Abstract
Normally-sighted subjects typically read words by placing the maximal acuity zone of their retina (i.e. the fovea) directly on the word. In contrast, patients with macular lesions (i.e. central scotomata) need to place the fovea outside the word itself, thus using peripheral areas of the retina to read. It is assumed that patients preferentially use a particular peripheral zone called the Preferred Retinal Location (PRL). The correlation between this PRL and reading performance is still controversial, giving rise to several studies with both patients and simulated subjects.
In this study, we present a bayesian “ideal observer” analysis of single-word reading in normal readers and central scotoma patients. In the latter case, numerous fixations can be necessary before recognition occurs. Our approach to reading assumes that the optimal reading strategy is the one that optimizes the “Expected Information Gain” for each future fixation. This gain is calculated on the basis of the image pixel values and takes into account the information provided by pixels about letter and word identities. Importantly, the model predicts the 2D spatio-temporal pattern of saccades during reading by using only pixel-based information, in contrast with theories that use the pre-processed letter-slot approaches to model reading.
The implementation of the model shows that vertical strategies (i.e., placing scotoma above or below the word) are optimal for word reading. We found a similar pattern of results in a word recognition experiment where macular blindness was simulated with a gaze-contingent paradigm in normally-sighted observers. More generally, our results support the idea that information processing models can help define optimal oculomotor strategies and provide important insights for visuo-motor rehabilitation methods.