This dissertation is a study of the prosodic system of Dakelh (Carrier), an Athapaskan language of central interior British Columbia, focusing primarily on the endangered Lheidli dialect spoken in the area of Prince George, B.C. The study is primarily based on original fieldwork data, elicited from three native speakers of Lheidli Dakelh, and partly on comparison with the Nak'azdli dialect as reported in Story (1989). This work contributes much-needed empirical data to the long-standing debate over the proper characterization of Dakelh prosody with respect to notions such as tone vs. pitch accent vs. stress. Under the general rubric of prosody, three topics are investigated in detail. The first is an analysis of syllable and foot structure, developed within the framework of Optimality Theory, which addresses such issues as word minimality, epenthesis patterns, syllabification, and the relationship between syllable structure and stress. For example, epenthesis and deletion are found to be highly sensitive to morphological factors and morphologically-defined domains. Secondly, a phonetic investigation of properties which are usually correlated with stress, namely increased pitch, duration, and/or amplitude, is undertaken. One of the findings is that in addition to word-final stress in verbs, manifested primarily in terms of duration, "prominence" in the form of increased pitch is typically also found on one of the earlier syllables in the word. The location of the latter is partly determined phonologically, and in part lexically; certain prefixes appear to carry lexical tone, as in many related languages. The third topic under examination is the phonological behaviour of tone. Though a lexical tone contrast cannot be established on the basis of isolation forms alone, evidence of such contrasts comes from sandhi processes. The Lheidli dialect is shown to differ significantly from the Nak'azdli dialect in the phonological realization of tone patterns; for example, the distribution of high tone is partly sensitive to the phonation type of a preceding consonant. The word-internal distribution patterns as well as the tone sandhi system of both dialects are analyzed in Optimality Theory. Tone sandhi is shown to be derivationally opaque and thus highly problematic for standard versions of the theory. / Arts, Faculty of / Linguistics, Department of / Graduate

