四縣客語音位排列與借字音韻學: 優選理論分析

全文

(1)國立臺灣師範大學英語學系碩士論文 Master Thesis Department of English National Taiwan Normal University. 四縣客語音位排列與借字音韻學:優選理論分析 Sixian-Hakka Phonotactics and Loanword Phonology: An Optimality-theoretic Analysis. 指導教授 : 林蕙珊博士 Advisor: Dr. Hui-Shan Lin 學生: 廖書儀 Student: Shu-Yi Liao 中華民國一○三年六月. June, 2014.

(2) 摘要. 本研究旨在探討四縣客語的音位排列 (phonotactics) 與其借字音韻學 (loanword phonology)。研究方法採用古典優選理論 (Classical Optimality Theory (McCarthy & Prince 1993, Prince & Smolensky 1993/2004)) 以及相關的衍生理論，包括聯合制約 (Local Conjunction (Smolensky 1993)) 和評價排序模組 (Rank-Ordering Model of EVAL (Coetzee 2006)) 。過去鮮少以制約排序的手法來分析四縣客語的音位排列與借字音韻學。音位排列方面會論及四種音韻現象:由 /, , /至 [, , ] 的衍生、韻母的形成、雙母音的結構，以及由 /, , / 至 [, , ] 的生成變化。關於借字音韻學，本論文探討源自於日語的借字。語料來源為台灣行政院客家委員會所發行的<<客語外來語：含原閩客國語互借詞(四縣腔)>>(2011a)。借字音韻學的分析著重於音段改適 (segmental adaptation)。音段改適是指語音在不同語言間借予的過程中所發生的變化。在借字的過程中，聽話者會將非母語語音錯誤聽辨成相似的母語語音。決定音位排列的制約，在討論音段改適是否恰當時扮演著極為關鍵的角色。針對音位排列所提出的數條制約，在應用於評斷借字的變化時，會聚合成一個制約組。此制約組內含許多標記制約 (markedness constraint)，總稱為 OK-σ。在討論借字音韻學的過程中，除了 OK-σ 之外，並還有許多信實制約 (faithfulness constraint) 作用於其中。信實制約的提出，目的在於準確預測最佳輸出值 (optimal output)，並且詳細描述音段在借予的過程中是如何改適。所有的信實制約皆和語音特徵 (phonetic feature) 有相關，故本文也提供日語和四縣客語的語音矩陣 (feature matrix)。就借字音韻學而論，本文研究結果與支持兩階段改適的觀點一致 (Kenstowicz 2005, Silverman 1992, Yip 2006)。第一階段的改適著重於非母語語音變化至母語語音的過程， i.

(3) 亦是上述所提的音段改適。第二階段的改適則著重評斷改適過後的音串(sound sequence) 是否滿足該語言的音位排列要求。. 關鍵字: 四縣客語、音位排列、借字音韻學、優選理論. ii.

(4) Abstract. This research studies Sixian-Hakka (S-Hakka) phonotactics and loanword phonology by adopting Optimality Theory (OT). Several theories related to OT are taken advantage of to account for various phenomena, including Classical Optimality Theory (McCarthy & Prince 1993, Prince & Smolensky 1993/2004), Local Conjunction (Smolensky 1993) and RankOrdering Model of EVAL (Coetzee 2006). Both S-Hakka phonotactics and loanword phonology have never been analyzed through constraint competition. For phonotactics, four phonological phenomena are in question. First, an argument for the derivation from /, , / to [, , ] is provided. Second and third are about rhymes and diphthongs. Fourth, the derivation from /, , / to [, , ] is explained. For loanwords, words that are originated in Japanese are collected from a loanword dictionary Keyu Wailaiyu: Han Yuan Min Ke Guo Yu Hujieci (Sixian Qiang) [Loanwords in Sixian-Hakka: Including words shared among Taiwanese, Hakka and Mandarin] (2011a). Issues about loanwords focus on segmental adaptations which refer to the non-native segments transform to the similar ones that are native in listeners’ inventory. Constraints found to be active in S-Hakka phonotactics play an essential role to judge whether a certain loanword adaptation is acceptable or not in the language. Constraints that are proposed for S-Hakka phonotactics are grouped together as one constraint package when they are applied to analyze loanword adaptations. The constraint package consists of several markedness constraints and is named as OK-σ. When loanword phonology is in question, in addition to OK-σ, a few faithfulness constraints are put forward in order to correctly predict the optimal output and elaborately describe how a segment is adapted when the loan process occurs. The faithfulness constraints are related to features so feature matrices of Japanese and S-Hakka inventories can be seen. Regarding loanword phonology, the findings of the research are consistent with the iii.

(5) perspective that claims two steps of transformations (Kenstowicz 2005, Silverman 1992, Yip 2006). First, non-native segments are adapted and turn out as the segments that exist in listeners’ phonetic inventory. Second, after non-native sounds are adapted, the whole sound sequences have to be judged by the phonotactics of the target language.. Key words: Sixian-Hakka, phonotactics, loanword phonology, Optimality Theory. iv.

(6) Acknowledgement. The completion of this thesis means the end of my graduate study and the embarking on another journey of my life. Before moving on to the next journey, I am here to show my utmost appreciation to people who have helped and supported me during my master program in NTNU. First of all, I was fortunate to have Prof. Hui-Shan Lin as my advisor who always provided constrictive suggestions when I came across questions. Her soothing voice and infinite patience led me to conduct and finish the research. Additionally, it was her course in Optimality Theory that stimulated my passion on phonology. I would also like to thank Prof. Chin-Wei Wu and Prof. Hsiu-Hsueh Liu as my committee members and they gave invaluable comments on my thesis. Next, my thankfulness goes to the teachers who have taught me in the linguistic classes in NTNU: Prof. Chun-Yin Doris Chen, Prof. Hui-Shan Lin, Prof. Jen-I Li, Prof. Miao-Hsia Chang , just to name a few. They have guided me to the different fields of linguistics and with their professional instructions I have enriched and broadened my knowledge. My gratitude goes to my lovely classmates and friends as well: Chia-Lin Hsieh, Chung-Yin Tsai, Wan-Ling Tang, Yan-Ru Hsieh, Yue-Zheng Wu, to name just a few. My classmates were always kind, supportive and inspiring in classes. With them, it has been joyful studying in NTNU. ChungYin and Wan-Ling as my roommate and flatmate respectively gave me the best experience of dormitory life and made our common room full of laughter. Yue-Zheng who for me was the liked-minded person even provided her care from abroad. Special thanks to Yan-Ru and her mother for being my participants. Finally, I am deeply indebted to my beloved family members, my mother, father, sister and brother. With their unconditional support and warm encouragement, I finished my graduate study successfully. Last and most of all, thankfulness from the bottom of my heart goes to all. v.

(7) people who love me, without your concern and support, it is impossible for me to pursue what I truly long for.. vi.

(8) Table of Contents. 摘要 ....................................................................................................................................... i Abstract .............................................................................................................................. iii Acknowledgement ................................................................................................................ v. Chapter 1 Introduction .......................................................................................................... 1 Chapter 2 Literature Review.................................................................................................. 3 2.1 Optimality Theory ........................................................................................................... 3 2.1.1 Classical Optimality Theory ............................................................................... 3 2.1.2 Local Conjunction .............................................................................................. 5 2.1.3 Rank-Ordering Model of EVAL ......................................................................... 6 2.2 Japanese Sound Inventory and Phonology ....................................................................... 7 2.3 Sixian-Hakka Sound Inventory ........................................................................................ 9 2.4 Loanword Phonology .................................................................................................... 11 Chapter 3 Sixian-Hakka Phonotactics .................................................................................. 13 3.1 Empty Nucleus .............................................................................................................. 13 3.1.1 The Phonetic Description for S-Hakka Alveolar-dentals ................................... 13 3.1.2 The Phonetic Description for S-Hakka Derived Vowel ..................................... 15 3.1.3 An Optimality-theoretic Analysis ..................................................................... 17 3.2 Rhymes ......................................................................................................................... 21 3.3 Diphthongs .................................................................................................................... 25 vii.

