Using Neural Networks to Explicate Human Category Learning: A Simulation of Concept Learning and Lexicalisation

Abstract

Presents a ?hybrid? neural network architecture comprising two Kohonen maps interrelated by Hebbian connections to perform a neural network based simulation of the development of a ?concept memory?, ?word lexicon? and ?concept lexicalisation? in an unsupervised learning environment using realistic psycholinguistic data. The results of the simulation demonstrate how neural networks, incorporating unsupervised learning mechanisms, can indeed simulate the learning of categories amongst children. The work demonstrates the efficacy of neural networks towards providing some insights into the elusive mechanisms that lead to the emergence of human categories and an explication of inherent conceptual categories