Iterated reproduction task reveals rhythmic priors associated with exposure to music

According to Bayesian theories of cognition, perception involves integration of noisy sensory information with probabilistic internal models. These internal models reflect the net sum of all of our prior experiences and assist in structuring perception in the presence of unreliable sensory input. Known as priors, the influence of internal models can most clearly be observed in situations where sensory input is weak. In these cases, the prior makes a much larger contribution to perception, effectively biasing perception toward events that are more commonly encountered.


Musical practices are observed throughout all cultures and each musical system emphasises different rhythmic signatures. Is it possible that exposure to music forms rhythmic priors that help structure our perception of auditory sequences, and if so, are these rhythmic priors influenced by culture?


To assess rhythmic priors, Nori Jacoby and Josh McDermott from MIT devised an iterated reproduction task wherein participants tapped in time with auditory sequences comprised of repeating three-interval rhythms (e.g., 3:2:1, 1:2:1). On each trial the researchers surreptitiously replaced the auditory sequence with the rhythm produced by the participant from the previous trial. The idea of this procedure is that if temporal priors help structure the perception of musical sequences, then the rhythms produced by participants over successive trials should gradually become biased toward these priors. Indeed, the authors showed that reproductions tended to drift from the initial sequence and then stabalise after only five trials.


However, to ensure that the task itself was not biased toward cultural norms, the initial rhythm was randomly generated. In western music, interval ratios are usually comprised of integers. So to prevent the task from being influenced by western music conventions, the initial trial was randomly selected from all possible interval ratios, including non-integer values.


Despite the rhythms being randomly generated, reproductions tended to converge toward sequences with integer ratios. Importantly this effect was observed in a range of control experiments designed to rule out the role of motor demands. For example, the result was not specific to the effector since an integer bias was found when participants provided a verbal response. Likewise, a ratio bias was observed when sequences were reproduced from memory, indicating that the effect was not due to auditory-motor entrainment associated with synchronisation tasks.


Indeed, the effect of priors was also apparent in perceptual discrimination tasks. Participants were presented sequences that varied along a continuum between 3:2:3 and 1:1:1 and performed a same-different judgement task on pairs of sequences. Discrimination performance showed a pattern characteristic of categorical perception, with increased sensitivity found for non-integer rhythms and decreased sensitivity for rhythms near to integer ratios. The loss of perceptual sensitivity near integer patterns is indicative of a prior drawing the perception of patterns toward integer rhythms.


Crucially, the integer bias uncovered by the iterated reproduction task was influenced by exposure to music. In American participants, biases were observed only for ratios commonly found in western music. Likewise, a remote Amazonian population – the Tsimane – also showed a bias for integer ratios, however in this case, biases were only shown for intervals found in Tsimane music. However, the effect of the priors appeared to reflect passive exposure to common rhythmic structures, as American musicians also showed the same pattern of integer bias as Americans with no musical training.


Although Amercian and Tsimane cultures differed in the profile of intervals associated with priors, both cultures showed preferences for integer ratios. The Tsimane are a remote population with almost no exposure to western culture so it is unlikely that cultural transmission can explain a preference for integer ratios in the Tsimane. So this begs the question, how is it that both groups show priors for integer rhythms? Although iterated reproductions are often used in social science to explore the dynamics associated with the formation of shared practices, attitudes and beliefs, the authors’ stress that the task used here does not recapitulate the development of rhythmic preferences. Instead they argue the task only uncovers pre-existing internal preferences. How widespread such preferences are across different cultures and why preferences for integer rhythms emerge remains to be seen.


Bronson Harry

The MARCS Institute, Western Sydney University

Twitter | ResearchGate | Web