Most four-year-olds know thousands of words and have a sense of grammar and syntax that would put even the most powerful AI programs to shame. How can so much linguistic sophistication come from listening to parental baby talk?
Just over three decades ago, the American linguistic Noam Chomsky came up with an answer: the world’s languages are all governed by the same universal grammar and each baby is born with a knowledge of that grammar. This idea rocked the then-dominant creed of behaviourism which said, in a nutshell, that animals must learn everything from scratch. But it raised as many questions as it answered. How did the universal grammar evolve? How is it encoded in genes and the neural structures of the brain? If language is rooted in biology, what is its relationship to mind and thought?
Now, linguists are joining forces with neuropsychologists and molecular geneticists to find some answers. And Steven Pinker, a cognitive psychologist at the Massachusetts Institute of Technology, who studies how infants learn language and was only an infant himself when Chomsky first outlined his theory, has attempted to explain this ambitious quest in a provocative book, The Language Instinct (Review, 26 February).
Advertisement
Why has it taken so long for Chomsky’s ideas to become so fashionable? Isn’t the current interest in language genes simply part of biology’s preoccupation with finding genetic explanations for behaviour and disease?
Certainly, people are receptive to the idea of there being genes for language. But just finding a gene tells you nothing about the logic of language and how it develops. Gene mania is only part of the story. Let me take a step back. The first enthusiasm for language being an innate capacity came in the 1960s when Chomsky set the scene and Eric Lenneberg wrote a book called The Biological Foundations of Language. Chomsky looks at sentences, he’s a paper and pencil theoretician. Lenneberg did the supportive work, he’s a neurologist. Lenneberg talked about the first recorded case of an inherited language impairment, about there being a critical period for language development in childhood, about links between language and retardation, about language skills being encoded biologically in the brain. Then in the 1970s the pendulum swung violently back in the other direction, towards a notion of language as something purely cultural and learnt.
Why?
In the 1960s there was a rush to embrace Chomsky’s theory as an instrument for disposing of behaviourism. Once everyone agreed that behaviourism was dead, ideas about language swung back to reflect the emphasis psychologists had always put on slow learning mechanisms being the force behind human intelligence. But in the past five to ten years, several developments have strengthened the claim that language is innate.
One is that studies of language development in children have produced important empirical findings supporting Chomsky. We now know in quite some detail how children mentally process the sequences of words in their parents’ speech to figure out the grammar of their parents’ language. And we know that they can’t do this without applying innate linguistic rules. This research means you can ask how you might build a computer program that could learn language the way children do.
Also from the biological end, Lenneberg has been largely corroborated by some interesting new case studies. There’s Chelsea, the deaf woman who, having been misdiagnosed as being retarded, never learnt language as a child. When she was finally fitted with hearing aids at the age of 30, she was a total loss at learning grammar and English – a very interesting demonstration of the critical period for language development.
Then there are people who suffer from Williams, or ‘chatterbox’, syndrome, a rare form of mental retardation in which young adults talk fluently, knowingly and grammatically but have an IQ of about 50 and are unable to tie their shoelaces or take something out of a cupboard. The condition seems to be associated with a defective gene on chromosome 11, and it shows that language can’t just be a manifestation of general intelligence.
In your book, you take these ideas a stage further. You argue that language ability evolved in our ancestors like a true biological instinct, a beneficial adaptation favoured by natural selection. This is controversial. Can you explain why?
Chomsky, in harmony with biologists such as Stephen J. Gould, says that not every trait is an adaptive product of natural selection and that this is true of language. It could have emerged as the by-product of evolving a large brain or analytical processing, or it could be the consequence of a universal physical law being applied to the brain. As an example of a by-product, you don’t need anything to posit a selective advantage for bones being white as opposed to green. Bones were selected for their rigidity. One way of making bones stronger is to make them out of calcium and calcium happens to be white. The whiteness is just a by-product. But I don’t think this argument works for a complex machine. It’s like saying a hurricane can blow through a junkyard and assemble a Boeing 707. Language is complex and useful and so it is surely adaptive.
Some behaviourists spend vast amounts of time trying to show that chimpanzees and other animals understand human language – that they, too, evolved a language instinct. Why are you so critical of this work in your book?
