8
LEARNING AND
PERCEIVING MUSICAL
STRUCTURES: FURTHER
INSIGHTS FROM ARTIFICIAL
NEURAL NETWORKS
, . ,
Abstract
Artificial neural networks expand our understanding of the acquisition and representation of
knowledge of our environment and the influence of this knowledge on perception. In the present
chapter, we review applications of artificial neural networks to music cognition, notably to the learn-
ing and perceiving of musical structures. We present a hierarchical self-organizing model that learns
basic regularities of the Western tonal system by mere exposure and simulates tonal acculturation.
After learning, the model simulates a variety of experiments dealing with the processing of tone,
chord, and key relationships. It provides a parsimonious account of these data sets by postulating
activation as the unifying mechanism underlying various cognitive tasks. The modelling of music
cognition presented in this chapter is restricted to behavioural data. Nevertheless, the computational
processes are based on fundamental neural constraints. Future developments of artificial networks
simulating neuropsychological cases and establishing direct links to neural correlates will contribute
to enhance our understanding of mechanisms of music perception.
Keywords:Music cognition; Connectionist modeling; Self-organization; Tonal acculturation
Introduction
Artificial neural networks are rooted in the biological structure of the neural system: they
consist of a set of artificial neurons (units) that are linked via synaptic connections of
different strengths. A unit does not represent one neuron, but simulates a population of
neurons. The goal of artificial networks is not to describe neural anatomy and physiology,
but to be founded on neural principles in order to simulate different levels of perceptual
and cognitive processing. Recent developments in artificial networks establish correspond-
ences to neural mechanisms by relating components of a model to brain structures,