Unsupervised Multimodal Neural Networks

The research described in this monograph is a significant milestone in computational modelling of cognitive processing using neural networks. The research integrates four key strands in neural network research, namely unsupervised learning, multimodality, temporal processing and neural multinets to come up with a computational framework that can be used effectively as a tool for investigating cognitive processes such as child language acquisition and second language acquisition. In-situ Hebbian-linked self-organising maps and counterpropagation networks are used to investigate static multimodal processing,whilst Kohonen''s Hypermap is extended to investigating temporal multimodal processing. These architectures are then integrated using multinet techniques to model child language acquisition from the one-word utterance stage to the two-word utterance stage. This monograph is very relevant to researchers working in cognitive processing,especially child language acquisition, as well as to researchers in neural networks.