UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Non-exhaustive parsing : phonetic and phonological evidence from St'át'imcets Caldecott, Marion Gerda 2009

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

Item Metadata

Download

Media
24-ubc_2009_spring_caldecott_marion.pdf [ 11.03MB ]
Metadata
JSON: 24-1.0067159.json
JSON-LD: 24-1.0067159-ld.json
RDF/XML (Pretty): 24-1.0067159-rdf.xml
RDF/JSON: 24-1.0067159-rdf.json
Turtle: 24-1.0067159-turtle.txt
N-Triples: 24-1.0067159-rdf-ntriples.txt
Original Record: 24-1.0067159-source.json
Full Text
24-1.0067159-fulltext.txt
Citation
24-1.0067159.ris

Full Text

NON-EXHAUSTIVE PARSING: PHONETIC AND PHONOLOGICAL EVIDENCE FROM ST’ÁT’IMCETS by MARION GERDA CALDECOTT B.A. (Hons.), The University of British Columbia, 1997 M.A., The University of British Columbia, 1999  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY  in  THE FACULTY OF GRADUATE STUDIES  (Linguistics)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) April 2009 © Marion Gerda Caldecott  Abstract This thesis tests the prediction of the Prosodic Hierarchy model that distinct phonological domains have distinct acoustic characteristics. Specifically, I test the prediction that if non-exhaustively parsed structures are permitted by the model, there should be converging phonological and phonetic evidence that distinguishes them from exhaustively parsed domains. Little phonological and no phonetic evidence has previously been provided to substantiate the proposed structures. I introduce the term ‘extrapod’ to refer to syllables that are not parsed at the foot level and present phonological and phonetic evidence from St’át’imcets (Lillooet Salish) that shows that extrapods are distinguished from exhaustively parsed domains, as predicted. Empirically, this thesis presents the first phonetic documentation of the prosody of St’át’imcets, including phrase-level intonation, stress and boundary effects. The results of this research are intended to both add to the documentation of an endangered language and benefit the community. Theoretically, this thesis confirms a previously untested prediction of current models of prosody, namely that non-exhaustively parsed structures are distinct from exhaustively parsed structures. Results from the investigation of segmental phonological processes, and acoustic correlates of prominence and boundary strength, support a weak but constrained interpretation of the phonetics-phonology mapping. Methodologically, this thesis develops protocols that elicit rigorous phonetic data while being suitable for fieldwork with First Nations elders, and that can be easily adapted into language teaching materials. The combination of verbal and pictoral contexts allows for controlled repetitions of tokens in natural linguistic contexts and provides materials that can be adapted for classroom use.  ii  Table of Contents Abstract ................................................................................................................................................................. ii Table of Contents................................................................................................................................................iii List of Tables ....................................................................................................................................................... vi List of Figures....................................................................................................................................................viii List of Abbreviations........................................................................................................................................... x Acknowledgements ............................................................................................................................................. xi Dedication ..........................................................................................................................................................xiii Chapter 1 Overview and Background............................................................................................................. 1 1.1 INTRODUCTION ........................................................................................................................................ 1 1.2 ORGANISATION OF THE THESIS............................................................................................................... 7 1.3 ST’ÁT’IMCETS IN THE CONTEXT OF THE SALISH LANGUAGE FAMILY .................................................. 7 1.4 GENERAL PHONOLOGY ......................................................................................................................... 10 1.4.1 Root structure ................................................................................................................................... 10 1.4.2 Consonant and vowel inventory ...................................................................................................... 11 1.4.3 Glottalised resonants ........................................................................................................................ 13 1.4.4 Stress: Phonology ............................................................................................................................. 16 1.4.5 Stress: Phonetics............................................................................................................................... 20 1.4.6 Intonation .......................................................................................................................................... 21 1.4.7 Prosodic domains: Phonology ......................................................................................................... 21 1.4.8 Prosodic domains: Phonetics ........................................................................................................... 23 1.5 ASSUMPTIONS MADE REGARDING THE PROSODIC HIERARCHY .......................................................... 26 1.5.1 Prosodic domains ............................................................................................................................. 27 1.5.2 Morpho-syntactic mapping .............................................................................................................. 27 1.5.3 Headedness/Prominence .................................................................................................................. 28 1.5.4 Boundaries ........................................................................................................................................ 29 1.5.5 Inventory ........................................................................................................................................... 30 1.6 METHODOLOGICAL, EMPIRICAL AND THEORETICAL CONTRIBUTIONS OF THIS THESIS .................... 31 1.6.1 Empirical contributions.................................................................................................................... 31 1.6.2 Theoretical contributions ................................................................................................................. 32 Chapter 2 St'át'imcets Intonation Contours ................................................................................................ 34 2.1 INTRODUCTION ...................................................................................................................................... 34 2.2 METHODOLOGY ..................................................................................................................................... 35 2.2.1 Speakers ............................................................................................................................................ 35 2.2.2 Stimuli and recording ....................................................................................................................... 35 2.3 RESULTS ................................................................................................................................................ 36 2.3.1 General results .................................................................................................................................. 36 2.3.2 Comparison of means and pitch ratios............................................................................................ 39 2.4 DISCUSSION ........................................................................................................................................... 42 2.5 CONCLUSION ......................................................................................................................................... 43 Chapter 3 Extrapods: History and Phonological Evidence ........................................................................ 44 3.1 3.2 3.3 3.4  INTRODUCTION ...................................................................................................................................... 44 PROSODIC HIERARCHY BACKGROUND ................................................................................................. 44 THE EVOLUTION OF THE STRICT LAYER HYPOTHESIS ........................................................................ 47 EXTRAPODS ........................................................................................................................................... 50  iii  3.4.1 Markedness ....................................................................................................................................... 52 3.4.2 Availability ....................................................................................................................................... 52 3.4.3 Size .................................................................................................................................................... 53 3.4.4 Location ............................................................................................................................................ 53 3.4.5 Visibility ........................................................................................................................................... 53 3.4.6 Constituent status ............................................................................................................................. 53 3.4.7 Language-specific (i.e. non-structural) characteristics .................................................................. 56 3.5 ST’ÁT’IMCETS DATA ............................................................................................................................. 57 3.5.1 Vowel alternating suffixes ............................................................................................................... 57 3.5.2 Transitivising paradigm glottal alternation..................................................................................... 63 3.5.3 Phonology summary......................................................................................................................... 74 3.6 POSSIBLE MODELS ................................................................................................................................ 74 3.6.1 Extrapod (a.k.a. Weak Layering (Itô & Mester 1992)).................................................................. 75 3.6.2 Degenerate foot ................................................................................................................................ 76 3.6.3 Ternary foot ...................................................................................................................................... 76 3.6.4 Recursive feet ................................................................................................................................... 77 3.7 APPLYING PHONOLOGICAL PREDICTIONS TO ST’ÁT’IMCETS .............................................................. 78 3.7.1 Foot-level processes: Glottal alternation ........................................................................................ 78 3.7.2 Word-level processes: Stress assignment ....................................................................................... 79 3.7.3 Phrase-level processes: Phrasal accent ........................................................................................... 80 3.7.4 Head Saliency maximisation: Vowel reduction ............................................................................. 83 3.8 PHONETIC PREDICTIONS ........................................................................................................................ 87 3.9 CONCLUSION ......................................................................................................................................... 90 Chapter 4 General Methodology and Protocol Development..................................................................... 92 4.1 INTRODUCTION ...................................................................................................................................... 92 4.2 EXPERIMENTAL CHOICES ...................................................................................................................... 92 4.2.1 Modality ............................................................................................................................................ 92 4.2.2 Specific correlates ............................................................................................................................ 93 4.3 INFORMATION ON SPEAKERS ................................................................................................................ 93 4.4 TOKEN SELECTION ................................................................................................................................ 94 4.5 MEASUREMENT AND ANALYSIS ........................................................................................................... 95 4.6 PROTOCOL DEVELOPMENT ................................................................................................................... 96 4.6.1 Experiment 1................................................................................................................................... 101 4.6.2 Experiment 2................................................................................................................................... 102 4.6.3 Advantages and disadvantages of the protocols........................................................................... 105 4.7 OTHER ELICITATION METHODS .......................................................................................................... 106 4.8 CONCLUSION ....................................................................................................................................... 110 Chapter 5 Prominence in St’át’imcets ......................................................................................................... 111 5.1 INTRODUCTION .................................................................................................................................... 111 5.2 PHONETIC BACKGROUND.................................................................................................................... 112 5.2.1 Hypotheses and predictions ........................................................................................................... 115 5.3 METHODOLOGY ................................................................................................................................... 116 5.3.1 Speakers .......................................................................................................................................... 116 5.3.2 Stimuli ............................................................................................................................................. 116 5.3.3 Measurements and analysis ........................................................................................................... 120 5.4 RESULTS .............................................................................................................................................. 122 5.4.1 Head versus non-head results ........................................................................................................ 125 5.4.2 Extrapods versus other heads ........................................................................................................ 129 5.5 DISCUSSION ......................................................................................................................................... 130 5.6 CONCLUSION ....................................................................................................................................... 132 Chapter 6 Prominence Distinctions: Unstressed Vowels versus Extrapods........................................... 133 6.1 6.2  INTRODUCTION .................................................................................................................................... 133 BACKGROUND ..................................................................................................................................... 134  iv  6.3 6.4 6.5 6.6 6.7 6.8  METHODOLOGY ................................................................................................................................... 137 RESULTS .............................................................................................................................................. 137 DISCUSSION ......................................................................................................................................... 139 WITHIN-WORD VERSUS ACROSS-WORD COMPARISONS: A CASE STUDY ........................................ 141 DISCUSSION OF WITHIN-WORD RESULTS .......................................................................................... 145 CONCLUSION ....................................................................................................................................... 146  Chapter 7 Prosodic Strengthening in St’át’imcets ..................................................................................... 151 7.1 INTRODUCTION .................................................................................................................................... 151 7.2 PROSODIC STRENGTHENING BACKGROUND ...................................................................................... 152 7.3 METHODOLOGY ................................................................................................................................... 156 7.3.1 Subjects ........................................................................................................................................... 157 7.3.2 Stimuli ............................................................................................................................................. 157 7.3.3 Equipment....................................................................................................................................... 160 7.3.4 Recording........................................................................................................................................ 161 7.3.5 Measurements and analysis ........................................................................................................... 164 7.4 RESULTS .............................................................................................................................................. 165 7.4.1 Domain-final unstressed vowels ................................................................................................... 165 7.4.2 Unstressed vowel discussion ......................................................................................................... 180 7.4.3 Domain-final extrapods ................................................................................................................. 183 7.4.4 Extrapod discussion........................................................................................................................ 194 7.5 CONCLUSION ....................................................................................................................................... 195 Chapter 8 Boundary Strength: Unstressed Vowels versus Extrapods.................................................... 196 8.1 INTRODUCTION .................................................................................................................................... 196 8.2 BACKGROUND ..................................................................................................................................... 197 8.3 METHODOLOGY REVIEW .................................................................................................................... 201 8.3.1 Subjects ........................................................................................................................................... 202 8.3.2 Stimuli ............................................................................................................................................. 202 8.3.3 Equipment....................................................................................................................................... 203 8.3.4 Recording........................................................................................................................................ 203 8.3.5 Measurements and analysis ........................................................................................................... 204 8.4 RESULTS .............................................................................................................................................. 204 8.4.1 Word-final unstressed versus extrapods ....................................................................................... 205 8.4.2 Phrase-final unstressed versus extrapods...................................................................................... 208 8.5 DISCUSSION ......................................................................................................................................... 214 8.6 CONCLUSION ....................................................................................................................................... 215 Chapter 9 Conclusion...................................................................................................................................... 217 Bibliography ..................................................................................................................................................... 234 Appendix A: Phoneme Conversion Chart ................................................................................................... 246 Appendix B: Other Transitivising Glottal Alternations............................................................................ 248 Appendix C: Prominence Means Graphs .................................................................................................... 258 Appendix D: Comparison of Initial versus Non-Initial Primary Stress ................................................. 263 Appendix E: Initial Extrapods from Experiment 2 ................................................................................... 265 Appendix F: PWord-initial versus PWord-final Extrapods..................................................................... 274 Appendix G: Boxplots for HD’s Peak Percentage...................................................................................... 281  v  List of Tables Table 1.1 The Salish language family.............................................................................. 8 Table 1.2 Consonant inventory (modified from van Eijk (1997))................................... 11 Table 1.3 Vowel inventory (modified from van Eijk (1997)) ......................................... 12 Table 1.4 Vowel inventory and allophones (Shahin & Blake: 315) ................................ 12 Table 1.5 Beck (1999) and Beck and Bennett’s (2007) definitions of domains in Lushootseed .......................................................................................................... 23 Table 2.1 Speakers and numbers of tokens .................................................................... 36 Table 3.1 De Lacy’s example 27 (2006: 228) ................................................................ 56 Table 3.2: Characteristics of extrapods.......................................................................... 57 Table 3.3 Vowel reduction in non-head tableau............................................................. 62 Table 3.4 Extrapod tableau............................................................................................ 62 Table 3.5 Tableau accounting for surface glottalisation ................................................. 68 Table 3.6 Tableau accounting for neutralisation following the stressed vowel ............... 69 Table 3.7 Accounting for neutralisation in unparsed foot............................................... 69 Table 3.8 Retention of glottalised resonant in root......................................................... 70 Table 3.9 Root constraints account for retention of glottalisation................................... 70 Table 3.10: How the models account for resonant glottalisation .................................... 78 Table 3.11 How the models account for stress assignment............................................. 79 Table 3.12 How the models account for phrase-level intonation contours...................... 82 Table 3.13 How the models account for non-head reduction.......................................... 83 Table 3.14 Summary of predictions............................................................................... 84 Table 3.15 Summary of extrapod attributes ................................................................... 86 Table 3.16 Summary of acoustic predictions ................................................................. 90 Table 4.1 Formant analysis by speaker and vowel ......................................................... 95 Table 5.1: Speaker information ................................................................................... 116 Table 5.2 Tokens for /a/ and /i/.................................................................................... 118 Table 5.3 Tokens for /u/ .............................................................................................. 119 Table 5.4 Sample target sentences (given in the orthography)...................................... 120 Table 5.5 Results for /a/ .............................................................................................. 123 Table 5.6 Results for /i/ ............................................................................................... 124 Table 5.7 Results for /u/ .............................................................................................. 125 Table 5.8 Primary stress > unstressed. ......................................................................... 126 Table 5.9 Primary stress > extrapod............................................................................. 127 Table 5.10 Primary stress = secondary stress ............................................................... 128 Table 5.11 Secondary stress > extrapod....................................................................... 129 Table 5.12 Non-initial primary stress > extrapod......................................................... 130 Table 6.1 Hypotheses .................................................................................................. 135 Table 6.2 Extrapod > unstressed.................................................................................. 138 Table 6.3 Direction of difference................................................................................. 139 Table 6.4 Mean relative differences between root and affix /a/ .................................... 142 Table 6.5 Mean relative differences between primary stress and secondary stress........ 142 Table 6.6 Relative within-foot differences between secondary stress versus unstressed and non-initial primary stress versus unstressed in /u/ tokens............................... 143 Table 6.7 Relative differences between unstressed vowels and extrapods for /u/.......... 145  vi  Table 7.1 Bessell (1997) formant values...................................................................... 154 Table 7.2 Shahin (2002: 224-226) formant value means .............................................. 155 Table 7.3 Unstressed vowel predictions....................................................................... 156 Table 7.4 Extrapod predictions.................................................................................... 156 Table 7.5 Speaker and token information .................................................................... 157 Table 7.6 Unstressed syllables across boundaries ........................................................ 158 Table 7.7 Final extrapods across boundaries................................................................ 159 Table 7.8 Tokens......................................................................................................... 160 Table 7.9 Northern dialect target sentences ................................................................. 162 Table 7.10 Southern dialect target sentences................................................................ 163 Table 7.11 AP’s unstressed vowel means .................................................................... 165 Table 7.12 CA’s unstressed vowel means.................................................................... 165 Table 7.13 HD’s unstressed vowel means.................................................................... 166 Table 7.14 LT’s unstressed vowel means .................................................................... 166 Table 7.15 AP’s unstressed vowel significant differences............................................ 166 Table 7.16 CA’s unstressed vowel significant differences ........................................... 170 Table 7.17 HD’s unstressed vowel significant differences ........................................... 173 Table 7.18 LT’s unstressed vowel significant differences ............................................ 176 Table 7.19 Summary of results.................................................................................... 180 Table 7.20 Extrapod means across boundaries............................................................. 183 Table 7.21 Extrapod significant differences................................................................. 184 Table 7.22 Results for predictions of peripherality in domain-final extrapods.............. 194 Table 8.1 Hypotheses and their predictions................................................................. 200 Table 8.2 Unstressed vowel versus extrapod predictions.............................................. 201 Table 8.3 Unstressed vowel versus extrapod structures................................................ 202 Table 8.4 Unstressed vowel and extrapod tokens......................................................... 202 Table 8.5 Northern dialect target sentences ................................................................. 203 Table 8.6 Southern dialect target sentences ................................................................. 204 Table 8.7 Word-final means and standard deviation .................................................... 205 Table 8.8 Word-final significant differences (unstressed > extrapod)........................... 205 Table 8.9 Phrase-final means and standard deviation................................................... 209 Table 8.10 Phrase-final (unstressed > extrapod) .......................................................... 209 Table 8.11 Word-final summary results....................................................................... 213 Table 8.12 Phrase-final summary results ..................................................................... 213 Table 8.13 Summary of boundary strength results ....................................................... 214 Table 9.1 Summary of results by speaker .................................................................... 219  vii  List of Figures Figure 2.1 AP’s answer, “Yes, he saw one dog (lit. “One was the dog he saw”) ............ 37 Figure 2.2 AP’s answer, “Yes, he saw one coyote.”....................................................... 37 Figure 2.3 AP’s question, “Did he see any chickens?”................................................... 38 Figure 2.4 AP’s question “Did she get new hats?”......................................................... 38 Figure 2.5 AP’s mean F0 values.................................................................................... 40 Figure 2.6 CA’s mean F0 values ................................................................................... 40 Figure 2.7 LT’s mean F0 values .................................................................................... 40 Figure 2.8 Mean pitch ratios between initial and nuclear vowel ..................................... 41 Figure 2.9 Mean pitch ratios between nuclear and final vowels ..................................... 42 Figure 3.1 “Oh, nilh sMichelle ta ats’xenána.” .............................................................. 80 Figure 3.2 “Oooh, cw7aoz, áts’xenas i a7en’wása sqlaw’.” ........................................... 81 Figure 3.3 “Iy, á7en’was i tsíkena áts’xenas.” ............................................................... 82 Figure 4.1 Slide used to elicit “Iy, kwánenskan ti ts’úqwaz’a.”.................................... 101 Figure 4.2 Slide used to elicit “T’ec ti/ta xúsuma múta7 i q’welápa.” .......................... 103 Figure 4.3 Slide used to elicit “Wá7lhkalh ít’em lhas pipántsek muta7 lhas sútik.”...... 104 Figure 4.4 Slide used to elicit “N’áscit ku músmustsu (kw)s Henry.” .......................... 105 Figure 4.5 Map used in ‘Map task’ .............................................................................. 108 Figure 4.6 “Put task” ................................................................................................... 109 Figure 6.1 Mean duration values for /u/....................................................................... 144 Figure 7.1 AP’s schwa, stressed /a/ and unstressed /a/ ................................................. 167 Figure 7.2 AP’s unstressed /a/ ..................................................................................... 168 Figure 7.3 AP’s schwa, stressed /u/ and unstressed /u/................................................. 169 Figure 7.4 AP’s unstressed /u/ ..................................................................................... 169 Figure 7.5 CA’s schwa, stressed /a/ and unstressed /a/................................................. 170 Figure 7.6 CA’s unstressed /a/..................................................................................... 171 Figure 7.7 CA’s schwa, stressed /u/ and unstressed /u/ ................................................ 172 Figure 7.8 CA’s unstressed /u/..................................................................................... 172 Figure 7.9 HD’s schwa, stressed /i/ and unstresed /i/ ................................................... 173 Figure 7.10 HD’s unstressed /i/ ................................................................................... 174 Figure 7.11 HD’s schwa, stressed /u/ and unstressed /u/ .............................................. 175 Figure 7.12 HD’s unstressed /u/ .................................................................................. 176 Figure 7.13 LT’s schwa, stressed /i/ and unstressed /i/................................................. 177 Figure 7.14 LT’s unstressed /i/ .................................................................................... 178 Figure 7.15 LT’s schwa, stressed /u/ and unstressed /u/ ............................................... 178 Figure 7.16 LT’s unstressed /u/ ................................................................................... 179 Figure 7.17 AP’s extrapod /a/, stressed /a/ and schwa.................................................. 185 Figure 7.18 AP’s extrapod /a/...................................................................................... 185 Figure 7.19 AP’s extrapod /u/, stressed /u/ and schwa ................................................. 186 Figure 7.20 AP’s extrapod /u/...................................................................................... 186 Figure 7.21 CA’s extrapod /a/, stressed /a/ and schwa ................................................. 187 Figure 7.22 CA’s extrapod /a/ ..................................................................................... 188 Figure 7.23 CA’s extrapod /u/, stressed /u/ and schwa ................................................. 188 Figure 7.24 CA’s extrapod /u/ ..................................................................................... 189  viii  Figure 7.25 HD’s extrapod /i/, stressed /i/ and schwa................................................... 190 Figure 7.26 HD’s extrapod /i/...................................................................................... 190 Figure 7.27 HD’s extrapod /u/, stressed /u/ and schwa................................................. 191 Figure 7.28 HD’s extrapod /u/ ..................................................................................... 191 Figure 7.29 LT’s extrpod /i/, stressed /i/ and schwa ..................................................... 192 Figure 7.30 LT’s extrapod /i/....................................................................................... 192 Figure 7.31 LT’s extrapod /u/, stressed /u/ and schwa.................................................. 