ABSTRACT
This chapter provides a review of recent work on statistical learning in the auditory and visual modalities by adults, infants, and nonhuman primates. Evolution has likely constrained learning mechanisms to be sensitive to information that has particular relevance to a species, and not all species require the same information. The core of authors' approach is that once visual information is filtered by low-level analyzers and parsed into a set of base elements, mid-level visual mechanisms extract co-occurrence statistics from these base elements and form new features. Mid-level Gestalt principles could serve as yet another powerful constraint on statistical learning, further reducing the computational complexity of the task. A variety of constraints appear to buffer the developing learner from this computational problem, including the presence of low-level feature detectors that are independent of experience. Many of the constraints on behavior arise from learning; that is, relatively permanent alterations in systems of behavior that result from both general and specific experiences.
