An adaptive function of the visual system is that it can flexibly update existing representations of objects upon changes in the environment. Moreover, these changes can alter the representations of other associated objects that are not directly visible. For example, the increasing size of headlights at night signals an approaching car, although the car may not be visible. What mechanism supports such inference? We propose that statistical learning provides a channel through which new information about one object can be transferred to related objects. Observers viewed a continuous sequence of circles grouped into color pairs (e.g., red always appeared before blue). Afterwards, the first circle in each pair increased or decreased in size. Observers recalled either the size of the second circle in the pair, or the size of a random circle that never followed the first one. We found that the size of the second circle was judged to be larger (or smaller) than the random circle if the first circle increased (or decreased) in size (Experiment 1). This suggests that changes in one object are automatically transferred to the object that previously reliably followed. This transfer may be facilitated by the fact that the first circle predicted the second circle, or the mere association between the two circles. To tease these ideas apart, in Experiment 2 the second circle increased or decreased in size, and observers recalled the size of the first circle, or a random circle. We found no difference between the judged size of the first circle and the random circle, suggesting that changes in one object are transferred to the predicted object, but not vice versa. No observer was explicitly aware of the color pairs. Thus, statistical learning implicitly and automatically updates the representation of objects upon changes to other objects via temporal prediction.
展开▼
