Acoustic analysis of /ti/ tokens produced in native words across a range of phonetic environments by two generations of Japanese speakers reveals a systematic influence of prosodic structure on the duration of frication following the release of consonant closure. The range of frication durations conditioned by prosodic structure in native words is partitioned into lexically contrastive sequences, ti~chi, in loanwords. Within speaker comparisons of words borrowed at different time periods suggests that this new contrast emerged, for some speakers, during adulthood. Implications of the data for a phonological theory of loanword adaptation are discussed.