Human language is a tool for communication, a medium of expression, and a mental scaffolding that organizes our thoughts in particular ways. But it is also something which is learned, and the surest thing that language scientists really know about how a first language is acquired is that our cultural myths about how learning happens—like the idea that parents and teachers somehow "instruct" children to speak—are basically completely wrong. The ultimate goal of theoretical linguistics is to give a precise, mathematical characterization of how this unconscious cognitive change—from an oblivious infant to the master of a system connecting sounds to meanings that's so intricate we still barely understand it—really works. I use Bayesian statistics as a formal device for capturing how the brain reasons under uncertainty, to look at learning in the areas of phonetics (how people pronounce and perceive words) and phonology (how those pronunciations are stored and processed in memory). By formalizing learning as reasoning under uncertainty, I believe we are moving substantially closer to the goal of having a precise characterization of the mysterious and changing mental landscapes of a child learning a first language.