193 Figure 7.32 LT’s extrapod /u/...................................................................................... 193 Figure 8.1 CA’s word-final unstressed /a/ versus extrapod /a/...................................... 206 Figure 8.2 CA’s word-final unstressed /u/ versus extrapod /u/ ..................................... 207 Figure 8.3 HD’s word-final unstressed /i/ versus extrapod /i/....................................... 207 Figure 8.4 LT’s word-final unstressed /i/ versus extrapod /i/ ....................................... 208 Figure 8.5 AP’s phrase-final unstressed /a/ versus extrapod /a/.................................... 210 Figure 8.6 AP’s phrase-final unstressed /u/ versus extrapod /u/ ................................... 211 Figure 8.7 HD’s phrase-final unstressed /u/ versus extrapod /u/................................... 212 Figure 8.8 LT’s phrase-final unstressed /i/ versus extrapod /i/ ..................................... 212  ix  List of Abbreviations √ = root ‘-‘ = suffix ‘=’ = clitic ‘ ´ ‘ = primary stress ‘ ` ’ = secondary stress ps = initial primary stress nips = non-initial primary stress ss = secondary stress us = unstressed ex = extrapod V = full vowel C = consonant 1 = first person 2= second person 3 = third person obj = object subj = subject sg = singular pl = plural ∆ = Designated Terminal Element (DTE) GR = glottalised resonant Ft = foot PWord = Prosodic Word PPhrase = Prosodic Phrase IP = Intonation Phrase () = Foot boundary [] = PWord boundary OT = Optimality Theory VSO = Verb-Subject-Object word order VOS = Verb-Object-Subject word order x  Acknowledgements This thesis could not have been written without help and support from a number of people. First and foremost, I’d like to thank my wonderful speakers (in alphabetical order). Without their knowledge, friendship, patience and generosity, this work could never have been undertaken, much less completed. I’d like to thank Aggie Patrick for her enthusiasm, insights and constructive comments about the language and the research, Carl Alexander for taking time from his busy and important work to help me and share some of his wisdom and Ceda Scotchman for bringing such a cheerful and positive outlook for work which must have seemed rather repetitive. Thanks to Mesísl (Herman Dan) for being the best teacher and substitute Grandpa, to Laura Thevarge for her patience and for helping me develop the methodology and to Rose Whitley for working so hard to make sure my pronunciation was correct and for our enjoyable visits. On the academic front, a number of professors helped me on my PhD journey. Thank you to Lisa Matthewson, who, in allowing me to help with her book, brought me back into the fold. Her support of my work and tolerance of my poor semantics attitude guide me in the kind of teacher I want to be. Courses from Gunnar Hansson and Martina Wiltschko made me realize that coming back to school was the right thing to do. Thanks also to Matt Bauer who put me on the course of Prosodic Strengthening research, which shaped a large part of the thesis, and to Pat Shaw, for insight into Salish phonology. My research and how it was written has been shaped and guided by my fabulous committee. My eventual goal of a triangulation of evidence was foreshadowed by the triangulation of expertise from these three researchers: technical/methodological, language/fieldwork and phonology/exposition. I thank Eric Vatikiotis-Bateson for the technical help, discussions about natural language elicitation, which came to be so important to me, and a healthy scepticism of phonological structure. I owe Henry Davis more than I can express here: without him, I wouldn’t have had the amazing opportunities of working with members of the community or teaching the language. I value his mentoring regarding fieldwork and giving back to the community, his attentive reading of the research, data checking, help with grant writing and for expecting the best of me. I’m especially grateful to Doug Pulleyblank, who agreed to supervise me a  xi  second time, who expertly guided my research, while still letting me take it where I wanted it to go and who insisted on clarity. For the encouragement and patience while I agonized over big and small details, I am grateful! This thesis would not be half of what it is without his efforts. This research was funded by grants from the Jacobs Research Fund and SSHRC grants to me, Doug Pulleyblank and Henry Davis. I would also like to thank my examining committee: May Bernhardt for her interest, thorough reading and helpful comments, Matthew Gordon for his positive and helpful comments and Bryan Gick for all of his advice and feedback and who gave me the motivation to start the project. In terms of peer support, I couldn’t have survived without Kristín Jóhansdóttir, who climbed with me and gave linguistic and moral support and Karsten Koch, who shared his ideas and insights about methodology, his materials and his knowledge of Salish languages. Thanks to Strang Burton for sharing his experience with and insights into methodology as well. On the Island, I thank Sonya Bird for much successful collaboration and phonetics mentoring. I am grateful to Carolyn Pytlyk for the huge favour of editing my thesis, and to Janet Leonard for her help with the exam presentation and for her Salish phonology insights. I owe special thanks to the two classes I taught, in Mission and at UVic. They gave me new perspectives on language and linguistics. I’m very lucky to have a wonderful support network outside of the university. Tanya Lewis and Toby Chernoff supported me by offering sympathetic ears, a place to stay, food to eat and awesome buttons on exam day. My Grandma, and Tim and Julia and their families, always asked how it was going but never when I’d be finished. My parents, Christel and Peter, encouraged me when I was down and celebrated with me when I was up. I could never have successfully completed my PhD without their amazing support. I’m so lucky to have such wonderful parents. And finally, I thank my favourite antipod(e), Ryan, who left his beloved Australia to support me while I undertook this journey. It takes a lot of strength and patience to be the spouse of a PhD student, and I will always be grateful.  xii  Dedication  To my parents, who always believed I could do it.  xiii  Chapter 1 Overview and Background  1.1  Introduction  This thesis is predicated upon one of the fundamental hypotheses that underlie current phonological theory, namely that there is a systematic mapping between prosodic domains and their phonetic correlates (e.g. Pierrehumbert 1980; Beckman & Edwards 1990, 1994; Fougeron & Keating 1997; Pierrehumbert et al. 2000). This thesis makes two assumptions about the phonetic-phonology interface: i) prosodic constituents are the domains of phonological processes; and ii) these domains are distinguished systematically by speakers through the phonetic manifestations of prosody, i.e. the acoustic and articulatory correlates to prominence, intonation, and boundary strength. A well-known example of this is that phrases in English (amongst other languages) are marked by final vowel lengthening (Klatt 1976; Cooper & Paccia-Cooper 1980; Beckman & Edwards 1990) and a low boundary tone (e.g. Pierrehumbert 1980). There is no suggestion that the phonetics-phonology mapping is simple, direct or implemented universally in the same way. Pierrehumbert et al. (2000) liken the mapping between phonological categories and sound percepts to the complexity of colour perception. The perception of colours, they suggest, is mediated by the nature of our cone cells, how the optical nerve functions, and higher-level cortical processing that permits coherency under different illumination conditions. Language-specific colour terms and categories are learned by speakers. Likewise, “[a]lthough the relationship between a sound percept and a phonological category may seem very direct to an individual listener, it still presents to the scientist a dazzling degree of complexity and abstractness” (Pierrehumbert et al. 2000: 284). In spite of its complexity, the general hypothesis is that prosodic domains do have a systematic mapping to the phonetics.1 1  The reader is referred to Shattuck-Hufnagel & Turk (1996) and Keating (2006) for an overview.  1  Much research has focused on the suprasegmental correlates of prosodic domains, such as pitch contours (e.g. Pierrehumbert 1980; Beckman 1986; Beckman & Pierrehumbert 1986; Pierrehumbert & Beckman 1988; on English and Japanese). Other research, namely the literature on Prosodic Strengthening, has shown that prosodic domains also have distinct acoustic correlates in the form of segmental articulation. For example, Fougeron & Keating (1997) began a prolific line of research with their paper examining boundary effects on English segments. In their electropalatograph (EPG) study, participants replaced arithmetical expressions with reiterative /no/ syllables. Speakers had longer contact duration for initial [n] segments at higher (i.e. larger) boundaries of the Prosodic Hierarchy than those at lower boundaries. This tendency has been shown to apply in other languages, such as French, Korean and Taiwanese (e.g. Fougeron 1998; Keating et al. 2003). Research into the mapping between prosodic domains and phonetic correlates is not limited to articulatory or acoustic correspondents. Keating et al. (2003) and Auer et al. (2004), examined the optical correlates associated with prominence and boundaries in English by tracking reflective markers adhered to the faces of participants using infrared tracking. They found that the duration of chin movement was longer around boundaries (as well as corresponding acoustic duration). They also found that perceivers were “generally good” at judging when speakers were producing boundaries and when they were not. Researchers also found that prominence was cued by lip and chin movement, and that a correlation between F0 and eyebrow movement existed. Perceivers performed above chance, and general correlations revealed chin movement to be the most accurate correlate. One of the results of the research into the phonetic correlates of prosodic domains is converging evidence for phonological structure. Hayes (2000) argues that because prosodic structure is not directly observable, converging evidence from phonetic and phonological studies is crucial for justifying particular hypotheses about those structures. Hayes & Lahiri (1991) found such converging evidence for PPhrases in Bengali. They show that the domain boundaries marked by intonational contours in the language are the same ones over which segmental processes such as r-assimilation and voicing assimilation fail to occur. Jun (1993) showed similar effects in Korean. Tonally defined  2  accentual and intonational phrases are shown to be the domains of segmental processes such as lenis stop voicing and obstruent nasalisation. With these issues in mind, this thesis searches for converging evidence of the sort mentioned above to substantiate structures that have been proposed and accepted but never tested. In particular, I examine those non-exhaustively parsed structures proposed under violations of EXHAUSTIVITY (Selkirk 1995). The current model of prosodic domain organisation is governed by a family of violable constraints referred to as the Strict Layering Hypothesis. Strict Layering was originally proposed as a universal and inviolable rule requiring the strict dominance of constituents at every level of the Hierarchy. However, research in the late 1980s and early 1990s provided evidence that non-strictly parsed structures must be permitted (Itô and Mester 1992; Inkelas 1989). Constituents that violate Strict Layering were often referred to as ‘extrametrical’, and so much of the discussion of non-exhaustively parsed structures has been solely around stress patterns. Hayes (1995) gives several examples of languages, Latin and Macedonian among them, which have a final syllable that does not seem to adhere to the stress pattern of the language. In these cases, an odd-numbered final unstressed syllable is predicted to be stressed by stress rules but is not. Hayes (1995) argues that such stress patterns are best described as having an extrametrical (i.e. non-footed) final unstressed syllable. As a result of the types of ‘theoretically incompatible observations’ (Ladd 1996) above, Strict Layering was reinterpreted as a set of violable constraints. One of these constraints, EXHAUSTIVITY, assigns violations to all constituents in the Prosodic Hierarchy that are not parsed into the constituent directly above them. At the syllablelevel, this makes a structural distinction between two types of unstressed syllables: those that are parsed into feet and those that are not. However, as mentioned by de Lacy (2002, 2006), the general assumption has been that all unstressed syllables are alike. This brings into question whether there are grounds for proposing a structure that makes a distinction between two apparently non-distinct unstressed syllables. To the best of my knowledge, there has been no phonetic and very little converging phonological research into substantiating non-exhaustive parsing (the aforementioned de Lacy 2002, 2006, discussed in Section 3.4, are the exceptions).  3  Given that there has been a strong research program testing the acoustic correlates of prosodic domains and that converging evidence is methodologically desirable, the lack of acoustic substantiation of non-exhaustively parsed structures presents a large gap in our knowledge. This thesis seeks to address this lacuna. If the current model of the Prosodic Hierarchy permits unfooted unstressed syllables, all else being equal, there should be converging evidence in the form of phonological processes and phonetic correlates that would distinguish them from footed unstressed syllables. In order to test this prediction, a language that makes a phonological distinction between footed and unfooted unstressed syllables needs to be considered. St’át’imcets (Lillooet Salish) is such a language.2 St’át’imcets has two phonological processes that distinguish unstressed footed syllables from unstressed unfooted syllables: 1) vowel reduction in unstressed footed syllables but not unstressed unfooted syllables (Section 3.5.1) and 2) resonant deglottalisation, which applies in unstressed unfooted syllables but not in unstressed footed syllables (Section 3.5.2). The question then arises as to what this mapping between non-exhaustively parsed syllables and their phonetic correlates should look like. A strong version of the mapping hypothesis predicts that all languages and speakers make the same mappings using the same correlates. This version is consistent with the SPE model of the phonetics-phonology interface: phonology is discrete and language specific while phonetics is continuous, gradient and universal (Chomsky & Halle 1968; Keating 1985, 1988; Cohn 1990). More recent research supports a weaker interpretation of the phonetics-phonology mapping. A weaker interpretation of the mapping can be interpreted in two ways: 1) language-specific (but systematic) phonetic implementation of phonological targets or rules (Keating 1985, 1988, 2006; Pierrehumbert et al. 2000; Bird et al. 2008 etc.) and 2) cases within languages where the presence of contrast is crucial and consistent but with variable phonetic manifestation (e.g. Fougeron & Keating 1997; Lieberman 1960). These two effects are discussed below.  2  Language background is presented in Section 1.3.  4  In the first interpretation, Keating (1988) proposes a sort of constrained variation: phonology chooses only a subset of phonetic dimensions for formal use as rules, and languages select these phonetic rules from a basic set of preferences. Take, for example, the sonority hierarchy. Keating suggests that while the hierarchy is available to all languages, the point at which nuclei are distinguished from margins phonologically is language specific (or even register specific, as shown by de Lacy 2002, 2006). Such patterns suggest a relatively weak interpretation of the mapping hypothesis: speakers represent phonological entities by mapping from a pool of phonetic rules dictated by our physical limitations. Pierrehumbert et al. (2000) describe the mapping between abstract phonemes and phonetic realisations in terms of strong analogies rather than absolutes: The theoretical entities that can be absolutely equated across languages are the continuous dimensions of articulatory control and perceptual contrast. Languages differ in how they bundle and divide the space made available by these dimensions. (Pierrehumbert et al. 2000: 286) There is no reason to think prosodic domain mapping is any different. The second interpretation is supported by research into prominence correlates and Prosodic Strengthening. Speakers do not consistently use the same phonetic correlates to mark either prominence or boundary strength. Lieberman (1960) showed that while stressed vowels are generally stronger (more acoustically prominent) than unstressed vowels in English, speakers make use of correlate trade-offs. Research by Fougeron and Keating (1997), Fougeron (1998), and Keating et al. (2003) etc. has presented crosslinguistic evidence that Prosodic Strengthening effects are language, segment, boundary and speaker specific but that the direction of the effect is consistently hierarchical. When a distinction is made, segments at higher (larger) domain boundaries are more strongly articulated than those at lower (smaller) domain boundaries, not vice-versa. Based on the research described above, we have a continuum of possible mappings:  5  (1)  strong (highly constrained by phonology, no phonetic variation)  weak but constrained  weak (highly variable, no constraint)  Based on previous cross-linguistic studies, the prediction is that the results in this thesis will support a weak but constrained mapping. In other words, which contrasts speakers make, and how those contrasts are manifested will be constrained by phonological principles and our physiological limitations but variability will occur. Variability will arise in cases where the weighting given to a particular constraining principle is not present or is overruled by other factors. This thesis looks for contrasts in the two functions that Keating (2006), citing Pierrehumbert (1999), uses to characterise prosody: grouping and prominence marking. Specifically, the thesis examines the acoustic correlates of prominence and boundary effects of two unstressed syllables in St’át’imcets: one footed and one not (the latter hitherto referred to as an ‘extrapod’). The predicted results are that i) speakers will make acoustic distinctions between the two domains; ii) they will use a combination of correlates seen in other languages; iii) not all speakers will use the same correlates in the same way; and iv) not all contrasts will necessarily be reflected by all speakers. In short, I expect them to follow the type of mapping proposed by Pierrehumbert et al., (2000: 284): neither arbitrary nor universal. To preview the main conclusions of the thesis, results show that speakers do indeed distinguish unstressed syllables from extrapods in both prominence marking and boundary strength. This supports the predictions of the model and further strengthens the hypothesis that there is a mapping between prosodic domains and acoustic correlates. The results reveal two types of contrast: direct mapping correlated with specified phonological contrasts (i.e. phonological heads are acoustically stronger than non-heads and segments at larger boundaries are stronger than those at smaller boundaries) and indirect mapping between elements that do not stand in a specified phonological relationship (i.e. extrapods and unstressed vowels). In the absence of a specified phonological relationship, the acoustic nature of extrapods is determined by other  6  potentially conflicting demands such as domain position and/or head/boundary maximisation.  1.2  Organisation of the Thesis  The remainder of this chapter presents the phonology of St’át’imcets within the context of other Salish languages and discusses the theoretical framework and contributions of this thesis. Chapter 2 presents the results of a pilot study examining phrase-level intonation contours as background to the rest of the thesis. Chapter 3 traces the history of the Prosodic Hierarchy and the evolution of the Strict Layering Constraint and presents the relevant theoretical assumptions underlying the thesis. It introduces extrapods and presents phonological evidence from St’át’imcets that supports distinguishing them from unstressed syllables. Chapter 4 presents the general methodology used for both phonetic experiments presented in the thesis. Chapter 5 presents the results from the first experiment, which examines the acoustic correlates of stress in the language. Chapter 6 whether and how speakers use prominence correlates to differentiate between unstressed vowels and extrapods. Chapter 7 presents the results of the second experiment, which examines boundary strength effects in St’át’imcets. Chapter 8 examines whether and how boundary effects are used by speakers to distinguish unstressed vowels and extrapods. Finally, Chapter 9 concludes the thesis.  1.3  St’át’imcets in the Context of the Salish Language Family  The major references for the pan-Salish information given in the following sections are Czaykowska-Higgins & Kinkade (1998) and two Salish bibliographies: van Eijk (2008) and van Eijk & Caldecott (2008). The references for St’át’imcets are van Eijk’s (1985, published 1997) grammar, his (1987) dictionary and Davis’ (forthcoming) grammar. St’át’imcets is a Northern Interior Salish language, also called Lillooet Salish. The Salish language family is part of the Northwest Coast Sprachbund (Beck 2000). It comprises 23 languages spoken (or formerly spoken) in British Columbia, Washington State, Idaho, Montana and a part of Oregon. The language family is split into five main branches as can be seen in the table below (modified from van Eijk 2008).3 3  See also Czaykowska-Higgins & Kinkade (1998) and Kroeber (1999) for similar groupings.  7  Table 1.1 The Salish language family (* = a moribund, extinct or sleeping language) I Bella Coola Division 1) Bella Coola [Nuxalk] (Kimsquit, Bella Coola, Kwatna, Tallheo) II Central [Coast] Division 2) Comox a) *Island Comox b) Mainland Comox [Sliammon=Lhaamen] (Homalco-Klahoose-Sliammon) 3) *Pentlatch 4) Sechelt [Shashishalhem] 5) Squamish=Skwxwú7mesh 6) Halkomelem a) Upriver Halkomelem [Upper Sto:lo] (Chilliwack, Chehalis, Tait) b) *Downriver Halkomelem (Musqueam, Kwantlen, Katzie) c) Island (Nanaimo, Chemainus, Cowichan) 7) *Nooksack 8) Northern Straits (Saanich [SENĆOŦEN], *Sooke, *Songhish=Songhees [Lkungen], *Lummi, *Samish, *Semiahmoo) 9) Clallam=Klallam (Western, Eastern, Becher Bay) 10) *Lushootseed [Puget (Sound) Salish] a) Northern (Skagit, Stillaguamish, Snohomish) b) Southern (Skykomish, Snoqualmie, Duwamish, Suquamish, Muckleshoot, Puyallup, Steilacoom, Nisqually, Sahewamish) 11) *Twana (Quilcene, Skokomish, Duhlelap, Hoodsport, Vance Creek) III *Tsamosan [Olympic] Division 12) Quinault (Queets, Quinault) 13) Lower Chehalis (Humptulips, Wynoochee, Westport-Shoalwater) 14) Upper Chehalis (Satsop, Oakville Chehalis, Tenino Chehalis) 15) Cowlitz IV Oregon Division 16) *Tillamook [Hutyéyu] a) Tillamook (Nehalem, Garibaldi-Nestucca) b) Siletz V Interior Division A) Northern 17) Lillooet [St’át’imcets] (Upper [Lillooet-Fountain], Lower [Mount Currie-Douglas]) 18) Shuswap [Secwepemctsín] a) Western (Fraser River, Canim Lake, Chu Chua, Pavilion-Bonaparte, Deadman’s CreekKamloops) b) Eastern (Shuswap Lake, Kinbasket [Athalmer], Enderby) 19) Thompson [N¬e÷kepmxcin] (Lytton, Thompson Canyon, Nicola Valley, Spuzzum-Boston Bar) B) Southern 20) Colville-Okanagan a) Northern (Head of the Lakes, Vernon, Penticton, Similkameen) b) Southern (Lakes-Colville-Inchelium, San Poil-Nespelem, Southern Okanogan, Methow) 21) Columbian (Chelan, Entiat, Wenatchee [Peskwaus=Pesquous], Moses Columbia) 22) Spokane-Kalispel-Flathead a) Spokane b) Kalispel (Chewelah, Kalispel, Pend d’Oreille) c) Flathead [Salish] 23) *Coeur d’Alene [Snchitsu’umshtsn]  St’át’imcets is most closely related to N¬e÷kepmxcin (Thompson) and Secwepemctsín (Shuswap) but also shows some Coast Salish influence from Skwxwú7mesh. 8  St’át’imcets language territory is roughly between Ts’k’wáylacw (Pavilion) in the north, Lil’wat7úl (Mount Currie) in the southwest, and Xáxtsa7 (Port Douglas) in the southeast (van Eijk 1997; Davis forthcoming). The number of St’át’imcets speakers is estimated at approximately 100, with most fluent speakers over the age of 60 (www.uslces.org/uslces.html). St’át’imcets has two major dialects: Northern (also called Upper or Fountain), spoken around Lillooet and Southern (also called Lower or Mt. Currie-Douglas), spoken around Mount Currie and the communities north of Harrison Lake. The Southern dialect may be further divided to into two dialects: that spoken around Mount Currie and that spoken around the community of Sqátin (Skookumchuck). In this thesis, I refer to these three dialects as Northern, Southern and Lower Southern. The dialects are mutually intelligible, with the main differences being i) phonetic—vowel quality, particularly /a/) (van Eijk 1997; Namdaran 2006, Henry Davis, p.c.); ii) phonological—retraction before /z, z’/ (van Eijk 1997; Namdaran 2006); iii) lexical/morphological—lexical items and contraction of the determiner and nominaliser (van Eijk 1997; Matthewson 2005; Davis forthcoming), and iv) syntactic—word order (van Eijk 1997; Matthewson 2005; Davis forthcoming) and the use of conjunctive inflection in nominalised elements (H. Davis p.c.). Syntactic and morphological characteristics of St’át’imcets are not the focus of this thesis, so they are discussed only briefly here.4 Like all other Salish languages, St’át’imcets is fundamentally predicate initial. While both VOS and VSO word order are possible in both dialects, VOS word order is unmarked in the Northern dialect, while VSO is more common in the Southern dialect(s) (Davis 1999; Matthewson 2005). Nouns, adjectives, adverbs and verbs can occur as predicates. Like other Salish languages, subject and object morphology is realised as affixes or clitics on the predicate, and possessive morphology is realised on the possessed head.5 Both person morphology and possessive morphology are used to provide some of the tokens for the experiments presented in Chapters 5-8. Like all Salish languages, morphemes in St’át’imcets have a particular linear  4 5  For more in-depth pan-Salish syntactic discussion, see Kroeber (1999). Possessed subjects in nominalised clauses are clitics that may be realized on an auxiliary (Davis 2000).  9  order. The basic Salish morpheme order is as follows: PS/S(-)ASP-LOC-RED-√ROOT-RED-PA-LS-TR/ITR/CTL-O-S/PS-ASP6 (Czaykowska-Higgins & Kinkade 1998: 23) The specific order for St’át’imcets is discussed in Section 1.4.7 below. St’át’imcets follows typical Salish patterns of morphological structure and strategies. Like other Salish languages, it has full words (containing lexical content and one primary stress) as well as clitics and affixes (Czaykowska-Higgins & Kinkade 1998: 22). Prefixes are limited and rarely stressed. Suffixes can be subclassified into grammatical and lexical (Czaykowska-Higgins & Kinkade 1998). Grammatical suffixes include transitivising suffixes and person suffixes, both of which play a role in this thesis. Lexical suffixes have specific lexical meanings and can be subclassified as strong (accented), weak (unaccented) or variable in terms of stress assignment. (Davis forthcoming; CzaykowskaHiggins 1993, 2004b; Czaykowska-Higgins & Kinkade 1998; Watt 1999; Watt et al 2000; Bar-el & Watt 2001; Blake 2000; Leonard 2006, 2007; Shaw et al. 2002; among others) Given the root-like stress behaviour of lexical suffixes, they are not included in the tokens used in either phonetic experiment. St’át’imcets also makes use of reduplication and other non-concatenative strategies, as other Salish languages do (see van Eijk (1997) and Czaykowska-Higgins & Kinkade (1998) for more information).  1.4  General Phonology  This section presents a discussion of phonological topics that are relevant to the thesis. 1.4.1 Root structure The predominant root type in St’át’imcets is CVC, comprising 65% of the lexicon (van Eijk 1997:32). CVCC roots comprise 18%; CCVC comprise 5%; and disyllabic roots comprise 5%. Residual types make up 7%.7  6  PS/S = possessive, transitive/intransitive subject clitics, ASP=aspect, LOC= locative, RED=reduplicant, ROOT= root, PA = primary affixes (which vary between languages), TR/ITR/CTL = transitive, intransitive, applicative and control markers, O=object, S=subject 7 See Shaw (2004a) for further discussion of the phonotactics of consonant clusters in St’át’imcets.  10  1.4.2 Consonant and vowel inventory In terms of phoneme inventory, St’át’imcets has the typically large Salish consonant inventory (44) and small vowel inventory (8). The consonant inventory is shown in Table 1.2: Table 1.2 Consonant inventory (modified from van Eijk (1997))8 Lab  DentalLateral  DentalPalatal  Velar  Dent Lat Dent Pal plain  p  ejective  €  Stop  Affricate  t  Ž  m glottalised   Glottal  ÷  k  k∑  q  q∑  ˚  ˚∑  œ  œ∑  sš  x  x∑  ⋲  ⋲w  y Á  • °  w „  ª ·  ª∑ ·∑  ç    Fricative plain  Labiouvular  cæ  plain ejective  Resonant  LabioUvular velar  n ˜  l,l Ò,Ò  z ¸  h  There has been much debate as to the nature of St’át’imcets retraction and retracted phonemes (van Eijk 1997; Remnant 1990; Bessell 1992, 1997, 1998; Shahin 1995, 2002; Namdaran 2006;). Shahin (2002) argues that /ª, ·, ª∑,·∑/ are better classified as gutturals, while /s, c, l, Ò/ as well as /z, ¸/ in the Lower dialect and the uvular and labio-uvular obstruents are better classed as emphatics. Namdaran (2006), on the other hand, gives ultrasound evidence that the articulation of the St’át’imcets uvular resonant “is equivalent to that of the uvular stop q, both involving tongue dorsum backing and raising towards the upper pharynx and tongue root retraction towards the lower pharynx” (2006: 143).9 The vowel inventory of St’át’imcets has also been a source of debate, particularly the status of schwa and retracted vowels in the language. 8  The phonemes presented above and in Table 1.2 below are given in NAPA. In his phonetic alphabet, Van Eijk uses /c/ and /s/ rather than /æ/ and /š/, and represents retraction with an under dot. The underline is used here for ease of transcription and is in keeping with the orthography. It should also be noted that, while he uses the symbol conventionally used to represent pharyngeals (ª), van Eijk consider them to be uvulars. This classification is confirmed by Namdaran (2006). IPA and orthographic conversions are given in Appendix A. 9 The reader is referred to Namdaran (2006) for a summary of acoustic and articulatory research on vowel and consonant retraction.  11  Table 1.3 Vowel inventory (modified from van Eijk (1997)) Front High i, i Mid ¢, ¢ Low a, a  Back u, u  Van Eijk (1997:3) describes the pronunciations of /a, i, u, ¢/ as [´, e, o, ¢] and /a, i, u, ¢/ as [a, ´/ė, ø, ʌ]. Namdaran (2006) reports vowel articulations (based on Remnant 1990, Bessell 1992, 1998; Shahin 2002) as ([Upper dialect]/[Lower dialect]): /i/ = [i]/[e], /u/ = [u]/[o], /a/ = [å]/[´] and epenthetic /¢/. In terms of the phonology of vowels, Shahin (2002:173) describes the underlying vowels as high front /I/, low /Æ/, and high back /U/ but later represents underlying vowels as: /I/ being [+HIGH], /U/ as [+LAB, +HIGH] and /Æ/ as [+LOW] (2002: 244). Van Eijk (1997) includes both schwa and the retracted vowels in the underlying inventory, while Matthewson (1994) argues that schwa is not underlying. Shahin (2002) and Shahin & Blake (2004) further distinguish between epenthetic and excrescent schwa on phonetic grounds.10 Shahin (2002) proposes that the surface allophones are distributed in the following way: Table 1.4 Vowel inventory and allophones (Shahin & Blake 2004: 315) (rd= round, phar’d=pharyngealised, uv’d = uvularised)  10  Shahin and Blake (2004:ftn 11) acknowledge methodology-dependent vowel quality effects. Vowels elicited in isolation showed distinct F1/F2 positions from those elicited in phrases. Beyond mentioning the distinction, these effects were not discussed. For further discussion of the status and/or behaviour of schwa in Salish languages, see Shaw (1996, 2001, 2002a, b, 2004a,b,c, 2005), Blake (1992, 2000, 2001, 2004) on Sliammon, Czaykowska-Higgins (1993, 1998, 2004a,b on Nxa’amxcín), Bianco (1996, 1998) on Cowichan-Halkomelem, Urbanczyk (2000), Shaw et al. (1999) on hen’q’emin’em’, as well as Kinkade (1998).  12  For Shahin (2002) and Shahin and Blake (2004), the pharyngealised vowels are allophones that occur preceding a retracted coronal consonant while the uvularised ones occur preceding a post-velar segment. Namdaran (2006) found retracted vowels occurred both preceding and following retracted segments (though the articulation varied dependent upon consonant context).11 In order to sidestep issues of coarticulation and harmony, only non-retracted full vowels in non-uvular and non-pharyngealised environments are included as tokens in the two phonetic experiments conducted. Hence, the vowels that I will consider are: /i/, /u/ and /a/. In terms of feature specifications, which become relevant in Chapters 7 and 8, I follow Shahin’s original description of underlying vowels as [+high], [+front] for /i/, [+high], [+back] for /u/ and [+low] for /a/. 1.4.3 Glottalised resonants St’át’imcets is one of only 52 languages in Maddieson’s (1984) study of 314 languages with glottalised (laryngealised) resonants.12 Bird & Caldecott (2004a,b, 2005) and Bird et al. (2008) describe the phonetic characteristics of glottalised resonants in St’át’imcets. Caldecott (1999, 2003, 2004) presents an overview of phonological issues surrounding glottalised resonants in St’át’imcets. St’át’imcets has both underlying and derived glottalised resonants. Underlying glottalised resonants in certain suffixes undergo alternation with plain resonants depending on parsing. This alternation will be discussed in detail in Section 3.5.2. Glottalised resonants can also be derived morphologically through two main sources: diminutive reduplication and glottalising lexical suffixes (which bears on Section 3.5.1).13 11  The reader is referred to Shahin (2002) and Namdaran (2006) for further discussion of vowel retraction effects and harmony. 12 Glottalised resonants are well represented in the languages of the Northwest Coast. See Caldecott (1999) and references therein for phonological treatments across Salish, (and specifically phonetic characteristics in SENĆOŦEN). Glottalised resonants in St’át’imcets are produced with varying phonation, from a full glottal stop to creaky voicing. ‘Laryngealised’ is a term that more accurately describes the latter phonation type. The term ‘glottalisation’ will be used in this thesis to refer to phonemically glottalised (either underlying or derived) resonants for both articulation types. For further discussion of speaker variation associated with glottalised resonant phonation, see Bird (2008). 13 Resonants following a root with an infixed glottal stop inchoative morpheme also surface as glottalised (H. Davis p.c.)  13  In terms of the resonant glottalisation accompanying diminutive reduplication, Van Eijk (1997: 61) describes it as fairly unpredictable. Davis (forthcoming: Chapters 37, 50) describes resonant glottalisation associated with diminutive reduplication as affecting the final voiced consonant in the stem. Following both van Eijk (1997) and Davis (forthcoming), Caldecott (2003) discusses the following examples, taken from van Eijk (1997, 1987). The following data illustrate the glottalisation patterns associated diminutive reduplication:14 (2)  NAPA  orth.  a. χzum  English  NAPA  orth.  English  xzum ‘big’  χzÂz¢Â  xzézem’  ‘a little bigger’  b. twit  twit  twi„t  twiw’t  ‘youth’  c. ß-qla„  sqlaw’ ‘beaver’  ß-qlÂl¢„  sqlélew’  ‘little beaver’  d. pála÷  pála7 ‘one’  pápla÷  pápla7  ‘one person’  e. ka¬áß  kalhás ‘three’  ka¬Â¬ß  kalhélhs  ‘three animals’  ‘hunter’  In general, the rightmost resonant surfaces as glottalised, as seen in examples (2a,b). 15 Example (2c) shows that if the rightmost resonant is already glottalised in the underlying form, it will absorb the glottalisation associated with the diminutive, so no other resonant will be glottalised (i.e.*ß-qlÂl’¢„). If the rightmost resonant in the word is a glottal stop, as in example (2d), no preceding resonants surface as glottalised (i.e. *pápl’a÷). Finally, if the word does not contain a resonant, as in example (2e) there is no glottalisation. Caldecott (2003) proposes that the diminutive is not only expressed by reduplication of the consonant preceding the stressed vowel but also a floating glottalisation feature [creak], which aligns with the rightmost resonant in the word. The second way in which glottalised resonants are morphologically derived is through lexical suffixes. There is a group of lexical suffixes (somatic lexical suffixes)  14  Unless otherwise specified, examples in this thesis are given in the North American Phonetic Alphabet (NAPA) and the orthography. 15 There are some exceptions: natx∑ ’tomorrow’  s-n’á-n’atx∑ ‘morning’ ÷álsәm÷á÷әl’sәm *÷á÷әlsәm’ ‘sick’ ‘a little bit sick’  14  that causes glottalisation on resonants. It is also following this group of somatic suffixes that the vowel reduction discussed in Section 3.5.1 occurs. Davis (forthcoming, Chapt. 46) lists the following as some of the glottalising lexical suffixes: • • • •  (n-...)-q∑ -æ (n-...)