Fig. 2: Identifying electrodes that show predictive recall.

a Tracking of neural prediction via Granger Causality: Prediction between runs is added to each auto-regressive model. If the neural signal acquires memory about upcoming states after listening to the story once, then data from run 2 should be able to improve the prediction of the auto-regressive model for run 1 (red arrows). Signal from run 1, on the other hand, should not be able to improve the prediction of the auto-regressive model for run 2 (blue arrows). The difference between those predictions across runs is interpreted as a measure of predictive recall. b Difference in F-values between the prediction of run 1 from run 2 and the prediction of run 2 from run 1, indicating neural evidence for predictive recall that emerges on the second run of listening. Part b shows electrodes selected for further analyses and part c shows all electrodes (the bottom row images are slightly rotated outwards for enhanced visibility of three-dimensional electrode positions).