Fig. 2: Enhanced encoding of word-level and phoneme-surprisal features in native speech.
From: Shared and language-specific phonological processing in the human temporal lobe
a, Spanish (left) and English (right) sentences annotated with phonemes, syllable boundaries, phoneme surprisal, word boundaries, frequencies and lengths. We hypothesized that neural encoding of speech representations requires increasing language experience (colour bar at right). b, Across electrodes, the variance in the neural response uniquely explained by word-level features is significantly higher in native versus foreign speech (LME model boundary ***P < 0.001, frequency ***P < 0.001, length ***P < 0.001; see Methods for formula) and phoneme surprisal (LME **P = 0.0012). Box plots show non-outlier maximum and minimum values (whiskers), median (centre line) and lower and upper quartiles (box limits) corresponding to electrodes surpassing a significance threshold for model R2 > 0.05 and ΔR2 > 0.001 (boundary n = 246, frequency n = 256, length n = 126). c, Unique variance for acoustic–phonetic features does not significantly differ in native versus foreign speech (LME peakRate P = 0.41, formant P = 0.60, consonant P = 0.31; see Methods for formula). Box plots show non-outlier maximum and minimum values (whiskers), median (centre line) and lower and upper quartiles (box limits) corresponding to electrodes surpassing a significance threshold for model R2 > 0.05 and ΔR2 > 0.001 (peakRate n = 502; formant n = 316; consonant n = 270). d, Electrodes with positive unique variance for word and phoneme-surprisal features primarily located in mid-STG; native (left) and foreign (right) speech. Electrode size indicates unique variance magnitude. Black scatters correspond to speech-responsive electrodes. e, Electrode counts with significant word and phoneme-surprisal feature unique variance (permutation test versus shuffled distribution) across native and foreign speech in each anatomically defined brain region. f, Proportion of speech-responsive electrodes with significant word and phoneme-surprisal feature unique variance (permutation test versus shuffled distribution) for speech conditions across hemispheres; native (left) and foreign (right) speech. g, Ninety-eight electrodes show positive unique variance for word and phoneme surprisal in both native and foreign languages; unique variance for word and phoneme surprisal is not significantly correlated across languages (Spearman r(98) = 0.16, P = 0.12). Inset plot shows electrode count in each quadrant of the scatter plot. NS, not significant.
