Across studies, child-directed, but not overheard, speech predicts children's later vocabularies in the first couple years of life. That speech appears systematically different from speech directed to other adults, though at some point the two types of speech must become equivalent. Adults alter their speech to children, but in what way, how calibrated their alterations are, and how they relate to learning, is unclear. To better understand the advantage of child-directed over adult-directed speech for learning, we use existing natural language corpora to capture the trajectories of multiple text-based metrics of complexity meaningfully related to learning, and discuss methods to capture aspects of speech complexity impossible to capture from transcripts alone.