碩士 / 國立交通大學 / 電機與控制工程系 / 88 / In this thesis, we investigate a new technique on a Mandarin text-to-speech (TTS) system. Our major effort is focused on prosodic information generation. New methodologies for constructing fuzzy rules on a prosodic model to simulate human’s brain are studied in this thesis. The proposed Recurrent Fuzzy Neural Network (RFNN) is a multi-layer recurrent neural network (RNN) which integrates a Self-cOnstructing Neural Fuzzy Inference Network (SONFIN) into a connectionist structure. The RFNN can be functionally partitioned into two parts. The first part adopts the SONFIN and is taken as a prosodic model to explore the relationship between high-level linguistic features and prosodic information by inferring fuzzy rule. The second part employs a five-layer network to generate all prosodic parameters by using the prosodic fuzzy rule from the first part and other important features of syllable fed in directly. Using the method proposed in our TTS system can overcome not only sandhi rule and the other prosodic phenomena existing in the traditional TTS systems but also to find out some rules about prosodic phrase structure. Hence, we can generate proper prosody parameters, including pitch means, pitch shapes, maximum energy levels, syllable duration and pause duration, to synthesis fluent speech. To verify the performance of this prosodic model, we modify a TTS system developed previously, based on time-domain pitch synchronous overlap add (TD-PSOLA) method, with a Mandarin monosyllable database. Through some listening test, the synthetic speech is more natural than previous version.

碩士 / 國立新竹教育大學 / 臺灣語言與語文教育研究所 / 97 / This thesis studies on the phonetic variation of Shoucheng dialect in Puli. Most people in Shoucheng belong to an Austronesian tribe, Kahabu. Their first language is Taiwan Minnan, and it is with some degree of Austronesian substratum effect. We found out that Shoucheng dialect is changed because of both phonetic and social factors; however, the factors that urge sound to change have different degree of impacts. This thesis analyzes four types of sound changes or variations in Shoucheng dialect, including iang/ing variation, o/oo neutralization, diminutive tone sandhi and non-high tone mixing. We use sociolinguistic theory to discuss the correlations of social factors and certain phonetic changes. Quantitative study and statistic analysis are adopted in the thesis. We collected data from sixty subjects that are evenly divided into three generations. Each generation has ten males and ten females. The results of this study are: 1. o/oo neutralization remains the most (75.58%), because it conforms to the symmetry principle of vowel system and coincident with neutralized tendency of Puli dialect and general Taiwan Minnan. 2. In phonetic point, some sound changes are related to the different phonetic environment, like iangand o. 3. In social factors, “age”, “education”, “living here by marriage or not” and “social network” have significant relations with certain sound changes. Phonetic preservation and change of different age forms an X-shaped figure, it means that Shoucheng sounds will decrease as people getting older, but Puli sounds will increase in younger generation. Overall, Shoucheng dialect is now “changing in progress” and “changing rapidly” towards Puli dialect.

碩士 / 國立新竹教育大學 / 臺灣語言與語文教育研究所 / 97 / The Vocabulary Study of Hailu Hakka Dialect in Shin-Wu Abstract Hakka People of Huizhou origin residents settled in Shin-wu region, north western area of Taiwan, represent a significant portion of the total population of the inhabitants. It owns the high population comparing with adjoining regions in the majority. Amongst the dialects of Hakka, the Hai-Lu Hakka dialect has been Commonly used as a prime dialect in Shin-wu, from where reserved a wide variety of sub-dialects of Hakka language. It brings a lot of attentions and acquired enormous research achievements. Regretfully, major studies focus on the research of the powerless dialects out of Hai-Lu dialect. In Taiwan, numerous studies of Hai-Lu dialect were deviated to the Hsinchu region. It has been a long time since Hai-Lu dialect has not been well treated and recognized in that of Shin-wu area. This research uses Hai-lu dialect in Shin-wu region is an investigating object. Through the arrangement of basic words and its phonological form and word formation analysis, we hope this study can compensate to a certain degree the discrepancy of previous research gaps. The objective of this study is to devote to establish a theoretical definition of “basic word list”. Based on the basic word list to develop overall analytical and statistical methods to the distinct items. The purpose of classification is to acquire more objective and consolidated information that illustrates phonetic and word formation scattering status. It is important that basic vocabulary data thus compensates for gap of Hai-lu dialect in Shin-wu region. The study of phonetic form of word is to develop statistical analytical items by its frequency of phoneme, structure of syllable and pronunciation of the Chinese Character. Proceed to the study of word from its tone sandhi of flow of speech and phonetic form of diminutive. The word formation research is to analyze morpheme and morpheme group according to simple words, derivational words and compound word to proceed word formation and the frequency statistics.

