Authors:

GARAGNANI, M., Wennekers, T. & PULVERMULLER, F.

Reference:

Year of publication:

2006

CBU number:

6288

Abstract:

We simulated learning of neural assemblies in a model of the left perisylvian language cortex. Six cortical areas were simulated: primary auditory cortex, auditory belt and parabelt areas (Wernicke areas), inferior prefrontal cortex and premotor cortex (Broca areas), and primary motor cortex. Each area (layer) consisted of a 25x25 grid of graded-response neurons, with between-neuron connection probabilities that decreased with increasing cortical distance, following a Gaussian distribution. Between-area connections were topographic and followed the same probability distribution. Local and global inhibition mechanisms controlled activity within each layer. The model was confronted with simultaneous patterns of activations in both its motor and auditory layers, as one would expect it in early speech production. We observed formation of cell assemblies which responded specifically to one of the input patterns (words). During the testing of the network, the following functional characteristics emerged: (1) When only the auditory layer was stimulated with one of the patterns, oscillatory activity of the relevant assembly developed in that layer and spread synchronously through all other six layers; (2) During continuous input with one pattern, initial strong activation of the assembly was followed by a steady-state of oscillatory activity (pulse train); (3) Auditory presentation of unknown pseudo-patterns (corresponding to meaningless pseudo-words) produced less activity in the network, causing inconsistent or only partial activation of cell-assemblies and, on average, significantly reduced oscillations in them. These results help explain neurophysiological observations about cortical processing of words and pseudo-words, and may assist in the further development of neurobiological models of language.