碩士 / 國立臺灣大學 / 語言學研究所 / 95 / This present study aims (1) to investigate the correlation of phonological representations and phonetic realizations of Ta-pu Hakka tones, and (2) to explore the tonal coarticulation in different kinds of contexts. There are two major Hakka sub-dialects spoken in Taiwan, the Hai-lu and the Suu-hsien systems. Another sub-dialect, Ta-pu Hakka, is spoken mainly in some townships in central Taiwan by about 60,000 people. Different sub-dialects of Hakka have their own tonal systems that have been described in detail in previous literature. However, most of the tonal representations used in the literature with regard to tones were based on auditive perception and phonological description. As a result, those phonological tonal representations were more impressionistic. This study first investigated the correlation between phonological representations and phonetic realizations of the Ta-pu Hakka tones by taking acoustic and statistical approaches. Six informants, including three males and three females, were recruited in this study. Two sets of inventories were adopted as the test stimuli: the first was a set of monosyllabic words, and the other was a set of disyllabic words. The monosyllabic words were used to verify the citation tones, while the disyllabic words were used to verify the sandhi tones. Meanwhile, the tonal representations in different prosodic positions, i.e., in word-initial (WI) and in word-final (WF) positions were also investigated. The second research question in this study was to explore tonal coarticulation in disyllabic words in Ta-pu Hakka. Tonal coarticulation refers to tonal variations in consecutive speech, which has been a great concern to many phoneticians (Xu, 1994, 1997, 1999, 2001, 2004; Gandour et al., 1994 & 1999; Peng, 1997; Shen, 1990 & 1992). In this study, we investigated tonal variations in different prosodic positions, tonal contexts, and in a context with different neighboring tones. Target Approximation model (Chen & Xu, 2006; Xu, 1997, 1999, 2004; Xu & Wang, 2001) was used to help explain the tonal variations found in this language. To answer the above two research questions, we measured the average pitch height (F0) of the beginning point (BP), the ending point (EP), the maximum pitch (F0_Peak), the minimum pitch (F0_valley), and the average pitch of the syllable (Mean_F0). Besides, we also calculated the mean slope and the mean duration of the whole syllable. For Yang-Ping tone, the F0 valley alignment was calculated. As for falling tones and checked tones, the F0 peak alignment was calculated instead. In investigating tonal representations, we adopted Fon and Chiang’s (1999) postulated formulae and Shi’s (1990) logarithmic function. In the first section with regard to verifying the tonal representations in Ta-pu Hakka, the revised T-scales (5-scaled tonal system) of the seven citation tones were suggested as follows: Yin-Ping as [33], Yang-Ping as [313], Shang as [31], Chü as [53], Yin-Ju as [3], Yang-Ju as [4], and Supra-Yin-Ping as [34], based on the data from monosyllabic words. As for the tones in disyllabic words, the T-scales showed great incongruence between the results obtained via the two calculation methods. The results suggested that tonal representations in disyllabic words should not be considered as the norms of tonal representations. As for the three sandhi tones (ST), we found that the revised T-scales via Shi’s function were more appropriate to show their phonetic realizations. The three sandhi tones were suggested as ST-Yin-Ping as [324], ST-Yang-Ping as [323], and ST-Chü as [44], respectively. However, the revised T-scales in fact represented the phonetic realizations of the sandhi tones under the influence of tonal coarticulation, so we would still adopt the original T-scales, ST-Yin-Ping as [35], ST-Yang-Ping as [33], and ST-Chü as [55], when sandhi tones were involved in investigating tonal coarticulation effect. The phonetic realizations of tones in Ta-pu Hakka showed that the seven citation tones included one level tone, two falling tones, two checked tones, one rising tone, and one concave tone. The tones in this language could be paired off, one belonged to the lower register, whereas the other belonged to the higher register. Note that the T-scales of the two checked tones were very close to each other. The statistical results of comparing means of the measured parameters in the two checked tones indicated that not only the pitch height but also the falling slope served as the criterion of distinguishing one from the other, within which the lower checked tone, Yin-Ju [3], had a steeper falling F0 slope. As for tonal coarticulation, we explored the tonal variations from different aspects. First we investigated the tonal variations of tones according to their position in disyllabic words, namely in WI and WF positions. Afterward, Xu’s (1994) classification of different tonal combinations into a compatible or a conflicting context was examined and modified based on the tonal combinations in Ta-pu Hakka. We proposed a third type of tonal context, which was termed as a contour compatible context, referring to a context where a rising tone followed by a falling tone, or a falling tone followed by a rising tone. Then the influence of adjacent tones on their preceding or following tone was compared and discussed. The results of tonal coarticulation in Ta-pu Hakka were summarized as follows. First, in regard with the position effect, final-lengthening was found in all the non-falling tones and the Yang-Ju [4]. However, the duration of Yin-Ju [3] and Shang [31] was longer in WI rather than in WF. Furthermore, the mean F0 slope of these two tones was steeper in WF position, indicating an abrupt fall at the end of the utterance. As for Supra-Yin-Ping [34], however, the mean slope was steeper when it was in SI than in SF. The most interesting finding was that the F0 valley alignment of Yang-Ping [313] was closer to the onset in WI, and it became closer to the middle of a syllable when in WF. With regard to the tonal variations in a compatible or conflicting context, we found that Shang [31] showed a steeper slope in a contour compatible context than in a conflicting context. The result also suggested that a contour compatible context was very similar to a compatible context. As for Supra-Yin-Ping [34], the slope was even steeper in a contour compatible context than in a compatible context. As for the results of the influence of adjacent tones on each other, carryover assimilation effect was found in Yin-Ping [33], Yang-Ping [313], Supra-Yin-Ping [34], Chü [53] and Yang-Ju [4]. Anticipatory assimilation was found in Shang [31], Chü [53] and Yin-Ju [3]. For Yang-Ju [4], anticipatory effect was also found, but the influence of its following tones on its tonal variations showed both assimilation and dissimilation effect. Furthermore, we also found a positive correlation between the offset F0 in syllable 1 (WI) and the onset F0 in syllable 2 (WF). Furthermore, the tonal variations found in Ta-pu Hakka could be explored via Xu’s Target Approximation model. Both anticipatory assimilation and carryover assimilation phenomena supported Xu’s claim that “when two pitch targets occur next to each other, if the offset of the first one is different from the onset of the second one, the second one will appear as if it has been assimilated or partially assimilated to the second” (Xu and Wang 2001, p. 329). As for the anticipatory dissimilation, the effect could be found in Yin-Ping sandhi rule ([33] turning [35]) and Yang-Ping sandhi rule ([313] turning [33]). On the other hand, the anticipatory assimilation could also be seen in Chü sandhi rule ([53] turning [43]). We stipulate that tone sandhi rules are in fact the historical products of tonal coarticulation, and the phonological aspect of tone sandhi rules is either to show greater contrast between adjacent tones or to ease the effort of articulation. To sum up, both the phonetic realizations of tonal representations and tonal coarticulation effect in Ta-pu Hakka aim to show either harmony or contrast. The phonetic realizations of tonal representations and coarticulation in this language also help group all the tones into different natural classes: such as non-falling vs. falling, and non-checked vs. checked.

