從輕聲與韻律外現象看漢語方言中虛化的功能詞 - 政大學術集成

全文

(1)國立政治大學語言學研究所博士學位論文 National Chengchi University Graduate Institute of Linguistics Doctoral Dissertation. 政治大. 立從輕聲與韻律外現象. ‧ 國. 學. 看漢語方言中虛化的功能詞. ‧ sit. y. Nat. Grammaticalized Function Words in Chinese Dialects:. n. al. er. io. Neutral Tone and Prosodic Invisibility. Ch. engchi. i Un. v. 指導教授：蕭宇超博士 Advisor: Yuchau E. Hsiao, Ph.D.. 研究生：黃子權撰 Student: Tzu-Chuan Huang. 中華民國：一〇九年六月 June, 2020. DOI:10.6814/NCCU202001638.

(2) 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i Un. v. ii. DOI:10.6814/NCCU202001638.

(3) Abstract This dissertation concerns the prosodization of function words in Chinese dialects in terms of neutral tone and prosodic invisibility. Neutral tone refers to the pitch pattern derived by complete loss of original lexical contrast; prosodic invisibility describes a phenomenon where a tone is extraprosodic and thus is excluded from some general tonal process. Both phenomena are considered derived from the status of prosodic clitic, meaning independent syntactic words that are not prosodized as independent prosodic. 政治大 between lexical words and function words. Words of lexical categories strongly 立. words, and the occurrence of such prosodic clitics is subject to a universal distinction. ‧ 國. 學. correlate with prosodic word status, whereas words of functional categories correspond to more than one prosodic type, ranging from independent prosodic word to free. ‧. prosodic clitic, and therefore members of functional categories may diverge from one. Nat. sit. y. another in whether to perform neutral tone and/or prosodic invisibility. Two prosodic. n. al. er. io. classes are found in the current survey, termed prominent function words on the one. i Un. v. hand, and non-prominent function words on the other. Members of the former class. Ch. engchi. exhibit a similar prosodization to lexical word while the class of non-prominent function words is vulnerable to the neutral tone process and prosodic invisibility. The central problem to be addressed in this dissertation is the formal characterization of prosodic distinctions between these subclasses within functional categories. The conventional response to this problem is to take an interface constraint approach grounded by morphosyntax, which assumes that different prosodizations of functional elements directly follow from differences in their morphosyntactic properties and language-specific constraint rankings. However, this straightforward correlation does not uniformly hold in the case that serves as a basis of the dissertation, the neutral tone. iii. DOI:10.6814/NCCU202001638.

(4) and prosodic invisibility of Chinese dialects, where function words exhibit a considerable overlap in their morphosyntactic distributions that precludes any differentiation rooted in morphosyntax. This dissertation, though employing a set of interface constraints too, proposes that the ranked interface constraints are grounded by not only morphosyntactic properties but semantic/pragmatic information. By taking an overall look at various types of function words which have potential to perform neutral tones and/or prosodic invisibility, I conclude that the distinction between prominent and non-prominent classes of function words can be drawn along the scale of grammaticalization. The idea is based on the general conception in the literature of. 政治大. grammaticalization that phonological erosion and semantic bleaching usually come. 立. hand in hand. As a function word becomes more semantically bleached during the. ‧ 國. 學. course of grammaticalization, it is more likely to be non-prominent, or weak, to license. ‧. certain constituents in the prosodic structure. By encoding this information of grammaticalization in the interface constraint, this dissertation offers a formalized. sit. y. Nat. io. er. mechanism to make use of both morphosyntactic properties and information from semantics-syntax interface. This approach thus leads to a unified account of prosodic. al. n. iv n C distinction between subclasses within hfunctional e n g c categories. hi U. Keywords: grammaticalization, prosodization, syntax-phonology interface. iv. DOI:10.6814/NCCU202001638.

(5) Table of Contents. Introduction .......................................................................................................... 1 1.1. The scope and aim ......................................................................................... 1. 1.2. Organization of the dissertation ..................................................................... 4. 1.3. Theoretical background ................................................................................. 5. 1.3.1. Prosodic phonology ............................................................................... 5. 1.3.2. Prosodic clitic theory ............................................................................. 7. 1.3.3. Constraints on prosodic structures ....................................................... 11. 1.3.4. A typology of function word parses..................................................... 15. 1.4. 1.4. Definition of neutral tone ................................................................... 17 Language background .................................................................................. 22. ‧. Prosodic Prominence and Degree of Grammaticalization ........................... 24. y. Nat. 2.2. Conflation of grammaticalized classes ...................................................... 28. 2.3. Alignment at different levels ....................................................................... 33. 2.4. Summary ...................................................................................................... 36. sit. Prosodic licensing by edge-alignment ....................................................... 24. io. 2.1. n. al. er. 2.. Neutral tone .................................................................................................. 17. 學. 1.4.1. 立. 政治大. ‧ 國. 1.. 3.. Ch. engchi. i Un. v. Grammaticalized function words in Cantonese and Sixian Hakka ............ 37 3.0. Introduction .................................................................................................. 37. 3.1. Minimal tonal neutralization ........................................................................ 38. 3.1.1. Non-neutralization in majority of functional classes ........................... 38. 3.1.2. Neutralization of sentence-final particles in Cantonese....................... 49. 3.1.3. Asymmetry between subclasses of sentence-final particles ................ 57. 3.1.4. The stresslessness................................................................................. 66. 3.2. Invisible function words in Miaoli Sixian Hakka ........................................ 68. v. DOI:10.6814/NCCU202001638.

(6) 3.2.1. T1 sandhi in Miaoli Sixian Hakka ..................................................... 69. 3.2.2. Underapplication of T1 sandhi ............................................................... 75. 3.3. Grammaticalized function words in Taiwanese and Standard Mandarin . 83 4.0. Introduction .................................................................................................. 83. 4.1. Tonal neutralization in Taiwan Southern Min ............................................. 84. 4.1.1. Predominant occurrence of tonal neutralization .................................. 86. 4.1.2. Problematic grammatical words .......................................................... 94. 4.1.3. The extrametricality of neutral tones ................................................. 101 4.1.3.1 4.1.3.2. ‧ 國. Distribution of the two types of neutral tones .................................... 112. 4.2.2. Neutralization at lexical vs. postlexical levels ................................. 119. ‧. 4.2.1. Summary .................................................................................................... 124. sit. y. Nat. 4.3 5.. Tonal neutralization in Standard Mandarin ............................................... 112. 學. 4.2. 政治大 Domain-finality 立 .......................................................................... 106 Invisibility to tone sandhi........................................................... 102. Grammaticalized function words in Shanghainese .................................... 126. io. er. 4.. Summary ...................................................................................................... 82. 5.0. Introduction ................................................................................................ 126. 5.1. Tonal neutralization in Shanghainese ........................................................ 127. n. al. Ch. engchi. i Un. v. 5.1.1. Maximal neutralization of grammatical words .................................. 128. 5.1.2. Neutralization in lexical vs. functional domains................................ 137. 5.2. Prosodic invisibility in Shanghainese ...................................................... 146. 5.2.1. The nature of melody redistribution .................................................. 146. 5.2.2. Blocking of redistribution .................................................................. 148. 5.2.3. Prosodic licensing of non-root C particle .......................................... 154. 5.3. Summary .................................................................................................. 159. vi. DOI:10.6814/NCCU202001638.

(7) 6.. 7.. Theoretical remarks on typology .................................................................. 161 6.1. Neutral tone ................................................................................................ 161. 6.2. Prosodic invisibility ................................................................................... 165. 6.3. Prosodic Well-formedness ......................................................................... 165. 6.4. Summary .................................................................................................... 168. Conclusions ....................................................................................................... 170. References ................................................................................................................. 174. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i Un. v. vii. DOI:10.6814/NCCU202001638.

