碩士 / 國立交通大學 / 電信工程系 / 89 / In this thesis, based on a Min-Nan speech TTS system proposed in previous study, some modifications were done to improve the quality of synthesized speech. The original Min-Nan speech TTS system contained three parts- text-parser, prosodic information generator and PSOLA synthesizer. It is well known that the tone sandhi will change the tones of most speech from the lexicon tone in Min-Nan. In order to get better pitch contour of synthesis speech, a unit is added after parser to change the tone information generated from parser according to the tone sandhi rules of Min-Nan. The POS(part of speech) is also added in the lexicon of parser to get more information in text and a better performance of prosodic information generation unit was achieved. Finally, in order to improve the performance of short word, a statistical prosodic information generating scheme is built for a word synthesizer.

博士 / 國立臺灣大學 / 資訊工程學研究所 / 97 / Taiwan Southern Min (Taiwanese) is an important language that has received only a little attention in the world. The characteristic of written Taiwanese is quite different from Mandarin or English in some respects. We will focus on Taiwanese processing techniques in this dissertation. POJ is an important script of Taiwanese. We introduce character code of POJ, and mention the numbered POJ as the interchange code for various POJ encodings. Then, we propose a two-stage search strategy for POJ text search, and propose POJ syllable query expansion. We also describe the display method for POJ, POJ word processing utilities and word segmentation method for HR mixed script. We propose a rule-based tone sandhi algorithm. We translate every word into Mandarin, and obtain the POS information. Using the POS data and tone sandhi rules, we then tag each syllable with its post-sandhi tone marker. Finally we implemented a Taiwanese tone sandhi processing system. Our system achieves 97.4% and 89.0% accuracy rate with training and test data, respectively. Additionally, we propose a POS tagging method. We develop a word alignment checker to help the two Taiwanese scripts word alignment work, select the most adequate Mandarin word using Hidden Markov probabilistic model, and finally tag the word using Maximal Entropy Markov Model classifier. We achieve an accuracy rate of 91.5% on Taiwanese POS tagging work. We have established some useful online written Taiwanese tools for past several years. Based on these tools and preliminary research results, we hope the written Taiwanese processing related research can be promoted.

碩士 / 國立清華大學 / 語言學系 / 84 / The thesis investigates Yinping Tone Sandhi (YTS) in Miouli Hakka. Direct Reference Hypothesis and Indirect Reference Hypothesis are developed to examine Yinping Tone Sandhi Rule (YTSR) at the interface between syntax and phonology. In this research, we adopt Kaisse''s (1985) Domain C-command to observe the direct syntactic influence upon phonology and find that the domain for Yinping Tone Sandhi cannot be defined by the relations between syntactic constituents. In addition, we discuss the application of Yinping Tone Sandhi Rule from the perspective of prosodic phonology. Based on Selkirk''s (1984, 1986) End-based Parameters, we suggest that the phonological phrase in Hakka is marked at the right edge of a non-adjunct maximal projection. Our investigations show that YTS is better conditioned by the domain of intonational phrase (IP) and whereby we propose a formation rule of IP. Finally, we examine trisyllabic or quadrisyllabic Yinping Tone Sandhi under Hsiao''s (1991,1994) Beat-counting Theory of foot formation, and revise the direction of beat scanning to account for alternative tonal readings.

碩士 / 國立政治大學 / 語言學研究所 / 102 / The grammar of Meixian and Bangkok Hakka tone sandhi has been analyzed from a rule based approach. Nevertheless, there are some questions and details that could not be solved by the analysis, such as the status of tones that do not undergo sandhi, triggers of the tone sandhi, and the structural relation between citation tones and their sandhi counterparts. Thus, the purpose of this study is to re-analyze the tone sandhi in Meixian and Bangkok Hakka under the constraint based framework, Optimality Theory (OT). There are two mechanisms of tonal alternations in the two dialects’ tone sandhi: assimilation, and dissimilation. So in order to capture the tonal alternations, the current analysis applies the concepts of the Obligatory Contour Principle, and the No-Jumping Principle. The constraints generated according to these principles work well with the application of the Constraint Conjunction approach. The conjoined constraints are needed since the tone alternations are highly marked, and depend a lot on the head/right syllable tone. Furthermore, this thesis also posits several faithfulness constraints that rank higher than the tone sandhi markedness constraint. The high ranked faithfulness constraints govern the preservation of several tones from any alternation, and preservation of some structures of the citation tones when they become sandhi tones (i.e. register and initial target). In conclusion, the positing of tone sandhi markedness constraints and undominated identity constraints presents a better input-output correspondence relation of the tone sandhi phenomena in Meixian and Bangkok Hakka. To conclude the thesis, a brief summary of the study and possible further issues are presented.

