Figure 3. Neural code for syllables in the left hemisphere.

Syllables are theorized to be composed during planning and used as frames for articulation and monitoring networks during speech execution. a. We analyze the neural code for the syllable frame using a syllable discrimination framework: ROI-grouped HG activations of VCV trials were contrasted (subtracted) with CVC trials. b. This analysis was carried out for each speech network and time points with significant syllable discrimination were found for planning, articulation and monitoring networks. Plots show mean ΔHG contrasts of VCV versus CVC for each network aligned to the response onset. Shaded area denotes standard error. Horizontal bars indicate durations of significant contrasts (across vs. within frame, one-sided permutation test with cluster correction, cluster forming threshold p<0.05, 1000 permutations). c. Histograms show electrode distributions of peak ΔHG contrasts per network: planning, articulation, and monitoring. Significant differences in peak ΔHG (one-way ANOVA, F_2,495 = 64.3, p < 0.001, two-sided), followed by post-hoc one-sided t-tests indicate pairwise significance in syllabic frame contrasts (ΔHG) between the networks indicate a temporal progression of syllable activations across ROI networks (T_Planning < T_{Articulation,} t₍₃₃₃₎ = 3.3, p = 0.001; T_Articulation < T_Monitoring, t₍₄₀₃₎ = 1.3, p = 0.2 (not significant), T_Planning < T_Monitoring, t₍₃₅₂₎ = 3.78, p < 0.001). d. μECoG captures spatio-temporally selective syllabic coding sites for speech planning and execution. High-density μECoG electrode arrays were implanted over planning (anterior) and execution (posterior) networks in two awake patients. Electrode arrays had either 256 electrodes with 1.72 mm inter-electrode distance for S1 (green) or 128 electrodes with 1.33 mm inter-electrode distance for S2 (pink). e. Neural code for the syllable was extracted by contrasting HG activations per syllabic structure (ΔHG: VCV – CVC) for each electrode. Heat maps show spatial activation patterns of onset and peak times of ΔHG activations for electrodes with significant syllable contrasts. Large circles reflect electrodes with significant syllable contrasts (one-sided permutation test with cluster correction, cluster forming threshold p < 0.05, 1000 permutations, S1: 100/201, S2: 37/63). Non-significant electrodes are indicated by grey dots. These significant contrasts indicate syllable coding on an electrode specific level and demonstrate spatial selectivity on the underlying neuroanatomy. f. A spatial correlation analysis of the syllable contrast onsets along the anterior-posterior axis shows a spatial progression of temporal coding for syllable codes starting in the anterior planning section and moving towards the posterior execution network (S1: R² = 0.05, F_2,92 = 6.02, p = 0.016, S2: R² = 0.08, F_2,28 = 6.02, p = 0.07, two-sided). This spatio-temporal progression was also seen in the peak times (F-statistic test against the constant model, S1: R² = 0.09, F_2,98 = 10.9, p < 0.001, S2: R² = 0.13, F_2,28 = 5.68, p = 0.024, two-sided). Dotted lines indicate the 95% confidence interval of the linear model fit. The presence of a spatial gradient, extracted using high-density μECoG electrode arrays, demonstrates a potential anatomical substrate for the transition from planning to execution for speech production.