A discriminative account of the learning, representation and processing of inflection systems

Abstract

ABSTRACT What kind of knowledge accounts for linguistic productivity? How is it acquired? For years, debate on these questions has focused on a seemingly obscure domain: inflectional morphology. On one side, theorists inspired by Rumelhart & McClelland’s classic error-driven learning model have sought to show how all morphological forms are the products of a single memory-based process, whereas the opposing theories have claimed that irregular forms are processed by qualitatively different mechanisms to rule-governed regulars. This review argues that while the main ideas put forward by Rumelhart & McClelland – that inflectional patterns are learned, and rule-like behaviour emerges from the distribution of forms – appear to be correct, the theory embodied in their model (and those following it) is incompatible with the discriminative nature of learning itself. An examination of the constraints error-driven learning mechanisms impose on theories of morphological processing – along with language learning and human communication itself – is presented.