# Temporal control in modelling eye fixations (TIME) ## Casimir J.H. Ludwig, Hans A. Trukenbrod, Ralf Engbert In natural vision, processing difficulty will change periodically during sequences of fixations. To ensure adaptive behaviour, it is important that fxation duration is adjusted to systematic changes in difficulty. We aim to understand these temporal control mechanisms by developing and fitting computational models to extended sequences of fixations. Observers generated large sequences of fixations in a gaze-contingent paradigm where the fixated item cued the direction of the next saccade. Items were embedded in visual noise, which changed periodically to increase or decrease processing difficulty. The key features of our data are: fixation duration adapts immediately to an increase in processing difficulty; adaptation to increasing difficulty initially "overshoots", but then decreases again to a new steady state; adaptation of fixation duration to decreasing difficulty tends to be delayed (but not always). Here we report simulations and Bayesian fits of a model based on evidence accumulation to a threshold. The model has three key mechanisms that operate on different timescales: (i) during the first fixation after *any* change in difficulty, foveal inhibition transiently reduces the drift rate; (ii) the drift rate itself varies with the quality of evidence presented during a given fixation; (iii) the decision threshold adjusts over the course of several fixation in response to the recently experienced processing difficulty. These mechanisms themselves are symmetric with respect to processing difficulty. However, their interaction over different timescales explains the asymmetry in adaptation between increasing and decreasing difficulty, as well as the initial overshoot in fixation duration after an increase in difficulty.