An Attentionally Constrained Model of Statistical Word Learning

Abstract

Recent research supports the notion that word learning can be conceptualized as a statistical learning process. As many have noted however, statistical learning is constrained by processes such as attention and memory. Here we tested an attentionally constrained framework of statistical word learning. We observed, through infant-perspective head cameras, infants’ visual input as parents labeled novel objects during an infant-parent object-play session. We then constructed statistical learning models that aggregate word-to-object associations. We fed a baseline model the word-to-object co-occurrence patterns obtained from parent-infant observations. We fed an attentionally constrained model weighted co-occurrence patterns based on the perceptual properties of the objects (i.e., object sizes from the infant’s view) at the time words were uttered. Models’ learning was compared to children’s forced-choice test results. Of interest is which of the two models best approximates children’s learning. Implications of these results for statistical learning accounts of word learning will be discussed.


Back to Table of Contents