Abstract
Selective attention can be biased towards particular locations or simple visual features (e.g., color) that have frequently been selected in the past, an effect called probability learning. Does probability learning operate based only on simple visual features or can it also influence attentional allocation to more complex and naturalistic stimuli, like real-world objects? We tested this in an experiment (N=27) in which participants were asked to find a real-world object that was either upright or rotated 180 degrees (upside-down) amongst other distractor objects that were tilted 45 degrees to either the left or right. All items were grayscale photographs of individual realistic objects varying in category and selected to have a canonical upright orientation. Each trial contained the same eight items, with the target item varying in identity from trial to trial. One of the eight items was more often the target. During a 348-trial training phase, this ‘rich’ object appeared as the target 50% of the time, while each other object only appeared as the target 7% of the time. A subsequent 174-trial testing phase presented each item as the target equally often. Participants were considerably faster at finding the target-rich object in both the training phase [mean advantage for target-rich objects = 246ms, t(26)=5.66, p < .00001, Cohen’s d = 1.09] and the testing phase [mean advantage = 121ms, t(26)=4.80, p < .0001, d = 0.93]. These results were replicated in a group of different participants (N=20), using a different set of 8 real-world objects. Based on these results, we conclude that long-term probability learning can affect search performance for real-world objects and not just simple visual stimuli.