LANGUAGE ACQUISITION ACCORDING TO CONNECTIONISTS: MODELS OF LEARNING MORPHOLOGY
Jenny Hayes · 2013
The challenge for supervised neural net models of morpho-syntax has been to demonstrate that language learning that appears to entail a data base of rules and exceptions can be simulated without the need for these structures to be present. This article reviews connectionist models of morpho-syntax which have attempted to meet this challenge. The article begins with a background description of how connectionist models work and then proceeds to explain the way in which both static and sequential models of the acquisition of morpho-syntax have been developed. That static connectionist models have been able to simulate the development of both verbal and nominal morphology is discussed in the context of how these models question the dual mechanism model of morphological processing (Pinker, 1999). The role of sequential connectionist models in understanding of the treatment of plural nouns in compound words is considered in detail. Finally, how morpho-syntax might be learnt in considered in an overview of developmental models.