Optimal coding of interval timing in expert drummers, string musicians and non-musical control subjects

Abstract

Jazayeri and Shadlen (Nat. Neurosc. 2010) recently reported that when human observers reproduce time intervals drawn from different distributions production times exhibit a systematic regression towards the mean. They explained and model their data with a performance-optimizing Bayesian model. We repeated their study on a population of expert drummers, string musicians and non-musical control subjects, with both visual and auditory stimuli. For auditory stimuli, no subjects showed regression towards the mean. For visual stimuli, the non-musical control subjects showed a strong regression, but the drummers maintained near-veridical performance with almost no regression to the mean. The string musicians showed intermediate behaviour. We measured separately temporal precision for auditory and visual stimuli with a bisection task. Auditory precision was higher than visual precision, and the drummers were more precise in both domains. Interestingly, there was a strong negative correlation between precision in the bisection task and regression towards the mean in the production (more precise subjects regressed less). We model these data with a modified performance-optimizing Bayesian model, where the prior is calculated from a running mean of previous trials.

EU, FP7 - STANIB.