以優選理論分析兩個客語方言之連讀變調 - 政大學術集成

全文

(1)國立政治大學語言學研究所碩士論文 National Chengchi University Graduate Institute of Linguistics Master Thesis. 指導教授：蕭宇超 Advisor：Yuchau E. Hsiao. 立. 政治大. ‧ 國. 學. 以優選理論分析兩個客語方言之連讀變調. ‧. An Optimality Theory Approach to the Tone Sandhi in Two Hakka Dialects. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. 研究生：陳煒翰撰 Student：Wei-Han Chen. 中華民國一百零二年一月 January, 2013. v.

(2) AN OPTIMALITY THEORY APPROACH TO THE TONE SANDHI IN TWO HAKKA DIALECTS. BY. 立. 治政 Wei-Han Chen 大. n. er. io. al. sit. y. ‧. ‧ 國. 學. Nat. A Thesis Submitted to the Graduate Institute of Linguistics in Partial Fulfillment of the Requirements for the Degree of Master of Arts. Ch. engchi. January, 2013. i n U. v.

(3) 記錄編號 : G0099555009. 國立政治大學博碩士論文全文上網授權書 National ChengChi University Letter of Author ization for Theses and Dissertations Full Text Upload. (提供授權人裝訂於紙本論文書名頁之次頁用) (Bind with paper copy thesis/dissertation following the title page). 本授權書所授權之論文為授權人在國立政治大學語言學研究所系所學年度第一學期取得碩士學位之論文。 T his 尬nnattests thatthe. 組 101. Division ofthe Department of Graduate Institute of Linguistics at 匕. National Chen C hi University h as received a Master degree t hesis/dissertation by t he u ndersigned in the. 扛. semester of101 academic year．. 政治大論文題目 ( Title ) : 以優選理論分析兩個客語方言之連讀變調 (A n Optimally Theory Approach 立 to the Tone Sandh n Two Hakka Dialects ). ‧ 國. 學. 指導教授 (Supervisor : 蕭宇超. ‧. 立書人同意非專屬、無償授權國立政治大學，將上列論文全文資料以數位化等各種方式車製後收錄於資料庫，透過單機、網際網路、無線網路或其他公開傳輸方式提供用戶進行線上檢索、瀏覽、下載、傳輸及列印。國立政治大學並得以再授權第三人進行上述之行為。. Nat. sit. y. T he u ndersigned g rants non-exclusive and g ratis authorization to National Chen C hi University, to re produce the above thesis/dissertation 血H text m aterial via digitalization or any other way, and ．to store E nthe. io. er. database f or u sers to access online search browse, download, transmit and print v ia single-m achine, the. n. al. v. Inter net, wireless I nternet or other public Im ethods. National Chen C hi University is entitled to reauthorize a. i n U. third party to perform the above actionsC． h. e nD c h i Full Text Uploading 仿orInternet Access ) : gissertation 論文全文上載網路公開之時間 (Time of Thesis 網際網路 (The Internet) [ 立即公開．立書人擔保本著作為立書人所創作之著作，有權依本授權書內容進行各項授權，且末侵害任何第三人之智慧財產權。 The undersigned guarantees thatth work is the original work ofthe undersigned, and is therefore eligible to g rant various authorizations according to this letter of authorization and does ．I]ot infringe any intellectual. property right of any third party．. ．依據96年9月 22日96學年度第i學期第i次教務會議決議，畢業論文既經考試委員評定完成，並已繳交至圖書館，應視為本校之檔案，不得再行抽換。關於授權事項亦採一經授權不得變更之原則辦理。 According to the resolution ofthe first Academic Af fairs Meeting ofthe first semester on September 22nd, 2007,Once the thesis/dissertation is passed afterthe off iciating examiner．s evaluation and sentto the library, it will be considered as the librarys record, t hereby changing and replacing ofthe record is disallowed. Forthe I]砌打eL OfauthOr切廿OIL OI]Ce the authO廿羽H On 泊餌anted tO thC 此L aI了， any 血山。L a几eL at沁n 泊 d泊a几Owed ，. 立書人 : 陳煒翰 &. ^(Signature) : & *% "^ Date Date of ofsignature signature :. >^. / Jc^^^f /. >^1. (dd/mm/yyyy).

(4) The members ofthe Committee approve the thesis of Wei-Han Chen defended on January 9th, 2013.. 77. 77 ^^. Processor Yuch^i E. Hsiao. Advisor. 立. □"' 。. 政治大. Professor Chin-wei Wu. ‧ 國. 學. Committee Member. ‧ y. Nat. n. al. er. io. sit. / du^ shan Professor Hui-shan Lin. v i n Ch e n g cCommittee h i U Member. Kawai Chui, Director, Graduate Institute of Linguistics. ^4.

(5) 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. Copyright © 2013 Wei-Han Chen All Rights Reserved iii. v.

(6) ACKNOWLEDGEMENT. 這一刻，剛好是口試後第二週，直到現在我才感覺到這一切即將結束……。. 作為整本論文的開頭，首先我要感謝我的指導教授蕭宇超老師。感謝老師帶. 治政大助我順利完成論文。此外，也感謝老師的激勵與指導，使我成為本門第一位兩年立我進入音韻和優選理論的世界，在修業過程中讓我體驗了許多神奇的事物，也幫. 半畢業的學生。. ‧ 國. 學. 感謝兩位口試委員，林蕙珊老師和吳瑾瑋老師。感謝兩位老師百忙中耐心地. ‧. 看完我的論文，並提供許多寶貴的意見，使我的論文內容能更完備。其中，我特. y. Nat. 別感謝林蕙珊老師，沒有老師的幫忙，我的香港之旅就無法成行，也無緣一睹音. IACL-20 提供我論文分析最重要的資訊。. al. er. io. sit. 韻大師們交手的畫面。另外，特別感謝林燕慧老師、包智明老師以及黃婷學姊在. n. v i n 除此之外，亦感謝每位老師對我的指導。感謝「立領」何萬順老師、「仙女」 Ch engchi U. 黃瓊之老師、「電眼」萬依萍老師、「星媽」徐嘉慧老師、「紳士」莫建清老師、「犀. 利」黃麗儀老師和「客家一姊」賴惠玲老師。每位老師對我都相當照顧，但我想特別感謝賴老師。老師除了提供我找回客家本色的機會，在我論文遇到瓶頸時，更是貼心的提供資訊，讓我能夠有驚無險的完成本篇論文，謝謝老師。最後，感謝「正妹助教」曾惠鈴一路上陪我走過這麼多風雨！工作室學長姊的鼓勵與關心是我研究所生涯中相當重要的部分。感謝從碩一就一直很照顧我的子權、感謝京都姊熱心的鼓勵與打氣、感謝一起闖蕩香港的詩敏學姊，沒有你們幫助和照顧，我絕對無法這麼快完成論文。另外，也感謝智仁、. v.

(7) 凱琳、明昌、綠茜、旺楨、書豪、平周、柏溫、侃彧、韶君、瑋萱、傻妞、承彬，有了你們讓我的生活多了許多樂趣。在研究所修業的過程中，感謝語言所 99 級的全體同學陪我一起努力打拚、一起吃喝玩樂、一起製造許多美好的回憶。感謝最 man 的阿 man 和阿中、感謝總是熱心的王維、感謝一起分享很多事的培禹、感謝一起在印度出生入死的印度國民女神婉君，感謝你們帶給許多美好的回憶。提及同袍戰友，當然要特別提一下大公寓的室友們：DA 花。感謝百變老頭大衛，感謝你總是像老師一樣教我很多事，像哥哥一樣帶著 HA 花到處玩，像朋友一樣跟我分享很多生活的大小事。感. 治政大受再當幸福小屁孩的感覺。也感謝你帶著我一起拚論文，這兩年半的奇蹟一半要立. 謝美食天后捧呆，感謝你總是幫大家規劃這麼多吃喝玩樂的行程，讓我們可以享. 歸功於你。感謝瘋狂麻吉涵絜，感謝你讓我在研究所還有人可以分享所有大事、. ‧ 國. 學. 小事、好事、壞事。感謝你許多神奇的點子，讓大公寓的生活變得回味無窮。感. ‧. 謝隱形室友縮縮英明，感謝你曾經幫我修電腦，陪我喇賽，參與大公寓的生活。. y. sit. io. er. 瘋狂的時光。. Nat. 感謝大公寓的各位，感謝你們陪我走過每個喜怒哀樂的時刻，我會想念我們一起. 感謝每位在我研究所期間陪伴過我的朋友，感謝 BFF、30 層美食團、YY 團、. al. n. v i n 以及許多高中和大學的好朋友，感謝你們繼續陪我創造回憶。其中，我想特別感 Ch engchi U 謝巫宛臻，感謝你當時陪我一起準備考試，雖然我們沒能一起完成當時的目標，但我很感謝你，我做到了！最後，我要感謝我的家人，爺爺陳阿畑先生、爸爸陳慶鑫先生、媽媽方麗凰女士以及弟弟陳品綸。感謝你們的包容，讓我能夠選擇自己想走的路。感謝你們一路對我的支持，沒有你們的支持和體諒，我無法完成這個學位。. 就在一切即將結束的此刻，我想說：「感謝大家一路上的照顧，我會繼續加油！」. vi.

