Generalized surface correspondence in reduplicative opacity

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content GENERALIZED SURFACE CORRESPONDENCE IN REDUPLICATIVE OPACITY by Yifan Yang A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (LINGUISTICS) August 2022 Copyright 2022 Yifan Yang ii Acknowledgements This dissertation, as well as my graduate study, cannot be finished without the help of many people. First and foremost, my deepest gratitude goes to the chairs of my dissertation committee, Rachel Walker and Stephanie Shih, for their tremendous help on this project over the years. Rachel had been my advisor at USC for six years before moving to UC Santa Cruz. She continued overseeing my dissertation and other projects in the last year of my graduate study. Rachel is the most meticulous, diligent, organized, punctual, and considerate person that I’ve ever met. I cannot tell how much I have benefited from her in the past years. I always got timely and insightful feedback, not only for this dissertation, but also for all my assignments, term papers, conference abstracts, presentations, and everything. I also felt so refreshed and encouraged after every meeting with her, especially when I was in despair. Rachel is an awesome advisor and she will continue to be my life-long role model in my future career as a teacher and scholar. I am grateful to Stephanie for her inspiring comments and questions. I am always impressed by her new ideas and her curiosity into language data and research methodology. Her seminars about tone, sound symbolism, and Pokémon, changed my old-school-way of thinking about phonology. Also, thank Stephanie for sharing her lovely lab room with me so that I could have a quiet place to write this dissertation! I am much indebted to Khalil Iskarous, who is also in my dissertation committee and chaired one of my screening projects. Khalil always has impressive ideas. He always guided me to see the broad picture and pushed me to think about greater theoretical implications. I especially iii benefited from his recent project on Ladin, which opened up an opportunity for my research on this language. During my study at USC, there are many other faculty members that gave me enormous help and support. I want to thank Louis Goldstein for his courses in phonetics and Articulatory Phonology, which made me exposed to a new exiting field of research that I didn’t know before. Andrew Simpson, Sandra Disner, Toben Mintz, Barry Schein, and Mary Washburn are both great teachers and great colleagues that I have worked with as a Teaching Assistant. Also, I am grateful to Karen Jesney, Brian Smith, and Sam Zukoff, for their advice and discussion about my research projects at various stages of my study. I would also like to thank Guillermo Ruiz, the department manager, for his generous help with all kinds of administrative stuff. The case studies in this dissertation cannot be finished without the help of many people outside of USC. I am deeply grateful to my consultants of Huozhou Chinese, without whom I could not collect data for my research or verify the fascinating patterns that were documented decades ago. I especially thank Mr. Jianguo Zhang and Ms. Jingyu Jia, who became my liaisons in Huozhou and introduced consultants to me. For the case study of Rapa Nui, I would like to thank Dr. Paulus Kieviet for generously sharing his corpus of Rapa Nui, as well as his insights on this language. Studying for a doctorate degree is a long journey, but my great peers at USC and my friends in Los Angeles made this journey less tedious. I thank my wonderful cohort, Betul Erbasi, Sarah Harper, Tanner Sorensen, Jesse Storbeck, as well as other classmates and colleagues in the Department of Linguistics. With Betul, I explored lots of Turkish food in LA; with Tommy and Yijing, I felt less lonely when studying in our “little dark room”. I still clearly remember when Haley Wei Wei took me to Santa Monica for the first time during my hectic first semester back in 2015. I also had good times with Yubin, Muxuan, and Ariela, who came to USC later, among many others in this department. My college friends and roommates in China were always available when I got stuck with reading or writing at night. Many, many thanks are due to Shanyi Wang, Jiang Wang, and Qisong iv Nie for their emotional support from a different time zone. Those conversations during many sleepless nights definitely made me mentally strong in this tough journey. I must also thank my undergraduate and graduate professors in China, especially Caimei Yang, who taught me Introduction to Linguistics in my junior year, and Jisheng Zhang, who introduced phonology to me. My career of linguistics would never start without them. Finally, I want to thank my parents for their unconditional support over the years. I cannot finish my study without their understanding. v Table of Contents Acknowledgements .............................................................................................................. ii List of Figures ...................................................................................................................... ix Abstract ................................................................................................................................ x Chapter 1 Introduction ........................................................................................................ 1 1.1. Reduplicative opacity and backcopying ................................................................................ 5 1.2. Theories of reduplication in the literature ........................................................................... 8 1.2.1. Theories of reduplication and the role of correspondence ....................................... 8 1.2.2. Reduplication as a type of nonconcatenative morphology ..................................... 15 1.3. Desiderata: a model of reduplication with Generalized Surface Correspondence ....... 19 1.4. Overview of the dissertation ................................................................................................. 22 Chapter 2 Theory ................................................................................................................ 25 2.1. Reduplication as prosodic node affixation .......................................................................... 26 2.2. Generalized Surface Correspondence .................................................................................. 30 2.2.1. The correspondence-requiring constraint CORR ....................................................... 33 2.2.2. Limiter constraints ......................................................................................................... 35 2.2.2.1. The identity-driving constraint IDENT-XX [𝛾 G] (F) .................................................... 35 2.2.2.2. ACCORD-XX .................................................................................................................... 39 2.2.3. Summary .......................................................................................................................... 43 2.3. The effects of Generalized Surface Correspondence in reduplication ........................... 44 2.3.1. Dominated MAX-IO-μ ...................................................................................................... 46 2.3.2. Dominated MAX-IO-seg .................................................................................................. 48 2.4. Summary ................................................................................................................................... 51 Chapter 3 Diminutive formation in Huozhou Chinese ....................................................... 54 3.1. Diminutive formation in Huozhou Chinese ........................................................................ 57 3.1.1. Data sources ..................................................................................................................... 57 vi 3.1.2. Aspects of Huozhou phonology .................................................................................... 57 3.1.3. Diminutive formation in West Huozhou .................................................................... 60 3.1.3.1. ŋ-ending and j-ending nouns ................................................................................... 61 3.1.3.2. w-ending nouns .......................................................................................................... 64 3.1.3.3. Nouns without ending and compounds ................................................................. 65 3.1.4. Summary of patterns ..................................................................................................... 66 3.2. Theoretical preliminaries ...................................................................................................... 68 3.3. Reduplication in Huozhou: mora affixation, backcopying, and vowel raising ............. 73 3.3.1. Reduplication as mora affixation ................................................................................. 74 3.3.2. Backcopying driven by SCorr ....................................................................................... 79 3.3.3. Issues of vowel raising ................................................................................................... 84 3.3.4. Summary .......................................................................................................................... 93 3.3.4.1. Summary of rankings ................................................................................................ 93 3.3.4.2. Limiting SCorr to reduplication ............................................................................... 96 3.4. Subtraction as an alternative to reduplication .................................................................. 98 3.4.1. Subtraction as mora affixation ..................................................................................... 98 3.4.2. Vowel raising in rime change ..................................................................................... 103 3.4.3. Summary ........................................................................................................................ 108 3.5. Interim summary .................................................................................................................. 108 3.6. Alternative approaches ........................................................................................................ 112 3.6.1. Alternative analyses of backcopying ......................................................................... 113 3.6.1.1. Base-Reduplicant Correspondence ....................................................................... 113 3.6.1.2. Stratal OT without SCorr ......................................................................................... 115 3.6.2. Alternative analyses of subtraction ........................................................................... 117 3.6.2.1. Prosodically Defective Morphemes ....................................................................... 118 3.6.2.2. Indexed Constraints and realization-based models ........................................... 121 3.6.3. Cophonologies and Morphological Doubling Theory ............................................. 124 3.6.3.1. Cophonology Theory and nonconcatenative allomorphy ................................ 124 3.6.3.2. Morphological Doubling Theory and backcopying ............................................ 126 3.6.4. Summary ........................................................................................................................ 129 3.7. Excursus: a typology of diminutive formation in nearby dialects ................................ 130 3.8. Summary ................................................................................................................................. 136 Appendix: Backness (dis)agreement across syllable boundaries ............................................... 137 Chapter 4 Reduplication in Rapa Nui ............................................................................... 141 4.1. Data sources ........................................................................................................................... 143 4.2. Language background .......................................................................................................... 143 vii 4.2.1. Metrical structure and stress patterns ..................................................................... 144 4.2.2. Deriving the metrical structure ................................................................................. 147 4.2.3. Issues about degenerate feet ...................................................................................... 149 4.3. Intensifying Reduplication in Rapa Nui ............................................................................. 151 4.3.1. Data ................................................................................................................................. 151 4.3.2. Template for intensifying reduplication .................................................................. 156 4.3.3. Vowel length alternation in intensifying reduplication ........................................ 159 4.3.3.1. Vowel Shortening driven by Surface Correspondence ...................................... 159 4.3.3.2. Vowel lengthening in derived environment ....................................................... 164 4.3.3.3. Summary of ranking ................................................................................................ 166 4.3.4. Alternative analyses without correspondence ........................................................ 168 4.3.4.1. Minimal Reduplication without SCorr .................................................................. 169 4.3.4.2. Serial Template Satisfaction ................................................................................... 170 4.3.5. Summary ........................................................................................................................ 177 4.4. Plural formation in Rapa Nui ............................................................................................... 178 4.4.1. Data ................................................................................................................................. 179 4.4.2. Analysis .......................................................................................................................... 183 4.5. Reduplication in Rapa Nui: an interim summary ............................................................. 187 4.6. Alternative analysis with BRCT ........................................................................................... 188 4.7. Summary ................................................................................................................................. 192 Chapter 5 Generalized Surface Correspondence revisited .............................................. 193 5.1. Similarity threshold and the role of CORR-XX ................................................................... 194 5.2. Local versus global evaluation of GSC constraints ........................................................... 197 Chapter 6 Typological survey and general discussion ..................................................... 201 6.1. Cases of truncatory backcopying ........................................................................................ 202 6.1.1. Tonkawa ......................................................................................................................... 203 6.1.2. Hawaiian ........................................................................................................................ 204 6.1.3. Guarijio ........................................................................................................................... 205 6.1.4. Huozhou ......................................................................................................................... 205 6.1.5. Summary of patterns ................................................................................................... 206 6.2. Generalized Surface Correspondence and truncatory backcopying ............................ 207 6.2.1. Mora deletion ................................................................................................................ 208 6.2.2. Segment deletion .......................................................................................................... 210 6.3. GSC in reduplication-related phenomena ......................................................................... 214 6.3.1. Previous proposals of correspondence-approach to copy epenthesis ................ 215 viii 6.3.2. Extending GSC to copy epenthesis: a preliminary discussion ............................... 217 Appendix: Augmentative backcopying ........................................................................................... 219 Chapter 7 Conclusion ....................................................................................................... 224 7.1. Summary of the dissertation ............................................................................................... 224 7.2. Implications ........................................................................................................................... 227 7.3. Future Directions ................................................................................................................... 231 7.3.1. Dissimilation in reduplication .................................................................................... 231 7.3.2. Issues about typology and typological predictions ................................................. 234 References ........................................................................................................................ 236 ix List of Figures Figure 1. Spectrograms of [pʰɑŋ 35 ] ‘plate’ and its diminutive forms ............................................... 63 x Abstract This dissertation focuses on the role of Surface Correspondence in handling backcopying, a type of reduplicative opacity. Backcopying is a phenomenon where the base undergoes unexpected phonological change in order to conform to the features or prosodic shape of the reduplicant. The analysis and interpretation of backcopying have been at the heart of the debate that whether or not correspondence is needed in handling reduplication-phonology interactions. This study argues for a correspondence-based approach to reduplication, drawing evidence from the truncatory backcopying patterns in new or recently-reported data from Huozhou Chinese and Rapa Nui. The ultimate goal of this study is to argue for a more economical grammar by demonstrating that the complicated patterns in reduplication can be readily handled with more general rather than specialized theoretical mechanisms. Motivated by the language data, the proposal of this study has two components, a model of Minimal Reduplication with Generalized Surface Correspondence (MR + GSC). These two components correspond to two issues central to the theories of reduplication: (i) the underlying phonological form that initiates reduplication, and (ii) the grammatical mechanism that predicts truncatory backcopying. The case studies of Huozhou Chinese diminutive formation and Rapa Nui reduplication aim to provide evidence for the utility of prosodic node affixation in handling nonconcatenative morphology and the importance of Surface Correspondence in reduplicative opacity, especially truncatory backcopying. The current study aims to make three major contributions to both language data and theory. First, this study argues for a correspondence-based approach with new and recently-reported xi language data, contributing to the debate on whether or not syntagmatic correspondence is needed in a theory of reduplication. Second, this study argues for the role of Surface Correspondence (SCorr) in reduplication and extends its utility to the domain of morphological reduplication, which has not been explored in detail in previous literature. Third, the data presented in this research also involve nonconcatenative allomorphy, which demonstrates the benefit of prosodic node affixation in nonconcatenative morphology. 1 Chapter 1 Introduction This dissertation is about the role of existing, independently-motivated theoretical mechanisms in reduplication, focusing on a phenomenon that is termed as truncatory backcopying in morphological reduplication. The general goal of this research is to argue for a more parsimonious architecture of grammar by demonstrating that the puzzling patterns in reduplication can be readily handled with more general rather than specialized theoretical mechanisms. The key proposal is to integrate the principles of Minimal Reduplication (Saba Kirchner 2010) and Generalized Surface Correspondence (after Walker 2000a, b, 2001, Hansson 2001/2010, Rose & Walker 2004, Bennett 2013, 2015a, b, Inkelas & Shih 2014, Shih & Inkelas 2019 a.o.), two theories that have been proposed for seemingly unrelated phenomena. Reduplication is a phenomenon where a surface form exhibits repetition of a portion of the segments. The copied or repeated portion is referred to as the Reduplicant, while the part that provides materials for copying is referred to as the Base, such as [pa] R -[pa.ta] B (where the reduplicant is underlined). As for truncatory backcopying, it refers to a case where the base is truncated by segment deletion or shortening to match the shape of the reduplicant (see §1.1 for details). The examples from Huozhou Chinese diminutive reduplication serve as an illustration of the pattern in question. The data are from the author’s fieldwork. The superscript digits indicate tones. 2 (1) Truncatory backcopying in Huozhou diminutive reduplication Noun Reduplicated Gloss pɑw 11 poː 11 .po 33 ‘bag, purse (DIM)’ tɕʰjɑw 35 tɕʰɥoː 35 .tɕʰɥo 55 ‘stick (DIM)’ tʰow 53 tʰuː 53 .tʰu 11 ‘bean (DIM)’ ljow 35 lyː 35 .ly 55 ‘glass ball (DIM)’ The puzzle of the data above is what drives the base (e.g., [pɑw 11 ]) to lose the final glide after reduplication (e.g., [poː 11 .po 33 ]). The reduplicant in Huozhou diminutive reduplication always ends with a vowel, and it will be argued that the shape of the base is influenced by that of the reduplicant, resulting in segment deletion. The driving force of truncation in such a case is at the heart of the debate on whether or not certain reduplication-specific mechanisms are necessary for a parallel grammar. In this study, the central hypothesis is that both reduplication and truncatory backcopying, such as Houzhou diminutive formation, are emergent and epiphenomenal. Further, a Surface Correspondence relation is active in reduplication and it enforces the base and the reduplicant to match in both segmental contents and shape. Meanwhile, an alternative derivational analysis, where segment deletion has to take place before reduplication, is rejected. Reduplication has contributed to many phonological and morphological theories, especially Prosodic Morphology (McCarthy & Prince 1986/1996, 1993) and Correspondence Theory (McCarthy & Prince 1995), both of which are seminal in the development of Optimality Theory (Prince & Smolensky 1993/2004). In many studies of reduplication, a distinction is drawn between morphological reduplication and phonological reduplication (e.g., Kawu 2000, Karabay 2004, Walker & Feng 2004, Inkelas & Zoll 2005, Yu 2005, Feng 2006, Inkelas 2008, 2014). Morphological reduplication is a word-formation process that uses reduplication to express meaning. On the other hand, copying of existing segments is a repair strategy to resolve certain markedness conditions in phonological reduplication. The current study focuses on 3 morphological reduplication, though phonological reduplication will be also briefly discussed in Chapter 6. In the theories of reduplication, two issues that are closely related need to be addressed, namely (i) the underlying phonological form that initiates reduplication, and (ii) the grammatical mechanism that predicts patterns such as truncatory backcopying. In this study, I argue that morphological reduplication and truncatory backcopying, a form of reduplicative opacity, are emergent and epiphenomenal. Truncatory backcopying in reduplication can be handled with prosodic node affixation (McCarthy 1981, Saba Kirchner 2010, 2013, see also Bermúdez-Otero 2012, Bye & Svenonius 2012), which follows the basic principles of Minimal Reduplication (MR, Saba Kirchner 2010, 2013), and Generalized Surface Correspondence (GSC, after Walker 2000a, b, 2001, Hansson 2001/2010, Rose & Walker 2004). Both MR and GSC are independently motivated theories. The current study argues that these two components, MR and GSC, are compatible and can be integrated to solve certain patterns in reduplication that are identified in understudied languages. An illustration of the MR + GSC model is previewed in (2), which will be further explained later in §1.3 and Chapter 2. (2) Minimal Reduplication with Generalized Surface Correspondence (MR + GSC) In (2), reduplication is treated as a repair strategy for an empty prosodic node. Meanwhile, there is correspondence relation between surface segments (SCorr) established to enforce the identity. In the following discussion, I will use “Generalized Surface Correspondence” or “GSC” as the p a t σ + σ Input: Prosodic node a!xation Surface Correspondence p x a y t z σ σ p x a y Output: Reduplication 4 name of the theory, while the term “Surface Correspondence” or “SCorr” is used to indicate the relation between surface segments that is assigned by GEN. Although SCorr relation was first proposed to handle long-distance assimilation/dissimilation as well as local interactions (Walker 2000a, b, 2001, Hansson 2001/2010, Rose & Walker 2004, Rhodes 2012, Bennett 2013, 2015, Inkelas & Shih 2014, Shih & Inkelas 2019 a.o.), I argue that it can be and should be extended to the treatment of reduplicative patterns. The current study aims to make three major contributions to both language data and theory. First, this study argues for a correspondence-based approach with new and recently-reported language data, contributing to the debate on whether or not surface-to-surface correspondence is needed in a theory of reduplication. Second, this study argues for the role of Surface Correspondence (SCorr) in reduplication and extends its utility to the domain of morphological reduplication, which has not been discussed in depth in previous literature to my knowledge (see Yu 2003, 2015, Inkelas 2008 for some relevant discussion). Third, the data presented in this research involve nonconcatenative allomorphy, which demonstrates the benefit of prosodic node affixation in nonconcatenative morphology. In general, the central goal of the research is to show how puzzling patterns in reduplication can be analyzed with minimal theoretical apparatus, leading to a parsimonious architecture of grammar. For the rest of this chapter, I will detail the background information that motivates and justifies the current research. Section 1.1 introduces reduplicative opacity and the puzzling patterns that are central to the theories of reduplication. Section 1.2 reviews the theories of reduplication and discusses the logic and rationale behind the current model. Section 1.3 summarizes the literature review and provides the desiderata of a model that can solve the conflicting issues raised in the literature. Finally, Section 1.4 provides an overview of the rest of the chapters of the dissertation. 5 1.1. Reduplicative opacity and backcopying Opacity refers to situations in which a phonological alternation fails to apply in the expected environment or exceptionally applies in an environment that does not meet the normal conditions (Kiparsky 1973). In terms of reduplicative opacity, a phonological process can be said to over- or underapply in either the base or the reduplicant (Wilbur 1973, McCarthy & Prince 1995, 1999 a.o.). This terminology, however, is carried over from rule-based analyses to a constraint-based framework. In the following discussion, I follow the way in which McCarthy and Prince (1995, 1999) characterize reduplicative opacity in a strictly descriptive and theory- neutral sense. Thus, overapplication refers to a situation in which a phonological mapping introduces a disparity between the reduplicative output and the lexical stem that is not expected merely on phonological grounds (McCarthy & Prince 1999:233). Backcopying is a special case of overapplication (or underapplication) that is controversial in the literature. Whether a case is interpreted as backcopying or not often depends on the posited underlying representation and the analytical mechanisms involved. This term tends to be specifically associated with the mechanisms of Base Reduplicant Correspondence Theory (BRCT, McCarthy & Prince 1995, 1999). A classic example from Tagalog illustrates this point. If an underlying form is posited to be /paN-RED-puːtul/, where N represents a placeless nasal, the output form [pamumuːtul] is said to exhibit backcopying (McCarthy & Prince 1999:252). In this example, the RED copies the onset /p/ of the following syllable, which is assimilated to and fused with its preceding nasal. The assimilation and fusion of [Np] as [m] in the output is copied back to the base, producing [pamumuːtul] rather than *[pamupuːtul], which is argued to be a case of backcopying (cf. Inkelas & Zoll 2005, §6.1). If the same surface pattern is instead interpreted as nasal place assimilation followed by deletion and infixation reduplication (i.e., /paN-puːtul/ → [pampuːtul] → [pamuːtul] → [pa-mu-muːtul]), it is not backcopying overapplication anymore. Therefore, in previous literature, whether or not a case is “backcopying” depends on the theoretical mechanisms that are adopted in the analysis. 6 In the current study, I provide a descriptive and theory-neutral definition of overapplication with a backcopying character as follows (based on the description of overapplication in McCarthy & Prince (1999:233)). (3) Overapplication with a backcopying character a. Overapplication refers to a situation where a phonological mapping introduces a disparity between the reduplicative output and the lexical stem that is not expected merely on phonological grounds. b. Overapplication with a backcopying character is a case where the disparity is located in the base part and the disparity in the base conforms to the reduplicant. The prose in (3) does not commit to any specific theoretical models. Thus, backcopying is purely characterized by the observed surface forms. For the example of Tagalog discussed above, no matter what underlying form is posited, the surface form [pa-mu-muːtul] abides by the definition of backcopying given in (3). In this example, the base part is [-muːtul] on the surface and the original [p] in the input lexical stem [-puːtul] becomes [m] to conform to the reduplicant, which has no phonological grounds. In the following discussion, I will simply use backcopying to indicate backcopying overapplication, and the cases that exhibit underapplication with a backcopying character (or backcopying underapplication) will be specified where relevant. 1 Most cases in the literature that fit the current definition of backcopying involve featural changes in the base part. In other words, the phonological process that causes a disparity usually targets certain features, such as [nasal] in the Tagalog example above. However, some other cases involve the modification of the prosodic shape in the base part. When the modification is truncatory, I call the pattern truncatory backcopying. In truncatory backcopying, segments or 1 There have been other definitions of backcopying in BRCT literature. Kiparsky (2010:3) defines backcopying as “overapplication in the reduplicant of a process triggered by the reduplicant in the base”. This definition limits backcopying to a smaller subset than the current definition in this study. Under this definition, the example of Malay nasal harmony is considered as backcopying: /RED + waŋɪ/ → [w̃ãŋɪ̃ -w̃ãŋɪ̃ ], since it is the reduplicant that causes nasal spreading in the base and the nasalized [w̃ã] in the base is copied back to the reduplicant. However, many other studies define “backcopying” in a less restrictive way, which is similar to the definition in (1). For example, Inkelas (2014:136) defines backcopying as a situation where “base conforms to red rather than the reverse”. Similarly, McCarthy, Kimper & Mullin (2012:209) describe “backcopying” as “a phonological process whose structural description is met in the reduplicant overapplies to the base”. 7 moras are deleted in the base to conform to the shape of the reduplicant. This pattern is illustrated with Tonkawa (Hoijer 1933, cited and adapted from Gouskova 2007:382) and Guarijio (Miller 1996, cited from Caballero 2006:278) in (4) and (5) respectively. (4) Truncatory backcopying in Tonkawa: vowel shortening naː.toʔs ‘I step on it’ na-na.toʔs ‘I step on it/REPETITIVE’ maː.koʔs ‘I weep’ ma-ma.ka.noʔs ‘I weep/REPETITIVE’ jaː.tsoʔs ‘I see him’ he-ja-ja.tse.woʔ ‘I see him/REPETITIVE’ (several looks at it) (5) Truncatory backcopying in Guarijio: segment deletion toní ‘to boil’ to-tó ‘to start boiling’ sibá ‘to scratch’ si-sí ‘to start scratching’ kusú ‘to sing (animals)’ ku-kú ‘to start singing’ suhku ‘to scratch body’ su-sú ‘to start scratching the body’ In Tonkawa, the shape of the reduplicant is a light syllable (CV). The long vowel in the base is shortened due to the influence of the reduplicative template (e.g., [na-na.toʔs], not *[naː- na.toʔs]). In Guarijio, the base part undergoes truncation so as to match the shape of the reduplicant, a light syllable template (e.g., [to-tó], not *[to-toní]). These patterns in Tonkawa and Guarijio abide by the definition of backcopying in (3), since the disparity in the base, either vowel shortening (i.e., mora deletion) or segment deletion, is not explained solely on phonological grounds in the reduplicative context. The base itself in each case is modified to conform to the reduplicant (i.e., vowel length or syllable structure). A relevant term in the BRCT literature is “templatic backcopying”, which is not adopted in this study. The issue of templatic backcopying and the Kager-Hamilton Conundrum (McCarthy & Prince 1999) is taken up in Chapter 6. The major concern of the current study is “truncatory backcopying” patterns. The implications of such pattern for theories of reduplication will be discussed in the following section, §1.2. 8 1.2. Theories of reduplication in the literature Reduplication has been on the frontier of the development of phonological and morphological theories. There are two issues at the heart of the theories of morphological reduplication. The first issue concerns representation, namely, what underlying phonological form initiates reduplication. Primary proposals include a RED morpheme, as in BRCT (McCarthy & Prince 1995, 1999), or empty prosodic templates (Saba Kirchner 2010, 2013, Bermúdez-Otero 2012, Bye & Svenonius 2012, McCarthy, Kimper & Mullin 2012, Zimmermann 2013, 2015, 2017b, Paschen 2018, a.o.). Second, theories of reduplication need to address what mechanisms enforce identity between the base and reduplicant, which can lead to reduplicative opacity, especially backcopying. In previous proposals, reduplicative opacity has been attributed either to surface- to-surface correspondence (e.g., BR correspondence, McCarthy & Prince 1995) or derivation (e.g., Level Ordering, Kiparsky 2010). The issues of representation and grammatical apparatus are closely connected and they are reflected in two lines of theories of reduplication. One of the approaches is represented by BRCT, in which RED and BR correspondence are essential elements, while another approach eliminates BR correspondence and views reduplication as prosodic node affixation. In what follows, I will review and discuss the theories of reduplication and the role of correspondence in §1.2.1. Next, in §1.2.2, I will examine the position of reduplication in morphology. Since morphological reduplication is a type of nonconcatenative morphology, it is worth discussing how the representation and grammar of reduplication could inform morphological theories in a broader theoretical context. This section serves as a background review of the current theories of reduplication and the problems raised by new and recently-reported data. 1.2.1. Theories of reduplication and the role of correspondence In this section I review issues surrounding the ongoing debate about whether or not we need BR correspondence and a RED morpheme that initiates such a relation. Identity-preserving interactions between reduplication and phonology were first identified by Wilbur (1973). The 9 most widely accepted theory in mainstream OT is BRCT in McCarthy & Prince (1995, 1999), where the notion of correspondence was proposed. The full model of BRCT is illustrated in (6). (6) Full model of BRCT (McCarthy & Prince 1995:4) In BRCT, a phonologically empty morpheme, RED, triggers reduplication and copies phonological material from the BASE. A correspondence relation exists between RED and BASE, namely, BR correspondence. Thus, when RED meets a structural description and undergoes a phonological change, an identity effect can be imposed back on the BASE through BR-identity. Backcopying, as a special case of reduplicative opacity, forms the strongest argument for BR correspondence and symmetric BR identity. A number of cases in the literature comply with the definition of backcopying, most of which served as support for BRCT. A familiar example of backcopying is Malay nasal spreading. Data from Malay are shown in (7), followed by the analysis with BR correspondence in (8) (Onn 1976, cited from McCarthy & Prince 1995:42). (7) Malay nasal spreading in reduplication hamə̃ hãmə̃-hãmə̃ ‘germ/germs’ waŋĩ w̃ãŋĩ-w̃ãŋĩ ‘fragrant/(intensified)’ aŋãn ãŋãn-ãŋãn ‘reverie/ambition’ aŋẽn ãŋẽn-ãŋẽn ‘wind/unconfirmed news’ (8) Malay nasal spreading: analysis with BRCT /RED + waŋi/ IDENT-BR(nas) *NV oral *Ṽ IDENT-IO(nas) ☞ a. w̃ãŋĩ-w̃ãŋĩ 6 3 b. waŋĩ-waŋĩ 1W 2L 1L c. waŋĩ-w̃ãŋĩ 2W 4L 3 Input Output / RED + X / [RED + BASE] IO-Faith BR-Correspondence IR-Faith 10 Malay has a pattern of rightward nasal spreading, where vowels, glides, and laryngeal consonants after a nasal undergo nasalization (e.g., [waŋĩ]). However, in a reduplicated form, relevant segments to the left of a nasal are also nasalized, such as [w̃ãŋĩ-w̃ãŋĩ] (*[waŋĩ-w̃ãŋĩ]), as shown in (7). This is interpreted as a case of backcopying, namely, the [wa] sequence in the base is nasalized due to the influence of the velar nasal in the reduplicant and the nasalized [w̃ã] sequence is transmitted back to the reduplicant through BR identity. This analysis is captured in (8), where the candidate in (8a) is selected as the winner, although it incurs severe violations of *Ṽ that penalizes nasalized vocoids (vowels or glides). In (8b), the sequence [wa] in the base fails to be nasalized by the preceding [ŋ], which his ruled out by *NV oral . In (8c), the nasalized [w̃ã] sequence in the base is not copied back to their counterparts in the reduplicant, fatally violating IDENT-BR(nas). To keep the discussion focused, I assume the reduplicant is prefixed (see McCarthy & Prince 1995:43–46 for additional discussion). The core insight of this dataset is that it is better analyzed by BR correspondence, while it poses a challenge to derivational theories (cf. Raimy 2000). If this pattern is attributed to two rules, a copying rule and a nasal spreading rule, either order cannot generate the expected pattern, as demonstrated in (9) (after McCarthy & Prince 1995:43). (9) Malay nasal spreading: analysis with rule-ordering a. UR /RED – waŋi/ b. UR /RED – waŋi/ Copying waŋi-waŋi Nasal spreading RED – waŋĩ Nasal spreading waŋĩ-w̃ãŋĩ Copying waŋĩ-waŋĩ SR *[waŋĩ-w̃ãŋĩ] SR *[waŋĩ-waŋĩ] In (9a), when copying precedes nasal spreading, the surface form exhibits normal application of nasalization. In comparison, when nasal spreading comes first, as in (9b), it results in underapplication since the vocoids in the base fail to be nasalized though the context is met. Thus, this pattern is considered to be the most powerful argument for BR correspondence. 11 However, some follow-up studies on reduplication cast doubt on the so-called backcopying patterns, and resultingly, the necessity for BRCT. Some reported cases of backcopying were found to be spurious or unreliable, including the pattern of Malay nasal spreading and Southern Paiute (Gurevich 2000). For Malay nasal spreading, overapplication of nasalization at the base- reduplicant juncture was not verified by native speakers (Kiparsky 2010, see also Inkelas and Zoll 2005:221). For Southern Paiute, McCarthy & Prince (1995:101–103) interpreted the data as a case of backcopying underapplication of [w] ~ [ŋʷ] alternation. However, the data were re-examined in Gurevich (2000) and it turned out that the data were misinterpreted in McCarthy & Prince (1995). Meanwhile, apart from the data issues, almost all the other cases of backcopying are viewed as epiphenomenal and can be reanalyzed without resorting to BR identity (Zoll 2002, Inkelas and Zoll 2005, Paschen 2018; see also Raimy 2000, McCarthy, Kimper & Mullin 2012). These two aspects of evidence undermine the role of BR correspondence and BR identity. Therefore, in the past two decades, other theories of reduplication have been proposed to replace BRCT, including Morphological Doubling Theory (Inkelas and Zoll 2005), Minimal Reduplication (Saba Kirchner 2010; see also Generalized Nonlinear Affixation, Bermúdez-Otero 2012), and Serial Templatic Satisfaction (McCarthy, Kimper & Mullin 2012). In (10) below, I give a (non-exhaustive) list of backcopying cases that have been discussed and reanalyzed. (10) Cases of backcopying in the literature Language with backcopying Analysis with BR correspondence (Re-)analysis without BR correspondence Axininca Campa McCarthy & Prince (1995) McCarthy, Kimper & Mullin (2012) Eastern Kadazan McCarthy & Prince (1995) Zoll (2002) Samala (Ineseño Chumash) McCarthy & Prince (1995) Inkelas & Zoll (2005); Saba Kirchner (2010) Chaha Banksira & Kenstowicz (1999) Inkelas & Zoll (2005) Tagalog McCarthy & Prince (1999) Inkelas & Zoll (2005) Seereer-Siin McLaughlin (2000) Paschen (2018) 12 The problems discussed above inspired another line of research that eliminates RED and BR correspondence in reduplication. Two significant proposals following this line include Saba Kirchner’s (2010) Minimal Reduplication and Serial Template Satisfaction (McCarthy, Kimper & Mullin 2012). 2 In both theories, morphological reduplication is initiated by an empty prosodic template in the input to phonology and results from segment fission. A schematic illustration is /σ + p 1 a 2 t 3 u 4 / → [p 1 a 2 -p 1 a 2 t 3 u 4 ], where the digits represent input-output mapping. Additional analyses in the same vein, namely, reduplication as segment fission to repair empty prosodic templates, include Bye & Svenonius (2012), Zimmermann (2013, 2015, 2017b), Paschen (2018), among others. A difference between Minimal Reduplication and Serial Template Satisfaction is that the former approach is analyzed in parallel OT while the latter is implemented in Harmonic Serialism (McCarthy 2000, 2002, 2007). In theories without BR correspondence, reduplicative opacity is often attributed to either level ordering or serial derivation in Harmonic Serialism. As for the backcopying cases that used to be analyzed with BRCT, most of them are subject to reinterpretation (e.g., Samala/Ineseño Chumash in Saba Kirchner 2010; Axininca Campa in McCarthy, Kimper & Mullin 2012). The Malay example in (7), which cannot be handled straightforwardly in a derivational approach, is suggested to be unreliable (Kiparsky 2010), and therefore does not pose a problem for Minimal Reduplication or Serial Template Satisfaction. Further, the questioning of BR correspondence even led to proposals that abandon the notion of correspondence in OT and turn to a containment-based theory of faithfulness (e.g., van Oostendorp 2004, 2006). Nevertheless, eliminating BR correspondence is not without problems. The debate on the role of correspondence in reduplication is far from settled. Some recent studies that argue in favor of a parallel approach with correspondence include Zukoff (2017), Wei & Walker (2020), and Yang (2022) (cf. Paschen 2018, 2021, Lamont 2021). Several points suggest that the role of surface-to-surface correspondence in reduplication is still important. First, some recently- 2 Another relevant approach to reduplication is Morphological Doubling Theory (Inkelas & Zoll 2005). This approach will be discussed in later chapters in the context of the case studies. 13 reported data call for a correspondence-based approach to handle opacity in reduplication. In Rapa Nui, for example, the base long vowel is shortened in concert with reduplication to match the length of its counterpart in the reduplicant, as shown in (11) (examples from Kieviet 2017:64, 94). (11) Truncatory backcopying (vowel shortening) in Rapa Nui reduplication vaːnaŋa vana-vanaŋa ‘to talk’ → ‘to chat’ hoːɾou hoɾo-hoɾou ‘hurry’ → ‘hurry very much’ In this type of reduplication, the shape of the reduplicant always amounts to two moras. It is straightforward to account for the vowel shortening pattern by positing a correspondence relation between the vowels. The base vowel is then shortened due to the enforcement of an IDENT constraint on vowel length. More details can be found in Chapter 4 and Yang (2022). Second, the correspondence-based approach is well supported in related phenomena besides reduplication. There have been proposals for the activity of correspondence in other domains such as aggressive reduplication (Zuraw 2002) and copy epenthesis (Kitto & de Lacy 1999, Stanton & Zukoff 2018, a.o.). In Zuraw’s (2002) work on aggressive reduplication, a correspondence-like relation, COUPLING, is required to exist between any two substrings of a word. This can be used to account for some over-/underapplication patterns in aggressive reduplication in Tagalog. See (12) for examples (Zuraw 2002:400). (12) Aggressive reduplication in Tagalog: intervocalic tapping a. transparent díːɾi ‘loathing’ b. overapplied ɾúːɾok ‘acme’ c. underapplied deːde ‘baby bottle’ The phonological process, intervocalic tapping ([ɾ]/V_V), is applied in an unexpected environment in (12b), but it fails to apply in (12c), where the context is met. Zuraw (2002) proposed that there is a correspondence-like relation between two portions of words that causes misapplication of intervocalic tapping through IDENT, such as [ɾúː]𝛼[ɾok]𝛼 and [deː]𝛼[de]𝛼 (Greek 14 letters indicate correspondence between strings). Besides aggressive reduplication, another domain that suggests the utility of correspondence is copy epenthesis, a phenomenon where the quality of an epenthetic segment depends on a nearby segment, sometimes viewed as a copy of the nearby segment. Stanton & Zukoff (2018) identified a case of backcopying in copy epenthesis in Ho-Chunk, which calls for a correspondence-based approach. For more details, see Stanton & Zukoff (2018:665–668). Finally, in Minimal Reduplication, although this theory does not employ a RED morpheme and BR correspondence, the analysis of Samala still needs to resort to a certain type of correspondence on the surface. Samala (previously known as Ineseño Chumash) overcopies the prefix in reduplicated forms. For instance, for the input /σ μμ , s-{ikmen}/, the output form is [siksikmen] instead of *[sikikmen], where the prefix [s] is overcopied (illustrated in bold). This augmentation process is motivated by the alignment constraint ALIGN-L(PStem-σ) that requires the left edge of a PStem (i.e., [-ikmen] in this case) to be aligned with the left-edge of a syllable (i.e., there must be an onset). However, another candidate that needs additional treatment is *[sikmikmen]. In this candidate, [m] rather than the prefix [s] is copied to make sure that the PStem aligns with the left edge of a syllable. To rule out this candidate, Saba Kirchner (2010:150) extended Zuraw’s (2002) REDUP constraint to the analysis of Samala. The definition of REDUP constraints is given in (13), followed by a tableau to demonstrate the effect of REDUP in (14). 3 Note that the brackets in the candidates of (14) indicate syllables in correspondence. (13) Constraints for Samala backcopying a. REDUP(σ): For some syllable in the output there is a distinct syllable in the output that stands in correspondence with it. (Saba Kirchner 2010:150) b. IDENT-𝜅𝜅: If a word contains two substrings S 1 and S 2 that stand in correspondence, and if a segment in S 1 corresponds to a segment in S 2 , the two segments must be identical. (after Zuraw 2002:404) 3 Saba Kirchner (2010:150-151) did not present the analysis in great detail. The tableau in (14) elaborates the idea therein with additional constraints to demonstrate the utility of REDUP. 15 (14) Input /σ μμ , s-{ikmen}/ (after Saba Kirchner 2010:151) /σ μμ , s-{ikmen}/ INTEGRITY REDUP(σ) IDENT-𝜅𝜅 ☞ a. [sik][sik]men 3 b. [sik][mik]men 3 1W c. sik[mik][men] 3 2W d. sikmikmen 3 1W The candidate *[sikmikmen] in (14d) fatally violates REDUP(σ) due to the lack of correspondence between some syllables. When [m] is copied to fulfill the role of onset, as in (14b) and (14c), the candidates will incur violations of IDENT-𝜅𝜅, leaving (14a) the winner. This analysis suggests that correspondence or an equivalent mechanism still plays a role in analyzing reduplication, though it is not necessarily BR correspondence. In sum, whether or not correspondence plays a role in reduplication is among the central issues in the theories of reduplication since the inception of BRCT. Although the debate is ongoing and several studies favor eliminating BR correspondence, it seems that a certain type of correspondence is still supported to handle reduplicative opacity, as in the analysis of Samala in Saba Kirchner (2010). In addition, some new data, including Huozhou Chinese (Chapter 3) and Rapa Nui (Chapter 4), also lend support for a correspondence-based approach to reduplication. Finally, related phenomena that show an identity effect in domains other than reduplication, such as aggressive reduplication and copy-epenthesis, are also better handled if a correspondence relation is posited. Thus, the role of correspondence is still important in dealing with reduplication, which is a major claim of the current study. 1.2.2. Reduplication as a type of nonconcatenative morphology This section reviews theories of reduplication from a different perspective. As mentioned earlier, a core debate regarding reduplication is the role of correspondence in handling reduplicative opacity and how such a correspondence relation is initiated (i.e., a RED morpheme), which has led to different theoretical proposals. In this section, I will zoom out and discuss the relation 16 between reduplication and other types of nonconcatenative morphology. I will focus on how different theories of reduplication in the literature could contribute to the understanding of nonconcatenative morphology and morphological theories in general. Most morphological constructions in languages are clearly formed through the sequential concatenation of phonological material that is paired with morphemes, such as English plural [dɔɡ-z]. Other types of morphological construction, however, are achieved by phonological processes such as copying, deleting, feature changing, etc. In Tohono O’odham, for instance, perfective verbs are formed by deleting the last segment of their imperfective counterparts (e.g., [hiːnk] ‘bark, imperfective’ → [hiːn] ‘bark, perfective’) (Anderson 1992). Cases of morphology that exhibit rule-like operations are called nonconcatenative morphology, or process morphology. Examples include morphological reduplication, subtractive morphology, morphological lengthening, morphological mutation, among others. These two types of morphology are crucial to a debate on morphological theories. Since most morphological constructions are concatenative, they can be treated by morpheme-based models, where the phonological material for each morpheme are arranged in a linear fashion (e.g., Lieber 1980, 1992, Selkirk 1982, Kiparsky 1982, Halle & Marantz 1993, Akinlabi 1996, Wolf 2007, a.o.). However, rule-like nonconcatenative morphology has been interpreted as evidence for realization-based models (e.g., Anderson 1992, Alderete 2001, Horwood 2001, Kurisu 2001), in which nonconcatenative morphology is treated as morphologically-conditioned phonology. In realization-based models, morphemes that exhibit nonconcatenative morphology do not necessarily have any prescribed phonological exponent, and the phonological realization of such morphemes depends on the grammar that they are associated with. Extensive research in the past decades has suggested that many instances of nonconcatenative morphology can be reanalyzed as the concatenation of phonological material, especially with the development of Autosegmental Phonology (Goldsmith 1976), Moraic Theory (Hyman 1984, 1985, Hayes 1989), and Prosodic Morphology (McCarthy & Prince 1986/1996, 1993). Reduplication, in particular, can be successfully analyzed as affixation. In BRCT (McCarthy & 17 Prince 1995), for instance, reduplication has been attributed to the affixation of RED; or, in another approach, reduplication has been attributed to affixation of prosodic nodes (e.g., Saba Kirchner 2010). The Generalized Nonlinear Affixation theory of morphology and phonology (GNA, Bermúdez-Otero 2012) pushes forward the idea that nonconcatenative morphology is emergent. It seeks to reanalyze all types of nonconcatenative morphology as the affixation of nonlinear phonological elements, such as floating features, tones, and/or prosodic nodes (mora, syllable, foot, etc.). In this way, nonconcatenative patterns are just epiphenomenal to affixal phonological materials. Some phenomena that have been reanalyzed as concatenation and construed as arguments for GNA include morphological lengthening/gemination (mora affixation, e.g., Davis and Ueda 2002, 2006), reduplication (prosodic node affixation, e.g., Saba Kirchner 2010, 2013, Bye and Svenonius 2012, Zimmermann 2013, 2015, 2017b), and morphological mutation (feature affixation, e.g., Wolf 2007, Paschen 2018). Among nonconcatenative morphology phenomena, subtractive morphology poses a serious challenge to morpheme-based models. In subtractive morphology, the meaning of a morphologically complex word is expressed by deleting phonological material from its morphologically simplex counterpart. Although subtraction appears opposite to additive morphology by nature, some recent studies advocate an additive view of subtractive morphology (Trommer 2011, Bye & Svenonius 2012, Trommer and Zimmermann 2014, Zimmermann 2017a, Köhnlein 2018), which advances the case of a unified view of concatenative and nonconcatenative morphology. Returning to theories of reduplication, the classic BRCT model diverges from the unified approach to nonconcatenative morphology of GNA. Although BRCT also analyzes reduplication as concatenation, reduplication is initiated by RED, which is a formal device that is specific to reduplication only. The specialized RED morpheme makes reduplication stand out compared to the other nonconcatenative morphology. In contrast, the analysis of nonconcatenative morphology as nonlinear affixation circumvents this issue. In GNA, nonconcatenative morphological patterns, including reduplication, are all viewed as emergent. The phonological 18 materials that initiate morphological realization (floating features, tones, and/or prosodic nodes) are independently motivated, all of them being basic, general units in phonology. This suggests that the theories that view reduplication as prosodic node affixation (e.g., Minimal Reduplication) present an advantage over BRCT, if we pursue a unified and emergent view of nonconcatenative morphology. The case studies to be presented in the following chapters support the emergent view because both languages involve allomorphic variations that exhibit different nonconcatenative patterns. In the case study of Huozhou Chinese (Chapter 3), for instance, diminutives can be formed by subtraction or reduplication, exemplified as follows (the superscript digits indicate tones; the reduplicant is underlined): (15) Diminutive formation in Huozhou Chinese Non-diminutive Diminutive Gloss a. [pʰɑŋ 35 ] [pʰaː 35 ] or [pʰɑŋ 35 .pʰa 55 ] ‘plate’ b. [saj 11 ] [saː 11 ] or [saj 11 .sa 33 ] ‘sieve’ The diminutive form in Huozhou Chinese can be either segment deletion (e.g., [pʰaː 35 ]) or reduplication (e.g., [pʰɑŋ 35 .pʰa 55 ]). Since both additive and subtractive processes can be variations of the same morphological construction, a unified treatment for both patterns is a desideratum. Therefore, in this case study, I propose that the phonological form of the Huozhou diminutive morpheme is a floating mora with no segmental content, and both forms in (15) are repair strategies for the deficient mora. To summarize, in this section I have briefly reviewed theories of reduplication in the theoretical context of nonconcatenative morphology. The development of morphological theories and new empirical data (e.g., Huozhou Chinese) lead to the further question that whether the analysis of reduplication as prosodic node affixation offers advantages over an approach in BRCT. 19 1.3. Desiderata: a model of reduplication with Generalized Surface Correspondence In the review of §1.2, the theories of reduplication in the literature are concerned with representation and grammar. On the one hand, there is a debate as to whether correspondence is needed in handling reduplicative opacity. Several pieces of evidence in §1.2.1, including new and recently-reported language data and phenomena in other domains (aggressive reduplication and copy epenthesis), suggest the benefits of a correspondence-based approach. On the other hand, theories of reduplication with prosodic node affixation have made progress towards a unified and emergent view of nonconcatenative morphology, as discussed in §1.2.2. These two aspects are currently instantiated in theories that are not compatible with each other. The classic correspondence-based theory of reduplication is BRCT (McCarthy & Prince 1995, 1999), where reduplication is initiated by a specialized RED morpheme. In contrast, another line of research, which eliminates RED and BR correspondence altogether, treats reduplication as prosodic node affixation. The current study pursues a model of reduplication that features two components, namely (i) reduplication as prosodic node affixation, and (ii) Generalized Surface Correspondence. This model pursues the line of Minimal Reduplication (Saba Kirchner 2010) and Generalized Nonlinear affixation (Bermúdez-Otero 2012), treating reduplication and other types of nonconcatenative morphology as the affixation of nonlinear phonological material (in particular, prosodic nodes). In the meanwhile, I argue that Surface Correspondence (SCorr), an abstract relation between surface segments, can be extended to the domain of reduplication beyond its current use in long-distance assimilation/dissimilation of segments and tones (Walker 2000a, b, 2001, Hansson 2001/2010, Rose & Walker 2004, Bennett 2013, 2015a, b, Inkelas & Shih 2014, Shih & Inkelas 2019 a.o.). The model in this study incorporates two components, Minimal Reduplication (MR) and Generalized Surface Correspondence (GSC), which have not been integrated in earlier studies. An illustration of MR + GSC is repeated in (16) below. 20 (16) Minimal Reduplication with Generalized Surface Correspondence (MR + GSC) The illustration in (16) contains the two components of the model. First, the input contains an empty syllable template, which is treated as deficient and ill-formed. This syllable template is repaired by populating with copied segments as segment fission (Saba Kirchner 2010, 2013, Bermudez-Otero 2012, Bye & Svenonius 2012, Zimmermann 2013, 2015, 2017b, see also Struijke 2002, Karabay 2004, Yu 2005, Feng 2006, a.o.). Meanwhile, there is Surface Correspondence (SCorr) structure provided by GEN in the reduplicated form, indicated by subscript letters (segments with identical index stand in correspondence). Note that GEN can provide various SCorr structures between segments, while the one in (16) only demonstrates one possibility, [p x a y t z p x a y ]. There can be other possible candidates such as [p x a x t x p x a x ], [p x a y t x p x a y ], [p x a y t z p i a j ], etc., which are subject to further evaluation of the identity-enforcement constraints. The SCorr structure in this model is flexibly assigned between any segments, and does not result from the transitivity of input-output correspondence. More details are provided in Chapter 2 (§2.2.2.2). As a summary of the theoretical issues that have been reviewed so far, there are three desiderata of pursuing the current model of reduplication, MR + GSC. Desideratum 1: A model that unifies reduplication and other types of nonconcatenative morphology. As a type of nonconcatenative morphology, reduplication should not be handled with a specialized formal device distinct from other nonconcatenative patterns, such as subtractive morphology, morphological lengthening, morphological mutation, and so on. In p a t σ + σ Input: Prosodic node a!xation Surface Correspondence p x a y t z σ σ p x a y Output: Reduplication 21 comparison, the BRCT model, which uses RED to initiate reduplication, differentiates reduplication from the other nonconcatenative patterns. Instead, the analysis of reduplication as prosodic node affixation provides a unified view of nonconcatenative morphology since other types of nonconcatenative morphology, not only reduplication, can be also treated as emergent with affixation of general and basic nonlinear elements in phonology. Desideratum 2: A model that accounts for reduplicative opacity as seen in new and recently- reported data. As briefly mentioned earlier, this dissertation analyzes recently reported data from Rapa Nui and new data from Huozhou Chinese. Both languages involve special patterns of reduplication-phonology interaction, i.e., truncatory backcopying, which cannot be properly addressed by cyclicity or other derivational models. In contrast, a model with surface-to-surface correspondence provides a more straightforward analysis. In the meanwhile, there are similar patterns of opacity in phenomena other than reduplication. Cases of overapplication have been reported in aggressive reduplication (recall (12), Zuraw 2002) and backcopying in copy epenthesis (Ho-Chunk, Stanton & Zukoff 2018). Thus, a correspondence-based model can offer a solution for the data examined in this dissertation and potentially bring all these patterns under the same umbrella. Desideratum 3: An economical model that uses independently-motivated mechanisms. This study argues for a correspondence-based model to handle reduplicative opacity. Instead of BR correspondence that is invoked by RED and specific to the domain of reduplication, this study proposes that Surface Correspondence, which has been proposed for long-distance assimilation/dissimilation, can be readily extended to reduplication. Surface Correspondence is provided by GEN, without any prerequisites, which is subject to the evaluation of universal constraints. Since Surface Correspondence has been well supported as a component of Universal Grammar, there is no reason to exclude it from the domain of reduplication. In other words, the rationale behind the proposal is that there should be no mechanisms that would “turn off” 22 Surface Correspondence when reduplication takes place. In addition, there has already been some limited discussion regarding the role of Surface Correspondence in reduplication in Inkelas & Zoll (2005), Yu (2005), and Inkelas (2008), among others. This study thus does not propose a new theory of reduplication, but rather focuses on how existing and independently-motivated grammatical mechanisms can be extended to the domain of reduplication with minimal modifications. The in-depth case studies in this dissertation have two major goals. First, they provide evidence for a unified model of nonconcatenative morphology. Second, Surface Correspondence, which has been proposed to handle assimilation and dissimilation patterns, can be readily extended to the domain of reduplication, making it a general mechanism in grammar. 1.4. Overview of the dissertation In what follows, I provide an overview of the dissertation chapters. Chapter 2 lays out the theories of reduplication and Surface Correspondence (SCorr) that I pursue in the dissertation. For reduplication, I adopt the general proposal of Minimal Reduplication (Saba Kirchner 2010, 2013), which is in line with the theory Generalized Nonlinear Affixation (Bermúdez-Otero 2012). For Surface Correspondence, I adapt a version of SCorr into Generalized Surface Correspondence (GSC), which is characterized by a feature-blind CORR-XX constraint and feature-restricted of IDENT-XX (e.g., Walker 2015, see also Hansson 2014, Bennett & DelBusso 2018). The core theme of the chapter is to demonstrate how these independently motivated theories can be extended to handle reduplicative opacity, which leads to a more parsimonious architecture of grammar. Chapter 3 presents the first in-depth case study of Huozhou Chinese diminutive formation, based on data from my fieldwork. There are two important aspects of Huozhou diminutive formation that are relevant to the two goals of this dissertation. First, Huozhou diminutive reduplication shows truncatory backcopying. For example, the reduplicated form of [pɑw] is 23 [poː.po] (*[pɑw.po] or *[pɑw.pa]), where the last segment in the base is deleted to match the prosodic shape of the reduplicant. Further, I identify the pattern in Huozhou as “conditional truncatory backcopying” since it only takes place when nouns end with a labiovelar glide [w]. These patterns support a correspondence-based approach to reduplication, and in the proposed account it is Surface Correspondence that fulfills the role. Second, Huozhou diminutive formation can be alternatively realized as subtraction. To account for the variation, I propose that the phonological form of the diminutive marker is a mora, which can be variably realized into reduplication of subtraction. Thus, this case study also serves as an argument for a unified approach to nonconcatenative morphology. Chapter 4 presents another case study in detail, focusing on the data from Rapa Nui. Similarly, the patterns in Rapa Nui reduplication also serve as arguments for the two components in the model of this dissertation. There are two types of reduplication in Rapa Nui, intensifying reduplication and plural formation. For intensifying reduplication, there is a puzzling process of vowel shortening during reduplication. When a root is of HLL shape (‘H’ for heavy syllable; ‘L’ for light syllable), such as [vaːnaŋa], the initial long vowel is shortened in the reduplicated output, i.e., [vanavanaŋa], which is truncatory backcopying with vowel shortening. I argue that the truncatory backcopying pattern in Rapa Nui supports a correspondence-based approach. For plural formation, the plural form of verbs can be produced either through reduplication (e.g., [ʔaɾa] → [ʔaʔaɾa]) or lengthening (e.g., [haʔuɾu] → [haːʔuɾu]), depending on the shape of the singular form. These variable plural forms on the surface are attributed to the affixation of a single mora, which further demonstrates the benefits of analyzing reduplication as prosodic node affixation. Chapter 5 revisits the proposal of Generalized Surface Correspondence. Several theoretical issues that have been debated in the literature of Agreement-by-Correspondence, including the purpose of Corr constraints and local versus global evaluation of IDENT-XX and CORR-XX, are discussed in this chapter. The analyses of Rapa Nui reduplication and Huozhou diminutive reduplication in previous chapters facilitate the discussion of these theoretical issues. 