Cette thèse consiste en une description du sun̄wadia, langue mélanésienne parlée dans l'île de Maewo, située dans la partie centre-nord du Vanuatu (Pacifique Sud). La description est basée sur deux séjours de terrain de trois mois chacun. La langue présente, à des stades plus ou moins poussés, les traits caractéristiques de la famille austronésienne: l'existence d'un article personnel, une tendance à l'omniprédicativité, la modification de la structure argumentale du verbe par quelques morphèmes hérités du système à applicatif des langues situées plus haut dans la généalogie de la famille. Elle présente également le système de marques personnelles très reconnaissable des langues austronésiennes.Il s'agit d'une langue relativement conservatrice sur le plan phonologique, mais dont la morphologie dérivationnelle apparaît assez érodée (il n'existe ni conjugaison verbale, ni déclinaison casuelle, ni marquage morphologique du nombre sur les noms). La thèse s'organise en six parties. La première étudie la phonologie, la morphophonologie, et la structure morphologique du mot sun̄wadia, ainsi que les phénomènes de sandhi. La seconde partie s'intéresse aux parties du discours et pose quelques bases syntaxiques sur lesquelles s'appuiera le reste de l'étude. Les quatre parties suivantes étudient successivement le fonctionnement du syntagme substantival, celui de la prédication (structurée par le recours fondamental aux constructions à verbes sériels), le système de repérage et de référence temporels, spatiaux et circonstanciels, et enfin l'organisation de l'énoncé. La thèse est accompagnée d'un extrait de corpus glosé d'une douzaine de pages. / This PHD thesis consists in a description of Sun̄wadia, a melanesian language spoken in Maewo Island, Central-NorthVanuatu (South-Pacific). It is based on two fieldworks of three months each.This language exhibits, in a variable extension, some of the typical characters of the austronesian languages: it has a personal article, tends towards omnipredicativity; the argumental structure of the verb undergoes modifications via a few morphemes related to the applicative systems of languages that are situated higher in the genetic tree of the family. The language also has the typical pronominal system of austronesian languages.Sun̄wadia is a relatively conservative language, on the phonemic level at least, but its morphology appears to be quite eroded (there aren't any verbal conjugation, nor any nominal declension, nor any morphological marking of number on the noun). This thesis is organised in six parts. The first one studies phonemics, morphophonemics, and the morphological structure of the Sun̄wadia word, as well as sandhi phenomenon. The second part concerns the parts of speech and lays down some syntaxic grounds that will be useful to the rest of the study. The last four parts review the substantival syntagm, the functioning of predication (which is mostly built around serial verbs constructions); the temporal, spatial, and circumstantial reference, and, lastly, the global organisation of the clause. The thesis comes with a short excerpt of our oral corpus (a dozen of pages).

碩士 / 國立臺灣師範大學 / 台灣文化及語言文學研究所在職進修碩士班 / 97 / The study introduces from the “traditional dialectology” to get an insight to the current status of the Taiwanese dialect in Penghu Shiyeu Towship, adopting the field survey methods, corpus analysis and language of comparative law, in order to record and analyze the phonological and lexical characteristics in the Shiyeu region. The purpose of the study aims at: 1. Investigate and compile phonological system of the Shiyeu Taiwanese dialect, in order to understand the voice characteristics of the dialect. 2. Compare the dialect with other secondary dialects related to Taiwanese dialect, compiling the sounding, rhyming and toning characteristics of the Shiyeu Taiwanese dialect. 3. Compile special vocabulary of Shiyeu Township through investigation and comparison of vocabulary. 4. Complete the vocabulary draft for the Shiyeu Township Taiwanese dialect to be used as significant reference language source for preparing the Taiwanese dialect teaching materials. The conclusion of the study includes the following:   1. The Shiyeu Township Taiwanese dialect tends to orient towards the Pienchuan Accent, while preserving a significant difference from the common accent in Kaohsiung City. 2. With respect to vowels, there are seven major vowels /a、o、e、e 、ö、i、u/ in Hejie, Zhumen and Xiaomen villages, and only the /e/ and /ö/ vowels have bee preserved; while there are only five major vowels /a、o、e、i、u/ in other villages, generally regardless of // and /o/, while “Gu”and “Gu” share the same sound. 3. With respect to consecutive tone sandhi, the intonation (yin 2) is modulated to middle tone /51/ > /35/, while the intonation (yang 5) and intonation (yang 7) are modulated to lower tone /13、33/ > /11/ ,The intonation (yin 3) and the intonation (yin 2) receiving /?// are modulated into two categories /11/ > /51、55/；/3/ > /51、55/ and quite different from other Pienchuan accent. 4. The “□Ah” tone-sandhi is mainly the “fixed low tone,” stem will be pronounced as the original tone-sandhi, while the “Ah” is pronounced as the fixed lower tone /11/, which is significantly different from the Taiwanese Youshi accent. 5. The “Epenthesis” phenomenon of affixed stem is one characteristics of the Shiyeu Taiwanese dialect. This “Epenthesis” phenomenon is however not found in Zhumen village, which shows the uniqueness of the secondary dialect in Zhumen village. 6. The Shiyeu folks tend to dwarf on their own language while possessing high tolerance to foreign accents

