從優選理論分析臺灣華語的韻母音組限制 - 政大學術集成

全文

(1)國立政治大學語言學研究所碩士學位論文 National Chengchi University Graduate Institute of Linguistics Master Thesis. 從優選理論分析臺灣華語的韻母音組限制治. 政. 大. An Optimality Theory 立Approach to Phonotactic Restrictions of. 學. al. sit. io. 指導教授：蕭宇超. er. Nat. y. ‧. ‧ 國. Taiwanese Mandarin Rhymes. n. iv n C U Advisor:h Yuchau E.i Hsiao engch 研究生：葉德偉撰 Student: Tei-wei Yeh. 中華民國一○九年六月 June, 2020. DOI:10.6814/NCCU202000482.

(2) An Optimality Theory Approach to Phonotactic Restrictions of Taiwanese Mandarin Rhymes. 立. 政治大. 學. ‧ 國. By. ‧. Tei-wei Yeh. n. er. io. sit. y. Nat. al. Ch. engchi. i Un. v. A Thesis Submitted to the Graduate Institute of Linguistics In Partial Fulfillment of the Requirements for the Degree of Master of Arts. June, 2020. DOI:10.6814/NCCU202000482.

(3) 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i Un. v. Copyright © 2020 Tei-wei Yeh All Rights Reserved. DOI:10.6814/NCCU202000482.

(4) ACKNOWLEGEMENTS 感覺像是完成了一件人生大事，雖然過程很辛苦，但是看到成品時的成就感讓我覺得一切都是值得的。能夠順利地完成這本論文，首先一定要再三地感謝蕭宇超老師，對蕭老師的感謝已經不僅僅是單純對指導教授的感謝，而是對於老師在我寫論文的期間是如何地為了讓我能夠準時畢業而犧牲了許多自己的時間一次又一次地解決我的疑難雜症。謝謝林蕙珊老師和吳貞慧老師所有寶貴的建議讓我的論文更完整，也謝謝兩位老師幾次的舟車勞頓來擔任我的口試委員。謝謝惠鈴學姐在行政上的支持與照顧，在我每次疏忽了重要程序的時候不厭. 政治大. 其煩的提點我；謝謝怡臻在我論文初期碰到困難時給了我很重要的意見；謝謝子. 立. 權從以前對我的學期報告給出的建議，如今都成了論文的一部分；謝謝旺楨在工. ‧ 國. 學. 作室的行政支持讓我可以安心寫論文；謝謝欣蓉在我兩次口試的時候都主動幫助我處理一些雜務；謝謝仲良在最關鍵的時刻提供我重要的語料。. ‧. 這段時間也非常感謝所有身邊朋友的陪伴。謝謝俊賢長期以來給我的歸屬感. y. Nat. io. sit. 跟溫暖；謝謝孟群、思妘和柏宇在我需要說話的夜裡用你們的聲音陪伴我；謝謝. n. al. er. 政矩和姿羽擔任我的心靈導師開導我；謝謝珮辰常在我需要放鬆的時候找我一起. Ch. i Un. v. 取暖；謝謝泰叡總是在我需要人陪的時候和我一起消磨時光；也謝謝徐臨嘉老師. engchi. 在我壓力最大的時候給了我精神上的支持，讓我可以渡過難關。最後謝謝我的家人，不管我做什麼樣的決定，都無條件地支持我，讓我可以沒有後顧之憂地完成碩士學位。最重要的，我要謝謝我自己，雖然在整個碩士生涯中曾經幾度想直接放棄算了，但要不是我後來還是撐過來了，今天就不會有這樣的感動。這段時間有太多不同的事情同時發生在我的生命中，說實在的，走得很辛苦，可是當我咬緊牙關走到今天，再回首一看，才發現，原來這就是人生精彩的所在。. v DOI:10.6814/NCCU202000482.

(5) VITA Education. 2020. M.A. in Graduate Institute of Linguistics, National Chengchi University. 2017. B.A. in Department of English, National Chengchi University. Grants and Scholarships. National Chengchi University. ‧. 2018-2020. Graduate Student Scholarship. 學. 2019. ‧ 國. 立. 政治大. Ministry of Science and Technology Scholarship. y. sit. n. al. National Chengchi University 2017. Ch. engchi. er. Theoretical Phonology Lab Assistantship. io. 2018-2020. Nat. Project No. MOST 107-2410-H-004-115-MY2. i Un. v. Graduate Student Scholarship. National Chengchi University. Publication. 2019. “On Some Irregular Patterns of Syllable Contraction in Taiwan Mandarin.” Paper presented at the 52nd International Conference on Sino-Tibetan Languages and Linguistics, The University of Sydney, Sydney, Australia, June 26 vi. DOI:10.6814/NCCU202000482.

(6) TABLE OF CONTENTS ACKNOWLEDGEMENTS VITA. v. vi. TABLE OF CONTENTS ABSTRACT. vii. ix. Chapter 1 Introduction. 1. 1.1 Preliminaries. 1. 1.2 Field of Research. 2. 政 2治大. 1.2.1 The Domain of Rhyme 3 立. 1.3 Thesis Layout. ‧ 國. 學. 5. Chapter 2 Theoretical Background and Literature Review. ‧. 2.1 Optimality Theory. 7. 7. 2.1.1 Constraint Ranking. y. Nat. 8. n. al. 2.1.3 Local Conjunction and Self Conjunction. Ch. engch 12 i. 2.2 Previous OT Analyses of Mandarin. i Un. 2.2.1 The Issue of Mid Vowel Assimilation 2.2.2 The Issue of Low Vowel Raising. 17. 21. 21. 3.2.2 The Low Front Vowel [a] 3.2.3 [jaj]. 19. 19. 3.2 Controversial Rhyme Structures 3.2.1 [e] and [o]. v11. 12. Chapter 3 Patterns of Rhymes in Taiwanese Mandarin 3.1 Principles and Charts. 10. er. io. 2.1.2 Categories and Definitions of the Constraints. sit. 1.2.2 Dialect. 22. 22 vii. DOI:10.6814/NCCU202000482.

(7) 3.3 A General Description Chapter 4 An OT Analysis. 23 28. 4.1 Rhymes without a Nasal Coda. 28. 4.1.1 Mid/Low Vowel Assimilation. 28. 4.1.2 The Phonotactic Restriction of Triphthongs 4.1.3 The Sequence of High Vocoids 4.1.4 Summary. 34. 40. 46. 4.2 Rhymes with a Nasal Coda. 47. 4.2.1 High Vowel Preceding a Nasal Coda. 47. 政治大. 4.2.1.1 The Regular Patterns [in], [wən] and [yn]. 立. 48. 4.2.2 [ən] and [əŋ] without a Prenuclear Glide. REFERENCES. 80. 74. al. n. Chapter 5 Conclusion. 68. er. io. 4.2.4 Summary. 64. y. Nat. 4.2.3.2 Low Vowel Raising. 77. 64. sit. 4.2.3.1 Low Vowel Backing. 62. ‧. 4.2.3 Low Vowel Preceding a Nasal Coda. 54. 學. ‧ 國. 4.2.1.2 The Irregular Outputs of /iŋ/, /uŋ/ and /yŋ/. Ch. engchi. i Un. v. . viii DOI:10.6814/NCCU202000482.

(8) 國. 立. 政. 治. 大. 學. 研. 究. 所. 碩. 士. 論. 文. 提. 要. 研究所別：語言學研究所論文名稱：從優選理論分析臺灣華語的韻母音組限制指導教授：蕭宇超研究生：葉德偉論文提要內容：（共一冊，16,505 字，分五章）. 立. 政治大. 華語的音節結構非常簡單，最多僅能容納 CGVX 四個音段（Duanmu，2007）。. ‧ 國. 學. 然而其內部卻有著非常複雜的音組限制。先前的研究（Duanmu，2007；Lin，. ‧. 2007/2015；Hsieh，2012）曾透過優選理論分析中元音的同化及低元音的提高現象，然這些研究並未針對臺灣口音中不同的特色做出考量；再者，除了這些音韻. y. Nat. io. sit. 規則以外，華語的韻母仍存在許多其他的限制。. n. al. er. 本論文以臺灣的口音為基礎，透過優選理論（Prince and Smolensky，1993）. Ch. i Un. v. 分析華語的韻母音組限制，包含無鼻音韻尾及有鼻音韻尾兩個部分。無鼻音韻尾. engchi. 的韻母主要的現象包括中/低元音在雙元音結構下的同化、介音與滑音韻尾的異化及介音後不得緊接高元音；含鼻音韻尾的韻母結構中，主要的現象包含高/低元音與鼻音韻尾的同化、央元音的插入及高元音的提高。除了這些現象外，臺灣口音包含了少數不規則的韻母，如[iŋ]及[oŋ]，本論文認為此方言擁有較高排序的制約排除[əŋ]的組合（*əŋ）及連續的後高音段（OCP-[+hi, +bk]）。經過本論文的分析，臺灣華語的韻母中，大部分的空缺皆具系統性，僅*[ɥa] 及*[ɥɑŋ]無其他對稱的空缺，故本論文將此二韻母視為偶然空缺。而除了韻母內部外，聲母及韻母的組合仍存在著其他音組限制，則有賴後續更深入的研究。 ix DOI:10.6814/NCCU202000482.

