What counts as effective input for language development? Decades of research have focused on features of child-directed speech in dyadic interactions, despite observations that such interactions represent exceptions to children’s language-learning environments worldwide. This focus may partly reflect our tendency to frame language learners as passive recipients of adult input. Here, I explore how reframing the child as a self-directed learner might give us purchase on the question of how the language they are exposed to relates to the language knowledge they develop. In particular, I will present work that draws on experimental, computational, and observational methodologies to understand the circumstances for children’s learning from overheard speech---a form of input common within the U.S. and across the globe---which we might consider an especially ecologically valid demonstration of children’s self-directed learning capacities. The experiments I will primarily discuss seek to address conflicting evidence in the literature for the utility of overheard speech: while toddlers readily learn from overheard speech in lab (e.g, in studies by Akhtar and colleagues), longitudinal studies consistently find no correlation between the overheard speech available to children in their daily environments and their later vocabularies (e.g., Shneidman & Goldin-Meadow, 2012). Pilot cross-cultural work and future directions will be discussed.