碩士 / 國立清華大學 / 統計學研究所 / 93 / Taiwanese is rich in tone sandhi. It is a two-part problem: when to “tone sandhi,” and how to “tone sandhi.” For multi-syllabic words, major rules exist for both parts of the problems. For a complete sentence or a phrase consisting of multiple words, the tone sandhi rules for word may not apply at the last syllable of each word. Traditional approach to this problem is by the syntactic analysis, and this paper studies the tone sandhi problem by statistical approach. Using as corpora the seven volumes of Buddhist Sutra, published and phonetically annotated in Taiwanese by a senior nun, we model the phonetic transcription by syllable-based Markov language model, and study specifically the tone sandhi problem. A unigram model gives 80% correct and bigram model 84%. Both results are computed using seven-fold cross-validation.

碩士 / 國立政治大學 / 語言學研究所 / 99 / This thesis is aimed at analyzing Yinping tone sandhi in Meinong Hakka from the perspective of Optimality Theory. The discussion is divided into two parts. First, this thesis explores the tonal domain in terms of the interface of syntax and phonology. The analysis indicates that the tonal domain is founded on an intonational phrase. The formation of an intonational phrase is based on the Sense Unit Condition (Selkirk, 1984) and is closely related to restructuring (Nespor and Vogel, 1986). The special constructions of IPs are also investigated. The data supports the hypothesis proposed by Nespor and Vogel (1986), namely, that these special constructions have the privilege to form an isolated intonational phrase. Secondly, Optimality Theory is adopted to analyze the data after exploring the tonal domain. Two set of constraints are proposed in this thesis. One is that of tonal constraints, which are mainly concerned with the processing of tone sandhi. The other is that of prosodic constraints, which are responsible for explaining the delimitation of the tonal domains of the data. The two sets of constraints can be used to account for the operation of the sandhi phenomena in ordinary sentences and also in special constructions. The alternative readings are explained by means of the re-ranking of *IP and ALIGN-R (Non-adjunct XP, IP) from the perspective of Cophonology Theory.

碩士 / 國立政治大學 / 語言學研究所 / 99 / This thesis offers an Optimality-theory approach to the tone sandhi of checked tones (TSC) in Shanghai, in an attempt to regulate its surface patterns which have long been considered anomaly as opposed to the tone sandhi of smooth tones (TSS). With a reanalysis in the present study, TSC and TSS arguably have in common that their process of tone mapping is both subject to ANCHOR-L(tσ1, Hd), by which the un- derlying tone of the initial syllable ends up at the stressed syllable in the output. It follows that TSC is different from TSS in the way that metrical head is assigned: all domains undergoing TSS are stressed on their initial syllable; only in domains of TSC does the stress move rightwards, either to the second place in the tone sandhi of Yinru (TSYI), or to the final syllable in the tone sandhi of Yangru (TSYA). Given the assumption that checked syllables and smooth syllables in Shanghai are light and heavy, respectively, in terms of moraicity, and that foot-parsing is binary and left-aligned in general, the one-syllable shift of stress in TSYI can be accounted for by positing COINCIDE(LightHd, FT-final), which sets up the preference for light stress- bearing syllables to be licensed foot-finally. On the other hand, given that Yangru in the citation forms represents a rising contour, different from the level tone of Yinru, a licensing constraint, namely COINCIDE(Contour, PrWd-final), is further posited so that the long-distance movement of metrical head observed in TSYA emerges to satisfy the requirement for the retained rising contour to be licensed word-finally. Taken together, this thesis instantiates a remarkable case of the mapping among multiple prominent positions, including heavy syllables, metrical head, contour tones, and edge positions. Also, the present analysis demonstrates that Shanghai tone sandhi involves an interaction between tone and stress, thus a contribution to the general OT tone-prominence typology literature (cf. Zhang 2001, Barnes 2002, De Lacy 2002).

