Research

My research focuses on phonetic detail within phonology and how it fits into representations, developing a phonological model for what that representation looks like and how it can account for observed synchronic and diachronic patterns.

To probe this model, I look at phonetic correlates of phonological contrasts and other predictors in production and test how they influence perception and how they can shift based on experimental manipulations. Many of these patterns provide an additional line of evidence establishing pathways of diachronic developments.

Phonetics in Phonology

I am developing a connectionist phonological model to account for the representation of phonetic details and the resulting patterns in speech perception and production, based on what factors are connected to phonetic details and which phonetic details are specified in that connection. This model can help explain both synchronic patterns and patterns of change.

Phonetic details in the phonological representation are linked not just with individual sounds but also connected to sounds with shared features. I demonstrate that shifts are generalized at the featural level: Exposing listeners to raised or lowered F1 or F2 in one vowel produces corresponding shifts in vowels that match in height or backness, respectively.
I also test how listeners normalize for different talkers' vowel spaces based on exposure to the same absolute formant value in vowels produced by different speakers. Listeners extend shifted formants across speakers; their subsequent perception of a novel talker reflects the relative formants of exposure stimuli based the training talker's apparent vowel space.

Factors beyond phonological contrasts also exhibit associated phonetic details, including emotional valence and part of speech. In an ongoing study, I show that longer vowel duration increases listeners' perception that a stimulus item is a noun rather than a verb, both for real words and for nonce words. The existence of the effect among nonce words cannot be explained by phonetic details associated with individual words. After accounting for factors such as part of speech, there is no clear evidence that a model of the phonological representation requires phonetic details to be linked to individual lexical items rather than being mediated by other factors. Speakers have a limited ability to learn new word-specific phonetic details; exposure to stimuli manipulated in opposing directions in different words does not provide evidence for word-specific shifts based on the phonetic details of the exposure stimuli. Instead, recent exposure to a word broadens the acceptable phonetic realizations of the phonological categories within that word; higher activation levels result in those words reaching full activation based on weaker phonetic matches.

Evidence for Sound Changes

Experimental data, in combination with traditional comparative methods, can elucidate questions about sound change, and in turn, patterns of sound change can further our understanding of biases in phonological interpretation of phonetic input. There are broad parallels between diachronic developments and synchronic patterns of perception and production, which make experimental data a useful source of complementary evidence for reconstructions.

One pattern that I have investigated is voicing-conditioned vowel duration; some correlates of coda voicing -- spectral tilt and intensity contour -- provide a likely pathway for how voicing-conditioned vowel duration develops. I examine how these vowel characteristics are influenced by coda voicing and how these characteristics in turn influence perceived vowel duration. While listeners compensate for the duration expected based on the coda, there are different effects when codas are removed. Vowels that had been produced with voiced codas are perceived as longer than vowels produced with voiceless codas.

I also use synchronic parallels to help test specific reconstructions, which is particularly useful for changes without many clear reconstructed parallels, such as the developments of the Proto-Indo-European laryngeals, which I investigated in my dissertation. I have also demonstrated a parallel for the sound change proposed in Winter's Law, with vowel lengthening before voiced unaspirated stops and not before voiced aspirated stops. While Hindi is often taken as representative of how voiced aspirated stops influence vowel duration, I demonstrate that Telugu differs from Hindi and provides a duration pattern that parallels Winter's Law. The different realizations of voiced aspirated stops in Hindi and Telugu further help illustrate crucial phonetic differences in sounds that are often treated as equivalent.

Language Documentation

I also contribute to a research group working on A'ingae, a South American language isolate; we have produced the first instrumentally based overview of its phonological system and continue to elucidate both areally motivated characteristics of its phonology and unique characteristics, such as narrative use of syllable-level falsetto.

As part of my collaboration with the A'ingae fieldwork project, we have been working on reconstructing phonological changes. Although A'ingae has no known linguistic relatives, synchronic allomorphy and limited distributions of sounds suggest several sound changes that substantially restructured the system, including development of contrastive prenasalized stops and development of vowel nasality.

Phonetics and Effects of Methodology

Because of restrictions on travel and in-person research during the Covid-19 pandemic, many researchers have started making remote recordings, using a range of methods. It is important to know how these methods might affect acoustic measurements, particularly for underdocumented languages, but also in any phonetic research. Variation might arise either because of different recording devices or because of the software used to make recordings. In a collaborative project, I led the Yale fieldwork group in investigating how varied recording methods impact phonetic measurements. Different recording devices and software programs have a substantial impact on many of the measurements that we tested, including measurements of duration, frequency, and aperiodic noise.

I have also examined how commonly used headphone checks impact results in online perception experiments. Headphone checks exclude many participants, which alters participant demographics such as gender, age, and geographic region. However, these exclusions do not increase effect sizes for most characteristics; for many perception studies, there is little evidence that headphone checks provide any benefit.

In addition to effects of how data is collected, methods of analysis can impact results. In a collaborative project, I demonstrated that some apparent individual differences in phonetic convergence are likely to be an artifact of the frequently used difference-in-difference method for measuring convergence.