What aspects of infants' prelinguistic communication are most valuable for learning to speak, and why? We test whether early vocalizations and gestures drive the transition to word use because, in addition to indicating motoric readiness, they (a) are early instances of intentional communication and (b) elicit verbal responses from caregivers. In study 1, 11 month olds (N = 134) were observed to coordinate vocalizations and gestures with gaze to their caregiver's face at above chance rates, indicating that they are plausibly intentionally communicative. Study 2 tested whether those infant communicative acts that were gaze-coordinated best predicted later expressive vocabulary. We report a novel procedure for predicting vocabulary via multi-model inference over a comprehensive set of infant behaviours produced at 11 and 12 months (n = 58). This makes it possible to establish the relative predictive value of different behaviours that are hierarchically organized by level of granularity. Gaze-coordinated vocalizations were the most valuable predictors of expressive vocabulary size up to 24 months. Study 3 established that caregivers were more likely to respond to gaze-coordinated behaviours. Moreover, the dyadic combination of infant gaze-coordinated vocalization and caregiver response was by far the best predictor of later vocabulary size. We conclude that practice with prelinguistic intentional communication facilitates the leap to symbol use. Learning is optimized when caregivers respond to intentional vocalizations with appropriate language.