Distributional Learning in Context: How Social Embedding Structures Infant-Directed Speech

Infants and toddlers typically hear words accompanied by a variety of direct and indirect cues to their meaning. To name just a few, words are embedded in frequently repeated linguistic constructions, they tend to co-occur with specific objects that they refer to, and they tend to be used in different social-interaction routines and activity contexts. Whereas children are capable of detecting several different types of cues and using them to facilitate word learning, we are only beginning to uncover the developmental processes by which words come to be embedded in multimodal, dynamic contexts that mark them as items to be learned and help children to discover their meaning.In this dissertation I address two broad questions. First, how do infants and caregivers co-construct interaction sequences in which words are accompanied with useful cues? In a series of observational studies of infant-mother dyads observed longitudinally from the age of 4 months to 12 months, I describe how infants’ increasing motor abilities enable them to elicit contingent caregiver responses containing object-naming and predictable sequences of other informative utterances. Second, what can we learn about what matters for word learning by using the contextual distributions of words to predict how early and in what combinations they will be learned? In these studies, I use a corpus of child-directed speech to construct a representation of each word’s distribution over syntactic and thematic usage contexts. Then, using a large open dataset of children’s parent-reported word production and comprehension, I show that both types of context distribution contribute over and above previously described factors in predicting both the age of acquisition of words and the degree to which word pairs tended to be learned together versus randomly. Taken together, the studies in this dissertation support a view of early word learning in which (1) multiple layers of social, linguistic, and sensorimotor contextual cues jointly facilitate word learning, (2) infants learn to participate actively in the responsive interactions that produce high-quality word exposures, and (3) although these processes are too complex to be replicated with full experimental control, they leave identifiable traces in the structure of children’s lexicons