UCL Discovery
UCL home » Library Services » Electronic resources » UCL Discovery

The dynamics of Japanese prosody

Lee, KLA; (2015) The dynamics of Japanese prosody. Doctoral thesis , UCL (University College London). Green open access


Download (6MB) | Preview


This dissertation explores aspects of Tokyo Japanese (Japanese henceforth) prosody through acoustic analysis and analysis-by-synthesis. It 1) revisits existing issues in Japanese prosody with the minimal use of abstract notions and 2) tests if the Parallel Encoding and Target Approximation (Xu, 2005) framework is suitable for Japanese, a pitch accent language. The first part of the dissertation considers the nature of lexical pitch accent through examining factors that affect the surface F0 realisation of an accent peak (Chapter 2) and establishing the articulatory domain that hosts a tonal target in Japanese (Chapter 3). Next, pitch accent interactions with other communicative functions are considered, specifically in terms of focus (Chapter 4) and sentence type (Chapter 5). Hypotheses using acoustic analyses from the previous Chapters are then verified through analysis-by-synthesis with articulatory synthesisers AMtrainer, PENTAtrainer1, and PENTAtrainer2 (Chapter 6). Chapter 2 provides conclusive evidence of Japanese as a two-tone language as opposed to bearing three underlying tones in its phonology, previously unresolved in existing literature. Proponents of the two-tone hypothesis gather evidence from perception: when stimuli are played in isolation, native listeners can only distinguish two tone levels (High and Low). On the other hand, production evidence reveals robustly three distinct surface F0 levels. Using a series of linear regression analyses, I show the third tone level could be interpreted as a result of pre-low raising, a common articulatory phenomenon. The F0 of an accent peak is inversely correlated with the F0 of the following low target, being an enhanced peak in preparation for the upcoming L. Interpreted together with native listeners’ inability to hear three tones when said in isolation, as repeatedly reported in previous studies, I establish Japanese has only H and L in its tonal inventory. Chapter 3 establishes the syllable as the tone-bearing unit in Japanese tonal articulation. Often described as a mora-timed language, it has been previously unclear whether articulatory tonal targets are hosted in a mora or a syllable in Japanese. When comparing accented words of various syllable structures I found that the F0 accent peak of CVCV wordsoccurs consistently earlier than that of CVn/CVCV words. CVCV words are longer in total duration so its earlier F0 peak is a result of a shorter tone-bearing unit (i.e. two consecutive short morae/syllables). CVn/CVV words on the other hand have a later peak F0 due to hosting an articulatory target as a long syllable, rather than two short morae. I further verified the syllable hypothesis using two articulatory synthesisers, PENTAtrainer1 and PENTAtrainer2. The syllable as a tone-bearing unit incurs fewer predictors but provides better learning accuracy. Chapter 4 explores focus prosody in declarative sentences. Using a newly collected corpus of 6251 sentences that controls for accent condition, focus condition, sentence type, and sentence length, I challenge the widely held idea that post-focus compression of F0 range is accent-independent. Currently it is generally accepted that regardless of the accent condition of the focused word, the excursion size of ‘initial rise’ that marks the beginning of the first word 4 after focus is shrunken. However, confining the notion of post-focus compression to initial-rise (usually extending across only two morae) sets Japanese apart from other languages like English or Mandarin, where such compression is robust across the entire post-focus domain. I show that when F0 range is measured across a wider domain, compression is absent. Where post-focus compression is absent, the F0 trajectory appears to be a result of articulatory carryover effects. This will be interpreted as a result of weak articulatory strength on the post focus domain, explaining the difference in F0 trajectories in long and short utterances. Chapter 5 builds on the previous Chapter to consider in addition the focus prosody in yes/no questions. I investigate what marks a yes/no question, and how focus prosody differs in declarative and interrogative utterances. Acoustic analyses show that questions are marked by a final rise, but the exact shape of such a rise depends on the accent condition of the sentence-final word. When compared to declarative sentences, the key differences in yes/no questions include: a higher F0 level; the absence of post-focus compression even in contexts otherwise observed in statements; and on-focus F0 raising as the only robust focus marker. These findings point to the fact that interrogative focus prosody is not an amalgamation of focus markers and question markers, and bear implication on the representation of Japanese intonation. Chapter 6 verifies observations established thus far through analysis-by-synthesis. I demonstrate comparative modeling as a means to adjudicate between competing theories using PENTAtrainer2, PENTAtrainer1 and AMtrainer. In terms of local fitting accuracy, AMtrainer yielded comparable synthesis accuracy to the PENTAtrainers. Finally, I further demonstrate the compatibility of PENTA with Japanese prosody showing highly accurate F0 predictive analysis (when trained with Chapter 2 production data), and highly satisfactory speaker-dependent synthesis accuracy (when trained with Chapter 4 and 5 sentential data). Naturalness judgment ratings show that the natural stimuli sound as natural as the synthetic stimuli, though questions generally sound less natural than statements. Reasons for this discrepancy are discussed with reference to the design of the stimuli.

Type: Thesis (Doctoral)
Title: The dynamics of Japanese prosody
Event: UCL (University College London)
Open access status: An open access version is available from UCL Discovery
Language: English
UCL classification: UCL > Provost and Vice Provost Offices
UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences
UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences > Faculty of Brain Sciences
UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences > Faculty of Brain Sciences > Div of Psychology and Lang Sciences
URI: https://discovery.ucl.ac.uk/id/eprint/1470249
Downloads since deposit
Download activity - last month
Download activity - last 12 months
Downloads by country - last 12 months

Archive Staff Only

View Item View Item