The next DaP meeting will take place on the 21th of April at 15:30 at the Laboratoire Parole et Langage, 5 Avenue Pasteur, 13100 Aix-en-Provence. During the meeting, Jean Baptiste Bernard from the Laboratoire de Psychologie Cognitive in Marseille will present a study on: Predicting oculomotor strategies in reading with normal and damaged visual fields.
Abstract : The distribution of fixation positions during reading shows that normally-sighted subjects typically read by successively placing their fovea next to the center of the words, in order to reduce the deleterious effects of the visual periphery such as reduced acuity or crowding. In this study, we present a Bayesian ideal observer of reading which is based on spatiotemporal characteristics of letter recognition across the visual field. Our approach to reading assumes that the optimal fixation is the one that optimizes the “Expected Information Gain” to identify the letters from the current word. At each fixation, Bayesian inference is used to combine prior knowledge with newly extracted information about letter identities. This is done by using letter recognition rates and letter confusion matrices across the visual field. Lexical inference is also used to update letter identity information at each fixation. The model predicts the 2D spatiotemporal pattern of saccades during reading, in contrast with theories that use 1D letter-slot approaches to model reading. This is critical to predict optimal reading strategies for readers who cannot extract visual information with their fovea, such as Age-related Macular Degenerescence patients. Model predictions for normal and damaged visual field are discussed.