The lexical mapping of abstract functional words like modal verbs is an open problem in acquisition (e.g., Gleitman et al. 2005). In diachronic linguistics it has been proposed that learner mapping errors are responsible for innovations in the historical record (see Kiparsky 1974; Roberts & Roussou 2003, among others). This suggests that child error patterns should be consistent with historical changes. I studied the acquisition of modal lexemes by flavor (e.g., ability, epistemic) in order to assess the validity of this proposal in relation to the mapping problem. A preference task and a sentence-repair task were designed to address the question: Do children make structural mapping errors that, if left unchecked, are compatible with the innovations we see in the historical record (e.g., deontic > epistemic)? This study provides experimental data on the acquisition of modal lexemes by flavor and some long-awaited preliminary support for the hypothesis that child learners drive historical change.