-q -x¢n  ‘head, animal, top, cover’ ‘mouth, speech, food, opening’ ‘bottom part, buttocks, legs’ ‘leg/foot, shoe, crotch’  The following examples discussed in Caldecott (2003) show the general pattern of glottalisation: (3)  NAPA a. √χzum b. √k¢m c. √pzan  orth. English xzum ‘big’ s-kem ‘area’ n-pzan ‘meet’  NAPA χzuÂ-q∑ ßk¢Â-æ n-pzá˜-q-a˜  d. √wic  wix  wíc-q∑-am’  e. √k¢l f. √pála÷  kel ‘area’ pála7 ‘one’  ‘comb’  n-k¢l’-q pál÷-aq∑  orth. English xzum’qw ‘big animal’ skem’ts ‘door’ nzpán’qan’ ‘catch up’ (Davis forthcoming) wíxqwam’ ‘to comb one’s hair’ nkel’q ‘to back up’ pál7aqw ‘one ball’ (van Eijk 1997)  Examples (3a), (3b), and (3e) show that the resonant immediately preceding the lexical suffix surfaces as glottalised. Example (3d) shows that the rightmost resonant only, in this case the final one, is glottalised (i.e. *w’ix-qw-am’). In (3c), we see that both the preceding and following resonants surface as glottalised. In (3f), we see that the glottal stop patterns as a resonant, as in the formation of the diminutive. Caldecott (2003) proposes that the lexical suffixes shown above have a floating [creak] feature associated with them. Following Davis (forthcoming), the feature is analysed as aligning with the nearest resonant(s). As can be seen in the data above, resonant glottalisation is independent of vowel quality and occurs following both full vowels and schwa. As such, it is also independent of the vowel alternation discussed in Chapter 3.5.1, which is shown to be dependent upon parsing. While glottalised resonants in St’át’imcets are fascinating, they are not the focus of this thesis and so are not discussed further.  15  1.4.4 Stress: Phonology Much phonological research on stress in Salish languages has been conducted, including Czaykowska-Higgins (1993a, 1995, 1998, 2004a, b) on Nxa’amxcín; Idsardi (1991) and Kuipers (1993) on Secwepemctsín; Carlson (1989) and Black (1996) on Spokane; Thompson & Thompson (1992) on N¬e÷kepmxcin; Bar-el and Watt (1998, 2001), Watt (1999, 2001) and Dyck (2004) on Skwxwu7mesh; Bianco (1996, 1998) on HalkomelemCowichan; Urbanczyk (1999) on Lushootseed; Shaw et al. (1999) and Shaw (2001), on hen’q’emin’em’; Blake (1992, 2000, 2001) on Sliammon; Montler (1986) and Leonard (2006, 2007) on SENĆOŦEN. Idsardi (1991) and Revithiadou (1999) both discuss stress in Interior Salish languages in the context of broader theoretical issues. Czaykowska-Higgins and Kinkade (1998) divide the Salish family into four broad stress systems: i) morphologically governed (Interior excluding St’át’imcets and Coeur d’Alene), which has main stress rightmost “given the lexically-specified stress properties of the component morphemes” (Czaykowska-Higgins & Kinkade, 1998: 15), ii) penultimate, subject to weight and morphology (St’át’imcets and Skwxwú7mesh), iii) penultimate, but subject to morphological factors only (SENĆOŦEN), and iv) fixed (everything else). 16 17 Regarding stress assignment in Salish languages, Czaykowska-Higgins and Kinkade (1998) propose the following four tendencies: i) languages tend to exhibit three degrees of stress (primary, secondary, unstressed), but secondary stress does not surface (or is not transcribed) consistently; ii) prefixes are generally not stressed, except when they are reduplicative morphemes; iii) schwas resist stress, surfacing as stressed only when there are no full vowels in the word; and iv) with the exception of Upper Chehalis and Cowlitz, unstressed full vowels are reduced to schwa or are completely deleted. Kroeber (1999) refines tendency iv) somewhat by claiming that unstressed full vowels reduce to schwa and in Interior languages are often deleted entirely.  16  Leonard (2006, 2007) argues that, rather than being morphologically complex, stress in SENĆO ŦEN is generally predictable assuming trochaic binary feet aligned to the right edge of a lexical stem. Stress in this language is also sensitive to a weight distinction between schwa and full vowels. 17 Brown & Thompson (2005, 2006a) examine pitch perturbations in Upriver Halkomelem and conclude that the language shows tonal characteristics.  16  The status of St’át’imcets with respect to the above tendencies is not clear. Acoustic research presented in Chapter 7 will show that unstressed vowels in St’át’imcets do not reduce to schwa in terms of vowel quality. Acoustic results in Chapters 5 and 6 also call the position of St’át’imcets with respect to the first tendency into question. Section 1.4.7 presents evidence from Davis (forthcoming) that shows that tendency iii) needs further refinement in St’át’imcets as well. 1.4.4.1 Stress in St’át’imcets The stress system of St’át’imcets is documented by van Eijk (1997) and analysed by Roberts (1993) and Roberts and Shaw (1994). According to van Eijk (1997:14), “[i]n both roots and root-suffix combinations, assignment of the stress follows two basic tendencies: (1) in words with at least one [full] vowel...the stress falls on the first [full vowel], whether or not that [full vowel] is preceded by [schwa]; in words with only [schwa], the stress falls on the first [schwa]; (2) from the base established by the first tendency, the stress moves two vowels at a time, as long as it does not fall on the last vowel.” Roberts (1993) presents a derivational analysis wherein he proposes the following rules: • Final moras are extrametrical • Parse the word (beginning with the underlying grid mark) from left to right into moraic trochees, i.e., feet containing two light syllables (the first being more prominent than the second) or a single heavy syllable • End Rule Right (at word level) • Only syllables with full vowels are stress-bearing. Neither schwa nor a moraic consonant may head a bimoraic foot, i.e., they must occupy the weak, second element of the moraic trochee. Roberts and Shaw (1994) frame the above generalisations in prosodic terminology and propose an Optimality Theory (henceforth OT) account: • • • • •  Feet are trochaic and assigned from left to right Prosodic words are right-headed The prosodic head does not fall on the final mora of a word Schwa is non-nuclear moraic and dispreffered as head Coda consonants are non-nuclear moraic  17  The data below illustrate the basic pattern of stress assignment. More complex data, illustrating the behaviour of schwa and the coda consonant clusters is discussed briefly thereafter.18 (4)  a. (æun) b. (æún-kax∑) c. (æún-tu)mu d. (æùn-tu)(mú-kax∑)  ‘to order’ ‘you ordered him’ ‘order us’ ‘you ordered us’  (5)  a. (æú-u˜) b. (æú-u˜)-kax∑ c. (æù-u˜)-(túmu) d. (æù-u˜)-(túmu)-kax∑ e. (æù-u˜)-(tùmu)-(káÒap)  ‘to point at’ ‘you pointed at him’ ‘point at us!’ ‘you pointed at us’ ‘you folks pointed at us’  The examples in (4a, b, c) and (5a, b) show primary stress on the initial vowel, which demonstrates that the stress pattern is trochaic and that feet are assigned from left to right. Roberts and Shaw (1994) account for these patterns with FOOT FORM, ALIGNMENT and PARSE constraints. In (4d, 5c, 5e) primary stress occurs on the rightmost foot of the word, indicating that the PWord is right-headed. This is also accounted for by an ALIGNMENT constraint. Secondary stress, while not marked in the orthography, occurs on the heads of pre-tonic feet. Examples (4c) and (5b, 5d) also show that stress may not fall on the final vowel or syllable of a word *((æu-u˜)-káx∑). Roberts & Shaw (1994) account for this through the interaction of a NON-FINALITY constraint prohibiting stress on the final mora and crucially ranking FOOT-BIN over PARSE, which crucially outranks ALIGN-FT-L, to rule out incomplete feet. As a result, following Roberts and Shaw (1994), I treat syllables like the final ones in (4c) and (5b, 5d) as unfooted. If these syllables were footed, we would expect them to receive primary stress, according to the rightmost generalisation. However, they are not described as receiving stress, primary or secondary, and thus are  18  These data are also found as example 3 in Roberts (1993) and originate in van Eijk (1981, 1985/1997). Roberts and Shaw (1994) re-elicited some forms from speakers (Roberts, 1993: 297; Roberts & Shaw, 1994: 2). They are presented in NAPA transcription here to be consistent with other data.  18  treated as unparsed at the foot level. The status of these syllables is discussed extensively in Chapter 3. As mentioned above, the basic stress pattern is complicated by the behaviour of schwa and coda consonants. No roots with schwa or lexical suffixes are included in the experimental tokens in this thesis, but the behaviour of schwa and consonants are discussed briefly here. If we first consider schwa, it receives main stress only when there are no full vowels in the word.19 Compare the examples in (6) with those in (7): (6)  a. (m¢¡æ-¢n) b. (l¢¡ª∑-¢n)  ‘Write it!' ‘Hide it!’  (7)  a. m¢æ -(xál) b. (l¢™ª∑-¢n)-(kán)  ‘to write (intr.)’ ‘I hid it’ (Roberts & Shaw 1994)  The examples in (6a, 6b) show that in words with two schwas, stress appears on the first vowel. Examples (7a, 7b) show that in a word with a full vowel, stress will surface on the full vowel.20 In addition to being morphologically dependent, the behaviour of schwa is also domain dependent, as discussed in Section 1.4.7. The examples in (8) show another interesting aspect of schwa behaviour: footing is assigned from the first full vowel rather than the first schwa. (8b) also shows that, while schwa is dispreferred as head, it still fulfils the binary requirement for feet. This is accounted for through a distinction between nuclear and non-nuclear moras and a Peak Prominence constraint. (8)  a. b.  ˚∑¢ (zúš¢m)¬kax∑ ˚∑¢ (zùš¢m)(káx∑ha)  ‘You worked’ ‘Did you work?’ (Roberts & Shaw 1994)  Second, roots that end in a consonant cluster behave as complete feet. Van Eijk (1997) describes the pattern in the following way:  19  Henry Davis (p.c.) points out that there are three morphologically conditioned exceptions: 1) the circumstantial morpheme, 2) the passive suffix, and 3) diminutive reduplication. However, given that full vowels are the focus of this experiment, they will not be discussed further. 20 In longer words, schwa does receive PWord-level stress. For example, [(pùp¢˜)(k¬tu÷)] ‘pùpen' kélh tu7’ ‘S/he might come across it’ (H. Davis p.c.). The full vowel does not receive stress in this case because it is outside the PStem domain. St’át’imcets domains are discussed further in Section 1.4.7  19  Certain consonant clusters count as vowels with regard to the distribution of stress. These clusters comprise either root-final clusters...final clusters on so-called lexical suffixes...or word-final clusters resulting from a lexical suffix of the shape C(C) added to a word. (van Eijk 1997:15) In other words, a two consonant cluster in a root or lexical suffix counts for stress.21 The examples in (9a, b) surface with primary stress on the second syllable and secondary stress on the first: (9)  a.  √(ptàkw¬)-(mín=aš)  ‘S/he told a legend about him/her’  b.  √(àt)-(mín=aš)  ‘S/he came back for him/her’ (Caldecott 2006b)  Robert and Shaw (1994) consider coda consonants to be non-nuclear moras, which accounts for their asymmetrical stress pattern. As moraic, they are available as non-heads of feet but as non-nuclear, they are not available as heads. As a result, the initial syllables above are binary under a moraic analysis. The interaction of the constraints proposed by Roberts and Shaw (1994) is not the focus of this thesis, so the analyses in sections 3.5.1 and 3.5.2 assume the above constraints and refer to them simply as a single super constraint: STRESS. 1.4.5 Stress: Phonetics Phonetic research examining the acoustic correlates of stress in Salish languages is extremely limited. Studies focusing on stress are limited to Watt et al., (2000) on Skwxwú7mesh and Benner (2006a, b) on SENĆOŦEN. These studies support the crosslinguistic generalisation that stressed vowels have greater duration, intensity and higher F0 than unstressed vowels (e.g. Laver 1994). The only acoustic investigation of St’át’imcets stress is Caldecott (2006a), a pilot study. This study was limited by not taking intonation and phrasal position into account. Nevertheless, it found that speakers use F0, duration and intensity, though not 21  A consonant cluster of four outside the root also counts as moraic for stress (Davis forthcoming: Chapter 46). Speakers vary to some extent in how many consonants a cluster requires in order to count as moraic. This phenomenon is left for further research.  20  consistently, to mark stress. Chapter 5 in this thesis addresses the limitations of the previous study and adds to the small pool of empirical data available on stress in Salish languages. 1.4.6 Intonation Research into the larger intonation contours in Salish languages is limited to the domain research discussed in Section 1.4.7.1 (i.e. Beck 1999; Beck & Bennett 2007 on Lushootseed; Barthmeier 2004 on Okanagan; Koch 2008 on N¬e÷kepmxcin), Benner (2006) on SENĆOŦEN and Jacobs (2007) on Skwxwu7mesh. Benner (2006) showed that the general intonation contour of declaratives in SENĆOŦEN features a left-aligned nuclear accent and pitch declination, as well as a low final boundary tone. Interrogatives do not show rising intonation to signal yes/no questions. Jacobs (2007) also shows no final pitch rise associated with yes/no questions. Both declaratives and interrogatives had an overall fall in pitch, while interrogatives started lower and ended higher than declaratives. A pilot study on St’át’imcets intonation is reported in Chapter 2. 1.4.7 Prosodic domains: Phonology There has been much research into the phonology of prosodic domains in Salish languages, including: Bar-el and Watt (1998, 2001) on Skwxwwú7mesh; Bates & Carlson (1989) on Spokane; Beck (1999) and Beck and Bennett (2007) on Lushootseed; Bianco (1996) on Halkomelem-Cowichan; Blake (1998, 2000, 2004) on Sliammon; Czaykowska-Higgins(1993, 1995, 1998, 2002, 2004a, b) on Nxa’amxcín; Shaw (2001, 2002a,b 2004a,b,c, 2005) on hen’q’imin’em’ and St’át’imcets; Shaw et al. (1999) and Shaw et al. (2002) on hen’q’imin’em’; Urbanczyk (1996, 1999) on Lushootseed; Bar-el and Watt (2001), Watt (1999, 2001) and Watt et al. (2000) on Skwxwú7mesh. Research into prosodic domains in Salish languages has focused on the Prosodic Word (henceforth PWord) and sub-PWord constituents, and draws heavily on the work by Czaykowska-Higgins mentioned above. Czaykowska-Higgins (1995, 1998) examines the mapping between morphological domains (M) and phonological domains (P). While both MWords and PWords are composed of three domains—Root, Stem, and Word, the 21  domains are not isomorphic. Each domain contains a specified series of morphemes and is the domain of certain phonological rules. Czaykowska-Higgins (1998) proposes the schema for PWords and Mwords in Nxa’amxcín: (10)  [MW ASP [MS LOC RED [MR ROOT]MR RED PA LS]MS LS TR O S]MW ITR ASP [PW ASP LOC RED[PS [PR ROOT]PR RED PA LS TR O S]PS]PW ITR ASP  Czaykowska-Higgins (2004a) examines stress, phonotactics, segmentism, retraction, and cluster simplification as converging evidence for the proposed phonological structures. The PRoot is found to be unique in terms of segments and shape. The PStem is the domain of retraction and stress assignment. The PWord contains the grammatical prefixes. Following on from Czaykowska-Higgins’ work, Davis (forthcoming Chapt. 46) defines the Prosodic Word domain as follows: (11) [PWord[PStem[PRootroot + lexical suffixes]other suffixes + pronominal enclitics]other enclitics] The prosodic domains above are defined by the different stress effects that are active within them.22 Prefixes are excluded from the PWord domain since they are not stressed.23 The weight of consonant clusters seen in section 1.4.4.1 above applies within the Prosodic Root only, and schwa is not stressed within the PStem domain unless there are no full vowels, or the schwa is the head of the final foot in longer words (H. Davis p.c.). Consider the following examples: (12)  a. b.  [[[Nepnep]PRoot wít]PStem]PWord “They are staying in the middle of the road” [[[Népnep]PRoot]PStem iz’]PWord “Those ones are staying on the road” Davis (forthcoming, Chapt. 46)  In the first example, the stem and word final vowel is stressed, not the initial schwa. In the second example, the clitic /iz’/ does not form part of the Prosodic stem and so is not stressed. Instead, the initial schwa surfaces as stressed. 22  Domains can also be defined in terms of retraction and morpho-syntactic characteristics. See Davis (forthcoming) for more discussion. 23 According to (van Eijk 1997:17) There are five prefixes which do take stress: lá- ‘in’, 7á- ‘towards’, lhlá- ‘from’, kná- ‘around’ and the formative 7í-). These prefixes form part of the PWord domain and follow basic stress assignment patterns. H. Davis (p.c.) notes that these prefixes only occur in deictic roots and seem fully integrated with them.  22  As mentioned above, in longer examples, schwa does have PWord-level stress: (13)  [[[(pùp]PRoot en') (kélh]Pstem tu7)]PWord  'S/he might come across it'  In the example above, the schwa in the second foot surfaces with primary PWord-level stress because the clitic /tu7/ again does not form part of the PStem. In terms of other domain diagnostics, evidence is less clear. Shaw (2004b) presents evidence that word-initial and -final edges sustain more robust phonotactics than word-internal domains, permitting greater cluster complexity than elsewhere.24 Larger domain distinctions, such as the distinction between PWord and Phrase in the language, are complicated, and will be discussed in Section 1.4.8.1 below. 1.4.8 Prosodic domains: Phonetics As with stress, acoustic research into prosodic domains in Salish languages is sparse. Research is limited to Beck (1999) and Beck and Bennett (2007) on Lushootseed; Barthmaier (2004) on Okanagan, Bennett (2006) on SENĆOŦEN and a recent thesis by Koch (2008a) on N¬e÷kepmxcin.25 Beck (1999) and Beck and Bennett (2007), working with recorded narratives, define Lushootseed Prosodic Groupings in the following way: Table 1.5 Beck (1999) and Beck and Bennett’s (2007) definitions of domains in Lushootseed PWord • Domain of phonological affixation and stress assignment • Words have only primary stress, which occurs on the leftmost full vowel • Non-words (functional morphemes) cannot bear stress PPhrase • Made up minimally of a PWord, maximally of a PWord + affixes and 2 clitics • Set off by pauses (50-100 ms, shorter in fast speech) • Lack of assimilation at boundaries • Headed by PWord (with amplitude peak) IPhrase  • Marked by peak on first full vowel, irrespective of stress • Set off by pauses, F0 declination, F0 reset  24  Prominence and nuclear weight also add to the complexity permitted in codas. See Shaw (2004b) for further discussion. 25 See also Brown & Thompson (2006a,b) for preliminary exploration of determiner clisis in Upriver Halkomelem.  23  Beck (1999) and Beck and Bennett (2007) also argue for the addition of a Phonological Paragraph below the Utterance level. They argue that this domain, reflecting subjecttopic shift, is marked by declination in F0 peaks and F0 reset. Another study that examines the correlates of prosodic domains is Koch (2008a). This thesis focuses on the interaction between intonation and focus in N¬e÷kepmxcin. For Koch, PWords are lexical items that are the domain of stress. Stress is morphologically governed and falls rightmost within morphologically determined sub-word domains. PPhrases are the domain of accent, and have the following acoustic correlates: (14)  a. b. c. d. e.  partial F0 declination reset H* pitch accent L- end boundary tone lack of phonological interactions across boundaries pauses in slower speech Koch (2008a:189)  Morpho-syntactically, Koch (2008a) defines a PPhrase as consisting of a PWord and its affixes and clitics. Like Beck (1999) and Beck and Bennett (2007), Koch parses predicates and nominal arguments into separate PPhrases. Person clitics, as with other clitics, are parsed into the same PPhrase as the verb. IPhrases have a right-aligned nuclear accent, which is correlated with reduced F0 and amplitude declination as well as lengthening of the syllable bearing the accent. Along similar lines to the research above, Benner (2006a, b) analyse prosodic domains in SENĆOŦEN. These studies are based on impressionistic rather than statistical observations of acoustic results. Benner notes that prosodic groupings (which she equates with Beck and Bennett’s PWords) are often marked by pauses, changes in pitch contour and/or voice quality. PPhrases (consisting of PWord + clitics) are leftheaded and distinguished by a lack of assimilation across boundaries. Unlike the research above, Barthmeier (2004) analyses the prosodic domains of Okanagan in terms of Intonational Units. Following Chafe (1994) he defines Intonational Units as prosodic units based on cognition, which represent “the linguistic output of the speaker’s focus of attention” (Barthmeier 2004: 36). The acoustic correlates to the Intonational Unit in Okanagan are a convergence of pauses, changes in pitch (i.e. contour  24  and reset) and duration of segments, and change in phonation type. Koch (2008a) equates them to PPhrases and IPhrases in the traditional Prosodic Hierarchy. As we saw in Sections 1.4.4 and 1.4.7, PWords in Salish are traditionally defined as the domain of stress (van Eijk 1997; Davis forthcoming; Roberts 1993; Roberts & Shaw 1994; Czaykowska-Higgins 1993, 1995, 1998) and include a lexical root plus affixes and clitics. In contrast, Koch (2008a), Beck (1999) and Beck and Bennett (2007) would consider the equivalent domain to be the PPhrase. The PWord/PPhrase distinction is clearer in many other Salish languages because unstressed vowels reduce to schwa or delete and stress does not occur on grammatical affixes or clitics. This follows the standard assumption that functional elements like clitics do not have inherent stress and do not bear phrasal accent (Selkirk 1995b on English). The difficulties in distinguishing PWords from PPhrases in St’át’imcets are discussed below. 1.4.8.1 St’át’imcets: PWord or PPhrase? No acoustic research into St’át’imcets higher prosodic domains or their morpho-syntactic mapping has been conducted. This is partly due to the lack of clear diagnostics to distinguish PWord and PPhrase. In St’át’imcets, unlike N¬e÷kepmxcin and Okanagan, stress routinely falls on clitics and affixes in longer words, functional and grammatical alike. As a result, the distinction between PWord and PPhrase is difficult to tease apart. The task is made more difficult because roots overwhelmingly occur in the language with suffixes or clitics, either the existential determiner or person morphology, even if it is a null pronoun (Davis forthcoming). To define prosodic boundaries in St’át’imcets, one criterion to which we might appeal is retraction. As discussed above, retraction is used by Czaykowska-Higgins as a diagnostic for PWord in Nxa’amxcín. St’át’imcets has two general types of retraction: progressive phonological retraction induced by roots and regressive uvular/pharyngeal retraction harmony (Shahin 2002). Both types have been considered to be limited to PWord boundaries in St’át’imcets (Davis forthcoming, Shahin 2002). However, Davis (forthcoming, Chapt. 51) notes that phonological retraction is highly variable between speakers, and is limited to root+in/transitive suffixes. Given that this domain is not one  25  of those defined in Section 1.4.7, we begin to see the complexity of the morphophonology in this language. Uvular/pharyngeal harmony is also not a possible diagnostic for PWord. Recordings collected by the author indicate that retraction occurs not just within PWords but also between PWords and determiners. For example, the /u/ of the determiner /ku/ surfaces as [kø] before uvular-initial nouns:26 (15)  a.[kø] as in “pù¬un¬káx∑ ha ku qu7?” (Did you boil some water? LT: 15.02.07) b.[ku] as in “ยn¬káx∑ ha ku kapúh?” (Did you buy any coats? LT: 01.02.07) The above result suggests that either the domain of harmony is in fact the  PPhrase rather than the PWord or that determiners in St’át’imcets are parsed to the PWord rather than the PPhrase (Selkirk (1995b) proposes that in English, determiners are parsed directly to the PPhrase). A complete discussion of the domain of retraction is outside the purview of this thesis; however, the idea of using retraction as a diagnostic for domain certainly warrants more in-depth research. The above finding rules out retraction as a simple diagnostic for PWord status. Perhaps prominence correlates could be used. It is possible that what has been traditionally considered primary stress in longer PWords is in fact PPhrase accent (Beckman & Edwards, 1994). This would be in line with Koch’s (2008a), Beck (1999) and Beck and Bennett’s (2007) classification of root + clitics and affixes as PPhrase rather than PWord. Chapters 5 and 6 investigate whether different prominence correlates can be used to distinguish word-level stress from phrase-level accent, following Sluijter and van Heuven (1996a,b). The next section lays out the assumptions regarding the Prosodic Hierarchy that are made in this thesis.  1.5  Assumptions Made regarding the Prosodic Hierarchy  This section discusses some points of on-going debate on the nature of the Prosodic Hierarchy and the assumptions made in the thesis regarding: i) prosodic domains, ii) 26  The retraction was judged aurally by both the author and Henry Davis (p.c.). Ultrasound research might shed light on whether the retraction of the vowel outside of the PWord differs in any way from that within the PWord, which could potentially be used as a diagnostic for PWord domain.  26  morpho-syntactic mapping, iii) headedness/prominence, iv) boundaries, and vi) inventory. 1.5.1 Prosodic domains Following research by Nespor and Vogel (1986) and Selkirk (1984, 1986, 1995a,b) on English; Pierrehumbert & Beckman (1988) on Japanese; Hayes & Lahiri (1991) on Bengali; and Jun (1993) on Korean, I assume that prosodically-defined constituents are the domains of phonological processes, and that any domain in which a phonological process occurs is a constituent that must be definable and referenceable by constraints. In addition to serving as the locus for phonological properties that subsequently receive a phonetic interpretation, prosodic domains can be marked as acoustically distinct from one another in terms of how prominent heads are cued (Beckman & Edwards 1990, 1994), intonation patterns (e.g. Pierrehumbert & Beckman 1988, Selkirk 1995a) and/or articulation of segments at boundaries (e.g. Fougeron & Keating 1997, Byrd & Saltzman 1998, Keating et al. 2003, Cho 2002, Tabain 2003a,b). Constituent grouping is governed by the Strict Layer Hypothesis, which I assume is a set of violable constraints following Selkirk (1981,1984, 1995b). 1.5.2 Morpho-syntactic mapping Following research by Nespor and Vogel (1986) and Selkirk (1984, 1986, 1995a,b) as well as Czaykowska-Higgins (1993, 1998) specifically on Salish languages, I assume that phonological domains are separate from but mapped to morphological domains. As such, phonological processes do not apply directly to morpho-syntactic constituents. I also assume that it is phonological constituents that are distinguished phonetically. It is possible that isomorphy between morpho-syntactic domains and phonological domains results in phonetic correlates that align with both; however, it is the phonological domains with which the phonetic realisations are associated. Prosodic constituents may be defined in morpho-syntactic terms (e.g. Selkirk 1984, 1986; Nespor & Vogel 1986) or intonational terms (e.g. Beckman & Pierrehumbert 1986, Jun 1993). In this thesis, I will assume the morpho-syntactic definitions laid out in Davis (forthcoming), because the theory must provide a way of mapping morpho-  27  syntactic categories to the prosodic constituents that are fed by them. The nature of this mapping in St’át’imcets is not the focus of the current research but is briefly discussed in Section 1.4.7. The category label of the domain PWord versus PPhrase is not crucial for either experiment conducted in this thesis. In the prominence experiment reported in Chapters 5 and 6, unstressed vowels and extrapods are both compared in final position within the same domain. As a result, it is predicted that any domain-final effects, whether PWord or PPhrase will affect both syllable-types equally. The possibility that what has traditionally been described as primary stress may in fact be PPhrase accent is discussed in Sections 6.6 and 6.7. In the boundary strength experiment reported in Chapters 7 and 8, the status of the domains considered should not affect the direction of the Prosodic Strengthening effect. Tokens are compared in Intonation Phrase (henceforth IP)-final and non-final position. It is predicted that segments at the IP boundary will be stronger than those that are not IP-final, regardless of whether they are PWords or PPhrases. 1.5.3 Headedness/Prominence Prominence has both phonological and phonetic definitions that overlap but are not coextensive. In phonology, relative prominence is assigned under specific phonological relationships, namely motherhood or sisterhood (e.g. Liberman and Prince 1977).27 Under such analyses, heads are more prominent than non-heads within a certain domain, and headship is associated with certain phonological properties, such as positional markedness effects and stress.28 The phonetic realisation of prosodic prominence or headship is sometimes also referred to as prominence: according Laver (1994), one syllable is more prominent than another (in English) if it has a combination of higher pitch, longer duration and greater intensity or articulatory excursion.  27  This is not meant to be exhaustive. ‘Prominence’ in phonology can refer to other relatively salient relationships, such as Beckman’s (1998) discussion of onsets as prosodically prominent or privileged positions, or the prominence of stem-initial position in Bantu (D. Pulleyblank p.c.). 28 I assume here, following Ladd (1996:5) that stress ‘reflects a set of prominence relations between the elements of the utterance. The stress pattern is manifested in a variety of phonetic cues, which are admittedly not well understood.”  28  In this thesis, I will use ‘prominence’ to refer to phonological headship and the associated phonetic correlates in terms of ‘strength’. Given the mapping in Section 1.1, the assumption is that the prominence correlates associated with phonological heads will be ‘stronger’ or ‘acoustically stronger’ than non-heads: they will have a combination of higher F0 and/or greater duration and/or intensity.29 Recent research has argued that prominences at different levels of the hierarchy are distinguished by different acoustic correlates in English (Beckman & Edwards 1990, 1994, Stevens 1994, Shattuck-Hufnagel et al. 1994). Beckman and Edwards (1990, 1994) found that while foot-level prominence in English was mainly correlated with increased duration, word-level prominence was correlated with increased pitch (see also Fry 1955, 1958 on English and Sluijter and van Heuven 1995a, b on English and Dutch). In this thesis, I will assume that: 1) all constituents must have heads (which are assigned by head-alignment constraints);30 2) the phonological heads of constituents are correlated with phonetic strength; 3) where the head is located and what acoustic correlates represent those heads are determined on a language-specific basis; and 4) heads at different levels of the hierarchy may be correlated with different acoustic characteristics. OT constraints may refer to heads or non-heads of prosodic domains as defined and proposed by de Lacy (2002, 2006). 1.5.4 Boundaries Early work on prosodic organisation in English, such as Prince (1983) and Selkirk (1984) excluded notions of prosodic constituency from the formal representation to focus on relative prominence. However, more recent work on a number of languages (e.g. Nespor & Vogel 1986, Hayes 1989; Beckman & Pierrehumbert 1986, Pierrehumbert & Beckman 1988, Selkirk 1986, Hayes and Lahiri 1991, Jun 1993) has returned to Selkirk (1978)’s older proposal that prosodic domains have well-defined boundaries.  29  It is possible that a better term for the phonetic realisation of prominence would be ‘saliency’. Ladd (1996) defines pitch accent as having a prominence-cuing effect, and notes that it often involves a local maximum OR minimum. In cases where the prominent element is cued by a drop in pitch, the head would not be considered ‘stronger’ under the definition used here, but would be more salient. 30 Note Crowhurst (1992), which provides evidence that morphological feet need not specify heads.  29  Following the above research, I assume that prosodic constituents have welldefined edges or boundaries and that these boundaries have acoustic reflexes in terms of intonation patterns (Pierrehumbert & Beckman 1988, Selkirk 1995a) and segment articulation (Fougeron & Keating 1997, Byrd & Saltzman 1998, Keating et al. 2003, Cho 2002, Tabain 2003a, b). These cues are determined on a language-specific basis. OT constraints may refer to the span or boundaries of domains (Selkirk 1980, Nespor & Vogel 1986). 1.5.5 Inventory There has been much debate in the literature about which constituents comprise the inventory of the Prosodic Hierarchy. As inventory issues are not the focus of this thesis, I only briefly discuss them here. The debate and the positions of its participants are welllaid out in Shattuck-Hufnagel & Turk (1996) and the reader is referred there for more indepth discussion. Disagreement is particularly focused on the middle of the Hierarchy. For example, Nespor and Vogel (1986) argue for the inclusion of Clitic Group directly above the PWord, while Beckman and Pierrehumbert (1986) and Pierrehumbert and Beckman (1988) split the PPhrase into Accentual and Intermediate Phrases. These divisions are not isomorphic. I will assume that UG gives us mechanics to form domains and rank them in a hierarchical manner, while the mapping of those domains is implemented on a language-specific basis. Some supporting evidence in the form of Prosodic Strengthening research is presented in Chapters 7 and 8. Whether or not the constituents in Salish behave like their cross-linguistic counterparts will be left for future research. In the interest of simplicity, I follow Selkirk’s (1978) model: moras-syllables-feet-PWord-PPhrase-IPhrase-Utterance. It has been argued that lexical domains should not be included in the Prosodic Hierarchy and that metrical and phonological domains should be considered separate entities (Inkelas 1989). I will assume that both sub-PWord and super-PWord constituents are included in the same Hierarchy and are thus governed by the same principles. Furthermore, I assume that the generalisations governing non-strict parsing will apply in the same manner at all levels of the Hierarchy. In other words, violations of the Strict Layer Hypothesis should be applicable at all levels of the hierarchy.  30  1.6  Methodological, Empirical and Theoretical Contributions  This thesis has three major goals: i) to develop protocols to elicit phonetic data for statistical analysis in a way that accommodates fieldwork with older First Nations speakers and is convertible to language teaching materials; ii) to contribute to the documentation and understanding of St’át’imcets, an endangered language; and iii) to contribute to our understanding of the Prosodic Hierarchy model and the phoneticsphonology mapping. The methodological contributions are discussed in detail in Chapter 4 while the empirical and theoretical contributions are outlined below. 1.6.1 Empirical contributions The major empirical contribution of this thesis is that it constitutes the first comprehensive documentation of prosody in St’át’imcets, including acoustic analyses of stress, intonation and boundary strength. While the phonological pattern of stress in the language has been well researched, this thesis comprises the first in-depth account of the acoustic correlates of prominence. It shows that speakers make a four-way distinction in prominence between stressed vowels in head feet, stressed vowels in non-head feet, unstressed vowels and extrapods. Acoustic results indicate that speakers clearly mark secondary stress on the heads of pre-tonic feet, as reported in Roberts and Shaw (1994). Different acoustic correlates of prominence are correlated with different domains. Foot-level prominence is realised mainly by increased duration as well as a small increase in F0 and intensity. Primary and secondary stress are distinguished by a small increase in F0 and duration. PPhrase/PWord-final feet are distinguished from other feet by an increase in the magnitude of contrast: the stressed vowel of the head foot has much higher pitch and greater duration than its non-head, and the magnitude of this contrast is greater than that between stressed vowels and unstressed vowels in non-head feet. The contrast between unstressed vowels and extrapods is similar in magnitude to head versus non-head relations but in conflicting directions In this thesis, I show that, like N¬e÷kepmxcin, St’át’imcets phrases have a H* nuclear accent associated with the final stressed vowel in the phrase. PPhrases (possibly  31  PWords) have a L% boundary tone realised on the non-stressed vowel(s) following the final stressed vowel. Domain-final lengthening applies to both final unstressed and extrapod vowels. As in SENĆOŦEN and Skwxwú7mesh, yes/no questions are not associated with a final rise. They are produced at an up-shifted register. In terms of vowel quality, this thesis shows that unstressed vowels do not reduce to schwa. Word-internal unstressed vowels in open syllables are more reduced than those in closed syllables. The thesis also shows that St’át’imcets results are similar to previously reported cross-linguistic Prosodic Strengthening results. Finally, the results in this thesis also reveal a great deal of inter- and intra-speaker variation. This result is relevant for researchers of elderly and/or limited populations, as well as future phonetic studies in which the same speakers might take part. 1.6.2 Theoretical contributions The most important theoretical result of this thesis is that it provides converging evidence for the non-exhaustively parsed structures predicted to exist by the Prosodic Hierarchy model. It introduces the term ‘extrapod’ and confirms the hitherto untested prediction of the model that extrapods are phonetically distinct when distinguished phonologically. The combined results of acoustic prominence correlates and boundary strength argue for the weak, but constrained interpretation of the phonetics-phonology interface, as laid out in Section 1.1. The results were both speaker and segment-specific. Two types of contrast emerged in this thesis: i) directionally constrained and ii) directionally unconstrained. Directionally constrained contrast was correlated with a phonologically-specified relationship between elements. In other words, a direct mapping between the phonetics and the phonology resulted in cases such as head versus non-head comparisons, and the hierarchically related domains of the Prosodic Hierarchy. The lack of a phonologically specified/specifiable relationship between unstressed vowels and extrapods was correlated with an indirect mapping: contrast, but not direction of contrast was crucial. The contrast between unstressed vowels and extrapods was determined by how individual speakers weighted other sometimes conflicting demands. Foot non-head reduction, head constituent maximisation, boundary enhancement and utterance position emerged as those conflicting demands.  