Western Austronesian languages have typologically rare but theoretically important voice systems that raise many questions about their learnability. While these languages have been featured prominently in the descriptive and typological literature, data on acquisition is sparse. In the current paper, we report on a variationist analysis of Tagalog child-directed speech using a newly collected corpus of caregiver-child interaction. We determined the constraints that condition voice use, voice selection, argument position, and thematic role assignment, thus providing the first quantitative analysis of verb argument structure variation in the language. We also examined whether children are sensitive to the constraints on variability. Our analyses showed that, despite the diversity of structures that children have to learn under Tagalog’s voice system, there are unique factors that strongly predict the speakers’ choice between the voice and word order alternations, with children’s choices related to structure alternations being similar to what is available in their input. The results thus suggest that input distributions provide many cues to the acquisition of the Tagalog voice system, making it eminently learnable despite its apparent complexity.