Date of Award

Spring 2022

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Linguistics

First Advisor

Bowern, Claire

Abstract

In naturalistic speech, the phonetic instantiation of phonological categories is often highly variable. Speakers have been observed to converge on patterns of phonetic variation that are consistent within languages but variable cross-linguistically for the same phonological phenomenon. Speakers are evidently sensitive to these sorts of patterns and learn the phonetic variation in a consistent way. Furthermore, the systematicity of this variation suggests that these patterns should change over time systematically as well. Most Australian languages assign lexical stress consistently on the first syllable of the word, raising the question of how the phonetics of stress varies across languages with this phonologically stable pattern. This dissertation presents an investigation into structured variation of the acoustic correlates of stress and prosody in sixteen Indigenous languages of Australia that all have consistent initial stress placement, with a focus on the source(s) of variation in these factors cross-linguistically. Acoustic correlates of stress, despite the phonological uniformity present among these languages, show significant cross-linguistic variation, both in the presence or absence of a particular cue to stress, as well as the size of these effects. The phonological uniformity of stress assignment allows for a more controlled comparison of the acoustic correlates of stress across these languages, since the placement of stress marking remains constant. Acoustic correlates investigated are vowel duration, pre-tonic and post-tonic consonant duration, intensity, f0 (maximum and range), and vowel peripherality. These cues are identified using a series of mixed effects linear regression models. To identify the source(s) of variation in acoustic correlates to stress, the population genetics tool Analysis of Molecular Variance (AMOVA) is used. This is a statistical tool created for analysis of genetic variance that has been applied to cultural evolution topics such as music and folktales. This model finds significant variation across languages, as well as substantial intra-speaker variation, similarly to the findings for both biological and cultural evolution, but no significant intra-language variation across speakers. These results are also supported by the investigation of inter- and intra-language variation using regresssion modeling. Another population genetics measure, fixation index, is used to create a network model of language relationships based on the phonetic correlates of lexical stress. This network shows clear relationships between the Pama Nyungan languages in this sample, as well as some Gunwinyguan languages, supporting the claim that the phonetic cues to stress are stable within language families and change according to the principles of diachronic language change. Smaller groupings in this network also indicate some contact-induced change or areal effects in these phonetic markers. Phrasal prosody is also investigated in this dissertation, using a toolkit for automated phrasal contour clustering. For each language, f0 is measured at regular intervals across the word, which is used as input to a complete-linkage clustering algorithm to identify major categories of phrasal contours. Results of this sort of automatic clustering provide testable hypotheses about phrasal types in each language, while avoiding some common pitfalls of impressionistic analyses of prosodic phrases. As with the investigation into lexical stress, this sort of automated typological work serves as a crucial complement to more detailed language-specific studies for the creation of well-rounded and well-supported theories. The data used in this dissertation are narrative speech recordings sourced from language archives, collected in varying field settings. In processing these data I have created a large corpus of these recordings force aligned at the segment level and have worked out post-hoc methods for controlling noise and variation in field-collected audio to create a comparable set of language data. I include in the dissertation a lengthy discussion of these methods, with the aim of providing a practical toolkit for the use of archival materials to address novel phonetic questions, as well as to aid in the creation of language revitalization resources.

Download
COinS