(9) ABSTRACT The syllable structure of Mandarin is quite simple which is described as CGVX (Duanmu, 2007), indicating that it can only contain four segments at most. However, there are complex restrictions within such simple structure. Previous research (Duanmu, 2007; Lin, 2007/2015; Hsieh, 2012) has proposed some analyses regarding to mid vowel assimilation and low vowel raising in terms of Optimality Theory. However, these studies do not focus on Taiwan dialect, which actually has some different rhymes from other dialects. Furthermore, there are still other phonotactic. 政治大 Optimality Theory (Prince. restrictions in addition to mid vowel assimilation and low vowel raising. This thesis proposes an. 立. and Smolensky, 1993). ‧ 國. 學. approach to analyze the phonotactic restrictions of Taiwanese Mandarin rhymes. The analysis is divided into two parts: One discusses the rhymes without a nasal coda and. ‧. the other the rhymes containing a nasal coda. The phenomena in the rhymes without a. sit. y. Nat. nasal coda mainly include the mid/low vowel assimilation, the dissimilation of. er. io. prenuclear and postnuclear glides and the prohibition of two adjacent high vocoids.. al. iv n C h e n vowel assimilating to the nasal coda, the central i U and low vowel raising. In g c hinsertion n. As for the rhymes with a nasal coda, the analyses mainly include high/low vowels. addition to these phenomena, there are some irregular rhymes in Taiwanese Mandarin such as [iŋ] and [oŋ]. These rhymes are surfaced out because of the high-ranked constraints such as *əŋ and OCP-[+hi, +bk]. Through the analysis, most absent rhymes in Taiwanese Mandarin is systematic except for *[ɥa] and *[ɥɑŋ]. Therefore, they are considered accidental gaps in this thesis. In addition to the phonotactic restrictions within the rhyme domain, there are much more constraints regarding to the combination of the onset and the rhyme, which relies on further studies. x DOI:10.6814/NCCU202000482.

(10) Chapter 1 Introduction 1.1 Preliminaries. The syllable structure of Mandarin is quite simple, which is composed of only four segments at most. However, many syllables that seem to be able to exist in Mandarin fail to surface out. That is, there are in fact complex phonotactic restrictions within Mandarin syllables.. 立. 政治大. Duanmu (2007) proposes the structure of CGVX to describe a Mandarin syllable,. ‧ 國. 學. where C represents a consonant, G a glide, V a vowel and X a nasal coda or an offglide. This model indicates that not only is a Mandarin syllable composed of only at. ‧. most four segments, the segment of each position also has a narrow inventory.. Nat. sit. y. Traditionally, a Mandarin syllable is divided into two parts: Onset (the domain of. n. al. er. io. C) and Rhyme (the domain of GVX)1. A lot of restrictions can be observed solely. i Un. v. within the rhymes. For example, *[jan], *[wow] and *[e]2 are all absent in Mandarin. Ch. engchi. even though [j, w] are allowed to appear in the position of G, [a, o, e] are allowed to appear in the position of V and [n, w] are allowed to appear in the position of X3. 1. Whether the prenuclear glides (i.e. the G of CGVX) belong to the onset or the rhyme has been quite. controversial. Duanmu (1990) proposes that a prenuclear glide is actually the secondary feature of the onset. For example, kuan ‘close’ is actually pronounced [kwan], where [k] has a secondary feature of [+round]. However, Hsu (2009), on the other hand, thinks that [j] belongs to the rhyme while [w] belongs to the onset. Lee (2001) also states that prenuclear glides of Mandarin possess a very special position, whose phonological feature is contradicted to Chinese phonological system. Here I clarify that the domain of rhyme in this thesis includes prenuclear glides, which will later be elaborated in 1.2.1. 2. Syllables like [e] and [o] appear as interjections or particles in Mandarin. However, they never appear. in other lexical words. The clarification is elaborated in section 3.2.1. 3. The inventories of each position will be displayed later in section 3.1. 1. DOI:10.6814/NCCU202000482.

(11) Although some of the inexistent rhymes can be considered accidental gaps, most of the absent patterns seem to be systematic. For example, no diphthongs4 can follow the glide [ɥ] to form a triphthong. Such systematic gaps indicate that there must be some rules and restrictions confining the combinations of Mandarin Rhymes. Hence, this thesis aims to figure out all the systematic restrictions within Mandarin rhymes, based on Taiwan accent, in terms of Optimality Theory, and establish a complete constraint ranking system.. 1.2 Field of Research. 立. 政治大. 1.2.1 The Domain of Rhyme. ‧ 國. 學 ‧. The domain of rhyme in this thesis includes the prenuclear glide, the vowel, and the nasal coda or the off-glide. Specifically, a glide-initial syllable is viewed as a zero-. y. Nat. io. sit. onset syllable in this thesis, where the whole syllable is its rhyme, for it is observed that. n. al. er. all glide-initial syllables in Mandarin can be preceded by another consonant to form a. Ch. new syllable, as the examples shown in (1).. engchi. i Un. v. (1) Examples of glide-initial and true-consonant-initial syllables. 4. a. yan [jen] ‘eye’. tian [thjen] ‘to lick’. b. yao [jɑw] ‘medicine’. miao [mjɑw] ‘temple’. c. wen [wən] ‘warm’. tun [thwən] ‘to swallow’. To analyze more accurately, the off-glides [j] and [w] are treated as codas in this thesis instead of parts. of the nuclei. However, the term “diphthong” is still used to call a sequence of a vowel and an off-glide. The use of the term “triphthong” is in the same way. For example, [ej] and [ɑw] are diphthongs; [wej] and [jɑw] are triphthongs. 2. DOI:10.6814/NCCU202000482.

(12) d. wei [wej] ‘tiny’. sui [swej] ‘to follow’. e. yong [joŋ] ‘servant’. xiong [ɕjoŋ] ‘elder brother’. f. yue [ɥe] ‘approximate’. xue [ɕɥe] ‘boot’. That is, a true-consonant-initial syllable and a glide-initial syllable only differ in whether there is a consonant filling in the C slot. To sum up, only syllable-initial consonants are not included in the discussion. The domain that is discussed in this thesis is illustrated in (2).. 政治大. (2) Illustration of the domain of rhyme (O = Onset, R = Rhyme) b. O C. G V X j e n yan ‘cigarette’. io. X n. an ‘safe’. n. al. G. Ch. engchi. R V a. X n. y. Nat. C G V s a san ‘three’. 1.2.2 Dialect. d. O C. sit. R. er. X n. qian ‘thousand’ c. O. R. ‧. ‧ 國. G j. 立. R V e. 學. a. O C tɕh. i Un. v. Mandarin is widely spoken all over the world, and therefore there are a lot of different dialects and accents. To specify, we will analyze Mandarin rhymes based on Taiwan accent. The different rhymes between Taiwan accent and Beijing accent are listed in (3).. 3. DOI:10.6814/NCCU202000482.

(13) (3) Special rhymes in Taiwanese Mandarin TM rhyme. Example. Gloss. Beijing accent. a. ə. [kə]. ge ‘song’. [kɤ]. b. iŋ. [ɕiŋ]. xing ‘star’. [ɕj ŋ]. c. oŋ. [(w)oŋ]. weng ‘old man’. [w ŋ]. [toŋ]. tung ‘winter’. [toŋ]. In Taiwan accent, a mid vowel appearing in the rhyme domain alone surfaces as [ə], which is similar to the mid central vowel preceding a nasal coda, illustrated in (3a).. 政治大. The spectrograms in (4 – 5) produced by the praat serve to support this observation.. 學. ‧ 國. 立. (4) The spectrograms of Taiwan accent. b. Tese ‘characteristic’ (F2 = 1399). ‧. a. Lengqi ‘air conditioner’ (F2 = 1396). n. er. io. sit. y. Nat. al. Ch. engchi. i Un. v. (5) The spectrograms of Beijing accent a. Lengqi ‘air conditioner’ (F2 = 1485). b. Tese ‘characteristic’ (F2 = 1348). 4. DOI:10.6814/NCCU202000482.

(14) In (4), the F2’s, which indicate the backness of a vowel, are close, which means that the vowels in the syllables leng and se are the same in Taiwan accent. On the other hand, in (5), the F2’s of the vowels in leng and se have a great distance, meaning that they are two different vowels. As for (3b), some may argue that [in] and [iŋ] have been neutralized in Taiwan accent. In spite of this fact, both [in] and [iŋ] are still heard in the dialect even though speakers may not choose the “correct” coda. For example, both xin ‘new’ and xing ‘star’ are able to be pronounced as [ɕin] or [ɕiŋ]. Such phenomenon of neutralization is also seen when it comes to the rhymes [ən] and [əŋ] (Fon et al, 2011).. 政治大. In Beijing dialect, the rhyme [oŋ] in (3c) must follow a consonantal onset.. 立. Otherwise, it will be pronounced as [wəŋ] (Lin, 2007), as shown in (3). However, there. ‧ 國. 學. is no such distinction for the rhyme [oŋ] in Taiwan dialect. Though sometimes. ‧. Taiwanese Mandarin speakers pronounce the world weng as [woŋ], it is undoubted that in this case the initial [w] is syllabified to the onset position since [woŋ] cannot follow. y. Nat. io. sit. any consonant. Therefore, the rhyme part of this syllable is still [oŋ].. n. al. er. In addition to the differences listed above, there is still another special part worth. Ch. i Un. v. mentioning. That is, the –r suffix is absent in Taiwanese Mandarin. Therefore, the –r. engchi. suffix is not going to be discussed. On the other hand, the rhyme [ɚ] (pinyin: er, Bopomofo: ㄦ) do exist in Taiwan dialect. However, it can only solely appear in a syllable without any other segments. Due to its special phonological status, this rhyme will not be discussed, either.. 1.3 Thesis Layout. This thesis consists of five chapters. Chapter One introduces the basic background of Mandarin syllable structure and Mandarin rhymes. Afterwards, we also specify what 5. DOI:10.6814/NCCU202000482.

(15) is going to be discussed in this thesis. In Chapter Two we introduce some theoretical backgrounds and review previous research on Mandarin phonotactic restrictions. Chapter Three lists the patterns of Taiwanese Mandarin rhymes and generally describes the patterns. Main analysis lies in Chapter Four. We first analyze the rhymes without nasal codas, and then go on to the rhymes with a nasal coda. The last chapter sums up the constraint ranking arranged in Chapter Four and ends up with further issues.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i Un. v. 6. DOI:10.6814/NCCU202000482.

