The origin of language
The past 30 years has witnessed a debate between the holders of two very different views about how humans are able to talk. The behaviourists, following B. F. Skinner, argue that we learn to talk in the same way that we learn any other skill. Children are rewarded when they speak correctly, and reproved when they make mistakes. We can talk, whereas chimpanzees cannot, because we are better at learning: there is nothing special about language. In contrast, Noam Chomsky and his followers have argued that humans have a peculiar competence for language, which is not merely an aspect of their general intelligence. We learn to utter, and to understand, an indefinitely large number of grammatical sentences, and to avoid an even larger number of ungrammatical ones, so we cannot possibly learn which sentences are grammatical by trial and error. Instead, we must learn the rules that generate grammatical sentences. These rules are of great subtlety, so that, although we acquire and apply them, we cannot formulate them explicitly. For example, consider the two following sentences: How do you know who he saw? (1) Who do you know how he saw? (2) How do you know who he saw? Who do you know how he saw? Every speaker of English knows at once that is grammatical, and is not. But what rule tells us this? No-one but a trained linguist would have any idea, any more than a non-biologist would know how the rate of beating of the heart is adjusted to meet changing demands. In section 17.3, we describe a hypothesis about the rule that tells us that is ungrammatical: it is a subtle rule, but as yet no-one has thought up a simpler one. It is hard to believe that we could so painlessly master such rules unless we were genetically predisposed to do so. More generally, it is still beyond the wit of linguists and computer scientists to write a language-translating programme, yet many 5-year-olds know two languages, do not mix them up, and can translate from one to the other. A second reason for thinking that we cannot learn to talk by trial and error lies in the poverty of the input on which a child must rely. After hearing a finite set of utterances, a child learns to generate an indefinitely large number of grammatical sentences. This implies that the child learns rules, and not merely a set of sentences.