Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Segmental and prosodic complexity in Nivaĉle : laryngeals, laterals, and metathesis Gutiérrez, Analía 2015

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2015_september_gutierrez_analia.pdf [ 21.04MB ]
Metadata
JSON: 24-1.0166445.json
JSON-LD: 24-1.0166445-ld.json
RDF/XML (Pretty): 24-1.0166445-rdf.xml
RDF/JSON: 24-1.0166445-rdf.json
Turtle: 24-1.0166445-turtle.txt
N-Triples: 24-1.0166445-rdf-ntriples.txt
Original Record: 24-1.0166445-source.json
Full Text
24-1.0166445-fulltext.txt
Citation
24-1.0166445.ris

Full Text

SEGMENTAL AND PROSODIC COMPLEXITY IN NIVAĈLE: LARYNGEALS, LATERALS, AND METATHESIS   by  Analía Gutiérrez    B.A., Universidad de Buenos Aires, 2002 B.A., Universidad de Buenos Aires, 2003 M.A., University of South Carolina, 2008      A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF     DOCTOR OF PHILOSOPHY  in  THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (LINGUISTICS)   THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) July 2015    © Analía Gutiérrez, 2015    ii Abstract    This dissertation investigates a series of phonological and phonetic aspects of Nivaĉle, a Mataguayan language spoken in the Argentinean and Paraguayan Chaco. The data is based on original fieldwork done by the author, with several Nivaĉle speakers in the communities of Uj’e Lhavos and Santa Teresita (Paraguay).  This research has a twofold contribution. On the one hand, it adds to the documentation of an endangered and understudied Chaco language. On the other hand, it deepens our understanding of Nivaĉle segmental phonology and advances an Optimality Theoretic analysis of Nivaĉle prosodic phonology. One of the central topics of this dissertation is the interaction between prosodic constituency, stress, and the realization of the constricted glottis ([c.g.]) feature in vowels. Contra Stell (1989), I propose that there is no phonological opposition between modal vowels vs glottalized vowels; rather, Nivaĉle glottalized vowels are sequences of /Vʔ/, a vowel plus moraic glottal stop with different prosodic parsings. A superficially complex stress system in Nivaĉle is shown to reduce to systematic regularities of three types. First, it is shown that stress is quantity-sensitive, with a consistent correlation between bimoraic weight (tautosyllabic /Vʔ/) and stress prominence. Secondly, primary/secondary stress patterns reflect competing edge-alignment constraints where prosodic foot domains align with internal morphological category (MCat) edges. Thirdly, it is argued that a CVC syllable, which constitutes the Minimal Prosodic Word in Nivaĉle, can function as a degenerate foot. The generalization that it characteristically surfaces with secondary (rather than primary) stress is shown to be an emergent consequence of independently motivated constraint rankings.  With regards to the Nivaĉle lateral obstruents, it is argued that the typologically rare velar lateral /k ͡l/ is a complex segment that is the diachronic result of lateral hardening of Proto-Mataguayan *l. Based on its phonological patterning, it is proposed that /k ͡l/ is specified for DORSAL and [lateral]. Integrating multiple facets of these prosodic and segmental analyses, vowel-consonant metathesis further deepens our understanding of the complex interplay of Nivaĉle phonological constraints. Metathesis is shown to be motivated by satisfaction of the Syllable Contact Law, interacting with constraints governing complex codas, derived complex onsets, epenthesis, and deglottalization.      iii Preface   This dissertation consists of original and independent work by the author, Analía Gutiérrez. The following is a list of presentations and publications in which various parts of this dissertation were first introduced.   1. Preliminary analysis included in Chapter 3 and 4 were presented in CILLA VI (2013), SSILA 2015, and WCCFL 33 (2015).  2. Parts of the discussion in Chapter 5 were presented at WECOL 2011, SSILA 2012, and published in Gutiérrez (2014).  3. Some aspects of the data and analysis included in Chapter 6 were presented at WSCLA 15 (2010), SSILA 2011, and published in Gutiérrez (2010, 2012).   The fieldwork undertaken for this dissertation was approved by the UBC Behavioural Research Ethics Board, under the Ethics Certificate Number: H11-00662.     iv Table of Contents  Abstract ........................................................................................................................................................ ii	  Preface ......................................................................................................................................................... iii	  Table of Contents ....................................................................................................................................... iv	  List of Tables............................................................................................................................................... ix	  List of Figures ............................................................................................................................................. xi	  List of Symbols ......................................................................................................................................... xiv	  List of Abbreviations................................................................................................................................. xv	  Acknowledgements.................................................................................................................................. xvii	  Dedication .................................................................................................................................................. xx	  Chapter 1: The Nivaĉle language: Background ....................................................................................... 1	  1.1 Goals ............................................................................................................................................... 1 1.2 The Nivaĉle language and its speakers .......................................................................................... 1 1.2.1 The Mataguayan family and the Gran Chaco......................................................................... 2 1.2.2 The regions and the sociolinguistic situation ......................................................................... 4 1.2.3 The typological profile of the Nivaĉle language: An overview............................................. 9 1.3 Previous linguistic research and descriptions on the Nivaĉle language ...................................... 10 1.4 Methodology ................................................................................................................................. 13 1.5 Representation of data and examples ........................................................................................... 16 1.5.1 Transcriptions and alphabets................................................................................................. 16 1.5.2 Conventions used in examples.............................................................................................. 18 1.6 Structure of the dissertation.......................................................................................................... 18	  Chapter 2: Overview of the Nivaĉle phonological system ..................................................................... 21	  2.1 Introduction................................................................................................................................... 21 2.2 Phonological inventory ................................................................................................................. 21 2.2.1 Consonants ............................................................................................................................ 21 2.2.1.1 Consonant distribution................................................................................................... 23 2.2.2 Vowels................................................................................................................................... 30   v 2.3 Nivaĉle syllable structure and phonotactics ................................................................................. 39 2.3.1 Overview of syllable structure .............................................................................................. 39 2.3.1.1 Onsets............................................................................................................................. 40 2.3.1.2 ʔ-epenthesis.................................................................................................................... 43 2.3.1.3 Codas ............................................................................................................................. 44 2.3.1.4 Nucleus .......................................................................................................................... 49 2.3.2 Sonority ................................................................................................................................. 49 2.4 Pronominal allomorphy ................................................................................................................ 57 2.5 Phonological processes in Nivaĉle ............................................................................................... 63 2.5.1 Palatalization ......................................................................................................................... 63 2.5.2 Epenthesis.............................................................................................................................. 67 2.5.3 Vowel harmony ..................................................................................................................... 70 2.5.4 Metathesis and deletion......................................................................................................... 72 2.5.5 Deglottalization of glottalized consonants and vowels ........................................................ 73 Chapter 3: The phonetics and phonology of Nivaĉle laryngeals........................................................... 75	  3.1 The problem.................................................................................................................................. 75 3.2 Overview of phonation types........................................................................................................ 81 3.2.1 On glottal stop and creaky voice in Nivaĉle ........................................................................ 83 3.2.2 The variable realization of Nivaĉle glottalized vowels ........................................................ 85 3.2.3 On the relationship between [c.g.] and the acoustic parameter of duration ........................ 89 3.3 Feature specification of Nivaĉle laryngeals ................................................................................. 95 3.3.1 On the ambiguous behavior of glottal stop .......................................................................... 95 3.3.2 Phonological patterning of [ʔ] .............................................................................................. 97 3.3.3 Glottal stops and syllable structure..................................................................................... 102 3.3.3.1 Glottal stop as an epenthetic onset.............................................................................. 102 3.3.3.2 Non-epenthetic glottal stop onset ................................................................................ 105   vi 3.3.3.3 Glottal stop as coda ..................................................................................................... 110 3.3.4 Feature specification of Nivaĉle glottal stop ...................................................................... 112 3.4 Prosodic representation of Nivaĉle glottalized vowels .............................................................. 120 3.5 Conclusions ................................................................................................................................. 127 Chapter 4: Prosodic structure of Nivaĉle.............................................................................................. 128	  4.1 Introduction................................................................................................................................. 128 4.2 Minimality: Nivaĉle CVC monosyllables .................................................................................. 132 4.3 Prosodic properties of /ʔ/ ............................................................................................................ 142 4.3.1 Syllabic status of /ʔ/ vs Glides............................................................................................ 144 4.3.2 Syllable structure parsing constraints for /ʔ/ ...................................................................... 146 4.4 Stress patterns in Nivaĉle ........................................................................................................... 153 4.4.1 Previous accounts................................................................................................................ 155 4.5 Stress and affixation processes ................................................................................................... 158 4.5.1 Nominal domain.................................................................................................................. 159 4.5.2 The Root .............................................................................................................................. 161 4.5.2.1 Loanword Phonology .................................................................................................. 171 4.5.3 MSt1: Derivational suffixes ................................................................................................ 172 4.5.4 The MSt2 domain: Possessive prefixation ......................................................................... 182 4.5.5 The MWd domain: The plural suffix.................................................................................. 194 4.5.6 Secondary stress & domain interaction .............................................................................. 196 4.5.7 Prefixation and suffixation.................................................................................................. 201 4.5.8 Verbal domain ..................................................................................................................... 206 4.6 Conclusions ................................................................................................................................. 212 Chapter 5: Nivaĉle laterals: Implications for typology and feature specification............................. 214	  5.1 Introduction................................................................................................................................. 214 5.2 Laterals: resonants and lateral obstruents................................................................................... 216   vii 5.2.1 Cross-linguistic typology of laterals ................................................................................... 216 5.2.2 Lateral affricates and complex segments............................................................................ 220 5.3 Phonological behavior of k ͡l........................................................................................................ 222 5.3.1 Phonotactic patterning of affricates, laterals and ejectives ................................................ 222 5.3.2 Phonological status of k ͡l ..................................................................................................... 229 5.4 Acoustic properties of Nivaĉle laterals ...................................................................................... 232 5.4.1 On Nivaĉle k ͡l ...................................................................................................................... 232 5.4.1.1 [k ͡ɫ] and [l].................................................................................................................... 233 5.4.1.2 [kɫ͡] and [k]................................................................................................................... 235 5.4.2 On Nivaĉle ɬ ........................................................................................................................ 238 5.4.2.1 FFT and LPC spectra................................................................................................... 239 5.5 On the featural representation of Nivaĉle laterals...................................................................... 244 5.6 Comparative data from Mataguayan languages ......................................................................... 251 5.7 Historical perspectives: The emergence of the complex segment k ͡l......................................... 256 5.7.1 Fortition ............................................................................................................................... 257 5.7.2 Prestopping.......................................................................................................................... 258 5.8 Conclusions ................................................................................................................................. 262 Chapter 6: Vowel-consonant metathesis in Nivaĉle ............................................................................. 263	  6.1 Introduction................................................................................................................................. 263 6.2 Alternating vs. non-alternating stem forms................................................................................ 266 6.2.1 Noun plurals: Non-alternating V-final noun stems ............................................................ 266 6.2.2 Noun plurals: Glottal-final stems........................................................................................ 271 6.2.3 Noun plurals: Metathesis in C-final stems ......................................................................... 272 6.2.4 Noun plurals: C-final stems with Vowel epenthesis .......................................................... 274 6.3 Driving forces behind metathesis: Syllable structure constraints .............................................. 278 6.3.1 *COMPLEXCODA ................................................................................................................. 278   viii 6.3.2 Syllable Contact Law .......................................................................................................... 280 6.3.3 Vowel epenthesis................................................................................................................. 285 6.4 Domain of metathesis ................................................................................................................. 288 6.5 Featural effects of metathesis ..................................................................................................... 293 6.5.1 Deglottalization of stops ..................................................................................................... 293 6.5.2 Deglottalization of vowels .................................................................................................. 295 6.5.3 Spirantization....................................................................................................................... 304 6.6 Metathesis and historical vowel deletion ................................................................................... 309 6.6.1 Previous accounts................................................................................................................ 309 6.6.2 Theoretical perspectives on metathesis............................................................................... 315 6.7 Conclusions ................................................................................................................................. 321	  Chapter 7: Conclusions........................................................................................................................... 323	  7.1 Summary and contributions........................................................................................................ 323 7.2 Future research directions........................................................................................................... 328 Bibliography ............................................................................................................................................ 331	     ix List of Tables  Table 1.1 Correspondences between orthographies and the IPA ............................................................... 17 Table 2.1 Nivaĉle consonants ...................................................................................................................... 22 Table 2.2 Nivaĉle vowels (adapted from Stell 1989: 57) ........................................................................... 30 Table 2.3   Nivaĉle vowels: Synchronic Proposal ...................................................................................... 39 Table 2.4 Nivaĉle core syllable structure and the role of ʔ ........................................................................ 40 Table 2.5 Initial CC clusters [Note: the +sign indicates that the sequence is only attested across morpheme boundaries]................................................................................................................................. 42 Table 2.6  Medial CC clusters .................................................................................................................... 48 Table 2.7 Co-occurrence constraints on MANNER: Onset CC Obstruents with equal sonority ................. 52 Table 2.8 Co-occurrence constraints on MANNER: Onset CC with rising sonority.................................... 53 Table 2.9 Co-occurrence constraints on PLACE........................................................................................... 53 Table 2.10 Co-occurrence constraints on MANNER..................................................................................... 54 Table 2.11 Nivaĉle pronominal possessive paradigm................................................................................. 58 Table 2.12 Palatalization: Comparison within the Mataguayan family...................................................... 64 Table 3.1  Cognates of Nivaĉle “glottalized vowels”................................................................................. 77 Table 3.2  Comparative duration of Vʔ and VT ......................................................................................... 94 Table 3.3 First person plural inclusive possessive prefixes...................................................................... 103 Table 3.4 Syllable types and glottal stop................................................................................................... 111 Table 3.5 Feature specification of glottal stop and ejectives .................................................................... 120 Table 4.1 Morphologically derived vs non-derived and relevant domains .............................................. 180 Table 4.2 Nivaĉle pronominal possessive paradigm (cf. Table 2.11)....................................................... 183 Table 4.3 Verbal paradigm of to wash ...................................................................................................... 209 Table 4.4 Verbal paradigm of to wash oneself ......................................................................................... 210 Table 5.1 Affricates, laterals and ejectives................................................................................................ 223   x Table 5.2 Initial CC clusters: Affricates, laterals and ejectives................................................................ 223 Table 5.3 Feature representation of fricatives, stops and k ͡l ..................................................................... 249 Table 5.4 The lateral systems of Mataguayan languages.......................................................................... 252 Table 5.5 Comparative evidence for *l > l and k ͡l................................................................................... 253 Table 5.6 Comparative evidence for *hl > ɬ............................................................................................ 254 Table 6.1 Vowel-alternation examples (adapted from Campbell & Grondona 2007:5) .......................... 310 Table 6.2 Vowel-deletion with loss of /ʔ/ (after Campbell & Grondona 2007:7) ................................... 313    xi List of Figures  Figure 1.1 The Gran Chaco region and the approximate location of the Uj’e Lhavos, Santa Teresita and other Nivaĉle communities. (Source: Wikipedia, public domain)................................................................ 5 Figure 1.2  Nivaĉle subgroups in Paraguay (1: chishamnee lhavos, 2: shichaam lhavos, 3: yita’ lhavos, 4: jotoy lhavos, 5: tavashay lhavos. Source: Stell 1989:24) ......................................................................... 6 Figure 2.1 Waveform and spectrogram of [napuʔ́] ‘two’ by male speaker FR.......................................... 28 Figure 2.2 Waveform and spectrogram of [nap’u]́ ‘myth’ by male speaker FR........................................ 28 Figure 2.3 Waveform and spectrogram of [tí] ‘that’ by male speaker FR................................................. 28 Figure 2.4 Waveform and spectrogram of [tʼíʔ] ‘broth’ by male speaker FR ............................................ 28 Figure 2.5 Waveform and spectrogram of [ɬakúʔ] ‘his cheek’ by female speaker TS .............................. 29 Figure 2.6 Waveform and spectrogram of [ɬakʼú] ‘his weapon’by female speaker TS.............................. 29 Figure 2.7  Waveform and spectrogram of [tik’ín] ‘small’ by male speaker FR....................................... 30 Figure 2.8 Waveform and spectrogram of  /xɑkʔin/: [xɑkʼín] ‘I am leaving’ by male speaker FR........... 30 Figure 2.9  Nivaĉle vowels charted in a two-dimensional unnormalized F1-F2 space, male speaker (FR) in blue, female speaker (TS) in pink ........................................................................................................... 35 Figure 3.1 Continuum of glottal constrictions (after Ladefoged 1971), reproduced from Gordon & Ladefoged (2001). ........................................................................................................................................ 82 Figure 3.2 Waveform of Nivaĉle voice qualities: modal [e], creaky [e ̰], and [eʔ], from male speaker FR.84 Figure 3.3 Waveform and spectrogram of [k ͡lóʔo ̰̰p] ‘winter’ by male speaker MV. .................................. 87 Figure 3.4 Waveform and spectrogram of [k ͡lo ̰́p] ‘winter’ by male speaker MV. ..................................... 87 Figure 3.5 Waveform and spectrogram of [fajxoʔ́] ‘charcoal’ by female speaker CS............................... 88 Figure 3.6 Duration results for creaky vs. modal vowel pairs across six Nivaĉle speakers: FAR, FR, GR (male) and CS, RF, TS (female) .................................................................................................................. 93 Figure 3.7 Waveform and spectrogram [t’akúmʔin] ‘s/he works’ by female speaker TS. ...................... 107 Figure 3.8 Waveform and spectrogram [kot ͡sxátʔe] ‘on the land’ by male speaker FR........................... 108   xii Figure 3.9 Waveform and spectrogram [kot ͡sxát’e] ‘on the land’ by male speaker FR. .......................... 108 Figure 3.10 Waveform and spectrogram of [xapɛ̀ʔjeʔéj] ‘I heard’ by male speaker FR ......................... 115 Figure 3.11 Prosodic representation of /Vʔ/.............................................................................................. 121 Figure 3.12 Waveform and spectrogram of [ɬaʔ́] ‘its fruit’ by male speaker FR..................................... 125 Figure 3.13 Waveform and spectrogram of [ta ̰́j] ‘it has fruit’ by male speaker FR ................................ 125 Figure 3.14 Waveform and spectrogram of [xawɑ̰́m] by male speaker FR ............................................. 126 Figure 3.15 Waveform and spectrogram of [xawɑ́ʔmeʃ] by male speaker FR ........................................ 126 Figure 3.16 Waveform and spectrogram of [ʔaβu ̰́n] ‘your roasted meat’ by female speaker TS............ 126 Figure 3.17 Waveform and spectrogram of [ʔaβuʔ́na] ‘there is no (your) roasted meatʼ by female speaker TS. .............................................................................................................................................................. 126 Figure 4.1 Prosodic representation of /Vʔ/................................................................................................ 147 Figure 4.2  Word-internal morphological relationships ............................................................................ 159 Figure 4.3 Morphological structure of the noun ....................................................................................... 160 Figure 4.4  Domains for stress assignment in a noun ............................................................................... 160 Figure 4.5. Morphological structure of the verb ....................................................................................... 207 Figure 5.1 Waveform and spectrogram of [ek ͡ɫe] ‘parrot’ by female speaker TS.................................... 233 Figure 5.2 Waveform and spectrogram of [ʔele] ‘missionary’ by female speaker TS. ........................... 234 Figure 5.3 Mean average of  F1, F2, and F3 at 7 timepoints of [ɫ] in [ek ͡ɫe] across 5 tokens................. 235 Figure 5.4 Mean average of  F1, F2, and F3 at 7 timepoints of [l] in [ele] across 5 tokens. .................. 235 Figure 5.5 Waveform and spectrogram of [xɑ̀ʔpəkl͡éɬ] ............................................................................. 236 Figure 5.6 Waveform and spectrogram of [xɑpék]................................................................................... 236 Figure 5.7 Waveform and spectrogram of [uk ͡lʔɑ́] .................................................................................... 237 Figure 5.8 Waveform and spectrogram of [uk’a]́...................................................................................... 238 Figure 5.9 FFT spectra of Nivaĉle fricatives /f s ʃ ɬ x/, male speaker FR................................................ 241 Figure 5.10 FFT spectra of Nivaĉle fricatives /f s ʃ ɬ x/, female speaker TS........................................... 242   xiii Figure 5.11 LPC spectra of Nivaĉle fricatives, male speaker FR............................................................. 243 Figure 5.12 Spectrogram of a prestopped lateral in Montana Salish (Ladefoged & Maddieson 1996:201).. ................................................................................................................................................. 259    xiv List of Symbols   µ = mora σ = syllable ́ = primary stress ̀ = secondary stress ‘.’= syllable boundary ‘-’ = affix boundary ‘=’ = clitic boundary ( ) = foot boundary    xv List of Abbreviations   1 = first person 2 = second person  3 = third person  alv. = alveolar A = agentive ANTIP = antipassive AR = area ART = artifact AUG = augmentative BEN = benefactive C = consonant CAUS = causative CISL = cislocative CLASS = classifier  CLPN= Linguistic Committee of the Nivaĉle People COM = comitative COL = colective CON = container COMP =complementizer COORD = coordinative dB = decibels DET = determiner DIST = distal EXCL = exclusive F = feminine Ft = foot F1 = first formant F2 = second formant Ft = foot Hz = Hertz H = homorganic HUM = human IMP = imperative INC = inclusive INDEF = indefinite INT = intransitivizer   xvi IPFV = imperfective IT = itive LOC = locative MAL = malefactive MARK = mark/trace MAT = material MED = mediative MCat = morphological category MSt = morphological stem MWd = morphological word ms = milliseconds N = nucleus N = non NEG = negation NMLZ = nominalizer O = obstruent O = object OT = Optimality Theory P = patient PL = plural PLANT = plant POSS = possessive PR =pronominal PRO = pronoun PUNC = punctual  PrWd = Prosodic Word R = resonant (sonorant) REL = relative REFL = reflexive RES = resultative S = subject s = second s.o. = someone s.t. = something T = stop  V = vowel VENT = ventive VBLZ = verbalizer   xvii Acknowledgements  I am a grateful person; it is a significant moment for me to show my appreciation to the people who have accompanied and helped me on this academic and personal journey.  First and foremost, I would like to thank all the Nivaĉle people I had the privilege and fortune to meet and work with. I am especially grateful to the Rojas Núñez family for they have always treated me as a family member: Sara, Rosalina, José, Inés, Clara, Elizabeth, and her little Lisandro. Sara Rojas, in particular, was my first Nivaĉle teacher and friend, and left this world too young; she is very much loved and missed. Don Félix Ramírez Flores has been my main consultant throughout this research project. I am deeply thankful for his patience and persistence, for all the wonderful work in teaching me the Nivaĉle language, and for sharing his passion for the Nivaĉle culture. Teresita Sánchez has also been a great and inspiring teacher. Her generosity, clarity, and intuitions have helped me greatly: jaichavalhch’e ti lhasfiyit can’um nôque’esh ts’ishamesh ti javanch’e cava shtanei jatsjanesh’in can’um tsitôi c’oya nôque’esh. My special gratitude goes to many other Nivaĉle consultants and teachers: Andrés Crespo, Faustino Ramírez, Graciano Ramírez, Myriam González, Raquel Fleitas González, José González, Francisco Fleitas, Agustín Juárez, Celestina Céspedes, Celestina Sánchez, Erasmo Pintos, Mauricio Valdéz and Sofía. I am especially indebted to the members of the Linguistic Committee of the Nivaĉle People for inviting me to form part of their language and culture promotion efforts. It has been a privilege working with such a wonderful and dedicated group of people. Thank you so much for your trust and for sharing abundant insights on the Nivaĉle language and culture: ts’ishamelh’ash chi’ pa Fitsôc’ôyich.  A special thanks to Miguel Fritz for sharing his experience working with the Nivaĉle people, and for introducing me to so many valuable people in Uj’e Lhavos and Filadelfia. I am equally thankful to Julia Isabel Gómez Giménez for sharing her vast knowledge on the Nivaĉle education system, for introducing me to many Nivaĉle teachers, and for offering a help hand during my visits and then in the distance. I would also like to thank Irma Hein for sharing her work on the Nivaĉle language. Thanks to Miguel del Puerto, Gundolf Niebuhr and Cristof Eitzen for their generosity in allowing me hold some elicitation sessions at the Uj’e Lhavos School, the Fernheim Colony Archives and the Kalofoné Studio, respectively. Sincere thanks to Verónica Villalba, Patricia Fernández, and Ana and Amadeo Benz, for their friendship and support during my stays in Filadelfia. They made me feel at home: aguyje peeme che pyaite guive.  