24 Chapter 6 focuses on the typology of truncatory backcopying and typological predictions of GSC. This chapter offers a typological survey of truncatory backcopying, as well as augmentative backcopying, with a focus on the former type. In addition, the predictions of Generalized Surface Correspondence are discussed with regard to the observed patterns. In particular, it turns out that the conditional backcopying in Huozhou Chinese is not an isolated instance of the ranking schema that can be predicted by GSC. A similar pattern outside of the reduplication domain can be analyzed with the same logic. Finally, this chapter provides a preliminary discussion of extending GSC to other reduplication-related phenomena, especially copy epenthesis. Chapter 7 summarizes and concludes the dissertation. Some further directions are discussed in this chapter. Some notable issues about the current model, including certain unattested predictions of GSC and the issue of dissimilation in reduplication, await further research. 25 Chapter 2 Theory As discussed in Chapter 1, the main theme of the dissertation is to argue for an economical architecture of grammar by demonstrating how existing, independently-motivated grammatical mechanisms, such as prosodic node affixation and Surface Correspondence, can be readily extended to analyze reduplicative opacity in languages. The theory of reduplication that is adopted in this dissertation has two major components. First, I treat morphological reduplication as emergent, which is attributed to prosodic node affixation, following proposals in the literature by McCarthy & Prince (1986) and Saba Kirchner (2010, 2013) (see also Bermúdez- Otero 2012, Bye & Svenonius 2012, a.o.). Second, I argue for the utility of correspondence in handling reduplicative opacity. The type of correspondence in this study is dubbed Generalized Surface Correspondence, which derives from Agreement-by-Correspondence (ABC, Walker 2000a, b, 2001 Hansson 2001/2010, Rose & Walker 2004, Bennett 2013 a.o.). In the current study, I am not proposing a new theory of Surface Correspondence nor arguing for the basic architecture of ABC in detail. What I am focusing on in the study is the role of Surface Correspondence in handling reduplicative opacity and how it can be integrated into theories of reduplication. Although there have been multiple versions of ABC in the literature, I follow one of them with certain modifications and extend it to the domain of reduplication. This chapter begins with the introduction of reduplication as prosodic node affixation in Section 2.1. No specialized RED morpheme or Base-Reduplicant correspondence is assumed. The fundamentals of Generalized Surface Correspondence will be laid out in Section 2.2. In Section 26 2.3, I will present the predictions of GSC in reduplication with hypothetical examples. Section 2.4 summarizes the chapter and discusses several general issues. 2.1. Reduplication as prosodic node affixation The current study argues that reduplication is emergent. In this approach, reduplication is a repair strategy for deficient prosodic templates, which is in line with Minimal Reduplication (MR, Saba Kirchner 2010, 2013, see also Bermúdez-Otero 2012, Bye & Svenonius 2012, a.o.) and Serial Template Satisfaction (STS, McCarthy et al. 2012). This idea dates back to McCarthy and Prince (1986). In both MR and STS, morphological reduplication is viewed as the affixation of a prosodic template, which is populated by copying or fission of existing segmental material in the base. The analysis in this dissertation adopts the basic mechanisms of MR, while STS will be discussed as an alternative where relevant. In the original proposal of MR (Saba Kirchner 2010), the phonological exponents of the morphemes that are realized as reduplication on the surface are treated as empty prosodic templates. The prosodic template in the input to phonology is fulfilled through fission of segments in the base of affixation, which gives rise to reduplication. The process is usually driven by the markedness constraints that penalize floating and unheaded prosodic units, including MAXFLT (Wolf 2007) and HEADEDNESS(X) (Selkirk 1995). Different from BRCT, (McCarthy & Prince 1995, 1999), which is the classic theory of reduplication in mainstream OT, MR does not assume an underlying RED morpheme nor BR correspondence. The main advantage of MR, as defended by Saba Kirchner (2010), is that MR does not require any reduplication-specific formal device such as a RED morpheme. In BRCT, BR correspondence was motivated as an identity-enforcing mechanism to handle various types of over- /underapplication patterns observed in reduplication. MR, in contrast, leaves open several possible ways of dealing with various types of phonology-reduplication interactions, even though BR correspondence is not adopted in this model. The analyses in Saba Kircher (2010) made use of Stratal OT (Kiparsky 2000, 2003, 2010, 2015, Bermúdez-Otero 2012, 2018, a.o.) and 27 REDUP (Zuraw 2002) to deal with overapplication in Samala, both of which are independently motivated grammatical mechanisms. The original proposal of MR has been adopted in several theories/theoretical frameworks with additional innovations. One recent theoretical proposal that relates to MR, which is first proposed by Bermúdez-Otero (2012), is known as Generalized Nonlinear Affixation (GNA). Generalized Nonlinear Affixation aims to capture various morphophonological phenomena, and the principle insight is that all non-concatenative morphological processes can be treated as the affixation of nonlinear phonological materials, such as floating features, tones, or prosodic nodes, and therefore, nonconcatenative morphology, including reduplication, is emergent. The reanalysis of reduplication as prosodic node affixation is a crucial part of GNA (Bermúdez-Otero 2012). Bermúdez-Otero (2012) adopts Saba Kirchner’s (2010) proposal of Minimal Reduplication to argue for GNA, and Stratal OT is incorporated into the analysis as the way to predict reduplicative opacity. Beyond reduplication, there have been many successful analytical studies supporting GNA, and it has been shown that the affixation of nonlinear phonological elements can result in various morphological processes, including subtractive morphology (Trommer 2011, Bye & Svenonius 2012, Trommer and Zimmermann 2014, Zimmermann 2017a), morphological mutation (Wolf 2007), morphological lengthening (Davis and Ueda 2002, 2006), etc. As laid out in Chapter 1, the analysis of reduplication as prosodic node affixation has additional benefits. Previous studies have shown the utility of prosodic node affixation in deriving various types of nonconcatenative morphology, including morphological lengthening and subtractive morphology. In Chapter 3 and Chapter 4, this advantage will be argued and evidenced by the case studies of Huozhou Chinese and Rapa Nui. In the current study, I follow the basic theoretical assumptions of Minimal Reduplication, summarized in (1). 28 (1) Morphological reduplication: the basic theoretical assumptions a. The phonological exponent of morphological reduplication involves deficient prosodic templates. b. Reduplication is emergent as a repair strategy for the deficient prosodic template. c. Reduplication is viewed as segment fission and incurs violations of INTEGRITY-IO (Struijke 2002). A pattern of reduplication in Rapa Nui serves to illustrate. Verbal pluralization in Rapa Nui (Austronesian) can be expressed through reduplication (Kieviet 2017). In most cases, the onset and nucleus of the initial syllable are copied to form an affixal light syllable, as exemplified in (2) (the reduplicant is underlined). (2) Rapa Nui plural reduplication Singular Gloss Plural Gloss a. ʔaɾa to wake up ʔaʔaɾa to wake up.PL b. eke to mount eeke to mount.PL c. tuɾu to go down tutuɾu to go down.PL d. hopu to bathe hohopu to bathe.PL e. kai to eat kakai to eat.PL To analyze the data with the theoretical assumptions in (1), I propose that the underlying phonological form of the plural marker is a mora (μ) that lacks segmental content. 4 In this way, reduplication is initiated to repair the deficient prosodic template and integrate the mora into a well-formed prosodic structure. See the demonstration in (3). 5 (3) Rapa Nui plural reduplication as mora affixation /μ + ho μ pu μ / HD(μ) DEP-IO INTEGRITY-IO ☞ a. ho μ .ho μ .pu μ 2 b. tə μ .ho μ .pu μ 2W L c. μ + ho μ .pu μ 1W L 4 Kieviet (2017:62) also mentioned the mora affixation analysis but it was not implemented into an OT analysis. 5 Another candidate, [o μ .ho μ .pu μ ], is omitted here. The copy of onset will be discussed in Chapter 3, Section 3.1. 29 In this case, an empty mora template in the input is non-headed since it is not associated with any segments. Thus, in (3a), reduplication is motivated by HEADEDNESS(μ), which penalizes non- headed moras, at the expense of INTEGRITY-IO because of the copied segments [ho]. In addition, epenthesis of default segments can be an alternative way to fulfill the mora template, as indicated by (3b), but it is ruled out by the higher-ranked DEP-IO. For (3c), the input floating mora leaves non-headed in the output, fatally violating HEADEDNESS(μ). Note that the prosodic structure of the input has been established at the stage of mora affixation, which is a common assumption in the approaches where reduplication is viewed as prosodic affixation. Various strategies have been adopted to ensure that prosodification occurs at an earlier stage. In the current study, I resort to Stratal OT where the roots are pre-optimized and prosodified in an early stratum during word construction. More details about the data and analyses will be presented in Chapter 4. Some follow-up studies on reduplication in the vein of MR incorporate more theoretical innovations, including Zimmermann (2013, 2015), Paschen (2018), among others, which feature an autosegmental analysis with Containment Theory (Prince & Smolensky 1993, van Oostendorp 2005, 2006, a.o.) instead of Correspondence Theory (McCarthy & Prince 1995). These studies argue that phonology-reduplication interactions are attributed to Level Ordering plus other autosegmental mechanisms rather than surface-to-surface correspondence relations on surface structures (such as BR correspondence). However, the role of correspondence in reduplication is overlooked in previous theories that take the line of MR. Even though one advantage of MR is that the lack of BR correspondence will not overpredict backcopying (Saba Kirchner 2010:17-19), the MR analysis of Samala still makes use of a correspondence-like structure and correspondence-driving constraints, namely, Zuraw’s (2002) REDUP for aggressive reduplication, to deal with the overcopy of prefix (Saba Kirchner 2010:150 ff., see also Chapter 1, §1.2.1). In the current study, one major theme is to examine the role of Surface Correspondence in reduplicative opacity, which is motivated by several reasons. First, the original proposal of MR does not explicitly reject correspondence as a potential way to deal with some patterns that 30 cannot be handled through level ordering. Second, Surface Correspondence has a similar function to BR correspondence and coupling in aggressive reduplication, and it is expected to have a similar effect in morphological reduplication. However, Surface Correspondence is different from BR correspondence because it is not initiated by any special formal device such as RED. Since Surface Correspondence has been supported by various assimilation/dissimilation phenomena, it is viewed as a part of Universal Grammar, and there is no reason that it is specifically excluded from reduplicative patterns. Third, there have been some studies that argue for the utility of Surface Correspondence in morphological reduplication (Inkelas & Zoll 2005) and phonological reduplication such as copy epenthesis or compensatory reduplication (Yu 2003, 2005, Inkelas 2008), but the role of Surface Correspondence in reduplicative opacity, especially backcopying in morphological reduplication, has not been scrutinized (see Yu 2005 for relevant discussion). 2.2. Generalized Surface Correspondence This section lays out the basic mechanisms of Generalized Surface Correspondence, which is based on Agreement-by-Correspondence (ABC, Hansson 2001/2010, Rose & Walker 2004, a.o.) and the Surface Correspondence Theory of Dissimilation (Bennett 2013, 2015). It is noteworthy that Generalized Surface Correspondence, which will be referred to as GSC in the following discussion, is not a new theory. It follows most fundamental mechanisms of ABC, and therefore, it is viewed as a version of ABC. The notion “correspondence” was introduced into Optimality Theory (Prince & Smolensky 1993/2004) as the core concept of Correspondence Theory (McCarthy & Prince 1995, see also McCarthy & Prince 1993). Correspondence is an abstract relation between representations, and the formalization in McCarthy & Prince (1995:14) is given below. 31 (4) Correspondence Given two strings S 1 and S 2 , correspondence is a relation ℛ from the elements of S 1 to those of S 2 . Elements 𝛼∈S 1 and 𝛽∈S 2 are referred to as correspondents of one another when 𝛼ℛ𝛽. Correspondence was first applied to input-output (IO) and base-reduplicant (BR) domains, where BR correspondence is a type of surface-to-surface correspondence relation. The introduction of correspondence led to other proposals of correspondence(-like) relations. One such proposal is Agreement-by-Correspondence (ABC, Hansson 2001/2010, Rose & Walker 2004, Bennett 2013, 2015, a.o.). The foundations of ABC include Walker (2000a, b, 2001), Hansson (2001), and Rose and Walker (2004). Prior to ABC, there had been other proposals of surface-to-surface correspondence relations, including Kitto & de Lacy’s (1999) BE correspondence for copy epenthesis and Zuraw’s (2002) coupling in Aggressive Reduplication. These relations will be discussed and compared to ABC in the following chapters where relevant. Although ABC was proposed for long-distance consonant agreement, in the subsequent studies, it has been extended to other relevant phenomena, including vowel harmony (e.g., Sasa 2009, Rhodes 2012, Walker 2015, 2018), consonant dissimilation (Bennett 2013, 2015a, b), tone assimilation/dissimilation (Shih & Inkelas 2019), and reduplication (Yu 2003, 2005, Inkelas 2008). Readers are referred to Rose & Walker (2004:487 ff.) for the cognitive basis and arguments for ABC. The core insight of Generalized Surface Correspondence, consistent with ABC, is that there is a correspondence relation between the phonological elements (usually consonants, vowels, and tones) on the surface within an output. This abstract correspondence relation is termed Surface Correspondence (SCorr), which is analogical to Base-Reduplicant correspondence. The word “generalized” indicates that Surface Correspondence is not specialized for certain configurations such as RED, which makes it different from the other correspondence(-like) relations such as BR correspondence. Instead, it is proposed to be a general mechanism that is 32 provided by GEN without any prerequisites. The main purpose of the current study is to show how this SCorr is able to make additional predictions that are attested in a variety of languages. Since SCorr relations between segments are provided by GEN, the possible SCorr structures vary, as demonstrated in (5) with a hypothetic word [pa.ta.ku]. Throughout the dissertation, SCorr is notated with subscript roman letters. The segments that share the same subscript form one correspondence set (see also “correspondence class” in Bennett 2013 et seq.). For the examples in (5), the number of correspondence sets ranges from one to six. Note that the correspondence sets for [pa.ta.ku] in (5) are not exhaustive. (5) SCorr relation between segments SCorr relations Correspondence Set(s) Remarks a. [p x a x .t x a x .k x u x ] {p x , a x , t x , a x , k x , u x } all the segments stand in SCorr b. [p x a i .t x a j .k x u k ] {p x , t x , k x }, {a i }, {a j }, {u k } all the consonants stand in SCorr c. [p x a i .t y a i .k z u i ] {p x }, {t y }, {k z }, {a i , a i , u i } all the vowels stand in SCorr d. [p x a i .t y a j .k z u k ] {p x }, {t y }, {k z }, {a i }, {a j }, {u k } none of the segments stands in SCorr with another … … … Following Bennett (2013 et seq.) and Bennett and Delbusso (2018), SCorr is a transitive, symmetric, and unitary relation. Transitivity indicates that if S 1 corresponds to S 2 and S 2 corresponds to S 3 , then S 1 is in correspondence with S 3 . Being symmetric means that if a segment S 1 is in correspondence with S 2 , then S 2 is in correspondence with S 1 (if S 1 ℛ S 2 , then S 2 ℛ S 1 ) (cf. Walker 2001, Hansson 2001/2010). More about the transitivity of SCorr will be discussed in Chapter 5. A unitary relation refers to the concept that if two segments stand in SCorr, then there is only one correspondence structure between them. A configuration such as [p x,i … p x,j ], where two ps are in correspondence (indicated by x) and not in correspondence (indicated by i, j) at the same time, is not expected to be generated. There are two central components in GSC, namely, a CORR constraint that requires SCorr between surface elements, and a set of limiter constraints that impels the corresponding 33 elements to be identical (e.g., IDENT-XX). In the following sections, I will introduce CORR (§2.2.1) and the limiter constraints (§2.2.2), including their definition and evaluation. Some general discussion of CORR and IDENT-XX will be reserved for Chapter 5, after the case studies. 2.2.1. The correspondence-requiring constraint CORR The CORR constraint is an essential part of the theory. Its basic function is to penalize pairs of segments that fail to establish a SCorr relation. The current theory of GSC pursues a unified feature-blind CORR constraint that requires correspondence between segments in general. This proposal follows the line of McCarthy (2010), Shih (2013), and Walker (2015) (see also Bennett & Delbusso 2018). 6 The definition of the constraint follows the one in Walker (2015): (6) CORR-XX: Assign a violation to any pair of segments that are not in correspondence in the output. The definition in (6) requires a pairwise evaluation, which reflects the transitivity and symmetry property of SCorr. For a pair of segments [p x … p y ], for instance, the non-correspondence between two segments only incurs one violation of CORR-XX rather than two. To further illustrate the evaluation of CORR-XX, a violation profile is given in (7), with a hypothetical input /patu/. 6 Another line of research pursues feature-specific CORR constraints, which can be formulated as CORR-XX[𝛼F]. See, for example, Section 2.3.2 in Bennett (2013) for detailed discussion on the definition and evaluation of this constraint. 34 (7) Violation profile of CORR-XX: patu CORR- XX Pairs that violate CORR-XX a. p x a x .t x u x none b. p x a y .t y u y 3 [p x … a y ], [p x ... t y ], [p x ... u y ] c. p x a x .t y u y 4 [p x ... t y ], [p x ... u y ], [a x ... t y ], [a x ... u y ] d. p x a y .t x u y 4 [p x ... a y ], [p x ... u y ], [a y ... t x ], [a y ... u y ] e. p x a y .t x u z 5 [p x ... a y ], [p x ... u y ], [a y ... t x ], [a y ... u z ], [t x ... u z ] f. p x a y .t z u w 6 [p x ... a y ], [p x ... t z ], [p x ... u w ], [a y ... t z ], [a y ... u w ], [t z ... u w ] Again, the evaluation of CORR-XX is pairwise, namely, the violations are assigned based on which pairs of segments are not in correspondence. Two extremes of SCorr structure can be provided by GEN for a four-segment input exemplified in (7). In (7a), all the segments stand in correspondence, meaning that they are in a single correspondence set, and therefore, no violation of CORR-XX is incurred by the candidate. On the other end of the spectrum, in (7f), no pair of segments stands in SCorr, meaning that each segment forms its own correspondence set. This candidate violates CORR-XX six times, i.e., 𝐶(4,2), due to the symmetric property of SCorr relation. For the other candidates in between, the violations of CORR-XX vary, depending on the number of correspondence sets and how the segments are partitioned into each correspondence set. The candidates in (7b), (7c), and (7d) contain two correspondence sets but they violate CORR-XX to different degrees. In sum, the function of CORR-XX is to drive SCorr between all the segments in an output string. The wellformedness of the corresponding segments is subject to the evaluation of other markedness constraints, which will be introduced shortly in §2.2.2. Some additional discussion on the feature-blind CORR constraint is reserved until the chapter summary (§2.4) and Chapter 5. 35 2.2.2. Limiter constraints Limiter constraints also play an essential role. Their function is to evaluate the wellformedness of segments that stand in SCorr. The term limiter constraints follows Bennett’s (2013, 2015a, b) “CC·Limiter”, which is specifically for consonants. These constraints are named as “limiter” since they are used to penalize corresponding segments that do not meet certain conditions. In what follows, two main limiter constraints that will be used in the analyses of this study will be introduced. These constraints are IDENT-XX [γG] (F) (§2.2.2.1) and ACCORD-XX (§2.2.2.2), the latter of which is a newly proposed constraint in this study. 2.2.2.1. The identity-driving constraint IDENT-XX [𝛾 G] (F) The limiter constraint, IDENT-XX [γG] (F) , requires identity between corresponding segments, which can be traced back to the prototype IDENT-XX(F) (e.g., Walker 2000a, b, Hansson 2001/2010, Rose & Walker 2004). The formulation adopted in the current study contains a trigger feature [G], which is based on Walker (2015). To start with, the definition of the most general version of IDENT-XX(F) is given in (8) (after Rose & Walker 2004:492): (8) IDENT-XX(F): Let X and Y be a consecutive pair of corresponding segments in the output. Assign a violation if X is [𝛼F] and Y is [𝛽F] (where 𝛼 ≠ 𝛽). The constraint IDENT-XX(F) is analogical to IDENT-BR(F) in McCarthy & Prince (1995). As mentioned earlier, the difference between SCorr and BR correspondence is that the latter one requires the presence of RED in the input, a specialized mechanism. For two segments that stand in SCorr, identity for feature [F] is invoked by this constraint, at the expense of IDENT-IO(F). As a demonstration, the violation profile of IDENT-XX(voice) and its interaction with IDENT-IO(voice) is given in (9): 36 (9) Violation profile of IDENT-XX(voice) s … z IDENT-XX(voice) IDENT-IO(voice) CORR-XX a. s x … z x 1 b. z x … z x 1 c. s x … s x 1 d. s x … z y 1 The pair in (9a) violates IDENT-XX(voice) since the corresponding segments (indicated by subscripts) do not match in the feature (voice). In both (9b) and (9c), the corresponding segments agree in their value for the feature (voice), which satisfies IDENT-XX(voice), but they both incur a violation of IDENT-IO(voice) by changing the value for the feature (voice) of the input segment. Note that the direction of feature agreement is opposite in (9b) and (9c), which will be revisited in §2.3. Finally, the pair in (9d) vacuously satisfies IDENT-XX(voice) since the segments do not stand in correspondence in the output, but it incurs one violation of CORR-XX due to non- correspondence. In the definition, consecutive refers to adjacent segments within the same correspondence set. Thus, for the sequence [s x 1 … z x 2 … z x 3 ] (the superscript numbers indicate the linear order of the segments), IDENT-XX(voice) is violated only once by [s x 1 … z x 2 ], since [s x 1 … z x 3 ] is not adjacent among the segments in the correspondence set. In comparison, the pair [s x 1 … z x 3 ] in [s x 1 … z y 2 … z x 3 ] is treated as adjacent, because the intervening segment is not in the same correspondence set as the other segments. Similar proposals that involve a local evaluation of IDENT-XX (IDENT-CC) include Hansson (2007), Rhodes (2012), Walker (2015, 2018), Shih & Inkelas (2019); see Bennett (2013 et seq.) for a global evaluation of IDENT-XX (IDENT-CC therein). 7 Beyond the basic formulation, a further developed version of IDENT-XX(F) is the feature- restricted IDENT-XX [γG] (F). The formulation and definition of this constraint is adapted from Walker (2015:6) (cf. Hansson 2014, Sande 2019): 7 The CORR-XX is evaluated globally in the current proposal. The issue of global vs local evaluation will be taken up in Chapter 5. 37 (10) IDENT-XX [γG] (F): Let X and Y be a consecutive pair of corresponding [𝛾G] segments in the output. Assign a violation if X is [𝛼F] and Y is [𝛽F] (where 𝛼 ≠ 𝛽). In this definition, consecutive again refers to adjacent segments within the same correspondence set, but it is further restricted by [𝛾G]. To demonstrate the function of this constraint, the violation profile of IDENT-XX [+cont] (voice) and IDENT-XX(voice) is as follows. (11) Violation profile of IDENT-XX [+cont] (voice) in comparison with IDENT-XX(voice) s 1 … p 2 … z 3 IDENT-XX [+cont] (voice) IDENT-XX(voice) CORR-XX a. s x 1 … p x 2 … z x 3 1 (s x 1 … z x 3 ) 1 (p x 2 … z x 3 ) b. z x 1 … p x 2 … z x 3 2 (z x 1 … p x 2 , p x 2 … z x 3 ) c. s x 1 … p x 2 … z y 3 2 (s x 1 … z y 3 , p x 2 … z y 3 ) The segments that stand in SCorr belong to a correspondence chain. However, within a chain of corresponding segments, there can be sub-chains that are defined by restrictive features. In (11a), for instance, all the segments form a chain of correspondence, but [s x 1 … z x 3 ] forms another chain defined by [+cont]. Thus, the feature-restricted constraint IDENT-XX [+cont] (voice) evaluates the subset of segments in a correspondence set that shares the feature [+cont] (i.e., [𝛾G] in the schema). This constraint requires that any consecutive pair of [+cont] segments must agree on [voice] (i.e., [F] in the schema). In (11a), the pair [s x 1 … z x 3 ] incurs one violation of IDENT-XX [+cont] (voice) since both [+cont] segments do not match in [voice]. These two segments are treated as consecutive since the intervening segment does not meet the condition [+cont]. Although this candidate also incurs one violation of IDENT-XX(voice), that violation is caused by [p x 2 … z x 3 ] rather than [s x 1 … z x 3 ], since the latter pair is not considered to be consecutive when a [+cont] restriction is not in play. These two constraints also differ in the evaluation of (11b). In (11b), the pairs [z x 1 … p x 2 ] and [p x 2 … z x 3 ] do not meet the condition [+cont], which vacuously satisfy 38 IDENT-XX [+cont] (voice). However, IDENT-XX(voice) is violated twice by both pairs. Finally, in (11c), the corresponding segments [s x 1 … p x 2 ] satisfies IDENT-XX(voice) and vacuously satisfies IDENT-XX [+cont] (voice) since they do not meet the condition [+cont]. This candidate violates CORR-XX twice, as indicated in the tableau. The difference between IDENT-XX [+cont] (voice) and IDENT-XX(voice) can be further illustrated (12) Local evaluation of IDENT-XX a. IDENT-XX [+cont] (voice): evaluating over adjacent corresponding [+cont] pairs b. IDENT-XX(voice): evaluating over adjacent corresponding pairs The illustrations in (12) show that the feature-restricted IDENT-XX [γG] (F) only evaluates the adjacent pairs of segments that are defined by feature [G] (i.e., [+cont]), as in (12a). In contrast, the general version, IDENT-XX(F), evaluates every adjacent pair of segments that stand in correspondence (12b). In the current study, both formulations will be used, depending on the actual scenario. One of the core insights of Surface Correspondence is that segments with greater similarity are more likely to interact with each other via SCorr. In other words, agreement between segments is sensitive to similarity, which is conditioned by shared features. In the current model, similarity is encoded in IDENT-XX [γG] (F) as [𝛾G]. The constraint is interpreted as requiring corresponding segments that are similar in terms of [𝛾G] to also be similar with regard to [F]. As a final note, I propose that the target [F] in both IDENT-XX(F) and IDENT-XX [γG] (F) can be either s x p x s x p x s x p x [–cont] [+cont] [–cont] [+cont] [–cont] [+cont] s x p x s x p x s x p x [–cont] [+cont] [–cont] [+cont] [–cont] [+cont] 39 segmental features or suprasegmental features, such as length and stress. For example, the constraint IDENT-XX [+vocalic] (length) or IDENT-VV(length) for short will be used to deal with length match in Chapter 4. 2.2.2.2. ACCORD-XX ACCORD-XX is a new limiter constraint proposed in this study. It follows the idea that surface-to- surface correspondence can be blocked in a pair that is not related by segment fission. The proposed definition of ACCORD-XX is given in (13). (13) ACCORD-XX: Let X i and X j be two segments in the input, and X i ′ and X j ′ be two segments in the output; and let X i ′ ℛ X i and X j ′ ℛ X j . Assign a violation if X i ′ and X j ′ form a consecutive pair that stands in Surface Correspondence. Given the definition, ACCORD-XX penalizes SCorr between two segments that do not share the same input, as in (14a) below. It is satisfied when two segments that stand in SCorr result from the fission of a single segment (14b), or when two segments do not stand in SCorr at all (14c-d). (14) Illustration of the effect of ACCORD-XX a. ACCORD violated b. ACCORD satisfied c. ACCORD satisfied d. ACCORD satisfied The violation profile of ACCORD-XX is further demonstrated in (15). In this example, input-output correspondence is indicated by digits while Surface Correspondence is notated with subscript letters. Input Output s 2 s x 1 … s x 2 s 1 s 1 s x 1 … s x 1 s 2 s x 1 … s y 2 s 1 s 1 s x 1 … s y 1 40 (15) Violation profile of ACCORD-XX s 1 … s 2 ACCORD-XX CORR-XX INTEGRITY-IO a. s x 1 … s x 1 … s x 2 1 (s x 1 … s x 2 ) 1 b. s x 1 … s x 1 … s y 2 2 1 c. s x 1 … s y 1 … s z 2 3 1 In (15), the output segments that result from segment fission are indicated by the same superscript digit (the constraints that drive segment fission are not included in this tableau). The evaluation of ACCORD-XX is consistent with that of IDENT-XX in that it evaluates each pair of corresponding segments locally. For (15a), all three segments stand in SCorr, where there are two local pairs in this candidate, [s x 1 … s x 1 ] and [s x 1 … s x 2 ]. The constraint is violated once by the pair [s x 1 … s x 2 ], since the segments that stand in SCorr do not stem from the same source in the input. In comparison, the candidate in (15b) satisfies ACCORD-XX since SCorr is only assigned to the pair [s x 1 … s x 1 ] for which the segments are related to the same input segment, while the candidate in (15c) vacuously satisfies ACCORD-XX because none of the segments stands in SCorr with others. Meanwhile, CORR-XX is violated in (15b) and (15c) by the segments that are not in the same SCorr set. This constraint is related to the idea that segment fission can give rise to a surface relation between segments, though in an indirect way. There has been a history in the literature that IO correspondence invokes a surface-to-surface correspondence between fissioned segments due to transitivity, as illustrated in (16). (16) Correspondence-by-transitivity: an illustration Input Output X [X’ + X] IO-Faith Correspondence IO-Faith 41 This type of correspondence relation is termed differently by authors, such as OO- correspondence (Raimy & Idsardi 1997, Kawu 2000), Surface Correspondence (Struijke 2002, Stanton & Zukoff 2015), and Relational Correspondence (Karabay 2004). In (16), the output segments (X′ and X) are required to be identical since they are in a relation with the same input segment. The identity between the corresponding output segments is enforced by IDENT that has been formulated in various ways, including IDENT ΣΣ (Struijke 2002), IDENT-RR (“R” for “relational”, Karabay 2004), and IDENT-SC (“SC” for “surface correspondence”, Stanton & Zukoff 2015), etc. These IDENT constraints on surface structures is able to account for misapplication of phonological processes in various domains. The constraint ACCORD does not penalize non-correspondence of segments from the same source. Instead, it penalizes corresponding segments that come from different sources, which can have an effect of favoring corresponding fissioned segments together with CORR-XX (recall the illustration in (14)). However, ACCORD brings up a further issue. As shown earlier, the effect of ACCORD on SCorr can lead to the structure in (14b), which is similar to (16). The issue is whether we can simply use correspondence-by-transitivity, rather than SCorr, to handle reduplicative identity. If we can end up with the same correspondence structure, “SCorr + ACCORD” seems to introduce redundancy. Nevertheless, there are two advantages of SCorr compared to transitivity. First, SCorr is more flexible and it is freely assigned by GEN over segments. Thus, it is possible to get non- correspondence on the surface. As argued by Bennett (2013, 2015a, b), the segments that do not stand in correspondence on the surface can result in featural dissimilation. Hence the flexibility of SCorr is able to handle the patterns where dissimilation is accompanied by reduplication. In comparison, correspondence-by-transitivity does not seem violable. Another advantage of using SCorr to handle reduplication is related to its prediction. SCorr is provided by GEN, which is not limited to the domain of reduplication. In the reduplicated form [papata], for example, it is possible to get a SCorr structure such as [p x a y .p x a y .t x a y ], where two ps correspond to t which falls outside of the reduplicative domain. This is able to drive truncatory backcopying with the 42 proper ranking between ACCORD, CORR, and MAX. This effect will be introduced in §2.3.2 shortly and further discussed in Chapter 6. As for transitivity, the only possible structure will be [p x a y .p x a y .t i a j ]. The joint effect of ACCORD-XX and CORR-XX leads to some special SCorr structures for reduplicated forms. In the current study, reduplication emerges as an epiphenomenon to fulfill an empty prosodic template. A hypothetical input-output mapping is given in (17) to illustrate the structure of IO correspondence in a reduplicated form, which follows the principle of Minimal Reduplication in (1). (17) Input-output mapping: /σ + pat/ → [pa.pat] For the output form in (17), the SCorr structures can be various, each of which is subject to the evaluation of CORR-XX and ACCORD-XX; see (18) below. (18) SCorr structures of [pa.pat] σ + p 1 a 2 t 3 ACCORD-XX CORR-XX a. p x 1 a x 2 .p x 1 a x 2 t x 3 4 [p x 1 …a x 2 ], [a x 2 …p x 1 ], [p x 1 …a x 2 ], [a x 2 …t x 3 ] b. p x 1 a x 2 .p x 1 a x 2 t y 3 3 [p x 1 …a x 2 ], [a x 2 …p x 1 ], [p x 1 …a x 2 ] 4 [p x 1 …t y 3 ], [a x 2 …t y 3 ], [p x 1 …t y 3 ], [a x 2 …t y 3 ] c. p x 1 a y 2 .p x 1 a y 2 t z 3 8 [p x 1 …a y 2 ], [p x 1 …a y 2 ], [p x 1 …t z 3 ], [a y 2 …p x 1 ], [a y 2 …t z 3 ], [p x 1 …a y 2 ], [p x 1 …t z 3 ], [a y 2 …t z 3 ] d. p x 1 a y 2 .p x 1 a j 2 t k 3 9 [p x 1 …a y 2 ], [p x 1 …a j 2 ], [p x 1 …t k 3 ], [a y 2 …p x 1 ], [a y 2 …a j 2 ], [a y 2 …t k 3 ], [p x 1 …a j 2 ], [p x 1 …t k 3 ], [a j 2 …t k 3 ] e. p x 1 a y 2 .p i 1 a j 2 t k 3 10 [p x 1 …a y 2 ], [p x 1 …p i 1 ], [p x 1 …a j 2 ], [p x 1 …t k 3 ], [a y 2 …p i 1 ], [a y 2 …a j 2 ], [a y 2 …t k 3 ], [p i 1 …a j 2 ], [p i 1 …t k 3 ], [a j 2 …t k 3 ] p 1 a 2 t 3 p 1 a 2 t 3 p 1 a 2 43 The constraint ACCORD-XX is violated by four local pairs in candidate (18a), although CORR-XX is satisfied by all segments being in a single correspondence set. For (18b), the coda [t] is excluded from the correspondence set, which results in fewer violations of ACCORD-XX, but it incurs 4 violations of CORR-XX. In comparison, all the other candidates in (18c-e) satisfy ACCORD-XX, but each of them violates CORR-XX to different degrees. If ACCORD-XX outranks CORR-XX, the grammar will select (18c) as the optimal SCorr structure for the reduplicated form. In (18c), only the fissioned segments stand in SCorr with each other, which is favored by ACCORD-XX. At the same time, this candidate incurs the least severe violation of CORR-XX, while satisfying ACCORD- XX. As we will see in the following sections, this SCorr structure is able to deal with some cases of truncatory backcopying, especially length match. Note that the candidates in (18d) and (18e) are harmonically bounded by (18c); the candidate in (18b) is harmonically bounded by either (18a) or (18c) depending on the actual ranking. Although these structures are not favored under any ranking of the current set of constraints, whether or not they are crucial for other predictions awaits further investigation. 2.2.3. Summary This section has introduced the crucial constraints of GSC, including the correspondence- requiring constraint CORR-XX and two limiter constraints that evaluate SCorr structures, IDENT-XX [γG] (F) and ACCORD-XX. Among these constraints, CORR-XX and IDENT-XX [γG] (F) are based on the proposal of Walker (2015), and ACCORD-XX is newly proposed in the current study. The function of the main constraints is summarized in (19). 44 (19) Constraints in GSC Constraint Function a. CORR-XX Requires Surface Correspondence between segments SCorr- requiring b. IDENT-XX(F) Impels identity between segments that are in Surface Correspondence. Limiter constraints c. IDENT-XX [𝛾 G] (F) Impels identity between [𝛾G] segments that are in Surface Correspondence. d. ACCORD-XX Penalizes SCorr among segments that are from distinct input sources. 2.3. The effects of Generalized Surface Correspondence in reduplication Surface Correspondence relation was first proposed to deal with long-distance segment agreement. The basic mechanism is that segments standing in SCorr are required by an IDENT- XX(F) constraint to be identical at the sacrifice of IDENT-IO(F). First of all, a classic example of Chaha (Semitic) consonant harmony can be used to briefly illustrate the basic function of GSC. In Chaha, [–cont] consonants (i.e., plosives) within a verb root are required to agree in [voice] such as [dɨɡɨs] (*[tɨɡɨs]) (Rose & Walker 2004, from Banksira 2000). The constraints needed to produce this pattern are given in (20), followed by a tableau in (21) (cf. McCarthy 2010, Shih 2013). (20) Constraints for voicing assimilation a. CORR-XX: Assign a violation to any pair of segments that are not in correspondence in the output. b. IDENT-XX [–cont] (voice): Let X and Y be a consecutive pair of corresponding [–cont] segments in the output. Assign a violation if X is [⍺voice] and Y is [–⍺voice]. c. IDENT-XX [+cons] (continuant): Let X and Y be a consecutive pair of corresponding [+cons] segments in the output. Assign a violation if X is [⍺continuant] and Y is [– ⍺continuant]. 45 d. IDENT-IO(F): Let X be a segment in the input and Y be a correspondent of X in the output. Assign a violation if X is [𝛼F] and Y is [𝛽F], where 𝛼 ≠ 𝛽. ([F] can be [voice] or [continuant] in this analysis). (21) Input /tɨɡɨs/: Chaha consonant harmony /tɨɡɨs/ ID-XX [+cons] (cont) ID-IO (cont) CORR- XX ID-XX [–cont] (voice) ID-IO (voice) ☞ a. d x ɨ x ɡ x ɨ x s y 4 1 b. t x ɨ x ɡ x ɨ x s y 4 1W L c. t x ɨ i ɡ y ɨ j s z 10W L d. d x ɨ x ɡ x ɨ x d x 1W L 2W e. d x ɨ x ɡ x ɨ x s x 1W L 1W 2W The candidates in (21a), (21b), and (21c) violate CORR-XX to different degrees, depending on the SCorr structure of each candidate, and (21c) is ruled out for the most severe violation. The candidate in (21a) is selected as the winner since the corresponding [–cont] segments agree in [voice] (i.e., d x … ɡ x ), satisfying IDENT-XX [–cont] (voice), although it violates IDENT-IO(voice). Although CORR-XX favors a single correspondence set, as in (21d) and (21e), these candidates involve corresponding segments that do not satisfy IDENT-XX [+cons] (cont) or IDENT-IO(cont). It is noteworthy that IDENT-XX [+cons] (cont) is only violated once in (21e) by the pair [ɡ x … s x ], since both segments are [+cons] and consecutive. In the analysis, the pressure of IDENT-XX [+cons] (cont) or IDENT-IO(cont) drives the grammar to leave the final /s/ outside of the same correspondence set with the other segments, while IDENT-XX [–cont] (voice) enforces the agreement in [voice] between the plosives. The crucial part of the ranking above is CORR-XX, IDENT-XX [𝛾 G] (F) >> IDENT-IO(F). It is worth mentioning that the candidate *[tɨkɨs], which shows a different directionality of voicing (left to right), is not included in the tableau. The directionality issue can be modeled by positing IDENT-X R X L (e.g., Rose & Walker 2004:500), a variation of IDENT-XX, which calls for leftward assimilation. As the main goal of this study, the identity effect observed in various reduplicative patterns can be achieved in a similar fashion. In the analysis of consonant harmony and vowel harmony, 46 the enforcement of featural identity between segments is achieved partly due to ranking CORR- XX, IDENT-XX(F) >> IDENT-IO(F), where input-output featural identity is ranked low. Nevertheless, the IO-faithfulness constraints that are involved in reduplicative patterns are not limited to IDENT-IO(F), especially when dealing with truncatory backcopying that involves shortening or segment deletion. In the following sections, I will illustrate that when the faithfulness constraints MAX-IO-μ (§2.3.1) and MAX-IO-seg (§2.3.2) come into interaction with GSC constraints, a grammar can give rise to truncatory backcopying with length match and truncatory backcopying with segment deletion respectively. 2.3.1. Dominated MAX-IO-μ A type of truncatory backcopying introduced in Chapter 1 is length match. If a mora (μ) is prefixed to /paː/, a root that contains a long vowel, a potential output is [pa.pa] (instead of *[pa.paː]), where a long vowel in the base is shortened to match the monomoraic reduplicative template. The current mechanism of GSC is able to derive this pattern by ranking the constraint MAX-IO-μ below a relevant IDENT-XX constraint. To demonstrate this effect, the relevant constraints that have not been previously introduced are defined in (22), followed by a tableau with a hypothetical input /μ + paː/ in (23). (22) Crucial constraints for length match a. IDENT-XX [+vocalic] (length): Let X and Y be a consecutive pair of corresponding [+vocalic] segments in the output. Assign a violation if X and Y are not associated with the same number of moras. (IDENT-VV(length)) b. MAX-IO-μ: Assign a violation for every mora in the input that does not have a correspondent in the output. (MAX-μ) 47 (23) Input /μ + paː/: vowel shortening /μ + p 1 aː 2 / CORR-XX ID-VV(length) MAX-μ ☞ a. p x 1 a x 2 .p x 1 a x 2 1 b. p x 1 a x 2 .p x 1 aː x 2 1W L c. p x 1 a y 2 .p x 1 aː y 2 4W 1W L d. p x 1 a y 2 .p x 1 a y 2 4W 1 e. p x 1 a y 2 .p i 1 aː j 2 6W L In (23), GEN provides various SCorr structures, and two of them, (23a) and (23b), satisfy CORR-XX. The corresponding segments are subject to the evaluation of ID-VV(length). For (23b), the corresponding vowels are not associated with an identical number of moras (i.e., they differ in vowel length). Therefore, it is ruled out. The winner (23a) complies with the requirements of CORR-XX and ID-VV(length) by shortening the long vowel in the base, which violates MAX-μ. As a side note, another candidate *[p x 1 aː y 2 .p x 1 aː y 2 ] involves an epenthetic mora in the reduplicated part to satisfy ID-VV(length). This candidate indicates that DEP-μ is ranked higher than ID- VV(length) to avoid mora epenthesis. When the constraint ACCORD-XX comes into interaction with the other SCorr constraints, a different output will be selected. Since ACCORD-XX requires a “match” between SCorr and IO correspondence, a side effect of this constraint is to restrict the enforcement of identity to reduplication only. See the demonstration in (24). (24) Input /μ + paː/: the effect of ACCORD-XX /μ + p 1 aː 2 / ACCORD- XX CORR-XX ID-VV (length) MAX-μ a. p x 1 a x 2 .p x 1 a x 2 3W L 1 b. p x 1 a x 2 .p x 1 aː x 2 3W L 1W L c. p x 1 a y 2 .p x 1 aː y 2 4 1W L ☞ d. p x 1 a y 2 .p x 1 a y 2 4 1 e. p x 1 a y 2 .p i 1 aː j 2 6W L 48 With the same set of constraints in addition to the dominating ACCORD-XX, the output in (24) is [p x 1 a y 2 .p x 1 a y 2 ] (24d) instead of [p x 1 a x 2 .p x 1 a x 2 ] (24a). Though both surface forms show vowel shortening, (24d) has a different SCorr structure, where only the segments that are related to the same input source stand in SCorr. The candidates in (24c) and (24d) maximally satisfy ACCORD-XX and minimally violate CORR-XX at the same time, and the higher-ranked ID-VV(length) motivates shortening in (24d). The effect of ACCORD-XX can be further demonstrated in (25), where the input does not contain any affixal prosodic templates that may initiate reduplication. (25) Input /pa.paː/: the effect of ACCORD-XX /p 1 a 2 .p 3 aː 4 / ACCORD- XX CORR-XX ID-VV (length) MAX-μ ☞ a. p x 1 a y 2 .p i 3 aː j 4 6 b. p x 1 a y 2 .p x 3 a y 4 2W 4L 1W c. p x 1 a y 2 .p x 3 aː y 4 2W 4L 1W d. p x 1 a x 2 .p x 3 a x 4 3W L 1W e. p x 1 a x 2 .p x 3 aː x 4 3W L 1W With the same ranking, the highly-ranked ACCORD-XX prevents overapplication of length match in a non-reduplicative context. The candidates in (25b-e) violate ACCORD-XX to different extents, and non-correspondence is favored when there is no segment fission taking place. More discussion on the ways of limiting SCorr to reduplication along this line will be presented in Chapter 4. 2.3.2. Dominated MAX-IO-seg When MAX-IO-seg is dominated by the other GSC constraints, the grammar will lead to segment deletion. This effect has been noticed in Inkelas & Shih (2014), where segment deletion can be viewed as a way to repair unstable Surface Correspondence, and segment deletion is viewed as dissimilation to an extreme point. Truncatory backcopying with segment deletion is a parallel 49 to this insight, which is the typical scenario of the Kager-Hamilton Conundrum (McCarthy & Prince 1999) (for more discussion on this point, see Chapter 6, §6.1.5). In general, the ranking schema for truncatory backcopying that exhibits segment deletion is that GSC constraints (CORR-XX, ACCORD-XX, IDENT-XX, etc.) outrank MAX-IO-seg (MAX-S). However, the relative position of ACCORD-XX leads to different output SCorr structures. In the following toy grammar, when the input /σ + p 1 a 2 t 3 / undergoes reduplication, the output is [p 1 a 2 .p 1 a 2 ]. In this case, the coda of the base is deleted in the output in order to comply with the reduplicative template of a syllable. Assume ACCORD-XX outranks CORR-XX and MAX-S. (26) Input /σ + pat/: templatic backcopying with segment deletion /σ + p 1 a 2 t 3 / ID-CC ACCORD-XX CORR-XX MAX-S ☞ a. p x 1 a y 2 .p x 1 a y 2 4 1 b. p x 1 a y 2 .p x 1 a y 2 t z 3 8W L c. p x 1 a x 2 .p x 1 a x 2 3W L 1 d. p x 1 a y 2 .p x 1 a y 2 t x 3 1W 1W 6W L e. p x 1 a x 2 .p x 1 a x 2 t x 3 1W 4W L L In this tableau, I use ID-CC as a shorthand for IDENT-XX [+cons] (ALLFEATURES), namely, all corresponding consonants are featurally identical. When CORR-XX is sandwiched between ACCORD-XX, ID-CC, and MAX-IO, segment deletion is a way to maximally reduce non- correspondence and minimally violate ACCORD-XX and ID-CC. Between (26a) and (26b), which fully satisfy ACCORD-XX and ID-CC, the candidate in (26a) is the winner since it has a smaller number of correspondence sets. The logic behind this effect is simple. The constraint CORR-XX imposes a pressure to put all segments into the same correspondence set, but the constraint ACCORD-XX requires the corresponding segments to be mapped to the same input. The winner (26a) strikes a balance between the forces of these two constraints. Note that another candidate with full reduplication [pat.pat] is ruled out by a more severe violation of multiple constraints. 50 The possible structure, [p x a y t z .p x a y t z ], which satisfies top-ranked ID-CC and ACCORD-XX, incurs more violation of CORR-XX and INTEGRITY. With the same set of constraints, another possible output with a different SCorr structure will be selected if ACCORD-XX is demoted, as demonstrated in (27). Note that the winner in (26) and (27) have an identical phonetic surface form. (27) Input /σ + pat/: templatic backcopying with segment deletion / σ + p 1 a 2 t 3 / CORR-XX ID-CC MAX-S ACCORD-XX a. p x 1 a y 2 .p x 1 a y 2 4W 1 L b. p x 1 a y 2 .p x 1 a y 2 t z 3 8W L L ☞ c. p x 1 a x 2 .p x 1 a x 2 1 3 d. p x 1 a y 2 .p x 1 a y 2 t x 3 6W 1W L 1L e. p x 1 a x 2 .p x 1 a x 2 t x 3 1W L 4W When ACCORD-XX is ranked low, segment deletion is motivated by the pressure of CORR-XX and ID-CC. Similarly, CORR-XX favors a single correspondence set, ruling out (27a), (27b), and (27d), but ID-CC penalizes the pair [p x 1 … t x 3 ] in (27e), and therefore, segment deletion is triggered as the repair strategy to this SCorr structure that does not meet the identity between consonants. It is worthy of mention that the candidate in (27d) also violates IDENT-XX-SROLE (e.g., Rose & Walker 2004, Bennett 2013) which requires that corresponding segments have identical syllable roles. The rankings in (26) and (27) provide a sketch of the interaction between MAX-S and GSC constraints, which leads to truncatory backcopying during reduplication. In this way, truncatory backcopying is viewed as emergent and epiphenomenal due to the effect of Surface Correspondence. The other constraints such as IDENT-XX-SROLE and IDENT-IO(F) will be also involved in the analyses of linguistic data where relevant. The two rankings that predict truncatory backcopying with segment deletion are summarized in (28). The constraint ID-CC is in parenthesis in (28a) since it is not the main driving force of deletion. 51 (28) Rankings that motivate segment deletion a. ACCORD-XX, (IDENT-CC) >> CORR-XX >> MAX-S b. CORR-XX, IDENT-CC >> MAX-S, ACCORD-XX The difference between the rankings in (28) is the position of ACCORD-XX. In the attested cases of templatic backcopying, the pattern is always that all the “extra segments” are deleted to make the base match the reduplicative template. See the examples from Guarijio (Caballero 2006:278, from Miller 1996). More discussions about the typology are in Chapter 6. (29) Guarijio toní ‘to boil’ to-tó ‘to start boiling’ sibá ‘to scratch’ si-sí ‘to start scratching’ kusú ‘to sing (animals)’ ku-kú ‘to start singing’ suhku ‘to scratch body’ su-sú ‘to start scratching the body’ In Guarijio, the base is trimmed to the exact size of the reduplicant, and no extra deletion is observed. This pattern can be achieved by the same ranking in (28a), namely, ACCORD-XX >> CORR- XX >> MAX-S. In this ranking, CORR-XX and ACCORD-XX are the driving force of truncatory backcopying. Nevertheless, in Chapter 3, I will present a case study of Huozhou Chinese, where the pattern is generated by the ranking in (28b), where CORR-XX and IDENT-XX are top-ranked and motivate segment deletion. I call this pattern “conditional truncatory backcopying". As discussed in §2.2.2.2, the prediction of truncatory backcopying with segment deletion is a special effect of SCorr. An approach that pursues surface-to-surface correspondence relations through transitivity is not able to make such a prediction. 2.4. Summary In this chapter, I laid out the theories of reduplication and Surface Correspondence that I will use in the study. For reduplication, I follow the general proposal of Minimal Reduplication (Saba Kirchner 2010, 2013), which is in line with the theory Generalized Nonlinear Affixation 52 (Bermúdez-Otero 2012). For Surface Correspondence, I adapt a version of ABC into Generalized Surface Correspondence (GSC), which is characterized by a feature-blind CORR-XX constraint and feature-restricted IDENT-XX [𝛾 G] (F), along with other crucial constraints. Again, the theoretical mechanisms that I pursue in this study are largely consistent with the previous proposals. The core theme of the study is to argue how the independently motivated theories can be extended to handle reduplicative opacity, which leads to a more economical architecture of grammar. For GSC, the major difference from classic ABC is that there is only a single CORR constraint that does not refer to any feature. As mentioned in §2.2.1, this idea is in line with the proposal of McCarthy (2010), Shih (2013), and Walker (2015). The major advantage of a feature-blind CORR- XX is that it reduces the burden of feature reference to IDENT-XX [𝛾 G] (F) only. As pointed out in McCarthy (2010), having two types of constraints that refer to features introduces redundancy. In addition, in harmony systems, the feature-blind CORR-XX and feature-restricted IDENT-XX [𝛾 G] (F) are able to handle the problem of discrete triggers (see Walker 2015 for more details, cf. Walker 2018). Although McCarthy (2010) has a more radical proposal that eliminates CORR and uses MAX- XX (MAX-CC therein) instead, this study does not follow this formalization. The main reason is that the purpose of MAX-XX is to make Surface Correspondence more analogical to the correspondence relations in other domains (e.g., Input-Output, Output-Output), the definition of MAX-XX in (McCarthy 2010:3) is not consistent with the definition of MAX (e.g., MAX-IO) in general. The elements of GSC will be further revisited in Chapter 5 after the case studies. The case studies in the following chapters argue for two major aspects of the theory that I pursue here. On the one hand, analyzing reduplication as prosodic node affixation has an advantage since prosodic node affixation is versatile and can lead to various forms on the surface, such as reduplication, subtraction, and lengthening, etc. This point will be supported by both Huozhou Chinese (Chapter 3) and Rapa Nui (Chapter 4). On the other hand, the proposal of prosodic node affixation eliminates RED and BR correspondence. Instead, the backcopying patterns observed in the case studies can be handled by Surface Correspondence, which is an independently motivated mechanism in grammar. 53 Finally, one issue about GSC in reduplication is the domain of application. GSC is purely phonological, but the identity effect in reduplication that is impelled by GSC is morphologically- conditioned. In other words, the grammar should be able to restrict the identity effect to reduplicative context only, without causing any problems of overprediction. As shown earlier, the constraint ACCORD can contribute to this issue. Additional discussion will be presented in the case studies. 54 Chapter 3 Diminutive formation in Huozhou Chinese This chapter presents an in-depth case study of diminutive formation in Huozhou Chinese, where the reduplication patterns show a pattern of conditional truncatory backcopying. A diminutive in Huozhou Chinese is formed either through rime change (sound changes in the rime for morphological purposes) or reduplication, exemplified as follows: (1) Diminutive formation in Huozhou Chinese Non-diminutive Gloss Diminutive [pʰɑŋ 35 ] ‘plate’ a. [pʰaː 35 ] Rime change b. [pʰɑŋ 35 .pʰa 55 ] Reduplication [pɑw 11 ] ‘bag’ c. [poː 11 ] Rime change d. [poː 11 .po 33 ] Reduplication There are two highlighted patterns in Huozhou diminutive formation. First, the diminutive form can be either rime change or reduplication. For the noun [pʰɑŋ 35 ] ‘plate’, for instance, the diminutive form can be either [pʰaː 35 ] (rime change) or [pʰɑŋ 35 .pʰa 55 ] (reduplication), while there is no clear semantic difference in these two forms. Second, diminutive reduplication shows a case of truncatory backcopying. The reduplicant in Huozhou Chinese is always an open syllable, as illustrated in (1b) and (1d). However, the shape of the reduplicant is carried over into the base when the noun ends with [w], as in (1d) [poː 11 .po 33 ], which is the focus of this case study. Recall Chapter 1, I call this pattern “truncatory backcopying” since the base is truncated by segments to match the shape of the reduplicant. Additionally, this case of backcopying is only exhibited 55 in nouns with a [w] offglide, comparing (1b) with (1d). Although the form in (1d) may seem to motivate an ordered application of rime change and reduplication (i.e., [pɑw 11 ] → [poː 11 ] → [poː 11 .po 33 ]), I will show that there are problems with such an approach and argue instead that this form is a case of reduplicative opacity which results from Surface Correspondence. With respect to the two highlighted patterns, the major goals of this chapter are twofold. First, the analyses will demonstrate that rime change and reduplication can be analyzed as the variable realizations of an affixal mora, which supports the model of reduplication as the affixation of prosodic nodes. Second, I will show how and why the pattern of full reduplication in (1d) is best analyzed as the enforcement of surface identity through Surface Correspondence, which forms a major argument of the main theme of the dissertation. In the following analyses, two groups of interacting constraints jointly generate the observed patterns. The first group handles the variable realization of mora affixation. Since the diminutive marker is treated as an underlying mora, a set of constraints, which will be referred to as mora wellformedness constraints, force the realization of the floating mora. The faithfulness constraints, MAX (MAX-IO-seg and MAX-IO-μ) and INTEGRITY, play an essential role in selecting the exact surface form. When the mora wellformedness constraints are top-ranked and when INTEGRITY dominates MAX, rime change (subtraction) is selected as the way to realize the diminutive morpheme. Instead, if MAX dominates INTEGRITY, reduplication is produced as the output. Another group of constraints is responsible for backcopying and featural changes, which includes SCorr constraints and featural identity constraints (e.g., MAX(Feature)). I will show that the truncatory backcopying pattern is emergent and attributed to the pressure of CORR and IDENT-XX. The MAX(feature) constraints, along with other relevant markedness constraints, drive the changes at the segmental level such as vowel raising. The details of analyses can be found in §3.3.3 and §3.3.4. This chapter contains eight sections and an appendix. Here I provide a brief preview for each of the sections. 56 • Section 3.1 offers the background information of Huozhou phonology. The data and patterns of Huozhou diminutive formation are presented in detail. • Section 3.2 introduces the theoretical preliminaries and sets the stage for the formal analyses. In this section, the proposal of mora affixation is made; the output moraic structures are proposed. • Section 3.3 focuses on Huozhou diminutive reduplication (e.g., 1b, 1d in the example above), which constitutes the main argument for the utility of SCorr in reduplication. In this section, the utility of SCorr constraints is demonstrated in generating truncatory backcopying as well as featural changes (1d). In the meanwhile, the same set of constraints can also account for the block of truncatory backcopying as in (1b). • Section 3.4 deals with Huozhou diminutive rime change (e.g., 1a, 1c), the alternative pattern of diminutive formation. The purpose of this section is to show how mora affixation can lead to subtraction. The featural changes in (1a) and (1c) are also analyzed. • Section 3.5 offers an interim summary of the analyses in §3.3 and §3.4. A unified grammar can be established by integrating the two groups of constraints that are introduced earlier. The two highlighted patterns of Huozhou diminutive formation are reviewed and discussed. • Section 3.6 compares the current proposal with other alternatives, with respect to the two highlighted patterns. First, the alternative approaches for truncatory backcopying are discussed to argue for the advantage of Surface Correspondence. Second, this section argues for the analysis of mora affixation by comparing the current proposal with other models of nonconcatenative morphology. • Section 3.7 provides additional support for the treatment of mora affixation from a typological perspective. The other rime change processes in nearby dialects can be also viewed as mora affixation. • Section 3.8 summarizes and concludes this chapter. 57 • The appendix deals with the (dis)agreement of backness in diminutive reduplication. Although it is not directly related to the main theme of the dissertation, the discussion in this section makes the analysis more thorough. 3.1. Diminutive formation in Huozhou Chinese 3.1.1. Data sources Houzhou is a city located in the valley of Fen River, southern Shanxi Province. 8 Huozhou Chinese is classified as a variety of Zhongyuan Mandarin (Hou & Wen 1993, Shen 2003), and there are four sub-varieties, i.e., West Huozhou, East Huozhou, North Huozhou, and Wangzhuang (Feng & Zhao 2014:10). 9 Huozhou diminutive formation was first introduced in Satoh & Feng (1989), followed by X.Tian (1992), Hou & Wen (1993), J.Tian (2009), Shen et al. (2010), and Feng & Zhao (2014), among others. The patterns are consistently described across the studies, despite some notational and sub-dialectal variations, as well as some minor divergences. The data presented in this chapter are based on my consultation with three informants (one male, two female; aged between 50 and 65) from the town of Bailong, a representative of West Huozhou. The divergences between my data and the description in previous literature will be mentioned where necessary. 3.1.2. Aspects of Huozhou phonology Huozhou Chinese has the same syllable structure as Standard Mandarin, and the maximal syllable template is (C)(G)V(X), in which G represents glide, and X can be either glide or nasal. I adopt the traditional view of Chinese syllable structure, where Final is a constituent that contains the prenuclear glide and rime (2a, b), and falling diphthongs are treated as a branching 8 Huozhou is a subordinate city within Linfen, a prefecture-level city in Shanxi Province. 9 Wangzhuang is a village in northern Huozhou, bordering Lingshi, Shanxi. The variety spoken here has some phonological properties of Jin Chinese (e.g., the preservation of checked syllable), and therefore, it is treated as a different variety from North Huozhou. 58 nucleus (2b) (e.g., Cheng 1966, Lin 1989, cf. Bao 1990, 1996, Duanmu 1990, 2007, 2008, van de Weijer & Zhang 2008, a.o.). (2) Maximal syllable structure of Huozhou Chinese a. b. In terms of syllable weight, I follow the proposal that all full Mandarin syllables ––– those that carry a full tone ––– are bimoraic (e.g., Duanmu 1990, 1993, 1994, 2007, Wu and Kenstowicz 2015). Thus, the nuclear vowel in an open syllable should be represented as a long vowel on the surface (3a) in order to bear the lexical tone. Meanwhile, only elements in the rime can be moraic, namely, moraicity does not extend to the prenuclear glide (3a-b). The reduplicant in nominal reduplication is usually treated as a weak syllable, which is monomoraic and neutral-toned, as shown in the underlined part in (3c) (e.g., Duanmu 1993, 1999, 2007, Sui 2013). (3) Moraic representation of Huozhou Chinese syllables (tones are omitted) a. b. c. e.g. [pʰaː] [tɕjaː] [pʰɑŋ] [tɕjɑŋ] [pʰɑŋ.pa] [kwɑŋ.kwa] Rime Onset σ C Final G Nuc Coda V C Rime Onset σ C Final G Nuc V G µ µ σ C (G) Vː µ µ σ C (G) V X µ µ σ C (G) V X µ σ C (G) V 59 I posit that Huozhou Chinese has six underlying vowels, /i, y, u, ə, a, o/ (but see Duanmu 2007 and Zhang to appear for the discussion of a five-vowel system in Standard Mandarin). Both onglides ([j], [ɥ], and [w]) and offglides ([j] and [w]) are derived from corresponding vowels. In traditional Chinese phonology, the constituent final is usually described as a whole unit, and the inventory of Huozhou finals is listed in (4). For each final, I give the corresponding underlying form in parentheses. All the surface forms without an offglide or nasal coda (labeled as “open”) are represented with a length mark (ː) for reasons that I explained above. Since the offglide is not treated as the coda, I use ending instead, and the labels ‘ŋ-/j-/w-ending’ are used for the purpose of categorization throughout the paper. As a side note, the symbols [ɿ] and [ʅ] are not standard IPA symbols, but they are often used to represent the “apical vowels” in various Chinese languages and dialects (see Lee & Zee 2017 for more information). In addition, the inventory of consonants is listed in (5). (4) Inventory of finals in Huozhou open ŋ-ending j-ending w-ending aː (/a/) ɤː (/ə/) ɚː (/ə/) ɑŋ (/aŋ/) əŋ (/əŋ/) aj (/ai/) ej (/əi/) ɑw (/au/) ow (/əu/) iː (/i/) jaː (/ia/) jeː (/iə/) ɿː (/i/) ʅː (/i/) ɨː (/i/) jɑŋ (/iaŋ/) iŋ (/iŋ/) jaj (/iai/) jɑw (/iau/) jow (/iəu/) uː (/u/) wɑː (/ua/) woː (/uə/) wɑŋ (/uaŋ/) uŋ (/uŋ/) waj (/uai/) wej (/uəi/) yː (/y/) ɥaː (/ya/) ɥeː (/yə/) ɥɑŋ (/yaŋ/) yŋ (/yŋ/) (5) Inventory of consonants in Huozhou labial labio- dental alveolar retroflex alveo- palatal palatal velar plosive p pʰ t tʰ k kʰ nasal m ȵ ŋ fricative f v s z ʂ ʐ ɕ x affricate ts tsʰ tʂ tʂʰ tɕ tɕʰ lateral l approximant (ɥ w) j ɥ w 60 Here are several remarks about the final inventory. Hou & Wen (1993:688) documented three additional finals [ɛ, wɛ, ɥo] that were neither reported in other work such as Feng & Zhao (2014) nor in my consultation. Hou & Wen (1993:689) also documented a series of nasalized finals, [õ, ɪ̃ , ʅ̃ , iɛ̃], but they mentioned that these finals only existed in the speech of the old generation. The informants I consulted with did not report this series. Also, the series [ɤː, jeː, woː, ɥeː] in (4) was documented by X.Tian (1992) as [ʌ, jʌ, wʌ, ɥʌ]. As for the underlying representation, mid vowels [e] and [o] on the surface are derived from underlying /ə/. Relevant discussions on the derivation of surface mid vowels in Mandarin can be found in Duanmu (2007), among others. There are five lexical tones in Huozhou Chinese, low-level (T1), rising (T2), mid-level (T3), falling (T4), and high-level (T5). In this dissertation, I transcribe the tones with Chao’s (1968) digits as ‘11’, ‘35’, ‘33’, ‘53’, and ‘55’. Some additional issues about tones will be introduced where relevant. 