(8) 1. Introduction. 1.1 The scope and aim This dissertation concerns the prosodization of function words in Chinese dialects in terms of neutral tone and prosodic invisibility. Neutral tone refers to the pitch pattern derived by complete loss of original lexical contrast; prosodic invisibility describes a phenomenon where a tone is extraprosodic and thus is excluded from some general. 政治大 meaning independent syntactic words that are not prosodized as independent prosodic 立. tonal process. Both phenomena are considered derived from the status of prosodic clitic,. ‧ 國. 學. words, and the occurrence of such prosodic clitics is subject to a universal distinction between lexical words (i.e. words belonging to ‘open’ classes such as nouns, verbs,. ‧. adverbs and adjectives) and function words (i.e. words ascribed to ‘closed’ categories. sit. y. Nat. like pronouns, determiners, auxiliaries, complementizers, and other sorts of particles. n. al. er. io. etc.). Words of lexical categories strongly correlate with prosodic word status, forming. i Un. v. a uniform prosodic class which is prominent and immune to neutral tone phenomenon. Ch. engchi. and prosodic invisibility, whereas words of functional categories exhibit a much more erratic pattern, generally corresponding to more than one prosodic type, ranging from independent prosodic word to free prosodic clitic, and therefore members of functional categories may diverge from one another in whether to perform neutral tone and/or prosodic invisibility. A typical example in Standard Mandarin is given in (1). While a lexical word, such as the verb mai2 ‘to buy’ in this case, uniformly assumes the status of prosodic word and thus always bears a full lexical tone (1a-d), distinct types of function words may corresponds to different prosodizations: an aspect marker is outside any instance of prosodic word and therefore bears a neutral tone (1b), a phasal. 1. DOI:10.6814/NCCU202001638.

(9) complement carries a full tone by building its own prosodic word (1c), and a directional complement has an option between the two (1d & 1e). Clearly, function words correspond to two prosodic classes, termed prominent function words on the one hand, and non-prominent function words on the other. Members of the former class exhibit a similar prosodization to lexical word, while the class of non-prominent function words is vulnerable to the neutral tone process and prosodic invisibility.. (1) Neutral tone in lexical words and different types of function words a. (Qu4) (Mai4) Go to.buy T T ‘Go buy it.’. 學. ‧ 國. 立. 政治大. ‧. b. (Mai4) le0 To.buy PFV T o ‘To have bought it already’. y. Nat. sit. n. al. er. io. c. (Mai4) (hao3) To.buy done.for T T ‘To (properly) done the shopping’. Ch. engchi. i Un. v. d. (Mai4) (lai2) To.buy DIR T T ‘To get it by buying.’ e. (Mai4) lai0 To.buy DIR T o ‘To get it by buying.’. The central problem to be addressed in this dissertation is the formal characterization of prosodic distinctions between these subclasses within functional categories, which. 2. DOI:10.6814/NCCU202001638.

(10) consists of several interrelated questions. First, how are those subclasses of function words correspond to different prosodizations? Second, what types of function words does each class comprise? What motivates the distinction between the subclasses? A relatively simple and straightforward response to this problem is presented in Selkirk (1995b), which takes an interface constraint approach grounded by morphosyntax, where all prosodic differences among English words, including those among distinct prosodic types within the functional class, are analyzed by positing a single set of ranked constraints on prosodic structures, and on their interface to morphosyntactic structures. Under this analysis, different prosodizations of functional. 政治大. elements directly follow from differences in their morphosyntactic properties and. 立. language-specific constraint rankings. However, what is proposed to be a. ‧ 國. 學. straightforward correlation between a function word’s prosody and its morphosyntax. ‧. does not hold generally. Little such correlation is found in the case that serves as a basis of the dissertation, the neutral tone and prosodic invisibility of Chinese dialects. In these. sit. y. Nat. io. er. cases, just as the examples in (1) show, function words exhibit a considerable overlap in their morphosyntactic distributions that precludes any differentiation rooted in. n. al. morphosyntax.. Ch. engchi. i Un. v. On the other hand, many formal accounts (e.g. Zec and Inkelas 1990, Zec 2005) see cliticness as arbitrary, derived from an underlying feature, classification, or subcategorization that distinguishes between specific types of function words. In this subcategorization approach, constraints on prosodizations refer to specific types or even tokens of function categories in an arbitrary way, and therefore it falls short on attaining a more general mechanism. This dissertation advocate the interface constraint approach and yet propose that the ranked interface constraints are grounded by not only morphosyntactic properties. 3. DOI:10.6814/NCCU202001638.

(11) but semantic/pragmatic information. By taking an overall look at various types of function words which have potential to perform neutral tones and/or prosodic invisibility, I conclude that the distinction between prominent and non-prominent classes of function words can be drawn along the scale of grammaticalization. The idea is based on the general conception in the literature of grammaticalization that phonological erosion and semantic bleaching usually come hand in hand. As a function word becomes more semantically bleached during the course of grammaticalization, it is more likely to be non-prominent, or weak, to license certain constituents in the prosodic structure.. 政治大. By encoding this information of grammaticalization in the interface constraint,. 立. this dissertation offers a formalized mechanism to make use of both morphosyntactic. ‧ 國. 學. properties and information from semantics-syntax interface. This approach thus leads. categories.. ‧. to a unified account of prosodic distinction between subclasses within functional. er. io. sit. y. Nat 1.2 Organization of the dissertation. al. n. iv n C The dissertation is organized as follows. of chapter 1, I provide the theoretical h eInnthegrest chi U framework of the dissertation (§1.3), and preliminary notes on some important terms and concepts (§1.4). Chapter 2 is devoted to the mechanisms justifications for the currently proposed framework, which serves as the central approach to the phenomena that we address in this study. I then proceed to the core analysis of neutral tone and prosodic invisibility in grammaticalized words in Chinese dialects. Chapter 3 investigates the case of Miaoli Sixian Hakka and Cantonese; Chapter 4 is devoted to Taiwanese and Standard. 4. DOI:10.6814/NCCU202001638.

(12) Mandarin, and Chapter 5 addresses the case of Shanghainese. Chapter 6 gives some theoretical remarks on the overall picture of the dialects according to what is established in the previous chapters. And chapter 7 summarizes the conclusion of this dissertation.. 1.3 Theoretical background 1.3.1 Prosodic phonology The prosodic framework to be used in this study is an adaptation of what may be called classical Prosodic Phonology (Selkirk 1978, 1981, 1984, 1986, Nespor and Vogel. 政治大. 1982, 1983, 1986, Hayes 1984, 1989). I’ll briefly describe here the important elements. 立. of the classical theory.. ‧ 國. 學. The Prosodic Phonology framework rests on at least four general premises:. ‧. . The Indirect Reference Hypothesis: the domains of phonological phenomena are. sit. y. Nat. . Phonological representations are organized into prosodic structures.. al. iv n C Prosodic structures are based on, isomorphic to, syntactic structures. h ebutnnot gchi U n. . io. . er. delimited by phonological representations, not syntactic representations.. Prosodic structures are built from a Prosodic Hierarchy of constituent types, and are combined according to principles of Strict Layering.. The motivation for distinct prosodic and syntactic structures consists in phonological phenomena that take effect within domains that may be partly syntactically determined, but which don’t necessarily correspond to any syntactically definable constituent. Nespor and Vogel (1986) undertake a comprehensive study of such phenomena, presenting cross-linguistic evidence for every constituent type of the Prosodic. 5. DOI:10.6814/NCCU202001638.

(13) Hierarchy. For the purposes of the present study, I’ll assume that the Prosodic Hierarchy consists of the types of constituent listed in (2).. (2) Prosodic Hierarchy Constituents Intonational phrase Phonological phrase Prosodic word metrical foot syllable mora. Abbreviations ι ϕ ω Σ. σ μ. 政治大 A constituent is omitted here, the clitic group, which is posited as intermediate between 立. ‧ 國. 學. prosodic word and phonological phrase, its function being to group clitics with prosodic words (Hayes 1984, Nespor and Vogel 1986, Nespor 1999). However, I follow others. ‧. in seeking descriptions of clitic structures in terms of elements not specific to clitics. Nat. sit. y. (Berendsen 1986, Zec and Inkelas 1991, Zec 1993, Selkirk 1995a, see next subsection).. n. al. er. io. Prosodic constituents in the classical theory are organized by Strict Layering. In. i Un. v. concise terms, Strict Layering requires that a constituent of level p on the Prosodic. Ch. engchi. Hierarchy be parsed exhaustively into constituents of level p–1. For example, an intonational phrase dominates only phonological phrases, which dominate only prosodic words, and so on. However, we will see that there are reasons to think that not all prosodic structures respect Strict Layering. Another important aspect of prosodic structures is their role in the organization of phonological prominence. The idea is that every constituent has exactly one head, which is the most prominent of its sub-constituents, and that every head is aligned to all the heads beneath it (Halle and Vergnaud 1987, Hayes 1995, Selkirk 2002). As for the mapping from syntactic to prosodic structures, there are several. 6. DOI:10.6814/NCCU202001638.

