Because saccades are used to acquire information localized in our visual environment they are conventionally regarded as being exclusively concerned with the spatial position of objects. However, recent studies have shown that they are also affected by the temporal regularities in dynamic environments. Here we ask whether contextual control of latencies in a search task can be established using reinforcement learning.
Five subjects made saccades within 80-750ms toward a target displayed among distractors. For each subject we constructed two classes of latencies, “short” and “long”, using the first and last quartiles of individual baseline distributions (e.g. [80;294] and [432;750]ms respectively). We then used a latency-contingent display paradigm in which finding the target was made contingent upon specific saccadic latencies. For group 1, the target was displayed only for “short” latencies with leftward saccades, and for “long” latencies with rightward saccades. The opposite was true for group 2.
When short- and long-latency saccades were reinforced (i.e. the target was displayed) depending on the saccade direction (9600 reinforcement trials), median latencies differed by 85ms on average (all outside the 98% null hypothesis CIs). Post-training, in the absence of reinforcement, we still observed considerable differences in latency distributions, i.e. on average 95ms for leftward versus rightward saccades.
Our results demonstrate the contextual control of saccadic latencies, further supporting the extent of reinforcement learning for saccades. This study reveals that saccade triggering is finely controlled by learned temporal and spatial properties of the environment.