Learning via prediction : mapping continuous stimuli to discrete symbols