(14) proposals for how this happens, including the mapping algorithms of Nespor and Vogel (1982, 1986), Selkirk’s (1986) end-based theory, and relation-based mapping (Zec & Inkelas 1990). For an overview, see Inkelas and Zec (1995). For present purposes, I’ll adopt the mapping constraints of Prosodic Clitic Theory, to which I turn next.. 1.3.2. Prosodic clitic theory. This dissertation adopts a version of Prosodic Phonology that’s based on several further hypotheses proposed in Selkirk (1995a, 1996, 2005) and Truckenbrodt (1995, 1999). Following some other scholars (Basri et al. 1998, Parker 1999), I’ll call this version of. 政治大. the classical framework Prosodic Clitic Theory (PCT).. 立. The distinguishing premises of PCT are non-strict layering, an explicit set of. ‧ 國. 學. syntax-prosody correspondences, and an implementation of the building of prosodic. ‧. structure by Optimality Theoretic constraints. PCT matters a lot to our study because the premises it proposes directly determine how lexical and functional categories can. sit. y. Nat. io. er. be distinct in the prosodic structure.. Selkirk (1995a) proposes to revise Strict Layering by decomposing it into four. al. n. iv n C separate principles, two of which aren’t By this revision, prosodic structures h eabsolute. ngchi U must obey the requirements of Headedness and Layeredness, but can sometimes violate those of Nonrecursivity and Exhaustivity:. (3) Principles Headedness Layeredness Nonrecursivity Exhaustivity. Violability inviolable inviolable violable violable. Definition a (constituent of level) p has a head q = p–1 a p doesn’t parse a q, q > p a p doesn’t parse a q, q = p a p doesn’t parse a q, q < p–1. With the violable principles, the emergence of level recursion and/or level skipping is. 7. DOI:10.6814/NCCU202001638.

(15) possible, as illustrated in (4). Note that given the nested ϕ-structure in this case, a distinction can be made between what Ito and Mester (2007, 2009) call maximal ϕ (marked by Φ) ― a ϕ not dominated by any other ϕ ― and non-maximal ϕ (marked by φ), the former being higher than the latter on the hierarchy. As discussed later in this paper, contour-tone licensing in Shanghai makes appeal to these subtypes.. (4) An instance violating Nonrecursivity and Exhaustivity. 政治大. 立. ‧ 國. 學. Another goal of PCT is an explicit and restrictive set of correspondences between. ‧. syntactic and prosodic structures. Selkirk (2005) proposes to limit these to the following one-way correspondences, and labels them the Syntactic Grounding Hypothesis:. er. io. sit. y. Nat (5) The Syntactic Grounding Hypothesis. n. al. Ch. syntactic constituent comma phrase (CmmP) lexical maximal projection (LexP) branching syntactic constituent lexical word (Lex). engchi. → → → →. i Un. v. prosodic constituent intonational phrase major phonological phrase minor phonological phrase prosodic word. In short, prosodic structure-building principles that relate syntactic and prosodic constituents can refer only to these pairs. Note that the division of phonological phrases into major phrase and minor phrase (Selkirk and Tateishi 1988) is omitted in this dissertation since it is of little importance to neutral tone and prosodic invisibility. Therefore, there is only one phonological phrase corresponding rule is available in this dissertation.. 8. DOI:10.6814/NCCU202001638.

(16) In regard to the grounding of intonational phrase, the “comma phrase” of (5) replaces Nespor and Vogel’s identification of intonational phrases with a heterogeneous set of constituents comprising root clauses, parentheticals, and non-restrictive relatives. The term comma phrase is introduced by Potts (2005) as a cover term for a similar class of intonational phrase determining constituents. He proposes that these constituents have in common a semantically significant feature COMMA, which is responsible for their semantic interpretation, and for their comma intonation. Precisely, Potts is concerned not with comma phrases in general, but with their intersection with another class of expressions (“supplements”) which consist of nonrestrictive relatives,. 政治大. parentheticals, and appositives. Therefore, Potts doesn’t claim that all comma phrases. 立. share a uniform syntax or semantics. Nonetheless, Potts’s treatment builds support for. ‧ 國. 學. a unified class of comma phrases, thereby improving on previous, less unified. ‧. characterizations of intonational phrase correspondents.. Selkirk (2011), however, proposes a new definition for determining intonational. y. Nat. er. io. sit. phrase that, which using the notion “clause” that is purely syntactically defined. She introduces at least two notions of clause come into play: “the standard clause” and the. al. n. iv n C ‘”illocutionary clause.” The standardhclause is understood e n g c h i U as “the constituent that is the complement of the functional head C. In modern syntactic theory, C, is commonly assumed to introduce the canonical sentence, which consists of an explicit or an implied subject, a predicate and a locus for Tense: CP [Comp0 [standard clause]].” The illocutionary clause “is the highest syntactic projection of the sentence and carries its illocutionary force, which determines its appropriateness in a discourse context (...); the syntactic structure for this clause type [is] assumed to be ForceP [Force0 [illocutionary clause]].” In a different paper (Selkirk, 2009), a slightly different formulation is given, where it is proposed that the term “clause” for interface mapping could apply to “the. 9. DOI:10.6814/NCCU202001638.

(17) complement of any functional head of the [Rizzi (1997)-style] ‘complementizer-layer’ ” (i.e. TopicP or FocusP). In Shanghainese, we will see that employing this syntacticallygrounded C to define intonational phrase results in different prosodization that is significant for the analysis. The Grounding Hypothesis also incorporates an idea that plays a crucial role in our analysis: prosodic structures are based only on lexical syntactic constituents, ignoring functional ones (Selkirk and Shen 1990). Truckenbrodt articulates this as the Lexical Category Condition:. 政治大. (6) The Lexical Category Condition (LCC) (Truckenbrodt 1999:226). 立. Constraints relating syntactic and prosodic categories apply to lexical syntactic. ‧ 國. 學. elements and their projections, but not to functional elements and their. ‧. projections, or to empty syntactic elements and their projections.. sit. y. Nat. io. er. This condition is based on the long observed phenomenon that function words, though being X0s in the syntax ― Ds and Cs among others ― they. al. n. iv n C consistently fail to assume the statush of prosodic word, U e n g c h i and thus tend to have a different prosody from lexical words. This idea has been carried over virtually wholesale into work making use of syntactically grounded interface constraints. The following discussion from (Selkirk, 2011, 453) is representative:. [I]t's likely that lexical and functional phrasal projections ― LexP and FncP ― have to be distinguished [...] The functional vs. lexical distinction is important for syntactic-prosodic correspondence at the word level (Fnc0 vs. Lex0): lexical category words are standardly parsed as prosodic words (°), while functional. 10. DOI:10.6814/NCCU202001638.

(18) category words like determiners, complementizers, prepositions, auxiliary verbs, etc. ― in particular the monosyllabic versions of these ― are not [...] If instead of a general Match XP this correspondence constraint were limited to lexical categories, then, on the basis of the syntactic structure [VP Verb [FncP Fnc NP]], the F-domain structure (FVerb Fnc (FNP)) would be predicted [...]. Similar claims can be found in Selkirk (1984, 1995, 2011); Hale and Selkirk (1987); Selkirk and Shen (1990); Truckenbrodt (2007); Chung (2003); Werle (2009); Selkirk and Lee (2015), among others. The common thread running through these works is that. 政治大. there is no impetus to parse function words as ωs. Yet the corollary of this ― that the. 立. phrasal projections of functional categories should not be parsed as Fs ― has been. ‧ 國. 學. challenged. For instance, Elfner (2012) shows that small clauses and TPs in Irish, both. ‧. of which are headed by a functional category, are preferentially mapped to ϕs. Furthermore, in the same vein, Tyler (2018) argues that mapping at the word level. sit. y. Nat. io. er. indiscriminately demands that all syntactic heads, lexical and functional, be mapped to prosodic words. In doing so, the word-leveling is brought in line with its fellow the. al. n. iv n C mapping at the phrase level, which Elfner applies to the phrasal projections h e nhasg argued chi U of both lexical and functional categories too. In this dissertation, I also adopt the same assumption that at either the word level or phrase level, mapping from morphosyntax to prosody does not distinguish between lexical categories and functional categories. In other words, the LCC is not assumed in this dissertation.. 1.3.3.. Constraints on prosodic structures. The last premise that distinguishes PCT is its implementation of interface conditions on syntax-prosody correspondences, as well as purely prosodic conditions. 11. DOI:10.6814/NCCU202001638.