32  Results support refining the definition of Prosodic Strengthening to allow a finely cumulative interpretation. Under this interpretation, the articulation and associated acoustic characteristics of segments are affected by how many boundaries they are adjacent to, not just how high or large the outermost boundary is. Southern dialect speakers showed consistently stronger unstressed vowels and extrapods. This result is predicted under an interpretation in which all boundaries, not just the highest/largest counts. Ultimately, the results of this thesis support the basic hypothesis upon which the thesis is based: distinct prosodic domains have distinct acoustic characteristics.  33  Chapter 2 St'át'imcets Intonation Contours  2.1  Introduction  This chapter represents the first study of intonation in St’át’imcets.31 It is a pilot study of the intonation contours of declaratives and yes/no questions. While intonation per se is not the focus of the thesis, an understanding of accent and pitch contours is crucial when investigating acoustic prominence correlates. The results are included here as background for the phonetic and phonological evidence presented in the thesis. This pilot experiment examined the pitch contours and F0 differences between the initial, penultimate and final vowels of yes/no questions and their declarative answers in St’át’imcets. It was designed to address two questions: i) Where is nuclear accent? and ii) Are yes/no questions associated with higher pitch? In terms of the first question, Koch (2008a) found nuclear accent in N¬e÷kepmxcin to be associated with the rightmost stressed vowel. This claim is based on a lower than expected declination in F0. Pierrehumbert’s (1979) and Sorensen and Cooper’s (1979) results (as reported in Pierrehumbert 1979) show a mean rate of declination over an 8-9 syllable phrase between 6.7 and 11.1 Hz/syllable. Koch found that the actual rate of declination in N¬e÷kepmxcin between the leftmost and rightmost stressed syllable to be considerably less, with the two speakers showing mean declination rates of 3.84 Hz/syllable and 3.15 Hz/syllable. Koch concludes that the decreased rate of declination relative to the expected declination is indicative of a H* nuclear accent associated with the rightmost stressed vowel. The second question was designed to test the Universality of Intonational Meaning Hypothesis, which proposes a non-arbitrary relation between meaning and sound in the case of question intonation (Bolinger 1978; Gussenhoven 2002; Chen 2005). 31  This section was presented as Caldecott (2007) at the 42nd International Conference on Salish and Neighbouring Languages (ICSNL). See also recent research by Oberg (2007) and Caldecott & Davis (2008) examining the acoustic correlates of WH phrases in the language.  34  Under this hypothesis, a biological predisposition associating high pitch with uncertainty or submissiveness has been universally grammaticalised to associate interrogativity with raised pitch.32 Acoustic studies of other Salish languages, namely SENĆOŦEN (Benner 2006a, b) and Skwxwú7mesh (Jacobs 2007), suggest that, contrary to the Universality of Intonational Meaning Hypothesis, no final pitch rise is associated with yes/no questions in those languages. Auditory descriptions of St’át’imcets suggest that yes/no questions are not associated with a rise either. Based on results from N¬e÷kepmxcin (Koch 2008a), which showed a rightmost nuclear accent and from Benner (2006a, b) and Jacobs (2007) that yes/no questions are not associated with a final pitch rise in SENĆOŦEN and Skwxwú7mesh, the preliminary hypothesis is that St’át’imcets will also have rightmost nuclear accent and no final pitch rise associated with yes/no questions. The results indicate that like N¬e÷kepmxcin, St’át’imcets has a H* right-headed nuclear accent and L% boundary tone. Like Skwxwú7mesh and SENĆOŦEN, yes/no questions do not have final pitch rise. Instead, yes/no questions are produced with an up-shifted F0 register.  2.2  Methodology  2.2.1 Speakers The three speakers who participated in this experiment, AP, CA and LT, also participated in the experiments reported in Chapters 5-8. For more information about the speakers, please see Chapter 4.  2.2.2 Stimuli and recording The task was a semi-scripted conversation: one speaker asked scripted questions presented in English, which he or she translated. The other speaker answered based on a photo they saw on a MS PowerPoint (PPT) slide. AP and CA questioned each other, while I played the role of second speaker for LT. 32  The biological motivation for the hypothesis comes from Ohala’s (1983, 1984, 1995) work on the Frequency Code. Please refer to those works for more discussion.  35  Three scenarios were presented to speakers: i) going to a farm, ii) going for a hike, and iii) going clothes shopping. Stimuli were questions in St’át’imcets, such as “Did s/he see any cows?” and “Did s/he buy three hats?”. Both questions and responses were recorded. Table 2.1 below presents the total number of tokens recorded for each speaker: Table 2.1 Speakers and numbers of tokens Speaker Total recorded Excluded  Answers  Questions  AP  120  48  33  44  LT  160  34  70  56  CA  123  30  53  40  Tokens were excluded for misspeaks, hesitations, bi-clausal answers, wh-questions and instrumental errors. The number of excluded tokens was high due to the unscripted nature of the responses and natural hesitations and misspeaks in natural speech. Speakers were recorded using a Marantz 670 PMB recorder and Audiotechnica AT-831b cardioid lavaliere microphones. They were then uploaded onto an Apple iBook G4 and analysed using Praat 4.5.12 (Boersma & Weenink 2007). Based on a visual inspection of a series of phrases from all speakers, the initial, penultimate and final vowels emerged as loci of pitch excursions. As a result, maximum pitch of the initial, penultimate and final vowel was extracted using Praat. Relative pitch ratios between the initial vowel and the penultimate vowel, and the penultimate vowel and final vowel were compared. As this was a pilot study, no statistical analyses were performed.  2.3  Results  The following sections present the results of the experiment. Section 2.3.1 gives general results while the comparisons of means and pitch change ratios are given in Section 2.3.2. 2.3.1 General results Figure 2.1 below shows a pitch tracing of AP’s answer to a yes/no question. From this figure, we see the highest pitch of the phrase is on the final stressed /a/ in the phrase.  36  Figure 2.1 AP’s answer, “Yes, he saw one dog (lit. “Yes, one was the dog he saw”) aa11_iy_pepla7_ta_sqax7a_atsxenas  (In this and subsequent figures, examples are given in the orthography, and ´ indicates primary stress.) 3.37065212 3.57238089 300 250 200 150 100 75 Iy,  pépla7 ta  sqáx7a  áts’xnas  á  a  0  4.479 Time (s)  We can also see a pitch drop on the final vowel. I propose that the pitch peak on the rightmost stressed vowel is the H* nuclear accent and that the pitch drop on the final vowel is the L% boundary tone. Note also the pitch contour on the initial word /iy/ ‘yes’ which has both a peak and a fall and is followed by a reset in F0 on the next word. Figure 2.2 demonstrates the pitch contour when the final word in a declarative phrase is monosyllabic (the final vowel in the phrase is stressed). aa23_iy_atsxenas_ta_pepla7_nkyap  4.07433205 Figure 2.2 AP’s answer, “Yes, he saw one coyote.”  0.239765004 300 250 200 150 100 75  Iy  áts’xnas  ta  pépl7a  nk’yáp  y 0.2398  á 4.074  Time (s)  37  The figure above shows that when the phrase-final vowel is stressed, the H* and the L% boundary tone can occur on the same vowel. This type of tone compression was not systematic, as we can see in Figure 2.4 below.33 The next two figures show AP’s pitch tracings for yes/no questions. Figure 2.3 shows a yes/no question that ends in the disyllabic word ‘tsíken’ (chicken). Again, the penultimate vowel, the final stressed one, has the highest pitch in the phrase. aq4_atsxenas_ha_ku_tsiken Figure 2.3 AP’s question, “Did he see any chickens?” 0.128072466 1.77922463 300 250 200 150 100 75 àts’xnás  ha  ku  tsíken  í  e  0.1281  1.779 Time (s)  The next Figure shows a yes/no question that ends in a monosyllabic word. aq45_kwam_ha_i_tsitsela_qmut  Figure 2.4 AP’s question “Did she get new hats?” 300 250  200 150 100 75 kwám ha  i  tsítsela  0.2623  qmút 2.86  Time (s)  This figure shows that the H* accent is still associated with the final stressed vowel in the word, but there is no following unstressed vowel to host the L% boundary tone, and the 33  No systematic pattern of compression versus truncation was identified. The possibility that these strategies are utterance-type dependent is left for future consideration.  38  tone is truncated. The above four figures represent the general intonation contours in St’át’imcets declaratives and yes/no questions: A H* phrasal accent occurs on the final stressed vowel of the phrase and a L% boundary tone occurs on the unstressed vowel following the phrasal accent. The above figures from AP are not completely representative of the three speakers who participated. She had the highest proportion of absolute maxima on the final stressed syllable: 66%. CA, her brother and also a Northern dialect speaker, had 57% when glottalisation was taken into account. One of the problems encountered in the measurements of pitch was that numerals in the language often contain glottal stops or glottalised resonants. Irregular pitch pulses associated with this type of glottal activity can be mismeasured as high pitch. CA, in particular, had a strong correlation between glottal segments and misread pitch. Once this effect was controlled for, CA showed a similar, though weaker, effect to AP. The third speaker who participated in the pilot study was LT, the Southern dialect speaker. She showed a pattern in which 40% of her tokens had highest pitch on the initial vowel, and only 23% on the final stressed vowel. Pitch range appeared to be gender-specific. The female speakers had a range of between 100 and 300 Hz while CA had a more compressed top line, with a range of approx. 100-200 Hz34. 2.3.2 Comparison of means and pitch ratios The following figures show the means for vowels in declaratives and yes/no questions. Only tokens that ended in disyllabic words were considered. The ‘initial vowel’ marks the first vowel in the phrase while the following two vowels come from the final word in the phrase. The ‘nuclear vowel’ is the final stressed vowel in the phrase, which is the penultimate vowel in all cases. The ‘final vowel’ is the unstressed vowel of the final word. The first two figures show that AP and CA have an upward-shifted register for yes/no questions but no final rise.  34  With only one male speaker, it is difficult to say whether this pattern is attributable to gender rather than individual differences. Anecdotal evidence suggests that male speakers do generally have a more limited range than female speakers (H. Davis, p.c.) but further research is required.  39  Figure 2.5 AP’s mean F0 values  Figure 2.6 CA’s mean F0 values  For LT, while both declaratives and yes/no questions begin with similar mean F0, the nuclear accent and final vowel show an upward shift. Figure 2.7 LT’s mean F0 values  40  Since intonation is a relative phenomenon, pitch ratios were also measured. Figure 2.8 below shows the change between the initial vowel and the nuclear accent as a ratio. The first data point represents the initial vowel with a ratio value of 1. The second data point represents the mean ratio of the nuclear vowel over the initial vowel. The ‘perceptible’ line represents the pitch ratio required to signal a perceptually salient change in pitch in this frequency range ('t Hart 1981). For example, LT’s nuclear accent vowel is 0.83 times as high as her initial vowel in her answers. The perceptibility line shows that a pitch ratio of .84 indicates a perceptually salient pitch change (following the values calculated in t’ Hart 1981). Figure 2.8 Mean pitch ratios between initial and nuclear vowel  From the above figure, we can see that AP and LT have significantly greater pitch changes between the initial vowel and the nuclear vowel in questions than answers. CA, on the other hand, shows no significant differences. The following figure shows that all speakers have a fall between the nuclear vowel (penultimate, stressed vowel) and the final vowel. AP and CA have greater falls to the final vowel in questions than answers. Most of these phrase-final falls are perceptible by the pitch ratio outlined above, and the degree of fall is greater than that of a normal stressed-unstressed vowel.  41  Figure 2.9 Mean pitch ratios between nuclear and final vowels  2.4  Discussion  The results of this pilot study show that St’át’imcets is similar to other Salish languages in terms of having a rightmost nuclear accent and no final pitch rise associated with yes/no questions. Results show that a H* phrasal nuclear accent occurs in both declaratives and yes/no questions on the rightmost stressed vowel for Northern dialect speakers. The Southern dialect speaker, LT, shows a H* nuclear accent on the final stressed vowel in yes/no questions only. In the case of declaratives, the initial vowel had the highest pitch. With only one Southern dialect speaker, it is not possible to determine whether LT’s lack of rightmost H* accent in declaratives is a dialectal or idiolectal phenomenon. In either case, an analysis which aligns the H* accent with the leftmost stressed syllable in the phrase could account for her results. Both types of phrases for both dialects are marked by a L% boundary tone. The relative results confirm the Universality of Intonational Meaning, in that yes/no questions are associated with raised pitch. For the Northern dialect speakers, yes/no questions are produced with a similar pitch contour but at a higher pitch register. For LT, yes/no questions had increased pitch associated with the final stressed vowel. The increase in pitch is greater between initial vowels and the nuclear accented vowel, and the fall from that vowel to the final vowel is greater in yes/no questions than declaratives for the two female speakers.  42  2.5  Conclusion  This chapter presented the results of Caldecott (2007), a pilot study of phrasal intonation in St’át’imcets. While more research into the intonation of prosodic domains is required, these results contribute to the growing body of documentation of prosody in Salish languages. This research has established that the final stressed and unstressed vowels in a phrase are crucial to F0 contours. As a result, this informs our experimental design and proposed parsing model. More research into the patterns of phrase-final monosyllabic words and whether tone truncation or compression is a more commonly used strategy needs to be conducted. However, this research does show that monosyllabic roots, in spite of being degenerate feet, can carry phrasal accent. The next chapter presents the phonological history of the Prosodic Hierarchy and the evolution of extrapods, along with phonological evidence that unstressed syllables and extrapods are phonologically distinct in the language. Based on this evidence, the phonetic predictions of the proposed extrapod model are laid out.  43  Chapter 3 Extrapods: History and Phonological Evidence  3.1  Introduction  This chapter presents the phonological motivation for the experiments reported in the following chapters. The experiments focus on the Prosodic Hierarchy model and the Strict Layer Hypothesis. The evolution of the model and the Strict Layer Hypothesis are discussed in Sections 3.2 and 3.3 respectively. Additional goals of this chapter are to: i) provide evidence that in St’át’imcets, there are two types of unstressed syllables; ii) argue that the only structure that supports the phonological facts is the extrapod model; iii) create an extrapod paradigm, including the evidence from St’át’imcets; and iv) lay out the phonetic predictions of the extrapod model that are tested in the following chapters.  3.2  Prosodic Hierarchy Background  Taking into account the general assumptions presented in the introduction, this section lays out the history of the Prosodic Hierarchy. Ladd (1996) reports Halliday (1960, 1967) as the first explicit organisation of prosodic structures into “a hierarchy of welldefined domain types” (1996: 237). In Halliday’s proposal, utterances consist of tone groups, which consist of feet, feet of syllables and syllables of phonemes. As early as the Halliday model, there was an assumption of Strict Layering that resulted in a parsing mismatch. In Halliday’s model, syllables must be parsed into feet, and feet must be headed by an initial strong syllable. When an unstressed syllable occurs initially, no head is present, and the syllable is hypothesised to be preceded by a silent, foot-initial ictus, which acts as head. Halliday’s model is illustrated in the example below, where [ˆ] represents the silent ictus, [/ ] represents foot boundaries and [/ /] represents tone group boundaries. 44  (1)  Utterance / \ Tone group Tone group / \ / | \ Foot Foot Foot Foot Foot / | \ / \ / \ / | \ | ˆSyl Syl Syl Syl ˆ Syl SylSyl Syl Syl / / ˆ When I / got there/ / ˆshe’d/ already / left// (modified from Ladd, 1996:237)  As we can see, words play no role in Halliday’s analysis, and the distinction between stressed (or salient) syllables (/got/, /al-/) and non-salient (/when/, she’d/) seems more reminiscent of an accented versus unaccented or lexical versus functional distinction. Ladd notes that the ictus at the beginning of the second tone group represents a pause, the existence of which is supported by converging phonetic evidence. In the case of the first ictus, however, no such converging evidence exists. As Ladd points out, this case would be considered a case of extrametricality or degenerate foot under a more current theory of prosodic organisation (Ladd, 1996: 298, ftn 14). Our current model of prosodic structure began to take shape in the late 1970’s and early 1980’s. Prosodic phonology during this time focused on showing that phonological rules applied not directly to syntactic constituents but rather to discrete prosodic constituents which formed part of a separate component of grammar subject to its own internal rules and organisation (Selkirk 1978, 1981, 1984; Nespor & Vogel 1986). In early versions of this model, phonological constituents were organised hierarchically like syntactic constituents but differed in that they were not recursive35. Concurrently, metrical theory sought to account for the relational, non-adjacent phonological properties of stress (Liberman 1975; Liberman & Prince 1977; Halle & Vergnaud 1987). These two theoretical streams converged with the development of the Prosodic Hierarchy. One of the first works to propose the integration of metrical structure and hierarchical prominence was Liberman and Prince (1977). For Liberman (1975) and Liberman and Prince (1977),‘perceived’ stress actually “reflects the combined influence of a constituent-structure pattern and its grid alignment” (1977: 249). In this model, stress is a segmental, rhythmical grouping feature while relative prominence is a function of constituent structure, defined in terms of sisterhood. 35  Some theories, such as Neeleman & van de Koot (2006), still argue that this is case.  45  Segments’ [+/-stress] features serve as the terminal elements of constituents. It is from these terminal elements that strong or weak nodes and branches in the metrical treestructure are projected. An element’s prominence is defined syntagmatically only, as being stronger (or weaker) than its sister. The term Designated Terminal Element (DTE) is used for the most prominent terminal element in a constituent. In any constituent in which a sisterhood relationship exists, the DTE of the strong sister is marked as metrically stronger on the concomitant grid than the weak sister. Metrical grid strength is represented by integers at different levels: (2)  4 123 execute | | | s ww \/ / s / \/ M36 (Liberman & Prince: 324.116a) Liberman and Prince (1977) are also the first to introduce the notion of ‘feet’,  ‘extrametricality’, and ‘stray’ syllables but did not develop these further. Their treatment of such syllables is discussed in Section 3.4. After Liberman and Prince (1977), models of metrical phonology underwent several alterations including tree-only proposals (Kiparsky 1979; Giegerich 1985), gridonly proposals (Prince 1983; Selkirk 1984), the introduction of metrical feet and bracketed grid notation (Halle & Vergnaud 1987), foot typology (Hayes 1981, 1995), and the incorporation of trees and grids (Hammond 1988; Hayes 1983). These models merged with prosodic phonology research (Selkirk 1980, 1984; Nespor & Vogel, 1986) and intonation theory (Pierrehumbert 1980; Beckman & Pierrehumbert 1986; Beckman & Edwards 1990, 1994) into our current model of the Prosodic Hierarchy.  36  Liberman and Prince (1977: 268 give the stress assignment for execute as:  execute +- + (where + is [+stress] and – is [-stress]. The final syllable does not project a strong node because it does not branch, and strong nodes must be branching. As a result, it is possible to have [+stress] syllables that are not metrically strong.  46  One of the first ‘modern’ models for English was proposed in Selkirk (1978): (3)  Utterance (Utt) intonation phrase (IP) phonological phrase (PhP) prosodic word (Wd) foot (Ft) syllable (Syl) (Selkirk 1995b ex. 3) As discussed in the Introduction, the domains included in the Prosodic Hierarchy  have been a topic of much debate. The organisation of the Prosodic Hierarchy is defined by the Strict Layer Hypothesis, discussed in the next section.  3.3  The Evolution of the Strict Layer Hypothesis  The Strict Layer Hypothesis was first formulated by (Selkirk 1981; 1984) as: (4)  ...a category of level i in the hierarchy immediately dominates a (sequence of) categories of level i-1 (Selkirk 1981). (Assuming syllable to be level 1, the others will be levels 2, ..., n.) (Selkirk 1984: 26)  As the quotation above indicates, the notion that constituents must be parsed directly into the level above them was part of the first formalisation of Strict Layering.37 Hayes (1984; 1989) reformulated the definition to include the idea that constituents grouped into the immediately dominating layer must be of the same type: (5)  The categories of the Prosodic Hierarchy may be ranked in a sequence C1, C2,... Cn such that a. all segmental material is directly dominated by the category Cn, and b. for all categories Ci, i ≠n, Ci directly dominates all and only constituents of the category Ci + 1 (Hayes 1984: 204)  In a precursor to OT-style constraints, Nespor and Vogel (1986) present Strict Layering as four principles: (6)  37  a. A given nonterminal unit of the prosodic hierarchy, Xp, is composed of one or more units of the immediately lower category, Xp-1.  See Ladd (1996: 237-241) for a discussion of the development of a fixed depth hierarchy.  47  b. A unit of a given level of the hierarchy is exhaustively contained in the superordinate unit of which it is part. c. The hierarchical structures of prosodic phonology are n-ary branching. d. The relative prominence relation defined for sister nodes is such that one node is assigned the value strong (s) and all the other nodes are assigned the value weak (w). (Nespor & Vogel 1986: 7) Principle (6b) is often referred to as a requirement on ‘strict’ or ‘direct’ dominance or succession. The term “Exhaustivity” is formally introduced by Halle and Vergnaud (1987) in their Exhaustivity Condition: (7)  The rules of constituent boundary construction ... apply exhaustively subject to the Recoverability Condition...38 (Halle & Vergnaud, 1987: 15)  The goal of this condition was to force foot construction at the edges of domains before higher structures could be constructed. In each of the above cases, the principles or rules governing prosodic organisation were considered to be inviolable and universal. The first challenge to the inviolability of Strict Layering came in the form of Ladd (1986), which showed that the Hierarchy must allow recursive IP levels. In an experiment using conjoined clauses, his results showed acoustic evidence of nested IPs. A second challenge came from Inkelas (1989), which showed that the Prosodic Hierarchy must allow both recursivity and violations of Strict Dominance. Itô & Mester (1992) showed further evidence that Strict Layering, in particular direct dominance (Exhaustivity), could not hold in Japanese. Japanese has a bimoraic foot, bimoraic minimal word constraint and a monosyllabic word ban. These constraints conflict in the case of loanword clippings, such as in their example (4) /daiyamondo/ ‘diamond’. The monosyllabic clipping /dai/ is ruled out due to the ban on monosyllabic words. The trimoraic /daiya/, consisting of a foot + light syllable word, results. Itô and Mester suggest three possible structures to account for this ‘1½ Foot problem’: Weak 38  The Recoverability Condition is defined as follows: Given the direction of government of the constituent heads in the grammar, the location of the metrical constituent boundaries must be unambiguously recoverable from the location of the heads, and conversely, the location of the heads must be recoverable from that of the boundaries (Halle & Vergnaud 1987: 10).  48  Layering (extrapods are parsed directly to the word), Superfoot (an additional level in the prosodic hierarchy, consisting of a foot plus a light syllable), or degenerate foot (the light syllable forms its own foot). They find no independent motivation for a Superfoot constituent.39 They argue that words with unfooted syllables pattern with mono-foot words rather than difoot words, so those unfooted syllables cannot be parsed as degenerate feet. As a result, they propose Weak Layering, a combination of three licensing requirements, as the best possible solution. The formalisation of Weak Layering is given in (8): (8)  a. Mora Confinement: mora is licensed only by syllable b. Proper Headedness: Every (nonterminal) prosodic category of level i must have a head, that is, it must immediately dominate a category of level i-1 c. Maximal Parsing: Prosodic Structure is maximally parsed, within the limits imposed by other (universal and language particular) constraints on prosodic form. (Ito & Mester 1992: 11-13)  Based on the evidence in the works cited above, and couched in OT terms, Selkirk (1995b) reformulated the Strict Layer Hypothesis as the following violable constraints: (9)  LAYEREDNESS: No Ci dominates a Cj j > 1 , e.g. “No σ dominates a Ft.” • HEADEDNESS: Any Ci must dominate a Ci-1 (except if Ci = σ), e.g. “A PWd must dominate a Ft.” • EXHAUSTIVITY: No Ci immediately dominates a constituent Cj, j < i-1, e.g. “No PWd immediately dominates a σ.” • NONRECURSIVITY: No Ci dominates Cj, j = i, e.g. “No Ft dominates a Ft.” (Selkirk 1995b: 443)  LAYEREDNESS is presumed to be part of GEN while the status of HEADEDNESS is unclear (McCarthy, 2008). The violability of EXHAUSTIVITY and NONRECURSIVITY shows that these are two constraints in CON and can be more lowly ranked. 39  See Hewitt (1992) for a derivational analysis of Superfoot constituents as bounded PWords, and prosodic evidence that supports them as combinatory prosodic constituents.  49  These constraints are actually families of constraints that apply at each level of the hierarchy. McCarthy (2008), for example, specifies the levels to which constraints such as EXHAUSTIVITY apply: EXHAUSTIVITY (foot) is a constraint that assigns a violation for any syllable that is parsed directly to the PWord. The use of the term “non-exhaustive parsing” in this thesis refers to prosodic elements which violate the EXHAUSTIVITY constraint above. It does not refer to Hayes’ (1981, 1995) Nonexhaustivity restriction on extrametricality (1995: 58), nor does it refer to Inkelas’ (1989) definition of exhaustive parsing as “[e]very utterance is exhaustively parsed by each domain in D” (1989: 7). This thesis examines violations of EXHAUSTIVITY, specifically the syllables that are unparsed at the foot level and how they are distinguished from parsed syllables. The next section introduces the term ‘extrapod’ to refer to unfooted syllable residue and discusses previous phonological treatments of such syllables.  3.4  Extrapods  EXHAUSTIVITY violations generally have been discussed in terms of ‘extrametricality’. According to Inkelas (1989: 47), the original purpose of extrametricality was “...to identify a small set of elements which are systematically ignored by the stress rules of various languages.” As a result, such elements have been previously identified and discussed only in terms of stress (Itô & Mester 1992 and Pulleyblank 1986, notwithstanding). As this thesis demonstrates, previous labels for and treatments of the elements that violate EXHAUSTIVITY are insufficient to accurately account for their phonological characteristics. In order to clearly and accurately define the specific elements addressed here, I introduce the term ‘extrapod’ to refer to non-exhaustively parsed syllables at the foot level. This avoids confusion with previous terminology and focuses on the fact that their behaviour falls out from their phonological structure. Non-exhaustively parsed syllables have previously been referred to as ‘stray’ (e.g. Liberman and Prince 1977), ‘extrametrical’ (e.g. Hayes 1981, 1995), ‘extraprosodic’ (Kiparsky 1985) or ‘invisible’ (Inkelas 1989). ‘Stray’ is structurally non-descript and was used in derivational analyses when syllables that were not parsed as part of a footing mechanism were subsequently parsed. ‘Extrametrical’, ‘extraprosodic’ and ‘invisible’  50  have been used to describe elements (usually coda consonants and syllables) that are invisible to phonological stress rules. Hayes (1995), among others, uses extrametricality as a rule to cause peripheral (usually final) syllables to be invisible to stress rules, independent of structure. The term ‘extrametrical’ is not appropriate for the elements considered here, because, as we shall see, extrapods are subject to word- and phrase-final prosodic processes and therefore are not outside the realm of prosody at all. In other proposals, extrametricality is extra-structural, such as the diacritic proposed by Hayes (1981), and the [+/-extrametricality] feature assigned to right-adjacent suffixes from roots in Nxa’amxcín by Czaykowska-Higgins (1993). A further proposal by Inkelas (1989) defines extrametricality in terms of a mismatch between morphological and prosodic word structure. In contrast to the above proposals, the non-exhaustively parsed elements considered here are defined in purely phonological terms on the basis of their structure only. Extrapods are characterised by their position in prosodic structure. Their phonological and phonetic characteristics fall out from their lack of foot-level status and lack of phonologically-specified relationship to some syllables, such as unstressed vowels.40 The proposed structure for extrapods (in a trochaic language) is given below: ( (=Foot level boundary, [ = PWord level boundary, V= stressed vowel, v=unstressed footed vowel, v=extrapod vowel)  (10)  PWord / \ F \ /\ \ [(CVCv)Cv] It is crucial at this point to emphasise that the unfooted syllable resulting from an  EXHAUSTIVITY violation is not definable as a ‘domain’. The extrapod is a syllable, just as a footed unstressed syllable is, and part of a PWord, just as a footed unstressed syllable is. However, the extrapod is not a domain that can be formally independently referenced from unstressed syllables. Rather, it is a sort of residue to which the label ‘extrapod’ is applied for the sake of convenience. That status of extrapods as residual elements rather 40  Extrapods do stand in a phonologically-specified relationship with the heads of PWords. This relationship is correlated with a phonetic contrast, whereby phonological heads are realised with higher F0 and intensity than extrapods, as we see in Chapter 5.  51  than definable domains is discussed further with reference to de Lacy (2002, 2006) later in this chapter. What follows is a discussion of the phonological characteristics of extrapods, all of which fall out from phonological constraints of the grammar. The predictions pertaining to acoustic realisation of prominence will be discussed in Section 3.8. 3.4.1 Markedness The need to recognise and formalise the violability of the Strict Layer Hypothesis derives from what Ladd (1996) calls ‘theoretically incompatible observations’. In other words, the existence of non-exhaustively parsed syllables is not predicted under an inviolable interpretation of Strict Layering. Rather, the model was modified to take into account their existence. As a result, we might predict that violations of EXHAUSTIVITY should be phonologically marked, and this in fact is true in a number of languages. Extrapods are either parsed as degenerate feet or left as residue at the foot level and are treated differently from properly formed feet. Languages often ‘repair’ the non-optimal elements by destressing, lengthening or rephrasing in the case of degenerate feet and lengthening, shortening, deletion or epenthesis in the case of extrametrical or stray syllables (Itô and Mester 1992; Mester 1994; Hayes 1995; Kager 1995 and references therein; Davis forthcoming). 3.4.2 Availability The availability of extrapods is dependent on the language-specific ranking of foot structure constraints. As a result, the prediction is that they will occur in languages where a conflict between the requirement to parse syllables and a limit on foot size occurs. Furthermore, they should in part be excluded in languages that do not permit Loose Minimal words (one foot plus any unfootable residue). 41 In languages where the minimal word size is subject to Strict Minimal word requirements (only one foot and no more), extrapods in shorter words are predicted not to surface.  41  McCarthy & Prince (1990) introduce an earlier version called the Maximum Minimal Word or MaxMinWord.  52  3.4.3 Size In terms of size, extrapods generally constitute less than a foot (as defined by languagespecific foot requirements), otherwise the criteria for footing would be satisfied. 42 In other words, extrapods cannot be large/heavy enough to constitute a foot within a particular language, regardless of whether feet are defined in moraic or syllabic terms. 3.4.4 Location Extrapods must peripheral (Hayes 1981, 1995). They cannot occur within feet, and most often (but not exclusively) occur at PWord peripheries: word-initial position in English (McCarthy & Prince 1993), Dutch (Kager 1989; de Lacy 2002) and Japanese (Itô & Mester 1992), word-final position in English (Liberman & Prince 1977) and Japanese (Itô & Mester 1992). In Dutch, extrapods occur word-medially (Kager 1989; de Lacy 2002, 2006)43. 3.4.5 Visibility Extrapods are invisible to foot-based processes (Pierrehumbert & Beckman 1988; Itô & Mester 1992). However, given that they are parsed to the PWord, they should be visible to PWord-level and higher level processes. 3.4.6 Constituent status Nespor and Vogel (1986) exclude extrapods from their model entirely through the strict inviolability of EXHAUSTIVITY. Since every syllable must be parsed into a foot, syllables that are not part of feet must be parsed as degenerate feet. In this theory, branching is nary, and feet are defined as “consisting of one relatively strong and any number of relatively weak syllables dominated by a single node” (1986:84). Extrapods would have to be parsed as either part of a ternary foot or as a weak (degenerate) foot.  42  An exception to this generalisation would be a language in which a Foot Alignment constraint (such as All-Foot-Left) outranks Parse-Syll. In such a case, a single foot at the left of right edge of the word would be parsed and the remaining syllables in the word would be unfooted. The unfooted residue in such a case would not be limited in size (D. Pulleyblank p.c.). 43 This is by no means an exhaustive list but rather is limited to languages discussed in this thesis.  