(9) 3.4 Palatalization ................................................................................................................. 29 3.5 Summary for Sixian-Hakka Phonotactics....................................................................... 32 Chapter 4 Sixian-Hakka Loanword Phonology .................................................................... 34 4.1 Adaptations of Vowels ................................................................................................... 34 4.1.1 Short Vowels Adapted from Long Counterparts ................................................ 35 4.1.2 Adaptations of [] ........................................................................................... 37 4.1.2.1 Central Unrounded [] Adapted from Back Unrounded [] ................... 37 4.1.2.2 Back Rounded [] Adapted from Back Unrounded [] ......................... 41 4.1.3 Summary for the Adaptations of Vowels ........................................................... 43 4.2 Adaptations of Onset Consonants .................................................................................. 44 4.2.1 Stops ................................................................................................................ 45 4.2.1.1 Voiced Stops ......................................................................................... 45 4.2.1.2 Voiceless Stops...................................................................................... 50 4.2.2 Fricatives ......................................................................................................... 55 4.2.2.1 Alveolar-dental [] Adapted from Alveolar []....................................... 55 4.2.2.2 Alveolar-palatal [] Adapted from Palatal-alveolar [] ........................... 57 4.2.2.3 Glottal [] Adapted from Palatal [] ...................................................... 59 4.2.2.4 Labiodental [] Adapted from Bilabial [] ............................................. 61 4.2.3 Affricates ......................................................................................................... 64 4.2.3.1 Alveolar-palatal [] Adapted from Voiceless Palatal-alveolar [] ......... 64 4.2.3.2 Alveolar-palatal [] Adapted from Voiced Palatal-alveolar [] ........... 66 viii.

(10) 4.2.4 Lateral [] Adapted from Flap [] ...................................................................... 69 4.2.5 Summary for the Adaptations of Onset Consonants .......................................... 71 4.3 Adaptations of Coda Consonants ................................................................................... 72 4.3.1 Word-middle Nasal Codas ................................................................................ 73 4.3.2 Word-final Nasal Codas.................................................................................... 78 4.3.3 Summary for the Adaptations of Nasal Codas................................................... 81 Chapter 5 Conclusion .......................................................................................................... 82 Reference ............................................................................................................................ 84. ix.

(11) Chapter 1 Introduction. Three branches of Sinitic languages exist in Taiwan: Mandarin Chinese, Taiwanese and Hakka. Loanwords in the first two languages have been widely investigated in the recent years, such as English loanwords in Mandarin (Guo 1999, Hsieh, Kenstowicz & Mou 2009, Lu 2005, Wang 2010, Wu C. 2006, Wu H. 2006) or Japanese loanwords in Taiwanese (Chang 1993, Hsieh 2006, Nien 2009). Nevertheless, loanword phonology for Hakka remains unclear. Only Jiang (1998) conducts a field work to examine loanwords in Dongshi-Hakka and Wu C. (2004), based on Jiang’s linguistic data, discusses the accent-to-tone adaptation. Therefore, one of the primary goals in the thesis is to explore Hakka loanword phonology. Several Hakka dialects can be distinguished in Taiwan, such as Sixian-Hakka, Hailu-Hakka or Dongshi-Hakka. Sixian-Hakka (S-Hakka) is chosen for the current work because, compared with other Hakka dialects, it is used by the largest population. When discussing loanword phonology, it is important to know the phonotactics of the source and the target languages. Little is known about S-Hakka phonotactics so it is another main subject in the present research. In a few words, two issues are the fundamental concerns in the thesis. First, S-Hakka phonotactics will be investigated. Second, based on the phonotactics, how loanwords transform when they are borrowed from Japanese to S-Hakka will be studied. Besides, Optimality Theory is adopted to conduct the analyses. For S-Hakka phonotactics, a number of markedness constraints are set up to regulate the sound sequences. As for the loanword adaptation, the current work especially pays attention to the segmental but not suprasegmental adaptations. Consistent with previous studies (Kenstowicz 2005, Silverman 1992, Yip 2006), the present results support two steps of transformations. Segments that do not exist in listeners’ native inventory would be altered with. 1.

(12) the least significant violations of constrains which are related to features. IDENT-IO[F] captures the change of feature values, MAX-IO[F] asks for no feature deletion and DEP-IO[F] requires no feature to be added in. After the segmental adaptations, the sound sequences have to be judged by the phonotactics of listeners’ native language. The thesis is arranged as follows. Chapter 1 introduces the motivation and the main research directions. Chapter 2 summarizes previous literature. Chapter 3 provides discussions about S-Hakka phonotactics, including how an empty nucleus position is filled, how rhymes and diphthongs are constructed and how consonants are palatalized. Chapter 4 analyzes how loanwords would be adapted when they are borrowed from Japanese into S-Hakka. The analyses of loanwords are separated into three parts and they are about vowels, onset consonants and coda consonants. Finally, Chapter 5 makes a conclusion for the thesis.. 2.

(13) Chapter 2 Literature Review. 2.1 Optimality Theory This section is going to introduce several theories that are related to Optimality Theory. Classical Optimality Theory (OT) (McCarthy & Prince 1993, Prince & Smolensky 1993/2004) will be presented in §2.1.1, Local Conjunction (Smolensky 1993) in §2.1.2 and Rank-Ordering Model of EVAL (ROE) (Coetzee 2006) in §2.1.3. The reason why Local Conjunction and ROE are taken advantage of is because data cannot be reasonable analyzed by adopting classical OT only. In the present study, Local Conjunction is mainly used to rule out impossible candidates and ROE is to capture the language variation. If only classical without Local Conjunction and ROE is applied to make analyses, a great number of serious problems would occur.. 2.1.1 Classical Optimality Theory Optimality Theory (OT) is first proposed by McCarthy & Prince (1993) and Prince & Smolensky (1993/2004). OT includes several basic concepts. First, there is a set of universal constrains shared by every language, which is called university; nevertheless, some constraints remain inactive in certain languages. Second, various constraint rankings lead to the diversity of languages around the world. Third, violability reveals the truth that constraints can be violated but violations can only be triggered under proper motivations; in other words, violations must be minimal and with reasonable accounts. OT schema is presented in (1). As illustrated, OT is an input-output mechanism. In the schema, candidates are derived from a linguistically well-formed input through Generator (GEN) and then they are evaluated parallelly by Evaluator (EVAL), which includes universal constraints ranked in a language-specific way. The final output is the optimal candidate that. 3.

(14) incurs the fewest or the least significant violations at the EVAL stage. It is notable that the output-based framework applies constraints in the way of parallelism. That means constraints in a ranking are simultaneously applied to a candidate. This is very different from the rulebased theory in which rules are applied step by step.. (1) OT schema Input. GEN. Candidate1. Candidate2. Candidate 3. Candidate 4....... EVAL {Constraint A >> Constraint B >> Constraint C, Constraint D ......}. Output. An optimality-theoretic analysis is demonstrated in (2) where the leftmost constraint (i.e. CA) is highest ranked while the rightmost one (i.e. CD) is lowest ranked. The dotted line indicates that CC and CD are at the same position in the ranking. Thus, both CC and CD are lowest ranked and they are ranked equally. The index symbol “” designates the optimal candidate which is the final output. Additionally, every violation is assigned by an asterisk mark “*” and the fatal violation is given an exclamation mark “!.” Both candidates (c) and (d) are excluded because they violate the constraints which are ranked high. Candidate (a) is more harmonic than candidate (b) because the former incurs fewer violations than the latter does. The shading cells suggest that the violations are no longer relevant for the following discussions.. 4.

