Integration of letter information across a simulated central scotoma

Purpose. C;m readers integrate visual information across a central scotoma.' If so, how much improvement in reading speed might he achieved? We examined whether an ideal observer model of reading, named Mr. Chips, wouid perform better when visual information is present on both sides of a central scotoma (two visual islands), or on just one side of the central scotoma (one visual island). We have also created a reading task in which we compared the performance of human readers, with simulated central scotomas, to the performance of Mr. Chips. Methods. Strings of text are sampled through a "retina" with regions of highresolulion (the visual span) in which letters can be identified, and low-resolution (periphery and scotomas) in which spaces can be distinguished from letters bin letters cannot be identified. The size of the visual span is set at 9 characters, and a central scotoma of 1 to 7 characters is embedded within this visual span. Subjects make simulated saccades b> pressing k<:ys on an electronic piano keyboard. Each key represents a movement of an integc r number of characters forward or backward in the text. The task of the human and t te ideal reader is to move through the text in the minimum number of saccades, iden itying every word. Suhiects are gi\en as much time as they need to make their s.iccade decisions. Results. When two islands of vision an' used rather than one, ihe mean saccade si/e of the ideal observer increases by ( 7%. Preliminary results indicate that human readers can approach the performance cf the ideal observer. The increase in mean saccade siJ.e when two islands of visiot arc present rather than one is 61% . Conclusions. Given unlimited time for planning simulated saccades, human readers can integrate information acros - a central scotoma and approach the level of performance of an ideal observer. If rc\ ders are asked to perform the task more quickly, integration of information acu ss a central scotoma becomes less likely.