Figure 4

Clear Reading Speed as a Control Measurement. Analysis of the population reading speed in the clear condition supports the hypotheses that significant methodological variations did not occur and that linguistic proficiency was not a factor in the learning process. (A) Population reading speed over time for the clear conditions, with mean (solid orange line) and standard deviation (filled area) pooled across font sizes. After a startup transient, there is no clear difference in learning when comparing the per-session early performance (open circles, sessions 6–10, corresponding roughly to the second week, labeled start) to final performance (closed circles, sessions 36–40, corresponding roughly to the final week, labeled end), as verified statistically (Wilcoxon rank sum test, p = 0.5). The two segments used in the comparison are highlighted (gray zones). (B) To additionally verify a lack of influence, the weekly mean performance on clear conditions versus phosphene conditions over time is plotted as a scattergram. The weekly performance was computed for each viewing condition on a per-subject basis, pooling across font sizes, and normalizing by the mean value. The range of normalized clear view performance was 0.9 to 1.1, and the range of normalized phosphene view performance was 0.5 to 1.9. Different colors represent the different phosphene viewing conditions as in Fig. 1. Open circles are the first comparison segment (skipping the startup transient as in A), points are intermediate weeks, and filled circles are the final comparison segment (also as in A). The gray line shows a linear regression with slope of 0.03, and R² of much less than 0.01, demonstrating that phosphene view performance was decoupled from clear view performance.