3.1.3. Diminutive formation in West Huozhou There are two ways to form a diminutive in Huozhou Chinese: diminutive rime change and diminutive reduplication. The so-called “diminutive rime change” is a morphophonological process where the rime part of a syllable undergoes regular sound changes to indicate diminutive, hypocoristic, and/or cuteness. Diminutive rime change can be found in many Chinese dialects, especially the ones in Zhongyuan Mandarin and Jin Chinese. The process in Huozhou Chinese was first reported in Satoh & Feng (1989), followed by X.Tian (1992), Hou & Wen (1993), J.Tian (2009), Shen, Feng & Tsumura (2010), and Feng & Zhao (2014), among others. The patterns of diminutive rime change are consistently described across the studies, despite some notational and sub-dialectal variations, as well as some minor divergences. Reduplication is another common way to form a diminutive in Chinese dialects and languages. Based on my consultation with informants, only nouns with ending ([j], [w], or [ŋ]) can undergo diminutive rime change in Huozhou Chinese, namely, nouns with a syllable shape other than (C)(G)Vː. The general pattern of diminutive rime change can be summarized as segment 61 subtraction with feature movement. In comparison, diminutive reduplication has no restrictions on syllable types. These observations are consistent with the description in previous literature. To illustrate the patterns of diminutive formation, I will start with ŋ-ending and j- ending nouns (§3.1.3.1), followed by w-ending nouns which show some special properties (§3.1.3.2). Then, some examples of nouns without endings and compound nouns will be introduced (§3.1.3.3). As I mentioned previously, the rime-changed forms are represented with a length mark to indicate bimoraicity since they still carry a full tone. For reduplication, the reduplicant is treated as a weak syllable and represented as monomoraic, which will be discussed shortly. 3.1.3.1. ŋ-ending and j-ending nouns Diminutive rime change in Huozhou Chinese is a regular process. After rime change, the last segment (glide or nasal) is dropped. In some cases, this is accompanied by featural changes of the nuclear vowel. This pattern is particularly clear for ŋ-ending and j-ending nouns, as exemplified in (6) and (7). 10 The examples throughout this section are surface representations. The reduplicant is underlined. Note that underlying /a/ surfaces as [ɑ] ([+back]) when it is followed by [ŋ] in a closed syllable or preceded by [w] in an open syllable, while it surfaces as [a] ([–back]) elsewhere. More details about backness agreement are in the appendix. 10 There is a divergence between the author’s consultation and a recent documentation in Feng & Zhao (2014). The final /-əŋ/ is sometimes changed to [-ɯ] in Feng & Zhao (2014:29) while the informants in this study produced [-u] instead. The variations need further investigation. 62 (6) Diminutive formation in Huozhou: ŋ-ending nouns Noun Rime-changed Reduplicated Gloss a. pʰɑŋ 35 pʰaː 35 pʰɑŋ 35 .pʰa 55 ‘plate’ b. lɑŋ 35 laː 35 lɑŋ 35 .la 55 ‘basket’ c. pʰjɑŋ 53 pʰjaː 53 pʰjɑŋ 53 .pʰja 11 ‘pieces’ d. tɕjɑŋ 33 – tɕjɑŋ 33 .tɕja 33 ‘scissors’ e. kwɑŋ 55 kwɑː 55 kwɑŋ 55 .kwɑ 33 ‘jar’ f. ɥɑŋ 53 ɥaː 53 – ‘garden, courtyard’ g. pʰəŋ 35 pʰuː 35 pʰəŋ 35 .pʰu 55 ‘basin’ h. kəŋ 11 kuː 11 kəŋ 11 .ku 33 ‘root’ i. tʰuŋ 33 tʰuː 33 tʰuŋ 33 .tʰu 33 ‘bucket’ j. tɕʰyŋ 35 – tɕʰyŋ 35 .tɕʰy 55 ‘skirt’ (7) Diminutive formation in Huozhou: j-ending nouns Noun Rime-changed Reduplicated Gloss a. saj 11 saː 11 saj 11 .sa 33 ‘sieve’ b. pʰaj 35 – pʰaj 35 .pʰa 55 ‘card’ c. kʰwaj 53 kʰwɑː 53 kʰwaj 53 .kʰwɑ 11 ‘piece, chunk’ d. kwaj 55 – kwaj 55 .kwɑ 33 ‘walking stick’ e. pej 53 puː 53 – ‘generation, lifetime’ f. pej 11 – pej 11 .pu 33 ‘cup, mug’ g. kʰwej 11 kʰuː 11 kʰwej 11 .kʰu 33 ‘bowl, container’ h. tʂʰwej 35 – tʂʰwej 35 .tʂʰu 55 ‘hammer’ In (6) and (7), every syllable becomes a (C)(G)Vː structure when rime change takes place. For reduplication, the examples above exhibit partial reduplication, i.e., only the part of the noun root excluding the ending is copied. Both rime change and reduplication are accompanied with certain segmental changes of features. In these examples, some nouns have both rime-changed and reduplicated forms, while for some others, only one form is commonly used by the informants. The gaps above are not likely to be phonologically conditioned, which is suggested by some highly similar pairs such as (7c) ~ (7d) and (7e) ~ (7f). Although not every noun has both rime-changed and reduplicated forms, the grammar should be able to generate both forms. The 63 nature of the gaps and the choice between the two forms need further investigation, but the current proposal and analysis will not be affected by these issues. Segment deletion is a typical characteristic of Huozhou diminutive rime change, which can be further verified by acoustic data, as shown in Figure 1. a. Non-diminutive [pʰɑŋ 35 ] b. Rime-changed [pʰaː 35 ] c. Reduplicated [pʰɑŋ 35 .pʰa 55 ] Figure 1. Spectrograms of [pʰɑŋ 35 ] ‘plate’ and its diminutive forms (produced by a 55-year-old female speaker) In Figure 1-a, there is a clear presence of the nasal coda; however, it is dropped in the rime- changed form (Figure 1-b). Also, the nasal coda is not copied when reduplication takes place (Figure 1-c). It is noteworthy that the acoustic data are not necessarily direct evidence of vowel length and moraic structures. Instead, the representation of [pʰaː 35 ] with a length mark (Figure 1-b) is a phonological assumption based on the previous studies of Chinese languages, as mentioned in §3.1.2. A comprehensive phonetic study is saved for future research. Regarding the segmental changes, mid vowel raising (and rounding) is a highlighted pattern in both rime change and reduplication. The low vowels (6a-f; 7a-d) and high vowels (6i-j) retain their height after rime change and reduplication, while mid vowels (6g-h; 7e-h) are raised to [u] 64 (e.g., [pʰəŋ 35 ] → [pʰuː 35 ]). These patterns will be analyzed as feature movement and revisited in §3.4.2. The rime-changed forms still bear the same lexical tone as their non-diminutive counterparts, and they can be used independently. Therefore, the rime-changed forms are viewed as heavy syllables on the surface, following the discussion in §3.1.2. For the reduplicated forms, the tonal patterns need some further explanation. In Standard Mandarin, tones are not copied in nominal reduplication, and the surface tone of the reduplicant is based on the tone of its preceding syllable (or the BASE), which is also true in Huozhou Chinese. This is why there appear to be tonal changes after reduplication. 3.1.3.2. w-ending nouns When a noun ends with a labiovelar glide [w], the patterns of diminutive formation show some special properties, as exemplified in (8): (8) Diminutive formation in Huozhou: w-ending nouns Noun Rime-changed Reduplicated Gloss a. pɑw 11 poː 11 poː 11 .po 33 ‘bag, purse’ b. tɑw 11 toː 11 toː 11 .to 33 ‘knife’ c. tɕʰjɑw 35 tɕʰɥoː 35 tɕʰɥoː 35 .tɕʰɥo 55 ‘stick’ d. tɕʰjɑw 53 tɕʰɥoː 53 – ‘sedan’ e. tʰow 53 tʰuː 53 tʰuː 53 .tʰu 11 ‘bean’ f. low 33 luː 53 luː 33 .lu 33 ‘basket’ g. ȵjow 35 – ȵyː 35 .ȵy 55 ‘insects’ h. ljow 35 – lyː 35 .ly 55 ‘glass ball’ For w-ending nouns, the rime-changed forms are still regular, with the appearance of segment deletion plus some subsegmental changes. Notably, these rime-changed forms exhibit stepwise raising in relation to the noun root, i.e., [ɑ] becomes [o], while [o] is raised to [u] or [y]. 11 This is different from the examples in (6) and (7), where a low vowel still remains low after rime change 11 There is one exception. The rime-changed form of /kəu/ ‘dog’ is [kɯː 33 ] rather than [kuː 33 ]. 65 (e.g., [saj 11 ] → [saː 11 ], [pʰɑŋ 35 ] → [pʰa 35 ]). The pattern of stepwise raising suggests a special treatment for w-ending nouns in formal analysis, which will be discussed in §3.3.3. For reduplication, the diminutive forms show full reduplication rather than partial reduplication, and it appears that both the base and the reduplicant undergo diminutive rime change, resulting in [poː 11 .po 33 ] rather than [pɑw 11 .po 33 ]. This pattern seems to be a case of backcopying, where the base is deviant from its input in order to conform to the copied portion. This property again makes w-ending nouns stand out, compared to those examples in (6) and (7). As mentioned earlier, this is the major focus of this chapter, and I propose that the identity effect here is enforced through Surface Correspondence. The details about this case will be discussed in §3.3.2. 3.1.3.3. Nouns without ending and compounds Since diminutive formation in the form of rime change only takes place for nouns that end with [ŋ, j, w], diminutive forms for nouns without an ending are always expressed through full reduplication, as exemplified in (9). (9) Diminutive formation in Huozhou: nouns without ending open-syllable nouns Noun Rime-changed Reduplicated Gloss a. tsɿː 11 – tsɿː 11 .tsɿ 33 ‘tree branch’ b. tsʰaː 11 – tsʰaː 11 .tsʰa 33 ‘pocket’ c. ŋɤː 35 – ŋɤː 35 .ŋɤ 55 ‘moth’ d. tɕjaː 11 – tɕjaː 11 .tɕja 33 ‘home’ e. xwɑː 11 – xwɑː 11 .xwɑ 33 ‘flower’ Finally, for some compounds, diminutive rime change only takes place on the second syllable (usually the head of the compound) if it ends with [j], [w], or [ŋ], while reduplication is not common. 12 See (10) for examples. 12 Some compounds do show reduplication as well. For example, the hypocoristic form of [je 35 .kɑw 55 ] ‘lamb’ is [je 35 .ko 55 ], and it can be also [je 35 .ko 55 .ko 33 ]. 66 (10) Diminutive formation in Huozhou: compounds. Noun Rime-changed Reduplicated Gloss a. tʂəŋ 55 .tʰow 33 tʂəŋ 55 .tʰuː 33 – ‘pillow’ b. pu: 55 .taj 33 pu: 55 .tʰaː 33 – ‘cloth bag’ c. zɿ: 35 .xwɑŋ 33 zɿ: 35 .xwɑː 33 – ‘ear ring’ d. ȵiŋ 35 .ɥɑŋ 55 ȵiŋ 35 .ɥaː 55 – ‘silver coin’ e. pɑŋ 11 .təŋ 33 pɑŋ 11 .tʰuː 33 – ‘stool’ f. tɕjow 33 .tsuŋ 11 tɕjow 33 .tsuː 11 – ‘shot glass’ g. wej 35 .tɕʰyŋ 55 wej 35 .tɕʰyː 55 – ‘apron’ 3.1.4. Summary of patterns Summing up, rime change and reduplication are two ways to form a diminutive in Huozhou Chinese. Rime change in Huozhou Chinese is a regular process where closed syllables are changed into open ones, and reduplication is used as an alternative way to form a diminutive, being partial or full depending on the type of syllable. Based on my consultation with informants, although there are no significant semantic differences between diminutive rime change and diminutive reduplication in most cases, the speakers mentioned a handful of pairs that do differ in meaning. For example, [mu 35 ], the rime-changed form of [məŋ 35 ] ‘door’, often refers to doors of houses/rooms, while the reduplicated form, [məŋ 35 .mu 55 ], specifically indicates the door of animal stalls (e.g., pigpen). However, this semantic difference is not prevalent among the data collected, nor has it been reported in previous literature (e.g., X.Tian 1992, Feng & Zhao 2014). The nature of the gaps and the choice between the two forms need further investigation, but the current proposal and analysis will not be affected by these issues. The patterns of diminutive formation are summarized in (11), where the ones in parentheses are marginal. For /-yŋ/, it is not as productive as previously reported. An example of rime change is found in a compound word, [wej 35 .tɕyː 55 ] ‘apron’ (10g) (underlying /uei 35 .tɕyŋ 35 /), compared to (6j) [tɕʰyŋ 35 .tɕʰy 55 ] ‘skirt’ with reduplication (note [tɕyː] in both words is the same morpheme). For /-iŋ/ and /-iəu/, their rime-changed forms have been documented in earlier literature (X.Tian 1992, Feng & Zhao 2014), but they were not reported by the consultants of the 67 current study; see (8g) and (8h). 13 Although the rime change process takes place in the rime domain, I list finals in (11) to show the potential influence of the prenuclear glide on vowel quality after rime change in several cases. Note that two possible rime-changed forms of /-iau/, [ɥoː] and [joː], are listed in (11), which I consider as free phonetic variations. Among all the non- open finals in (4), /-iai/ does not have any diminutive forms. 14 (11) Patterns of diminutive formation Rime change Reduplication Closed syllables ŋ-ending /aŋ/ → [aː] /iaŋ/ → [jaː] /uaŋ/ → [wɑː] /yaŋ/ → [ɥaː] /əŋ/ → [uː] (/iŋ/ → [iː]) /uŋ/ → [uː] (/yŋ/ → [yː]) partial reduplication (e.g., [pʰɑŋ 35 .pʰa 55 ]) j-ending /ai/ → [aː] /uai/ → [wɑː] /ei/ → [uː] /uei/ → [uː] w-ending /au/ → [oː] /iau/ → [ɥoː]/[joː] /əu/ → [uː] (/iəu/ → [yː]) full reduplication (e.g., [poː 11 .po 33 ]) Open syllables n.a. full reduplication (e.g., [xwɑː 11 .xwɑ 33 ]) The patterns of diminutive formation can be categorized as changes at the prosodic level and changes at the segmental level. At the prosodic level, the patterns can be described and generalized as follows. For closed syllables (except for /-iai/ final), both rime change and reduplication are possible ways to form a diminutive, as in (6), (7), and (8). The process of rime change in Huozhou transforms closed syllables into open ones, with the appearance of segment deletion. For reduplication, partial or full reduplication is chosen depending on the endings ([ŋ, j] versus [w]). In open syllables, full reduplication is used to form a diminutive while no rime change is observed, as in (9). At the segmental level, there are also regular phonological processes, as sketched in previous sections, and the most highlighted pattern is vowel raising. In both rime change and 13 In Feng & Zhao (2014:29), /-iŋ/ is changed to [-iː] or [-iɯ]; /-iəu/ is changed to [-iɯ]. 14 The syllables with [-jaj] final are very rare in general in Huozhou Chinese. 68 reduplication, w-ending nouns show stepwise vowel raising for both low and mid vowels, while ŋ-/j-ending nouns show mid vowel raising only. In addition, w-ending nouns show full reduplication, and the vowel quality of the base is changed to conform to that of the reduplicant, which differs from ŋ-/j-ending nouns. So far in the discussion, the description does not entail any theoretical assumptions about how those forms are generated in grammar. However, in some previous literature (e.g., X.Tian 1992), the description implies that partial reduplication results from a serial application of full reduplication and rime change, e.g., [pɑŋ 35 ] → [pɑŋ 35 .pɑŋ 55 ] → [pɑŋ 35 .pa 55 ]. In addition, full reduplication of w-ending nouns is described as “multiple application of rime change”, i.e., [pɑw 11 ] → [pɑw 11 .pɑw 33 ] → [poː 11 .po 33 ]. However, this idea will run into some problems in formal analyses, which will be revisited in §3.6.1.2. The patterns of Huozhou diminutive formation involve various aspects of prosodic and segmental changes, as summarized above. Among the data introduced above, there are two highlighted patterns of this morphophonological process. First, the diminutive form can be either rime change or reduplication, and rime change can be viewed as a case of subtractive morphology. Second, diminutive reduplication shows a case of truncatory backcopying. The reduplicant in Huozhou Chinese is always an open syllable, while the shape of the reduplicant is carried over into the base when a noun ends with [w], as in (8). These two patterns are the focus of this case study, which will be examined in detail in the following sections. 3.2. Theoretical preliminaries This section provides additional theoretical background for the following analyses. The proposal is that the underlying phonological form of Huozhou diminutive marker is a mora, which can be realized into either reduplication or subtraction scenarios. In the following discussion, I provide the rationale of the proposal and briefly show different surface forms can be attributed to mora affixation. 69 Diminutive formation in Huozhou Chinese is derivational rather than inflectional, and it can be treated as the affixation of a DIMINUTIVE morpheme in morphosyntax. As discussed above, two processes are variably involved, rime change and reduplication. There is a long history of analyzing reduplication as concatenation of an element, either an abstract RED (McCarthy & Prince 1995) or a prosodic template (e.g., McCarthy 1981, Saba Kirchner 2010, 2013, McCarthy, Kimper & Mullin 2012). Rime change in Huozhou Chinese, on the other hand, can be viewed as a case of subtractive morphology, in which the coda or offglide is deleted for morphological purposes. The key proposal of the current analysis is that the underlying phonological representation of the diminutive morpheme is a floating mora, and both reduplication and segment subtraction (rime change) are ways to repair the deficient mora. The treatment of reduplication as resolving deficient prosodic templates is in line with Minimal Reduplication (Saba Kirchner 2010, 2013, see also Bermúdez-Otero 2012, Bye & Svenonius 2012) and Serial Template Satisfaction (McCarthy et al. 2012). In these approaches, a prosodic template is populated by copying the existing segmental materials. The idea that subtractive morphology can be re-analyzed as affixation is in accordance with the proposals of Bye & Svenonius (2012), Trommer (2011), Trommer & Zimmermann (2014), Zimmermann (2017a), and Köhnlein (2018). Although previous analyses involve different theoretical apparatus (e.g., Colored Containment versus Correspondence Theory), the core insight is the same, i.e., subtractive morphology can emerge as a way of realizing the affixal feature or prosodic node to satisfy phonological wellformedness. The differences between the current proposal and the one in Trommer & Zimmermann (2014) will be discussed in §3.6.2.1. With the underlying affixal mora being proposed, the representations of Huozhou diminutive formation are as follows. The superscript digits indicate input-output segment correspondence rather than tones. For the reduplicated form in (12b), the digits are used to keep track of segment fission, i.e., the segments with the identical index are fissioned from the same 70 segment in the input. The alternation between [a] ([–back]) and [ɑ] ([+back]) is attributed to local assimilation of backness under the influence of [ŋ], which will be revisited in the appendix. (12) Diminutive formation as mora affixation (tones are not marked) Input Output a. Reduplication b. Subtraction → /lɑ μ ŋ μ + μ/ [lɑ μ ŋ μ .la μ ] [laː μμ ] In (12), as well as the analyses below, the phonological form of the diminutive morpheme is indicated by a mora in bold and italic for the sake of readability. There are two ways to realize the floating mora with segmental contents. For (12a), it is populated by segment fission; 15 in (12b), the affixal mora is realized by linking to the nuclear vowel, while triggering segment (and mora) deletion. Although not every noun has both forms, as discussed earlier, the gaps are less likely to be phonologically conditioned, and certain forms are possibly blocked for reasons other than phonological wellformedness (e.g., some reduplicated forms sound childish). Therefore, I assume that the phonological grammar should be able to predict both reduplication and subtraction for every input. I will show how these forms can be generated through constraint ranking in §3.3.1 and §3.4.1. The phenomenon of lexical exceptionality in these datasets will be addressed in future research. The variation is proposed to be dealt with by Partially Ordered Constraints (Anttila 1997, 2002, 2006) (see §3.4.1 and §3.5 for details). 15 Interestingly, Shen et al. (2010) described that the reduplicant in diminutives is lengthened. I speculate that this might be due to how the data were elicited. According to my consultation, the reduplicant is perceived to be longer if the words are pronounced in isolation, which is probably due to a final lengthening effect. However, there does not seem to be obvious lengthening when the words are embedded in a carrier sentence. This issue needs further acoustic investigation, but it will not cause any serious problems to the current formal analysis. Even if the reduplicant is lengthened and needs to be represented with two moras, e.g., [lɑ μ ŋ μ .laː μμ ], the additional mora in the reduplicant can be motivated and epenthesized by grammar rather than underlyingly specified. l 1 ɑ 2 ŋ 3 µ σ + µ µ l 1 ɑ 2 ŋ 3 µ σ µ µ σ l 1 a 2 l 1 aː 2 µ σ µ 71 The illustration in (12) suggests that the prosodic structure of the root has been established at the stage of mora affixation, which is a common assumption in the approaches where reduplication is viewed as prosodic affixation. Various strategies have been adopted to ensure that prosodification occurs at an earlier stage. For instance, Stratal OT (e.g., Kiparsky 2000, 2003, 2010, 2015; Trommer 2011; Bermúdez-Otero 2018 a.o.) is used in Bermúdez-Otero (2012) (see also Trommer & Zimmermann 2014). In Serial Template Satisfaction (McCarthy et al. 2012), the prosodic template is affixed to a structure that has been fully prosodified at an earlier step in Harmonic Serialism (e.g., McCarthy 2000b, 2007). In the current analysis, I resort to Stratal OT to model the layering effect, and roots are pre- optimized and prosodified in an early stratum during word construction. The basic tenet of Stratal OT is that word construction includes multiple levels that are organized in a feed-forward way. Each level has its own phonological grammar, which is a parallel constraint ranking system as in classic OT. Specifically, mora assignment of noun roots in Huozhou is attributed to the grammar applied at the root level, before diminutive formation takes place (see Trommer 2011 for the proposal of root-level stratum). I will refer to this stratum as level 0. At the root level, I assume that moras are assigned to coda consonants/offglides through WEIGHT-BY-POSITION (cf. Hayes 1989). In addition, since no underlying contrastive long vowels are assumed in Huozhou Chinese, and all full syllables are heavy to carry their underlying lexical tones, I use a constraint *FULLTONE/σ μ to lengthen underlying short vowels. 16 These constraints are defined in (13), followed by a tableau in (14) to demonstrate moraification. In this case, mora assignment is handled by a grammar that consists of the abovementioned constraints, and it is applied to the root level (the innermost) during morphological construction. 16 This constraint is motivated by a set of principles about weight, stress, and tone in the literature. Duanmu’s (1993) study argues that only a stressed syllable can retain its underlying tones, and a Tone-Stress Principle is proposed in Duanmu (2007:249), stating that “a stressed syllable can be assigned a lexical tone or pitch accent; an unstressed syllable is not assigned a lexical tone or pitch accent”. In addition, the Stress-to-Weight Principle (Myers 1987, Riad 1992, Kager 1999) requires stressed syllables to be heavy (bimoraic). Thus, in Chinese, a syllable that bears a full tone needs to be heavy/bimoraic. 72 (13) Constraints for mora assignment a. WEIGHT-BY-POSITION (WBP): Assign a violation for every consonant/offglide in the rime that is not moraic. b. *FULLTONE/σ μ : Assign a violation for every light syllable that is associated with a full lexical tone. c. DEP-μ: Assign a violation for every mora in the output that does not have a correspondent in the input. (14) Mora assignment of Huozhou Chinese (level 0; every syllable is associated with a full tone) 17 kai 55 WBP *FULLTONE/σ μ DEP-μ ☞ a. ka μ j μ 55 2 b. kaː μμ j μ 55 3W c. kaː μμ j 55 1W 2 pʰaŋ 35 WBP *FULLTONE/σ μ DEP-μ ☞ d. pʰɑ μ ŋ μ 35 2 e. pʰɑː μμ ŋ μ 35 3W f. pʰɑː μμ ŋ 35 1W 2 sa 55 WBP *FULLTONE/σ μ DEP-μ ☞ g. saː μμ 55 2 h. sa μ 55 1W 1L At the root level, moras are assigned through constraint interaction. The candidates (14c) and (14f) both violate WBP, although both are heavy and satisfy *FULLTONE/σ μ . The trimoraic candidates (14b) and (14e) incurs a more severe violation of DEP-μ, as well as *σ μμμ , which will be introduced in the following section. For (14g), the vowel is assigned with two moras to make it able to bear fthe lexical tone, and (14h) violates *FULLTONE/σ μ . To keep the analysis concise and focused, I will not show the process of morification in the following analyses, and the input in the tableaux hereafter will contain a noun root with moraic structure plus an affixal mora as the exponent of the diminutive morpheme, resembling the 17 I assume that some syllable-internal optimization, such as backness assimilation of underlying /a/, also takes place at this level. 73 input structure of (12). In §3.3 and §3.4, I will show how the proposed representations along with other theoretical mechanisms can generate both reduplication and subtraction. 3.3. Reduplication in Huozhou: mora affixation, backcopying, and vowel raising This section focuses on reduplication in Huozhou Chinese, the major argument for the utility of Surface Correspondence in reduplicative opacity. Below in (15), some key data are repeated to illustrate the main patterns that are to be accounted for in this section. I give the underlying representation of the non-diminutive forms to better illustrate the change of vowel quality. (15) Diminutive reduplication in Huozhou Non-diminutive Reduplicated a. [pɑw 11 ] /pau 11 / [poː 11 .po 33 ] ‘bag’ w-ending b. [tʰow 53 ] /tʰəu 53 / [tʰuː 53 .tʰu 11 ] ‘bean’ c. [saj 11 ] /sai 11 / [saj 11 .sa 33 ] ‘sieve’ j-ending d. [pʰɑŋ 35 ] /pʰaŋ 35 / [pʰɑŋ 35 .pʰa 55 ] ‘plate’ ŋ-ending There are three remarkable patterns in these data. First, an important property of diminutive reduplication is truncatory backcopying (15a-b). For w-ending nouns, the root also undergoes featural changes to conform to the copied portion, i.e., [poː 11 .po 33 ] (*[pɑw 11 .po 33 ]), giving an appearance that the shape of the reduplicant is copied back to the root. This pattern is not found in j-ending and ŋ-ending nouns (15c-d) (*[saː 11 .sa 33 ], *[pʰaː 35 .pʰa 55 ]). Thus, I call this case a conditional truncatory backcopying. Second, there is an asymmetry between w-ending nouns and the others in terms of low vowel raising, i.e., the low vowel is raised in the reduplicant only when it is followed by [w] (/u/) in the input (15a), in contrast to (15c) and (15d). Third, the low vowel followed by [w] (/u/) shows a case of stepwise raising: it is only raised to mid (15a). There are three major goals in this section with regard to the data. First, §3.3.1 shows how reduplication is produced with mora affixation. Second, in §3.3.2, I will show that Surface Correspondence and the IDENT-XX constraints with proper conditions can motivate or block the 74 identity effect for different inputs, resulting in the truncatory backcopying pattern of w-ending nouns. Finally, the issues of vowel raising will be discussed in §3.3.3. As a preview, the patterns and relevant sections are summarized in (16), which will also appear in §3.3.4. (16) Preview of the analysis Main patterns explained Relevant Section a. Truncatory backcopying (w-ending nouns) §3.3.2 b. Truncatory backcopying blocked (j-/ŋ-ending nouns) §3.3.2 c. Stepwise vowel raising (w-ending nouns) §3.3.3 d. Issues of vowel quality change (j-/ŋ-ending nouns) §3.3.3 3.3.1. Reduplication as mora affixation The analysis starts from diminutive reduplication. Reduplication is a common way to realize an affixal prosodic template, where the input deficient prosodic units is populated by copying the existing segments. The idea was proposed as early as The Prosodic Morphology Hypothesis in McCarthy & Prince (1986/1996), and it was developed in Saba Kirchner’s (2010, 2013) Minimal Reduplication with a constraint-based implementation (cf. McCarthy et al. 2012). The common prosodic templates for reduplication include mora, syllable, and foot. A deficient prosodic unit in the input is considered as marked. The general driving force to repair such a marked structure is the constraint HEADEDNESS(μ), which belongs to the constraint family HEADEDNESS(X) (Selkirk 1995). This constraint requires that every mora in the input have segmental content in the output. In addition, the presence of the affixal mora in the output is enforced by MAXFLT (Wolf 2006). These constraints are defined in (17). (17) a. MAXFLT: All autosegments that are floating in the input have output correspondents. b. HEADEDNESS(μ): Assign a violation mark for every mora in the input that does not dominate a segment. For an input such as /lɑ μ ŋ μ + μ/, if the floating mora is deleted as a way to vacuously satisfy HEADEDNESS(μ), a violation of MAXFLT will be incurred. Note that MAXFLT is not equivalent to MAX- 75 μ, which evaluates the general input-output mapping of moras. In the case of an underlying floating mora, one violation of MAXFLT entails one violation of MAX-μ, but not vice versa. The constraint MAX-μ will not be shown in the following tableaux. In order to satisfy HEADEDNESS(μ), a straightforward way is to dock the floating mora on an existing segment, which would cause vowel lengthening. This scenario will violate *σ μμμ that penalizes a super-heavy trimoraic syllable (e.g., McCarthy 2005), defined in (18). Note that the floating mora could also replace the mora on the coda. This issue will be addressed later in §3.4. (18) *σ μμμ : Assign a violation mark for any trimoraic syllable. Under the pressure of MAXFLT, HEADEDNESS(μ), and *σ μμμ , reduplication can be selected as the repair strategy to the floating mora, at the expense of violating INTEGRITY-S. Since the copying of segments is viewed as segment fission, INTEGRITY-S is violated by multiple correspondence relations between input and output segments. The constraint is defined as follows (after McCarthy & Prince 1995:124): (19) INTEGRITY-S: Assign a violation mark for every segment in the input that has multiple correspondents in the output. The interaction between MAXFLT, HEADEDNESS(μ), *σ μμμ , and INTEGRITY-S is able to drive reduplication, as in (20). The segments are subscripted to indicate input-output correspondence. Again, I assume that prosodification of the noun root has been completed in the relevant stratum (i.e., level 0, recall §3.2.1) and the input noun root is assigned with two moras in the following tableau. Further, note that the local agreement on backness in the input (i.e., agreement between [ɑ] and [ŋ]), as well as in the candidates, has been done in the previous level. Tones are not marked throughout the analysis. 76 (20) Input /lɑ μ ŋ μ + μ/, ‘basket + DIM’ l 1 ɑ 2 μ ŋ 3 μ + μ MAXFLT HEADEDNESS(μ) *σ μμμ INTEGRITY-S ☞ a. l 1 ɑ 2 μ ŋ 3 μ .l 1 a 2 μ 2 b. l 1 ɑ 2 μμ ŋ 3 μ 1W L c. l 1 ɑ 2 μ ŋ 3 μ + μ 1W L d. l 1 ɑ 2 μ ŋ 3 μ 1W L The floating mora is realized as reduplication in (20a) at the expense of INTEGRITY-S, since two segments are copied from the input. The candidate in (20c) leaves the floating mora deficient while the one in (20d) deletes the floating mora, and therefore, they are ruled out by HEADEDNESS(μ) and MAXFLT respectively. In addition, a highly-ranked DEP-S can block segment epenthesis as another repair strategy to the floating mora (e.g., *[l 1 ɑ 2 μ ŋ 3 μ .l 1 ə 4 μ ]), which is omitted here. In (20a), the vowel [a] with [–back] is due to the markedness restriction that low back vowel only cooccurs with velar consonants or labiovelar glide. This markedness restriction can be formalized as a constraint *ɑ] σ , namely, a back low vowel is penalized in an open syllable, and it should dominate IDENT-IO(back). One issue in the analysis is how to ensure the copying of onset. In the example above, another loser candidate [l 1 ɑ 2 μ ŋ 3 μ .a 2 μ ] should be considered. Also, if the input contains [k 1 w 2 ɑ 3 μ ŋ 4 μ ] ‘jar’, both [k 1 ] and [w 2 ] need to be present in the copied string [k 1 w 2 ɑ 3 μ ŋ 4 μ .k 1 w 2 ɑ 3 μ ]. To solve this issue, I resort to ONSET and a faithfulness constraint that regulates string contiguity, following the treatment in McCarthy & Prince (1995:12-13) and Bye & Svenonius (2012:453-457). The relevant constraints are defined in (21). (21) Copying of onset a. ONSET: Assign a violation for every syllable that does not have an onset. b. I-CONTIGUITY: Let S be a string in the input. Assign a violation if the portion of S standing correspondence does not form a contiguous string in the output. (after McCarthy & Prince 1995:123) “No skipping is allowed.” 77 The constraint ONSET motivates the copy of onset consonant. I-CONTIGUITY enforces contiguity in the corresponding output string. It is violated by an input-output mapping such as /a 1 b 2 c 3 / → [a 1 c 3 ]. The effects of these two constraints are shown in (22). (22) Input /kwɑ μ ŋ μ + μ/ ‘jar + DIM’ k 1 w 2 ɑ 3 μ ŋ 4 μ + μ I-CONTIGUITY ONSET INTEGRITY-S ☞ a. k 1 w 2 ɑ 3 μ ŋ 4 μ .k 1 w 2 ɑ 3 μ 3 b. k 1 w 2 ɑ 3 μ ŋ 4 μ .a 3 μ 1W 1L c. k 1 w 2 ɑ 3 μ ŋ 4 μ .k 1 a 3 μ 1W 2L The winner, (22a), satisfies both I-CONTIGUITY and ONSET, although it copies the most segments among the candidates, which incurs three violations of INTEGRITY-S. Candidate (22b) does not copy the onset so it is ruled out by ONSET. Candidate (22c) violates I-CONTIGUITY by skipping [w] in the output. The alternation between [a] and [ɑ] remains the same as discussed earlier in (20). Additionally, Since this language allows onsetless syllables in general, the constraint ONSET needs to be dominated by DEP-S. Thus, the ranking DEP-S >> ONSET >> INTEGRITY-S indicates that onsetless syllables can be only repaired by segment fission. 18 The constraint DEP-S will not be included in the analysis for conciseness. The input of (22) brings another problem. Since the surface glides are treated as underlying vowels, it is possible to use /w/ in the input to populate the floating mora, producing *[k 1 w 2 ɑ 3 μ ŋ 4 μ .k 1 u 2 μ ]. In addition, it is also possible to produce an output *[k 1 w 2 ɑ 3 μ ŋ 4 μ .w 2 ɑ 3 μ ], where [w] serves as the onset. These candidates will satisfy both ONSET and I-CONTIGUITY, and they violate INTEGRITY-S only twice. This problem can be solved by limiting a segment’s syllable role through IDENT-IO-SROLE (after McCarthy & Prince 1993, 1994a, Suzuki 1999, Gafos 1996, 1998, Rose & Walker 2004, Bye & Svenonius 2012). 19 18 For a hypothetical input [aŋ], this ranking would predict [a 1 ŋ 2 .ŋ 2 a 1 ŋ 2 ] after reduplication, where the velar nasal is copied to fulfill the onset position. This candidate can be ruled out by IDENT-IO-SROLE introduced in (23) since [ŋ] in onset position is copied from a coda. 19 This constraint is defined in the input-output domain rather than between surface segments. 78 (23) IDENT-IO-SROLE: For every pair of segments that stand in input-output correspondence, assign a violation if they do not have identical syllable roles. Although the role of a prenuclear glide in Chinese syllables is controversial, it clearly does not serve as the role of the nucleus. Therefore, the corresponding pair [w 2 ~ u 2 ] in *[k 1 w 2 ɑ 3 μ ŋ 4 μ .k 1 u 2 μ ], as well as [w 2 ] in *[k 1 w 2 ɑ 3 μ ŋ 4 μ .w 2 ɑ 3 μ ], fatally violates IDENT-IO-SROLE. This constraint needs to be ranked above INTEGRITY-S. Given the assumption made in §3.3.2, prosodification has been done before affixation. Thus, the syllable structure of each candidate is shown in the tableau. (24) Input /kwɑ μ ŋ μ + μ/ ‘jar + DIM’ k 1 w 2 ɑ 3 μ ŋ 4 μ + μ ID-IO-SROLE INTEGRITY-S ☞ a. k 1 w 2 ɑ 3 μ ŋ 4 μ .k 1 w 2 ɑ 3 μ 3 b. k 1 w 2 ɑ 3 μ ŋ 4 μ .k 1 u 2 μ 1W 2L c. k 1 w 2 ɑ 3 μ ŋ 4 μ .w 2 ɑ 2 μ 1W 2L In sum, the constraint interactions so far are demonstrated in the Hasse diagram below: (25) Hasse diagram 1. (20a) ≻ (20b), (20c), (20d); 2. (22a) ≻ (22b), (22c); (24a) ≻ (24b) Due to the low ranking of INTEGRITY-S, the pressure of realizing the floating mora is resolved by reduplication. Although there are other ways to avoid a headless floating mora, those repair strategies will be suppressed by ranking relevant faithfulness constraints higher than INTEGRITY- S. For example, the floating mora can be also docked on an epenthetic segment, which will violate the higher-ranked DEP-S, as mentioned earlier. MAXFLT *σ µµµ HEADEDNESS(µ) INTEGRITY-S ONSET CONTIGUITY IDENT-SROLE 1 2 79 3.3.2. Backcopying driven by SCorr The most prominent process during diminutive reduplication is the backcopying pattern, especially for w-ending nouns. Some examples are repeated in (26). (26) Backcopying in diminutive reduplication Non-diminutive Rime-changed Reduplicated a. [pɑw 11 ] /pau 11 / [poː 11 ] [poː 11 .po 33 ] ‘bag’ w-ending b. [tʰow 53 ] /tʰəu 53 / [tʰuː 53 ] [tʰuː 53 .tʰu 11 ] ‘bean’ c. [saj 11 ] /sai 11 / [saː 11 ] [saj 11 .sa 33 ] ‘sieve’ j-ending d. [pʰɑŋ 35 ] /pʰaŋ 35 / [pʰaː 35 ] [pʰɑŋ 35 .pʰa 55 ] ‘plate’ ŋ-ending A remarkable property of diminutive reduplication is truncatory backcopying (26a-b), which is the focus of the current section. For w-ending nouns, the root also undergoes changes to conform to the copied portion, i.e., [poː 11 .po 33 ] (*[pɑw 11 .po 33 ]), giving an appearance that the shape of the reduplicant is copied back to the root. This pattern is not found in j-ending or ŋ- ending nouns (26c-d) (*[saː 11 .sa 33 ], *[pʰaː 35 .pʰa 55 ]). Thus, I call this case a conditional truncatory backcopying. Although, at first sight, (26a) and (26b) appear to be the product of an ordered application of rime change and reduplication, some issues will arise with this analysis, which will be discussed in §3.5 and §3.6. For the patterns in (26), I will discuss w-ending nouns and j- /ŋ-ending nouns separately. Again, the core purpose of this section is to argue for the role of Generalized Surface Correspondence in handling backcopying in Huozhou diminutive reduplication. The backcopying pattern in Huozhou reduplication is interpreted as a case of long-distance assimilation, which can be managed by Surface Correspondence. Reduplication in Huozhou exhibits truncatory backcopying with segment deletion, since the syllable shape is also identical between these two strings (both are C(G)V). Meanwhile, this case of truncatory backcopying is conditional because only w-ending nouns trigger such a process. In this analysis, long-distance agreement across syllable boundaries is attributed to Generalized Surface Correspondence, more specifically, the enforcement CORR-XX and IDENT-XX. 80 In this case, I further elaborate the original definition of IDENT-XX in Chapter 2. I adopt Krämer’s (2003) proposal that syntagmatic identity can be enforced between moraic elements, which is implemented as IDENT-XX [μ, 𝛾 G] (F). The purpose of limiting surface identity over moraic segments is to exclude the onglide from triggering any unexpected assimilation (recall §3.1.2 for the discussion of Chinese syllable structure), which will be detailed in the following sections. The definitions of relevant constraints are given in (27). (27) Constraints: a. CORR-XX: Assign a violation to any pair of segments in the stem domain that are not in correspondence in the output. b. IDENT-XX [μ, +back] (round): Let X and Y be a consecutive pair of corresponding [+back] segments that are moraic in the output. Assign a violation if X and Y do not match in value for [round]. c. IDENT-XX [+round] (high): Let X and Y be a consecutive pair of corresponding [+round] segments in the output. Assign a violation if X and Y do not match in value for [high]. d. IDENT-IO(F): Let 𝛼 be a segment in the input and 𝛽 be any correspondent of 𝛼 in the output. Assign a violation if 𝛼 and 𝛽 do not match in feature [F]. In (27), CORR-XX impels SCorr between segments; the constraints IDENT-XX [μ, +back] (round) and IDENT-XX [+round] (high) drive the corresponding segments to be identical, both of which are feature-restricted identity constraints in the format proposed in Walker (2015). Note that all the limiter constraints are evaluated locally over adjacent pairs. The joint effect of SCorr and MAX-S is able to account for the patterns in (26a) and (26b), illustrated in (28). The constraint INTEGRITY-S is omitted. Notation-wise, the superscript digits indicate input-output segmental correspondence, while the segments that share the same subscript letters stand in Surface Correspondence (the moraic structures are not shown). As a reminder, the input has undergone an earlier level for mora assignment and syllable-internal optimization. 81 (28) Input /pɑ μ w μ + μ/ ‘bag + DIM’: backcopying pɑ 1 w 2 + μ CORR- XX ID-XX [μ, +bk] (rd) ID-XX [+rd] (hi) MAX- S ID (lo) ☞ a. p x oː x 1 .p x o x 1 1 2 b. p x o x 1 w x 2 .p x o x 1 2W L 2 c. p x ɑ x 1 w x 2 .p x a x 1 1W L L d. p x ɑ x 1 w y 2 .p x a x 1 4W L L e. p x ɑ x 1 w x 2 .p x ɑ x 1 w x 2 3W L L The crucial comparisons in this analysis are between (28a) versus (28b) and (28c), all of which satisfy CORR-XX, and for which the evaluation of the IDENT-XX constraints plays an essential role. Candidate (28b) violates ID-XX [+rd] (high), which requires that [+round] segments be identical in height. In this candidate, the segments with [+round] in the chain include [o x 1 ~ w x 2 ~ o x 1 ], and ID-XX [+rd] (high) is violated twice by [o x 1 ~ w x 2 ] and [w x 2 ~ o x 1 ]. The candidate in (28c) violates ID- XX [μ, +bk] (round) once because the pair [ɑ x ~ w x ] in the chain agrees on backness but does not match in roundness. This candidate is exempt from ID-XX [+rd] (high) since only [w x 2 ] in the chain meets the condition [+round]. Instead, the winner (28a) uses segment deletion as a way to satisfy surface identity due to the ranking schema CORR-XX, IDENT-XX >> MAX-S, IDENT-IO(F); the constraint CORR-XX, which requires that any two segments stand in correspondence, rules out (28d). Finally, the candidate (28e) violates ID-XX [μ, +bk] (round) three times by [ɑ x 1 ~ w x 2 ], [w x 2 ~ ɑ x 1 ], and [ɑ x 1 ~ w x 2 ] since none of the members in each pair agrees on roundness. Also, as a repetition of §3.3.1, a back low vowel is penalized in an open syllable (*ɑ] σ ). Thus, the candidates in (28c-d) end with a [–back] low vowel [a]. Another candidate, [p x a x 1 w x 2 .p x a x 1 ], is worthy of mention, though it is not included in the tableau. This candidate escapes violating ID-XX [μ, +bk] (round) due to the lack of adjacent [+back] pairs in the chain. However, it is ruled out since [a x 1 ~ w x 2 ] violates the local assimilation requirement. More specifically, segments in the rime domain of a Chinese syllable need to agree on backness (i.e., [pɑw] rather than *[paw]). Local assimilation can be also achieved by Surface 82 Correspondence (e.g., Shih & Inkelas 2014). The parameters of IDENT-XX [𝛾 G] (F) can be further modified to limit the evaluation to local segments only. Here I follow the notation of Shih & Inkelas (2019:140) and propose a variant of the limiter constraint, IDENT-X::X [𝛾 G] (F), defined in (29). (29) IDENT-X::X [𝛾 G] (F) Assign a violation for every consecutive pair of corresponding segments (X 1 , X 2 ) if: a. X 1 and X 2 are both [𝛾G]; and b. X 1 and X 2 are immediately adjacent; and c. X 1 is [𝛼F] and X 2 is [𝛽F] (where 𝛼 ≠ 𝛽). The pair of segments [a x 1 ~ w x 2 ] in the candidate *[p x a x 1 w x 2 .p x a x 1 ] will be ruled out by IDENT- X::X [μ] (back) since they are immediately adjacent moraic segments but do not agree on [back]. Similarly, the candidate [p x e x 1 w x 2 .p x e x 1 ] fatally violates various local assimilation constraints, including IDENT-X::X [μ] (back) and IDENT-X::X [μ] (round), although it satisfies all the IDENT-XX constraints in the tableau. More details on local assimilation with can be found in the appendix of this chapter. Partial reduplication is found in j-/ŋ-ending nouns, and the backcopying pattern is blocked. Some examples are repeated in (30). (30) Partial reduplication: j-/ŋ-ending nouns Non-diminutive Reduplicated a. [saj 11 ] /sai 11 / [saj 11 .sa 33 ] ‘sieve’ b. [pʰɑŋ 3 ] /pʰaŋ 35 / [pʰɑŋ 35 .pʰa 55 ] ‘plate’ In these examples, backcopying is not observed. The grammar established so far is able to account for partial reduplication of the j-/ŋ-ending nouns with a low nuclear vowel. An example is given in (31). The constraint INTEGRITY is low-ranked. 83 (31) Input /sa μ j μ + μ/ ‘sieve + DIM’ sa 1 j 2 + μ CORR-XX ID-XX [μ, +bk] (rd) ID-XX [+rd] (hi) MAX-S ID (hi) ID (lo) INTEG ☞ a. s x a x 1 j x 2 .s x a x 1 2 b. s x aː x 1 .s x a x 1 1W 2 c. s x a x 1 j y 2 .s x a x 1 4W (CORR-XX) 2 The ranking of these constraints remains the same as shown in (28) and (31). Due to the lack of round segments, ID-XX [+rd] (high) is vacuously satisfied. Also, since all the corresponding segments that have a role of rime are [–round], ID-XX [μ, +bk] (round) is satisfied. The candidate in (31c) is ruled out by CORR-XX. For (31a) and (31b), although both of them satisfy the top-tier constraints, the candidate in (31a) is favored over (31b) by MAX-S. As shown earlier, the main driving force of deletion is the ID-XX constraints, and both of them are not in action when the corresponding segments do not meet the conditions. Further, since the main motivation of vowel raising, ID-XX [+rd] (high), does not exert any effect, some other losing candidates with vowel raising such as *[saj.su] are harmonically bounded by (31a) due to their violations of relevant IDENT-IO(F). Summing up, this analysis shows the joint effect of CORR-XX and IDENT-XX. Another constraint discussed in Chapter 2, ACCORD, is actually ranked low and does not play a role here. The reason is that this pattern requires correspondence between segments that are not fissioned from the same input segment. In particular, the crucial motivation of truncatory backcopying in this case is that [w] and other vocalic segments are put in the same correspondence chain defined by a certain feature, such as [o x 1 ~ w x 2 ~ o x 1 ]. A way to fixed this ill-formed SCorr structure is to delete a member by ranking MAX-S low. In addition, this process is blocked for j-/ŋ-ending nouns that lack [+round] or [+back] feature. The analysis so far concentrates on how truncatory backcopying is predicted by SCorr. Further details about the featural changes (e.g., [poː.po] rather than *[puː.pu]) will be presented in Section 3.3. 84 3.3.3. Issues of vowel raising This section turns to some details about the featural changes during reduplication. Recall the data in (32). (32) Featural changes in diminutive reduplication Non-diminutive Reduplicated a. [pɑw 11 ] /pau 11 / [poː 11 .po 33 ] ‘bag’ w-ending b. [tʰow 53 ] /tʰəu 53 / [tʰuː 53 .tʰu 11 ] ‘bean’ The analysis so far does not explain why the low vowel in (32a) is raised to [o] while the mid vowel in (32b) is raised to [u]. To solve this issue, I adopt the autosegmental theory of featural faithfulness, where features are viewed as autosegmental elements and stand in input-output correspondence. I propose that the nuclear vowel raising is attributed to the preservation of the [+high, +round] features of the original labiovelar glide. Featural correspondence and preservation are regulated by MAX(feature), which was first proposed by McCarthy & Prince (1995), and was further discussed by Lombardi (1995/2001), Causley (1997), McCarthy (2000), Zhang (2000), Walker (2001), McCarthy (2008), etc. Although there has been some debate on whether MAX(feature) can obviate IDENT-IO(feature), some studies show the utility of both types of featural faithfulness in grammar (e.g., Walker 2001, Wolf 2006, cf. Zhang 2000). (33) MAX(feature): Assign a violation for every input feature [F] that does not have a correspondent in the output. The idea that vowel change is due to the preservation of features in [w] can be further illustrated in (34). When [w] is deleted, the [+round] and/or [+high] features are retained by MAX(+round) and MAX(+high), resulting in rounding and raising of the input low vowel. More details will be presented. 85 (34) Segment deletion with feature movement a. [poː] (from [pɑw]) b. [tʰuː] (from [tʰow]) Further, the issue of stepwise raising needs additional treatment. To block the raising from low to high, I make use of local conjunction (e.g., Smolensky 1993, Itô & Mester 1998, Łubowicz 2002, Smolensky 2006) which has been used in analyzing stepwise raising (e.g., Kirchner 1996; Walker 2005, 2011). Local constraint conjunction forms complex constraints from simple ones. I follow the theoretical assumptions of Itô & Mester (1998:10): (35) Local Conjunction of Constraints (definition from Itô & Mester 1998:10) a. Definition Local conjunction is an operation on the constraint set forming composite constraints: Let C 1 and C 2 be members of the constraint set CON. Then their local conjunction C 1 &C 2 is also a member of CON. b. Interpretation The local conjunction C 1 &𝛿C 2 is violated iff both *C 1 and *C 2 are violated in the same domain 𝛿. c. Ranking (universal) C 1 &𝛿C 2 >> {C 1 , C 2 } The composite constraint in effect for Huozhou raising is IDENT-IO(high)& S IDENT-IO(low). The domain of this conjoined constraint is segment, indicated by S. p o µ σ µ <w> µ [+round] [+high] t h u µ σ µ <w> µ [+round] [+high] 86 (36) IDENT-IO(high)& S IDENT-IO(low): Assign a violation mark if IDENT-IO(high) and IDENT- IO(low) are violated with respect to the same segment. If a segment in the output violates both IDENT-IO(high) and IDENT-IO(low), the conjoined constraint ID(high)& S ID(low) (short-handed version) will be violated. This constraint penalizes the input-output mapping between a low vowel and a high vowel which skips one level of height and changes [+low, –high] to [–low, +high]. With these additional mechanisms, the pattern of stepwise raising can be produced. The effect of MAX(feature) and ID(high)& S ID(low) is illustrated in (37). (37) Input /pɑ μ w μ + μ/ ‘bag + DIM’: stepwise raising pɑ 1 w 2 + μ CORR-XX ID-XX [μ, +bk] (rd) ID-XX [+rd] (hi) ID(hi)& s ID(lo) MAX(+rd) MAX(+hi) MAX- S ID (hi) ID (lo) ☞ a. p x oː x 1 .p x o x 1 1 1 2 b. p x aː x 1 .p x a x 1 1W 1 1 L c. p x uː x 1 .p x u x 1 1W L 1 2W 2 d. p x u x 1 w x 2 .p x u x 1 1W L L 2W 2 e. p x o x 1 w x 2 .p x e x 1 1W (ID-XX [+rd] (hi)) L L 2 In (37), the SCorr constraints are among the top-tier and ranked higher than MAX(+high) and MAX-S. The winner (37a) raises the low vowel to [o], which violates MAX(+high) but satisfies MAX(+round). Since only ID-IO(low) is violated, the winner escapes violating the conjoined constraint. Candidate (37b) does not retain [+round] in the output, fatally violating MAX(+round). When a low vowel becomes high, as in (37c), (37d), and (37e), it satisfies both MAX(+high) and MAX(+round) at the expense of ID-IO(high) and ID-IO(low), which triggers ID(high)& S ID(low). Note that the candidate in (37e) also violates IDENT-XX [+rd] (high) for a similar reason as in (28b). 87 The evaluation indicates that low vowel raising is actually epiphenomenal. MAX(+high) is expected to promote vowel raising, but the higher-ranked ID(high)& S ID(low) blocks a two-step raising the low vowel to [u], which involves a simultaneous change of [high] and [low]. Further, the requirement of MAX(+round) drives the low vowel to [o] rather than [ɒ], since a low back rounded vowel [ɒ] is not permitted in Standard Mandarin (Duanmu 2007, cf. Zhang, to appear), which also holds in Huozhou Chinese. The candidate *[puw.pu] (37d) harmonically bounds *[puː.pu] (37c). Although candidate (37d) does not affect the evaluation here, a similar scenario makes a difference when the input is /tʰo μ w μ + μ/. In such a case, the expected output is [tuː.tu] rather than *[tuw.tu]. To tackle this issue, I propose that the ill-formed candidate *[tuw.tu] is ruled out by an OCP constraint *V [+high] G, a general phonotactic restriction in Chinese syllables. 20 See (38) and (39). (38) *V [+high] G: Assign a violation if a high vowel is followed by a tautosyllabic glide. (39) Input /tʰo μ w μ + μ/ ‘bean + DIM’: stepwise raising tʰo 1 w 2 + μ *V [+high] G MAX (+rd) MAX (+hi) MAX-S ID(hi) ☞ a. tʰuː 1 .tʰu 1 1 2 b. tʰoː 1 .tʰo 1 1W 1 L c. tʰəː 1 .tʰə 1 1W 1W 1 L d. tʰu 1 w 2 .tʰu 1 1W L 2 The candidates in (39b) and (39c) are ruled out by MAX(+round) and/or MAX(+high). For (39d), it fatally violates *V [+high] G, though no segment deletion takes place in this candidate. This evaluation indicates that *V [+high] G >> MAX-S. The nouns with j-/ŋ-ending can be analyzed in the same manner. To facilitate discussion, some data are repeated in (40). 20 This constraint can be viewed as a version of *[𝛼high][𝛼high]. 88 (40) Featural changes in diminutive reduplication: j-/ŋ-ending nouns Non-diminutive Reduplicated a. [saj 11 ] /sai 11 / [saj 11 .sa 33 ] ‘sieve’ b. [pʰɑŋ 3 ] /pʰaŋ 35 / [pʰɑŋ 35 .pʰa 55 ] ‘plate’ c. [pej 11 ] /pəi 11 / [pej 11 .pu 33 ] ‘cup, mug’ d. [kʰwej 11 ] /kʰwəi 11 / [kʰwej 11 .kʰu 33 ] ‘bowl, container’ e. [pʰəŋ 35 ] /pʰəŋ 35 / [pʰəŋ 35 .pʰu 55 ] ‘basin’ f. [kəŋ 11 ] /kəŋ 11 / [kəŋ 11 .ku 33 ] ‘root’ The low vowel in each noun root remains low, as in (40a) and (40b). However, the mid vowel is raised to [u] in the reduplicant only, as in (40c-f), demonstrating a root-affix asymmetry. For nouns with a low vowel, the constraints MAX(+high) and MAX(+round) do not take effect since no vowel raising is motivated; recall the analysis in (31). In the analyses so far, a question is why /pɑ μ w μ + μ/ → [poː μμ .po μ ] is not treated as coalescence (e.g., /pɑ 1 μ w 2 μ + μ/ → [poː 1,2 μμ .po 1 μ ]). If this alternative is pursued, MAX(feature) is no longer needed. However, if rime change is viewed as coalescence rather than feature movement, other constraints need to be promoted in the gram- mar to block coalescence of j-ending and ŋ-ending nouns, such as /sa 1 μ j 2 μ + μ/ → *[seː 1,2 μμ .se μ ]. Instead, under the current analysis, the patterns can be unified as segment subtraction with feature movement. However, the j-/ŋ-ending nouns with a mid nuclear vowel present a challenge to the analysis so far. There are three finals falling into this category, [əŋ], [wej], and [ej]. The three finals will be discussed in sequence. Although some changes involving the [back] feature are also noticed (e.g., [kʰwej 11 .kʰu 33 ]), the details are not directly relevant to the main theme. The issue of backness agreement will be presented in the Appendix as a supplement. Final [əŋ] In general, there should be an additional driving force to trigger mid vowel raising (e.g., [pʰəŋ.pʰu]) but prevent a low vowel from raising (e.g., *[saj.su]). I propose that the identity- enforcing constraint in action here is IDENT-XX [–low] (high). Also, the root-affix asymmetry is 89 attributed to a positional faithfulness IDENT-IO(high) ROOT . I assume the velar nasal is associated with a set of dorsal features, [+high, –low, +back] (e.g., Hall 2007:332). (41) Constraints for asymmetrical mid vowel raising a. IDENT-XX [–low] (high): Assign a violation to corresponding [–low] segments in the output that do not match in value for [high]. Evaluated over pairs that are adjacent in the chain defined by [–low]. b. IDENT-IO(high) ROOT : Let 𝛼 be a segment in S1 and 𝛽 be any correspondent of 𝛼 in S2, and let both S1 and S2 be in the root. If 𝛼 is [𝛾high], then 𝛽 is [𝛾high]. (after McCarthy & Prince 1994a, Urbanczyk 1996) (42) Input /pʰə μ ŋ μ + μ/ ‘basin + DIM’ pʰə 1 ŋ 2 + μ CORR-XX ID-XX [μ, +bk] (rd) ID-XX [+rd] (hi) *V [+high] G MAX-S ID (hi) RT ID-XX [–lo] (hi) ID (hi) ☞ a. pʰ x ə x 1 ŋ x 2 .pʰ x u x 1 1 1 b. pʰ x ə x 1 ŋ x 2 .pʰ x ə x 1 2W L c. pʰ x u x 1 ŋ x 2 .pʰ x u x 1 1W L 2W d. pʰ x uː x 1 .pʰ x u x 1 1W 1W L 2W For j-/ŋ-ending nouns, segment deletion is always blocked. The candidate in (42d) is ruled out by MAX-S. In (42a), since [ə x 1 ] and [ŋ x 2 ] belong to the root, they are faithful due to ID-IO(high) ROOT . However, [ŋ x 2 ] and [u x 1 ] agree for [high] because they are chain-adjacent with regard to [–low] and enforced by ID-XX [–lo] (high). The nuclear vowel in the reduplicant is raised to [+high] due to the faithful [ŋ] that is protected by ID-IO(high) ROOT . The pair [ə x 1 ~ u x 1 ] in the chain [ə x 1 ~ ŋ x 2 ~ u x 1 ] is not enforced by ID-XX [–lo] (high) since the constraint is evaluated locally. For the constraint ID- MM [+bk] (rd), [ŋ] is included into the chain since it is moraic. Nevertheless, I posit that [ŋ] does not have a [round] feature, and it blocks the potential regressive rounding assimilation in the chain [ə x 1 ~ ŋ x 2 ~ u x 1 ] of (42a). 21 In this way, the constraint ID-XX [μ, +bk] (rd) is vacuously satisfied in (42a), 21 Another solution is to posit a markedness restriction on a rounded [ŋ], such as *ŋʷ, which dominates ID-XX [μ, +bk] (rd). In this way, regressive rounding assimilation can be also blocked. 90 as well as in (42b) and (42c). In comparison, the vowels in (42b) remain faithful but incur a more severe violation of ID-XX [–low] (high). For (42c), a perfect match in vowel height is penalized by the higher-ranked ID-IO(high) ROOT . In this analysis, the mid vowel is raised to a high back vowel rather than a high front one (e.g., *[pʰəŋ.pi]), which is due to the influence of the velar nasal. Some side issues about the agreement on backness will be discussed in the appendix of this chapter. The introduction of ID-IO(high) RT does not affect the evaluation of w-ending nouns in §3.3.2, when it is properly ranked with MAX(+high). See (43) below. (43) Input /tʰo μ w μ + μ/ ‘bean + DIM’: the position of ID-IO(high) RT tʰo 1 w 2 + μ MAX(+hi) ID(hi) RT MAX-S ☞ a. tʰ x uː x 1 .tʰ x u x 1 1 1 b. tʰ x oː x 1 .tʰ x o x 1 1W L 1 For the input /tʰə μ 1 u μ 2 + μ/, the mid vowel in the root (base) triggers raising in the reduplicated output due to the influence of IDENT-XX and MAX-S. The candidates in (43) indicate that ID- IO(high) RT needs to be ranked lower than MAX(+high). Final [wej] The nouns with [wej] final can be analyzed in a similar fashion. An example is given in (44). I assume that the output [kʰwej.kʰwu] is phonetically interpreted as [kʰwej.kʰu]. 91 (44) Input /kʰwe μ j μ + μ/ ‘container + DIM’ kʰw 1 e 2 j 3 + μ CORR-XX ID-XX [μ, +bk] (rd) ID-XX [+rd] (hi) *V [+high] G MAX -S ID (hi) RT ID-XX [–lo] (hi) ID (hi) ☞ a. kʰ x w x 1 e x 2 j x 3 .kʰ x w x 1 u x 2 2 1 b. kʰ x w x 1 o x 2 j x 3 .kʰ x w x 1 o x 2 3W (ID-XX [+rd] (hi)) 3W L c. kʰ x w x 1 u x 2 .kʰ x w x 1 u x 2 1W 1W L 2W d. kʰ x w x 1 u x 2 j x 3 .kʰ x w x 1 u x 2 1W (*V [+high] G) 1W L 2W The situation is more complicated due to the involvement of the prenuclear glide, but the grammar is still able to select the correct output. Both (44a) and (44b) satisfy ID-XX [μ, +bk] (rd) since the adjacent moraic segments in [e x 2 ~ j x 3 ~ u x 2 ] (44a) and [o x 2 ~ j x 3 ~ o x 2 ] (44b) fail to meet the condition [+back]. In addition, (44b) incurs more violations of ID-XX [+rd] (hi). The evaluation of these two candidates are illustrated in (45). (45) The evaluation of ID-XX [μ, +bk] (rd) and ID-XX [–lo] (hi) a. [w x 1 ~ e x 2 ~ j x 3 ~ w x 1 ~ u x 2 ] Candidate (44a) moraic moraic moraic [+bk] [–bk] [–bk] [+bk] [+bk] [–lo] [–lo] [–lo] [–lo] [–lo] [+hi] [–hi] [+hi] [+hi] [+hi] [+rd] [–rd] [–rd] [+rd] [+rd] b. [w x 1 ~ o x 2 ~ j x 3 ~ w x 1 ~ o x 2 ] Candidate (44b) moraic moraic moraic [+bk] [+bk] [–bk] [+bk] [+bk] [–lo] [–lo] [–lo] [–lo] [–lo] [+hi] [–hi] [+hi] [+hi] [–hi] [+rd] [+rd] [–rd] [+rd] [+rd] 92 The illustration in (45) shows how ID-XX [μ, +bk] (rd), ID-XX [+rd] (hi), and ID-XX [–low] (hi) are evaluated for (44a) and (44b). In (45a), the constraint evaluates chain-adjacent segments that are both moraic and [+back], but no such chains exist in this candidate. In (45b), the chain that satisfies both conditions is [o x 2 ~ o x 2 ]. Thus, the constraint ID-XX [μ, +bk] (rd) is satisfied by both candidates. Nevertheless, the candidate in (44b) is less favored due to its violation of ID-XX [+rd] (hi) and ID-XX [-low] (hi). In this candidate, the chain defined by [+round] is [w x 1 ~ o x 2 ~ w x 1 ~ o x 2 ], where there are three adjacent pairs of segments do not match in [high] ([w x 1 ~ o x 2 ] ,[o x 2 ~ w x 1 ] and [w x 1 ~ o x 2 ]). Further, the chain defined by [–low] in the same candidate is [w x 1 ~ o x 2 ~ j x 3 ~ w x 1 ~ o x 2 ], where there are three pairs of segments do not match in height ([w x 1 ~ o x 2 ] ,[o x 2 ~ j x 3 ] and [w x 1 ~ o x 2 ]). For the other candidates, the one in (44c) is ruled out by MAX-S and ID-IO(high) ROOT , and the sequence [u x j x ] violates *V [+high] G in (44d). Final [ej] The last final that involves a mid nuclear vowel is [ej] (/əi/). Similar to the previous cases, the raising from the mid vowel to [u] during reduplication is attributed to the effect of ID-XX [-low] (high). An illustration is given in (46). (46) Input /pʰe μ j μ + μ/ ‘cup + DIM’ pʰe 1 j 2 + μ CORR-XX ID-XX [μ, +bk] (rd) ID-XX [+rd] (hi) *V [+high] G MAX- S ID (hi) RT ID-XX [–lo] (hi) ID (hi) ☞ a. pʰ x e x 1 j x 2 .pʰ x u x 1 1 1 b. pʰ x e x 1 j x 2 .pʰ x ə x 1 2W L c. pʰ x uː x 1 .pʰ x u x 1 1W 1W L 2W d. pʰ x u x 1 j x 2 .pʰ x u x 1 1w (*V [+high] G) 1W L 2W 93 Again, the candidate in (46d) is ruled out by *V [+high] G. The candidate in (46c) incurs fatal violations of MAX-S and ID-IO(high) ROOT . In comparison to (46b), the winner (46a) satisfies ID- IO(high) ROOT while incurring fewer violations of ID-XX [–low] (high). Finally, though not shown in the tableau, the reason why the output is [pej.pu] rather than *[pej.pi] can be the result of consonant-vowel interaction. It is common that labial consonants cause the following vowel to be round (see Padgett 2011:1764 for a review). When reduplication takes place, the copied underlying mid vowel is optimized under the influence of the labial onset, since no mid vowel is allowed in an open syllable with a labial onset (*[pə]). 3.3.4. Summary 3.3.4.1. Summary of rankings This section shows how reduplication can result from mora affixation (§3.3.1), and more importantly, the analysis demonstrates how reduplicative opacity in Huozhou diminutive formation can be achieved through Surface Correspondence (§3.3.2), as well as the issues of vowel raising during reduplication (§3.3.3). Several groups of constraints are responsible for the observed patterns. In §3.3.1, the markedness constraints MAXFLT, HEADEDNESS(μ), and *σ μμμ , force the floating mora to be realized with segmental contents; the low-ranked INTEGRITY favors reduplication as the way to realize the mora template. Further, the additional constraints L-ANCHOR, I-CONTIGUITY, and IDENT-IO-SROLE regulate the shape of the reduplicant. The Hasse diagram of the reduplication-inducing constraints is repeated in (47). 94 (47) Hasse diagram: reduplication-inducing constraints 1. (20a) ≻ (20b), (20c), (20d); 2. (22a) ≻ (22b), (22c); (24a) ≻ (24b) In §3.3.2 and §3.3.3, backcopying and vowel changes are attributed to the joint force of Surface Correspondence and MAX(feature). The Hasse diagram in (48) summarizes the constraint interaction in §3.3.2 and §3.3.3. Some constraints are grouped for the sake of conciseness. (48) Hasse diagram: backcopying and vowel raising in reduplication 1. (28a) ≻ (28b, (28c), (28d); 2. (37a) ≻ (37b); 3. (37a) ≻ (37c), (37d), (37e); 4. (39a) ≻ (39d); 5. (42a) ≻ (42c); 6. (42a) ≻ (42d); 7. (42a) ≻ (42b); 8. (43a) ≻ (43b) MAXFLT *σ µµµ HEADEDNESS(µ) INTEGRITY-S ONSET CONTIGUITY IDENT-SROLE 1 2 MAX-S MAX(+hi) MAX(+rd) ID-XX [–lo] (hi) ID(hi) RT ID(lo) ID(hi) *V [+hi] G 4 8 5 6 2 7 CORR-XX ID(hi)& S ID(lo) ID-XX [+rd] (hi) ID-XX [μ, +bk] (rd) 1 3 95 The constraint interactions in (48) account for various patterns found in Huozhou diminutive reduplication. These constraints can be categorized into several groups, and each group is mainly responsible for one pattern, despite some overlapping. A summary of the patterns is given in (49), followed by detailed explanations. (49) Summary of constraint interactions Main patterns explained Major constraint interactions in (48) Relevant Section a. Truncatory backcopying (w-ending nouns) 1 (and 6, 7) §3.3.2 b. Truncatory backcopying blocked (j-/ŋ-ending nouns) 1 (and 6, 7) §3.3.2 c. Stepwise vowel raising (w-ending nouns) 1, 2, 3, 4 §3.3.3 d. Asymmetrical mid vowel raising (j-/ŋ-ending nouns) 1, 5, 6, 7, 8 §3.3.3 • Backcopying in w-ending nouns (49a): The top-ranked CORR-XX drives correspondence between segments; the IDENT-XX constraints are satisfied at the sacrifice of MAX-S, giving an appearance of backcopying. Also, the IDENT-XX constraints dominate the IDENT- IO(feature) constraints to allow vowel raising. (Example: [pɑw 11 ] → [po: 11 .po 33 ] ‘bag’.) • Backcopying blocked in j-/ŋ-ending nouns (49b): This accounts for The same constraint interactions predict that backcopying does not take place when the input contains a j- /ŋ-ending noun. Due to the lack of [+round] segments, these nouns are exempt from the top-tier constraints, ID-XX [μ, +bk] (rd) and ID-XX [+rd] (high). (Example: [saj 11 ] → [saj 11 .sa 33 ] ‘sieve’, not *[sa.sa].) • Stepwise vowel raising in w-ending nouns (49c): MAX(+high) drives mid vowel raising and MAX(+round) drives low vowel raising; the local conjunction ID(high)& seg ID(low) blocks the two-step raising from a low vowel to a high vowel. (Example: [pɑw 11 ] → [po: 11 .po 33 ] ‘bag’, not *[pu: 11 .pu 33 ]) 96 • Asymmetrical mid vowel raising in j-/ŋ-ending nouns (49d): The constraint ID-XX [– low] (high) drives mid vowel raising while the positional faithfulness constraint ID- IO(high) ROOT protects the root from raising. (Example: [pʰəŋ 35 ] → [pʰəŋ 35 .pu 55 ], not *[pʰuŋ 35 .pu 55 ] or *[pʰəŋ 35 .pə 55 ].) This section has shown the utility of Surface Correspondence in reduplicative opacity. Among the constraint interactions summarized in (48) and (49), the essential ranking schema is {CORR-XX, IDENT-XX} >> MAX-S >> IDENT-IO, and both segment deletion and featural changes can serve as the strategies to satisfy surface identity. This pattern is called conditional truncatory backcopying since backcopying is only limited to nouns with offglide [w]. This condition is encoded in feature-restricted ID-XX [μ, +bk] (rd) and ID-XX [+rd] (hi). Further, several vowel raising processes that target the low vowel and the mid vowel are in concert with reduplication, but the driving force is various. For w-ending nouns, vowel raising is attributed to ID-XX [+rd] (high), MAX(+high), and MAX(+round). For j-/ŋ-ending nouns, vowel raising is due to ID-XX [–lo] (high). These markedness constraints interleave with ID-IO(high) ROOT , ID- IO(high), and ID-IO(low) to generate various vowel raising patterns. Finally, the constraints in (47) are also interacting with the ones in (48), which is taken up in §3.5. 3.3.4.2. Limiting SCorr to reduplication In this analysis, truncation driven by Surface Correspondence should only take place in diminutive reduplication. In other words, the identity effect driven by SCorr is only activated when the diminutive affix shows up in the input, which can be viewed as a case of morphologically-conditioned phonology. This point can be further demonstrated by the examples in (50). 