Supplementary Fig. 1: Neurotypical contraction bias is replicated in M-Turk data.

M-Turk (Perspect. Psychol. Sci. 6, 3–5, 2011) has been shown to be a reliable tool to acquire quality data, even for very demanding experiments (PLoS One 8, e57410, 2013; Behav. Res. Methods 43, 155–167, 2011). In common hardware configurations and web browsers, the latency between executing the code to present a sound and the actual onset of sound presentation, and the difference between the actual and intended duration of sounds have both small variability (STDs < 1 ms) (Behav. Res. Methods 48, 897–908, 2016; Behav. Res. Methods 47, 649–665, 2015). Still, we first verified the validity of the M-Turk experiment for our study by assessing the contraction bias of the 125 participants who performed the task with the broad uniform distribution (3 octaves, Methods; Supplementary Table S2). (a) We quantified the bias by measuring the difference between performance in Bias+ and Bias- trials (Fig. 1b). The bias was positive for 114/125 participants (91.2%) and was highly significant (t₁₂₄ = 2.7; p < 10⁻²⁴, Cohen’s d = 1.14, for a paired t-test). Each filled circle shows the mean accuracy of one participant in Bias- (x axis) and Bias+ trials (y-axis). Color denotes mean accuracy of the participant across all trials (side-bar). The bias is manifested in the elevation from the diagonal (higher accuracies in Bias+ trials). Bottom inset image compares the mean accuracy across participants for Bias+ (blue) and Bias- trials (red). Black points indicate the mean values; error bars indicate standard error of the mean. (b) Regressing participants’ decision (n = 125) using a GLM Probit regression (Methods) - the weights of each of the estimated bias components. The most prominent contributors to the sensory bias are the stimuli of one trial back, and the mean of all frequencies. Error bars indicate standard deviation. Buhrmester, M. et al. Amazon’s Mechanical Turk: A New Source of Inexpensive, Yet High-Quality, Data? Perspect. Psychol. Sci. 6, 3–5 (2011). Crump, M. J. C., Mcdonnell, J. V & Gureckis, T. M. Evaluating Amazon ’ s Mechanical Turk as a Tool for Experimental Behavioral Research. PLoS One 8, e57410 (2013). Sprouse, J. A validation of Amazon Mechanical Turk for the collection of acceptability judgements in linguistic theory. Behav. Res. Methods 43, 155–167 (2011). Reimers, S. & Stewart, N. Auditory presentation and synchronization in Adobe Flash and HTML5 / JavaScript Web experiments. Behav. Res. Methods 897–908 (2016). Babjack, D. L. et al. Reducing audio stimulus presentation latencies across studies, laboratories, and hardware and operating system configurations. Behav. Res. Methods 47, 649–665 (2015).