Abstract
Understanding the relative contributions of goal-directed and stimulus-responsive attention is a critical problem in visual cognition. To pit the two processes against one another, we develop an experiment where both are elicited by the task structure and the stimulus set, respectively. Specifically, we aim uncover the relative influence of salient, distracting information while participants learn how to sort visual stimuli into four different categories using less salient, but informative features of the stimulus. In this regard, we can examine both learned goal-directed attention and reflexive, stimulus-responsive attention to salient distractors simultaneously, thereby examining their interactions over learning.
The 3 informative features of the stimulus are learned through trial-and-error. The salience condition presents irrelevant features that act as salient distractors; and the irrelevant features in the baseline condition are all equally non-salient. In contrast with predictions from theories of salience in visual attention, we find that those in the salient condition display more efficient eye movements by minimizing fixations to unimportant features relative to the baseline condition as measured by an optimization score (Blair, Watson & Meier, 2009) that reflects the proportion of fixations allocated to informative versus irrelevant features. Findings suggest that theories predicting eye movements from bottom-up information do not wholly account for oculo-motor activity during category learning problems of this nature. However, salience must enact some influence on the programming of saccades, otherwise both conditions would exhibit a similar optimization score. Our interpretations of these data are: 1) in the short duration preceding a saccade, goal-directed and stimulus-responsive attention both act to influence the selected saccade target; and 2) hybrid theories of goal-directed/stimulus-responsive processes provide the most plausible account of attention as it is observed in eye movements during a learning task.
Meeting abstract presented at VSS 2013