GENERALIZATION AND DISCRIMINATION EMERGE FROM A SELF-ORGANIZING COMPONENTIAL NETWORK - A SPEECH EXAMPLE

It is demonstrated that a componential code emerges when a self-organi sing neural network is exposed to continuous speech. The code's compon ents correspond to substructures that occur relatively independently o f one another: words and phones. A capability for generalisation and d iscrimination develops without having been optimised explicitly. The c omponential structure is revealed by optimising a necessarily complica ted nonlinear moment of the data's distribution, equal to the mean-squ ared output response of a multi-layered network of simple threshold ne urons. Earlier analytical work had predicted that componential codes, generalisation and discrimination should emerge from the self-organisa tion of threshold neurons of this form, assuming certain properties of the pattern-space distribution of the data.