(15) (2) An application of classical OT Input CA  a. Cand1 b. Cand2 c. Cand3 d. Cand4 *!. CB. CC *!. CD * *. *!. 2.1.2 Local Conjunction The idea of Local Conjunction is proposed by Smolensky (1993). Under the model of Local Conjunction, two constraints are conjoined together as a “conjoined constraint.” The conjoined constraint is violable if and only if its two component constraints are violated at the same time and within the same domain. An application of Local Conjunction is demonstrated in (3). In (3), the symbol “δ” refers to a certain “domain,” such as a segment, a syllable or a morpheme, etc. The conjoined constraint (i.e. [CA & CB]δ) is violated by candidate (d) due to the fact that the candidate in the meantime violates the two component constraints (i.e. CA and CB). Comparatively, candidate (c) violates CA but not CB, and candidate (b) violates CB but not CA; as a result, the two candidates incur no violation of the conjoined constraint. The optimal output is candidate (a) that violates no constraints.. (3) An application of Local Conjunction Input [CA & CB]δ  a. Candi1 b. Candi2 c. Candi3 d. Candi4 *!. CA. CB *!. *! *. *. Regarding the constraint ranking, two things are worth noting. First, the conjoined constraint and its two component constraints can separately show in the ranking. The need for the conjoined constraint does not mean the two component constraints cannot be independently applied. Second, universally the conjoined constraint outranks its two components. As for the 5.

(16) two component constraints, whether they rank equally or one outranks the other is based on the language in question.. 2.1.3 Rank-Ordering Model of EVAL Rank-Ordering Model of EVAL (ROE) is proposed by Coetzee (2006) and it is mainly adopted to deal with language variations. Coetzee emphasizes that ROE can predict the relative frequency of occurrence for different variants. In the model, there is a “critical cut-off” to separate constraints into two strata: constraints above the cut-off and constraints below the cutoff. Coetzee suggests that constraints above the cut-off are like those in the classical OT so violations of those constraints are fatal enough to rule out candidates. However, violations of constraints below the cut-off are not sever enough to discard candidates. That means if there are two candidates whose violations are below the cut-off, both of them will be the possible outputs, which consequently results in the condition of language variation. It should be noted that since the violation below the cut-off is not fatal at all, the exclamation mark “!” standing for “fatal violation” cannot appear below the cut-off. An application of ROE is presented in (4).. (4) An application of ROE cut-off Input 1 a. Cand1 2 b. Cand2 c. Cand3 d. Cand4. CA. CB. CC. CD *. * *! *!. In (4), the cut-off is marked by a broad stroke. Both candidates (c) and (d) are ruled out since they violate the constraints above the cut-off. There are two possible outputs, candidate (a) and candidate (b). For candidates (a) and (b), they are able to be the real outputs because they violate constraints below the cut-off. Aside from the fact that two candidates are possible in the outputs, another fact is that candidate (a) is more well-formed than candidate (b) because 6.

(17) candidate (a) violates CD which is lowest ranked. Owing to the more well-formedness of candidate (a), it is predicted that candidate (a) would be observed more frequently than candidate (b) would be. In other words, statistically candidate (a) accounts for higher percentage whereas candidate (b) accounts for lower ones.. 2.2 Japanese Sound Inventory and Phonology Japanese sound inventory adopted in the present research is based on Tsujimura (2007) and all the phonetic symbols are transcribed in the form of International Phonetic Alphabet (IPA). Japanese includes ten phonemic vowels which are given in (5). As shown in (5), Japanese contrasts with vowel length. For example, [] means ‘ten’ while [] means ‘door.’. (5) Japanese vowels Front. Central. Back.   .   . 1  . High Mid Low. (6) Japanese consonants Bilabial −vd  Stop +vd  −vd  Fricative +vd   Affricate −vd +vd  +vd  Nasal +vd  Flap +vd  Glide. Alveolar         . Palatal-alveolar         . Palatal         . Velar Uvular Glottal                           . Japanese high back vowel // has several phonetic variants according to different dialects. Standard Japanese, which is Tokyo Japanese, has [] and other dialects have [], [] or [] (Labrune 2012, Vance 2008). Compared with eastern dialects (one of which is Standard Japanese), western dialects are slightly rounded (Shibatani 1990). In connected speech, // is reported to be more advanced and thus becomes [] (Hisagi, Nishi & Strange 2008, Vance 2008). 1. 7.

(18) With regard to consonants provided in (6), Japanese has fourteen consonants as phonemes and nine consonants as allophones which appear in parentheses. Except nasal sounds which will be elaborated in the next paragraph, the seven allophones (i.e. []) can be divided into five different groups because they are generated from five different phonemes. First, both [] and [] are the allophones derived from the phoneme //. The former one occurs when // is followed by [], while the latter one shows up when // is followed by []. Second, both [] and [] are generated from the phoneme //. The allophone [] arises when // is in front of [], while [] occurs when // is in front of []. Third, the segment [] as the voiced counterpart of [] is derived when // precedes []. Forth, when // precedes [], palatalization is applied and // turns out to be []. Finally, when // precedes [] and undergoes palatalization, the allophone [] appears. In a few words, as revealed in (7), allophones [] must co-occur with a following [], and [] have to precede [] Allophones occur with [] are taken as palatalized consonants since [] is a high front vowel with the place of articulation at hard palate.. (7) Derivations of Japanese obstruent allophones Phonetic context Phoneme -/ -/  -/ -/. . .     .   . With respect to nasals, Japanese sets up a well-known nasal assimilation process. In the underlying representation, // and // can serve as onsets but only // can be the coda. Every nasal coda // would be coarticulated with its following consonant. That is to say, the underlying nasal coda would become [] when preceding a labial segment such as [] or [], 8.

(19) become [] when preceding a velar segment such as [] or [], become [] when being at the word-final position (Vance 1986), or remain unchanged as [] elsewhere. The allophones [] are generated through different types of regressive assimilations, as given in (8).. (8) Derivations of Japanese nasal allophones Phonological Labialized process Phoneme / → [-]/____[+labial]. Velarized. Uvularized. →[-]/____[+back]. →[]/____#. 2.3 Sixian-Hakka Sound Inventory S-Hakka sound inventory adopted in the current study is mainly based on Chung (2004), Gu, He & Liu (2002), and the textbook Keyu Nengli Renzheng: Jiben Cihui — Chuji (Sixian Qiang) [Basic Vocabulary for Sixian-Hakka Proficiency Test — Elementary Level] (2011b). The textbook is published by the Taiwanese government2 for the formal S-Hakka proficiency test. The present study transcribes S-Hakka inventory in the form of IPA. S-Hakka contains five phonemic vowels that are given in (9).. (9) S-Hakka vowels Front High Mid Low.   . Central. Back.   .   . Except the five vowels, many researches on S-Hakka articulatory phonetics (Chung 2004, Gu 2005, Gu et al. 2002) claim that the vowel [] also exists in the language, which is also supported by several acoustic studies as well (Liang 2004, Tan 2010, Zheng 2010). The issue. 2. Council for Hakka Affairs, Executive Yuan, Taiwan. 9.

