The purpose of this paper is to provide a comprehensive understanding of how contrastive phonation is produced and perceived by native speakers of a language with both tonal and phonation contrasts, Yi. In the production experiment, we measure a wide scope of relevant physiological and acoustic parameters, which show substantial physiological-acoustic coupling: a Contact Quotient (CQ) distinction is the essential property of the phonation contrast, while H1*-H2* and H1*-A1*, which are significantly correlated with CQ, are the best acoustic measures for the phonation contrast. The bandwidth of the first formant (B1) and the Cepstral peak prominence (CPP) are effective acoustic cues too. In addition to the well-established contributions of the vocal folds, this study gives insight into the role of supraglottal settings for the phonation contrast. A consistent F1 difference for the phonation contrast in Yi indicates a shape change in the vocal tract, supporting the multidimensional phonation model proposed by Edmondson & Esling (2006). Such a supraglottal effect is similar to that in ATR (Advanced Tongue Root) languages, related to a change in pharyngeal size, though in Yi the direction is RTR (Retracted Tongue Root). With more detailed discussion about variations in the tense vs. lax contrast across speakers as well as across languages, the relativity of phonation contrasts is highlighted. The perception experiment looks at the effectiveness of voice quality, F0 and F1 for the perception of phonation categories by Yi listeners. We draw a perception map from a perceptual confusion matrix and then compare that with a production map. It is found that listeners heavily rely on F1 for the perception of the tense vs. lax contrast, though it is a secondary cue in production. However, the perception map generally agrees with the production map, though the perception map varies across listeners. The variation found in the production experiment suggests a possible sound change in the phonation contrast. The variation in the perception map further indicates the future direction of the sound change: Vowel quality might eventually take over as the distinctive feature of the tense vs. lax contrast.

Phonation is used in many tonal languages, but how it should be incorporated into tonal systems is not well understood. The purpose of this dissertation thus is to examine the role of phonation in tonal contrasts, and to investigate how phonation and pitch interact in the tonal space. This dissertation presents close studies of tonal contrasts from three language families that provide different answers to these questions.

From the case of Yi languages, where the tense/lax phonation ("register") contrast is orthogonal to the tone contrast, we show that phonation production can be independent from tonal production. Phonation in these languages is not only phonologically contrastive, but phonetically completely independent. Tone is purely pitch, and phonation is purely voice quality.

From the case of Mandarin, where non-modal phonation (specifically, creaky voice) is known to be an allophonic cue to the low-dipping Tone 3, we show that the presence of creak in Mandarin is purely driven by pitch range, and can occur with any of the low-pitch targets. Moreover, we show that in a corpus of pitch sweeps that rose or fell over large pitch ranges, Mandarin speakers' voice quality co-vary with pitch in a wedge-shaped function.

Finally, we investigate the case of Black Miao, a language with five-level-tone contrasts, the case of maximum use of pitch contrast attested among languages. Due to limitations on pitch production and perception, it should be very hard to make these contrasts purely by pitch. Both production and perception experiments show that non-modal phonations provide very important cues in both tonal production and perception, but in different ways. On the one hand, the three mid-range tones (T22, T33, T44) have very similar pitch cues, but T33 is quite distinct from T22 and T44 by its breathy voice quality. On the other hand, the extra-high (T55) and extra-low (T11) tones show the same kind of pitch-dependent voice quality that was found in Mandarin: the extra-low pitches tend to be creaky while the extra-high pitches tend to be tense.

Taking these cases together, we establish: 1) there are two different uses of phonation types, based on their relationship with pitch: pitch-dependent and pitch-independent, which can either enhance the perceptual distinctiveness of extreme pitch targets, or serve as a contrastive dimension for tones with similar pitch values; 2) tonal contrast is multidimensional, and the dimensionality of tonal contrasts in part depends on the size of the tonal inventory. We propose a new tone space model, and principles for good dispersion of tonal contrasts. This dissertation thus extends our knowledge about tonal contrasts, and provides a better understanding of the interaction between phonation and pitch in tone languages.

This study compared the contrastive phonation types of four languages—Gujarati (modal vs. breathy), Hmong (modal vs. breathy vs. creaky), Mazatec (modal vs. breathy vs. creaky), and Yi (tense vs. lax)—on several acoustic measures, within and across languages. For Gujarati, Hmong, and Yi, two electroglottographic (EGG) measures were also compared; a Contact Quotient measure distinguished the within-language phonation types in all three languages. While several acoustic measures distinguished phonation types within each language, only H1*-H2* did so in all four languages. However, when each within-language phonation category was then compared across languages, each category was found to differ from language to language on multiple acoustic measures, e.g. breathy in Hmong is distinct from breathy in Gujarati. This unexpected result suggests that language/speaker differences in voice quality are larger than phonation category differences. This suggestion finds support in a Multi-Dimensional Scaling analysis of the acoustic measures. A three-dimensional space turns out to mostly distinguish the languages, less so the phonations. The phonation categories do not form clusters in this space across languages as might have been expected, but they do occupy separate regions along the third dimension of the space, a dimension correlated especially with H1*-H2*. Thus H1*-H2* is again seen to be the most important measure of phonation contrasts across languages.