Figure 5. Phonological units are sequentially articulated during execution.

a. Graphical representation of sequential execution that depicts phonological units articulated in their correct order during the utterance. Colored histograms show distribution times of decoding peaks for each positional phonological unit within the ROI network. Positional phonological unit codes were decoded in the correct serial order (pairwise one-sided Mann-Whitney U test, p < 0.05, for exact p-values see Main text) b. Time-resolved decoding models were trained on neural signals aligned to the utterance to decode phonological units at each position within the pseudo-word. Each colored trace indicates temporal evolution of positional phonological units in time for a ROI (mean across 20 decoding instances). Results are shown for all ROIs in the left hemisphere, grouped by planning, articulation, and monitoring networks. Significant phonological unit decoding for all positions was observed only in execution ROIs (articulation and monitoring), suggesting that temporal sequencing is carried out as part of the execution of speech sounds rather than during planning. Horizontal bars indicate durations of significant decoding (one-sided temporal permutation cluster test, cluster forming threshold p < 0.05; 250 shuffled decoders). Background shading indicates ROI networks for planning, articulation and monitoring. The presence of ordering demonstrates the serial execution of positional phonological unit codes by the execution network. c. Continuous phonotactic transitions are present in motor cortex during articulation. μECoG electrode array was implanted over sensorimotor cortex in an awake patient. The electrode array had 128 electrodes with 1.33 mm inter-electrode distance (purple) d. Graphical schematic demonstrating that phonological unit sequence execution may involve both discrete phonological units as well as their underlying transitionary statistics (phonotactics). Forward phonotactic transition from C1 to V2 is denoted by P_fwd1, and from V2 to C3 is denoted by P_fwd2. Time-resolved decoding revealed accurate phonological unit sequencing for CVC trials (C1: blue, V2: orange, C3: yellow). Temporal decoding models trained to predict the phonotactic transitions (3-way: low, medium, and high phonotactic transitions) demonstrated significant decoding for P_fwd1 (dark brown) and P_fwd2 (light brown). Horizontal bars indicate durations of significant decoding (one-sided temporal permutation cluster test, cluster forming threshold p < 0.05; 250 shuffled decoders). Colored histograms indicate distribution times of peak decoding onsets for positional phonological units and phonotactic transitions, revealing significant temporal ordering of phonological units and phonotactic transitions (pairwise one-sided Mann-Whitney U test, p < 0.05, T_C1 < T_V2, p < 0.001; T_V2 < T_C3 p < 0.001, T_Pfwd1 < T_Pfwd2, p < 0.001). Further, the timing of the forward transitions was embedded between the phonological unit positions (pairwise one-sided Mann-Whitney U test, p < 0.05, T_C1 < T_Pfwd1, p = 3.7e-08; T_Pfwd1 < T_V2, p < 0.001, T_V2 < T_Pfwd2, p < 0.001, T_Pfwd2 < T_C3, p < 0.001). This significant temporal embedding demonstrates the existence of both discrete and continuous motor coding during the execution of sequences during speech production.