(20) concerned with S-Hakka [] is very controversial among various studies. Only two things are unanimously agreed by researchers: [] can appear only after [, , ] and it is not a phoneme. The questions about [] will be discussed and analyzed in §3.1. In addition to vowels, S-Hakka includes sixteen consonants as phonemes and six consonants as allophones. All the consonants are provided in (10), and allophones are given with parentheses.. −vd. −SG +SG. Fricative. −vd +vd. Affricate. −vd. −SG +SG. Nasal Liquid Glide. +vd +vd +vd.         .         .         .         .         . Glottal. Velar. Palatal.         . Alveolar -palatal.         . Alveolar. Alcelardental3. Stop. Labio -dental. Bilabial. (10) S-Hakka consonants.         . The allophones [] are generated via palatalization when the phonemes // are followed by []. For the allophone [], from which phoneme it is derived from is controversial. Hashimoto (1973) and Chung (2004) argue that [] can be derived when // or // precedes []. In contrast, Yang (1971) and Gu et al. (2002) claim that [] is derived when // precedes []. The allophone [] can only be onset but the thesis find no data in which [] stands for onset consonant. Therefore, the present study will not discuss more about which phoneme [] is. In the previous literature, the three alveolar-dentals are not transcribed with the diacritic mark “”, which stands for “dental”, but the thesis transcribes them with the diacritic in order to make their phonetic quality more clear. For example, in terms of the feature value, alveolar [] is assigned [-distributed] while alveolar-dental [] is given [+distributed]. 3. 10.

(21) generated from. Regarding allophones [j, w], when S-Hakka diphthongs are in their surface representations, one of the vowel is in the form of semi-vowel which is either [j] or [w]. For example, // which means “pitcher” is realized as [].. 2.4 Loanword Phonology Loanword phonology occurs when words from a language enter another one and, through the loan process, non-native sounds undergo certain degrees of transformations. Non-native sounds include both segments and suprasegments but since the present work focuses on segments, suprasegments are not mentioned anymore. The language where loanwords originally come from is called “donor language” or “source language” (henceforth LS). For the language where loanwords enter, it is called “recipient language” or “target language” (henceforth LT). According to a great amount of literature concerned with loanword phonology, three distinct perspectives are going to be introduced. First, adaptations happen primarily because of misperception (Dupoux, Kakehim, Hirose, Pallier & Mehler 1999, Peperkamp & Dupoux 2003). The LT users misperceive segments in the LS and thus adaptations can be observed. Generally, segments that do not exist in the LT would be wrongly perceived. The wrongly perceived segments, in fact, are perceived as the most similar segments that exist in the LT. Second, adaptations occur mainly due to the grammar that is not allowed in the LT. More specifically, the violation of the LT phonology results in the adaptations (Jacobs & Gussenhoven 2000, LaCharite & Paradis 2005). In LaCharite & Paradis’ research, they argues that if there is a given LS phonological category (i.e. feature combination) that also can be found in the LT, the loaned sound sequence will be preserved without any transformations, even though there are phonetic differences. On the other hand, if the given LS phonological category cannot be found in the LT, which means the phonology category is ungrammatical in the LT,. 11.

(22) the loaned sound sequence will be adapted based on the most similar phonological category in the LT. Third, adaptations are conducted based on both perception and grammar evaluation. (Kenstowicz 2005, Silverman 1992, Yip 2006,). This viewpoint is the intermediate between the aforementioned two perspectives. Through the perception, identical segments in the LS and LT remain unchanged; non-identical or non-native segments would be replaced by the most similar ones that exist in the LT. After the segmental modification, loanwords are checked again in order to satisfy the phonotactics established by the LT. Silverman (1992) suggests there are two modification levels for words borrowed from one language to another. The first is “Perceptual Level” dealing with inventories, both sound and tonal inventories. Adaptations having nothing to do with inventories but being related to phonotactics would be handled in the second level, “Operative Level.” The output of the Perceptual Level would be applied to the Operative Level so as to judge whether it violates phonotactic constraints or not. Silverman’s model of loanword phonology (1992:293) is given in (11).. (11) Silverman’s loanword adaptation model (1992:293) incoming acoustic signal. Operative Level representation. Perceptual Level representation. output. Operative Level processes. native phonotactic constraints and preferences. native segment and tonal inventory constraints. 12.

(23) Chapter 3 Sixian-Hakka Phonotactics. This chapter manages four phonotactic phenomena that can be observed in S-Hakka. The first one is how an empty nucleus position is filled, which will be presented in §3.1. The second one is about rhymes, the combination of a nucleus vowel and a coda consonant in a tautosyllable, which will be shown in §3.2. The third one is related to diphthongs, the combination of two vowels at the nucleus position, which will be dealt with in §3.3. Finally, palatalization will be discussed in §3.4.. 3.1 Empty Nucleus In S-Hakka, the underlying forms /, , / are realized as [, , ] in their surface representations (Chung 2004:72, 159-161). Regarding the derivation, this section aims to provide an optimality-theoretic account to explain how an empty nucleus position is filled and which segment is going to fill the vacancy. In §3.1.1 and §3.1.2, the phonetic descriptions for alveolar-dentals [, , ] and the derived vowel [] will be presented respectively, and several related researches will be reviewed as well. In §3.1.3, there will be an optimality-theoretic analysis.. 3.1.1 The Phonetic Description for S-Hakka Alveolar-dentals With respect to the natural class of S-Hakka alveolar-dentals (i.e. [, , ]) , Chung (2004:62) based on the place of articulation states that [t] and [t] are more like alveolars while [] is more like dentals. In contrast, Gu (2005:125) takes them as dentals and Liang (2004:6) considers them as alveolar-fronts (i.e. the place of articulation is slightly more advanced than the one of general alveolars). Chung (2004) assigns [+apical] to [, , ] 13.

(24) because they are pronounced with the tongue tip. Nevertheless, it is important to note that “apical” is not taken as a feature in The Sound Pattern of English (SPE) (Chomsky & Halle 1968/1991). Based on SPE (Chomsky & Halle 1968/1991), the feature [apical] which is adopted by Chung (2004) is better to be replaced by [distributed]. Roca & Johnson (1999) provide a clear definition for [distributed]. “The feature [distributed] refers to the “distribution” of the tongue over the passive articulator” (Roca & Johnson 1999:106). Roca & Johnson elaborately explain that only coronals are associated with [distributed] and its feature value, plus or minus, lies in the tongue body that precisely involves in. In other words, the feature value of [distributed] relies on the size of air obstruction in the vocal tract. An obstruction that is shorter along the airflow (e.g. alveolars [, ]) is [-distributed], whereas an obstruction that is considerable longer along the airflow (e.g. palatal-alvelars [, ]) is [+distributed]. It is worth noting that Roca & Johnson use the term “coronal apical” for sounds that are [-distributed] and “coronal laminal” for sounds that are [+distributed]. The term “apical” adopted by Roca & Johnson (1999) and Chung (2004) is a little be different. Chung uses “apical” as a feature and assumes [+apical] is assigned when sounds are articulated with tongue tip. For example, Chung assigns [+apical] to both alveolars (e.g. [, , , ] in S-Hakka) that are articulated with the tongue against the alveolar ridge, and alveolardentals (e.g. [, , ] in S-Hakka) that are articulated with the tongue against the back of upper teeth and along the alveolar ridge. On the other hand, Roca & Johnson do not use “apical” as a feature but use the term to describe sounds with the air obstruction which is caused by tongue tip and occurs at a particular point in the mouth like alveolar ridge. For instance, Roca & Johnson assign [-distributed] to alveolar [t] while [+distributed] to alveolar-dental []. According to the place of articulation, alveolars and retroflexes are [-distributed] whereas dentals, alveolar-dentals, palatal-alveolars and palatals are [+distributed]. From now on, the 14.

