From Feature to Dimension: Generalization of Value-Driven Attentional Capture into a Multidimensional World

Abstract

Reward can help teach selective attention what is relevant in a complex world. Previousstudies have demonstrated that specific features consistently associated with high reward capture attention more than those associated with low reward. How reward influences attentional capture at the level of entire feature dimensions remains unclear. In this thesis, I test the hypothesis that learning to attend to highly rewarding features in one dimension generalizes to attentional capture by unrewarded features within that same dimension. I used two variants of a behavioral paradigm named the Token Task. In this task, participants learned to associate certain features from the dimensions of color and orientation with high reward, and others with low reward. Participants were randomly assigned to either a “color” or an “orientation” group, and all highly rewarded features were drawn from the corresponding dimension. To probe attention, I occasionally interrupted the learning task with a visual search task. Each search array contained both a color singleton and an orientation singleton, one of which was randomly chosen each trial to be the search target. I predicted that a learned dimensional attention bias would facilitate pop out and thus search for singleton targets in the dimension containing the highly rewarded feature. The findings suggest that reward can drive attentional capture at the dimensional level, at least for color. Additionally, the findings provide evidence of separate learning styles during the task that can differentially drive attentional capture. An experimental replication with variation demonstrated that the design of the visual search array is crucial to detecting an effect of dimensional attention.