What to Expect When the Unexpected Becomes Expected: Harmonic Surprise and Preference Over Time in Popular Music

Abstract

Previous work demonstrates that music with more surprising chords tends to be perceived as more enjoyable than music with more conventional harmonic structures. In that work, harmonic surprise was computed based upon a static distribution of chords. This would assume that harmonic surprise is constant over time, and the effect of harmonic surprise on music preference is similarly static. In this study we assess that assumption and establish that the relationship between harmonic surprise (as measured according to a specific time period) and music preference is not constant as time goes on. Analyses of harmonic surprise and preference from 1958 to 1991 showed increased harmonic surprise over time, and that this increase was significantly more pronounced in preferred songs. Separate analyses showed similar increases over the years from 2000 to 2019. As such, these findings provide evidence that the human perception of tonality is influenced by exposure. Baseline harmonic expectations that were developed through listening to the music of “yesterday” are violated in the music of “today,” leading to preference. Then, once the music of “today” provides the baseline expectations for the music of “tomorrow,” more pronounced violations—and with them, higher harmonic surprise values—become associated with preference formation. We call this phenomenon the “Inflationary-Surprise Hypothesis.” Support for this hypothesis could impact the understanding of how the perception of tonality, and other statistical regularities, are developed in the human brain.

Does Amount of Information Support Aesthetic Values?

“…The most common way to apply the syntactic approach to information mentioned in the previous section has been through principles of Information Theory ( Shannon, 1948 ; MacKay, 2003 ; Cover and Thomas, 2006 ). Multiple examples of such applications exist in the area of aesthetic values in both the visual ( Arnheim, 1974 ; Stamps<suffix>III.</suffix>, 2002 ; Rigau et al, 2007 , 2008 ; Aleem et al, 2017 ; Correa-Herran et al, 2020 ) and auditory ( Meyer, 1957 ; Knopoff and Hutchinson, 1981 ; Temperley, 2007 ; Rohrmeier and Koelsch, 2012 ; Agres et al, 2013 ; Miles et al, 2017 , 2021a ; Delplanque et al, 2019 ) domains.…”

“…To explain in more detail how to use Shannon entropy in the visual domain, we follow the presentation of Aleem et al (2017) . For a similar presentation in the auditory domain, see the work of Miles et al (2017 , 2021a) . They wanted to measure complexity in visual images, especially in paintings.…”

“…This happens because surprise or violation of expectation is equivalent to “something is wrong here.” Thus, part of the pleasure in music is thought to be due to the discharge from the tension prompted by surprise. Second, surprise appears as a quantity ( Miles et al, 2017 , 2021a ). Basically, a surprise is a violation of expectation and thus, one can define it precisely in terms of an underlying probability distribution ( Egermann et al, 2013 ).…”

“…In general, one cannot talk about the entropy of a sample because it may only have one member. An example of a sample used in the literature is the set of chords in a section of a song ( Miles et al, 2021a , b ). In this case, the population probabilities could arise from “all” songs in a year or integrated over several years.…”

“…However, in the case of amount of surprise over temporally integrated expectations, one instead uses statistics that accumulate over time and thus, possibly changes continuously. For example, when Miles et al (2021a) studied surprise in songs released in, say, 1980, they compared them to songs released for many years before that time point. Because they did the same for 1990, the underlying statistics could be different.…”