In this work, we examine a context in which a conflict arises between two roles that infant-directed speech (IDS) plays: making language structure salient and modeling the adult form of a language. Vowel devoicing in fluent adult Japanese creates violations of the canonical Japanese consonant-vowel word structure pattern by systematically devoicing particular vowels, yielding surface consonant clusters. We measured vowel devoicing rates in a corpus of infant- and adult-directed Japanese speech, for both read and spontaneous speech, and found that the mothers in our study preserve the fluent adult form of the language and mask underlying phonological structure by devoicing vowels in infant-directed speech at virtually the same rates as those for adult-directed speech. The results highlight the complex interrelationships among the modifications to adult speech that comprise infant-directed speech, and that form the input from which infants begin to build the eventual mature form of their native language.