Capturing the complete language environment: What could children learn by 'listening in'?

Inspired by qualitative studies typically limited to child-directed speech, we develop a coding scheme designed to characterize all utterances accessible to two English-learning children in terms of their relative utility for word-learning. We focus in particular on contributors to referential transparency as a well-established and meaningful dimension of language learnability in context. These include the spatial positions of the caregivers and children, the caregivers' use of gaze or gesture to illustrate their meaning, the child's visual access to the caregiver or to the referent of the utterance, and the caregiver's use of modified prosody. As a proof of concept, we apply this coding scheme to existing naturalistic video corpora for one English-learning child whose language development is well-documented. We find that both speech directed to the child and speech overhearable by her are highly variable along the qualitative dimensions we coded, and identify the heterogeneity of overheard speech as a source of noise in previous investigations. While irrelevant as a referential cue, our results suggest caregivers' prosodic modification may play a functional role in marking speech intended for the child — especially given the significant qualitative overlap between overhead and child-directed speech along other dimensions. In spite of the frequent similarity between overheard and child-directed speech, overheard utterances were significantly less associated with child attention. Taken together, our results shed light on how adults and children co-structure the early language environment, and promise to provide similar insights when applied to naturalistic video corpora for children across the world.