Taking Conjunction Search to a Higher Dimension

Abstract

In visual search for conjunctions, targets are defined by combinations of features. Thus, targets might be RED VERTICAL items among RED horizontal and green VERTICAL distractors. Guided Search and similar models argue that features are conjoined only once an object is attended. However, efficient conjunction search is possible because observers can guide attention toward red and toward vertical. Real-world objects may be defined by the conjunction of many features, with no one feature defining the target. Can attention be guided to higher order conjunctions? We created 6D conjunction stimuli and varied the number of features shared between targets and distractors and the number of distractor types. RT x set size slopes range from 1.4 msec/item when two distractor types each share three features with the target to 66 msec/item when six distractor types each share five features with the target. Using different 3D conjunctions, we held distributions of features constant (1/3 of each of three colors, shapes, and orientations). We varied the grouping of distractors in order to look for evidence of simultaneous guidance by multiple attributes versus sequential grouping of subsets (Find the odd orientation in this color subset). We find that guidance and grouping may both contribute to efficient conjunction search.