(16) Chapter 2 Theoretical Background and Literature Review 2.1 Optimality Theory. Different from traditional rule-based phonology, Optimality Theory (henceforth OT), first developed by Prince and Smolensky (1993), emphasizes that it is a string of constraints that confines the surface forms of a certain language. A typical example proposed by Kager (2004) is illustrated in (1) below.. 政治大. 立. *VNASAL. *!. IDENT-IO(nasal). *. al. er. io. sit. y. *. Nat. b. → sæ̃nd. *VORALN. ‧. a. sænd. ‧ 國. /sænd/. 學. (1) /sænd/ → [sæ̃nd] (Kager, 2004). n. The constraints can be paraphrased as in (2 – 4), in terms of McCarthy’s (2008) definition formula.. Ch. engchi. i Un. v. (2) *VORALN Assign one violation mark for every oral vowel preceding a nasal segment.. (3) *VNASAL Assign one violation mark for every nasalized vowel.. 7. DOI:10.6814/NCCU202000482.

(17) (4) IDENT-IO(nasal) Assign one violation mark for every segment in the output that has a different value for the [nasal] feature from its correspondent input.. 2.1.1 Constraint Ranking. As is shown in (1), there is a certain order for the string of constraints, which is called constraint ranking. Constraints ranked higher are less violable than those ranked lower. In example (1), the highest ranked constraint is *VORALN, and then. 政治大. *VNASAL. The lowest one is IDENT-IO(nasal). Such constraint ranking is represented as. 立. 學. ‧ 國. below:. *VORALN >> *VNASAL >> IDENT-IO(nasal). ‧. If the constraint is violated, an asterisk is marked in the tableau. An exclamation mark following the asterisks indicates that the candidate is ruled out by that constraint.. y. Nat. io. sit. Let’s take a look at the example in tableau (1). Candidate (a) [sænd] violates the. n. al. er. constraint *VORALN since there is an oral vowel [æ] preceding the nasal [n]. Therefore,. Ch. i Un. v. an asterisk is marked to indicate the violation. Likewise, candidate (b) [sæ̃nd] violates. engchi. the other two constraints since there is a nasal vowel [æ̃] in the output and the vowel has different values for the [nasal] feature between its input and output. Asterisks are therefore marked in the correspondent boxes. For these three constraints, Candidate (a) only violates one constraint while candidate (b) violates two. However, *VORALN, which is violated by candidate (a) and obeyed by candidate (b), is ranked the highest among the three; therefore, candidate (a) is first ruled out. An exclamation mark is thus added after the asterisk. The arrow preceding candidate (b) indicates that it is the “winner.” The shadowed boxes indicate that the constraints are not active because the winner has been selected. 8. DOI:10.6814/NCCU202000482.

(18) The constraint ranking can also be illustrated as a Hasse Diagram (McCarthy, 2008) as (5) below.. (5) The Hasse Diagram of the example *VORALN. *VNASAL. IDENT-IO(nasal). 立. 政治大. In diagram (5), the vertical lines indicate domination, where we can interpret this. ‧ 國. 學. diagram as “*VORALN dominates *VNASAL, and *VNASAL dominates IDENT-IO(nasal).”. high.. ‧. There might be some cases where there are two or more constraints ranked the same. y. Nat. io. sit. Different constraint domination relations are illustrated in (6) below. (6a) indicates. n. al. er. that both MAX and DEP dominate ONSET, but there is no domination relation between. Ch. i Un. v. MAX and DEP. (6b) indicates that *V# dominates both IDENT(long) and MAX, but there is. engchi. no domination relation between IDENT(long) and MAX. (6c) indicates that both IDENT(long) and MAX are dominated by *COMP-SYLL, *Cunsyll and MAX-C simultaneously, but there is no domination relation between IDENT(long) and MAX and among *COMPSYLL, *Cunsyll and MAX-C.. 9. DOI:10.6814/NCCU202000482.

(19) (6) Hasse Diagrams of different constraint domination relations (McCarthy, 2008) a. MAX, DEP >> ONSET MAX. DEP. ONSET b. *V# >> IDENT(long), MAX *V#. IDENT(long). MAX. 政治大. c. *COMP-SYLL, *Cunsyll, MAX-C >> IDENT(long), MAX. 立 *C. *Cunsyll. OMP-SYLL. MAX-C. MAX. ‧. ‧ 國. 學. IDENT(long). y. Nat. n. er. io. al. sit. 2.1.2 Categories and Definitions of the Constraints. Ch. i Un. v. To take a view of the definitions of constraints (2 – 4), it is found that constraint. engchi. (4) confines the output with a different aspect from constraints (2) and (3). Constraints (2) and (3) define what should a surface form of this language be like, which have nothing to do with the input. This category of constraints is called markedness constraint. Markedness constraints do not mention the input form or the input-output mapping. They focus on the presence of the output form (McCarthy, 2008). Constraint (4) mentions the input-output mapping, which is different from constraints (2) and (3). This category of constraint is called faithfulness constraint. McCarthy and Prince (1995/1999) propose the correspondence theory, which states that each candidate that is evaluated supplies an output form that has correspondence 10. DOI:10.6814/NCCU202000482.

(20) relations with the input. Faithfulness constraints focus on the correspondence relations between the input and the output. In some situations, one constraint can be violated not only once by a single candidate. An example proposed by McCarthy (2008) is shown in (7).. (7) /bad/ → [bat] (McCarthy, 2008) /bad/ a. → bat b. bad. *VOICE. IDENT([voice]). *. *. **!. 立. 政治大. In tableau (7), constraint *VOICE, prohibiting voiced consonants to surface in the. ‧ 國. 學. output, is violated by candidate (b) twice because [b] and [d] are both voiced. However,. ‧. it is necessary to carefully define how violation marks should be given. Think that if. y. Nat. we consider that constraint *VOICE directly rules out candidates that have voiced. er. io. sit. consonants, the winner of example (7) will be candidate (b) because both candidates violate *VOICE but only candidate (a) violates IDENT([voice]). The correct output is not. n. al. Ch. i Un. v. predicted in this case. To avoid this situation, McCarthy (2008) advises that a constraint. engchi. be defined starting with “Assign one violation mark for every…” This thesis also adopts this suggestion to define every active constraint in the paper.. 2.1.3 Local Conjunction and Self Conjunction. In OT, two constraints can be conjoined together to form a local conjunction. When a local conjunction conjoins constraint A and constraint B, it is violated only if the candidate violates both constraints A and B in a given domain δ. For simplification, 11. DOI:10.6814/NCCU202000482.

(21) a local conjunction conjoining constraint A and constraint B in domain δ is noted as {A&B}δ in this thesis. Universally, {A&B}δ is ranked higher than both A and B. Self-conjunction, like local conjunction, is a conjoined constraint which is violated when a certain constraint is violated twice in a given domain. Tableau (8) is an example of self-conjunction proposed by Ito and Mester (1998). A self-conjunction of constraint A in domain δ is noted as A2δ. Like local conjunction, A2δ is universally ranked higher than A, usually with another constraint ranked in between.. (8) /bhidh/ → [bhid] or [bidh] (hypothetical) (Ito and Mester, 1998)5 /bhidh/. 立. ** *. ‧. * **!. * *. n. 2.2 Previous OT Analyses a of Mandarin. er. io. sit. y. Nat. d. bid. *!. 學. c. → bhid. *ASP. DENT. ‧ 國. a. bhidh b. → bidh. 政治I 大 ([asp]). *ASP2σ. iv l C n hengchi U. 2.2.1 The Issue of Mid Vowel Assimilation. Previous research has done some analyses about the phonotactic restrictions of Mandarin. Duanmu (2007) has analyzed the phonotactic restrictions of Mandarin rhymes through OT. In his analysis, he derives two constraints to evaluate Mandarin rhymes: NC-Harmony and GN-Harmony. The former indicates that the nucleus and the. 5 ASP stands for aspiration. In OT, if there are two winners among the candidates, they might indicate different speech variations. 12. DOI:10.6814/NCCU202000482.

(22) coda of a same syllable should share the same value for [back] and [round] feature. The latter one indicates that the nucleus and the prenuclear glide of a same syllable should share the same value for [back] and [round] feature. Between the two constraints, NCHarmony is ranked higher than GN-Harmony. The examples by Duanmu are shown in (9) and (10) below.. (9) /ɥə/ → [ɥe] (Duanmu, 2007) /ɥə/. NC-Harmony. a. ɥə b. → ɥe. {GN-[back], GN-[round]}. 政治大. 立. c. ɥø. Avoid-[ø]. *. *. *!. ‧ 國. 學. c.. woj. *. *!. Ch. engchi. sit. y. al. er. → wej. *. n. b.. *!. io. wəj. GN-Harmony. Nat. a.. NC-Harmony. ‧. (10) /wəj/ → [wej] (Duanmu, 2007) /wəj/. *!. i Un. v. In tableau (9), the input /ɥə/ has no coda, so NC-Harmony is not violated by any candidates. This situation is called vacuous satisfaction, and the definition is in (11) below.. 13. DOI:10.6814/NCCU202000482.