(19) on the well-formedness of prosodic structures, as Optimality Theoretic (OT) constraints (Prince and Smolensky 1993). That is, these conditions are ranked and violable, and are assumed to comply with whatever parameters define well-formed OT constraints. In addition to Non-recursivity (NRC) and Exhaustivity (EXH), the ALIGN and WRAP constraint schemata are proposed, which can be specified to apply to various types of prosodic constituents (P-cats) and syntactic categories (S-cats). The ALIGN schema for prosodic structure determination (Selkirk 1995a) is essentially a renaming, in terms of Generalized Alignment (McCarthy and Prince 1993b), of the nonOptimality Theoretic end-based theory of prosodic structure determination (Selkirk. 政治大. 1986). The WRAP schema is argued by Truckenbrodt (1995, 1999) to be necessary to. 立. account for P-phrase construction in several languages.. ‧. ‧ 國. 學. (7) OT schemata for Prosodic Clitic Theory constraints NRC(p): a p-cat of level p doesn’t parse a p-cat of level p.. b.. EXH(p, q): a p-cat of level p doesn’t parse a p-cat of level q, q < p–1.. c.. WRAP(s, p): every s-cat of category s is contained in some p-cat of level p.. y. sit. er. io. al. iv n C ALIGN(x, L/R, y, L/R): the left/right of every constituent of type x is aligned h e nedge gchi U n. d.. Nat. a.. to the left/right edge of some constituent of type y.. Later on, Selkirk (2011) proposes a new version of constraints on the syntax-prosody mapping relation Match Theory, following Alignment Theory. In Selkirk's (2011: 451) original definition reproduced in (8), MATCH is actually not a new type of constraint, but simply two-sided Alignment.. 12. DOI:10.6814/NCCU202001638.

(20) (8) Match Theory a.. MATCH(α,π) [= SP faithfulness] The left and right edges of a constituent of type α in the input syntactic representation must correspond to the left and right edges of a constituent of type π in the output phonological representation.. b.. MATCH(π,α) [= PS faithfulness] The left and right edges of a constituent of type π in the output phonological representation must correspond to the left and right edges of a constituent of type α in the input syntactic representation.. 立. 政治大. This schemata of Match constraints has been revised at two dimensions, however. First,. ‧ 國. 學. Elfner (2012: 28), in a move away from the gradient evaluation implied in the. ‧. alignment-based conception of MATCH, proposes an all-or-nothing categorical version of MATCH-PHRASE given in (9). The subscript "T" indicates that the constraint is stated. sit. y. Nat. io. al. n. (8).. er. with reference to terminal nodes, which overcomes some problems with the version in. Ch. engchi. i Un. v. (9) MATCH-PHRASE Suppose there is a syntactic phrase (XP) in the syntactic representation that exhaustively dominates a set of one or more terminal nodes α. Assign one violation mark if there is no phonological phrase (ϕ) in the phonological representation that exhaustively dominates all and only the phonological exponents of the terminal nodes in α.. As a categorical constraint, this terminal-node-based version is easy to evaluate and is. 13. DOI:10.6814/NCCU202001638.

(21) thus assumed in this dissertation. The other revision is argued by Ito & Mester (2019). They point out that MATCHconstraints as in (8) or (9) create a serious redundancy within OT-phonology since the theory already contains not only the (semi-)equivalent edge Alignment constraints, but also a fully-worked-out subsystem of faithfulness constraints that militates against all conceivable kinds of input-output discrepancies, and syntax-prosody correspondence is just one kind of correspondence relation There is no need for MATCH constraints to duplicate their work. Accordingly, they suggest to replace the current conception of MATCH by a purely existential conception, and the constraints can thus be replaced by. 政治大. the familiar MAX/DEP constraints of General Correspondence Theory, as applied to the. 立. syntax-prosody relation. As such, SP: MAX/DEP constraints require nothing but the. ‧ 國. 學. existence of a correspondent in the output, as defined in (10), whereas IDENT and other. ‧. faithfulness constraints deal with detailed aspects of correspondence, together with the usual one-sided Alignment constraints.. er. io. sit. y. Nat (10) SP-Correspondence Constraints. al. n. iv n C syntactic hrepresentation and e n g c h i U P its. Let S be an input. corresponding output. phonological representation. a.. SP:MAX: A constituent of type α with phonological content in S corresponds to some constituent of type π in P.. b.. PS:DEP: A constituent of type π in P corresponds to some constituent of type α in S.. This revised existential schemata is proved to be more precise and applicable, and thus it is also applied throughout this dissertation.. 14. DOI:10.6814/NCCU202001638.

(22) 1.3.4.. A typology of function word parses. The combined assumptions of PCT result in a particular conception of prosodic clitics. Non-strict layering permits several different prosodic analyses of clitic parsing. Selkirk (1995a) proposes three potential analyses, exemplified in (11) with English phrases appearing in a “function word+lexical word” configuration (examples abstracted from Ito & Mester 2019). By this proposal, a function word can share a prosodic word with a host (‘internal clitic’), be parsed with a host in a recursive prosodic word (‘affixal clitic’), or be parsed directly by a higher constituent like phonological phrase (‘free clitic’).. 立. 政治大. (11) Three kinds of prosodic clitics. ‧ 國. 學. Internal clitics. the students ϕ}. y. Nat. ( the students. ω). sit. ( the students. ‧. {. ω). io. n. al. er. a.. b.. Affixal clitics {. c.. Ch. engchi. i Un. v. could ( stay ω) ϕ}. ( could ( stay ω). ω). ( could ( stay ω). ω). Free clitics {. at ( home ω) ϕ}. (. at ( home ω) ω). (. at ( home ω) ω). 15. DOI:10.6814/NCCU202001638.

(23) Here the crucial constraints are at the word level, where lexical status is not distinguished from functional status, as opposed to the version proposed by Ito and Mester (2019).. (12) Word-level correspondence constraints (cf. SP:MAX-Lex of Ito & Mester 2019) a.. SP:MAX-X0: A constituent of type X0 (syntactic word) with phonological content in S corresponds to some constituent of type ω (prosodic word) in P.. b.. PS:DEP-ω:. 政治大. A constituent of type ω (prosodic word) in P corresponds to some constituent of. 立. type X0 (syntactic word) in S.. ‧ 國. 學 ‧. SP:MAX-X0 is only completely fulfilled in candidate (11b), because both the lexical word and function word have their respective correspondents ωs. Although the ω built. y. Nat. er. io. sit. for the function word is no exact match, with the right edge misaligned, SP:MAX-X0 is not violated since it only cares about the existence of the correspondent ω. Therefore,. al. n. iv n 0 C candidates (11a) and (11c) each violate AX-X once, for the lack of correspondent h eSP:M ngchi U. ω built on the function words. Note that exact correspondence (preservation of edges, no deletion, no insertion, uniqueness of mapping, order preservation, etc.) is enforced by the other faithfulness constraints (IDENT, LINEARITY, UNIFORMITY, INTEGRITY, etc.) and one-sided edge ALIGN, so in (11b), the misaligned right edge of the ω built on the function word is enforced by high-ranking Align-R(ω, Lex), which dictates that the right edge of ω be licensed by the right edge of some lexical word. As we will see in the following chapters, this is crucial for the distinction between lexical and functional categories given the non-discriminated MATCH version.. 16. DOI:10.6814/NCCU202001638.