(8) VITA. Education M.A. in Graduate Institute of Linguistics, January 2013 National Chengchi University B.A. in Department of Foreign Languages, June 2010 National Chiayi University. 學. ‧ 國. 立. 政治大. Grants and Scholarships. Graduate Institute of Linguistics Scholarship. ‧. 2012. National Chengchi University. y. Nat. sit. io. National Science Council Travel Grant for attending the 20th Annual Conference of the International Association of Chinese Linguistics (IACL-20) Project No. NSC-101-2922-I-004-064. n. al. er. 2012/8. Ch. engchi. i n U. v. 2011/8-2012/7. Research Assistant for the National Science Council Project “On Hakka genericity-characterizing constructions: Lexical semantics, event frames and information structure (1/2)” Project No. NSC 100-2140-H-004-185-MY2. 2011/9-2013/1. Teaching Assistant for Language and World Civilization National Chengchi University. 2011. Graduate Institute of Linguistics Scholarship National Chengchi University. vii.

(9) 2011. The committee member of the 3rd Theoretical Phonology Conference (TPC-3). 2011/2-2011/6. Teaching Assistant for Morphology National Chengchi University. 2011/10. 2010-2013. National Chengchi University Travel Grant for attending the 44th International Conference on Sino-Tibetan Languages and Linguistics (ICSTLL-44) Theoretical Phonology Laboratory Research Assistantship National Chengchi University. 立. Publication. 政治大. ‧ 國. 學. ‧. Wei-Han Chen. 2012. Tone Sandhi in Ningdu Tiantou Hakka. Paper accepted for poster presentation at the Fourth Theoretical Phonology Conference (TPC-4). Taipei, Taiwan: National Chengchi University.. y. Nat. sit. n. al. er. io. Wei-Han Chen. 2012. Disyllabic Tone Sandhi in Taiwan Liujia Raoping Hakka. Paper presented at the 20th Annual Conference of the International Association of Chinese Linguistics (IACL-20). Hong Kong, China: Hong Kong Polytechnic University, August 29.. Ch. engchi. i n U. v. Wei-Han Chen. 2012. Tone Sandhi and Merged Tones in Raoping Hakka. Paper presented at the Student Workshop on East Asian Linguistics 2012 (SWEAL-2012). Hsinchu, Taiwan: National Tsing Hua University, August 4. Wei-Han Chen. 2011. Phonological Alternations of the Schwa-Nasal Rhymes in Taiwanese Accented Mandarin. Paper presented at the 44th International Conference on Sino-Tibetan Languages and Linguistics (ICSTLL-44). Mysore, India: Central Institute of Indian Languages, October 9.. viii.

(10) TABLE OF CONTENTS. Acknowledgement ........................................................................................................ v VITA............................................................................................................................ vii Table of Contents ........................................................................................................ ix. 治政 Chinese Abstract ....................................................................................................... xiii 大立 ‧. ‧ 國. 學. English Abstract ......................................................................................................... xv. CHAPTER 1 INTRODUCTION ............................................................................. 1. sit. y. Nat. er. io. CHAPTER 2 LITERATURE REVIEW ................................................................. 5. al. v i n C h ................................................................................. 2.1.1 Optimality Theory 5 engchi U n. 2.1 Theoretical Background ................................................................................... 5. 2.1.2 Correspondence Theory ........................................................................ 7 2.1.3 Indexed constraint approach ................................................................. 8 2.1.4 Local conjunction................................................................................ 10 2.2 Tone ................................................................................................................ 12 2.2.1 Tonal markedness ............................................................................... 12 2.2.2 Internal structure of tone ..................................................................... 14 2.3 Previous studies .............................................................................................. 16 2.3.1 Bao (2011) .......................................................................................... 17 2.3.2 Zhang (1999) ....................................................................................... 18 ix.

(11) 2.3.3 Hsieh (2005)........................................................................................ 19 2.3.4 Lin (2011, 2012) ................................................................................. 19. CHAPTER 3 LIUJIA RAOPING HAKKA ......................................................... 23 3.1 Tone inventory................................................................................................ 23 3.2 Tone sandhi patterns ....................................................................................... 25 3.3 OT analysis ..................................................................................................... 32 3.3.1 Positional tone sandhi ......................................................................... 32 3.3.2 Contextual tone sandhi of Yangping .................................................. 39. 治政 3.3.2.2 Yangping-Yinshang contextual tone 大 sandhi ............................. 47 立 3.3.2.3 Yangping-Yinru contextual tone sandhi ................................... 53. 3.3.2.1 Yangping-Yinping contextual tone sandhi ................................ 39. ‧ 國. 學. 3.3.3 Ru tone sandhi .................................................................................... 59. ‧. 3.4 Summary......................................................................................................... 62. Nat. sit. y. CHAPTER 4 NINGDU TIANTOU HAKKA ....................................................... 65. io. er. 4.1 Tone inventory................................................................................................ 65. al. v i n 4.2.1 Tone sandhi patternsCofhconstruction .............................................. 70 engchi U n. 4.2 Tone sandhi patterns ....................................................................................... 69. 4.2.2 Tone sandhi patterns of construction. .............................................. 73. 4.3 OT analysis ..................................................................................................... 77 4.3.1 OT analysis of Tone sandhi in construction .................................... 77 4.3.2 OT analysis of Tone sandhi in construction .................................... 86 4.3.2.1 The second syllable tone sandhi ............................................... 86 4.3.2.2 The first syllable tone sandhi.................................................... 97 4.3.2.3 Both syllables tone sandhi ...................................................... 103 4.3.2.4 Wrong predictions of [*IsRise & *T]ADJ ................................ 105 4.3.2.5 Ru tone sandhi ........................................................................ 117 x.

(12) 4.3.3 OT analysis of Tone sandhi in construction. and. ....................... 125. 4.4 Summary....................................................................................................... 127. CHAPTER 5. CONCLUSION .............................................................................. 129. 5.1 Summary of the thesis .................................................................................. 129 5.2 Future issues ................................................................................................. 134. REFERENCES......................................................................................................... 135. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. xi. i n U. v.

(13) 國. 立. 政. 治. 大. 學. 研. 究. 所. 碩. 士. 論. 文. 題. 要. 研究所別：語言學研究所論文名稱：以優選理論分析兩個客語方言之連讀變調指導教授：蕭宇超研究生：陳煒翰. 政治大. 論文提要內容：(共一冊，20,688 字，分五章). 立. ‧ 國. 學. 本論文以優選理論分析六家饒平客語和寧都田頭客語雙音節詞之連讀變調，. ‧. 並藉由聲調內部結構說明變調的動機。在六家饒平客語方面，主要是呈現位置變. sit. y. Nat. io. n. al. er. 調(positional tone sandhi)，較有標(more marked)的聲調位於前字時發生變調。另. i n U. v. 外，此方言的陽平和陽去因歷史演變，兩者的本調(citation tone)同為 HH，但兩. Ch. engchi. 者因不同的變調形式而產生不同的變調(sandhi tone)。陽去的變調形式為位置變調，而陽平的變調形式為環境變調(contextual tone sandhi)。本文採用聯合制約 (Local Constraint Conjunction)捕捉環境變調制約運作的環境。另外，運用「詞素特定音韻」(morpheme-specific phonology)標記聯合制約，解釋陽平和陽去不同的變調規則。在寧都田頭客語方面，變調受詞法結構影響，且兩個音節皆有可能發生變調。本文利用標記制約理論(indexed constraints approach)解釋不同結構的變調情形。另藉由位置信實制約不同的排序，嘗試說明聲調的保留屬於類型差異 xiii.

(14) (typological differences)。而此方言的變調類型包含位置變調和環境變調，同樣要求較有標的聲調改變，並使用聯合制約說明在環境變調的情況下制約運作的環境。. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. xiv. i n U. v.

(15) ABSTRACT. This thesis investigates the tone sandhi in Liujia Raoping Hakka and Ningdu Tiantou Hakka under the framework of Optimality Theory. The internal structure of. 政治大. the tone reveals the motivation and the mechanism of tone sandhi. In terms of the tone. 立. sandhi in Liujia Raoping Hakka, the universal tonal markedness tendency could be. ‧ 國. 學. regarded as the motivation triggering the positional tone sandhi in the left syllable.. ‧. Moreover, the morpheme-specific phonology is adapted to account for the tone sandhi. Nat. io. sit. y. of historical merged tones, Yangping and Yangqu. Yangping and Yangqu are both high. er. level tones; however, they display different tone sandhi patterns. Yangping displays. al. n. v i n C h Yangqu displays the contextual tone sandhi whereas e n g c h i U the positional tone sandhi. Local conjunction constraints are posited to restrict markedness constraints to specific contexts in order to account for the mechanism of contextual tone sandhi. On the other hand, in terms of the tone sandhi in Ningdu Tiantou Hakka, the tone sandhi is construction sensitive and takes place in both syllables. The indexed constraint. approach is adopted to explain the tone sandhi in different constructions. In addition, the preservation of tone is argued to be the typological difference according to xv.

(16) different rankings of the positional faithfulness constraints. Finally, the tonal markedness tendency motivates the positional tone sandhi and the conjoined constraints are posited to govern the contextual tone sandhi in this dialect.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. xvi. i n U. v.

