Abstract
High-level features of objects, such as semantic information, have been shown to bias attention, even when task-irrelevant. However, it remains unclear the exact mechanism by which this attentional guidance is instantiated. We hypothesized that task-irrelevant semantic information organizes visual input through mechanisms of grouping. Similar to grouping by similarity in low-level features, we predict that semantic information organizes visual input by semantic relatedness. Specifically, when presented with multiple task-irrelevant objects, attention is guided, or prioritized, to a subset of objects that are semantically related, creating a grouping-like effect. In the present studies, participants were presented with an array of 4 or 6 objects. The objects were either colored squares (low-level information only) or grayscale real-world objects (high-level information only). On any given trial half of the objects were related to a single category (e.g., clothing or blue squares) while the other half was chosen randomly from other semantic or color categories, respectively. A target was presented randomly on one of the objects, independent of relatedness, rendering color and semantics task-irrelevant for each experiment. For both colored squares and real-world objects, when each group had equal number of members (grouped by color or by semantic relatedness), target identification was faster and more accurate when targets were presented on the color- or semantically-related objects. Interestingly, when the size of the related group increased (e.g., three related objects and one non related object in set size four) performance was slower for targets presented on the objects that shared membership in the larger group versus objects in the smaller group. Taken together, these results support the semantic grouping hypothesis such that semantic information, just as color, organizes visual input to enable efficient attentional allocation. Importantly, given task-irrelevant nature of semantic information, our results suggest that task-irrelevant semantic grouping is an automatic process.