Towards a Unified Model of Language Acquisition

Abstract

In this theoretical paper, we first review and rebut standard criticisms against distributional approaches to language acquisition. We then present two closely-related models that use distributional analysis. The first deals with the acquisition of vocabulary, the second with grammatical development. We show how these two models can be combined with a semantic network grown using Hebbian learning, and briefly illustrate the advantages of this combination. An important feature of this hybrid system is that it combines two different types of distributional learning, the first based on order, and the second based on co-occurrences within a context.