(17) CHAPTER 1 INTRODUCTION. 政治大. This thesis investigates the disyllabic tone sandhi in two Hakka dialects under. 立. the theoretical framework of Optimality Theory (hereafter OT). Tone sandhi refers to. ‧ 國. 學. the tonal alternation in certain contexts in a connected speech. It is one of the issues. ‧. which has attracted great interests from phonologists, such as Hsiao (1991, 1995,. Nat. io. sit. y. 2000, 2006), Chen (2000), Lin (2006, 2008, 2011, 2012), Bao (2011), among others.. er. This thesis discusses the tone sandhi in two Hakka dialects, Liujia Raoping Hakka and. al. n. v i n C hof the internal structure Ningdu Tiantou Hakka, in terms e n g c h i U of tone (Bao 1999).. Liujia Raoping Hakka is a Hakka dialect spoken in Hsinchu City, Taiwan, used only among family members. Most of the speakers are bilingual. They are capable of communicating in a more dominant Hakka dialect, Sixian Hakka or Hailu Hakka, in public. Tone sandhi in this language is interesting because there are two tones merging in a single citation form but remaining distinct in the sandhi forms. It is due to the historical changes of tones (Xu 2008). Yangping HH and Yangqu HH are both high 1.

(18) 2. level tones as a citation tone. In terms of the sandhi tone, Yangping HH changes to HM when it is followed by LL, HM or M. On the other hand, Yangqu HH becomes MM regardless of the following tones. Previous studies, such as Xu (2005, 2008), did not discuss the motivation and mechanism of tone sandhi in this language. Besides, it brings the challenge for classic OT to deal with the situation which two output forms are generated from the identical input. These issues are the main concerns in this thesis.. 立. 政治大. In terms of Ningdu Tiantou Hakka, it is a Hakka dialect spoken in Jiangxi. ‧ 國. 學. Province, China. Tone sandhi in this language is interesting because it is construction. ‧. sensitive. The verb-object construction displays different tone sandhi patterns from. Nat. io. sit. y. the modifier-head and conjunction constructions. Tone sandhi in the verb-object. al. er. construction only takes place in the first syllable while it may occur in both syllables. n. v i n Cconstructions. in the modifier-head and conjunction of citation tones U h e n g c hThei preservation in the tone sandhi processes in different constructions supports the concept that the preservation of tone contrast should be of typological difference instead of the assignment of stress. Besides, this chapter illustrates how conjunction constraints trigger tone sandhi and how they preserve the unchanged tonal pairs. The issues which will be discussed in this thesis are as follows. First, what are the motivation and mechanism of tone sandhi in both Hakka dialects? Second, how.

(19) 3. does OT deal with the historical merged tones in Liujia Raoping Hakka? The organization of this thesis is as follows. Chapter 1 introduces the research problems. Chapter 2 reviews some relevant studies on the tonal structure and tone sandhi. Chapter 3 discusses how OT deals with the tone sandhi and merged tones in Liujia Raoping Hakka. Chapter 4 discusses how OT deals with the tone sandhi in Ningdu Tiantou Hakka. Chapter 5 concludes this thesis.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v.

(20) 4. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v.

(21) CHAPTER 2 LITERATURE REVIEW. 政治大. This chapter comprises four sections. Section 2.1 discusses related theories,. 立. including Optimality Theory, Correspondence Theory, Indexed constraint approach,. ‧ 國. 學. and Local conjunction. Section 2.2 discusses the tonal markedness and the internal. ‧. structure of tone. Section 2.3 reviews previous studies on tone sandhi.. io. sit. y. Nat. n. al. er. 2.1 Theoretical Background 2.1.1 Optimality Theory. Ch. engchi. i n U. v. Optimality Theory is a constraint-based framework (Prince and Smolensky 1993/2004, McCarthy and Prince 1993, McCarthy 2008) which is different from traditional derivational approach. In the framework of OT, two main systems are involved, namely Generator (GEN) and Evaluator (EVAL). The Generator (GEN) produces infinite candidates for an input. Then the candidates are evaluated by hierarchically ranked constraints, as shown in (1). 5.

(22) 6. (1) OT Schema (Kager 1999:8). GEN. EVAL. Unlike the derivational approach, these constraints are universal and violable. Constraints exist universally but. 立. 治政 they could be ranked 大differently. in different. ‧ 國. 學. languages. The violation of the higher ranked constraints is fatal. The candidate having the minimal violation is selected as the optimal output.. ‧. The operation of OT is demonstrated with tableaux as shown in (2).. sit. n. al. er. io. /Input/. y. Nat (2) Operation of OT. Constraint A. Ch. Candidate (c). v. i engch Constraint C. Constraint B. ☞Candidate (a) Candidate (b). i n U. Constraint D. * *!. * *!. In (2), the input is placed in the left top space and the possible candidates generated by GEN are placed in the left column. In terms of the right columns, each constraint is placed in an individual column. The constraint ranked left represents that it is ranked higher. On the other hand, the constraint near the right side of the tableau.

(23) 7. represents that it is ranked lower, i.e. constraint A is ranked higher than constraint B. The dotted line between constraint C and constraint D represents that there is no crucial ranking between these two constraints. The finger marker points out the optimal output form. Every asterisk mark represents one violation. The exclamation mark highlights the fatal violation. The shade area represents that the violation does not influence the elimination because there is a fatal violation for the high-ranked. 政治大. constraint.. 立. Take the tableau in (2) as an example, candidate (b) is ruled out because it. ‧ 國. 學. violates the undominated constraint A. Candidate (c) violates the highly ranked. ‧. constraint B. In this case, even though candidate (a) violates constraint C, it is still. Nat. n. al. er. io. sit. y. selected as the optimal output.. Ch. 2.1.2 Correspondence Theory. engchi. i n U. v. The concept of Correspondence Theory was developed by McCarthy and Prince (1995) which defines the identification between two language representations, such as input and output. The schema of Correspondence is shown as (3).. (3) The schema of Correspondence (McCarthy and Prince 1995:262) Given two strings S1 and S2, correspondence is a relation Ʀ from the elements of.

(24) 8. S1 to those of S2. Element α ϵ S1 and β ϵ S2 are referred to as correspondence of one another when α Ʀ β.. Based on the schema, each relationship between two representations is governed by a faithfulness constraint. The faithfulness constraints governing the relationship between strings are given in (4).. 政治大. 立. (4) a. MAX (maximality): Every segment in S1 has a correspondent in S2.. ‧ 國. 學. b. DEP (dependence): Every segment in S2 has a correspondent in S1.. ‧. c. IDENT[F] (identity): Correspondent segments are identical with respect to. n. al. er. io. sit. y. Nat. feature F.. Ch. 2.1.3 Indexed constraint approach. engchi. i n U. v. According to Inkelas (2007), the phonology of a language might not be absolutely uniform. It could be influenced by factors, such as lexical stratum (e.g., native vs. foreign), and part of speech. Within OT, the indexed constraint approach is one of the approaches to capture the language internal variation. In the indexed constraint approach, instead of constraint reranking, there is only one fix constraint ranking to capture the morphologically conditioned phonology. The.

(25) 9. constraints could be divided into several different indexed constraints. Alderete (2001) indexed ¬OO-MAX-(Accent) as ¬OOdom-MAX-(Accent) in dealing with the Japanese high tone. The indexed ¬OOdom-MAX-(Accent) targets the specific domain where there is a dominant suffix in the input. Additionally, Itô and Mester (1999) also used the indexed constraint approach in dealing with the four lexical strata in Japanese. Itô and Mester (1999) mentioned that. 政治大. Japanese lexicon could be divided into four different strata according to the. 立. phonotactic differences. They proposed FaithUF, FaithAF, FaithSJ and FaithY and. ‧ 國. 學. ranked them in one fix constraint ranking to capture the different preservation of. ‧. different lexical strata.. Nat. io. sit. y. Besides, Pater (2000, 2010) pointed out that both the markedness and. al. er. faithfulness constraints could be indexed. Moreover, Pater (2000, 2007, 2010) named. n. v i n C h “morpheme specific the indexed constraint approach e n g c h i U phonology”. Pater (2010) used. this approach in handling the syncope in Yine. It is the condition that indistinguishable morphemes exhibit different phonological behaviors, such as undergo different phonological processes. The indexed constraints suggest that the speakers own the intuition to distinguish the morphemes. In this thesis, we will follow the concept of Pater (2010) because the merged tones are indistinguishable and undergo different phonological processes..

(26) 10. 2.1.4 Local conjunction In terms of local conjunction, it is proposed by Smolensky (1993, 1995) that every constraint in CON could be locally conjoined with another constraint. Lubowicz (2005) mentioned that the idea behind local conjunction is for constraints to exclude the worst of the worst. Itô and Mester (1998) defined the local conjunction, as shown in (5).. 政治大. 立. (5) Local conjunction of Constraints (Itô and Mester 1998:10). ‧. ‧ 國. 學. a. Definition. Local conjunction is an operation on the constraint set forming composite. io. sit. y. Nat. constraints:. n. al. er. Let C1 and C2 be members of the constraint set Con. Then their local conjunction. Ch. C1 & C2 is also a member of Con.. engchi. i n U. v. b. Interpretation The local conjunction C1 & C2 is violated if and only if both *C1 and *C2 are violated in some domain δ..

(27) 11. c. Ranking (universal) C1 & C2 ». C1. C1 & C2 ». C2. In (5b), it diaplays that the conjunction constraint will be violated if the component constraints are all violated. In (5c), it is the conjunction schema that the. 政治大. conjunction constraints have to dominate the component constraints universally.. 立. Besides, Lubowicz (2005) mentioned that there should be some restrictions of the. ‧ 國. 學. component constraints; therefore, she proposed a restricted version of local. ‧. conjunction.. n. al. Ch. C=C1&C2 is violated iff. engchi. er. io. sit. y. Nat (6) Restricted local conjunction (Lubowicz 2005:259). i n U. v. a. LOCC1 ∩ LOCC2 ≠ ∅ b. C1 results in C2 if C1 is Faithfulness and C2 is Markedness.. In (6), Lubowicz (2005) imposed two restrictions on the component constraints. First, the loci of violation of both component constraints could not be empty, i.e. the loci of violation of both component constraints should intersect. Second, if there is a.

