Fig. 5: Population neural activity in human STG supports categorical representation of lexical tones shaped by language experience.

a Native Mandarin speakers and English speakers listen to the same Mandarin speech stimuli. b The scatterplot of the linearity versus the modulation depth for the tuning curves of relative pitch height in STG electrodes from native Mandarin speakers (red) versus native English speakers (blue). More negatively tuned electrodes are found in native Mandarin speakers (shaded gray area). c The averaged temporal receptive field (absolute beta weights of the encoding regression model) for relative pitch height encoding electrodes in Mandarin (red) and English speakers (blue). (mean ± s.e.m., black markers indicate time points with significant difference between the two groups, FDR corrected α < 0.05, two-sample two-sided t test). d Same as b, but for pitch change. e Same as c, but for pitch change. f The pair-wise tone classification accuracy relative to the accuracy in the acoustic space, using speech-sensitive electrodes (red: native Mandarin speakers; blue: native English speakers; dark red: native Mandarin speakers excluding strong negative tuning electrodes shown in b); error bars indicate standard error of the mean estimation from n = 200 times bootstrapping. g Using partitions based on averaged relative pitch height and averaged pitch change respectively, representational similarity analysis (RSA) is performed on acoustic space and the population STG activity in native Mandarin speakers and native English speakers, respectively. Categorical index = mean between-tone classification accuracy (red blocks) – mean within-tone classification accuracy (cyan blocks). h The categorical index based on acoustic relative pitch contour, STG population space in native Mandarin speakers (red) and STG subset excluding negative tuning electrodes (dark red), as well as in native English speakers (blue) (***p < 0.005, n.s. p > 0.1, permutation test).