Abstract
Reading speed decreases with eccentricity, even when the stimulus size and spacing are scaled. This is puzzling given that the periphery is faster by most measures. This discrepancy has been attributed to slower feature integration in the periphery. In the periphery, the rate of feature integration depends on both the number of features and eccentricity (Cantone, Tillman, & Pelli, VSS 2008). Here, we ask which features matter: the number available or only those actually used. We measure reading speed for four-letter words in the fovea and the periphery. We vary the set size from which the words are chosen: either the 26 most frequent words or all 2200 four-letter words. Less information and hence fewer features would be needed to identify words selected from the smaller set (Shannon & Weaver, 1948). Hick's law (1952) states that reaction time is proportional to the logarithm of set size and hence to the amount of information. However, the number of features available in any given word is the same in both sets. We find that set size has a large effect on reading speed, indicating that what matters is the number of features used. When words are selected from the larger set, reading speed is slower in the fovea and falls much faster (twice as fast) with eccentricity than when they are selected from the smaller set. We also find that changing the font to change the perimetric complexity (perimeter2/area, a measure of how many features are available) has no effect on reading speed. Thus we conclude that the slowing of feature integration with eccentricity depends on the number of features used, not the number available.