Abstract
We discuss a class of computational models that provide promising explanations of the processes underlying music cognition. These models, called neural net, connectionist, or parallel distributed models, are suited to music cognition because they can learn from passive exposure to the structural regularities of a musical culture. They have the potential to account for (1) the development, in the mind of the average listener, of cognitive schemas for music and (2) the subsequent generation of musical expectations based on these schemas. Using Western harmony and Indian rāgs as examples, we illustrate how one can simulate the expectancies of a native of one culture listening to the music of another. We show how a network can be constructed according to known music-theoretic constraints in order to study how some properties emerge from others. Finally, we review the results of experiments that test predictions about expectancies generated by these models.