53  Pierrehumbert and Beckman (1988) cite unpublished data that convince them to move away from extrametricality as a diacritic, to extrametricality as a structural representation. Under their new definition, extrapods would be defined as adjoined “to a node at the next higher level than is usual for that type of constituent” (1988:148). Because of the undeniable existence of extrapods, they adjust their definition of tree structure to appeal to uniform height rather than uniform depth and define extrametrical nodes formally as: (11)  a. A node ni having height p is extrametrical if it is directly dominated by a node nj having height q≥ p +2 b. For any extrametrical node ni directly dominated by nj, and for all nonextrametrical daughter nk of nj, either ni < nk, or nk< ni (Pierrehumbert & Beckman 1988:150 ex. 11).  The fact that nodes cannot dominate only extrametrical materials falls out from their definition of Extrametricality and Uniform Height. In positing a formal definition of extrametrical domains, Pierrehumbert and Beckman (1988) makes the prediction that extrapods should be referenceable independently of other structures. This is contra de Lacy (2002, 2006), discussed below. One of the most recent, and for this research, most relevant treatments of extrapods is de Lacy’s (2002; 2006) analysis of Kager’s (1989) Dutch Semi-Formal vowel reduction.44 In that register, /u, i/ undergo reduction in footed unstressed syllables but not in what he calls stray unstressed syllables (what I call extrapods). By making this distinction in footing, he can account for an alternation in two unstressed syllables that have been previously assumed to be non-distinct. According to de Lacy’s proposal, every node in the Prosodic Hierarchy is a head (Designated Terminal Element or DTE) or a non-head (Non-designated Terminal Element or non-DTE). The terms are defined below:  44  De Lacy (2006) also presents an analysis of vowel reduction in Berguener Romansh non-stressed vowels. Pre-tonic unfooted unstressed vowels, which are not subject to foot –DTE constraints, do not permit schwa. Post-tonic footed unstressed vowels, which are subject to foot –DTE, constraints do. Given that the St’át’imcets phonological process is more similar to the Dutch data (in that it deals with non-initial posttonic extrapods), it is the Dutch data that will be discussed in detail. The reader is directed to de Lacy (2006) for further discussion of Berguener Romansh.  54  (12)  pDTE∏=def  A node n of prosodic type p is the DTE of prosodic category ∏ iff the path from n to ∏ consists entirely of prosodic heads. pnon-DTE∏=def  A node n of prosodic category p is a non-DTE iff i) n is (transitively) associated to ∏ and ii) n is not a pDTE∏ (de Lacy 2006:64) By appealing to the above definitions, de Lacy can distinguish between two types of unstressed syllables: those that are part of a foot, and those that are not. A point that will prove to be important in the following discussion is that the term ‘foot-non-DTE’ is not synonymous with ‘unstressed syllable’. Unstressed syllables that are not parsed into feet are not foot non-DTEs-they have no DTE status at all with respect to feet. This follows from the definition of non-DTE: to be a non-DTE of a foot, a segment must be dominated by a FT node. (Where FT=the prosodic category 'foot') (de Lacy 2002:113) As previously mentioned, this model makes the prediction that phonological processes cannot target extrapods without also targeting unstressed syllables. Since extrapods have no foot-level standing, they are characteriseable only as Word Non-DTEs. Footed unstressed syllables are also Word Non-DTEs. As a result, processes that target Word Non-DTEs apply to both extrapods and footed unstressed syllables while Foot nonDTE processes apply to footed unstressed syllables only. In the case of the Dutch semi-formal register, vowels in the –DTE of a foot reduce, while those outside of feet (extrapods and heads of feet) do not. Motivation for such position-specific reduction comes from Kager’s generalisation, which states that, “[f]or vowels whose reducibility depends on position, reduction is generally easier in adjunct [i.e. foot non-DTE] positions than in stray positions” (Kager 1989:313). De Lacy argues that the Dutch data represent an example of positional markedness. In this case, the sonority of vowels is prosodic position dependent. There is pressure to maximise the saliency head of the foot, and one way to achieve this is to minimise the sonority of the non-head. As a result, schwa is the most marked segment  55  for DTEs but the least marked segment for non-DTEs. Extrapods, which are not part of feet, are not subject to such reduction pressures. To account for the Dutch pattern, de Lacy proposes a constraint which assigns a violation for every instance of a vowel of equal or greater sonority to /i, u/ in the nonDTE of the foot. This constraint is crucially ranked above IDENT [+high]: (13)  /individu/[(ìnde)vi(dú)] *[(ìnde)ve(dú)]  Table 3.1 De Lacy’s example 27 (2006: 228) /individu/ *-∆Ft≥{i,u} (a) (ìndi)vi(dú) *! (b) (ìnd¢)¢(dú) (c) (ìnd¢)vi(dú)  IDENT [+high] **! *  By distinguishing between footed and unfooted syllables, de Lacy reveals the importance of appealing to parsing rather than stress in defining extrapods. 3.4.7 Language-specific (i.e. non-structural) characteristics The characteristics of extrapods that do not fall out from structural requirements tend to be phonotactic restrictions. As a result, the shape of extrapods is determined on a language specific basis. In English, for example, Liberman and Prince (1977) report that extrapods are word-final sonorants following consonants, or the non-syllabic glide /–y/ in a number of suffixes (Liberman & Prince 1977: 292-295). In the informal register of Dutch, there is a stricter sonority restriction on unstressed syllables than on extrapods. As a result, while extrapods do not permit the full range of vowels (no low vowels), they permit both high and mid-high vowels, which are not permitted in unstressed footed syllables. As mentioned in 3.4.3, language specific foot-binarity requirements affect how heavy/large an extrapod might be. For example, Japanese permits only monomoraic extrapods (Ito & Mester 1992). In summary, we have the following generalisations regarding the phonological characteristics of extrapods:  56  Table 3.2 Characteristics of extrapods Universal 1) Markedness: Subject to language-specific repair strategies 2) Availability: Only in cases where a parsing conflict occurs 3) Size: Must constitute less than a foot 4) Location: Peripheral (at the foot level, though more usually at the word or phrase level) 5) Visibility: Not visible to foot-level phonological processes 6) Constituent status: Does not constitute separate, degenerate foot. -Distinct from unstressed syllables but cannot be independently referenced  Language specific -Japanese: must be monomoraic -Dutch: no low vowels - English: sonorant following consonant or affixal nonsyllabic /–y/.  The next section lays out the evidence that in St’át’imcets, as in Dutch, all unstressed syllables are not equal. Section 3.6 introduces alternative models and Section 3.7 presents evidence that the extrapod model is the best solution to account for the data.  3.5  St’át’imcets Data  St’át’imcets, like the informal register of Dutch presented above, shows phonological evidence of two types of unstressed syllables: those immediately following a stressed syllable and those following another unstressed syllable (removed from the stressed syllable by one). There are two phonological processes that support positing a distinction between these two types of unstressed syllables: i) vowel alternating suffixes and ii) glottal alternating suffixes. Vowel alternating suffixes are addressed first. 3.5.1 Vowel alternating suffixes The middle suffix /-Vm/ and the n-transitivising suffix /-Vn/, when following a somatic lexical suffix, alternate between [-am] ~ [-em] and [-an]~[-en]. The alternation is described by van Eijk (1997) and Davis (forthcoming) as being dependent on stress. If the  57  vowel of the suffix is stressed, as in examples (14) and (15), it surfaces as an /a/ (the vowel in question is bolded): 45 NAPA  Orthography  (14) a. x∑ù«-q∑-áÂ=¬kan... b. mùl-x¢n-áÂ=¬kan c. ßùp-q∑áÂ=¬kan (15) a. x∑ù«-q∑-á˜=¬kan... b. mùl-x¢n-á˜=¬kan... c. ßùp-q∑-á˜=¬kan...  English  Cut’qwám’lhkan i tsícwan I took my hat off when I lámcal went to church (lit. I bared Mulcenám’lhkan Supqwám’lhkan Cut’qwán’lhkan ti/ta sqáycwa Mulcenán’lhkan ti/ta sk’úk’wm’ita Supqwán’lhkan ti/ta nkwtámtsa  my head) I put my feet in the water I scratched myself on the head I took the man’s hat off I put the child’s foot in the water I scratched my husband on the head  If the vowel of the middle/transitive suffix is unstressed, it surfaces as a schwa:46 NAPA  Orthography  English  (16)  a. ßùp-ki˜-úß-Åm=¬kan b. t¢q∑-úß-Åm=¬kan  Supkin’úsemlhkan. Teqúsemlhkan.  I scratched my forehead I touched my face  (17)  a. ßùp-ki˜-úß-Ån=aß b. ßùp-aná÷-Ån=¬kan  Supkin’úsenas. Supaná7enlhkan.  She scratched his forehead I scratched his ear  However, when the suffix is word final, we see an alternation that cannot be accounted for by a purely stress-based analysis. The suffixes in examples (18) and (19) are wordfinal and unstressed and the vowel surfaces as schwa. The same suffix vowels in examples (20) and (21) are also word-final and unstressed but surface as full vowels:  45  All examples are given in a broad NAPA transcription, orthography, and English, and were elicited by the author, unless otherwise specified. Glottalisation of the resonant in these cases is dependent on the presence or absence of a glottalising suffix, as discussed in Section 1.4.3. The schwa in (14b), (15b), (17a) and the initial schwa in (16b) are also are often omitted by speakers (H. Davis p.c.). 46 One possible exception to this is the form /t¢q-al-uß-¢m=¬kan/. The form was recorded by the author as [t¢qalúߢm¬kan] but subsequently from the same speaker as [t¢qáluߢm¬kan] by H. Davis (p.c.). This variation in pronunciation may be due to the reluctance of some speakers to put stress on the connector /-al/ (H. Davis p.c.). The same variation occurred in /t¢q-al-uß-¢n=¬kan/.  58  NAPA  Orthography Wá7lhkan supkin’úsem Wá7lhkan teqúsem  English I’m scratching my forehead Yes, I touched my face47  b. ...ßùp-ki˜-úß-Ån  Wá7lhkan supaná7en ti/ta sqáx7a Cúz’lhkan supkin’úsen  c. ...t¢q∑-úß-Ån  Wá7lhkan teqúsen  I’m scratching the dog’s ear I’m going to scratch his forehead Yes, I touched his face.  a. ...x∑ú«-q∑-a  Cúz’lhkan cút’qwam’  b. ...múl-x¢n-a  Cúz’lhkan múlcenam’  c. ...ßúp-q∑-a  Cúz’lhkan súpqwam’  a. ...x∑ú«-q∑-a˜...  Cúz’lhkan cút’qwan’ ti/ta sqáycwa Cúz’lhkan múlcenan’ ti/ta sk’úk’wm’ita c. Páptkan súpqwan’ ti/ta nkwtámtsa  (18)  a. ...ßùp-ki˜-úß-Åm b. ...t¢q∑-úß-Åm  (19)  a. ...ßùp-aná÷-Ån...  (20)  (21)  b. ...múl-x¢n-a˜... c. ...ßúp-q∑-a˜...  I’m going to take my hat off. I’m going to put my feet in the water I’m going to scratch myself on the head I’m going to take the man’s hat off I’m going to put the child’s foot in the water I always scratch my husband on the head  As we can see from the examples above, describing the domain of vowel alternation is not as simple as stressed versus unstressed. Our analysis must account for why the unstressed suffix vowels in (18) and (19) are different from the ones in (20) and (21). Van Eijk (1997:119) describes the variation as “depending on the preceding somatic suffix” while Davis (forthcoming) describes the domain of alternation in somewhat more detail: ...in combination with a lexical suffix, the middle suffix shows up as either –am(’) or –em(’), depending on where the stress might fall. If the middle suffix could be stressed (for example, if more suffixes were added), then it shows up as -am(’); otherwise, it shows up as –em(’)... (Davis forthcoming: 44)  47  The translation given by the consultant was an unusual one and not the only translation available for such sentences. See Davis (forthcoming) for more information.  59  He uses the same description for the transitiviser –an: “The choice between –en and –an following a lexical suffix is largely dictated by stress, just as with the middle suffix...” (Davis forthcoming: 47) What Davis means by ‘depending on where stress might fall’ concerns the position of the syllable in the word. The final syllable of a syllabically odd-numbered word is often referred to by van Eijk and Davis as ‘potentially stressed’. Van Eijk and Davis distinguish between syllables that are stressed (the first syllable of a word and every second syllable thereafter, excluding final ones), those that cannot be stressed (every even numbered syllable following the stressed one) and ‘potentially stressed’ syllables (those word-final odd-numbered syllables which would have stress in longer words). Potentially stressed syllables are those left unfooted by the constraints proposed in Roberts & Shaw (1994). In longer derivations of a word ending in a potentially stressed syllable, that same syllable is no longer final but rather the head of a foot. In the first example below, /=kax∑/ is final, unfooted and unstressed. In the second, the same enclitic is not word final but carries primary stress as the head of the rightmost foot. (22)  a. (æù¬u˜)(túmu¬)kax∑ Tsulhun’túmulhkacw. b. (æù¬u˜)(tùmu¬)(káx∑kŬ) Tsulhun’tumulhkácw kelh.  You pointed as us48 You will point at us (Roberts & Shaw 1994)  Following Roberts and Shaw’s (1994) analysis, I consider the ‘potentially stressed’ syllable as unfooted (i.e. an extrapod). With this parsing in mind we can account for the phonological distinction between a final full vowel and schwa in two unstressed syllables. Specifically, vowel reduction occurs when the unstressed vowel is parsed as the non-head or -DTE of a foot. (23)  a. b.  [(ßùpki)(˜úßÅm)¬kan] [(ßùpki)(˜úßÅm)]  48  This example, cited in Roberts (1993) and Roberts and Shaw (1994) was re-elicited. The orthography has been changed in keeping with the other examples.  60  (24)  a. b.  [(ßùpki)(˜úßÅn)aß] 49 [(ßùpki)(˜úßÅn)]  A full vowel occurs elsewhere: in the head of the foot (25a, 26a) and when the unstressed vowel is in an extrapod syllable (25b, 26b): (25)  a. b.  [(mùlx¢)(ná¬kan)] [(múlx¢)naÂ]  (26)  a. b.  [(mùlx¢)(nᘬkan)] [(múlx¢)na˜]  The St’át’imcets data can be accounted for by constraints similar to those proposed by de Lacy to account for the Dutch data. First, a markedness constraint ensures no full vowels surface in the non-head of the foot: (27)  *-∆Ft > schwasuffix  Assign a violation for every instance of foot non-head with sonority greater than schwa (in the middle and transitive Vn suffixes following somatic lexical suffixes)  This markedness constraint must outrank the vocalic faithfulness constraints. De Lacy (2002: 135) uses IDENT V to collectively refer to the faithfulness constraints that preserve peripherality of vowels (e.g. IDENT [+low] for /a/). I will use the same constraint: (28)  IDENT V  Correspondent segments in input and output have identical peripherality feature values (my interpretation)  The stress assignment constraints in the language, which guarantee parsing remains constant, must also outrank the markedness constraint. (29)  STRESS:  Be trochaic, binary and right-headed; disprefer schwas and consonant clusters as heads; etc.  This constraint interaction can be seen in the tableaux below:  49  As mentioned above, the schwas in these examples, particularly (24a) are often entirely elided.  61  Table 3.3 Vowel reduction in non-head tableau /ßupki˜ußam/ a. [(ßùpki)(˜úߢm)] b. [(ßùpki)(˜úßam)] c. [(ßùpki)˜u(ßám))]  STRESS  *-∆Ft> schwasuffix  IDENT V *  *! *!  Table 3.4 Extrapod tableau /mulx¢naÂ/ STRESS a. [(múlx¢)naÂ] b. [(múlx¢)n¢Â] *! c. [(múl)(x¢naÂ)]  *-∆Ft> schwasuffix  IDENT V *!  In the first tableau, the winning candidate is selected, in spite of violating the lowest ranked constraint. The other two candidates are ruled out due to violations of the crucially higher ranked constraints. In the second tableau, the winning candidate incurs no violations while the second candidate violates IDENT V by reducing the vowel outside of foot non-head position and the third candidate satisfies *-∆Ft> schwasuffix at the expense of violating STRESS. The tableaux above show that appealing to constraints that refer to foot domains allows for a simple and elegant analysis of St’át’imcets vowel alternation. In summary, St’át’imcets vowel reduction is similar to that of Dutch, and can be accounted for by the same constraints. A positional markedness constraint, motivated by head maximisation, prohibits vowels more sonorant than schwa in foot non-head position. The extrapod vowel, to which foot-level constraints do not apply, does not reduce. The above analysis is parsing rather than stress, dependent and comprehensively and elegantly accounts for the data. The second piece of evidence that supports the claim that in St’át’imcets, extrapods and unstressed syllables are discrete phonologically is glottal alternation in the transitive paradigm.  62  3.5.2 Transitivising paradigm glottal alternation50 In St’át’imcets, resonants in seven suffixes associated with the transitive paradigm alternate between glottalised and non-glottalised forms in the same environments as mentioned in Section 3.5.1. Specifically, a resonant following a footed, unstressed vowel surfaces as glottalised and a non-glottalised resonant occurs elsewhere. The suffixes that undergo alternation come from three categories: transitivising suffixes, object suffixes and a non-topical subject marker. St’át’imcets has four types of transitivisers: s-type, min-type, cit-type and n-type. S-type transitivisers are variously referred to as causative, non-control or neutral. According to Davis (forthcoming), they are used whenever n-type (directive/full-control) transitivisers are blocked. Min-type are referred to in the Salish literature as relational. They are used with mental state, perception, joint-action, and transfer predicates or to indicate indirect goals. Cit-type are indirectives, used to create ditransitive predicates. N-type transitivisers comprise two subtypes: -Vn types are directive or full control while nun-types, which are very infrequent, are limited control. –Vn-type transitivisers are generally interpreted as “accomplishments with controlling agents” (Davis forthcoming: Chapt. 39).51 Of the four types of transitivising suffixes in St’át’imcets, two undergo glottal alternation and a third selects object suffixes that undergo the alternation. In min-type and n-type transitivisers the final resonant in the suffix undergoes alternation: [-min’]~[min], [an]~[Vn]~[Vn’]52 and [nun’]~[nun]. S-type transitivisers select special object suffixes, three of which contain a resonant that undergoes glottal alternation in exactly the same environments as n- and min-type suffixes [-tum’x]~[tumx] (1sgObj), [-tum’i(n)]~ [tumi(n)]53 (2sgObj) and [-tan’i]~[tani] (3plObj co-occurring with 1sgSubj)54. The final suffix which undergoes alternation is /-tali/ ([-tal’i]~[-tali]). This suffix marks a transitive subject as non-topical, and can occur with any of the transitivisers,  50  A version of this section appeared as Caldecott (2006b). Davis, forthcoming (Chapter 39) gives non-agentive passives and experiencer predicates as exceptions to this general interpretation. 52 V represents a full vowel copy of the root vowel. Vowel alternations are not addressed here. 53 The (n) in the 2sgObj morpheme represents the fact that /n/ is often dropped preceding other object suffixes but always retained word-finally. 54 Sg=singular, pl=plural, Obj=object, Subj=subj. 51  63  even -cit, which does not undergo the glottal alternation itself, nor in its object suffixes (Davis, forthcoming: Chapt. 46). This section focuses on the –min’ transitivising suffix, which represents the general pattern for all of the suffixes.55 The glottal alternation in this suffix occurs in the same prosodic contexts as the vowel alternations in Section 3.5.1: The suffix final [n’] surfaces as glottalised following an unstressed vowel (as in 30), and plain following a stressed vowel (as in 31):56 (30) a. «íq-mi˜ b. «íq-mi˜-aß c. «ìq-mi˜-ítaß d. ßq∑áÒ-mi˜ e. ßq∑áÒ-mi˜-aß f. ßq∑àÒ-mi˜-ítaß  T’íqmin’! T’íqmin’as. T’iqmin’ítas Sqwál’min’! Sqwál’min’as. Sqwal’min’ítas.  Come and get it! S/he arrived for it. They arrived for it. Report on him/her! S/he reported on him/her. They reported on him/her.  (31) a. ÷ìwa÷-mín-aß b. ÷ì÷wa÷-mín-itaß c. k∑ù¬¢n-mín-aß d. k∑ù¬¢n-mín-itaß  I7wa7mínas. I7wa7mínitas. Kulhenmínas. Kulhenmínitas.  S/he went with him/her. They went with him/her. S/he borrowed it. They borrowed it.  This basic pattern is again complicated by data in which the transitive suffix occurs wordfinally following an unstressed vowel and is two syllables removed from the stressed vowel. In (32) below, a plain resonant surfaces following an unstressed vowel: (32)  a. ÷í÷wa÷-min b. k∑ú¬¢n-min  Í7wa7min! Kúlhenmin!  Accompany him/her! Borrow it!  We can rephrase the generalisations by appealing to foot structure. Glottalisation surfaces only when the resonant is foot final (33a, b). When the resonant occurs outside of this position (i.e. the head syllable or extrapod), it surfaces as non-glottalised (34a, b): (33)  a. b.  [(«íqmi˜)] [(«íqmi˜)aß]  55  The reader is directed to Appendix B for the application of the analysis to the other suffixes. Jules (2008) reports that in Secwepemc, the glottalisation of the –min suffix resonants reflects an accidental interpretation of the predicate: i) Temtmíns re kyé7es. ‘She sat beside her grandmother’s bedside’ versus ii) Temtmín’s re qmut.s ‘She accidentally sat on her hat’. No such interpretation has been reported for St’át’imcets. 56  64  (34)  a. b.  [(÷ì÷wa÷)(mínaß)] [(÷í÷wa÷)min]  The generalisations proposed in above rely on one crucial assumption: that the intervocalic glottalised resonant is parsed as part of the preceding foot and syllable as in (35a) rather than wholly as the onset of the following syllable as in (35b): (35)  Í Í /| \/ |\ a. («íq-mi˜)a š  Í Í /| /|\ b. («íq-mi)˜aš  By positing that the resonant is ambisyllabic, and therefore, part of the previous foot, the generalisation is clear. Resonants surface as glottalised foot-finally, and plain elsewhere. If we were to adopt the traditional parsing of the resonant (as in 35b) the generalisation is less clear: resonants surface as glottalised when in the onset of an unstressed syllable that is two syllables away from the stressed vowel (33b) or when in coda position of an unstressed syllable that follows a stressed syllable (33a). By positing that the glottalised resonant is ambisyllabic, we are able to propose a simple and elegant environment for alternation. Prior to discussing the analysis, we must first motivate ambisyllabicity. 3.5.2.1 Ambisyllabicity The proposal that glottalised resonants are ambisyllabic is not one that has previously been made for St’át’imcets. Most research dealing with syllable structure in the language has focused on the parsing of clusters (Matthewson 1994; Roberts and Shaw 1994; Shaw 2004b). In these works, the possibility of ambisyllabic consonants is not raised and intervocalic resonants are parsed as the onset of the following syllable (due to a highlyranked ONSET constraint). However, there are two types of evidence that support the syllabification in (35a) over that in (35b). First, there is a trend in the language towards loss of glottalisation on resonants (Bird & Caldecott 2004a). Glottalised resonants are neutralised significantly more frequently in onset (including word-initial and post-C onsets) than in coda position. The fact that glottalised resonants are NOT lost in intervocalic position can be taken as evidence that they may not be strictly parsed as onset resonants. There is a phonotactic  65  restriction against word-initial glottalised resonants (with the exception of reduplicated forms) and no glottalised resonant initial suffixes, suggesting a trend toward glottalisation being licensed only in coda position, perhaps by a mora. The second piece of evidence comes from phonetics. Bird & Caldecott (2004a,b) and Bird et al. (2008) found that onset glottalised resonants in St’át’imcets are preglottalised (meaning the sub-oral articulation occurs preceding or concurrent with resonant onset but ceases before resonant offset) while coda glottalised resonants were post-glottalised (meaning sub-oral articulation followed resonant onset but continued until or beyond resonant offset). Intervocalic glottalised resonants were articulated with a closure in the middle of the resonant (or sub-oral articulation follows resonant onset and ceases before resonant offset). 57 Given that the timing of intervocalic glottalised resonants is a combination of both onset and coda timing, we may interpret intervocalic glottalised resonants as ambisyllabic: neither purely coda nor onset but a combination of both. The evidence presented above, while not conclusive, leans towards ambisyllabicity. The underlying motivation for ambisyllabicity may be to enhance foot boundaries. For example, Bye (2007b) reports both historical and synchronic lengthening (gemination or consonant gradation) of obstruents at foot boundaries in Inari Saami. Bye (2007a) motivates consonant gradation by appealing to ‘enhancement’. Closed syllable rhymes enhance the distinction between syllables in that they are more acoustically salient than open ones.58 Following this line of reasoning, an ambisyllabic resonant could increase the contrast between the preceding unstressed syllable and the following one, thus maximising foot boundary contrasts. At this point, it is unclear whether glottalised resonants are alone in being ambisyllabic, perhaps due to their phonetic manifestation, or whether other consonants might also be ambisyllabic. To the best of my knowledge, there is no evidence against positing such structures. Pending further research, I assume that the intervocalic 57  In some speakers glottalisation on resonants is manifested as overlapping sub-oral and oral gestures, i.e. creaky-voiced throughout. See Bird and Caldecott (2004a,b) and Bird (2008) for further discussion. 58 In Inari Saami, consonant gradation can be seen as maximising the coda of the head of the foot (the stressed syllable in the foot). In the data reported in Bye (2007b), gradation occurs between the initial stressed syllable and an immediately following syllable, which has secondary stress. Coda maximisation is a cross-linguistic strategy used by languages to maximise the head syllable of the foot (Bye & De Lacy 2008).  66  glottalised resonants are ambisyllabic, and therefore fill the coda position of the preceding syllable and foot. 3.5.2.2 Analysis The data in 3.5.2 can be accounted for by an interaction between a context-free markedness constraint and a positional faithfulness constraint (Beckman 1998). As with the previous analysis, the stress assignment constraints of the languages are collectively represented as in (36): (36)  STRESS:  Be trochaic, binary and right-headed; disprefer schwas and consonant clusters as heads, etc.  The context-free markedness constraint that disfavours glottalisation on a resonant is as follows: (37)  *CG/RES  No CG on a resonant  *CG/RES is based on a ‘relative harmony’ scale for constricted glottis (glottalisation) (Howe & Pulleyblank 2004).59 This harmonic ranking is supported by typological evidence that glottalised resonants are more marked than ejectives. If we consider this scale in terms of markedness, we can express the harmonic scale as: *CG/RES>>*CG/OBS. As ejectives do not undergo the glottal alternation, nor partake in other morphological processes that involve resonant glottalisation (such as diminutive reduplication), or even the trend towards loss of glottalisation on resonants, *CG/OBS must be very lowly ranked. At issue here is the interaction between constraints that reference constricted glottis on resonants, so constraints pertaining to (CG) on obstruents are not discussed (see Howe & Pulleyblank (2004) for further discussion). In terms of what the relevant faithfulness constraint should be, a complication arises. Reference to the subsegmental features of coda consonants in the margins of the non-head syllable is not possible under de Lacy’s analysis of positional markedness (2002:118) but perhaps we can adapt the terminology to apply to the prosodic faithfulness constraint in this case.60 If, following Roberts (1993), Roberts and Shaw 59 Howe 60  and Pulleyblank (2004) use CG/SON AND CG/STOP. According to de Lacy’s Scale Structure Combination Restriction, markedness constraints referring to  67  (1994) and Matthewson (1994), we assume that coda consonants in the language are nonnuclear moraic, we could modify de Lacy’s theory to permit reference to all non-nuclear moraic root nodes in -∆{Ft,σ} (syllable –DTEs of feet or the non-head or unstressed syllable in the foot):61 (38)  IDENT (CG/µ;-∆{ Ft,σ})  For all -∆{Ft,σ}, if an input non-nuclear moraic root node has (CG) associated with it, then the corresponding output root node has (CG) associated with it  The constraint above protects (CG) in the coda position of the non-head syllable of the foot. Both the markedness and faithfulness constraints must also be ranked with the STRESS  (39)  constraint proposed above, in the following manner: STRESS, IDENT (CG/-∆/µ;-∆{Ft,σ})>>*CG/RES  We can see the constraint interaction in the tableaux below. The tableau in Table 3.5 shows the crucial ranking of both STRESS and IDENT (CG/-∆/µ;-∆{Ft,σ}) above *CG/RES: Table 3.5 Tableau accounting for surface glottalisation62 /«iq-mi˜-as/ STRESS IDENT (CG/-∆/µ;-∆{Ft,σ}) a.  [(«íq-mi˜)aß] b. [(«íq-min)aß] c. [«iq-(mí˜aß)]  *CG/RES *  *! *!  *  Candidate (a) wins because both STRESS and IDENT (CG/-∆/µ;-∆{Ft,σ}) outrank *CG/RES, protecting the glottalised resonant in the weak part of the foot. Candidate (b) satisfies *CG/RES but violates the positional faithfulness constraint, and is thus ruled out. Candidate (c) satisfies IDENT (CG/-∆/µ;-∆{Ft,σ}) but violates the highly ranked STRESS constraint and is ruled out. The tableau above demonstrates that by appealing to a foot  subsegmental feature scales cannot refer to prosodic structure: “For example, there can be no constraint *σ¡/{dorsal}, militating against dorsal segments (i.e. velar consonants, back vowels) in stressed syllables” (de Lacy 2002: 13). 61 Under de Lacy’s model, moraic coda consonants are defined as -∆(σ,µ) or non-head mora of the syllable (2002: 44; 2006: 65). For the sake of clarity, I will represent them in the constraint simply as µ. 62 The glottalised resonant is considered ambisyllabic.  68  non-head positional faithfulness constraint, we can account for the retention of glottalisation. The next tableau shows that the proposed analysis can also account for the neutralisation of glottalisation in head position. Table 3.6 demonstrates that our constraint ranking also accounts for neutralisation in the head of the foot: Table 3.6 Tableau accounting for neutralisation following the stressed vowel /÷i÷wa÷mi˜aß/ STRESS IDENT (CG/-∆/µ;-∆{Ft,σ}) *CG/RES a. (÷ì÷wa÷)(mí˜aß)] *! b.  [(÷ì÷wa÷)(mínaß)] c. [÷i÷(wá÷mi˜)aß] *! * Candidate (b) is our winning candidate with no violations. Candidate (a) violates *CG/RES and candidate (c), in an attempt to satisfy IDENT (CG/-∆/µ;-∆{Ft,σ}) by shifting foot structure, violates the highly ranked STRESS constraint. The tableau above shows that the proposed analysis accounts for neutralisation outside of the foot non-head. The tableau below shows that the analysis also accounts for neutralisation in the extrapod. Table 3.7 Accounting for neutralisation in unparsed foot /÷i÷wa÷mi˜/ STRESS IDENT (CG/-∆/µ;-∆{Ft,σ}) a. [(÷í÷wa÷)mi˜] b.  [(÷í÷wa÷)min] c. [÷i÷(wá÷mi˜)] *!  *CG/RES *! *  Candidate (b) is our winning candidate with no violations. Candidate (a), which remains faithful to the underlying representation, violates *CG/RES. Candidate (c), which attempts to satisfy IDENT (CG/-∆/µ;-∆{Ft,σ}) at the expense of stress is ruled out by the higher ranked constraint. Thus, we are able to account for the neutralisation of glottalisation in extrapod position. To summarise, by ranking a positional faithfulness condition, which specifically targets codas in foot non-head syllables, above a context-free markedness constraint, we are able to account for the glottal alternation above. An analysis that relies on stress rather than parsing could not account for neutralisation in unstressed extrapods but not footed, unstressed syllables. One adjustment to the constraint ranking proposed above must be made. Consider the examples below: 69  (40) a. b.  €á„t-min š°á™-mi„  'Come back for it!' 'Report on him/her'  As we can see from the examples above, glottalised resonants in roots can surface immediately following a stressed vowel, unlike the glottalised resonants in the suffixes in (31). These data force us to re-examine our constraint ranking, as we can see in the tableau below. Table 3.8 Retention of glottalised resonant in root ( marks unintended winner)  /š°a™-mi„/  STRESS  IDENT (CG/-∆/µ;-∆{Ft,σ})  a.  (š°á™-mi„) b.  (š°ál-mi„)  *CG/RES **! *  Candidate (a) retains glottalisation on both resonants but is ruled out by two violations of *CG/RES, to candidate (b)'s one violation. The preservation of glottalisation on the resonant in candidate (a) is outside of the protected environment, so our constraints cannot account for its retention. This asymmetry between faithfulness to glottalisation in roots and faithfulness to glottalisation in suffixes needs to be addressed in our constraint ranking. Affixes in general in the language have restricted phonotactics with respect to roots (van Eijk 1997). This asymmetry is not surprising, as according to Beckman (1998), roots rate as a privileged position for faithfulness. Therefore, in order to address this asymmetry and to ensure that roots with glottalised resonants retain glottalisation even when they are not in the weak position of the foot, we must posit a highly ranked IDENT (CG/ROOT) constraint. (41)  IDENT (CG/ROOT)  For all (CG), if (CG) is associated with a root node in an input root, then (CG) must be present in that root node in the output root. We can see this ranking at work in the tableau below: Table 3.9 Root constraints account for retention of glottalisation /√š°a™-mi„/ IDENT STRESS IDENT (CG/-∆/µ;-∆{Ft,σ}) *CG/RES (CG/ROOT) a.  (√š°á™-mi„) ** b. (√š°ál-mi„) *! * c. (√š°ál-min) *! * d. (√š°á™-min) *! *  70  Candidate (a), our winning candidate, is faithful to glottalisation in the root and in the suffix. It incurs two violations of *CG/RES but wins out over Candidate (b), which violates the highly ranked IDENT (CG/ROOT) constraint and * CG/RES. Candidate (c) violates both the highly ranked IDENT constraints and is ruled out, while candidate (d) violates IDENT (CG/-∆/µ;-∆{FT, σ}) and *CG/RES. To summarise, the addition of the highly ranked root-specific faithfulness constraint allows for a straightforward analysis of all of the data. In summary, this section has proposed a positional faithfulness constraint using de Lacy’s terminology and demonstrated that it can be used to account for the mintransitivising suffix alternation. By appealing to the distinction between foot non-heads and extrapods, which have no foot standing, we can account for the neutralisation of glottalisation in extrapods but not in footed, unstressed syllables. The IDENT (CG/-∆/µ;-∆{Ft,σ}) constraint above is motivated as a positional faithfulness constraint. However, the position of faithfulness in this case is not one that is normally considered prosodically privileged. The preservation of the marked segment in a weak position and the neutralisation in a relatively strong position is contra what is predicted under positional faithfulness. Beckman (1998) motivates positional faithfulness by referencing prosodically privileged positions. Certain positions are ‘privileged’ because they have a psycholinguistic or phonetic prominence, such as roots, root-initial syllables, stressed syllables, long vowels, and onsets. Other, non-privileged positions do not have this advantage. Beckman argues that the above positions are privileged by increased saliency, which allows them to convey a wider range of marked features and segments. In fact, she argues “that this perceptual salience is exploited directly in the phonological component of the grammar, via faithfulness constraints” (1998:185). She goes on to say, “[i]n circumstances of positional neutralisation, it is always the perceptually non-prominent position which undergoes reduction, while the prominent positions preserve a full range of contrasts” (1998:4). In fact, one of the phonological diagnostics of a privileged position is that there is “maintenance of contrasts which are neutralised elsewhere” (1998:1). Beckman accounts for such phonological behaviour by ranking a positional  71  faithfulness constraint above a context-free faithfulness constraint and context-free markedness constraints: (42)  IDENT POSITION (F) >> IDENT F, *F When we consider the glottal alternation seen above, the phonological diagnostic  for a privileged position is present. A more marked segment is maintained in a weak position and not in other contexts. The constraints used in the analysis mirror Beckman’s: Positional faithfulness >> context free faithfulness, markedness. However, the position that retains glottalisation is one that matches four of Beckman’s unprivileged positions: • • • •  non-initial syllable unstressed syllable syllable coda affixes/inflections/function words  Given that the glottalised resonants in these suffixes occur in a phonetically unprivileged position, the fact that they demonstrate one of the phonological diagnostics for a privileged position is unexpected. The implication of this is that, based on the St’át’imcets data presented here, we must assume one of the following three options: 1) St’át’imcets footed unstressed syllables are phonetically salient in some different way that makes them privileged. 