97 (50) Morphologically-conditioned identity effect a. Diminutive affix [[pɑw] root + [ μ ] diminutive affix ] stem SCorr activated b. Root only [pɑw] root no SCorr c. Compound [[pɑw] root ] word + [[tʂɑw] root ] word no SCorr As shown in (50a), SCorr takes effect when the diminutive affix presents in the construction, while in the other constructions such as (50b) and (50c), the pressure of CORR-XX and IDENT-XX that could lead to deletion should not be activated. Thus, the issue in (50) is how to make SCorr morphologically-conditioned and limited to the reduplicative context only. In Chapter 2, §2.3.1, the higher-ranked ACCORD could have a side effect that restricts SCorr to reduplication only. Nevertheless, this option is not viable in the case of Huozhou diminutive reduplication since ACCORD needs to be ranked low (recall the discussion in §3.3.2 above). A solution is to make use of Level Ordering again that has been adopted earlier in §3.2. Recall that I resort to Level Ordering to handle prosodification prior to mora affixation. Meanwhile, Level Ordering, including Stratal OT (Kiparsky 2000, 2008, 2010), is a common way to deal with morphologically-conditioned phonology since each level of morphological construction can have its own phonological grammar. In Chinese, morphological constructions such as affixation are very limited in general, so I will assume that the diminutive morpheme is on its own stratum (level 1) while a handful of the other affixes (e.g., the aspect marker) belong to a later stratum. Each level has its own phonological grammar, as illustrated in (51). (51) Level ordering and morphologically-conditioned SCorr Level Phonological grammar level 0 Root level IDENT-XX, MAX-S >> CORR-XX; also prosodification etc. (recall §3.2) level 1 Diminutive affix CORR-XX, IDENT-XX >> MAX-S level 2 Other affixes IDENT-XX, MAX-S >> CORR-XX … … … 98 For the diminutive affix, CORR-XX is ranked high to activate the identity effect. In other morphological constructions, CORR-XX is ranked lower than MAX-S, which will not lead to segment deletion. When a single root presents in the input without the diminutive marker, as in (52), segment deletion driven by SCorr is not enforced. (52) Non-reduplicative context: no SCorr required pɑ 1 w 2 ID-XX [μ, +bk] (rd) ID-XX [+rd] (hi) MAX-S ID (hi, lo) CORR-XX ☞ a. p x ɑ y 1 w z 2 3 b. p x oː x 1 1W 1W L c. p x o x 1 w x 2 1W 1W L d. p x ɑ x 1 w x 2 1W L In (52), no SCorr relation is required due to the low ranking of CORR-XX. The candidate in (52a) vacuously satisfies all the IDENT-XX constraints. Therefore, no segment deletion is motivated in this construction. Note that it is Surface Correspondence that plays the key role in the analysis of truncatory backcopying. If we only resort to level ordering without SCorr, the analysis will run into problems. This point will be further explored in §3.6.1.2. 3.4. Subtraction as an alternative to reduplication 3.4.1. Subtraction as mora affixation This section shows how segment subtraction in rime change can be derived from mora affixation. Since the underlying phonological form of Huozhou diminutive morpheme is a floating mora, the constraints MAXFLT and HEADEDNESS(μ) enforce the presence and realization of the input floating mora in the output. The same set of constraints, together with *σ μμμ , is able to drive coda deletion at the expense of MAX-S. A tableau is given in (53). 99 (53) Input /lɑ μ ŋ μ + μ/, ‘basket + DIM’ lɑ μ ŋ μ + μ MAXFLT HD(μ) *σ μμμ MAX-S MAX-μ ☞ a. laː μμ 1 1 b. lɑː μμ ŋ μ 1W L L c. lɑ μ ŋ μ + μ 1W L L d. lɑ μ ŋ μ 1W L 1 The winner (53a) realizes the floating mora by docking it on the nucleus vowel, and the deletion of the coda in (53a) avoids a super heavy syllable at the expense of MAX-S and MAX-μ, compared to (53b). Another candidate (53c) leaves the deficient mora headless, violating HEADEDNESS(μ). Finally, the candidate in (53d) fails to preserve the floating mora, i.e., the exponent of the diminutive morpheme, which is ruled out by MAXFLT. In sum, MAXFLT and HEADEDNESS(μ) drive the realization of the floating mora, and *σ μμμ motivates coda deletion in the output. Note that the vowel [a] in candidate (53a) is due to the markedness restriction that the low back vowel only cooccurs with velar consonants or labiovelar glide (Recall §3.3.2). The agreement on backness in the other candidates is achieved through local assimilation during pre-optimization. A challenge is posed by two additional candidates, [lɑ μ ŋ μ ] and [lɑ μ ŋ μ ], where the affixal mora replaces an existing mora in the input. These two candidates harmonically bound the expected winner [laː μμ ] since neither of them violates MAX-S. I propose that the candidates with mora replacement are ruled out by NOVACUOUSDOCKING(μ). This constraint is an extension of NOVACUOUSDOCKING(feature) proposed by Wolf (2006), which requires that “floating features cannot dock onto segments that already bore the same feature value in the input” (Wolf 2006:21). In terms of moras, when a floating mora replaces an existing mora in the output, it is considered as vacuous. In other words, the docking of the floating mora is expected to cause phonological consequences, which is comparable to Kurisu’s (2001) REALIZEMORPHEME. The proposed NOVACUOUSDOCKING(μ) is defined as follows: 22 22 See Saba Kirchner (2010:48-49) for a different definition of NOVACUOUSDOCKING(μ). 100 (54) NOVACUOUSDOCKING(μ): ∀μ 1 ∈ I: [¬[∃S ∈ I such that S is a segment and μ 1 is docked on S]] → [∃μ 2 ∈ O such that μ 1 ℛμ 2 and μ 2 is docked on a segment S’ ∈ O] → [¬[∃S ∈ I such that SℛS’ and S links to the same amount of moras as S’]] “Do not replace an existing mora with a floating mora such that the segment to which the floating µ is associated and its corresponding input segment are associated with the same number of µs.” With the constraint of NOVACUOUSDOCKING(μ), the candidates [lɑ μ ŋ μ ] and [lɑ μ ŋ μ ] can be ruled out. The effect of this constraint is illustrated in (55). (55) Input /lɑ μ ŋ μ + μ/, ‘basket + DIM’ lɑ μ ŋ μ + μ NOVACDOC(μ) MAX-S ☞ a. laː μμ 1 b. lɑ μ ŋ μ 1W L c. lɑ μ ŋ μ 1W L The ranking established in (55) is NOVACDOCK(μ) >> MAX-S. The constraint NOVACDOCK(μ) is violated by (55b) and (55c), where the floating mora replaces an existing mora in the input. When the floating mora docks on the nuclear vowel, it results in a heavy open syllable, which also satisfies *FULLTONE/σ μ proposed earlier. Nevertheless, this constraint is only satisfied accidentally, and it is not one of the forces that drive segment subtraction, as illustrated in (56): (56) Input /lɑ μ ŋ μ + μ/, ‘basket + DIM’; the role of *FULLTONE/σ μ lɑ μ ŋ μ + μ *FULLTONE/σ μ MAXFLT NOVACDOC(μ) MAX-S ☞ a laː μμ 1 b laː μμ 1W 1 c la μ 1W 1W 1W 1 The winner (56a) satisfies all the other constraints except for MAX-S. The candidate (56b) is ruled out by MAXFLT since the floating mora is deleted. For (56c), the violation of *FULLTONE/σ μ is not necessarily decisive, since this candidate also disobeys NOVACDOC(μ). 101 The constraints introduced so far are able to predict segment subtraction in rime change, with the ranking MAXFLT, *σ μμμ , HEADEDNESS(μ), NOVACDOC(μ) >> MAX-S. The affixal mora is thus forced to be realized on the surface, and the pressure of *σ μμμ and NOVACDOCK(μ) trigger coda deletion. In this analysis, the specific phonological instantiation of mora affixation is determined by the position of various faithfulness constraints. When MAX-S is ranked lower, segment subtraction is produced. In contrast, if INTEGRITY-S is ranked below MAX-S, reduplication will be the surface realization of the floating mora. The analyses indicate that grammar is able to predict different surface forms, either reduplication or subtraction, through the position of relevant faithfulness constraints. Here I adopt Partially Ordered Constraints (Anttila 1997, et seq.), which allows freely ranked constraints in grammar in order to account for the variation between reduplication and subtraction. The unordered INTEGRITY-S and {MAX-S, MAX-μ} give rise to two possible rankings, as demonstrated in (57) and (58). Some constraints are omitted in each tableau, which is indicated by the wavy line. (57) Input /lɑ μ ŋ μ + μ/ ‘basket + DIM’: MAX-S >> INTEGRITY lɑ μ ŋ μ + μ MAXFLT MAX-S MAX-μ INTEGRITY-S ☞ a. lɑ μ ŋ μ .la μ 2 b. laː μμ 1W 1W L c. la μ ŋ μ 1W 1W L (58) Input / lɑ μ ŋ μ + μ/ ‘basket + DIM’: INTEGRITY >> MAX-S lɑ μ ŋ μ + μ MAXFLT INTEGRITY-S MAX-S MAX-μ a. lɑ μ ŋ μ .la μ 2W L L ☞ b. laː μμ 1 1 c. lɑ μ ŋ μ 1W L 1 The winners (57) and (58) are selected depending on the position of {MAX-S, MAX-μ} and INTEGRITY-S. When {MAX-S, MAX-μ} outrank INTEGRITY-S, reduplication is selected as the way to fix the floating mora (57). When the ranking is reversed to INTEGRITY-S >> {MAX-S, MAX-μ}, 102 subtraction is predicted instead (58a). Although the ranking argument so far shows that at least one of MAX-S and MAX-μ should outrank INTEGRITY-S to produce reduplication, I assume that both of them dominate INTEGRITY-S in (57) for now. The function of MAX-S and MAX-μ and their role in reduplication will be further discussed in §3.5. The last issue of this section is about the open syllables. Recalling (9), when the input is an open syllable, only reduplication is allowed. This fact can be captured by the established grammar: (59) Input / ŋɤː μμ + μ/ ‘moth + DIM’ ŋɤː μμ + μ MAXFLT NOVACDOC(μ) INTEGRITY-S MAX-S MAX-μ ☞ a. ŋɤː μμ .ŋɤ μ 2 b. ŋɤː μμ 1W L 1W c. ŋɤː μμ 1W L 1W d. ɤː μμ 1W L 1W 1W In (59), reduplication is the only way to realize the floating mora in the current analysis, as in (59a). The candidates (59b) and (59c) are ruled out by MAXFLT and NOVACDOC(μ). The deletion of any input segment will result in an ill-formed output, such as (59d). 23 In sum, on the prosodic level, both subtraction and reduplication result from the pressure of MAXFLT, HEADEDNESS(μ), *σ μμμ , and NOVACDOC(μ). I will refer to this set of constraints as mora wellformedness constraints in the following discussion. The choice of subtraction and reduplication is due to the relative ranking of INTEGRITY-S and {MAX-S, MAX-μ}. Although a ŋ- ending noun is used for demonstration above, this same analysis applies to j-/w-ending nouns as well. The Hasse diagram of the analysis in §3.4.1 is summarized in (60). The diagram in (60a) indicates the ranking that generates subtraction; the one in (60b), however, further elaborates 23 The candidate (59d) also violates the positional faithfulness constraint MAX ONSET . In addition, if the vowel is deleted, it will violate HNUC (Prince & Smolensky 1993/2004). 103 the ranking of (47), which predicts reduplication. These two rankings differ in the order between INTEGRITY-S and {MAX-S, MAX-μ}. (60) Hasse diagram a. Subtraction (during rime change) b. Reduplication 1 and 2. (20a) ≻ (20b, c), (53a) ≻ (53b, c, d), (55a) ≻ (55b, c); 3. (58b) ≻ (58a); 4. (57a) ≻ (57b); 5. (22a) ≻ (22b, c), (24a) ≻ (24b); 3.4.2. Vowel raising in rime change This section deals with the subsegmental processes during rime change. The most prominent process is the change of vowel height, especially for w-ending nouns. Some examples are repeated in (61). (61) Vowel raising in diminutive rime change Non-diminutive Rime-changed Reduplicated gloss a. [pɑw 11 ] /pau 11 / [poː 11 ] [poː 11 .po 33 ] ‘bag’ w-ending b. [tʰow 53 ] /tʰəu 53 / [tʰuː 53 ] [tʰuː 53 .tʰu 11 ] ‘bean’ c. [saj 11 ] /sai 11 / [saː 11 ] [saj 11 .sa 33 ] ‘sieve’ j-ending d. [pej 53 ] /pəi 53 / [puː 53 ] – ‘lifetime’ e. [pʰɑŋ 35 ] /pʰaŋ 35 / [pʰaː 35 ] [pʰɑŋ 35 .pʰa 55 ] ‘plate’ ŋ-ending f. [pʰəŋ 35 ] /pʰəŋ 35 / [pʰuː 35 ] – ‘basin’ The patterns are similar to the ones in reduplication. First, the low vowel is raised only when it is followed by [w] (61a), comparing to [aj] (61c) and [ɑŋ] (61e). Second, the low vowel followed by [w] can be only raised to mid (61a), which is a case of stepwise raising. Third, the mid vowel is raised to high regardless of what ending it is followed by (61b, d, f). INTEGRITY-S 1 3 MAX-S MAX-µ MAXFLT HEADEDNESS(µ) *σ µµµ NOVACDOC(μ) ONSET CONTIGUITY IDENT-SROLE INTEGRITY-S 2 4 5 MAXFLT HEADEDNESS(µ) *σ µµµ NOVACDOC(μ) ONSET CONTIGUITY IDENT-SROLE MAX-S MAX-µ 104 These patterns can be readily explained with the same mechanisms introduced in §3.3. The main motivation of stepwise raising is attributed to MAX(feature), as illustrated in (62). (62) Input /to μ w μ + μ/ ‘bean + DIM’: raising and rounding to 1 μ w 2 μ + μ MAXFLT *σ μμμ HD(μ) NOVACDOC(μ) MAX (+rd) MAX (+hi) ID (hi) ☞ a. tuː 1 μμ 1 b. toː 1 μμ 1W L c. tiː 1 μμ 1W 1 d. to 1 μ w 2 μ 1W (MAXFLT) L In (62), the pressure of MAX(+round) and MAX(+high) forces the output to preserve the relevant features in the input, ruling out (62b) and (62c) respectively. Subtraction is motivated by the set of constraints introduced in §3.4.1, and (62d) is ruled out by MAXFLT. Like in the analysis of reduplication, the issue of stepwise and asymmetrical raising of the low vowel in diminutive rime change results from IDENT-IO(high)& S IDENT-IO(low). See the tableau in (63). (63) Input /pɑ μ w μ + μ/ ‘bag + DIM’: stepwise raising pɑ 1 μ w 2 μ + μ MAXFLT *σ μμμ HD(μ) NOVACDOC(μ) MAX (+rd) ID(hi)& S ID(lo) MAX (+hi) ID (hi) ID (lo) ☞ a. poː 1 μμ 1 1 b. puː 1 μμ 1W L 1W 1 c. paː 1 μμ 1W 1 L d. pɑ 1 μ w 2 μ 1W (MAXFLT) L L The winner (63a) raises the input low vowel to [o], which violates MAX(+high) but satisfies MAX(+round). Since only IDENT-IO(low) is violated, the winner escapes violating the conjoined 105 constraint. When a low vowel becomes high, as in (63b), it satisfies both MAX(+high) and MAX(+round) at the expense of IDENT-IO(high) and IDENT-IO(low), which triggers ID-IO(high)& S ID- IO(low). Candidate (63c) does not retain [+round] in the output, fatally violating MAX(+round). Finally, (63d) violates the set of constraints for subtraction. This analysis is able to account for the asymmetrical pattern of vowel raising. For j-ending and ŋ-ending nouns, vowel raising from low to high will also be blocked by the conjoined constraint, while MAX(+round) cannot motivate any epiphenomenal raising, as in (66). I assume the velar nasal is [+high, –low] (e.g., Hall 2007:332). (64) Input /pʰɑ μ ŋ μ + μ/ ‘plate + DIM’: no raising pʰɑ 1 μ ŋ 2 μ + μ MAX (+rd) ID(hi)& S ID(lo) MAX (+hi) ID (hi) ID (lo) ☞ a. pʰaː 1 μμ 1 b. pʰoː 1 μμ 1 1W c. pʰuː 1 μμ 1W L 1W 1W With the same ranking, the grammar predicts no raising of ŋ-ending nouns after rime change (same for j-ending nouns). The constraint ID-IO(high)& S ID-IO(low) rules out (64c), while the winner (64a) is favored over (64b) by IDENT-IO(low). Note that the winner is [pʰaː] with a nonback [a] rather than [pʰɑː]. This is due to the interaction between *ɑ] σ and IDENT-IO(back), as will be discussed in the appendix of this chapter (see also §3.3.2). The last issue of the subsegmental processes during rime change is the raising of mid vowels in j-ending and ŋ-ending nouns, including [əŋ] (/əŋ/), [wej] (/uəi/), and [ej] (/əi/), all of which become [uː] after rime change. The case of [əŋ] is in line with the analysis above, and [əŋ] → [uː] can be attributed to the preservation of [+high] and [+back] of /ŋ/ after rime change. A constraint MAX(+back) is needed for this purpose but it is not included in the analysis. For [wej], I propose that [wej] → [uː] is conditioned by both MAX(+high) and local agreement. On the one hand, mid vowel raising is motivated by MAX(+high) during rime change; on the other 106 hand, the onglide [w] can influence the nuclear vowel through a local agreement constraint (IDENT-X::X [–cons] (back); see additional discussion in the appendix of this chapter). However, taking [kʰwe μ j μ ] as an example, the output could be *[kʰwoː 1 μμ ], where the feature [+high] of /j/ in the input is vacuously docked on [w] in the output. Compare (65a) with (65b) below. (65) Preservation of [+high] of [j] Input [kʰwe μ j μ + μ] a. Output [kʰwuː μμ ] b. Output *[kʰwoː μμ ] In (65a), when [j] is deleted, the [+high] feature is preserved by MAX(+high) and docked on the nuclear vowel, which leads to vowel raising. In (65b), the [+high] feature of [j] replaces the original [+high] of [w]. In this case, no vowel raising is motivated, though the nuclear vowel becomes rounded due to local assimilation. To prevent (65b), the constraint NOVACDOC(F) (after Wolf 2007:2) needs to be included. See the definition in (66) and the tableau in (67). The structures in (65a) and (65b) reflect the candidates in (67a) and (67b) respectively. Again, I posit that [kʰwuː] is phonetically realized as [kʰuː] as in §3.3.2. 24 (66) NOVACUOUSDOCKING(F): Features cannot dock onto segments that already bore the same feature value in the input. 24 Local assimilation of backness and roundness can be achieved by IDENT-X::X(F) constraints introduced in (29). In a non-derived environment, MAX(+high) is vacuously satisfied. For example, a monomorphemic input /kʰuə/ surfaces as [kʰwo] or [kʰwə] rather than *[kʰwu]. See additional discussion in the appendix of this chapter. k h w e µ σ µ j + µ [+high x [+high y k h w o µ σ µ <j> µ [+high y [+high x k h w u µ σ µ <j> µ [+high y [+high x 107 (67) Input /kʰwe μ j μ + μ/ ‘plate + DIM’: no raising kʰwe 1 μ j 2 μ + μ w [+high] x j [+high] y NOVACDOC(F) ID(hi) ☞ a. kʰwuː 1 μμ w [+high] x u [+high] y 1 b. kʰwoː 1 μμ w [+high] y o [–high] 1W L The relevant features are listed along with the input and candidates for better illustration. The candidate in (67b) is ruled out by NOVACDOC(F), because the [+high] y feature of [j] in the input docks on [w] and replaces the feature with the same value. Although it is also possible to have a candidate [kʰwuː] that fuses two [+high] features, where [w] has [+high] x,y . This candidate can be readily ruled out by UNIFORMITY. Finally, for [ej] (/əi/), the only syllable that has diminutive forms is [pej] /pəi/. Similar to the discussion in §3.3.2, the reason why the output is [puː] rather than [piː] can be the result of consonant-vowel interaction. In sum, vowel raising in rime change is mainly attributed to a set of MAX(feature) constraints. The Hasse diagram of the subsegmental processes in rime change is given in (68). (68) Hasse diagram 1. (62a) ≻ (62b); 2. (63a) ≻ (63b); 3. (63a) ≻ (63c); 4. (63a) ≻ (63d); 5. (67a) ≻ (67c) MAX(+hi) ID(lo) ID(hi) MAX(+rd) ID(hi)& S ID(lo) 1 2 4 5 NOVACDOC(F) 3 MAXFLT HEADEDNESS(µ) *σ µµµ NOVACDOC(μ) 108 3.4.3. Summary This section has shown that how subtraction is produced, with MAX-S being ranked lower than INTEGRITY-S (§3.4.1), and how the patterns of vowel raising during rime change are generated (§3.4.2). All these patterns can be explained by the same set of constraints that have been used in §3.3, plus NOVACDOC(μ) and NOVACDOC(F). These constraints can be categorized into two groups. First, the mora wellformedness constraints interact with MAX-S and INTEGRITY-S to predict subtraction or reduplication, as in (60). Second, the mora wellformedness constraints interact with the featural faithfulness constraints to produce vowel raising during rime change, as in (68). More discussion about the interaction and connection between these constraints will be presented in §3.5. 3.5. Interim summary This section provides a general discussion on the connection and interaction between the analyses in §3.3 and §3.4. The analysis developed so far offers a thorough explanation of Huozhou diminutive formation. First, it shows how reduplication and subtraction can be variably chosen as the way to realize the floating mora. Both subtraction and reduplication are repair strategies of the input deficient mora, which gives a unified account of these processes in a morpheme-based model and provides support for the main claim of Generalized Nonlinear Affixation (Bermúdez-Otero 2012). The mora wellformedness constraints, MAXFLOAT, HEADEDNESS(μ), *σ μμμ , and NOVACDOC(μ), trigger subtraction or reduplication, depending on the relative ranking between {MAX-S, MAX-μ} and INTEGRITY-S. Second, the utility of Surface Correspondence in reduplicative opacity is demonstrated in the analysis above. In reduplication, backcopying and vowel raising are primarily driven by IDENT-XX [+round] (high) and IDENT-XX [μ, +back] (round) and. Meanwhile, the constraints MAX(+round), MAX(+high), and the conjoined constraint ID(high)& S ID(low) play a major role in stepwise vowel raising during rime change. 109 The Hasse diagrams of each subsystem were presented in (48), (60), and (68) in previous sections. However, the interaction between the previously introduced constraints and subsystems warrants examination, as shown in (69). The mora wellformedness constraints, namely, MAXFLOAT, HEADEDNESS(μ), *σ μμμ , and NOVACDOC(μ), are in the top tier and omitted. Given the discussion in previous sections, when INTEGRITY-S >> {MAX-S, MAX-μ}, rime change is motivated; when the ranking is reversed to {MAX-S, MAX-μ} >> INTEGRITY-S, the floating mora is realized as reduplication. (69) Input /pɑ μ w μ + μ/ ‘bag + DIM’, /tʰo μ w μ + μ/ ‘bean + DIM’, and /sa μ j μ + μ/ ‘sieve + DIM’ pɑ μ w μ + μ MAX-μ MAX-S INTEG-S ID(hi) ID(lo) ☞ a. p x oː μμ x .p x o x μ 1 2 2 b. p x oː x μμ 1W 1 L 1L tʰo μ u μ + μ MAX-μ MAX-S INTEG-S ID(hi) ID(lo) ☞ c. tʰ x uː μμ x .tʰ x u x μ 1 2 2 d. tʰ x uː x μμ 1W 1 L 1L sa μ j μ + μ MAX-μ MAX-S INTEG-S ID(hi) ID(lo) ☞ e. s x a x μ j x μ .s x a x μ 2 f. s x aː x μμ 1W 1W L For the inputs in (69), the candidates show either subtraction (69b, d, f) or reduplication (69a, c, e), depending on the relative ranking between INTEGRITY-S, MAX-S, and MAX-μ in the grammar; currently, the grammar selects reduplication as the output. All the candidates in (69) satisfy the top-tier mora wellformedness constraints and SCorr constraints, and SCorr only has an effect in diminutive reduplication. 25 The crucial rankings shown in this tableau are that both MAX-S and MAX-μ need to dominate INTEGRITY-S, based on (69a) ≻ (69b), (69c) ≻ (69d), and (69e) ≻ (69f). Further, the pairs (69a) ≻ (69b) and (69c) ≻ (69d) indicate that MAX-μ must dominate IDENT- IO(high) and IDENT-IO(low). 25 Crucially, SCorr is also in force during subtraction, but it has no effect when the output is a [CVː] syllable. The only potential problem is when the output is [CGVː] where G is [w], such as /kwɑj + μ/ → [kwɑː] ‘jar (diminutive)’. Nevertheless, [k x w x ɑː x ] is not subject to the evaluation of ID-XX [μ, +bk] (rd) because the prenuclear glide is non-moraic. 110 As a summary, the patterns of Huozhou diminutive formation and the major constraints that are responsible for each pattern are recapitulated in (70): (70) Huozhou diminutive formation: the patterns involved Patterns Major constraints a. Reduplication as mora affixation Mora wellformedness, MAX, INTEGRITY b. Subtraction as mora affixation Mora wellformedness, MAX, INTEGRITY c. Backcopying in reduplication SCorr, MAX(F), MAX-S, ID-IO(F) d. Vowel changes in reduplication SCorr, MAX(F), ID-IO(F) e. Vowel raising in subtraction MAX(F), ID-IO(F) Two groups of constraints are responsible for these patterns, as indicated in (70). On the one hand, a set of constraints regulates mora wellformedness and prompts either subtraction or reduplication as the strategy to realize the floating mora, as in (70a-b), where MAX (MAX-S and MAX-μ) and INTEGRITY-S are re-rankable. On the other hand, another set of constraints determines the vowel changes of the output, as in (70c-e). Note that the patterns of (70c) and (70d) can be further elaborated as in (49) in §3.3.4. These constraints are put together in (71) and (72). The rankings are modified for clarity, while the relative relationship of domination in (48), (60), and (68) remain unchanged. Note that INTEGRITY-S and {MAX-S, MAX-μ} are rerankable to predict the variation between rime change and reduplication. 111 (71) Hasse diagram: Rime Change (subtraction), INTEGRITY-S >> {MAX-S, MAX-μ} (72) Hasse diagram: Reduplication, {MAX-S, MAX-μ} >> INTEGRITY-S MAXFLT HEADEDNESS(µ) *σ µµµ NOVACDOC(μ) NOVACDOC(F) *V [+hi] G MAX-S MAX-μ MAX(+hi) MAX(+rd) ID(lo) ID(hi) INTEGRITY-S ID(hi)& S ID(lo) CORR-XX ID(hi)& S ID(lo) ID-XX [+rd] (hi) ID-XX [μ, +bk] (rd) ONSET CONTIGUITY IDENT-SROLE ID(hi) RT ID-XX [–lo] (hi) MAXFLT HEADEDNESS(µ) *σ µµµ NOVACDOC(μ) NOVACDOC(F) *V [+hi] G MAX(+hi) MAX(+rd) ID(lo) ID(hi) ID(hi)& S ID(lo) CORR-XX ID(hi)& S ID(lo) ID-XX [+rd] (hi) ID-XX [μ, +bk] (rd) ONSET CONTIGUITY IDENT-SROLE ID(hi) RT ID-XX [–lo] (hi) INTEGRITY-S MAX-S MAX-μ 112 In (71) and (72), the bold and red lines indicate the constraint interactions that are responsible for resolving the floating mora. When the mora wellformedness constraints are top-ranked and INTEGRITY-S dominates {MAX-S, MAX-μ}, as in (71), rime change is selected as the way to realize the diminutive morpheme. Since reduplication is blocked in this scenario, the SCorr constraints do not take effect (or are vacuously satisfied), and the MAX(feature) constraints along with other relevant markedness constraints drive the changes at the segmental level such as vowel raising. Instead, when {MAX-S, MAX-μ} dominates INTEGRITY-S, as in (72), reduplication is produced as the realization of the diminutive morpheme. Under this ranking, segment deletion is triggered not because of the mora wellformedness constraints, but due to the pressure of SCorr as a concession to satisfy the IDENT-XX constraints, which results in backcopying. In addition, the ranking of the other constraints remain the same, and they jointly motivate vowel changes. 3.6. Alternative approaches The analyses in previous sections focused on the two highlighted patterns of Huozhou diminutive formation. First and foremost, the analysis of Huozhou diminutive reduplication aims to demonstrate the role of Surface Correspondence in generating backcopying and vowel changes. In this case, truncatory backcopying is viewed as a repair strategy to fulfill long- distance segment assimilation. Second, the analysis of Huozhou diminutive formation provides arguments for a morpheme-based model by showing that both reduplication and subtraction can result from mora affixation. In this section, several alternative approaches to these two aspects, truncatory backcopying and subtractive morphology, will be discussed. In what follows, Section §3.6.1 presents the alternative analyses of truncatory backcopying; Section §3.6.2 turns to the discussion on the other approaches of subtractive morphology. In §3.6.3, I focus on Morphological Doubling Theory and Cophonology Theory, which are closely linked and can potentially handle both aspects of Huozhou diminutive formation. 113 3.6.1. Alternative analyses of backcopying 3.6.1.1. Base-Reduplicant Correspondence Backcopying is a crucial prediction made by Base-Reduplicant Correspondence Theory (‘BRCT’, McCarthy & Prince 1995). The full model of BRCT is repeated in (73): (73) Full model of BRCT (McCarthy & Prince 1995:4) In BRCT, a phonologically empty morpheme, RED, triggers reduplication and copies phonological material from the BASE. A symmetric correspondence relation exists between RED and BASE (BR- Correspondence). Thus, when RED meets a structural description and undergoes a phonological change, the identity effect can be imposed back on the BASE through BR-Identity. Backcopying forms the strongest argument for BRCT (recall Chapter 1). The diminutive reduplication of Huozhou Chinese involves backcopying at two dimensions. Taking [pɑw] → [poː.po] for example, both segmental properties and prosodic shape of the base are changed to conform to the reduplicant. For the change of segmental properties, the motivation for vowel raising (and rounding) can be still attributed to MAX(+high) and MAX(+round) in BRCT. For the change in prosodic shape, the identity between the prosodic structures of two strings can be readily handled by MAX-BR. The relevant constraints are defined in (74). A demonstrative analysis is given in (75). (74) a. MAX-BR: Every segment of BASE has a correspondent in RED. b. IDENT-BR(F): Let X be a segment in BASE and Y be any correspondent of 𝛼 in RED. Assign a violation if X is [𝛼F] and Y is [𝛽F] (where 𝛼 ≠ 𝛽). Input Output / RED + X / [RED + BASE] IO-Faith BR-Correspondence IR-Faith 114 (75) Input /pɑw + RED/ ‘bag + DIM’: a BRCT account pɑ 1 w 2 + RED MAX- BR ID- BR MAX (+rd) ID(hi)& S ID(lo) MAX- S MAX (+hi) ID (hi) ID (lo) ☞ a. poː 1 .po 1 1 1 2 b. puː 1 .pu 1 1W 1 L 2W 2 c. paː 1 .pa 1 1W 1 1 L d. paː 1 .po 1 1W 1 1 1L e. pɑ 1 w 2 .pa 1 1W L L L f. po 1 w 2 .po 1 1W L L 2 Apart from MAX-BR and IDENT-BR, the other constraints in (75) remain the same as in the SCorr analysis in the earlier sections. The candidates in (75e) and (75f) are ruled out by MAX-BR, since [w] in the base does not have a correspondent in the reduplicant. For the other candidates which satisfy MAX-BR, (75d) violates IDENT-BR because the corresponding segments are not identical in features; (75b) and (75c) are ruled out by MAX(+round) and ID(high)& S ID(low) respectively. The general ranking schema to generate the backcopying pattern is summarized in (76): (76) BRCT analysis of Huozhou diminutive formation a. Backcopying of segmental properties: IDENT-BR(F), MAX(F), ID(hi)& S ID(lo) >> IDENT-IO(F) b. Backcopying of prosodic shape: MAX-BR >> MAX-S The logic of the analysis with BRCT is the same as the SCorr account in §3.3.2. On the one hand, the MAX(feature) constraints and ID(high)& S ID(low) motivate step-wise vowel changes (76a); on the other hand, the ranking MAX-BR >> MAX-S drives segment deletion (76a-d). A problem for this analysis is that it would overpredict backcopying for the inputs other than w-ending nouns. The ranking MAX-BR >> MAX-S is blind to the type of input. Nouns with j-/ŋ- ending will end up with unexpected backcopying, such as *[saː.sa] instead of [saj.sa]. To circumvent this issue, we could propose a MAX-[j/ŋ] constraint that is specific to certain 115 segments and rank it higher than MAX-BR, which will make backcopying only take place for w- ending nouns. See the demonstration in (77). (77) Input /pɑw + RED/ ‘bag + DIM’; /saj/ ‘sieve + DIM’. pɑ 1 w 2 + RED MAX-[j/ŋ] MAX-BR MAX-S ☞ a. poː 1 .po 1 1 b. pɑ 1 w 2 .pa 1 1W L sa 1 j 2 + RED MAX-[j/ŋ] MAX-BR MAX-S ☞ c. sa 1 j 2 .sa 1 1 d. saː 1 .sa 1 1W L 1W As shown in (77), the segment-specific version of MAX prevents backcopying from taking place in j-/ŋ-ending nouns (77c). Nevertheless, the potential solution with MAX-[j/ŋ] does not necessarily mean that the BRCT approach is more advantageous than SCorr. As discussed in §3.3.3, the analysis of asymmetrical mid vowel raising needs to resort to SCorr anyway, even if a BRCT analysis is pursued. Furthermore, SCorr can be used to handle local assimilation as well (§3.3.2, see also Appendix of Chapter 3), which turns out to be a more versatile mechanism than BR correspondence. Thus, if SCorr can predict both backcopying and the other accompanying long-distance assimilation processes with a single mechanism, BR correspondence can be obviated. 3.6.1.2. Stratal OT without SCorr Stratal OT is a model that accounts for the interleaving between morphology and phonology. There can be a phonology (a constraint ranking) associated with each stratum (stem, word, etc.), which results in a derivational effect and can be a potential source of opacity (Kiparsky 2000, Bermúdez-Otero 1997, 2007, et seq.). The utility of strata in predicting various kinds of reduplicative opacity has been advocated for by Kiparsky (2010), Bermúdez-Otero (2012), etc. A typical example to illustrate the idea is the overapplication of nasal harmony in Madurese reduplication (after Inkelas 2014:137): 116 (78) Nasal harmony in Madurese reduplication input output Stratum 1 /mowa/ [mõw̃ã] Stratum 2 [RED - mõw̃ã] [w̃ã-mõw̃ã] Madurese nasal harmony operates from left to right, but it overapplies in the reduplicated form [w̃ã-mõw̃ã], where the prefixed reduplicant does not meet the context of nasal harmony. The stratal analysis gives rise to the opaque effect and predicts overapplication. Huozhou diminutive reduplication may appear to be derivational. Previous literature (e.g., X.Tian 1992) described Huozhou diminutive reduplication as an ordered application of rime change and reduplication. However, multiple problems will arise with this analysis if opacity is purely attributed to Level Ordering, without resorting to additional mechanisms such as SCorr. If two strata are assumed, the potential application of phonological grammars can be illustrated as follows. I will describe the phonology of each stratum as ‘rime change’ and ‘reduplication’ for demonstrative purposes. Note that I still assume that syllable-internal optimization has been done in an earlier stage of derivation. (79) Huozhou diminutive reduplication: [pɑw] → [poːpo] input output Stratum 1: Rime change [pɑw] [poː] Stratum 2: Reduplication [poː] [poː.po] (80) Huozhou diminutive reduplication: [saj] → [saj.sa] input output Stratum 1: Rime change [saj] [saː] Stratum 2: Reduplication [saː] *[saː.sa] The first problem is that there cannot be a unified stratal ordering between these grammars. To generate the expected output in (79), rime change is applied before reduplication, but this ordered application results in a wrong output in (80). Thus, to produce the correct outputs, we need to assume different stratal orderings for w-ending nouns and j-/ŋ-ending nouns. 117 Second, the derivation [pɑw] → [poː] → [poː.po] will be a case of multiple exponence (see Harris 2017 for a review), i.e., both reduplication and subtraction are applied to realize the diminutive morpheme. That being said, if the current model of mora affixation is pursued, two moras are needed underlyingly for the construction: one for reduplication, another for subtraction. However, it will run into a problem when dealing with j-/ŋ-ending nouns. See the illustration in (81) and (82). (81) [ [pɑw + μ] STRATUM1 + μ] STRATUM2 [pɑ μ w μ + μ x ] + μ y → [poː μμx ] + μ y → [poː μμx .po μy ] (82) [ [saj + μ] STRATUM1 + μ] STRATUM2 [sa μ j μ + μ x ] + μ y → [sa μ j μ + sa μx ] + μ y → *[saː μμy + sa μx ] or *[sa μ j μ + saː μxμy ] For w-ending nouns, it is clear that we need to realize the first affixal mora as subtraction, followed by reduplication, as in (81). However, for j-/ŋ-ending nouns in (82), when reduplication comes first, the second floating mora could be also realized as subtraction on the first syllable or lengthening of the second syllable. This will still result in the wrong output. 3.6.2. Alternative analyses of subtraction In this section, several alternative approaches to subtractive morphology are discussed and compared with the current proposal. I will start with a morpheme-based model of subtractive morphology in Trommer & Zimmermann (2014) and Zimmermann (2017a). Next, I will consider several realization-based models, including Realizational Morphology Theory (RMT, Kurisu 2001) and Transderivational Anti-faithfulness (TAF, Alderete 2001, Horwood 2001), both of which treat nonconcatenative morphology as morphologically-conditioned phonology. Both RMT and TAF make use of Indexed Constraints. 118 3.6.2.1. Prosodically Defective Morphemes Trommer & Zimmermann (2014) proposed that subtractive morphology (consonant/vowel deletion, shortening, length polarity, etc.) can be achieved by mora affixation. This proposal was further developed into the framework of Prosodically Defective Morphemes (PDM) in Zimmermann (2017a), where there is a comprehensive discussion on the effects of prosodic node affixation. The commonality between PDM and the current analysis in this paper is that subtractive morphology can be emergent due to mora affixation. A crucial component in PDM is Colored Containment (van Oostendorp 2006), a development of the original Containment Theory in classic OT (Prince & Smolensky 1993/2004), in contrast to correspondence-theoretic OT assumed in the current analysis. The basic principle of containment is that phonological material in the output is never deleted. Instead, some underlying material can be marked as phonetically uninterpreted, or “silent”, on the surface, which can give an appearance of subtraction. The rime change pattern in Huozhou Chinese can be represented as in (83) to convey the main points of this model. Below I follow the proposal and representation in Trommer & Zimmermann (2014). (83) Subtraction with covert structure (simplified) [saː] The structure in (83) is phonetically realized as [saː] on the surface, where [j] in the output is marked as silent (indicated by the double slash and angle brackets), but not deleted. In this representation, the affixal mora (μ) is not fully incorporated into the structure, so it is unpronounceable, and it exists as a drifting mora in the output. The structure in (83) is able to be generated with the constraints proposed in Trommer & Zimmermann (2014) and s a <j> µ µ σ µ 119 Zimmermann (2017a). As argued by Zimmermann (2017a), PDM can predict a full range of phenomena that involve Morphological Length-Manipulation, as well as reduplication. However, the strict assumption of containment rather than correspondence could cause some problems when dealing with Huozhou diminutive formation. An important part in the analysis of Huozhou diminutive formation is the backcopying pattern of w-ending nouns during reduplication (§3.3.2). In this chapter, this is interpreted as a case of long-distance assimilation that can be resolved by Surface Correspondence (SCorr). Nevertheless, SCorr does not seem compatible with PDM since featural changes in PDM are attributed to operations on autosegmental association lines rather than abstract correspondence relations. Thus, SCorr is not available in PDM and the backcopying pattern poses a challenge to this approach. As pointed out in Zimmermann (2017:65 ff.), reduplication in PDM is generated by populating deficient prosodic nodes through copying segments, which is the same mechanism pursued in this paper. Since the model of PDM makes use of an enriched autosegmental representation, featural changes such as the ones observed in Huozhou diminutive formation rely on autosegmental operations. With regard to vowel raising and rounding, it can be achieved through feature spreading, but what motivates backcopying remains unsolved, as illustrated in (84). (84) Reduplication in PDM: w-ending nouns in Huozhou diminutive formation a. b. [poː.po] *[pow.po] To generate the correct surface form [poː.po] for the input /pɑw + μ/, we may expect a representation as in (84a). In (84a), [+round] spreads to the other segments, resulting in p o <w> µ µ σ p o µ σ [+rd] p o w µ µ σ p o µ σ [+rd] 120 rounding, while [w] is marked as phonetically invisible, compared to (84b). The issue in this model is that the “deletion” (phonetic non-interpretation) of [w] lacks motivation. Zimmermann (2017:67) points out that the copied segments are required to be identical to the ones in the base. It is assumed that there is a phonetically invisible association line that links the original segments and their copied counterparts, as illustrated in (85) (the double slashes indicate phonetic invisibility) (85) Horizontal association line in PDM (simplified illustration) [saj.sa] It is not clear whether this horizontal association line is subject to any evaluation in CON (e.g., CORR-XX), although, as mentioned in Zimmermann (2017:67), it is “conceptually similar” to the surface-to-surface correspondence relation as in Walker (2000a, b, 2001), Rose & Walker (2004), and Hansson (2001/2010). The main purpose of this horizontal association line is to distinguish copied segments from epenthetic segments, since the latter ones do not associate with any existing segments. The horizontal line does not fulfill the same role as SCorr in that it is only established between copied segments, which is similar to the idea of correspondence-by-transitivity. In comparison, [w] deletion with SCorr is partially motivated by the correspondence relation between any segments in the string. Thus, due to the lack of a similar correspondence-like mechanism, [w] “deletion” in (84) needs to be attributed to other reasons. Recalling the structure in (83), subtraction can be motivated by mora affixation. If the same strategy is used for [w] deletion in (84a), an extra mora needs to be present underlyingly. Thus, one mora is s a j µ µ σ s a µ σ 121 responsible for reduplication while another one contributes to subtraction. The same issue of multiple exponence arises, as discussed in §3.6.1.2 earlier. Although both PDM and the current analysis strive for the same goal of a morpheme-based model for nonconcatenative morphology, the correspondence-theoretic analysis in this dissertation can easily handle the backcopying pattern in Huozhou diminutive reduplication. 3.6.2.2. Indexed Constraints and realization-based models Now I turn to several realization-based models. Realizational Morphology Theory (RMT, Kurisu 2001) and Transderivational Anti-faithfulness (TAF, Alderete 2001; Horwood 2001) make use of indexed constraints. Since both theories share certain similarities, I will focus on RMT as a demonstration. In RMT, the driving force of nonconcatenative morphology is the constraint REALIZEMORPHEME which requires that the output must be distinct from the input. The constraint is defined in (86). (86) REALIZEMORPHEME (RM): Let α be a morphological form, β be a morphosyntactic category, and F(α) be the phonological form from which F(α+β) is derived to express a morphosyntactic category β. Then RM is satisfied with respect to β iff F(α+β)≠F(α) phonologically. (Kurisu 2001:39) This model can be demonstrated with Huozhou diminutive formation. In this system, REALIZEMORPHEME should be sandwiched by general faithfulness constraints and indexed faithfulness constraints, such as MAX-S DIM . The definition of MAX-S DIM is given in (87) as an example of how indexed constraints are defined in this proposal. (87) MAX-S DIM Let S 1 be a string of segments in the input and S 2 be a string of segments in the output that contains the diminutive morpheme. Assign a violation for every segment in S 1 that does not have a correspondent in S 2 . 122 In this definition, the indexed constraint is evaluated for an entire output that contains a diminutive morpheme. A tableau for demonstration is given in (88). (88) Subtraction driven by REALIZEMORPHEME /lɑ μ ŋ μ / + DIM MAX-μ *σ μμμ RM INTEGRITY DIM MAX-S DIM ☞ a. laː μμ 1 b. lɑ μ ŋ μ .la μ 2W L c. lɑ μ ŋ μ .lɑ μ ŋ μ 3W L d. lɑ μ ŋ μ 1W L e. lɑː μμ ŋ μ 1W L f. la μ 1W L All the candidates in (88) contain both the root and the diminutive morpheme. To satisfy RM, the output in (88) must be phonologically different from the input. The specific surface form of DIM is determined by the relative ranking between RM and a set of indexed faithfulness constraints. In (88), the constraint MAX-S DIM is ranked the lowest, indicating that deletion is the least antagonistic way to realize the abstract morpheme DIM (88a), rather than lengthening (88e) or reduplication (88b-c). The monomoraic candidate in (88f), [la μ ], can be ruled out by MAX-μ. 26 In general, RMT can capture both subtraction and reduplication in Huozhou diminutive formation. However, this type of theory runs into a too-many-solution problem in terms of typological predictions. Under the force of RM, a morpheme such as DIMINUTIVE can be realized through any phonological operation (tonal changes, featural changes, etc.). Thus, the grammar needs to restrict random changes by ranking all the other indexed faithfulness constraints, except for INTEGRITY DIM and MAX-S DIM , above RM, which also causes a problem of constraint proliferation. One way to restrict the possible phonological changes is to specify the diminutive morpheme with underlying phonological material, such as a mora. If so, this treatment makes RMT no 26 It is not clear how this theory deals with free variation between subtraction and reduplication. We may posit that MAX-S DIM and INTEGRITY DIM are freely ranked in (86) so that either (86a) or (86b) can be the winner. 123 different from the model pursued in the current analysis. With an underlying mora specified in the input, any candidate that fails to realize the affixal mora will be simply ruled out by the top- tier MAXFLOAT and HEADEDNESS(μ), which could obviate the constraint RM. In sum, the current analysis leads to a more parsimonious grammar comparing to a realization-based model with indexed constraints (RMT, as well as TAF), where all faithfulness constraints need to be indexed. Further, with an underlying mora specified in the current analysis, the possible realizations of mora affixation are limited and typologically attested. This point will be further supported from a typological perspective in Section 7 of this chapter. Along the lines of Indexed Constraints, there can be another alternative analysis, where the markedness constraint NOCODA is indexed to DIMINUTIVE as a trigger of subtraction (assume offglides [w, j] are also coda for simplicity). When MAX-S is sandwiched by NOCODA DIM and NOCODA, subtractive morphology will be realized: (89) Indexed Constraint: NOCODA DIM >> MAX-S >> NOCODA /lɑŋ/ + DIM NOCODA DIM MAX-S NOCODA ☞ a. laː 1 b. lɑŋ 1W L 1W /lɑŋ/ NOCODA DIM MAX-S NOCODA ☞ c. lɑŋ 1 d. laː 1W L In (89), NOCODA DIM is enforced in a structure with DIMINUTIVE, so subtraction is produced (89a). When DIMINUTIVE does not appear in the input, MAX-S protects segment deletion in general (89c). An issue here is how to handle reduplication as another possible output. First, it is less clear which markedness constraint can motivate reduplication, since reduplication is usually attributed to a specific underlying representation, either a prosodic template or RED. Second, if reduplication is treated as concatenative, such as the affix of RED in Base Reduplicant Correspondence Theory (McCarthy & Prince 1995), then we end up with a hybrid model that is both realization-based and morpheme-based. 124 3.6.3. Cophonologies and Morphological Doubling Theory This section centers on Cophonology Theory (e.g., Orgun 1996, Inkelas, Orgun, & Zoll 1997, Anttila 1997, 2002, 2006) and Morphology Doubling Theory (Inkelas & Zoll 2005), which are closely connected. I will discuss how the patterns of Huozhou Chinese, truncatory backcopying and nonconcatenative allomorphy, pose challenges to this line of theory. I will start from Cophonology Theory and its role in nonconcatenative morphology, followed by the discussion of Morphology Doubling Theory, a theory of reduplication where Cophonology Theory plays a crucial role. 3.6.3.1. Cophonology Theory and nonconcatenative allomorphy Cophonology Theory assumes that multiple sub-grammars coexist in the same language, and certain morphological constructions can subscribe to their own sub-grammars, or cophonologies. During the construction of complex words, each subconstituent (root, stem, etc.) can have its own cophonology. The virtue of Cophonology Theory is that it captures the overlaps between morphologically-conditioned phonology and nonconcatenative morphology and analyzes them in a single model. Cophonology Theory views nonconcatenative morphology as morphologically-conditioned phonology. In the case of Huozhou diminutive rime change, segment subtraction can be attributed to a cophonology where NOCODA dominates MAX-S. This ranking is specific to the DIMINUTIVE morpheme only. The cophonologies are listed in (90), and the consequences are shown in (91) and (92). (90) Cophonologies in Huozhou diminutive rime change a. Diminutive: NOCODA >> MAX-S b. Non-diminutive: MAX-S >> NOCODA 125 (91) Diminutive context: NOCODA >> MAX-S /lɑŋ/ + DIM NOCODA MAX-S ☞ a. laː 1 d. lɑŋ 1W L (92) Non-diminutive context: MAX-S >> NOCODA /lɑŋ/ MAX-S NOCODA ☞ a. lɑŋ 1 b. laː 1W L In (91) and (92), two sub-grammars are applied to different morphological contexts, resulting in subtraction in diminutive formation. A problem for this approach is that we need to find a general phonological process for each type of nonconcatenative morphology, an issue also identified in Zimmermann (2017a). Although NOCODA >> MAX-S can be the motivation for Huozhou diminutive rime change, as will be seen in §3.7, a range of rime change patterns is found in nearby dialects, including lengthening and epenthesis. It is less clear what markedness constraints would motivate lengthening and epenthesis, if they are viewed as morphologically-conditioned phonology. For example, for the rime change process in Heshun (a nearby town of Huozhou), it is always the nuclear vowel that is lengthened (e.g., /liŋ/ → [liːŋ]). It is possible to propose a constraint such as HAVELONGVOWEL to require an extra mora in the output to motivate morphological lengthening, but this constraint also implies that there should be a general phonological process that requires vowel lengthening in languages. With an underlying mora being posited, we have a benefit that is similar to the one discussed in §3.6.2.2, namely, the possible surface forms are more restricted. 126 3.6.3.2. Morphological Doubling Theory and backcopying Morphological Doubling Theory (MDT) is a theory of reduplication developed in Inkelas & Zoll (2005). MDT treats morphological reduplication as the copying of morphosyntactic feature bundles. A schematic representation is shown in (93) (Inkelas & Zoll 2005:7). (93) Schematic illustration of MDT where [F] = semantic feature bundle. In this illustration, the output node and the input nodes are referred to as “mother node” and “daughter nodes” respectively. The copying of morphosyntactic feature bundles results in two identical daughters of the reduplicated output, or full reduplication. However, each daughter node, as well as the mother node, can undergo phonological deviation from the underlying representation. In MDT, the phonological alternations on each node are attributed to the interaction of three cophonologies: (94) Cophonologies in reduplication (after Inkelas & Zoll 2005:19) Thus, various patterns of reduplication-phonology interaction result from these cophonologies. Also, partial reduplication can be viewed as a cophonology of truncation applied to a daughter node. Huozhou diminutive reduplication can be modeled with MDT as follows: [output] [F + some added meaning] [input] [F] [input] [F] [zzz] [xxx] [yyy] Cophonology Z Cophonology Y Cophonology X Stem Stem 127 (95) Huozhou diminutive reduplication with MDT a. b. The proposal in (95) assumes that the phonological changes take place at the daughter nodes. For (95a), in order to generate the expected output [poː.po], both daughters need to undergo certain phonological processes. In this specific case, the daughter on the left is traditionally considered the base while the one on the right is reduplicative affix. Two cophonologies, Cophonology X and Cophonology Y, target “base” and “reduplicant” respectively. Furthermore, given the asymmetrical patterns in Huozhou diminutive reduplication, this proposal requires that Cophonology X triggers [w] deletion in (95a) but does not take effect on [j] or [ŋ] such as [saj] (95b); in comparison, Cophonology Y that trims the reduplicant to an open syllable needs to be applied in every condition, as in both (95a) and (95b). In sum, although both Cophonology X and Cophonology Y involve segment deletion, Cophonology Y drives coda deletion in every context but Cophonology X only deletes offglide [w] in the “base”. The motivation of segment deletion for both cophonologies can be NOCODA, but Cophonology X and Cophonology Y need to behave differently. As discussed above, Cophonology X should target w-ending nouns only. A potential solution to this asymmetrical pattern is to posit a MAX- [j/ŋ] constraint as the one used in (75). The sketches of Cophonology X and Cophonology Y are [poː.po] [DIM] [paw] [po] Cophonology Z Cophonology Y Cophonology X /pɑw/ /pɑw/ [saj.sa] [DIM] [saj] [sa] Cophonology Z Cophonology Y Cophonology X /saj/ /sai/ 128 demonstrated in (96) and (97) below. Also, I will set aside the issue of vowel length in these cophonologies. (96) Cophonology X pɑw MAX-[j/ŋ] NOCODA MAX-S ☞ a. poː 1 b. pɑw 1W L saj MAX-[j/ŋ] NOCODA MAX-S ☞ c. saj 1 d. saː 1W L 1W lɑŋ MAX-[j/ŋ] NOCODA MAX-S ☞ e. lɑŋ 1 f. laː 1W L 1W (97) Cophonology Y (no compensatory lengthening after segment deletion) pɑw NOCODA MAX-[j/ŋ] MAX-S ☞ a. po 1W b. pɑw 1W saj NOCODA MAX-[j/ŋ] MAX-S c. saj 1W L L ☞ d. sa 1 1 lɑŋ NOCODA MAX-[j/ŋ] MAX-S e. lɑŋ 1W L L ☞ f. la 1 1 In (96), deletion is banned in j-/ŋ-ending nouns, which is indicated by the ranking MAX-[j/ŋ] >> NOCODA in Cophonology X, then (96c) and (96e) are selected as the winners. However, NOCODA drives segment deletion in w-ending nouns, as in (96a). In Cophonology Y, however, a unified effect for w-/j-/ŋ-ending nouns is required. In this case, deletion is chosen as the way to resolve NOCODA, where MAX-S is ranked low. Therefore, the winners (97a), (97d), and (97g) all undergo deletion, while the featural changes can be achieved 129 by MAX(feature) as introduced in previous sections. The MDT model is able to predict the reduplicative patterns in Huozhou diminutive formation. However, a general critique about this model, which is relevant to the case here, concerns its predictions. MDT relies on cophonologies to regulate the shape of the reduplicant, such as NOCODA for a CV output. As pointed out in Saba Kirchner (2010:128), many languages have CV reduplication but it seems more common to prohibit codas in reduplicants rather than to prohibit codas in all contexts of the same language. To sum up, taking §3.6.3.1 and §3.6.3.2 into consideration, the analyses of rime change (subtraction) and reduplication with Cophonology Theory and MDT do not seem to be a unified one. For rime change, it is viewed as a morphologically-conditioned phonology, or an epiphenomenon of phonological grammar. Instead, the MDT model of reduplication with cophonologies treats reduplication as the copying of semantic feature bundle, which operates on morphosyntactic level rather than phonological level. This results in discrepancies between the account of rime change and reduplication in Huozhou diminutive formation, in contrast to a unified account with mora affixation. In addition, it is unclear how this proposal can handle the variation between rime change (subtraction) and reduplication and make the patterns of Huozhou diminutive formation a unified model. 3.6.4. Summary This section has discussed several alternative analyses from two aspects, corresponding to the two goals of this chapter. On the one hand, the backcopying pattern can be potentially re- analyzed with BRCT, Stratal OT, or MDT. On the other hand, several alternative models can deal with rime change (subtractive morphology). However, some drawbacks remain on each of these approaches when considering the overall patterns in Huozhou diminutive formation. The analyses with BRCT and Stratal OT in §3.6.1 show that a certain kind of output-output correspondence is needed to generate the backcopying pattern rather than a solely cyclic 130 treatment. However, BR-Correspondence runs into a problem of overprediction, i.e., all types of input will undergo backcopying under the ranking MAX-BR >> MAX-S. The discussion in §3.6.2 is about the approaches to nonconcatenative morphology, and each of them is able to predict rime change and reduplication in Huozhou diminutive formation. However, Prosodic Deficient Morpheme does not assume correspondence, which has problems with producing the backcopying pattern. The other realizational models, RMT and TAF, face the too-many-solution problem, as discussed in §3.6.2.2. Finally, the MDT model of reduplication and the Cophonology Theory of rime change result in some discrepancies of the patterns under investigation, based on §3.6.3. The discussion of the alternative analyses favor an approach that employs Surface Correspondence to deal with the backcopying patterns, as well as other segmental changes, observed in Huozhou diminutive formation. Also, the treatment of Huozhou diminutive formation as mora affixation offers a unified, restrictive, and economical approach, comparing to the other realizational models of process morphology. 3.7. Excursus: a typology of diminutive formation in nearby dialects The data and analysis for Huozhou diminutive formation show that segment subtraction and reduplication can be the phonological realization of an affixal mora. This proposal is further supported by the other rime change processes in nearby dialects. Rime change is a typical morphophonological characteristic of many varieties of Zhongyuan Mandarin and neighboring Jin Chinese. Most cases of rime change have certain morphological functions, such as denoting diminutive, but the functions are not necessarily identical across dialects. There has been abundant descriptive work on rime change of Chinese dialects (e.g., He 1982, Tian 1986, Hou & Wen 1993), and some systematic and in-depth formal analyses of rime change have been done by Lin (1989, 1993, 2001, 2004, 2008, 2010, etc.) and Duanmu (1990), among others. 131 An important insight of Lin’s (1993, et seq.) analysis is that many rime change patterns can be treated as the affixation of a mora and/or floating features, which is consistent with the analysis of Huozhou diminutive rime change. A case study in Lin (1993) is of Heshun, a variety of Jin Chinese in eastern Shanxi Province: (98) Heshun diminutive rime change (Tian 1986, cited and adapted from Lin 1993:658) noun root diminutive gloss noun root diminutive gloss a. /lu 22 / [luː 22 ] ‘stove’ d. /ɕiəu 44 / [ɕjəːw 44 ] ‘sleeve’ b. /tai 44 / [taːj 44 ] ‘bag’ e. /liŋ 35 / [liːŋ 35 ] ‘collar’ c. /tsua 35 / [tswaː 35 ] ‘claw’ f. /ʂəŋ 31 / [ʂəːŋ 31 ] ‘body’ The generalization of the dataset is that the nuclear vowel in each noun root is lengthened to form a diminutive, and therefore, Heshun diminutive rime change can be viewed as mora affixation. Lin’s (1993) analysis is couched in a rule-based approach with some constraint-based features. The derivations are as follows (cf. Duanmu 1990): (99) Derivation of Heshun diminutive rime change: /lu/ → [luː] → [luː] a. b. (100) Derivation of Heshun diminutive rime change: /liŋ/ → [liːŋ] → → [liːŋ] a. b. c. In (99a), the affixal mora is incorporated into the noun root, resulting in lengthening (99b). In (100), the incorporation of the affixal mora results in a trimoraic syllable at an intermediate µ σ u l + µ µ σ uː l µ µ σ l i ŋ + µ µ µ σ µ µ l iː ŋ µ σ µ l iː ŋ 132 stage (100b), and a templatic constraint [μμ] σ motivates mora deletion and segment reassociation in (100c). The crucial insight of this analysis is that Heshun diminutive rime change surfaces as vowel lengthening through mora affixation. However, several remarks about the analysis are in order. Following studies of Standard Chinese, as introduced in Section 2, Chinese syllables that carry a full tone are bimoraic, and the underlying short vowel in an open syllable is lengthened on the surface. Thus, the non-diminutive form of ‘stove’ on the surface is supposed to be [luː], which raises an issue about the representation after mora affixation. The diminutive form should in principle contrast with its non-diminutive counterpart, but it is not clear how it will be represented. Though we may speculate that the diminutives could be indeed trimoraic (e.g., [luː μμμ ], [liː μμ ŋ μ ]), more investigation is needed to support this hypothesis. Nevertheless, this issue does not undermine the idea that morphological lengthening can be interpreted as mora affixation in Chinese dialects (cf. an alternative analysis in Duanmu 1990). Apart from Heshun, the same lengthening pattern is also found in Yuncheng (Lyu 1991), a variety of Zhongyuan Mandarin, which is spoken in southern Shanxi Province (close to Huozhou). If this analysis is implemented in Optimality Theory, lengthening will be attributed to mora affixation at the expense of IDENT-IO(length). Another representative example is Huojia D rime change. Huojia D is a class of rime change in Huojia Chinese, a variety of Jin Chinese in northwestern Henan Province, which is close to Shanxi Province. This type of rime change specifically applies to place names rather than common nouns, as well as verbs, adjectives, and adverbs (He 1982). When it is applied to place names (usually village names), it also denotes a diminutive/hypocoristic meaning. The patterns are given in (101), where only rimes are listed (prenuclear glide is excluded; Huojia Chinese has the same syllable structure as Standard Mandarin). 133 (101) Huojia D rime change (adapted from He 1982:27, see also Lin 1993:670) a. Root D-changed b. Root D-changed c. Root D-changed /u/ [wəː] /in/ [jɛ̃ː] /au/, /əu/ [ɔː] /i/ [jɛː] /un/ [wɛ̃ː] /ai/, /ei/ [ɛː] /y/ [ɥɛː] /yn/ [ɥɛ̃ː] /an/ [ɑ̃ː] /iŋ/ [jɔ̃ː] /ən/ [ɛ̃ː] /uŋ/ [wɔ̃ː] /aŋ/ [ɔ̃ː] /yŋ/ [ɥɔ̃ː] /əŋ/ [ɔ̃ː] Lin (1993) proposed that Huojia D rime change can be treated as mora affixation, and the affixal mora is realized as a default epenthetic segment [ə] (cf. Lin 2001, 2004, 2008, 2010). Therefore, for open rimes, a mid vowel is inserted after D rime change (101a). 27 When a nasal coda is preceded by a high vowel, it is merged with the epenthetic mid vowel (101b). When a nonhigh vowel is followed by a high vowel or nasal, segment merging takes place without epenthesis (101c). The same idea can be implemented with the mechanisms in this paper (cf. Lin 2001, 2004, 2008, 2010). In Huojia D rime change, the affixal mora can be realized as an epenthetic segment, where DEP-S plays a role. A sketch of the core analysis is presented below. Several top-ranked constraints that regulate mora realization are omitted (MAXFLT, HEADEDNESS, *σ μμμ ). The constraint *Ṽ [+high] penalizes nasalized high vowels (after Lin 2001); UNIFORMITY bans coalescence of input segments. 27 According to He (1982), another two open rimes in Huojia, /ɿ/ and /ʅ/, become [ɐ] in D rime change. This pattern appears to be an exception so it is not discussed here. Further investigation is needed. 134 (102) Huojia D rime change; input /u + μ/, /un + μ/ uː 1 μμ + μ NOVACDOC(μ) *Ṽ [+high] MAX-S DEP-S UNIFORMITY ☞ a. w 1 μ ə 2 μ 1 b. uː 1,2 μμ 1W 1 c. uː 1 μμ 1W L u 1 μ n 2 μ + μ NOVACDOC(μ) *Ṽ [+high] MAX-S DEP-S UNIFORMITY ☞ d. w 1 ɛ̃ː 2,3 μμ 1 e. uː 1 μμ 1W L f. ũː 1,2 μμ 1W L 1W (103) Huojia D rime change; input /au + μ/ and /an + μ/ a 1 μ u 2 μ + μ NOVACDOC(μ) *Ṽ [+high] MAX-S DEP-S UNIFORMITY ☞ a. ɔː 1,2 μμ 1 b. a 1 μ ə 2,3 μ 1W L c. aː 1 μμ 1W L a 1 μ n 2 μ + μ NOVACDOC(μ) *Ṽ [+high] MAX-S DEP-S UNIFORMITY ☞ d. ɛ̃ː 1,2 μμ 1 e. a 1 μ ɛ̃ 2,3 μ 1W L f. aː 1 μμ 1W L Contrary to Huozhou diminutive rime change, segment deletion is banned in Huojia D, which is indicated by the higher-ranked MAX-S. When the input contains a high vowel, as in (102), the affixal mora is realized as epenthesis since coalescence is blocked by either NOVACDOC(μ) (102b, c) or *Ṽ [+high] (102f). The winner (102d) does not incur any violations of UNIFORMITY since it is the epenthetic vowel that coalesces with an input vowel. In both (102a) and (102d), the high vowel in the input forms a glide after epenthesis to resolve a hiatus. In contrast, when the input contains a nonhigh vowel, as in (103), the surface forms show coalescence (103a, d), since the epenthetic vowel cannot be properly fitted into the root. Note that (103a, d) do not violate NOVACDOC(μ), because the corresponding segments are associated with different numbers of moras, according to the definition of NOVACDOC(μ). Several additional mechanisms also assist with the selection of coalescence. First, a nonhigh vowel is less preferred 135 to form a glide for hiatus resolution (see Casali 1996 for relevant discussion), which helps rule out forms such as *[A 1 əː 2,3 μμ ] and *[A 1 ɛ̃ː 2,3 μμ ] (‘A’ represents a low glide; not shown in the tableaux). This explains why epenthesis is only allowed in (101a-b), where there is an underlying high vowel. Second, the output must be monosyllabic, which was also pointed out by Lin (1993), and the forms such as *[a 1 μ u 2 μ .ə 3 μ ] and *[a 1 μ .n 2 ə 3 μ ] are ruled out (not shown in the tableaux). The regulation on output shape can be achieved by, for example, ALIGN[σ] in Feng (2006). 28 Diminutive formation in Huozhou and other (diminutive) rime change patterns in nearby dialects, such as Heshun, Yuncheng, and Huojia, show the versatility of mora affixation. It has been discussed in the literature that mora affixation can result in morphological lengthening, gemination, reduplication, subtraction, epenthesis, and metathesis, etc. (recall §3.2), and the data from Chinese dialects in a particular region cover most of the typical realizations of mora affixation, as summarized in (104). For each type of realization, a crucial faithfulness constraint is in operation: (104) Realization of mora affixation in Chinese dialects Type Example Faithfulness in operation a. subtraction Huozhou diminutive MAX-S, MAX-μ b. reduplication Huozhou diminutive INTEGRITY-S c. lengthening Heshun and Yuncheng diminutive IDENT-IO(length) d. epenthesis Huojia D DEP-S e. mutation (coalescence) Huojia D UNIFORMITY Although the utility of mora affixation, as well as other prosodic nodes, has been supported by various cross-linguistic phenomena (e.g., Zimmermann 2017a), the typology observed in closely related Chinese dialects further shows the advantage of a unified morpheme-based approach towards non-concatenative morphology. With a single underlying representation, 28 However, the sketch here does not account for an asymmetrical pattern in Huojia D rime change. When /aŋ/ and /əŋ/ undergo D rime change, the contrast is lost, and the rime-changed form of both rimes is [ɔ̃]. See Lin (2008, 2010) for an explanation with contrast preservation. 136 various realizations of morphological processes are treated as the consequences of constraint permutation in phonology, where faithfulness constraints play a crucial role in particular. 3.8. Summary Overall, this chapter has presented a case study of Huozhou diminutive formation, where there are two processes involved, diminutive rime change and diminutive reduplication. First and foremost, the patterns of reduplication demonstrate the role of Surface Correspondence. The analysis shows that SCorr can motivate segment deletion with the ranking CORR-XX, IDENT-XX >> MAX-S, which results in an appearance of backcopying. Also, various segmental changes including vowel raising and backing further argue for the utility of SCorr in dealing with the observed patterns. Second, the case study of Huozhou Chinese also offers arguments for the prosodic affixation approach to process morphology, where mora affixation can lead to either subtraction (rime change) or reduplication. 