Music perception in historical audiences: Towards predictive models of music perception in historical audiences

Abstract

Background in Historical Musicology. In addition to making inferences about historical performance practice, it is interesting to ask questions about the experience of historical listeners. In particular, how might their perception vary from that of present-day listeners (and listeners at other time points, more generally) as a function of the music to which they were exposed throughout their lives. Background in Music Cognition. To illustrate the approach, we focus on the cognitive process of expectation, which has long been of interest to musicians and music psychologists, partly because it is thought to be one of the processes supporting the induction of emotion by music. Recent work has established models of expectation based on probabilistic learning of statistical regularities in the music to which an individual is exposed. This raises the possibility of developing simulations of historical listeners by training models on the music to which they might have been exposed. Aims. First, we aim to develop a framework for creating and testing simulated perceptual models of historical listeners. Second, we aim to provide simple but concrete illustrations of how the simulations can be applied in a preliminary approach. These are intended as illustrative feasibility studies to provide a springboard for further discussion and development rather than fully fledged experiments in their own right. Third, we aim to appeal to the expertise of historical musicologists in identifying useful research questions and appropriate constraints for the simulations, so these can be used to complement existing evidence on the perception of music by historical listeners. Main contribution. Our primary contribution is to develop and illustrate a framework which we believe can shed light on the perception of music by historical listeners and, in particular, how listeners of different periods might have generated different predictions to music as a function of differences in their musical experiences. The framework we develop involves several steps. First, identifying a research question; second, selecting a corpus (or corpora) to represent the musical experience of the listener(s) we want to simulate; third, identify the central musical features of interest and use them to develop a representation scheme for the selected compositions; finally, the model parameters are selected and the models are trained on the selected corpora to simulate particular listeners. We identify and discuss the decisions that must be made at each step. Finally, we illustrate the framework by training models on a range of corpora from different stylistic traditions from different locations and points in history, including analyses at the level of entire collections, individual compositions, and individual events. Implications. The results of our illustrative analyses suggest that the trained models behave as we hypothesised, demonstrating sensitivity to stylistic similarities which could illuminate how listeners from different eras might have experienced musical structures. However, the approach is in need of expertise in historical musicology to establish clear and relevant research questions and to select appropriate parameters for the simulations. With such additional input, we believe simulated listeners will provide important insights, alongside other evidence, into the question of how our forebears experienced the music of their time.