2) Positional faithfulness need not motivated by phonetic privilege 3) This is a case of positional markedness, not faithfulness. If we consider the first option, we must make a distinction between privileged footed unstressed syllables on the one hand and unprivileged unparsed unstressed syllables and stressed syllables on the other. It is difficult to imagine what sort of phonetic characteristic that might be. The second option is to remove the underlying motivation for positional faithfulness. Given the strong evidence Beckman presents that positions of increased saliency are loci of faithfulness while other positions are not, we should not reject this  72  motivation on the basis of one exception. More evidence indicating positional faithfulness in non-prosodically-strong positions would be required before rejecting this underlying principle. The final option is to appeal to positional markedness rather than faithfulness. If we assume, following (Zec, 1994), that glottalised resonants are lower on the sonority scale than plain resonants, we could account for why they are maintained in foot-final position. Being lower in sonority, glottalised resonants (GRs) would be less marked in syllable coda position than plain resonants. The reduced sonority of the glottalised resonant would maximise the foot boundary contrast. This is the most plausible of the three options. Motivation for the glottal alternation is then similar to the reduction process in Section 3.5.1: maximisation of contrast. Just as schwa is preferred in foot margins to maximise the saliency of the head, the phonetic realisation of glottalised resonants in foot-final position maximises the foot boundary. As mentioned in Section 3.5.2.1 above, foot-final glottalised resonants, whether in coda position or ambisyllabic, are often realised as modal followed by the engagement of the sub-oral articulators. As a result, there is a salient auditory break between the end of the foot and the following syllable. If maximising the contrast at a foot boundary is the underlying motivation for the retention of glottalisation, the analysis of neutralisation elsewhere is straightforward: the head syllable of the foot and the extrapod are not at foot boundaries, so there is no motivation to retain glottalisation.63 If glottalisation is signalling contrast at a foot boundary, this makes the prediction that increased glottalisation correlates should be present in larger domain-final feet: In other words, that IP-final GRs should be more strongly articulated than PWord or footfinal GRs. This sort of effect is a manifestation of Prosodic Strengthening, which is discussed in more detail in Chapters 7 and 8. More research into the interaction between foot structure, morphology and contrast is required.  63  By positing “foot-final boundary” as the locus of alteration, we could potentially remove the need to specify intervocalic glottalised resonants as ambisyllabic. Harris (2003) argues that reference to foot boundaries in English, Danish and Ibibio more accurately captures data generalisations than positing ambisyllabicity. In those examples, however, it is the initial foot boundary that is referenced, which Harris motivates through a cross-linguistic preference for head prominence maximisation.  73  3.5.3 Phonology summary The sections above have presented evidence that St’át’imcets phonology makes a distinction between footed and unfooted unstressed syllables.64 This distinction is motivated by two phonological processes that target unstressed syllables but not extrapods. An alternative analysis might be to argue that the extrapod is, in fact, a degenerate foot, which would simplify the domain generalisation in both processes. The next section lays out the additional structures available under the current model and evaluates them against St’át’imcets phonological data.  3.6  Possible Models  The analysis above has argued that St’át’imcets has two distinct types of unstressed syllables. It was argued that the extrapod analysis was the best way to account for this data without presenting any alternatives. The model as it stands can make use of the following three alternative options to account for the sort of data seen here: 1) Degenerate Feet, 2) Ternary Feet, or 3) Recursive Feet. Each of these makes testable phonological predictions about the nature of the constituents involved. This section presents these predictions and evaluates them against St’át’imcets data to confirm that extrapods are, in fact, the optimal model. Given that the term extrapod is not neutral, the syllable in question is referred to throughout this section as ‘other’. The predictions made by each of the four models are categorised under two headings that are relevant to the foot status of the ‘other’ syllable: 1) process application and 2) head maximisation, as explained below: 1) Application of phonological processes The predictions of the models with respect to how phonological processes at the Footlevel, PWord-level, and PPhrase-level apply to ‘other’ syllables are evaluated in this section. In particular, resonant deglottalisation (foot-level) stress (word-level prominence), and intonational effects (phrase-level prominence) will be examined.  64  Thank you to Matthew Gordon for pointing out the relevance of Vaysman (2009), which discusses metrically-condidtioned segment alternations. Unfortunately, I saw this paper at the point of filing, but will consider its relevance in the future.  74  2) Head maximisation: The data in 3.5.1 were argued to be best analysed as motivated by a foot-internal head maximisation process. Based on the Dutch and St’át’imcets data, we might expect to see more general reduction in non-head position but no reduction in positions outside the foot. The general push within feet for the head to be more salient than the non-head is achievable in one of two ways: maximising the acoustic saliency of head syllable, or reducing that of the non-head (or both). Based on Beckman (1998) we might also expect maintenance of markedness in prosodically privileged positions (i.e. in heads). The next section will lay out the structure and predictions of all of the possible models considered here. 3.6.1 Extrapod (a.k.a. Weak Layering (Itô & Mester 1992)) Extrapods are structurally represented as below: ( (=Foot level boundary, [ = PWord level boundary, V= stressed vowel, v=unstressed footed vowel, v=extrapod vowel)  (42)  PWord / \ F \ /\ \ [(CVCv)Cv]  In terms of processes, foot-level (and foot-final) process should not apply to the extrapod since it is not part of a foot. However, since the extrapod is directly parsed to the PWord, word-level (including word-final) processes should apply. In terms of intonational effects, neither word-level nor phrase-level prominence should occur on the extrapod. Based on the hypothesis that stress and phrasal accent both occur only on heads, neither can associate with the nucleus of the extrapod. In terms of head maximisation, we would predict that there would be no motivation for reduction or maximisation in the extrapod vowel, as it neither in the head nor the non-head of the foot.  75  3.6.2 Degenerate foot The second possible analysis of the ‘other’ syllable is that it is a degenerate foot: parsed at the foot level but forms a foot on its own. This is, in fact, the solution alluded to by Van Eijk (1997) and Davis (forthcoming) in their descriptions of the phonological process mentioned above. In referring to the syllable as ‘potentially’ stressed, they are implying that the syllable is head-like. ( (=Syllable [=foot V=secondary stress, v=unstressed V=primary stress):  (43)  PWd / \ F F /\ | [(CVCv)(CV)]  In terms of processes, foot-level phonological processes that apply to heads of feet or foot boundaries should apply to the ‘other’ syllable.65 The ‘other’ syllable should be available as the locus for both word-level and phrase-level prominence. Since the syllable in a degenerate foot is by definition the head of that foot, we would predict that as a priveleged position, contrasts should be maintained and there should be no reduction. 3.6.3 Ternary foot A third analysis is the one available under Nespor and Vogel’s (1986) assertion that branching is n-ary, namely a ternary foot: (44)  PWd | Ft / | \ [(CVCvCv)]  65  This prediction follows under the standard assumption that prosodic feet must have heads. If the ‘other’ syllable is parsed as a degenerate foot, the vowel must be the head of the foot. In languages with stress patterns like St’át’imcets, Hayes (1995) distinguishes degenerate feet from unparsed syllables by whether the syllable in question attracts main stress or not. It is possible, due to a Non-finality constraint, that the ‘other’ syllable will not bear main stress, but it is predicted to carry secondary stress in this case.  76  If the ‘other’ syllable is parsed as the third syllable of a ternary foot, foot-level processes should apply to both the unstressed syllable and ‘other’ syllable. It also means that footfinal and PWord-final processes should apply equally to final syllables in disyllabic words and trisyllabic words but not the middle syllable in trisyllabic words. In terms of intonation, neither the second nor the third syllable is a head, so neither should host stress or phrasal accent. In terms of head maximisation, depending on how the domains are defined, two different predictions are made. The ‘other’ syllable is not in a privileged position, so there is no predicted markedness maintenance. In terms of reduction, if the domain is defined as –DTE of foot, then both the vowel in the second and third syllables should be reduced. However, if we define the process in terms of foot margins, we can make a distinction between the second (non-foot margin) and third vowel (foot-margin). 3.6.4 Recursive feet The final alternative analysis is recursive feet. Since Ladd (1986) showed that NONRECURSIVITY must be violable, the option of Recursive Feet is also available to us: (45)  PWd | F /\ F \ /\ \ [((CVCv)Cv)]  In terms of process application, foot-level including foot-final processes should affect both the unstressed middle vowel and the final vowel since both are foot-final or –DTEs. Since both are foot final, there is no way to distinguish these two vowels from each other at the foot level. In terms of word and phrase-level intonation processes neither vowel is the head of a foot, so neither should host word-level or phrase-level prominence. Since  77  both vowels are in –DTE position, any pressure to reduce syllable margins should apply equally to both66. As we can see when we consider the models above, each makes different predictions in terms of headedness and boundaries. The next section lays out how these predictions can account for the St’át’imcets data.  3.7  Applying Phonological Predictions to St’át’imcets  Now that the predictions of the four models have been laid out, we can apply them to the St’át’imcets data. The diagnostic for a foot-level process is resonant glottalisation, for word-level process is stress assignment and for phrase-level process is intonational contours. In terms of head maximisation, we consider the vowel-alternating suffixes seen above. 3.7.1 Foot-level processes: Glottal alternation The example of foot-level processes considered here is the glottal alternation discussed in Section 3.5.2. Recall from above that the underlying glottalised resonant in the suffix /-min’/ surfaces as glottalised following unstressed vowels in even numbered syllables and as non-glottalised elsewhere. Table 3.10 demonstrates how each model accounts for the process. Table 3.10 How the models account for resonant glottalisation [=word, (=foot, σ=syllable  Extrapod Degenerate foot  [σ-min’] √ √  Ternary foot Recursive foot  √ √  [σσ-min] √ √ as long as process is defined as  [σσ-mínσ] √ √  targeting heads  x [(σσmin’)] x [(σσ)min’)]  √ √  Both the extrapod and the degenerate foot analysis can account for the data, under different definitions of the process. For extrapods, the context of alternation focuses on retention of [CG] in foot-final resonants. For the degenerate foot analysis to work, the process must apply to the foot. head This would provide a simple, easy solution. 66  Hewitt’s (1992) MaxMinWord does not make the same predictions as the Recursive foot model here. Hewitt’s MaxMinWord more closely resembles our extrapod analysis.  78  Neither the ternary, nor the recursive foot analysis can correctly predict deglottalisation in the final syllables. In both cases, when glottalisation is defined as a foot-final process, it must apply to the ‘other’ syllable. Even if we argue for deglottalisation in the head of the foot, these structures would not predict the correct surface form since the ‘other’ syllable is not a head in these models. 3.7.2 Word-level processes: Stress assignment This section evaluates how the predictions of the various models account for St’át’imcets word-level prominence (stress) assignment. First, let us review the generalisations about the St’át’imcets stress system: • • • •  Feet are trochaic, binary and assigned from left to right Prosodic words are right-headed Schwas and coda consonants are non-nuclear moraic The PWord head (primary stress) does not occur on the final mora  The table below evaluates how the predictions of each of the models account for St’át’imcets word-level stress assignment. Table 3.11 How the models account for stress assignment [(æù¬u˜)(túmu¬)kax∑] Extrapod [(æù¬u˜)(túmu¬)kax∑]  Correct √  Degenerate foot  [(æù¬u˜)(tùmu¬)(káx∑)]  Ternary foot  [(æù¬u˜)(túmu¬kax∑)]  x √  Recursive foot  [(æù¬u˜)((túmu¬)kax∑)]  √  The table above shows that only the degenerate foot analysis cannot predict the correct surface stress pattern. Monosyllabic words in the language are forced to be degenerate feet and carry main stress, as in the case of: [(nk’yáp)] ‘coyote’ from Figure 2.2.67 It is possible that a high-ranking NON-FINALITY constraint prohibits primary stress from ocurring on the final syllable.68 Such an anlaysis predicts that the final syllable will have secondary stress: [(æù¬u˜)(túmu¬)(kàx∑)]. No previous descriptions of the language have reported secondary stress on the final syllable (van Eijk 1997, Roberts 67  Some lexical suffixes also surface with stress in word-final position: e.g. qwenuxw-álhcw ‘hospital’ (lit. sick place) (Davis, forthcoming Chapt. 46). 68 Thank you to Pat Shaw and Matthew Gordon for pointing out this possible alternative.  79  1993, Roberts and Shaw 1994, Davis in press). However, the possibility is tested acoustically in Chapter 5. Since the word-final syllable does not carry main stress, we can conclude that it is not a degenerate foot. Extrapods, ternary and recursive feet predict the correct stress pattern. 3.7.3 Phrase-level processes: Phrasal accent As we saw in Chapter 2, St’át’imcets phrases have phrasal accent (increased pitch, in some cases, the highest in the phrase) on the rightmost stressed syllable and a low boundary tone at the end of the phrase. Based on these generalisations, the diagnostic for phrase-level processes are the location of the H* accent and L% boundary tone. The pitch tracking in the figure below demonstrates the general pattern of the pitch contour on a phrase that ends in a multi-syllabic word. The penultimate /a/ in [(àts’xe)(nána)] is the rightmost stressed vowel in the phrase: Figure 3.1 “Oh, nilh sMichelle ta ats’xenána.” (‘Oh, it’s Michelle that I see’) Elicited as the answer to the question “Who do you see?” Phrasal accent marked with H. Low boundary tone marked with L#. The sentences in this and subsequent figures are written in the orthography.  As we can see from the diagram above, a pitch peak occurs on the penultimate /a/ [àts’xenána] (with the peak also overlapping on to the resonant)—the rightmost stressed vowel in the phrase. The drop in pitch associated with the low boundary tone is present at  80  the end of the phrase on the final vowel. If we follow the hypothesis that phrasal accent can only occur on the heads of feet, then we can assume that the presence of phrasal accent indicates the head of the foot. Support for this assumption comes from phrase-final monosyllabic nouns. These degenerate feet host the H phrasal accent, as evident in the phrase-final /sqláw’/ in Figure 3.2: Figure 3.2 “Oooh, cw7aoz, áts’xenas i a7en’wása sqlaw’.” (‘Oh, no, he saw two beavers’) Elicited as a response the question “Did he see four beavers?”  The figure above shows that monosyllabic degenerate foot words carry the H* phrasal accent. The question then is, do extrapods pattern like the final degenerate foot above, or more like the final unstressed vowels in Figure 3.1? If we consider figure below, which illustrates a phrase-final ‘other’ syllable, we can see that the pattern is more similar to that in Figure 3.1 than in Figure 3.2: we do not see a pitch peak on the final syllable of [(áts’xe)nas]:  81  Figure 3.3 “Iy, á7en’was i tsíkena áts’xenas.” (‘Yes, he saw two chickens.’) Elicited as a response to “Did he see two chickens?”  In the above phrase, we see no pitch peak on the final /a/ in the ‘other’ syllable, unlike on the degenerate foot in Figure 3.2. The final syllable does, however, show the low boundary tone, which was associated with the final unstressed vowel in Figure 3.1 This provides evidence that ‘other’ syllables are NOT heads of feet. The following table summarises how the models account for this data: Table 3.12 How the models account for phrase-level intonation contours H L Correct | | (CVCv)Cv √ Extrapod H L | | (CVCv)Cv Degenerate foot H x | (CVCv)(CV) √ Ternary foot H L | | (CVCvCv) √ Recursive foot H L | | ((CVCv)Cv)  82  The table above shows that, as in Section. 3.7.2, only the degenerate foot proposal fails to account for the intonational pattern. As we saw in Figure 3.1 the representation for a forced degenerate foot such as [(sqlaw’)] is: H | (CCCVR) Given that the correct intonational pattern observed is not this one, the degenerate foot analysis fails. 3.7.4 Head Maximisation: Vowel reduction This section demonstrates how the models account for the notion of head maximisation. In St’át’imcets, the process to which we apply the predictions is the vowel reduction seen in Section 3.5.1. Recall that in middle/transitive suffixes a schwa surfaces in the –DTE of a foot while the full vowel surfaces elsewhere (i.e. DTE of the foot, and in ‘other’ syllables). In order to account for this reduction in the weak part of a foot, the models must distinguish unstressed syllables from ‘other’ syllables. Table 3.13 How the models account for non-head reduction [(σÅn)] [(σσ)an] √ √ Extrapod √ √ Degenerate foot √ Ternary foot x predicts: [(σσÅn)] √ Recursive foot x predicts [((σσ)Ån)]  [(σσ)(ánσ)] √ √ √ √  In the case of non-head vowel reduction, both extrapods and degenerate feet make the correct prediction: in neither case is the ‘other’ vowel the non-head of a foot, so it is not reduced. The ternary foot model, on the other hand, predicts that only the stressed vowel should surface as full. Since both the unstressed and ‘other’ syllables are –DTEs of the foot, the motivation for reduction applies equally to both. The Recursive Foot model also predicts reduction since both vowels are in –DTE position. Even if we were to assume that the vowel reduction alternation is not foot dependent, how the ternary and recursive foot model could account for the alternation is unclear. How would these analyses distinguish between reduction in unstressed vowels  83  in even numbered syllables but not odd numbered syllables? Indeed, they cannot, and therefore a foot-based analysis is simpler and more elegant. Table 3.14 summarises how well the various models succeed in accounting for the phonological data. Table 3.14 Summary of predictions Model Extrapod PWord / \ F \ /\ \ [(CVCv)Cv] Degenerate PWd foot / \ F F /\ | [(CVCv)(CV)] Ternary foot PWd | Ft / | \ [(CVCvCv)] Recursive PWd feet | F /\ F \ /\ \ [((CVCv)Cv)]  1. 2. 3. 4.  Predictions Foot-level processes Word-level processes Phrase-level processes Positional privilege  Success?  1. 2. 3. 4.  Foot-level processes Word-level processes Phrase-level processes Positional privilege  1. 2. 3. 4.  Foot-level processes Word-level processes Phrase-level processes Positional privilege  x  1. 2. 3. 4.  Foot level processes Word-level processes Phrase-level processes Positional privilege  x  √ √ √ √  √  x x √  √ √  x  √ √  x  In short, only the extrapod analysis can account for all of the St’át’imcets phonological data presented here. Having established that the extrapod analysis is the correct one for St’át’imcets and that extrapods and unstressed vowels are phonologically distinct, we can discuss them in terms of the characteristics of extrapods presented in Table 3.2. Extrapods in St’át’imcets demonstrate the same general characteristics as the nonexhaustively parsed elements discussed in Table 3.2: they are marked, present in a case of parsing conflict, smaller than a foot, located at the periphery, visible to word- and phraselevel but not foot level processes and are not heads.  84  First, in terms of markedness, extrapods in St’át’imcets are subject to languagespecific repair strategies. Davis (forthcoming) notes that speakers in fluent speech will insert an epenthetic schwa at the end of an extrapod-final word, which means the extrapod is refooted (into head position of a foot). Second, as we saw in Section 3.7.2, extrapods occur as a result of a conflict between parsing and foot-binarity. Third, in terms of size, St’át’imcets extrapods permit heavy syllables, but only to a point. As mentioned above, a four consonant cluster outside a root acts as moraic for some speakers. For these speakers, a four-consonant cluster in an extrapod would presumably not be permitted, since the extrapod would then be a binary foot. Fourth, extrapods occur at peripheries. In the cases discussed here, extrapods occur word-finally but they also occur initially at the PWord level ([pu(láka7)] Lower dialect ‘drum’) and PPhrase level, in the form of determiner proclitics (e.g. ti/ta, ku), the proclitic complementiser (i=) and prefixes (e.g. kens-).69 Fifth, extrapods are not visible to foot-level processes, or to processes which target heads (e.g. phrasal accent). They are available for boundary tones in St’át’imcets since tones are assigned at the PPhrase-level in the language. Finally, extrapods are not degenerate heads. They do not host stress or phrasal accent. Following the assumption that phrase-level heads can only occur on lower-level heads, the failure of extrapods to host phrase-level prominence is due to the fact they are not heads. The only phonotactic restrictions on extrapods are those that apply to all syllables in the language, and because final extrapods are invariably suffixes or enclitics, those that apply to suffixes/enclitics, namely reduced consonant inventory and no underlyingly retracted vowels (van Eijk 1997). Taking the St’át’imcets extrapods into account, we can form a more fully developed picture of extrapods, outlining those attributes that are universal (i.e. a requirement of the model) and those that are language-specific. There are no modeldriven requirements that extrapods be a certain shape or restricted to certain phonemes. 69  This is following Selkirk (1995b), who proposes that function words in English are clitics, which are parsed directly to the PPhrase level. Experiment 2 originally examined the vowel quality of Pword- and Pphrase-initial extrapods. Results are located in Appendices E and F.  85  While more cross-linguistic research needs to be conducted, our starting points are summarised in the table below: Table 3.15 Summary of extrapod attributes Universal (required by model) 1) Marked: Subject to language-specific repair strategies (due to PARSE-SYLL violations) 2) Availability: Only in cases where a parsing conflict occurs, and where repair strategies are not available 3) Size: Must constitute less than a foot (as defined by language-specific foot requirements) in iterative footing languages, otherwise FT BIN and PARSESYLL would be satisfied  Language Specific (shape, phonotactics) 1) None: St’át’imcets 2) Foot structure driven: Japanese permits mono-moraic only (due to a bimoraic FT BIN constraint). 3) Sonority driven: Dutch permits no low vowels. 4) Phonotactics: English has generally post-consonant sonorants or affixal nonsyllabic /–y/.  4) Location: external. Extrapods cannot break up a foot. They must occur external to feet, but are not necessarily peripheral at word-/phrase-level. 5) Visibility: by definition, they are not footed, and so are not subject to processes that affect feet. - Are visible to PWord and PPhrase-level processes: e.g. boundary tones in languages in which boundary tones are assigned at word-level or above 6) Constituent Status: Extrapods are not degenerate feet. They are not heads, and so are not available as anchors for stress or phrase-level prominence - Are distinct from footed unstressed syllables, but cannot be independently referenced  With the phonological attributes of extrapods in hand, let us now consider the acoustic predictions based on the model above.  86  3.8  Phonetic predictions  Since non-exhaustively parsed syllables such as extrapods are permitted by the model, and given the converging phonological evidence presented in Sections 3.5.1 and 3.5.2 that unstressed syllables and extrapods are distinct phonologically, the prediction is that speakers will distinguish them acoustically. This prediction is based on the set of assumptions about the phonetics-phonology mapping laid out in Section 1.6. If: a) prosodic constituents are the domains of phonological processes; b) prosodic constituents are distinguished phonetically; and c) there is a non-random mapping between phonetics and phonology, then two syllables that are shown to be in different phonological domains should be distinguished phonetically by speakers. The predictions about how unstressed syllables and extrapods will differ phonetically are guided by Pierrehumbert’s (1999) definition of prosody as the ‘...grouping and relative prominence of the elements making up the speech signal.” In the two experiments presented in the following chapters, I examine the acoustic correlates of relative prominence (stress) and boundary effects (Prosodic Strengthening). While there has been much research on the acoustic nature of prominence (i.e. head versus non-head), no research to my knowledge has explored the acoustic characteristics of differently parsed unstressed syllables. As such, while there have been various proposals about the relative prominence of unstressed syllables and extrapods, there has been no converging phonetic evidence to support these hypotheses. Nor has Prosodic Strengthening research considered non-exhaustively parsed domains. This section lays out the predictions that have been made but not tested by previous research, as well the four possible hypotheses available under the model. Predictions about the acoustic nature of extrapods are limited to untested phonologically-based relative prominence claims and generally fall into two major categories: partial and full specification. Nespor and Vogel’s (1987) analysis could be defined as partial specification. Recall that extrapods must be parsed as part of a ternary syllable or as a weak foot (i.e. degenerate foot) under this analysis. Under ternary syllable parsing, weak syllables within a foot are indistinct from one another. Apart from the one required strong syllable, “[a]ll other syllables are weak, and there are no further predictions about relative 87  prominence relations among them” (1987:86). Under weak foot parsing, they argue that only the head of the strong foot is perceived as stressed. In contrast, full specification is proposed by both Liberman and Prince (1977) and Beckman and Pierrehumbert (1988). In both of these models, relative prominence is structurally assigned. Liberman and Prince (1997), while admitting that they know of “little evidence that bears on the details of this adjunction” propose that the stray syllable [extrapod] is adjoined as “weak sister to the nearest maximal left foot” (1977: 294). According to them, “[t]here is nothing to prevent further differentiation among the sequence of “weak” elements, but neither is there anything to require it” (1977:324). They note that, in the case of ‘execute’ (presented as example 3.2), “it may seem odd not to make the terminal syllable [extrapod], which is [+stress] metrically stronger than the preceding unstressed syllable” and that “[s]ome people feel that the last element in the sequence may be stronger than the other weak elements.” In both cases, they dismiss these concerns due to the fact that previous theories and poets do not differentiate between the unstressed syllables; so, in effect, neither will they. Relative prominence between non-sister nodes can be seen as ‘extra-structural’: a value is assigned to weak syllables independently of structure. While still proposing full specification, Pierrehumbert and Beckman (1988) make a distinction between Liberman and Princes’ discrete relations and their own gradient relations, based on phrase-level tonal saliency. For Pierrehumbert and Beckman (1988), weak nodes carry a percentage value defining their subordination in relation to the head node, NOT their strong sister. As a result, only weak syllables under the same head can be compared. In both cases, there is no structural requirement for specific prominence relations and individual speakers determine the ranking. This predicts that the acoustic interpretation of relative prominence between extrapods and unstressed syllables is not fixed but variable. A further prediction is that the acoustic characteristics of extrapods will be determined by other factors. How individual speakers weight potentially conflicting demands such as head maximisation (both the non-head minimisation presented in this chapter and head constituent maximisation, as in the FSN strategy discussed below), boundary prominence maximisation and utterance position effects such  88  as final lengthening, will result in variation in the contrast between unstressed vowels and extrapods. The research presented above suggests two hypotheses regarding the acoustic properties of extrapods. Under the first hypothesis, extrapods are non-distinct from unstressed syllables. Under the second hypothesis extrapods are distinct but there is no consistent direction of difference. The Dutch and St’át’imcets vowel reduction data present us with a third hypothesis: that extrapods will be acoustically stronger than unstressed syllables, since extrapods are not subject to foot-internal head maximisation constraints. The final hypothesis considered completes the paradigm: unstressed syllables will be acoustically stronger than extrapods. Motivation for this hypothesis comes from Prosodic Strengthening research. Segments at larger (or higher) domain boundaries of the Prosodic Hierarchy have been found to be articulated more strongly than those at smaller or lower boundaries in several languages: English (Fougeron & Keating 1997; Cho 2005), French (Tabain 2003a,b); Taiwanese (Keating et al. 2003); Korean (Cho 2002); and Dutch (Cho & McQueen 2005). While Prosodic Strengthening research has not considered non-exhaustively parsed domains, a ‘finely cumulative’ interpretation would predict that that the more boundaries a segment is adjacent to, the stronger it is. In other words, unstressed syllables are more prominent than extrapods due to their being parsed into an additional constituent. This hypothesis is referred to as Footed-is-Stronger-thanNot (FSN). To summarise, Table 3.13 presents the four hypotheses tested in the following chapters along with the acoustic predictions they make in terms of prominence correlates and boundary strength.  89  Table 3.16 Summary of acoustic predictions (us = unstressed, ex = extrapod)  Name  Hypothesis  Acoustic prediction  Traditional  us =ex  No prominence correlate or boundary distinctions -All unstressed syllables are alike and non-distinct.  Variable  us ~ ex  Maximise Prominence (MaxProm)  ex > us  Extrapods and unstressed syllables are distinct in terms of prominence correlates and boundary effects, but the difference is not predictable or consistent. -Other factors motivate phonetic distinctions. Extrapods have stronger prominence correlates and boundary effects than unstressed syllables as a result of not being subject to foot-internal head maximisation constraints.  Footed Stronger than Not (FSN)  us > ex  Unstressed vowels have stronger prominence correlates and boundary effects than extrapods as result of being properly parsed.  Chapters 5-8 test these predictions in St’át’imcets. Results from an experiment testing prominence correlate predictions are presented in Chapters 5 and 6. Boundary strength results are presented in Chapters 7 and 8.  3.9  Conclusion  This chapter discusses the history of the Prosodic Hierarchy and the evolution of the Strict Layer Hypothesis. It introduces the term ‘extrapod’ to refer to syllables that violate EXHAUSTIVITY at the foot level and presents the phonological attributes and acoustic predictions of such a model. It was argued that St’át’imcets is a language that contains extrapods, thus setting the groundwork for acoustic experiments on the language reported in subsequent chapters. While extrapods are not predicted by the theory as ‘full fee paying members’, their existence is difficult to deny. By being defined as marked violations of the Strict Layer Hypothesis, they are accounted for in a coherent manner. This predicts that extrapods should be phonologically marked relative to exhaustively parsed constituents, which this chapter has shown to be the case. However, no converging acoustic evidence has been presented. After the general methodology is presented in Chapter 4, the  90  remaining chapters of this thesis examine St’át’imcets extrapods for phonetic evidence that substantiates the structures motivated on phonological grounds. Chapter 5 examines the pitch, duration and intensity of stressed, unstressed and extrapod vowels to see how stress is cued in the language. Chapter 6 compares prominence correlates to determine whether and how extrapods are distinct from footed unstressed vowels. Chapter 7 examines the formant structure of domain-final vowels to determine if Prosodic Strengthening effects reported for other languages obtain in St’át’imcets as well. Finally, Chapter 8 compares the boundary strength effects of extrapod and footed unstressed vowels to determine whether and how they are acoustically distinct.  91  Chapter 4 General Methodology and Protocol Development  4.1  Introduction  This chapter introduces the speakers who participated in the two experiments reported in this thesis and discusses the general methodology and protocol development. One of the goals of this thesis is to develop protocols that elicit natural speech (as opposed to lab speech) that can be easily adapted to language teaching materials. The successes and failures of this methodology are discussed in Section 4.5.3, following the general methodology discussion, which addresses the rationale behind the development of the protocol developed here.  4.2  Experimental choices  This thesis focuses on production rather than perception. Practically speaking, the dearth of research on prosody in St’át’imcets and the small number of speakers mean that a production study is the more suitable experiment type at this point. A perception study that makes use of the correlates tested here would provide a better understanding of how hearers judge prominence correlates and how they distinguish unstressed vowels from extrapods. The results of the production study undertaken lay the groundwork for both perception studies of this language and production studies in others. 4.2.1 Modality The experiments in this study focus on acoustic rather than articulatory distinctions. Much of the phonetic research on prosody cross-linguistically and most in the Salish literature is acoustic. In order to be comparable to previous research, this study focuses on acoustics rather than articulation. Significant patterns in acoustic contrasts could act  92  as a starting point for future articulatory research, such as ultrasound, to examine tongue movement as a correlate to boundary strength. 4.2.2 Specific correlates The choice of acoustic correlates tested was based on Pierrehumbert’s (1999) definition of prosody as the grouping and prominence of the sound system. Three traditional acoustic correlates of prominence in English were chosen (F0, duration, intensity) based on the wide body of previous research conducted on stress. Other acoustic correlates could also be examined, including vowel-consonant coarticulation (Tabain 2003a), and aspiration or glottal parameters (Sluijter & van Heuven 1996a,b). Vowel quality, (i.e. position on F1/F2 plane), is chosen as the acoustic correlate to boundary strength based on Cho’s (2005) research on English. Although this phonetic research has mainly been conducted on English, it provides a frame of reference for the examination of similar effects in St’át’imcets.  4.3  Information on Speakers  The six speakers that participated in one or both of the experiments reported in this thesis are fluent first language speakers of St’át’imcets. Of the six speakers, four are women and two are men. Four speak the Northern dialect of St’át’imcets while two speak the Southern dialect. All speakers are also fluent in English. More detailed information about the six speakers is given below. AP is a 64-year-old fluent speaker of the Northern dialect. She is the sister of CA (below). Both grew up in Tsal’álh and both of their parents were Northern dialect speakers. AP currently speaks St’át’imcets when she teaches it three times a week and when she sees elders (on average three times a week). She has upper dentures, and no bottom teeth. CA is a 67-year-old male fluent speaker of the Northern dialect and AP’s brother. He uses the language every day in his role as a spiritual leader in the community. He has top and bottom dentures.  