Abstract
Although massive amounts of information constantly flood our senses, we seem to effortlessly direct attention to relevant information and ignore information that may distract us. A surge of recent studies has demonstrated that this efficient selection to a large extent relies on the extraction of regularities (e.g., within a kitchen, a pan is more likely to be found on the stove than on the ground), often referred to as statistical learning. While it is clear that objects and their parts do not appear at random locations within scenes, it is also plausible that within a given object one can also learn which location is relevant and which is not. In a set of experiments, we investigated this within-object statistical learning by (implicitly) learning participants that particular locations within an object were more likely to contain relevant information than other object locations. We show that learned within object prioritization not only exists, but once established also remains in place even when the object itself was presented at a completely different location within visual space, in which learning never took place. Strikingly, whereas space-based statistical learning typically arises very rapidly and then remains stable, object-based statistical learning gradually emerged over the course of the experiment. Moreover, during early stages of learning learned priority appeared to generalize to other neutral objects without a spatial bias, and only became specific to the biased object at later stages of statistical learning. No such dissociation was observed when the participants were explicitly informed about the regularity. Together our findings demonstrate that learned attentional enhancement is not exclusively tied to the retinotopic location where learning took place, but instead is tied to specific locations within objects independent of spatial appearance.