Statistical learning and probabilistic prediction in music cognition: mechanisms of stylistic enculturation

Abstract

Music perception depends on internal psychological models derived through exposure to a musical culture. It is hypothesized that this musical enculturation depends on two cognitive processes: (1) statistical learning, in which listeners acquire internal cognitive models of statistical regularities present in the music to which they are exposed; and (2) probabilistic prediction based on these learned models that enables listeners to organize and process their mental representations of music. To corroborate these hypotheses, I review research that uses a computational model of probabilistic prediction based on statistical learning (the information dynamics of music (IDyOM) model) to simulate data from empirical studies of human listeners. The results show that a broad range of psychological processes involved in music perception-expectation, emotion, memory, similarity, segmentation, and meter-can be understood in terms of a single, underlying process of probabilistic prediction using learned statistical models. Furthermore, IDyOM simulations of listeners from different musical cultures demonstrate that statistical learning can plausibly predict causal effects of differential cultural exposure to musical styles, providing a quantitative model of cultural distance. Understanding the neural basis of musical enculturation will benefit from close coordination between empirical neuroimaging and computational modeling of underlying mechanisms, as outlined here.

Musical prediction error responses similarly reduced by predictive uncertainty in musicians and non-musicians

“…LE melodies were transposed to two different octaves to cover approximately the same pitch range as HE melodies. The uncertainty of the stimuli was estimated with Information Dynamics of Music (IDyOM), a variable-order Markov model of auditory expectation (Pearce, 2005(Pearce, , 2018. When predicting pitch continuations based on a training corpus of Western tonal hymns and folk songs, this model confirmed higher mean entropy (which is a measure of uncertainty) and information content (which is a measure of surprise) for HE as compared to LE melodies (see Quiroga-Martinez et al, 2019 for more details).…”

“…Prediction is fundamental for the perception of auditory sequences. When listening to a series of sounds, the brain generates expectations about future events partly based on the statistical regularities of the context and long-term knowledge of acoustic signals (Huron, 2006;Pearce, 2018). The violation of these expectations generates neural prediction error responses (den Ouden, Kok, & de Lange, 2012).…”

Music predictability and liking enhance pupil dilation and promote motor learning in non-musicians

“…activity -as they conform to listeners' prior expectations and allow them to form precise predictions 243 about the incoming stimulus 1 . Such effect of predictability may explain effects of greater sensitivity and 244 memory of music from one's own culture, or of simple more than complex excerpts 31,32 .…”

“…Through passive exposure to music, we implicitly develop models about its structure 1,2 . These models 27 allow both listeners to generate expectations about upcoming musical events 3-5 , and trained musicians 28…”

“…Similar cognitive mechanisms are likely to play a role in musical perception and performance. When 47 listening to music, humans entertain a number of predictions, or hypotheses, about future musical 48 events 1,28 , and these predictions are subject to refinement and learning on different time scales and at 49 different levels of sophistication. For example, musical predictions vary as a function of the musical 50 culture one is exposed to [29][30][31] and can adapt to novel musical styles 32 .…”