Coherence and connectivity analyses

The timeseries of the frontal ([−46, 2, 28]) and temporal ([−56, −34, 12]) sources of interest (Fig. 5b) were extracted between 0 and 912 ms after every spoken word using the function spm_eeg_inv_extract. The condition-averaged waveform (i.e. the evoked response) in each source was then subtracted from every trial to result in data with zero-mean and approximate stationarity within the time window of interest. The Fourier spectra were then computed in FieldTrip using multitapers with a ±4 Hz smoothing box. This decomposition was then subjected to separate FieldTrip connectivity analyses with either imaginary coherence or Granger causality. This same procedure was repeated 1000 times with the trial labels in each region shuffled to create a null distribution. Statistical assessment of the presence of coherence or connectivity at each frequency involved the comparison of the observed data against the null distribution (Fig. 8a, c). Between-group comparisons of imaginary coherence employed unpaired t-tests with unequal variance (the normality assumption was not violated), cluster corrected for multiple comparisons (Fig. 8b). To compare the strength of Granger causal relationships between regions, we first corrected for differences in signal to noise ratio between participants and regions by dividing the magnitude of each frequency value by the across-frequency mean for that individual-region pair. This created a profile of relative influence for each region at each frequency, corrected for overall differences in signal strength. At each frequency, the significance of ‘directionality’ (i.e. temporal to frontal vs frontal to temporal) was assessed with a repeated measures general linear model, and the output corrected for multiple comparisons (Fig. 8d).