This work would not have been written without the unwavering support from my doctoral supervisor, Patricia A. Shaw. Since the first day I arrived at UBC, Pat trusted my project on the Nivaĉle language and helped me in many different ways to make my research happen. During the write-up of my dissertation, Pat was incredibly instrumental in guiding me to pen and (importantly) complete this project, which has greatly benefited from her vast knowledge in phonology and linguistic fieldwork. Pat has guided me through every step along the academic process; and has offered support in personal matters, with invaluable patience, generosity and kindness. I am deeply grateful and indebted to her.        xviii I am also enormously grateful to Gunnar Ó. Hansson and Molly Babel, members of my dissertation committee. Gunnar trusted my dissertation project, carefully read my thesis, formulated continuous thought-provoking questions, shared countless insights on phonology and historical linguistics, and overall was very supportive and encouraging during this process. Molly offered constant help with the phonetic analysis of my data, challenged my working thoughts and ideas in productive ways, enthusiastically pushed me to get things done, and was always very encouraging. To my committee: thank you so much for your knowledge and generosity in guiding me through this academic journey! I am very privileged to have had Megan J. Crowhurst as my external examiner. I am deeply thankful for her detailed, careful and thoughtful comments and questions. Many thanks to Patrick Moore and Candace K. Galla for having served as my university examiners. I appreciate their feedback, questions and gentleness during my defence.  My fieldwork research was funded by the Jacobs Research Funds (Whatcom Museum, 2010), the Bottom Billiom Fieldwork Fund (Liu Institute for Global Issues, UBC, 2011), and a Small Grant from ELPD (SOAS, 2012). I sincerely thank each of these sources for supporting my work.  In addition to my committee members, I have benefited from the teachings and help of a number of professors during my Ph.D. at UBC. Many thanks to Strang Burton, Henry Davis, Brian Gick, Lisa Matthewson, Amanda Miller, Douglas Pulleyblank, Hotze Rullmann, Joe Stemberger, and Martina Wiltschko. Special thanks are due to Edna Dharmaratne and Shaine Meghji not only for their constant help in administrative matters, but also for their warm presence, advice and kindness.  I would like to extend my sincere gratitude to Héctor Frattini, for his continuous support and for parallel analyses, and to Gareth Sirotnik, for reminding me about the stream that has always been here.  Thanks to my cohort for helping me navigate the first two years in the PhD. program: Majed Al Solami, Ruby Arkoh, Chenhao Chiu, Pat Littell, Megan Louie, Scott Mackie, Masaki Noguchi, Abigail Scott, and Audra Vincent. I would also like to thank my other peers from the UBC Linguistics Department, particularly: Jen Abel, Alexis Black, Mario Chávez-Peón, Raphael Girard, Maria Amélia Reis Silva, Beth Stelle, Anita Szakay, Noriko Yamane, Sonja Thoma and Natalie Weber. Also, thanks to Emily Elfner and Yuwen Lai. I would like to restate my thanks to Chenhao, Mario and Masaki for their friendship and companionship: Chenhao, for always lending me a hand with a smile; Mario, for sharing his thoughts on glottalization and all the encouragement; Masaki, for his help, hikes and chelas.  I owe a lot of gratitude to Lucía A. Golluscio, Alejandra Vidal and Hebe González for their inspiring work, their academic training and all their support at different stages in my career. I also thank Verónica Nercesian, for sharing her knowledge and insights on the Wichí language, for all the lively discussions, and for her friendship.  A number of dear people have made my days in Vancouver sunnier. I am thankful to Mike Brown for the stimulating conversations, bike rides and movies. Thanks to Manu Fonseca for his kind support, and for reminding me that I would get there. I am grateful to Oralia Gómez-Ramírez for being such a loving, wise, and generously supportive friend. Muchas gracias a Ana Vivaldi, Rafael Wainer, Franka and Ramona, for providing me with a sense of familiar warmth. Obrigada a Amélia Reis Silva   xix for her generosity and help, for making me feel at home, and for all the memorable times we spent together (also thanks to Rafa and Tomás). Merci beaucoup, Raphael Girard, compañero…por todas las conexiones, las percepciones y las emociones compartidas. My caring friends in Buenos Aires have cheered me up and welcomed me back again and again. Many thanks to Gaby, Mariela, Ana, Daniela, Manu and Ivi. And to Gus, for the thesis-writing and pre-defence music.  I would like to thank Matthew Bennett for his encouragement, and for believing in my capacity to start and finish a Ph.D. abroad, and in a different language, despite all the difficulties I had to face.  Last, but not least, I wish to express my thankfulness to my family for their unconditional help and love. Thanks to my niece, Violeta, and my nephew, Simón, for the steps we made together and for their healing energy and love; Titi los adora. I am thankful to my inspiring sister, Mariana. Muchas gracias, hermana, por acompañarme con tus palabras sabias, tu abrazo apretado y tu corazón luminoso, siempre cerca. My special gratitude to my dear mother and father, Amparo and Oscar. Muchas gracias mamá y papá, de todo corazón, por ser mi pilar más fuerte en la vida y por estar siempre a mi lado. Su constante apoyo, compañía, paciencia, confianza y amor me han mantenido siempre en pie y me han impulsado a seguir pedaleando.     xx Dedication   Para mis padres, Amparo y Oscar, por su fortaleza y amor  1 Chapter 1: The Nivaĉle language: Background   1.1 Goals  This dissertation investigates a series of phonological and phonetic aspects of Nivaĉle, a Mataguayan language spoken in the Argentinean and Paraguayan Chaco. The data and generalizations presented here are based on original fieldwork done by the author between 2009-2013. This study focuses on several phenomena that deal with the representation, distribution, and organization of the sounds in this language, which are of interest from both typological and theoretical perspectives. Specifically, the phenomena investigated in this thesis include the phonological status of glottal stop and glottalized vowels, the lateral obstruents [ɬ] and [k ͡l], and some morpho-phonological processes such as VC-metathesis. One of the central topics of this dissertation is the interaction between prosodic constituency, stress and the realization of the constricted glottis ([c.g.]) feature in vowels.   The goals of this dissertation are twofold. First, the description and analysis of the abovementioned phenomena aim to deepen our understanding of the segmental and prosodic phonology of the Nivaĉle language. Also, it is the hope that the implications of my analysis of glottalization can set the basis for a comparative phonological study within the Mataguayan language family. Secondly, as this study explores the relationship between glottalization and metrical structure in phonological theory, it hopes to contribute to the literature on glottalization in languages of the Americas, and the relationship between glottalization and prosody (England 1983, Macaulay & Salmons 1995, Gerfen 1999, Blankenship 2002, Avelino 2004, 2011, Kehrein & Goldston 2004, Picanço 2005, Gerfen & Baker 2005, Stenzel 2007, Arellanes 2009, Elías-Ulloa 2009, Frazier 2009, Herrera Zendejas 2009, Baird 2011, Baird & Pascual 2012, Storto & Demolin 2012).  1.2 The Nivaĉle language and its speakers      This section provides an overview of the Nivaĉle language and the people who speak it. The family relations of the Nivaĉle language are reported in §1.2.1. Geographic and sociolinguistic aspects of the Nivaĉle language are summarized in §1.2.2.    2 1.2.1 The Mataguayan family and the Gran Chaco Nivaĉle [niβaˈk ͡le] (ISO 639-3: cag) is a Mataguayan language spoken in the Argentinean and Paraguayan Chaco by approximately 16,350 speakers in Paraguay (DGEEC 2012) and 553 in Argentina (INDEC, 2004-2005). The word Nivaĉle means “human being” in a broad sense (Chase Sardi 1990: 7); for the Nivaĉle people, it means “person” and “man” (Fritz 1994: 35). According to Stell (1989:17) the first reference of the name Nivaĉle in the literature can be found in Susnik (1961:47), who maintains that “niwaqli” means “men”, and that this name includes the whole (Nivaĉle) nation.  The Nivaĉle language has also been referred to in the literature as Gentuse/Wentusi/Wentusix (Espínola 1794, Greenberg 1956, Loukotka 1968, as cited in Stell 1989:20, Ashlushlay (Nordenskiöld 1910, Henry 1939, Wicke & Chase-Sardi 1969, Stell 1972), Chorupí (Lehmann-Nitsche 1936) Churupí (Schmidt 1940), Chulupí (Junker, Wilkskamp & Seelwsiche, 1968; Stell 1989), Chunupí or Suhin (Hunt 1915, 1924), and Chunupí (Palavecino 1936, Mason 1950, Tovar 1964).1 While Chulupí is commonly used in Argentina, Nivaĉle is the term used in Paraguay.2 Here I adopt the spelling Nivaĉle, rather than Nivaclé, Nivakle, or Niwakle, following the conventions established during the II Nivaĉle Linguistic Conference (Uj’e Lhavos, Paraguay, December 3-5 2010). During that conference the Linguistic Committee of the Nivaĉle People  (Comisión Linguística del Pueblo Nivaĉle, CLPN) was created.  One of the goals of the CLPN, formed by Nivaĉle teachers and specialists on the Nivaĉle language and culture, was to revise and consolidate the two Nivaĉle orthographies, one proposed by the Catholic missionaries and the other proposed by the Mennonite missionaries. Besides Nivaĉle, the Mataguayan language family (Swadesh 1959, Najlis 1984, Fabre 2005) comprises three other languages: Chorote, Maká, and Wichí.  This language family has also received alternative names in the literature, such as Mataco (Loukotka 1968: 53-55, Voegelin & Voegelin 1977:                                                 1 This name has caused some confusion in the literature because Chunupí is an alternative name of Vilela (Lule-Vilela), a genetically unrelated Chaco language.  2 According to Hunt (1915) and Stell (1989:17) Chulupí derives from Wichí tsonape/sonape “shepperd”. According to Nordenskiöld (1910), the name Ashlushlay was given by the Chorote people: from [aɬu] lizard, [ɬaj] “people or fruits”, meaning “the people that eat lizard”. The name Wentusix has been attributed to the Maká people (Stell 1989, Fabre 2014): “to cut one’s hair”.  3 223-224) Mataco-Mataguayan (Tovar 1951: 400, 1961, 1964), Mataco-Maka (Kaufman 1990:46), and Matacoan (Campbell 2012).  Within the Mataguayan family, Tovar (1964: 371) proposes that Wichí and Chorote are more closely related with each other than any other language of the family because of 50% shared vocabulary. It is also worth mentioning some interesting phonological similarities between these two languages; e.g., the presence of preglottalized resonants (Carol 2014, Nercesian 2014a).    Based on Tovar’s lexical study of the Mataguayan languages (Tovar 1964), Fabre (2005:3) proposes the existence of two branches within the Mataguayan family; on the one hand, Chorote and Wichí, and on the other hand, Nivaĉle and Maká (which share 43% of their vocabulary). Yet, this classification deserves a word of caution, as there are morphosyntactic features, for instance, the determiner system, that makes Chorote more similar to Maká and Nivaĉle (Fabre 2005, Carol 2014). According to Campbell (2012: 98), Mataguayan languages are “diversified on a scale similar to Germanic languages”. In sum, even though the relationships between these languages are clearly supported, a thorough and systematic comparative study within this language family is yet to be done.  The location of the Mataguayan languages and peoples spans across Northeastern Argentina, Southeastern Bolivia, and Southwestern Paraguay; a region known as the Gran Chaco (from Quechua chaku “hunting land”). The Gran Chaco comprises about 1,000,000 square kilometers divided between Northern Argentina, Eastern Bolivia, Western Paraguay and South-Eastern Brazil. Approximately twenty-nine languages belonging to seven language families with different degrees of vitality (Arawakan, Guaycuruan, Lule-Vilela, Mataguayan, Tupí-Guaraní, Maskoyan (or Enlhet-Enenlhet) and Zamucoan) and two language isolates, Chiquitano (or Besiro) and Guató, are spoken in this region (Golluscio & Vidal 2009-2010). With regards to the relationship between the Mataguayan and other language families, several proposals have linked the Mataguayan and Guaycuruan language families in a Macro-Guaycuruan group  4 (Henry 1939, Mason 1950, Greenberg 1956, 1987, Kaufman 1990, Viegas Barros 1993, 2004).3 According to Greenberg (1987), the Macro-Guaycuruan group belongs to the Macro-Panoan branch, which, along with Macro-Carib and Macro-Ge-Bororo comprise the Ge-Pano-Carib phylum. Besides the hypotheses regarding the genetic affiliations of the Mataguayan and Guaycuruan languages, it has been proposed that the Chaco languages share morphosyntactic and phonological features that stem from their extensive historical contact (Comrie et. al 2010, Golluscio & Vidal 2009-2010, González 2014), and thus constitute a linguistic area.  1.2.2 The regions and the sociolinguistic situation It is believed that the original, pre-contact territory of the Nivaĉle people used to be located between the Bermejo and the Pilcomayo rivers (Hunt 1913-15, Fritz 1994, Andrés Crespo, p.c., (cf. Figure 1.1). At the beginning of the 20th century, due to pressures from European colonizers, the Argentine military, and numerous battles with the Toba people, the Nivaĉle people retreated to the Pilcomayo River and crossed it.4  Later (1920-1930), the Nivaĉle people would be invaded and fought by the Bolivian militaries prior to the Chaco War (1932-1935), where Paraguay and Bolivia disputed what is now the Paraguayan Chaco.   In 1925 and 1927, the first Catholic Missions, San José de Esteros and San Leonardo (Fischat), were established in Nivaĉle territory, next to the Pilcomayo River; it is considered that the first prolonged contact between the Nivaĉle and the white people occurred there (Andrés Crespo, p.c.). During the Chaco War, the Catholic missionaries protected the Nivaĉle who were in the firing line of Bolivian and Paraguayan troops (Fritz 1994: 29). It was also during the wartime that many Nivaĉle people abandoned their villages and migrated to Argentina to work in the sugar plantations (Stell 1989:8). After the war, the migration reversed direction and the Nivaĉle began their annual migration to                                                 3 The Guaycuruan languages comprise Mocoví, Pilagá, Toba (or Qom), Kadiwéu (or Caduveo) and Abipón, no longer spoken.  4 According to Hunt (1913-15: 258), cited in Stell (1989:8), in 1913 “the last village on the Bermejo River was broken up and its members joined their compatriots across the Pilcomayo River”.  5 (and settled down around) the Mennonite colonies of Neuland and Fernheim, Boquerón County, Paraguay, in search of agricultural work (Chase Sardi 1972: 26, cited in Stell 1989:9). 5   Nowadays, the Nivaĉle language is spoken across twenty-four communities in the Boquerón and Presidente Hayes Department in Paraguay (DGEEC, http://www.dgeec.gov.py), and in the provinces of Salta and Formosa in Argentina (UNICEF, http://www.unicef.org):   Figure 1.1 The Gran Chaco region and the approximate location of the Uj’e Lhavos, Santa Teresita and other Nivaĉle communities. (Source: Wikipedia, public domain)6   There has not been complete agreement about the number of subgroups that constitute the Nivaĉle people, not only within the literature but also among the Nivaĉle people. Klein & Stark (1977: 392) maintain that there are two groups: the inland or ‘bush’ Chulupí, and the ‘river’ Chulupí.  In contrast, Chase-Sardi (1981) and Stell (1989), maintain that there are five groups (see map in Figure 1.3):                                                  5 The Mennonite settlers arrived in the Paraguayan Chaco in the late 1920; the Meno, Fernheim, and Neuland colonies were established near present day Filadelfia. It is estimated that more than 10,000 Mennonites live in the Paraguayan Chaco; they are the largest employers of the indigenous peoples of the Central Chaco and approximately 60% of the Paraguayan indigenous population inhabit their colonies and surrounding [or “nearby”] areas (Miller 1999:17).   6 This image is on the public domain due to its age; its copyright has expired. The editor has released the changed image in the public domain. http://commons.wikimedia.org/wiki/File:GranChacoApproximate.jpg  6 (1) chishamne’ lhavos ‘the highlanders (Upriver)’, who live around the Pedro P. Peña area (Paraguay) and Salta (Argentina).  (2) shichaam lhavos ‘the lowlanders (Downriver)’, who live around the Missions of San José de Esteros and San Leonardo/Fischat. Both (i) and (ii) belong to tovoc lhavos ‘people of the (Pilcomayo) river’.  (3) yita’ lhavos ‘people of the scrub’, who live in the Mission Santa Teresita (Mariscal Estigarribia). This group is also known as c’utjaan lhavos ‘people of the thorns.’ (4) jotoy lhavos ‘people of the feathergrass’ who live in the communities around Campo Loa,  58 kms Southeast of Mariscal Estigarriba (FR, p.c).  (5) tavashay lhavos ‘people from the inland’, who live north of San José de Esteros, and Southeast of Filadelfia, close to the Mennonites colonies.    In turn, Fritz (1994) and Siffredi (1989) maintain there are only three groups (i), (ii), and (iii). Below, I present the map with the aforementioned five Nivaĉle subgroups presented by Stell:   Figure 1.2  Nivaĉle subgroups in Paraguay (1: chishamnee lhavos, 2: shichaam lhavos, 3: yita’ lhavos, 4: jotoy lhavos, 5: tavashay lhavos. Source: Stell 1989:24)   Based on my fieldwork and interviews with Nivaĉle speakers, those five major regional subgroups are recognized within the Nivaĉle people. However, it is not very clear what the systematic  7 linguistic differences are (if any) between the jotoy lhavos, the tavashay lhavos and the other subgroups. I can report some regional variants pertaining to the following subgroups: chishamnee lhavos, shichaam lhavos and yita’ lhavos. The regional dialectal differences mostly consist of (i) vocabulary, this is quite evident between the chishamne’ lhavos and the shichaam lhavos speakers, (ii) phonology: (a) in the jotoy lhavos variety there is no low back unrounded vowel /ɑ/ contrasting with /a/ (see minimal pairs in §2.2.2), (b) in the yita’ lhavos variety, the sequence /k ͡lʔ/ is pronounced as [kʼ], rather than [k ͡lʔ] (§3.3.4), (iii) phonetics: the epenthetic vowel [e] is mostly used in in the chishamnee lhavos variety in contrast to [i] in the shichaam lhavos variety (§6.2.4).  Besides the regional variants, I have documented a number of morphosyntactic and lexical differences that have been arising between younger and older generations. In fact, these intergenerational differences have been a concern among older speakers and teachers of the semi-urban community of Uj’e Lhavos (§1.4) as there is the feeling that young people “do not speak the language very well and mix it with Spanish”. In turn, the younger speakers (in their twenties) I have consulted with, mentioned that sometimes they do not understand certain words or expressions used by their grandparents, or that their grandparents say “things differently”.   It is worthy of mention that many of the Nivaĉle traditional practices have been abandoned in the semi-urban communities such as Uj’e Lhavos where there is a closer contact with the Paraguayan and the Mennonite societies; for instance: “elders are no longer telling the myths to the kids at night (…) we started to forget things because we are inside of a town, the town of the white people” (FR, p.c.).7 The role of media is also signaled as a threat: “there is no time now, we have television, we listen to the radio and the news; it is impossible to remember things from the past” (FR, p.c). Also, traditional Nivaĉle practices such as hunting, honey harvesting, the celebration of female rites of passage along with the associated festivities (traditional dances, games, and drinks) are domains of intertwined cultural and language knowledge and use that are not directly accessed by younger generations anymore. It is                                                 7 Note that all “p.c.” quotes and citations from Stell (1989) are given in my own translation from Spanish.   8 thus felt within the community that along with the loss of cultural practices, and due to the pressure from Spanish, the Nivaĉle language will start shrinking. Remarkably, all the speakers of younger generations I worked with, who are between 20 and 45 years old are bilinguals; they are fluent in Spanish as well as in Nivaĉle. This is not exactly the case for speakers that are more than 60 years old; ‘monolinguals’ in Nivaĉle can be found within this age group, some of whom can understand Spanish but who mostly rely on their children and grandchildren to interact with the ‘samto’, the white people.    Nevertheless, a number of crucial factors that promote the maintenance of this language in Paraguay should be highlighted.    First, the Nivaĉle language is still transmitted, spoken at home – by 99% of community members (Melià 2010) – and in the community. Second, even though some Nivaĉle communities live together with the Manjui (the Nivaĉle name given to the Chorote people, and the one used in Paraguay to refer to this group), Enlhet-Enenlhet and Guaraní peoples, bilingualism levels are low and have not motivated language shift to the indigenous languages in contact. For instance, only 6.53% of the Nivaĉle people speak Guaraní, making it one of the indigenous groups in Paraguay that speak Guaraní the least (Melià 2010). Different is the sociolinguistic situation in the multicultural Misión La Paz (Salta, Argentina). There is linguistic exogamy and everyday interactions are made in Nivaĉle, Chorote and Wichí (Campbell & Grondona 2010). The language is not taught in schools, and so the Nivaĉle people from Misión La Paz have asked the Nivaĉle peoples in Paraguay for literacy materials and to help them in the preservation of their language (Erasmo Pintos, p.c.). Third, the Nivaĉle people have positive attitudes towards their language, especially the middle-aged and older generations.8 Fourth, in the Paraguayan Chaco, Nivaĉle writing and reading skills are taught until the sixth grade of Catholic primary school and until the third grade of Mennonite Schools.9 Education in Nivaĉle is either supervised by the Catholics, the Apostolic Vicariate of the Pilcomayo (VAP)                                                 8 This positive attitude has been noted in the literature: “It is felt that the Chulupí are the proudest tribe in the Chaco as far as language and culture are concerned” (Klein & Stark 1977:391-392). 9 Very recently, Nivaĉle teachers have become “national teachers” and so they can teach until grade 4 and 5 of Mennonite schools, though this has not been implemented in all the schools (Wilmar Stahl and Gundolf Niebuhr, p.c.)  9 (http://www.vicpilcomayo.org.py), or the Association for the cooperative services between indigenous communities and the Mennonites (ASCIM http://www.ascim.org) in the areas administered by the Central Mennonite Committee; for example Uj’e Lhavos, Fernheim Colony (Filadelfia).   Regarding the second and third factors, two Nivaĉle linguistic conferences, organized by Nivaĉle teachers in collaboration with VAP and ASCIM, took place in Filadelfia in 2006 and 2010.  During the last conference, the Linguistic Committee of the Nivaĉle People (CLPN) was created with the goals of: (i) revising and unifying the orthographies that were used in Catholic and Mennonites schools,  (ii) creating a new dictionary on the basis on Seelwische’s (1980) original dictionary (§1.3), (iii) developing new literacy materials, and  (iv) promoting the use and preservation of the Nivaĉle language and culture.  1.2.3 The typological profile of the Nivaĉle language: An overview It is considered that Nivaĉle tends to polysynthesis and agglutination (Fabre 2014). The morphology (especially on the verb) is very rich, with both inflectional and derivational prefixes, suffixes and clitics.  It is a head marking language and there are no adpositions, that is, prepositions or postpositions; locative functions are signaled by applicatives, relational names and verb serialization (Fabre 2014). There are five verbal conjugations (Stell 1989, Fabre 2014) based on the semantics and syntax of the verb. Verb marking displays both active and inverse (hierarchical) alignment in this language.  Like many of the Chaco languages, the basic word order is SVO. Another common trait is that Nivaĉle makes a distinction between alienable nouns and inalienably possessed nouns, and inclusive versus exclusive first person plural pronouns.   The determiner system is very complex and besides tense interpretations (Stell 1989) evidentiality distinctions are made through determiner choice (Gutiérrez 2011).   With regards to the phonological features, like in all Mataguayan languages, there is a distinction between plain and ejective stops, no opposition between voiceless and voiced obstruents, and the presence of lateral obstruents.   10 1.3 Previous linguistic research and descriptions on the Nivaĉle language This section provides an overview of the linguistic research done on the Nivaĉle language, in chronological order.   As mentioned in §1.2.2, the contact between the Nivaĉle people and the religious missionaries occurred fairly late in comparison to other indigenous peoples of the area. The traditional territory of the Nivaĉle, along the north margins of the Pilcomayo River, remained unreached from the Spanish colonizers and missionary settlements during the 17th and 18th centuries (Stell 1989:4). Even though the Anglicans made contact with the Nivaĉle people at the end of the 19th century and tried to establish a mission in 1899, it floundered and was abandoned. Likewise, the second mission, Nanawa, established around 1916, failed shortly after its inception (Fritz 1994:26-27). However, the earliest sources on the Nivaĉle language were written by an Anglican preacher, Richard Hunt (1915, 1924), who stayed for two months in the Nanawa mission.10 He published a Chunupí or Suhin-English vocabulary, and a Chunupi or Suhin grammar with lessons. In the vocabulary, Hunt presents the orthographic systems used to represent the sounds, a Chunupí-English vocabulary and a general overview of the language. In the grammar, Hunt also presents the sound system of the language. Interestingly, Hunt makes a distinction between short and long vowels. He also presents palatalized alternants [fj hj kj nj] (fy, hy, ky, ñ) of [f h k n] preceding front vowels, and notes “the hy sound is very strongly aspirated, closely akin to sh and fy” (Hunt 1915:1). Also, it is important to note that, besides [h], Hunt includes j: “guttural as in Spanish”.   Judging by the examples he cites, the difference between short and long vowels mentioned by Hunt must correspond to the difference between modal and glottalized vowels (cf. Chapter 3), though it is not clear if glottalization was present at this diachronic stage of the language. Whereas the palatalized variants of the labiodental fricative and the nasal have not been attested in my fieldwork, the status of the patalalized laryngeal and velar will be later discussed (§2.5.1). The Chunupi or Suhin grammar comprises a description of the primary parts of speech along with exercises and test questions for “the student”.                                                 10  Hunt also did work on Chorote and Wichí, and developed the first Wichi alphabet (1937).  11  The Missionary Oblates of Mary Immaculate were also pioneers in the study and description of the Nivaĉle language in Paraguay. Junker, Wilkskamp & Seelwische (1968) published a Nivaĉle grammar (123 pp.), with lessons and a vocabulary, which was the antecedent of Priest José Seelwische’s (1975a) Nivaĉle-Spanish pedagogical grammar. Another fundamental piece of work made by Seelwische was the first Nivaĉle-Spanish Dictionary (1980, 1990), and a number of pedagogical materials and texts for formal education, some in collaboration with Nivaĉle teachers (Seelwische & Avalos 1972; Seelwische 1975a, b, 1980, 1990, 1993 a, b). These constitute the published work on the language currently available to the Nivaĉle communities in Paraguay.   In turn, a Slovenian linguist based in Paraguay, Branislava Sušnik, published a descriptive grammatical sketch (1954, 60 pp.), and two articles on different aspects of the language (1954, 1959), such as the phonological system, morphological processes, and similarities between the Nivaĉle and the Maká verb.  There have also been some brief comparative lexical studies between the Nivaĉle and other Chaco languages (Henry 1939, Tovar 1962, 1964, Loukotka 1964).  At the end of the 1960s, an Argentine linguist, Nélida N. Stell, started doing fieldwork in Salta and published a phonological sketch (1972). This sketch consists of a list of phones with their (i) articulatory description, (ii) phonological environment and (iii) relevant examples. Stell also presents the Nivaĉle phonemic inventory, where she claims a contrast between modal and glottalized vowels (cf. Chapter 4), with some notes on stress and consonant clusters. Stell’s (1989) doctoral dissertation on the Nivaĉle grammar is based on original fieldwork carried out in Misión Chaqueña “El algarrobal”, Misión La Paz (Salta, Argentina), and Misión San Leonardo de Escalante (Paraguay). This descriptive grammar is divided in five sections: phonology, morpho-phonology, morphology, syntax and texts. The phonology section provides examples with each phoneme, some minimal pairs, lists of words illustrating the types of consonant clusters and syllable types, and a generalization about stress. The morpho-phonology section presents the attested morphophonemic alternations in the language with numerous examples. The morphology section focuses on nominal and verbal paradigms, and presents a  12 list of derivational and inflectional affixes with their possible combinations. The syntax section discusses the nominal and verbal phrases and subordinate clauses. At the end of the grammar, Stell presents 10 texts with their translations. Stell’s work represents a remarkably valuable and quite extensive description of the Nivaĉle language.   The most recent linguistic publications on the Nivaĉle language include work by Campbell & Grondona (2007, 2010, 2012) and Campbell (2013) on historical reconstruction in Nivaĉle, the sociolinguistic situation of the Nivaĉle in Misión La Paz, and a general overview of the language (Campbell & Grondona 2012:625-633). In turn, Fabre studied several aspects of the verbal morphology, such as inverse alignment (Fabre 2009-2010, 2012) and has proposed a grammatical sketch of the language (Fabre 2014). Aside from the present dissertation, I have done work on the determiner system and studied some aspects of the phonology, such as metathesis, glottalized vowels, and the status of the complex segment k ͡l (Gutiérrez 2010, 2011, 2012, 2014).11  It is worth mentioning that Stell and Campbell & Grondona worked with the chishamne lhavos group. In addition, Stell (1989) worked with some shichaam lhavos consultants. However, independently of my own research reported on here, no documentation of the yita’ lhavos variety has been carried out. This is the subgroup that includes the Nivaĉle community of Santa Teresita, where part of my fieldwork has been carried out.  My major complementary source of data comes from Stell’s (1989) doctoral thesis. The main differences between Stell’s and my data lies in my careful attention to documentation of the phonetic realization of vowel-glottal coda (§3.2.2), and the locus of primary and secondary stress in some forms (§4.4.1). Another difference is the distribution of /k ͡l/ in the shichaam lhavos variety. Whereas Stell had noticed that /k ͡l/ consistently delateralized to [k] in coda position, I show that /k ͡l/ can occur before /ʔ/ (cf. Chapter 5). Further, given that my fieldwork included the documentation of the regional variant                                                 11 Also, there is an ongoing documentation research project on the Nivaĉle language: “Documentation and Comparative Lexicon and Morphosyntax of Nivaĉle and Pilagá, of Northern Argentina”, National Science Foundation Research Grant BCS-DEL 1263817 (PIs: Alejandra Vidal & Doris Payne).   13 yita’ lhavos, another novel observation is that /k ͡l/ +/ʔ/ is realized as k’ in the speech of yit’a lhavos speakers. However, the metathesis data which I documentated was entirely consistent with the data documented by Stell (cf. Chapter 6).  1.4 Methodology The major source of data for this study comes from my own fieldwork with both female and male native speakers of Nivaĉle (mostly shichaam lhavos who migrated to Uj’e Lhavos, and yita’ lhavos speakers). Complementary sources of data are Stell’s (1989) doctoral thesis, and Seelwische’s (1990) dictionary; whenever an example is taken from either of these two sources, an appropriate citation will be used.   My fieldwork was primarily conducted in two Nivaĉle communities of the Paraguayan Chaco: Uj’e Lhavos “people that live in the big (place)” and Santa Teresita.   Uj’e Lhavos is a semi-urban community located less than 1 km West of Filadelfia (Fernheim Colony), the capital of the Boquerón Department.12 According to the last Indigenous Census (DGEEC 2010), there are 1,772 Nivaĉle living in this community. Interestingly, 97% of the interviewed people use Nivaĉle, 2% Spanish, and only 0,4 % speak Guaraní. As previously mentioned, and despite being settled in a semi-urban environment, the Nivaĉle still, and pervasively, speak their language. The majority of the Nivaĉle people residing in Uj’e Lhavos come from Misión San José de Esteros and Misión San Leonardo (Fischat), and they recognize themselves as shichaam lhavos (‘Lowlanders’). As discussed in §1.2.2, the Nivaĉle started migrating to the Mennonite colonies in the middle of the 20th century in search for agricultural work. Nowadays, the economy of the Nivaĉle, and other indigenous people, depends on Mennonite sources of employment. Most of the Nivaĉle men work at Mennonite                                                 12 Despite its multiculturality, there is not much interaction between the different cultures in Filadelfia. This situation can be seen in the space distribution: each culture is placed in a separate area of the town. There are three indigenous communities: Guaraní (Yvopey Renda), Nivaĉle (Uj’e Lhavos) and Enlhet (Macheto), situated in non-adjacent outskirts of Filadelfia. Only the Nivaĉle people have ownership deed (since 2005). Also, there are a number of Ayoreo families that dwell at the entrance of the town. The center and north of the town is occupied by Mennonite families, who are members of the Mennonite Colony, and on the east side of the town, the “latinos” neighbourhood. The “latinos” is a term used by the Mennonites to refer to the Paraguayan and the Brazilians, who started migrating to Filadelfia twenty years ago. There are thus eight languages spoken: Western Guaraní, Nivaĉle, Enlhet, Ayoreo, Plattdeutsch (Low German), Spanish, Paraguayan Guaraní and Portuguese. With a total of approximately 16,000 inhabitants, only 37% are non-indigenous peoples.    14 farms and industries, and the young and middle-aged nivacche “women” work full-time or part-time as maids in Mennonite households.13   I visited Uj’e Lhavos for several weeks from 2009 to 2013, for a total of seven months. In July 2009, I introduced my work to the leader of the Uj’e Lhavos community, Paulino Chávez, and asked for permission to work in the community. I met several families and started working with Félix Ramírez Flores (FR, 67 years old, shichaam lhavos), and Sara Rojas Núñez (SR, 29 years old, shichaam lhavos). FR has been my main consultant for this study; he is now a retired Nivaĉle teacher who collaborated with Priest Seelwische in the elaboration of the Nivaĉle dictionary and pedagogical materials. FR has also been one of the advocates for the organization of the Nivaĉle Linguistic Conferences and the CLPN.   