93  CS is an 85-year-old female fluent speaker of the Northern dialect. She grew up in Lillooet and both parents were Northern dialect speakers. She works with the Language Authority and speaks with elders when she sees them in town. She, too, has top and bottom dentures. HD is a 70-year-old male fluent speaker of the Lower Southern dialect. He was born at Seabird Island but lived until age 15 in Sqátin (Skookumchuck). His mother was a Halkomelem speaker. He currently uses the language in weekly language classes and when speaking to other elders approximately twice a year when he returns home. He has no teeth at all. He also has a swelling on his lower lip that he has had since 1958. LT is a 77-year-old female fluent speaker of the Southern dialect. She lived and spoke the language daily in Mt. Currie until age 25. She continued to use the language to speak to her mother, another Southern dialect speaker, until her mother passed away. She currently uses the language at two weekly conversation classes and has both top and bottom dentures. RW is a 77-year-old female fluent speaker of the Northern dialect. Until recently, RW spoke to her mother in the language and is very involved with the local Language Authority. She has top and bottom dentures.  4.4  Token Selection  Tokens in both experiments were selected to get as close to minimal pairs as possible, which proved challenging in a language with 44 consonants. To minimise differences and avoid the effects of coarticulation or pitch perturbations, no tokens with preceding or following ejectives, glottalised resonants, or uvulars/gutturals/pharyngeals/emphatics were included. Crucially, unstressed vowel and extrapod comparisons for both experiments were based on vowels in identical morphemes and situated in the same morpho-syntactic and intonation contexts. As a result, any significant differences between the two can only be attributable to a difference in parsing.  94  4.5  Measurement and Analysis  The vowels in both experiments were generally segmented according to Peterson & Lehiste (1967:192-196). For vowels preceding glottal stop, the offset was measured as the last regular pitch pulse, before dropping off below 75 Hz.70 The vowels in experiment 1 were coded for vowel type and stress type. The vowels in experiment 2 were coded for vowel type, syllable, boundary and position. Each token was visually inspected to examine the pitch, intensity or formant tracing. The analysis of the formants of the vowels proved unexpectedly challenging. Speakers were analysed separately, taking into account speaker differences. It was also necessary to measure vowel types separately, as the formant tracker in Praat could not consistently measure different vowels at the same settings accurately. As a result, the following speaker-specific formant analyses were performed: Table 4.1 Formant analysis by speaker and vowel (LPC order/max Hz. N/A indicates vowels were not collected from that speaker.)  Speaker AP CA HD LT  /a/ 8/3500Hz 6/4000 Hz N/A N/A  /i/ N/A N/A 6/3500 10/4000  /u/ 8/4000 8 /4000 4/3500 9/4500  /ә/ 6/3500 6/3500 6/4000 7/4000  I first sought to follow Harrison’s (2004) results, which found that Praat made the most accurate and consistent F1 and F2 measurements when 5 formant measurements (LPC order 10) were used. However, this setting returned unreliable and inconsistent results. Measurements made with the adjusted settings produced less variability (i.e. were more consistent) than ones performed at the default settings. Praat scripts were executed to obtain measurements in both experiments, and a selection of results were cross-checked with the on-screen tracings to ensure reliability. In order to control for different inherent vowel baselines, no across-vowel comparisons were made. Vowels in English have been shown to have inherent differences in F0 and intensity (Fry 1955, 1958; Lehiste & Peterson 1959; Lieberman 1960; House 1961; Crystal & House 1990). Caldecott (2006a) showed that St’át’imcets 70  Vowel-initial roots in the language are, in fact, glottal stop initial (van Eijk 1997; Matthewson 2005). As such, there were cases where the vowel in question preceded a following glottal-stop initial word.  95  vowels show similar results. The vowel /a/ had significantly higher intensity and/or significantly lower pitch than /i/ in similar segmental contexts. For one speaker, it was also significantly longer. Given the small number of speakers, no across-speaker comparisons were made either. In terms of statistical analyses, multiple t-tests were used in both experiments. Ttests were selected over ANOVAs for two reasons: i) to eliminate (some) Type I error arising under an ANOVA analysis, and ii) to be comparable to previous research, namely Koch (2008a). First, ANOVAs include all data points for every calculation of a given variable, even when the comparison of a subset is the goal. The data points considered here all derive from the same speaker, and in some cases, the same utterances. As such, an ANOVA runs the risk of finding significance where there is none; that is, it “is likely to underestimate the true variability of the results, leading to tests that are biased towards rejection of null hypotheses” (Keppel & Wickens 2004: 142-3, citing Scariano & Davenport 1987). The use of t-tests enabled selecting no more than one data point per utterance for a number of calculations, thus reducing violations of the assumption of independence. One of the disadvantages of using multiple t-tests is the increase in familywise errors. This issue was offset by using a Bonferroni correction and is discussed in more detail in Chapters 5 and 7.71 The second motivation for using t-tests is in order for the results to be comparable to Koch (2008a). Koch (2008a) also opts for t-tests over ANOVAs to avoid increased violations of the assumption of independence. Given the dearth of acoustic studies of this nature in Salish languages, using similar methodology is desirable. It means that the results of these experiments are comparable with the only other major acoustic study of intonation in Salish languages.  4.6  Protocol Development  As mentioned in Chapters 1 and 2, there has been little research conducted on the higher order prosody of First Nations languages. For the researcher, this means a paucity of both information and research methodology.72 One of the goals of this thesis was to  71 72  My thanks to Karsten Koch for much helpful discussion on this section. See Caldecott and Koch (2007), Koch & Caldecott (2007), and Koch (2008a,b) for further discussion.  96  develop protocols that would be rigorous enough to collect data suitable for phonetic analysis but that would be appropriate for working with First Nations elders. Another goal was to elicit these data in such a way that would preserve the ‘naturalness’ of the speech recorded as much as possible. Recording data for phonetic experiments requires numerous repetitions in highly controlled contexts. Most acoustic studies similar to this one are conducted by recording speakers reading sentences from paper or a computer screen. In addition to controlling the speakers’ output, a reading task avoids any intonation influence by the elicitor. The question of influence from English was a concern during the development of these protocols since in most cases the elicitor was a person with a limited fluency in St’át’imcets. Having speakers read would eliminate any influence that a non-fluent speaking partner might have. Traditional data collection in the form of reading tasks provides challenges for fieldworkers working with an elderly, limited, or remote population. As these are older speakers, deteriorating vision means reading might be difficult. Furthermore, having elderly speakers read sentences for any period of time is tiring and becomes uninteresting quickly. It is important to remember that we rely on the patience and generosity of busy and/or potentially ailing older people, and we do not want to unnecessarily bore or tire them. In addition, since many languages have only recently been orthographised, some of the older speakers did not learn, or choose not to use the orthography. As a result, the protocol developed had to: i) not be reliant on reading, ii) be entertaining, iii) take into account the stamina of elders, but iv) still encourage speakers to repeat sentences many times in as natural a way as possible. Defining natural speech is complicated and can refer to syntactic, semantic, intonational and/or articulatory characteristics. In terms of phonetics, this is often categorised as the difference between ‘lab speech’, which is usually produced in the lab by reading repetitions of sentences, and more natural speech produced in conversation. Some advantages and disadvantages of eliciting lab speech through a reading task are mentioned above. Other advantages of lab speech are that they allow the researcher to control the data speakers produce in terms of segmental and sentential context and that it produces a large number of tokens in a controlled acoustic environment.  97  One way in which experimenters control segment and/or intonation context is to use nonsense words or nonsense sentences in ‘out-of-the-blue’ contexts. For example, in Cho (2005), sentences like “No, Little Bah bopped the girl” were used to elicit /a/ in strictly bilabial contexts. Another form of eliciting out-of-the-blue contexts is straight translation tasks, in which speakers translate English sentences repeatedly. There are major issues with using nonsense words and out-of-the-blue contexts. First, nonsense sentences or words, while perfectly acceptable in English, were not produced by the speakers I have had the privilege of working with. They refuse to produce nonsense words point blank and strongly prefer sentences with some context (see Matthewson (2004) on the importance of using context in semantic fieldwork). Second, there is the question of the influence of English intonation. Matthewson (2004) showed the use of a metalanguage (e.g. English) to elicit semantic data in First Nations languages is legitimate and that by presenting a discourse context in a metalanguage, some complications of doing so in the object language are avoided.73 Her claim does not extend to articulation or prosody, which may be affected by a metalanguage. The possible “contamination” from English, along with the likelihood of list effects, means straight translation tasks were not desirable. Finally, to my mind, the greatest weakness of lab speech is the lack of context. Whenever speech is taken out of context, it ceases to have linguistic or communicative value. 74 Lab speech removes the context of how we normally use speech (with a partner, to achieve some goal, sociolinguistic or otherwise) and also our connection with what is being said. While the nature of speech out of context is not well understood, it should be pointed out that recent research examining the differences between traditional elicitation techniques and those designed to record more ‘natural’ speech has shown that there is little difference in acoustic characteristics between them (Koch 2008b). Koch found the effects of elicitation method were minimal in N¬e÷kepmxcin. He compared a number of 73  Matthewson (2004) exemplifies this with a case of determining whether clefts are used naturally in a particular context in St’át’imcets. Using the target language to set up the discourse context forces the elicitor to choose either a cleft or neutral structure in the set-up, potentially influencing the speaker’s output. Matthewson points out that with much thought (and fluency in the language) elicitors could construct target-language examples that avoid these sorts of problems. 74 Thanks to Eric Vatikiotis-Bateson for emphasising this point.  98  acoustic characteristics of speech recorded in three different contexts: spontaneous conversation, scripted conversation and single sentence elicitation. He found no significant differences in speech rate, F0 declination rate, amplitude declination rate, duration, amplitude or F0 peaks between sentences elicited by spontaneous speech and single sentence elicitation. He did, however, find differences in intensity and some marginal differences in F0 peak between scripted data on the one hand and spontaneous speech and single sentence elicitation on the other. This is consistent with the findings listed in Klatt (1976), which found that comparisons of read speech, spontaneous speech and nonsense speech had more similarities than differences. In my research, I have found that speakers give a more natural translation in conjunction with a pictoral context. By this, I mean that they use discourse particles when presented with a picture and asked for a translation that are not present in ‘out-ofthe-blue’ context translations. This supports Matthewson’s (2004) recommendation that a combination of verbal and non-verbal contexts is the ideal for elicitation. The protocols developed here build on Burton (2005) and Koch (2008a,b) which develop multimedia aids for elicitation with First Nations speakers and for curriculum materials. The protocols used in the thesis involved showing the speakers a randomised set of slides in MS PowerPoint (henceforth PPT) to minimise the need for translation and maximise ‘natural’ speech. In addition, a verbal context was supplied. As a result, the methodology employed is a combination of verbal cues and non-verbal cues (pictures on slides) to present a coherent context for the speakers. The main methods used were partly scripted and picture-prompted. Two other methods used for the general intonation research, free conversation (speakers talked about whatever they wished) and task-based elicitation (speakers interacted with me while doing a task) are discussed in Section 4.7. In the partly-scripted sessions, I asked questions in St’át’imcets and the speakers responded spontaneously based on what they saw on PPT slides.75 In the picture-prompting sessions speakers saw slides depicting a limited set of circumstances to prompt them to say the target sentence, with or without an English prompt. 75  Questions were asked using St’át’imcets intonation. While I cannot be certain that my intonation was completely accurate, when asked about the quality of my questions, speakers did not indicate they heard anything unnatural.  99  The slides in both cases contained a picture of an action and an object. In experiment 1, speakers responded to a yes/no question based on the slide. They were asked to answer as fully as they could. In experiment 2, speakers gave the St’át’imcets sentence for what was depicted on the slide. When the speakers’ volunteered forms did not match the target, they were asked if the target was an acceptable form if used in conjunction with the picture, and if yes, they were asked to use that form every time they saw the appropriate slide. If not, either the picture was changed to suit the target or an alternative target sentence was constructed that suited the picture. The number of different slides needed to be limited for two reasons: i) so that speakers in the second experiment could remember which sentence they were being asked to pronounce with limited English prompting, and ii) to limit the amount of drawing by the author.76 However, slides still had to be varied enough to hold the speakers’ attention. To this end, a limited number of verbs with varied objects were used. This allowed the slides to be visually interesting and less taxing on the speakers’ memories. Tokens were situated in target sentences with similar (in the case of unstressed vowel and extrapod comparisons, identical) segmental, morpho-syntactic and intonation effects. Recording sessions began with speakers telling a story in St’át’imcets or a short conversation with me as a warm-up. A training session was also held for each part of the experiment. In the first session, this was about 10-15 slides and diminished for subsequent sessions. Speakers decided for themselves whether the session was self-paced (pressing the enter key on the computer themselves) or not (the author pressed the button). For most speakers, the experiments were conducted in multiple sessions. A portion of each section was recorded in each recording session so that comparisons were balanced. No significant differences (based on t-tests) were found between data recorded on different dates. More specific methodology is given below.  76  This limitation may at first not seem valid. In an ideal elicitation environment, researchers would have unlimited time and ability to draw the 87 different slides required in these experiments. However, in reality, such time and technological constraints are all too pressing. One of the goals was to achieve the maximum flexibility with the minimum input under real world limitations.  100  4.6.1 Experiment 1 This experiment examined how St’át’imcets speakers mark prominence acoustically in the language and is reported in Chapters 5 and 6. Speakers produced the target sentences as responses to yes/no questions asked in St’át’imcets by the author, based on a scene they saw on PPT slides.77 Speakers were shown 25 randomised repetitions of each token. In 3 of the 25 slides, the object shown was not the object in the presented yes/no question. Speakers were asked to respond, as fully as possible, to the question based on what they saw on the screen. They were asked to pay attention to the pictures in case they did not match up with the question. These mismatches served to keep the speakers attention and provided some welcome levity. A sample slide is shown in Figure 4.1 below: Figure 4.1 Slide used to elicit “Iy, kwánenskan ti ts’úqwaz’a.” (Yes, I caught the fish.) In this and subsequent figures, the caption is written in the orthography  77  Specific methodology, including tokens, is reported in Chapter 5.  101  4.6.2 Experiment 2 This experiment examined boundary effects in the language, and how they might be used to characterise extrapods. Aside from the comparisons between unstressed vowels at different boundaries and domain-final extrapods at different boundaries reported in Chapter 7, and the comparison of unstressed vowels versus extrapods reported in Chapter 8, two other sets of comparisons were also carried out. First, St’át’imcets also has PWord-initial (root-initial) extrapods, which would potentially allow for a comparison of extrapods that violate the Strict Layer Hypothesis only once, and those that violate it twice (i.e. determiners). Selkirk (1995b) considers determiners to be directly parsed to the PPhrase level, thus lacking both a foot and word boundary. By comparing determiner proclitics to root-initial extrapods, we should be able to determine whether lack of boundary (or increased violations of the Strict Layer Hypothesis) also results in a cumulative acoustic effect. Second, I compared PWord-initial extrapods to PWord-final extrapods to give us a better understanding of how position affects the acoustic characteristics of these domains and how generalisations about prosodic privilege apply to extrapods. The results of these two sub-experiments, while interesting, are not reported in Chapters 7 and 8. While these comparisons were recorded and analysed, the consonantal contexts proved too dissimilar to allow confidence in the results. They are presented as documentation of an endangered language in Appendices E and F. Experiment 2 was divided into four parts. Speakers were shown the slides and asked for the St’át’imcets sentence describing the picture on the slide. As mentioned above, if the description did not match the target, speakers were asked if the target was appropriate and if so, asked to use it when presented with that slide. If not, the target or the slide was changed. After the initial training session speakers generally required no English prompting for parts I, II and III, though they could always ask for an English prompt if necessary. Due to dialect-specific word-order differences between Northern and Southern speakers, part IV was changed and somewhat simplified for the Northern speakers. As a result, they required little or no English prompting while the Southern speakers required it for every slide. 102  Part I elicited stressed vowels and unstressed foot-internal tokens in a traditional carrier sentence: ‘Tsut sLisa ‘x’ inátcwas’ (Lisa said ‘x’ yesterday). The context provided was that Lisa is a little girl who is learning to talk. The speaker is telling me what words she said yesterday. Speakers remembered the carrier sentence and put in the word of the picture on the screen (token). This traditional method, while not ideal, was chosen because it was the easiest way to get tokens in correct morphological and intonational context78. Part II elicited unstressed foot-final vowels and determiners. A sample slide is given below: Figure 4.2 Slide used to elicit “T’ec ti/ta xúsuma múta7 i q’welápa.” (The xusum and the strawberries are sweet)  Part III elicited initial extrapods and foot-final unstressed syllables that did not fit neatly into the scenario for Part II. A sample slide is given in Figure 4.3:  78  Traditional carrier sentences such as these cause speakers to put a heavy focus on the token, often separating it from the rest of the sentence with pauses and marking it with higher intensity.  103  Figure 4.3 Slide used to elicit “Wá7lhkalh ít’em lhas pipántsek muta7 lhas sútik.” (We sing in the summer and in the winter)  Finally, part IV proved challenging for speakers. The target construction was difficult to remember, and while acceptable for most speakers, it is not the most common word order for Northern speakers.79 To ensure natural word order, the scenario was modified for the Northern speakers. In both cases, speakers instructed me to give something to someone. Southern speakers ordered me to give them or Henry one of my things. For Northern speakers the context was a hold-up: speakers ordered me to give them or Henry one of their (non-visible, but assumed to be possessed) things. Below is a sample slide Northern dialect speakers were given:  79  Word order was crucial here to avoid locating the token in either nuclear accent position or phrase-final position.  104  Figure 4.4 Slide used to elicit “N’áscit ku músmustsu (kw)s Henry.” (Give one of your (non-referential, non-visible) cows to Henry.)  4.6.3 Advantages and disadvantages of the protocols Using the slide-show protocol proved successful in a number of ways. First, once all of the drawings were scanned (or downloaded) into the computer they were alterable for many contexts, and vocabulary could be changed on the fly. This is valuable for fieldworkers who do not have access to the internet, or printers/scanners to produce new materials on location. Second, speakers found the drawings less distracting than photos and were entertained by them. Basic drawings are also easier to modify on the fly than photos. The slides were varied enough to keep the speakers’ attention while still prompting them for the desired response. In terms of downsides, the work is very intensive up front, and it requires electricity and a separate computer or recording device during the session (see also Caldecott & Koch 2007; Koch & Caldecott 2007). Another downside is the difficulty in constructing pictures that accurately represent the target sentences. For example, in the slide in Figure 4.4 the original target sentence was unacceptable for speaker CA. The onsite repair involved a determiner used when the object is not visible or is non-referential. This required coming up with a context and picture that would still prompt the speaker to remember what they had to say while implying that the object was not visible. While the resulting slide is not ideal, it was the best solution possible under the circumstances.  105  4.7  Other Elicitation Methods  The intonational generalisations reported in this thesis were gained through three other types of elicitation: partly scripted conversation, free conversation and task-based elicitation (see Caldecott and Koch (2007) and Koch and Caldecott (2007) for more discussion). The partly scripted conversations used to elicit the data in Chapter 2 involved one speaker reading a set of questions that the other speaker answered by looking at PPT slides. Eliciting with two speakers means the researcher is able to control the questions the speakers ask while allowing spontaneous answers and not introducing any English or non-native speaker elements. One advantage of this method is that it can be adapted to situations when only one speaker is present as in Experiment 1, when the questioner was the author. Free conversation also proved very successful, although limited by the fact that two fluent speakers are required. This method was not used to elicit the data presented in the main experiments in this thesis due to the lack of consistent or sufficient tokens. However, free conversation elicitation did result in a large volume of data that might otherwise not be available. Recall from Chapter 2 that partly scripted conversations showed that yes/no questions in the language had no final pitch rise. It was concluded that St’át’imcets speakers do not use pitch rises to signal questions. However, conversational data revealed that speakers DO use a pitch rise in echo questions. It would have been difficult to find this information using traditional elicitation techniques. To further examine the pitch rise associated with echo questions, a task-based methodology was developed. Task-based elicitation is another way to control the sort of data speakers produce while allowing them to produce it spontaneously. One of the best known sources for linguistic task-based corpora is the HCRC Map Task Corpora (Anderson et al. 1991). The HCRC design requires two speakers with slightly different maps to interact, and for one to guide the other to a set position on the map. This proved too complicated for my elicitations for several reasons: i) my language skills were not good enough to play the  106  second speaker; ii) maps are difficult to draw; iii) how places are represented can be culturally specific; and iv) not everyone is a map reader.80 The goal of my map task was for the speaker to ask echo questions based on the directions given by me. For example, I would say ‘Go left at the church’ in such a way that ‘church’ might be mumbled or difficult to understand (in this type of task, not being a native speaker is an advantage). The speaker would then ask something like ‘At the church?’ or ‘Where?’ I based my design on the HCRC task but with locations that were relevant and fit my drawing and language abilities. My map is given below:  80  Some of these complications could be resolved by having two speakers participate rather than one speaker plus the non-fluent elicitor. While this would be ideal, the reality often is that we are lucky to have one speaker to work with, much less two who feel comfortable working together and who are physically able to leave their homes. The ideal protocol would be one that is flexible enough to use with either one or more fluent speakers and could involve a less than fluent elicitor when necessary.  107  Figure 4.5 Map used in ‘Map task’  This task proved frustrating for my speaker and did not result in any productions of the target echo questions. Far more successful was the ‘Put task’. This task consisted of a board with coloured objects in squares, along with a number of cut-out coloured animals. The speaker was asked to put an animal on a particular square. The objects were designed to be confusing in order to prompt the speaker to ask echo questions. For example, the word for blanket ‘slap’’ is similar to the word for tree ‘srap’. When mumbled, the question, “Put the chicken on the ....” prompted the speaker to ask ‘Lti stám’a?’ (On the what?), which was accompanied by a pitch rise. The task also allowed me to elicit simple  108  wh-questions. Having two blue houses on the board meant the speaker asked ‘Which blue house?’ when asked to put the animal on the blue house. The task is laid out below: Figure 4.6 “Put task”  These three general ways of eliciting intonation have proven successful and are easily adaptable for learning materials. The partly scripted conversations using slides can be used for practice in vocabulary, asking questions or even as a testing tool. Packaged with sound files, these materials could be good for practising at home. The conversations could be turned into scripts for students to practise, though consulting with the speakers in advance would be crucial. Task-based elicitation, such as the ‘Put task’, would be good practice for command giving, colours, and vocabulary. The ‘Map task’, when drawn properly, might also be useful for practice giving directions.  109  4.8  Conclusion  A large part of the work on this thesis was spent developing the best protocols under the limitations discussed above that would result in rigorous phonetic data, that would be interesting and appropriate for speakers, and that could be meaningfully adapted to learning materials. Many drafts of slides were constructed, tested and discarded for several reasons, the main one being that the picture did not accurately portray the target. This could be because the drawing was poor or ambiguous, or because it did not elicit the correct word order, the right determiner, or because it was not different enough from a previous slide, or the action was not semantically compatible with the object, among other problems. In the end, however, I believe that developing a partly verbal, partly non-verbal context through pictures and conversation resulted in the best tokens of the most natural speech possible under the artificial circumstances of phonetic elicitation while still providing enough tokens in a controlled environment to be statistically meaningful. These materials are also basic enough and flexible enough to be useful in a classroom setting. The next four chapters present the results of the two experiments described above. Chapter 5 presents the part of Experiment 1 that examines how speakers mark stress in St’át’imcets.  110  Chapter 5 Prominence in St’át’imcets  5.1  Introduction  This chapter examines three traditional acoustic correlates of prominence (F0, duration and intensity) to examine how speakers in St’át’imcets mark stress. The results presented here are used as background to test the hypothesis that extrapods and unstressed syllables should be distinct in terms of these prominence correlates. This hypothesis is investigated in the next chapter. Before the prominence correlate differences between unstressed vowels and extrapods can be compared, how speakers of the language implement the phonetic distinction between heads and non-heads must be understood. This includes stressed versus non-stressed vowels (i.e. unstressed vowels and extrapods), as well as primary versus secondary stress. To address possible categorisations of extrapods as bearing secondary stress, comparisons of extrapods to vowels with secondary stress and noninitial primary stress are also made. Results show that speakers use a combination of F0, duration and intensity to make prominence distinctions. However, speakers make fewer distinctions between heads of various types than between heads and non-heads. The direction of difference is as predicted: phonological heads are acoustically stronger than non-heads. Results also show that extrapods are not heads: they differ acoustically both from vowels with primary and secondary stress. The next section gives some background into the acoustic correlates of stress.  111  5.2  Phonetic Background81  As we saw in Chapter 3, phonological stress is considered a relative phenomenon, dependent on structure—relative strength is only defined under a sisterhood or motherhood relation. In the phonetic literature, stressed syllables are defined in relative rather than absolute terms as well— a stressed syllable is generally understood to be more perceptually ‘prominent’ than an unstressed syllable. Laver (1994) gives the following definition of stress (in English): “Other things being equal, one syllable is more prominent than another to the extent that its constituent segments display higher pitch, greater loudness, longer duration or greater articulatory excursion from the neutral disposition of the vocal tract” (Laver 1994: 450). As discussed in Section 1.5.3, the acoustic correlates of prominence will be discussed in terms of ‘stronger than’ rather than ‘more prominent than’. The correlates of stress, while separable, are interdependent. The complexity of the interrelations can be seen from the conflicting results in Fry (1955, 1958) and Lieberman (1960). Fry (1955, 1958) showed that both duration and amplitude affected stress judgements with synthetic speech in English and that change in duration but not amplitude was enough to alter judgements. Fry (1958) reports that the magnitude of the difference in F0 between stressed and unstressed syllables in the same word had no effect, as long as the difference was discernible. He concluded also that phrase-level intonation (i.e. F0) is an ‘overriding factor’ in the perception of stress. The implication is that speakers use different cues to signal word and phrase-level prominence and that duration is a stronger cue at the word level.82 Lieberman (1960), on the other hand, found that stressed vowels in English had a higher F0 72% of the time, higher amplitude 90% of the time and longer duration 70% of the time when compared to unstressed vowels across words. When compared to the unstressed vowel within the same word, stressed vowels had higher F0 90% of the time, higher amplitude 87% of the time and longer duration 66% of the time. Lieberman found a ‘trading effect’ where no stressed vowel had both a lowered F0 and lowered  81 82  Part of this section is taken from Caldecott (2006b). See also Beckman & Edwards (1990, 1994).  112  amplitude compared to unstressed vowels. He concludes that increased F0 and amplitude are stronger correlates of stress than duration based on these results. Such conflicting results lead Sluijter & van Heuven (1996a, b) to conclude that much of the confusion regarding stress correlates is due to the covariation of accent and stress. For them, stress is a property of syllables in words while accent is associated with focus. Their results indicate that duration, glottal parameters (e.g. high frequency emphasis and glottal leakage) and to a lesser extent vowel quality are consistent correlates of stress in English. F0 and intensity, on the other hand, are correlates of accent.83 While there has been extensive research on the acoustic correlates of stress in English, there have been few acoustic analyses of stress in North American indigenous languages. One of these is Gordon (2004) on Chickasaw. 84 In keeping with the English results above, Gordon (2004) found that stressed vowels have higher F0, intensity and duration than unstressed syllables. These results are based on across-word, not withinword measurements. Based on the above research, it seems reasonable to begin with the assumption that St’át’imcets speakers will use a combination of F0, duration and intensity to mark stress. The only acoustic studies of stress in Salish languages are Watt et al. (2000) on Skwxwú7mesh (Coast Salish), an observational (i.e. non-statistical) study on SENĆOŦEN (Benner 2006a, b) and a pilot study by the author on St’át’imcets (Caldecott 2006a). Results in Watt et al. (2000) are consistent with Gordon (2004) and the English studies above, at least for disyllabic roots. Watt et al. studied the production of one male speaker. They found that his stressed vowels had higher F0, duration and intensity than his unstressed vowels. In trisyllabic roots, the initial stressed vowel had higher means for  83  They use the term focal accent, but as far as I can tell, also make the assumption that right-headed phrasal accent is focal accent. Koch (2008a) discusses this in detail and shows that phrasal accent and focal accent are not coextensive in N¬e÷kepmxcin. 84 Gordon (2004) in his acoustic study of stress in Chickasaw lists only the following acoustic studies on American indigenous languages: Seiler (1957) on Cahuilla; Michelson (1983) on Mohawk and Oneida; Doherty (1993) on Cayuga; Tuttle (1998) on Tanana; Hargus (2001) on Witsuwit’en and Martin & Johnson (2002) on Creek.  113  F0, duration and amplitude than both the second and third vowels. No significant differences between the second and third vowels were found.85 Benner’s (2006a, b) findings suggest that pitch is an unreliable correlate of stress in SENĆOŦEN. Trisyllabic word results are comparable to Watt et al. (2000): the initial stressed syllable is higher pitched than the two others, but the final syllable, posited to carry secondary stress, is distinguished from the intervening unstressed syllable only by loudness.86 Caldecott (2006a) presents the results of a pilot study examining the acoustic correlates of stress in St’át’imcets. The study is limited to /a/ and /i/ and does not take phrase-level intonation or utterance position into account. Results are vowel-dependent and indicated a three-way distinction between stressed, unstressed and extrapods rather than a simple stressed-unstressed distinction.87 The pilot featured three of the same speakers that participated in the experiment reported here: AP, LT, and RW. Speakers showed no primary stress versus secondary stress distinctions in either vowel. Other results were vowel-dependent. For /a/, the three speakers used different correlates to mark prominence: AP had significantly higher F0; LT had longer duration and increased intensity; and RW had only increased intensity. All speakers marked stress with significantly greater F0 and intensity for /i/. LT and RW had significantly greater duration and intensity extrapod /a/ than unstressed /a/. While AP made no unstressed versus extrapod distinction for /a/, she was the only speaker to do so for /i/, with significantly higher pitched extrapods than unstressed vowels. Since neither unstressed vowels nor extrapods are heads (and therefore neither was assumed to stand in a prominence relation to the other), Caldecott concluded that the acoustic distinctions were only attributable to the difference in parsing.  