Statistical learning: From acquiring specific items to forming general rules

Abstract

Statistical learning is a rapid and robust mechanism that enables adults and infants to extract patterns of stimulation embedded in both language and visual domains. Importantly, statistical learning operates implicitly, without instruction, through mere exposure to a set of input stimuli. However, much of what learners must acquire about a structured domain consists of principles or rules that can be applied to novel inputs. Although it has been claimed that statistical learning and rule learning are separate mechanisms, here we review evidence and provide a unifying perspective that argues for a single mechanism of statistical learning that accounts for both the learning of the input stimuli and the generalization to novel instances. The balance between instance-learning and generalization is based on two factors: the strength of perceptual biases that highlight structural regularities, and the consistency of unique versus overlapping contexts in the input.

Keywords: generalization; infants; rule learning; statistical learning.

Figure 1

The design of Saffran, Aslin, and Newport (1996).

Figure 2

Distributions of (A) transitional probabilities (from Saffran et al., 1996) and (B) phonetic tokens (from Maye et al., 2002). The blue distribution in (B) is unimodal and the red distribution is bimodal.

Figure 3

The design of Marcus et al. (1999) and the two sets of 4 words (red column and blue diagonal) used by Gerken (2006).