Cross-regional word duration patterns in Mandarin

Duration contrasts can convey many types of information, including language background, word structure, word frequency, speech genre, intention, and emotions. An understanding of duration lays the foundation for many aspects of speech technology since duration plays a major role in speech production and perception. This dissertation explores the duration patterns of Mandarin words among three Mandarin dialectal regions---Beijing, Taiwan, and Malaysia. This dissertation brings diverse methodologies on speech data collection, annotation, and corpus construction to investigate linguistic pattern. Three speech production studies are conducted to explore the duration patterns of words with different length and internal structures. These studies reveal the general duration patterns of Mandarin Words. First of all, all the multi-syllabic words demonstrate the disyllabic long-short metrical form. Second, linguistic factors---syllable structure, positions (syllable position, word position, and sentence position), word frequency, word category, word internal structure, particle attachment, speech rate of sentence have significant effects on syllable duration. Thirdly, social factor---region interacts with multiple linguistic factors (word structure, syllable position, and particle attachment) and plays an important role in duration prediction. Quantitative data from these studies reveal that there are regional differences in rhythmic contrast among different Mandarin speaking regions. Beijing Mandarin speakers are more sensitive to the length change of linguistic unit and show stronger rhythmic contrast than speakers from Taiwan and Malaysia Mandarins. The results also display that Malaysia Mandarin speakers show the similar rhythmic pattern as Beijing Mandarin speakers. The investigation of duration patterns in this dissertation provides a detailed description of word duration in Mandarin. This dissertation also provides the foundation for further research on duration pattern related super-segmental feature. A comprehensive understanding of duration pattern with linguistic and social factors is helpful to improve the quality of durational models used in speech technology