The dynamics of neural populations have provided insight into the structure and function of many neural circuits 6,26,27,29,32,33
. To determine the dynamics of phonetic representations, we investigated how state-space trajectories for CVs entered and departed target regions for phonetic clusters. Trajectories of individual CV syllables were visualized by plotting their locations in the first two PC dimensions versus time (, PC1 & PC2; for Subject 1).
Dynamics of Phonetic Representations
We first examined how trajectories of different consonants transitioned to a single vowel /u/ (). The cortical state-space was initially unstructured, and then individual trajectories converged within phonetic clusters (e.g. labial, front tongue, dorsal tongue, and sibilant), while simultaneously cluster trajectories diverged from one another. These convergent and divergent dynamics gradually increased the separability of different phonetic clusters. Later, as each consonant transitioned to /u/, trajectories converged to a compact target region for the vowel. Finally, trajectories diverged randomly, presumably as articulators returned to neutral position. Analogous dynamics were observed during the production of a single consonant cluster (e.g. labials) transitioning to different vowels (/a/, /i/, and /u/)().
We quantified the internal dynamical properties of the cortical state-space by calculating cluster separability, which measures the mean difference of between-cluster and within-cluster distances. The time course of cluster separability, averaged across subjects and CVs, is plotted in : separability peaked ~200 ms before the CV transition for consonants (onset, ~−300 ms), and at +250 ms for vowels (onset, ~50 ms). We further examined the dynamics of correlations between the structure of the cortical state-space and phonetic features (averaged across subjects), which is plotted in . Across subjects, we found that cluster separability and the correlation between cortical state-space organization and phonetic features were tightly linked for both consonants and vowels in a time-dependent fashion (R2 range = [0.42–0.98], P < 10−10 for all). This demonstrates that the dynamics of clustering in the cortical state-space is strongly coupled to the degree to which the cortical state reflects the phonetic structure of the vocalization.
The dynamic structure of the cortical state-space during production of all CV syllables is summarized in . In this visualization, the center of each colored tube is located at the centroid of the corresponding phonetic cluster. Tube diameter corresponds to cluster density and color saturation represents the correlation between the structure of the cortical state-space and phonetic features. This visualization highlights that, as the cortical state comes to reflect phonetic structure (coloring), different phonetic clusters diverged from one another, while the trajectories within clusters converged. Furthermore, we observed correlates of the earlier articulatory specification for sibilants (red, e.g. /sh/, /z/, /s/). Additionally, with all CVs on the same axes, we observed that consonants occupy a substantially distinct region of cortical state-space compared to vowels, despite sharing the same articulators. The distribution of distances comparing consonant and vowel representations was significantly greater than the consonant-consonant comparison or vowel-vowel comparison (P < 10−10
for all comparisons, WSRT, n = 4623 for all, Supplementary Figure 11
). Finally, the consonant-to-vowel sequence reveals a periodic structure, which is sub-specified for consonant and vowel features.