Language and the learning curve: A new theory of syntactic development (review)

This exercise in bridge-building brings together ideas from the theory of syntactic structure, experimental psychology, and complexity theory in order to offer a novel theory of language development supported by empirical evidence from the psycholinguistics of language development, the author's main area of expertise.¹ The result is interesting, provocative, and suggestive; even its weaknesses are instructive in revealing the profound difficulty of building bridges across disciplinary boundaries.

The first chapter focuses on the linguistic notion of 'valency', or subcategorization. Ninio argues that the minimalist operation Merge is equivalent to the recognition of a dependency relation in dependency grammar, so syntactic structure consists of nothing but the links between individual words, with larger phrases doing little or no work. This similarity has been noticed within linguistics not only by dependency grammarians, but also by some advocates of the minimalist program itself (Chametzky 2003), but it is probably much less familiar to psychologists. N draws the important conclusion that syntax is actually much easier to learn than it would be if the full apparatus of phrase structure had to be induced, because syntax is merely the satisfaction of whatever dependents each word needs. Furthermore, she supports the theoretical claim by arguing that children's two-word combinations almost always consist of a head-word combined with a dependent. Moreover, when children produce longer utterances, they nearly always keep heads and dependents next to each other. For instance, their noun phrases are more likely to contain a determiner when functioning as objects than as subjects, which is as expected if determiners are heads.

The discussion is convincing as far as it goes, but it ignores three important areas of syntax. The first two are adjuncts and what one might call 'complicated syntax'-all of the cases where the dependent is moved away from the head, such as extraction in relative clauses and WH-questions. These omissions are understandable in a book about the speech of very young children, but they seriously undermine the general claim that syntax is easy to learn because 'adult syntax is simple' (2). The third omission is more important: N's treatment of grammatical relations. In Ch. 1 these are simply taken for granted, which again undermines the claim that syntax is easy. Yes, it is easy to learn once you have learned to distinguish subjects, objects, and so on. But how easy is it to learn these distinctions, based as they are on a complex combination of more concrete properties? The problem is only increased if, as N argues in Ch. 4, the child gets no guidance at all from semantics.

Ch. 2 gives its name to the book: 'The learning curve'. It introduces nonpsychologist readers to the POWER LAW OF PRACTICE, which N describes as 'one of the great success stories of cognitive psychology' (39). This 'law', which rests on a great deal of empirical evidence, asserts that 'the speed of performance of a task increases as a power-law function of the number of times the task is performed' (37). The law goes beyond the common-sense idea that practice makes perfect by defining rather precisely the degree of perfection that a given amount of practice produces. When plotted on a graph as a 'learning curve', progress does not show as a straight line from zero to whatever is the maximum possible, but follows a 'power-law' curve in which performance is proportional to some power (e.g. to the square) of the amount of practice. The observed patterns are sufficiently precise to allow a debate within psychology as to whether the curve is best defined by a power law or by an exponential formula (in which the amount of practice is itself the power of some constant).

The idea that the power law of practice applies to the learning of language is an important one that has been developed, apparently independently, by others (e.g. Ellis 2002), since...