A Pitch in Time: an Artificial Neural Network of Melodic Expectancy

The development of expectancies during the unfolding of auditory patterns in time is a recognized but poorly understood aspect of human cognition. This study investigates development of pitch-based expectancies in melody prediction. Two artificial neural networks tested the hypothesis that expectancies, such as pitch proximity and pitch reversal as proposed by Narmour [12], can be learned from exposure to a musical environment. A multi-layered back-propagation network and a perceptron both performed at better than chance level. The way in pitch relations are acquired and represented in networks is discussed together with implications for future experiments and network models of musical pitch development.