Carlo (MCMC) method. As its name suggests, this spins the softwareequivalent of a roulette wheel to generate a random tree, then examineshow snugly the branches of that tree fit the modern foliage. It thenspins the wheel again, to tweak the first tree ever so slightly, atrandom. If the new tree is a better fit for the leaves, it is taken asthe starting point for the next spin. If not, the process takes a stepback to the previous best fit. The wheel whirrs millions of times untilsuch random tweaking has no discernible effect on the outcome.

When Dr Dunn fed the languages he had chosen into the MCMC casino,the result was several hundred equally probable family trees. Next, hethrew eight grammatical features, all related to word order, into themix, and ran the game again.

The results were unexpected. Not one correlation persisted acrossall language families, and only two were found in more than one family.It looks, then, as if the correlations between grammatical featuresnoticed by previous researchers are actually fossilised coincidencespassed down the generations as part of linguistic culture. Nurture, inother words, rather than nature. If Dr Dunn is correct, that leaves DrChomsky’s ideas in tatters, and raises questions about the veryexistence of a language organ. You may be sure, though, that theChomskyan heavy artillery will be making its first ranging shots inreply, even as you read this article. Watch this space for furtherdevelopments.