(28) 12. faithfulness constraint, the faithfulness constraint has to lead to the violation of the markedness constraint.. 2.2 Tone In this section, two issues will be discussed, namely tonal markedness and the internal structure of tone.. 立. 2.2.1 Tonal markedness. 政治大. ‧ 國. 學. Universally, contour tones are more marked than level tones. There are three. ‧. main rationales to claim it. First of all, according to Zhang (2001), Yip (2001, 2002). Nat. io. sit. y. and Bao (2003), languages are more common to have level tones than contour tones,. al. er. i.e. level tones are typologically more unmarked than contour tones. Second, in terms. n. v i n of acquisition, contour tones tend C to h be level off (Yip 2001). e n g c h i U It suggests that level tones are relatively easier. Last, in terms of articulation, contour tones require more articulatory efforts and longer duration (Zhang 1999). Accordingly, Yip (2001, 2002). proposed a tonal markedness tendency based on minimizing the articulatory effort, as shown in (7)..

(29) 13. (7) Minimize articulatory effort (Yip 2001:315) a. Contour tones are more marked than level tones: *CONTOUR b. Rising tones are more marked than falling tones: *RISE » *FALL c. High tones are more marked than low tones: *H » *L. In (7), it shows that contour tones are more marked than level tones in terms of the. 政治大. articulatory effort. Furthermore, rising tones cost more efforts than falling tones do. 立. while high level tones cost more efforts than low level tones do.. ‧ 國. 學. Nonetheless, de Lacy (1999, 2002) posited the positional tonal markedness. ‧. tendency, as shown in (8) and (9), to account for the stress system of Ayutla Mixtec.. n. al. *HD/L » *HD/M. Ch. engchi. er. io. sit. y. Nat (8) Tonal preference in the head position. i n U. v. (9) Tonal preference in the non-head position *NonHD/H » *NonHD/M. In (8), it shows that the low level tone is the least preferred in the prosodic head position while in (9), the high level tone is the least preferred in the prosodic non-head.

(30) 14. position. The positional tonal markedness tendency was originally used to discuss the interaction between stress and tones. Since the existence of stress in Chinese dialects is controversial, it will not be considered in this thesis. The tonal markedness tendency proposed by Yip (2001, 2002) will be applied and regarded as the motivation of tone sandhi in Liujia Raoping Hakka and Ningdu Tiantou Hakka.. 2.2.2 Internal structure of tone. 立. 政治大. It is believed that tones have their internal structures. The four different possible. ‧ 國. 學. models of tonal geometry summarized by Yip (2002) will be discussed as follows.. ‧. Nat. sit. n. al. er. io. H σ l. y. (10) Tonal geometry 1 (Yip 1980, Hyman 1993). Ch. engchi. i n U. v. h. In (10), the tone melodies are not regarded as one unit. They are independent of each other. Moreover, they are not dominated by a tonal node. In this model, the register is free to spread alone. However, the whole tone could not spread as a unit and the contour could not spread without the register..

(31) 15. (11) Tonal geometry 2 (Clements 1981, Duanmu 1990, 1994, Snider 1990) σ o. or o. H lH h. μ. μ. o. o. Tonal node. H lH h. In (11), two tone melodies are dominated by a tonal node and each of them is. 政治大. still independent of each other. In terms of spreading, the model in (11) exhibits the. 立. same case as the model in (10) does.. ‧. ‧ 國. 學. (12) Tonal geometry 3 (Yip 1989, Hyman 1993). sit. io. Tonal node. n. al. er. H. y. Nat. σ. l. h. Ch. engchi. i n U. v. In (12), the register node is the tonal node. Also the tone melodies are dominated by the tonal node. In this model, the whole tone is allowed to spread as a unit. However, the register and the contour could not spread without each other..

(32) 16. (13) Tonal geometry 4 (Bao 1990, 1999, Snider 1999) σ o. Tonal node. H Contour l. h. 政治大. In the model in (13), the tone melodies are dominated by the contour node. Both. 立. contour node and register node are dominated by the tonal node. In this model, unlike. ‧ 國. 學. the models discussed above, each node is able to spread alone; besides, the whole tone. ‧. could still spread as a unit. In this case, the model in (13) will be adopted with some. Nat. er. io. sit. y. labeling differences in the analysis in chapter three and four, as in (14).. n. al. i n C (14) Labeling the tonal structure proposed (1990, 1999) h einnBao gchi U T Register level. v. Tonal level. Register Contour t. t. Contour level Tone melody level. 2.3 Previous studies In this section, some of the previous studies focusing on tone sandhi issues will.

(33) 17. be discussed.. 2.3.1 Bao (2011) Based on previous studies (Chen 1987, 2000), Bao (2011) classified Chinese tone sandhi patterns into four types, namely contextual tone sandhi, positional tone sandhi, templatic tone sandhi and tone spread. In terms of the contextual tone sandhi,. 政治大. the tone sandhi is sensitive to the adjacent tone elements. Besides, this type of tone. 立. sandhi could target the register and contour separately. The tone sandhi targets the. ‧ 國. 學. register in languages such as Luoyang (He 1996) while the tone sandhi targets the. ‧. contour in languages such as Pingyao (Hou 1980). Moreover, contextual tone sandhi. Nat. io. sit. y. could target both register and contour in languages such as Gaomi (Li 2004). Second,. al. er. positional tone sandhi refers to the condition which tones undergoing tone sandhi are. n. v i n conditioned by the position.CThe sandhi in the Southern Min dialect of Xiamen h etonen g chi U. (Chen 2000) is positional. All the citation tones in the non-utterance final position undergo tone sandhi. In addition to the contextual and positional tone sandhi, templatic tone sandhi is another tone sandhi type. It occurs in the polysyllabic compounds or phrases. The sandhi tones could not be derived from the citation tones in the input. The sandhi tones are prespecified. Harbin Mandarin (Jiang 1997) belongs to this type. Last, unlike the contextual and templatic tone sandhi, the citation tone.

(34) 18. and sandhi tone in tone spread are not confined in a specific syllable. The sandhi tone spreads to neighboring syllables in tone spread. Wu dialects (Chen 2000) are examples of this type of tone sandhi. In chapter three, the analyses reveal that the tone sandhi in Liujia Raoping Hakka tends to be positional. The contextual tone sandhi only occurs when the citation tone is the historically merged Yangping HH1. Furthermore, the tone sandhi in Ningdu. 政治大. Tiantou Hakka illustrates both positional and contextual tone sandhi.. 立. ‧ 國. 學. 2.3.2 Zhang (1999). ‧. Zhang (1999) proposed the duration constraint family to account for the tone. Nat. io. sit. y. sandhi in Pingyao. This constraint family comprises three sub families, i.e. duration. al. er. (rime), duration (word) and duration (boundary). Duration (rime) pointed out the. n. v i n different rime duration between ruC tones tones. Zhang (1999) mentioned h eand n gnon-ru chi U that it is how ru tones and non-ru tones are distinct from each other. Duration (word). confined the numbers of the tonal inflection point in disyllabic words. The tonal inflection points remain the tonal contrast; however, the total numbers of the inflection point is confined in disyllabic words owing to the limited word length. Last, duration (boundary) defined the restrictions at the syllable boundary. In terms of articulation, the duration between two rimes is considerably shorter than the average.

(35) 19. syllable. Therefore, it is hard to realize the abrupt pitch change at the syllable boundary. The constraint targets the pitch change at the syllable boundary will be applied to handle the tone sandhi in Ningdu Tiantou Hakka in chapter four.. 2.3.3 Hsieh (2005) Hsieh (2005) investigated the tonal chain shift in the tone sandhi in two dialects. 政治大. of Southern Min, Coastal Taiwanese and Mainstream Taiwanese. In both dialects, tone. 立. sandhi only occurs in the non-phrase final position. The tone in the phrase final. ‧ 國. 學. position never undergoes tone sandhi. Hsieh (2005) mentioned that it is because the. ‧. phrase final syllable is 40% longer than the phrase internal syllable (Peng 1996). The. Nat. io. sit. y. tone in the longer syllable tends to stay static. It could account for the tone sandhi in. al. er. these two dialects; however, it could not explain the disyllabic tone sandhi in Ningdu. n. v i n Tiantou Hakka. Since both C syllables disyllabic h e ningthec h i U tone sandhi in Ningdu Tiantou Hakka may undergo tone sandhi, the preservation of tone in the phrase final position may not always due to the syllable length. Hence, in this thesis, the preservation of tone is argued to be the typological difference among languages.. 2.3.4 Lin (2011, 2012) Lin (2011) analyzed the disyllabic tone sandhi in Dongshi Hakka under the.

