Different results for a statistical learning task depending on a different statistical learning task

What is it about?

Statistical learning refers to the ability to learn from repeating patterns in the world. It is often applied to language, and in particular, to aspects of language that a learner hears. In this study, four-year-old children heard patterns of consonants, like the /mk/ in 'tomcat'. They heard those consonants in nonsense words like "dimmkes" and "sammkef". In previous studies, we found that four-year-olds can produce /mk/ more accurately in a word like "foamkep" if they first hear lots of examples of /mk/ in different words. Here, we generally replicated that previous finding, but we also observed something unexpected. Children completed two experiments: this one focused on consonant sequences as well as a different statistical learning experiment that focused on prosody, or patterns of stressed and unstressed syllables. When children completed the prosody experiment first, they actually were significantly less accurate to produce /mk/ if they heard lots of examples of /mk/. We interpret the finding as evidence that learning from the two experiments interacted.

Featured Image

Why is it important?

Statistical learning generally assumes that human beings do a good job of learning patterns in the world and, when appropriate, applying those patterns to their own behavior. If our interpretation of the results is correct, and our four-year-olds conflated patterns from the two different experiments, it calls into question the assumption that humans are good at knowing when to apply a learned pattern. In fact,

Perspectives