Modeling the Relationship among Linguistic Typological Features with Hierarchical Dirichlet Process

We propose that topic models can be used to represent the relationship among linguistic typological features. Typological features are typically analyzed in terms of universal implications. We argue that topic models can better capture some phenomena, such as universal tendencies, which are hard to be explained by implications. We conduct experiments to evaluate the predictive accuracy of our Hierarchical Dirichlet Process (HDP) model on the WALS dataset. We discover some interesting findings. Topics regarding word order types are recognized. We also find a topic that regards areal tendency.