Research

*As of September 2022, my career has shifted exclusively to teaching. I am employed at the University of Wisconsin–Madison.*

I study phonetics. My primary focus is vowel inherent spectral change, which examines how the acoustics of a given vowel change in systematic ways throughout its duration. One of my goals is to connect this line of research with work on modularity and levels of representation – the idea that a sound is represented in the mind in several ways. But primarily I seek to inform sociophonetic methodology.

I administer reading tasks consisting of dozens of repetitions of English's vowels in order to compute several detailed formant contours for each vowel.

I also conduct listening experiments using computer-synthesized vowel stimuli in order to test people's sensitivity to fine-grained acoustics reported in the production literature – to test how much of the acoustic changes which are measurable are also perceptible.

Vowel modeling

Vowel inherent spectral change (VISC) argues that vowels are best modeled dynamically. The best-supported perspective is the dual-target model, consisting of onset and offset formant values (i.e. F1 and F2 sampled at 20% and 80% of duration). But there has also been evidence that vowels consist of onset formant values plus a contour of specified direction but not specified length or spectral slope. Several of my projects explore this issue. One study directly tested the major models of VISC using an experiment proposed by Geoffrey Stewart Morrison. Another study replicates work by Amy Neel with robust phonetic inputs for the stimuli.

Self-correction, vowel contours, and cue weighting

As mentioned above, VISC has found support for the importance of vowel dynamics as important, even without midpoint specification. But recent work has shown that speakers self-correct deviant vowel tokens, holding that a speaker's midpoint formant values are the phonetic target of vowels. This would suggest that vowel dynamics are secondary to midpoint targets (since the contour would change based on an utterance's initial values in relation to the average midpoint). I seek to reconcile these lines of research using highly repetitive production data. I am testing the hypothesis that self-correction is more about 'getting back on track' than about 'hitting the mark'. I am also probing whether dips in formant variability at specific time points reflect more important cues to vowel identity.

Interrogating Pillai scores

The Pillai score is an output of MANOVA and MANCOVA models, and it indicates the degree of overlap of two multivariate distributions. It has increasingly been used as a standardized metric for vowel merger, but no work so far has investigated its veridicality empirically. I compare individual differences in speakers' Pillai scorse for a given merger in production to their ability to detect the vowels as distinct in perception. I also do model selection with Pillai scores that incorporate dynamic spectral information in a variety of ways to identify the optimal model.

Sociophonetics

Vowel dynamics are becoming increasingly important to sociophonetic research. The vowels in two dialects can differ in their position in formant space, but they can also differ in the shape of their trajectories. One of my projects seeks to tease apart the relative importance those cues with synthesized vowel stimuli. Another focuses on dialect classification and asks whether (and when) more detailed formant contours aid listeners in the task. These studies help us understand the detail of indexical representations, complementing my other research on the detail of phonemic representations.