(24) 1.4 Preliminaries of neutral tone In complex tonal languages such as Beijing Mandarin, F0 contours are utilized to lexically contrast every syllable, known as lexical tones. Yet syllables in unstressed positions, typically being grammatical morphemes, or the second syllable of certain disyllabic or trisyllabic words, may surface with none of the lexical tones; instead, they are referred to as having the so-called neutral tone, known in Chinese as qingsheng ‘light tone.’ The term neutral tone was first introduced by Chao in the 1920s to describe the pitch pattern of unstressed syllables in Beijing Mandarin which shows variability as a function of the preceding lexical tone (1932, 1933, 1968). To illustrate, the. 政治大. directional verb lai ‘to come’ bears the rising lexical tone [35] when it is stressed. As a. 立. post-verbal complement in guo-lai ‘to come over here’ and chu-lai ‘to come out, where. ‧ 國. 學. it often gets destressed, lai has a neutral tone with a surface mid-to-low pitch and a. ‧. surface high-to-mid pitch, respectively, depending on which lexical tone the preceding syllable is associated. Since the F0 realization of neutral tone is largely determined by. sit. y. Nat. io. er. the lexical tone of the preceding syllable, a neutral-toned syllable is traditionally considered toneless/ targetless (Yip 1980; Wang 2000, 2003; Li 2004) or “unspecified. n. al. C h terms. U n i for tone” (Duanmu 2000) in phonological engchi. v. In one word, stresslessness and tonelessness are generally accepted to be the intrinsic characteristics of neutral tone. In the remainder of this section, however, I will argue that, although stresslessness indeed catches the nature of neutral tone, tonelessness is better to be replaced by complete (paradigmatic) neutralization. Besides, a preliminary formalization of neutral tone based on the currently proposed definition is also attained.. 1.4.1. Definition of neutral tone. Since the initialization of Chao (1968), a cover term such as “stresslessness” or the like. 17. DOI:10.6814/NCCU202001638.

(25) has often been used to explain why underlying tones are lost or neutralized in neutraltoned syllables. The idea is that neutral-toned syllables usually have the phonetic properties which are regarded as the phonetic correlates associated with stresslessness. One of the properties is shortened duration (probably the most-mentioned one as regards the nature of neutral tone). A substantial literature (Lin & Yan 1980; Cao 1986; Chen & Xu 2006; Lee & Zee 2008; Hsieh & Chuang 2008; Chen 2015, among others) indicates that, at least for most Mandarin dialects, the mean duration of neutral-toned syllables is about 50%-60% of the full lexical-toned syllables. Another significant phonetic property of neutral tone for stresslessness is weak articulatory strength (Chen. 政治大. & Xu 2006; Huang 2015). For example, Huang (2015) contends that neutral tone in. 立. Taiwan Mandarin, though not consistently shortened, is still “unaccented” because its. ‧ 國. 學. articulatory strength is weaker than the full lexical tones in the similar pitch range.. ‧. Taken together, neutral-toned syllables are often realized with clear acoustic reductions. Specifically, at the segmental level, vowels in neutral-toned syllables are often. sit. y. Nat. io. er. centralized, devoiced, or even elided (Zadoenko 1958; Chao 1968; Cheng 1973; Gao 1980; Lin and Yan 1980, 1990; Lin 1983; Cao 1986; Chen 1986; Li 1990; Yang 1991;. n. al. Duanmu 2000; Wang 2004;. iv n C Lee and 2008). Tonal neutralization is therefore h eZee ngchi U. nothing but a sign of reduction motivated by stresslessness at the tonal level (Hsieh & Chuang 2008). As regards the processing of tonal neutralization itself, the view has been widely held that neutral-toned syllables are toneless at some point during the derivation, because it can better account for the variable surface pitch pattern and its dependency on the preceding lexical tones. To illustrate this point, let us consider two types of neutral tones in derivational terms. The first type, as depicted in (1), is inherent neutral tone, which refers to the case where syllables are always neutral-toned, even in. 18. DOI:10.6814/NCCU202001638.

(26) supposedly stressed positions. That is, they have lexicalized the status of being neutraltoned. Morphemes of this type in Beijing Mandarin include the sentence-final particles, such as ba, and ma, and the possessive marker de, among many others. These syllables have no lexical tones and thus are analyzed as underlyingly toneless. Consequently, the last resort for the toneless syllable to surface with some pitch is either spreading from the preceding lexical tone (Yip 1980), or from the interpolation between the preceding lexical tone and a boundary tone (represented as T%, see Li 2003). Either way, tonelessness serves as a motivation for the mechanism deriving the surface pitch pattern of inherent neutral tone.. 立. 政治大. (13) Inherent neutral tone: underlying tonelessness. ‧ 國. σ. 學. σ. σ. σ. TT. ‧. →. T T (T%). er. io. sit. y. Nat. al. iv n C U Mandarin. These verbs are h equn‘toggo’ the directional verbs lai ‘to come’ and c hin iBeijing n. The other type of neutral tone is non-inherent, as sketched in (14). Take for example. underlyingly specified for tone in citation, the former being [35] and the latter [51], whereas as an unstressed post-verbal complement in chu-lai ‘to come out’ and chu-qu ‘to go out,’ their tonal contrast are neutralized through two stages: tone loss in unstressed positions and tonal spreading/interpolation. In this case, the tonelessness of non-inherent neutral tone is not underlying, in contrast to the case of inherent neutral tone in (13); rather, the tonelessness of non-inherent neutral tone is derived as a result of stresslessness, and it is this derived tonelessness that feeds the tonal spreading/interpolation, in the very same way as the underlying tonelessness in (13). 19. DOI:10.6814/NCCU202001638.

(27) does. In other words, tonelessness, in conjunction with stresslessness, contribute to the surface pitch pattern of non-inherent neutral tone.. (14) Non-inherent neutral tone: derived tonelessness σ. σ. σ. σ. TTTT. σ. σ. →. → TT. T T (T%). Note that in both cases (13) and (14), tonelessness is never a surface-true statement. Especially in the latter case (14), the derived tonelessness in unstressed syllables is at. 政治大. an intermediate stage in the derivation. Therefore, this non-surface true tonelessness. 立. will run into difficulties fitting in with a surface-oriented non-derivational framework,. ‧ 國. 學. such as Optimality Theory (henceforth OT). Specifically, the only mechanism in OT. ‧. that can capture tonelessness derived in weak positions is positional markedness, as. sit. y. Nat. indicated in the constraint ranking: *TONE/σ̆ ≫ MAX(TONE), which prohibits unstressed. io. er. syllables associated with a tone. And yet, this ranking wrongly rules out the surface. al. iv n C hiteexists). lexical tone (and the boundary tone if h i isUwhy the previous accounts (e.g. n g cThis n. form in (14), where the unstressed neutral-toned syllable is linked to the preceding. Li 2003) for the tonelessness in OT resort to positional faithfulness instead, along with an economy constraint (i.e. *STRUC): MAX(TONE)/ˈσ ≫ *STRUC (TONE) ≫ MAX(TONE). Despite the applicability, this approach is still far from flawless in two regards. First, it attributes the derived tonelessness in (14) to the effect of * STRUC(TONE), failing to capture the insight in the derivation that the derived tonelessness is motivated by stresslessness. Second, as Gouskova (2003) argues, the economy constraints like *STRUC(TONE) should be excluded from CON, given the principle that “for every markedness constraint, there is at least one non-null structure that fully satisfies it.”. 20. DOI:10.6814/NCCU202001638.

(28) (Gouskova 2003:17) Accordingly, we may say that tonelessness, more precisely the derived tonelessness, is incompatible with a surface-oriented theory like OT. Besides the theoretical incompatibility, the tonelessness account also acts counter to experimental results that the F0 realization of neutral tone does exhibit its own tonal target (e.g. Chen & Xu 2006 for Beijing Mandarin; Huang 2018 for Taiwan Mandarin; Li & Chen 2019 for Tianjin Mandarin). For example, in Chen & Xu (2006), the number of neutral-toned syllables was manipulated from one to three. Their results showed that, when there was only one neutral-toned syllable, the F0 realization of neutral tone was varied greatly as a function of previous full lexical tones. In contrast, as the number of. 政治大. neutral-toned syllables increased, the F0 realization of neutral tone showed a clear. 立. tendency of merging into a stable mid-low target of its own, regardless of the preceding. ‧ 國. 學. and following full lexical tones. The authors argue that, due to occurring in unstressed. ‧. positions, neutral tone in Beijing Mandarin is implemented with much weaker articulatory strength compared to the full lexical tones. Therefore, neutral tone is less. sit. y. Nat. io. er. effective in overcoming the influence of the surrounding full lexical tones, and also takes more than one neutral-toned syllable to approach its tonal target, resulting in more. al. n. iv n C surface F0 variability observed in the hneutral-tone i U e n g c hrealization.. Given both the theoretical problems and the experimental evidence, I propose a new definition for neutral tones based on surface-orientedness, in replacement of the non-surface true “tonelessness” view, as stated below.. (15) Surface-oriented definition of neutral tones a.. Neutral-toned syllables are unstressed.. b.. The lexical tones are completely (paradigmatically) neutralized in neutral-toned syllables.. 21. DOI:10.6814/NCCU202001638.