(36) 20. framework of OT. Lin (2011) posited NoJump-t to capture the assimilation at the tone melody level. Besides, OCP-T(11), OCP-c( ) and OCP-C( ) are posited to capture the dissimilation at the tonal level and the contour level. However, the dissimilation of the. tone melody and the assimilation of adjacent tone melodies only occur when. there is a low register tone in the head position. Therefore, Lin (2011) posited conjunction constraints, such as [NoJump-t & *HD/Lr]ADJ, to specified the contexts. 政治大. where the markedness constraints are active. The situation is also found in both. 立. dialects discussed in this thesis. The triggers of tone sandhi in merged tones in Liujia. ‧ 國. 學. Raoping Hakka and the contextual tone sandhi in Ningdu Tiantou Hakka are confined. ‧. to certain contexts. Accordingly, the conjunction constraints will be applied in this. Nat. sit. al. er. io. adjacent tones.. y. thesis. Besides, the domain of the conjunction constraints will be defined as the. n. v i n C h the construction Additionally, Lin (2012) investigated e n g c h i U sensitive tone sandhi in. Pingyao. In Pingyao, the register is influential in dealing with the tonal preservation in the head syllable, i.e. high register tones and low register tones display different cases of preservation in the head syllable. In this case, IDENT-IO-T-HD(Hr) and IDENT-IO-T-HD(Lr) were posited. Since the register is influential in the tonal preservation in both Hakka dialects in this thesis as well, constraints targeting the register and tonal preservation are applied in the analyses in the following chapters..

(37) 21. Moreover, in Pingyao, the subject-predicate and verb-object construction exhibit the. same. tone. sandhi. pattern. whereas. the. modifier-head,. conjunction,. verb-complement construction and the reduplicated nouns exhibit the same tone sandhi pattern. Nevertheless, this correspondence relation is different from the one in Ningdu Tiantou Hakka. The subject-predicate and verb-complement construction do not undergo tone sandhi in Ningdu Tiantou Hakka whose tone sandhi is construction. 政治大. sensitive as well. Therefore, the correspondence relation between tone sandhi patterns. 立. and constructions is argued to be the language specific phenomenon.. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v.

(38) 22. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v.

(39) CHAPTER 3 LIUJIA RAOPING HAKKA. 政治大. This chapter investigates the disyllabic tone sandhi in Taiwan Liujia Raoping. 立. Hakka. The data are mainly from Lü (1993), Hsu (2005, 2008) and Lin (2007). These. ‧ 國. 學. data were confirmed by the teaching materials1 provided by Hakka Affairs Council,. ‧. Executive Yuan.. Nat. io. sit. y. The rest of this chapter will be arranged as follows. Section 3.1 introduces the. er. tone inventory and section 3.2 introduces tone sandhi patterns. Section 3.3 provides. al. n. v i n the OT analysis for the toneC sandhi h epatterns. i U 3.4 provides the summary of this n g c hSection chapter.. 3.1 Tone Inventory There are six citation tones in Raoping Hakka, namely Yinping LL, Yinshang HM, Yinru M2, Yangping HH, Yangqu HH3, and Yangru H. In Chao’s (1930) system, tones 1. 2. The link of the material is as follows: http://elearning.hakka.gov.tw/attclass/cprofile.aspx?param=10000127 The italic and underlined tones are ru tones. Ru tones end with voiceless unreleased stops. 23.

(40) 24. are ranged from 1 to 5. 1 represents the lowest tone while 5 represents the highest tone. In this thesis, tones are converted into Yip’s (2001) system. We marked 4 and 5 as H, marked 3 as M and marked 1 and 2 as L. In terms of the tone categories, the historical tonal changes of Raoping Hakka are different from other Hakka dialects. Yangshang has partially merged into Yinping and partially merged into Yangqu. Yinqu had merged into Yinshang in the diachronic. 政治大. process (Hsu 2008). On the other hand, what is interesting in Liujia Raoping Hakka is. 立. the identical citation tones of Yangping and Yangqu. Both Yangping and Yangqu are. ‧ 國. 學. high level tones, HH. However, their sandhi tones are different when they precede Yin. ‧. tones. When they precede Yin tones, Yangping becomes HM and Yangqu becomes. Nat. io. sit. y. MM. Hsu (2008) points out that it is due to the diachronic tonal change.. n. al. er. Since the internal structure of tone will be concerned in dealing with the tone. i n C sandhi, the internal structures of tones h are presented as follows: engchi U. 3. v. Yangqu in Lü (1993) was specified as MM; however, according to the teaching materials, Hsu (2005, 2008) and Lin (2007), the citation tone of Yangqu should be HH..

(41) 25. (1) The internal structure of Liujia Raoping Hakka tones. 學. ‧ 國. 立. 政治大. 3.2 Tone Sandhi Patterns. ‧. The disyllabic tone sandhi patterns of Liujia Raoping Hakka are illustrated as. n. al. er. io. sit. y. Nat. follows:. Ch. engchi. i n U. v.

(42) 26. (2) Liujia Raoping tone sandhi patterns Yinshang HM [Hr, hl] LL-HM Lr-Hr l-hl MM-HM Lr-Hr h-hl H-HM Hr-Hr h-hl HM-HM Hr-Hr hl-hl MM-HM Lr-Hr h-hl M-HM Lr-Hr h-hl. 立. Yinru M [Lr, h] LL-M Lr-Lr l-h MM-M Lr-Lr h-h H-M Hr-Lr h-h HM-M Hr-Lr hl-h MM-M Lr-Lr h-h M-M Lr-Lr h-h. Yangping HH1 [Hr, h] LL-HH Lr- Hr l-h MM-HH Lr-Hr h-h H-HH Hr-Hr h-h MM-HH Lr-Hr h-h MM-HH Lr-Hr h-h M-HH Lr-Hr h-h. Yangqu HH2 [Hr, h] LL-HH Lr-Hr l-h MM-HH Lr-Hr h-h H-HH Hr-Hr h-h MM-HH Lr-Hr h-h MM-HH Lr-Hr h-h M-H Lr-Hr h-h. 政治大. 學. 1st σ Yinping LL [Lr, l] Yinshang HM [Hr, hl] Yinru M [Lr, h] Yangping HH1 [H, h] Yangqu HH2 [Hr, h] Yangru H [Hr, h]. Yinping LL [Lr, l] LL-LL Lr-Lr l-l MM-LL Lr-Lr h-l H-LL Hr-Lr h-l HM-LL Hr-Lr hl-l MM-LL Lr-Lr h-l M-LL Lr-Lr h-l. Yangru H [Hr, h] LL-H Lr-Hr l-h MM-H Lr-Hr h-h H-H Hr-Hr h-h MM-H Lr-Hr h-h MM-H Lr-Hr h-h M-H Lr-Hr h-h. ‧. ‧ 國. 2ndσ. y. Nat. er. io. sit. The bold tones are ones which undergo tone sandhi. Yangping and Yangqu are indexed as HH1 and HH2 respectively in order to avoid confusion. Among the six. n. al. Ch. engchi. i n U. v. tones, LL never changes, but the other five tones change in the first syllable. The tone sandhi process does not occur in the second syllable. In this case, tone sandhi in this language can basically be regarded as the positional tone sandhi (Chen 1987, 2000, Bao 2011). Positional tone sandhi refers to the sandhi behavior which is conditioned by the position of the tone undergoing tone sandhi. That HM and HH2 becomes MM is consistent with the universal tonal markedness, i.e. falling tones and high tones are relatively more marked tones in.

(43) 27. terms of minimizing articulatory effort (Yip 2001, 2002). This universal tendency could be regarded as the motivation of tone sandhi. In this language, the requirement of no falling tone and no high level tone is the trigger of tone sandhi. The mechanism of how the output MM is produced will be illustrated in (3).. (3) Yinshang and Yangqu tone sandhi mechanism a. Yinshang + Yinping. 政治大. HM-LL → MM-LL [fo tʃʰa] ‘train’. 立. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. b. Yangqu + Yangping. Ch. HH2- HH1 → MM-HH1 [kʰiu. engchi. i n U. v. ŋien] ‘last year’. In (3), example (a) illustrates Yinshang HM followed by Yinping LL and example.

(44) 28. (b) illustrates Yangqu HH2 followed by Yangping LL. Both of them follow the requirement of no falling tone and no high level tone; therefore, they will not map into HM and HH. In (3a), the HM [Hr, hl] becomes MM [Lr, h] instead of LL [Lr, l] because of preserving the left tone melody of the citation tone. Also in (3b), the HH2 becomes MM owning to the preservation of the left tone melody. The tonal markedness tendency and the left tone melody preservation require MM to be a low. 政治大. register high tone, [Lr, h], in this language. I argue that MM is a low register high tone,. 立. [Lr, h], in this language. If it is a high register low tone [Hr, l], HM [Hr, hl] would. ‧ 國. 學. become LL [Lr, L]. In addition, the preservation is based on Yip’s (2001) argument of. ‧. tonal target. Yip (2001) mentioned that most contour tones in Chinese languages only. Nat. io. sit. y. have the initial target, and the contour moves away from the target. According to the. al. er. spetrogram in (4), the falling tone here is lack of the final target. In (4), the falling. n. v i n tone HM has a short high plateauC in h the beginning but U e n g c h i no low plateau in the final position. Hence, it is reasonable to believe that contour tones in this language have only initial target in this language. In this case, the left tone melody is preserved since a tone (contour or level) always has the initial tonal target, but only the level tone has two targets..