碩士 / 國立高雄師範大學 / 英語學系 / 91 / Makung Southern Min has long been considered idiolect because of its specific phonemes, lexemes, and tone sandhi patterns. This is the reason the previous studies focused mainly on data collection. When dealing with the syllabification, a traditional syllable structure used to be the only choice. Such a theory is inadequate in differentiating VC from VG, in explaining labial dissimilation and nasal assimilation, and in proposing a complete set of phonotactic constraints. Therefore, the idiolect syllable performances in SMM are often regarded as free variations. The purpose of the study is to explore the syllabification of Makung Southern Min under the framework of Optimality Theory. We began the study by discussing the most unmarked phenomena: the Sonority Sequence Principle and the asymmetrical attitudes toward the CV and VC structures. The Sonority Sequence Principle helps construct the Southern Min syllable template and define the peak of the syllable. In Southern Min, either a single vowel or a rising diphthong is to occupy the position of the peak. In addition, there is only one slot for the post-peak segment. Thus we are able to differentiate a rising diphthong from a falling diphthong and to explain why the structure ui or iu is always considered a falling diphthong in the previous studies. Referring to the different attitudes between CV and VC, the former is preferable in natural languages, while the latter is much more restricted. This phenomenon is apparent in Southern Min. In the following steps, we try to figure out the constraints percolating within each marked domain. In SMM, there is no deletion for the peak. This higher ranked constraint makes SMM and other sub-dialects different. At last, the Southern Min contraction is discussed. . The contracted forms provide empirical examples for us to observe the mechanism of phonotactic constraints. Under the framework of Optimality Theory, the syllabification of SMM is constructed with an explanatory set of hierarchically ranking constraints. The differences between SMM and other sub-dialects result from different ranking hierarchies, instead of free variations.

碩士 / 國立政治大學 / 語言學研究所 / 105 / The goal of this thesis is to explore the tone sandhi domain, tone group (hereafter, TG), of Taiwan Southern Min of younger generation. A corpus is built and analyzed within the framework of Prosodic Theory, a derived theory from the Phonology-Syntax Interface Theory. In the literature, TG boundary # was marked at the right edge of an XP which is neither an adjunct nor a clitic. However, the TG boundary # of younger generation is marked at the right edges of a maximal XP, an adjunctive XP and an X-head. Moreover, language variation is observed in the TG parsing: the presence of TG boundary # is optional. There are four TG parsings not observed in the literature: (1) V # NP # (2) V # QP # (3) QP # A # (4) AdvP # V #. TG is redefined as a Phonological Phrase which syntactically corresponds to a maximal XP, an adjunct XP, an X-head but not to a clitic XP. The four TG parsings are then analyzed with Rank-Ordering Model of EVAL (ROE), a subtheory within the Optimality Theory domain which particularly addresses language variation. According to ROE, constraints on the right side of critical cut-off line participate in the prediction of variation. Three constraints involving in variation are proposed: Align-R (XP-Fnc, ϕ), ϕ-min and Align-R (X-head, ϕ). Align-R (XP-Fnc, ϕ) distinguishes lexical projection and functional projection, ϕ-min requires a Phonological Phrase to be minimally disyllabic, and Align-R (X-head, ϕ) predicts # to occur at the right edge of an X-head. The analysis centers around the three constraints’ relation with the critical cut-off line. The constraint rankings are: Ident-T-Rϕ | Align-R (XP-Fnc, ϕ) >> ϕ-min >> Align-R (X-head, ϕ) >> *T >> Align-R (XP-Lex, ϕ). Except for the faithfulness constraint Ident-T-Rϕ, the other markedness constraints are arranged at the right side of the cut-off line in order to predict language variation.