In subsequent fieldtrips (2010-2013) I worked with the following speakers: Andrés Crespo (AC), Faustino Ramírez (FAR), Rosalina Rojas (RR), José Rojas (JR), José González (JG), Francisco Fleitas (FF), Celestina Céspedes (CC), Teresita Sánchez (TS), Graciano Ramírez (GR), Agustín Juárez (AJ), Raquel Fleitas González (RF), Myriam González (MG), Celestina Sánchez (CS), Mauricio Valdéz (MV) and Sofía (SD).14   During my stay in 2010, I was invited to collaborate with the organization of the Second Linguistic Nivaĉle Conference, which took place in Uj’e Lhavos at the end of that year, and in 2011, I was invited to form part of the CLPN constituted during that conference (December 2010). My main role in the group was to assist the CLPN in the discussion of the Nivaĉle sound system and the orthography revisions. As mentioned in §1.2.1, one of the goals of the CLPN was to consolidate the orthographies that were being used in the Catholic and Mennonite schools.                                                  13 Note that “Nivaĉle” refers to the Nivaĉle people, but also to Nivaĉle men. “Nivacche” refers only to women that belong to the Nivaĉle people.   14 Here I present information about the (approximate) age of the consultants at the time of the recording (for some of the consultants) and the (self assessed) dialectal variety:  FR: 67, shichaam lhavos, AC: 65, shichaam lhavos; FAR: 66 shichaam lhavos; RR: 70 shichaam lhavos; JR: 70, shichaam lhavos; JG: ~ 85, yita’ lhavos; FF: ~70 yita’ lhavos; CC: 64, shichaam lhavos, TS: 49, shichaam lhavos, AJ: 47, shichaam lhavos; SR: 29 shichaam lhavos, MG: 23, yita’ lhavos; RF: 30, yita’ lhavos, CS: 28, tavashay lhavos, MV: 24, shichaam lhavos, SD; 20. Note that I received permission to reveal all this information.  15  Between February and June 2012, I conducted fieldwork in Uj’e Lhavos and the rural community of Santa Teresita. Besides collecting data, I participated in bimonthly meetings with the Nivaĉle Linguistic Team and helped in the organization of the first training workshop for Nivaĉle teachers (May 17-19, 2012, Uj’e Lhavos). In June 2013, I returned to the field to work on the transcriptions of various texts with the help of Teresita Sánchez and Elizabeth Rojas, and to double-check previously collected data.     The methodology followed during fieldwork consisted of: (i) semi-structured interviews (in some sessions one of the Nivaĉle speakers would make the interview or ask for a specific vocabulary item or construction) (ii) text collection of different genres (myths, personal narratives, prayers, and speeches, description-explanations of traditional cultural practices) (iii) translations from/to Spanish (iv) description of photographs and storyboards (v) elicitation of target words in different contexts (word lists and paradigmatically related sets of data). This was the most frequently used methodology for the elicitation of targeted phonological phenomena. The general methodology used for phonetic data collection would consist of asking the speaker to repeat the target word five times and then to give an example of that target word in a sentence.   Audio recordings of elicitation sessions with consultants were done with a digital recorder (Zoom H4N), an AT803 Omnidirectional Condenser Lavalier Microphone, a Countryman lapel microphone (phantom power), and a Superlux E-523 stereo microphone (for recording of conversations). For the most critical phonetic recordings, I was able to work in a quiet small room in the school of Uj’e Lhavos, outside class hours, and in a recording studio in Filadelfia. The editing of the recordings, the data segmentation, and acoustic analysis were done in Praat for Mac (Boersma & Weenink 2014). Statistical analyses were done in R (R Core Team 2013).   In this thesis, certain aspects of the phonological analysis will be presented within the approach of Optimality Theory (OT, Prince & Smolensky 1993/2004, McCarthy 2002, 2007).  16 1.5 Representation of data and examples This section provides a brief overview of the conventions used in this thesis for representing linguistic data. Section 1.5.1 presents the basic correspondences between the International Phonetic Alphabet (IPA), the orthography used in Seelwische’s (1990) dictionary and the modifications introduced by the CLPN. Section 1.5.2 explains the presentation of linguistic examples.  1.5.1 Transcriptions and alphabets The default level of transcriptions will be broad and phonemic. When necessary, a distinction between a broad phonemic transcription or and input/underlying representation / / and a narrow phonetic transcription or an output/surface representation [ ] will be made. For clarity purposes, primary stress will be represented with an acute accent and secondary stress with a grave accent. In the body of the text and in certain examples, the orthographic transcription will be used. However, given the general goal of this thesis, the majority of the examples will only have a phonemic transcription.   Table 1.1 below presents the basic correspondences between the orthography proposed by Seelwische, the International Phonetic Alphabet (IPA), and the orthography proposed by the CLPN. Note that the orthography developed by Seelwische is influenced by both the Spanish and English orthographies; e.g., the use of “qu” before /i e/,  “sh”= /ʃ/, respectively. The CLPN did not introduce many modifications to Seelwische’s orthrography. Basically, the letters ĉ and ĵ are introduced as variants of c and j to represent uvular articulations. Importantly, the CLPN decided to change cl to ĉl in order to differentiate it from the Spanish consonant cluster [kl] (§5.3.2). Whereas the representation of the glide in onset position remains the same, that is, as y, consensus about the glide in syllable final position has not been reached yet; it is either represented as y or as the vowel i.       17 Table 1.1 Correspondences between orthographies and the IPA Seelwische’s orthography IPA Orthography ELN p [p] p t [t] t c + [a ɑ o u] [k]  c  [q] ĉ qu (+ [e i]) [k] c p’ [p’] p’ t’ [t’] t’ c’ (+ [a ɑ o u]) [k’] c’  [q’] ĉ’ quʼ (+ [e i]) [k’] c’ ch [t ͡ʃ] ch ts [t ͡s] ts chʼ [t ͡ʃʼ] chʼ tsʼ [t ͡sʼ] tsʼ cl [k ͡l]  ~ [q ͡l] ĉl f [f] f s [s] s sh [ʃ] sh lh [ɬ] lh m [m] m n [n] n j [x] j  [χ] ĵ w [w] ~ [β] v y/i  [j] y/i ’ [ʔ] ’ i [i]  ~ [ɪ] i e [e] ~ [ɛ] e a [a]  ~ [ə] a ô [ɑ]  ô o [o]  ~ [ɔ] o u [u]  ~ [ʊ] u VV [V ̰] ~ [Vʔv ̰] VV  18  1.5.2 Conventions used in examples  To a large extent, the linguistic examples in this thesis follow the Leipzig Glossing Rules (Comrie et al. 2008). Most textual examples consist of three lines. The first line is the phonemic broad transcription. When relevant, periods (.) are used to represent syllable boundaries; hyphens (-) are used to represent morpheme boundaries, i.e., between prefixes, suffixes, and their bases. The equal sign (=) is used to represent clitic boundaries. The second line of each example is aligned word by word with the second line and provides morpheme-gloss of the material in the first line. Spanish words are given in italics. See the list of abbreviations (p. xi) for values of grammatical abbreviations used in glosses. The third line of an example gives an English translation of the example, in single quotes. As mentioned in §1.5.1, for certain examples, a very first line with the phonetic transcription, and a line with the Nivaĉle orthography (between the phonemic transcription and the morpheme-gloss lines) are included. The source of each example is from my fieldwork; otherwise the source is noted next to or beneath the translation line.  1.6 Structure of the dissertation   In this chapter, I provided an overview of the Nivaĉle language and its speakers, and I presented the methodology used in this research.    Chapter 2 presents an overview of the Nivaĉle phonological system building on Stell (1989) with proposed modifications based on my own fieldwork. This overview includes the Nivaĉle consonant and vowel inventories, the phonotactic constraints in the language such as syllable structure and consonant clusters, and an overview of phonological processes, such as palatalization, epenthesis, vowel harmony, and deglottalization of ejectives and glottalized vowels, which will be investigated in detail in subsequent chapters.   Chapter 3 examines the featural specification and prosodic representation of the Nivaĉle glottal stop and the glottalized vowels. It is proposed that the glottal stop is a moraic root node specified for [c.g.] and that it is unspecified for PLACE. One of the most important outcomes of this chapter is the  19 proposal that Nivaĉle glottalized vowels are not contrastive phonemes, contra Stell (1989). I propose a prosodic representation of these vowels as /Vʔ/ sequences where the glottal stop is hosted by a mora. This mora is parsed to the Nucleus of the syllable, and depending on prosodic context rearticulated/creaky vowels and vowel-glottal coda result. Importantly, Nivaĉle glottalized vowels deglottalize when the [c.g.] feature is not parsed to the head foot.   Chapter 4 examines the internal structure of the Prosodic Word and stress assignment in Nivaĉle in both the nominal and verbal domains. It is proposed that the Minimal Word is a closed monosyllable: CVC, and that glottalized vowels bear weight, that is, they only surface under stress. Further, it is posited that Nivaĉle has a quantity sensitive stress system, and that the rhythmic type is iambic. The following foot types are thus attested LL, H, and LH. The role of prefixes and suffixes with regards to stress placement is discussed, as there are L-edge and R-edge stress generalizations that apply; e.g. there is a different pattern found in alienable and inalienable nouns. A major generalization at the uppermost prosodic level is that primary stress falls on the rightmost foot of the PrWd.  Chapter 5 advances an analysis of the Nivaĉle lateral system, composed of two lateral obstruents: the lateral fricative /ɬ/ and the complex segment /k ͡l/. First,  it is proposed that /k ͡l/, which is the diachronic result of Proto-Mataguayan *l, is a complex segment, specified for DORSAL and [lateral]. Regarding the phonetic explanations behind said sound change, I hypothesize that the lateral approximant was realized with a brief stop closure which was misinterpreted as a real stop burst and reanalyzed as a laterally released stop. Further, the development of *l into [k ͡l] and not into [t ͡l] can be explained by the perceptually ambiguous nature of laterals in consonant clusters; it has been shown that the lateral release has a substantial effect on the acoustics of coronal stops, shifting them acoustically closer to velars (Kawasaki 1982, Flemming 2007, Hallé, Best & Bachrach 2003). Chapter 6 provides an Optimality Theory account of syllable-sensitive processes in Nivaĉle such as of vowel-consonant metathesis and vowel epenthesis. My major claim in this chapter is that Nivaĉle metathesis is driven by syllable requirements: (a) the avoidance of complex codas, and (b) the  20 satisfaction of the Syllable Contact Law (Hooper 1976, Murray & Vennemann 1983, Vennemann 1988, Gouskova 2004). Vowel epenthesis occurs when VC-metathesis would yield illicit consonant clusters. I also compare my proposal with the historical vowel deletion account presented by Campbell & Grondona (2007), and discuss alternative analyses such as pseudometathesis (Blevins & Garrett 2004).  Chapter 7 concludes with the major outcomes of this thesis, and establishes the topics for future research.   21 Chapter 2: Overview of the Nivaĉle phonological system  2.1 Introduction   This chapter provides an overview of the Nivaĉle phonological system and phonological processes, following and contrasting both Stell’s (1989) thesis and my own fieldwork data. The goal of this chapter, then, is to present the Nivaĉle segmental inventory (§2.2), the phonotactic patterns in the language (§2.3, 2.4), and various phonological processes (§2.5) that will serve as a background for the rest of the thesis.  2.2 Phonological inventory 2.2.1 Consonants The phonemic inventory of consonants is presented in Table 2.1. Of special interest for this study are, on the one hand, the series of ejective stops and affricates and the presence of a glottal stop (cf. Chapter 3), and, on the other hand, the status of a complex segment /k ͡l/, which I argue  is specified for DORSAL place of articulation (cf. Chapter 5). Note that segments in square brackets represent allophonic variants of the segments to their left, the variation being indicated by the ~ symbol.  22 Table 2.1 Nivaĉle consonants  labial dent-alv. palato-alv. palatal velar uvular glottal p t     k        ~   [q]  plain ejective p’ t’         k’       ~   [q’]    ʔ  stop        laterally released      k ͡l       ~   [q ͡l]  nasal m n     affricate plain ejective    t ͡s t ͡s’ tʃ͡   tʃ͡’     fricative  f  ~ [ɸ] s       ʃ      ɬ                x        ~   [χ]   ~   [h] approximants w ~ [β]   j     Similar to other Mataguayan languages (Chorote, Maká and Wichí), Nivaĉle has a two-way laryngeal distinction in stops and affricates (plain vs. ejectives) and no voicing contrast (only voiceless) within this class (§1.2.3). Unlike Wichí (Nercesian 2014a), aspiration is not a contrastive feature in stops, affricates and sonorants; and unlike Maká (Gerzenstein 1994), Wichí (Nercesian 2014a), and Chorote (Carol 2014), [h] does not have a clear phonemic status. A remarkable contrast with Chorote, Maká and Wichí is that Nivaĉle has a palato-alveolar fricative /ʃ/ (§ 2.5.1).   With regard to the laterals, there is an interesting cross-linguistic difference within this class across the family. While the other Mataguayan languages have a lateral fricative /ɬ/ and a sonorant lateral /l/, Nivaĉle has the former but not the latter. By contrast, there is a complex segment /k ͡l/, which “has a simultaneous articulation and release of a dorsal and a lateral” (Stell 1989:58). Comparative data show that Nivaĉle /k ͡l/ corresponds to /l/ in the other Mataguayan languages. The analysis of the development of *l into [k ͡l] will be discussed in §5.7.   The glides [w] and [j] pattern with consonants; they can occupy onset and coda position. They can precede and follow all of the six Nivaĉle vowels (§2.2.2) but *VwC and *VjC are unattested tautosyllabic sequences in the language. I conclude therefore that /w/ and /j/ do not function as off-glides in a diphthongal relationship with the preceding vowel nucleus. Rather, they function as a  23 consonantal “coda”, fulfilling the final C position of the Nivaĉle CVC Minimal Foot constraint (§4.2). It is significant as well that there are (to my knowledge) no vowel-glide alternations in the language: the glides /j/ and /w/ do not vocalize. The hypothesis that /j/ and /w/ are fundamentally consonantal receives further support from vowel epenthesis behaviour. Specifically, given an inviolable constraint against complex codas in Nivaĉle (§2.3.1), it is noteworthy that glide-final stems pattern with C-final stems (§2.3.1) in triggering V-epenthesis if a single C suffix is added. A further question is whether the approximants /j/ and /w/ pattern with the laryngeal /ʔ/. On the one hand, all three can be parsed into the final consonant (“coda”) position of a syllable. On the other hand, however, /ʔ/ has a unique relationship with a prosodically prominent (stressed) nucleus in that, under conditions outlined in §3.4, the [c.g.] feature of /ʔ/ can be incorporated into the nucleus. 2.2.1.1 Consonant distribution   Instances of phonemic contrasts for onset and, in some cases, coda position, are given through the illustrative minimal and near-minimal pairs in (1)-(14): Stops, affricates, and /ʔ /: Contrasts in PLACE and in Plain vs. Ejective [c.g.]  (1)   a. /paɬa/         /p/ ≠ /p’/    ‘in a little while’   b. /p’aɬaʔ/     ‘myth’   c. /napuʔ/    ‘two’    d. /nap’u/    ‘to lick’  (2)   a.  /ti/           /t/ ≠ /t’/    ‘that’   b. /t’iʔ/    ‘broth’   c. /jitex/    ‘carob’    24   d. /ji-tʼeʃ/    3S-say    ‘s/he says’  (3)   a.  /ɬa-kuʔ/          /k/ ≠ /k’/    3POSS-cheek    ‘his/her cheek’   b. /ɬa-k’uʔ/    3POSS-weapon     ‘his/her weapon’   c. /kus/    ‘hot’   d.  /k’us/    ‘happy’  (4)   a.  /amʔɑ/             /ʔ/ ≠ /p/    ‘rat’   b. /ampa/    ‘nothing’  (5)   a.  /t ͡sut/          /t ͡s/ ≠ /t ͡s’/    ‘lucky you!’   b. /t ͡s’ut/    ‘tasteless’         Stell (1989:96)  (6)   a.  /ɑjt ͡ʃe/          /t ͡ʃ/ ≠ /t ͡ʃʼ/    ‘accompany (you)!    b.  /ɑjt ͡ʃʼe/    ‘feel it (you)’!         Stell (1989:96)   Fricatives: Contrasts in PLACE  (7)   a.  /k’uʦ.fa-s/         /f/ ≠ /x/    friend-PL    ‘friends’    25   b. /k’ut ͡s.xas/    elder-PL    ‘elders’  (8)      a. /xa/           /x/ ≠ /ɬ/    DET15   b. /ɬa/    F.DET  (9)   a. /a-sas/          /s/ ≠ /ʃ/    2S-dirty.bad.ugly    ‘you are dirty’   b. /a-ʃaʔ/     2POSS-price.salary     ‘your salary’  (10)   a.  x-an          /x/ ≠ /ʃ/     1S-put    ‘I put’   b. xan    REL.PR    ‘the one that’   c. ∅-ʃan    3s-silence/quiet    ‘it is quiet’  Laterals (11)   a.  /xa-ɬɑn/          /k ͡l/ ≠ /ɬ/    1S-light    ‘I light’   b. /xa-k ͡lɑn/    1S-kill    ‘I kill’                                                   15 The Nivaĉle determiner system consists of four morphemes: na, xa, ka, and pa. In Gutiérrez (2011), I propose that Nivaĉle determiners encode both (i) evidential and (ii) deictic information. For ease of exposition, though, I only use “DET” in the gloss of the examples throughout this thesis.   26 (12)    a.  /tkamk ͡lɑj/         /k ͡l/ ≠ /k/    ‘s/he makes s.o. suffer’   b.  /tkamkɑj/    ‘s/he makes flour’  Nasals (13)    a. /jim/          /m/ ≠/ n/    ‘empty’   b. /jin/    ‘s/he paints’   c.  /namat ͡ʃ/     ‘axe’   d. /xa-nat ͡ʃ/    1S-come     ‘I come from’        Stell (1989:97)    Glides (14)    a. /ji-waʔʃ/         /w/≠ /j/    1POSS-den    ‘den’   b. /ji-jaʔʃ/    1POSS-quality       ‘my quality’        Stell (1989:97)   A central component of my thesis relies on the analysis of phonological processes that involve a constricted glottis feature [c.g.]. Let us, therefore, consider the realization of the series of ejective obstruents in Nivaĉle.  According to the documented cross-linguistic surveys cited, ejective sounds are found in 18% of the world’s languages (Ladefoged & Maddieson (1996:78), and in approximately 12% of South American Indigenous languages (González 2003). Specifically, as previously mentioned, the contrast between plain and ejective non-continuant obstruents is a characteristic feature of Mataguayan languages.  Ejective sounds are produced with an approximately simultaneous closure at the oral cavity and at the glottis (Ladefoged & Johnson 2011:137). The larynx is raised during both closures and increases  27 the air pressure trapped in the vocal tract area, between the two closures. When the oral closure is released, the compressed air is generally expelled in a way that causes the release burst to have high amplitude, auditorily associated with a ‘popping’ sound not found in plain stops. The glottis remains closed at the point of oral release, and opens some time after. The relative timing between the oral and glottal releases has motivated one of the distinctions between strong and weak ejectives (Lindau 1984, Kingston 1985, Bird 2002, Wright et al. 2002). For example, Kingston (1985) proposes that “strong” or “fortis” ejectives have longer VOT. Yet this parameter can be subject to a place-based constraint; a dorsal ejective [k’] tends to be stronger than labial or coronal ejectives (Bird 2002, Hajek & Stevens 2005). An initial (non-statistically validated) inspection of spectrograms of recordings, shows that in the velar ejective there is a longer delay between the release of the glottal closure and the onset of the vowel voicing than in labial and coronal ejectives.  Figures 2.1-2.2, 2.3-2.4, 2.5-2.6, show waveforms and spectrograms of the contrast between plain and ejective stops, in minimal and near minimal pairs with bilabial, alveolar and velar place of articulation, respectively. Note the stronger burst of the ejectives in comparison to the burst of the voiceless stops, and laryngealization of the vowel preceding the ejective stops, visible as irregular spacing between the glottal pulses (see Figures 2.2, 2.4 and 2.6).     28   Figure 2.1 Waveform and spectrogram of [napuʔ́] ‘two’ by male speaker FR Figure 2.2 Waveform and spectrogram of [nap’u ́] ‘myth’ by male speaker FR       Figure 2.3 Waveform and spectrogram of [tí] ‘that’ by male speaker FR Figure 2.4 Waveform and spectrogram of [tʼíʔ] ‘broth’ by male speaker FR  29    Figure 2.5 Waveform and spectrogram of [ɬakúʔ] ‘his cheek’ by female speaker TS Figure 2.6 Waveform and spectrogram of [ɬakʼú] ‘his weapon’by female speaker TS   The morphological concatenation of glottal-initial suffixes, such as the imperfective [-ʔin] (16a) causes a preceding stop or affricate to emerge as an ejective (16b), (§3.3.2.2). A comparative acoustic study between underived (15) and derived (16) ejective stops and affricates goes beyond the scope of this dissertation yet, an initial (non-statistically validated) inspection of spectrograms reveals an inconsequential minimal difference between these two sounds.   (15)      tik’ín    ‘small’  (16)   a.  x-ɑk-ʔín    1S-go-IMPFV    ‘I am leaving’   b.  xɑk’ín   30   Figure 2.7  Waveform and spectrogram of [tik’ín] ‘small’ by male speaker FR Figure 2.8 Waveform and spectrogram of  /xɑkʔin/: [xɑkʼín] ‘I am leaving’ by male speaker FR   2.2.2 Vowels Stell (1989:97) posits a phonemic distinction between plain vowels /i e a ɑ o u/ and  “glottalized” vowels /ỉ ẻ ả ɑ ̉ỏ ủ /. Table 2 below presents the vowel inventory based on Stell’s thesis. Square brackets enclose allophonic variants of the vowels. Note that [ɪ] is a variant of /i/ and /e/, and [ə] is a variant of /a/.  Table 2.2 Nivaĉle vowels (adapted from Stell 1989: 57)    front central back  plain glot. plain                  glot.  plain glot. high      i      [ɪ] iʔ  u               uʔ mid e    eʔ       o                oʔ  low               [ə]  a                        a’  ɑ   ɑʔ  31 The status of glottalized vowels will be discussed at length in Chapters 3 and 4. However, two observations should be made at this point. First, voice quality, i.e. laryngealization or creakiness is not reported as a contrastive value in the vowel systems of other Mataguayan languages. In this sense, Nivaĉle would constitute an exception or an innovation in this language family. Second, note that vowel length is not proposed to be contrastive either in Stell’s analysis or in mine.16 Importantly, even though I posit that vowel length is not contrastive synchronically in Nivaĉle, I report durational differences between modal and glottalized vowels that are statistically significant (§3.2.3). Glottalized vowels are roughly double the duration of modal vowels. In this context, non-modal phonation has been documented cross-linguistically as being associated with phonemic long vowels but not their short counterparts – e.g., in Hupa (Golla 1977, Gordon 1998:101, Gordon & Ladefoged 2001:18). Stell (1989:84) correlates cases of phonetic vowel length on modal vowels to speech styles (careful and/or emphatic speech). Based on my fieldwork data, I am in accord with this generalization. Vowel length is, for instance, profusely used in storytelling to indicate emphasis, or in everyday conversations to express one’s emotions. Recall that I represent primary stress with an acute accent and secondary stress with a grave accent (§1.5.2). Note also that the vowel /e/ usually gets realized more open, i.e., [ɛ] in closed syllables:  (17)      [pa    músika    sit ͡séx ʔəpɛ́ːʔ]  /pa=musika         sit ͡séx  apeʔ́/  DET=music   loud   too  ‘the music is too loud!’  (18)   [t ͡ʃa ́ːχpi    ɬu ̀t ͡sχəjítʃ͡  ʔɪsíːs]  tʃ͡a   xapi  ɬut ͡sxa-jitʃ͡    is-is  EXCL  DET-PL  young.woman-PL pretty-PL  ‘wow, the girls are (very) pretty!’                                                  16 Recall, however, that Hunt (1924) posited a distinction between long and short vowels (§1.3).  32  Another phenomenon worth mentioning in the vowel inventory system is vowel reduction. It has been observed cross-linguistically that, for some languages,  the difference in articulation of stressed vs. unstressed vowels is also reflected in the tendency of unstressed vowels to become more centralized (Gussenhoven 2004). In (18) it can be observed that a front low vowel /a/ is reduced to [ə] in an unstressed position, and that it is deleted in the determiner.  In fact, in Nivaĉle, a front unstressed vowel /i/ or /e/ is often reduced to [ɪ] (19, 20b), a low central unstressed vowel /a/ (21) is often reduced to a mid vowel [ə] or [ʌ], and a high back unstressed vowel /u/ is sometimes produced as a near-close near-back vowel [ʊ] (22). As will be discussed in subsequent chapters (§3.3.4, §6.6.2), unstressed vowels can also be the target for translaryngeal vowel harmony and deletion processes:   (19)  [tɪk’ín]    /tik’in/   ‘small’  (20)   a. sisé       ‘cane’   b. [sisɪt ͡ʃát]    /sise-t ͡ʃat/    cane-COL    ‘cane fieldʼ  (21)  [xiβe ́ʔk ͡lə]   /xiβéʔk ͡la/   ‘moon’  (22)  [-nʊt ͡sa ̰x]   /-nut ͡sa ̰́x/    ‘angry’    The miminal pairs in (23)-(27) illustrate some of the phonemic contrasts found between the Nivaĉle vowels.    33 (23)   a.  /is/              /i/ ≠ /u/     ‘nice’   b.  /u-s/     big-PL  (24)   a. /in/       /i/ ≠  /e/      ‘paint! (IMP)’   b. /en/      ‘love! (IMP)’  (25)   a. [naβa]       /a/ ≠ /ɑ/        /na-wa/     DET-PL       ‘the’  b. [naβɑ]   /nawɑ/   ‘pollen’   c.  /xa-tan/   1S-unhappy   ‘I am unhappy’  d. /xa-tɑn/   1S-chew   ‘I chew’  e.  /win.tax/   ‘treeʼ  f.  /win.tɑx/   ‘brown snail’  (26)   a. /wat-sɑʔ/      /o/ ≠ /ɑ/ INDEF.POSS-scabbies ‘someone’s scabbies’  b.  /wat-soʔ/ INDEF.POSS-penis ‘someone’s penis’  (27)   a.  /kak ͡lek /                      /e/ ≠ /o/ ≠ /ɑ        ‘heavy’   34   b.  /k’ak ͡lok/     ‘twisted’   c.  /kak ͡lɑk/     ‘quiet’  In the following figure, the vowel plot for a male (FR) and a female speaker (TS) are presented. Each of the six Nivaĉle contrastive vowels /i e a ɑ o u/ were recorded in the context of a preceding plain alveolar stop in a stressed syllable.   (28)   a. tí   ‘that’      b.   jitéx   ‘carob’      c. táta   ‘dad’   d.  itɑ́x    ‘fire’  e. to ́s   ‘snake’  f. tu ́ɬ   ‘night’  The midpoint of each vowel was measured in Praat using LPC analysis with a series of overlapping Gaussian 50 ms windows and a 25 ms step size. Formant values are given in Hertz.  35   Figure 2.9  Nivaĉle vowels charted in a two-dimensional unnormalized F1-F2 space, male speaker (FR) in blue, female speaker (TS) in pink  Given the formant values in Figure 2.9, the vowel [a] can be characterized as relatively central. However, this vowel patterns with front vowels in processes of palatalization, while /ɑ/ systematically patterns with back vowels (§2.5.1).   The vowel [ɑ] is the one that displays more variability both across dialects and in its diphthongal variants. In a few examples (29-32), the vowel /ɑ/ alternates with a diphthongal variant – [ɑɔ] or [ɒ] in fast speech. The vowel [ɑɔ] is represented in the Nivaĉle dictionary (Seelwische 1980) as a sequence of two vowels: aô. What Seelwische represents as aô corresponds to three different realizations. First, as shown in the data of (29) below, what Seelwische represents as aô corresponds to the presence of a glottal in between two heterosyllabic vowels. Second, in data like (30), it just  36 corresponds to the glottalized counterpart of /ɑ/ or /a/, analyzed in this work as the sequence /ɑʔ/ or /aʔ/ in a closed syllable:  (29)   [naʔɑ́]    /naʔɑ/  na-ô     (Seelwische 1990)  DET-DIST     ‘that’    (30)   a.  [jɑ̰́s] /jɑʔs/   y-aôs        (Seelwische 1990)   1POSS-son   ‘my son’   b.  [jika ̰́txok]     /ji-ka-tʼxok/     yi-caôtjoc   (Seelwische 1990) 1POSS-POSS.CLASS-uncle ‘my brother-in-law’   One of my main consultants, FR, who collaborated in the elaboration of Seelwische’s dictionary, explicitly mentioned that the vowel in ‘son’ does not have two different vowels; rather, it was his orthographic opinion that it should be written as yôôs, which correctly focuses on the salience of extended duration in the context of glottalized vowels.  Third, what Seelwische represents as aô represents the low back vowel /ɑ/ followed by a a glide /w/ (note that the [β] is the /w/, represented in the orthography as “v”). This can be variously realized as as [ɑɔ̯] or as a farther back low [ɒ].17                                                   17 A front vowel /a/ before /w/ does not undergo a comparable process.    37 (31)   [tɑɔ̯βk ͡láx] ~ [tɒβk ͡láx]  taôvĉlaj  ‘infantʼ   (32)    [ʔɑɔ̯βte ́χ]  aôvtej  ‘(it) hurts’   The low back vowel [ɑ] is subject to regional variation. In the yita’ lhavos ‘people of the sandy spot’ variety (a.k.a. c’utjaan lhavos ‘people of the thorny bushes’), there is no low back-unrounded vowel [ɑ]; [a] is ubiquitously found instead. According to RF, a Nivaĉle primary school teacher in Misión Santa Teresita (yita’ lhavos), the vowel [ɑ] is only produced when reading texts at school or during mass, otherwise the [a] has replaced the [ɑ] in everyday life (p.c).18  (33)   yita’ lhavos        shichaam lhavos  [xák]        [xɑ́k]   j-ôc   1S-go  ‘I go’  (34)    [taje ̰́x]        [tɑje ̰́x]  tôyeej  ‘shaman’  (35)   [a-ɬa ́n]         [aɬɑ́n]  a-lhôn    2IMP-light  ‘light(IMP)!’  (36)    [xak ͡la ̰́p]       [xak ͡lɑ̰́p]   ja-ĉlôôp   1S-to.have.on.lap  ‘I have (sb.) on my lap’                                                 18 MG, another female yita’ lhavos speaker, also confirms this observation; she states that [ɑ] does not exist in her speech except for certain words such as [ʔɑxɑk ͡lɑ] bird (where the dorsal consonants might be causing the [a] vowel to become [+back]).   38 (37)      [ina ̰́t]         [inɑ̰́t]   inôôt  ‘water’  (38)      [towa ́k]        [towɑ́k]  tovôc  ‘river’  (39)      [ɬa ́si]        [ɬɑ́se]   lhôs-e  3POSS-son-FEM  ‘his/her daughter’   Another feature of the yita’ lhavos variety is that in certain words, the front vowel [e] is pronounced as a high front vowel [i], as seen in (39) above and in some of the vowel correspondents in (40-42) below:19  (40)   yita’ lhavos       shichaam lhavos    [t ͡ʃʼitʃ͡ʼí]       [t ͡ʃʼetʃ͡ʼé]  ch’ech’e  ‘parrot’  (41)   [kek ͡lejt ͡ʃí]            [kek ͡lejt ͡ʃé]  queĉlei-che  ?-FEM  ‘bean’  (42)   [nìkxak-é]            [ne ̀kxɑké]   necjôqu-e   boy-F  ‘girl’                                                 19 Another case of vowel raising seems to occur with back vowels as well, though I have only documented the following example: (i)  kuʦxa ̰́t (yita’ lhavos)    koʦxa ̰́t (shichaam lhavos)   ‘land’   39 Whereas one could hypothesize that vowel raising in some forms is due to adjacency with the preceding alveopalatal affricate, e.g. in (40) and (41), it is not very clear why [e] would raise to [i] in (39) and (42). Interestingly, Hunt (1924:2) already noted an alternation between  [e] ~ [i] in the feminine marker: “it is not always easy to determine the final vowels e and i, e.g. the feminine termination as in somto-ke or somto-ki, ‘white woman’”. Note in that regard that the [e] ~ [i] in (39) is the feminine marker, though this same alternation is not registered in (42). It is worth mentioning that [i] can occur in final position in the shichaam lhavos variety, e.g. [mimi] ‘mother’; [k ͡limʃi] ‘flour’; thus, this is not a case of absolute neutralization.  In sum, given the issues that have been discussed so far, I propose that the contrastive and non-constrastive vowel inventory for Nivaĉle could be organized as follows: Table 2.3   Nivaĉle vowels: Synchronic Proposal Phonemes yita’ lhavos shichaam lhavos Variants /i/ /i/ /i/ [i] ~ [ɪ] /e/   /e/ ~ /i/ /e/ [e] ~ [ɛ] /a/   /a/ /a/ [a] ~ [ə] ~ [ʌ] /ɑ/   (/ɑ/>/a/) /ɑ/  [ɑ] ~ [ɑɔ̰] ~ [ɒ] /o/   /o/ /o/ [ɔ] /u/   /u/ /u/ [ʊ]   2.3 Nivaĉle syllable structure and phonotactics  The phonotactics of a language comprise restrictions on the permissible combinations of sound elements. Specifically, phonotactic constraints define the syllable structures, consonant clusters, and vowel sequences that are allowed in a language. This section presents an overview of the syllable structure in Nivaĉle in order to serve as the background for the discussion of phonotactic constraints and morpho-phonological processes.  2.3.1 Overview of syllable structure  The core syllable structure in Nivaĉle consists of the following structures: CV, CVC, CCV and CCVC.   40 Table 2.4 Nivaĉle core syllable structure and the role of ʔ Syllable  CV CVC CCV CCVC  CV CVC  CCV CCVC  ʔV CVʔ *CʔV *CʔVC   ʔVʔ *ʔCV *ʔCVC       CCVʔ  In the following subsections, I analyze the different constraints on syllable structure in this language. 2.3.1.1 Onsets Contrary to Stell (1989:116, 117), I claim that there are no onsetless syllables in the language neither word-initially, nor word-medially. That is, the constraints ONSET and *VV are undominated: (43)   ONSET  *[σV (Syllables must have onsets)  (Itô ̂ 1989, Prince & Smolensky 1993) (44)  *VV:  There cannot be two adjacent vowels.     All Nivaĉle consonants may appear as singleton onsets. Complex onsets are allowed in the language but only word initially. The following examples show alienable nominal roots, that is, roots that do not require the presence of an obligatory possessive prefix (45), and predicative verbs (46):  (45)   a.  txo ́p    ‘temperate’   b. ʃk ͡lɑkxa ́j ~  sk ͡lɑkxa ́j    ‘wild cat’   c. ʃnawɑ́p    ‘spring’   d. xpɑ̰́k    ‘straw’   e. pxuxu ́k    ‘cactus’   f. ft ͡su ̰́k    ‘palm tree’     41 (46)      a. fk’at ͡sáx    ‘wide’     b.  kxám    ‘just’  Note that examples with initial #pC, #fC, #xC are provided for completeness here, but in fact are extremely rare clusters. The pervasive generalization is that C1 of an initial CC cluster is COR, consistent with Morelli (1999, 2003). Ejectives /p’ t’ k’ t ͡s’ tʃ͡’/ cannot occur as the first member of a complex onset, but can occur as the second member, as seen in (46a), above. Further, given that CCC clusters are not allowed, the initial onset cluster [fk’] in (46a) provides evidence against treating ejective consonants as C+ʔ sequences.20  Table 2.5 below shows the CC co-occurrences in word initial position. Data was taken from Seelwische’s dictionary and my own fieldwork. White cells indicate attested CC sequences; when there are fewer than 4, I indicate the number of attested examples. Grey cells indicate unattested combinations. The plus (+) sign indicates that the sequence is only attested across a morpheme boundary. It can be seen that complex onsets are quite restricted and that coronals [t] and [ɬ] are the preferred C1. The preferred C2 is [x].                                                 20 The following examples provide additional evidence: [tt ͡ʃʼakfaj] ‘s/he is married (with children)” and [tt ͡sʼak ͡laj] ‘s/he is poor’.   42 Table 2.5 Initial CC clusters [Note: the +sign indicates that the sequence is only attested across morpheme boundaries]  stops ejectives affricates fricatives nasals glides          C2 C1 p t k ʔ pʼ tʼ kʼ t ͡s’ t ͡ʃ’ t ͡s t ͡ʃ k ͡l ɬ f s ʃ x m n j w p                       t +  +  + 1  + + 1 +  + 1 + 2   +   3    3   k                      stop ʔ  eject.  k ͡l  t ͡s    aff. t ͡ʃ  ɬ + + +  + + + + + + + +  + + + + + + + + f    1   3  1  s  3 2  1 2  1 1  ʃ    1  1    fricatives x   1  nasals   glides    43 Word-medial complex onsets are not allowed; inalienable CCVC nominal roots have the first consonant of the complex onset syllabified as the coda of the preceding syllable, which contains a possessive prefix:  (47)   ji-f.xúx      1POSS-toe       ‘my toe’  (48)   k’a-k.xúʔ  1A.2P-greet  ‘I greet you’  In Section 2.3.1.3, I will discuss the kind of consonants that can occur in coda position and CC medial clusters (cf. Table 2.6). 2.3.1.2 ʔ-epenthesis In the previous section, I claim that ONSET (43) is an undominated constraint in the language. Consider the following data:  (49)       a.  [ʔak ͡le]    a’.  [kat ͡sakle]  a’’. *[kat ͡sʔak ͡le] *[kasʔak ͡le]  *[kat ͡si-ʔakle] (§2.4)      /ak ͡le/     /kat ͡si-akle/    ‘scalp’     1POSS.PL-scalp          ‘our scalp’   b. [ʔɑ́ʔseɬ]    /ɑʔs-eɬ/    son-PR.PL    ‘your sonʼ  cf.    c.  [ʔɑ̀ʔseʔe ́ɬ]    /ɑʔs-e-eɬ/    son-F-PR.PL    ‘your daughter’    44 An epenthetic glottal stop is inserted in both word initial position and word medial position to ensure satisfaction of ONSET. A descriptive generalization that falls out from ONSET » DEP-IO-ʔ is that there cannot be two adjacent vowels:   (50)   DEP-IO-ʔ: An output glottal stop must have an input correspondent (‘No ʔ-epenthesis’). (51)   ONSET » DEP-IO-ʔ The epenthetic status of the glottal stop can be determined by comparing (49b) with (49c). The root in (49a) does not start with a glottal stop because the first person possessive plural form is not [kat ͡sʔak ͡le] or *[kasʔak ͡le], but rather, [kat ͡sakle] ‘our scalp’.  Interestingly, a glottal stop is not epenthesized in order to repair a vowel hiatus at the pre-root prefix domain: *[kat ͡si-ʔakle]. As will be discussed in §2.4, the final vowel of a prefix consistently deletes when added to a vowel-initial root. In contrast, in a post-root suffix domain, glottal stop epenthesis is enforced in order to not violate *VV (44). The epenthetic status of the glottal stop will be discussed in §3.3.3.1. 2.3.1.3 Codas  While closed syllables are a frequently found syllable structure type, complex codas are not allowed in this language, and so *CC]σ is an undominated constraint. In contrast, NOCODA has to be low ranked: (52)   *COMPLEXCODA   *CC]σ   ‘Codas are simple’               (Kager 1999)  (53)   NOCODA     *C]σ     ‘Syllables are open’     (Kager 1999)   All consonants may appear in coda position except for the ejectives /p’ t’ k’ t ͡s’ tʃ͡’/ which can neither serve as word-internal nor word-final codas. In that regard, several authors (Itô 1986, Itô & Mester 1994; Lombardi 1991, 1995) have pointed at the restriction against LARYNGEAL and PLACE jointly occurring in coda position. Similarly, in Nivaĉle, ejectives lose their laryngeal feature in coda  45 position (see §2.5.5 and §6.5.1). Steriade’s (1997) Licensing by Cue approach presents a perceptual motivation for this constraint: “an optimal identification of an ejective (…) will depend on the nature of the right hand context, i.e. on the presence of a vowel or a sonorant” (78). Because none of the ejective obstruents can occur as the first member of an initial consonant cluster in Nivaĉle, this fact can be interpreted as an argument for Steriade’s (1997) perceptual explanation of laryngeal neutralization: glottalized obstruents neutralize in the absence of a following sonorant, regardless of whether or not they are in the ‘same’ syllable.    From the affricate set, /t ͡s/ can only occur in onsets (54b): (54)   a. -fé.tas   ‘root’  b. -fe.ta.t ͡s-íj    root-PL   ‘roots’  The affricate /t ͡s/ can also occur in word-internal codas (55b), but only before [x] (cf. Table 2.6). There is only one example where /t ͡s/ can also occur before /f/, as seen in (7) [kut ͡sfas] ‘friends’:  (55)   a. -ɑ.fí ̰s   ‘reach’  b.  -ɑ.fit ͡s-xan   reach-INT   ‘to reachʼ       (Stell 1989:123)   (56)   a. xaj-kút 1S-steal ‘I steal’ b. ku ̀t ͡s-xanáx  steal-NMLZ   ‘thief’      46 The palato-alveolar affricate [t ͡ʃ] occurs in onset and word final coda position. In word medial coda position, it only occurs before [ʔ], [ʃ] and [x] initial suffixes. Recall that the morphological concatenation of glottal-initial suffixes causes a preceding stop or affricate to emerge as an ejective, as seen in (58a).  (57)   xokita ́tʃ͡  ‘lapacho (tree)ʼ  (58)   a. xa-wa ̀n-t’ɑfít ͡ʃ-ʔin    ~     xawa ̀nt’ɑfít ͡ʃʼin         1S-REF-hide-IMPFV    ‘I am hiding’   b.  xa-wa ̀n-t’ɑfí ̰t ͡ʃ-ʃi      na=jitáʔ    1S-REF-hide-IMPFV-LOC     DET=scrubland    ‘I hide (inside) the scrubland’   c. t’a-k ͡la ̰t ͡ʃ-xop    t’a-claach-jop                (Seelwische 1990)    3POSS-song-FOR    ‘(his/her clothing)for singing’            A rare and interesting alternation, discussed in detail in Chapter 5, pertains to the complex segment /k ͡l/, which consistently neutralizes to [k], when it is not preceded by a vowel (§5.3), but, interestingly, it is retained before a glottal stop (§3.3.4). This special behavior of /k ͡l/ suggests a perceptual, rather than a syllabic, explanation behind complex segment neutralization.  In sum, there are two types of complexity under consideration: one is sequential complexity (i.e. consonant clusters); the other is internal segmental complexity, e.g. affricates and glottalized consonants. As previously mentioned, whereas there are a number of complex onsets, complexity never occurs in coda position, the constraint *COMPLEXCODA is undominated.   A major set of distributional generalizations that plays an important role in my analysis is that where there is a word-internal coda, the following onset is always of equal or lesser sonority (§2.3.2 and §6.3.2); that is, obstruent(O)-resonant(R) sequences are not attested at the MStem1 domain: *O.R. I  47 consider the Morphological Stem 1 to consist of the root and derivational suffixes (cf. Chapter 4, §4.5). In contrast, OR sequences are attested between prefixes and roots or across a word-enclitic boundary. However, there does not seem to exist an internal sonority hierarchy among the obstruents. Table 2.6 below summarizes the CC medial clusters combinations. Similarly to Table 2.5, white cells mean attested, grey cells mean unattested, “1, 2, 3” indicates the number of examples. Further, for relevant cases, I indicate the specific morpheme in contact.  Light grey cells indicate neutralization; the outcome is indicated in the relevant cell.  48  Table 2.6  Medial CC clusters stops ejectives affricates fricatives nasals glides          C2 C1         p t k ʔ pʼ tʼ kʼ t ͡s’ t ͡ʃ’ t ͡s t ͡ʃ k ͡l ɬ f s ʃ x m n j w p            1 3   -ʃi  1      t   ɬ                 wat- wat- wat- wat- k                    = = stop ʔ                      ejective  k ͡l k k k  k k k k k k k k k k k k k   = = t ͡s 21             1        aff. t ͡ʃ                -ʃi    = = ɬ 1 3   2    t ͡ʃ’e       1      f         t ͡ʃ’e    1   -ʃi      s                  kas- kas- kas- kas- ʃ         t ͡ʃ’e    1   -ʃi    = = fricatives x         t ͡ʃ’e    3   -ʃi -xi   = = m         t ͡ʃ’e =tsex   3   -ʃi     = nasals n         t ͡ʃ’e             j                     2 glides w                   1     p t k ʔ pʼ tʼ kʼ t ͡s’ t ͡ʃ’ t ͡s t ͡ʃ k ͡l ɬ f s ʃ x m n j w                                                 21 /t ͡s/ simplifies to [s] and [t].  49 2.3.1.4 Nucleus So far, I have presented an analysis of the Nivaĉle internal syllable structure that accommodates the notions of Onset and Coda not as prime constituents, but rather as prosodic domain edges. Further, I assume an internal syllable structure that has a Nucleus as a constituent (Shaw 1992, 1994), specifically, as the Prosodic Head of the syllable. The mora (Hyman 1985, McCarthy & Prince 1986, Zec 1988, Hayes 1989), which serves as the “primitive subsyllabic constituent and as a measure of syllable weight” (Zec 1995:85), gets parsed to the Nucleus of a syllable and therefore plays a crucial role in the assignment of stress (cf. Chapter 4) and in the realization of [c.g.] (cf. Chapter 3).  In this dissertation, I argue that besides vowels, glottal stop is associated with a mora (§3.3.4), and this mora is parsed to the Nucleus of the syllable (59). All other (non-moraic) segments are parsed directly to the syllable (61). (59)   Parse-µ-Nuc (Parseµ): Moras are parsed into the Nucleus of a syllable. (60)   Parse-Nuc-σ: (ParseNuc): The Nucleus is parsed into a syllable. (61)   Parse-SEG-σ: Segments are parsed into syllables (exhaustive parsing). 2.3.2 Sonority  It has been traditionally assumed that the organization of segments within the syllable is driven by the sonority hierarchy, a ranking of segments from least to most sonorous. Several proposals have been put forward to define and formalize the concept of ‘sonority’ (Sievers 1876/1893, Jespersen 1904, Murray & Vennemann 1983, Vennemann 1988, Keating 1988, Clements 1990, Goldsmith 1990, Clements & Hume 1995, Gouskova 2004, Parker 2002, 2008, 2012, among others).  Sonority has been defined in terms of openness or aperture of the oral articulation (Kirchner 1988, Goldsmith 1990, Howe & Pulleyblank 2001), relative loudness of speech sounds (Sievers 1876/1893), or inherent or perceived loudness (Ladefoged 1975, Clements 1990, Laver 1994), and so correlated with acoustic notions such as energy (Ladefoged 1971, Keating 1988, Goldsmith 1990, Wright 2004) and segmental intensity levels (Parker 2002, 2008). Even if the concept of sonority has been questioned as a linguistic construct per se (Parker 2008: 56) or its existence has been denied  50 (Ohala 1990), this concept has had a crucial impact on distinctive feature theory (Chomsky & Halle 1968, Steriade 1982; Clements 1988, 1990, Clements & Hume 1995). Further, a number of cross-linguistic tendencies in the distribution and sequencing of segments are explained with reference to sonority hierarchies – though not without some problems (see Henke, Kaisse & Wright 2012, and Parker 2012 for useful discussion of this issue). When major phonological groupings of sounds are considered the sonority hierarchy with the five classes identified in (62) is generally assumed (Bell & Hooper 1978, Harris 1983, Clements 1990, Kenstowicz 1994, de Lacy 1997), although both more and less inclusive classes have been proposed for particular languages (Jespersen 1904, Goldsmith 1990, Gouskova 1999, de Lacy 2002).  Stops, fricatives, and affricates are the lowest ranked segments in regards to the sonority hierarchy whereas vowels are the most sonorous segments:  (62)   Sonority Hierarchy (SH): Vowels > Glides > Liquids > Nasals > Obstruents   An intimately related concept is the Sonority Sequencing Principle, which predicts that segments comprising the onset should rise in sonority until they reach the nucleus of the syllable (e.g. vowel), and that the nucleus should be more sonorous than the onset and coda.   (63)   Sonority Sequencing Principle (SSP): Sonority increases towards the syllable peak and  decreases towards the syllable margins (Clements 1990).    Let us take a look at consonant clusters in Nivaĉle. As seen in §2.3.1, at most two consonants can occur in word-initial position. For instance: (a) obstruent + obstruent: [kxam] ‘exact’, [pxuxuk] ‘type of cactus’, [fk’at ͡sax] ‘wide’, [ftʃ͡inax] ‘ray’, [ft ͡su ̰k] ‘palm tree’, [sxet ͡sitʃ͡] ‘owl’ (b) obstruent + sonorant: smitka ‘peanut’, [ʃnawap] ‘spring’.    Consonant clusters can also emerge through morpheme concatenation. Certain verbs surface with a word initial complex onset when the third person subject prefix /t-/ is attached:  51 (64)   t-ka-ʃaʔ́-j-a    3S-MED-price-VBLZ-PUNC   ‘she/he sells’   (65)   t-t ͡ʃ’ak.fá-j            3S-husband-VBLZ            ‘s/he is married’   Also, two reduplicative forms have CCVC shapes:   (66)   txux-txux-ʔin     ‘narrow’  (67)   tʃim-tʃim-tʃ͡ʼe  tshimtshimch’e            ‘(a road) full of potholes’              In these two cases, the first member of the consonant cluster is an alveolar stop, and the second member a fricative. Interestingly, the consonant cluster in (67) is not affricate. I specifically asked the consultants about the pronunciation of this word; they confirmed that there are two consonants, and that they would write it down as t +sh.  Table 2.7 and 2.8 below presents co-occurrence constraints on MANNER of CC-initial clusters. This compilation of attested Nivaĉle sequences is non-exhaustive. However, it established that the Nivaĉle data show a relative freedom of co-occurrence across MANNER categories with “equal” sonority within the “Obstruent” class of the general hierarchy in (62). Only affricate-stop and affricate-affricate sequences are not attested. However, in the context of morpheme concatenation only an alveolar stop can occur before another stop or an affricate; no CC-initial root composed of a STOP-STOP or STOP-AFFRICATE is ever attested. Fricatives are the least constrained members of CC-initial clusters. Of this set, the velar fricative [x] is the most frequent. Note that the shaded cells indicate non-licit or unattested consonant clusters. Affricates cannot occur as the first member of a CC-initial cluster except for [t ͡s] before [x].   52 Table 2.7 Co-occurrence constraints on MANNER: Onset CC Obstruents with equal sonority C1↓ C2→ STOP AFFRICATE FRICATIVE STOP t-ka-mkɑ-j 3S-?-flour.VLBZ t-t ͡ʃ’akfa-j 3S-husband-VBLZ ‘to be married’  kxatux  ‘cactus’  tʃaʔnu  (cf. 67) ‘rain’  pxotʃi ‘soft’ AFFRICATE    t ͡sxot’atax  ‘kingfisher’ FRICATIVE  ʃtɑtax mushroom  fk’at ͡sax ‘wide’  ftʃ͡inax  ‘ray’  ft ͡su ̰k ‘palm tree’  sxet ͡sitʃ͡ ‘owl’  Also note the total and systematic absence of Sonority Reversal sequences in Onset, namely, *Nasal/Glide-Obstruent.  In turn, Table 2.8 presents the co-occurrences constraints on MANNER with CC-initial clusters that involve a rise in sonority. These clusters are very rare. Only /t/ can occur as C1 and /n/ as C2. The only OBSTRUENT-m cluster is [smitka], and related forms. The only OBSTRUENT-GLIDE sequence is [swuk ͡lax].       53 Table 2.8 Co-occurrence constraints on MANNER: Onset CC with rising sonority C1↓ C2→ NASAL GLIDE STOP t-nijkɑ-j-xan  3S-thread-VBLZ-CAUS ‘s/he makes threads’ tnɑxke ‘vase’      AFFRICATE   FRICATIVE ʃnɑwɑp  ‘spring’ smitka ‘peanut’ swuk ͡lax ‘anteater’   Finally, Table 2.9 considers co-occurrences constraints on place of initial CC clusters. Except for *LABIAL-LABIAL, and *DORSAL-CORONAL, the specific place-place categorizations establish that there seems to be significant freedom of occurrence of place in both derived and non-derived OO clusters at the left edge of the word. Once again, however, the voiceless alveolar stop /t/ is the only CORONAL consonant that can occur before a LABIAL consonant.   Table 2.9 Co-occurrence constraints on PLACE C1↓ C2→ LAB COR DOR LAB  ft ͡sɑnax ‘suncho treeʼ pxuxuk  ‘cactus’ fxos ‘steep’ COR tpik ͡la ~  pik ͡la ‘fish’ t-fuj-an 3S-blow-CAUS ‘s/he blows’ tʃaʔnu ‘rain’ t-ʃajk’u-j 3S-lay.eggs-INT ‘it lays eggs’ t ͡sxap  ‘s/he tackles’ DOR xpɑ̰k   ‘straw’  kxatux  ‘type of cactus’  54  Let us now turn to the consideration of medial CC clusters formed by derivational suffixation on the end of a ROOT. Table 2.10, below, recapitulates the data presented in Table 2.6 and reorganizes the relevant information in terms of sonority (across MANNER). Note that “H” stands for homorganic and that exceptions to the pattern established in the relevant cell are indicated between parenthesis. Table 2.10 Co-occurrence constraints on MANNER OBSTRUENTS RESONANTS LATERALS NASALS GLIDES C1↓   C2→  ʔ STOP AFF EJ. FRIC k ͡l ɬ m n  j w ʔ            STOP  (*H)    (*kk ͡l)      AFF      (t ͡s.x)       EJEC            FRIC     (*H) (*xk ͡l)      k ͡l            OBSTRUENTS  ɬ      (ɬ.x)       m            NAS AA n            j           2 GLI w  (*wp)  (*wp’)     1    Several patterns are worth mentioning: First, the glottal stop can occur before all consonants except before another glottal stop or an ejective. Second, with considerable less freedom of occurrence are ejectives, affricates and the complex segment /k ͡l/, which are the most restricted as C1. From this set, there are two interesting cases. On the one hand, as previously mentioned (§2.3.1), the affricate [t ͡s] can only occur before [x]. On the other hand, [k ͡l] can also only occur as C1 before [ʔ] (§ 5.3.1), but neither [x], nor [k] can occur before [k ͡l].   Third, the situation within the resonant class is different: /m/ and /w/ show the most restricted distribution, and thus appear to be the most marked segments. This is not surprising given the lack of co-occurrence between labial obstruents in both medial and initial CC clusters. Nasals do not co-ccour,  55 and co-occurrences of glides are restricted. It is not very clear why /n/ can occur before the palatal glide but the reverse is not attested [*jn], or more specifically, when such contact happens the glide is deleted: In contrast with obstruents, though, there is a clear preference for glides to occur as C1.  In order to account for the Nivaĉle data, certain adjacent categories of the Sonority Hierarchy in (62) can be collapsed. I assume the sonority hierarchy in (68) (for further discussion see §6.3.2). Recall that there is no sonorant lateral or rhotic in Nivaĉle. Hence, the category of liquids is null, indicated informally by parentheses in the full version in (62) below:   (68) Vowels > Resonants > Obstruents       cf. (62)  Sonority Hierarchy (SH): Vowels > Glides > (Liquids) > Nasals > Obstruents As will be discussed in Chapter 6 (§6.3.2), when morpheme concatenation results in a ‘bad syllable contact’ – namely, when sonority rises across a syllable boundary – VC metathesis takes place. Because it is more optimal for syllable onsets to have lesser sonority, and syllable codas greater sonority; “a sonority reversal” is where a coda is less sonorous than the following onset (i.e., obstruent-resonant). When there is a sonority reversal between a coda and a following onset, metathesis functions as a repair process, reversing the sequence of the coda and the immediately preceding vowel. In other words, metathesis emerges as a repair process (69b).  (69)  a. tɑje ̰́x   ‘shaman’   b. tɑj.xe-met ͡ʃ       b’. *tɑjex̰metʃ͡  shaman-POWER.OVER  ‘shaman that has power over other shamans’  What these examples show is that metathesis functions to repair the “bad syllable contact” […ex.m…] by effectively relocating the low sonority coda segment, [x], from coda position and putting it into the onset position of its host syllable, thus optimizing the syllabic parse to […xe.m…]. Note that  56 vowel-consonant methathesis occurs with all vowels and a range of consonants. For a full account of metathesis see Chapter 6.  The notion of sonority also provides a window into the fact that different layers of affixes define different prosodic domains, these being motivated by different patterns of patterning.  Whereas the data in (69), along with many more examples presented in Chapter 6, illustrate that an obstruent-resonant “sonority reversal” sequence across a syllable boundary is actively and systematically avoided within the phonology of Nivaĉle, there are other contexts where such contact persists without being subject to metathesis (or any other repair strategy).   (70)   ɬ-te ́ʃ̰=ji  2S-say-1O  ‘you tell me’  In (70), it can be seen that the palatoalveolar fricative /ʃ/ in the coda precedes the palatal glide /j/ in the following onset (a bad syllable contact). What I hypothesize is that in cases where such “bad syllable contact” sequences persist, they are permitted to do so because there is a stronger prosodic boundary between them. Specifically, note in (70), that the first person object pronoun is a clitic, the prosodic domain difference being signified by the clitic boundary marker [=]. The generalization governing metathesis is that it operates within an inner prosodic domain, identifed as the MSt1, whereas it does not function in the outer domain of clitics embraced within the higher Prosodic Word domain.    The fact that there are a diversity of prosodically-sensitive phonological constraints – the Syllable Contact Law and metathesis (§6.3.2) – that all demonstrably apply within a well-defined morpho-prosodic domain (the Morphological Stem) that is not fully co-extensive with, but rather is internal to the Prosodic Word, constitutes a significant body of empirical evidence that the prosodic phonology needs ‘inside access’ to morphological domain structure in the sense of Shaw (2009).  57 2.4 Pronominal allomorphy  Here I will describe an interaction between the possessive personal pronouns and the syllable well-formedness constraints introduced in §2.3.1. This discussion will serve as the background for issues like stress, which will be the subject matter in Chapter 3.   In the nominal domain, the morphemes that can precede the lexical root are possessives and possessive classifiers. Like the other Mataguayan languages and several languages of the Chaco region, Nivaĉle distinguishes between inalienable (a.k.a. relational) and alienable (a.k.a. non-relational) nouns. Inalienable nouns denote entities that are inherently possessed, and thus obligatorily require the presence of possessives. In contrast, alienable nouns are not obligatorily prefixed by possessives. On the one hand, Nivaĉle inalienable nouns comprise: body parts, family relationships, clothing /accessories. On the other hand, alienable nouns comprise: objects, animals/plants, elements from nature (sun, tree, thunder, river), human beings (man, woman, girl, etc).   The possessive pronominal prefix paradigm is summarized in Table 2.11 below.22 When applicable, the possessive plural suffixes [-ʔeɬ] ~ [-eɬ] are included.                                                     22 Here and in the rest of the thesis, I will be representing the indefinite possessive prefix with an intial [β] instead of [w], its ‘basic’ form, as /w/ consistenly surfaces as [β] in this context. Also note that there exists another set of indefinite possessive prefixes whose occurrence is very restricted: [n- ~ na- ~ tin- ~ tn-], e.g. (i). However, they also serve to derive an inalienable noun and create a related alienable noun (ii), (iii): (i)  na-kfíj    ~ watá-kfij  INDEF.POSS-shoe   INDEF.POSS-shoe  ‘(someone’s) shoe’  ‘(someone’s) shoe’ (ii) tin-βa ̰́k  INDEF.POSS-intestine  ‘reed’ (iii) tin-ɑ́x  INDEF.POSS-skin  ‘leather’   58 Table 2.11 Nivaĉle pronominal possessive paradigm PERSON SINGULAR PLURAL 1INC. ji- ~  j- kat ͡si- ~ kat ͡sʼi-, ~ kat ͡s- ~ kat ͡s’-, kas-  1EXCL.                                   ji- ~  j- _____ -ʔeɬ ~ -eɬ 2 ʔa- ~ ʔ- ʔa- ____-ʔeɬ ~ -eɬ  3 ɬ- ~ ɬa- / tʼ- ~ tʼa-  ɬ- ~ ɬa- / tʼ-~ tʼa- ____   + ʦiβeʔ INDEF. wat- ~ wat’-, wata- ~ wat’a, ~ wa- wat- ~ wat’-, wata- ~ wat’a, ~ wa-____-ʔeɬ ~ -eɬ  Some patterns of alternation can be noticed. First, there is variation in the syllable shape of the prefixes: C ~ CV, CVC ~ CVCV. These alternating forms are motivated by syllable structure constraints; namely ONSET, *VV, and and *COMPLEXONSET (cf. §2.3.1). Whereas C-final prefixes attach to V-initial roots, CV- prefixes attach to C-initial and CC-initial roots.23 Similarly, CVCV-prefixes, i.e., [kat ͡si] (74a,b) and [βata] (79), attach to CC-initial roots in order to not violate COMPLEXONSET. Illustrative examples are given below.   (71)   First Person Singular: [ji-] ~ [j-]  a. ji-t’o ́x             1POSS-aunt       ‘my aunt’  b.  ji-k.fíj   1POSS-shoe   ‘my shoe’  c.  j-ɑ́k      1POSS-food      ‘my food’   (72)   Second Person Singular: [ʔa-] ~ [ʔ-]   a. ʔa-tí.niʃ   2POSS-necklace   ‘your necklace’                                                   23 The only C-prefix that attaches to a C-initial root is the third person possessive prefix [ɬ-], see example (73c) and discussion in pages 61-62.   59  b.  ʔa-k.tʼéʔ   2POSS-grandmother   ‘your grandmother’  c.  ʔ-éj   2POSS-name   ‘your name’  (73)   Third Person Singular: [ɬa-] ~ [ɬ-]   a. ɬa-k.te ̰́tʃ͡   3POSS-grandfather   ‘his/her grandfather’  b.  ɬ-ɑ.s-e    3POSS-son-F   ‘his/her daughter’  c.  ɬ-t’o ́x    3POSS-aunt   ‘his/her aunt’  (74)   First Person Inclusive Plural: [kat ͡si-] ~ [kas-] ~  [kat ͡s-]   a.  ka.t ͡sí-f.xux   1POSS.INCL.PL-toe   ‘our toe’  b.  ka.t ͡s’í-k.teʔ    1POSS.INCL.PL-grandmother   ‘our grandomother’ (75)   a.  kas-fín   1POSS.INCL.PL-sling   ‘our sling’  b.  kas-ka ̰́t   1POSS.INCL.PL-chin   ‘our chin’  (76)   ka.t ͡s-a ́.k ͡le       1POSS.INCL.PL-sebum  ‘our sebum’    60 (77)   First Person Exclusive Plural: [ji-] ~ [j-] NB: this prefix co-occurs with PR.PL suffix [-(ʔ)eɬ]  a.  ji-ko ̰̀ts.xat-[i]s-éɬ   1POSS-land-PL-PR.PL   ‘our (excl.) lands’  b.  ji-ka.tʃ͡i.k ͡lɑ-[ʔ]eɬ   1POSS-earring-PR.PL   ‘our (excl.) earrings’      (Stell 1989:185)  c.  j-ɑk-éɬ   1POSS-food-PR.PL   ‘our foods’  (78)   Second person plural: [ʔeɬ] ~ [-eɬ]   a. ʔa-tʃ͡a ̀k.fa-[ʔ]éɬ   2POSS-husband-PR.PL   ‘your (pl) husband’  b.  ʔ-a ̰̀p.ku.nɑ.k ͡l-e ́ɬ   2POSS-traditional.salad-PR.PL   ‘your traditional salad’  (79)   Indefinite Person: [βata] ~[βat-]  a.  βa.tá-f.tiɬ   INDEF.POSS-thread   ‘someone’s thread’  b.  βat-a ̰́j   INDEF.POSS-traditional.purse   ‘someone’s purse’  c.  βa ̀t-ka.t-[í]s   INDEF.POSS-chin-PL   ‘someone’s chins’  What is shown in the above data is that the final vowel of a prefix, whether that vowel is /i/ or /a/, systematically deletes when affixed to a vowel-initial root. Deletion of the initial vowel in an underlying V1V2 cluster (in accordance with generalizations noted by Casali 1997) functions, therefore, as a strategy to avoid *VV sequences and to optimize CV syllable structure.  What is interesting to note is that although *VV is a pervasive constraint in Nivaĉle, different strategies are invoked to resolve  61 violations of *VV in the pre-root prefix domain as opposed to the post-root suffix domain. In the pre-root domain, as we have seen here, *VV violations are resolved by deletion of V1. In contrast, a sequence of two adjacent vowels at the root-suffix boundary does not surface due to glottal stop epenthesis, as shown in (77b) and (78a). Interestingly, a parallel set of complementary repair mechanisms occurs in Campa languages (Arawakan, Peruvian Amazon): like Nivaĉle, V1 is deleted in prefixes, but after the root, rather than V-deletion, [t] insertion applies instead to resolve the vowel hiatus (Megan Crowhurst, p.c.). The alternation in the first person plural (inclusive) possessive [kat ͡si ~ kat ͡s ~ kas] deserves some comments. First, I posit that the underlying representation must have the palato-alveolar affricate and not the alveolar fricative, that is, /kat ͡si/ and not /kas/. As, described in §2.3.1, /t ͡s/ simplifies to /s/ before a consonant, except before a velar fricative /x/. The vowel /i/ gets deleted before V-initial roots to avoid a violation of ONSET. Interestingly, this vowel also gets deleted before C-initial roots. While the motivation behind vowel deletion in this context is not very well understood, one of my main consultants stated that the [kat ͡si] ~ [kas] is based on intergenerational variants. According to FR, the use of [kas-] before consonant-initial roots is characteristic of Nivaĉle younger generations, who in turn only use [kat ͡si] with CC-roots to avoid illicit triconsonantal clusters of three consonants. In contrast, [kat ͡si] is mostly used by older generations, both before C- and CC-initial roots. According to Stell’s (1989:185) description, [kat ͡si] is exclusively used before C-initial roots. The alternation between [kas] and [kat ͡s] is definitely a reflection of a language change in progress. During my fieldwork, I found variation of these two allomorphs across speakers of different generations.  The only case in which a C-initial prefix attaches to a C-initial root is displayed by the lateral fricative [ɬ-] (73c). In (73), the third person possessive prefix [ɬa] (73a,b) alternates with [ɬ], (73c). The lateral fricative [ɬ] appears to be the only ‘syllabic’ consonant in the language.24 However, given the                                                 24 Note that in Wichí, a related language, the cognate third person possessive prefix /la-/ gets reduced to [l] in casual speech: e.g., [l.’wu] ‘his neck’. Nercesian (2011:138) treats this ‘reduced’ allomorphic variant as “syllabic” and states that this syllabic lateral is longer than its non-syllabic counterpart (no duration measurements are given). In contrast, I have not found any significant duration differences in Nivaĉle.  62 constraints postulated in §2.3.1, only moras get parsed into the Nucleus; segments get parsed into the syllable. The lateral fricative is not associated with a mora; therefore, it is parsed directly to the syllable.   It is posited that the vowel [a] in [ɬa] (73) gets deleted, in accordance with the previously discussed vowel hiatus repair mechanism. Also, it is worthy of mention that the vowel [a] also gets deleted in the following form: /ɬakomʔa/ → [ɬkomʔa] ‘everybody’, suggesting that vowel deletion occurs after a /ɬ/ in other domains. Moreover, during my fieldwork, I noticed another case of syllable reduction; the first person subject /xa-/ reduces to /x/ in fast speech, for instance /xa-k ͡le ̰ʃ́/ ‘I wash’⇒ [x.k ͡leʃ̰] (cf. §2.5.4).  A parallel situation to the one described in the nominal domain can be found in the verbal domain: the second person subject [ɬa] alternates with [ɬ].   (80)    a. ɬa-k.tʃ͡aʔ́   2S-paddle   ‘you paddle’  b.  ɬ-ám   2s-come   ‘you come’  c.  ɬ-péʔ.ja   2S-listen   ‘you listen’   An alternative analysis is to propose that the lateral fricative forms a complex onset with the following syllable: e.g., [ɬna ̰ʃ̰́] ‘his/her nose’ [ɬpéʔja] ‘you listen’. Nothing in principle would rule out this possibility, especially when considering that, if sonority were to be taken into account, no violation for sonority reversal would be incurred. Note that the possessive prefixes [kas-] and [βat-] occur before sonorant-initial roots. The Syllable Contact Law operates in the MStem domain (root + derivational suffixes), (§4.5).   Another analytical approach would be to hypothesize that the initial [ɬ-] is extrametrical, i.e., that it is outside of the metrical/syllabic parsing of the segmental string. However, I do not claim that  63 this consonant is not parsed into the syllable. Rather, I adduce evidence in Chapter 4 to support the hypothesis that it is counted towards foot formation. Foot formation, stress assignment and the licensing of glottalized vowels provide the evidence for treating the lateral fricative as a different syllable, similarly to other CV possessive prefixes.  2.5 Phonological processes in Nivaĉle There are several phonological processes that will be referred to in subsequent chapters; it thus will be useful to include a brief characterization of them here.  2.5.1 Palatalization Palatalization, a cover term for a variety of processes that arise through the bidirectional interaction of a high front vowel or palatal glide with consonants, has been proposed as an areal feature of the Chaco languages (Klein 1992, Messineo 2003, González 2014). Specifically, González (in press) argues that full palatalization, where “a consonant shifts its primary place of articulation and often its manner of articulation while moving toward the palatal region of the vocal tract” (Bateman 2007:2), is a common feature of the Chaco languages. In the Mataguayan languages, palatalization affects dorsal consonants (González, in press). For example, in Maká there is a contrast between a uvular stop /q/ and a velar stop /k/, which gets realized as a palatal stop [kj] before /e/ and /a/ (Gerzenstein 1994: 47). Carol (2014) claims that there is a phonemic contrast between /k/ and /kj/. There is no palato-alveolar affricate [t ͡ʃ] in the aforementioned languages. In contrast, the Bermejo variety of Wichi (Nercesian 2014a) has the affricate /t ͡ʃ/ in the phonological inventory (only in onset position). Recall that in Nivaĉle there is a velar stop /k/, in alternation with a uvular stop [q], plus the presence of /t ͡ʃ/ (§2.2.1). Also, Nivaĉle has dorsals [x], [χ] and the palato-alveolar /ʃ/. None of the other Mataguayan languages have [ʃ] except for Wichí; [x] palatalizes into [ʃ] when preceding [i] and [e] across morpheme boundaries (Nercesian 2011:153).     64 Table 2.12 Palatalization: Comparison within the Mataguayan family  Chorote Wichí Maká Nivaĉle Gloss ihnyetak natek xunxetek xunʃatat ͡ ʃ  ‘tusca tree’ -hetek -etek  -etek  -ʃatet ͡ ʃ   ‘head’ tetik   titet ͡ ʃ  ‘plate’ -k’inix -t ͡ ʃ ʼ ihnɑ -k ʼinχaʔ  -t ͡ ʃ ʼ inxɑ ‘younger sister’   k’ek’e t ͡ ʃ ʼe t ͡ ʃ ʼe ‘parrot’   -ekaχ -t ͡ ʃa ̰x ‘bring’ k  vs. tʃ͡ -ki -ke ~ -t ͡ ʃe -ki -ke ~ -t ͡ ʃe FEM -hetek  xiɬa -ʃatet ͡ʃ  ‘head’  ʔilex   ʔileʃ-ex wash-INST  -k ͡leʃ̰  ‘wash’   xinawap ʃnawɑp ‘spring’ -hi -hi -xi -xi ~ -ʃ i INSIDE x. vs. ʃ   -xem -xam ~ -ʃam THROUGH  Besides the existence of roots with /ʃ/ and /tʃ͡/, there exists an alternation between the Nivaĉle palatal – [t ͡ʃ], [ʃ] – and velar-initial suffixes /k/, /x/ – for example: [ʃam] ~ [xam] ‘LOC (on (top of)/up/through)’; [ʃane] ~ [xane] ‘LOC (down)’; [t ͡ʃiʃam] ~ [kiʃam] ‘on’; [ʃi] ~ [xi] ‘RES’, ‘LOC’; [tʃ͡’e] ~ [k’e] ‘LOC (interior) /intensive’).25 The velar vs. palatal realization of the consonant-initial suffix is motivated by the vowel quality of the rightmost vowel of the preceding root. If there is a front vowel, the palatal variant is used. As noted previously in the discussion of the vowel inventory (§ ̙2.2.2), these data show that whereas [a] patterns with front vowels, [ɑ] patterns with back vowels. Interestingly, note that the trigger (vowel) and the target (consonant) are not necessarily adjacent: there can be labials, (83a), coronals (82a), (84a), dorsals (82), (84b) and a glottal stop (81b), (83b), before the palato-alveolar affricate and fricative.                                                  25 Until I find more conclusive evidence, I will refer to these morphemes as suffixes. However, there could some arguments for treating these morphemes as clitics. First, they have freedom of host selection; they can attach to both nominal and verbal stems. Second,  these morphemes behave differently from derivational suffixes in terms of stress assignment, i.e. they do not necessarily “shift” stress to the last syllable (see §3.3.2, §4.5.3).  65 (81)    a. ɬ-ne ̰́t-ʃam   2s-get-LOC(up)   ‘you get up’  b.  jitaʔ́-ʃam   scrub-LOC(up)   ‘very thick scrublandʼ   (82)       xaxu ̰́x-xam   ɬa=tʼún  1S-bite-LOC(on)  F.DET=galleta(type of bread)  ‘I bit the cracker’       (83)   a.  k ͡lím-ʃi   white-RES   ‘flour’  b.  tʃ͡i-jɑ́ʔ-xi   INDEF.S-drink-RES   ‘it is drunk’   (84)  a.  is-t ͡ʃ’é   nice-INTENS   ‘beautiful’  b. ux-k’é   na=nik ͡lɑ̰́t ͡sitʃ͡   big-INTENS  DET=corn   ‘the corn is very big/thick’  (85)      a.  ɬ-ɑ̰́w-xané   2s-be-LOC(down)   ‘you sit down’  b.  ɬ-ɑ.β-e ̰́ɬ-ʃané   2-be-PR.PL-LOC(down)   ‘you sit down’  c.  j-íʔ-ʃané   3-be-LOC(down)   ‘s/he sits down’  Stell (1989:301) documented ɬ-aw-xané-ʔeɬ for the second person plural rather than the form in (85b). Whereas I am not sure about the reason behind the different placement of the pronominal plural  66 [ʔeɬ], it is interesting to note –consistent with the Palatalization analysis documented here – that in the (85b) form where [-ʔeɬ] precedes the locative (LOC), the front vowel in [-ʔeɬ] triggers the use of the ‘palatalized’ alternant [-ʃane]. The collective and feminine suffixes in the following examples also show the velar and palato-alveolar alternation: (86)  a. smìt.ka-t ͡ʃát    peanut.tree-COL   ‘stand of peanut trees’  b.  tìs.xu-kát   quebracho-COL   ‘stand of quebracho trees’   (87)  a.  niβak-t ͡ʃé   man-F   ‘woman’  b.  samto-ké   white.man.F   ‘white woman’   Despite broad-based empirical support for interpreting these alternations in terms of an active synchronic palatalization process governed by the front vs. back vowel quality of the preceding root there are some forms in my database that do not conform to the expected palatalization generalizations:  (88)   a. -tɑ̰ɬ́-ʃam   ‘to come from’   cf. (82) -xam b. jikxu ́s̰-ʃam  ‘on my kneeʼ    cf. (82) -xam c. t ͡sikɑ̰́t-ʃane   ‘I fell down’   cf. (85) -ʃane       d. t ͡sa ̀nku-t ͡ʃát   ‘stand of duraznillo trees’ cf. (86a) -tʃ͡at  What is unexpected in each of these cases is that the palatalized alternants, [ʃ] and [t ͡ʃ], occur after a back vowel, rather than a front vowel.  In the first three cases above, one might hypothesize that the intervening coronal consonant (ɬ, s, t) might be exerting a local assimilatory influence on the realization of the following segment, e.g. as perseverating the coronal articulation. This would not,  67 however, explain the form in (88d). The analysis of non-adjacent palatalization patterns in Nivaĉle go beyond the scope of this dissertation but open an interesting venue of research. 