Neurobiological studies have generated new ways of thinking about development of brain structure and function. Development involves more than just growth from simple to complex structures. The initial over-abundance of neurons and synaptic connections is subsequently pruned of those that are non-functional. In addition, as behavioural and cognitive functions emerge and become automatized, the underlying brain representations are reorganized. In this paper, I shall argue that these different modes of neurodevelopmental change provide a useful metaphor for examining language acquisition. It will be argued that language acquisition can involve learning to ignore and inhibit irrelevant information, as well as forming new ways of representing complex information economically. Modular organization is not present from the outset, but develops gradually. This analysis suggests a new way of assessing specific language impairment (SLI). There has been much debate as to whether children with SLI lack specific modular components of a language processing system. I propose instead that these children persist in using inefficient ways of representing language. Finally, I consider what we know about the neurobiological basis of such a deficit. There is mounting evidence that children with SLI have subtle structural anomalies affecting the language areas of the brain, which are largely genetically determined. We should not, however, conclude that the language difficulties are immutable.