碩士 / 國立新竹教育大學 / 進修部語文教學碩士班 / 93 / Abstract Speech sound of Hu-Si village in PongHu has such characteristics: 1. Pronounce O as IO. 2. /ə/ vowel in Old Chuan accent. 3. Yin-shang tone turn to Yang-ping in tone sandhi. The three characteristics changed gradually by the conversion of life style, the convenient transportation and influential mass media. The purpose of this research is to study the mutational situation of the three characteristics. There’re many factors that influence speech mutation. This research did survey according to different age and sex, and we found are followed： 1. In the vowel /IO/： A. The /IO/ decreased gradually as the decreasing of age group. Teenagers of fifteen to nineteen rarely pronounce /io/ particularly. B. The /io/ in different sex doesn’t have prominent difference. 2. In the vowel /ə/. A. /ə/had changed to /e/ or /ue/, however /e/ is pronounced more. B. We can hardly find /ə/ pronounced in youths of twenty to thirty-nine and teenagers of fifteen to nineteen. Female changed faster than male, especially teenager female of fifteen to nineteen. 3. In tone sandhi Yin-shang tone turn to Yang-ping. A. It was pronounced Yang-ping tone mainly, however there was a tendency to Yin-ping tone, especially female of fifteen to nineteen. The three situation of speech mutation is the phenomena of lexis diffusion. /io/and /ə/ vowels diffused early, most of them had come to the step of V4. The tone sandhi of Yang-Shang most were in V1 step. Beside the important factors of different age group and sex, other social factors that influence the speech mutation are: 1. People of Peng-Hu are tolerate to languages, however had tendency to depreciate speech sound of their own. 2. The decreasing of ability South-Min dialect in teenagers fifteen to nineteen is mainly because of Mandarin eduation. 3. Beside the inner structure and outer social factors of speech, the most important concern is the reflection of speakers toward the speech consciousness, such consciousness reflection was most obvious in /io/ vowel. Language has practical values but also is the infusion of feelings. When there’re conflicts between its practical value and feelings, we hope people should keep comparable identification to dialects of our own. Language beauty is in its local characteristics.

碩士 / 國立交通大學 / 電信工程系所 / 95 / In the thesis, a syllable recognition system for Taiwanese was established and applied to a real application - Intelligent Transportation System(ITS). First , a syllable-based Taiwanese speech recognition system was implemented. And the effect on the recognition performance of the syllable with entering tone was carefully exiamined. Based on the linguistics knowledge, the syllable with entering tone in database was re-labeled according to the tone sandhi of Taiwanese in order to improve the recognition rate. In addition, Kullback Leibler distance between the acoustic model before and after re-labeling was examined to verify the linguistics knowledge. And the syllable bigram language model to the recognition system will obtain the rate improvement. Finally, the Taiwanese Speech Recognition System was applied to an Intelligent Transportation System(ITS)

Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Linguistics and Philosophy, 2003. / Includes bibliographical references (p. 283-295). / This dissertation concerns both phonetic and phonological aspects of tone mapping in various Chinese languages. The central issue addressed is the role of contrast and positional prominence and neutralization in the realization of tone. The inventory of tonal contrasts constrains the outputs of contextual neutralization as well as the location of pitch targets in phonetic implementation. Two prominent phonological positions in the tone sandhi domain are distinguished: peripheral (initial and final) positions and metrically strong positions. Input tones occupying different prominent positions in the input are preserved in the output; their realization in the output can be determined by the location of stress. A typology of diverse patterns of tone preservation and realization emerge from the interaction of positional faithfulness and positional markedness constraints. The research findings reported here have implications for both phonetics and phonoloy. / by Zhiqiang Li. / Ph.D.