137 Appendix: Backness (dis)agreement across syllable boundaries Since the main thread of the dissertation is to show the effect of SCorr in driving truncatory backcopying, the following discussion that provides additional evidence of the utility of SCorr is not included in the main text above. To provide a full account of the featural changes during reduplication, a closer examination is needed on the issue of backness agreement. Some data are repeated in (1). Recall that [a] is [– back] while [ɑ] is [+back] on the surface. (1) Backness (dis)agreement in reduplication Non-diminutive UR Non-diminutive SR Reduplicated a. /sai 11 / [saj 11 ] [saj 11 .sa 33 ] ‘sieve’ b. /pʰaŋ 35 / [pʰɑŋ 35 ] [pʰɑŋ 35 .pʰa 55 ] ‘plate’ c. /pəi 11 / [pej 11 ] [pej 11 .pu 33 ] ‘cup, mug’ d. /pʰəŋ 35 / [pʰəŋ 35 ] [pʰəŋ 35 .pʰu 55 ] ‘basin’ e. /tɕiŋ 33 / [tɕiŋ 33 ] – ‘water well’ f. /tʂuŋ 11 / [tʂuŋ 11 ] – ‘shot glass’ g. /tɕʰyŋ 35 / [tɕʰyŋ 35 ] [tɕʰyŋ 35 .tɕʰy 55 ] ‘skirt’ Backness agreement is a common syllable-internal optimization in Chinese dialects. It usually targets the low vowel /a/ and the mid vowel /ə/. In (1), both /a/ and /ə/ in the non-diminutive forms are assimilated to the nasal coda or offglide on the surface (1a-d), while the high vowels remain faithful (1e-g). This can be explained by the local SCorr constraints that will be introduced later. For the reduplicated forms, the low vowel in the reduplicant is not backed in (1a) and (1b), while the mid vowel becomes back in (1c) and (1d). To generalize, both the low vowel and the mid vowel undergo backness agreement within the same syllable, while only the mid vowel becomes [+back] in the reduplicant. The high vowels do not undergo backness agreement in any contexts. To account for these patterns, I propose that the assimilation of backness across syllable boundaries is mainly attributed to IDENT-XX [μ] (back), plus a contextual markedness constraint 138 *ɑ] σ . The markedness constraint *ɑ] σ is proposed based on the observation of Chinese phonotactics, where it does not occur in an open syllable unless it is preceded by a labiovelar onglide. (2) a. IDENT-XX [μ] (back): Assign a violation to corresponding moraic segments in the output that do not match in value for [back]. Evaluated over pairs that are adjacent in the chain. b. *ɑ] σ : Assign a violation to the back low vowel in an open syllable. In addition, two constraints that regulate local syllable-internal assimilation are proposed in (3), following the definition in (29) in §3.3.2. (3) a. IDENT-X::X [μ] (back): Let X 1 and X 2 be two moraic segments that are immediately adjacent. Assign a violation if they stand in correspondence but do not match in value for [back]. b. IDENT-X::X [–cons] (back): Let X 1 and X 2 be two [–cons] segments that are immediately adjacent. Assign a violation if they stand in correspondence but do not match in value for [back]. Local assimilation can be analyzed by Surface Correspondence for at least some patterns (Shih & Inkelas 2014). In the current analysis, I make use of SCorr to deal with local assimilation in Chinese syllables, which was not adopted in previous literature (e.g., Duamnu 2007, Lin 2015). Recall that the maximal structure of a Chinese syllable is [C[G[VX] R ] F ] σ (‘R’ for ‘rime’; ‘F’ for ‘final’), and there are highly restricted phonotactic rules within a syllable. In Standard Mandarin, agreement on backness (and rounding) is enforced in the rime domain, as well as between the onglide and the nucleus. The constraint in (3a) targets moraic segments, namely, the segments in the rime domain. The constraint in (3b) targets immediately adjacent [–cons] segments, which is able to evaluate the GV sequence in [C[G[VX] R ] F ] σ . The constraints in (2) interact with the local assimilation constraints in (3) to generate the observed patterns, as illustrated in (4). 139 (4) Input /pʰɑ μ ŋ μ + μ/ and /pʰə μ 1 ŋ μ 2 + μ/ pʰɑ 1 ŋ 2 + μ MAX-S ID-X::X [μ] (bk) *ɑ] σ ID (bk) ID-XX [μ] (bk) ☞ a. pʰ x ɑ x 1 ŋ x 2 .pʰ x a x 1 1 1 b. pʰ x ɑ x 1 ŋ x 2 .pʰ x ɑ x 1 1W L L c. pʰ x a x 1 ŋ x 2 .pʰ x a x 1 1W 2W 2W d. pʰ x a x 1 .pʰ x a x 1 1W 2W L pʰə 1 ŋ 2 + μ MAX-S ID-X::X [μ] (bk) *ɑ] σ ID (bk) ID- XX [μ] (bk) ☞ e. pʰ x ə x 1 ŋ x 2 .pʰ x u x 1 f. pʰ x ə x 1 ŋ x 2 .pʰ x i x 1 1W 1W The candidate in (4d) indicates that ID-XX [μ] (back) is ranked below MAX-S. The candidate in (4a) satisfies both ID-X::X [μ] (back) and *ɑ] σ but it incurs one violation of ID-XX [μ] (back) because the correspondence pair [ŋ x 2 ~ a x 1 ] is not identical for [back]. This candidate also illustrates the different effect between ID-X::X [μ] (back) and ID-XX [μ] (back), where the former targets immediately adjacent segments only. The candidate in (4b) is ruled out by *ɑ] σ , although it satisfies ID-XX [μ] (back). For (4c), it violates both the local assimilation constraint ID-X::X [μ] (back) and the long-distance assimilation constraint ID-XX [μ] (back). The role of ID-XX [μ] (back) is crucial for the candidate in (4e), where the mid vowel becomes back in the reduplicant due to the influence of [ŋ] (recall that [ə] is specified as [+back]). When the onglide [w] appears, it is not subject to the evaluation of ID-XX [μ] (back) since it is not moraic, but the onglide motivates local assimilation through IDENT-X::X [–cons] (back). See (5) for an illustration. Again, the candidate [kʰ x w x e x 1 j x 2 .kʰ x w x u x 1 ] is assumed to be phonetically realized as [kʰwej.kʰu]. 140 (5) Input /kʰwɑ μ ŋ μ + μ/ and /kʰwe μ 1 j μ 2 + μ/ kʰwɑ 1 ŋ 2 + μ MAX-S ID-X::X [–cons] (bk) *ɑ] σ ID (bk) ID-XX [μ] (bk) ☞ a. kʰ x w x ɑ x 1 ŋ x 2 .kʰ x w x ɑ x 1 1 b. kʰ x w x ɑ x 1 ŋ x 2 .kʰ x w x a x 1 1W L 1 1W c. kʰ x w x ɑ x 1 .kʰ x w x ɑ x 1 1W 1 kʰwe 1 j 2 + μ MAX-S ID-X::X [–cons] (bk) *ɑ] σ ID (bk) ID-XX [μ] (bk) ☞ d. kʰ x w x e x 1 j x 2 .kʰ x w x u x 1 1 1 1 e. kʰ x w x e x 1 j x 2 .kʰ x w x i x 1 2W L L f. kʰ x w x u x 1 .kʰ x w x u x 1 1W L 2W L Again, the candidates in both (5c) and (5f) are ruled out by MAX-S. Although the candidate in (5a) violates *ɑ] σ , the low vowel in the reduplicant is backed by ID-X::X [–cons] (bk). Also, the candidate in (5e) indicates that ID-X::X [–cons] (bk) outranks ID-XX [μ] (bk), and therefore, (5d) is selected as the winner. The tableaux in (4) and (5) establish the ranking MAX-S >> ID-X::X [μ] (bk), ID-X::X [–cons] (bk) >> *ɑ] σ >> ID-IO(back), ID-XX [μ] (back). 29 29 Local assimilation in the rime domain is prioritized in Chinese syllables. For instance, the segments in the rime of [kwej] agree on [back] when there are both onglide and offglide. 141 Chapter 4 Reduplication in Rapa Nui Rapa Nui (glottocode: rapa1244) is an Austronesian language spoken on Easter Island, Chile. Though the study of Rapa Nui grammar dates back to the early 20 th century, a recent investigation conducted a thorough study on the grammar of Rapa Nui, with the help of an extensive corpus (Kieviet 2017). Similar to many other Austronesian/Polynesian languages, Rapa Nui has a rich pattern of reduplication. The study of Kieviet (2017) identifies and summarizes two types of reduplication, plural formation and intensifying reduplication. 30 The major patterns of interest are overviewed in (1): (1) Reduplication patterns in Rapa Nui (Kieviet 2017:62, 64, 94, 555) Intensifying reduplication a. vaːnaŋa vanavanaŋa ‘to chat’ b. hoːɾou hoɾohoɾou ‘hurry very much’ Plural formation c. ʔaɾa ʔaʔaɾa ‘to wake up (PL)’ d. haʔuɾu haːʔuɾu ‘to sleep (PL)’ The key patterns to be investigated in this chapter are summarized in (1). In Rapa Nui, there is a puzzling process of vowel shortening during intensifying reduplication. When a root is of HLL shape (“H” for heavy syllable; “L” for light syllable), the initial long vowel is shortened in the reduplicated output (1a-b), indicated by boldface. This pattern is another case of truncatory 30 These two types are named as “Type 1 reduplication” and “Type 2 reduplication” respectively in Kieviet (2017). 142 backcopying, according to the definition and discussion in Chapter 1, which motivates a correspondence-based approach to reduplication. In addition, the plural form of verbs can be produced either through reduplication (1c) or lengthening (1d), depending on the shape of the singular form. This pattern suggests a potential analysis where a single underlying mora (prosodic node) can be variably realized in different forms on the surface. These two patterns correspond to the two main themes of the dissertation. In this chapter, for the intensifying reduplication, I propose that vowel shortening is due to the enforcement of the surface identity effect via Surface Correspondence. For plural formation, I propose that the variable plural forms on the surface are attributed to the affixation of a single mora. This case study seeks to argue for the utility of correspondence as the identity-enforcing mechanism in reduplication, in comparison with theories that resort to level ordering or Harmonic Serialism. In addition, the analyses of both intensifying reduplication and plural formation jointly provide support for the current proposal with SCorr over Base Reduplicant Correspondence Theory, though both are correspondence-base approaches. The rest of the chapter is organized as follows. Section 4.1 introduces the data sources for the case study; Section 4.2 presents the language background, with a focus on the metrical structure and stress patterns, which provides foundation for understanding the patterns of reduplication. Section 4.3 analyses vowel shortening in intensifying reduplication, which shows the utility of Surface Correspondence in handling truncatory backcopying. Also in this section, two theories of reduplication without correspondence are evaluated in detail as additional arguments for a correspondence-based approach. Section 4.4 shows how plural forms are variably realized as reduplication or lengthening, with an analysis of mora affixation, followed by an interim summary in Section 4.5. Section 4.6 compares the current model with BRCT, another correspondence-based approach to reduplication. Finally, a summary is given in Section 4.7. 143 4.1. Data sources The Rapa Nui data discussed here are drawn from various sources. The main description and generalizations for Rapa Nui phonology and morphology are from A Grammar of Rapa Nui (Kieviet 2017). This comprehensive grammar was written based on the study of an extensive corpus, which contains around 580,000 words. A portion of the corpus that contains around 240,000 words was made available for the current research. The corpus only contains the raw texts, which are unglossed. The glosses of the words that appear in this chapter are verified with a cross-reference to different sources. Most of the data are drawn from the documentation of Kieviet (2017). The words from this source are labeled as “Kxxx” (K + page number). Several other sources were also consulted, including a Rapa Nui- Spanish/English dictionary (Fuentes 1960) and a thesis on Rapa Nui reduplication (Johnston 1978). The words/glosses from these sources are labeled as “Fxxx” (F + page number) and “Jxxx” (J + page number) respectively. Some words extracted from the corpus are labeled with their ID in the corpus, such as “R616.620” and “Mtx-4-01.38”. For interpretation of the label, see Kieviet (2017:23–25) and Appendix B therein for details. 4.2. Language background The orthography of Rapa Nui is transparent. In the orthography of Rapa Nui (Weber & Weber 1985, 2005) that has been adopted in Kieviet (2017), long vowels are distinguished from their contrastive short counterparts by a macron (e.g., ā vs. a), and all (pseudo-)reduplicated forms 31 are marked by a hyphen (e.g., mana-mana'u), where the apostrophe represents a glottal stop. Though some earlier text materials of Rapa Nui were not written in the standard orthography, the texts in the corpus used for this study were processed and made uniform in written form, including vowel length and (lexical) reduplicated forms. 31 Pseudo-reduplicated forms refer to the words that contain some repeated sounds but lack a base form, e.g., hā'umu-'umu [haːʔumu-ʔumu] ‘to whisper’. 144 Rapa Nui has 12 consonant phonemes (/p, t, k, ʔ, m, n, ŋ, h, v, ɾ, f, s/) 32 and 10 vowel phonemes (/i, u, e, o, a, iː, uː, eː, oː, aː/). The syllable structure of Rapa Nui is (C)V(ː) in general, and codas are not allowed except in loan words. In terms of syllable weight, (C)V is light while (C)Vː is heavy. There are no diphthongs in Rapa Nui, so vowel sequences belong to separate syllables, such as [ˈho.a] ‘to throw’ (Kieviet 2017:28–37). The most relevant aspect of Rapa Nui phonology for this investigation of reduplication is the metrical structure and stress patterns, introduced in the rest of this section. 4.2.1. Metrical structure and stress patterns The foot type in Rapa Nui is a moraic trochee. Feet are constructed from right to left, and the rightmost foot of a word receives the primary stress (Kieviet 2017:45). Some examples are listed in (2). When the last syllable is light, the penultimate syllable receives primary stress (2a) and (2b). When the last syllable of a word is heavy (bimoraic), the ultimate syllable forms a foot and receives primary stress, as in (2c) and (2d). Further, the strong mora of other feet receives secondary stress (2d-e). For certain words that are longer than three syllables, the secondary stress can freely vary between the first and the second syllable, such as in the reduplicated word [va.(ˌna.va).(ˈna.ŋa)] and [(ˌva.na).va.(ˈna.ŋa)] ‘to chat’ (2g), but no variation is reported for another word of the same shape, [(ˌo.ro).ma.(ˈtu.ʔa)] ‘priest’ (2h). The variation of secondary stress observed in longer words will be revisited shortly. Furthermore, I propose that degenerate feet are not allowed in Rapa Nui, so foot binarity is given priority over parsing all syllables into feet, which will be discussed in detail in §4.2.3. 32 The phonemes /f/ and /s/ are only found in loanwords (Kieviet 2017:28). 145 (2) Stress patterns of Rapa Nui Word Transcription μ count Gloss Source a. noho [(ˈno.ho)] 2 ‘to sit’ K45 b. mauku [ma.(ˈu.ku)] 3 ‘grass’ K45 c. maŋō [ma.(ˈŋoː)] 3 ‘shark’ K45 d. keretū [(ˌke.re).(ˈtuː)] 4 ‘pumice’ K45 e. haŋupotu [(ˌha.ŋu).(ˈpo.tu)] 4 ‘younger child’ K45 f. pāpa’i [(ˌpaː).(ˈpa.ʔi)] 4 ‘to write’ K45 g. vana-vanaŋa [va.(ˌna.va).(ˈna.ŋa)] or 5 ‘to chat’ K45 [(ˌva.na).va.(ˈna.ŋa)] h. oromatu’a [(ˌo.ɾo).ma.(ˈtu.ʔa)] 5 ‘priest’ K45 i. hānautama [(ˌhaː).(ˌna.u).(ˈta.ma)] 6 ‘pregnant’ K45 Given the stress patterns outlined above, for words containing an even-number of moras, secondary stress falls on every other mora before the main-stressed mora. The metrical structure of all possible word shapes that contain an even-number of moras is summarized in (3), and all the syllables in these words can be exhaustively parsed. Note that certain word shapes are unattested in Rapa Nui, including HL, LHLH, HLLL, HLHL, and HLLLH (Kieviet 2017:39) (“H” for heavy syllable; “L” for light syllable), all of which contain heavy syllables that are followed by an odd number of moras. (3) Metrical structure of words with even-numbered moras Metrical structure μ count Metrical structure μ count (ˈH) 2 (ˌH)(ˌH)(ˈH) 6 (ˈLL) 2 (ˌLL)(ˌLL)(ˈH) 6 (ˌLL)(ˈLL) 4 (ˌLL)(ˌH)(ˈLL) 6 (ˌH)(ˈLL) 4 (ˌLL)(ˌH)(ˈH) 6 (ˌLL)(ˈH) 4 (ˌH)(ˌLL)(ˈH) 6 (ˌH)(ˈH) 4 (ˌH)(ˌH)(ˌH)(ˈLL) 8 (ˌH)(ˌLL)(ˈLL) 6 (ˌLL)(ˌH)(ˌLL)(ˈLL) 8 (ˌLL)(ˌLL)(ˈLL) 6 (ˌH)(ˌLL)(ˌH)(ˈLL) 8 (ˌH)(ˌH)(ˈLL) 6 For words that have an odd-number of moras and are longer than three syllables, there can be variations of secondary stress, according to the description in Kieviet (2017:45). However, this 146 variation is not entirely clear due to the lack of data. In Rapa Nui, the only attested word shapes that may show such variation include LLLLL, LLLH, and LLLHLL. Given the assumption that degenerate feet are not allowed, possible footing structures for these words have either a medial stray syllable (e.g., [(ˌva.na).va.(ˈna.ŋa)]) or an initial stray syllable (e.g., [va.(ˌna.va).(ˈna.ŋa)]), as in (4). (4) Metrical structure of words with odd-numbered moras (longer than three syllables) Metrical structure μ count word count in corpus L(ˌLL)(ˈLL) or (ˌLL)L(ˈLL) 5 101 L(ˌLL)(ˈH)? or (ˌLL)L(ˈH)? 5 10 L(ˌLL)(H)(ˈLL)? or (ˌLL)L(ˌH)(ˈLL)? 7 1 Although for some words, especially those of LLLLL type, there can be two options for the position of secondary stress, there are three areas of uncertainty about the variation of secondary stress. First, it is unclear if LLLH and LLLHLL words have similar variations (indicated by question marks), since Kieviet’s (2017) grammar did not provide any examples of stress assignment for these words. The reason behind this is probably due to the low frequency of LLLH and LLLHLL words in the corpus, with only 10 LLLH words and 1 LLLHLL word reported in Kieviet (2017:40). Second, for LLLLL words, it is uncertain if two patterns of secondary stress, “medial stray syllable” and “initial stray syllable”, are equal in status in this language. Third, whether the variation is purely phonological or it is morphologically sensitive remains indecisive, since the only example provided in the grammar is a reduplicated word. Again, the latter two issues are also possibly related to the lack of sufficient examples, since stress is not marked in the text corpus. As stated in Kieviet (2017:45), “more study is needed to determine which factors determine levels of lower-order stress”. Nevertheless, a similar stress pattern is found in some other Polynesian languages, such as Hawaiian (East Polynesian, glottocode: hawa1245). Hawaiian demonstrates a typical Polynesian type of metrical structure, and the basic rules of stress assignment are almost the same as those 147 in Rapa Nui (Elbert & Pukui 1979, Senturia 1998). In Hawaiian, for longer words that contain a sequence of odd-numbered light syllables before the primary stress (the head foot in Hawaiian is also right-aligned), the initial syllable usually receives a secondary stress, resulting in the same medial-stray-syllable pattern as Rapa Nui, such as [(ˌʔe.le).ma.(ˈku.le)] ‘old man’ (Alderete & MacMillan 2015:3). Although some words also show the initial-stray-syllable pattern in Hawaiian, as in [ma.(ˌku.a).(ˈhi.ne)] ‘mother’, Alderete & MacMillan (2015:4) pointed out that the medial-stray-syllable pattern seems to be dominant. Additionally, the medial-stray-syllable pattern is likely related to morphological structure, according to the study by Senturia (1998). In sum, since it is not clear if this kind of variation in Rapa Nui and Hawaiian is a typical Polynesian pattern, I will reserve this issue for future study. In the following analysis, I will assume that the “medial stray syllable” option is also dominant in Rapa Nui, similar to Hawaiian. Even though the grammar is able to predict both options, as will be shown in the following section, this variation will be put aside since it is not crucial to the core topic of the current study, namely, surface identity in reduplication. 4.2.2. Deriving the metrical structure This section gives an analysis of the metrical structure, which is related to the later discussion on reduplication patterns. The metrical structure of Rapa Nui presented above can be derived through the interaction of a set of standard metrical constraints (McCarthy & Prince 1993, Prince & Smolensky 1993/2004), along with a faithfulness constraint, as listed in (5). The evaluation is given in (6). (5) Constraints a. FTBIN: Assign a violation mark for each foot that is not binary under moraic or syllabic analysis. b. PARSE: Assign a violation mark for each syllable that is not parsed by a foot. c. ALL-FT-R: Assign a violation for every foot that does not stand at the right edge of the prosodic word (The distance between a foot and the right edge of prosodic word is counted by syllables). 148 d. ALIGN(PRWD, L, FT, L): Assign a violation for every prosodic word that does not begin with a foot (Misalignment is counted by syllable). (ALIGN-L) e. RIGHTMOST: The head foot is rightmost in the prosodic word (Misalignment is counted by syllable). f. DEP-IO-μ: Assign a violation for every mora in the output that does not have a correspondent in the input (DEP-μ). (6) Input /oɾomatuʔa/ ‘priest’, /haːnautama/ ‘pregnant’, and /mauku/ ‘grass’ /oɾomatuʔa/ DEP-μ RTMOST PARSE ALIGN-L ALL-FT-R ☞ a. (ˌo.ɾo).ma.(ˈtu.ʔa) 1 3 b. o.(ˌɾo.ma).(ˈtu.ʔa) 1 1W 2L c. (ˌoː).(ˌɾo.ma).(ˈtu.ʔa) 1W L 6W /haːnautama/ DEP-μ RTMOST PARSE ALIGN-L ALL-FT-R ☞ d. (ˌhaː).(ˌna.u).(ˈta.ma) 6 e. (ˌhaː).na.u.(ˈta.ma) 2W 4L /mauku/ DEP-μ RTMOST PARSE ALIGN-L ALL-FT-R ☞ f. ma.(ˈu.ku) 1 1 g. (ˈma.u).ku 1W 1 L 1W h. (ˌmaː).(ˈu.ku) 1W L L 2W First, the candidate in (6b) fatally violates ALIGN-L in comparison with to (6a), indicating ALIGN- L >> ALL-FT-R. The candidate in (6c) which lengthens the initial syllable is ruled out by the ranking DEP-μ >> PARSE. For another input, /haːnautama/ ‘pregnant’, the fully-parsed candidate (6d) is favored over (6e), which incurs fewer violations of ALL-FT-R, indicating PARSE >> ALL-FT-R. For the input /mauku/, since (6g) satisfies ALIGN-L but violates RIGHTMOST, RIGHTMOST >> ALIGN-L is required to make (6f) the winner. Further, the ranking DEP-μ >> ALIGN-L is established by (6f) ≻ (6h), the latter of which parses every syllable into feet by lengthening the first syllable, violating DEP-μ. In general, mora insertion or deletion is not allowed to avoid unparsed syllables 33 . Finally, since degenerate feet are disallowed, FTBIN is ranked over PARSE, which is 33 Kieviet (2017:42) mentions that there seems to be a tendency to lengthen an initial vowel so that there will be complete feet, but it does not appear to be a regular phonological rule of this language. As also pointed out by Kieviet (2017:42) “the pressure toward whole feet is not sufficiently strong to prevent the occurrence of many hundreds of LLL words… [LLL] is the third most common pattern overall.” Therefore, for now, I prefer a conservative 149 omitted in the tableaux above. The issue of degenerate feet is discussed in more detail in the following section. As mentioned earlier, although the variation of secondary stress position is put aside, the grammar is able to predict this variation in various ways. One potential solution is to use partially ranked constraints (Anttila 1997). When the ranking between ALIGN-L and ALL-FT-R is reversed to ALL-FT-R >> ALIGN-L, the winner will be [va.(ˌna.va).(ˈna.ŋa)] instead of [(ˌva.na).va.(ˈna.ŋa)]. Again, this variation will not be discussed in detail throughout the following sections. Summing up, the general stress patterns and metrical structure of Rapa Nui can be derived by the constraints introduced above, and the rankings are summarized in (7). (7) Hasse diagram 1. (6a) ≻ (6c) 2. (6f) ≻ (6h) 3. (6f) ≻ (6g) 4. (6d) ≻ (6e) 5. (6a) ≻ (6b) 4.2.3. Issues about degenerate feet In the analysis above, I propose that FTBIN >> PARSE, and degenerate feet are not allowed in Rapa Nui. However, Kieviet (2017:38) assumes that syllables must be exhaustively parsed despite the violation of FTBIN, with the following arguments. Since the first syllable of a LLLLL word can be stressed (the medial-stray-syllable option discussed earlier), the first syllable should be parsed into a foot. Additionally, since the penultimate syllable receives the primary stress, the last two syllables should form a foot. Therefore, Kieviet (2017:65) considered that there can be three analysis that limits lengthening to reduplication only (which will be introduced in §4.4), but the current grammar can be further modified given a more in-depth study of this language. PARSE ALL-FT-R ALIGN-L RIGHTMOST DEP-μ FTBIN 1 2 3 4 5 150 possibilities for footing, i.e., (L)(LL)(LL), (LL)(L)(LL), or (LL)L(LL), while only the first structure seems possible for the variable patterns shown in (1g), i.e., (ˌL)(LL)(ˈLL) and (L)(ˌLL)(ˈLL). Kieviet’s (2017) analysis above is problematic for three reasons. First, a degenerate foot is a marked structure across languages, and it is more common for languages to ban degenerate feet, according to Hayes’ (1995) typological study of metrical stress. It is found that degenerate feet are only allowed in languages that also allow degenerate-size monosyllabic content words (Hayes 1995:88, see also McCarthy & Prince 1986, 1990). In Rapa Nui, however, words that contain only a light syllable are only limited to certain grammatical particles (Kieviet 2017:39), and therefore, it does not fulfill this expectation for a degenerate foot. Second, the proposed footing (ˌL)(LL)(ˈLL) is problematic because there should not be any foot structure for a lapse, a sequence of unstressed syllables (see, for example, Elenbaas & Kager 1999). That is, a foot structure such as (ˌLL)L(ˈLL), or less likely (ˌL)LL(ˈLL), is expected instead. Third, the variation of secondary stress is not a strong argument for the existence of degenerate feet in this language. Even if degenerate feet are disallowed in Rapa Nui, as I proposed in the analysis above, the observed variations can still be predicted by the grammar. Besides, a similar variation is found in Hawaiian, which is closely related to Rapa Nui, as well as in Spanish (Harris 1983, 1989, Hayes 1995, Buckley 2016). The analyses of Hawaiian (Alderete & McMillan 2015) and Spanish (e.g., Buckley 2016) do not assume any degenerate feet but are still able to generate the observed patterns. In sum, there is no evidence to support the claim that degenerate feet exist in Rapa Nui, and I will assume FTBIN >> PARSE in the analysis throughout the chapter. To conclude, the metrical structure and stress patterns of Rapa Nui are summarized as follows. The foot type in Rapa Nui is a moraic trochee, and no degenerate feet are allowed. Feet are constructed from right to left, and the rightmost foot of a word receives the primary stress. Although some words show variable positions of secondary stress, with either an medial stray syllable or an initial stray syllable, sufficient data is lacking to further explore the variation. However, the variation of secondary stress does not affect the core issue of the current study, so it is put aside in the following discussion. 151 4.3. Intensifying Reduplication in Rapa Nui This section focuses on a type of reduplication in Rapa Nui, which is derivational and used for intensification, repetition, or conversion (Kieviet 2017:69-72), referred to as intensifying reduplication in this dissertation. 34 This analysis shows the utility of Surface Correspondence in reduplication, which is further supported by the discussion of alternative theories. In what follows, §4.3.1 presents the data; §4.3.2 and §4.3.3 give the analyses of the observed patterns with the proposed model. Two alternative analyses will be discussed in §4.3.4, followed by a summary in §4.3.5. 4.3.1. Data The main function of intensifying reduplication in Rapa Nui is to mean intensification or repetition of an action. In some cases, a verb can be converted to a noun or adjective through this type of reduplication with a sentiment of intensification (e.g., [paː.ho.no] ‘answer’ → [paː.ho.no.ho.no] ‘argumentative’). The size of the reduplicant for intensifying reduplication always amounts to two moras: either a heavy syllable (H) or two light syllables (L). For monosyllabic and disyllabic roots, full reduplication is observed, as exemplified in (8). 35 (8) Intensifying reduplication: monosyllabic and disyllabic roots 36 Root Gloss Reduplicated Gloss Source a. paː ‘to fold’ (ˌpaː).(ˈpaː) ‘to fold repeatedly’ K64 b. kiː ‘to say’ (ˌkiː).(ˈkiː) ‘to say repeatedly’ K64 c. ho.a ‘to throw’ (ˌho.a).(ˈho.a) ‘to throw various things’ K70 d. ho.no ‘to patch’ (ˌho.no).(ˈho.no) ‘to patch various things’ K70 e. va.hi ‘to divide’ (ˌva.hi).(ˈva.hi) ‘to divide in various parts’ K70 34 It is labeled as “Type 2 reduplication” in Kieviet (2017:69). 35 Most monosyllabic words in Rapa Nui are bimoraic, with the exception of some functional particles (Kieviet 2017:39). 36 The reduplicant is not underlined in this set of examples because full reduplication is observed and it is unclear which part is the copied string. 152 For trisyllabic roots, however, either the first or the last two syllables of the root can be copied, referred to here as “left-edge copying” or “right-edge copying”. There are two major configurations for trisyllabic roots in Rapa Nui, HLL and LLL, as exemplified in (9) and (10). In the following table, the sources of each root-reduplicated pair are listed separately. The stresses for each example are marked based on the rules introduced in §4.2.1, since no stresses are marked in the original data. (9) Intensifying reduplication: HLL roots Root Source Gloss Reduplicated Gloss Source a. vaː.na.ŋa K64 ‘to talk’ (ˌva.na).va.(ˈna.ŋa) ‘to chat’ K64 (ˌvaː).(ˌna.ŋa).(ˈna.ŋa) K64 b. maː.ʔe.a K64 ‘stone’ (ˌma.ʔe).ma.(ˈʔe.a) ‘stony, rocky’ K64 c. paː.ho.no K64 ‘answer’ (ˌpaː).(ˌho.no).(ˈho.no) ‘argumentative’ K64 d. hoː.ɾo.u K94 ‘hurry’ (ˌho.ɾo).ho.(ˈɾo.u) ‘hurry very much’ K555 (ˌhoː).(ˈɾo.u).(ˈɾo.u) R616.620 e. taː.ta.ke K144 ‘to argue’ (ta.ta).ta.(ta.ke) ‘to argue (FREQ.)’ 37 R539-3.214 f. maː.ɾo.a K151 ‘to stand’ (ˌma.ɾo).ma.(ˈɾo.a) ‘to stand (FREQ.)’ Mtx-4-01.38 37 The exact meaning of the reduplicated form is not clear. Here I note the gloss with ‘freq.’ (‘frequentative’) to indicate that the meaning has changed after intensifying reduplication, following the same practice in Alderete & MacMillan (2015) for a similar reduplication process in Hawaiian. 153 (10) Intensifying reduplication: LLL roots Root Source Gloss Reduplicated Gloss Source a. ha.ʔe.ɾe K64 ‘to walk’ (ˌhaː).(ˌʔe.ɾe).(ˈʔe.ɾe) ‘to stroll’ K64 (ˌha.ʔe).ha.(ˈʔe.ɾe) K64 b. ma.na.ʔu K64 ‘to think’ (ˌmaː).(ˌna.ʔu).(ˈna.ʔu) ‘to be worried’ K64 c. ŋa.e.ʔi K69 ‘to move’ (ˌŋaː).(ˌe.ʔi).(ˈe.ʔi) ‘to move back and forth’ K69 d. po.ɾe.ko K70 ‘to be born’ (ˌpoː).(ˌɾe.ko).(ˈɾe.ko) ‘(different kids) to be born’ K70 (ˌpo.ɾe).po.(ˈɾe.ko) Mtx-5-05.008 e. ti.ŋa.ʔi K70 ‘to kill’ (ˌtiː).(ˌŋa.ʔi).(ˈŋa.ʔi) ‘to kill several people’ K70 f. ŋa.ɾu.ɾu F717 ‘fainting’ (ŋaː).(ɾu.ɾu).(ɾu.ɾu) ‘knocked out’ R615.72, F717 g. ŋa.ʔa.ha K204 ‘burst’ (ˌŋaː).(ˈʔa.ha).(ˈʔa.ha) ‘broken into pieces’ R334.300, J50 h. pa.ɾe.he K267 ‘piece’ (ˌpaː).(ˌɾe.he).(ˈɾe.he) ‘pieces’ K340 For some roots, both left-edge copying and right-edge copying exist (e.g., 9a, 9d, 10a, 10d), while for some others, only one of the two patterns is observed. This variation is likely to be lexically determined and it is not the focus of the current discussion, but regardless of whether the base is HLL or LLL, left-edge copying always yields LLLLL and right-edge copying always yields HLLLL. One major concern of this study lies in the vowel shortening process in the root when a reduplicative prefix is present. When a HLL root undergoes left-edge copying, as illustrated in (9), the long vowel in the root is shortened after reduplication, such as /hoː.ro.u/ → [(ˌho.ro).ho.(ˈro.u)] (*[(ˌho.ro).(ˌhoː).(ˈro.u)]) ‘hurry very much’. In contrast, when right-edge copying takes place, as in (9a), (9c), and (9d), no modification of syllable weight is observed. This form is of special interest since the prosodic template of the reduplicant seemingly affects the vowel length in the base, resulting in shortening of the input long vowel. In comparison, for the LLL roots in (10), when a root undergoes right-edge copying, such as /ha.ʔe.re/ → [(ˌhaː).(ˌʔe.re).(ˈʔe.re)] ‘to stroll’, the vowel in the first syllable of the root is lengthened, making it a complete foot. When the root undergoes left-edge copying, such as /ha.ʔe.re/ → 154 [(ˌha.ʔe).ha.(ˈʔe.re)] ‘to stroll’, no quantity change is observed in the root, nor is the first vowel of the reduplicant lengthened. The pattern in (9) is another example of truncatory backcopying. However, different from Huozhou diminutive reduplication, the truncatory backcopying pattern in Rapa Nui exhibits vowel shortening (mora deletion) rather than segment deletion. Even though Kieviet (2017:63) points out that right-edge copying is more common in Rapa Nui, the vowel shortening pattern occurring in left-edge copying of HLL roots is “remarkable” and “surprising” (Kieviet 2017:66), and therefore cannot be ignored. Although the examples with vowel shortening in reduplication are relatively uncommon in the corpus, vowel shortening is still considered by Kievet to be a phonological process rather than lexicalization (Kieviet p.c.). Beyond this, a similar pattern is also found in Hawaiian, a Polynesian language that is close to Rapa Nui: (11) Reduplication of CVːCV(C)V (HLL) roots in Hawaiian (Alderete & McMillan 2015:7-8) a. right-edge copy; without shortening CVːCVCV-CVCV 320/406 tokens e.g., kaː.wa.la → (ˌkaː).(ˌwa.la)-(ˈwa.la) ‘speech of which little is understood’ b. left-edge copy; with shortening CVCV-CVCV(C)V 25/406 tokens e.g., liː.ha.u → (ˌli.ha)-li.(ˈha.u) ‘gentle, cool rain (FREQ.)’ c. infixation; optional shortening CV(V)-CV-CVCV 34/406 tokens e.g., moː.hi.o → (ˌmoː)-hi-(ˈhi.o) ‘draft, gust of wind (FREQ.)’ e.g., puː.no.hu → (ˌpu-no)-(ˈno.hu) ‘to rise as smoke or mist (FREQ.)’ (12) Reduplication of CVCVCV (LLL) roots in Hawaiian (Alderete & McMillan 2015:7) a. right-edge copy; with lengthening CVːCVCV-CVCV 138/292 tokens e.g., ki.o.la → (ˌkiː).(ˌo.la)-(ˈo.la) ‘tossing back and forth, or up and down’ b. left-edge copy; without lengthening CVCV-CVCVCV 62/292 tokens e.g., ke.ʔe.hi → (ˌke.ʔe)-ke.(ˈʔe.hi) ‘to stamp, tramp (FREQ.)’ c. infixation; without lengthening CV-CV-CVCV 36/292 tokens e.g., po.lu.hi → (ˌpo-lu)-(ˈlu.hi) ‘dull, sleep (FREQ.)’ 155 As shown in (11) and (12), the patterns of Hawaiian are comparable to those of Rapa Nui, except that Hawaiian roots can also undergo infixation reduplication. For CVːCV(C)V roots (or HLL roots), the long vowel is shortened when left-edge copying takes place. In contrast, for CVCVCV roots (or LLL roots), the vowel in the initial syllable is lengthened when there is right- edge copying. Again, given the corpus study of Alderete & McMillan (2015), vowel shortening in Hawaiian reduplication is relatively uncommon, with 25 out of 406 CVːCV(C)V roots exhibiting left-edge copying with vowel shortening, but vowel shortening is still treated as a systematic sub-pattern enforced by the phonological grammar. So far, the reduplication patterns of HLL and LLL roots in Rapa Nui have been introduced, while another three types of trisyllabic roots, LLH, HHH, and LHH, are relatively rare. For LLH words, only left-edge copying is observed, with only two examples being found, as shown in (13). No reduplicated forms are identified for HHH and LHH words. (13) Intensifying reduplication: LLH roots Root Source Gloss Reduplicated Gloss Source a. ʔa.u.eː K64 ‘to cry out’ (ˌʔa.u).(ˌʔa.u).(ˈeː) ‘to cry repeatedly’ K64 b. ke.ɾe.tuː K39 ‘pumice’ (ˌke.ɾe).(ˌke.ɾe).(ˈtuː) ‘pumice’ 38 K40 Based on the discussion, the characteristics of intensifying reduplication in Rapa Nui are summarized below: 38 The meaning of the reduplicated form remains the same as the root, according to the description in Kieviet (2017:40). 156 (14) Intensifying reduplication in Rapa Nui: a summary Root shape Schematic representation Left-edge copying Right-edge copying a. H CVː CVː-CVː b. LL CV.CV CV.CV-CV.CV c. HLL CVː.CV.CV CV.CV-CV.CV.CV CVː.CV.CV-CV.CV d. LLL CV.CV.CV CV.CV-CV.CV.CV CVː.CV.CV-CV.CV e. LLH CV.CV.CVː CV.CV-CV.CV.CVː n.a. First, the reduplicant always amounts to two moras, and it copies as many segments as possible (14a-b). Second, for trisyllabic roots HLL and LLL, either left-edge copying or right-edge copying is possible, and some roots can have both forms. Third, reduplication is accompanied by a change of vowel length. When a HLL root undergoes left-edge copying, as in (14c), the long vowel in the root is shortened. This is a case of truncatory backcopying, and the focus of this chapter. When a LLL root undergoes right-edge copying, as in (14d), the short vowel in the first syllable is lengthened. The analysis in the following section will center on the vowel length alternation in HLL and LLL roots (§4.3.3.1 and §4.3.3.2 respectively). To keep the analysis focused, the mechanism that controls left- and right-edge copying is not discussed in the following analysis. 4.3.2. Template for intensifying reduplication The main theme of the dissertation is to argue for the role of Surface Correspondence in reduplication, which is evidenced by left-edge copying of HLL roots. Recalling the discussion in Chapter 1 and Chapter 2, reduplication is viewed as a repair strategy for affixed prosodic nodes. With regard to the observations in §4.3.1, I propose that intensifying reduplication in Rapa Nui is the realization of a derivational morpheme, INTENSE, and the underlying phonological form of INTENSE is a bimoraic template (μμ). 39 This section centers on how a bimoraic template can generate the observed shapes of reduplicant in intensifying reduplication (one heavy syllable or 39 Kieviet (2017:64-67) has the same proposal. 157 two light syllables). The specific advantage of prosodic node affixation will be supported by Rapa Nui plural reduplication, which will be discussed in §4.4 shortly. The summary in (14) shows that the reduplicant always amounts to two moras, but it can be realized as one heavy syllable or two light syllables depending on what the root shape is. The general tendency is to copy as many segments in the root as possible, and the bimoraic template is only realized as a heavy syllable when the root is a bimoraic heavy syllable (e.g., [paː] ‘to fold’ → [paː.paː] ‘to fold repeatedly’; *[pa.pa.paː]). In comparison, the bimoraic template always surfaces as two light syllables when combined with the other types of roots, such as [ho.a.ho.a] ‘to throw various things’ (*[hoː.ho.a]) and [va.na.va.na.ŋa] ‘to chat’ (*[vaː.va.na.ŋa]). It could be speculated that the template is a foot rather than two moras, since a foot in Rapa Nui is bimoraic. However, the discussion and analysis in §4.2 show that the reduplicant is not always coextensive with a foot, such as [va.(ˌna.va).(ˈna.ŋa)] ‘to chat’, even though this type of footing is not considered in this chapter due to the reasons discussed in §4.2.1. Therefore, a bimoraic template gives more flexibility in the analysis of Rapa Nui intensifying reduplication. 40 Although a bimoraic template is not a typical prosodic unit in Prosodic Morphology (McCarthy & Prince 1986/1996), there have been some studies of reduplication that make use of this template (two moras rather than a bimoraic foot), such as the analyses of Mbe (Saba Kirchner 2010:14-15) and Paiwan (Yeh 2008, Bye & Svenonius 2012:465-466). A key issue in the analysis of intensifying reduplication is how to ensure the bimoraic template is properly realized as a heavy syllable or two light syllables. To be specific, there should be some constraints that can rule out candidates such as *[vaː.vaː.na.ŋa] and *[pa.pa.paː], and the constraints, *Vː AFFIX and a self-conjoined constraint INTEGRITY& S INTEGRITY (INTEG 2 ), are proposed to solve this issue: 40 The reduplication patterns in Hawaiian are similar to those in Rapa Nui, and a minimal word template is proposed to account for the Hawaiian data. See Alderete & McMillian (2015) for more discussion. 158 (15) Constraints for the shape of the reduplicant a. *Vː AFFIX : Assign a violation for every long vowel in an affix. b. INTEGRITY& Seg INTEGRITY: Assign a violation if and only if there is more than one violation of INTEGRITY for the same segment. (INTEG 2 ) The constraint *Vː AFFIX is a context-specific version of *Vː, and *Vː is a well-established constraint in literature (McCarthy & Prince 1994b, Rosenthall 1994, Davis & Ueda 2006, a.o.). In general, a long vowel is viewed as marked (e.g., Urbanczyk 1996:211), and there has been a specific constraint to penalize any long vowel in the reduplicant, *LONGVOWEL RED (Gouskova 2007:378). This generalization about markedness also holds for Rapa Nui, where the INTENSE morpheme always copies two light syllables from the root unless the root is a single heavy syllable, such as [paː.paː] ‘to fold repeatedly’. In order to deal with the exception case [paː.paː], the constraint INTEG 2 is used to block double reduplication (e.g., *[pa.pa.paː]). The intention of the constraint here is to forbid input elements that are copied multiple times to satisfy the template, such as [pa.pa.paː]. The conflict between these two constraints gives rise to different surface shapes of the INTENSE morpheme, as illustrated in (16). Note that the tableau in (16) only shows how the shape of the reduplicant is derived, while the motivation for vowel shortening in the root is saved for the following section. (16) Input /μμ + vaːnaŋa/ ‘INTENSE + talk’; /paː/ ‘INTENSE + fold’ /μμ + vaːnaŋa/ INTEG 2 *Vː AFFIX PARSE MAX-μ INTEG ☞ a. (ˌva.na).va.(ˈna.ŋa) 1 1 4 (vana) b. (ˌvaː).(ˌvaː).(ˈna.ŋa) 1W L L 2 (va) L c. (ˌva.va).va.(ˈna.ŋa) 2W L 1 2 (va), 2 (va) W /μμ + paː/ INTEG 2 *Vː AFFIX PARSE MAX-μ INTEG ☞ d. (ˌpaː).(ˈpaː) 1 2 e. (ˌpa.pa).(ˈpaː) 2W L 2 (pa), 2 (pa) W When the input is more than one syllable, both double reduplication and copying of a heavy syllable are blocked by the higher-ranked INTEG 2 and *Vː AFFIX , as indicated by the winner (16a). The 159 candidates in (16b-c) are ruled out by INTEG 2 and *Vː AFFIX , respectively. For (16c), since each of [va] is copied twice, incurring two violations of INTEGRITY, this candidate violates INTEG 2 twice. When the input only contains a heavy syllable, however, the output violates *Vː AFFIX to satisfy INTEG 2 , as shown in (16d). Summing up, the discussion above proposes a solution to deal with the variable shapes of the INTENSE morpheme on the surface. For simplicity, these constraints and relevant candidates will be omitted in the following sections. 4.3.3. Vowel length alternation in intensifying reduplication 4.3.3.1. Vowel Shortening driven by Surface Correspondence This section shows that the enforcement of identity through correspondence plays a crucial role to account for vowel shortening in (9), a case of truncatory backcopying. To facilitate the discussion in this section, some data of HLL left-edge copying are repeated in (17). In the interests of focus on the identity effect, the discussion in this section centers on left-edge copying of HLL roots. Vowel lengthening in LLL roots will be revisited in §4.3.3.2. (17) Left-edge copying of HLL roots Root Gloss Reduplicated Gloss a. vaː.na.ŋa ‘to talk’ (ˌva.na).va.(ˈna.ŋa) ‘to chat’ b. maː.ʔe.a ‘stone’ (ˌma.ʔe).ma.(ˈʔe.a) ‘stony, rocky’ c. hoː.ɾo.u ‘hurry’ (ˌho.ɾo).ho.(ˈɾo.u) ‘hurry very much’ d. taː.ta.ke ‘to argue’ (ˌta.ta).ta.(ˈta.ke) ‘to argue (FREQ.)’ e. maː.ɾo.a ‘to stand’ (ˌma.ɾo).ma.(ˈɾo.a) ‘to stand (FREQ.)’ When HLL roots undergo left-edge copying, the long vowel in the base is shortened to conform to the shape of the reduplicant. In the current proposal, there is a correspondence relation between vowels in the reduplicated form [(ˌva x .na y ).va x .(ˈna y .ŋa z )], and IDENT-VV(length) enforces a length match between corresponding vowels. 160 Following the proposal in Chapter 2, segments in the output are required to stand in correspondence by CORR-XX, and corresponding segments are subject to the evaluation of IDENT-XX [+vocalic] (length), which can be written as IDENT-VV(length). The constraints are defined in (18). (18) Crucial constraints a. CORR-XX: Assign a violation to any pair of segments that are not in correspondence in the output. b. IDENT-XX [+vocalic] (length): Assign a violation if a consecutive pair of corresponding vowels is not associated with an equal number of moras. At the heart of the analysis, the crucial ranking IDENT-VV(length) >> MAX-μ drives vowel shortening when left-edge copying takes place. See the tableau in (19) below. For the sake of simplicity, the onsets are not marked with subscripts. The constraint CORR-XX only evaluates vowels. (19) Input: /μμ + vaːnaŋa/ ‘INTENSE + talk’ /μμ + vaːnaŋa/ CORR- XX IDENT- VV(length) DEP-μ MAX-μ ☞ a. va x .na y .va x .na y .ŋa z 8 1 b. vaː x .na y .vaː x .na y .ŋa z 8 1W L c. va x .na y .vaː x .na y .ŋa z 8 1W L d. va x .na y .vaː i .na j .ŋa k 10W L The candidate in (19a) undergoes vowel shortening in the base under the pressure of IDENT- VV(length), at the cost of MAX-μ. The candidate in (19b) is ruled out by DEP-μ since the reduplicant is augmented by one mora in order to satisfy IDENT-VV(length). Although the base in the candidate (19c) remains faithful, it fatally violates IDENT-VV(length) since the corresponding vowels [a x ~ a x ] do not match in length. Finally, the candidate in (19d) is ruled out since it incurs more violations of CORR-XX by having extra correspondence classes. 161 The tableau in (19) captures the essential part of the proposal, namely, length match is driven by correspondence. However, in certain circumstances, a correct output string of segments can be generated via an unintentional correspondence relation. For instance, another two candidates in (19) can be [va x .na x .va x .na x .ŋa y ] and [va x .na x .va x .na x .ŋa x ], both of which satisfy IDENT-VV(length) at the cost of MAX-μ. Additionally, these two candidates also incur fewer violations of CORR-XX, which makes them harmonically bound the current winner (19a). The issue of unexpected correspondence does not seem perilous for the input /μμ + vaːnaŋa/, but it can lead to a wrong output in other circumstances. For instance, for a hypothetical input /μμ + tatataː/, the expected output should be [ta.ta.ta.ta.taː], where the last long vowel remains long, following the general patterns of Rapa Nui (e.g., /μμ + ʔa.u.eː/ → [ʔa.u.ʔa.u.eː], recall (13)). However, a wrong output *[ta.ta.ta.ta.ta] would be produced. See (20) for illustration. (20) Hypothetical input: /μμ + tatataː/ /μμ + tatataː/ CORR-XX ID-VV (length) MAX-μ L a. ta x .ta x -ta x .ta x .ta x 1 b. ta x .ta x -ta x .ta x .taː x 1W L c. ta x .ta y -ta x .ta y .taː z 8W L Given the current ranking, (20a) is selected as the winner. In (20a), all the segments (vowels) are in a single SCorr relationship in order to satisfy CORR-xx. In this correspondence structure, vowel shortening is triggered through the enforcement of IDENT-VV(length), producing [ta.ta.ta.ta.ta] rather than [ta.ta.ta.ta.taː]. This issue can be avoided by ACCORD-XX that evaluates SCorr structures based on input- output mapping. The definition of ACCORD-XX is repeated in (21). (21) ACCORD-XX: Let X i and X j be two segments in the input, and X i ′ and X j ′ be two segments in the output; and let X i ′ ℜ X i and X j ′ ℜ X j . Assign a violation if X i ′ and X j ′ form a consecutive pair that stands in Surface Correspondence. 162 The constraint ACCORD-XX penalizes SCorr between two segments that do not share the same input (recall the discussion in Chapter 2, §2.2.2.2). This constraint favors [va x .na y .va x .na y .ŋa z ] over [va x .na x .va x .na x .ŋa x ], which avoids the issue of unexpected correspondence. See the illustration in (22). The superscript digits indicate input-output correspondence. (22) Input: /μμ + vaːnaŋa/ ‘INTENSE + talk’ /μμ + vaː 1 na 2 ŋa 3 / ACCORD- XX CORR- XX IDENT- VV(length) MAX-μ ☞ a. va x 1 .na y 2 .va x 1 .na y 2 .ŋa z 3 8 1 b. va x 1 .na y 2 .vaː x 1 .na y 2 .ŋa z 3 8 1W L c. va x 1 .na y 2 .va x 1 .na y 2 .ŋa x 3 1W 6L 1 d. va x 1 .na x 2 .va x 1 .na x 2 .ŋa x 3 4W L 1 With the top-ranked ACCORD-XX, candidates (22c) and (22d) are ruled out since both contain corresponding output segments that are not fissioned from the same input source, though they incur fewer violations of CORR-XX. Thus, the SCorr structure in (22a), [va x .na y .va x .na y .ŋa z ], is favored by the ranking ACCORD-XX >> CORR-XX. Likely, the candidate [ta x 1 .ta x 2 -ta x 1 .ta x 2 .ta x 3 ] in (20a) also incurs four violations of ACCORD-XX, which will be ruled out. The constraint ACCORD-XX has a side effect. It restricts the enforcement of vowel length match to reduplication only. When the morpheme INTENSE does not appear in the input, no surface identity is required by the grammar. For an input such as /taːtake/ ‘to argue’, the grammar should not overpredict and generate *[ta.ta.ke]. This issue can be also avoided by ACCORD-XX. See (23) for an illustration. (23) Input: /taːtake/ ‘to argue’ /taː 1 ta 2 ke 3 / ACCORD- XX CORR- XX IDENT- VV(length) MAX-μ ☞ a. taː x 1 .ta y 2 .ke z 3 3 b. ta x 1 .ta x 2 .ke x 3 2W L 1W c. taː x 1 .ta x 2 .ke x 3 2W L 1W 163 The top-ranked ACCORD-XX rules out (23b) and (23c) where SCorr is established between non- fissioned segments. The candidate (23a) is exempt from the enforcement of IDENT-VV(length) since all the vowels are in different correspondence classes. Intensifying reduplication is the only pattern in the language that requires vowel shortening in the base, and the current ranking limits SCorr to this specific context only. Finally, the interaction between surface correspondence and metrical well-formedness needs also to be considered. Recall the constraint ranking in (7): PARSE is ranked below DEP-μ to allow stray syllables. The interaction between PARSE and SCorr constraints is illustrated in (24), where the candidates have the same correspondence relations as (24a) and (24b). (24) Input: /μμ + vaːnaŋa/ ‘INTENSE + talk’ /μμ + vaːnaŋa/ IDENT-VV(length) MAX-μ PARSE ☞ a. (ˌva x .na y ).va x .(ˈna y .ŋa z ) 1 1 b. (ˌva x .na y ).(ˌvaː x ).(ˈna y .ŋa z ) 1W L L In (24), the losing candidate in (24b) without vowel length modification violates IDENT-VV(length) but satisfies PARSE. However, the winner obeys IDENT-VV(length) at the cost of PARSE, indicating the ranking IDENT-VV(length) >> PARSE. In other words, mora deletion is enforced by surface identity at the sacrifice of metrical well-formedness. In sum, the analysis demonstrates how CORR-XX, ACCORD-XX, and IDENT-VV(length) jointly result in vowel shortening in Rapa Nui intensifying reduplication. The ranking that generates the truncatory backcopying pattern is IDENT-VV(length) >> MAX-μ. The Hasse diagram of constraint ranking discussed so far is given in (25). 164 (25) Hasse diagram: truncatory backcopying with vowel shortening 1. (19a) ≻ (19c, d); (24a) ≻ (24b) 2. (22a) ≻ (22c, d) 3. (19a) ≻ (19b); 4.3.3.2. Vowel lengthening in derived environment To complete the analysis of the observed patterns in §4.3.1, this section turns to the issue of vowel lengthening of the root in a derived environment. Vowel lengthening in the root in intensifying reduplication takes place when an LLL root undergoes right-edge copying. Some examples are repeated in (26). (26) Vowel lengthening in reduplication; LLL roots Root Reduplicated Edge of copying Gloss a. ha.ʔe.ɾe (ˌhaː).(ˌʔe.ɾe).(ˈʔe.ɾe) R ‘to stroll’ (ˌha.ʔe).ha.(ˈʔe.ɾe) L b. po.ɾe.ko (ˌpoː).(ˌɾe.ko).(ˈɾe.ko) R ‘(different kids) to be born’ (ˌpo.ɾe).po.(ˈɾe.ko) L c. ma.na.ʔu (ˌmaː).(ˌna.ʔu).(ˈna.ʔu) R ‘to be worried’ d. ŋa.e.ʔi (ˌŋaː).(ˌe.ʔi).(ˈe.ʔi) R ‘to move back and forth’ When right-edge copying takes place, the grammar drives vowel lengthening in the first syllable. I propose that this process is motivated by metrical well-formedness, since a stray initial syllable can be avoided through lengthening. In order to limit these alternations to derived environments only, I make use of output- output faithfulness (OO-FAITH) with recursive evaluation (e.g., Benua 1997), following the MAX-μ, PARSE DEP-μ CORR-XX, ID-VV(length) ACCORD-XX 1 3 2 165 treatment of a similar pattern in Hawaiian by Alderete and MacMillan (2015). One crucial constraint in the analysis is OO-PROSMATCH, defined in (27). (27) OO-PROSMATCH: The left edge of the main stress foot in the underived stem must have a correspondent at the left edge of some foot in the base of the reduplicated word (OO- PM, Alderete and MacMillan 2015:19). The derived-environment lengthening is demonstrated in (28) with recursions of a single grammar. I assume familiarity with the mechanism of recursive evaluation. The recursions are labeled as (A) and (B) respectively. Again, to keep the analysis simple and focused, the mechanism that controls left- and right-edge copying is not discussed. I simply assume that the input has suffixed mora templates when right-edge copying takes place. 41 (28) Input: /haʔeɾe + μμ/ ‘to stroll + INTENSE’, right-edge copying (A) haʔeɾe OO-PM DEP-μ PARSE ☞ a. ha.(ˈʔe.ɾe) 1 b. ha.(ˈʔe.ɾe) 1 c. (ˌhaː).(ˈʔe.ɾe) 1W ➔ (B) haʔeɾe + μμ OO-PM DEP-μ PARSE a’. (ˌhaː).(ˌʔe.ɾe).(ˈʔe.ɾe) 1 b’. (ˌha.ʔe).ɾe.(ˈʔe.ɾe) 1W L 1W c’. (ˌhaː).(ˌʔe.ɾe).(ˈʔe.ɾe) 1 The higher-ranked DEP-μ bans lengthening in the underived form. Paradigm (28c) is thus ruled out due to its violation of DEP-μ in the dominant recursion. For (28b), where there is no lengthening, the left edge of the head foot in [ha.(ˈʔe.ɾe)] (non-derivative) corresponds to a foot- medial segment in the base of [(ˌha.ʔe).ɾe.(ˈʔe.ɾe)] (derivative), and therefore, paradigm (28b) loses to the winner (28a) on the basis of OO-PROSMATCH. 41 Recall §4.2.1 that I assume words with five light syllables have a default pattern of (ˌLL)L(ˈLL) rather than L(ˌLL)(ˈLL). Therefore, I assume the ranking Align-L >> All-Ft-R. The candidate [ha.(ˌʔe.ha).(ˈʔe.ɾe)] is not included here. 166 It is worth mentioning that when LLL roots undergo left-edge copying, vowel lengthening is readily blocked by the current ranking. For instance, the output of /μμ + haʔeɾe/ is [(ˌha.ʔe).ha.(ˈʔe.ɾe)] rather than *[(ˌha.ʔe).(ˌhaː).(ˈʔe.ɾe)]. This candidate *[(ˌha.ʔe).(ˌhaː).(ˈʔe.ɾe)] is ruled out by DEP-μ, since OO-PROSMATCH is satisfied and does not play a role here. See (29) for illustration. (29) Input: /μμ + haʔeɾe/ ‘to stroll + INTENSE’, left-edge copying (A) haʔeɾe OO-PM DEP-μ PARSE ☞ a. ha.(ˈʔe.ɾe) 1 b. ha.(ˈʔe.ɾe) 1 c. (ˌhaː).(ˈʔe.ɾe) 1W L ➔ (B) μμ + haʔeɾe OO-PM DEP-μ PARSE a’. (ˌha.ʔe).ha.(ˈʔe.ɾe) 1 b’. (ˌha.ʔe).(ˌhaː).(ˈʔe.ɾe) 1W L c’. (ˌha.ʔe).(ˌhaː).(ˈʔe.ɾe) 1W L In (29), the paradigm (29c) is ruled out by DEP-μ in the first iteration. For the other paradigms, the constraint OO-PROSMATCH does not play a crucial role in left-edge copying, since the head foot in the base of the derivative form does not alternate. Instead, vowel lengthening is also blocked in this case by DEP-μ in the second iteration. Note that the paradigm in (29b) also violates ID-VV(length). 4.3.3.3. Summary of ranking To sum up, the discussion in §4.3.3.1 and §4.3.3.2 demonstrates how grammar generates vowel shortening and vowel lengthening patterns in Rapa Nui intensifying reduplication. The analysis of vowel shortening has shown that that the enforcement of surface identity is crucial in the case of truncatory backcopying. The patterns of intensifying reduplication of trisyllabic roots are repeated in (30), where the shaded patterns are the focus of the discussion above. The Hasse diagram of the ranking established in the earlier sections is given in (31). 167 (30) Summary of intensifying reduplication patterns Root shape Schematic representation Left-edge copying Right-edge copying a. HLL CVː.CV.CV CV.CV-CV.CV.CV CVː.CV.CV-CV.CV b. LLL CV.CV.CV CV.CV-CV.CV.CV CVː.CV.CV-CV.CV c. LLH CV.CV.CVː CV.CV-CV.CV.CVː n.a. (31) Hasse diagram 1. (19a) ≻ (19c, d); (24a) ≻ (24b) 2. (22a) ≻ (22c, d) 3. (19a) ≻ (19b) 4. (28a) ≻ (28b) In this grammar, the top-ranked ACCORD prevents unexpected correspondence, as discussed in §4.3.3.1. The other highly-ranked constraints, CORR-XX and IDENT-VV(length), enforce the surface identity between corresponding elements so that there is a match between vowel length. The ranking between OO-PM >> DEP-μ >> PARSE motivates derived-environment vowel lengthening when LLL roots undergo right-edge copying, as discussed in §4.3.3.2. Although I make use of transderivational recursive evaluation for vowel lengthening, it does not affect the analysis in §4.3.3.1. earlier. The only case that is left unanalyzed is (30c). When a LLH root undergoes left-edge copying, the same ranking can clearly predict the expected output, since the form [(ˌCV.CV).(ˌCV.CV).(ˈCVː)] fully satisfies PARSE, and MAX-μ. A tableau with the input /ʔa.u.eː/ is given in (32). MAX-μ, PARSE DEP-μ CORR-XX, ID-VV(length) ACCORD-XX 1 3 2 OO-PROSMATCH 4 168 (32) Input /μμ + ʔa.u.eː/, ‘to cry out’ /μμ + ʔa.u.eː/ MAX-μ PARSE ALL-FT-R ☞ a. (ʔa.u).(ʔa.u).(eː) 4 b. (ʔa.u).ʔa.(u.e) 1W 1W 3L As shown in (32), when left-edge copying takes place for a LLH root, the output contains three complete feet, which satisfies both MAX-μ and PARSE. Any modification of vowel length, such as (32b), will be ruled out by these higher-ranked constraints, though they may incur fewer violations of of the lower-ranked ALL-FT-R. Note that it is unclear if LLH roots exhibit right-edge copying as well, since only two examples are found in the corpus, both of which only show left- edge copying (recall (13) for examples). 4.3.4. Alternative analyses without correspondence The purpose of the current section is to argue for the role of Surface Correspondence in handling the truncatory backcopying pattern of Rapa Nui intensifying reduplication. In §4.3.3, I have demonstrated how vowel shortening in the base can be attributed to IDENT-VV(length) through SCorr. In this section, I will argue for the advantage of a correspondence-based approach to reduplication by showing that the two theories of reduplication without correspondence fail to account for the observed patterns. In both theories, neither metrical wellformedness nor derivational models could provide a better solution than a correspondence-based approach. For the rest of the section, I will discuss the model of Minimal Reduplication (Saba Kirchner 2010, 2013) without SCorr in §4.3.4.1, demonstrating that we cannot simply use Level Ordering or other ad hoc constraints to account for truncatory backcopying. In §4.3.4.2, I will turn to another theory of reduplication, Serial Template Satisfaction (McCarthy, Kimper & Mullin 2012), where reduplicative opacity is attributed to derivation in Harmonic Serialism. In both cases, the motivation for vowel shortening is hard to identify. 169 4.3.4.1. Minimal Reduplication without SCorr First, I will examine if vowel shortening can be motivated simply based on the metrical wellformedness constraints or Level Ordering in Minimal Reduplication, without resorting to SCorr. For Rapa Nui, the phonological exponent of intensifying reduplication is still analyzed as an empty bimoraic template (μμ). Since correspondence-based surface identity constraints are not available in this approach, a motivation for vowel shortening is lacking. Consider the tableau in (33), assuming left-edge copying takes place: (33) /μμ + vaːnaŋa/ and non-reduplicated root /vaːnaŋa/ /vaːnaŋa/ *Vː AFFIX PARSE ALIGN-L ALL-FT-R MAX-μ ☞ a. (ˌvaː).(ˈna.ŋa) 2 b. va.(ˈna.ŋa) 1W 1W L 1W /μμ + vaːnaŋa/ *Vː AFFIX PARSE ALIGN-L ALL-FT-R MAX-μ ☞ c. (ˌva.na).va.(ˈna.ŋa) 1 3 1 d. va.(ˌna.va).(ˈna.ŋa) 1 1W 2L 1 e. (ˌvaː).(ˌvaː).(ˈna.ŋa) 1W L 5W L M f. (ˌva.na).(ˌvaː).(ˈna.ŋa) L 5W L In this approach, the empty moras are forced to be realized with segmental content through a highly-ranked constraint *MAXFLT (Wolf 2007), similar to what is used in previous discussions (§3.3.4). Meanwhile, segment copying incurs violations of INTEGRITY, which is ranked low (both *MAXFLT and INTEGRITY are not shown in the tableau). In this case, the grammar is required to produce a faithful mapping for a monomorphemic input /vaːnaŋa/ but vowel shortening in reduplication. The candidate [(ˌvaː).(ˌvaː).(ˈna.ŋa)] in (33f) can be ruled out by the same constraint *Vː AFFIX that penalizes long vowels in the surface form of the INTENSE morpheme (recall §4.3.2). However, the rankings for both inputs in (33) cannot be reconciled. To make (33a) the winner, the constraint PARSE should be ranked at least higher than ALL-FT-R, which will fail to select (33c) as the winner at the same time. 170 One way to circumvent this issue is to introduce a constraint that bans both [(ˌvaː).(ˌvaː).(ˈna.ŋa)] (33e) and [(ˌva.na).(ˌvaː).(ˈna.ŋa)] (33f) without affecting the other candidates. There is an observation in Kieviet (2017:42, 66) that heavy syllables are more common word-initially, and therefore, (33e-f) could be disfavored by violating this tendency. Nevertheless, as also noted in Kieviet (2017:42), “[of] all 329 three- or four-foot words …, 164 have initial H, 47 have one or two medial H, while 35 have final H”. The statistics suggest that the tendency to avoid noninitial heavy syllables is not very strong for longer words (164 initial H vs. 82 noninitial H). Furthermore, some reduplicated words do end with a heavy syllable. For a LLH-type root such as [ʔa.u.eː] ‘to cry out’, the reduplicated form is [ʔa.u.ʔa.u.eː] ‘to cry repeatedly’ (Kieviet 2017:63). This direction therefore does not seem promising. Another avenue to pursue is to assume a different ranking for reduplication. In the original proposal of MR, Stratal OT (Kiparsky 2000 a.o.) can be adopted for reduplication-phonology interaction (Saba Kirchner 2010, Bermúdez-Otero 2012). Furthermore, stratal ordering is adopted in the current analysis in Chapter 3. Assuming reduplication is a word-level stratum, for the same set of constraints in (33), we can rank ALIGN-L and ALL-FT-R above MAX-μ in this stratum, and therefore, (33c) can be selected as the winner. However, this approach will overproduce shortening in other scenarios due to the pressure of ALL-FT-R. For instance, when /vaːnaŋa/ ‘to talk’ undergoes right-edge copying, vowel shortening will be triggered, i.e., *[(ˌva.na).ŋa.(ˈna.ŋa)] rather than [(ˌvaː).(ˌna.ŋa).(ˈna.ŋa)], since the former candidate incurs fewer violations of ALL-FT-R. Also, the reduplicated form of /ʔa.u.eː/ ‘to cry out’ will be predicted to be *[(ˌʔa.u).ʔa.(ˈu.e)]. Therefore, an account that relies solely on metrical constraints and stratal ordering faces difficulties without surface-to-surface correspondence. 4.3.4.2. Serial Template Satisfaction Another theory that involves an affixal prosodic template without reduplication-specific correspondence is Serial Templatic Satisfaction (“STS”, McCarthy, Kimper & Mullin. 2012). In 171 STS, reduplication comes about through the step-wise realization of empty prosodic templates in Harmonic Serialism (McCarthy 2000, 2002, 2007). In what follows, I will show how this theory of reduplication also faces problems in generating vowel shortening if Surface Correspondence is not available. In STS, one mechanism to populate the affixal prosodic template is COPY, and it is motivated by a markedness constraint, HEADEDNESS(X), which penalizes any prosodic constituent of type X without a head of type (X–1). The input is gradually altered in each step. Below is an example to illustrate the core COPY mechanism of STS. (34) The step of copying in STS (adapted, McCarthy et al. 2012:193) σ + pa.ta HEADEDNESS(σ) *COPY(seg) *CODA ☞ a. pa.pa.ta 1 b. pat.pa.ta 1 1W c. σ + pa.ta 1W L In this example, the input of the current step contains an empty syllable template. The dominating HEADEDNESS constraint drives reduplication and eliminates (34c) where the syllable template is not headed. The template is filled by copying segments from the root, as in the winner in (34a), incurring a violation of *COPY(seg), while the constraint *CODA prevents a CVC reduplicant, in (34b). Note that *COPY(seg) is defined over strings. Thus, a single operation of copy may copy a string of multiple segments, which incurs only one violation of *COPY(seg), no matter how many segments are contained in the copied string. Regarding the reduplicative patterns of Rapa Nui, the situation is complicated by vowel shortening in the process of reduplication. Rapa Nui shows an interaction between metrical structure and reduplication, and the vowel shortening process may alter the metrical structure during the derivation in Harmonic Serialism. Therefore, the major issue for the STS analysis lies in when and why vowel shortening should take place. On the topic of foot construction in STS, McCarthy et al. say the following (2012:183-184, footnote 8): 172 “The assumption that reduplicative templates are affixed to fully prosodified stems is, of course, just the traditional assumption that prosodic structure is assigned cyclically (e.g., Ito 1986; Kiparsky 1979). In STS, it means that the constraints favoring bottom-up parsing (PARSE-SEGMENT, PARSE-SYLLABLE) dominate the constraints that favor filling the template, top-down. Indeed, it is hard to see how it could be otherwise: it is not possible to concatenate a template consisting solely of prosodic structure with a stem that lacks prosodic structure. For further discussion, see Sect. 6.2. Also see Wolf (2008) for a general theory of cyclicity in an HS-like system.” For all the tableaux in McCarthy, Kimper & Mullin (2012), the derivation always starts with the prosodic template affixed to a fully prosodified stem, since this is the “relevant part of the HS derivation” (McCarthy, Kimper & Mullin 2012:183). Nevertheless, the actual underlying representation, or the input to phonology, should contain an unparsed string of segments plus a prosodic template for reduplication. This point is hinted in McCarthy, Kimper & Mullin (2012:202), where the underlying representation of Chumash reduplication is given as /k-σ- ʔaniš/. Thus, a complete derivation should begin with syllable parsing and foot construction, before reduplication takes place. In STS, the driving force of reduplication is the constraint HEADEDNESS which requires a prosodic template to be headed. Therefore, for the input /μμ + vaːnaŋa/ in Rapa Nui, the foot structure of the root needs to be built first, before reduplication takes place, and the constraint HEADEDNESS is dominated by other constraints like PARSE. Based on the theoretical proposal for STS, the potential derivations of Rapa Nui intensifying reduplication are illustrated in (35) (syllable structures are omitted), and vowel shortening could possibly take place at different steps, either before or after reduplication. In the derivations below, I follow the Strict Inheritance assumption regarding foot construction in Harmonic Serialism, namely, GEN can only build one foot at each iteration and it cannot remove or alter any previously built feet (Pruitt 2010:486). 173 (35) Potential derivations in STS: vowel shortening of HLL roots during left-edge copying a. b. c. Again, the prosodic template for Rapa Nui intensifying reduplication is treated as two moras (μμ). Though it is not entirely clear how a mora template works in the current model of STS, the major issue in the derivations above is related to the shortening process and all the derivations in (35) run into certain problems. For (35a-b), shortening takes place before copying, which may eventually produce the expected output [(ˌva.na).va.(ˈna.ŋa)]. However, the constraint(s) that drive vowel shortening, need to be ranked above the constraint that triggers reduplication, i.e., HEADEDNESS(μ), because the ranking of markedness constraints regulates the steps of derivation in Harmonic Serialism. This will cause an issue of overprediction, namely, an input with a bare HLL root outside reduplication will undergo vowel shortening as well. Consider the tableaux below: shortening copy → → → → template "lled (output) input to phonology build foot vaː na ŋa ft μ μ μ μ μμ + va na ŋa ft μ μ μμ + μ va na ŋa ft μ μ μμ + μ va na ŋa ft μ μ μ ft va na μ μ … shortening va na ŋa ft μ μ μ vaː na ŋa ft μ μ μ μ copy → → → → … template "lled (output) va na ŋa ft μ μ μ ft va na μ μ input to phonology build foot vaː na ŋa ft μ μ μ μ μμ + μμ + μμ + input to phonology build foot vaː na ŋa ft μ μ μ μ vaː na ŋa ft μ μ μ μ template !lled → → … build foot → vaː na ŋa ft μ μ μ μ ft → shortening → ft μ μ vaː na ŋa μ μ ft va na ŋa ft μ μ μ ft va na μ μ ft ft va na μ μ → … … copy μμ + μμ + μμ + 174 (36) Potential grammar for (35a) Step 1: shortening μμ + vaːnaŋa *Vː MAX-μ PARSE HD(μ) ☞ a. μμ + va.na.ŋa 1 3 2 b. μμ + vaː.(ˈna.ŋa) 1W L 1L 2 c. μμ + vaː.na.ŋa 1W L 3 2 Step 2: build foot μμ + va.na.ŋa *Vː MAX-μ PARSE HD(μ) ☞ d. μμ + va.(ˈna.ŋa) 1 2 e. μμ + va.na.ŋa 3W 2 … (following steps are omitted) In (36), a possible driving force of shortening is *Vː, which has to dominate MAX-μ in order to motivate vowel shortening in an early step. However, the grammar in (36) faces the problem of overprediction. If we assume *Vː >> MAX-μ, as in (36), the grammar is not specific to the reduplicative context, and the issue of overprediction will arise. When the input only contains /vaːnaŋa/, for instance, vowel shortening will also take place. See the illustration in (37). (37) Overprediction of *Vː >> MAX-μ Step 1: shortening vaːnaŋa *Vː MAX-μ PARSE HD(μ) ☞ a. va.na.ŋa 1 3 b. vaː.(ˈna.ŋa) 1W L 1L c. vaː.na.ŋa 1W L 3 Step 2: build foot va.na.ŋa *Vː MAX-μ PARSE HD(μ) L d. va.(ˈna.ŋa) 1 e. va.na.ŋa 3W If we posit such a grammar where the driving force of truncation is ranked high, vowel shortening will take place in every context, not just limited to the reduplicative context. 