Figure 3.

Experiment 1, regularity modulated pupil size. A, B, The gap detection task showed worse performance for RAND compared with REG sequences. Sensitivity (d') to the gap was significantly higher, and RT shorter for REG relative to RAND sequences. Circles represent individual data points. Error bar shows ±1 SEM. Plots (C–F) show averaged normalized pupil diameter over time, baseline corrected (−1- to 0-s preonset). The shaded area shows ±1 SEM. The horizontal bars show time intervals during which significant differences (bootstrap statistics) were observed. The black bar shows the original results, the gray bars show the significant time intervals after adjusting for the subject-wise difference (RAND-REG) in RT (mid-gray) and d-prime (light-gray). C, Averaged pupil diameter for all conditions. D–F, Average pupil diameters separated by alphabet size 5, 10, and 15 (left to right) showed sustained larger pupil diameters for random conditions (red, orange, and yellow) than regular conditions (shades of blue). D, Alphabet size 5 showed significant differences between REG5 and RAND5 from 2 to 3 s onwards. E, For alphabet size 10, REG10 separates from RAND10 from 3 s onwards with a sustained significant difference from ∼7 to 8 s. F, For alphabet size 15, REG15 separates from RAND15 from 4 s, and is significantly different from 6 s onwards. For E, F, the significant effects at onset are likely artefacts of regressing out the behavioral measures, resulting from low variability between participants at the onset time points.