碩士 / 國立中央大學 / 客家語文研究所 / 101 / This thesis deals with the phonology of Gutian township in Shanghang County, FujianProvince, China. Gutian dialect is a kind of Western Fujian Hakka. The thesis is divided into six chapters. The first chapter is an introduction that consists of the motive and objective, methodology, literature review, and facts and figures about theShanghang County. The second chapter focuses on the phonology of Gutian Hakka dialect,including initials, finals, tones and tone sandhi phenomena. We will show in the third chapter the correspondences between Gutian dialect and the Middle Chinese classificatory system. In chapter four we put the emphasis on the characteristics of Gutian Hakka phonology. Chapter five offers a phonological comparison between Gutian and seven other Western Fujian Hakka dialects (Ninghua, Qingliu, Liancheng, Changting, Yongding, Wuping, and Shanghang). Chapter six is the conclusion.

Cette thèse est une étude sur les quatre règles de base qui pourraient constituer un corpus cohérent pour l'enseignement du breton. Les trois premières, concernant les mots en eux-mêmes, sont présentées dans le premier volume : accent tonique, longueur sous l'accent et dévoisement de la consonne en finale absolue. La proposition d'un accent unique standard est la clef réaliste qui permet de donner cohérence à l'ensemble. Le second volume traite des phénomènes de sandhi. Après une large enquête menée dans les écoles bilingues du CE2 au Lycée, de Mai 2007 à Avril 2008, on constate une pénétration importante du modèle français dans la prononciation scolaire. Sont mis en évidence plusieurs points faibles de l'enseignement du breton : 1) plus de 50% des ouvrages répertoriés ne mentionnent pas la question des liaisons, fonctionnant pourtant le plus souvent à l'inverse du français. 2) Ce qui intervient pour beaucoup dans l'extension des liaisons “à la française” chez les apprenants, est la différence faite à la finale, depuis 1901, entre les substantifs et les autres espèces de mots. Cette décision, qui ne tenait pas compte de l'existence de deux sortes de suffixes en breton, les neutres et les durcissants, entraîne en effet une multiplication artificielle des consonnes sourdes visuelles en finales. On peut mesurer aujourd'hui les conséquences de cette décision sur le terrain. 3) Outre son impact sur les liaisons compte-tenu de l'effet Buben, la décision de 1901, reconduite jusqu'à l'accord orthographique de 1941, désorganise aussi le décodage de la longueur de la voyelle sous l'accent dans bon nombre de mots, autres que les substantifs. La thèse pose donc clairement la question de la pertinence de cette règle vieille de plus d'un siècle, à l'heure où l'enseignement du breton tente de se structurer

博士 / 長庚大學 / 電機工程學研究所 / 96 / In this dissertation, we review research efforts in text-to-speech (TTS), pronunciation error detection for computer assisted language learning, and speech corpus design for Taiwanese, or Min-nan – a major native language spoken in Taiwan. Following an introduction of the orthography and phonetic structure of Taiwanese, we describe the various databases used for these tasks, including the Formosa Lexicon (ForLex) – a phonetically transcribed database using Formosa Alphabet (ForPA), an alphabet system designed with Taiwan’s multi-lingual applications in mind – and the Formosa Speech Database (ForSDat) – a speech corpus made up of microphone and telephone speech. For speech synthesis, through the use of the bi-lingual lexicon information, the Taiwanese TTS system is made up of three functional modules: a text analysis module, a prosody module, and a waveform synthesis module. An experiment conducted to evaluate the text analysis and tone-sandhi modules reveals about 90% labeling and 65% tone-sandhi accuracies. Multiple-level unit selection for a limited domain application of TTS is also proposed to improve the naturalness of synthesized speech. For computer assisted language learning (CALL), we focus on how to find the real pronunciation of the user. The data-driven based method was used to generate pronunciation errors hypotheses incorporating knowledge-based method. In the experiment results, the detection rate of pronunciation detection can achieve 77.2%. Finally, we applied this technique into our CALL system.