Controlling target selection for saccadic eye movements by informational value
Humans make up to three saccadic eye movements per second, with each saccade changing which part of a visual scene is projected on the relatively small retinal fovea, where visual information is processed with the highest visual acuity, and which parts of scenes are projected onto the relatively lar...
|PDF Full Text
No Tags, Be the first to tag this record!
|Humans make up to three saccadic eye movements per second, with each saccade changing which part of a visual scene is projected on the relatively small retinal fovea, where visual information is processed with the highest visual acuity, and which parts of scenes are projected onto the relatively large peripheral part of the retina, where visual acuity is comparatively low. Due to the small size of the fovea, each saccade is necessarily led by a decision about which part of a visual scene to choose as the next eye movement target and followed by an evaluation of the eye movement’s spatial accuracy (i.e., was the saccade spatially accurate enough to allow foveation of relevant visual information?). Since visual scenes, however, are often complex and contain an abundance of visual information, two of the major challenges that the oculomotor system faces are to decide which environmental factors to take into account when choosing a saccade target, and what information in the postsaccadic visual input to consider for evaluation of eye movement accuracy. In three studies, the current dissertation investigated how the magnitude of task relevant information that can be gained at a saccade target (i.e., a saccade’s informational value) influences target selection for saccadic eye movements as well as target selection for saccade adaptation, i.e., the mechanism that maintains eye movement accuracy.
Study I tested how the informational value and the costs of saccades are balanced for saccade target selection in a combined visual search and perceptual discrimination task. We instructed participants to find a target amidst distractors, and to discriminate a feature on the target. Participants were rewarded for correct discriminations and were given a limited amount of time to complete as many trials as possible. Informational value was manipulated by showing, in each trial, two targets with different discrimination difficulty. Participants were instructed that they could freely choose which target they wanted to search and discriminate. Costs of saccades were manipulated by, in each trial, varying the relative number of distractors that shared a feature with one or the other target. Since the available time to complete trials was limited, participants had to trade-off search costs (i.e., how much time to invest to find a target) against informational value (i.e., searching for an easy- or difficult-to-discriminate target) to optimize target selection, and their individual monetary gain per unit of time. We found that participants generally considered both informational value and search costs, when choosing which target to search and discriminate. However, participants accumulated less reward than was predicted by an ideal observer model that assumed an optimal trade-off between informational value and search costs for target selection. A generative probabilistic model revealed that this was due to noise at two levels: noise at the decision level corrupted which target participants chose to search for, resulting in occasional choices for lower gain targets, while noise at the fixation level corrupted target selection during search, resulting in occasional fixations on elements that were not instrumental to find the eventually chosen target quickly. Thus, although participants generally traded-off informational value and search costs for target selection, their final performance was constrained by noise in decisions about what to search for and how to search for it.
Study II tested if informational value can facilitate the critical time-window in which the oculomotor system evaluates eye movement accuracy (Bahcall & Kowler, 2000; Fujita et al., 2002; Shafer et al., 2000). Participants were asked to make a saccade towards a neutral eye movement target. Upon saccade onset, two differently oriented stimuli were presented at opposite locations near the eye movement target, and participants were instructed that they have to report the orientation of one of the stimuli at the end of the trial. Critically, participants did not know upon saccade offset which stimulus was task-relevant (i.e., had higher informational value), but this was only revealed several seconds after saccade termination. Thus, eye movement accuracy (i.e., did the saccade accurately guide the fovea to a location close to the task-relevant visual information?) could not be evaluated upon saccade offset, but only based on visual working memory representation of the two stimuli, several seconds after saccade termination. We observed that informational value allowed for a delayed evaluation of eye movement accuracy, resulting in systematic trial-to-trial changes in the saccade amplitude that compensated for the experienced movement error relative to the task-relevant stimulus. This effect could not be explained by a lingering attentional bias, and disappeared when the informational value of stimuli was set to zero, i.e., when participants were not required to report their orientation. Thus, saccade adaptation considered the relative informational value of stimuli to select a target for saccade adaptation on a trial-to-trial basis, and informational value allowed for an evaluation of eye movement accuracy even long after saccade offset.
Study III tested how eye movement accuracy is maintained in dynamic environments, where two stimuli with varying informational value, whose onset was separated in time, generated two conflicting error signals about movement accuracy. Participants were instructed to make a saccade towards a neutral eye movement target (a small square). Upon saccade onset, two candidate stimuli were shown at opposite locations near the eye movement target. Critically, the candidate stimuli were not shown at the same time, but their onset was separated by a temporal delay. Depending on the condition, participants were instructed either to discriminate a feature on the first candidate stimulus, whose error signal was available upon saccade offset, or on the second candidate stimulus, whose error signal became available only sometime after saccade termination. We found that saccade adaptation obligatorily evaluated movement accuracy based on the first available error signal after a saccade, irrespective of the informational value of the corresponding stimulus. This obligatory effect of the first available error signal disappeared when we, in a separate experiment, used a large circle instead of a small square as saccade target. However, although a large eye movement target allowed for greater flexibility when evaluating eye movement accuracy, it did not always allow for lasting saccade adaptation towards the location of task-relevant visual information, and it promoted a stronger contribution of an explicit targeting process, deployed voluntarily to bring gaze close to the location of task-relevant visual information. Thus, properties of the visual environment not only modulate whether saccade adaptation does or does not consider the relative informational value of two competing error signals when selecting a target for saccade adaptation, but environmental properties also influence how experienced movement errors are corrected.
In sum, the present dissertation demonstrates that the oculomotor system considers the magnitude of task-relevant visual information at the eye movement target for target selection during visual search (Study I), and for selection of a target against which eye movement accuracy is evaluated (Studies II and III). This influence of informational value was observed on a trial-to-trial basis (Studies I and II), and on a longer time scale (Study III). However, the influence of informational value was not obligatory. Instead, the present dissertation shows that the oculomotor system aims to balance to benefits of saccades, in regard to the information that can be gained at the eye movement target, against the temporal costs of saccades (Study I), and it, furthermore, demonstrates that properties of visual environments modulate whether the relative informational value of stimuli is considered for evaluation of eye movement accuracy (Study III).