2.5.2 Epenthesis  Some phonotactic constraints in Nivaĉle reflect the prohibition against onsetless syllables complex codas and the syllable contact law (SCL) (§2.3.2, §6.3.2). In terms of strategies to repair violations of syllable structure, it is notable that underlying segments are hardly ever deleted. Rather, ʔ-epenthesis and V-epenthesis work as syllable repair strategies. This leads, therefore, to an over-arching observation that MAX-IO faithfulness constraints are higher ranked than DEP-IO constraints in Nivaĉle.  As is discussed in greater detail in Chapter 3, the default epenthetic onset is the glottal stop.  Glottal stop epenthesis  (89)   a. tsi-kú=[ʔ]a  pa=cocido   1S-like-3O  DET=cocido   ‘I would like to have a cocido (mate tea)’  b. k’-uʔ-é.ʃ=a   pa=fitsɑkʼɑjit ͡ʃ  1S-believe-INST=3O DET= God        ‘I believe in God’  (90)   a.  xa ̀-t-pe.k ͡l-e ́j       1S-DIR-go-DIR        ‘I return’  b.  xa-peʔ̀.je-[ʔ]e ́j       1S-hear-DIR        ‘I hear (from the distance)’  As illustrated in these examples, ʔ-epenthesis functions as a pervasive strategy to provide otherwise onsetless syllables with a consonantal onset. Explicit argumentation is adduced in Chapter 3 for the featural representation of /ʔ/. Its place-less specification allows it to function as the optimal epenthetic consonant to effect non-violation of the syllable structure markedness constraints ONSET.    68 Vowel epenthesis  As will be developed in subsequent chapters, vowel epenthesis is enforced in order not to violate higher-ranked markedness constraints, e.g.*COMPLEXCODA and the SCL (§6.3.2). As is illustrated by a broad diversity of suffixes below, the default epenthetic vowel is [i].  (91)    a. woxó    ‘pecari’  b. kasús    ‘pumpkin’ +-k     PL woxók ‘pecaris’  kasus[í]k  ‘pumpkins’     *kasusk (92)    a. aɬú    ‘lizard’  b. βat-ma ́tʃ͡     INDEF.POSS-meal  c. oβa ́j    ‘guavirami (fruit)’ aɬús ‘lizards’  βa ̀tmatʃ͡[í]s ‘meals’  oβaj[í]s  ‘fruits’     *βatmat ͡ʃs   *toss  (93)    a. wɑkɑ ́   ‘cow’  b. wosók   ‘butterfly’ + -s       PL        + -tax      AUG wɑkɑta ́x ‘zebu’  woso ̀k ͡l[i]ta ́x ‘big butterfly’    *wosoktax   (94)    a. ɬaβɑ́    ‘flower’  b. ɬu ̰́p     ‘nest’ +-tʃ͡at     COL ɬaβɑt ͡ʃa ́t ‘garden’  ɬup[i]t ͡ʃa ́t   ‘group of nests’     *ɬu ̰ptʃ͡at (95)    a. tɑwɑ́     ‘Maká people’     b. k’afok   ‘crow’ +ɬaj   GROUP tɑwɑɬa ́j ‘group of Maká people’  k’afo ̀k[i]ɬa ́j ‘Argentine militaries’    *k’afokɬaj  69 (96)   a. wot ͡só    ‘lechiguana’   b. saxétʃ͡     ‘fish’  +-met ͡ʃ SHAMAN wot ͡so-me ́tʃ͡ ‘shaman of the lechiguana’   saxe ̀tʃ͡[i]me ́tʃ͡ ‘shaman of the fish’       *saxetʃ͡met ͡ʃ  (97)   a. kumok ͡lú    chañar(tree)’  b. ɬaft ͡su ̰́k   ‘palm tree’ +-nɑk       RES   kumo ̀k ͡lunɑ́k  ‘chañar wine’  ɬaft ͡su ̀k[i]nɑ́k  ‘palm tree wine’     *ɬaft ͡su ̰́knɑk (98)   a. isí    ‘clean’  b. sás    ‘dirty’  +-nat    CAUS   isinát ‘to make s.t. clean’  sas[i]na ́t ‘to make s.t. dirty’     *sasnat (99)  a. k ͡lesá    ‘knife’  b. kot ͡sxa ̰́t    ‘soil/ground/land’ +-niɬ  MAT k ͡lesaníɬ ‘made of iron’  kot ͡sxa ̀t[i]níɬ ‘made of soil’     *kot ͡sxa ̰́tniɬ  (100)  a. naɬú    ‘day’  b. k ͡lapʼa ̰́f   ‘bold’  +-jan     CAUS xana ̀ɬuja ́n ‘I light’  xak ͡la ̀pʼaf[i]ja ́n ‘I shave’  (101)  a. ji-pa ́stʃ͡e    1POSS-finger      ‘my finger’  b. p’o ́k    ‘arrow’ +-waʃ      MARK  jipas̀tʃ͡ewa ́ʃ  ‘my fingerprint’   p’ok[i]wáʃ  ‘mark of an arrow’    *xak ͡lapʼa ̰́fjan        *p’okβaʃ   70 2.5.3 Vowel harmony In Nivaĉle, spreading of vocalic features can be observed across epenthetic and non-epenthetic glottal stops at morpheme boundaries. When two vowels are adjacent in the input due to morpheme concatenation, a glottal stop is inserted and there is regressive or progressive vowel harmony. The examples in (102-103) show regressive assimilation, where a stressed vowel [e] is the trigger for vowel harmony and low vowels /ɑ a/ are the target. I have also found one example with the vowel /u/ as the target (103c).  (102)  [meʔeɬ́]   /mɑ-eɬ/    go-PR.PL   ‘go (pl)!’  (103) a.  [xapeʔ̀jeʔe ́j]   /xapeʔj-a-ej/   pa=tʼɑ̰́j             1S-hear-PUNC-LOC  DET=noise             ‘I heard noise (from the distance)’   b.  [ʔaβa ̰̀tʃ͡eʔeɬ́]   /a-βaʔtʃ͡a-eɬ/   2POSS-property-PR.PL   ‘your(pl) property/you (pl)   c.  [xanu ̀keʔeɬ́]   /xa-nuku-eɬ/  1S-drop-PR.PL  ‘We (excl.) drop’   The data in (102) and (103a,b) show unstressed [ɑ] and [a] undergoing total vowel assimilation to [e]. In (103c) we can also see a high back vowel undergoing assimilation of /e/. In contrast, example (104) shows progressive vowel harmony across an underlying glottal stop.     71 (104) a.  [xapʼo ̀ʔeʔe ́n]        /xa-pʼoʔ-e-ʔin/   1S-close-LOC-IPFV   ‘I am closing’  b.  *[xa-pʼo ̀-ʔi-ʔín]  c. *[xa-pʼo-ʔe-ʔín]  The examples just presented will be discussed in Chapter 3, §3.3.3. Notably, vowel harmony can also be marginally found in epenthetic vowels across coronal and dorsal consonants (§2.6.2, cf. Chapter 6):26  (105) a.  -sɑ̰́t   ‘vein’  b.  -sɑ.t-[ɑ́]j  vein-PL       ‘veins’  (106) a.  ji-pɑ́ʔ.kɑt   1POSS-hand   ‘my hand’  b.  ji-pɑ̀ʔ.kɑt-[ɑ́]j   1POSS-hand-PL   ‘my hands’  (107) a.  βat-sa ́ʃ̰   INDEF.POSS-hair   ‘someone’s hair’  b.  βa ̀t-sa.ʃ-[á]j   INDEF.POSS-hair-PL   ‘someone’s hairs’   (108) a.  ʔan.ko ̰́k        ‘s/he has a limp’                                                  26 Note, as will be explained in Chapter 4, that the glottalized vowels deglottalize when not receiving stress (105b, 107b, and 108b).  72  b.  kas-ʔa ̀n.ko.x-[o ́]j     1POSS.PL-limp-PL     ‘we have a limp’  (109) a.  ji-βóʔ.mat   1POSS-wound   ‘my wound’  b.  ji-βoʔ̀.mat-[a ́]s   1POSS-wound-PL   ‘my wounds’  (110) a. ji-ka ́-̰t.xok   1POSS-POSS.CLASS-uncle   ‘my brother-in-law’  b.  ji-ka ̰-̀t.xo.k-[o]-βo ́t   1POSS-POSS.CLASS-uncle-FAM.PL   ‘my brothers-in-law’  Vowel harmony in epenthetic vowels, though, is restricted to a small number of examples. Further, I have documented fluctuations in the vowel quality of the epenthetic vowel. I posit that these alternations might be due to a change in progress; because determiners inflect for number, the plural marker of the noun can be dropped. As a consequence, plurality on the nouns is not consistently marked and thus hesitations arise.  (111) a.  ji-pɑ̀ʔ.kɑ.t-[ɑ́]j      ~  ji-pɑ̀ʔ.kɑ.t-[í]j ~   na-wá=ji-pɑʔ́.kɑt     1POSS-hand-PL  1POSS-hand-PL   DET-PL=1POSS-hand-PL    ‘my hands’                 ‘my hands’   2.5.4 Metathesis and deletion When morpheme concatenation would result in an illicit syllable structure or a bad syllable contact (§2.3.2), vowel-consonant metathesis may come into play as a repair strategy (cf. Chapter 6).   (112) a.  ɑ.tɑ́x     ‘cranium’  73  b.  ɑt.xɑ-́s     b’. * ɑtɑxs    cranium-PL   ‘craniaʼ  (113) a.  ji.jɑ̰́x   ‘puma’  b.  jij.xɑ-́s     b’. jijɑ̰xs   puma-PL   ‘pumas’  c.  jìj.xɑ-me ́t ͡ʃ    c’. jijɑ̰xmet ͡ʃ   puma-SHAMAN   ‘shaman that has power over the pumas’     Unstressed vowels also tend to delete (or reduce, cf. Chapter 4) in fast speech or storytelling (§2.3.1); specifically, they occur in open pretonic syllables:  (114) a. [ɬ.kómʔa]    /ɬakomʔa/   ‘everybody’  b.  [ɬ.pe ́ʃ]   /ɬapeʃ/   ‘a long time ago.’   (115) a.  [xtɑ́ʔɬej]      /xa-tɑʔɬ-ej/   1S-come-DIR   ‘I come from’  b. [x-k ͡le ̰ʃ́]    /xa-k ͡leʔʃ/   1S-wash   ‘I wash’  2.5.5 Deglottalization of glottalized consonants and vowels A pervasive and complex set of phenomena in Nivaĉle involves the “deglottalization” of both consonants and vowels under what initially appears to be a diversity of circumstances. A major goal of the present research is to show that the realization of the distinctive feature of [constricted glottis] is  74 fully systematic, and that its surface alternations are governed by the interaction of constraints on both segmental and prosodic structure in Nivaĉle.  As will be explained in detail in Chapters 3 and 6, the realization of the [constricted glottis] feature in consonants is perceptually motivated: ejectives can only occur before vowels, as this context allows for the optimal identification of the glottalized release. In turn, the realization of the [constricted glottis] feature with vowels is tied to prosodic prominence: rearticulated/creaky vowels or the alternant [Vʔ] realization can only occur in a stressed syllable. Where the requisite conditions for the realization of [c.g.] are not met, the segment will “deglottalize”. For example, forms like the following (repeated from (107) above) illustrate deglottalization of the vowel [a ̰́] in the root for ‘hair’ if it does not surface in a stressed syllable.   (116)  a.  βat-sa ́ʃ̰   INDEF.POSS-hair   ‘someone’s hair’  b.  βa ̀t-saʃ[a ́]-j   INDEF.POSS-hair-PL   ‘someone’s hairs’  The deglottalization processes will be discussed at length in Chapters 3, 4 and 6.  75 Chapter 3: The phonetics and phonology of Nivaĉle laryngeals    3.1 The problem Issues related to the distribution and phonemic status of the glottal stop, its surface realizations, and glottalization on vowels, have challenged the phonological analyses of languages of the Americas such as Mayan (England 1983, Frazier 2009, Avelino 2011), Mazatec (Silverman et al. 1995, Silverman 1997, Blankenship 2002), Mixtecan (Hollenbach 1984, Macaulay & Salmons 1995, Gerfen 1999, Gerfen & Baker 2005, DiCanio 2008, Herrera Zendejas 2009), Tukanoan (Miller 1999, Stenzel 2004, 2007, Silva 2012) and Zapotec (Munro & Lopez 1999, Avelino 2004, Arellanes 2009, Chávez-Peón 2010), among many others.   Most of the challenges posed by the glottal stop arise from its ambiguous patterning. On the one hand, the glottal stop can pattern with either stops or sonorants. On the other hand, it can get realized as a full segment or as glottalization in the same language (Zoll 1998 [1996]). In that regard, what is commonly referred to in phonological inventories as ‘glottal stop’ has been variously analyzed as: i) a full independent segment (e.g., Yalálag Zapotec, cf. Avelino 2004; Desano, cf. Miller 1999; Wanano, cf. Waltz & Waltz 2000, and Stenzel 2004; Chalcatongo Mixtec, cf. Macaulay 1987), ii) a constricted glottis ([c.g.]) feature on vowels (e.g. Desano, cf. Kaye 1970; Mixtec, cf. Bradley 1970, Hinton et. al 1992, Gerfen 1999), iii) a floating constricted glottis feature (e.g., Yawelmani, cf. Archangeli & Pulleyblank 1994; Mixtec, cf. Macaulay & Salmons 1995; Blackfoot, cf. Peterson 2004), and as a floating tone (e.g., Southern Min dialects, cf. Chung 1996 as cited in Kavitskaya 2002;  Tukano, cf. Ramírez 1997 as cited in Stenzel 2007).  Further, unlike other features, which are posited to have a unique structural dominance affiliation within a given feature hierarchy model, the [c.g.] feature has been variously analyzed as directly dominated by a mora in Mixtec (Macaulay & Salmons 1995) and Wanano (Stenzel 2007), exclusively by a non-nuclear mora in Blackfoot (Peterson 2004), by a root node (Zoll 1996), or a laryngeal ([LAR]) node (Clements 1985, Clements & Hume 1995, Picanço 2005).  76  The relationship between the glottal stop and vowels in the Mataguayan languages has not been thoroughly studied. What seems to be consistent in the previous literature, though, is its treatment as a consonant rather than as a vocalic feature. Gerzenstein (1983, 1994) includes the glottal stop in the consonantal phonemic inventory of Chorote and Maká. The author states that this consonant can occur in word initial, medial (between homorganic vowels) and final position. However, neither the featural representation of the glottal stop nor the nature of the relationship with the homorganic vowels that are being interrupted is addressed. In turn, Nercesian (2011:92) specifies the syllabic affiliation of the glottal stop in Wichí by claiming that it can serve as an onset in word initial and medial position, and as a word medial and final coda. She further notes that glottal stop onsets cause laryngealization – and elongation – of the following vowel. In summary, then, glottal articulation has been analyzed as a consonant in other Mataguayan languages.  Recall, however, from §2.2.2 that Stell (1989) posits that there is a distinction in Nivaĉle between plain and glottalized vowels. Given that glottalization or creakiness in vowels has not been reported as a contrastive value in any of the other Mataguayan languages, it is illuminating to compare the representation of cognates across the different languages. Note in the following cognate forms that what Stell represents as a contrastively glottalized vowel corresponds to what is represented as two homorganic vowels interrupted or followed by a glottal closure in Chorote and Wichí:          77 Table 3.1  Cognates of Nivaĉle “glottalized vowels” Languages Transcription Gloss Source Nivaĉle /fảj/ [fa ̰j] ~ [faʔa ̰j] ‘algarroba pods’ (Stell 1989: 192) (my fieldwork notes) Chorote fwaaʔ waʔaj ‘algarroba pods’ (Gerzenstein 1983:128) (Campbell and Grondona 2007:19) Wichí fwaʔaj ‘algarroba pods’ (Nercesian 2011:92)     Nivaĉle jikféʔ ‘my ear’ (my fieldwork notes) Maká jikfiʔ ‘my ear’ (Gerzenstein 1994:71)     Nivaĉle ɬawoʔ xawoʔ ‘worm’ ‘I fish’  (my fieldwork notes) Chorote awoʔ ʔa-woʔo ‘worm’ ‘I fish’ (Gerzenstein 1983:50) (Carol 2014:78)   Within the literature on the Nivaĉle language there is no consensus – either explicit or tacit – about the distinction and relationship between glottal stops and glottalized vowels. In his Nivaĉle-Spanish dictionary, Seelwische (1990) represents glottalized vowels in closed syllables as double vowels, e.g. c’utsaaj, cf. ‘old person’ [k’ut ͡sá ̰x] (my transcription); and elsewhere, i.e. in word-medial and final position, as a vowel followed by an apostrophe, e.g. yo’nis ‘fox’, cf.  [joʔnis] (my transcription), and nu’ ‘bone’, cf. [núʔ] (my transcription). With respect to Seelwische’s double vowel representation, the reader will recall (§2.2.2) that there is no vowel length distinction in Nivaĉle independently of a modal/glottalized distinction.  Even though there is phonetic validity to there being greater vowel length in glottalized vowels (§3.2.1), Seelwische’s double-vowel representation gives no overt acknowledgment of laryngeal articulation in these vowels.27 The glottal stop – represented with an apostrophe [’] – does not head an entry anywhere in the dictionary. In other words, Seelwische does not                                                 27 Nivaĉle teachers have also represented the Nivaĉle glottalized vowels as two vowels separated by an apostrophe: e.g. c’utsa’aj. The representation of glottalized vowels has thus been object of debate within the Nivaĉle educational community. The Linguistic Committee of the Nivaĉle People (CLPN) has decided to continue using Seelwische’s representation: c’utsaaj.   78 give independent recognition to [’] as a distinctive segment in the language, aside from its post-vocalic realization. Even though suffixes are listed and certain suffixes like the imperfective [ʔin] would need to be analyzed as glottal-initial since glottal realization here is distinctive, not predictable, Seelwische lists this suffix as -in. 28  In turn, Stell (1989: 92) posits the existence of a phonemic contrast between plain vowels /i e a ɑ o u/ and their glottalized /ỉ ẻ ả ɑ̉ ỏ ủ/ counterparts. As well, she treats the glottal stop as an independent consonantal phoneme in the language, /ʔ/.   Stell represents glottalized vowels with a hook on top of the vowels. I will adopt a similar transcription only in this section to illustrate the different ways glottalized vowels are represented in the Nivaĉle literature. Elsewhere I represent what Stell characterizes as “glottalized vowels” as [V ̰] ~ [Vʔv ̰] in closed syllables and as [Vʔ] in open syllables (see discussion below), and as /Vʔ/ in input representation:                                                     28 The CLPN has undertaken a revision of Seelwische’s dictionary: this suffix is listed as ’in but under the entry for the vowel i.  (117)  a. is           ‘nice’ b. ỉs            ‘write (IMP)’ (118)  a. jitex       ‘carob’        b. jitẻx       ‘grass’    (119)  a. -saʃ         ‘mucus’ b. -sảʃ           ‘wool’ (120)  a. k ͡lɑp       ‘fast’ b. -k ͡lɑ̉p     ‘to be seated on the lap’ (121)  a. k ͡lop       ‘white/larva.’ b. k ͡lỏp        ‘winter’ (122)  a. ji-f.xux      1POSS-toe    ‘my toe’ b. ji-f.xủx    1POSS-stick    ‘my stick’  79  From the data set in (117)-(122), we can observe that all of the listed minimal pairs involve closed syllables. However, Stell notes that all glottalized vowels can also occur in final position (123)-(125). She provides the following auditory characterization (1989:61): “glottalized vowels are clearly perceived as two identical vowels separated by a glottal closure [ʔ]. The second vowel is shorter in initial and interconsonantal position. In absolute final position, the second vowel is voiceless” [my translation from Spanish/AG].  Importantly, Stell (1989:62) also notes that the glottalized vowel “may lose its second voiceless vowel when the following word starts with a vowel [(123)], a consonant [(124)], or when suffixation occurs [(125)]. In the last case, it also loses its glottalized property” [my translation from Spanish/AG]:                 I have not encountered the phenomenon described in (123) and (124): all those forms are realized in my data as a vowel-glottal sequence regardless of the following context. The deglottalization phenomenon exemplified in (125) is an interesting and pervasive process that I will discuss in §6.5.30                                                  29 For clarity, I have added the UR of the glottalized vowels, in accordance with Stell’s analysis.  30 Given Stell’s general characterization of the glottalized vowels, it is worth pointing out the presence of some transcription inconsistencies in her grammar. Sometimes glottalized vowels in word final position are transcribed as (123)              /ɬ-kả/ a. ɬa      ɬ-kả                    DET   3POSS-fruit     ‘his/her fruit’             / ɬ-kả/ b. ɬa      ɬ-kaʔ              ux      DET   3POSS-fruit     big     ‘his/her fruit is big’ (124)  β.    /ji-kfẻ/ χ. a. ji-kfẻ  δ.    1POSS-ear    ‘my ear’     /ji-kfẻ/ b. ji-kfeʔ           tik’in      1POSS-ear      small      ‘my ear is small’       (125)      /tisủx/ a. tisủx      ‘quebracho’      /tisủx/ b.  tisxu-j      ‘quebracho (pl)’                            Stell (1989:62)29  80  In contrast, in their article on internal reconstruction in Nivaĉle, Campbell and Grondona (2007:5) present only six vowels /i e a ɑ o u/ as well as the glottal stop in the Nivaĉle phonological inventory. Recall from Chapter 1 that both Stell (1989) and Campbell and Grondona (2007) worked with the same group – the chishamnee lhavos ‘highlanders’ (Upriver dialect) group, though with a difference of 30 years. In addition, Stell (1989) worked with some shichaam lhavos ‘lowlanders’ (Downriver dialect) consultants. Recall, in that regard, that I worked with both shichaam lhavos and yita’ lhavos speakers (‘people of the scrub’).  While Campbell & Grondona do not explicitly discuss the status and representation of glottalized vowels, it can be observed by comparing their transcription with Stell’s in (126) and (127) that Stell’s ‘glottalized vowel’ [ả] gets variantly transcribed by Campbell & Grondona as Vʔ (126b) and as VʔV (127b):  (126) a.    k’ut ͡sảx         ‘old man’                 (Stell 1989:141)       b.    k’ut ͡sa ʔx       ‘old man’                   (Campbell & Grondona 2007:6)   (127) a.   fảjuk           ‘algarrobo tree’           (Stell 1989:192)        b.   ɸa ʔayuk      ‘algarrobo tree’          (Campbell & Grondona 2007:6)    Given the representational divergences in the cited literature (Seelwische 1980, Stell 1989, and Campbell & Grondona 2007), one crucial topic to be analyzed is the phonetic realization and                                                                                                                                                        “glottalized”, with a hook on the vowel: (ia) - (iiia), and sometimes as a sequence of a vowel and a glottal stop: (ib) - (iib) or a vʔv sequence (iiib). It is not clear whether these are typos or whether this reflects some sort of variation in the actual realization of the glottalized vowels being documented. (i) a. ɬkủ                        (Stell 1989:149)     ‘load’   b.  ɬkuʔ                 (Stell 1989:65)     ‘load’ (ii) a. konxả    (Stell 1989:78)    ‘smooth-billed ani’   β. b. konxaʔ              (Stell 1989:133) χ.     ‘smooth-billed ani’  δ.  (iii) a. nủ                         (Stell 1989:116)    ‘dog’ b. nuʔu                  (Stell 1989:103)    ‘dog’    81 phonological status of Nivaĉle glottalized vowels and the glottal stop. The goal of this chapter is to address the featural and prosodic representation of the alleged phonemic glottalized vowels (Stell 1989) and the glottal stop.    Section 3.2 presents an overview of phonation types in the languages of the world, with focus on creaky and glottal phonation and its variable manifestation in Nivaĉle vowels. Two basic categories of realization of Nivaĉle ‘glottalized’ vowels are proposed: the first, I call a rearticulated/creaky vowel, represented variably as [Vʔv ̰] (careful speech) ~ [V ̰] (fast speech), and, the second, I call a vowel-glottal coda, represented as [Vʔ]. Further, based on data collected in the field, it is shown that duration is a statistically significant variable that distinguishes modal from rearticulated/creaky vowels. Section 3.3 advances a proposal in which the glottal stop is specified for [c.g.], but not for place. First, a glottal stop can occur, as an independent segment, in syllable onset position. Second, if a vowel + glottal stop sequence occurs in a syllable closed by another consonant, the glottal stop, here analyzed as underlyingly moraic (see §3.4 and §4.3), can be parsed to the vocalic nucleus of the syllable and hence form part of a complex nucleus – phonetically realized as a rearticulated/creaky vowel. In turn, if there is no other consonant in coda position, the [c.g.] feature will be realized post-vocalically as a glottal stop (vowel-glottal coda). It is argued here that these diverse glottal realizations are rooted in a set of prosodic constraints. Rearticulated/creaky vowel and vowel-glottal coda, thus, are variants that occur in complementary distribution due to the different parsing of the [c.g.] feature. In Section 3.4, it is shown that Nivaĉle glottalized vowels must occupy a prominent position, that is, they occur only in a stressed syllable, i.e. the head of a foot (cf. Chapter 4, §4.3). Section 3.5 concludes with the main findings of this chapter. 3.2 Overview of phonation types Phonation types – or voice quality – refer to the manner in which the vocal folds vibrate. Several proposals have been advanced in order to explain the different ways in which the vocal folds are configured in the production of speech (Catford 1964, 1977, Ladefoged 1971, 1973, Laver 1980, Ladefoged & Maddieson 1996, Gordon & Ladefoged 2001). Based on glottal constriction, that is, the  82 degree of aperture between the arytenoid cartilages, Ladefoged (1971:17) proposes a continuum of phonation types.31 This continuum ranges from voiceless (arytenoid cartilages furthest apart, no vibration of the vocal folds) going through breathy voiced, to regular modal voicing, then to creaky voice, and finally to glottal closure (arytenoid cartilages closest together, no vibration of the vocal folds). Airflow rate is inversely related to the degree of glottal constriction (Ladefoged & Maddieson 1996:48).    Figure 3.1 Continuum of glottal constrictions (after Ladefoged 1971), reproduced from Gordon & Ladefoged (2001).   Phonation contrasts in vowels can be found in several languages, ranging from two-way to four-way contrasts. For instance, in terms of two-way contrasts, Gujarati (Fischer-Jørgensen 1967) and Kedang (Samely 1991) distinguish between modal and breathy vowels, whereas Mundurukú (Picanço 2005) contrasts modal and creaky voice on vowels. Three-way contrasts across the categories of breathy, modal, and creaky can be mostly found in Otomanguean languages (Kirk et al. 1993, Silverman 1995, 1997, Blankenship 2002, Esposito 2010, among others). Four-way contrasts – modal, breathy, creaky and interrupted – can be found in San Lucas Quiaviní Zapotec (Munro & López 1999, Chávez Peón 2010).32 The phonetic correlate of ‘glottalized’ or ‘laryngeal’ vowels has been described in the literature as creaky or laryngealized voice. This type of non-modal phonation is “typically associated with vocal folds that are tightly adducted but open enough along a portion of their length [the anterior portion] to                                                 31 Note that Ladefoged’s (1971) original proposal consisted of nine phonation types or states: voiceless, breathy voice, murmur, lax voice, voice, tense voice, creaky voice, creak, and glottal stop.  32 Interrupted vowels are defined as modal voice followed by a glottal closure. Interrupted is also defined by Chávez-Peón (2010:12) as “glottalized voice”.  83 allow for voicing” (Gordon & Ladefoged 2001:386), which [often, but not always] results in a typical low fundamental frequency (Ladefoged 1971, Laver 1980).  In turn, other proposals move beyond the glottal states and make reference to supra-glottal mechanisms involving “a number of valves that represent a synergistic and hierarchical system of laryngeal articulations” (Edmondson & Esling 2006:157) that may create distinctive phonation types (see also Moisik & Esling 2011; Moisik, Czaykowska-Higgins & Esling 2012). For instance, based on contrastive phonological behaviour in Dinak, Bai, and !Xo ́o ̃, Edmondson and Esling (2006) add ‘faucalised’, ‘harsh’ and ‘strident’ voice to the range of phonation types.  3.2.1  On glottal stop and creaky voice in Nivaĉle Aperiodicity (Gordon & Ladefoged 2001) in the signal is one of the traditional acoustic characteristics of creaky voice. Figure 3.2 presents waveforms of the Nivaĉle modal vowel [e], in [ɬk ͡le ́s] ‘her children’, the glottalized vowel [e ̰], in [ɬk ͡le ́ʃ̰]  ‘she washes’, and the modal vowel followed by a glottal stop [eʔ], in [jikféʔ] ‘my ear’, all pronounced by FR. Periodicity differences between these three phonation types can be observed. While the pitch cycles for the modal vowels are quite regularly spaced (a), the waveform for the glottalized vowel displays irregularly spaced pulses (b). In contrast, no vibration and no pulses are present in the waveform of a glottal stop (c).       84  a. MODAL    b. CREAKY     c. GLOTTAL          Figure 3.2 Waveform of Nivaĉle voice qualities: modal [e], creaky [e ̰], and [eʔ], from male speaker FR.  Periodicity can be calculated as the relative absence of jitter, the latter referring to the presence of irregularly spaced vocal pulses, or the variation in the duration of successive f0 cycles, which translates into the characteristic auditory impression of creak (Laver 1980:124) as “a rapid series of taps, like a stick being run along a railing” Catford (1964:32).33 Jitter, also known as pitch perturbation, has been used as a parameter to establish differences in phonation types (Gordon & Ladefoged 2001). The increased length of the pitch cycles is indicative of a lowered fundamental frequency (the acoustic                                                 33 Laver (1980:127) refers to jitter as ‘aperiodicity of the fundamental frequency’, Ní Chasaide and Gobl (1997:41) as ‘random variation in F0’.  85 correlate of the perceptual property of pitch) for creaky voice relative to modal voice (Gordon 2001). In the following section, I discuss the manifestation of these parameters in the Nivaĉle glottalized vowels. 3.2.2 The variable realization of Nivaĉle glottalized vowels   On the basis of my fieldwork, I observed two basic categories of realization of Nivaĉle ‘glottalized’ vowels: the first, I call a rearticulated/creaky vowel, represented variably as [Vʔv ̰] (careful speech) ~ [V ̰] (fast speech), and, the second I will refer to as a vowel-glottal coda, represented as [Vʔ].    It has been noted that the implementation of ‘glottalized’ vowels is subject to variation within and between speakers across languages (Gordon & Ladefoged 2001, Avelino 2004, Peterson 2004, Gerfen & Baker 2005, Picanço 2005, Munro, Lillehaugen & Lopez 2008).34 The Nivaĉle glottalized vowels follow this pattern. What I refer to as rearticulated/creaky vowels tend to consist of a modal vowel portion followed by either: (i) a glottal closure released into a short voiceless or creaky vowel [Vʔv ̥] ~ [Vʔv ̰] or (ii) a period of glottalization/laryngealization or creak [VV ̰], respectively. Given the latter description, it is worth mentioning that cross-linguistically ‘glottalized/laryngealized’ and ‘creaky’ vowels are not necessarily interchangeable tems. For instance, Blankenship (2002) makes a distinction between vowel laryngealization and creaky phonation. She points out that Mazatec contrastively laryngealized vowels do not consistently have an audible creak or display irregular glottal pulses on a spectrogram (Blankenship 2002:164). As will be later shown, this is also the case for some realizations of Nivaĉle glottalized vowels: aperiodicity is not always present in the signal (cf. Figure 3.16). However, while aperiodicity is normally present, a consistent acoustic difference between modal and glottalized vowels is one of duration.35 One further point is that the description in (i), i.e. “a full short glottal closure released into a short voiceless or creaky vowel [Vʔv ̥] ~ [Vʔv ̰]” is similar to what are                                                 34 For instance, gender has been noted as a factor in the realization of phonation types. Gordon and Ladefoged  (2001:10), and Munro, Lillehaugen and López (2008:35) report that creaky vowels produced by Quiaviní Zapotec men sound creakier than those produced by women. Speech rate has also been correlated to variation in the implementation of phonation types (Esposito 2003, Picanço 2005:37). 35 This consistent durational realization lays a strong foundation for positing that the glottal stop is associated with a mora (cf. Chapter 4, §4.3).  86 sometimes referred to as echo vowels. An echo vowel has the same vowel quality as the vowel preceding the glottal stop (Gerfen & Baker 2005:312) but its formants are weaker.   The alternation between (i) [Vʔv ̥] ~ [Vʔv ̰] and (ii) [V ̰], is, according to my fieldwork research, mostly due to speech style factors. The words containing glottalized vowels were recorded in a different set of contexts: in isolation (the word was repeated five times), in sentences, and in conversations/narratives. Whereas the rearticulated variant [Vʔv ̰] ~ [Vʔv ̥] tokens typically occurred in careful speech, the creaky variants, [V ̰], were typically found in fast or casual speech tokens. Yet overall, despite the variability in the production of Nivaĉle glottalized vowels, these vowels involve a sequencing of modal phonation and laryngealization, similar to Coatzospan Mixtec laryngealized vowels (Gerfen & Baker 2005).   There are two important observations about the distribution of the Nivaĉle glottalized vowels. First, they never occur in an unstressed context. Secondly, in the case of rearticulated vowels, stress is consistently realized on the first, not the second (or “rearticulated”) portion of these sequences. Based on this observation, I claim that the “rearticulated” portion does not constitute a second, separate syllable. Rather, these “rearticulated” vowels constitute a single complex bimoraic syllable nucleus (§3.4). The diverse realizations of the rearticulated vowels are generally represented in the data presentation throughout this thesis as [V ̰], unless the more specific phonetic realization is particularly relevant to the discussion at hand.   Figures 3.3 and 3.4 illustrate the alternative realization of rearticulated and creaky vowels, respectively. Note that both the rearticulated [Vʔv ̰] and the creaky variant [V ̰] have approximately identical duration: 200 ms.  87  Figure 3.3 Waveform and spectrogram of [k ͡lóʔo ̰̰p] ‘winter’ by male speaker MV.   Figure 3.4 Waveform and spectrogram of [k ͡lo ̰́p] ‘winter’ by male speaker MV.  88 In Figure 3.3, three different phases can be clearly observed: modal phonation followed by a glottal closure, followed by aperiodicity in the glottal pulses, which translates into a creaky and (and lower amplitude) vowel. Figure 3.4 shows an initial period of modal phonation followed by aperiodicity.   Let us turn to an acoustic consideration of what are referred to as the Nivaĉle “vowel-glottal coda” cases. Recall that these are represented as [Vʔ], and occur when there is no (other) coda consonant in the syllable.  Figure 3.5 Waveform and spectrogram of [fajxoʔ́] ‘charcoal’ by female speaker CS.  As seen in Figure 3.5, a vowel-glottal coda consists of a modal vowel portion followed by a full glottal closure. The last part of the vowel can be creaky due to the adjacency with the glottal stop. Besides aperiodicity, Gordon and Ladefoged (2001) propose a number of acoustic properties that distinguish between modal and non-modal phonation types, specifically: acoustic intensity, spectral tilt, fundamental frequency, formant frequencies, duration, and airflow. From the analyses of the Nivaĉle data I collected in the field, duration is a relevant acoustic property that merits discussion, which I turn to in the following section.  89 3.2.3 On the relationship between [c.g.] and the acoustic parameter of duration At the interface of phonology and phonetics lie issues of contrastiveness and discreteness, and  articulatory complexity, gestural overlap, and variability. In other words, the complex interplay between distinctive features and their physical manifestation in speech have posed long standing questions at the intersection of these fields (Pierrehumbert 1990, Ladd 2014).  A major endeavor in phonological theory has been the postulation of distinctive features as primitive components of segments and phonological patterns in sound systems (Trubetzkoy 1939, Jakobson, Fant & Halle 1951, Chomsky & Halle 1968, Schane 1973, Kenstowicz & Kisseberth 1979, Clements 1985, Clements & Hume 1995, among others).  According to a universalist approach (Chomsky & Halle 1968, Clements 1985), all phonological contrasts and sound patterns are described by a set of universal features that are provided by Universal Grammar. By contrast, an emergent feature approach (Mielke 2008 [2004], 2005) posits that features are abstract categories based on generalizations that emerge from phonological patterns. In other words, different phonetic properties can be relevant for defining sound patterns, and as such, some degree of cross-linguistic variation, or subtle phonetic differences between languages, is expected. In actuality, these are some of the challenges posed for the universalist approach: “Ladefoged (1984) observed that many facts of language-specific phonetics are consistent within a given speech community, but are not explainable from universal principles of phonology or phonetics” (Hume & Mielke 2006: 729). The example under discussion in Hume & Mielke (loc. cit.) pertains to vowel systems of Italian and Yoruba. Even though these languages have the ‘same’ seven vowels /i e ɛ a o ᴐ u/, the Yoruba vowels are not as evenly distributed as in Italian. Ladefoged (1984: 85-86) attributed this variability to different lip shapes and mouth opening between the two groups, what Ladd et al. (2008) refer to as individual biases.36 It can be understood by these observations, that the correspondence between phonological and phonetic                                                 36 By individual biases Ladd et al. (2008:118) mean “anything in a given individual’s genetic makeup that somehow inclines the individual to acquire, perceive and/or produce a given linguistic phenomenon in preference to some alternative. Such biases could include a range of cognitive/perceptual and anatomical/physiological factors”.  