(25) thesis adopts [distributed] proposed by Roca & Johnson, rather than [apical] put forward by Chung, to do the feature specification because [distributed] is more widely adopted when researchers discuss phonetic features. In short, based on the definition of [distributed], S-Hakka alveolar-dentals [, , ] and alveolar-palatals [, , , ] are [+distributed].. 3.1.2 The Phonetic Description for S-Hakka Derived Vowel The vowel [] is named as “apical vowel” by researchers whose studies are for different languages. Chung (2004) who studies several Hakka dialects calls it an “apical vowel” because the vowel only appears after alveolar-dentals which are articulated with tongue tip. Gu (2005:130) who also focuses on Hakka dialects more specifically calls it an “apical front vowel” and describes that it is pronounced with tongue tip approaching the back of upper teeth and with lip spreading4. Tseng & Chou (1999) conduct a research about the phonetic transcriptions for languages used in Taiwan, including Mandarin Chinese, Taiwanese and Hakka, and they take [] as apical vowel as well. Huang (2011) whose study is concerned with epenthetic vowels in Squliq Ayayal, a language in Austronesian family, also names [] as apical vowel; furthermore, in her study, the apical vowel also has to co-occur with onset [], [] or [], which is similar to S-Hakka requiring [] to appear with onset [], [ or []. With regard to S-Hakka vowel quadrangle, the one suggested by Chung (2004) is presented in (12). Chung assumes that the location of [] in the vowel quadrangle has to be more advanced than [], which means [] is [-back]. His argument is based on the fact that [] can only occur behind alveolar-dentals, so the place of articulation for [] approaches to the back of upper teeth and alveolar ridge rather than palatal which is the place of articulation for front vowels (e.g. //).. 4. Gu (2005:130) does not transcribe the S-Hakka apical vowel as [] but as [ï]. 15.

(26) (12) S-Hakka vowel quadrangle (Chung 2004:71) Front Central Back    High    Mid    Low. However, according to the acoustic analyses of S-Hakka inventory (Liang 2004, Tan 2010, Zheng 2010), the location of [] is not like what Chung assumes to be in front of [], but it locates behind the vowel []. The experimental result in Liang’s study is presented in (13). Besides, in Tan’s and Zheng’s studies, the vowel [] is found to gradually merge into the vowel [] by young language users. The mergence is carrying out because [] is very restrictive in distribution, [] is hard to be independently pronounced, and [] is not easy to be correctly perceived when being compared with [] and [].. (13) S-Hakka vowel quadrangle (F2: F1) (Liang 2004:97). For the present work, vowel [] is considered as a central vowel and thus is [+back], which follows the acoustic analyses but differs from Chung’s assumption. In conformity with the acoustic experimental result makes analyses in the current study more logical and reasonable. The vowel quadrangle adopted by the thesis is shown in (14).. (14) S-Hakka vowel quadrangle (The thesis) Front Central Back    High    Mid    Low 16.

(27) 3.1.3 An Optimality-theoretic Analysis Although phonologically it is said that nucleus is the only obligatory component in a syllable, S-Hakka consists of a syllable type in which only onset is filled in the underlying representation. In Chung’s study (2004:72, 159-161), S-Hakka allows three consonants that have no need to combine with vowels and can become as independent syllables: /, , /. Chung also provides their surface representations: [, , ]. By comparing the underlying with their surface forms, it can be noticed that [] occurs. The occurrence of [] indicates that S-Hakka does not allow the “empty nucleus” which refers to the condition that the nucleus position is not filled by any segment; therefore, the language figures out its own way to solve the problem. The current study uses FILLNuc in (15) to capture the fact that the language requires a segment at the nucleus position.. (15) FILLNuc: Nucleus positions are filled with segmental materials.. Chung assumes that the empty nucleus position is occupied by [] for the reason that [+distributed]5 spreads from the preceding alveolar-dental consonant (i.e. [, , ]) to the following nucleus position. If based on Chung’s assumption a constraint is proposed in order to describe the derivation, AGREE[distributed] is given in (16) which asks a derived vowel to agree with its preceding consonant on [distributed].. (16) AGREE[distributed]: A derived vowel must agree with its preceding consonant on the feature value of [distributed]. However, AGREE[distributed] is not enough, since [, , ], [, , ] and [, , ]. 5. In Chung’s (2004) study, he adopts [+apical] but it is replaced by [+distributed] in the thesis. Please refer to §3.1.1 for more details. 17.

(28) all satisfy the constraint due to the fact that vowels [, , ] are all [+distributed], as can be seen in (17). Then, why it is [] rather than [] or [] that is derived? The present work takes vowel markedness into consideration.. (17) Feature matrices for S-Hakka vowels         Dorsal High + + + Low Back + +    Coronal Anterior 6 Distributed + + +  Labial Round +.   +.   + +.  +. With respect to markedness, Lombardi (2002) investigates epenthetic vowels from the perspective of markedness and typology and the research results are significantly important. The markedness relationships among vowels found by Lombardi are presented in (18).. (18) Markedness relationships among vowels (Lombardi 2002:5): i. Languages may vary in whether low or nonlow vowels are less marked. If low vowels are less marked, the epenthetic vowel will be //. *low, *nonlow: re-rankable ii. If nonlow vowels are less marked, other constraints choose among them: Back vowels are less marked than front vowels: *Front >> *Back Mid vowels are marked: *Mid Round vowels are marked: *[+round] >> *[-round]. Several things about Lombardi’s research are worth noting. First, Lombardi assumes that it is language-specific whether low or nonlow vowels are marked; as a result, in the markedness hierarchy, the top-ranking constraints usually are used to determine in a language whether it is low or nonlow vowel that is marked. Second, being concerned with tongue height, vowels are. The segment [] is [+distributed], which is based on Chung (2004). The segments [, ] are [+distributed], which is based on Clements & Hume (1995). 6. 18.

(29) specified [nonlow] (e.g. [, , , , ] in Lombardi’s study), [mid] (e.g. [, ] in Lombardi’s study) or [low] (e.g. [] in Lombardi’s study). Third, Lombardi finds that mid vowels are universally marked but *[mid] is freely ranked among other markedness constraints (Lombardi 2002:11). Fourth, front vowels (e.g. [, e] in Lombardi’s study) are universally more marked than back vowels (e.g. [, , , , ] in Lombardi’s study), and round vowels are more marked than unround vowels. In the thesis, constraint rankings put forward by Lombardi are adjusted to fit the approach of binary values of phonetic features in order to have a consistent analytic approach with the succeeding sections. Thus, [nonlow] in Lombardi’s study is replaced by [low], [low] by [+low], and [mid] by [-high, -low]. Furthermore, [front] in Lombardi’s study is substituted by [-back], and [back] by [+back]. Being concerned with S-Hakka markedness relationship of vowels, constraints with their rankings are provided in (19). It is inferred from the derived [] that S-Hakka makes vowels of [+low] more marked, rather than vowels of [low].. (19) Markedness relationships among S-Hakka vowels (adjusted from Lombardi 2002:5): i. Languages may vary in whether [+low] or [-low] vowels are less marked: S-Hakka: *[+low] >> *[-low] (language-specific) ii. If *[-low] vowels are less marked, other constraints choose among them: *[-back] >> *[+back] (language universal) *[+round] >> *[-round] (language universal) *[-high, -low] (language universal) (This constraint is freely ranked among other markedness constraints (Lombardi 2002:11)). The analysis through constraint competing is presented in (21) where how the empty vowel position is filled and which segment is going to fill in the vacancy are explained. Another constraint, IDENT-IO-V in (20), is given so as to make sure that the constraint hierarchy is applied for derived vowels.. 19.