(29) c.. Neutral-tone realization is acquired by: i.. The influence of surrounding tones; and/or. ii.. The emergence of a (default) reduced tone.. Note that this definition disassociates neutral tone from shorter duration and weaker energy. That is to say, neutral tone can be characteristic of the two properties, and yet neither of these properties are sufficient or necessary to define whether it is a neutral tone or not. Consequently, neutral tone in this dissertation comprises not only the typical type as in Beijing Mandarin, which is shorter and weaker, it also covers a few cases that. 政治大. is traditionally termed tonal neutralization. For example, the default or variable tonal. 立. manifestation of (some) sentence-final particles in Cantonese and Hakka Chinese, and. ‧ 國. 學. the tone loss and spreading/redistribution of the preceding tone involved in normal tone. ‧. sandhi in Shanghainese. These cases are under the scope of neutral tone in this dissertation because they involve stresslessness, complete neutralization of. sit. y. Nat. io. er. paradigmatic contrast, and the surface pitch in these cases is acquired by spreading of the offset of the preceding full tone, emergence of a default reduced form, or association. n. al. with the boundary tone.. Ch. engchi. i Un. v. 1.5 Language background There are five Chinese dialects/languages investigated in this dissertation, each representing one of the major Sinitic families. Cantonese is the prestige variety of Yue Chinese originating from the city of Guangzhou. Miaoli Sixian Hakka is the prevailing subdialect of Northern Sixian Hakka spoken in Miaoli County, Taiwan. Standard Mandarin is the official language in both mainland China and Taiwan, and is thus the. 22. DOI:10.6814/NCCU202001638.

(30) representative of Mandarin Chinese. Taiwanese is a famous variety of Southern Min Chinese in Taiwan, and Shanghainese is the best-known dialect of (northern) Wu Chinese.. (16) Languages investigated Languages/dialects Miaoli Sixian Hakka Cantonese Standard Mandarin Taiwanese Shanghainese. Affiliations Hakka Chinese Yue Chinese Mandarin Chinese (Southern) Min Chinese (Northern) Wu Chinese. Distribution (Northern) Taiwan Guangdong, Hong Kong Mainland China, Taiwan Taiwan, Xiamen Shanghai. 政治大. Note in particular that Standard Mandarin comprises two varieties in this dissertation.. 立. One is known as Putonghua (普通話), which is the standard version in mainland China.. ‧ 國. 學. The other is called Guoyu (國語), a variety of Standard Mandarin localized in Taiwan.. ‧. I will distinguish the two where necessary by dubbing the former mainland “Standard. sit. y. Nat. Mandarin” and the latter “Taiwan Mandarin.” This contrast is important because the. al. n. current analyses.. er. io. two varieties show differences in the patterns of neutral tones that are significant to the. Ch. engchi. 23. i Un. v. DOI:10.6814/NCCU202001638.

(31) 2. Prosodic Prominence and Degree of Grammaticalization. 2.1 Prosodic licensing by edge-alignment In virtually all the prosodic analyses making use of Match theory or its predecessors, it is usually taken for granted in the syntax-phonology mapping principles that there is a conventional dichotomy between lexical and functional categories. Only members of. 政治大. lexical categories can uniformly assume the status of an independent prosodic. 立. constituent. Members of functional categories exhibit a much more erratic pattern. ‧ 國. 學. cross-linguistically, generally corresponding to more than one prosodic type which. ‧. allegedly follows straightforward from differences in their syntactic distribution. However, this dichotomy does not hold generally in the case which serves as a basis of. sit. y. Nat. io. er. this dissertation, that of Chinese dialects. In a range of branches of Chinese family, there are certain dialects, such as those in Yue Chinese and Hakka, where functional. al. n. iv n C elements are as prosodically robust as in terms of tonal phonology, and hlexical e n gelements chi U hence no need to distinguish between the two categories. On the other hand, in many Chinese dialects, including Standard Mandarin, words of functional category correspond to two prosodic classes with respect to particular phonological processes, prosodically prominent function words on the one hand, and prosodically nonprominent function words on the other. Members of the former class are parsed into the prosodic hierarchy on a par with words of lexical categories, that is, assuming a prosodically salient status by inclusion in a crucial prosodic constituent. This salience protects them from particular phonological reduction. In contrast, the class of non-. 24. DOI:10.6814/NCCU202001638.

(32) prominent function words is characterized by absence of such salience due to its being excluded from the very prosodic constituent; therefore, members of this class are closer than those of the prominent class to what is traditionally defined as function words in the sense that they are reduced in form and/or eligibility for taking part in normal phonology. Significantly, the two classes of function words in Chinese dialects exhibit a considerable overlap in their morphosyntactic distributions. This case, together with the former one, where the lexical/functional contrast is unavailable, jointly calls for a more sophisticated system to deal with the function words which fare the same way as lexical ones in prosody, while keeping the distinction between lexical category and. 政治大. (phonologically reduced) functional category.. 立. In this dissertation, I propose a formal mechanism that not only captures the two. ‧ 國. 學. prosodic asymmetries empirically attested ‒ the asymmetry between words of lexical. ‧. and functional categories, and between the two prosodic classes of functional elements ‒ but also accommodates the situation where there is no asymmetry at all. The core of. sit. y. Nat. io. er. the proposal is as follows. In principle, the interface grammar grants words of functional category a correspondent in the prosodic structure which is equivalent to that of words. n. al. of lexical category, contra. iv n C the convention h e n g cinh itheUliterature.. Nevertheless, the. correspondents of these categories does not fare alike generally with respect to the prosodic markedness. While lexical correspondent is always prosodically well-formed, the correspondent of functional category is prone to be incompatible with the prosodic structure, and the incompatibility of the two prosodic classes of functional elements differs in an implicational way: if it is illegitimate for the non-prominent class to have a correspondent, then so is it for the prominent class, not vice versa. In other words, lexical category and the two classes of functional elements are in a ordering relation, forming. This can be schematically represented by a hierarchy regarding prosodic. 25. DOI:10.6814/NCCU202001638.

(33) salience, as in (1), with lexical category at the most salient end, the non-prominent function word at the opposite end, and the prominent function word in between, where “A › B” is read as “A is more salient than B,” the details of which will be discussed in the next section.. (1) Lexical word › prominent function word › non-prominent function word. The conception is cast in Optimality Theory by a set of interface constraints, following the revised version of Match Theory argued by Ito and Mester (2019), as laid out below.. 立. (2) Interface constraints. 政治大. ‧ 國. 學. Let S be an input syntactic representation and P its corresponding output. Assign a violation mark for every constituent of type α. Nat. y. SP-MAX(α, π):. sit. a.. ‧. phonological representation.. with phonological content in S that does not correspond. er. io. n. atol some constituent of type π iinv P. C ha violation markUfornevery constituent of type π ALIGN-L/R(π, α): Assign engchi. b.. that is not left- or right-aligned with its corresponding constituent of type α in S.. This system principally consists of two well-known families of constraints. The SPMAX in (2a) is a faithfulness constraint requiring Syntax-to-Prosody Maximality, and as such, it differs from the generally adopted MATCH constraint, according to Ito and Mester (2019), in having a purely existential conception. That is, for an element given in the input syntactic representation, SP:MAX insists merely on the existence of some. 26. DOI:10.6814/NCCU202001638.

(34) corresponding prosodic constituent in the output phonological representation, rather than on exact correspondence, which is enforced by the other SP-faithfulness constraints. What is crucial is that this existential correspondence constraint is indifferent to whether the input element belongs to lexical category or functional category. The potential asymmetry between the categories is then governed by the Prosody-to-Syntax edge Alignment in (2b), the other family of constraints that principally makes up our system, which requires an output prosodic constituent in alignment on one side of some particular type of syntactic constituent. To capture the ordering effect in hierarchy (1), the value of the variable α for edge alignment in (2b). 政治大. refers to contiguous ranges in the hierarchy, resulting in a set of stringency constraints. 立. (see De Lacy 2002), as illustrated in the following tableau, where “ω” stands for. ‧ 國. 學. prosodic word, “Lex” for lexical word, “FncPrm” for prominent function words,. ‧. “FncNonPrm” for non-prominent function words, and “X0” for morphosyntactic word.. sit. y. Nat. io. n. al. Ch. a. (ω Lex)(ω FncPrm) (ω FncNonPrm) b. (ω Lex)(ω FncPrm) FncNonPrm c. (ω Lex) FncPrm FncNonPrm. ALIGN-R (ω, Lex) ** *. er. (3) Edge Alignment with stringency formulation. ALIGN-R (ω, Lex-FncPrm) *. n U engchi. iv. ALIGN-R (ω, X0). With these two sets of constraints, the different mapping status can be treated as a result of the interaction of the SP-faithfulness and the stringent edge Alignment. The permutation of these two families of constraints produces the typology in (4), including the asymmetry between lexical and functional categories (4c), the asymmetry between the two classes of functional words (4b), and even no asymmetry (4a). As we can see, the conventional lexical/functional dichotomy arises as long as ALIGN-R(ω, Lex) is ranked above SP:MAX-X0. If SP:MAX-X0 is undominated, there is no asymmetry. 27. DOI:10.6814/NCCU202001638.