(45) 29. (4) The initail target of Yinshang [ho] HM ‘good’. 立. 政治大. ‧ 國. 學. The two ru tones alternate with each other in the first syllable. It follows the positional tone sandhi pattern. The motivation for the ru tone sandhi may be the. ‧. requirement which prevents citation ru tones from surfacing. This kind of motivation. sit. y. Nat. io. al. n. non-final position.. er. can be seen in the Taiwanese tone sandhi which forbids citation tone surfacing in the. Ch. engchi. i n U. v. As for the Yangping tone sandhi, it is influenced by the following Yin tones. The mechanisms of how the Yangping tone sandhi is influenced by the context are illustrated in (5) and (6) respectively..

(46) 30. (5) Yangping-Yinping contextual tone sandhi HH1-LL → HM-LL [vu ʃin] ‘body’. 政治大. In (5), Yangping is followed by Yinping. In this context, the intersyllabic. 立. -melody of Yangping is assimilated by the -melody of Yinping; therefore, the. ‧ 國. 學. HH-to-HM mapping is an assimilation process. Furthermore, when the second. ‧. syllable does not contain an. tone melody in the intersyllabic position, this. y. Nat. al. er. io. dissimilation at the tone melody level when HH1 is followed by. HM and M. Adjacent. n. There is an. sit. assimilation process will not occur.. Ch. engchi. i n U. tone melodies are not allowed, so the. v. tone melody in the. first syllable is dissimilated to . The mechanism is shown in (6a) and (6b)..

(47) 31. (6) Dissimilation tone sandhi mechanism a. HH1-HM → HM-HM geu] ‘yellow dog’. b. HH1-M → HM-M ŋit] ‘tomorrow’. 立. 政治大. 學 ‧. ‧ 國. [ʃau. io. sit. y. Nat. al. n. Both (6a) and (6b) show that two adjacent. Ch. engchi U. er. [voŋ. melodies in the intersyllabic. v ni. position are forbidden in these two particular situations. The right tone melody of the first tone is then dissimilated to . What is interesting in (5-6) is that the two phonological processes, assimilation and dissimilation, take place at the same level, and it correctly accounts for the tone sandhi patterns..

(48) 32. 3.3 OT analysis 3.3.1 Positional tone sandhi According to the analyses in section 3.2, the tone sandhi is mainly positional and the tonal markedness tendency is the motivation. In terms of the positional tone sandhi, tone sandhi processes only occur in the first syllable in Liujia Raoping Hakka, i.e. the second syllable never undergoes tone sandhi. Under the framework of OT, we. 政治大. need a constraint to prevent the tone in the right syllable from alternating. Therefore,. 立. the constraint, IDENT-IO-T-R, is needed. IDENT-IO-T-R is adopted from Hsiao. ‧ 國. 學. (2000). This constraint penalizes candidates whose tone in the second syllable is not. ‧. identical with its corresponding tone in the output. Its definition will be shown in (7).. al. er. io. sit. y. Nat (7) IDENT-IO-T-R (ID-T-R):. n. v i n Assign one violation mark forC every the right edge which is not h einput n g ctoneh iat U. identical with its corresponding tone in the output (Hsiao 2000).. The constraint in (7) follows the concept of edge prominence (Li 2003). Lin (2011) investigated the disyllabic tone sandhi in Dongshi-Hakka and posited IDENT-IO-T-HD to capture the preservation of the tone in the right syllable. This constraint is not adopted here because the term “head” may easily mislead readers to.

(49) 33. the idea of stress. The syllable with stress may be regarded as head in languages with stress. Duanmu (1990) proposed the non-head stress principle (NHS) to govern the stress in Chinese. In NHS, the stress is assigned to the syntactic non-head position. It may account for the tone sandhi which is sensitive to the syntactic structure. However, Raoping tone sandhi is syntactically insensitive, which makes it hard to determine the stress assignment according to the NHS. Since the existence of stress in Chinese. 政治大. dialects is still controversial, the term “head” is not adopted in order to avoid. 立. confusion.. ‧ 國. 學. In this thesis, the preservation of tone is argued to be a typologically different. ‧. phenomenon. Languages retaining the tone in the left syllable are attested, such as. Nat. io. sit. y. Hakha-Lai (Lin 2005a) and Chengdu (Lin 2006). Besides, languages retaining the. al. er. tone in the right syllable are attested as well. Beijing Mandarin (Hsiao 2000, Lin. n. v i n C h Boshan (Lin 2004), 2008), Taiwanese (Hsiao 2000), e n g c h i U Sixian-Hakka (Lin 2005b) and Dongshi-Hakka (Lin 2011) are all languages which preserve the tone in the right syllable in the disyllabic tone sandhi process. In this case, the retained tone may be considered to be typologically different. The analyses in chapter four will support this argument. In addition to the position, the tonal markedness may be regarded as one of the factors influencing the tone sandhi. Yip (2001) mentioned that falling tones and high.

(50) 34. tones are comparatively more marked than mid and low tones in terms of minimizing the articulatory effort. Accordingly, *Fall and *HH are posited. The definitions will be shown in (8) and (9).. (8) *Fall Assign one violation mark for every output falling tone.. 立. Assign one violation mark for every output high level tone.. 學. ‧ 國. (9) *HH. 政治大. ‧. Nat. in (8) and (9). Its definition will be shown in (10).. n. al. Ch. engchi. er. io. sit. y. Furthermore, IDENT-IO-T is needed to interact with the markedness constraints. i n U. v. (10) IDENT-IO-T (ID-T): Assign one violation mark for every input tone which is not identical with its corresponding tone in the output (Hsiao 2000, Lin 2011).. The interactions among constraint (7), (8), (9) and (10) is displayed in (11)..

(51) 35. (11) Temporal ranking 1 ID-T-R *Fall. *HH ID-T. ID-T-R is always undominated in this language because the tone in the right syllable never undergoes tone sandhi. All the candidates whose tone in the second. 政治大. syllable is changed will be ruled out by ID-T-R. Due to the undominated ranking,. 立. these situations will no longer be discussed in the rest of this chapter. The ranking. ‧ 國. 學. ID-T-R » *Fall, *HH illustrates the feature of positional tone sandhi and universal. ‧. tonal markedness. ID-T-R placed undominated exhibits that tone sandhi always. y. Nat. al. er. io. sit. occurs in the left syllable. As for the universal tonal markedness, *Fall and *HH are. n. comparatively more marked due to the ease of articulation. In this case, it is easier to. Ch. engchi. i n U. v. preserve unmarked tones, MM and LL. Furthermore, As for markedness constraints, according to Yip (2001), *Fall should have outranked *HH. However, in this language, there is no crucial ranking between *Fall and *HH nonetheless. The markedness conflation proposed by de Lacy (2004) is then considered, i.e. in this language, *Fall and *HH are equally ranked. Yinshang HM will be the example to display the violation of constraints in (11) and the tableau is in (12)..

(52) 36. (12) Yinshang HM in the temporal ranking 1 HM-LL → MM-LL [fo. tʃʰa]. ‘train’. HM-LL [Hr, hl]-[Lr, l]. *Fall. a.HM-LL [Hr, hl]-[Lr, l]. *!. b.ML-LL [Lr, hl]-[Lr, l]. *!. c.HH-LL [Hr, h]-[Lr, l]. *HH. ID-T. *!. *. ? d.LL-LL [Lr, l]-[Lr, l]. 立. ‧. ‧ 國. *. 學. ☞e.MM-LL [Lr, h]-[Lr, l]. 政治大. *. In (12), tones such as MH and LM are not considered because they are not. sit. y. Nat. io. n. al. er. allowed to appear in the output in this language. They may automatically be ruled out. i n U. v. by the undominated phonotactic constraint, *Rise. Moreover, although the ML in. Ch. engchi. candidate (b) does not exist in this language, either, it is unnecessary to posit a constraint like *ML. It is because the ML will be rule out by *Fall as it is shown in (12). Since ML and HM perform identically in terms of the constraint violation, only HM which is in the tone inventory will be discussed in the following tableaux. Besides, the ru tone candidates will be discussed in the next section. In this tableau, *Fall rules out candidate (a) and (b) due to the falling tones whereas *HH ruled out candidate (c) due to the high level tone. The constraints in (11) illustrate the.

(53) 37. motivation and the preservation of the tone sandhi clearly, yet this temporal ranking fails to choose the optimal output between candidate (d) and candidate (e). Hence, the preservation mechanism in (3) should be considered. In (3), the left tone melody is preserved in tone sandhi based on the tonal target proposed by Yip (2001). According to Yip (2001) and the spectrogram in (4), the falling tone only has the initial target. The tonal target needs to be preserved in the tone sandhi process. In this case, the. 政治大. constraint IDENT-IO-t-L is important here. Its definition is in (13).. (13) IDENT-IO-t-L (ID-t-L):. 學. ‧ 國. 立. ‧. Assign one violation mark for every input tone whose left tone melody is different. Nat. al. er. io. sit. y. from its corresponding tone melody in the output (Lin 2011).. n. v i n C h posited by Lin The ID-t-L in (13) is originally e n g c h i U (2011). Lin (2011) proposed this constraint based on Beckman (1998) and Nelson (1998, 2003) who consider the left element preservation to be the universal tendency. However, in this thesis, the preservation of the tone element is regarded as the typological difference. That languages have different sides of syllable undego tone sandhi is due to different rankings of preservation constraints, as it will be presented in chapter four. Therefore, the constraint in (13) is based on Yip (2001) and adapted to preserve the initial tonal.

