Each language has its own phonotactics– rules or constraints that govern the language’s possible syllables. We implicitly know that “hing” is a possible English syllable, but “ngih” is not. Phonotactic knowledge guides language production, as revealed in speech errors. Even though a slip is an error, it nonetheless tends to be phonotactically legal. When people produce sets of syllables that follow artificial phonotactic constraints (e.g. a “rule” that the consonant “f” is not allowed at the end of a syllable) they implicitly learn those artificial constraints. The learning is revealed in their speech errors (e.g. their slips would rarely put an “f” at the end of a syllable). I present a new theory of this learning process. It attempts to explain how speech errors arise during production, how this production process changes with experience. Two particularly interesting aspects of this learning are (1) the need for consolidation of learning for some rules and (2) constraints on learnability that arise from the speaker's prior linguistic experience.