Imitating and Evaluating Real and Transformed Musical Performances

Expression in musical performance is usually accounted for in terms of a generative model that takes a structural description as input and produces expressive modifications of rhythmic structure as output. This idea is examined in two experiments. In the first, piano players try to imitate heard performances that are either real performances by another pianist, or one of three kinds of transformations of the original performance that disrupt the relationship between structure and expression in different ways. The results show that the more the relationship is disrupted, the more inaccurate and unstable is the imitation attempt, thus providing support for the generative model. The players have some ability to imitate the disrupted versions, however, demonstrating that the generative model is not a complete account, and some explanations are offered for this. The second experiment assesses listeners' responses to the materials of the first experiment and shows that their preference ratings for the various versions of the melodies follow essentially the same pattern as do the accuracy/stability measures for the imitation attempts. This is taken as further evidence for the generative model in both performing and listening and for the central role of a structural representation in understanding and evaluating musical expression.