(54) 38. target. Including ID-t-L, the temporal ranking should be revised as in (14).. (14) Temporal ranking 2 *Rise ID-T-R *Fall. ID-t-L. *HH. ID-T. 政治大. 立. In (14), there is no interaction among ID-t-L and other constraints so far in this. ‧ 國. 學. temporal ranking. The ranking of ID-t-L will be revised in the next section. With. ‧. ID-t-L, LL will be ruled out and MM will successfully be chosen as the optimal. Nat. n. al. er. io. sit. y. output. The tableau is given in (15).. Ch. (15) Yinshang HM in the temporal ranking 2 HM-LL → MM-LL tʃʰa]. ‘train’. HM-LL [Hr, hl]-[Lr, l]. *Fall. [fo. *HH. engchi. ID-T. ID-t-L. c.LL-LL [Lr, l]-[Lr, l]. *. *!. ☞d.MM-LL [Lr, h]-[Lr, l]. *. i n U. v. In (15), candidate (c) is ruled out by ID-t-L because the -melody of LL is.

(55) 39. different from the -melody of HM. Therefore, MM is selected as the optimal output. The tableau in (15) explains why the comparatively more marked MM, rather than the least marked LL, is chosen as the sandhi tone. In (15), it suggests that the tonal markedness tendency could be regarded as the motivation of the positional tone sandhi (Chen 1987, 2000, Bao 2011). In this language, *Fall and *HH are ranked equally which may be considered to be the. 政治大. markedness conflation (de Lacy 2004). Moreover, in the tone sandhi process, the left. 立. tone melody, the initial tonal target, is preserved.. ‧. ‧ 國. 學. 3.3.2 Contextual tone sandhi of Yangping. Nat. dissimilation process proposed in (5) and (6).. n. al. Ch. engchi. er. io. sit. y. The residual issue about non-ru tones is how OT deals with the assimilation and. i n U. v. 3.3.2.1 Yangping-Yinping contextual tone sandhi In (5), Yangping HH1 illustrates the assimilation at the tone melody level when it is followed by LL. The constraint ranking in (14) fails to select the optimal output for HH1-LL. The optimal output of HH1-LL is HM-LL which will be ruled out by *Fall in the temporal ranking. Since it is an assimilation process, the constraint, Agree-t (Lin 2011), is considered..

(56) 40. (16) Agree-t: Assign one violation mark for adjacent tones which have different tone melodies in the intersyllabic position.. As for the ranking of Agree-t, it has to be ranked at the lowest level with ID-T and dominated by ID-t-L. If Agree-t is ranked higher, it will predict the wrong output. 政治大. forms for unchanged tonal pairs, such as LL-HM or LL-M. In order to illustrate the. 立. assimilation, a higher ranked constraint is needed. Notice that Agree-t has to be. ‧ 國. 學. confined to a specific context, i.e. where HH1 precedes LL; otherwise, HM-HM, the. ‧. optimal output of HH1-HM, will be wrongly eliminated. To solve this problem,. Nat. n. al. er. io. sit. y. [Agree-t & *LL]ADJ is posited.. (17) [Agree-t & *LL]ADJ:. Ch. engchi. i n U. v. Assign one violation mark for adjacent tones which violate both Agree-t and *LL.. In (17), the component constraint, *LL, has to be ranked lower than ID-T so that the unchanged tonal pair, LL-LL, will not be incorrectly eliminated. [Agree-t & *LL]ADJ penalizes adjacent tones which violate both Agree-t and *LL. Therefore, only adjacent tones including a low level tone may violate this constraint. The domain, two.

(57) 41. adjacent tones, is based on Lin (2011) and Lubowicz (2002). Lin (2011) has proposed constraints, such as [NoJump-t & *HD/ ]ADJ, to account for Dongshi Hakka. The domain of this constraint focuses on two adjacent tones. Therefore, the constraint in (17) follows Lin’s (2011) idea and its domain is defined as two adjacent tones. This constraint has to dominate *Fall to preserve the optimal output HM-LL. Tableau (18) displays how [Agree-t & *LL]ADJ works.. 政治大 (18) [Agree-t & *LL] 立 predicts the output of HH1-LL ADJ. ʃin] ‘body’. b.HH-LL [Hr, h]-[Lr, l] c.LL-LL [Lr, l]-[Lr, l] d.MM-LL [Lr, h]-[Lr, l]. *!. engchi *. *!. er. i n U *. *. *! *. y. ‧. *. *LL. ☞a.HM-LL [Hr, hl]-[Lr, l]. Ch. ID-T. Agree-t. *HH. n. al. *Fall. [Agree-t & *LL]ADJ. o ati NID-t-L. HH1-LL [Hr, h]-[Lr, l]. sit. ‧ 國. [vu. 學. HH1-LL → HM-LL. v. * *. *. **. *. *. [Agree-t & *LL]ADJ has to outrank *Fall for producing the optimal output, HM..

(58) 42. Also, it has to be dominated by ID-T-R in order not to rule out unchanged tonal pairs, such as LL-HM or LL-HH. Besides, ID-t-L dominates *Fall in order to prevent HH1 and HH2 from mapping to LL. Candidates (b) and (d) violate [Agree-t & *LL]ADJ because there is a low level tone and the intersyllabic tone melodies are different. For both candidates, the right tone melody of the first syllable is. which shows no. agreement with the left tone melody of the second syllable, . Candidate (c) is ruled. 政治大. out by ID-t-L because the -melody of LL is not identical with the left tone melody of. 立. the input. This conjunction constraint successfully selects the optimal output; it would. ‧ 國. 學. have the wrong prediction if the input were not Yangping HH1. The example is given. ‧. io. sit. y. Nat. n. al. er. in (19).. Ch. engchi. i n U. v.

(59) 43. (19) [Agree-t & *LL]ADJ has the wrong prediction HH2-LL → MM-LL [tʰien. tʃʰa] ‘tram’. 立. *. *. **. *?. *. *. *. 學. ‧ 國. *. 政* 治大* *. *!. ☞d.MM-LL [Lr, h]-[Lr, l]. *LL. *!. ID-T. Agree-t. *. b.HH-LL [Hr, h]-[Lr, l] c.LL-LL [Lr, l]-[Lr, l]. *HH. ?a.HM-LL [Hr, hl]-[Lr, l]. *Fall. [Agree-t & *LL]ADJ. ID-t-L. HH2-LL [Hr, h]-[Lr, l]. ‧ sit. y. Nat. n. al. er. io. In (19), the input is Yangqu HH2 which is identical with Yangping in the citation. Ch. i n U. v. form, but their sandhi tones are different. HH2 always changes to MM in the first. engchi. syllable. Then [Agree-t & *LL]ADJ fails to select the right output. In order to solve this problem, [Agree-t & *LL]ADJ should be indexed to Yangping, i.e. only when the input contains Yangping HH1, [Agree-t & *LL]ADJ works. Accordingly, [Agree-t & *LL]ADJ then should be [Agree-t & *LL]ADJ(Yangping). The constraint dominating *Fall is actually [Agree-t & *LL]ADJ(Yangping) instead of [Agree-t & *LL]ADJ. This constraint is indexed and specifically targeting the input containing Yangping. It follows the idea.

(60) 44. of Pater’s (2007, 2010) morpheme specific phonology. Pater (2007, 2010) argued that if two indistinguishable morphemes undergo or trigger different processes, it is hard to explain purely phonologically. The issue here is not about indistinguishable morphemes but the diachronically merged tones. Since Yangping and Yangqu are indistinguishable and they undergo or trigger different processes, they conform to the requirement of Pater’s (2007, 2010) morpheme specific phonology. The indexed. 政治大. constraint suggests that the distinction between HH1 and HH2 does exist in native. 立. speakers’ intuition.. ‧ 國. 學. Moreover, Fukuzawa (1999) and t and Mester (1999, 2001) believe that the. ‧. indexed approach can only be applied on faithfulness constraints. However, Pater. Nat. io. sit. y. (2010) investigated the Yine syncope and argued that both markedness and. n. al. er. faithfulness constraints are able to be indexed. Hence, the markedness constraint in. Ch. (17) is defined as an indexed constraint.. engchi. i n U. v. To deal with the morpheme specific issue, researchers, such as Orgun (1996, 1998, 1999), Anttila (1997, 2002), Inkelas (1998), Anttila and Cho (1998), Orgun and Inkelas (2002), Inkelas and Zoll (2005, 2007) among others, argue for the morpheme specific rankings, namely cophonologies. Nevertheless, this approach is not appropriate for predicting historically merged tones. Cophonology requires the morphological motivation or the partially unranked constraints. The merged tones do.

(61) 45. not provide the morphological motivation. Moreover, there are no partially unranked constraints; thus, the indexed approach is preferred in this thesis. The situation in (19) is no longer a problem in (20).. (20) [Agree-t & *LL]ADJ(Yangping) predicts the output of HH2-LL HH2-LL → MM-LL tʃʰa] ‘tram’. *. *. n. al. sit. y. *. ‧. ‧ 國. 學. *LL. ID-T. Agree-t. [Agree-t & *LL]ADJ. *HH. *Fall. io. ☞d.MM-LL [Lr, h]-[Lr, l]. *!. Nat. a.HM-LL [Hr, hl]-[Lr, l]. [Agree-t & *LL]ADJ. (Yangping). 立. ID-t-L. HH2-LL [Hr, h]-[Lr, l]. 政治大. *. *. *. er. [tʰien. Ch. engchi. i n U. v. In (20), [Agree-t & *LL]ADJ(Yangping) has to outrank *Fall whereas *Fall has to outrank [Agree-t & *LL]ADJ. [Agree-t & *LL]ADJ(Yangping) outranks *Fall in order to deal with the situation where HH1 is the input, as in (18). *Fall has to outrank [Agree-t & *LL]ADJ; otherwise, candidate (d), MM-LL, cannot be the optimal output. Moreover, according to the indexed approach schema (Fukuzawa 1999, t and Mester 1999, 2001, Pater 2000, 2007, 2010), the indexed constraint should outrank.