As a theatrical demonstration that language is not unique to humans, it’s misguided. It’s clear that you can get chimps to do something that has some similarities to human language, but how you interpret this depends on what we mean by language. The meaningful scientific question is whether a chimp’s language is homologous to ours – descended from a shared ancestor – or whether it simply has something in common with our language. This hasn’t been tested, and even if it were I don’t think we would find any evidence for homology. It’s ironic that people try to ‘elevate’ apes by forcing our communication system on them as if that is the measure of biological worth.
If language really is an instinct, why does it take most infants as long as three years to learn to talk?
Another way of putting the question is: why isn’t the baby born talking? There are probably two answers. One is simply that the structures of the brain are not completely assembled and developed at birth. The baby has to be squeezed out into the world before its head gets too big to pass through the pelvis. Another answer is that learning is an essential part of language because by its very nature language has to be a shared code. If you spoke a language of one you might as well not speak at all. The learning period synchronises the language ability of each child to that of everyone else around them.
In some wild animals, it’s true, the communication system is completely hard-wired. Some birds, for instance, are born with a song that is genetically determined and impervious to external influence. But our language is infinitely more complex. It’s hard to imagine how a single, hard-wired version of it could stay in register across all members of the community from one generation to the next. The genes encoding the language would be subject to all the usual pressures to change and diversify. Besides, there’s no way that you could encode 60 000 words – the vocabulary of an average high-school graduate – in a genome consisting of 50 000 to 100 000 genes. Vocabulary has to be learnt.
The other side of the learning coin is that different communities can evolve different languages over time provided they are geographically separated and language innovations don’t spread from one to the other. If the learning of language was perfect then language wouldn’t change. I think a lot of innovation comes when a parent uses a language signal that is ambiguous and a child analyses it in a different way. The word ‘orange’, for example, was originally ‘norange’, from the Spanish ‘naranja’. But the phrase ‘a norange’ can also be cut perceptually as ‘an orange’.
Visitors in foreign countries sometimes end up arguing over whether people think in any particular language. Do the English think in English, the Japanese in Japanese, and so on? But you maintain that thinking is largely free of linguistic constraints. Can you explain why?
There are several reasons. One of them is that when computer scientists first tried to model common-sense knowledge, they came unstuck because they tried to do it in English. English is way too ambiguous, words and phrases have multiple meanings, and it is also very inexplicit. When I’m talking to you I know that you can fill in for a lot of what I don’t say. If I have a sentence with three meanings, I can count on you to pick the right one. If I leave 37 steps between A and B I can probably count on you to fill them in, to do all the necessary deduction and elimination. That’s why I think the language of thought is likely to be much richer and more explicit than any particular language. Any sentence is much more a suggestion than a bit of knowledge.
Yet many people might still find it hard to accept that they don’t think in any particular language. What does thought consist of if not words and sentences?
We think in several different media, using different mental devices for representing the world – visual images, more likely graphics, big maps, auditory images, and so on. We think in abstract logical propositions, and language is the medium for externalising these for the benefit of a listener. There is a lot of common sense in this. How would you ever coin a new word if thoughts crucially depended on words? How would you borrow a word from another language unless you already had a mental representation of the concept behind the word and felt the need to express what you’d discovered?
But where does this leave education? If language doesn’t influence thought, how can any university tutor claim to be on a mission to ‘teach students to think’?
What you learn at school and university are historical conventions accumulated over thousands of years. Mental skills and customs for which we haven’t evolved an innate capacity have to be recreated generation after generation. In evolutionary terms, for example, reading is a recent cultural invention and so it has to be actively learnt rather than passively acquired. But what we don’t learn to do at school is how to have thoughts or ideas.
Turning to an analogy with computers, many people might say: ‘Sure, I agree that my computer microprocessor has a built-in operating system and what I see on my screen is a language roughly connected to that. But in between is software, and this has to be designed and debugged. What’s the interface between what I’m saying to you and my mental microprocessor? Who or what designs and debugs the software?’
Education and life experience. But education isn’t simply a question of filling the RAM of the computer with data, just dumping it in. It’s about exploiting in new domains our innate abilities to process information. Education is the process of taking a mind adapted to throwing rocks around, cracking open bones and shaking coconuts off trees, and plugging in abstract variables. Suddenly you’re thinking about the motions of the planets instead of the motions of rocks, you’re thinking about spaces of possible states of gas as opposed to physical spaces.