90 descriptions are not necessarily based on a one-to-one correspondence in terms of a postulated set of universal features. The problem laid out by Ladefoged can be also related to the one pointed out by Ladd (2014: 30-31): that phonological theory has been grounded on a specific theory of phonetics, namely,  ‘systematic phonetics’. Ladd states that systematic phonetics – “embodied in the principles of the International Phonetic Association” – is based on two premises: “the segmental idealization and the universal categorization assumption” (31). The goal of systematic phonetics is to provide a “universally valid taxonomy of speech sounds” (Ladd 2014:41). Such a goal is challenged when considering, for instance, the case of Kera (Ladd 2014:42-43). This Chadic language has been described as having a voicing distinction in stops, three distinctive tones, and a number of co-occurrence restrictions between laryngeal features of stop consonants and pitch properties of the tonal distinctions. For example, voiced stops predominantly occur before low tone. However, it has been shown “that VOT is extremely variable in all stops, and co-varies with pitch” (43). In sum, even though VOT is not distinctive in Kera, VOT is, in fact, one of the phonetic cues to the phonological category of tone.   A similar situation can be observed in Nivaĉle. One of the major proposals in this chapter is that a Nivaĉle glottalized vowel consists phonemically of a vowel plus glottal sequence, where the glottal is lexically specified for [c.g.]. As mentioned in §3.2.1, the post-vocalic [c.g.] feature is manifested as aperiodicity in the signal or as a full glottal closure. In addition, a consistent acoustic difference between modal and glottalized vowels is one of duration. Even though duration is not lexically distinctive in the Nivaĉle vowel inventory, it can be posited that duration is one of the phonetic cues to glottalization in as much as VOT is one of the phonetic cues to tone in Kera. In this vein, even though I am proposing that the glottal stop can be phonologically understood in terms of the [c.g.] feature and the association with a mora (cf. Chapter 4), the phonetic manifestation is not a one-to-one correspondence between phonemic (128a) and featural (128b) representations, and phonetic reality (128c).   91 (128) a.     /Vʔ/                  b.                         µ                               µ                               |                     |                                                                    .                   .                 /  \                        /              [+son]   [-cons]                 LAR                                                                |                                                  [c.g.]  c.                                   /CVʔC/  With these observations in mind, let us turn to a consideration of the phonetic manifestation of duration as non-modal phonation types have been commonly associated with longer duration relative to modal phonation types (Gordon & Ladefoged 2001:18, Blankenship 2002:185,189). For instance, contrastive creaky vowels in Jalapa Mazatec (Kirk et al. 1984, Kirk et al. 1993, Silverman et al. 1995, Silverman 1997) and glottalized vowels in Chuxnabán Mixe (Jany 2007) are longer than corresponding modal vowels. However, laryngealized vowels in Coatzospan Mixtec are reported to be shorter than their modal counterparts (Gerfen & Baker 2005: 321). Any claim about durational differences should take into consideration both vowel quality and the phonological status of the vowels under study. In this sense, Gerfen & Baker (2005:329) conclude that, cross-linguistically, laryngealized vowels can be longer, shorter, or equal in duration to their modal counterparts. In Nivaĉle, although glottalized vowels are consistently and significantly longer than their modal counterparts, vowel duration before a glottal coda is variable.  Similarly, non-modal phonation has been documented as being associated with phonemic long vowels but not their short counterparts, e.g., in Hupa (Golla 1977, Gordon 1998:101, Gordon & Ladefoged 2001:18). Romero-Méndez (2008:49) states that length and laryngeal features are intertwined in Ayutla Mixe. There are two types of glottalized vowels in Ayutla Mixe: [Vʔ] is a short modal vowel followed by a glottal constriction and [VʔV] is a long vowel with medial constriction (141).  It has also been observed that creaky phonation is usually confined to a portion of the vowel (Silverman 1997, Gordon 1998, Gordon & Ladefoged 2001) due to perceptual reasons; non-modal  92 vowels are less perceptually salient than modal vowels.37 The presence of a modal voiced portion serves to enhance the salience of a non-modal vowel (Gordon 1998) and/or to manifest tonal contrasts (Silverman 1997).   Interestingly, it has been noted that non-modal phonation may be associated with prosodic properties, and, more specifically, with stress. Gerfen (1996:130) posits a strong correlation between vowel glottalization in Coatzospan Mixtec and stress, more specifically, he claims that glottalized vowels are licensed by stress.38  One of the central claims in this chapter is that there is a relationship between stress, duration and the optimal acoustic context for the realization and perception of glottalization in Nivaĉle vowels [VʔV]̰. In order to test this hypothesis, five repetitions of each of the following words were recorded with six Nivaĉle speakers.                                                             37 To the best of my knowledge, modal voice always precedes creaky voice.  38 In contrast, Macaulay & Salmons (1995:54) do not associate glottalization in Mixtec with stress.  (129)  a. [ʔ]ís        ‘nice’ b. ʔí ̰s        ‘write!’  (130)  a. jitéx    ‘carob’        b. jité ̰̰x    ‘grass’     (131)  a. ɬ-sa ́ʃ     ‘mucus’ b. ɬ-sáʃ̰      ‘his wool’   (132)  a. k ͡lɑ́p    ‘fast’ b. k ͡lɑ̰́p     ‘to be seated on the lap’  (133)  a. k ͡lo ́p    ‘white/larva’ b. k ͡lo ̰ṕ ‘winter’  (134)  a. ji-f.xu ́x   1POSS-stick ‘my toe’ b.  ji-f.xu ̰́x   1POSS-stick ‘my stick’  93 The words were recorded in isolation and measurements were done in Praat for Mac (version 5.3.08; Boersma & Weenink 2014). Results were compiled and statistics were run in R for Mac (R Core Team 2013). Figure 3.6 presents the durational differences for the modal and rearticulated/creaky vowels as produced by male speakers FR, FAR, GR, and females speakers CS, TS, and RF.  Figure 3.6 Duration results for creaky vs. modal vowel pairs across six Nivaĉle speakers: FAR, FR, GR (male) and CS, RF, TS (female)  A repeated measures ANOVA was conducted with duration as the dependent variable and vowel quality as within-speaker variable with six levels /i e a ɑ o u/ and glottalization as another within-speaker variable with two levels (plain vs. glottalized). The analysis yields significant main effects of vowel quality (F(5, 25) = 7.99, p < 0.001) and glottalization (F(1, 5) = 119.5, p < 0.001). The analysis did not show a significant interaction of vowel quality and glottalization (F(5,25) = 0.552, p = 0.73).  94 In summary, this study shows that duration is a statistically significant acoustic property that differentiates modal vs. glottalized contrast within the Nivaĉle vowel system; creaky vowels are roughly twice as long as their modal counterparts.   The duration of a vowel before a glottal stop, that is, what I call vowel-glottal coda (cf. Figure 3.5) has not yet been systematically studied in a comparable way; word shape and size of forms containing Vʔ (vowel + glottal stop)  and VT (vowel + oral stop) are factors that could not be rigorously controlled during fieldwork and certainly merits future investigation. From an impressionistic point of view, the duration of a vowel in a [Vʔ] context is not as long as the duration of a creaky/rearticulated vowel, but appears to be slightly longer than a vowel before a stop [VT]. Let us examine some data: Table 3.2  Comparative duration of Vʔ and VT  Speaker Token Vowel Mean (ms) SD ji-β.ɬíʔ ‘my rib’ (n=6) 76.35  20  TS  xa-f.ɬít ‘I blow’ (n=6)  i  58.50  4 ji-k ͡lɑ́ʔ ‘my toy’ (n=6) 80.15  14  xa-k ͡lɑ́t ‘I escape’ (n=6)   ɑ  69 10 ɬaβoʔ́ ‘resident’ (n=6) 93  25  FR  toβo ́k ‘river’ (n=6)   o 67 3.6  If we compare the durational differences between [Vʔ] and [VT] presented in Table 3.2, a trend can be noticed at this point: a vowel in a vowel-glottal sequence appears to be longer than a vowel followed by an oral stop.     95 3.3  Feature specification of Nivaĉle laryngeals In this section I present several arguments linking the phonetic attributes and phonological behaviour of these glottal/glottalized realizations to the underlying presence of a glottal stop /ʔ/ which, I propose, is featuraly specified as [c.g.]. A further claim (anticipating the detailed discussion in §3.4) is that /ʔ/ is consistently moraic in post-vocalic tauto-syllabic position. 3.3.1 On the ambiguous behavior of glottal stop Cross-linguistically, glottal stops often pattern differently from supralaryngeal consonants. This asymmetry has been mostly characterized in terms of different featural configurations, namely, that glottals are placeless or do not have an oral articulator (Steriade 1987, Cohn 1990, Bessell & Czaykowska-Higgins 1992, Buckley 1994, Rose 1996, Ola Orie & Bricker 2000, Broselow 2001). Some phonological patterns particular to the glottal stop are laryngeal transparency to the spreading of vocalic place features (135a) or nasalization (135b); debuccalization of final stops and fricatives (136); and epenthesis/hiatus-resolution processes (137, see also Shaw (1991) and Borroff (2007).  (135) Arbore (Cushitic) a. /ɡereʔa/   [ɡereʔe] ‘it is a belly’              (Steriade 1987)       Sundanese (Austronesian) b. /niʔis/ [niʔ̃i ̃s]  ‘relax in a cool place’   (Cohn 1993)                    (136) Kelantan (Austronesian) a. /ʔasap/ [ʔasaʔ]  ‘smoke’    (Trigo 1991:124)  b. /kilat/ [kilaʔ]  ‘lightening’ c. /balas/ [balah]  ‘finish’  (137) Malay (Austronesian)   a. /di-daki/ [didaki]  ‘to climb [PASS]’                (Lombardi 2002: 228)  b. /di-ukir/ [diʔuke] ‘to carve [PASS]’  In some cases, glottal stops have been treated as a type of pharyngeal (McCarthy 1991). Following McCarthy (1994), Lombardi (2002:221) adopts the hypothesis that glottal stops have  96 pharyngeal place and extends the Place Markedness hierarchy (Prince & Smolensky 1993) by adding PHARYNGEAL as the least marked place: *DOR, *LAB » *COR » *PHAR. This representation would then, according to Lombardi, account for the unmarked status of the glottal stop and its role in the aforementioned phenomena of transparency, neutralization and epenthesis. According to Rose (1996) the characterization of laryngeals as bearing pharyngeal place, though, depends on the presence of guttural consonants (i.e. pharyngeals and uvulars) in the phonemic inventory of the language in question. Note, however, that a non-guttural system like Yucatec Maya has placeless laryngeals /h ʔ/, but also another laryngeal “h2” specified for dorsal place that is the product of historical change: *x > h2 (Ola Orie and Bricker 2000). These authors find that, on the one hand, Yucatec Maya laryngeals /h ʔ/, unlike other consonants in the language, can take part in processes of transparency, deletion, debuccalization and epenthesis. They thus behave like placeless consonants. On the other hand, Yucatec Maya “h2” can resist deletion (140), and thus behaves like consonants specified for PLACE (140). Whereas in (138), deletion targets the [j] consonant of the affix /uj-/, in (139) deletion targets the initial [ʔ] or [h] consonant of the root: the hypothesis is that these /ʔ/ and /h/ laryngeals are not specified for PLACE. In contrast with the examples in (139), note in (140) that even though the first consonant of the root is a laryngeal [h], deletion targets the consonant of the prefix rather than this [h]. Because this surface [h], called “h2” to differentiate it, behaves like a consonant with a PLACE node (cf. the consonant-initial roots in the data of (138)).   Input    Output  Gloss (138) uj-sa ̀ak’   u-sa ̀ak’   ‘the itchy one’  uj-kìik    u-ki ̀ik   ‘his older sister’  (139) uj-ʔal     uj-al   ‘the heavy one’  uj-heʔel kʼuʔ   uj-eʔel kʼuʔ  ‘the nest’s egg’  (140) uj-ha ̀ah    u-ha ̀ah   ‘the true one’   97  The dual behavior of /h/ found in Yucatec Maya gives further support to the proposal first made by Shaw (1991): a laryngeal consonant may have two different representations in the same language.  Another facet of the complex status of glottal stops is that they have been analyzed variously as (i) segmental or (ii) suprasegmental phenomena. When considered full segments, glottals have been treated as obstruents (Ladefoged 1971, Hyman 1975, Bessell 1992) or sonorants (Chomsky & Halle 1968) and so patterning with glides (Kenstowicz & Kisseberth 1979, Kavitskaya 2002). In addition, there has been debate as to whether glottal stops are [+consonantal] (Hyman 1985) or not (Hume & Odden 1996).  Two other kinds of patterns have led to the analysis of glottals as suprasegmentals: specifically, glottal stops may be implemented as creaky phonation overlaid on the realization of other segments, and underlying creaky phonation may be realized as glottal stop (Avelino 2004:181). For instance, whereas the glottal stop of Yatzachi Zapotec is sometimes realized as creakiness on the surface (as a prosodically conditioned variant realization), other related languages – Jalapa Mazatec, Comaltepec, Chinantec and Copala Trique – simply have phonemic creakiness with no surface glottal stop realization (Borroff 2007:39). To recapitulate the discussion in §3.2.3, there is not a necessary one-to-one correspondence between phonemic representations and phonetic reality. 3.3.2 Phonological patterning of [ʔ] Let us turn, then, to an investigation of the phonemic status and feature specification of glottal stop in Nivaĉle. In terms of morphological and prosodic distribution, [ʔ] patterns with plain stops in that both occur in onsets and codas throughout the word domain.  In contrast, the series of ejective obstruents /p’ t’ k’ t ͡sʼ tʃ͡ʼ/ can only occur in onset position. They are never attested in coda position. Interestingly, if an independently motivated morphophonemic process in the language (i.e., vowel-consonant metathesis, cf. Chapter 6) results in an underlying  98 ejective obstruent being parsed into coda position (141a), it systematically neutralizes to its plain counterpart (141b):  (141) a.  a.p’ax   ‘yarara’  b.  ap.xa-s   b’. *ap’.xa-s   yarara-PL   ‘yararas’  The conclusion then (which will be discussed further in §3.3.3.3), is that glottal stop and ejective obstruents do not pattern together in coda position. Specifically, ejective stops and affricates are not allowed in coda position (§6.5.1). However, in onset position, glottal stop patterns with ejectives. Not only do both freely occur as onsets (as previously shown in §2.3), but also there exists a non-adjacent glottal dissimilatory process whereby any rearticulated/creaky vowel surfaces as deglottalized when the onset of the immediately following syllable consists of either an ejective or a glottal stop.  Glottal dissimilation (142) a.  [k ͡lat ͡sʼu ̰́s]     /k ͡lat ͡sʼuʔs na=kot ͡sxaʔt/    slippery DET=ground   ‘the ground is slippery’   b. [k ͡lat ͡sʼúsʔe]        /k ͡lat ͡sʼuʔs-ʔe     na=ji-xpɑjit ͡ʃ /   slippery-LOC(ON)    DET=1POSS-house   ‘it is slippery on the house ’   c.  [k ͡lat ͡sʼústʃ͡ʼə]        /k ͡lat ͡sʼuʔs-tʃ͡ʼe     na=nɑjiʃ/   slippery-LOC(AROUND)    DET=road   ‘it is slippery around the road’ cf.     99  d.  [k ͡lat ͡sʼú ̰sʃi]            /k ͡lat ͡sʼuʔs-ʃi     ɬa=wiʃini/   slippery-LOC(INSIDE)    F.DET=lagoon   ‘it is slippery inside the lagoon’   (143) a.  [βɑ̰̀ɬjɪt ͡ʃʼe ́]          /∅-βɑʔɬ-ji-t ͡ʃʼe     ɬ-pa=siwɑk ͡lɑk/   3S-walk-1O-LOC(AROUND)   F-DET=spider    ‘a spider is walking on/around’       cf.  b. [t ͡siβɑ́ɬt ͡ʃʼə]         /t ͡si-βɑʔɬ-t ͡ʃʼe     ɬ-a=siwɑk ͡lɑk/ 3A.1P-walk-LOC(AROUND)   F-DET=spider  ‘a spider is walking on/around (me)’         (144) a.  [k ͡lapʼa ̰́f]  /∅-k ͡lapʼaʔf/   3S-bald   ‘he is bald’  b. [k ͡lapʼa ́fʔe]   /∅-k ͡lapʼaʔf-ʔe    na=ɬ-tako/   3S-bald-LOC   DET=3POSS-face   ‘his face is bald’  (145) a. [xak ͡le ̰ʃ́] /xak ͡leʔʃ   na=titetʃ͡/       1S-clean   DET=plate       ‘I clean the plate’  b.  [xak ͡leʃ́tʃ͡ʼə]      /xak ͡leʔʃ-tʃ͡ʼe   na=ji-jeʔs/            1S-clean-LOC(AROUND)  DET=1POSS-hair        ‘I wash my hair’      100 (146) a.   [xajβu ̰́j] /xaj-βuʔj/ 1S-cover ‘I cover myself’   b.  [xajβu ́jʔin] /xaj-βuʔj-ʔin/ 1S-cover-IPFV        ‘I cover myself repeatedly’  (147) a.  [ɬantʃ͡a ̰́xjij] /ɬa-n-tʃ͡aʔx=ji-j    na=jukuβe/ 2S-CISL-take=1O-DIST   DET=bread ‘you bring me bread’  b.  [xantʃ͡áxʔaj] /xa-n-tʃ͡aʔx=ʔa-j   na=jukuβe/ 1S-CISL-take=2O-DIST   DET=bread ‘I bring you bread’   (148) a. [xaβku ́ʔxi]   xa-wkuʔ-xi    na=hamaca  1S-swing-LOC(INSIDE)   DET=hammock      ‘I swing/swang in the hammock’ cf. b.   [xaβku ́ʔin]       /xa-wkuʔ-ʔin/  1S-swing-IPFV  ‘I am swinging’  Unlike other deglottalization patterns explained in Nivaĉle that systematically occur when a syllable containing a glottalized vowel or a vowel-glottal coda is unstressed (see §2.6.5, §4.5); the above data above show that this Glottal Dissimilation process regularly occurs even in stressed syllables. The trigger in this case is an ejective or glottal stop onset in the following syllable. Compare the different effect on a non-ejective onset in (142d, 147a, 148a).   101 A further interesting property of this Glottal Dissimilation process, as illustrated in the data above, is that the trigger is not necessarily directly adjacent to the target, for another consonant can intervene between them. For example, in /k ͡lat ͡sʼuʔs-ʔe/, the /uʔ/ target is separated from the following onset /ʔ/ trigger by an intervening (coda) /s/, and yet the deglottalization takes place nonetheless, resulting in the output [k ͡la.t ͡sʼús.ʔe] (where [.] marks syllable boundaries). From the available data {f, t, s, ɬ, ʃ, j, x} are attested as possible intervening consonants, that is, labial, coronal and dorsal consonants. Of great significance is the directionality of this Glottal Dissimilation process, in that there can be an ejective in the onset immediately preceding a glottalized vowel, as in (142d) [k ͡lat ͡sʼú ̰s-ʃi], and it does not trigger deglottalization.  The generalization that arises from this puzzling phenomenon is (i) a post-vocalic glottal stop is the target and (ii) either another glottal stop or an ejective in the following onset is the trigger. This can be interpreted as a type of Obligatory Contour Principle (OCP) effect (Odden 1986) whereby there cannot be two adjacent [c.g.] feature specifications (cf. (148b) or two [c.g.] feature specifications across a segment specified for [+cons]:   (149)   .            ( .  [+cons] )   .             /    |    |       LAR             PLACE              LAR           |        |                      [c.g.]                 [c.g.]    In contrast, there can be two [c.g.] feature specifications across a segment that is not specified for [+cons], that is, in cases where there is an intervening vowel, as the following examples illustrate:  (150) a.  tʃ͡ʼetʃ͡ʼe      T’VT’ ‘parrot’  b. xa-p’ó-ʔej na=ɬaʃí  T’Vʔ    1S-close-DIR DET=door   ‘I close the doorʼ   102  c. j-i-ʔeʔ́  na=ɬa-x.pɑ́.jitʃ͡  ʔVʔ   3S-be-LOC   ‘s/he is at home’  To summarize the generalizations that have been established in this section about how ejectives and glottal stop pattern phonologically, it has been shown that glottal stop patterns with all consonants in its distributional freedom of occurrence in (simplex) onset and coda positions.39 In onset position, glottal stop patterns with ejectives as a trigger for Glottal Dissimilation. In coda position, glottal stop does not pattern with ejectives: like plain stops, a glottal stop can occur in a coda, but an ejective cannot. 3.3.3 Glottal stops and syllable structure In the literature on glottalized vowels in other languages, most of the arguments against treating the glottal stop as a phonemic segment rely on its defective distribution, e.g., the glottal stop may be the only coda in a language, e.g. in Mixe (Macaulay & Salmons 1995), and/or the glottal stop may not occur or be contrastive in initial position (e.g. in Quiaviní Zapotec; Chávez-Peón 2010). As will be shown in the following sections, the Nivaĉle glottal does not fit this picture; [ʔ] is contrastive in onset position and I will argue that, importantly, it can be parsed to coda position.  3.3.3.1 Glottal stop as an epenthetic onset An initial question related to the interplay of syllable structure constraints and the role of epenthetic glottal stop is whether onsetless syllables ever occur in Nivaĉle. Stell (1989:116-117) claims that V syllables are licit syllable structures in Nivaĉle; she illustrates her point through the following examples: (151)  o-sej-k ͡la   ‘prickly pearʼ (152)  ɬa-n-ku-a   3S-DIR-desire-3O   ‘he desires (s.t.)’                Stell (1989:116-117)                                                 39 It will be recalled from Chapter 2, Table 2.4, that neither glottal stop nor ejectives have freedom of occurrence in word-initial onset clusters. Neither can occur as C1 in such clusters. With respect to C2, there are no occurrences of glottal stop and only three attested examples with an ejective.  103 Based on the data gathered in the context of my own fieldwork, I will argue that onsetless syllables are neither allowed at the beginning (151), nor inside of the word (152); the constraint ONSET is undominated (§2.3.1). An epenthetic glottal stop is inserted to ensure satisfaction of this constraint, thus violating DEP-IO-ʔ:    (153)  [ʔ]osejk ͡lá   ‘prickly pearʼ (154)  ɬan-kú=[ʔ]a   3S-desire=3O   ‘he desires (s.t.)’  (155) ONSET » DEP-IO-ʔ   (156)         /osejk ͡la/ ONSET DEP-IO-ʔ  a.      osejk ͡la *!   b.  ʔosejk ͡la  *   The allomorphic alternation between [kas-] and [kats-] also provides an argument for treating the glottal stop in data such as (157a) as epenthetic. In her grammar, Stell (1989) presents three allomorphs for the first person possessive prefix; these are claimed to be phonologically conditioned (§2.5): Table 3.3 First person plural inclusive possessive prefixes kas- before CV-initial roots kat ͡s- before V-initial roots kat ͡si- before CC-initial  roots        Because [kat ͡s-] – not [kas-] – is prefixed to the root in cases like (157b), it must be inferred that there is no underlying glottal stop in root initial position:   104 (157) a. [ʔ]a.si.nɑ ́      speech-BEN      ‘word/speech’  b.  ka.t ͡s-a ̀si.nɑ ́ 1POSS-speech       ‘our speech’  c.  *kas-ʔasi.nɑ  1POSS-speech      ‘our speech’   Further, Stell notes a series of allomorphic alternations involving glottal stops. There exist a number of suffixes that alternate between being vowel-initial and [ʔ]-initial, as seen in the (d)-(e) and (f)-(g) pairs in (158), as well as parallel alternations between the vowel-final and [ʔ]-final prefixes, as in the (a)-(b) pairs. Rather than treating such cases as allomorphic alternations, I treat them as phonologically-governed alternations. For example, if the root for ‘love’ is posited to be V-initial, /en/, rather than glottal-initial, then the surface occurrence of [ʔ], and the /x-/ form of the first person subject follow as phonological generalizations. (158) Morpheme boundary epenthetic onset  a.  ni-n-fós       NEG-3S-bury    ‘s/he does not not bury’ b.  ni[ʔ]-én                cf.     c.    x-én     NEG-love                              1S-love     ‘you do not love’                  ‘I love’  d.  ni-j-e ́n-eɬ     NEG-1S-love-PR.PL     ‘we do not love’  e.  x-ɑ.t ͡sí-[ʔ]eɬ     1S-pour-PR.PL     ‘we pour’                    Stell (1989:258) f.  xa-̀t-pek ͡l-éj     1S-DIR-return-DIR      ‘I return’ g.  xa-peʔ̀-ja-[ʔ]e ́j     1S-hear-PUNCT-DIR      ‘I hear (from the distance)’  What is seen here is that vowel sequences that arise through morpheme concatenation are systematically avoided by epenthesis of a glottal stop. From the perspective of syllabification of the segmental sequence, this [ʔ] functions to provide an onset for the otherwise onsetless vowel-initial  105 syllable. To illustrate this, consider the syllabification of the form for ‘you do not love’ in (158b) with – as opposed to without – the epenthetic [ʔ]: (159)              σ          σ                                    σ        * σ              N         N      N           N         n   i     [ʔ]  e  n n    i           e   n  What has been argued in this section is that there are a diversity of morphophonemic alternations in Nivaĉle which receive a more systematic interpretation within an analysis that recognizes a role for epenthetic glottal stop. There are two basic contexts in which [ʔ] can be epenthesized to repair ill-formed surface sequences. One is to provide an ONSET to all otherwise V-initial words (§2.3.1). The other is to avoid a word-internal sequence of two vowels in a row, *VV, as in (159). The further question then is what kind of evidence can be adduced for whether a surface [ʔ] in either of these contexts is underlying. This is addressed in the next section. 3.3.3.2 Non-epenthetic glottal stop onset  Non-epenthetic glottal stop can occur clitic/suffix-initially (160a), (161), and (162). A crucial piece of evidence for the phonemic status of glottal stop in onset position is the contrast between the second person object /ʔa/ (160a) and the third person object /a/ (160b):  (160) a. k’-uʔ-éʃ=ʔa             1S-believe-INST=2O          ‘I believe in you’  b.    k’-uʔ-éʃ=a    pa=fit ͡sɑkʼɑjit ͡ʃ  1S-believe-INST=3O  DET= God        ‘I believe in God’  Other grammatical suffixes such as the locative [-ʔe] and the iterative/imperfective [-ʔin] consist of a lexically specified glottal stop before the vowel. In contrast with the directional /-ej/ in (158d, f), when these suffixes get attached to a consonant-final root, the glottal stop of the locative  [-ʔe] and the  106 iterative/imperfective [-ʔin] is parsed into onset position. Also note that, similarly to the examples in (142)-(148) there is glottal dissimilation in examples (161) and (162).    (161) [t’a-kúm-ʔɪn]             aʼ. *t’a.ku.min    /tʼa-kuʔm-ʔ in/       3S-work-IPFV            ‘He is working’.     (162) a.  kla.t ͡s’ús-ʔe       aʼ. *kla.t ͡s’u.s-e  slippery-LOC  b. nɑke x-an-ʔé  naβa=ji-tɑ̀s.xe-ʃi.j-[í]s    bʼ. *xane   here 1S-put-LOC DET.PL=1POSS-eye-REC-PL   ‘I put the glasses hereʼ  cf. (158d,f) repeated here for convenience:  (163) a. ni-j-én-eɬ                          NEG-1S-love-PR.PL      ‘we do not love’     (Stell 1989:258)   b. xa ̀-t-pek ͡l-e ́j ̤ ̤   na=Filadelfia      1S-CISL-return-DIR  DET=Filadelfia        ‘I return to Filadelfia’  As seen in Figure 3.7 below, the presence of the suffix-initial [ʔ] from the example in (161) shows clearly in the waveform as aperiodicity and low amplitude in the signal.    107       Figure 3.7 Waveform and spectrogram [t’akúmʔin] ‘s/he works’ by female speaker TS.  During fast speech, the ʔ-initial suffix overlaps with the articulation of a preceding non-continuant obstruent, e.g., a root-final stop, such that an ejective stop results:   (164)  ji-ʔé  na=kot ͡s.xát-ʔe  ~  kot ͡s.xá.t’e  be-LOC  DET=land-LOC  ‘It is on the land’   Below, Figures 3.8 and 3.9 show the alternation between the forms in (164). Note the long glottal stop closure in Figure 3.8; this is characteristically found in a very emphasized stop-glottal sequence in a citation context.  108  Figure 3.8 Waveform and spectrogram [kot ͡sxátʔe] ‘on the land’ by male speaker FR.  Figure 3.9 Waveform and spectrogram [kot ͡sxát’e] ‘on the land’ by male speaker FR. In the first version (Figure 3.8), FR emphasized the presence of a glottal stop in a very careful pronounciation of ‘on the land’. The second version (Figure 3.9) is characteristic of casual speech. The important point here is that the glottalization that is realized as either [tʔ] or [t’] can only result from  109 there being a phonemic /ʔ/ in the input; [kot ͡s.xa.te] is not attested as a possible output. The alternative hypothesis that the root-final consonant is an ejective stop /kot ͡sxaʔtʼ/ is not plausible as it would not account for the […tʔ…] realization. Nor would it account for the fact that in other contexts when an epenthetic vowel is inserted, as shown in (165), there is no ejective in the output.   (165) a.   kot ͡sxat-[í]s      land-PL            ‘lands’           b. *kot ͡sxat’-[i]s  Besides serving as suffix-initial onsets, the following examples show that glottal stop can also serve as a lexically-specified (i.e. non-epenthetic) root-internal onset.   (166) a.   kan.ʔút     ‘yesterday’    b.  nu.ʔú   ‘dog’  c.  ʃniɬ.ʔɑ́  ‘small lizard’  d.  mis.ʔa ́   ‘scarlet-headed blackbird’  e.  ɬu ̀m.ʔa.ʃí     ‘tomorrow’  f.  kum.ʔɑ́    cf.    f’.  kum.xɑ́  ‘crowned eagle’       ‘aloja (alcoholic drink)’  g.  am.ʔɑ́     cf.   g’. am.pá   ‘rat’        ‘nothing’    j.  ka.jin.ʔɑ́    ‘hummingbird’   k.  k ͡li.sa.ʔa ́    ‘blue-black grassquit’        (167) a.  fak ͡l.ʔu ́   ‘brother-in-lawʼ  110  b.  fak ͡l.ʔa ́    ‘nephew’ c. fak ͡l.ʔís      ‘batʼ  d.   uk ͡l.ʔɑ́   ‘type of dove’  Albeit not exhaustive, this is a representative list of cases in which the glottal stop surfaces as a root-internal onset. A closer look reveals that these might not all be considered mono-morphemic roots; the [ʔ] might in fact be morpheme-initial, especially when considering the similarity between the kinship terms and that many forms are names of animals. Nevertheless, whether the above examples involve instances of glottal stop behaving as root-internal onset or not, these data clearly support the claim that glottal stops can behave as contrastive onsets in Nivaĉle. Recall, in this regard, the minimal pairs listed in (160a-b) and (166f-f’), (166g-g’).   It is worth commenting on examples (167) where the complex segment [k ͡l] is parsed as a coda before a glottal stop onset. As seen in Chapter 2 and as will be discussed in Chapter 5, it is normally the case that the complex segment [k ͡l] consistently neutralizes to [k] in coda position. However, the only context in which [k ͡l] does not undergo this neutralization to [k] is before a tautomorphemic glottal stop (§5.3.1). This ‘exceptional’ syllabic behavior of [k ͡l] will become relevant in the discussion of the feature specification of glottal stop (Section 3.3.4).  3.3.3.3 Glottal stop as coda Let us turn now to a consideration of contexts where glottal stop can be interpreted as serving as a word-medial (168) or word-final coda (169). These are the contexts where the theoretical assumptions behind the featural representation of both glottalized vowels – more precisely, what I have previously referred to as vowel-glottal coda – and glottal stops per se will be motivated in detail.  In Chapter 2, I showed that CVC is an attested (and frequent) syllable type in the language, and, in Chaper 4, I claim that the minimal foot in Nivaĉle is CVC (§4.2). Let us investigate now the distribution of the glottal stop with respect to the final coda C in these CVC syllables. Examples in  111 (168) show that a coda containing a glottal stop can precede both obstruents and sonorants. However, it cannot precede another glottal stop or an ejective (see also §2.3.1.3, Table 2.9).  (168) Word-internal coda a. ji-pɑ́ʔ.kɑt ‘my hand’ b. xi.βéʔ.k ͡la ‘moon’ c. βeʔ́.ɬa ‘one’ d. tɑ́ʔ.ɬɑs ‘vase’ e. nɑ́ʔ.ni ‘girl’ f. xa-peʔ́.j-a   ‘I hear’  In addition, examples of word-final glottal stops are presented in (169); they occur after all vowel qualities.  (169) Word-final coda a. tʃ͡íʔ  ‘and’ b. t’íʔ  ‘broth’ c.  me ́ʔ  ‘otter’ d.  ji-k.t’eʔ́  ‘my grandmother’  d.   ji.taʔ́  ‘scrubland’ e.  kas-k ͡lɑʔ́ ‘our toy’   f.  ji-mɑ́ʔ  ‘s/he sleeps’ g.  faj.xoʔ́  ‘charcoal’ h.  tʃ͡a.ɬuʔ́  ‘short’ i.  xa-β.kúʔ ‘I swing’ j.   k’ak,xúʔ ‘ I greet you’  In sum, it has been shown that Nivaĉle glottal stop can occur in both onset and coda position;  Table 3.4 summarizes the possible syllable parsings of the glottal stop.  Table 3.4 Syllable types and glottal stop Syllable types CV CVC CCV CCVC ʔV CVʔ *CʔV *CʔVC  ʔVC *ʔCV *ʔCVC  ʔVʔ   CCVʔ  112 This broad base of distribution, parallel to other major classes of consonants, motivates the representation of /ʔ/ as having an independent root node. This allows it to be parsed into what have to this point been informally referred to as “onset” and “coda” position.  Although issues of prosodic structure will be discussed in greater detail in Chapter 4, it is useful at this point to briefly clarify the conception of syllable structure being adopted here. Of particular relevance, “onset” and “coda” are not prime concepts, but rather represent segments identifiable in relation to the left or right edge, respectively, of the prosodic domain of a syllable. The prime prosodic constituents within the syllable (abbreviated σ) are the Nucleus (abbreviated N), which functions as the Head of a syllable, and the mora (abbreviated µ), “a unit which functions variously as the prosodic host for segments with significant sonority, weight, duration, and/or tone” (Shaw, p.c.). Because Nivaĉle vowels all have weight, they are moraic, and because they have greater sonority than all other classes of segments (see §2.3.2) in the language, they are parsed to the Nucleus. Major questions to be addressed in the rest of this chapter and in Chapter 4 relate to how the various realizations of “glottal stop” are prosodically parsed.   3.3.4 Feature specification of Nivaĉle glottal stop In light of the evidence related to the phonotactic patterning of the glottal stop, I will discuss feature representation for this segment. The first hypothesis advanced here is that the glottal stop is unspecified for the place features. Three supporting arguments will be discussed:  (i) laryngeal transparency (vowel harmony across a glottal stop) (ii) parsing of the glottal stop in coda position (as opposed to ejectives, which are specified for place features), (iii) lack of delateralization of [k ͡l] before a glottal stop onset (cf. (167) above; Chapter 5) Laryngeal transparency has been advanced as an argument for the lack of internal place of articulation structure of laryngeals in comparison to other consonants; that is, for glottal stops being placeless in non-guttural systems. In autosegmental phonology terms: due to laryngeal transparency,  113 vocalic features can spread across a glottal stop because no crossing of an intervening consonantal place specification is involved (Goldsmith 1976, Sagey 1988). In Nivaĉle, spreading of vocalic features can be observed across non-epenthetic (170-172) and epenthetic glottal stops (174-175) at morpheme boundaries (§2.5.3). Specifically, progressive vowel harmony has been attested with the imperfective /-ʔin/ and the locative /-ʔe/ suffixes; as discussed in §3.3.3.2 these glottal initial morphemes are not epenthetic.  (170)       /…e-ʔin/ […e-ʔen] a.  [nixak ͡leʃ̀tʃ͡ʼəʔe ́n]   /ni=xa-k ͡leʔʃ-t ͡ʃ’e-ʔin/       NEG=1S-wash-LOC-IPFV   ‘I do not (regularly) do the cleaning’     b. [xaj-ku ̀m-ʔe-ʔe ́n]   /xaj-kuʔm-ʔe-ʔin/     1S-work-LOC-IPFV   ‘I am/was working’  (171) /…a-ʔin/ […a-ʔan] [ɬpeʔ̀jaʔa ́n]  /ɬ-peʔj ̤a-ʔin/ 2S-hear-IPFV ‘you are hearing’  (172) /…ɑ-ʔin/ […ɑ-ʔɑn]  [jifɑ̀ʔjɑʔɑ́n]  /ji-fɑʔjɑ-ʔin/  3S-fly-IPFV  ‘s/he is/was flying’    114 In (170)-(172) the high front vowel /i/ is realized harmonically as a front or back non-high non-rounded vowel across an underlying glottal stop. I have not found examples of non-low back rounded vowels triggering harmony: *[oʔon], *[uʔun]:  (173) [xaβkúʔin]   *[xaβkúʔun]  /xa-βkuʔ-ʔin/ 1S-swing-IMPFV ‘I am swinging’  Vowel harmony is also attested across epenthetic glottal stops. When two vowels are adjacent in the input due to morpheme concatenation, a glottal stop is inserted (§2.3.1.2) and there is regressive vowel harmony: the vowel following the glottal stop spreads its place features to the preceding vowel. This vowel harmony process occurs when the trigger is a [-back, -low] vowel and the target is a [+low] vowel. Examples (174)-(175) illustrate this phenomenon:  (174) a. [meʔéɬ]   /mɑ-eɬ/                IMP.go-PR.PL            ‘Go you all!’   b. [meʔéj]   /mɑ-ej/ IMP.go-DIR   ‘Go (you singular) there!’  (175) a.   [xapɛ́ʔj-a]      /xapeʔj-a/   pa=ɬaβí ̰m                 1S-hear-PUNC   DET=wind                ‘I heard the wind’      115  b.  [xapɛ̀ʔjeʔéj]    /xa-peʔj-a-ej/    pa=tʼɑ̰́j                  1S-hear-PUNC-DIR  DET=noise                 ‘I heard noise (from the distance)’   The /VʔV/ sequence presented in (175b) differs from creaky/rearticulated vowels in that the stress system (see Chapter 4) counts it as two syllables instead of one. Further, and concomitantly, the second vowel here is somewhat longer than the first one, contrary to the case of glottalized vowels:   Figure 3.10 Waveform and spectrogram of [xapɛ̀ʔjeʔéj] ‘I heard’ by male speaker FR  Interestingly, the two attested types of vowel harmony processes can be shown with the predicative verb ‘to be nearby’. On the one hand, we see progressive spreading of vowel features across an underlying glottal stop (176). On the other hand, we see regressive vowel harmony across an epenthetic glottal stop (177):         116 (176) Progressive Vowel harmony      [ʃaʔ̀ɬaʔá]   na=niβak ͡lé  ɬa-βt ͡sa ̰́t   /∅-ʃaʔɬa-ʔe/  DET=nivaĉle 3POSS-village  3-close-LOC  ‘the Nivaĉle community is nearby’   (177) Regressive vowel harmony   [ʔaʃaʔ̀ɬeʔe ́ ɬ]  /a-ʃaʔɬa-eɬ/  2S-close-PR.PL  ‘you (pl.) are nearby’   Vowel harmony occurs across a glottal stop, as seen in (177) but not across a consonant specified for PLACE, as the examples below illustrate:  (178) a. ∅-ʃaʔ̀ɬa-xuɬ́   3S-close-VEN   ‘s/he is getting close (to the deictic centre)’  b.  ∅-ʃaʔ̀ɬa-tʃ͡ʼe    3S-close-IT        ‘s/he is still close (but s/he is moving away from the deictic centre ’ (adapted from Seelwische 1990:169)  In summary, the attested cases of laryngeal transparency to vowel harmony processes thus provide support for the analysis of the glottal stop as placeless. The different patterns (progressive vs regressive) vowel harmony associated to underlying vs. epenthetic glottals merits further investigation.  Let us turn to a second argument for PLACE feature(s) not being part of the lexical representation of  glottal stop. It has been observed that both PLACE and LARYNGEAL features are often restricted in coda position (Itô 1986, Mester & Itô 1989, Itô & Mester 1994; Lombardi 1991, 1995). The lack of specification for PLACE, then, might therefore explain the asymmetric behaviour between glottal stop and ejectives in coda position. Recall from Chapter 2 that ejectives are banned from occurring in  117 coda position. In descriptive terms, the generalization appears to be that when [c.g.] is functioning as a “secondary” feature (i.e. on ejective obstruents) in Nivaĉle, it is not tolerated in coda position. A plain glottal stop, however, can – and quite pervasively does – occur as a coda. I propose a structural account for this: whereas both ejectives and glottal stop in Nivaĉle are specified with a [c.g.] feature, only ejectives are specified for place – LABIAL, CORONAL, and DORSAL. Glottal stop is not: it is literally place-less. In order to capture this different patterning, the following coda constraints can be locally conjoined:40  (179) *PLACE]σ : A segment in the coda of a syllable cannot be marked for [PLACE]  (180) *[c.g.]]σ : A segment in the coda of a syllable cannot be marked for [c.g.]  (181) [*PLACE & [c.g.]]σ:  A segment in the coda of a syllable cannot be marked for both PLACE and    [c.g.]  According to Smolensky (1993) a locally-conjoined constraint is violated if both of its conjuncts C1 and C2 are violated in a local domain D (181). 41 Whereas ejective stops and affricates (T’) in coda position would violate the two coda constraints in (179) and (180), and also the locally-conjoined constraint in (181)  – see candidate (a) in (182) – the glottal stop would only violate (180) and not (181) because /ʔ/ is not specified for PLACE in Nivaĉle (see (183a)). (182)          /T’.CV/  [*PLACE & *[c.g.]]σ *PLACE]σ *[c.g.]]σ  a.      T’.CV *! * *  b.      T.CV  *   (183)          /ʔ.CV/  [*PLACE & *[c.g.]]σ *PLACE]σ *[c.g.]]σ  a.      ʔ.CV   *  b.      T.CV  *                                                  40 I am grateful to Megan Crowhurst for this suggestion.  41 In this vein, a locally-conjoined constraint can be formally defined as follows (Lubowicz 2005:254): (i) C=[C1&C2]D is violated iff both C1 and C2 are violated in a local domain D   118 Similarly, a plain stop or affricate (T) would only violate one constraint (179), and not (181) – see candidates (182b) and (183b).   It has been established, then, that the glottal stop can be parsed as a coda, in contrast with ejectives. One supporting argument for the glottal stop being parsed to coda position is word minimality. As will be advanced in Chapter 4 (§4.2), the minimal monosyllabic word in Nivaĉle is CVC. Open syllable CV or CCV words are not attested: a well-formed Minimal Foot needs to be closed by a coda consonant.42 Given that CVC constitutes a Minimal Word in Nivaĉle (see data in (184a,b) below) and given that CVʔ words are well-formed (see data in (184c,d) below), it follows that the glottal stop is functioning as a coda consonant.  (184) a. tós ‘snakeʼ    b.   ∅-túx     3S-eat     ‘s/he eats’ c.    méʔ ‘otter’  d.   ɬ-áʔ     3POSS-fruit     ‘fruit (of the tree)’   Finally, the third argument favouring the lack of oral place of articulation of the glottal stop comes from the phonotactic behaviour of /k ͡l/. This complex segment only occurs before vowels and it pervasively neutralizes to [k] in final coda position (185) or word internal coda position (186), before another consonant (cf. Table 2.6, §5.3.1).                                                   42 One might question whether CVC could be considered bimoraic, and hence satisfy FTBIN-µ. Although it is argued in §3.4 that /ʔ/ is moraic and therefore a CVʔ word would indeed meet a FTBIN-µ condition, there is no evidence that any other consonants in Nivaĉle are moraic. Consequently, in the absence of independent evidence for any consonant other than [ʔ] bearing weight, the appropriate generalization for the CVC MinWord in Nivaĉle is that it must be closed by a C. In Optimality Theory, this conforms to the constraint FINAL-C (McCarthy 1991:203, 1993:176).  119 (185) a.  wo.sók    ‘butterfly’   b.  wo.so.k ͡l-ís   butterfly-PL   ‘butterflies’  (186) a. xa-tʼuʔ̀.k ͡l[i].ja ́n   1S-obstruct-CAUS    ‘I obstruct’  b.  ∅-t’u ̰́k-ʃi      3S-obstruct-LOC(inside)    ‘it is obstructed’  Nevertheless, there is one particular context in which [k ͡l] is preserved in coda position, namely before glottal stops root internally. Compare, in this regard, (187a) with (187b), where the glottal stop onset is not part of the root (§3.3.2 ; §3.3.3.2)  (187) a. uk ͡l.ʔɑ́           ‘turtle doveʼ cf.   b. [xatpék’in]   xa-t-pek ͡l-ʔin   1S-CISL-return-IPFV   ‘I return (more than once)’  Here I argue that the fact that [k ͡l] can only occur as a coda before [ʔ] highlights the place-less specification of glottal stop. The fact that [k ͡l] does not occur before consonants but pervasively before vowels – and before tautomorphemic glottal stop – suggests a relationship between glottal stop and vowel-like properties. This special behaviour of [k ͡l] favours a Licensing by cue approach (Steriade 1997) (cf. Chapter 5.3). The generalizations arrived at in this section form the basis for the following feature specification of the glottal stop and ejectives. Note that I am assuming unary features.   120 Table 3.5 Feature specification of glottal stop and ejectives  ʔ T’ PLACE  LAB/COR/DOR CONSTRICTED GLOTTIS    Briefly, the data and phenomena analyzed so far are predicted by the ∅ place specification hypothesis for /ʔ/. The phonologically active distinctive features below the root node that function to define a glottal stop in Nivaĉle is [c.g]. In the following section, arguments will be presented that, at the prosodic level, /ʔ/ is moraic. 3.4  Prosodic representation of Nivaĉle glottalized vowels Let us now turn to a consideration of the representation of the Nivaĉle glottalized vowels, which are variantly realized as (i) [Vʔv ̰] ~ [V ̰] and (ii)  [Vʔ] (§3.2.1).  The major argument advanced in this chapter is that a Nivaĉle ‘glottalized’ vowel consists of a vowel plus glottal sequence – where the glottal is lexically specified for [c.g.]. As a consequence, I differ from Stell (1989) in that I do not consider the [c.g.] to be a contrastive feature within the vowel inventory. Instead, I analyze the [c.g.] to be a distinctive component of an independent /ʔ/ segment that occurs postvocalically (§3.3.4). This is, I claim, the same lexically distinctive /ʔ/ segment that can occur in Onset position. Its particular phonetic realization is dependent on how it is parsed into prosodic structure. In other words, the implication of this proposal is that, contrary to Stell, I do not postulate the existence of twelve vowels in Nivaĉle. Rather, I hypothesize that the alleged phonemic contrast between modal and glottalized vowels (Stell 1989) is actually a contrast between a modal vowel /V/ and a sequence of a vowel and a glottal stop: /Vʔ/.  What is interesting is that the prosodic constraints of the language allow the glottal stop to have variable syllabic parsing and thus different realizations emerge: (i) [Vʔv ̰] ~ [V ̰], and (ii) [Vʔ]. Specifically, I posit that the variable parsing of the [c.g.] feature is tied to prosodic context. First, given that there is a consistent correlation between glottalized vowels and the locus of stress, I propose that the glottal stop is, like vowels, underlyingly moraic (cf. Chapter 4, §4.3). The WEIGHT-TO-STRESS  121 PRINCIPLE (Prince & Smolensky 1993) states that heavy (bimoraic) syllables are required to be stressed. Second, if the glottal segment is aligned with the right edge of the syllable domain, the glottal stop will be parsed directly to the syllable node, as a coda, and the mora will be parsed to the Nucleus (cf. Figure 3.11a). Alternatively, if there is another consonant intervening between the glottal stop segment and the right edge of the syllable, then the mora (and its associated /ʔ/ features) will be parsed directly into the Nucleus of the syllable (cf. Figure 3.11b). In other words, only if the coda position is already filled by another consonant will the glottal stop be parsed into the Nucleus; a complex nucleus emerges at the expense of not creating a complex coda.     Figure 3.11 Prosodic representation of /Vʔ/   It is proposed, then, that the various prosodic roles the glottal stop has can be captured by interplay between prosodic markedness constraints and faithfulness constraints outlined in (188)-(194). A more detailed explanation will be given in §4.3.2.  (188) ONSET:  * [σV (Syllables must have onsets).  (Itô ̂ 1989, Prince & Smolensky 1993) (189) DEP-IO-ʔ: An output glottal stop must have an input correspondents (‘No ʔ-epenthesis’). (190) NOCODA: Syllables are open.    (Kager 1999) *C]σ  122 (191) *COMPLEXCODA       (Kager 1999) * CC]σ  ‘Codas are simple’.  (192) *COMPLEXNUC: No more than one segment may associate to the nucleus  (Prince & Smolensky 1993) (193)  MAX-IO-ʔ: A glottal stop in the input is present in the output. (194) MAX-IO-µ: A mora in the input is present in the output. First, as already discussed in §3.3.3.1 (see also §2.3.1.1) there are no onsetless syllables in Nivaĉle: ONSET is higher ranked than DEP-IO-ʔ. An epenthetic glottal stop is inserted to avoid onset-less syllables or vowel-vowel sequences: ONSET » DEP-IO-ʔ. Second, *COMPLEXNUC is higher ranked than NOCODA. This ranking explains why the glottal stop is parsed into coda position, if possible (i.e. if there is not another consonant in the coda). Recall that, because the glottal stop is not specified for place features (§3.3), it can be realized in coda position, in contrast with ejectives, which are specified for both [c.g.] and PLACE.  (195)  *COMPLEXNUC » *NOCODA  (196)    /      fajxoʔ      / *COMPLEXNUC *NOCODA  a.      faj.xo ̰ *!   b.   fajxoʔ  *  Candidate (b) wins over candidate (a) because it is more preferable to have a coda than a complex nucleus.  The tableau in (198) illustrates the parsing of [c.g.] into the Nucleus of the syllable (Figure 3.11b), which results in a complex bimoraic glottalized vowel. Because of the presence of an adjacent consonant to the right, the glottal stop is parsed into the Nucleus in order to not violate *CC]σ, an undominated constraint in the language (§2.3.1.3,  §6.3.1).   (197) *CC]σ , MAX-IO [ʔ],MAX-IO-µ: » *COMPLEXNUC     123 (198)           /k ͡loʔp / *CC]σ MAX-IO [ʔ] MAX-IO [µ] *COMPLEXNUC  a.        k ͡loʔp *!     b.        k ͡lop  *! *   c.     k ͡lo ̰p    *  The winning candidate (c) violates *COMPLEXNUC, at the expense of not violating higher ranked *CC]σ or the constraints MAX-IO-[ʔ] and MAX-IO-[µ] (see further discussion of these two constraints on §4.5.2). As a result, a more marked syllable nucleus emerges by parsing the [c.g.] feature into the nucleus of the syllable. The tableaux in (196) and (198) will be expanded and further discussed in §4.3.2.  As previously mentioned, the complementary distribution of a glottalized vowel and a vowel-glottal sequence is based on the interaction of COMPLEXCODA, COMPLEXNUC, and NOCODA. The different parsing of the glottal stop, which is argued to be dependent on prosodic context, can be illustrated through the following pairs of related forms:  (199) a.  ɬa-k.feʔ́ 3POSS-ear ‘his/her ear’  b. ta-k.fé-̰j 3S-ear-VBLZ(to.have) ‘it has ear’ (lit. meaning) /‘mug’   (200) a.  ɬa-ʃaʔ́  3POSS-salary  ‘his/her salary’ b.  xaj-ʃa ́-̰j  1S-salary-VBLZ(to.have)  ‘I have salary’  (201) a.  ɬ-aʔ́ 3POSS-fruit ‘its fruit’   124  b. t-á-̰j 3S-fruit-VBLZ(to.have)   ‘it has fruit’  (202) a.  ji-nuʔ́ 3POSS-bone ‘my boneʼ  b.   ta-nu ́-̰j ̰ 3S-bone-VBLZ(to.have) ‘it is bony’   In (199-202) a stative verb ‘to have X/X’s property’ is created by suffixation of the verbalizer /-j/ to a ʔ-final nominal root; ‘ear’, ‘salary’, ‘fruit’ and ‘bone’, respectively. In all these cases, the glottal stop is parsed into the complex nucleus of the syllable and the verbalizer is parsed into the coda. Similarly, the following data show variable parsing of /ʔ/:   (203) a.  xa-wɑ̰́m  xa=ji-k ͡lú.t ͡seʃ     [ʔ]ɑ.xɑ̀.k ͡lɑ-βo ́1S-lose   DET=1POSS-shotgun     bird-ART  ‘I lost my shotgun’  b. xa-wɑ́ʔ.m=eʃ  xa=wɑ.kɑ  ɬ-t ͡sʼo ̰́s 1S-lose =3O  DET=cow  3POSS-milk ‘I discarded the cow’s milk’   (204) a. k’ut.xá ̰n ‘thorn’ b. k’ut.xa ́ʔ.n=a  thorn=NEG  ‘there is no thorn’  (205) a. [ʔ]a-βu ̰́n                   2POSS-roasted.meat              ‘your roasted meat’        b.   [ʔ]a-βúʔ.n=a            2POSS-roasted.meat=NEG                ‘there is no (your) roasted meatʼ   125 Whereas the (a) examples in (203)-(205) show contexts where the glottal stop is parsed to the Nucleus of the syllable, the (b) examples show comparable contexts where the glottal stop is parsed to coda position, and the final consonant of the root is parsed into the onset of the subsequent syllable. The negative clitic does not impact stress assignment and so the postvocalic glottal stop can get realized (§4.3.2). The following spectrograms illustrate this variable realization of the glottal stop.    Figure 3.12 Waveform and spectrogram of [ɬaʔ́] ‘its fruit’ by male speaker FR Figure 3.13 Waveform and spectrogram of [ta ̰́j] ‘it has fruit’ by male speaker FR  Figures 3.12 and 3.13 present the spectrograms for the pair in (201). Under the analysis proposed here, the glottal stop of the vowel-glottal coda in Figure 3.12 is realized as nuclear creakiness in Figure 3.13 when the verbalizer [-j] is parsed into coda position. A mirror example is demonstrated in the following figures:  126   Figure 3.14 Waveform and spectrogram of [xawɑ̰́m] by male speaker FR Figure 3.15 Waveform and spectrogram of [xawɑ́ʔmeʃ] by male speaker FR   Figures 3.14 and 3.16 show a form where, in the analysis proposed here, the glottal stop is parsed to the Nucleus of the syllable; as a result, a Complex Nucleus arises. In contrast, the Figures 3.15   Figure 3.16 Waveform and spectrogram of [ʔaβu ̰́n] ‘your roasted meat’ by female speaker TS Figure 3.17 Waveform and spectrogram of [ʔaβuʔ́na] ‘there is no (your) roasted meatʼ by female speaker TS.  127 and 3.17 show forms where the glottal stop being parsed to Coda position since the final consonant of the root is parsed into the Onset of the following syllable; these are examples of what I have called vowel-glotal coda.  Note that the glottalized vowel produced by TS (Figure 3.16) does not display irregular glottal pulses on the spectrogram. As briefly mentioned in §3.2.1, for some speakers, glottalized vowels are not constantly produced with creak. The production of glottalized vowels by TS sometimes involves long vowels rather than rearticulated or creaky vowels. This phenomenon raises the question, again, about the absence of a one-to-one correlation between phonology and phonetics.  Recapitulating, the fundamental claim being advanced is that the underlying sequence of a /V/ followed by a glottal stop /ʔ/ is realized either (i) as a [V] plus glottal stop coda, or, (ii) as a ‘creaky/rearticulated’ vowel, if the syllable is closed by another consonant. 3.5 Conclusions  Relating the syllable and phonotactic restrictions laid out in Chapter 2, this chapter has established the featural and prosodic representations of the glottal stop and the so-called ‘glottalized’ vowels in Nivaĉle (Stell 1989). It has been proposed that the glottal stop is unspecified for place features, but specified for [c.g.]. Glottalized vowels are underlying vowel-glottal sequences: /Vʔ/. As such, they consist of a vowel followed by glottal stop, which is itself defined as a moraic root node specified for [c.g.]. This moraic root node can attach to (i) the nucleus of the syllable and form part of a complex nucleus – phonetically realized as a ‘rearticulated/creaky’ vowel – or (ii) the syllable node as coda and thus get realized as a glottal stop. Phonetic evidence for the alternation and relationship between rearticulated and vowel-glottal coda has been provided in the effect of affixation processes on syllabic parsing.  In addition, a crucial argument has been proposed: glottalized vowels are bimoraic. In Chapter 4, I will develop this proposal by showing that glottalized vowels occupy a prominent position, that is, the head of a foot, and thus bear stress.   128 Chapter 4: Prosodic structure of Nivaĉle  4.1 Introduction Prosodic structure refers to the organization of segments in terms of universal prosodic units: mora, syllable, foot, prosodic word. The goal of a theory of prosodic structure is to capture linguistically significant generalizations about “suprasegmental” phonological properties such as weight, duration, stress, tone, and intonation, among others.  An essential claim in metrical stress theory is that prosodic units are organized in a hierarchical relation; the Prosodic Hierarchy (Selkirk 1980, 1984, 1986, McCarthy & Prince 1986, Nespor & Vogel 1986/2007) serves as the basis for the analysis of prominence or stress assignment in a language. The Strict Layering Hypothesis (Selkirk 1984) states that all prosodic constituents at a particular level consist exclusively of constituents from the level below (prosodic levels cannot be skipped or repeated). The Headedness Principle requires that each prosodic unit dominates a head at the next lower level of the Prosodic Hierarchy: that is, every prosodic word (PrWd) is headed by a foot (Ft), every foot is headed by a syllable (σ), and every syllable is headed by a mora (µ). Shaw (1992) argues for an intermediate level between the syllable and the mora: a Nucleus constituent, in order to account for the two distinct types of bimoraic syllables (CVµCµ and CVµVµ) in templatic morphology. Given this distinction, Shaw proposes a contrast between nuclear moras and non-nuclear moras. As discussed in Chapter 3 of this thesis, I adopt the notion of the Nucleus as head of the syllable: in Nivaĉle, the Nucleus functions as the prosodic unit that hosts all and only the moraic units of the language. Specifically, I propose that: (i) Both /V/ and /ʔ/ are moraic in Nivaĉle. (ii) A mora is always parsed into the Nucleus in Nivaĉle.  (iii) Pre-nuclear (“Onset”) segments are not moraic.  (iv) Because there is no phonemic contrast in vowel length in Nivaĉle, the phonetic realization of length in a Nucleus derives from the mora from a glottal stop /ʔ/ being parsed into the Nucleus.  129 (v) If the full segmental content of the /ʔ/ is also parsed into the Nucleus, then the surface realization is of a creaky/rearticulated vowel. (vi) If the segmental content of the /ʔ/ is disassociated/delinked from its mora so that the /ʔ/ can be realized as a coda, then (in accordance with (ii) above) the mora remains in the Nucleus. (vii) The bimoraic status of the Nucleus in the context of glottalized [Vʔv ̰] ~ [VV ̰] and vowel-glottal [Vʔ] coda accounts for the ubiquituous association of these syllables with stress through the direct correlation of the Weight-to-Stress Principle (Prince & Smolensky 1993).  To clarify then, I am adopting a prosodic hierarchy model structured as follows: (206) Prosodic hierarchy (adapted from Selkirk 1980, Itô 1986, McCarthy & Prince 1986, Shaw  1992).      PrWd       Prosodic Word                 |      Ft    Foot                 |                          σ   Syllable                 |       N   Nucleus        |        µ   Mora   The mora (Hyman 1985, McCarthy & Prince 1986/1996, 1990, Zec 1988, Hayes 1989) serves as a “primitive subsyllabic constituent and as a measure of syllable weight” (Zec 1995: 85). Crosslinguistically, syllables commonly subclassify into (a) light monomoraic, and (b) heavy bimoraic.43 Stress assignment (along with tone assignment, compensatory lengthening, and reduplication, among other phonological phenomena) has been proposed as a diagnostic for syllable weight (Prince 1980, McCarthy & Prince 1986/1996, Zec 1988, Hayes 1989, Gordon 2006).  The moraic status of coda consonants is subject to cross-linguistic variation. There is no evidence suggesting that coda consonants are moraic in Nivaĉle. It has been hypothesized that the                                                 43 Less commonly attested are trimoraic syllables (e.g. German (Féry 1997), Hindi (Broselow et al 1997) etc.) and non-moraic or weightless syllables (e.g. həәn ̓q ̓əәminə̓әm ̓ Salish (Shaw et al. 1999); Mohawk (Piggott 1995), etc.).  130 weight of coda consonants is parameterized on a language-specific basis (Hyman 1985, Zec 1988, Hayes 1989, Zec 1995). In this way, a typology for moraic consonants can be found:  (i) Languages where CVC syllables pattern with CVV syllables and are thus heavier than CV light syllables; e.g. Finnish (Sadeniemi 1949), Hindi (Ohala 1986), Latin (Allen 1973), Yana (Sapir & Swadesh 1960).  (ii) Languages where CVC syllables pattern with CV syllables as against CVV syllables; e.g. Araucanian (Echeverría & Contreras 1965), Khalkha Mongolian (Walker 1997), Malayalam (Broselow et al. 1997).  Besides vowel quantity and moraic codas, prominence (Hayes 1995) has been proposed for distinguishing light from heavy syllables. For instance, “rhymes with full vowels, high tone, lower (and thus more sonorous) vowels, or rhymes with a complex vowel involving glottalization count as “more prominent” than their respective counterparts” (van der Hulst 1999:10 [my underline/AG]). Furthering the line of research into the relative prominence of different vowels within a given phonological system, Kenstowicz (1997) examines languages where prominence is determined by the quality of the syllabic nucleus:44 what he advances is phonological evidence that “lower vowels are more prominent than higher vowels and peripheral vowels are more prominent than central vowels” (157).     In their analysis of the complex stress system of Nanti (Kampa, Arawakan), Crowhurst and Michael (2005) show that, besides being sensitive to rhythmic factors, syllable quantity, and vowel quality, stress is, interestingly, sensitive to whether a syllable is open or closed (§4.2). Similarly, Munshi and Crowhurst (2012) show that in Koshur (Kashmiri) closed syllables “are preferred as stress peaks over [open] syllables with vowels of the same length” (427). The central point to be drawn from the Nanti and Kashur (Kashmiri) analyses is that a coda consonant that is arguably not moraic can nonetheless contribute to a closed CVC syllable having greater prosodic prominence than an otherwise comparable open CV syllable. Here it is argued that CVC syllables in Nivaĉle, despite being                                                 44 The languages under study are: Kobon (Davies 1981), Chuckchee (Skorik 1961), Aljutor (Kodzasov & Muravjova 1980), Mari (Gruzov 1960),  and Mordwin (Tsigankin & Debaev 1975).   131 monomoraic and monosyllabic, play an active role in the stress system, particularly as a locus of secondary stress in polysyllabic words.  Post-vocalic glottal stop, however, has a unique role in terms of prominence. Based on the observation that glottalized vowels are consistently stressed in Nivaĉle, one of the central claims advanced in this thesis is that the glottal stop is moraic (§4.3); this mora attaches to the nucleus of the syllable. That is, both creaky/rearticulated vowels and a Vʔ] rhyme are prominent. As expected from the cross-linguistic generalization that Onsets do not standardly contribute prosodic weight, a glottal stop onset in Nivaĉle is not moraic; only postvocalic glottal stop functions to contribute weight to a syllable. This follows from two claims in the present analysis. First, /ʔ/ is underlyingly moraic.45 Secondly, a mora in Nivaĉle can only be parsed into the Nucleus. Nivaĉle then behaves like a Type (ii) language in the bipartite moraic languages typology discussed above, where CVC syllables pattern with CV syllables: coda consonants do not bear weight. Notably, in distinction from /ʔ/, ejective obstruents do not bear weight.  The only weight-bearing, i.e. moraic, segments are vowels and glottal stop. Moreover, only the Nucleus can host weight. The brief characterization of Nivaĉle stress that is found in Stell (1989: 81-83) will be discussed in Section 4.4.1. No details of the Nivaĉle prosodic system, such as the moraicity of its segments, the minimal word, foot types, and phonological domains are addressed in Stell’s grammar nor have there been any other studies which have offered a detailed, prosodically motivated analysis of the stress system of Nivaĉle.  The goal of this chapter is to account for the internal structure of the Prosodic Word and of stress assignment in Nivaĉle, focusing on the nominal and verbal domains. Section 4.2 analyzes the smallest prosodic domain, that is, monosyllabic words, and proposes that the Minimal Word is a closed monosyllable: CVC. Section 4.3 analyzes the case of glottalized vowels, provides empirical                                                 45 This is intended to function as a simplifying assumption, rather than as a crucial tenet of the analysis. An alternative hypothesis is that vowels and /ʔ/ (but no other segments) acquire moraic status through the relative ranking of parsing constraints such as PARSE-ʔ-to-µ and PARSE-V-to-µ. Under either analysis, the pervasive generalization is that (unless deleted or parsed into an onset) a /ʔ/ consistently contributes weight to a syllable.  132 argumentation in support of the present claim that they bear weight and illustrates their consistent conformity to the WEIGHT-TO-STRESS Principle (Prince & Smolensky 1993).  In section 4.4, the stress patterns in Nivaĉle are discussed. I propose that Nivaĉle has a quantity sensitive stress system, and that the rhythmic type is iambic. The following foot types are thus attested: Heavy (H), Light-Light (LL), and Light-Heavy (LH). In addition, it is advanced that stress assignment is edge-based, aligned with the edges of hierarchically nested morphological categories (MCat), specifically: Root (Rt), Morphological Stem 1 and 2 (MSt1, MSt2), and Morphological Word (MWd). In Section 4.5, the internal morphological structure of nominal and verbal forms are presented; the role of prefixes and suffixes with regards to stress placement is discussed, as there are Left-edge and Right-edge stress generalizations that apply. A major generalization at the uppermost prosodic level is that primary stress falls on the rightmost foot of the PrWd.  4.2 Minimality: Nivaĉle CVC monosyllables In many languages, there is a restriction on the minimum prosodic size of a word (McCarthy & Prince 1986). Further, the minimal content word of a language has been equated with the minimal foot allowed in that language (Prince 1980, McCarthy & Prince 1986).46 In that vein, certain languages require that every content word either contain at least two moras or two syllables. This minimality requirement corresponds to the claim that feet must be binary under moraic or syllabic analysis (McCarthy & Prince 1993:90):  (207) FT-BIN-µ: Feet are binary at the moraic level.      (Kager 1999) (208) FT-BIN-σ: Feet are binary at the syllabic level.      (Kager 1999)  Nivaĉle presents an interesting case study in this regard. Specifically, in Nivaĉle, what is found is that a CVC syllable can stand alone as a word. However, because in the present analysis it is argued                                                 46 Nevertheless, several studies have shown that there is not necessarily a one-to-one correlation between minimal word requirements in a language and the minimal stress foot (Hayes 1999, Garrett 1999, Gordon 1999).   133 that there is no independent prosodic evidence for coda consonants (other than [ʔ]) being moraic, a Nivaĉle CVC word does not meet the bimoraic constraint (207). Nor, clearly, does a CVC word meet the bisyllabic constraint (208). Thus the smallest word in Nivaĉle does not conform transparently to the Binarity generalization of the prosodically-defined notion of a “Minimal Word”. It is proposed nonetheless that CVC in Nivaĉle, despite being monomoraic and monosyllabic, constitutes a Minimal Word and functions also as a minimal well-formed foot.  Crucially, a CV syllable never stands alone as a word. Nor does a CV syllable function as a foot in the stress system, whereas a CVC syllable can do so. The conclusion therefore is that Nivaĉle provides evidence that a monomoraic CVC syllable plays a fundamental prosodic role, both as a Minimal Word and as a stress foot.   There are two types of arguments in favor of claiming that the Minimal Word in Nivaĉle consists of a CVC syllable.   First, as illustrated in (209), (210), and (211), both CVC nominal and verbal roots can occur as independent words. Indeed, the majority of the Nivaĉle nominal and verbal roots consist of CVC monosyllables. Moreover, because all lexical items bear stress, these forms illustrate a minimal stress foot as well.47 Recall that primary stress is marked as an acute accent on the stressed syllable.  (209) a.  pʼók    ‘arrow’   b.  t ͡sók    ‘plastic’   c.  ɬéɬ    ‘snail’   d. tós    ‘snake’                                                    47 Examples of a (CVC) foot occurring in longer polysyllabic words are presented later, in §4.5.6.  134   e.  t’u ́n    ‘hard/galleta (type of bread)’       Alienable nominal roots can occur independently as a monosyllabic CVC foot as in (209a-e) above. Inalienable roots require the presence of a possessive prefix, as seen in (210). What these V-initial roots prefixed by a single-C possessive pronoun further establish is that a polymorphemic word can also be realized simply as CVC.  (210) a.  ɬ-ɑ́k    3POSS-food    ‘his/her food’   b.   j-ɑ́x    1POSS-skin    ‘my skin’   c.  j-éj    1POSS-name    ‘my name’     The verb forms below show a fully parallel pattern; CVC is also well-formed foot and Minimal Word. In the forms in (211), ∅ is represented to indicate the non-surface realization of a third person subject.  (211) a. ∅-k ͡lóp    3S-white         ‘s/he is white’   b. ∅-sás    3S-bad/ugly    ‘s/he is bad/ugly’       c. ∅-βáf    3S-pass.away    ‘s/he passed away’    d. ∅-túx    3S-eat    ‘s/he eats’  135 Similarly to (211), the following data show verbal roots that require the presence of a person prefix: single consonant prefixes attach to vowel-initial roots (212) and CV prefixes attach to C-initial roots (213):   (212) a. x-én       pa=Jesus     1S-love    DET=Jesus     ‘I love Jesus’     b.    ɬ-ám     2S-arrive     ‘You arrive’   c.  j-ít ͡ʃ   ka=tós     3S-go  DET=serpent     ‘a serpent passed by’  (213) a.    ji-t ͡ʃét    3S-upset.stomach    ‘s/he has an upset stomach’  b.  ji-mɑ́ʔ    3S-sleep    ‘s/he sleeps’  c.  xa-t’ó.wɑs=ʔakfí   ɬa=[ʔ]a ̰̀k.xi.ju ́k    1S-cut=LOC(inside)  F.DET=tree    ‘I cut down the tree’  What is particularly significant across both the nominal and verbal contexts is the fundamental generalization that Nivaĉle lacks CV words. Whereas polysyllabic words can be V-final, e.g., bisyllabic CV.CV (214a), or CVC.CV (214b), monosyllabic words cannot: *CV.  (214) a.    wa.wó    ‘maned wolf’  b.  k’ak.xó    ‘armadillo’   c.  ju.ku.βe ́    ‘bread’  136  d.  ni=ji-ka ́.ku   pa=Jesus    neg=1S-distrust  DET-Jesus    ‘I believe in Jesus’  e.   xa-fu ̀.ju-k’e ́      na=ji-k ͡la ́.nat ͡ʃ    1S-play(an instrument)-LOC  DET=1POSS-DRUM    ‘I play the drum’  Given that CVC – and not CV – is the minimal word in Nivaĉle and, as will be further argued later, coda consonants are not moraic, an interesting implication for the prosodic analysis of glottals emerges: as seen in (215), glottals can occupy coda position and fulfill the CVC MinWord condition.    (215) a.  ɬ-áʔ     3POSS-fruit     ‘fruit (of the tree)’   b.  ɬ-úʔ   ɬa=pelota       2S-throw  DET=ball     ‘you throw the ball’        c.     me ́ʔ      ‘otter’        d.    βíʔ      ‘caterpillar’     e.  máʔ     ‘frog’   There is no phonological evidence that a glottal stop is ever inserted to satisfy MinWord requirements. For instance, recall from Chapter 3 (§3.4) that a glottal final root like the one in (215) can be parsed to the Nucleus when a suffix is added; a rearticulated/creaky vowel arises: ta ̰́j ‘it has fruit’ (cf. (201b).  Two further empirical issues arise. First, while all the examples above illustrate CVC words with modal vowels, the following data establish that the nucleus of a CVC word may alternatively be a glottalized vowel, i.e. CV ̰C:    137 (216) a.  k ͡lo ̰́p    ‘winter’    b.  wɑ̰́s    ‘sky’      Secondly, it is well established that onset complexity does not normally play a prosodic role in MinWd or Foot-form constraints, and it is not surprising therefore that some monosyllabic words have a complex onset:   (217) a.   xpɑ̰́k        ‘straw’   b.  kxát    ‘fruit of the cactus’   But, what is significant in showing that onset complexity does not entail weight is that, like *CV,  *CCV is not a possible MinWd. The important criterion for a monosyllabic MinWd is that it be a closed CVC – or, as in (217) above, CCVC – syllable.   Also note that there are no free-standing VC words; the constraint ONSET is undominated, enforced by [ʔ] epenthesis (§2.3.1).   Some illustrative examples from (211a), (216a) and (215a) are presented in (218) with their moraic analysis.  (218)   a. Monomoraic  CVC foot: [k ͡lóp]                              138    b. Bimoraic foot: heavy bimoraic nucleus: [k ͡lo ̰́p]                                c. Bimoraic foot: heavy bimoraic nucleus with [ʔ] coda    In contrast, a CV or CCV (d) is not a well-formed attested foot type.       d. *(C)CV foot                                                     139 The details of the featural and prosodic representation of glottalized vowels (218b) and the glottal stop (218c) have been discussed in §3.4 and are recapitulated in §4.3.1.   