PATHWAYS TO LEARNING OBJECT NAMES THROUGH EVERYDAY EXPERIENCE WITH VISUAL OBJECTS

Abstract

To learn an object name, infants must segment the visual features of an object from the surrounding world and form a memory of that object. Infants must also parse the name from the flow of the speech stream and form a memory of that name. Finally, infants must map the name to referent in memory. The field has limited understanding of the learning mechanisms through which infants form and retain mappings before the first birthday, though the evidence shows that they do. These mechanisms cannot be determined without defining the properties of infants’ everyday environments which specify the real-world learning task that any putative mechanism must solve. The visual side of the learning problem has been understudied in investigations of the natural data for learning. In this dissertation, I characterized infants’ everyday learning environments—particularly the visual data—thus raising new proposals as to learning mechanisms. The studies presented in this dissertation employed data collected from infants wearing head-mounted cameras at home. In Study 1, we measured the frequency of visual objects and found a right skewed frequency distribution in which a small set of object categories with early-learned names were repeatedly present and a larger set of objects were rare. In Study 2, we measured the frequency of individual category members and found that a single focal exemplar was overrepresented in infants’ experience with each early-learned category examined. In Study 3, we measured the frequency of object names and their co-occurrence with referents and found that whereas visual objects appeared repeatedly across events, spoken names did not. Collectively, these findings implicate a memory mechanism not previously considered in the study of word learning in which memory representations of objects—slowly acquired through repeated visual experience—may serve as anchors such that infants can learn object-name mappings from limited co-occurrences.