博士 / 國立成功大學 / 外國語文學系 / 105 / The purpose of this study is to investigate the acquisition of Tone Three Sandhi (T3S) by focusing on Mandarin-acquiring children’s productions and comprehension of Tone Three Sandhi Avoidance (T3SA). The fact that T3SA overlaps with the optional T3S poses a learnability problem. That is, when children encounter an instance of T3SA, how do they know that the instance is a real case of T3SA but not a case of optional T3S where the surface tone happens to be a citation tone (i.e. Tone 3)? Three experiments were conducted to shed light on this learnability problem. The first experiment investigated the surface tone patterns of Mandarin-speaking adults when the critical/first syllable of two adjacent Tone 3 syllables was a topic or the noun being modified by a clause (i.e. a relative clause). Thirty-seven adults participated the recitation task. They are invited to recite a list of sixty sentences, including ten sentences with topics, twenty sentences with relative clauses and thirty filler sentences. The results showed that adults generally produced Tone 3 for the target syllables, showing that T3S was generally avoided among topics and relative clauses. Although sentence length had an effect on surface tone preference, the phenomenon could be accommodated by Chen’s (2000) Minimal Rhythmic Units (MRU) analysis because the formation of MRUs can be affected by sentence length. After the adults’ T3SA norm was established, the second experiment investigated Mandarin-acquiring children’s production of T3SA. Specifically, the longitudinal data from three children aged 1;6 to 3;11 were transcribed and analyzed. The results indicated that the average T3S accuracy rate was 98.55%. Additionally, three T3SA utterances involving topics and two T3SA utterances involving relative clauses were found. The coder of the one consistently reported that all the target syllables were Tone 3, indicating that children acquired T3SA at early stages, supporting the view of early acquisition of T3S. As I argued, Wang’s (2011) failure in observing young children’s adult-like T3S productions was due to her complex experimental designs. In this connection, in the third experiment, I adopted a comprehension task, which consumed less children’s processing capacity. In this experiment, twenty children aged between 3;9 to 4;3 (Mean: 4;0) and fifty-six adults were invited to judge if the heard sentence was an accurate description of a given picture. The knowledge in T3SA would affect their acceptability judgment. The results demonstrated that 4-year-olds comprehended T3SA involving topics in the same way as adults did, adding strong support to the view of early T3S acquisition. Additionally, the current results support the generative approach to language acquisition because the theory successfully predicted the development of T3S acquisition in Mandarin Chinese. Based on the experimental results, the study concludes that: 1) Chen’s (2000) MRU analysis is able to accommodate both T3S and T3SA phenomena in Mandarin Chinese; 2) Wang (2011) failed to observe children’s early T3S performance because her experiments consumed much of children’s processing burden, which in turn affected their linguistic performance (c.f. Liu ＆ Lee, 2014; Trueswell et al., 1999); 3) The usage-based approach to language acquisition (e.g. Tomasello, 2003) could not accommodate the current results as the environment of T3SA is a subset of the optional T3S and therefore T3SA cannot be acquired early solely by experience; 4) The generative approach to language acquisition (e.g. Chomsky, 1981) could accommodate the current results because abstract syntactic categories are considered innate knowledge, which could assist children’s acquisition of T3SA in the early developmental stages.

碩士 / 國立清華大學 / 臺灣語言研究與教學研究所 / 107 / In this study, to assess whether the preferred oral form, produced by each subject in off-line confrontation picture naming, is more easily processed, EEG was recorded while the subjects made the on-line picture-sound matching. Different tone-sandhi forms, used in two major Taiwanese subdialects, were manipulated to induce different responses shown in their event-related potentials. In addition to the two tone-sandhi conditions, there were two conditions irrelevant to tone-sandhi as fillers. The event-related stimuli consisted of four conditions in total. Behavioral results in the on-line experiments showed a pretty high acceptance rate for either one of the two tone-sandhi varieties given the same picture. It is rarely revealed in their behaviors of off-line confrontation picture naming. Grouped by their preferred forms produced orally in off-line confrontation picture naming, 7 subjects tended to use subdialect A, and 15 subjects tended to use subdialect B. However, the preferred tone-sandhi condition didn’t elicit smaller N400 as predicted in each group. Furthermore, the results from source localization analyses of their event-related potentials also confirmed the N400 results. The findings in this study that the preferred sandhi form, produced orally by each subject in off-line confrontation picture naming, was not unequivocally processed more easily than the other tone-sandhi form in on-line picture-sound matching, suggests a necessity to give a very careful consideration in delineating the boundaries of subdialects and trying to understand the mechanisms of dialectal amalgamations and merges.

碩士 / 國立臺灣師範大學 / 台灣文化及語言文學研究所 / 96 /