(30) (20) IDENT-IO-V: Assign one violation mark for every vowel change from input to output correspondent.. * * * *. *! *! *! *!. * *! * * * *. * *. * [+back]. * [-back]. *[-low]. *[+low]. *[-high, -low]. AGREE [distributed]. IDENT-IO-V. i. Input: /, , /  a.   b.   c.   d.   e.   f.   g.  ii. Input: // a.  b.  c.  d.  e.   f. . FILLNuc. (21) Input: /, , / → Output: [, , ]. *. *! *! *! *! *! *!. * * * *. * * * *. * *. *. * * * * * *. Candidate (21i.g) is ruled out because there is no segment at the nucleus position, which violates FILLNuc. Candidates (21i.d) to (21i.f) are penalized by AGREE[distributed] since [, , ] are not associated with [distributed]. By comparing candidates (21i.a) to (21i.c), it is found that the markedness constraints for vowels proposed by Lombardi (2002) can correctly predict the output which is (21i.a). Candidate (21ii.f) is chosen as the optimal output among other candidates, which results from the top-ranking of IDENT-IO-V. Overall, the underlying forms /, , / are realized as [, , ]. The occurrence of a derived vowel is because of FILLNuc which is ranked highest. Furthermore, by virtue of AGREE[distributed] and other markedness constraints that are related to vowels, it is [] rather than other vowels that show up.. 20.

(31) 3.2 Rhymes A rhyme refers to the combination of segments at the nucleus and coda positions. S-Hakka only allows voiceless stops and nasals to serve as codas, so consonants other than [, , , , , ] cannot be at the coda position. In (22), possible S-Hakka rhymes are presented, and those that is given a check “” indicates it is grammatical whereas a cross “” indicates unacceptable.. (22) Rhymes in S-Hakka                           . As shown in (22), alveolar codas [] are able to combine with every vowel to form rhymes. However, labial codas [] can only join to vowels that are not [+round], which is captured by OCPR[labial] in (23). Moreover, velar codas [] that are [+high] can only join to vowels whose place of articulation is not linked to coronal, which is regulated by *Cor-HiR in (24). These two constraints are directly related to feature values of involved segments, so feature specifications of related sounds are provided in (25).. (23) OCPR[labial]: Avoid a rhyme in which two segments are labial sounds.. (24) *Cor-HiR: Avoid a rhyme in which a coronal sound is followed by [+high].. 21.

(32) (25) Feature matrices for segments involved in S-Hakka rhymes              Dorsal High + + + Low + Back + + + +    Coronal Anterior Distributed + + +    Labial Round + + -. .   + +.  + -. Applications of OT are demonstrated as follows. In order to make tableaux easier to be comprehended, they are separated based on different vowels. Additionally, in each tableau, there may be more than one optimal outputs. It is because the constraints function as phonotactic constraints, and candidates that do not violate phonotactics do appear in the language in question. The outcomes from the following tableaux, (26) to (31), correspond with table (22).. (26) Input: /iC/ OCPR[labial]  . a. b.  c. . *Cor-HiR. *!. In (26), candidate (c) is penalized by *Cor-HiR for the reason that [] is a coronal sound and its following segment [] is [+high]. As for candidates (26a) and (26b), they incur no violation marks and hence are grammatical in S-Hakka. The constraints correctly predict that labial and alveolar codas, but not velar ones, can follow the vowel [].. 22.

(33) (27) Input: /eC/ OCPR[labial]  . a. b.  c. . *Cor-HiR. *!. In (27), candidate (c) is excluded. Candidate (27c) does not reach the requirement of *Cor-HiR because [] is a coronal segment and precedes [] which is [+high]. Candidates (27a) and (27b) satisfy the two constraints so without doubt they can be observed in the real language use.. (28) Input:C/ OCPR[labial]. *Cor-HiR.  a.  b.   c. . In (28), none of the candidates violate any of the constraints and thus the three candidates are all grammatical and acceptable in the language.. (29) Input: /uC/ a.  b.   c. . OCPR[labial] *!. *Cor-HiR. In (29), candidate (a) is ruled out due to the fact that both [] and [] are labial sounds but OCPR[labial] aims to avoid two adjacent labial segments within a rhyme. For candidates (29b) and (29c), they are the real outputs since no violation marks are assigned to them.. 23.

(34) (30) Input: /C/ a.  b.   c. . OCPR[labial] *!. *Cor-HiR. In (30), OCPR[labial] plays a key role to cast away candidate (a) in which [] and its following sound [] are articulated with labials. No violations are brought out by candidates (30b) and (30c); as a result, both of them are real outputs.. * * *. * * *. *Cor-HiR. OCPR[labial]. * [+back]. * [-back]. *[-low]. *[+low]. *[-high, -low]. AGREE [distributed]. i. Input: C  a.   b.  c.  d.  e.  f.  g.  h.  i.  j.  k.  l.  m.  n.  o.  p.  q.  r.  s. . FILLNuc. (31) Input: /CCC/. *!. *! *! *. *! *! *! *!. * * * *! *. * *!. *! *! *!. * * *. * * *! *. *. * * *. *. 24. * *. * *. *. *! *! *! *!. * *. * * * * *. * * * * *.

(35) *Cor-HiR. OCPR[labial]. * [+back]. * [-back]. *[-low]. *[+low]. *[-high, -low]. AGREE [distributed]. ii. Input:C  a.   b.  c.  d.  iii. Input:C  a.   b.  c.  d. . FILLNuc. (31) Input: /CCC/ (Cont.). * * *. * * *. *!. * * *. * * *. *!. *!. *!. In (31), how the vowel [] is derived through AGREE[distributed] and the markedness constraints please refer to §3.1 for details. Candidate (31d) that has alveolar-dental as its onset contains no vowel to fill the empty nucleus position and hence it violates FILLNuc. Concerned with candidate (31c), it incurs the violation mark of *Cor-HiR for the reason that [] belongs to coronal sounds and is followed by [] which is [+high]. The constraints then correctly predict candidates (31a) and (31b) are real outputs since they satisfy both OCPR[labial] and *Cor-HiR. To sum up, a permissible rhyme in S-Hakka has to obey both OCPR[labial] and *Cor-HiR. A candidate could be chosen as the optimal output only when it incurs no violations of the two constraints. In this way, there may be more than one real outputs in a tableau and also there may be no ideal outputs in a tableau.. 3.3 Diphthongs A diphthong refers to the combination of two vowels at the nucleus position. Permissible diphthongs in S-Hakka are given below. In (32), the combination that is assigned with a check “” is grammatical whereas with a cross “” is ungrammatical. 25.

(36) (32) Diphthongs in S-Hakka       7                .      .      . By observing (32), three constrains function to judge the grammaticality and ungrammaticality of vowel combinations. First, two identical vowels are not allowed to be adjacent within a nucleus position so the ungrammaticality of [, , , , ] can be captured by *LONG/VOWEL in (33). Second, segments that are [-high] cannot appear together; thus, OCPV[-high] in (34) is proposed to prohibit [, , , , , ]. Third, [OCP[back] & OCP[round]]V in (35), which requires two adjacent vowels not to have the same feature values on [back] and [round], is given to forbid [, , ,]. Notice that the constraints from (33) to (35) cannot be replaced by constraints that refer to the sonority hierarchy because SHakka includes both falling diphthongs (e.g. [, ]) and rising diphthongs (e.g. [, ]).. (33) *LONG/VOWEL: Assign one violation mark for every long or geminated vowel. (34) OCPV[-high]: Avoid a diphthong in which two vowels are both specified as [-high]. (35) [OCP[back] & OCP[round]]V: A violation mark is given if and only if, within a diphthong, OCP[back] that avoids two segments to have identical feature values on [back] and OCP[round] that avoids two segments to have identical feature values on [round] are simultaneously violated.. Following are several tableaux to show how acceptable diphthongs in S-Hakka are chosen through the constraint competition. In order to make analyses easier to be comprehended,. Whether the diphthong [] exists in S-Hakka is a controversial issue. No data in the present study show the combination so it is taken as ungrammatical. For the various researches discussing [] in S-Hakka, please refer to Chung (2004:161-164). 7. 26.