碩士 / 國立交通大學 / 電機與控制工程系 / 87 / This thesis investigates a new technique using a recurrent neural network enhanced by a fuzzy-rule inference engine in the prosodic model of a text-to-speech (TTS) system. The TTS system generates more natural synthetic speech with the enhanced recurrent-neural-network (RNN)-based prosodic model than with the traditional rule-based one. Using the method proposed in our TTS system can overcome sandhi and other prosodic phenomena existing in the traditional TTS systems which always had trouble in generating fluent speech. In other words, the proposal method provides a total solution to solve the problem of prosodic information synthesis. The enhanced RNN-based prosodic model employs a five-layer network and a fuzzy inference engine with twenty-five fuzzy rules to generate the prosodic information, including pitch means, pitch shapes, maximum energy levels, syllable duration and pause duration. Experimental results show that the enhanced RNN-based prosodic model can automatically learn prosody phonological rules of human beings. To verify the performance of this prosodic model, we implement a Mandarin TTS system which is based on time-domain pitch-synchronous-overlap-add (TD-PSOLA) method, with a Mandarin monosyllable database and an enhanced RNN-based prosodic model. Through subjective tests by the native laboratory members, it is shown that the synthetic speech is more natural than one synthesized by the traditional rule-based method.

The main topic of the thesis is a discussion of the contraction of esse (-st < est , -'s < -es), a widespread phenomenon in Latin comedy, generally known as ‘prodelision’. The thesis collects evidence for contracted forms from a large variety of sources (e.g. inscriptions, Sabellian languages and ancient grammarians) and analyses their transmission in the manuscript tradition of Latin authors. Contracted forms appear to be widespread but are neglected by or puzzling to editors. They are not misspellings, abbreviations, or representations of a sandhi phenomenon related to elision (‘prodelision’), but are clitic forms of the verb esse, showing phonetic reduction. The thesis discusses their linguistic significance, relying in particular on analysis of their behaviour in Terence: auxiliary or copula esse is clitic and its standard position seems to be clause-final, attached to the participle or predicate noun or adjective. Second, the thesis analyses another complex phenomenon of early Latin, the prosodical omission of final -s before a consonant (sigmatic ecthlipsis), relying on a database of all Terence's lines potentially involving it. Metrical evidence appears to be limited to cases in which a participle or predicate noun or adjective precedes a form of esse beginning with s-, thus confirming the prosodical distinctiveness of such a sequence. The evidence of sigmatic ecthlipsis thus ties in closely with that of contraction. The findings of the thesis have philological and editorial implications. First, they throw light on orthographic, phonetic, prosodic, and syntactic aspects of the verb ‘to be’ in Latin. It has a strong bond with the participle and predicate noun or adjective and may either be reduced phonetically in combination with such hosts, or participate in a simplification of -ss- which it may share with its host. Second, transmitted contracted spellings or readings betraying a misunderstanding of them should be given full consideration by editors and not disregarded. Contracted spellings might be restored in other conditions since manuscripts are often not trustworthy when transmitting uncontracted forms. Contracted forms may be stylistically motivated. They occur in comedy particularly in spoken metres, and in late Latin become a spelling archaism. Finally, sigmatic ecthlipsis in Terence occurs for certain only in restricted conditions: the editorial practice of printing it wherever it is possible even if not necessary is misleading.

