Márton Sóskuthy (SocioPhonAus3 Keynote Speaker)
Thursday, December 15th, 2022, 10.00am – 11:00am
Vowel dynamics refers to time-varying patterns in vowel production and perception. In recent decades, there has been increasing interest in this area, driven partly by methodological and computational advances that make it possible to model and visualise complex dynamic patterns in large data sets. Despite these developments, the study of vowel dynamics is currently hampered by two issues: (1) a lack of theory-building and (2) an unduly heavy emphasis on hypothesis testing to the detriment of exploratory research. These issues are unpacked in more detail below.
(1) Much work in vowel dynamics is presented as a counterpoint to approaches that aim to reduce vowels to a single static target (e.g. specifying vowels through a steady-state portion). However, the details of alternative, dynamic models are often unclear: in what way would a dynamically specified vowel look different from a statically specified one? This is further exacerbated by the confounding effects of coarticulation, prosody and biomechanics on vowel trajectory shapes. As a result, it is often difficult to evaluate the validity of dynamic approaches.
(2) Like many other fields, the study of vowel dynamics is biased towards confirmatory hypothesis testing (as opposed to exploratory research). Methods for hypothesis testing typically only offer an artificially narrow view of variation in dynamics. As a result, we are still not in a position to answer basic questions about the envelope of variation in vowel dynamics such as the following: Are there patterns of interest beyond midpoints and onset / offset values? What are the constraints on the shape of vowel trajectories? Not knowing the answers to these questions makes it difficult to build explanatory theories of vowel dynamics.
In this talk, I argue that we need to use exploratory methods to build a broader evidential foundation for the study of vowel dynamics, and to develop explicit models based on this foundation; hypothesis testing can only begin once these two are in place. I use data from Derby English and New Zealand English to illustrate the issues outlined above and some potential solutions. I also aim to showcase a range of different statistical methods such as parametric curve analysis, GAMMs, and functional PCA that can provide a fuller picture of dynamic patterns when used together.