(62) 46. the general constraint. On the other hand, [Agree-t & *LL]ADJ must dominate Agree-t, according to the conjunction schema (Smolensky 1993) that component constraints has to ranked lower than the conjunction constraint. Last, in order to preserve the unchanged tonal combination, such as LL-HH1, IDENT-LL is proposed. It prevents LL from changing. Since LL never undergoes tone sandhi, IDENT-LL is undominated. The constraint ranking is revised as in (21).. 立. 政治大. 學. *Rise. ID-LL. ‧. ID-T-R. io. sit. y. Nat [Agree-t & *LL]ADJ(Yangping). n. al. *Fall. Ch. i n U. e n g c h i *HH. [Agree-t & *LL]ADJ. Agree-t. er. ID-t-L. ‧ 國. (21) Temporal constraint ranking 3. ID-T. *LL. v.

(63) 47. 3.3.2.2 Yangping-Yinshang contextual tone sandhi There is a dissimilation process at the contour level when Yangping HH1 is followed by Yinshang HM. In OT, the dissimilation of high tone melodies can be captured by OCP- .. (22) OCP- :. 政治tone melody. 大. Assign one violation mark for adjacent. 立. ‧ 國. 學. OCP- has to be dominated by Agree-t; otherwise, it will incorrectly rule out. ‧. MM-HM, the optimal output of HH2-HM. However, if it is ranked this low, the. Nat. n. al. er. io. sit. y. optimal output of HH1-HM will be wrongly predicted, as in (23).. Ch. engchi. i n U. v.

(64) 48. (23) OCP- has the wrong prediction HH1-HM → HM-HM [voŋ geu] ‘yellow dog’. *. *. *. *. * *. *. ‧. ‧ 國. *. *LL. * * 治政大 *!. 學. [Lr, h]-[Hr, hl]. *. OCP-. ?d.MM-HM. *!. ID-T. c.LL-HM [Lr, l]-[Hr, hl]. 立*. Agree-t. b.HH-HM [Hr, h]-[Hr, hl]. [Agree-t & *LL]ADJ. **?. *HH. ☞a.HM-HM [Hr, hl]-[Hr, hl]. *Fall. [Agree-t & *LL]ADJ. (Yangping). ID-t-L. HH1-HM [Hr, h]-[Hr, hl]. n. er. io. sit. y. Nat. al. i n U. v. In (23), candidate (c) violates ID-t-L because the left tone melody of LL is. Ch. engchi. different from the left tone melody of HH1. Candidate (b) is ruled out because it violates both *Fall and *HH. Candidate (a) should have been the optimal output, but it consists of two falling tones and violates *Fall twice. In this case, candidate (d) is wrongly selected as the optimal output. In order to capture the dissimilation and promote the effect of OCP- , OCP- needs to target this specific context, i.e., where HH1 followed by HM. Therefore, [OCP- & *Fall]ADJ is posited..

(65) 49. (24) [OCP- & *Fall]ADJ: Assign one violation mark for adjacent tones which violate both OCP- and *Fall.. *Fall has to be dominated by [OCP- & *Fall]ADJ(Yangping) based on the constraint conjunction schema (Smolensky 1993). Besides, *Fall restricts the domain of OCPwhen they are conjoined. [OCP- & *Fall]ADJ has to be dominated by ID-T-R in order. 政治大. to confirm that the tone in the second syllable will not be changed.. 立. As in (18) and (19), the conjunction constraints could account for the specific. ‧ 國. 學. situation, but it could rule out other optimal output forms whose input are not. ‧. Yangping. Tableaux in (25) and (26) will illustrate the situations which [OCP- &. Nat. n. al. er. io. sit. y. *Fall]ADJ may have a right or wrong prediction.. Ch. engchi. i n U. v.

(66) 50. (25) [OCP- & *Fall]ADJ predicts the output of HH1-HM HH1-HM → HM-HM [voŋ. geu] ‘yellow dog’. *. *!. *. *. * *. *. *. *. *. *. ‧. ‧ 國. *. *LL. *. OCP-. ** 治 * 政大. 學. [Lr, h]-[Hr, hl]. ID-T. d.MM-HM. *!. Agree-t. c.LL-HM [Lr, l]-[Hr, hl]. 立. *!. [Agree-t & *LL]ADJ. b.HH-HM [Hr, h]-[Hr, hl]. *HH. ☞a.HM-HM [Hr, hl]-[Hr, hl]. *Fall. [Agree-t & *LL]ADJ. (Yangping). [OCP- & *Fall]ADJ. ID-t-L. HH1-HM [Hr, h]-[Hr, hl]. Nat. al. er. io. sit. y. *. n. In (25), it displays that [OCP- & *Fall]ADJ has to dominate *Fall in order not to. Ch. engchi. i n U. v. rule out the optimal output. Candidate (c) violates ID-t-L because the left tone melody of LL is different from the left tone melody of input HH1. Candidate (b) and (d) are ruled out by [OCP- & *Fall]ADJ because they contain HM and the adjacent tone melodies are ’s. In this case, candidate (a) is successfully selected as the optimal output. This is where [OCP- & *Fall]ADJ has the correct prediction. However, it faces a challenge in (26)..

(67) 51. (26) [OCP- & *Fall]ADJ has the wrong prediction HH2-HM → MM-HM [ŋi. ʃiu] ‘second-hand’. *. *. *. ‧. *?. *. 學. ‧ 國. *. *LL. ID-T *. OCP-. Agree-t. * *. [Lr, h]-[Hr, hl]. *. 政 **治大. *!. ☞d.MM-HM. [Agree-t & *LL]ADJ. *HH. c.LL-HM [Lr, l]-[Hr, hl]. *Fall. 立*!. b.HH-HM [Hr, h]-[Hr, hl]. [Agree-t & *LL]ADJ. ?a.HM-HM [Hr, hl]-[Hr, hl]. (Yangping). [OCP- & *Fall]ADJ. ID-t-L. HH2-HM [Hr, h]-[Hr, hl]. *. *. *. Nat. al. er. io. sit. y. *. n. In (26), [OCP- & *Fall]ADJ has the wrong prediction when the input is Yangqu. Ch. engchi. i n U. v. HH2 instead of Yangping HH1. The optimal output of Yangqu HH2 is MM-HM. However, it is eliminated by [OCP- & *Fall]ADJ due to the existing contour tone and the adjacent. tone melody. Candidate (a) is thus incorrectly predicted as the optimal. output. Hence, [OCP-. & *Fall]ADJ should be confined to the situation which. Yangping is the input form. In this case, it should be indexed as [OCP-. &. *Fall]ADJ(Yangping) based on the concept of morpheme specific phonology (Pater 2007, 2010). Including [OCP- & *Fall]ADJ(Yangping), the situation in (26) will be solved, as in.

(68) 52. (27).. (27) [OCP- & *Fall]ADJ(Yangping) predicts the output of HH2-HM HH2-HM → MM-HM ʃiu] ‘second-hand’. OCP-. *. 學. *. *. *. n. Ch. engchi. *. * *. *. y. *. *. sit. *. *!. *. er. *. ‧. io. *. *LL. ID-T. *. 政治大 **. al. Agree-t. [Agree-t & *LL]ADJ. *HH. *Fall. [OCP- & *Fall]ADJ. *!. [Agree-t & *LL]ADJ. (Yangping). [OCP- &. ☞d.MM-HM [Lr, h]-[Hr, hl]. 立. Nat. c.LL-HM [Lr, l]-[Hr, hl]. *Fall]ADJ. b.HH-HM [Hr, h]-[Hr, hl]. (Yangping). a.HM-HM [Hr, hl]-[Hr, hl]. ID-t-L. HH2-HM [Hr, h]-[Hr, hl]. ‧ 國. [ŋi. i n U. v. In (27), that [OCP- & *Fall]ADJ(Yangping) dominates [OCP- & *Ctr]ADJ follows the indexed approach schema (Fukuzawa 1999, t and Mester 1999, 2001, Pater 2000, 2007, 2010). The indexed constraint need to dominate the original one. On the other hand, that [OCP- & *Fall]ADJ(Yangping) dominates *Fall follows the constraint conjunction schema (Smolensky 1993), i.e. the conjunction constraint needs to dominate the component constraints. Moreover, this ranking prevents the optimal.

(69) 53. output of HH1-HM from being eliminated by *Fall, as it is shown in (26). The temporal constraint ranking is revised and illustrated in (28).. (28) Temporal ranking 4 *Rise. ID-t-L. ID-LL. 立. 政治大. [Agree-t & *LL]ADJ(Yangping). ID-T-R. [OCP- & *Fall]ADJ(Yangping). ‧ 國. 學 ‧. [OCP- & *Fall]ADJ. sit. y. Nat. io. n. al. *HH. er. *Fall. Ch. [Agree-t & *LL]ADJ. engchi. Agree-t. i n U. v. ID-T. *LL OCP-. 3.3.2.3 Yangping-Yinru contextual tone sandhi As for the Yangping-Yinru contextual tone sandhi, there is a dissimilation at the intersyllabic tone melody level. The mechanism was shown in (6b). It encounters.