(23) (11) The definition of “vacuous satisfaction” (McCarthy, 2008) When a candidate does not contain a structure that a certain constraint is against, it is stated that the constraint is vacuously satisfied by the candidate. For instance, an open syllable does vacuously satisfy a constraint requiring the coda to be voiceless. In OT analysis, vacuous satisfaction is treated the same as nonvacuous satisfaction, though.. Let’s get back to example (9). All candidates satisfy NC-Harmony, so none is ruled out. Among the three candidates, candidate (c) completely satisfies GN-Harmony, but. 政治大. it contains a front rounded vowel [ø], which is not licensed in Mandarin. Therefore, it. 立. is ruled out by a higher-ranked constraint Avoid-[ø]. As for candidates (a) and (b), (a). ‧ 國. 學. violates GN-Harmony twice because the prenuclear glide and the nucleus has different. ‧. values for both [back] and [round] features while (b) only violates GN-Harmony once, with its prenuclear glide and nucleus only differing in [round] feature. Therefore,. y. Nat. io. sit. candidate (b) is the best choice for the output form. Note that Avoid-[ø] must be ranked. n. al. er. higher than GN-Harmony, or candidate (c) will be wrongly predicted as the winner.. Ch. i Un. v. Example (10) illustrates how NC-Harmony is ranked higher than GN-Harmony.. engchi. Comparing candidates (b) and (c), the former violates GN-Harmony while the latter NC-Harmony. The attested output is (b) [wej], so NC-Harmony is less violable. Lin (2015) also analyzes the mid vowel assimilation phenomenon of different Mandarin dialects with OT approach. The research states that different dialects have different constraint rankings. Among them, the constraint ranking of Standard Mandarin is shown in (12).. 14. DOI:10.6814/NCCU202000482.

(24) (12) The constraint ranking about mid vowel assimilation of Standard Mandarin (Lin, 2015)6 *ø >> RIME-HARMONY, GV-HARMONY >> *o >> *e. Likewise, Lin ranked the constraint *ø on the top to avoid such segment. Although Lin accidentally ranked RIME-HARMONY and GV-HARMONY the same, the used constraints are almost identical to those used by Duanmu. Both of them high-rank a constraint to restrict that V and X in a same syllable have identical values for the [back] (and also [round]) feature. However, it is observed that there are different phenomena when the. 政治大. X slot is filled with a nasal coda or an off-glide, as is illustrated in (13).. 立. ‧ 國. 學. (13) The mid vowel assimilation when X is a nasal coda or an off-glide. y. sit. al. er. [n].. n. /əŋ/ → [əŋ]. /ə/ remains as a central vowel despite the following [-back] nasal. io. /əw/ → [ow]. NC-Harmony.. Nat. /ən/ → [ən]. /ə/ becomes [e] because of the [-back] feature of [j], which satisfies. ‧. /əj/ → [ej]. Ch. i Un. v. /ə/ becomes [o] because of the [+back] feature of [w], which. engchi. satisfies NC-Harmony.. /ə/ remains as a central vowel, which still satisfies NC-Harmony because of the following [+back] nasal [ŋ].. According to (13), it is assumed that the mid vowel has a different requirement of assimilation to a nasal coda from that to an off-glide. Therefore, it is necessary to re-. 6. Lin defines the term “rime” as the structure of VX. Therefore, her RIME-HARMONY constraint has the. same definition as Duanmu’s NC-Harmony in (9) and (10). 15. DOI:10.6814/NCCU202000482.

(25) rank the constraints or introduce new ones in order to fulfill the two different phenomena in table (13). Lin (2007) has also discussed about the difference of the mid vowel assimilation when it is preceded by a nasal coda or an off-glide. The problem does not lie in the segment filled in the X slot itself because it is observed that low vowels also assimilate to nasal codas. For example, shan [ʂan] ‘mountain’ and shang [ʂɑŋ] ‘injury’ are allowed, but *[ʂaŋ] and *[ʂɑn] are not. Therefore, it is unreasonable to argue that nasal codas do not trigger the assimilation. To solve this problem, Lin proposes a hypothesis: there are two main kinds of VX structure in Mandarin; one is diphthong and the other is. 政治大. “high/low vowel + nasal.” She excludes the structure of “mid vowel + nasal” in the. 立. underlying form. In her theory, a mid vowel only precedes a nasal coda due to the. ‧ 國. 學. condition and rule described in (14). There are no mid vowels preceding a nasal coda. ‧. in the underlying forms.. y. Nat. n. al. er. io. a. Environment. sit. (14) The [ə]-insertion condition. Ch. i Un. v. When a high vowel and its following nasal coda have different values for the. engchi. [back] feature, an [ə] is inserted between them. b. Rule ∅→ə/. -cons αbk. __. +nas βbk. c. Examples (The glide formation rule is omitted.) (i). /lun/ → [lwən] ‘wheel’. (ii). /piŋ/ → [pjəŋ] ‘ice’. 16. DOI:10.6814/NCCU202000482.

(26) Lin’s hypothesis seems to successfully explain why mid vowels do not assimilate to nasal codas. However, she does not clearly explain why there are syllables like shen [ʂən] ‘body’ and seng [səŋ] ‘monk’, where there are no high vocoids in the syllables to trigger the [ə]-insertion rule. Another problem lies in the example (ii) in (14c). In Taiwan accent, the word bing ‘ice’ is pronounced as [piŋ]. Other words that have the same rhyme are also pronounced this way. This pronunciation violates Lin’s RIME-HARMONY constraint in (12) and her [ə]-insertion rule in (14) because [i] is [-back] while [ŋ] is [+back]. Therefore, it is necessary to re-rank the constraints or propose new ones to fulfill the dialect spoken in Taiwan.. 學. ‧ 國. 立. 政治大. 2.2.2 The Issue of Low Vowel Raising. ‧. Low vowel raising in Mandarin is a special phonological phenomenon that it is. y. Nat. n. al. er. io. in (15).. sit. hard to find another similar rule in other contexts. The simplified formula is illustrated. (15) The low vowel raising rule. Ch. engchi. i Un. v. a → e / j, ɥ __ n. Aiming at this phonological phenomenon, Hsieh (2012) has analyzed it via rulebased phonology and OT. In his research, he proposes a self-conjunction constraint called “AGREE[back]2” to rule out the ill-formed *[jan] and *[ɥan], and successfully predicts the well-formed [jen] and [ɥen]. The definition of the constraint is shown in (16). 17. DOI:10.6814/NCCU202000482.

(27) (16) AGREE[back]2 (Flemming, 2003; Hsieh, 2012) Assign one violation mark for every syllable only if where there is a vowel not having the same value for the [back] feature as both of its adjacent consonants.. In Hsieh’s research, constraint (16) rules out *[jan] but is satisfied by [jen]. He follows the SPE tradition to regard the low front vowel [a] as [+back]. This theory does support constraint (16) to rule out *[jan] because [a] is [+back] while [j] and [n] are [back]. However, if we specify [a] as [+back], other problems might be caused, as elaborated below.. 政治大. First, the mid vowel assimilation and low vowel assimilation in diphthongs are no. 立. longer the same issue. As what is mentioned in section 2.2.1, the nucleus must have the. ‧ 國. 學. same value for the [back] feature as the off-glide. The fact is that we have the rhyme. ‧. [aj] (cf. [ej]) in Taiwanese Mandarin, which obviously violates the mentioned constraint if [a] is regarded as [+back]. Second, it will be difficult to explain why a low vowel is. y. Nat. io. sit. surfaced “even more retracted” as [ɑ] when preceding [+back] codas like [w] and [ŋ]. n. al. er. to form rhymes like [ɑw] and [ɑŋ] because [a] is already specified as [+back]. In other. Ch. i Un. v. words, now that both [a] and [ɑ] are both [+back], then what feature distinguishes these segments?. engchi. Due to these problems, it is not suitable to specify [a] as [+back] because it will lead to more obstacles for analysis, and the constraint AGREE[back]2, therefore, cannot rule out the ill-formed *[jan] and *[ɥan]. In order to take other phonological phenomena (e.g. [aj] and [ej] are composed of a front vowel and [j]; the two segments agree with each other in [back] in rhymes like [ɑw] and [ɑŋ]) into consideration, a new constraint instead of AGREE[back]2 should be proposed for low vowel raising.. 18. DOI:10.6814/NCCU202000482.

(28) Chapter 3 Patterns of Rhymes in Taiwanese Mandarin 3.1 Principles and Charts. This section lists all the rhymes that are used in Taiwanese Mandarin. The inventories of each position in the GVX structure are listed in (1).. (1) The list of phonetic segments for each position. X. ∅. j. w ɥ. (i, u, y). 立. a ɑ e ə. o. w n ŋ. ‧. j. 學. V. ∅. ‧ 國. G. 政治大. sit. y. Nat. As (1) shows, there are three glides can fill in the G slot, namely [j], [w], and [ɥ].. er. io. Since a syllable can have no prenuclear glide, this position can also be filled with a ∅.. There are multiple choices for the V slot. Among the high vowels [i], [u] and [y], only. al. n. iv n C the one correspondent to the prenuclear to [i], h eglide i U[w] to [u] and [ɥ] to [y]) can fill n g([j]c h in the V slot. As for the low vowels, [a] and [ɑ] are derived from a same phoneme /a/.. Mid vowels in the V position is also this case. The underlying form of the mid vowel is hypothesized as /ə/. The X slot can be filled with an off-glide [j]/[w], or a nasal coda [n]/[ŋ]. Like the G position, the X slot can also be filled with a ∅ to lead to an open syllable. Charts (2 – 5) illustrate all the existent and inexistent rhymes in Taiwanese Mandarin.. 19. DOI:10.6814/NCCU202000482.

(29) (2) G = ∅. V\X. j. w. n. ŋ. e. ∅. *e. ej. *ew. *en. *eŋ. ə. ə. *əj. *əw. ən. əŋ. o. *o. *oj. ow. *on. oŋ. a. a. aj. *aw. an. *aŋ. ɑ. *ɑ. *ɑj. ɑw. *ɑn. ɑŋ. n. ŋ. in. iŋ. je. *jej. *jew. *jə. *jəj. *jo ja. *jɑ. n. al. jen. *jeŋ. *jəw. *jən. *jəŋ. *joj. jow. *jon. joŋ. *jaj. *jaw. *jaŋ. *jɑj. jɑw. Ch. engchi U. y. ɑ. io. a. Nat. o. *iw = *ju. ‧. ə. ij (ji) = i. i. sit. e. 立. 學. i. ∅. ‧ 國. V\X. 治政 j w 大. *jan. er. (3) G = j. v ni. *jɑn. jɑŋ. (4) G = w V\X. j. w. n. ŋ. u. ∅ u. *uj = *wi. uw (wu) = u. *un. *uŋ. e. *we. wej. *wew. *wen. *weŋ. ə. *wə. *wəj. *wəw. wən. *wəŋ. o. wo. *woj. *wow. *won. *woŋ. a. *wa. waj. *waw. wan. *waŋ. ɑ. wɑ. *wɑj. *wɑw. *wɑn. wɑŋ. 20. DOI:10.6814/NCCU202000482.

