Recurrent changes in related languages which can not be attributed to chance, universals or diffusion, have been categorized as instances of drift. In this talk two robust cases of drift in the Austronesian language family are identified. In the first case, general syncope of short unstressed vowels in VC_CV contexts is common in certain subgroups, but absent in others. In the second case, syncope of unstressed vowels between identical consonants is common where general syncope is absent, and rare where general syncope is found. I argue that the most significant structural feature in predicting general syncope is the pre-existence of closed syllables. This analysis follows from a more general hypothesis regarding the role of sound patterns in determining sound change at the level of the individual language learner. The same general hypothesis may also be able to account for the distribution of geminate-producing syncope, once pre-existing segmental length contrasts are taken into consideration. The general model makes predictions which go beyond drift: general structure-preservation effects in sound change like that seen for syncope in Austronesian are associated with cases where ambiguity in phonological analysis of phonetic tokens is not subject to innate perceptual biases. To the extent that these predictions are upheld, this account of drift contributes to a general and restrictive theory of sound change.