Why we linger: Memory encoding, rather than visual processing demand, drives fixation timing on natural scenes - evidence from a large-scale MEG dataset

Prior to each of the ~200k daily eye movements, our brain needs to decide how long to remain fixated at the current location while novel information awaits elsewhere. To elucidate this process, here we report on the results of a large-scale scene-viewing experiment [4,080 natural scenes, 5 participants] that combines magnetoencephalography (MEG), eye tracking, and a semantic scene captioning task. Using multivariate analysis techniques of MEG source-space patterns, behavioural analyses and artificial neural network (ANN) modelling, we demonstrate that longer fixations do not arise from prolonged visual processing, but rather relate to downstream memory encoding. First, the temporal variability of ventral stream representational dynamics does not explain variability in fixation duration. Second, fixation durations are anti-correlated with patch complexity, as estimated from task-trained ANN models. Third, fixation durations correlate positively with ANN-predicted patch memorability and caption-inclusion, and co-occur with increased theta-gamma phase-amplitude coupling, particularly in frontal, and hippocampal regions.