(30) (5) G = ɥ V\X. j. w. n. ŋ. y. ∅ y. *yj = *ɥi. *yw = *ɥu. yn. *yŋ. e. ɥe. *ɥej. *ɥew. ɥen. *ɥeŋ. ə. *ɥə. *ɥəj. *ɥəw. *ɥən. *ɥəŋ. o. *ɥo. *ɥoj. *ɥow. *ɥon. *ɥoŋ. a. *ɥa. *ɥaj. *ɥaw. *ɥan. *ɥaŋ. ɑ. *ɥɑ. *ɥɑj. *ɥɑw. *ɥɑn. *ɥɑŋ. 治政大 in Taiwanese Mandarin. Charts (2 – 5) list all the present and absent rhymes 立 However, some inexistent rhymes might be controversial, which will be clarified in the ‧. ‧ 國. 學. next section.. sit. n. al. er. io. 3.2.1 [e] and [o]. y. Nat. 3.2 Controversial Rhyme Structures. Ch. engchi. i Un. v. The mid vowels [e] and [o] (see the leftmost column in chart (2)) only appear in interjections or particles in Taiwanese Mandarin no matter they are in onsetless syllables or follow a consonant. Never do these rhymes appear in lexical words. Therefore, the surface of the rhymes [e] and [o] is viewed as exceptions in this thesis, which are still considered inexistent.. 21. DOI:10.6814/NCCU202000482.

(31) 3.2.2 The Low Front Vowel [a]. According to Lin (2007), the low vowel [a] is considered [-back] in Mandarin since there are only two low vowels, [a] and [ɑ], in the patterns, and [a] is more advanced than [ɑ]. When there is no other segment in the rhyme domain except for a low vowel, the front [a] is surfaced (e.g. ba [pa] ‘eight’; sha [ʂa] ‘sand’), as is shown in the leftmost column in chart (2). Therefore, the low front vowel [a] is viewed as the underlying form of the low vowel.. 3.2.3 [jaj]. 立. 政治大. ‧ 國. 學. The rhyme [jaj] in chart (3) is considered absent in Taiwanese Mandarin. Some. ‧. may argue that this rhyme appears in the word yai [jaj] ‘cliff’. However, this is the only word that contains this rhyme. Hsieh (2012) states that Mandarin speakers tend to. y. Nat. io. sit. pronounce this word as [ja] in casual speech while Duanmu (2007) thinks people. n. al. er. pronounce this word as either [ja] or [aj].. Ch. i Un. v. In order to confirm how Taiwanese Mandarin speakers pronounce this word in. engchi. casual speech, a simple experiment has been done. Six native speakers were asked to read several sentences casually, and one of them contained the word yai. Consequently, five of the participants pronounced the word as [aj], and the other pronounced it as [jaj], but further admitted that this is actually a word hard to pronounce when the conductor made sure whether she really pronounced this way in casual speech. Though this experiment, different from what has been proposed by Hsieh and Duanmu, does not attest the output form of [ja], it proves that [jaj] can be reasonably viewed as an exception. 22. DOI:10.6814/NCCU202000482.

(32) 3.3 A General Description. According to previous analyses (Hashimoto, 1970; Duanmu, 2007; Lin, 2007), there are only one mid vowel phoneme and one low vowel phoneme in Mandarin. That is, different mid vowels in Mandarin are derived from a same phoneme in the underlying form, and so are different low vowels. With this theoretical knowledge, it can be assumed that a mid (low) vowel will not have minimal pairs with one another. For example, among [wej], *[wəj] and *[woj] in chart (4), only one of them is existent because the three mid vowels should not have minimal pairs. Chart (6) labels the. 政治大. “phonemic group” with bold lines based on this theory. In (6), a block squared by bold. 立. lines should theoretically have only one legal pattern.. ‧ 國. 學. ə o. w. n. sit. y. ŋ. er. al. n. e. io. (i, u, y). j. ∅. Nat. V\X. ‧. (6) Phonemic group indication chart (G = any). Ch. engchi. i Un. v. a ɑ. For example, if we apply chart (6) to the rhyme patterns without a prenuclear glide (G = ∅), chart (7) is obtained as below. Theoretically, each bold-lined square should. have only one pattern allowed. Therefore, among [ej], *[əj] and *[oj], only [ej] is allowed. Likewise, [ɑw] is allowed while *[aw] is not. If this rule is violated, then there 23. DOI:10.6814/NCCU202000482.

(33) should be a problem to be dealt with. For example, in chart (7), both [əŋ] and [oŋ] are allowed even though they are in the same bold-lined square. This irregular phenomenon thus deserves discussions.. (7) Phonemic group indication chart (G = ∅) V\X. w. n. ŋ. *e. ej. *ew. *en. *eŋ. ə. ə. *əj. *əw. ən. əŋ. o. *o. *oj. *on. oŋ. a. a. ow 治政 aj *aw 大. an. *aŋ. ɑ. *ɑ. *ɑj. *ɑn. ɑŋ. ‧ 國. 立. ɑw. 學. j. e. ∅. ‧. However, this way of division is still too general. Remember that in section 2.2.1,. sit. y. Nat. it has been mentioned that Lin (2007) states that mid vowels preceding a nasal coda are. er. io. inserted, whose underlying form is in fact a null, as is shown in (8). Therefore, the. al. border between V = ∅ and the mid vowels in the nasal-coda columns should be broken.. n. iv n C U the broken ones. hthe Chart (9) is the revision of (6), where e ndotted h i indicate g clines (8) The [ə]-insertion rule ∅→ə/. -cons αbk. __. +nas βbk. 24. DOI:10.6814/NCCU202000482.

(34) (9) Phonemic group indication chart (G = any) (revised) V\X (i, u, y). j. ∅. w. n. ŋ. e ə. o a ɑ. 政治大 If we adopt the model in chart (9) to each chart in (2 – 5), several counterexamples 立. ‧ 國. 學. may appear:. (i) In chart (2), both [əŋ] and [oŋ] are allowed. The two mid vowels have minimal. ‧. pairs.. sit. y. Nat. (ii) In chart (3), no mid vowels can be put between two [j]’s. *[jej], *[jəj] and *[joj]. al. er. io. are all not allowed. This seems to indicate that a mid vowel cannot surface in such. v. n. environment. The same case is found when it comes to the low vowels. Neither [a] nor. Ch. [ɑ] can appear between two [j]’s.. engchi. i Un. (iii) In chart (3), both [in] and [jen] are allowed. This seems to violate Lin’s [ə]insertion rule (see 2.2.1). However, no low vowels can be between [j] and [n]. The similar case is found when it comes to the pair [iŋ] and [joŋ], whereas we have [jɑŋ] this time. (iv) Like (ii), neither mid vowels nor low vowels can appear between to [w]’s in chart (4). (v) In chart (4), no matter whether a mid vowel is inserted or not, *[uŋ], *[weŋ], *[wəŋ] or *[woŋ] are all absent in the patterns. This phenomenon is also observed in 25. DOI:10.6814/NCCU202000482.

(35) the patterns *[yŋ], *[ɥeŋ], *[ɥəŋ] and *[ɥoŋ] in chart (5). (vi) Like (iii), both [yn] and [ɥen] are allowed. However, no low vowels appear between [ɥ] and [n] in chart (5). In fact, not only when there is a coda [n], [ɥ] cannot precede any low vowels. (vii) When there is a glide coda [j] or [w], neither mid vowels nor low vowels can surface after [ɥ] in chart (5). That is, [ɥ] cannot precede any diphthongs. (viii) Two adjacent high vocoids like *[ju] in (3), *[wi] in (4), and *[ɥi] and *[ɥu]7 in (5) are all unallowed. To make it clearer, (10) lists these violations of the phonemic group division in a brief way.. 立. 政治大. (10) The list of violations of the phonemic group division. ‧ 國. 學. a. [əŋ] and [oŋ] are both allowed. (see (2)). ‧. b. *[jej], *[jəj] and *[joj] are all not allowed. (see (3)) c. *[jaj] and *[jɑj] are both not allowed. (see (3)). y. Nat. n. al. Ch. f. [iŋ] and [joŋ] are both allowed. (see (3)). engchi. er. io. e. *[jan] and *[jɑn] are both not allowed. (see (3)). sit. d. [in] and [jen] are both allowed. (see (3)). i Un. v. g. *[wew], *[wəw] and *[wow] are all not allowed. (see (4)) h. *[waw] and *[wɑw] are both not allowed. (see (4)) i. *[uŋ], *[weŋ], *[wəŋ] and *[woŋ] are all not allowed. (see (4)) j. *[ɥa] and *[ɥɑ] are both not allowed. (see (5)). 7. Based on Hashimoto’s theory (see 3.1), it is the prenuclear glide instead of the postnuclear one that. spreads to the V position. Therefore, the sequence of two high vocoids is more properly transcribed as “vowel + glide.” However, here we transcribe a sequence of two high vocoids as “glide + vowel”, for it is more traditional and easier to read. Also, the form of “glide + vowel” is assumed to be the underlying form of a sequence of two high vocoids. To avoid being controversial both of the transcriptions are available in chart (2 – 5). 26. DOI:10.6814/NCCU202000482.

