Learning the Dutch plural in Optimality Theory

Abstract

This thesis deals with language acquisition within the framework of stochastic Optimality Theory. Stochastic Optimality Theory was specifically designed to model variation and optionality, which traditionally are critical areas for standard Optimality Theory. Using a learning algorithm that comes with stochastic OT – the Gradual Learning Algorithm proposed by Boersma & Hayes (2001) – I will try to learn the variation in Dutch noun pluralisation. The data set is taken from Van Wijk (2007) and contains the distribution of the two default plural affixes /-en/ and /-s/ over various phonological contexts. In some phonological contexts, Dutch noun pluralisation displays variation. This acquisition experiment aims to test the Gradual Learning Algorithm as a plausible device for language acquisition. If the algorithm succeeds in learning the correct grammar – or constraint hierarchy in Optimality Theoretic terms – it will prove to be yet a more interesting approach to consider in language acquisition research. Results show however that the Gradual Learning Algorithm has severe difficulties in constructing a grammar that correctly models the variation in Dutch noun pluralisation.