Abstract
Saccades and smooth pursuit eye movements adapt efficiently to environmental regularities. In the presence of a probabilistic bias favoring a particular visual motion direction, a robust linear correlation between human anticipatory smooth eye movements (aSEM) and direction probability has been previously reported. In real life, regularities are not immutable, and relevant behavioral contingences vary depending on the context. Importantly, sensorimotor learning depends on the nature of the specific cues signaling the context (Azadi and Harwood, 2014; Howard et al. 2013). Here we ask whether aSEM adapt efficiently to a probabilistic direction bias signaled by either a color cue or a motion-direction cue. Participants had to accurately track a small visual target moving either rightward or leftward (11°/s). In two experiments, unbeknownst to the participant, the probability of target motion direction (say P(Right)) was paired with the cue type and was manipulated across blocks. In practice, in experiment 1, when the fixation stimulus (color cue) was green the probability of rightward motion was P(Right|green)=0.5,0.25 or 0.75, while with the red cue the probability was the complementary one, P(Right|red)=1-P(Right|green). In a second experiment, the probability of target rightward motion was paired with the target vertical motion direction during the previous motion epoch: P(Right|target UP)=0.5,0.25 or 0.75; P(Right|target DOWN)=1- P(Right|target UP). Our data revealed two radically different oculomotor behaviors. aSEM did adapt to the contextual probabilistic contingency schedule signaled by the motion-direction cue but not when using the color cue. These results are well predicted by alternative probability-learning models based on different learnt probabilistic functions. Importantly, the initiation of visually-guided horizontal smooth pursuit was also modulated differently by the cue-conditional probability in the two experiments. Overall, these results highlight the complexity of contextual visuomotor adaptation and call for a new theoretical framework encompassing both probabilistic learning and cue efficacy for visuomotor learning.