Proceedings of the Annual Meeting of the Cognitive Science Society

Philippine-type Austronesian languages are well known for their rich voice systems, which include up to four voices that differ in which argument is the subject, rather than a simple active-passive contrast. These languages present a challenge to language acquisition theories, because they lack the robust relationship between causal agents and sentential subjects, a foundational assumption of both Nativist and Empiricist theories. We focus on one key issue: verbs differ in the voices they allow or prefer, raising questions about how children may acquire these patterns. We examine voice–semantics associations in three Austronesian languages spoken in Taiwan: Atayal, Paiwan, and Yami, using hierarchical Bayesian modeling to estimate probabilistic voice preferences for individual verbs while accounting for data sparsity. Hierarchical clustering of posterior distributions reveals overall biases in voice usage and partial alignment between semantic themes and voice-marking patterns, suggesting that voice-semantics distributions are systematic in ways learners may exploit during acquisition.