Proceedings of the Annual Meeting of the Cognitive Science Society

There is a strong relationship between book-rich environments and vocabulary size in early childhood, and for preschoolers, storybook sharing remains an important source of new vocabulary. In this work we ask, what makes a story an effective tool for word learning and vocabulary enrichment? We use data from 49 preschoolers sharing AI-generated picture books with a caregiver, with each set of stories containing 20 target words (nouns, verbs, and adjectives), to assess the feasibility of using both linguistic and visual contextual informativeness metrics to evaluate the quality of stories as support for learning new vocabulary. Results show that contextual informativeness impacts learning differently across word types: visual metrics and linguistic ground truth measures both correlate with learning for nouns and verbs (but not adjectives), and our automated metric for approximating linguistic informativeness shows significant predictive performance specifically for verbs. These results speak to the importance of considering different sources of information—including linguistic and visual information—when designing materials that support learning. We discuss these findings in the context of improving automatically generated child-directed stories to support the learning of a variety of different types of words.