碩士 / 國立中央大學 / 客家語文研究所 / 104 / Peitian village is located at Liancheng county which borders the area of Min dialects in the western Fujian province. In this area, the dialects have some phonological features due to the interaction between Hakka and Min dialects. In addition, She languages as substratum languages also had effect on phonological changes to some extent. The particular background caused the unusal appearance of Peitian Hakka dialect. The thesis is divided into six chapters.The first chapter is an introduction that consists of the motive and purpose, methodology, literature review. Besides, a brief description of historical and geographical background of Liancheng County are shown as well. The second chapter represents the phonetic system of Peitian Hakka dialect that is obtained from the field reseach, including the initials, finals, tones and the special tone sandhi. In the third chaper, a comparison is made between Peitian Hakka dialect and the middle Chinese phonetic system to discuss the sound evolvement and development of Peitian Hakka dialect. The fourth chapter focuses on the phonological features of the main research material. The fifth chapter demonstrates the study of vocabulary characteristics and the comparison among other western Fujian Hakka dialects such as Wuping and Yungding Hakka dialects. The last chapter is the conclusion of the research results, also, examination of the shortcomings of this study and possible future advices are provided to make further development of Peitian hakka dialect.

碩士 / 國立中央大學 / 客家語文研究所 / 104 / Zhongbao dialect researched in this paper is widely spoken in Zhongbao Village of Gutian Town, Liancheng County, Western Fujian. Western Fujian is located at the junction of Fujian, Guangdong and Jiangxi, where is a transition area for dialect. But, dialect of Liancheng County is complex and comes to front especially and the county is known as “Different pronunciations within 5 miles and different tones within 10 miles”. The reason of numerous dialects and difference in Liancheng County is not only related to different dialects introduced by early different immigrants, but also related to the contact with national minority, isolation caused by mountainous terrain, intermarriage between ethnic groups and nature evolution of dialect self. Besides, the left portion of Liancheng County is adjacent to Changting County of Chunke County, while the right portion is adjacent to Yong’an City of Min Chinese Fujian area. So, affect caused by language contact is one reason resulting in complex dialect, which also is the reason that Zhongbao dialect merges the features of Min Chinese, Hakka dialect and underlying language together, along with complex phonetic layers. The whole paper includes 6 chapters. Chapter I is introduction introducing the research motivation, research purpose and research method and following relevant literature review, localism and speaker introduction. Chapter II states Zhongbao dialect investigated by the writer in Zhongbao, presenting its phonology system, feature of tonal sandhi and literary and colloquial reading. Chapter III contrasts Zhongbao dialect with ancient Chinese phonology to view historical phonetics change. Chapter IV points out phonetic features of Zhongbao dialect according to Chapter III and observes its synchronic phonetic change based on horizontal comparison among dialects. Chapter V is comparison of the words, mainly including Wuping, Changting and Ninghua. Chapter VI is a conclusion explaining the research achievement of the paper and putting forward advices and making a self-criticism.

碩士 / 國立交通大學 / 外國文學與語言學研究所 / 96 / The purpose of this thesis is to study the function of the Taiwanese ka in the sentence. It is widely believed that Taiwanese ka is the counter part of Mandarin ba. They share many similar syntactic and semantic features. Although ka is treated as the counter part of Mandarin ba, there still remain differences between them. For instance, ka is able to assign thematic role to its argument, but ba is incapable to assign thematic role to the argument. Also, in the ka construction, bare verb is tolerated under some circumstance. This never happens in the ba construction. Unlike the previous researches simply treat ka as the counter part of ba, we propose that ka has two functions. One is to be the counter part of ba and the other is to be the applicative marker. When ka is treated as an applicative marker, ka is able to introduce an additional argument into an argument-saturated sentence. Based on the analysis proposed by Pylkkanen (2005), we believe that Taiwanese ka construction belongs to the high applicative construction. As for the reason that the bare verb is not tolerated in the Mandarin ba construction, Government-Nucleus Stress Rule (G-NSR) raised by Feng (2005) will be applied to explain the phenomenon. In Taiwanese, we believe that because Taiwanese has a very complicated tone-sandhi system to help speakers to deicide the stress, it need not follow the G-NSR. Therefore, we believe that what really affects the existence of the bare verb in Taiwanese may still be the boundness of a verb.