(36) k. *[ɥej], *[ɥəj] and *[ɥoj] are all not allowed. (see (5)) l. *[ɥaj] and *[ɥɑj] are both not allowed. (see (5)) m. *[ɥew], *[ɥəw] and *[ɥow] are all not allowed. (see (5)) n. *[ɥaw] and *[ɥɑw] are both not allowed. (see (5)) o. [yn] and [ɥen] are both allowed. (see (5)) p. *[ɥan] and *[ɥɑn] are both not allowed. (see (5)) q. *[yŋ], *[ɥeŋ], *[ɥəŋ] and *[ɥoŋ] are all not allowed. (see (5)) r. *[ɥaŋ] and *[ɥɑŋ] are both not allowed. (see (5)) s. Each of *[ju], *[wi], *[ɥi] and *[ɥu] is not allowed. (see (2 – 5)). 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i Un. v. 27. DOI:10.6814/NCCU202000482.

(37) Chapter 4 An OT Analysis In this chapter, an OT analysis will be displayed and divided into two sections. First we will discuss about rhymes without a nasal coda, including the following structures: V, GV, VG, GVG. The other section will focus on rhymes containing a nasal coda, including the structures of VN and GVN. For the structural indicating capital letters, the G preceding V indicates the prenuclear glides and the G following V indicates the postnuclear ones. For other symbols, V stands for vowel and N stands for nasal coda.. 學. ‧ 國. 立. 政治大. 4.1 Rhymes without a Nasal Coda. ‧ er. io. sit. y. Nat. 4.1.1 Mid/Low Vowel Assimilation. As is mentioned in section 2.2.1, the phenomenon of mid/low vowel assimilation. al. n. iv n C has been widely discussed in previous h e nresearch i U 2007; Lin, 2015). Let’s g c h(Duanmu, review their constraints about this phonological phenomenon in (1 – 2).. (1) NC-Harmony (Duanmu, 2007) or RIME-HARMONY (Lin, 2015) Assign one violation mark for every syllable whose nucleus and coda have different values for [back] and [round] features.. 28. DOI:10.6814/NCCU202000482.

(38) (2) GN-Harmony (Duanmu, 2007) or GV-HARMONY (Lin, 2015) Assign one violation mark for every syllable whose nucleus and prenuclear glide have different values for [back] and [round] features.. Both of these analyses mention the [back] and [round] features in a single constraint. However, in order to avoid predicting the wrong outputs *[ɥə], Duanmu assigns two violation marks for each of the two features, as (3) shows.. (3) /ɥə/ → [ɥe] (Duanmu, 2007) /ɥə/. 立. a. ɥə. *!. ‧ 國. *!. *. Nat. y. ‧. c. ɥø. *. 學. b. → ɥe. 治 {GN-[back], GN-[round]} 政 Avoid-[ø] 大. NC-Harmony. er. io. sit. The analysis in (3) rules out *[ɥə] because [ɥ] and [ə] do not agree in not only [round] feature but also [back] feature. Therefore, I think it will be better if we separate. al. n. iv n C the constraint into two, one assigning mark for the [back] feature and the h ea violation ngchi U other for the [round] feature.. Moreover, in Duaumu’s analysis, there is a high-ranked constraint “Avoid-[ø]” to prevent the ill-formed *[ɥø] from surfacing out. In Taiwanese Mandarin though, in addition to [ø], we still need to avoid segments like [ɤ] or [ɒ]. Therefore, we can just high-rank a constraint to avoid all the segments that are not listed in the inventory instead of introducing such specific constraints, which is displayed in (5). Constraints (6 – 9) are separated from Duaumu’s and Lin’s based on the different features. Constraint (10) is developed to ensure it is the vowel instead of the glide that changes 29. DOI:10.6814/NCCU202000482.

(39) its feature. Constraint (11) is the violated faithfulness constraints. The descriptive generalization is in (4).. (4) The descriptive generalization for mid/low vowel assimilation The vowel should agree with the prenuclear glide in [back] and [round] features, but if there is a glide coda following it, it should agree with the following glide. This requirement is enforced by vowel fronting/backing and rounding.. (5) PHONOTACTICS. 政治大. Assign one violation mark for every ø, ɶ, ɯ, ɤ, ɒ and ɰ in the output.. 立. ‧ 國. 學. (6) AGREE[bk]-VG. ‧. Assign one violation mark for every syllable whose vowel and glide coda do not agree in the [back] feature.. n. al. er. io. sit. y. Nat (7) AGREE[rd]-VG. Ch. i Un. v. Assign one violation mark for every syllable whose vowel and glide coda do not agree in the [round] feature.. engchi. (8) AGREE[bk]-GV Assign one violation mark for every syllable whose vowel and prenuclear glide do not agree in the [back] feature.. (9) AGREE[rd]-GV Assign one violation mark for every syllable whose vowel and prenuclear glide do not agree in the [round] feature.. 30. DOI:10.6814/NCCU202000482.

(40) (10) IDENT[F]-G Assign one violation mark for every glide which has a different value for the [F] feature from its input, where [F] = [back], [round] or [nasal].. (11) IDENT[F]-V Assign one violation mark for every vowel which has a different value for the [F] feature from its input, where [F] = [back] or [round].. Since the [back] feature and the [round] feature should be agreed in together,. 政治大. constraints (6) and (7) are ranked the same high, and so are constraints (8) and (9).. 立. Constraint (5) is never violated, so it is ranked the highest. AGREE[F]-VG8 is always. ‧ 國. 學. prior to AGREE[F]-GV, so (6) and (7) are ranked higher than (8) and (9). In order to. ‧. satisfy the constraints above, the vowel, instead of the glide, changes its value, so (10) is ranked higher than (11). Tableaux (12 – 14) illustrate how these constraints work.. y. Nat. io. sit. The parenthesized exclamation marks indicate that either violation is fatal. The Hasse. n. al. er. Diagram is illustrated in (15).. (12) /ɥə/ → [ɥe] /ɥə/. PHONOTACTICS. Ch. engchi. AGREE[F]-VG [bk]. [rd]. i Un. AGREE[F]-GV [bk]. a. → ɥe. 8. [rd]. IDENT[F]-V. *. b. ɥə c. ɥø. v. (*!). *. (*!). *!. *. The F in the bracket stands for Feature, which can be substituted into [back] or [round]. 31. DOI:10.6814/NCCU202000482.

(41) (13) /jəw/ → [jow] /jəw/. PHONOTACTICS. AGREE[F]-VG. AGREE[F]-GV. [bk]. [bk]. [rd]. *. *. [rd]. a. → jow b. jəw. *!. c. jew. (*!). IDENT[F]-V *. *. (*!). *. (14) /aw/ → [ɑw] IDENT[F]-G. 立. [bk]. TACTICS. [rd]. IDENT[F]-V. * (*!). *. (*!). *!. *!. *. io. sit. y. Nat. d. aj. AGREE[F]-GV. ‧. c. ɒw. 治政 [bk] 大 [rd]. 學. b. aw. ‧ 國. a. → ɑw. AGREE[F]-VG. PHONO-. (15) The Hasse Diagram of mid/low vowel assimilation. n. al. er. /aw/. C h PHONOTACTICS U n i engchi. v. AGREE[bk]-VG. AGREE[rd]-VG. AGREE[bk]-GV. AGREE[rd]-GV. IDENT[F]-G. IDENT[F]-V. 32. DOI:10.6814/NCCU202000482.

(42) Example (12) shows how AGREE[bk]-GV and AGREE[rd]-GV both assign violation marks to rule out the candidate *[ɥə]. Example (13) proves that AGREE[F]-VG is ranked higher than AGREE[F]-GV, for candidate (13c) will be wrongly predicted as the winner if AGREE[F]-GV is ranked higher. Example (14) proves that PHONOTACTICS should be ranked higher than AGREE[F]-VG because it will be a problem to determine whether [ɑw] or *[ɒw] is the optimal candidate if these constraints are ranked the same high. Note that the constraint IDENT[F]-G is only illustrated in example (14) because satisfying this constraint will lead to [ɰ] in examples (12) and (13), which also violates the highranked constraint PHONOTACTICS. For simplicity, IDENT[F]-G is only listed when it is active.. 政治大. 立. These constraints and their ranking successfully predict the correct output from each. ‧ 國. 學. “phonemic group” (see 3.3). However, there are still some gaps that “should be legal”. ‧. throughout this constraint ranking. For example, given an input /wəw/, an optimal output *[wow] should be predicted, but it is not attested in Taiwanese Mandarin, as is. y. Nat. io. sit. illustrated in (16). Another problem lies in the rhymes with a high vowel following a. n. al. er. prenuclear glide. For example, given an input /ju/, this constraint ranking will predict. Ch. i Un. v. its output as [ji], which may be interpreted as a lengthened [i]. However, it seems not. engchi. to be the fact, as shown in (17). These two problems will be later discussed in sections 4.1.2 and 4.1.3.. (16) /wəw/ → *[wow] /wəw/. PHONOTACTICS. AGREE[F]-VG. AGREE[F]-GV. [bk]. [bk]. [rd]. [rd]. a. 💣 wow. IDENT[F]-V *. b. wəw. *!. *. 33. DOI:10.6814/NCCU202000482.

(43) (17) /ju/ → ?[ji] /ju/. PHONOTACTICS. AGREE[F]-VG. AGREE[F]-GV. [bk]. [bk]. [rd]. [rd]. IDENT[F]-V. a. 💣 ji. *. b. ju. (*!). (*!). c. jy. *!. *. 4.1.2 The Phonotactic Restriction of Triphthongs. 政治大. As what is mentioned in 4.1.1, rhymes such as *[jej], *[jaj] (see 3.2.3), *[wow],. 立. *[wɑw], *[ɥej], *[ɥaj], *[ɥow] and *[ɥɑw] should be surfaced out because all of these. ‧ 國. 學. patterns are actually the optimal choices throughout the constraint ranking. It seems. ‧. that there are some more constraints which are active to these triphthongs.. y. Nat. To observe these absent patterns, it is found that glide [j] cannot precede a. er. io. sit. diphthong with a coda [j], glide [w] cannot precede a diphthong with a coda [w], and glide [ɥ] cannot precede any diphthongs, as is simplified in (18) below.. n. al. Ch. (18) Triphthongs that are not allowed. engchi. i Un. v. a. *jVj b. *wVw c. *ɥVj d. *ɥVw. Regarding to this phenomenon, Duanmu (2007) has also proposed an analysis. In his opinion, there is a “dissimilation effect” interfering between the prenuclear glide 34. DOI:10.6814/NCCU202000482.