175 If the derivation in (35b) is considered, a potential grammar is illustrated in (38). (38) Potential grammar for (35b) Step 1: build foot μμ + vaːnaŋa ALL-FT-R MAX-μ *Vː PARSE HD(μ) ☞ a. μμ + vaː.(ˈna.ŋa) 1 1 2 b. μμ + vaː.na.ŋa 1 3W 2 c. μμ + va.na.ŋa 1W L 3W 2 Step 2: shortening μμ + vaː.(ˈna.ŋa) ALL-FT-R MAX-μ *Vː PARSE HD(μ) d. μμ + va.(ˈna.ŋa) 1W L 1 2 L e. μμ + vaː.(ˈna.ŋa) 1 1 2 f. μμ + (ˌvaː).(ˈna.ŋa) 2W 1 L 2 … (following steps are omitted) For (38), to ensure shortening does not take place in the first step, MAX-μ needs to dominate *Vː so that (38c) is ruled out. However, with the ranking MAX-μ >> *Vː, no vowel shortening can be triggered in the second step. Thus, the candidate in (38e) will be the winner of the second step, though (38d) is the intended output. The derivation (35b) therefore does not seem viable, either. Finally, I will turn to the derivation in (35c). For this derivation, the motivation for vowel shortening is problematic as well, which can be further illustrated by the tableaux below: (39) Potential grammar for the derivation in (35c) Step 1: build foot μμ + vaːnaŋa MAX-μ FTBIN PARSE HD(μ) ☞ a. μμ + vaː.(ˈna.ŋa) 1 2 b. μμ + vaː.na.ŋa 3W 2 c. μμ + va.na.ŋa 1W 3W 2 Step 2: build foot μμ + vaː.(ˈna.ŋa) MAX-μ FTBIN PARSE HD(μ) ☞ d. μμ + (ˌvaː).(ˈna.ŋa) 2 e. μμ + va.(ˈna.ŋa) 1W 1W 2 176 … copying and filling the template (omitted) Step n: shortening (ˌva.na).(ˌvaː).(ˈna.ŋa) MAX-μ FTBIN PARSE HD(μ) L f. (ˌva.na).(ˌvaː).(ˈna.ŋa) g. (ˌva.na).(ˌva).(ˈna.ŋa) 1 1 In Step 1 and Step 2, feet of the root are gradually constructed while vowel shortening is prevented by the top-ranked MAX-μ. Given the current constraint ranking, however, the structure cannot be further improved in later stages through vowel shortening, as shown in the last tableau in (39). More specifically, there needs to be another constraint which can make (39g) [(va.na).(va).(na.ŋa)] favored over (39f) [(va.na).(vaː).(na.ŋa)], and this constraint should outrank MAX-μ and FTBIN. 42 However, this runs into a similar issue to the one in §4.3.4.1 (recall the tableaux in (33) above) since the markedness constraint that would drive mora deletion is not well motivated. Though there is some speculation made in §4.3.4.1, it does not seem satisfactory to eliminate the candidate [(ˌva.na).(ˌvaː).(ˈna.ŋa)] with ad hoc constraints. In addition, given the Strict Inheritance assumption of GEN (Pruitt 2010:486), [(ˌva.na).va.(ˈna.ŋa)], can never be produced in the derivation of (35c), since none of the established feet in [(va.na).(vaː).(na.ŋa)] is allowed to be removed in further steps. Summing up, it is unclear when and why vowel shortening takes place in the derivation. If vowel shortening occurs early, as in (35a-b), a problem of overprediction will arise. If the process occurs later, as in (35c), there is no satisfactory motivation, and the Strict Inheritance assumption of GEN makes the grammar unable to produce the expected output [(ˌva.na).va.(ˈna.ŋa)]. Admittedly, this assumption is not uncontroversial, and some analyses chose to allow GEN to make multiple changes in terms of stress and metrical structure (e.g., McCarthy 2008). Nevertheless, whichever assumption about GEN is made, the abovementioned problem of overprediction and the puzzling motivation for vowel shortening still persists. 42 The candidate in (39g) [(ˌva.na).(ˌva).(ˈna.ŋa)] also violates *CLASH. 177 4.3.5. Summary This section has provided an analysis of Rapa Nui intensifying reduplication, focusing on the motivation of vowel shortening. The patterns of intensifying reduplication are repeated below in (40). (40) Summary of intensifying reduplication patterns Root shape Schematic representation Left-edge copying Right-edge copying a. H CVː CVː-CVː b. LL CV.CV CV.CV-CV.CV c. HLL CVː.CV.CV CV.CV-CV.CV.CV CVː.CV.CV-CV.CV d. LLL CV.CV.CV CV.CV-CV.CV.CV CVː.CV.CV-CV.CV e. LLH CV.CV.CVː CV.CV-CV.CV.CVː n.a. The template of intensifying reduplication is two moras. When the root is monosyllabic or disyllabic (40a-b), the reduplicated form is either CVː-CVː or CVCV-CVCV. When the root is trisyllabic HLL or LLL, either left-edge copying or right-edge copying is possible, and some roots can have both forms (40c-d). When a HLL root undergoes left-edge copying, the long vowel in the root is shortened, which is a case of truncatory backcopying and the focus of this chapter (§4.3.3.1). When a LLL root undergoes right-edge copying, the initial short vowel is lengthened to create a well-formed foot (§4.3.3.2). For another shape of trisyllabic root, LLH, only left-edge copying is found. The constraints in this section, as well as their interaction with the metrical constraints introduced in §4.2, are summarized in (41). 178 (41) Hasse diagram The key issue in this section is how to generate vowel shortening in the base during reduplication. As presented in §4.3.3, I demonstrated that it can be attributed to the enforcement of IDENT-VV(length) via a SCorr relation established between surface vowels. In addition, it does not seem possible to generate this pattern without resorting to correspondence, as discussed in §4.3.4. Therefore, this section provides an argument for the correspondence- based approach and the utility of Generalized Surface Correspondence in reduplication. Nevertheless, the discussion so far does not compare the current proposal with another correspondence-based approach to reduplication, Base-Reduplicant Correspondence Theory. In theory, this pattern can be also analyzed with BR correspondence (see Yang 2022 for details). The comparison between the current proposal and BRCT will be presented shortly in §4.6 after the analysis of Rapa Nui plural formation, based on an overall consideration of intensifying reduplication and plural formation. 4.4. Plural formation in Rapa Nui This section introduces and analyzes the plural formation in Rapa Nui. Some highlighted data are previewed in (42). ALL-FT-R ALIGN-L RIGHTMOST DEP-μ OO-PM CORR-XX, ID-VV(length) ACCORD-XX MAX-μ, PARSE 179 (42) Plural formation in Rapa Nui Root Plural form gloss a. ʔaɾa ʔaʔaɾa ‘to wake up (PL)’ b. haʔuɾu haːʔuɾu ‘to sleep (PL)’ Rapa Nui verbal plurals can be formed by either reduplication (42a) or lengthening (42b), largely depending on the shape of the root. In this analysis, I will show how this pattern is best analyzed as mora affixation, namely, an underlying affixal mora can lead to both reduplication and lengthening. This analysis aims to achieve another goal of the current research, i.e., to demonstrate the advantage of mora affixation in dealing with non-concatenative morphology. For the rest of this section, Section 4.4.1 presents the data; Section 4.4.2 gives the analyses; Section 4.4.3. compares the current proposal with Base Reduplicant Correspondence Theory. A summary and some closing remarks are made in Section 4.4.4. 4.4.1. Data The phenomenon under investigation in this section is referred to as “plural formation” which mainly applies on verbs to express plurality. Though some adjectives also exhibit the same pattern, I will focus on the formation of plural verbs in the current discussion. An example is given in (43) to illustrate the typical function of this morphological construction. Note that Rapa Nui has VSO word order (Chapin 1978). (43) He to-topa o mātou ki raro. NTR 43 PL:descend of 1PL.EXCL to below ‘We went down.’ (K68) In this example, the first syllable of the verb topa [ˈto.pa] ‘to descend’ is reduplicated to agree with the plural subject mātou [ˌmaː.ˈto.u] ‘we (exclusive)’. 43 NTR is short for Neutral Aspect. 180 The formation of plural verbs is regular. In general, plural formation is only observed in two shapes of verbs, LL and LLL 44 , as exemplified in (44). (44) Plural formation in Rapa Nui (transcribed in IPA; stresses are not marked) Singular Gloss Plural Gloss Source a. word shape: LL ʔaɾa to wake up ʔaʔaɾa to wake up (PL) K62 eke to mount eeke to mount (PL) K62 tuɾu to go down tutuɾu to go down (PL) K62 hopu to bathe hohopu to bathe (PL) K172 kai to eat kakai to eat (PL) K231 b. word shape: LL mate to die maːmate to die (PL) K62 piko to hide piːpiko to hide (PL) K62 teɾe to run teːteɾe to run (PL) K62 neʔi to defecate neːneʔi to defecate (PL) K72 hiŋa to fall hiːhiŋa to fall (PL) K164 c. word shape: LLL haʔuɾu to sleep haːʔuɾu to sleep (PL) K62 haʔeɾe to walk haːʔeɾe to walk (PL) K62 tahuti to run taːhuti to run (PL) K62 For the singular forms in (44a), the first syllable of a LL verb is reduplicated to indicate plurality. For those in (44b), the first syllable of a LL verb is reduplicated and lengthened, making the output a HLL word. For the LLL verbs in (44c), no reduplication is observed. Instead, the vowel in the first syllable is lengthened to indicate plurality. In the examples above, the primary observation is that word shape determines whether reduplication or lengthening should be chosen to indicate plurality, i.e., only words of LL shape can be reduplicated while longer words of LLL shape choose vowel lengthening. Secondly, some LL words undergo both reduplication and lengthening, as in those in (44b). Kieviet (2017:61) pointed out that whether the reduplicated 44 There is only one LLLL word that undergoes this process, [paŋahaʔa] ‘heavy’ → [pa-paŋahaʔa] ‘heavy (PL)’. It is not included in the discussion since it is an adjective. 181 syllable is lengthened (44b) or not (44a) is lexically determined. The verbs in (44a) and (44b) are mutually exclusive. Nevertheless, the verbs that have a plural form like those in (44) only account for a small portion in the lexical database (Kieviet 2017:68). Many verbs do not have any distinct plural forms at all. See some examples in (45). 45 (45) Verbs without a distinct plural form Verb Gloss Source (evidence) aŋa to make K231 noho to stay K202 manaʔu to think K480 tikeʔa to see K497 vaːnaŋa to talk K543 Furthermore, for another set of verbs, though they exhibit the same pattern of reduplication, the meaning after reduplication is somewhat divergent from the meaning of the base. Some examples are given in (46). (46) Derived meaning after reduplication Base Gloss Reduplicated Gloss Source 46 a. moɾe to be cut, wounded momoɾe to harvest, pick; to break K68 b. puhi to blow pupuhi to shoot (with a weapon) K68 c. ɾehu to be forgotten ɾeɾehu to faint K68 d. viɾi to wrap viːviɾi to roll K99/K97 45 Kieviet (2017) did not explicitly list the verbs that do not have a plural form. However, some of these verbs can be identified based on their syntactic contexts. For example (Kieviet 2017:497): He tikeʔa mātou e tahi kāiŋa ʔitiʔiti … NTR see 1PL.EXCL NUM one homeland small:RED … ‘We saw a small island … ’ In this example, the subject is plural (first person plural exclusive) and the verb ‘tikeʔa’ does not involve any lengthening or reduplication. Further, the verb ‘tikeʔa’ can be verified to be the infinite form in the dictionary of Fuentes (1960:863) (As a side note, the word is documented as ‘tikea’ in Fuentes 1960:863, meaning ‘to look’, ‘to meet’, ‘to demonstrate’, etc.) 46 The sources of base and reduplicated form are listed before and after the slash respectively. 182 This set of verbs is different from the examples in (44) with respect to the function. As mentioned earlier, the verbs in (44) undergo phonological change to agree with the subject/agent in number, which can be treated as the phonological realization of an inflectional morpheme PLURAL. The function of the reduplicated verbs in (46), however, is more likely to be derivational. One piece of evidence is that the reduplicated forms in (46) do not necessarily co- occur with a plural subject, as shown in (47). (47) a. … he vī-viri te henua. NTR roll ART land. ‘… the land rolls.’ (K97) b. ’ina a Tiare kai mate; ko re-rehu nō ’ā NEG PROP Tiare NEG die PRF faint just CONT ‘Tiare was not dead; she had just fainted.’ (K343) In (47), both examples contain a reduplicated verb, vī-viri [ˌviːˈviɾi] ‘to roll’ and re-rehu [ɾeˈɾehu] ‘to faint’, but neither subject is plural, indicating an apparently different function from the example in (43). Based on the description of plural formation in Kieviet (2017) and the discussion above, verbs in Rapa Nui can be categorized into four subsets: (48) Categorization of Rapa Nui verbs Subsets Description Subset A verbs (LL and LLL) that undergo plural formation, choosing either reduplication or lengthening; (44a) and (44c) Subset B verbs (LL) that undergo plural formation; both reduplication and lengthening take place; (44b) Subset C verbs that do not have a plural form that is distinct from the singular form; (45) Subset D verbs that have a derived meaning after reduplication; not necessarily plural; (46) 183 Among the four subsets in (48), Subsets A, B, and C are related to plurality, which can be treated as different realizations of an inflectional morpheme PLURAL. For Subset C, rather than describing the verbs as “no plural form”, it is reasonable to posit that the plural marker, no matter what the underlying phonological form is, surfaces as zero realization when it co-occurs with the verbs in this group, and this group accounts for most of the verbs in Rapa Nui (Kieviet 2017:68). It is not yet clear what criteria can be used to distinguish these subsets. The study of Kieviet (2017) suggests that the categorization of Subsets A, B, and C is lexically determined. 47 The observation that is crucial to the current study, however, is that the verbs in Subset A show regular processes of plural formation, and the choice between reduplication and lengthening can be predicted by the phonological grammar depending on word shape. In the following analysis, I will focus on the patterns of Subset A, since the variation between reduplication and lengthening can be attributed purely to phonological factors. This variation is also essential to the main theme of the research. I will leave the analyses of other subsets for future research. 4.4.2. Analysis Based on the data presented in the previous section, plural formation in Rapa Nui forms a sub- regularity, i.e., among each group of verbs that undergo this process, the pattern is regular. One insight provided in (Kieviet 2017:62-63) about plural formation in Rapa Nui is that it can be treated as mora affixation. More specifically, in the context of the current research, the phonological form of the plural morpheme is a prefixed mora (μ), and it can be either realized as reduplication, lengthening, or null output under different conditions. Previous studies have shown how an affixal mora can trigger various morphophonological processes, especially vowel lengthening (Davis & Ueda 2002, 2006, Wolf 2006, a.o.) and reduplication (Saba Kirchner 2010, 47 Although Kieviet (2017) did not mention this, the membership of Subset C should be partly determined by phonological factors. Given the description of patterns, if a verb root has a shape of HLL, it will not undergo either reduplication or lengthening. Thus, HLL roots should be Subset C. 184 Bye & Svenonius 2012, a.o.), and it is not uncommon that a single affixal mora can exhibit different realizations in different phonological contexts (e.g., Paster 2010). For Rapa Nui, when a mora representing a plural morpheme is in the phonological input, the possible outputs for the verbs in Subset A are illustrated in (49). (49) Possible outputs of mora affixation: PLURAL + √VERB (Subset A) Pattern Example Gloss a. (C)V reduplication (LL words) ‘to go down’ b. Vowel lengthening in verb (LLL words) ‘to sleep’ The phonological form of the plural morpheme, an affixal mora, is either populated through reduplication (49a) or incorporated into the verb (49b). The choice between these realizations is purely phonologically determined. In the following analysis, the realization of the affixal mora is proposed to be driven by the constraint MAXFLT (Wolf 2007:2). Recall that MAXFLT differs from MAX-μ, which has been discussed in Chapter 3, §3.3.1. (50) MAXFLT: All autosegments that are floating in the input have output correspondents. For Subset A, the grammar is able to realize the plural marker in two ways, depending on the input word shape. When the input is a LLL word, only lengthening is allowed. I propose that the realization of the affixal mora is enforced by MAXFLT, in conflict with two other constraints, INTEGRITY and IDENT-IO(length), which ban reduplication and lengthening respectively. The relevant constraints are defined in (51), followed by a tableau in (52). tu ɾu μ μ μ + tu tu ɾu μ μ → μ ha ʔu ɾu μ μ μ + → μ haː ʔu ɾu μ μ μ μ 185 (51) Constraints a. INTEGRITY: Assign a violation for every segment in the input that has multiple correspondents in the output. b. IDENT-IO(length): Assign a violation if two vowels standing in input-output correspondence are not associated with an equal number of moras. (52) Input /μ + haʔuɾu/ ‘to sleep’ /μ + haʔuru/ MAXFLT INTEG ID-IO(length) ☞ a. (ˌhaː).(ˈʔu.ɾu) 1 b. (ˌha.ha).(ˈʔu.ɾu) 2W L c. ha.(ˈʔu.ɾu) 1W L For a LLL word in Subset A, lengthening is chosen instead of reduplication, as in (52a), which incurs a violation of the lower-ranked IDENT-IO(length). The candidate in (52b) undergoes reduplication and fatally violates INTEGRITY. If the floating mora is deleted, as in (52c), this candidate is ruled out by MAXFLT. It is noteworthy that the evaluation of IDENT-IO(length) is not conflated with MAXFLT, and their functions can be distinguished by the definitions. IDENT- IO(length) evaluates the corresponding vowels and checks if they are associated with the same number of moras. MAXFLT, however, targets floating elements in the input. 48 When the input is a LL word, only reduplication is allowed. If the mora is incorporated into the root, the output will be a HL form instead of LLL. Kieviet (2017:38-39) mentioned that HL words are ill-formed and unattested in Rapa Nui. The potential footing of a HL word, (ˈH)L, violates RIGHTMOST. See the tableau in (53). The constraints RIGHTMOST and PARSE have been introduced earlier in §4.2. 48 This constraint is also different from MAX-μ, which requires that every mora in the input must have a correspondent in the output. Instead, MAXFLT only evaluates the floating autosegmental elements in the input. 186 (53) Input /μ + tuɾu/ ‘to go down’ /μ + tuɾu/ MAXFLT RTMOST PARSE INTEG ID-IO (length) ☞ a. tu.(ˈtu.ɾu) 1 2 b. (ˈtuː).ɾu 1W 1 L 1W c. (ˈtu.ɾu) 1W L L For a LL word in Subset A, lengthening is blocked by RIGHTMOST, which rules out (53b). Again, the candidate in (53c) that deletes the floating mora in the input fatally violates MAXFLT. Therefore, reduplication is the only viable repair strategy for the input floating mora. Additionally, MAXFLT needs to dominate PARSE, given (53a) ≻ (53c). Further, another candidate to be considered is [(ˌtuː).(ˈtu.ɾu)], where lengthening is applied after reduplication to create a complete foot. This candidate is eliminated due to the ranking DEP-μ >> PARSE, which is consistent with the discussion in §4.2 and §4.3. See (54) for demonstration. (54) Input /μ + tuɾu/ ‘to go down’ /μ + tuɾu/ DEP-μ PARSE INTEG ☞ a. tu.(ˈtu.ɾu) 1 2 b. (ˌtuː).(ˈtu.ɾu) 1W L 2 In sum, the analysis above shows that the choice between reduplication and lengthening is purely phonological. The ranking INTEGRITY >> IDENT-IO(length) indicates that lengthening is given priority, and reduplication takes place when the output is not an ill-formed HL word and satisfies RIGHTMOST, or lengthening will be adopted instead. This example shows the efficacy of mora affixation, and its flexibility to surface as various forms under different conditions. For Subset A, the constraint interaction is summarized in the Hasse diagram below. 187 (55) Hasse diagram 1. (52a) ≻ (52c) 2. (52a) ≻ (52b) 3. (53a) ≻ (53b) 4. (54a) ≻ (54b) 4.5. Reduplication in Rapa Nui: an interim summary The previous sections have discussed two morphological processes in Rapa Nui, intensifying reduplication and plural formation, both of which involve reduplication. For intensifying reduplication, the highlighted pattern is vowel shortening in the base, or truncatory backcopying, which forms an argument for a correspondence-based approach. In the meanwhile, initial vowel lengthening has also been analyzed. For plural formation, the focused pattern is the alternation between reduplication and lengthening. The analysis shows that both plural forms can be attributed to an underlying affixal mora. The realization is phonologically determined by the shape of the verb root. The patterns and constraint rankings are summarized in (56). IDENT-IO(length) PARSE INTEGRITY RIGHTMOST MAXFLOAT DEP-μ 1 2 3 4 188 (56) Reduplication in Rapa Nui UR Surface patterns Grammar a. Intensifying reduplication μμ Reduplication with shortening in initial syllable of root Reduplication with lengthening in initial syllable of reduplicant b. Plural formation μ Reduplication Lengthening 4.6. Alternative analysis with BRCT A major goal of this dissertation is to argue for a correspondence-based approach to reduplication. The specific type of correspondence in reduplication that I argue for is Generalized Surface Correspondence. However, the classic approach to reduplication, Base Reduplicant Correspondence Theory, is another correspondence-based theory that needs to be discussed. Although two theories (MR without SCorr and STS) have been visited earlier in the chapter (§4.3.4), the discussion of BRCT has been reserved for the current section, after the analyses of both intensifying reduplication and plural formation in Rapa Nui. In what follows, I will discuss how the patterns and analyses of Rapa Nui so far favor the current MR + GSC model over BRCT, based on an overall consideration of both morphological processes in Rapa Nui. BRCT has two properties. First, it analyzes reduplication as the concatenation of RED, a specialized mechanism for reduplication. Second, it has a form of correspondence among MAX-μ, PARSE DEP-μ CORR-XX, ID-VV(length) ACCORD-XX OO-PROSMATCH IDENT-IO(length) PARSE INTEGRITY RIGHTMOST MAXFLOAT DEP-μ 189 surface elements, namely, BR Correspondence, which is initiated by the presence of RED in the input. In the current analysis, MR + GSC, these two aspects are replaced with affixal prosodic templates and Surface Correspondence respectively. The advantage of analyzing reduplication as epiphenomenal to empty prosodic templates lies in the analysis of Rapa Nui plural formation. As shown in §4.4.2, the alternation between lengthening and reduplication can be simply attributed to a single mora underlyingly. The actual realization is based on the shape of the root (LL or LLL). On the other hand, an apparent alternative is to assume that there is a RED morpheme for reduplication, if BRCT is pursued. In this case, reduplication and lengthening need to be attributed to two different underlying forms, i.e., RED for reduplication and an affixal mora (μ) for lengthening. Thus, plural formation in Rapa Nui is essentially a case of Phonologically Conditioned Suppletive Allomorphy (PCSA, Carstairs 1988, 1990, among many others). With this proposal, the two URs need to be listed in the input to phonology (e.g., Mascaró 2007), and the selection between RED and μ is determined by the grammar. An illustration is given in (53), following the treatment of PCSA in Mascaró (2007). (57) Allomorphy selection in Rapa Nui plural formation /{RED, μ} + tuɾu/ RTMOST INTEG ID-IO(length) ☞ a. tu.(ˈtu.ɾu) 2 b. (ˈtuː).ɾu 1W L 1W /{RED, μ} + haʔuru/ RTMOST INTEG ID-IO(length) ☞ c. (ˌhaː).(ˈʔu.ɾu) 1 d. (ˌha.ha).(ˈʔu.ɾu) 2W L The sample analysis in (57) is exactly the same as the grammar in (55). If we analyze reduplication with an underlying RED, two URs need to be listed in the lexicon. Since assuming two URs does not simplify the grammar as a trade-off, the analysis with RED introduces redundancy, compared to the single-UR analysis proposed in this study (i.e., mora affixation). This forms another argument for the emergent view of reduplication, together with the case 190 study of Huozhou Chinese in Chapter 3, where reduplication and subtraction are both epiphenomenal to mora affixation. The second property of BRCT is BR correspondence, a form of correspondence among surface elements. In §4.3.3, I have demonstrated that Surface Correspondence can be used to handle the vowel shortening pattern in Rapa Nui intensifying reduplication. In addition, the alternative analyses with Level Ordering or Harmonic Serialism, which do not resort to correspondence, cannot generate the expected patterns, at least not without further supplementation. This supports a correspondence-based approach to reduplication. However, vowel shortening can be also well handled by BR correspondence. See the illustration in (58) (after Yang 2022). (58) Vowel shortening driven by BR correspondence RED + vaːnaŋa ID-BR (length) MAX-μ PARSE ALL-FT-R ☞ a. (ˌva 1 .na 2 ).va 1 .(ˈna 2 .ŋa 3 ) 1 1 3 b. (ˌva 1 .na 2 ).(ˌvaː 1 ).(ˈna 2 .ŋa 3 ) 1W L L 5W The abridged tableau in (58) shows the essential analysis of vowel shortening with BR correspondence. The shape of the reduplicant is attributed to RED = Ft in this analysis, which is omitted here. With BR correspondence established between RED and BASE, the constraint IDENT- BR(length) drives vowel shortening when it is ranked higher than MAX-μ. The mechanism in (58) is similar to the analysis with SCorr in §4.3.3.1, but the difference between these two analyses is worth mentioning. In the analysis with BRCT, BR correspondence is invoked by the presence of RED in the input. The shape of the reduplicant can be either emergent (Urbanczyk 1996, McCarthy & Prince 1999, Downing 2001, a.o.) or determined by template constraint (e.g., RED = Ft, McCarthy & Prince 1995). If SCorr is adopted instead, Surface Correspondence is freely assigned between segments without any prerequisite, but the constraint ACCORD needs to be included to avoid unexpected SCorr structures, as discussed in 191 §4.3.3.1. The SCorr analysis requires a means of restricting correspondence to reduplicative context only, although ACCORD can partially accomplish the goal. Taking all these considerations together, the SCorr analysis still has advantages over BRCT. As a main argument in this dissertation, SCorr is a more general mechanism in grammar, which is not proposed for a specific phenomenon. Its ability to handle long-distance assimilation and dissimilation, as well as its utility in compensatory reduplication (e.g., Yu 2005, Inkelas 2008), shows the versatility of this theoretical mechanism. Thus, if SCorr can serve the same function as BR correspondence, a more general mechanism is more favorable than a highly specialized one. Although a new constraint ACCORD needs to be included in the current analysis, it captures the potential connection between segment fission and correspondence on the surface, which has long been noted in the literature (e.g., Raimy & Idsardi 1997, Kawu 1999, Struijke 2002, Karabay 2004). In this way, the constraint ACCORD is well motivated instead of an ad hoc solution to rescue the issue of unexpected correspondence. Finally, in the SCorr approach we need to specifically restrict Surface Correspondence to a reduplicative context in order to avoid overprediction. Although this issue seems to be a potential drawback, there are several ways to solve this problem without effort. In the analysis of Rapa Nui, limiting SCorr to reduplication can be treated as a side effect of ACCORD. In other cases, such as Huozhou diminutive reduplication, the morpheme-specific surface identity is attributed to Level Ordering, since ACCORD needs to be ranked low in this case. The point is that any theoretical proposals that aim to deal with morphologically-conditioned phonology are potential solutions to this problem, be it Level Ordering or cophonologies. Nevertheless, no matter which strategy is adopted to solve the issue, a surface-to-surface correspondence is needed to handle reduplicative opacity. Summing up, although the SCorr approach needs an additional ACCORD as a limiter of correspondence structure, it is more versatile in general, and therefore, the SCorr approach reduces the formal device that is reduplication-specific. In addition, the single-UR analysis of plural formation in Rapa Nui also suggests that it is possible to obviate RED as the underlying 192 representation of reduplication. The overall comparison between these two correspondence- based approaches favors the SCorr approach over BRCT. 4.7. Summary This chapter has presented the reduplication patterns in Rapa Nui, with a focus on the vowel shortening process in intensifying reduplication (when HLL-type roots undergo left-edge copying) and plural formation. The data for intensifying reduplication show another case of truncatory backcopying, which supports a correspondence-based approach to reduplication. The proposed analysis with SCorr shows that vowel shortening in Rapa Nui reduplication is straightforward when it is attributed to the ranking of IDENT-VV(length) >> MAX-μ in a parallel analysis. The motivation for vowel shortening is problematic for theories that resort to level ordering or Harmonic Serialism, such as the original proposal of Minimal Reduplication and Serial Template Satisfaction. Additionally, though Base Reduplicant Correspondence Theory is another correspondence-based approach to reduplication, the model with SCorr in the current study is more advantageous than BRCT, when taking an overall consideration of the analyses in this chapter. 193 Chapter 5 Generalized Surface Correspondence revisited This chapter reflects on the proposed model of Generalized Surface Correspondence. Some issues that were mentioned in Chapter 2 (§2.2) have been reserved for the current chapter, after the analyses of Huozhou diminutive reduplication and Rapa Nui intensifying reduplication. In this dissertation, Generalized Surface Correspondence is proposed as a version of Agreement-by-Correspondence (ABC, Hansson 2001/2010, Rose & Walker 2004, a.o.). In almost every version of ABC, there is a set of constraints invoking Surface Correspondence (usually CORR) and another set of constraints impelling the identity between corresponding segments (IDENT- XX) (cf. McCarthy 2010, Shih 2013, Hansson 2014). The constraints adopted in this dissertation are repeated in (1). The constraints CORR-XX and IDENT-XX follow the proposals in Walker (2015), while ACCORD-XX is newly proposed in this dissertation. (1) Constraints in GSC a. CORR-XX: Assign a violation to any pair of segments that are not in correspondence in the output. b. IDENT-XX [ 𝛾 G] (F): Let X and Y be a consecutive pair of corresponding [𝛾G] segments in the output. Assign a violation if X is [𝛼F] and Y is [𝛽F] (where 𝛼 ≠ 𝛽). c. ACCORD-XX: Let X i and X j be two segments in the input, and X i ′ and X j ′ be two segments in the output; and let X i ′ ℜ X i and X j ′ ℜ X j . Assign a violation if X i ′ and X j ′ form a consecutive pair that stands in Surface Correspondence. 194 CORR and IDENT-XX constraints exist in almost every version of ABC, which forms the essential mechanisms of the theory, but the ways to define these constraints vary in the literature. The current proposal follows Walker’s (2015) model, where similarity is encoded in the feature-restricted IDENT-XX [𝛾 G] (F) constraints. The function of CORR-XX is minimal and diverges from traditional CORR constraints in earlier literature (Hansson 2001/2010, Rose & Walker 2004, a.o.). Also, the limiter constraints in this model, including both IDENT-XX [𝛾 G] (F) and ACCORD-XX, evaluate local pairs of corresponding segments, but CORR-XX is defined in a global and transitive way. In what follows, I will discuss the rationale of the current model, drawing evidence from previous research and the case studies in this dissertation. Section 1 focuses on the issue of feature-restricted IDENT-XX [𝛾 G] (F) and why CORR-XX is needed. Section 2 turns to the issue of local versus global evaluation of the constraints in GSC. 5.1. Similarity threshold and the role of CORR-XX The core insight of ABC is that Surface Correspondence is favored between similar segments while the corresponding segments are impelled to be identical by IDENT-XX constraints. A related issue is how to encode the similarity threshold in the constraints. In most ABC literature, similarity is encoded in CORR constraints (Hansson 2001/2010, Rose & Walker 2004, Bennett 2013, Shih & Inkelas 2019 a.o.). I use Bennett’s (2013) proposal as an example. (2) Constraints in Bennett (2013:55, 72) a. CORR-D·XX[𝛼F]: ‘if two [𝛼F] segments are in the same domain D, they must correspond.’ For each distinct pair of output consonants, X & Y, assign a violation if: i. X & Y both have the feature specification [𝛼F], and ii. X & Y are both in the same domain D, and iii. X & Y are not in a surface correspondence relationship. 195 b. IDENT-XX[F] ‘If two distinct segments correspond, then they agree on [F]’ i. For each distinct pair of output consonants X & Y, assign a violation if: ii. a. X & Y are in a surface correspondence relationship, and iii. b. X and Y do not agree in feature [F]. The similarity threshold is encoded in CORR-XX[𝛼F] as [𝛼F]. This constraint is in the same line as Rose & Walker (2004), where CORR is formulated as CORR-X↔Y (X and Y represent segments that share certain features). For a string [s … m … p … l], the constraint CORR-XX[labial] favors a SCorr relation between [m] and [p] that share the feature [labial], i.e., [s … m x … p x … l]. This is how the similarity threshold is controlled by CORR-XX[𝛼F]. In another line of research, however, similarity is encoded as *[𝛼F][–𝛼F] [𝛽 G, 𝛾 H] that is conceptually similar to the current IDENT-XX [𝛾 G] (F) constraints (Hansson 2014, Walker 2015; see illustrations in Lionnet 2016, Sande 2019). Meanwhile, CORR is abandoned (Hansson 2014, cf. McCarthy 2010, Shih 2013) or its function is reduced to a minimum (Walker 2015). Feature- specific CORR constraints, such as CORR-XX[𝛼F], were argued to be redundant and unmotivated by McCarthy (2010). Further, Hansson’s (2014) proposal of Agreement-by-Projection (ABP) eliminates CORR constraints, only leaving *[𝛼F][–𝛼F] [𝛽 G, 𝛾 H] in the system. The definition of *[𝛼F][-𝛼F] [𝛽 G, 𝛾 H] is given in (3) (after Sande 2019:484). (3) *[𝛼F][–𝛼F] [𝛽 G, 𝛾 H] : A segment with a given set of feature values may not directly precede another segment with a different set of feature values in the ordered set of output segments that are [𝛽G, 𝛾H]. Assign one violation for each output form where at least one pair of vowels or consonants meets these criteria. The ABP constraint *[𝛼F][–𝛼F] [𝛽 G, 𝛾 H] has been favored by Hansson (2014) and adopted in several works (Lionnet 2016, Sande 2019). Walker (2015) argues for feature-restricted IDENT-XX [𝛾 G] (F), which is conceptually similar to *[𝛼F][–𝛼F] [𝛽 G, 𝛾 H] in ABP, to handle the problem of discrete triggers in a harmony pattern found in the Pasiego Montañes dialect (cf. Walker 2018). The 196 arguments in these works offer additional support for transferring (at least partially) the labor of CORR-XX[𝛼F] to IDENT-XX. Therefore, the current proposal of GSC follows the same line of research and adopts feature-restricted IDENT-XX [𝛾 G] (F) in the model. Although Walker (2015) offers evidence for the benefits of feature-restricted IDENT-XX [𝛾 G] (F), whether the correspondence-invoking constraint CORR-XX can be eliminated altogether remained an open question. Nevertheless, the utility of the correspondence-invoking constraint can be seen in the case study of Rapa Nui vowel shortening. In this case, the output exhibits vowel shortening, which is attributed to IDENT-XX [+vocalic] (length) or IDENT-VV(length) as a shorthand. The role of CORR in this case, teamed up with ACCORD, is to assign “proper” correspondence structures to the candidates. If such a mechanism is lacking, the analysis results in a similar problem of overprediction as discussed in Chapter 4. For instance, if the ABP-style constraint *[–long][+long] [+vocalic] is used for Rapa Nui vowel shortening (assume [long] for vowel length for now), we would expect an analysis as follows. (4) Input /μμ + vaːnaŋa/ and hypothetical input /μμ + tatataː/ /μμ + vaːnaŋa/ *[–long][+long] [+vocalic] MAX-μ ☞ a. va.na-va.na.ŋa 1 b. va.na-vaː.na.ŋa 1W L /μμ + tatataː/ *[–long][+long] [+vocalic] MAX-μ L c. ta.ta-ta.ta.ta 1 d. ta.ta-ta.ta.taː 1W L Recall that Rapa Nui requires vowel shortening in the base after left-edge reduplication unless the long vowel is the last one in the base. Therefore, the ranking *[–long][+long] [+vocalic] >> MAX-μ leads to vowel shortening in (4a). However, without a correspondence-invoking constraint CORR- XX and ACCORD-XX that restricts the relation between SCorr and input-output correspondence, the constraint *[–long][+long] [+vocalic] overpredicts vowel shortening in (4c), producing [ta.ta.ta.ta.ta] rather than [ta.ta.ta.ta.taː]. 197 In sum, along the line of previous research, there have been some supportive arguments for IDENT-XX [𝛾 G] (F) or *[𝛼F][–𝛼F] [𝛽 G, 𝛾 H] in ABP (Hansson 2014, Walker 2015; see also Lionnet 2016, Sande 2019), but the role of violable CORR-XX that invokes correspondence between segments is still needed, evidenced by the demonstration of Rapa Nui in (4). Nevertheless, the function of CORR-XX is reduced to a minimum since it does not refer to any features, which partially addresses McCarthy’s (2010) concern about the redundancy of feature-specific CORR constraints. 5.2. Local versus global evaluation of GSC constraints The second issue is whether the evaluation of CORR-XX and IDENT-XX should be local or global. The exact way of evaluation and its consequences have been a debate in the literature of ABC. Local evaluation of IDENT-XX has been supported by Hansson (2007) and adopted in subsequent works (Inkelas & Shih 2014, Walker 2015, Shih & Inkelas 2019, a.o.). In the chain [s x 1 … s x 2 … z x 3 ], for example, if agreement is evaluated locally by IDENT-XX, only the pairs [s x 1 … s x 2 ] and [s x 2 … z x 3 ] will be relevant but not [s x 1 … z x 3 ]. Instead, all the pairs, [s x 1 … s x 2 ], [s x 2 … z x 3 ] and [s x 1 … z x 3 ], will be evaluated if IDENT-XX is globally defined. For CORR constraints, global evaluation indicates transitivity and global correspondence, which was adopted in Bennett (2013, 2015) and Walker (2015). Local nontransitive evaluation of CORR was supported by Rhodes (2012) and Shih & Inkelas (2019). For instance, in the chain [s x 1 … s x,y 2 … s y 3 ], which follows the notation in Shih & Inkelas (2019), [s x 1 ] and [s y 3 ] do not stand in correspondence though both segments correspond to [s x,y 2 ]. The definitions of GSC constraints in (1) appear to mix and match local and global evaluation. Both IDENT-XX [𝛾 G] (F) and ACCORD-XX suggest local evaluation on adjacent pairs while CORR-XX indicates a transitive and unitary correspondence relation between segments (recall Chapter 2, section 2). In this section, I will show the reason why the limiter constraints, especially IDENT- XX [𝛾 G] (F), are evaluated locally. Then, the issues about CORR-XX will be discussed. Hansson (2007) argued that nonlocal evaluation of IDENT-XX would lead to problematic blocking effects. Therefore, subsequent studies mostly adopt the local evaluation of IDENT-XX 198 (Rhodes 2012, Walker 2015, Shih & Inkelas 2019), including the evaluation of IDENT-XX [𝛾 G] (F) in the current proposal. In fact, local evaluation of IDENT-XX [𝛾 G] (F) is also evidenced in the case study of Huozhou diminutive reduplication. In Chapter 3, I discussed how mid vowel raising is achieved IDENT-XX [𝛾 G] (F). An abridged tableau is repeated in (5) to exemplify. (5) Input /pʰə μ ŋ μ + μ/ ‘basin + DIM’; local evaluation of IDENT-XX [–lo] (high) pʰə 1 ŋ 2 + μ ID(hi) RT ID-XX [–lo] (hi) ID(hi) ☞ a. pʰ x ə x 1 ŋ x 2 .pʰ x u x 1 1 (ə x 1 ~ ŋ x 2 ) 1 b. pʰ x ə x 1 ŋ x 2 .pʰ x ə x 1 2W (ə x 1 ~ ŋ x 2 ), (ŋ x 2 ~ u x 1 ) L c. pʰ x u x 1 ŋ x 2 .pʰ x u x 1 1W L 2W In this example, I propose that mid vowel raising in the reduplicant is attributed to the influence of [ŋ] with a raised tongue dorsum. The constraint IDENT-XX [–lo] (high) favors (5a) over (5b) since two local pairs in (5b) violate IDENT-XX [–lo] (high), as marked in the tableau. However, if IDENT-XX [–lo] (high) evaluates globally, a wrong output will be selected. See the demonstration in (6). (6) Input /pʰə μ ŋ μ + μ/ ‘basin + DIM’; global evaluation of IDENT-XX [–lo] (high) pʰə 1 ŋ 2 + μ ID(hi) RT ID-XX [–lo] (hi) ID(hi) a. pʰ x ə x 1 ŋ x 2 .pʰ x u x 1 2 (ə x 1 ~ ŋ x 2 ), (ŋ x 2 ~ u x 1 ) 1 L b. pʰ x ə x 1 ŋ x 2 .pʰ x ə x 1 2 (ə x 1 ~ ŋ x 2 ), (ŋ x 2 ~ u x 1 ) c. pʰ x u x 1 ŋ x 2 .pʰ x u x 1 1 2 With globally defined IDENT-XX [–lo] (high), both pairs are subject to evaluation. In (6), both (6a) and (6b) incur two violations of IDENT-XX [–lo] (high). The candidate in (6b) harmonically bounds 199 (6a), which is not the expected output. In sum, the vowel changes during Huozhou diminutive reduplication offers additional support for local IDENT-XX constraints. Now I turn to the issue of global CORR-XX, which departs from the limiter constraints in GSC. Different from the traditional feature-specific CORR-XX[𝛼F], the only function of CORR-XX in the current system is to provide correspondence relations between segments. If feature-restricted IDENT-XX [𝛾 G] (F) is pursued in this model, a local feature-blind CORR-XX would run into problems. Suppose a local, nontransitive version of CORR-XX is defined in the following way. (7) CORR-XX: Assign a violation for every adjacent pair of segments that do not stand in Surface Correspondence. The key problem in such a definition is how to properly define “adjacent” (or “local”) if CORR- XX does not include any additional parameters. If the same example from (6) is used here, the candidate [pʰ x ə x,y ŋ y,z .pʰ z,a ə a,b ] would satisfy CORR-XX but IDENT-XX [–lo] (high) will not drive vowel raising in the reduplicant since [ŋ y,z ] and [ə a,b ] do not stand in correspondence. To generate the correct output, we expect a candidate such as [pʰ ə x ŋ x,y .pʰu y ]. However, the SCorr structure therein can be only established when additional parameters are encoded in CORR-XX (e.g., CORR- X μ X μ ). In other words, when a feature-blind CORR teams up with feature-specific IDENT-XX, there is only one way to define CORR-XX, i.e., the definition in (1a), in order to make the system work. Beyond this issue, Walker (2015:8-9) discussed two problems that would arise with nontransitive SCorr relations, which will not be repeated in detail here. To sum up, this Chapter revisits the proposal of GSC in this dissertation and addresses several theoretical issues based on the case studies. First, although feature-specific IDENT-XX [𝛾 G] (F) and its predecessor, *[𝛼F][–𝛼F] [𝛽 G, 𝛾 H] in ABP, have been well supported in the literature, the correspondence-invoking constraint, CORR-XX, is still needed in the system. The function of CORR-XX is minimal. Second, local evaluation of IDENT-XX [𝛾 G] (F) can be evidenced by the vowel 200 raising pattern in Huozhou diminutive reduplication. However, a global, transitive CORR-XX is the only viable definition when it is feature-blind. 201 Chapter 6 Typological survey and general discussion Based on the in-depth case studies in previous chapters, this chapter turns to a typological survey of truncatory backcopying and some discussion of general issues involving GSC. There are three goals in this chapter. First, it offers a typological survey of backcopying, with a focus on truncatory backcopying. Second, the predictions of Generalized Surface Correspondence will be discussed with regard to the observed patterns. In particular, it turns out that the conditional backcopying in Huozhou Chinese is not an isolated instance of the ranking schema that can be predicted by GSC. Beyond backcopying, a similar pattern outside of the reduplication domain can be analyzed with the same logic. Third, this chapter provides a preliminary discussion of extending GSC to other reduplication-related phenomena, especially copy epenthesis. Although copy epenthesis differs from morphological reduplication discussed in the case studies, it can be categorized as “phonological reduplication”, which is phonologically motivated to repair certain marked structures. In Chapter 1, I give a descriptive and theory-neutral definition of overapplication with a backcopying character, which is repeated in (1) below. (1) Overapplication with a backcopying character a. Overapplication refers to the situation where a phonological mapping introduces a disparity between the reduplicative output and the non-reduplicative lexical stem that is not expected merely on phonological grounds. 202 b. Overapplication with a backcopying character is a case where the disparity is located in the base part and the disparity in the base conforms to the reduplicant. As discussed earlier, some cases of backcopying involve the truncation or augmentation in the base part. In truncatory backcopying, segments or moras are deleted from the base to conform to the shape of the reduplicant. In reverse, when the base part is augmented so as to be identical to the reduplication, it is a case of augmentative backcopying. The case studies of Huozhou Chinese and Rapa Nui in previous chapters represent two kinds of truncatory backcopying in reduplication, segment truncation and mora deletion (vowel shortening). In both languages, I have shown the utility of Surface Correspondence in predicting these patterns, in comparison with the other approaches to reduplication. Huozhou Chinese, in particular, demonstrates a case of conditional backcopying, where the process is sensitive to segment features and only inputs with a [w] offglide undergo such a process. This pattern is not predicted in a BRCT model where BR correspondence does not refer to segment features. In what follows, I will start with a survey of truncatory backcopying in Section 6.1, followed by the evaluation of GSC with regard to the predictions in Section 6.2. In Section 6.3, the potential extension of GSC to copy epenthesis will be discussed. Finally, a brief survey of augmentative backcopying is presented in the appendix, while a detailed analysis and discussion of augmentative backcopying will be saved for future research. 6.1. Cases of truncatory backcopying This section presents a brief typological survey of truncatory backcopying. These patterns are uncommon cross-linguistically in general, but many of the following cases have contributed to the debate in the development of theories of reduplication, especially the controversy over the way templates and correspondence are understood. Some of the following data can be interpreted in various ways as arguments for different proposals, however, the surface patterns 203 still adhere to the theory-neutral definition of backcopying in (1) from a purely descriptive perspective. A set of truncatory backcopying cases involves vowel length alternation, where a long vowel in the base is shortened to conform to its counterpart in the reduplicant. Apart from Rapa Nui, a case study presented in Chapter 3, two similar cases are found in the literature, which are Tonkawa (Gouskova 2007) and Hawaiian (Alderete & McMillian 2015). Another set of cases, however, involves the deletion of segments in the base. To my knowledge, the only well- established example other than Huozhou Chinese in the previous literature is Guarijio (Caballero 2006). Although Downing (2000) also mentioned that there is a similar case in Hausa, it is not believed to be morphological reduplication according to Riggle (2006:880-881). Thus, I will not present the data of Hausa in further analysis and discussion. 6.1.1. Tonkawa Gouskova (2007) uses the data of Tonkawa (glottocode: tonk1249), an extinct Native American language, to argue for the utility of reduplicant-specific templates. In Tonkawa, when a word with an initial syllable that has a shape of CVV- (2a-c) or CVVC- (2d-f) is reduplicated, the long vowel in the syllable is unexpectedly shortened to match the CV template of the reduplicant. Tonkawa has a regular shortening process, where a long vowel following a word-initial CV syllable is prohibited in general, such as /ke-jaːloːna-oʔ/ → [(ké-ja)(lóː)(nóʔ)] ‘he kills me’. 49 This structure is repaired through vowel shortening. However, in the reduplicative context, especially in (2c) [he-ja-ja.tse.woʔ] (*[he-ja-jaː.tse.woʔ]) and (2f) [he-mʔe-mʔej.tsoʔ] (*[he-mʔe- mʔeːj.tsoʔ]), we cannot explain the shortening in the third syllable of each form (i.e. the first syllable of the base) as an application of regular shortening, because a long vowel is usually allowed in the third syllable, such as [(ké-ja)(lóː)(nóʔ)] ‘he kills me’. 49 Long vowels are notated as double vowels (e.g., [aa]) in Gouskova (2007) . Here I use the IPA length mark for long vowels (e.g., [aː]). 204 (2) Tonkawa reduplication Base Gloss Reduplicated Gloss a. naː.toʔs ‘I step on it’ na-na.toʔs ‘I step on it/REPETITIVE’ b. maː.koʔs ‘I weep’ ma-ma.ka.noʔs ‘I weep/REPETITIVE’ c. jaː.tsoʔs ‘I see him’ he-ja-ja.tse.woʔ ‘I see him/REPETITIVE’ (‘several look at it’) d. soːp.koʔ ‘he swells up’ so-sop.koʔs ‘I swell up/REPETITIVE’ e. tsoː.lʔoʔs ‘I defecate’ tso-tso.lʔoʔ ‘several defecate’ f. mʔeej.tsoʔ ‘he urinates’ he-mʔe-mʔej.tsoʔ ‘I urinate/REPETITIVE’ In sum, Tonkawa shows a clear example of truncatory backcopying, where the long vowel in the base is shortened in order to conform to the shape requirement of the reduplicant, which is similar to the case of Rapa Nui. 6.1.2. Hawaiian Both Hawaiian (glottocode: hawa1245) and Rapa Nui are East Polynesian languages. The examples below are cited from Alderete & McMillan (2015:8). The stress patterns and reduplication patterns in Hawaiian are highly similar to those of Rapa Nui since they are genetically related. The reduplicated form has a meaning of frequency or intensification, which also resembles the function of Rapa Nui reduplication. In (3), the initial long vowel in the base is shortened and matches its counterpart in the reduplicant. Vowel shortening is not an independently motivated process outside of the reduplicative context, supporting an interpretation of the shortening as driven by conformity with the reduplicant. (3) Hawaiian reduplication, CVːCV(C)V base (Alderete & McMillan 2015:8) Base Reduplicated Gloss of the reduplicated form ʔoːlapa ʔòla-ʔolápa ‘to flash, blaze suddenly (freq.)’ kuːʔai kùʔa-kuʔái ‘trade; for sale; to sell repeatedly’ liːhau lìha-liháu ‘gentle, cool rain (freq.)’ 205 6.1.3. Guarijio Caballero (2006) discussed and analyzed abbreviated reduplication in Guarijio (or “Huarijio”; Uto-Aztecan, glottocode: huar1255), which is an apparent case of truncatory backcopying. Some examples that are cited from Caballero (2006:278) are given in (4), which were originally reported in Miller (1996). In these examples, the reduplicated form indicates the inceptive aspect, and the base is truncated to the prosodic shape that matches the reduplicant, a CV syllable. Truncation is not an independent phonological process in this language. It is only found in the reduplicative context. (4) Guarijio abbreviated reduplication Verbal root Gloss Reduplicated Gloss toní ‘to boil’ to-tó ‘to start boiling’ sibá ‘to scratch’ si-sí ‘to start scratching’ čonó ‘to fry’ čo-čó ‘to start frying’ nogá ‘to move’ no-nó ‘to start moving’ kusú ‘to sing (animals)’ ku-kú ‘to start singing’ suhku ‘to scratch body’ su-sú ‘to start scratching the body’ muhíba ‘to throw’ mu-mú ‘to start throwing’ 6.1.4. Huozhou The cases introduced above all exhibit truncatory backcopying because the base has a mora or segments truncated. It is worth mentioning that the pattern of Huozhou diminutive reduplication (Chapter 3) is somewhat special when compared to the examples in this section. In Huozhou, the reduplicative template is proposed to be a single mora, which results in a reduplicant shape of C(G)V. Instead of imposing the mora template back to the base, the base remains heavy (bimoraic) while the [w] is deleted and the remaining vowel is lengthened to make the vacated mora well-formed. Some data are repeated in (5). 206 (5) Huozhou diminutive reduplication: w-ending nouns Noun Reduplicated Gloss pɑw 11 poː 11 .po 33 ‘bag, purse (DIM)’ tɕʰjɑw 35 tɕʰɥoː 35 .tɕʰɥo 55 ‘stick (DIM)’ tʰow 53 tʰuː 53 .tʰu 11 ‘bean (DIM)’ ljow 35 lyː 35 .ly 55 ‘glass ball (DIM)’ 6.1.5. Summary of patterns The cases reviewed above, as well as Rapa Nui and Huozhou Chinese, are summarized in (6). Though the known cases are not very abundant, the pattern is still attested in several languages that belong to different language families. (6) Cases of truncatory backcopying Language Reduplicant shape Consequence in the base Tonkawa monomoraic mora deletion Hawaiian bimoraic (two light syllables) mora deletion Rapa Nui bimoraic (two light syllables) mora deletion Guarijio monosyllabic segment deletion Huozhou monomoraic segment deletion In general, truncatory backcopying exhibits either shortening (mora deletion) or segment deletion. In addition, it is noteworthy that the patterns of truncatory backcopying in Guarijio and Huozhou are different in that the one in Huozhou is conditioned, namely, only w-ending nouns undergo segment deletion after reduplication while the other types of nouns, those with j/ŋ-ending, resist segment deletion. For some of the patterns above, the term “templatic backcopying” has been used in the literature, especially for the cases that resemble Guarijio in the strict sense. However, it is also loosely used to describe patterns like that in Tonkawa. The reasons why the term “templatic backcopying” is not used in the current study are as follows. First of all, I use “truncatory backcopying” to avoid ambiguity since it covers all the cases that involve either segment deletion or shortening (i.e., mora deletion) as reviewed in the sections above. What the term 207 “templatic backcopying” actually refers to does not seem to be consistent in the literature. To my knowledge, it is often used to indicate the scenario as in (4), where the base is truncated by segments, but Gouskova (2007) argues that Tonkawa vowel shortening is also a case of templatic backcopying (recall the data in §6.1.1). Although she acknowledged that the Tonkawa example is not a canonical case of templatic backcopying, it is viewed as a case of templatic backcopying in the loose sense since the reduplicative template of CV (a light syllable) and BR identity (IDENT- BR-μ) jointly result in vowel shortening when they dominate IDENT-IO-μ (Gouskova 2007:392). In a strict sense, however, only the case of Guarijio (Caballero 2013) complies with the situation presented by Kager and Hamilton. 50 Second, the use of “truncatory backcopying” as a theory- neutral term distances the current discussion from BRCT. The term “templatic backcopying” has been specifically associated with BRCT where there are templatic constraints and BR identity. In fact, the cases of truncatory backcopying presented earlier are not necessarily interpreted as the result of BR identity. The data of Huozhou Chinese, in particular, run into problems with a BRCT analysis (recall Chapter 3, §3.6.1.1). Thus, the use of “templatic backcopying” for the patterns above is not ideal. 6.2. Generalized Surface Correspondence and truncatory backcopying With the typological survey presented in §6.1, this section evaluates to what extent GSC can predict the observed patterns of truncatory and augmentative backcopying. It is worth mentioning that a prerequisite is to limit GSC to reduplicative context only. As has been discussed in Chapter 2 and the case studies, several strategies can be adopted. For the rest of this section, §6.2.1 and §6.2.2 examine what patterns GSC can predict, focusing on truncatory backcopying. As discussed in detail in previous chapters (Chapter 2, §2.3), the pressure of GSC constraints can lead to either mora deletion (shortening) or segment deletion, by ranking relevant GSC 50 Downing (2000) mentioned that Hausa ideophone reduplication is a case of Templatic backcopying. However, this pattern is not believed to be morphological reduplication according to Riggle (2006:880-881). 208 constraints above MAX-IO-seg (MAX-S) or MAX-IO-μ (MAX-μ). The predictions cover the two types of truncatory backcopying reviewed in §6.1. 6.2.1. Mora deletion To predict mora deletion in reduplication, the ranking is as the one in (7), where the main driving force of shortening is IDENT-VV(length), a shorthand for IDENT-XX [+vocalic] (length). The definition of ACCORD is recapitulated in (8). (7) Ranking schema of shortening ACCORD >> CORR-XX, IDENT-VV(length) >> MAX-μ (8) ACCORD-XX: Let X i and X j be two vocalic segments in the input, and X i ′ and X j ′ be two vocalic segments in the output; and let X i ′ ℛ X i and X j ′ ℛ X j . Assign a violation if X i ′ and X j ′ form a consecutive pair that stands in Surface Correspondence. Vowel shortening in Rapa Nui intensifying reduplication is analyzed with this ranking, which has been discussed in detail in Chapter 4. In (7), with the Surface Correspondence relation invoked by CORR-XX, the limiter constraint IDENT-VV(length) impels the corresponding vowels to match in length. The role of ACCORD is important here. This constraint is included to avoid unexpected correspondence relations that may accidentally overpredict vowel shortening. Since ACCORD penalizes any corresponding segments in the output that are not fissioned from the same input, the desirable output structure should only parse fissioned segments into the same correspondence class, such as [ke x 1 re y 2 -ke x 1 re y 2 tuː z 3 ] rather than [ke x 1 re x 2 -ke x 1 re x 2 tuː x 3 ], as illustrated in (9). Again, I will only mark vowels in the following tableau for conciseness. 209 (9) Rapa Nui intensifying reduplication: input /μμ + keretuː/ and /μμ + vaːnaŋa/ /μμ + ke 1 re 2 tuː 3 / ACCORD CORR-XX IDENT- VV(length) MAX-μ ☞ a. ke x 1 re y 2 -ke x 1 re y 2 tuː z 3 8 b. ke x 1 re x 2 -ke x 1 re x 2 tuː x 3 4W L 1W c. ke x 1 re x 2 -ke x 1 re x 2 tu x 3 4W L 1W /μμ + vaː 1 na 2 ŋa 3 / ACCORD CORR-XX IDENT- VV(length) MAX-μ ☞ d. va x 1 na y 2 -va x 1 na y 2 ŋa z 3 8 1 e. va x 1 na y 2 -vaː x 1 na y 2 ŋa z 3 8 1W L f. va x 1 na x 2 -va x 1 na x 2 ŋa x 3 4W L 1 The candidate in (9c) is emphasized. Since all the vowels in (9c) are in a single correspondence class, satisfying CORR-XX, the pair [re x 2 ~ tu x 3 ] is subject to the enforcement of IDENT-VV(length), which overpredicts mora deletion for this input. Due to the top-ranked ACCORD, this unexpected correspondence relation is ruled out. In comparison, when the input is /μμ + vaːnaŋa/, although the surface form in (9f) is accidentally identical to the winner (9d), length match is between the pair [na x 2 ~ va x 1 ] rather than [va x 1 ~ va x 1 ] as in (9d). In sum, the major role of ACCORD is to eliminate unexpected SCorr structure that may lead to the overprediction of mora deletion as exemplified in (9c). In addition, the top-ranked ACCORD has a side effect in the analysis of Rapa Nui vowel shortening, which is to limit the enforcement of IDENT-VV(length) to the reduplicative context only. See Chapter 4 for a detailed discussion on this point. The patterns of Tonkawa and Hawaiian in §6.1.1 and §6.1.2 can be potentially handled with the same logic. In both cases, a long vowel in the base is shortened to match the length requirement in the reduplicative template. The ranking in (7), where IDENT-VV(length) >> MAX- μ, is still crucial to these patterns, given a SCorr structure initiated by CORR and ACCORD. Both Tonkawa and Hawaiian have been handled with BRCT in the literature (Gouskova 2007, Alderete & McMillan 2015) where BR correspondence plays a key role. The detailed implementation of GSC in analyzing these data, as well as a comparison with BRCT, will not be presented in this 210 study. Nevertheless, the crucial role of correspondence in driving vowel shortening still holds in both cases. 6.2.2. Segment deletion When MAX-IO-seg (MAX-S) is dominated by relevant SCorr constraints, segment deletion will be predicted. According to the survey in §6.1, both Guarijio and Huozhou Chinese show truncatory backcopying that features segment deletion in the base. Segment deletion in these two cases, however, is not identical. For Guarijio, truncation is general and non-conditioned. All the segments that fall out of the reduplicative domain are deleted, no matter what the segments are (e.g., /toní/ → [to-tó] ‘to start boiling’; /muhíba/ → [mu-mú] ‘to start throwing’). For Huozhou Chinese, the pattern of truncatory backcopying is conditioned since only w-ending nouns undergo segment deletion after reduplication while the other types of nouns, those with a j/ŋ- ending, resist segment deletion. I will discuss how GSC constraints generate each type of truncatory backcopying that involves segment deletion for the rest of the section. Non-conditioned segment deletion can be achieved through the interaction between CORR- XX, ACCORD-XX, and MAX-S. To demonstrate the effect of constraint interaction, a hypothetical grammar with three constraints is illustrated in (10). (10) Truncatory backcopying with segment deletion: violation profile σ + papa CORR-XX ACCORD-XX MAX-S a. p x a i -p x a i 4 2 b. p x a x -p x a x 3 2 c. p x a x -p x a x p x a x 5 d. p x a i -p x a i p y a j 13 In the toy grammar, the input contains an empty syllable template that is realized as CV reduplication, which is the underlined portion in the candidates. The candidates in (10a) and (10b) exhibit truncatory backcopying; the ones in (10c) and (10d) are faithful in terms of MAX-S. 211 The potential rankings to select each of the candidates are as follows. The four scenarios below exhaust all the possible rankings and outputs of the three constraints. (11) Truncatory backcopying with segment deletion: rankings a. p x a i -p x a i ACCORD-XX >> CORR-XX >> MAX-S b. p x a x -p x a x CORR-XX >> ACCORD-XX >> MAX-S c. p x a x -p x a x p x a x CORR-XX, MAX-S >> ACCORD-XX d. p x a i -p x a i p y a j ACCORD-XX, MAX-S >> CORR-XX To predict truncatory backcopying, both CORR-XX and ACCORD-XX need to outrank MAX-S, but the relative ranking between CORR-XX and ACCORD-XX results in different SCorr structures. When ACCORD is top-ranked, as in (11a), it forces the segments that are not fissioned from the same input source to belong to separate correspondence classes. The deletion of the extra segments that are not in SCorr with any others leads to truncatory backcopying, which is a way to keep the number of correspondence classes to a minimum. When CORR-XX is higher-ranked, as in (11b), on top of the satisfaction of CORR-XX, segments are deleted in order to incur minimal violation of ACCORD-XX. It is noteworthy that the rankings in (11) are slightly different from the one that leads to mora deletion in (7). In (11), no IDENT-XX constraint such as IDENT-VV(length) is needed to motivate segment deletion. The effect of CORR and ACCORD alone can lead to truncation in the output form, without any reference to features. An issue here is which of the structures in (11a) and (11b) is appropriate for this type of truncatory backcopying, since both candidates have identical surface phonetic forms. The actual output requires a closer examination of the particular language in question. Huozhou Chinese instead exhibits a case of conditional truncatory backcopying. The process of segment deletion only applies in a specific phonological environment, namely, nouns that end with a labiovelar glide [w]. To achieve this, several IDENT-XX constraints set limitations in addition to CORR-XX and ACCORD-XX, as analyzed in detail in Chapter 3. The tableau that demonstrates the essential ranking is repeated in (12). The IDENT constraints are restricted to moraic members, indicated by [μ] (recall Chapter 3, §3.3.2). 212 (12) Conditional truncatory backcopying in Huozhou Chinese pɑ 1 w 2 + μ CORR- XX ID-XX [μ, +bk] (rd) ID-XX [+rd] (hi) MAX-S ID (hi) ID (lo) ☞ a. p x oː x 1 .p x o x 1 1 2 b. p x o x 1 w x 2 .p x o x 1 2W L 2 c. p x ɑ x 1 w x 2 .p x a x 1 1W L L d. p x ɑ x 1 w y 2 .p x a x 1 4W L L sa 1 j 2 + μ CORR- XX ID-XX [μ, +bk] (rd) ID-XX [+rd] (hi) MAX-S ID (hi) ID (lo) ☞ a. s x a x 1 j x 2 .s x a x 1 b. s x aː x 1 .s x a x 1 1W c. s x a x 1 j y 2 .s x a x 1 4W The crucial ranking in the tableaux is CORR-XX, ID-XX >> MAX-S. The constraints IDENT-XX [μ, +bk] (round) and IDENT-XX [+rd] (high) drive segment deletion in order to maximally satisfy CORR-XX. More importantly, the limiter constraints are restricted with features so that only the segment that is [+back] and [+round], namely [w], will be subject to the pressure. Thus, all the candidates for the input /sa 1 j 2 + μ/ escape violating these two constraints, and no deletion is motivated. The conditional truncatory backcopying in Huozhou diminutive reduplication is not the only case that manifests the joint effect of CORR and IDENT-XX in driving segment deletion. This ranking has been also used to handle other similar cases other than reduplication, which are characterized as “extreme end point of dissimilation” in Inkelas & Shih (2014). Inkelas & Shih argued for the role of Surface Correspondence in resolving nasal-consonant (NC) clusters. In Lithuanian, for instance, the NC cluster is resolved by either nasal place assimilation or deletion. When a nasal is immediately followed by a plosive, it undergoes place assimilation (13a), but the nasal is deleted with compensatory lengthening if it is followed by a continuant and/or sonorant consonant (13b). The data below are cited from Inkelas & Shih (2014:7), which are presented in Lithuanian orthography except for [ŋ]. The vowel [a̧] with a diacritic is long. 213 (13) Lithuanian nasal deletion a. sám-būris ‘assembly’ b. sá̧-skambis ‘harmony’ sán-dora ‘covenant’ sá̧-rašas ‘list, register’ sa[ŋ]-kaba ‘coupling, clamp’ sá̧-žinė ‘conscience’ sá̧-voka ‘idea’ To illustrate the effect of SCorr in driving nasal deletion, I give a brief analysis in (14), which is based on Inkelas & Shih (2014:7). I use IDENT-X::X [+cons] (cont, place), i.e., ID-C::C(cont, place), that requires feature matching between adjacent consonants (cf. CORR-C::C in Inkelas & Shih 2014, 2019 and Hansson 2001/2010, among others). (14) Lithuanian nasal deletion: sán + voka → sá̧-voka /sán + voka/ CORR-XX ID-C::C (cont, place) ID-IO (cont, nas) MAX-S ☞ a. s x á̧ x -v x o x k x a x 1 b. s x á x m x -b x o x k x a x 1W L c. s x á x w̃ x -v x o x k x a x 1W L d. s x á x m x -v x o x k x a x 1W L e. s x á x n x -v y o x k x a x 6W L The top-ranked CORR-CC calls for SCorr between consonants, ruling out (14e). The pressure of IDENT-C::C(cont, place) that requires adjacent consonants to be identical in [cont] and [place] motivates repair of the marked structure. Since MAX-S is ranked lower than IDENT-IO(cont, nas) (short for IDENT-IO(cont) and IDENT-IO(nas)), segment deletion is the best repair strategy in this analysis rather than assimilation, leaving (14a) as the winner. The analysis draws a parallel between Lithuanian nasal deletion and conditional truncatory backcopying in reduplication since both processes are motivated under the pressure of CORR and IDENT-CC/IDENT-XX. Thus, the analysis of Huozhou Chinese and the crucial ranking therein can be also extended to phenomena outside of reduplication. Although, in the literature on harmony systems, the primary role of Surface Correspondence is to facilitate long-distance assimilation by ranking CORR-XX and IDENT-XX(F) above input-output featural faithfulness IDENT-IO(F), 214 segment deletion is also a reasonable and attested prediction if MAX-IO interacts with the SCorr constraints. 