Abstract
Title of talk: Predicting where to attend in reading and scene perception
Due to our foveated visual system, we constantly need to predict where to attend next, while at the same time cognitively processing the stimulus at fixation as well as previously attended stimuli.
Eye tracking in combination with gaze-contingent display manipulations allows deep insights into how these parallel processes are coordinated. I will present recent research from my lab that addresses how ongoing higher-level cognitive processes guide the prediction of where to attend during reading and scene perception. In reading, the perceptual span gives a real-time readout of how far ahead in the text we attend. We show that the perceptual span is dynamically influenced by ongoing word processing, grows with development, and is longitudinally predicted by executive functioning. We also provide evidence that word predictability from context has a strong influence on how long we need to fixate a given word, suggesting that we constantly generate predictions when reading a sentence. Interestingly, such linguistic predictions also seem to guide where we attend in images during scene perception. We computed linguistic statistics on verbal scene descriptions obtained from humans or large language models and show that these statistics predict where and in which sequence an independent sample of participants attends to the image. Whereas earlier studies of scene perception focused on low-level visual sali