(35) available, lexical and functional categories are equally salient by assuming the same prosodic status. With SP-MAX-X0 above ALIGN-R(ω, Lex), while being dominated by ALIGN-R(ω, Lex-FncPrm), the asymmetry between the two classes of function words is observed.. (4) Factorial typology language Type a. (ω Lex)(ω FncPrm) (ω FncNonPrm). Grammar SP:MAX-X0 ≫ ALIGN-R(ω, Lex), ALIGN-R(ω, Lex-FncPrm). b. (ω Lex)(ω FncPrm) FncNonPrm. ALIGN-R(ω, Lex-FncPrm), ≫ SP:MAX-X0 ≫ ALIGN-R(ω, Lex). c. (ω Lex) FncPrm FncNonPrm. 立. 治政 A -R(ω, Lex-Fnc 大 ), A SP:M -X LIGN AX. Prm. LIGN-R(ω,. Lex) ≫. 0. ‧. ‧ 國. 學. ALIGN-R(ω, Lex) ≫ SP:MAX-X0 ≫ ALIGNR(ω, Lex-FncPrm). sit. y. Nat. 2.2 Conflation of grammaticality1. n. al. er. io. One might think that the subclasses of function words, namely the prominent function. i Un. v. words and the non-prominent function words, is as arbitrary as the subcategorization. Ch. engchi. approach, in the sense that the classification appears to be stipulated based on nothing but the phonological behavior of function words, and that members of the two classes may be different from one dialect to another. This inconsistency is shown in table (5), which gives a cross-dialectal summary of the members of the two prosodic classes within functional category, based on the (non-)application of tonal reduction. As we can see, the prominent class contains the largest range of distinct types of function words in Sixian Hakka, but the smallest range in Shanghai. The non-prominent class. 1. The term designates the degree of grammaticalization in Lehmann’s (2015) sense.. 28. DOI:10.6814/NCCU202001638.

(36) exhibits the reverse.. (5) Prominent and non-prominent function words NUM2. Sixian Hakka Taiwan Mandarin Southern Min Shanghainese. Prm. PHA. CL. PRO. DIR. Prm Prm NonPrm NonPrm. NUM = Numerals CL = Classifiers DIR = Directional complements MOD = Modifier markers. ASP. MOD. SFP. Prm. NonPrm. NonPrm. PHA = Phasal complements PRO = Object pronouns ASP = Aspect markers SFP = Sentence final particles. Prm = Prominent function words NonPrm = Non-prominent function words. 政治大 Inconsistent and variable as it appears, the classification is clearly implicational. In 立. ‧ 國. 學. other words, these distinct types of function words form an implicational hierarchy. The reduction of a given type in the scale entails the reduction of all the types to its left.. ‧. This is by no means accidental. As I shall argue here, the ordering of these types reflects. sit. y. Nat. io. As one may notice, numerals do not form a typical functional category and are sometimes termed. er. 2. al. iv n C both lexical and functional properties: (a) they h ebelong h i U class of elements (new numerals n gtoca semi-open n. “semi-lexical” (Corver and van Riemsdijk 2001) or “lexical operator” (Keizer 2007) because they show. can be added, but new numerals are not as common as new nouns or verbs), (b) they are phonologically and morphologically independent, (c) they only combine with nouns, (d) they do not assign theta roles (though they do take arguments), and (e) they cannot be separated from their complements, two of which being lexical properties (class membership and independence) and three being functional properties (complements, theta roles, separability). By defining functional category in a broad sense, we suggest that numerals as having such a semi-lexical status should be taken as a member of functional category, just as pronouns, which is another instance of lexical operator in Keizer (2007). Another justification for grouping numerals into functional category is formal. In generative syntax, numerals are treated as the head of a functional projection “number phrase,” or NumP, which is parallel to IP in CP domain.. 29. DOI:10.6814/NCCU202001638.

(37) their grammaticality, that is, the degree of being grammaticalized. This idea is not as surprising as it may appear. In the literature of grammaticalization, there has long been a consensus that phonological attrition and semantic depletion, or bleaching, go hand in hand, as dubbed by the term: the coevolution of meaning and form. Accordingly, I argue in this section that the classification of prominent and non-prominent function words complies the degree of grammaticality, such that members of prominent class tend to be “less grammatical” than those of non-prominent class. Take the types of function words in (5) for example. I suggest a grammaticality scale for these types, which exactly follows the ordering that they are arranged in (5), as depicted below,. 政治大. where “A < B” is read as “A is less grammatical than B”.. 立. NUM. < PHA, CL < PRO, DIR < ASP < MOD < SFP. Nat. y. ‧. ‧ 國. 學. (6) Grammaticality Scale. er. io. sit. This scale can be justified using Lehmann’s (1995) well-known parameters of grammaticalization, which are widely employed as a set of criteria to determine which. al. n. iv n C of two linguistic items is more grammatical the other. Lehmann’s model is based h e n gthan chi U on three principled aspects of the autonomy of a linguistic sign, weight, cohesion and variability, which can be analyzed from a paradigmatic and syntagmatic point of view. This results in six parameters, or six criteria, yet it is argued that only some of which are applicable to Chinese (Bisang 2020), as listed in (7).. (7) Parameters of grammaticality a.. Integrity: desemanticization (loss of semantic substance),. b.. Bondedness: univerbation (boundary loss). 30. DOI:10.6814/NCCU202001638.

(38) c.. Syntagmatic variability: fixation (decrease in syntactic freedom). Of the three parameters, bondedness and syntagmatic variability distinguish only limited types on the scale in (6). For example, the parameter of syntagmatic variability tells that numerals are less grammatical than all the other types, because they have much higher freedom in syntactic distribution. But it fails to explain why classifiers, having an affix-like status, can be less grammatical than object pronouns, which are relatively free in syntactic distribution. By the same token, the related parameter, bondedness, though successfully recognizes the relatively lower grammaticality of phasal. 政治大. complements than aspect markers through the test of separability, it cannot tell us why. 立. phasal complements are less grammatical than directional complements on the one hand,. ‧ 國. 學. and equally grammatical to classifiers on the other, given the fact that directional. ‧. complements are as much separable as them from the preceding main verb, while classifiers are inseparable from the preceding numerals. In contrast, the last parameter,. sit. y. Nat. io. er. Integrity, is more reliable than the other two. This parameter manifests itself in desemanticization. The term either means loss of contentive meaning, or addition of. al. n. iv n C more abstract meaning or function, h where abstractnessU e n g c h i can be roughly defined as lack. of mental image. This parameter helps distinguish the otherwise problematic pairs. For instance, phasal complements and classifiers are less grammatical than object pronouns and directional complements because they still keep contentive substance to some extent, while object pronouns and directional complements only has referential, deictic or aspectual function. With the grammaticality scale in hand, the two prosodic classes of function words are therefore built on the hierarchy by ignoring, or “conflating” in De Lacy’s (2002) term, some of the distinctions between the distinct types with different grammaticality.. 31. DOI:10.6814/NCCU202001638.