6.3. GSC in reduplication-related phenomena As mentioned throughout the discussion, Surface Correspondence was first proposed to handle long-distance consonant assimilation and was since extended to vowel harmony, dissimilation, and long-distance interaction between tones. A further phenomenon that also shows the utility of Surface Correspondence is copy epenthesis. Copy epenthesis refers to a case where the quality of an epenthetic segment depends on a nearby segment. In some accounts, the epenthetic segment is viewed as a copy of the nearby segment. Different from morphological reduplication, copy epenthesis is usually phonologically motivated to repair a marked structure. For instance, a way to avoid an onset cluster in [pra] is to insert a copied vowel, producing [para]. This phenomenon has been categorized in some prior studies as “phonological reduplication” (Inkelas & Zoll 2005, Inkelas 2008) or “compensatory reduplication” (Yu 2005). There have been studies that argue for a correspondence approach to copy epenthesis, namely, the epenthetic segment stands in correspondence with another segment in the output string, and the identity between these two segments is enforced through surface identity constraints (Kitto & de Lacy 1999, Yu 2003, 2005, Inkelas 2008, Stanton & Zukoff 2016, 2018). The correspondence relations to handle copy epenthesis vary from one proposal to anothe. They include BE-correspondence (Kitto & de Lacy 1999), Surface Correspondence (Yu 2005, Inkelas 2008), Correspondence-by-transitivity (Stanton & Zukoff 2016), and HE-correspondence (Stanton & Zukoff 2018). Though the core mechanism is similar among these proposals, the exact implementation varies or remains an open question. Nevertheless, it is reasonable and promising to draw a parallel between copy epenthesis and reduplication. In this section, I provide some speculative discussion on the role of Generalized Surface Correspondence in handling copy epenthesis, with the aim of pushing forward the unification of copy epenthesis, reduplication, and other phenomena with long-distance identity effects. 215 6.3.1. Previous proposals of correspondence-approach to copy epenthesis B(ase)E(penthesis)-correspondence, proposed by Kitto & de Lacy (1999), specifically targets copy epenthesis. BE-correspondence is an abstract relation provided by GEN between an epenthetic segment and any other segment in the base. Non-correspondence is also an option. For a hypothetical input /pra/, the candidates that resolve the consonant cluster can be [pa x ra x ] and [pə x ra y ]. In [pə x ra y ], the second vowel represents a default epenthetic segment that does not stand in BE correspondence with other segments in the candidate. This candidate violates the constraint BE-CORR (Kitto & de Lacy 1999:12). Yu (2005) and Inkelas (2008) extended Surface Correspondence to copy epenthesis, following the proposal of Hansson (2001/2010) and Rose & Walker (2004), among others. The core insight is the same as the GSC model I pursue here, though the implementation is slightly different. In this approach, a SCorr relation is initiated between segments by CORR, and identity is enforced by IDENT-XX. Meanwhile, the ranking between DEP-S and INTEGRITY-S is able to distinguish copy epenthesis from default epenthesis. An illustration adapted from Inkelas (2008:368) is given in (15). In this case, it is a consonant that is copied and epenthesized to fill in the onset position. Note that both CORR-CC and IDENT-CC are evaluated locally in the tableau, following Inkelas (2008:368). (15) Copy epenthesis with Surface Correspondence /aɡat/ IDENT- IO DEP- S INTEG- S IDENT- CC CORR-CC ☞ a. ɡ x aɡ x at y 1 1 (ɡ x ~ t y ) b. t x aɡ y at x 1 2 (t x ~ ɡ y , ɡ y ~ t x ) c. ɡ x aɡ x at x 1 1W L d. t x aɡ x at x 1 2W L e. ʔ x aɡ y at z 1W L 3W f. t x at x at x 1W 1 L 216 In (15), default epenthesis (15e) is ruled out by DEP-S. The SCorr structure between consonants is initiated by CORR-CC. For (15c-d), the epenthetic consonant is not identical to its correspondents, violating IDENT-CC. Due to the local evaluation of CORR-CC, a local copy (15a) is favored over a non-local one (15b), making [ɡ x aɡ x at y ] the winner. For (15f), the copy- epenthesized segment and the other output segments undergo long-distance assimilation to satisfy IDENT-CC. This candidate is ruled out by the higher-ranked IDENT-IO. A recent study on copy epenthesis (focusing on vowel epenthesis) by Stanton & Zukoff (2018) argues convincingly for the correspondence approach through various length and stress match patterns in copy epenthesis. They propose that there is a mandatory correspondence relation, HE correspondence, between the host and the epenthetic vowel. The proposal is similar to Kitto & de Lacy’s (1999) BE correspondence except that HE-Corr does not assume a non- correspondence option. In other words, any epenthetic vowel, be it copied or default, needs to stand in correspondence with some segment in the output string. The previous proposals share similarities but are not consistent in several aspects regarding implementation, especially how the correspondence structure is initiated. A related issue is that there is no agreement on whether a default epenthetic segment should also stand in correspondence with another segment. In (16), I summarize the divergences on these issues. (16) Proposals for copy epenthesis BE Corr HE Corr SCorr (GSC) Correspondence evaluated by … BE-CORR – CORR-XX Correspondence with default epenthesis … not obligatory obligatory not obligatory Domain of application specific; epenthesis specific; epenthesis general; long-distance assimilation and reduplication 217 6.3.2. Extending GSC to copy epenthesis: a preliminary discussion The summary in (16) indicates that SCorr is a more general type of correspondence that can handle both long-distance assimilation and reduplication, as discussed in the current study. Thus, it is desirable to extend SCorr to the domain of copy epenthesis, which could lead to a more parsimonious grammar. However, Stanton & Zukoff (2018) argue that HE correspondence should be distinct from Surface Correspondence. In what follows, I will discuss the arguments in Stanton & Zukoff (2018) and show how the current GSC model used in the current study can potentially solve the issues raised therein. First, Stanton & Zukoff (2018) reject the hypothesis that HE correspondence is the same as Surface Correspondence by nature due to evidence from Selayarese. Some examples are given in (17) (Stanton & Zukoff 2018:642). (17) Selayarese copy epenthesis and stress misapplication a. [sóːᵐbala] d. [kíːkiri] b. [bóːtolo] e. [láːᵐbere] c. [béːrasa] f. [túːlisi] Selayarese has penultimate stress (e.g., [kasúːᵐba], ‘dye for coloring clothes or cake’), while the words that have copy epenthesis in (17) show antepenultimate stress (e.g., [túːlisi], ‘write’, *[tuːlísi]). This pattern is driven by HE-IDENT[stress] that requires the epenthetic vowel to be identical in stress with its host. Thus, [túːli x si x ] is the winner since the corresponding vowels (indicated by subscript x) are both unstressed, and therefore, it satisfies HE-IDENT[stress]. In comparison, the form *[tuːlí x si x ] with normal penultimate stress fatally violates HE-IDENT[stress]. Stanton & Zukoff (2018) pointed out a potential problem of implementing the analysis in the framework of Agreement-by-Correspondence (Hansson 2001/2010, Rose & Walker 2004, a.o.). In the ABC framework, surface-to-surface correspondence between segments is impelled by CORR constraints, which refers to the similarity of surface segments. However, the earlier mechanisms in ABC lack a way to distinguish epenthetic versus non-epenthetic vowels in the output, which 218 could lead to overprediction of the identity effect. Stanton & Zukoff (2018:646) provided two examples in Selayarese to illustrate this issue. (18) Stress difference in near-minimal pair Antepenultimate stress [jáːɡuru] copy-epenthetic ‘to box, punch’ Penultimate stress [kaʔmúːru] underlying ‘nose’ If SCorr is blind to the status of vowels, then CORR impels correspondence in both forms as [jáːɡu x ru x ] and [kaʔmúː x ru x ]. The second example will also violate IDENT-VV[stress]. In GSC, however, Surface Correspondence is favored by either featural similarity or segment fission. If we view copy epenthesis as segment fission, which is parallel to the mechanisms for reduplication, the constraint ACCORD can distinguish these two examples. When ACCORD is ranked higher, the SCorr structures for the examples in (18) will be different. For (18a), the optimal SCorr structure is [jáː x ɡu y ru y ], since both us are fissioned from the same segment in the input. For (18c), however, [ka x ʔmúː y ru z ] is favored since there are no fissioned segments in this form. The violation profile of the possible SCorr structures for each example is given in (19). For simplicity, I only mark correspondence relation between vowels. (19) Violation profile of ACCORD and CORR-VV jaːɡur ACCORD CORR-XX ☞ a. jáː x ɡu y ru y 2 b. jáː x ɡu y ru z 3W c. jáː x ɡu x ru x 1W L kaʔmuːru ACCORD CORR-XX ☞ d. ka x ʔmúː y ru z 3 e. ka x ʔmúː y ru y 1W 2L f. ka x ʔmúː x ru x 2W L The constraint ACCORD is proposed to handle a similar issue in reduplicative identity, as discussed in Chapter 2. In this way, if we also view copy epenthesis as segment fission, we can draw a 219 parallel between copy epenthesis and reduplication. The constraint ACCORD can be readily used to distinguish the underlying status of surface segments. Appendix: Augmentative backcopying This chapter and the case studies in earlier chapters focus on the issues of truncatory backcopying. However, another type of reduplicative opacity that involves backcopying is termed as augmentative backcopying, as discussed at the beginning of the current chapter. Augmentative backcopying mirrors truncatory backcopying, where the base is unexpectedly augmented and matches the reduplicant. In this appendix, the well-established cases of augmentative backcopying are surveyed. The cases in the literature that fall into this category include Ndebele, Kikuyu, Japanese, and Samala (or Ineseño Chumash). Some of these patterns have been analyzed in detail in different theoretical works, while others were only briefly mentioned. The cases that are classified as augmentative backcopying in the current study do not comply with the definition of “templatic backcopying” discussed in §6.1.5. These cases are simply referred to as backcopying in the literature, although they mirror the patterns of truncatory backcopying in many ways. For instance, both truncation and augmentation in the base are related to the requirement of prosodic well-formedness of the reduplicant. Ndebele Hyman, Inkelas & Sibanda (1999/2009) discussed a case of backcopying in Ndebele (Southern Bantu, glottocode: ngun1276). The reduplicant in Ndebele is required to be exactly bisyllabic. For a monosyllabic verb stem, it is augmented by attaching a meaningless syllable -yi to satisfy the requirement of minimality after reduplication. For instance, in the reduplicated form [dl-a-yi+dl-a] (the boundary between base and reduplicant is marked by ‘+’), the reduplicant has a -yi suffix to fulfill the bisyllabic requirement, while the base [dl-a] is not 220 subject to the minimality condition. Note that the whole verb [dl-a-yi+dl-a] is also required to be minimally bisyllabic, which apparently satisfies the condition. The pattern in Ndebele that exhibits augmentative backcopying is an alternation realization of reduplicated imperatives of subminimal verbs. Some data are given in (1) and (2) (Hyman, Inkelas & Sibanda 1999/2009:11-12). (1) Ndebele reduplicated imperatives Bare stem Imperative Reduplicated imperative Gloss -dl-a yi-dl-a yi-dl-a+yi-dl-a ‘eat!’ -m-a yi-m-a yi-m-a+yi-m-a ‘stand!’ -lw-a yi-lw-a yi-lw-a+yi-lw-a ‘fight!’ -z-a yi-z-a yi-z-a+yi-z-a ‘come!’ (2) Ndebele reduplicated imperatives: consonant root with the perfective suffix -ile a. (ba-) m-ile ‘(they) stood’ *(ba)-yi-m-ile b. (ba-) m-a-yi+m-ile single, suffixed -yi in reduplicant c. (ba-) yi-m-a+yi-m-ile double, prefixed yi- in reduplicant and base In (1), the imperative, as a verb, needs to satisfy the bisyllabic condition, and therefore, it is prefixed by yi-. The reduplicated form, however, has two yi- prefixes, which is interpreted as a case of backcopying. Since it is the reduplicant rather than the base that needs to meet the bisyllabic condition, the augmented prefix yi- is required by the reduplicant and overapplied to the base, instead of the other way around. We may question why the reduplicated form in (1) cannot be derived from the imperative form that has been yi-prefixed, which would not be backcopying. Further evidence of the backcopying interpretation is shown in (2). When the consonant root m- is suffixed with the perfective marker -ile, it satisfies the bisyllabic condition, and the yi- prefix is not necessary, as in (2a) *[(ba)-yi-m-ile]. Nevertheless, when reduplication takes place, there are two possible alternations, as in (2b) and (2c). For (2b), the consonant root is copied, then a vowel -a and a 221 suffix -yi are added to make the reduplicant bisyllabic. Note that the perfective suffix -ile is outside of the domain of reduplication. In this variant, the suffix -yi does not show up in the base. Another variant in (2c), however, augments the reduplicant with a prefix yi-, which also appears in the base. This alternation is consistent with the pattern in (1), where a prefixed yi- is copied back to the base. If the reduplicated pattern in (1) is derived from the imperative form, then it is difficult to explain why *[(ba)-yi-m-ile] in (2a) is ill-formed. Kikuyu Kikuyu (glottocode: kiku1240) is also a Southern Bantu language that is related to Ndebele. The augmentative backcopying in Kikuyu is similar to the pattern in Ndebele. The data are only briefly discussed in Downing (2000:20, 34). In (3), the verb stems are monosyllabic (note that [o] and [e] in [goa] and [hea] are glides, according to Downing (2000:34)). Since Kikuyu also has a bisyllabic condition on the reduplicant, there is lengthening in the reduplicant to make it meet the bisyllabic requirement. What is special about the pattern in (3) is that the base also undergoes lengthening to meet the condition. Nevertheless, base lengthening is not obligatory unless it is required by the identity effect, since it is already well-formed in the infinitive form. This pattern resembles base shortening in Tonkawa, but with lengthening, as pointed out in Gouskova (2007:392). (3) Kikuyu verbal reduplication Infinitive Reduplicated Gloss ko-goa ko-goo.a-goo.a *ko-goa-goa *ko-goo.a-goa ‘fall’ ko-hea ko-hee.a-hee.a *ko-hea-hea *ko-hee.a-hea ‘be burnt’ Japanese Renyookei reduplication in Japanese applies to verbs to describe an action that is performed simultaneously (Poser 1990:93, see also Itô 1990, Kurisu 2005, and Saba Kirchner 2010). The pattern has prosodic augmentation in the reduplicant, which is overapplied to the base. The data in (4) and (5) are cited from Kurisu (2005:179). Japanese renyookei is formed by full reduplication, as shown in (4a-d). When a verb stem is monomoraic, as in (4e-h), both the 222 reduplicant and the base are augmented to bimoraic. Since lengthening is prohibited in a non- derived monomoraic verb stem and only required in renyookei reduplication, as in (5), Kurisu (2005) interprets that lengthening is required by the reduplicant to satisfy a bimoraic minimality condition on this constituent. The base is augmented by a mora to match the shape of the reduplicant, though it does not meet the context of lengthening. (4) Japanese renyookei reduplication Verb stem (≥μμ) Reduplicated Gloss a. tabe tabe-tabe ‘while eating’ b. kaki kaki-kaki ‘while writing’ c. aruki aruki-aruki ‘while walking’ d. utumuki utumuki-utumuki ‘while looking down’ Verb stem (=μ) Reduplicated Gloss e. si sii-sii ‘while doing’ f. mi mii-mii ‘while seeing’ g. ne nee-nee ‘while sleeping’ h. de dee-dee ‘while getting out’ (5) Lengthening in renyookei reduplication only a. SeNtaku-o si/*sii, sigoto-o hazimeta washing-ACC do work-ACC started ‘After doing laundry, I started my job.’ b. SeNtaku-o sii-sii, sigoto-o hazimeta washing-ACC do work-ACC started ‘I started my job, doing laundry.’ Samala (Ineseño Chumash) Samala is formerly known as Ineseño Chumash, a language of the Chumashan family, which is native to California. 51 Samala is among the most often-discussed patterns in the reduplication literature. This language is a case study that McCarthy & Prince 51 For more information about Samala, see https://cla.berkeley.edu/languages/ineseno.php 223 (1995) use to argue for BRCT, where it is simply referred to as Chumash. The data in (6) are adapted from Inkelas & Zoll (2005:186, 190), which came from Applegate (1976). (6) Samala (Ineseño Chumash) reduplication with a prefix Plain Reduplicated Gloss a. s-ikuk sik-sikuk *s-ik-ikuk ‘he is chopping, hacking’ b. s-iš-expeč ši-šex-šexpeč *š-iš-ex-expeč ‘they two are singing’ c. k-ʔanɨɨš k’an-k’anɨɨš *k-ʔan-ʔanɨɨš (*k’an-ʔanɨɨš) ‘my paternal uncles’ d. s-ʔamɨn’ s’am-s’amɨn’ *s-ʔam-ʔamɨn’ (*s’am-ʔamɨn’) ‘he is naked’ e. s-iy-eqwel si-yeq-yeqwel *s-iy-eq-eqwel ‘they are doing it’/’they are making …’ The examples in (6) show the reduplicated forms of vowel-initial stems with a preceding prefix. In (6a-d), the stem is preceded by a personal prefix (s-, k-), while in (6e), the stem is immediately preceded by -iy-. In the reduplicated forms of (6), the final consonant of any preceding prefix will be resyllabified with the vowel-initial stem and copied back to the base. This pattern is referred to as “overcopy” in McCarthy & Prince (1995, 1999). 224 Chapter 7 Conclusion In this chapter, I summarize the major claims and findings of the dissertation (Section 7.1). Then, I will discuss the implications of the current model (Section 7.2) and future directions (Section 7.3). 7.1. Summary of the dissertation In this dissertation, I focused on two issues central to the theory of reduplication: (i) the underlying phonological form that initiates reduplication, and (ii) the grammatical mechanism that predicts truncatory backcopying and related patterns. With respect to these issues, I argue for the utility of prosodic node affixation in handling nonconcatenative morphology, including reduplication, subtraction, and lengthening, and the importance of Surface Correspondence in reduplicative opacity, especially truncatory backcopying. In this dissertation, I have advocated for a model of Minimal Reduplication with Generalized Surface Correspondence (MR + GSC). Two components, prosodic node affixation (as in MR) and Surface Correspondence (as in GSC), are independently-motivated and well-established mechanisms in grammar. The central goal was to show that there is no need to stipulate reduplication-specific machinery, which leads to a more parsimonious grammar. In Chapter 1, I introduced the desiderata of a model of reduplication with Generalized Surface Correspondence, which are repeated below: 225 (1) Desiderata (from Chapter 1) a. A model of reduplication that unifies reduplication and other types of nonconcatenative morphology; b. A model that accounts for reduplicative opacity, as seen in new and recently- reported data; c. An economical model that uses independently-motivated mechanisms. The rationale behind the desiderata is as follows. First of all, reduplication is a type of nonconcatenative morphology, together with subtractive morphology, morphological lengthening, morphological mutation, and so on. A model that can unify all these types of morphological processes is preferable to one that posits a specific mechanism for each of the processes (such as RED for reduplication and TRUNC for truncation). The proposal made in this dissertation builds on an existing line of research that seeks to pursue this goal (e.g., Bermúdez- Otero 2012), and the benefits of viewing nonconcatenative morphology as emergent and epiphenomenal have been well established in previous research (Davis and Ueda 2002, 2006, Saba Kirchner 2010, 2013, Bye and Svenonius 2012, Zimmermann 2013, 2015, 2017a, Trommer and Zimmermann 2014, Köhnlein 2018, a.o.). Nevertheless, the lack of RED eliminates an important mechanism in reduplication, BR correspondence, but a correspondence-based approach seems to still be important in analyzing reduplicative opacity (recall Chapter 1, §1.2.1). Therefore, an existing, independently-motivated type of correspondence, Surface Correspondence (e.g., Hansson 2001/2010, Rose & Walker 2004), is proposed to take over the role of BR correspondence as a more general theoretical mechanism in the domain of reduplication. In fact, Surface Correspondence is freely assigned between surface segments by GEN without any prerequisites, so there is no reason to block the effect of SCorr in a reduplicative context. The arguments for MR + GSC lie in the two major case studies, Huozhou Chinese diminutive formation and Rapa Nui reduplication, both of which exhibit truncatory backcopying. On the one hand, the patterns of truncatory backcopying in both cases show the utility and importance of a correspondence-based approach to reduplication, in particular, Surface Correspondence. 226 On the other hand, the alternation of nonconcatenative morphology in each language supports the versatility of prosodic node affixation. The main patterns analyzed in the languages and the arguments for each component of MR + GSC are summarized in (2). (2) Summary of patterns in the case studies Pattern Analysis with … Chapter 3: Huozhou Chinese Truncatory backcopying (conditional, segment deletion) SCorr Alternation between reduplication and subtraction mora affixation Chapter 4: Rapa Nui Truncatory backcopying in intensifying reduplication (mora deletion) SCorr Alternation between reduplication and lengthening in plural formation mora affixation The phenomena in each of the case studies involve two aspects, corresponding to the two components of the MR + GSC model. In Huozhou Chinese, when w-ending nouns undergo diminutive reduplication, it exhibits a pattern of conditional truncatory backcopying (e.g., [pɑw] ‘bag’ → [poː.po] ‘bag DIM’), which can be analyzed with Surface Correspondence. Meanwhile, the Huozhou diminutive morpheme can either be realized as reduplication or subtraction. Mora affixation provides a unified approach in the analysis of this alternation, because a single underlying mora is able to surface as either reduplication or subtraction. For Rapa Nui, I first discussed base vowel shortening in intensifying reduplication, which is also a case of truncatory backcopying. I demonstrated that this pattern is better analyzed with a correspondence-based approach in comparison to derivational approaches that resort to level ordering or serialism instead of correspondence. In addition, the plural formation in Rapa Nui surfaces as either lengthening or reduplication, suggesting that mora affixation provides a more economical solution to this alternation in comparison to BRCT, which requires a RED morpheme to initiate reduplication. Also, models of nonconcatenative morphology, such as realizational models (e.g., 227 Anderson 1992, Alderete 2001, Horwood 2001, Kurisu 2001), were evaluated, compared to the approach of prosodic node affixation in this study. In sum, the case studies in the dissertation center on the issue of what mechanism is involved in reduplicative opacity and the issue of what representation initiates reduplication. The MR + GSC model fulfills the three desiderata discussed in (1). This model brings reduplication and other nonconcatenative morphology under the same umbrella, accounts for the new and recently-reported data in this research, and makes use of independently-motivated mechanisms rather than stipulative ones. 7.2. Implications Since the model MR + GSC contains two components, I will discuss the implications and consequences of this model from two aspects. One major contribution of the current study is the proposal that Surface Correspondence plays a central role in handling reduplicative opacity. Although Yu (2005) and Inkelas (2008) mentioned that SCorr can be readily used to deal with “phonological reduplication”, there is no reason to assume that SCorr is blocked in morphological reduplication. The extension of SCorr to the domain of reduplication, together with its ability to analyze long-distance assimilation/dissimilation of segments and tones, as well as some local assimilation patterns, lends further support to the utility and versatility of this theoretical mechanism. As I have argued, SCorr can replace BR correspondence as a more general mechanism. It can also account for certain patterns where BR correspondence runs into problems, such as Huozhou conditional truncatory backcopying. The utility of SCorr in the domain of reduplication also implies that we may reduce the types of surface-to-surface correspondence in grammar. As briefly discussed in Chapter 2, there have been many different types of correspondence proposed for different domains, including Correspondence-by-Transitivity (Raimy & Idsardi 1997, Kawu 1999, Struijke 2002, Karabay 2004, Stanton & Zukoff 2015 etc., see also Bat-El 2002, 2005, Yu 2003), B(ase)E(penthesis)- 228 correspondence (Kitto & de Lacy 2000, see also Stanton & Zukoff 2018), and coupling in Aggressive Reduplication (Zuraw 2002). These types of correspondence relations are inherently related and formulated in a similar way. There are also some overlaps regarding the functions, especially for reduplication/copying-related phenomena. A summary table is given in (3) to compare the main proposal of each surface-to-surface correspondence relation. (3) Proposals of surface-to-surface correspondence Proposal Prerequisite Evaluating constraint Relation between … a. BR correspondence RED n.a. strings b. Correspondence-by-Transitivity segment fission n.a. segments c. BE correspondence n.a. BE-CORR segments d. Coupling n.a. REDUP strings e. Agreement-by-Correspondence/ Generalized Surface Correspondence n.a. CORR segments & subsegments 52 These proposals are all about some kind of abstract relation between surface structures, but differences are observed in three respects, namely, prerequisites, evaluating constraint, and domain. Establishing BR correspondence (3a) and Correspondence-by-Transitivity (3b) requires specific configurations, i.e., the presence of RED morpheme in the input for BR correspondence and segment fission for Correspondence-by-Transitivity. The consequence of the prerequisites is that these correspondence relations are not subject to any evaluation in CON. The other proposals, BE-Correspondence (Kitto & de Lacy 1999), coupling in aggressive reduplication (Zuraw 2002), and Agreement-by-Correspondence (Hansson 2001; Rose & Walker 2004, a.o.), do not require any specific configurations to invoke correspondence on the surface. In these proposals, GEN freely assigns correspondence to surface structures, and the candidates that do not contain any surface-to-surface correspondence relations are penalized by constraints such as BE-CORR, REDUP, and CORR. 52 Surface Correspondence can be established between subsegments as in Q Theory (Inkelas & Shih 2016, Shih & Inkelas 2019). 229 In addition, the surface-to-surface correspondence relations in (3) are established in different domains. BR correspondence and coupling operate between strings, although they can further entail correspondence at the segmental level. For instance, the form [p 1 a 2 ] RED [p 1 a 2 t 3 ] BASE has BR correspondence between RED and BASE, which yields a one-to-one correspondence between the segments (note that t 3 in the BASE has no correspondent in the RED). However, the other types of correspondence target segments or subsegments, and they do not necessarily bring about correspondence on the string level. All of the correspondence relations in (3) are provided by GEN, and the main functions are similar, i.e., to maintain or reinforce the identity between surface representations. Nevertheless, for each of the correspondence relations, the evaluating constraint and the predictions vary. BR correspondence and Correspondence-by-Transitivity mainly deal with reduplication, while coupling in Aggressive Reduplication provides a tool to analyze the identity effect in lexical reduplication (or “pseudo-reduplication”). BE correspondence is specifically proposed for copy epenthesis. Finally, Surface Correspondence (SCorr) was originally a mechanism proposed for long-distance agreement. Among all of these proposals, Surface Correspondence is the most versatile, with extensions of the application of SCorr to phenomena such as copy epenthesis (or complementary reduplication, Yu 2003, 2005; Inkelas 2008), disagreement (Bennett 2013, 2015), subsegmental phenomena such as prenasalization (Inkelas & Shih 2016, 2017), and tonal phenomena (Shih 2013; Shih & Inkelas 2019). As discussed in previous chapters, SCorr (GSC) can be used to handle reduplicative opacity, especially templatic backcopying. In comparison, proposals such as BR correspondence and BE correspondence are highly specialized and specific to a certain structure. The fact that SCorr is able to account for a wide spectrum of phenomena provokes the question of how many different types of surface-to-surface correspondence are needed in grammar. The case studies in this dissertation support an outlook in which Surface Correspondence serves as a general tool to handle surface identity in different domains, and it may potentially obviate the other types of correspondence altogether. 230 However, a single, unified SCorr to handle the phenomena in multiple domains is not free of potential problems. As has been discussed in Chapter 3 and Chapter 4, it is necessary to consider how to limit the effect of SCorr-driven identity enforcement to reduplicative contexts only. In the current analysis, I resort to level ordering and ACCORD respectively. Whether there exists a general way to limit SCorr to a specific domain awaits further investigation. Another aspect that I argue for in this dissertation is the benefits of prosodic node affixation. There have been efforts to reanalyze all types of nonconcatenative morphology as the affixation of nonlinear phonological elements, such as floating features, tones, and/or prosodic nodes (mora, syllable, foot, etc.). Some phenomena that have been reanalyzed as concatenation include morphological lengthening/gemination (e.g., Davis and Ueda 2002, 2006), reduplication (e.g., Saba Kirchner 2010, 2013, Bye and Svenonius 2012, Zimmermann 2013, 2015, 2017a), morphological mutation (e.g., Wolf 2007, Paschen 2018), and subtractive morphology (Trommer 2011, Bye & Svenonius 2012, Trommer and Zimmermann 2014, Zimmermann 2017b, Köhnlein 2018). The case studies in this dissertation, Huozhou diminutive formation and Rapa Nui reduplication, bring further support for this point of view. Following this line, nonconcatenative morphology emerges as an epiphenomenon driven by phonology in order to repair the deficient prosodic nodes in the input. This approach has two major implications. First, all types of morphological processes, be they concatenative or nonconcatenative, are unified in the sense that all morphological processes are concatenative by nature. For morphemes that have segmental content as the phonological exponent, they are simply put together during morphological construction, such as the English plural form [dɔɡ-z]. For morphemes that exhibit reduplication, subtraction, or other rule-like patterns, their underlying forms are viewed as nonlinear phonological elements (feature, tone, prosodic nodes, etc), such as Huozhou diminutive rime change [pɑw + μ] → [poː]. In this way, concatenation is the sole basic type of morphological operation. This idea has been advocated by Bermúdez-Otero (2012) and Bye & Svenonius (2012), among others. The case studies in this dissertation offer additional support for this point of view. 231 Additionally, the approach of prosodic node affixation, which is a morpheme-based model (recall Chapter 3), adds some restrictions on the underlying representation. That is to say, morphemes that exhibit nonconcatenative patterns on the surface are also specified with phonological forms underlyingly other than segments. The realizational models of nonconcatenative morphology that are visited in Chapter 3 often run into a too-many-solutions problem. When there is no restriction on representation, the actual phonological instantiations of nonconcatenativity depend solely on grammar, and all the other possible forms need to be eliminated by indexed faithfulness constraints. In contrast, although a morpheme-based model involves complication in representation, it comes to a compromise, or strikes a balance, between representation and grammar, resulting in a more economical grammar and restrictive predictions. 7.3. Future Directions This dissertation argues for the role of Surface Correspondence in reduplication with two in- depth case studies and some discussion of typological predictions. However, some issues await further investigation to further evaluate the proposal. 7.3.1. Dissimilation in reduplication The first issue is cases that show dissimilation in reduplication. It has been argued that SCorr can handle long-distance dissimilation (Bennett 2013, 2015; see also Bennett & DelBusso 2018). Since I have argued that SCorr can be extended to the domain of reduplication, a follow-up question is how to deal with cases where dissimilation interacts with reduplication. Although the original proposal of Bennett (2013, 2015) is based on a feature-specific version CORR-CC(𝛼F), the model pursued in this dissertation, a generic CORR-XX and feature-restricted IDENT-XX [𝛾 G] (F), can equivalently give rise to the same pattern (see the discussion in Bennett & DelBusso 2018). For a hypothetical sequence /d … z/ in the input, the output form [d … s] that dissimilates for [voice] can be analyzed as follows (adapted from Bennett & DelBusso 2018:24). 232 (4) Input /d … z/ (hypothetical) d … z CORR-XX ID-XX [+voice] (cont) ID-IO (cont) ID-IO (voice) ☞ a. d x … s x 1 b. d x … d x 1W L c. d x … z x 1W L d. d x … z y 1W L In this analysis, the winner [d x … s x ] (4a) undergoes voice dissimilation so as to vacuously satisfy the feature-restricted constraint IDENT-XX [+voice] (cont), compared to (4c). The ranking IDENT- IO(cont) >> IDENT-IO(voice) drives dissimilation for [voice] rather than [continuant]. The analysis of dissimilation with feature-restricted IDENT-XX runs into a potential problem. It will give a successful prediction of dissimilation only when the input segments differ in feature [F] but need to dissimilate for feature [G]. In this way, the ranking schema for dissimilation is CORR-XX, IDENT-XX [𝛾 G] (F) >> IDENT-IO(F) >> IDENT-XX(G), as in (4). However, this ranking would not work for an identical input sequence such as /d … d/, if dissimilation is also required to take place in this case. Even though we may expect that the SCorr approach could provide a simple and unified analysis when dissimilation interacts with reduplication, the issue introduced above poses a challenge to the potential solution. Yimas has a dissimilation process, which is not blocked when reduplication takes place, as exemplified in (5) (Foley 1991:54). Note that Foley (1991) did not specify which part is the base in the reduplicated form. (5) Yimas dissimilation and reduplication Non-reduplicated Reduplicated gloss iray- iratay- ‘cry’ wark- waratɨk- ‘make’ park- paratɨk- ‘cut up’ yara- yarata- ‘pick up’ 233 A dissimilation phenomenon in Yimas avoids word-internal [r]s in two adjacent syllables that are separated only by one vowel. In such a case, the second [r] is dissimilated to [t] (e.g., iratay-, *iraray). In this case, the input contains a single /r/, which fissions into two in the output, as illustrated in (6). A tableau is accompanied with the illustration in (7). (6) Yimas dissimilation and reduplication (7) Yimas dissimilation and reduplication; input: /iray-/ ir 1 ay- CORR-XX ID-IO(rhotic) L a. r x 1 … r x 2 b. r x 1 … t x 2 1 c. r x 1 … r y 2 1 1 This illustration suggests that the dissimilation process differs from the hypothetical example in (4). If the same logic in (4) is followed, the expected output [r x … t x ] (7b) will be harmonically bounded by *[r x … r x ] (7a). There is no IDENT-XX [𝛾 G] (F) to rule out (7a) and drive dissimilation in this case. This is a common issue in the interaction between dissimilation and reduplication since the output segments that stand in SCorr are always from the same source in the input. The cases that involve dissimilation and reduplication are not common. Some cases have been identified in Bennett’s (2013) work on dissimilation, such as Korean (Kim 1995), Mayali (Evans 1995), Chaha (Banksira 1997), Salish (Fallon 2002), and Western Bade (Schuh 2002). A typological survey and detailed examination of data need to be done in order to better understand the potential interaction between these two processes. Nevertheless, the issue identified in this section seems to be a general issue for the generic CORR-XX and feature- restricted IDENT-XX [𝛾 G] (F) when handling dissimilation. i r 1 a 2 y 3 - i r 1 a 2 t 1 a 2 y 3 - 234 7.3.2. Issues about typology and typological predictions The second issue relates to typology and typological predictions. Although I have argued that SCorr can obviate BR correspondence as a more general theoretical mechanism, it makes some of the same unwanted predictions as BR correspondence. One such prediction is the so-called “internal junctural effect” (e.g., McCarthy & Prince 1995, Inkelas & Zoll 2005, Inkelas 2014). If the base-reduplicant juncture creates a context of a phonological rule and causes alternation in the reduplicant, it can be copied back to the base. A hypothetical example of the internal junctural effect is as follows. Suppose an input /RED + panit/ creates a context of nasal place assimilation in the output, the output can be [pam-pamit], if BR identity is enforced through correspondence. This case can be also viewed as a case of backcopying, given the definition in Chapter 1 (though it is termed as “recopying” in Kiparsky 2010). A tableau is given in (4) to illustrate the process. (8) Input /RED + panit/ (hypothetical) RED + panit AGREE(Place) ID-BR(Place) ID-IO(Place) ☞ a. pam-pamit 1 b. pam-panit 1W L c. pan-panit 1W L This internal junctural backcopying is predicted when AGREE(Place), IDENT-BR(Place) >> IDENT-IO(Place). However, such a pattern is unattested, and it has been used as an argument against BRCT since it indicates that BRCT overpredicts (e.g., Kiparsky 2010, Inkelas 2014). However, if SCorr replaces BR correspondence as the identity-enforcing mechanism in grammar, it results in the same prediction. In fact, since the mechanisms of SCorr and BR correspondence are highly similar, SCorr is able to predict various kinds of problematic predictions that BR correspondence does. Nevertheless, a future direction is to investigate constraint permutation of SCorr versus BR correspondence in reduplication, which could potentially predict that one pattern is more 235 frequent than another. From a typological perspective, with the constraints and candidates in (8), we would expect that the three types of outputs in (8) are equally common in world languages, which is not the case. Even though I have discussed some attested patterns of reduplicative opacity, including truncatory backcopying, these cases are rare in general, and only a handful of cases are identified, as discussed in Chapter 6. If SCorr is adopted instead, more SCorr structures can be generated for the same input, which could potentially influence the typological predictions and the frequency of each output form. To sum up, although there could be some unexpected predictions when SCorr is introduced into reduplication, the case studies earlier have shown its utility and benefits. In comparison to BR correspondence, one major advantage of SCorr is that it offers a more general and versatile theoretical mechanism that draws similarity between reduplicative identity and other phenomena with surface identity effects. Thus, if SCorr can serve the same function as BR correspondence, a more general mechanism is to be preferred over a highly specialized one. 236 References Alderete, John D. 2001. Dominance effects as transderivational anti-faithfulness. Phonology 18(2). 201–253. Alderete, John & Kayleigh MacMillan. 2015. Reduplication in Hawaiian: variations on a theme of minimal word. Natural Language & Linguistic Theory 33(1). 1–45. Anderson, Stephen R. 1992. A-Morphous morphology. Cambridge: Cambridge University Press. Anttila, Arto. 1997. Deriving variation from grammar. In Frans L. Hinskens, Roeland van Hout, and Leo Wetzels (eds.), Variation, Change and Phonological Theory, 35–68. Amsterdam: John Benjamins. Anttila, Arto. 2002. Morphologically conditioned phonological alternations. Natural Language & Linguistic Theory 20(1). 1–42. Anttila, Arto. 2006. Variation and Opacity. Natural Language & Linguistic Theory 24(4). 893-944. Banksira, Degif Petros. 1997. The sound system of Chaha. Université du Quebec à Montreal dissertation. Bao, Zhiming. 1990. On the nature of tone. Cambridge, MA: Massachusetts Institute of Technology dissertation. Bao, Zhiming. 1996. The syllable in Chinese. Journal of Chinese Linguistics 24(2). 312–353. 237 Bat-El, Outi. 2005. Parsing Forms with Identical Consonants: Hebrew Reduplication. In Dorit Diskin Ravid & Hava Bat-Zeev Shyldkrot (eds.), Perspectives on Language and Language Development, 25–34. Springer. Bennett, William G. 2013. Dissimilation, consonant harmony, and surface correspondence. New Brunswick, NJ: Rutgers University dissertation. Bennett, William G. 2015a. The Phonology of Consonants: Harmony, Dissimilation and Correspondence. Cambridge University Press. Bennett, William G. 2015b. Assimilation, dissimilation, and surface correspondence in Sundanese. Natural Language & Linguistic Theory 33(2). 371–415. Bennett, William G. & Natalie DelBusso. 2018. The typological effects of ABC constraint definitions. Phonology 35(01). 1–37. Bermúdez-Otero, Ricardo. 2012. The architecture of grammar and the division of labour in exponence. In Jochen Trommer (ed.), The Morphology and Phonology of Exponence, 8–83. Oxford: Oxford University Press. Bermúdez-Otero, Ricardo. 2018. Stratal Phonology. In S.J. Hannahs & Anna R. K. Bosch (eds.), The Routledge handbook of phonological theory, 100-134. Abingdon: Routledge. Buckley, Eugene. 1997. Integrity and correspondence in Manam double reduplication. North East Linguistic Society (NELS) 28. 59–67. Buckley, Eugene. 2016. Foot Alignment in Spanish Secondary Stress. In Harry van der Hulst, Jeffrey Heinz & Rob Goedemans (eds.), Dimensions of Phonological Stress, 79–100. Cambridge: Cambridge University Press. Bye, Patrick & Peter Svenonius. 2012. Non-concatenative morphology as epiphenomenon. In by Jochen Trommer (ed.), The Morphology and Phonology of Exponence, 427-495. Oxford: Oxford University Press. Caballero, Gabriela. 2006. “Templatic backcopying” in Guarijio abbreviated reduplication. Morphology 16(2). 273–289. 238 Caballero, Gabriela & Sharon Inkelas. 2013. Word construction: tracing an optimal path through the lexicon. Morphology 23(2). 103–143. Carstairs, Andrew. 1988. Some implications of phonologically-conditioned suppletion. In Geert Booij and Jaap van Marle (eds). Yearbook of morphology 1988, 67-94. Dordreht: Foris. Carstairs, Andrew. 1990. Phonlogically conditioned suppletion. In Wolfgang U. Dressler, Hans C. Luschütsky, Oskar E. Pfeiffer and John R. Rennison (eds). Contemporary Morphology, 17-23. Berlin: Mouton de Gruyter. Casali, Roderic F. 1996. Resolving hiatus. Los Angeles, CA: University of California dissertation. Causley, Trisha. 1997. Identity and Featural Correspondence: the Athapaskan Case. North East Linguistic Society (NELS) 27. 93–105. Chapin, Paul G. 1978. Easter Island: A characteristic VSO language. In Winfred P. Lehmann (ed.), Syntactic typology: Studies in the phenomenology of language, 139–168. Austin: University of Texas Press. Cheng, Robert L. 1966. Mandarin Phonological Structure. Journal of Linguistics 2(2). 135–158. Downing, Laura J. 2000. Morphological and prosodic constraints on Kinande verbal reduplication. Phonology 17(1). 1–38. Downing, Laura J. 2006. Canonical Forms in Prosodic Morphology. New York: Oxford University Press. Duanmu, San. 1990. A formal study of syllable, tone, stress and domain in Chinese languages. Cambridge, MA: Massachusetts Institute of Technology dissertation. Duanmu, San. 1993. Rime length, stress, and association domains. Journal of East Asian Linguistics 2(1). 1–44. Duanmu, San. 1994. Syllabic weight and syllabic duration: A correlation between phonology and phonetics. Phonology 11(01). 1–24. Duanmu, San. 1999. Metrical structure and tone: evidence from Mandarin and Shanghai. Journal of East Asian Linguistics 8(1). 1–38. Duanmu, San. 2007. The phonology of standard Chinese (2 nd edn). New York: Oxford University Press. 239 Duanmu, San. 2008. Syllable Structure: The Limits of Variation: The Limits of Variation. New York: Oxford University Press. Davis, Stuart & Isao Ueda. 2002. Mora augmentation processes in Japanese. Journal of Japanese Linguistics 18, 1–23. Davis, Stuart & Isao Ueda. 2006. Prosodic vs. morphological mora augmentation. Lexicon Forum 2, 121–143. Elbert, Samuel H., & Mary Kawena Pukui. 1979. Hawaiian grammar. Honolulu: The University Press of Hawaii. Elenbaas, Nine & René Kager. 1999. Ternary Rhythm and the Lapse Constraint. Phonology. 16(3). 273–329. Evans, Nicholas. 1995. Current Issues in the Phonology of Australian Languages. In John Goldsmith (ed.), The Handbook of Phonological Theory, 723-761. Cambridge, MA: Blackwell. Fallon, Paul D. 2002. The synchronic and diachronic phonology of ejectives. New York: Routledge. Feng, Guanjun. 2006. Morpheme recognition in prosodic morphology. Los Angeles, CA: University of Southern California dissertation. Feng, Liangzhen & Xueling Zhao. 2014. The Study of Huozhou Dialect [!"#$%&'(]. Taiyuan: Beiyue Literature & Art Publishing House. Fuentes, Jordi. 1960. Diccionario y gramática de la lengua de la Isla de Pascua. Santiago de Chile: Editorial Andrés Bello. Gafos, Diamandis. 1996. The articulatory basis of locality in phonology. Baltimore, MD: Johns Hopkins University dissertation. Gafos, Diamandis. 1998. Eliminating long-distance consonantal spreading. Natural Language & Linguistic Theory 16. 223–278. Goldsmith, John. 1976. Autosegmental Phonology. Cambridge, MA: MIT dissertation. Gurevich, Naomi. 2000. Reduplication in Southern Paiute and Correspondence Theory. West Coast Conference on Formal Linguistics (WCCFL) 19, 167-177. Somerville: Cascadilla Press. 240 Halle, Morris & Alec Marantz. 1993. Distributed morphology and the pieces of inflection. In Kenneth Hale and Stephen Jay Keyser (eds.), The view from Building 20, 111-176. Cambridge, MA: MIT Press. Hansson, Gunnar Ólafur. 2001. Theoretical and typological issues in consonant harmony. Berkeley, CA: University of California, Berkeley dissertation. Hansson, Gunnar Ólafur. 2010. Consonant harmony: long-distance interaction in phonology (University of California Publications in Linguistics ; v. 145). Berkeley: University of California Press. Hansson, Gunnar Ólafur. 2014. (Dis)Agreement by (Non)Correspondence - Inspecting the Foundations. Talk presented at the ABC-Conference. University of California, Berkeley. Harris, Alice C. 2017. Multiple exponence. New York: Oxford University Press. Harris, James W. 1983. Syllable Structure and Stress in Spanish: a Nonlinear Analysis (Linguistic Inquiry Monograph 8), Cambridge, MA: MIT Press. Harris, James W. 1989. The Stress Erasure Convention and Cliticization in Spanish. Linguistic Inquiry 20(3). 339–363. Hayes, Bruce. 1989. Compensatory lengthening in moraic phonology. Linguistic inquiry 20(2). 253–306. He, Wei. 1982. Classification of Huojia finals [)*$%+,-./]. Dialect [!$%(]. 22-36. Hoijer, Harry. 1933. Tonkawa: an Indian language of Texas. In Franz Boas & Harry Hoijer (eds), Handbook of American Indian languages 3, 1–148. New York: J.J. Augustin. Horwood, Graham. 2001. Anti-faithfulness and subtractive morphology. [ROA 466]. Hou, Jingyi & Duanzheng Wen. 1993. Survey of dialects in Shanxi Province [!01$%23&'4 5(]. Taiyuan: United Press of Universities in Shanxi. Hyman, Larry M. 1984. On the weightlessness of syllable onsets. Annual Meeting of the Berkeley Linguistics Society 10. 1–14. Hyman, Larry M. 1985. A Theory of Phonological Weight. Dordrecht: Foris. 241 Hyman, Larry M., Sharon Inkelas, & Galen Sibanda 1999/2009. Morphosyntactic correspondence in Bantu reduplication. In Kristin Hanson and Sharon Inkelas (eds), The nature of the word: essays in honor of Paul Kiparsky, 273-309. Cambridge, MA: MIT Press. Inkelas, Sharon. 2008. The dual theory of reduplication. Linguistics 46(2) 351–401. Inkelas, Sharon. 2014. The Interplay of Morphology and Phonology. New York: Oxford University Press. Inkelas, Sharon & Cheryl Zoll. 2005. Reduplication: Doubling in Morphology. Cambridge: Cambridge University Press. Inkelas, Sharon, Orhan Orgun, & Cheryl Zoll. 1997. Exceptions and static phonological patterns: cophonologies vs. prespecification. In Derivations and constraints in phonology, ed by. Iggy Roca. 393–418. Oxford: Clarendon Press. Inkelas, Sharon & Stephanie S Shih. 2014. Unstable surface correspondence as the source of local conspiracies. North East Linguistic Society (NELS) 44, vol. 1, 191–204. Amherst: UMass GLSA. Inkelas, Sharon & Stephanie S. Shih. 2016. Re-representing phonology: consequences of Q Theory. North East Linguistic Society (NELS) 46, vol. 2, 161–174. Amherst, MA: UMass GLSA. Inkelas, Sharon & Stephanie S. Shih. 2017. Looking into Segments. Annual Meetings on Phonology (AMP). Itô, Junko. 1990. Prosodic Minimality in Japanese. Chicago Linguistic Society (CLS) 26: Papers from the Parasession on the Syllable in Phonetics and Phonology, 213–239. Itô, Junko & Armin Mester. 1998. Markedness and word structure: OCP effects in Japanese. Ms., University of California, Santa Cruz. [ROA255]. Johnston, Marla Speas. 1978. Some syntactic and semantic patterns in reduplicated words in Rapa Nui. University of Wyoming MA Thesis. Kager, Réne. 1999. Optimality Theory. Cambridge University Press. Karabay, Fetiye. 2004. Exploiting motivations of reduplication: the Turkish case. The Worksop on Altaic Formal Linguistics. MITWPL 46, 149-161. 242 Kawu, Ahmadu Ndanusa. 2000. Structural markedness and nonreduplicative copying. North East Linguistics Society 30. Kieviet, Paulus. 2017. A grammar of Rapa Nui. Language Science Press. Kiparsky, Paul. 1973. Phonological representations. In O. Fujimura et al. (eds.), Three dimensions of linguistic theory, 1-135. Tokyo: TEC. Kiparsky, Paul. 1982. From cyclic phonology to lexical phonology. In Harry van der Hulst & Norval S. H. Smith (eds), The structure of phonological representations Part I, 131–175. Dordrecht: Foris. Kiparsky, Paul. 2000. Opacity and cyclicity. The Linguistic Review 17(2). 351-366. Kiparsky, Paul. 2003. Finnish Noun Inflection. In D. Nelson & S. Manninen (eds.), Generative Approaches to Finnic Linguistics, 109-161. Stanford: CSLI Publications. Kiparsky, Paul. 2010. Reduplication in stratal OT. In Linda Uyechi and Lian Hee Wee (eds.), Reality exploration and discovery: pattern interaction in language & life, 125–142. Stanford: CSLI Publications. Kiparsky, Paul. 2015. Stratal OT: a synopsis and FAQs. In Yuchau E. Hsiao & Lian-Hee Wee (eds.), Capturing phonological shades within and across languages, 2–44. Newcastle upon Tyne: Cambridge Scholars Publishing. Kitto, Catherine & Paul de Lacy. 1999. Correspondence and epenthetic quality. In C.Kitto & C.Smallwood (eds.), Proceedings of AFLA VI. Holland: Holland Academic Graphics. Kirchner, Robert. 1996. Synchronic chain shifts in Optimality Theory. Linguistic Inquiry 27(2). 341–350. Köhnlein, Björn. 2018. A morpheme-based approach to subtractive pluralization in German dialects. Phonology 35(4). 617–647. Kurisu, Kazutaka. 2001. The phonology of morpheme realization. University of California, Santa Cruz dissertation. Kurisu, Kazutaka. 2005. Gradient Prosody in Japanese. Journal of East Asian Linguistics 14(3). 175– 226. 243 Lamont, Andrew. 2021. Serial Reduplication Is Empirically Adequate and Typologically Restrictive. Linguistic Inquiry 1–58. Lieber, Rochelle. 1980. On the organization of the lexicon. Cambridge, MA: MIT dissertation. Lieber, Rochelle. 1992. Deconstructing morphology: word formation in syntactic theory. Chicago: University of Chicago Press. Lin, Yen-Hwei. 1989. Autosegmental treatment of segmental processes in Chinese phonology. Austin, TX: University of Texas dissertation. Lin, Yen-Hwei. 1993. Degenerate affixes and templatic constraints: Rime change in Chinese. Language (69)4. 649–682. Lin, Yen-Hwei. 2001. Toward a unified account of three classes of Huojia affixed words. In Debao Xu (ed), Chinese phonology in generative grammar, 223–240. San Diego: Academic Press. Lin, Yen-Hwei. 2004. Chinese affixal phonology: Some analytical and theoretical issues. Language and Linguistics 5(4). 1019–1046. Lin, Yen-Hwei. 2008. A systemic approach to morphophonological alternations: Diminutive rime change in Huojia and Jiyuan. The 11th International Symposium of Chinese Languages and Linguistics (IsCLL-11). National Chiao Tung University, Hsinchu, Taiwan. Lin, Yen-Hwei. 2010. Unexpected Morphophonological Outputs. NACCL-22 & IACL-18: Proceedings of the 22nd North American Conference on Chinese Linguistics and the 18th International Conference on Chinese Linguistics, vol. 1, 363–382. Lombardi, Linda. 1999/2001. Why Place and Voice Are Different: Constraint-Specific Alternations in Optimality Theory. In Linda Lombardi (ed.), Segmental Phonology in Optimality Theory: Constraints and Representations, 13–45. Cambridge: Cambridge University Press. Lubowicz, Anna. 2002. Derived environment effects in Optimality Theory. Lingua 112: 243–280. Lyu, Zhenjia. 1991. Yuncheng Dialect [!67$%8(]. Taiyuan: United Press of Universities in Shanxi. Mascaró, Joan. 2007. External allomorphy and lexical representation. Linguistic Inquiry 38(4). 715–735. 244 Métraux, Alfred. 1971. Ethnology of Easter Island (Bernice P. Bishop Museum Bulletin 160). Honolulu: Bernice P. Bishop Museum. McCarthy, John. 1981. A prosodic theory of nonconcatenative morphology. Linguistic Inquiry 373– 418. McCarthy, John J. 2000. Faithfulness and prosodic circumscription. In Joost Dekkers, Frank van der Leeuw, and Jeroen van de Weijer (eds), Optimality Theory: Phonology, Syntax, and Acquisition, 151–189. Oxford: Oxford University Press, McCarthy, John J. 2002. Research surveys in linguistics: A thematic guide to Optimality Theory. Cambridge: Cambridge University Press. McCarthy, John J. 2007. Restraint of analysis. In Silvia Blaho, Patrick Bye and Martin Krämer (eds.), Studies in generative grammar Vol. 95: Freedom of analysis, 203–231. Berlin: Mouton de Gruyter. McCarthy, John J. 2008. Doing optimality theory: applying theory to data. Malden, MA: Blackwell. McCarthy, John J. 2010. Agreement by correspondence without CORR constraints. Ms. University of Massachusetts, Amherst. McCarthy, John J. & Alan Prince. 1986/1996. Prosodic Morphology 1986. Technical Report. Rutgers University Center for Cognitive Science, New Brunswick, NJ. McCarthy, John J. & Alan Prince. 1993. Prosodic Morphology: Constraint Interaction and Satisfaction. Technical Report. Rutgers University Center for Cognitive Science, New Brunswick, NJ. McCarthy, John J. & Alan S. Prince. 1994a. The Emergence of the Unmarked: Optimality in Prosodic Morphology. In Proceedings of the North East linguistic society, ed by. Mercè Gonzàlez. 333–379. Amherst: GLSA Publications. McCarthy, John J. & Alan Prince. 1994b. Prosodic Morphology: An Overview, talks presented at the OTS/HIL Workshop on Prosodic Morphology, University of Utrecht. McCarthy, John J. & Alan Prince. 1995. Faithfulness and reduplicative identity. In University of Massachusetts Occasional Papers in Linguistics 18: Papers in Optimality Theory, ed. by Jill N. Beckman, Laura Walsh Dickey and Suzanne Urbanczyk, 249-384. Amherst, MA: GLSA. 245 McCarthy, John & Alan Prince. 1999. Faithfulness and identity in prosodic morphology. The Prosody-Morphology Interface 218–309. McCarthy, John J., Wendell Kimper & Kevin Mullin. 2012. Reduplication in Harmonic Serialism. Morphology 22(2). 173–232. McLaughlin, Fiona. 2000. Consonant mutation and reduplication in Seereer-Siin. Phonology 17(3). 333–363. Miller, Wick R. 1996. Guarijío: Gramática, textos y vocabularios. Mexico: UNAM, Instituto de Investigaciones Antropológicas. Myers, Scott. 1987. Vowel shortening in English. Natural Language and Linguistic Theory 5. 485–518. Nespor, Marina & Irene Vogel. 1986. Prosodic Phonology. Dordrecht: Foris. Nelson, Nicole. 2005. Wrong side reduplication is epiphenomenal: Evidence from Yoruba. In Bernhard Hurch and Vronika Mattes (eds.), Studies on Reduplication, 135–160. Berlin: Walter De Gruyter. Onn, Farid M. 1976. Aspects of Malay phonology and morphology: A generative approach. University of Illinois, Urbana-Champaign dissertation. Orgun, Cemil Orhan. 1996. Sign-Based Morphology and Phonology with special attention to Optimality Theory. Berkeley, CA: University of California dissertation. Padgett, Jaye. 2011. Consonant-Vowel Place Feature Interactions. In Marc van Oostendorp, Colin J Ewen, Elizabeth Hume, and Keren Rice (eds), The Blackwell Companion to Phonology Vol III, 1761-1786. Malden, MA: Wiley-Blackwell. Paschen, Ludger. 2018. The interaction of reduplication and segmental mutation: A phonological account. University of Leipzig dissertation. Paschen, Ludger. 2021. Trigger poverty and reduplicative identity in Lakota. Natural Language & Linguistic Theory. https://doi.org/10.1007/s11049-021-09525-y Paster, Mary. 2010. The verbal morphology and phonology of Asante Twi. Studies in African Linguistics, 39, 77–120. 246 Prince, Alan & Paul Smolensky. 1993/2004. Optimality Theory: Constraint Interaction in Generative Grammar. Malden, MA: Blackwell. Pruitt, Kathryn. 2010. Serialism and locality in constraint-based metrical parsing. Phonology 27(3). 481–526. Raimy, Eric & William Idsardi. 1997. A minimalist approach to reduplication in Optimality Theory. North East Linguistics Society 27. Raimy, Eric. 2000. The phonology and morphology of reduplication. Berlin, New York: de Gruyter. Rhodes, Russell. 2012. Vowel harmony as agreement by correspondence. In UC Berkeley Phonology Lab Annual Report. Berkeley: University of California. Riad, Tomas. 1992. Structures in Germanic prosody: a diachronic study with special reference to Nordic languages. Stockholm University dissertation. Rose, Sharon & Rachel Walker. 2004. A typology of consonant agreement as correspondence. Language 80(3). 475–531. Rosenthall, Samuel. 1994. Vowel/Glide alternation in a theory of constraint interaction. Amerst, MA: University of Massachusetts dissertation. Saba Kirchner, Jesse. 2010. Minimal reduplication. Santa Cruz, CA: University of California dissertation. Saba Kirchner, Jesse. 2013. Minimal reduplication and reduplicative exponence. Morphology 23(2):227–243. Sasa, Tomomasa. 2009. Treatment of vowel harmony in Optimality Theory. University of Iowa dissertation. Sande, Hannah. 2019. A unified account of conditioned phonological alternations: Evidence from Guébie. Language 95(3). 456–497. Satoh, Akira & Liangzhen Feng. 1989. Literal and colloquial readings in the Huozhou dialect of Shanxi Province [01"9$%:;<=>]. Chuugoku Gogaku 236. 24-34. Schuh, Russell G. 2002. Voicing dissimilation in Western Bade. Unpublished manuscript, UCLA. 247 Selkirk, Elisabeth. 1982. The syntax of words. Cambridge, MA: MIT Press. Selkirk, Elisabeth. 1995. The prosodic structure of function words. In Jill N. Beckman, Laura Walsh Dickey and Suzanne Urbanczyk (eds.), University of Massachusetts Occasional Papers in Linguistics 18: Papers in Optimality Theory, 439–470. Amherst, MA: GLSA. Senturia, Martha. 1998. A prosodic theory of hiatus resolution. San Diego, CA: University of California dissertation. Shen, Ming. 2003. Diminutives in Shanxi dialects [01$%-?@]. Fangyan 4. 335-351. Shen, Ming, Liangzhen Feng, and Hiroomi Tsumura. 2010. The decline of vowel harmony in Huozhou dialect, Shanxi Province: the perspective of GIS [A GIS.B01"#$%CDE FG-HIJK]. Journal of Chinese Linguistics 14. 137-157. Shih, Stephanie S. 2013. Consonant-tone interaction as agreement by correspondence. Ms. University of California, Berkeley. http://cogsci.ucmerced.edu/shih/shih-ctoneABC-draftms_1- 18-13.pdf Shih, Stephanie S & Sharon Inkelas. 2019. Autosegmental Aims in Surface-Optimizing Phonology. Linguistic Inquiry 50(1). 137–196. Stanton, Juliet and Sam Zukoff. 2016. Prosodic effects of segmental correspondence. Chicago Linguistic Society (CLS) 51, 501–515. Stanton, Juliet & Sam Zukoff. 2019. Prosodic identity in copy epenthesis. Natural Language & Linguistic Theory 36(2). 637–684. Smolensky, Paul. 1993. Harmony, markedness, and phonological activity. Paper presented at the First Rutgers Optimality Workshop (ROW 1), Rutgers University, New Brunswick, New Jersey. Smolensky, Paul. 2006. Optimality in phonology II: Harmonic completeness, local constraint conjunction, and feature domain markedness. In Paul Smolensky and Géraldine Legendre (eds), The harmonic mind: From neural computation to optimality-theoretic grammar Vol II, 27–160. Cambridge, MA: MIT Press. Struijke, Caro. 2002. Existential Faithfullness: A Study of Reduplicative TETU, Feature Movement and Dissimulation. Routledge. 248 Sui, Yanyan. 2013. Phonological and phonetic evidence for trochaic metrical structure in Standard Chinese. Philadelphia, PA: University of Pennsylvania dissertation. Suzuki, Keiichiro. 1999. Identity avoidance vs. identity preference: The case of Sundanese. Paper presented at the Linguistic Society of America, Los Angeles, January 1999. Tian, Juan. 2009. Phonology of Shanxi Huozhou Dialect [《01"#$%LD&'(]. Xi’an: Shaanxi Normal University MA thesis. Tian, Xicheng. 1986. Zi-changed finals in the Heshun dialect of Shanxi Province [01EM$% -NI+,]. Zhongguo Yuwen. 371-373 Tian, Xicheng. 1992. Diminutive rime change of Huozhou Dialect ["#$%-?@I+]. Journal of Shanxi University [!01OPP4(] 1. 1-8. Trommer, Jochen. 2011. Phonological aspects of Western Nilotic mutation morphology. Habilitationsschrift. University of Leipzig. Trommer, Jochen & Eva Zimmermann. 2014. Generalized mora affixation and quantity- manipulating morphology. Phonology 31(3). 463–510. Urbanczyk, Suzanne. 1996. Patterns of Reduplication in Lushootseed, Doctoral Dissertation, University of Massachusetts at Amherst. van de Weijer, Jeroen and Jisheng Zhang. 2008. An X-bar approach to the syllable structure of Mandarin. Lingua 118(9). 1416–1428. van Oostendorp, Marc. 2005. The theory of faithfulness. Ms. Meertens Institute, Amsterdam. van Oostendorp, Marc. 2006. A theory of morphosyntactic colours. Ms. Meertens Institute, Amsterdam. Walker, Rachel. 2000a. Yaka nasal harmony: Spreading or segmental correspondence? Berkeley Linguistics Society 26: 321–332. Walker, Rachel. 2000b. Long-distance Consonantal Identity Effects. West Coast Conference on Formal Linguistics 19: 532–545. 249 Walker, Rachel. 2001. Consonantal correspondence. In Robert Kirchner, Joe Pater & Wolf Wikeley (eds.), Proceedings of the workshop on the lexicon in phonetics and phonology: Papers in experimental and theoretical linguistics 6, 73–84. Edmonton: Department of Linguistics, University of Alberta. Walker, Rachel. 2005. Weak Triggers in Vowel Harmony. Natural Language & Linguistic Theory 23(4). 917–989. Walker, Rachel. 2011. Vowel Patterns in Language. Cambridge: Cambridge University Press. Walker, Rachel. 2014. Prominence-control and multiple triggers in vowel harmony: An ABC analysis. In Stephanie S Shih and Sharon Inkelas (eds), ABC↔Conference Archive (UC Berkeley Phonology Lab Annual Report), 202–213. University of California, Berkeley. Walker, Rachel. 2015. Surface Correspondence and Discrete Harmony Triggers. Annual Meetings on Phonology 2014 (AMP). Walker, Rachel. 2018. Feature identity and icy targets in Menominee vowel harmony. In Ryan Bennett, Andrew Angeles, Adrian Brasoveanu, Dhyana Buckley, Nick Kalivoda, Shigeto Kawahara, Grant McGuire & Jaye Padgett (eds.), Hana-bana: A Festschrift for Junko Ito and Armin Mester. Walker, Rachel & Bella Feng. 2004. A ternary model of morphology-phonology correspondence. West Coast Conference on Formal Linguistics (WCCFL) vol. 23, 773–786. Citeseer. Weber, Robert & Nancy Weber. 1985. Hacia el establecimiento de un sistema escrito parael rapa nui, lengua de la Isla de Pascua. Ms. Valparaíso, Chile: Universidad Católica de Valparaíso. Weber, Nancy & Robert Weber. 2005. Te ti ꞌaraꞌa tano mo pāpa ꞌi o te reꞌo rapa nui (Las mejores letras para escribir el idioma rapa nui). Isla de Pascua: ꞌŪmaŋa Hatu Re ꞌo = Academia de la Lengua Rapa Nui. Wei, Wei & Rachel Walker. 2020. A lookahead effect in Mbe reduplication: Implications for Harmonic Serialism. Linguistic Inquiry 51:845–859. Wilbur, Ronnie Bring. 1973. The Phonology of Reduplication. University of Illinois at Urbana- Champaign dissertation. 250 Wolf, Matthew. 2007. For an autosegmental theory of mutation. [ROA754]. Yang, Yifan. 2022. Rapa Nui: a case for correspondence in reduplication. Linguistic Inquiry. https://doi.org/10.1162/ling_a_00444 Yeh, Shih-chi. 2008. Suffixal Reduplication in Paiwan. In M. Bane, J. Holle, T. Grano, A. Grotberg, and Y.McNabb (eds.) Proceedings of CLS 44 (The Main Session). Chicago, IL: Chicago Linguistic Society, 225–38. Yu, Alan C L. 2003. The morphology and phonology of infixation. Berkeley, CA: University of California dissertation. Yu, Alan C L. 2005. Toward a Typology of Compensatory Reduplication. West Coast Conference on Formal Linguistics (WCCFL), 397-405. Somerville, MA: Cascadilla Proceedings Project. Zhang, Jisheng. forthcoming. On the Chinese vowel phonemes [QRLCDDS]. Contemporary Linguistics. Zhang, Jie. 2000. Non-contrastive features and categorical patterning in Chinese diminutive suffixation: MAX[F] or IDENT[F]? Phonology 17(03). 427–478. Zimmermann, Eva. 2013. Non-concatenative allomorphy is generalized prosodic affixation: The case of Upriver Halkomelem. Lingua 134. 1–26. Zimmermann, Eva. 2015. The power of a single representation: morphological tone and allomorphy. Morphology 26(3–4). 269–294. Zimmermann, Eva. 2017a. Morphological Length and Prosodically Defective Morphemes. Oxford: Oxford University Press. Zimmermann, Eva. 2017b. Multiple reduplication as non-segmental affixation. North East Linguistic Society (NELS) 47, 285–294. Amherst, MA: GLSA Zukoff, Sam. 2017. Actually, Serial Template Satisfaction does predict medial coda skipping in reduplication. Annual Meetings on Phonology (AMP). Zuraw, Kie. 2002. Aggressive reduplication. Phonology 19(03). 395–439.