Even in ordinary language, it’s amazing how much we use metaphor to talk about abstract entities. We verbalise time in terms of space, and human states in terms of space. Saying ‘John went from being sick to being well’ – that’s a state expressed in terms of space. Even children show that at some level they are conceptualising these abstract things. And it’s a very concrete machinery. Children make distinctive errors like, ‘Can we have any meeting in front of dinner, because behind dinner there won’t be any space left for a story’. Or ‘the fairy took him into a little person’, meaning changed him into a little person, as if size is a parameter along a mental X-axis and shrinking consists of displacement along it.
Proving that language and thought really are two separate things would presumably require the discovery of genes that influence grammar – but not intelligence. Four years ago, a storm of controversy greeted what seemed to be a discovery of exactly this kind. Myrna Gopnik and her colleagues at McGill University in Montreal published an analysis of a language impairment running in a family in London. Most afflicted members of the family have average nonverbal IQs, but speak slowly, frequently misusing pronouns and suffixes involved in forming plurals and the past tense. The pattern of inheritance suggests a trait controlled by a single dominant gene. Can it really be that simple?
First of all, the impairment turns out to be harder to pin down than the initial reports suggested. The afflicted members of this family certainly have problems with the ‘wug test’. A wug is a nonsense creature. Yet prompt any four or five year old to give you the plural and they blurt out ‘wugs’. In this family, however, otherwise intelligent adults treat it as a difficult puzzle. They also have problems in pronunciation which they seem to outgrow. In fact, they probably have graded problems in all aspects of grammar and vocabulary. So the inherited impairment isn’t crisp.
There are probably better cases of purely grammatical impairments. For example, Heather Vanderlay at Birkbeck College in London has carefully selected a group of kids with what seem to be specific grammar disorders, and is now comparing them with normal kids matched for age and intelligence. There’s one boy of 14 or 15 who’s highly intelligent in the way he uses language with computers and so on, but he, too, is almost reduced to tears by the wug test. So I think the notion of innate language disorder is real. It may or may not be inherited in the crisp Mendelian pedigree, and it may not always be limited to grammar as opposed to pronunciation or conceptual distinctions linked to grammar.
The implication of this research is that a defect in just one of our tens of thousands of genes can profoundly affect our linguistic skills. How likely is that?
It’s pretty plausible because anything that’s a complex machine – like language – can be brought to a halt by a defect in one of its many parts. How many ways are there to prevent a car from starting? As many as there are parts of an engine. But you can’t say that language or grammar is based on a single gene just because it can be disrupted by a single gene.
In your book, you lament the fact that as many as 90 per cent of the world’s 5000 or so languages are on the verge of extinction. But language barriers also help to keep people divided and at war with each other. Should we really be concerned about the extinction of languages? Isn’t it inevitable that the telecommunications revolution will sweep dominant languages like English across the globe?
Language extinction does matter. It’s a complex ethical and practical issue. It would be easy for me just to have a crusade based on aesthetics – save Plinget, save Apache, and so on, because they’re beautiful. But there are other reasons for saving languages, just as there are for saving species. They’re scientifically highly informative, in terms of both human prehistory and psychology. By looking at similarities and differences among neighbouring languages, you can trace back the history of migrations of populations to find out, for example, how the Japanese got into Japan and how the Asians first spread out over the Americas. In language diversity, you can also see how the mental machinery of language, the universal grammar, can be stretched and what its limits are.
What practical steps could be taken to save languages?
Children can easily be encouraged to learn an indigenous language and a national language simultaneously. Having schools, electronic media and publishing industries in that language would help. One of my colleagues at MIT, Alan Prince, has suggested, semi-seriously, that if we could work out what universal grammar is, then it would suffice to have tapes of parents talking to their children. The universal grammar would provide the algorithm needed to take that input and generate the whole language. But that’s still science fiction – we won’t be at that stage for a long time yet.
Further reading The Language Instinct, Steven Pinker, William Morrow in the US, Allen Lane in the UK, 493 pages $23/ £20
