UC Berkeley

This thesis explores the role of uniformity of speech articulation in shaping phonological systems of contrast and their phonetic implementations. The observable effect of uniformity for an individual speaker is that a given phonological primitive (such as a distinctive feature value or gesture, depending on one's theoretical framework) tends to be implemented with maximum articulatory similarity across the speech sounds sharing that primitive. Although less discussed than other organizing principles in substance-based phonology such as phonetic dispersion (Liljencrants and Lindblom, 1972), focalization due to quantal effects (Stevens and Keyser, 1989; Schwartz et al., 1997b), or articulatory ease (Martinet, 1955; Lindblom, 1990), uniformity has been observed in a range of the world's languages, mainly in the timing of laryngeal articulations in stop inventories (Keating, 2003; Chodroff and Wilson, 2017) but also in place-of-articulation primitives (Maddieson, 1996; Chodroff, 2017).

However, uniformity has typically been formulated as a purely linguistic constraint. A primary aim of this dissertation is to motivate uniformity as emerging from domain-general biases that shape complex systems of goal-oriented action more broadly, thereby shedding light on the substantive basis and structure of phonological systems. To this end, I describe a model in which articulatory uniformity emerges from articulatory reuse during learning. During the language acquisition process, a learner's internal model (mapping the effects of motor controls applied to the speech articulators to their outcomes) is not yet fully developed. Under these conditions, a "model-free" learning strategy based on bootstrapping off of the learner's already-mastered skills (exploitation, rather than exploration) may predominate, such that phonological categories whose outputs are perceptually similar may come to be produced with the same articulatory primitives.

This thesis tests aspects of the model of uniformity-through-reuse with an experiment on Sūzhōu Chinese, whose fricative vowels are known to somewhat resemble alveolopalatal fricative consonants in their tongue-palate constriction patterns and fricative noise production targets (Ling, 2009). Ultrasound tongue imaging was used to characterize the typical fricative vowel and alveolopalatal fricative consonant productions of 43 Sūzhōu Chinese speakers.

Analysis reveals that most Sūzhōu Chinese speakers typically use a single tongue posture uniformly across the fricative vowels and consonants examined, while a minority of speakers deviate from uniformity to an idiosyncratic extent. The extent to which a speaker deviates from a uniform strategy is shown to be unrelated to demographic characteristics and language ability in Sūzhōu Chinese and Standard Chinese. This pattern at the population level suggests that "speaker-side" factors, such as articulatory reuse, are primarily responsible for shaping the "synchronic pool of variation" (Ohala, 1989) for this set of Sūzhōu Chinese segments.