视觉的统计学习

Visual statistical learning is the automatic, implicit acquisition of regularities in the visual environment—patterns in how objects appear together in space and time. Humans learn these statistical structures quickly and without conscious awareness. For example, a coffee cup frequently appears near a computer, creating predictable associations that support object recognition and visual processing.

The incidental encoding paradigm exposes participants to nonsense object triplets displayed sequentially for 250 milliseconds each. A cover task—asking participants to respond to gray objects and withhold responses to red ones—distracts from the underlying statistical structure. This unrelated task ensures learning occurs implicitly, without participants' conscious awareness of the object patterns.

Transitional probability measures the likelihood that one object will follow another in a sequence. In visual statistical learning, objects within a triplet always appear in the same order, creating a transitional probability of 1.0 between those elements. Unrelated objects have much lower transitional probabilities, establishing the statistical structure participants implicitly learn.

After the encoding phase, participants complete a familiarity task where they view pairs of triplets—previously seen triplets and newly generated foils—and indicate which looks more familiar. The dependent variable is the number of correct identifications of familiar triplets. If learning occurred, participants select familiar triplets significantly more often than the 50% chance rate.

Visual statistical learning is demonstrated when participants correctly identify familiar triplets at rates significantly above chance performance of 50%. In typical experiments, participants achieve approximately 70% accuracy on the familiarity task, indicating they have implicitly learned the statistical structure of the object sequences during the brief 10-minute encoding period.

The visual statistical learning paradigm translates to broader sensory learning mechanisms, including auditory domains. Infants and children use auditory statistics in early language formation, learning reliable statistical relationships between sounds and letters. Similarly, researchers have shown that letter-color associations can be implicitly learned through exposure to customized texts with distinctly colored letters.

A cover task is essential to ensure learning occurs implicitly, without participants' conscious awareness of the statistical structure. By directing attention to an unrelated task—such as responding to colored objects—researchers prevent participants from deliberately searching for or consciously encoding the object sequences, thereby measuring true implicit learning mechanisms.