In sum, CVC functions as a well-formed MinWd in this language, regardless of moraic content. Interestingly, therefore, in the case of CVC the criterial restriction is on syllable shape rather than weight. If a lexical unit does not meet either the Ft-Bin-µ (207) or Ft-Bin-σ (208) constraints, then it can nonetheless qualify for MinWd status if it is a single closed syllable, i.e. if there is a “coda” consonant. This restriction on the syllable shape of the Minimal Word can be interpreted in three ways.  First, one could hypothesize that there is a constraint that requires a Prosodic Word to end in a consonant:   (219) FINAL-C: A Prosodic Word (PrWd) cannot end in a vowel     (McCarthy 1993)  *V]PrWd  However, it is clear that Final-C is not an undominated constraint, as polysyllabic words can be V-final (220), or C-final (221):  (220) a.  k ͡le.sá     ‘knife’   b.   wa.wó          ‘maned wolf’  c. tʃ͡a.xa.ní   ‘rodent’ (221) a.   ma.ko ́k ‘frog’  b. ma.ko.k-[í]s   frog-PL   ‘frogs’     What we can conclude is that the requirement to be C-final functions as a well-formedness MinWd constraint on monosyllabic words (where possible, the optimal Head foot of the PrWd will conform to Ft-Bin-σ (208)). In this vein, the following well-formedness constraint can be proposed:   140 (222) MINFT=CVC:   The Minimal Foot is a CVC syllable.  [violated by *(CV ́) *(CV ̀)]  This constraint would connect the observation that the minimal word in Nivaĉle is a foot that can be as small as (CVC). In addition, note that in polysyllabic words, an initial CVC foot receives secondary stress as shown in (223a-c) whereas otherwise comparable forms with an initial CV syllable (223d-e) do not have secondary stress, as an open CV syllable does nto satisfy the MinFt constraint in (222):  (223) a. ([ʔ]ɑ̀k).(xek ͡lɑ́)    (CVC)   ‘woman’ b.   (puʔ̀).(xana ́)    (CVʔ)       ‘threeʼ  c.  (fa ̰̀j)(ku-kát)    (CV ̰C)   carob-TREE.CLASS-COL   ‘a stand of algarrobo trees’  d.  si(βok ͡lo ́k)   ‘spider’  e.  ni(βak ͡le ́)   ‘man’  The discussion of CVC functioning as a foot will be shown in §4.5.6. The fact that Nivaĉle minimal word is not sensitive to weight but to whether a syllable is closed or not could be captured by two alternative analyses.  First, one could posit that CVC minimal words acquire a mora in order to satisfy foot binarity. This would constitute a case of variable closed-syllable weight, namely, light syllables become contextually heavy to satisfy a higher ranked constraint (Rosenthall & van der Hulst 1999, Morén 1999). For instance, CVC syllables do not count as heavy for stress assignment unless a satisfaction of word minimality or the avoidance of some critical constraints violation is at issue. In this case, a mora would be inserted – in violation of low ranked DEP-µ – in order to satisfy FT-BIN-µ.   141 (224) DEP-µ: A mora in the output has a correspondent in the input. (225) FT-BIN-µ: Feet are binary at the moraic level.  (226) FT-BIN-µ » DEP-µ  Given that there is no independent motivation for coda consonants having weight, this ‘variable closed syllable weight’ analysis would be entirely ad hoc, serving only to satisfy a FT-BIN-µ constraint. What I have posited so far is that the special ranking of FT-BIN-σ, FINAL-C and MINFT=CVC constitute relevant constraints to account for foot construction and stress assignment.   Second, an alternative analysis, in line with Crowhurst & Michael (2005) and Munshi & Crowhurst (2012), would posit that syllable codas play a role in stress assignment independently of moraic weight (contra Rosenthall & van der Hulst (1999) and Morén’s (1999) above mentioned accounts). Under this approach, it is argued that the prosodic role of codas in contributing to CVC syllables being preferentially stressed over CV syllables can be attributed to mora branchingness (Munshi & Crowhurst 2012: 430). These authors claim that “branching is another property of the mora with metrical significance in some languages”:  (227) Branching mora    Nevertheless, this branching mora approach cannot account for the Nivaĉle data because under the syllable representation in (227), a glottal stop coda and any other coda would be expected to pattern together. However, only the glottal stop is associated with a mora. The present analysis proposes that this mora is parsed to the Nucleus of the syllable (recall from §4.1 that only the Nucleus can host weight). In contrast, there is no phonological evidence of other coda consonants contributing to weight,  142 or to stress prominence in any way. In sum, it is argued that a syllable representation that includes the Nucleus as the prosodic head of the syllable has greater explanatory value as it captures the different patterning of glottals and other coda consonants.  In this chapter, I will posit that: (i) CVC syllables are not heavy, that is, there is no Weight-by-Position, neither consistent nor variable WBP, (ii) CVC syllables satisfy word minimality, and (iii) CVC can function as a foot in the stress system of Nivaĉle. However, what will be argued in §4.5.6 is that a CVC foot is not an optimal foot. In a polysyllabic word, a foot that satisfies the FT-BIN-σ constraint receives primary stress, while a CVC foot receives secondary stress. Consequently, I propose that the Head of the PrWd is optimally binary at the syllabic level. 4.3 Prosodic properties of /ʔ / It has been posited that languages differ in the parameter used to set the boundary along the sonority hierarchy between moraic vs. non-moraic coda consonants (Zec 1988, 1995). Specifically, the analysis of glottals as being moraic displays some variability. This is not an unexpected phenomenon; much of the ambiguity of glottals as bearing vs. not bearing a mora comes from its variable featural and prosodic representation across languages. For example, it has been argued that the special status of glottals derives from a structural difference between glottals and other segments, such as the absence of a supralaryngeal node (Cohn 1993). Glottals can pattern with either obstruents (Ladefoged 1971, Hyman 1975, Lass 1976, Bessell 1992) or sonorants (Chomsky & Halle 1968, Kenstowicz & Kisseberth 1979, Kavitskaya 2002). Moreover, whereas in some languages glottals have been argued to bear place features (McCarthy 1991, 1994, Ola Orie & Bricker 2000, Lombardi 2002), in others they do not (Steriade 1987, Cohn 1990, Bessell & Czaykowska-Higgins 1992). See §3.3.4 for a more detailed discussion of the featural specification of glottal stops.   Alongside of these featural differences, various proposals have been advanced with respect to the moraic status of glottal stops. Just to cite some examples, the glottal stop has been analyzed as moraic in Blackfoot (Peterson 2004, Elfner 2006), but as the only non-moraic consonant (along with /h/)  143 in Tehrani Farsi (Darzi 1991), and the only weightless consonant in Bella Coola/Nuxalk (Bagemihl 1991, 1998).   Kavitskaya (2002) argues that in languages where the loss of glottals results in compensatory lengthening alternations, these glottals need to be analyzed as phonological approximants and as moraic. She thus predicts the existence of two types of glottals based on their relationship with compensatory lengthening. If deletion of a glottal does not trigger compensatory lengthening, this glottal is predicted to be weightless and to pattern with (non-moraic) stops. In contrast, glottal approximants are predicted to have weight, and their deletion results in compensatory lengthening.    There is another variable that one should consider in the relationship between glottals and moras. The particular surface realization of the glottal stop – as an independent segment vs. glottalized realization concomitant with a vowel – can be conditioned by foot type and foot size (Elías-Ulloa 2006).  Elías-Ulloa (2006) analyzes a special case of contextual syllable weight in closed syllables. In Shipibo and Capanahua (Panoan), the weight of closed syllables changes according to the position in which they occur within the prosodic structure. For instance, “closed syllables are light in Capanahua when they occur unfooted or as the initial syllable of a disyllabic foot; elsewhere closed syllables are heavy” (2006:19).  Interestingly, variable closed syllable weight and prosodic structure in Capanahua is related to the occurrence of the glottal as a full segment, or as creakiness in the preceding vowel. Coda glottal stops can only surface in head syllables because this privileged or prominent prosodic position inhibits glottal coalescence into a preceding vowel nucleus. In contrast, in non-head positions, coda glottal stops fuse with the preceding vowels and are thus realized as glottalized/creaky vowels. In other words, Capanahua’s glottal coalescence is analyzed as a strategy to adjust the weight of its syllables to the metrical context in which they occur while respecting the disyllabic size of their feet (Elías-Ulloa 2006:9).  144  In this thesis, I propose that the occurrence of glottalized vowels in Nivaĉle is also correlated with the prosodic structure. On the one hand, creaky/rearticulated vowels and coda glottal stops surface only under stress, i.e. in prosodic head position: thus, it is posited that there is a direct correlation between these syllables being bimoraic and their being stressed. On the other hand, in non-head position, both the moraic value and the [c.g.] of an underlying glottal stop (whether in coda position or incorporated into a glottalized vowel) are lost: the pervasive generalization is that coda glottal stops and glottalized vowels are not realized except under stress. To recapitulate the Nivaĉle analysis, the glottal stop in Nivaĉle displays some interesting phonological behaviour. On the one hand, it is argued that it can be realized as a glottal stop in post-vocalic “coda” position. On the other hand, it can form part of the syllabic nucleus. It, thus, displays a dual patterning:  (i) like vowels it can be part of the nucleus. (ii) like consonants it can be realized as a full glottal stop and be parsed into coda position (§3.3.3.3).  4.3.1 Syllabic status of /ʔ / vs Glides The glottal has a dual patterning in terms of syllabic parsing that is not shared with other segments. For example, whereas glides can occupy onset or coda position, an analysis of tautosyllabic vowel-glide sequences establishes that they do not ever form part of a complex nucleus.  Given this asymmetric pattern, let us examine what featural representations can appropriately capture the natural class behaviours of the vowels, glides, the glottal stop, and glottalized vowels. Following  Levi (2008, 2011), I hypothesize that glides are [+consonantal]. There are three arguments in favor of the claim that glides /w/ and /j/ in Nivaĉle are [+consonantal].  First, as mentioned in §2.2.1, in terms of distribution, glides can occur in onset and coda position, before and after all vowel environments, and thus are not restricted to co-occurrence with a small number of vowels.   145 Second, there are no tautosyllabic vowel-glide-consonant (VGC) sequences in Nivaĉle. Because it has been established that complex codas are not allowed in Nivaĉle (§3.3.1.3), the only potential parsing of a tautosyllabic VGC string would therefore be if [VG] were a possible complex nucleus, i.e. a diphthong. Given that […VGC]σ strings are unattested, whereas […VG]σ strings are attested, it follows that the glides (G) pattern with other consonants in being parsed into the syllable coda, and not into a complex nucleus. Thirdly, in addition to these distributional constraints, active phonological evidence that glides are parsed into the coda comes from the fact that glide-final roots behave like other consonant-final roots in terms of triggering vowel epenthesis when suffixed by a single –C plural marker. That is, as the following examples of plural suffixation show, when a consonant-initial suffix is added to either a glide-final (228-229) or consonant-final stem (230-231), there is vowel epenthesis in order not to incur violations of *COMPLEXCODA:  (228) a.  ʃk ͡lɑk.xɑ́j   ‘partridge’  b.  ʃk ͡lɑ̀k.xɑj-[í]s    b’. *ʃk ͡lɑkxɑjs   partridge-PL   ‘partridges’  (229) a.  [ʔ]o.βa ́j ‘guavirami (fruit)’  b.  [ʔ]o.βa.j-[í]s    b’. *[ʔ]oβajs guavirami (fruit)-PL   ‘guavirami fruits’  (230) a.  βa.tʼ-ɑ́x INDEF.POSS-skin ‘skin’ b.   βa.tʼ-ɑ.x-[í]s    b’. *βatʼoxs   INDEF.POSS-skin-PL ‘skins’  146  (231) a.  k’ut.xa ̰́n ‘thorn’  b.  k’u ̀t.xa.n-[í]s    b’. *k’utxa ̰ns   thorn-PL   ‘thorns’ cf.   (232) a.  [ʔ]a.ɬu ́‘lizardʼ  b.  [ʔ]a.ɬu-́s   lizard-PL   ‘lizards’ Consequently, the generalization is that glides pattern with consonants and not with vowels (232). 4.3.2 Syllable structure parsing constraints for /ʔ / Let us turn now to a consideration of the patterning of the glottal stop and the vowels. It is proposed here that there is a direct and consistent correlation between the presence of [c.g.] in a syllable nucleus or as a coda [ʔ], and the locus of stress.   The prosodic representation of two words, one with a tautosyllabic vowel-glottal sequence, the other with a creaky/rearticulated vowel, is schematized in Figure 4.1 below (see also Chapter 3, §3.4)    147  Figure 4.1 Prosodic representation of /Vʔ/  To summarize, it is proposed that the Nivaĉle glottal stop patterns (i) with vowels in being moraic and parsed to the Nucleus node, and (ii) with consonants in potentially functioning as a coda and being parsed directly to the syllable node. First, /ʔ/ consistently contributes weight to a Nucleus and attracts stress: therefore it is claimed to be a mora-bearing unit. Under the hypothesis that Nivaĉle is a language (like the other Type (ii) languages referred to in §4.1) where only the Nucleus of a syllable can license weight, the Parsing constraint in (233), in conformity with the Strict Layering Hypothesis of the Prosodic Hierarchy, will effectively select only vowels and /ʔ/ to parse into a Nucleus.  (233)  PARSE-µ-TO-NUC (PARSEµ):  Moras must be parsed into the Nucleus of a syllable.  Secondly, the complementary patterning of /ʔ/ with segments other than vowels follows from the general, highly ranked parsing constraint:  (234) PARSE-SEG-TO-σ (PARSE-SEG): Segments must be parsed into syllable structure.          (cf. Prince & Smolensky 1993)  148 The constraint in (234) functions to ensure exhaustive parsing to the syllable level of all segments in a string. Depending on its relative position in a phonological string, a glottal stop can thus be parsed into the Nucleus or be parsed as an independent segment to either edge of a syllable, i.e. to either an “onset” or a “coda” position,48 e.g. as in (232a) [ʔaɬú] ‘lizard’ or (215e) [máʔ] ‘frog’, respectively. 49 The most fundamental analytical question with respect to a post-vocalic /ʔ/ is what constraints govern whether it is parsed into a Nucleus or a Coda. The most basic generalization to be captured is that /ʔ/ is parsed into the coda if and only if there is no other consonant parsed into coda position, i.e. in post-Nuclear position at the right edge of the syllable. Otherwise, the /ʔ/ will be parsed into the Nucleus.50 What this generalization reflects is the fundamental role of the FINAL-C constraint (see (219) in §4.2) in Nivaĉle. Its crucial ranking above NOCODA, normally a high-ranking markedness constraint, is shown first in the tableau in (235a): (235) a.  βíʔ ‘caterpillar’         / β i ʔ /  FINAL-C NOCODA  a.     (βí ̰) *!   b.  (βíʔ)  *   b.  βeʔɬa ‘one’        / βeʔɬa /  FINAL-C NOCODA a.      (βe ̰)ɬa *!  b.  (βéʔ)ɬa  *                                                 48 To reiterate, the terms “onset” and “coda” are used here, as elsewhere, not as primitive prosodic units but rather simply as informal terms of reference to designate segments at the left or right periphery of a syllable domain. The sole syllabic units that are formally posited to be prosodic primes in the framework assumed here are syllable (σ), Nucleus (N), and mora (µ). 49 Under the assumption that /ʔ/ is underlying moraic, it is further assumed that if /ʔ/ is parsed into an onset position, the universal markedness constraint  (unviolated in Nivaĉle) reflecting the cross-linguistic generalization that onsets are not moraic (*ONSET-µ) will effectively result in the deletion of the mora. This would entail a concomitant violation of MAX-µ. As the focus of the present analysis is on the manifest weight properties of /ʔ/ in non-onset position, a more detailed formal treatment of this issue is not directly relevant and hence is not pursued further. 50 Recall that a glottalized vowel or a tautosyllabic [Vʔ] sequence is only realized in a stressed syllable. In the tableaux here, this condition is met and therefore the constraint implications of WEIGHT-TO-STRESS, MAX-µ, MAX-[ʔ] are not shown.  149 With respect to the tableau in (235b), recall from §4.2 that a CVC syllable functions not only as a MinWd, but also – under the generalization MinWd=PrWd – as a minimal stress domain. As is argued in the subsequent sections of this chapter, the Nivaĉle stress system is based on FT-FORM=IAMBIC, but feet may be parsed from the left edge or the right edge of a word, depending on a well-defined hierarchy of morphologically-defined domains. What is relevant to the present discussion is that there can therefore be word-internal syllables with a final [ʔ] coda: the prediction of the present analysis is that such non-word-final [CVʔ] syllables will only be found under conditions of stress prominence. The tableau in (235b) illustrates this with the word [βe ́ʔɬa] ‘one’, where the initial syllable (βéʔ) functions as a left-aligned CVC foot, and the final syllable [ɬa] is stray, i.e. unparsed to the foot level. What is important to note is that in the initial syllable, which is functioning as a stress domain, it is optimal for the /ʔ/ to be parsed as a coda, thus satisfying the higher-ranked FINAL-C constraint, as opposed to its being parsed into the Nucleus, which would satisfy the lower-ranked NOCODA constraint. As both the tableaux above apply to monosyllabic feet with a final /Vʔ/, let us consider what the role of these postulated constraints is in the case of a bisyllabic iambic foot without /Vʔ/ in the head syllable of the foot. In the form for ‘armadillo’ (237), both candidates satisfy FT-FORM=IAMBIC and FT-BIN-σ, so these are not included in the tableau here. Of present relevance is that both FINAL-C and NO-CODA are seen to be violated in the winning candidate. What is more important for well-formedness here is that no segments are deleted, in violation of MAX-SEG. (236) MAX-SEG: Every segment in the input must have a correspondent in the output.  (237) k’akxó   ‘armadillo’  (238)        / k’akxo /     MAX-SEG FINAL-C NOCODA  a.  (k’ak.x o ́)  * *  b.     (k’a.x o ́) *! *    150 Given ‘armadillo’ as a basis of comparison, let us turn to examine a word with a final /ʔ/ that is parsed not to a monosyllabic foot as in (239), but rather into a bisyllabic iambic foot: (239)  jijeʔ́   ‘caraguata’  (240)       /   jijeʔ /      MAX-SEG FINAL-C NOCODA  a.  (ji.jeʔ́)   *  b.     (ji.je ̰́ )  *!   c.     (ji.je ́ ) *!     Importantly, what these tableaux in (238) and (240) establish is that even though satisfaction of the FINAL-C constraint is not “necessary” to create a well-formed bisyllabic iambic foot (238), it is better (240) for the glottal component to be parsed as a coda consonant (despite violating NOCODA) than it is for it to be parsed into the Nucleus, which results in violating the higher ranked FINAL-C constraint.  Having examined a diversity of contexts where /ʔ/ is parsed into the coda of a syllable, consider now the complementary set of cases where a /ʔ/ is parsed into the Nucleus. These are cases like /wɑʔs/ [wɑ̰́s] ‘sky’ where another C follows the /ʔ/ in the input string and is parsed to the coda position.   (241) a. wɑ̰́s ‘sky’ (216b)  (242)  /wɑʔs / *CC]σ MAX-SEG FINAL-C *COMPLEXNUC NOCODA  a.    wɑ́ʔs  *!    *  b.    wɑ́ʔ  *!   *  c.     wɑ̰́    *! * *   d. wɑ̰́s    * *   The crucial constraints are: (243) FINAL-C  »  NOCODA  (244) *COMPLEXCODA (*CC]σ),   MAX-SEG  »  *COMPLEXNUC (245) MAX-SEG »  NOCODA  (see tableaux for ‘armadillo’ (238) and ‘caraguata’ (240))  151 As was established in Chapter 2, Nivaĉle does not tolerate complex codas. As seen in (242), the ranking of both *COMPLEXCODA (*CC]σ) and FINAL-C above NOCODA accommodates this. What a comparison of candidates (a) and (d) in the tableau in (242) also shows is that Nivaĉle tolerates a *COMPLEXNUC violation over a *COMPLEXCODA violation. A comparison of candidates (b) and (d) further illustrates that incorporating the glottal into the nucleus avoids deletion of the underlying consonant /s/ (which would entail a MAX-SEG violation). In addition to incurring this same MAX-SEG violation, what a comparison of candidate (c) with the winning candidate (d) further establishes is that enforcing NOCODA, through either deletion or nuclear parsing of /ʔ/, is not optimal. The ranking relations that have been established thus far are summarized in (243)-(245). What remains to be clarified is the moraic status of the glottal affiliation within the syllable. The generalizations that have been discussed here present an interesting paradox. To recapitulate, it has been shown that there is a consistent correlation between [Vʔ] (whether realized as a glottalized vowel or as a vowel plus [ʔ] coda) and stress. On the one hand, a syllable with nuclear or coda glottal realization is always stressed, regardless of its position in a word:   (246) a. ɬpeʔ̀.ja.ʔa ́n ‘you are hearing’  b. t’u ̰́k.ʃi ‘obstructed’  c. ku ̰̀k.ti.nís ‘thunder (PL)’  d. xi.βeʔ́.k ͡la ‘moon’  e. ji.fɑ̀ʔ.jɑ.ʔɑ́n ‘s/he is/was flying’  f. faj.xoʔ́ ‘charcoal’  g. jik.t ͡su ̰́k ‘silk floss tree’   h. ka ̀s.fe.tá ̰s ‘our root/medicine’   152 On the other hand, all cases where an underlying /Vʔ/ sequence is ‘deglottalized’ (under any of the various circumstances discussed in §2.5.5, §4.5, §6.5, etc.), occur in unstressed syllables. In conformity with cross-linguistic generalizations, it is hypothesized in the present analysis, first, that /ʔ/ is a weight-bearing, i.e. moraic, segment, and secondly, that Nivaĉle stress is a weight-sensitive system. The bimoraic status of a tautosyllabic vowel plus glottal sequence accounts for the ubiquituous association of these syllables with stress, in conformity with the WEIGHT-TO-STRESS constraint (to be discussed further in §4.5.2.1):  (247) WEIGHT-TO-STRESS PRINCIPLE (WSP): Heavy (i.e. bimoraic) syllables are stressed.                  (Prince & Smolensky 1993)   Concomitantly, the fact that deglottalization phenomena in Nivaĉle are consistently correlated with a lack of stress entails deletion of the entire lexical representation for /ʔ/: both its featural and its moraic specification delete. Here is the paradox. On the one hand, it has been argued in terms of Word Minimality constraints (§4.2) that [ʔ] can and will function as a coda if there is no other consonant parsed into that position. On the other hand, however, no other coda consonant in Nivaĉle contributes to syllable weight. Aside from [ʔ], Nivaĉle can be categorized cross-linguistically as a Type (ii) language (as discussed in §4.1 above) where coda consonants are systematically non-moraic. Because /ʔ/ is consistently weight-bearing, the claim here is that, like vowels, a /ʔ/ in Nivaĉle is underlyingly moraic, like vowels: thus, its moraic status is part of its lexical representation, as opposed to being contextually determined through the postulated WEIGHT-BY-POSITION constraint, which claims that “Coda consonants are moraic” (Hayes 1989, Sherer 1994, Kager 1999:147). No coda consonants in Nivaĉle satisfy WEIGHT-BY-POSITION. The representational question, however, is this: when a glottal is parsed into a coda position (evidence for which is the fact that such cases satisfy the FINAL-C constraint: i.e. a [CVʔ] word satisfies  153 MinWd, and a [CVʔ] syllable can bear secondary stress), what happens to its affiliation with its underlying mora? While acknowledging that there are various possible interpretations of this, the present analysis seeks to foreground the following broad-based generalizations in Nivaĉle (see §4.1) by adopting the hypothesis that the mora needs to be parsed into the Nucleus: 1. Coda consonants do not bear weight.  2. The only moraic segments are vowels and glottal stop.  3. Only the Nucleus can host weight. 4. Stress prominence is sensitive to Nuclear weight. 5. MinWd is sensitive to closed syllable status (FINAL-C) independent of syllable weight. Let us turn now to a detailed examination of how these factors interact with other properties of the Nivaĉle stress system. 4.4 Stress patterns in Nivaĉle Stress in Nivaĉle is associated with the following phonological and phonetic properties. First, all lexical words have primary stress (the ‘obligatoriness’ parameter; Hyman 2006:231) with one syllable bearing the highest degree of prominence (the ‘culminativity’ parameter; Hyman 2006:231). Second, on the basis of samplings included in this dissertation, Nivaĉle stressed vowels are slightly but consistently longer than unstressed vowels; increased duration is correlated with stressed vowels. Another acoustic correlate of stress is higher pitch. Further, because glottalized vowels are bimoraic, they are always stressed (in accordance to the WEIGHT-TO-STRESS Principle (247)) and are characteristically double the duration of modal vowels (§3.2.3). Third, unstressed vowels may be reduced or deleted. As the conditions (e.g. fast speech) governing the reduction/elision behaviour of prosodically weak vowels are not investigated in detail in this work, the discussion presented here (unless otherwise noted) focusses on the level of analysis that entails full specification of these vowels and, as such, faithfully represents the attested speech of my consultants (in a range of speech styles ranging from ‘careful’ to ‘relatively informal’, but not focussing on ‘fast’ or ‘very informal’). Fourth, unstressed vowels undergo translaryngeal vowel harmony (§2.6.3, §3.3.4).  154  Metrical theory (Hayes 1985, 1987, 1995; McCarthy & Prince 1986, Prince 1990, Kager 1999) assumes a small universal inventory of foot types: the quantity-insensitive syllabic trochee and the quantity-sensitive “moraic trochee” (head-initial or left foot prominence), and the quantity-sensitive iamb (head-final or right foot prominence).   (248) Inventory of foot types: a. Syllabic trochee (quantity-insensitive): (σσ) b. Moraic trochee (quantity-sensitive): (LL) (H) c. Iamb (quantity-sensitive): (LL) (H) (LH)    In this section, I consider the different word-stress patterns in Nivaĉle and propose that (i) the foot type is iambic, and (ii) the Nivaĉle language has a quantity-sensitive stress system, where the moraic weight of /ʔ/ is consistently correlated with stress prominence, and (iii) a CVC syllable in word initial position – and only in this position of convergent morphological and prosodic prominence – is stressed, this correlating with the monosyllabic CVC foot that was argued in Section 4.2 to constitute the MinWd in Nivaĉle.51 As argued in Section 4.3, glottalized vowels and glottal codas contribute additional weight; the glottal stop /ʔ/ is underlyingly moraic. This mora is parsed to the Nucleus of the syllable. In addition, I advance the hypothesis that stress assignment in the nominal and verbal domains varies according to whether prefixes or suffixes are attached. The prosodic system that emerges is analyzed in terms of systematic constraints on foot construction from both the left and right edges of words where the prosodic domains are defined by hierarchical morpho-syntactic processes of prefixation and suffixation.                                                 51 An initial CVC syllable is optionally stressed as secondary stress assignment can be overridden by other factors.  155 4.4.1 Previous accounts The only previous analysis of the Nivaĉle stress system can be found in Stell (1989:81-83). The author provides a brief characterization of the stress patterns in the language. Below, I present her main arguments and discuss the examples she provides.  First, Stell (1989:81) characterizes the locus of primary stress as follows:  Primary stress falls on the root and in several inflectional and derivational affixes. In disyllabic or trisyllabic roots, primary stress mainly falls on the last syllable [249b; 250a]. In morphological constructions constituted by mono- and disyllabic roots, certain inflectional and derivational suffixes shift primary stress to the last syllable [249a,b; 250a,b]. In morphological constructions of four or more syllables, primary stress mainly stays on the root and on certain inflectional and derivational suffixes [my translation from Spanish/AG].  Some illustrative examples follow that description. In the left hand column, I include a faithful copy of Stell’s examples, on the right side my interpretation/analysis of her representations (including the morphemic breakdown). Note that Stell uses the standard IPA conventions for the representation of primary and secondary stress, while I use an acute and grave accent respectively.  Stell AG (249)  a. ˈis    ‘good’  ε.  φ. b. isˈis   γ.    ‘good (pl)’   a’. ∅-[ʔ]ís     3S-good     ‘s/he/it is good’  b’. ∅-[ʔ]is-[í]s      3S-good-PL      ‘they are good(pl)’    (250)  a. kaˈsus    ‘pumpkin’  b. kasuˈsik η.      ‘pumpkins’ a’. kasús      ‘pumpkin’  b’.  kasus-[í]k       pumpkin-PL        ‘pumpkins’     156 (251)  a. xaˈnỉs    ‘I tattoo’   b. xanisˈʔin ι.     ‘I am tattooing’ a’     xa-n-í ̰s        1S-REF-write/mark       ‘I tattoo myself’    b’.   xa ̀ʔ-n-is-ʔín          1S-REF-write/mark-IPFV        ‘I am tattooing myself’ 52  (252)  a.  ˈweʔɬa      ‘oneʼ a’.  [βɛ́ʔɬə]        ‘one’  The data in my corpus agree with Stell’s documentation of primary stress as presented in (249)-(252). Interestingly, Stell shows an example, viz. (252),  that does not follow the word final stress pattern that she claims is ‘mainly’ found in disyllabic and trisyllabic roots. I propose that primary stress falls on the penultimate syllable in this and other similar cases (see §4.5.2) because glottals are moraic and thus attract stress. While example (252) could in principle be an exception to the main stress pattern for disyllabic words, no reference to the special status of the glottal stop is ever mentioned or discussed by Stell.  Second, Stell (1989:83) claims that “secondary stress falls on the second and fourth syllable of morphological constructions of four or more syllables, as long as it does not coincide with primary stress (…) and in certain inflectional and derivational affixes (prefixes and suffixes)”. According to the author, the distribution of primary and secondary stress accounts for the fact that Nivaĉle does not allow sequences of more than two unstressed syllables (see below).   Stell (1989:83) illustrates this brief characterization of secondary stress with the following examples (my morphemic breakdown/AG; note that in my transliteration of her data, secondary stress is marked as a grave accent and primary stress as an acute accent):                                                  52 It is not very clear to me why there is a vowel-glottal coda in the first syllable of ‘I am tattooing’. However, I present this data as faithful to my field records.  157 (253) a. wa ̀.t-a.si.nɑ́         INDEF.POSS-word     ‘someone’s word’         b.   wa ̀.t-a si.nɑ-n.-jaʃ́              INDEF.POSS-speak-?-NLMLZ     ‘someone’s way of speaking’  (254)  a.  wat’-éj.xat ͡s.xan.xát     INDEF.POSS-teaching   ‘someone’s teaching/education’       b.  wat’-éj-xat ͡s.xa ̀n.xat-és                INDEF.POSS-word-teaching-PL           ‘someone’s educations’  (255)  a.  -tɑ.xɑ́x     ‘to grind’           b.   ji-tɑ́.xɑx.-ke-nɑ́.kʼ-eɬ̀     1POSS-grind-LOC-RES-EXCL.PL     ‘our pastas’  I am not in accord with Stell’s documentation of these primary and secondary stress examples, nor with certain aspects of her discussion.    First, my data differ in that secondary stress falls on the second, and not the first syllable of (253a) ‘someone’s word’; in my analysis, an iambic foot is formed with reference to the left-edge of the possessive prefix; the final foot bears primary stress. Even though I posit that the possessive prefixes [kas] and [βat] can bear stress (§4.5.4), it is only before consonant-initial roots (among other conditions). Before vowel-initial roots as in (253), the last consonant of [βat] gets parsed as the onset of the following root [βa.t-a ̀.sinɑ́]. In addition, I am not in agreement with where she has marked stress in (253b); I documented [βa.ta ̀.si.nɑ-n.jaʃ́]. Further, Stell’s example in (253b) shows a sequence of three unstressed syllables, which actually contradicts her aforementioned observation that there cannot be more than two unstressed syllables in a row in this language. Note that the stress pattern that I  158 documented is in accord with the generalization that there are no sequences of three or more unstressed syllables.  Second, given the data in (254b), it is not clear why ‘teaching’ has two primary stresses at both the left and right edge and secondary stress at the middle of the word (on the fourth syllable). According to my documentation, there is only one primary stress (the rightmost one here), and the other prominent syllables receive secondary stress. Third, even though I have not documented the form in (255), this example presents some unresolved issues: (i) secondary stress is found in this form in final position (without any explanation), and (ii) there is adjacent stress (treated as “stress clash” in the present analysis) in the last two syllables. In the following sections, I present evidence that stress clash is not allowed in Nivaĉle: e.g. some final syllables are left unparsed in order to not violate *CLASH.   Yet, from Stell’s brief description and puzzling presentation of secondary stress, it can be interpreted that stress is related to morphological constructions. The following section explores this issue. 4.5  Stress and affixation processes It is proposed in the present analysis that in Nivaĉle, there are four basic domains for stress assignment. The Root (Rt), the Morphological Stem 1 (MSt1), the Morphological Stem 2 (MSt2), and the Morphological Word (MWd). The root plus an “inner” level of suffixes defines the MSt1. Prefixation onto MSt1 defines the next higher domain: MSt2. An outer layer of suffixation to the MSt2 defines the MWd.  The basic claim advanced here is that each of these morphological categories (MCat) defines a prosodic domain that is relevant to identifying the locus of stress. Stress in Nivaĉle is edge-based and quantity-sensitive, building feet consistent with FOOTFORM=IAMBIC: a series of ranked alignment constraints and the WEIGHT-TO-STRESS Principle are the main constraints responsible for stress assignment.   159  It is important to highlight that the MCats are hierarchically layered. Even if a domain that is an MSt1 does not undergo any further prefixation or suffixation, it is nonetheless parsed and labelled within the nested structure as an MSt2 and as an MWd. Whether affixation does or does not occur is relevant to whether the output form meets a characterization of morphologically derived or non-derived (§4.5.3, §4.5.7), and, concomitantly, whether that output form is subject to the particular alignment constraint associated with the domain in question.    PRWD       MWD                        MST2            -SUFn      MST1                    (PREF1-)    (-PREF2-) (-)ROOT     (-SUF1)    (-SUFn)  Figure 4.2  Word-internal morphological relationships  4.5.1 Nominal domain Nivaĉle nouns can be modified by a number of affixes that express inflectional and derivational categories. A templ