(44) and the postnuclear glide. In Taiwanese Mandarin (as well as other dialects), the prenuclear glide and the postnuclear glide should have different values for both of their [back] and [round] features. This is the reason why there are no such rhymes in Taiwanese Mandarin where the prenuclear glide and the postnuclear glide are identical, and [ɥ] cannot precede any diphthongs, for it is [-back] as [j] and [+round] as [w]. To deal with this dissimilation phenomenon, the Obligatory Contour Principle (henceforth OCP) is usually used. OCP was first developed by Leben (1973) for tonal dissimilation in African languages. The definition is shown in (19) below.. 政治大. (19) The Obligatory Contour Principle (Ito and Mester, 1998). 立. Adjacent identical elements are not allowed at the melodic level.. ‧ 國. 學 ‧. As (19) shows, OCP can only confine that two adjacent elements should not be identical (at least in the value of a certain feature). However, the problem going to be. y. Nat. io. sit. coped with here does not lie in an adjacent pairs, but two glides with a vowel in between.. n. al. er. To solve this problem, I refer to Ito and Mester’s (1998) interpretation to OCP. In their. Ch. opinion, OCP is in fact a self-conjunction effect.. engchi. i Un. v. Take a look at Hsiao’s (2000) example about tone sandhi in Southern Min. In Southern Min, only the last syllable in a tone sandhi domain remains its base tone; other syllables will undergo tone sandhi. This phenomenon seems to be able to be predicted with an OCP constraint as (20) shows, where T stands for base tone. However, this is awkward because OCP is used to prevent two adjacent segments from having identical “tone level” or “tone contour”, while the base tones for each syllable are not the same.. 35. DOI:10.6814/NCCU202000482.

(45) (20) OCP-T Assign one violation mark for every adjacent pair of segments which remain the base tone.. Nevertheless, if we adopt Ito and Mester’s interpretation to revise this OCP constraint as a self-conjunction, as (21) shows, the original definition of OCP will not be violated, and the constraint still remains its essence.. (21) *T2δ. 政治大. Assign one violation mark for every tone sandhi domain where there are two. 立. syllables of the base tone.. ‧ 國. 學 ‧. Adopting this theory, the dissimilation phenomenon of the two glides in a same syllable can be enforced by a self-conjunction. This way, we can rule out the ill-formed. y. Nat. io. sit. outputs without violating the original definition of OCP.. n. al. er. Reviewing the unallowed triphthongs listed in (18) and Duanmu’s analysis that. Ch. i Un. v. the prenuclear and postnuclear glides should not share the same value for either [back]. engchi. or [round] feature, we can introduce a self-conjunction to prohibit a syllable from having two glides that have the same value for the [back]/[round] feature, as is in (23), which is self-conjoined by constraint (22) below.. (22) *GαF Assign one violation mark for every glide of [αF] feature, where F = [back] or [round].. 36. DOI:10.6814/NCCU202000482.

(46) (23) *[G2αF]σ Assign one violation mark for every syllable which contains two glides of [αF] feature, where F = [back] or [round].. Constraint (23) successfully rules out all the ill-formed triphthongs. However, when these candidates are ruled out, there should be another candidate to be selected as an optimal output. To solve this problem, here we have to mention the experiment in 3.2.3 again. The result of the experiment shows that Taiwanese Mandarin speakers tend to pronounce the word yai ‘cliff’ [jaj] as [aj] instead of [ja], which indicates that it is the. 政治大. prenuclear glide that is dropped when there are two glides having the same value for. 立. [back] or [round] feature in a syllable. The descriptive generalization is shown in (24).. ‧ 國. 學. In addition to constraints (22) and (23) introduced above, there are still other constraints. ‧. active, as below. Constraint (25) indicates the strategy of deleting the prenuclear glide. Constraint (26) is also listed to show another possible strategy to repair the ill-formed. y. Nat. io. sit. triphthongs. In addition to the strategy of deleting the coda, changing the coda might. n. al. er. also be a method to satisfy the higher-ranked constraints. Therefore, IDENT[F]-G is also active here.. Ch. engchi. i Un. v. (24) The descriptive generalization for the phonotactic restriction of triphthongs The prenuclear glide and the postnuclear glide of a syllable should have different values for both [back] and [round] feature. This requirement is enforced by prenuclear glide deletion.. (25) MAX Assign one violation mark for every segment in the input which does not have a correspondent output.. 37. DOI:10.6814/NCCU202000482.

(47) (26) MAX-CODA Assign one violation mark for every coda in the input which does not have a correspondent output.. Constraints (25) and (26) both assign a violation mark for deletion; however, (26) has a stricter definition, which is only violated when a coda is deleted, while (25) is violated no matter which segment does not surface out. Logically, it is known that when a candidate violates constraint (26), it must violate (25) as well; but not vice versa. Therefore, it is certain that (26) should be absolutely ranked higher than (25).. 政治大. To observe the result of the [jaj]-to-[aj] experiment, it is observed that there should. 立. not be two [j]’s in the syllable, so *[G2αF]σ is satisfied, and it absolutely dominates *GαF,. ‧ 國. 學. for it is a self-conjunction (see. 2.1.3). In fact, MAX should dominate *GαF because *GαF. ‧. is violated by every candidate containing a glide. Therefore, MAX should be ranked higher than *GαF to avoid glide deletion. The prenuclear glide [j] is deleted, so MAX is. y. Nat. io. sit. violated, but the coda [j] is not deleted, so MAX-CODA is satisfied. Both of the glides do. n. al. er. not change any features, so IDENT[F]-G is satisfied. Through these descriptions, we can. Ch. i Un. v. rank the satisfied constraints higher and the violated ones lower, as tableau (27) shows.. engchi. 38. DOI:10.6814/NCCU202000482.

(48) (27) /jaj/ → [aj] *[G2αF]σ. /jaj/. IDENT[F]-G. MAX-CODA. MAX. *GαF. *. *. a. → aj b. jaj. *!. **. c. ja. *!. *. *. d. waj. *!. **. e. jɑw. *!. **. f. a. *!. **. 治政 Note that candidate (e) in tableau (27) changes its 大 vowel from the input. Here we 立 do not have a candidate *[jaw], which violates I [F]-G as well but remain the same DENT. ‧ 國. 學. vowel as its input because *[jaw] is absolutely a worse choice than [jɑw] no matter how. ‧. these constraints are ranked, for we have already discussed the low vowel assimilation. y. sit. io. er. illustrated in (28).. Nat. phenomenon in 4.1.1. The Hasse Diagram of the triphthong phonotactic restriction is. al. n. iv n C (28) The Hasse Diagram of the phonotactic of triphthongs h e n grestriction chi U *[G2αF]σ. IDENT[F]-G MAX-CODA. MAX. *GαF. As (28) shows, the example in tableau (27) shows that the optimal output for /jaj/ is [aj], which violates MAX (and of course *GαF). This indicates that the violations of 39. DOI:10.6814/NCCU202000482.

(49) *[G2αF], IDENT[F]-G and MAX-CODA are not allowed. Therefore, all of these constraints dominate MAX. Furthermore, since glides should not be deleted in most contexts (for example, /ja/ does not surface as [a]), MAX dominates *GαF. The constraint ranking developed in this section may also explain the absence of the high vocoid sequences *[ɥi] and *[ɥu], for the two high vocoids which have the same values for [back] or [round] features are in the same syllable no matter whether there is a mid/low vowel between them or not. However, this analysis is controversial because the constraint *[G2αF]σ only prohibits two glides that have the same values for [back] or [round] features from appearing in the same syllable, while [i] and [u] are. 政治大. actually vowels. If we transcribe the sequence as two glides, it is still awkward for there. 立. will be no peak in the syllable. Moreover, in addition to *[ɥi] and *[ɥu], *[ju] and *[wi],. ‧ 國. 學. which are also sequences of two high vocoids, are also not allowed to surface. The. ‧. absence of these patterns is going to be discussed in the next section.. y. Nat. n. er. io. al. sit. 4.1.3 The Sequence of High Vocoids. Ch. i Un. v. As the previous section mentions, here we are going to discuss patterns composed. engchi. of two high vocoids, none of which is surfaced out. Duanmu (2007) considers *[ju] and *[wi] to be accidental gaps, but it is apparent that the absence of these patterns are systematic. Regarding to this phenomenon, Lin (2007) has mentioned that “glides cannot be followed by a high vowel which has a different value for the [back] feature.” This statement here does not rule out the ill-formed pattern *[ɥi]. In fact, it is weird for Lin to specify “what kind of high vowel cannot follow a glide” because there is actually no high vowel following a glide. Via revising Lin’s statement, we can generalize that two high vocoids cannot be adjacent, or glides cannot be followed by a high vowel. Since it is two adjacent 40. DOI:10.6814/NCCU202000482.

(50) segments that are prohibited to share a certain feature, OCP, which is mentioned in 4.1.2, can be introduced as a constraint here, as in (29).. (29) OCP-HIVOCOID Assign one violation mark for every pair of adjacent high vocoids.. In fact, it is not necessary to specifically prohibit two adjacent “high vocoids” based on the analysis in this section. Instead, we can just prohibit two adjacent segments that are both [+high]. However, if we substitute a more general constraint called “OCP-. 政治大. HI” for constraint (29), it will wrongly predict the output of /iŋ/ as *[eŋ], which will be. 立. elaborated in 4.2.1.2. Therefore, in order to take the analyses of the following sections. ‧ 國. 學. into account, here we have to specifically define that it is two high vocoids that cannot. ‧. be adjacent.. In the previous section, we predict that the ill-formed triphthong *[jaj] is repaired. y. Nat. io. sit. by deleting the prenuclear glide according to how Taiwanese Mandarin speakers. n. al. er. pronounce the word yai ‘cliff’. Likewise, now that the sequences of two high vocoids. Ch. i Un. v. have been ruled out by constraint (29), there should be optimal outputs for these inputs.. engchi. Then, how do the sequences of two high vocoids surface out? The answer can be found from the evidence of loanwords and translated names. The examples are shown in (30 – 31) below.. 41. DOI:10.6814/NCCU202000482.

