From ga-ga to articulate

Children's acquisition of language is surely one of the natural wonders of the world. Within an astonishingly short space of time, children acquire a working vocabulary close to that of adults (typically some 60,000 words), at the peak literally hoovering up new words at the astonishing rate of one new word every two hours. Linguists have, however, always been less impressed by such facts than by children's acquisition of grammar. It is, after all, grammar that allows us to arrange words in sentences of near infinite complexity of meaning.

Linguists are divided on the question of how children acquire grammar. The "outside-in" view supposes that children are predisposed to attend to the general features of speech but learn their grammatical rules by generalising what they hear in their speech community. This at least has the merit of explaining in one step how children from different linguistic cultures learn radically different kinds of grammar. The alternative view of "inside-out" theories supposes that children have an innate grammar capacity that allows them to identify the relevant grammatical components of the speech they hear. They then extract local grammatical forms by paying attention to the more specific features of sentence structure. It is this latter view that is most closely associated with Noam Chomsky.

Perhaps inevitably, linguists have focused on language production when trying to unravel the processes involved. On the grounds that comprehension must precede production, Kathy Hirsh-Pasek and Roberta Golinkoff argue that we may be able to learn much more about the mechanisms of grammar acquisition by studying children's understanding of speech. To do this, they developed a neat experimental paradigm that involved measuring the direction of children's gaze in response to verbal questions or prompts.

Children as young as 13 months (the one-word stage) faced two television screens that offered alternative action film of people or familiar TV characters. They would then be asked to "Find X doing Y". Significant preferences in gaze direction for the correct TV screen was taken to imply a correct understanding of the grammatical structure of the sentence.

This rather ingenious procedure allowed Hirsh-Pasek and Golinkoff to show that children as young as 13-15 months (with a maximum vocabulary of 25 words) are already aware that the words in sentences form functional units months before they can use them. By 17 months (at the 70-100 word stage) they appear to be aware that word order is an important guide to meaning (in this case, that recipients of actions follow the verb in English).

By 24 months (by which time they can produce two-four word sentences using a 200-300 word vocabulary), they are beginning to differentiate between transitive and intransitive verbs (specifically that the former imply causal events while the latter do not). As has previously been noted in other contexts, girls are significantly ahead of boys in this respect.

The learning processes involved remain complex. Their data suggest that, as early as 24 months, children store verbs with their syntax and that, once they have enough in the lexicon, they use this information to generalise to unfamiliar words in the same frame. In effect, they use phrase structural information in combination with the real world events they observe to predict verb meaning long before they can use the words in speech.

This leads Hirsh-Pasek and Golinkoff to favour a developmental model of language acquisition that combines elements of both "inside-out" and "outside-in" theories. It views infants as rising progressively through a series of stages by dint of bootstrapping mechanisms. The first stage (roughly the first nine months of life) involves an innate responsiveness to prosodic information in the speech they hear (much of which is specially structured by the speaker to reduce information content and maximise redundancy - the now-familiar "motherese"). Experience at this level allows infants to identify the structural units in the speech stream (ie words).

This then allows children in the second (semantic) phase (roughly 9-24 months) to map the words they have learned to recognise onto the real world events they observe: the regularities they note enable them to learn the meanings of words. Because motherese forces the prosodic, semantic, social and syntactic systems to concur, the infant is guided into comprehension of word order. It is still shaky on the finer details of syntax, but it can cope with basic grammatical form provided there is plenty of redundancy. Finally, building on this platform, the third phase allows full syntactic speech to develop with surprisingly little effort. Local syntactic forms then can be integrated into the processing mechanism relatively easily.

This is an excellent little book, though, like motherese, it contains a great deal of redundancy. (Its contents would probably have made a useful pair of more focused papers.) Still, for those whose eyes are inclined to glaze over at the mere mention of transformational grammars, it provides a fine introduction to some of the key controversies of the moment as well as a valuable explanation of a very nice experimental paradigm. More importantly, the findings it reports will go some way towards forcing us to rethink our approaches to language learning.

Robin Dunbar is professor of psychology, University of Liverpool.