Guest Lecture: Nori Jacoby "Perceptual Priors in Music and Speech Revealed by Iterated Learning"
![](https://ccrma.stanford.edu/sites/default/files/user/takako/norijacoby_image001.png?1471304853)
Bio:
Nori Jacoby studies the role of culture in auditory perception. His current work uses iterated learning alongside classical psychophysical methods to characterize perceptual biases in music and speech rhythms in various populations ranging from Westerners to the Tsimané, an Amazonian foraging-farming society in Bolivia. He is also working on computational modeling of synchronization and entrainment in jembe drum ensembles in Mali. Nori completed a Ph.D. at the Hebrew University of Jerusalem and was a postdoc at Josh McDermott's Computational Audition Lab at MIT and at Tom Griffiths's Computational Cognitive Science Lab at UC Berkeley. Starting in Fall 2016, he will be a Presidential Scholar In Society And Neuroscience at Columbia University.
Talk title: Perceptual Priors in Music and Speech Revealed by Iterated Learning
Nori Jacoby1,2, Josh McDermott1
1 Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA
2 Department of Psychology, University of California, Berkeley, USA
Abstract
Western musical notation is built on temporal intervals related by simple integer ratios. In contrast, speech features non-periodic prosodic rhythms that are influenced by linguistic context and phonological constraints (Turk and Shattuck-Hufnagel 2014). We introduce a novel procedure adapted from computational concept modeling (Bartlett 1932, Griffiths and Kalish 2005) in which a listener's reproduction of 2- or 3- interval rhythms is used to infer the internal "prior" probability distribution constraining human perception. Listeners are presented with a random seed (interval ratios drawn from a uniform distribution), and reproduce it by tapping, vocalizing, or speaking. The reproduction is then substituted for the seed, and the process is iterated. After a few iterations the reproductions become dominated by the listener's internal biases, and converge to their prior.When tested on musical rhythms in this way, Westerners consistently converged to attractors near integer-ratios. To explore whether these biases are due to passive exposure to Western music, we tested members of the Tsimané, an Amazonian hunter/gatherer society, on the same task. Despite profoundly different cultural exposure, the Tsimané also exhibited a prior favoring integer ratios, though the prominence of certain ratios differed from Westerners. In contrast, when applying the same paradigm to speech rhythms we also obtained non-integer ratio attractors associated with interpretable phonological constraints.
Our results provide evidence for the universality of integer ratios as perceptual priors in music. However, we found that the relative strength of rhythmic perceptual categories depends both on intra-group exposure and on context (speech/music). These findings were enabled by our method, which provides high-resolution measurement of rhythmic priors.