(51) (30) Loanwords and translated names containing the sequence of [ju] in the input English. Taiwanese Mandarin. a. UFO. [ju.foʊ]. youfu. [jow.fu]. b. Ulysses. [ju.lɪ.siz]. Youlixisi. [jow.li.ɕi.sz̩]. c. New Zealand. [nju zi.lænd]. Niuxilan. [njow.ɕi.lan]. d. Hugh Jackman. [hju dʒæk.mən]. Xiu Jiekeman. [ɕjow tɕje.khə.man]. (31) Loanwords and translated names containing the sequence of [wɪ]9 in the input English. Taiwanese Mandarin. 政治 weishiji 大. a. whisky. [hwɪs.kɪ]. b. Will Smith. [wɪl smɪθ]. Weier Shimisi. [khwɪn]. Kuiyin. [khwɪ.nə.khɹin]. kuinakelin. ‧ 國. [khwej.in] [khwej.na.khə.lin]. ‧. d. quinacrine. [wej.ɚ ʂʐ̩.mi.sz̩]. 學. c. Quinn. 立. [wej.ʂʐ̩.tɕi]. y. Nat. io. sit. Examples in (30) and (31) show that when a loanword or a translated name has a. n. al. er. sequence of two high vocoids in its English origin, there will be a mid vowel inserted. Ch. i Un. v. between the high vocoids in the translated form, and of course, the vowel will undergo. engchi. a glide formation, for there should not be two vowels in a single syllable. The descriptive generalization is shown in (32).. (32) The descriptive generalization for the sequence of high vocoids a. Two high vocoids cannot be adjacent. This requirement is enforced by mid vowel insertion.. 9. All of the examples in (31) have a lax [ɪ] following [w]. However, vowels do not contrast in tenseness. in Mandarin, so the output will not be influenced by the tenseness of the vowel in the input. 42. DOI:10.6814/NCCU202000482.

(52) b. There is only one vowel in a single syllable. This requirement is enforced by glide formation of a high vowel.. According to (32a), we developed constraint (33) as a violated faithfulness constraint for an output satisfying OCP-HIVOCOID. Constraint (34) is developed to ensure the inserted vowel is a mid one instead of a low one. Constraints (35 – 36) are developed based on (32b). Constraint (35) is based on McCarthy’s (2008) list of common markedness constraint, which includes constraints called “COMP-ONSET” and “COMPCODA” assigning violation marks for consonant cluster in the onset or coda position.. 政治大. Here the candidates containing a vowel cluster should be ruled out, so a new constraint. 立. “COMP-NUC” is introduced, where NUC stands for nucleus. Constraint (36) is used for. ‧ 國. 學. vowel gliding since a vowel changes its value for the [syllabic] feature when it becomes. ‧. a glide. In addition to the constraints mentioned in this paragraph, MAX is also active here. Tableau (37) provides an example of how a sequence of two high vocoids surfaces. n. al. er. io. sit. y. Nat. out.. (33) DEP-MIDV. Ch. engchi. i Un. v. Assign one violation mark for every mid vowel in the output which does not have a correspondent input.. (34) DEP-LOWV Assign one violation mark for every low vowel in the output which does not have a correspondent input.. (35) *COMP-NUC Assign one violation mark for every vowel cluster in the nucleus position. 43. DOI:10.6814/NCCU202000482.

(53) (36) IDENT[syl] Assign one violation mark for every segment which has a different value for the [syllabic] feature from its input.. (37) /ju/ → [jow] /ju/. OCPHIVOCOID. DEP-LOWV. MAX. a. → jow. 立. *. 政治 *! 大 *. 學. ‧ 國. *!. IDENT[syl] *. *!. c. u. e. jou. NUC. *. b. jɑw. d. ju. *COMP-. DEP-MIDV. *!. ‧. y. Nat. Different from previous tableaux, this one contains a doubled frame line, which. er. io. sit. indicates that the constraints left to it and right to it have no domination relations. Let’s first take a look at candidates (37a – d). The optimal candidate is (37a) [jow], which. al. n. iv n C has an inserted mid vowel. Other candidates than (37a) include (37b) [jɑw], h e n g cworse hi U. which has an inserted low vowel, (37c) [u], which deletes the prenuclear glide, and (37d) *[ju], which has a sequence of two high vocoids. Among these candidates, the one violates DEP-MIDV is the most tolerable. Therefore, we can have a result that OCPHIVOCOID, DEP-LOWV and MAX all dominate DEP-MIDV. On the other hand, if we compare candidates (37a) and (37e), the violation differs in the fact that the former violates IDENT[syl] while the latter violates *COMP-NUC. The better output is (37a) [jow], which proves that *COMP-NUC dominates IDENT[syl]. Here I have to clarify that the transcription of (37e) [jou] is actually widely used in previous 44. DOI:10.6814/NCCU202000482.

(54) studies. In this thesis, however, it is necessary to view a diphthong as a vowel plus a glide coda because it has been discussed that the prenuclear glide and the postnuclear glide should not have the same value for their [back] and [round] features (see 4.1.2). If we transcribe the postnuclear coda as a vowel, it will be considered as a part of the nucleus, and then the constraint prohibiting the prenuclear glide and the postnuclear glide to have the same value for their [back] and [round] features will become weird, for it is therefore regulating a “segment” and a “part of segment” to share the same value for the features. As a result, in order not to be contradicted, here we introduce these constraints to ensure the vowel gliding rule.. 政治大. Now two constraint rankings have been developed, as the doubled frame line in. 立. tableau (37) indicates. The fact is that we have no evidence to prove that there are. ‧ 國. 學. domination relations between the constraints left to the doubled frame line and the ones. ‧. right to it. To illustrate the constraint ranking clearer, the Hasse Diagram is available in (38). It does not matter for the candidates to be evaluated by the ranking on the left side. y. Nat. n. al. er. io. out.. sit. or the right side first because [jow], the optimal output is always the last one to be ruled. Ch. engchi. i Un. v. (38) The Hasse Diagram of the sequence of high vocoids OCP-HIVOCOID. DEP-LOWV. DEP-MIDV. MAX. *COMP-NUC. IDENT[syl]. As (38) shows, to repair a rhyme composed of two adjacent high vocoids, the best policy is to insert a mid vowel to violate DEP-MIDV. DEP-MIDV is thus dominated by OCP-HIVOCOID, DEP-LOWV and MAX. On the other hand, in order to make sure that a 45. DOI:10.6814/NCCU202000482.

(55) high vowel becomes a glide when there is another vowel, *COMP-NUC should dominate IDENT[syl].. 4.1.4 Summary. We have discussed all the rhyme patterns that do not contain a nasal coda in Taiwanese Mandarin through the analyses so far, and have obtained the conclusions as below: (i). Mid vowels and low vowels agree with the glide coda in [back] and [round] feature. (see 4.1.1). 立. Mid vowels and low vowels agree with the prenuclear glide in [back] and. 學. ‧ 國. (ii). 政治大. [round] feature when there is no glide coda. (see 4.1.1) When two glides that share the same values for [back] or [round] feature. ‧. (iii). are in the same syllable, the prenuclear one fail to surface out. (see 4.1.2). y. Nat. A mid vowel is inserted between two adjacent high vocoids. (see 4.1.3). io. sit. (iv). n. al. er. These results can almost explain the presence and the absence of the rhymes. Ch. i Un. v. without a nasal coda listed in 3.1. As for the constraint ranking, we have developed the. engchi. Hasse Diagrams at the end of each subsection. However, the constraint rankings should not be independent. After all, they are all evaluating the patterns of the same language, and even the same dialect. In addition, the constraints MAX and IDENT[F]-G are active in not only one analysis. Therefore, it is possible to combine the Hasse Diagrams in each subsection. Though not all of them have domination relations with another, the system is clearer if we have all the active constraints in one Hasse Diagram, as is illustrated in (39) below.. 46. DOI:10.6814/NCCU202000482.

(56) (39) The Hasse Diagram of rhymes without a nasal coda PHONOTACTICS. *COMP-NUC. AGREE[bk]-VG AGREE[rd]-VG. IDENT[syl]. AGREE[bk]-GV AGREE[rd]-GV. IDENT[F]-G. IDENT[F]-V. MAX. 立. MAX-CODA *[G2αF]σ. OCP-HIVOCOID DEP-LOWV. 政治大D. *GαF. EP-MIDV. ‧ 國. 學. To have an overview of all the rhyme patterns without a nasal coda, the constraint. ‧. ranking in diagram (39) can correctly predict their presence and absence, except for an unallowed rhyme *[ɥa]. Because there are no other absent patterns that seem to be. y. Nat. n. al. er. io. gap.. sit. symmetric to this rhyme, it is reasonable to consider the absence of *[ɥa] an accidental. Ch. 4.2 Rhymes with a Nasal Coda. engchi. i Un. v. 4.2.1 High Vowel Preceding a Nasal Coda. In section 2.2.1, it is mentioned that Lin (2007) has a hypothesis stating that there are two forms of the VX structure, diphthongs and “high/low vowel + nasal.” Based on her assumption, mid vowels never precede nasal codas in the underlying representation. In other words, all the mid vowels preceding a nasal coda in the surface representations are inserted or derived from other underlying forms. Her theory asserts that an [ə] is 47. DOI:10.6814/NCCU202000482.