Asset Metadata

Creator Yang, Yifan (author)

Core Title Generalized surface correspondence in reduplicative opacity

Contributor Electronically uploaded by the author (provenance)

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Linguistics

Degree Conferral Date 2022-08

Publication Date 07/22/2022

Defense Date 05/20/2022

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag backcopying,correspondence,morphology-phonology interface,OAI-PMH Harvest,reduplication

Format application/pdf (imt)

Language English

Advisor Shih, Stephanie (committee chair), Walker, Rachel (committee chair), Iskarous, Khalil (committee member), Mintz, Toben (committee member)

Creator Email yangyifa@usc.edu,yfn.yang@outlook.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC111373996

Unique identifier UC111373996

Legacy Identifier etd-YangYifan-10926

Document Type Dissertation

Format application/pdf (imt)

Rights Yang, Yifan

Type texts

Source 20220722-usctheses-batch-961 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email uscdl@usc.edu

Abstract (if available)

Abstract This dissertation focuses on the role of Surface Correspondence in handling backcopying, a type of reduplicative opacity. Backcopying is a phenomenon where the base undergoes unexpected phonological change in order to conform to the features or prosodic shape of the reduplicant. The analysis and interpretation of backcopying have been at the heart of the debate that whether or not correspondence is needed in handling reduplication-phonology interactions. This study argues for a correspondence-based approach to reduplication, drawing evidence from the truncatory backcopying patterns in new or recently-reported data from Huozhou Chinese and Rapa Nui. The ultimate goal of this study is to argue for a more economical grammar by demonstrating that the complicated patterns in reduplication can be readily handled with more general rather than specialized theoretical mechanisms.
Motivated by the language data, the proposal of this study has two components, a model of Minimal Reduplication with Generalized Surface Correspondence (MR + GSC). These two components correspond to two issues central to the theories of reduplication: (i) the underlying phonological form that initiates reduplication, and (ii) the grammatical mechanism that predicts truncatory backcopying. The case studies of Huozhou Chinese diminutive formation and Rapa Nui reduplication aim to provide evidence for the utility of prosodic node affixation in handling nonconcatenative morphology and the importance of Surface Correspondence in reduplicative opacity, especially truncatory backcopying.
The current study aims to make three major contributions to both language data and theory. First, this study argues for a correspondence-based approach with new and recently-reported language data, contributing to the debate on whether or not syntagmatic correspondence is needed in a theory of reduplication. Second, this study argues for the role of Surface Correspondence (SCorr) in reduplication and extends its utility to the domain of morphological reduplication, which has not been explored in detail in previous literature. Third, the data presented in this research also involve nonconcatenative allomorphy, which demonstrates the benefit of prosodic node affixation in nonconcatenative morphology.