(37) tableaux are separated according to the first vowel. Outcomes in tableaux (36) to (40) are able to correspond with table (32).. (36) Input: /V/.   . a.  b.  c.  d. e. . *LONG/VOWEL *!. OCPV[-high]. [OCP[back] & OCP[round]]V * *!. In (36), three candidates, (c) to (e), reach the requirements of the constraints so they are acceptable diphthongs in the language. For candidate (36a), it is ruled out owing to the violation of *LONG/VOWEL. Besides, if a candidate violates *LONG/VOWEL, it will violate [OCP[back] & OCP[round]]V as well since two identical segments must have the same feature values on [back] and [round]. Candidates violate *LONG/VOWEL and [OCP[back] & OCP[round]]V at the same time can be also observed in (37d), (38b), (39e) and (40c) below. As for candidate (36b), it is cast away by [OCP[back] & OCP[round]]V because [] and [] are both specified as [-back] and their articulations involve no labials in.. (37) Input: /V/ *LONG/VOWEL   . a.  b.  c.  d. e. . OCPV[-high]. *!. [OCP[back] & OCP[round]]V. * *!. In (37), two candidates are discarded. Candidate (37d) incurs a violation mark of *LONG/VOWEL and the two identical vowels also result in the violation of [OCP[back] & OCP[round]]V. With respect to candidate (37e), because [] and [] have the same feature. 27.

(38) values on [back] and [round], the candidate does not satisfy the requirement asked by [OCP[back] & OCP[round]]V. For candidates (37a) to (37c), they can be found in the real language use because they give rise to no violations.. (38) Input: /V/. . a.  b.  c.  d. e.. *LONG/VOWEL. OCPV[-high]. *!. * *!. [OCP[back] & OCP[round]]V *! *. *!. In (38), only candidate (d) is chosen as the output because no violation is incurred by it. Concerned with candidate (38a), the diphthong is [] in which the two vowels are given [back] and do not make use of labials when being articulated, so it is discarded by [OCP[back] & OCP[round]]V. Regarding candidate (38b), as aforementioned, it violates *LONG/VOWEL and thus violates [OCP[back] & OCP[round]]V as well; furthermore, the candidate violates OCPV[-high] since [] is [-high]. OCPV[-high] is also not obeyed by candidates (38c) and (38e) for the reason that [], [] and [] are all [-high].. (39) Input: /V/ *LONG/VOWEL . a.  b.  c.  d. e. . OCPV[-high]. [OCP[back] & OCP[round]]V. *! *! *!. *! *. *. In (39), [], [] and [] are [-high] so candidates (b), (c) and (e) violate OCPV[-high]. Besides, candidate (39e) in which there are two identical vowels brings out the violations of *LONG/VOWEL and [OCP[back] & OCP[round]]V. Regarding candidate (39d), both [] and [] 28.

(39) are specified as not only [+back] but also [+round] and thus the candidate violates [OCP[back] & OCP[round]]V. For candidate (39a), it is the optimal output since it satisfies all constraints.. (40) Input: /V/ . . *LONG/VOWEL. OCPV[-high]. [OCP[back] & OCP[round]]V. *!. *! *. *. a.  b.  c.  d. e. . *!. In (40), candidates (a) and (d) are the most harmonic outputs by virtue of no constraints being violated. With regard to candidate (40b), because [] and [] are specified as [-high], the requirement of OCPV[-high] is not fulfilled. For candidate (40c), it is a diphthong in which two identical vowels are included, so doubtless it violets *LONG/VOWEL and [OCP[back] & OCP[round]]V; additionally, for the reason that [] is [-high], the candidate also disrupts OCPV[-high]. Finally, candidate (40e) disobeys OCPV[-high] because [] and [] are given minus value on the feature [high]. Conclusively, S-Hakka phonotactics about diphthongs are set up by *LONG/VOWEL, OCPV[-high] and [OCP[back] & OCP[round]]V. All these constraints are markedness ones. Any candidate that violates one of the three constraints causes the ungrammaticality and hence has no way to be realized in the language. Candidates that do not violate the three constraints can be the real use in the language. In this way, a tableau may have more than one real outputs but may have no optimal outputs as well.. 3.4 Palatalization Palatalization can be found in S-Hakka. The allophones [, , ] occur via palatalization when their phonemes /, , / are followed by // (Chung 2004). In other words, the phonetic 29.

(40) sequences [, , ] never appear since they must undergo palatalization and become their variants. In the literature, [] is also considered to be derived through palatalization. Nevertheless, from which phoneme it is derived remains controversial. Some studies assume that both // and // are surfaced as [] (Chung 2004, Hashimoto 1973), but others claim that it is // which is realized as [] (Gu et al. 2002, Yang 1971). The present study does find [] in the surface representations, but finds no data in which [] serves as onset. Therefore, how [] is derived will not be discussed and for more details please see Chung (2004:67-68, 168174). Being inferred from the previous paragraph, it is only sure that [, , ] occur when /, , / appear before [] so the derivation of palatalization is from /, , / to [, , ]. Three constraints are given to regulate the phenomenon. First, as can be seen in the derivation, the input [] does not change; therefore, IDENT-IO-V which has been presented in (20) is necessary. The second one is *[+anterior, +stident]/which is defined in (41). The contextsensitive constraint *[+anterior, +strident]/ is proposed in order to prohibit [, , ] from appearing in the surface forms. The third one is IDENT-IO[anterior] in (42) which asks the feature value of [anterior] not to change from the input to its output correspondent. Because *[+anterior, +strident]/ and IDENT-IO[anterior] have something to do with features, feature matrices of related segments are listed in (43) and an application of OT can be seen in (44).. (41) *[+anterior, +strident]/: Any segment of [+anterior, +strident] preceding [] is prohibited.. (42) IDENT-IO[anterior]: Assign one violation mark for every feature change of [anterior] from input to output correspondent.. 30.

(41) (43) Feature matrices for segments involved in S-Hakka palatalization  Alveolar-dentals Alveolar Alveolar-palatals        Strident + + + + + +        Coronal Anterior + + + + Distributed + + + + + + Dorsal High Low Back (44) Input: /, , / → Output: [, , ] IDENT-IO-V *[+anterior, +strident]/ i. Input:/  a.  b.  c.  d.  ii. Input: // a.  b. t c. t d. t iii. Input: // a.  b.  c.  d. .    +  + -. Vowels    +  +.    +  + +. IDENT-IO[anterior]. *! * *! *!. * *! *. *! *!. * *! *. *! *!. *. In (44), candidate (a) contains [, , ] which are [+anterior, +strident] and precede [] so it is cast away by *[+anterior, +strident]/. For candidates (44c) and (44d), they change input // to its output correspondent [] and thus they are excluded by IDENT-IO-V. As for candidate (44b), it is the real output and it wins out because IDENT-IO[anterior] is ranked below the cutoff, which will be proven in §4.2.4. In short, with regard to the process of palatalization, since IDENT-IO-V is highest ranked, vowels cannot be altered but onsets can. IDENT-IO[anterior] is at the expense for the onset change. Besides, the ungrammaticality of [, , ] is captured by *[+anterior, +strident]/. 31.

