Abstract
Previous studies showed that attention, eye movements and visual short-term memory operate on (partly) context-dependent representations of stimuli. Specifically: When observers have to search for a target with particular features (e.g., medium orange), attention is usually tuned to the relative size and colour (e.g., largest, reddest; ‘relational search’) rather than the physical features (e.g., medium, orange). Attention can also be tuned to the specific features of the target, but feature-specific search is more effortful and slower. Importantly, it is currently unknown whether information about relative features is derived from lower-level neurons that respond to specific features, or whether visual inputs are first encoded relationally, with feature-specific codes extracted later. The present study addressed this question using functional magnetic resonance imaging (fMRI) in a colour search task in which we enforced relational vs. feature-specific search. Our current findings support the first possibility, with inputs being first processed in a feature-specific manner, and later relationally: In V1, repetition suppression was most pronounced in the feature-specific condition, indicating that these neurons respond to specific feature values. In V2, repetition suppression was equally strong for both conditions, but in later areas (V3, parietal and frontal areas), the result reversed, with stronger repetition suppression for relational search. Surprisingly, these results were obtained even when both the target and nontarget colours changed on a trial-by-trial-basis in relational search, and only the nontarget colour in feature-specific search. These findings show that repetition suppression is not always tightly linked to repetitions of the stimulus input, but can depend on top-down search goals, especially during later processing stages. Moreover, while V1 seems to respond to specific features, relational information is apparently derived as early as V3, and dominates throughout the visual processing hierarchy. This dominance may explain why relational search is more efficient and generally preferred to feature-specific search.
Acknowledgement: Australian Research Council (ARC)