Discovering place and manner features—What can be learned from acoustic and articulatory data?

Theories of phonological acquisition often assume that features are innate and universal, and establishing distinctive features requires the availability of minimal pairs. In this study, we explore an inductive approach to feature discovery without resorting to minimal pairs. Our basic assumption is that learners have access to a wide variety of phonetic information, and so feature induction can proceed through recursive clustering of the data. To test our hypothesis, we used as our data acoustic and articulatory measurements made simultaneously from a set of CVC syllables. The articulatory data are obtained with ultrasound imaging, and the vocal tract is characterized by a vector of cross distances from the tongue body to a reconstructed palate. The raw articulatory data are then mapped to a lower dimensional space using principle components, on which clustering is carried out. Experiments using acoustic and articulatory data suggest that these two sources of information lead to the discovery of place an...