(39) This processing of conflation is formalized using the stringent edge alignment proposed in 2.1. To begin with, let us suppose that each level in the hierarchy has a distinctive feature, or degree, in the form of a number that is greater or equals to 1. The greater the number, the higher the grammaticality. Therefore, lexical words have the feature value 0. The feature specifications shown in (8). It follows that the stringent edge alignment refers to contiguous ranges, following the revised definition given in (9). These can be seen as finer-grained divisions of ALIGN-R(ω, Lex-FncPrm). To illustrate, ALIGN-R(ω, Fnc≤2) refers to the edge alignment of ω with a numeral, phasal complement or classifier.. 立. Fnc=5. Fnc=6. ASP. MOD. SFP. sit. n. al. er. io (9) ALIGN-L/R(π, Fnc≤n):. y. Fnc=4. ‧. Fnc=3 PRO/DIR. Nat. NUM. Fnc=2 PHA/CL. 學. Fnc=1. ‧ 國. (8). 政治大. Ch. engchi. i Un. v. Assign a violation mark for every constituent of type π that is not left- or rightaligned with a syntactic word of the type Fnc≤n in S.. The proposed constraint system captures both the universal and language-specific nature of the grammaticality–reduction relation. On the universal side, the system captures the fact that all else being equal, phonological reduction never seeks out a less grammaticalized function word in preference to a more grammaticalized one. On the. 32. DOI:10.6814/NCCU202001638.

(40) language-specific side, the constraints in (3) allow conflation. Various conflations of grammaticality across Chinese dialects arise from the interaction between the stringent edge Alignment and SP-faithfulness. When SP-MAX is crucially outranked by edge Alignment referring to a range with maximum degree x, all the types of function words with grammatical degree ≤ x are conflated into the prominent class, and the rest types are the non-prominent class. For instance, in the case of Taiwan Mandarin, the boundary of prominent/non-prominent distinction is located between grammaticality degree 3 and 4; this amounts to the ranking: ALIGN-R(ω, Fnc≤3) ≫ SP-MAX.. 2.3 Alignment at different levels. 立. 政治大. The stringent edge Alignment proposed for the present study serves as a mechanism of. ‧ 國. 學. licensing function words with different degrees of grammaticality within a certain. ‧. prosodic constituent. While higher-grammaticality function words have the priority to be licensed, lower-grammaticality function words tend to be excluded. This licensing. sit. y. Nat. io. er. is parameterized to be sensitive to the edges of a constituent at two distinct levels in the prosodic hierarchy, prosodic word and phonological phrase, as proposed below.. n. al. (10) a.. ALIGN-L/R(ω, Fnc≤n). b.. ALIGN-L/R(ϕ, Fnc≤n). Ch. engchi. i Un. v. Word-level Alignment as formalized in (10a) may position non-prominent function words outside prosodic word, allowing for only lexical word and prominent function words at that position, through the interaction with SP-faithfulness. The prosodic (non)licensing effect is observable in the presence/absence of tonal neutralization, given the assumption that tonal neutralization is derived from stresslessness, and that a. 33. DOI:10.6814/NCCU202001638.

(41) syllable is unstressed for not being parsed in a prosodic word. Tonal neutralization thus serves as one of the basis for the word-level licensing. For example, in Taiwan Southern Min, function words with grammaticality degree greater than 2 are neutral-toned, which can be explained by the ranking ALIGN-R(ω, Fnc≤2) ≫ SP-MAX-X0.. (11) Word-level licensing Constraint ranking: ALIGN-R(ω, Fnc≤2) ≫ SP-MAX-X0 Subclasses Prosodization Tonal neutralization. Prominent (Fnc≤2) (ω Lex)(ω Fnc≤2) N/A. Non-prominent (Fnc>2) (ω Lex) Fnc>2 Neutralized. 政治大 Phrase-level licensing effect 立 can be observed through invisibility. This is a phenomenon. ‧ 國. 學. where function words are invisible to the normal tone sandhi. An example is Sixian Hakka, in which case the modifier marker ge, a degree 5 function word, carries a full. ‧. lexical tone which is a potential trigger of the processing of tone sandhi; however, the. Nat. sit. y. tone sandhi turns out to be blocked. Given that the tone sandhi is bound by the right. n. al. er. io. edge of phonological phrase, the invisibility of modifier marker can be accounted for by ranking ALIGN-R(ϕ, Fnc≤4) above SP-MAX-X0.. Ch. engchi. i Un. v. (12) Phrase-level licensing Constraint ranking: ALIGN-R(ϕ, Fnc≤4) ≫ SP-MAX-X0 Subclasses Prosodization Tone sandhi. Prominent (Fnc≤4) (ϕ Lex Fnc≤4) Applied. Non-prominent (Fnc>4) (ϕ Lex) Fnc>4 Blocked. Word-level and phrase-level licensing effects may co-occur in the same language. Shanghainese serves as a good example. In the language, all the function words but numerals are tonally neutralized, with sentence final particles further showing. 34. DOI:10.6814/NCCU202001638.

(42) significant invisibility by blocking the processing of tone spread. This results in different conflations. At the word-level prosody, function words with grammaticality degree greater than 1 are conflated into non-prominent class, undergoing tonal neutralization, as shown in (13). At the phrase-level, on the other hand, the nonprominent class only contains sentence final particles, the most grammatical function word in the proposed scale, as shown in (14). Therefore, most function words whose grammaticality is in between are partially reduced, in the sense that they only undergo one of the phonological reduction.. 政治大. 立. Prominent (Fnc≤1) (ω Lex)(ω Fnc≤1) N/A. Non-prominent (Fnc>1) (ω Lex) Fnc>1 Neutralized. ‧. Subclasses Prosodization Tonal neutralization. 學. ‧ 國. (13) Word-level: ALIGN-R(ω, Fnc≤1) ≫ SP-MAX-X0. n. al. Ch. engchi. y. sit. Prominent (Fnc≤5) (ϕmax Lex Fnc≤5) Applied. Non-prominent (Fnc>5) (ϕmax Lex) Fnc>4 Blocked. er. io. Subclasses Prosodization Tone spread. Nat. (14) Phrase-level: ALIGN-R(ϕmax, Fnc≤5) ≫ SP-MAX-X0. i Un. v. From this case emerge two theoretical implications. Firstly, the more grammatical a function words is, the more reduced it may be, and the degree of reduction can be measured by how many processes and/or which processes of phonological attrition applies to the function word. Secondly, the conflation of grammaticality is not only language-specific, but also process-specific. Different processes of reduction may call for different scenarios of conflation.. 35. DOI:10.6814/NCCU202001638.

(43) 2.4 Summary This chapter proposes the theoretical model. Several major claims are made here. Function words do not form a monolithic and homogeneous group in prosody. They should be divided along the scale of grammaticalization degree into at least two prosodic classes: the prominent class, which is eligible for licensing the edge of prosodic constituents at either word level or phrase level, and the non-prominent class, which does not enjoy that licensing privilege and is therefore extraprosodic at either level, being vulnerable to tonal neutralization and/or prosodic invisibility. This distinction is formalized by the schemata of constraint ranking ALIGN-R(ω, Fnc≤n).≫. 政治大. SP:MAX-X0 and ALIGN-R(Pcat, Fnc≤n) ≫ ALIGN-R(XP, Pcat).. 立. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. 36. i Un. v. DOI:10.6814/NCCU202001638.

(44) 3. Grammaticalized Function Words in Cantonese and Sixian Hakka. 3.0 Introduction Yue Chinese and Hakka Chinese are known for having relative scarce neutral tone compared to all the other branches of Sinitic languages (Cheng and Tseng 1997). Certain. 政治大. languages in the two families, such as Cantonese and Miaoli Sixian Hakka, are reported. 立. to even have no neutral tones (e.g. Matthews and Yip 1994, Liu 2004); that is, in these. ‧ 國. 學. languages each syllable, even those belonging to functional categories that are typical. ‧. target for tonal neutralization in Beijing Mandarin, carries a distinct lexical tone. Cantonese and Miaoli Sixian Hakka are therefore taken in this dissertation as. sit. y. Nat. representatives of languages at the lowest end of the neutralization scale, and will be. er. io. addressed in this very chapter. aHowever, instead of following the no-neutralization view,. n. iv l C n I shall argue that the category of sentence-final h e n g cparticles h i U serves as the one and only site. for tonal neutralization, with the apparent full lexical tone carried by (a subset of) sentence-final particles analyzed as resulting from a boundary tone or default tone. Under this analysis, grammatical words in both Cantonese and Miaoli Sixian Hakka are held to be minimally neutralized. Only the most grammaticalized function words (i.e. sentence-final particles) get to undergo tonal neutralization. Besides tonal neutralization, Tone sandhi in Miaoli Sixian Hakka also involves phenomenon called invisibility or extrametricality in this dissertation. That is, when. 37. DOI:10.6814/NCCU202001638.