(42) 3.5 Summary for Sixian-Hakka Phonotactics In §3, several markedness constraints and two faithfulness constraints are put forward in order to elaborate the phonotactic phenomena in S-Hakka. The correspondence between phonotactic conditions and constraints are revealed in (45).. (45) The correspondence between phonotactic conditions and constraints Phonotactic Conditions Constraints §3.1 Filling an empty nucleus position (15) FILLNuc (16) AGREE[distributed] (19) *[-high, -low] *[+low] >> *[-low] *[-back] >> *[+back] (20) IDENT-IO-V §3.2 Rhymes (23) OCPR[labial] (24) *Cor-HiR §3.3 Diphthongs (33) *LONG/VOWEL (34) OCPV[-high] (35) [OCP[back] & OCP[round]]V §3.4 Palatalization (20) IDENT-IO-V (41) *[+anterior, +strident]/ (42) IDENT-IO[anterior] In §3.1, FILLNuc functions to explain the occurrence of the derived vowel in the derivation from /, , / to [, , ]. Besides, AGREE[distributed] and other markedness constrains concerned with vowels predict the nucleus position is filled by [] rather than other vowels. In §3.2, any combination of a nucleus vowel and a coda consonant has to obey OCPR[labial] and *Cor-HiR. In §3.3, two vowels are allowed to be adjacent at the nucleus position only if they do not violate *LONG/VOWEL, OCPV[-high] and [OCP[back] & OCP[round]]V. Finally, in §3.4, the derivation from /, , / to [, , ] is formulated by IDENT-IO-V and *[+anterior, +strident]/; furthermore, IDENT-IO[anterior] is at the expense for the process of palatalization. All the constraints mentioned in the last paragraph are related to S-Hakka syllable structures and segmental combinations. The constraints listed in (45), except IDENT-IO-V and IDENT-IO[anterior], are grouped together as one constraint package which is called “OK-σ.” 32.

(43) OK-σ is dominant when it is applied to S-Hakka loanword phonology which is going to be presented in §4. The two faithfulness constraints (i.e. IDENT-IO-V and IDENT-IO[anterior]) are violable as loanwords are adapted and hence they are not included in OK-σ. It is worth noting that IDENT-IO-V is replaced by other faithfulness constraints that are related to vowel features (e.g. IDENT-IO[long] or IDENT-IO[back]) when loanword phonology is discussed. The replacement can better describe how a non-native vowel is adapted to a native one.. 33.

(44) Chapter 4 Sixian-Hakka Loanword Phonology. In this chapter, analyses will focus on the segmental adaptations from Japanese as the LS to S-Hakka as the LT. The Japanese loanwords are collected from a loanword dictionary Keyu Wailaiyu: Han Yuan Min Ke Guo Yu Hujieci (Sixian Qiang) [Loanwords in Sixian-Hakka: Including words shared among Taiwanese, Hakka and Mandarin] (2011a), which is published by Taiwanese government8. Issues regarding vowels are presented in §4.1. Those with regard to onset consonants are brought out in §4.2 and coda consonants in §4.3. Only a few cases involving deletions of segments. Some of them will be accounted for but there are still others being taken as exceptions that will be mentioned in footnotes.. 4.1 Adaptations of Vowels Japanese vowels are discriminated by vowel length and totally the language includes ten vowels (i.e. [, , , , , , , , , ]). On the other hand, S-Hakka consists of six vowels (i.e. [, , , , , ]) but vowel length is not a distinctive feature in the language. With the brief comparison, it can be inferred that when the loan process occurs, the vowels [, , , ] undergo no transformations because they exist in both languages. Therefore, there will be no discussions about [, , , ] in the thesis. However, as can be observed from the comparison, the feature [long] is contrastive only in Japanese but not in S-Hakka and thus the difference will be discussed in §4.1.1. Furthermore, the back unrounded vowel [] in Japanese is illicit in SHakka and will be analyzed in §4.1.2.. 8. Council for Hakka Affairs, Executive Yuan, Taiwan. 34.

(45) 4.1.1 Short Vowels Adapted from Long Counterparts In comparison of Japanese and S-Hakka vowel systems, an obvious difference is that vowel length is a distinctive feature in Japanese but the feature has no such function in S-Hakka. Thus, long vowels must undergo certain degrees of transformations during the loan process. The prohibition of long vowels in the LT is first mentioned in §3.3 where *LONG/VOWEL is brought out in (33), and the constraint is later included in OK-σ in §3.5. OK-σ is a constraint package in which all constraints are set up in order to make syllable structures of outputs acceptable in the LT. Therefore, OK-σ must not be dominated in any constraint rankings. OKσ is restated here in (46).. (46) OK-σ i. FILLNuc (original in (15)) ii. AGREE[distributed] (original in (16)) iii. *[-high, -low] (original in (19)) *[+low] >> *[-low] (original in (19)) *[-back] >> *[+back] (original in (19)) iv. OCPR[labial] (original in (23)) v. *Cor-HiR (original in (24)) vi. *LONG/VOWEL (original in (33)) vii. OCPV[-high] (original in (34)) viii. [OCP[back] & OCP[round]]V (original in (35)) ix. *[+anterior, +strident]/i (original in (41)). Statistics concerned with the long vowel adaptation are given in (47) where two facts emerge. First, Japanese long vowels are realized as their short counterparts when they are borrowed into S-Hakka and some examples are provided in (48). *LONG/VOWEL is able to capture the first fact that long vowels have to transform. Then, a prediction can be made that IDENT-IO[F] (i.e. [long] in the case) in (49) is violable so as to obey *LONG/VOWEL.. (47) Statistics for the adaptation of long vowels Japanese S-Hakka Number Total source correspondent V 124 124 V. 35. Percentage 100%.

(46) (48) Examples for the adaptation of long vowels Japanese loanwords S-Hakka correspondents   a   b   c   d   e. Gloss motorcycle motor lunch box teacher; doctor water pipe. Second, throughout the loan process, it is clear that segmental adaptations are preferred compared with deletions, since long vowels which do not exist in the LT could be deleted rather than being replaced by short vowels. Regarding the fact, MAX-IO-Seg is proposed in (50) asking for the preservation of every segment during the loan process. The application of these three constrains is demonstrated in (51).. (49) IDENT-IO[F]: Assign one violation mark for every feature change from input to output correspondent.. (50) MAX-IO-Seg: Assign one violation mark for every segment in the input that has no correspondent in the output.. (51) Input: [] → Output: [] ‘motorcycle’ OK-σ MAX-IO-Seg a.   *! b.  *! c. . IDENT-IO[long] *. ((46vi.) *LONG/VOWEL). In (51), candidate (c) is cast away because [] violates *LONG/VOWEL. Candidate (51b) is ruled out since [] in the input has no correspondent in the output, which does not obey the requirement of MAX-IO-Seg. OK-σ and MAX-IO-Seg outrank IDENT-IO[long]; thus, the optimal output is candidate (51a) that incurs the least fatal violation of IDENT-IO[long]. In sum, geminated vowels in the LS undergo shortening once they enter the LT. During the shortening process, IDENT-IO[long] must be at the expense so it is lowest ranked. Additionally,. 36.

(47) according to the statistics in (47) and others that will be provided in the following sections, MAX-IO-Seg plays an important role in the constraint ranking for loanword phonology due to the fact that preservation is favored rather than deletion.. 4.1.2 Adaptations of [] The current study compares Japanese with S-Hakka vowel inventories again. Aside from the long vowel adaptation, Japanese back unrounded vowel [] in loanwords must be altered as well for the reason that it is not a member in S-Hakka phonetic inventory. The ungrammaticality of [] in the output is given in (52). Two adapters originated from [] are discovered in different phonetic environments. It is [] when the preceding consonants are alveolar-dentals, such as [s, ]. On the other hand, it is [u] elsewhere. The adapter [] will be accounted for in §4.1.2.1 and [u] will be handled in §4.1.2.2.. (52) : Assign one violation mark for every back unrounded high vowel [].. 4.1.2.1 Central Unrounded [] Adapted from Back Unrounded [] As aforementioned, when the loaning happens, Japanese [] becomes S-Hakka [] on condition that the preceding consonant is an alveolar-dental sound (i.e. [s, ] in S-Hakka). For a closer look, statistics are provided in (53) and a few examples are listed in (54).. 37.