Issues in Uyghur backness harmony: Corpus, experimental, and computational studies

This dissertation investigates backness harmony in Uyghur (Turkic: China) from a variety of methodological and analytical perspectives. Backness harmony is a phenomenon where suffix forms must agree in backness with the roots to which they are attached. This dissertation demonstrates that the harmony system in Uyghur consists of a productive phonological core with many lexicalized components that have emerged as the result of sound change and extensive borrowing. That is, backness harmony in Uyghur has many of the properties of an inflectional class system with strong phonological correlates, rather than a purely phonological phenomenon. I explore the implications of this observation for theories of phonological opacity, phonetic biases in phonological learning, and the mathematical complexity of phonological patterns. The first part of the dissertation presents corpus and experimental studies that investigate the role lexicalization plays in the backness harmony system. Chapter 3 looks at an opaque interaction between backness harmony and an independent process of vowel reduction. A large scale corpus study reveals that this opacity is of a type hitherto unattested, exhibiting gradient rates of opacity that are correlated with root frequency. I argue that this behavior is not predicted by standard theories that treat opacity as an ordering relationship between two phonological processes, and is best analyzed as a parallel interaction between general phonological markedness constraints on harmony and lexically-indexed constraints that specify the harmonic class of a root. Chapter 4 presents the results of an experiment that examines how Uyghur speakers generalize backness harmony to novel roots. Responses display sensitivities to the phonetic properties of backness harmony that are not evident in attested forms. I suggest that this discrepancy between attested and novel words can be explained by phonetically-driven learning biases whose effect is obscured by lexical listing of the harmonizing class of attested roots, but which become evident in novel roots for which no such listing exists. Finally, Chapter 5 is a phonetic study that evaluates whether roots with no harmonizing segments display evidence of a covert backness contrast in their vowels that drives their harmonizing behavior, as previously suggested in the literature. Although evidence of coarticulatory effects from neighboring segments is found, there is no correlation between vowel backness and harmonizing behavior. This again supports an analysis where the harmonizing behavior of these roots is lexically specified, rather than phonologically determined. The second part of the dissertation, Chapters 6 and 7, look at Uyghur backness harmony from the perspective of formal language theory. Previous work suggests that patterns in segmental phonology tend to occupy a particular region of the subregular hierarchy (that is, they can be generated by mathematical models which are less powerful than finite-state automata). More complex patterns have been shown to pose learnability problems in artificial grammar learning studies. Chapter 6 demonstrates that Uyghur backness harmony exceeds an upper bound of complexity that has been proposed for segmental patterns. Chapter 7 presents a probabilistic extension of a language class commonly applied to phonological patterns, which is used to model the novel root data from Chapter 4. This model is used to test the hypothesis that the discrepancy between wug and corpus patterns presented in Chapter 4 results from a bias towards computationally simpler patterns. It is demonstrated that the productive pattern learned by speakers is not computationally simpler than the pattern found in attested words, presenting evidence against claims of learning biases towards computationally simpler patterns