85  Watt et al.’s results for trisyllabic roots may also be complicated by other factors. First, they considered only four examples, two of which had schwa as the final vowel, and one of which had schwa as the initial vowel. There is also some disagreement about the parsing of trisyllabic roots. Dyck (2004) proposes that roots are right headed, so rather than the first vowel being stressed, the second one would be. The status of the first vowel is not clear under that analysis. 86 It is unclear what evidence supports parsing the final syllable as a domain head rather than an extrapod. See Leonard (2007) for discussion of parsing in SENĆOŦEN. 87 Caldecott (2006a) refers to them as ‘other’ syllables rather than extrapods.  114  The experiment reported in this chapter seeks to address the limitations of Caldecott (2006a) by controlling for both phrasal intonation and morpho-syntactic environment. The results provide more accurate information on how stress is realised in St’át’imcets. 5.2.1 Hypotheses and predictions The experiment reported in this chapter asks two main questions. First, how do speakers mark prominence in the language? Second, is there an acoustic distinction between secondary stress and extrapods? The first question is addressed by comparing primary stressed with non-stressed vowels (i.e. unstressed vowels and extrapods) as well as primary versus secondary stress. Based on the research cited above and the hypothesis that prominence is defined under sisterhood/motherhood, the prediction is that the correlates of prominence for primary stressed vowels will be acoustically stronger than both unstressed vowels and extrapods (i.e. a combination of higher F0 and/or increased duration and/or intensity). Based on the results in Caldecott (2006a), the prediction is that speakers will make no distinction between primary and secondary stress. In terms of the second question, recall that van Eijk (1997) and Davis (forthcoming) implied that extrapods were similar to syllables with secondary stress due to phonological similarities seen in Chapter 3. However, based on the intonation evidence presented in Chapter 2 and 3, it is predicted that extrapods will not show the same acoustic characteristics as non-initial primary or secondary stressed vowels, and support the proposed parsing.88 If extrapods show no significant prominence correlate distinctions from vowels with primary or secondary stress, then we must rethink our parsing of extrapods and derive another solution to account for the intonational evidence. The next section presents the methodology used in the experiment.  88  Recall that PWord prominence is rightmost, so in words of two feet or more, primary stress appears on the head of the rightmost foot while the initial vowel has secondary stress.  115  5.3  Methodology  The specific methodology of this section is laid out below. A more general methodology, such as speaker background, can be found in the Chapter 4. 5.3.1 Speakers The speakers that participated in the study, their dialect and the number of tokens used and excluded are given in Table 5.1 below: Table 5.1 Speaker information Speaker  Dialect  Number of tokens used (excluded)89  AP  Northern  332 (12)  CS  Northern  340 (21)  HD  Lower Southern  410 (7)  LT  Southern  341 (4)  RW  Northern  339 (20)  Data were excluded for several reasons: measurement error (as determined in SPSS as being 3 standard deviations away from the median in boxplots), misspeaking (including complete deletion of the token), or being preceded or followed by a pause. 5.3.2 Stimuli Stimuli selection was complex, and took into account segmental context, morpho-syntax and speaker familiarity as well as controlling for stress position. Tokens were also selected to be common to both dialects and of a high enough frequency that speakers were comfortable producing them. As mentioned in Chapter 4, the actions of the stimuli had to be semantically compatible with a number of objects and easy to represent and interpret graphically. The ideal token would have been a morpheme that occurred in primary stressed, secondary stressed, unstressed and extrapod position. However, since monosyllabic roots are by far the most common root type (comprising at least 88% of the lexicon (van Eijk 89  These numbers represent the total tokens used and excluded from experiment 1, which is reported in this chapter and the subsequent one.  116  1997)), most roots contain only a stressed vowel. As a result, stimuli had to include roots with full vowels as well as suffixes or enclitics with full vowels. In a language with 44 consonants, finding minimal pairs in roots and other morphemes within the same word was impossible, so across-word rather than within-word comparisons are made. In other words, since the unavailability of tokens of the sort [(CViCvi)Cvi ] rules out Vi versus vi versus vi comparisons, Vi versus vj in tokens such as [(C ViCv)] and [(CVC vj)] were made. The only root+suffix/enclitic combinations that provided near-minimal segmental contexts were verb stem + person suffixes/enclitics. Stimuli were created from van Eijk (1987, 1985/1997) and Davis (forthcoming) and checked with speakers. The person paradigm was selected for two reasons: i) person suffixes/enclitics containing all three full vowels can occur in non-initial primary stress and word-final unstressed and extrapod positions, depending on the shape of the preceding root or root+suffix combination, and ii) employing person markers limits the type of morpheme to one paradigm and avoids mixing functional and lexical suffixes/enclitics. The person suffixes/clitics used were the enclitics 1sg. subj /=lhkan/ and 3 pl. subj /=wit/, and the 2 pl. object suffix /-tumulh/. Based on the fact that person enclitics pattern more like person suffixes than second position enclitics, I assume that the morpho-syntactic distinction between suffixes and enclitics does not affect how prominence is marked (van Eijk 1997, Davis forthcoming). Two further distinctions between stimuli containing /a/ and /i/ on the one hand and /u/ on the other exist. First, the suffix containing /u/ is disyllabic rather than monosyllabic (no monosyllabic person suffix containing /u/ exists). Second is the issue of argument structure and present versus non-present morphology. The enclitics containing /a/ and /i/ are subject morphemes while the suffix containing /u/ is an object morpheme. While I assume that argument structure does not affect how prominence is marked, there is also a difference in the type of syntactic constituents. In the tokens used to test /u/, the subject enclitic is attached to the pre-predicative auxiliary (It is necessary to use constructions of this type for stress placement reasons). In the tokens used to test /a/ and /i/, no overt object morphology was present but the construction is interpreted as having null 3sg. object morphology (Davis forthcoming). As a result, while both types of tokens lack some person morphology, it is possible that these morphemes still affect  117  how speakers parse them. Moreover, it is possible that the presence or absence of person morphology affects whether tokens are parsed as PWords or PPhrases. For the purposes of the experiment, I assume that all tokens are both PWords and PPhrases. Recall that Davis (forthcoming), Roberts (1993) and Roberts & Shaw (1994) treat the root + its affixes and enclitics as a PWord, while Koch (2008a), Beck (1999) and Beck & Bennett (2007) treat similar morpho-syntactic constituents in other Interior Salish languages as PPhrases. Given that Sluijter and van Heuven (1996a) found that duration signals lexical stress, while F0 and intensity are correlates of phrase-level prominence, it may be that what was previously considered primary stress in words of two feet or greater is in fact phrasal accent. Although this distinction is not the focus of this experiment, it is touched upon in this chapter and the next. As a result of the above distinctions, it is possible that /u/ will pattern differently from /a/ and /i/, but whether such a distinction would be attributable to affix type, argument morphology or prosodic shape requires further research. Tokens are given below, with the bolded syllable in the structure being the one targeted. Table 5.2 Tokens for /a/ and /i/ PWd=Prosodic Word, Ft = foot, ps = initial primary stress, nips= non-initial primary stress, ss= secondary stress, us= unstressed, ex=extrapod. Stimuli were not elicited in isolation and the three dots indicate the location in the target sentence.  unstressed  a i  Ph |\ PWd | Ft /\ ps us tsíx∑=kan get.there=1sg.subj ‘I got there...’ ˜áß = wit go=3pl.subj. ‘They went...’  primary stress & extrapod Ph |\ PWd / \ Ft \ / \ \ ps us ex k∑á n-¢nß=kan catch-tr-tr=1sg.subj ‘I caught...’ æít x∑-¢m = wit house-mid=3pl.subj ‘They camped...’  secondary stress & noninitial primary stress Ph | \ PWd / \ Ft Ft / \ / \ ss us nips us k∑à n-¢nß=kán=k¢¬ catch-tr-tr=1sg.subj=fut ‘I will catch...’ æìt x∑-em= wít =k¢¬ house-mid=3pl.subj=fut ‘They will camp...’  118  Table 5.3 Tokens for /u/ Primary stress  u  Ph | \ PWd / \ Ft \ / \ \ ps us ex æú ¬-u˜ =¬kan... point-tr.=1sg.subj ‘I pointed at...’  Secondary stress & non-initial primary stress & unstressed Ph | \ PWd / \ Ft Ft / \ / \ ss us nips us æù ¬-u˜-tú mu¬ point-tr-1.pl.obj ‘..pointed at us’  Extrapod Ph /| PWd / \ \ Ft Ft \ / \ / \ \ ss us ps us ex pà q∑u÷-mín-tu mu¬ frighten-tr-2pl.obj ‘...frightened of us’  All tokens were recorded using a Marantz solid-state pmd 660 recorder, on mono channel, at pcm 44.1 kHz and resampled at 16 kHz. Recording was done using a Shure head-worn condenser microphone WH30XLR. The tokens were framed in a target sentence that controlled for phrase-structure, intonation and phrasal accent as laid out in Chapter 2. Sample target sentences are given in Table 5.4 below.  119  Table 5.4 Sample target sentences (given in the orthography) a i Iy, kwánenskan Iy, tsítcwemwit Primary stress/ ti/ta múlca. (l)áku7 ti tsal’álha. extrapod ‘Yes, I caught a ‘Yes, they camped stick.’ at the lake.’ Iy, tsícwkan áku7 Iy, n’áswit áta7 Bingo. lamcalálhcwa. Unstressed ‘Yes, I went to ‘Yes, they went to Bingo.’ the church.’ Iy, tsìtcwemwít Iy, kwànenskán kelh láku7 ti Secondary kelh ti múlca. tsal’álha. stress/non-initial primary stress ‘Yes, I will catch ‘Yes, they will the stick.’ camp at the lake.’ Secondary stress/non-initial primary stress/unstressed Extrapod  u Iy, tsúlhun’lhkan ti/ta sqáx7a. ‘Yes, I pointed at the dog.’  Iy, wá7lhkan tsùlhun’túmulh lhkúnsa. ‘Yes, I’m pointing at you guys now.’ Iy, cúz’lhkan pàqu7míntumulh natcw. ‘Yes, I’m going to be afraid of you tomorrow.’  Target sentences were randomly presented to speakers 22 times in sessions of not more than two hours. 5.3.3 Measurements and analysis Following Gordon (2004) and Watt et al. (2000) all of the comparisons made in the first section are across-word and not relative.90 This means that, for example, the values of stressed vowels in every token were measured, and the mean was then compared to the mean of unstressed vowels across words. In addition to permitting the comparison of identical vowels, across-word comparisons were necessary to compare unstressed vowels and extrapods. Since extrapods occur only domain-finally, in order to control for word position effects, word-final unstressed vowels also had to be considered.91 As such, final 90  As discussed in this chapter and in Chapter 4.4, this was due to inherent vowel baselines and the lack of tokens that permitted within word comparisons. 91 Extrapods do occur word-initially in St’át’imcets and are discussed in Appendices E and F.  120  unstressed vowels and final extrapods cannot occur in the same word. However, since stress is generally considered a relative phenomenon, relative within-word comparisons for one speaker were also made. In these cases, the relative differences between vowels within the same word were compared. The caveat for these comparisons is that inherent vowel differences and segmental context could not be controlled for. Target sentences were segmented in Praat and a script was executed to measure duration, midpoint F0 and midpoint intensity.92 The duration of the vowel rather than the syllable, was measured based on findings in English (Crystal & House 1990). Crystal and House showed that average vowel duration is independent of the number and order of phones in a syllable while the duration of a syllable is a function of stress and the number of phones. Statistical analyses were performed in SPSS on the following comparisons: i)  Head versus non-head • Initial primary stress vs. unstressed • Initial primary stress vs. extrapod • Initial primary stress vs. secondary stress  ii)  Other stress versus extrapod • Secondary stress vs. extrapod • Non-initial primary stress vs. extrapod  Following the reasoning laid out in Chapter 4, t-tests for the 5 comparisons above for the 3 dependent variables (F0, duration, intensity) were performed. A Bonferroni correction was applied to avoid inflated family-wise errors, resulting in a significance value of p< .003.93 Values of .003≤p≤.01 (representing a 75% confidence interval) are reported as marginally significant.  92  Mean and Maximum F0 and intensity were also measured. Mean values were considered an indicator of trends and are not reported on. Maximum and midpoint values for both F0 and intensity reflect the same overall pattern, so the results reported here are the midpoints ones. 93 T-tests were also carried out for unstressed versus extrapod, as reported in the next chapter, as well as secondary stress versus non-initial primary stress and initial versus non-initial primary stress. As a result, the p value is based on a Bonferonni correction that is calculated over eight conditions rather than six. Comparisons were made within speakers and within vowels only.  121  5.4  Results  This section presents the results of the experiment in two sections, corresponding to the two questions posed above: i)  Heads versus non-heads • primary stress versus unstressed • primary stress versus extrapod • initial primary stress versus secondary stress  ii)  Other heads versus extrapods • secondary stress versus extrapod • non-initial primary-stress versus extrapod  Results are discussed cumulatively in Section 5.5. The first three tables give descriptive statistics by vowel, including the mean F0, duration and intensity and standard deviation.94 The first table shows the results for /a/. The second table shows the results for /i/ and the third table shows the results for /u/.  94  Graphical representations of results can be found in Appendix C.  122  Table 5.5 Results for /a/ (In this and subsequent tables, N= number of tokens, values represent means and the number in parentheses following the mean represents standard deviation. ps= initial primary stress, nips=non-initial primary stress, ss=secondary stress, us=unstressed, ex=extrapod) Syll N F0 (Hz) Duration (ms) Intensity (dB)  AP  ps nips ss us ex  22 25 25 24 23  164 (6) 160 (5) 162 (4) 157 (4) 154 (5)  80 (10) 118 (14) 70 (12) 89 (15) 80 (16)  81 (2) 79 (2) 80 (1) 79 (2) 79 (2)  CS  ps nips ss us ex  18 28 23 27 22  193 (9) 198 (8) 182 (7) 184 (11) 191 (14)  83 (16) 88 (15) 109 (25) 73 (16) 65 (12)  79 (2) 77 (1) 77 (1) 78 (3) 76 (3)  HD  ps nips ss us ex  24 26 26 25 19  134 (6) 152 (6) 136 (6) 145 (9) 134 (6)  81 (13) 107 (10) 90 (14) 63 (15) 63 (9)  81 (3) 82 (2) 80 (2) 81 (3) 78 (2)  LT  ps nips ss us ex  22 24 24 23 21  214 (10) 219 (8) 214 (7) 209 (9) 205 (8)  83 (9) 82 (12) 81 (9) 56 (5) 59 (8)  87 (.7) 85 (.4) 86 (1) 86 (1) 86 (1)  RW  ps nips ss us ex  28 22 21 23 30  170 (9) 180 (7) 169 (5) 180 (8) 176 (6)  90 (16) 100 (13) 91 (13) 84 (22) 81 (15)  79 (3) 77 (1) 78 (2) 77 (2) 77 (2)  123  Table 5.6 Results for /i/ Syll ps nips ss us ex  N 20 21 21 22 20  F0 176 (6) 171 (5) 177 (7) 161 (10-) 159 (6)  Duration 40 (6) 55 (9) 38 (8) 42 (8) 52 (11)  Intensity 81 (2) 83 (2) 81 (2) 79 (1) 81 (2)  CS  ps nips ss us ex  27 19 19 23 25  197 (7) 189 (6) 197 (10) 183 (7) 181 (8)  124 (17) 66 (14) 96 (18) 47 (13) 52 (14)  81 (3) 77 (1) 79 (2) 76 (2) 78 (4)  HD  ps nips ss us ex  32 25 25 34 32  137 (9) 147 (7) 140 (9) 139 (6) 134 (7)  70 (14) 55 (11) 69 (15) 59 (13) 60 (14)  78 (2) 82 (2) 81 (2) 75 (2) 76 (2)  LT  ps nips ss us ex  24 24 24 24 24  229 (11) 220 (9) 233 (11) 198 (9) 190 (6)  47 (8) 38 (6) 43 (8) 32 (13) 32 (9)  87 (1) 84 (1) 81 (2) 87 (1) 87 (.4)  RW  ps nips ss us ex  196 (11) 183 (8) 194 (8) 180 (9) 169 (7)  91 (25) 70 (18) 76 (10) 74 (19) 66 (18)  79 (2) 77 (1) 78 (2) 77 (2) 75 (.4)  AP  22 20 20 21 18  124  Table 5.7 Results for /u/ Syll ps nips ss us ex  N 23 21 21 21 23  F0 171 (7) 169 (7) 164 (5) 166 (7) 145 (5)  Duration 60 (8) 67 (11) 60 (14) 74 (12) 85 (11)  Intensity 80 (2) 82 (1) 82 (2) 80 (1) 79 (2)  CS  ps nips ss us ex  23 20 21 21 22  197 (7) 188 (5) 197 (9) 159 (10) 166 (11)  122 (36) 86 (14) 82 (7) 96 (18) 107 (23)  81 (3) 79 (2) 77 (2) 76 (2) 75 (2)  HD  ps nips ss us ex  23 30 30 30 31  138 (10) 143 (6) 145 (5) 127 (4) 120 (5)  81 (16) 84 (13) 68 (9) 63 (10) 62 (13)  82 (3) 82 (2) 80 (2) 79 (2) 78 (2)  LT  ps nips ss us ex  18 22 22 22 22  228 (8) 203 (5) 210 (4) 180 (6) 178 (6)  63 (6) 92 (10) 58 (6) 76 (8) 64 (8)  87 (1) 82 (1) 81 (1) 81 (1) 80 (1)  RW  ps nips ss us ex  24 22 22 22 23  184 (7) 172 (5) 180 (6) 163 (6) 160 (7)  108 (19) 69 (16) 72 (12) 68 (18) 67 (14)  78 (2) 69 (1) 71 (1) 70 (1) 68 (1)  AP  The tables above present the values and standard deviation for the F0, duration and intensity of vowels in initial and non-initial primary stressed position, secondary stressed position, unstressed position and extrapods. The following sections present the comparisons of the above values to determine whether and how speakers make significant differences between the various stressed vowels. 5.4.1 Head versus non-head results This section presents the significant differences for the head versus non-head comparisons given above. The first sub-section compares initial primary stressed vowels with unstressed vowels. The prediction for this comparison was that the prominence  125  correlates for primary stressed vowels would be stronger than for unstressed vowels. As we can see from the results below, this was generally the case: Table 5.8 Primary stress > unstressed. (In this and subsequent tables, only significant differences are listed. Shaded cells indicate no significant differences. A negative t-value indicates a significant difference in the direction opposite to the prediction. The superscript 1 indicates marginal significant difference.)  Speaker AP  CS  HD  LT  RW  Vowel a i u a i u a i u a i u a i u  F0 t-value 4.9 5.6  p-value <.001 <.001  Duration t-value p-value  -4.6  <.001  17.5 3.1 4.5 3.5 5.4 12 4.8 -5.5  <.001 .0041 <.001 .001 <.001 <.001 <.001 <.001  2.5 7.3  .011 <.001  Intensity t-value p-value  1  2.70 7.2 16.9 -5  .01 <.001 <.001 <.001  4.9  <.001  10.4 21.5 -4.0 5.2 10.4  <.001 <.001 <.001 <.001 <.001  7.3 8.7  <.001 <.001  5.4 3.2 3.6  <.001 .003 .001  19.4  <.001  3.1 15.8  .0041 <.001  All speakers made significant acoustic distinctions between initial primary stressed and unstressed vowels. The majority of the differences were strong, as demonstrated by the high number of p values that were less than .001. Stressed vowels generally had significantly higher pitch and intensity than unstressed vowels, with the exception of AP. AP showed no significant differences in intensity, and both she and LT had significantly longer unstressed /u/ than stressed /u/ (indicated in the table by the negative t-value). HD and RW showed higher F0 in unstressed /a/ than stressed /a/. This is most likely due to late pitch peak realisation. HD in particular had pitch peaks, which were realised late in the vowel, and this potentially carries over onto the following vowel. The boxplots of HD’s pitch peaks are given in Appendix G. If we compare the results of AP, LT and RW to their pilot results, we see that LT and RW made significant F0 distinctions in this experiment in addition to the duration and/or intensity distinctions they made in the pilot. AP was consistent in her use of correlates. No trade-off patterns of the type found in Lieberman (1960) were observed.  126  The next table compares initial primary stress with domain-final extrapods. The prediction was that since primary stress was the head of the domain, the prominence correlates for stressed vowels would be stronger than for extrapods. This, again, proved generally true with the exception of AP’s duration. Table 5.9 Primary stress > extrapod Speaker Vowel F0 t-value p-value AP a 5.7 <.001 i 9.2 <.001 u 14.1 <.001 CS a i 7.3 <.001 u 14.9 <.001 HD a i u 8.2 <.001 LT a 3.1 .0041 i 14.8 <.001 u 22.6 <.001 RW a -3.0 .0041 i 10.2 <.001 u 11.1 <.001  Duration t-value p-value -4.1 -8.8 3.9 16.3  <.001 <.001 <.001 <.001  5.2 3.1 4.7 9.2 6.4  <.001 .0031 <.001 <.001 <.001  3.7 8.6  .001 <.001  Intensity t-value p-value 3.3 .002  3.0 3.4 9.2 2.8 3.3 5.2 3.8 -3.4 23 3.2 9.2 18.4  .0051 .001 <.001 .0071 .002 <.001 <.001 .002 <.001 .002 <.001 <.001  The strong p-values demonstrated in the table above confirm the prediction that St’át’imcets speakers have acoustically stronger stressed vowels than extrapods. Most speakers, when they make a distinction, have significantly higher F0 and intensity stressed vowels than extrapods. Duration for speakers other than AP was longer in stressed vowels than extrapods. Unlike other speakers, AP’s /i/ and /u/ extrapods were significantly longer than stressed vowels. The next table compares heads of words with heads of feet. Caldecott (2006a) showed that St’át’imcets speakers made no acoustic distinction between initial primary and secondary stress. Therefore, the prediction is that there will be no acoustic distinctions between primary and secondary stress vowels.  127  Table 5.10 Primary stress = secondary stress (-indicates ss>ps)  Speaker  Vowel  F0 t-value  AP  CS  HD  LT  RW  a i u a i u a i u a i u a i u  p-value  3.3 4.3  .002 <.001  6.8  -3  8.6  Duration t-value p-value 3 .0041  <.001  -3.8 5.3 5  .001 <.001 <.001  .0051  3.5  <.001  <.001  Intensity t-value p-value  -4.1 3.4  <.001 .002  6.1  <.001  -4.7  <.001 .001 <.001 <.001  <.001  2.8  .0081  3.6 13.9 18.5  7.6  <.001  13.1  In the table above, the number of shaded cells indicates that speakers make fewer distinctions between primary and secondary stress than between stressed and non-stressed vowels. For example, RW makes no significant distinctions in pitch, and only makes significant distinctions in duration and intensity for /u/. Most speakers make significant distinctions in F0 only for /u/. To summarise, speakers make a clear directional distinction between stressed and non-stressed vowels and fewer, but still directional, differences between primary stress and secondary stress. Primary and secondary stress comparisons seem to suggest that duration and intensity play a larger role than F0 in cueing word/phrase-level prominence. The next section compares the prominence correlates of other stressed vowels and extrapods.  128  5.4.2 Extrapods versus other heads The following section addresses the second research question, namely whether extrapods are in fact acoustically equivalent to secondary stress or non-initial primary stress, as implied by van Eijk (1997) and Davis (forthcoming). Here, the prediction is that secondary stress, as a head, will have stronger prominence correlates (higher F0, greater duration, greater intensity) than extrapods, which are non-heads. Table 5.11 Secondary stress > extrapod Speaker Vowel F0 t-value p-value AP a 5.9 <.001 i 8.2 <.001 u 12.6 <.001 CS a i 6.2 <.001 u 12 <.001 HD a i 2.6 .001 u 19.8 <.001 LT a 3.8 <.001 i 17.3 <.001 u 20.4 <.001 RW a -4.3 <.001 i 11.1 <.001 u 10 <.001  Duration t-value p-value -4.5 -6.6 7.4 8.8 -4.8 7.9  <.001 <.001  8.6 4.5  <.001 <.001  <.001 <.001 <.001  Intensity t-value p-value 3.2 .0031 6.7  <.001  3.8 3.4 8.2 4  .001 <.001 <.001 <.001  -17.3  <.001  2.6 6.9 6.6  .011 <.001 <.001  The above table shows that speakers make significant distinctions between secondary stress and extrapods primarily in F0 and intensity. Notice again that AP’s extrapods are longer than secondary stressed vowels. For most speakers, though, secondary stressed vowels are stronger than extrapods. This supports our categorisation of ‘other’ syllables as extrapods over an analysis that parses them as secondary stress. Take LT, for example. Her secondary stressed vowels have consistently higher F0 than extrapods, longer duration for /a/ and /i/ but secondary stressed /i/ is significantly less intense than extrapod /i/. The next table compares non-initial primary stress with extrapods. Recall that van Eijk (1997) and Davis (forthcoming) consider extrapods to be ‘potentially stressed’. To rule out extrapods showing similar acoustic characteristics to non-initial primary  129  stress, the following table shows these comparisons. The prediction is that non-initial primary stressed vowels, as heads of the larger domain, will be stronger than extrapods. Table 5.12 Non-initial primary stress > extrapod Speaker Vowel F0 Duration t-value p-value t-value p-value AP a 3.4 .001 8.7 <.001 i 6.9 <.001 u 12.8 <.001 -5.4 <.001 CS a i 3.8 <.001 3.1 .0031 u 6.3 <.001 -3.6 .001 HD a 9.1 <.001 15.2 <.001 i 6.9 <.001 u 16 <.001 6.4 <.001 LT a 5.7 <.001 7.6 <.001 i 13.4 <.001 u 15.1 <.001 10.4 <.001 RW a 4.6 <.001 i 6.1 <.001 u 6.4 <.001  Intensity t-value p-value 2.8 8.1  .0071 <.001  3.1 7.8 11.1 8.6  .0041 <.001 <.001 <.001  -12.1 5.6  0<.001 <.0010  5.7  <.001  The number of p-values that are less than .001 in the table above confirms that, for most speakers, non-initial primary stress has stronger prominence correlates than extrapods. The results for F0 are more consistent than those for duration and intensity. AP again had significantly longer extrapod /u/s than non-initial primary stress /u/s. This pattern was also demonstrated by CS for the duration of /u/, and by LT for the intensity of /i/.  5.5  Discussion  Section 5.4 shows that St’át’imcets speakers use a combination of traditional acoustic correlates of stress to mark prominence in the language. Stressed vowels have higher F0, greater duration and greater intensity than both unstressed vowel and extrapods, as predicted by the model. Two exceptions to the above generalisation exist. One is the previously addressed higher F0 in HD and RW’s unstressed /a/. The other is that /u/ patterns differently from /a/ and /i/ for some speakers. One possible explanation for this might be a distinction, based on utterance length, between PWords and PPhrases. Prieto (2005) proposes a maximality constraint of two PWords on PPhrases in Catalan. If there is a similar  130  constraint on length in PWords in St’át’imcets, the longer tokens containing /u/ could be distinguished by speakers from those containing /a/ and /i/. The greater duration of /u/ could be interpreted as final lengthening applying in a different domain (e.g. phrase-final) than /a/ and /i/ (e.g. word-final). The directional differences between stressed vowels and extrapods support the prediction of the model, with the exception of AP’s duration results. AP’s extrapods have lower F0 but greater duration than stressed vowels. AP’s duration results might also be indicative of final lengthening, which is discussed further in Section 6.6. These results confirm that extrapods are acoustically distinct from initial primary stress in the same way that unstressed vowels are. The number of distinctions between primary stress and secondary stress was fewer than that made between heads and non-heads.95 Interestingly, more significant distinctions were made for /u/ than for /a/ or /i/, which is particularly evident in the F0 results. One possible interpretation of this assymetrical pattern is that the tokens for /u/ are different domains to those for /a/ and /i/, as mentioned above. If, following Sluijter and van Heuven’s (1996a, b) research on English and Dutch, F0 distinctions are correlated phrase-level prominence, it is possible that the long /u/ tokens are parsed as PPhrases, while the shorter /a/ and /i/ tokens are parsed as PWords. Such an analysis makes the prediction that the root+subject clitic tokens for /a/ and /i/ do not count as PPhrases while root+ subject+object clitic tokens for /u/ do. Further investigation is required to determine whether it is the length, the argument structure or the presence of overt morphology that is the crucial factor. The results of this study confirm the second prediction that extrapods are not heads. The number and strength of significant differences support an analysis of extrapods as distinct from both primary and secondary stressed syllables. The results of the current experiment differ somewhat from the results of the pilot. In the pilot study, speakers made no distinction between primary and secondary stress. However, utterance position of the tokens was not controlled. The differences in results could be due to the effects of phrasal intonation. More discussion of the interaction between stress and larger intonation contours is given in Chapter 6. 95  Further head comparisons are presented in Appendix D.  131  A further example of inter-experiment differences can be seen in Bessell (1997), in which RW also participated. Bessell (1997) found the duration of stressed /a/s for RW to be considerably higher than the results in this experiment and those of the pilot. Bessell’s means for RW were 153 ms compared to 134 ms in the pilot and 90 ms in this study. Bessell had speakers reading words from lists, as compared to words in sentences as in the pilot, without taking into account prosody and intonation, as in this study. These different results for one speaker show the importance of recording in context and controlling for intonation. The single word utterances of Bessell would have put the vowel in question under word-level stress and phrasal prominence, perhaps accounting for the increased duration.  5.6  Conclusion  This chapter has presented the results of an experiment examining how St’át’imcets speakers mark stress. The results were generally as predicted, with speakers using increased traditional prominence correlates (F0, duration and intensity) to mark the contrast between heads and non-heads. The effect was stronger between heads and nonheads of feet than between primary and secondary stress. It also confirmed that extrapods are acoustically distinct from both secondary stressed and non-initial primary stressed vowels. This chapter has established that St’át’imcets speakers use traditional correlates to mark prominence. The next chapter compares the prominence correlates of unstressed vowels and extrapods to test the hypothesis that they are acoustically distinct.  132  Chapter 6 Prominence Distinctions: Unstressed Vowels versus Extrapods  6.1  Introduction  This chapter presents the results for the second half of experiment 1, namely the comparison of the prominence correlates (F0, duration, intensity) between unstressed vowels and extrapods. This chapter uses the same methodology and tokens as the previous chapter to test the hypothesis that distinct phonological domains should show distinct acoustic characteristics. The direction of difference is predicted to be that of the MaxProm hypothesis presented in Chapter 3. Specifically, extrapods should have stronger prominence correlates than unstressed vowels. The most important result of this experiment is that phonetic contrasts reflect the prominence predictions of the model. Speakers use a combination of F0, duration and intensity to distinguish extrapods and unstressed vowels. However, they make fewer distinctions than between stressed and non-stressed vowels and reveal no set direction of contrast. These results support a model that assigns relative prominence under a phonologically-specified relationship (i.e. sisterhood/motherhood) only and is consistent with the Variable hypothesis. In the absence of the directionality constraint correlated with phonologically-specified relationships, other conflicting pressures emerge. The fact that the presence of contrast rather than the direction, is crucial supports a weak interpretation of the phonetics-phonology mapping. Across-word results are compared with within-word results for one speaker. This comparison shows the importance of considering syntagmatic comparisons as well as paradigmatic ones.  133  6.2  Background  Chapter 3 presents evidence that there is a phonological distinction between unstressed syllables and extrapods in St’át’imcets. In other words, the two γs in (1a) and (1b) below show different phonological characteristics. (1a)  (b) W | Ft /\ αγ  W /\ Ft \ /\ \ α β γ  [(α γ)]  [(α β) γ]96  In the first example, γ is both foot- and word-final (i.e. unstressed), while in the second, γ is only word-final (extrapod). All else being equal, such distinct phonological domains should show distinct acoustic characteristics and one dimension of difference might be prominence correlates (e.g. F0, duration, intensity). St’át’imcets gives us the opportunity to test these predictions and to reveal some of the acoustic characteristics of extrapods. As we saw in the previous chapter, the person enclitics /=lhkan/ (1sg.subj.) and /=wit/ (3pl.subj.), and the 2pl.obj. suffix /-tumulh/ occur in both extrapod and unstressed position, as in (2): (2a)  (b) W | Ft /\ α γ [(˜áßwit)] go=3pl.subj  W /\ Ft \ /\ \ α β γ [(æítx∑¢m)wit] house-middle=3pl.subj  The root+enclitic/suffix tokens allow a comparison of unstressed and extrapod vowels in identical segmental, morpho-syntactic and intonational contexts. One of the limitations of the experiment reported in the previous chapter was that exact minimal pairs across all comparisons were impossible. The fact that the unstressed versus extrapod comparison  96  Recall from the previous chapter that γ can be the same suffix/enclitic in both cases.  134  involves exact minimal pairs means that any significant differences between them will be particularly strong. Based on the above hypothesis, the prediction is that unstressed vowels and extrapods should be acoustically distinct in St’át’imcets. Recall that four possible hypotheses regarding the relative strength of extrapods in relation to unstressed vowels were available under the Prosodic Hierarchy model. These hypotheses are summarised in Table 6.1: Table 6.1 Hypotheses Name  Hypothesis  Acoustic prediction  Traditional  us = ex  There are no prominence correlate distinctions.  Variable  us ~ ex  Maximise Prominence (MaxProm)  ex > us  Extrapods and unstressed syllables are distinct in terms of prominence correlates but the difference is not predictable or consistent. -other factors motivate acoustic distinctions Extrapods have stronger prominence correlates than unstressed syllables as a result of not being subject to foot-internal head maximisation constraints.  Footed Stronger than Not (FSN)  us > ex  Unstressed vowels have stronger prominence correlates than extrapods as result of being properly parsed.  The Traditional hypothesis predicts there should be no prominence correlate distinctions because neither is the head of a domain. The other three hypotheses all test the prediction of the model that distinct phonological domains should have distinct acoustic characteristics. The Variable hypothesis predicts that there should a contrast but not necessarily a consistent one. Recall that Liberman and Prince (1977) and Pierrehumbert and Beckman (1988) propose that relative prominence is a function of structure, and assigned under sisterhood or motherhood. Because no phonologically-specified relation between 135  unstressed vowels and extrapods exists, there is no constrained direction of difference. The realisation of extrapods under this hypothesis emerges as the result of conflicting pressures such as head or boundary maximisation and positional prominence effects. MaxProm predicts that prominence correlates (F0, duration, intensity) for unstressed vowels will be weaker than those for extrapods. This is motivated by a footinternal constraint that minimises foot-non-heads to maximise the head. The constraint does not apply to extrapods because they are outside of the foot. FSN predicts the opposite: prominence correlates for unstressed vowels will be stronger than for extrapods. The head constituent of the word is more prominent than the non-head constituent. Under this hypothesis, unstressed vowels, being parsed into the head constituent of the word, should show stronger correlates than extrapods, which are not in the head constituent of the word. Given the results of the pilot study reported in Chapter 5, the prediction is that extrapods will be show stronger prominence correlates (higher F0, duration, intensity) than unstressed vowels. If extrapods and unstressed vowels are distinct, we have evidence that supports non-exhaustive parsing. If unstressed vowels and extrapods are distinct in a consistent manner, this would support a stronger version of the phoneticsphonology mapping. If unstressed vowels are acoustically stronger than extrapods, this would support Liberman and Prince’s (1977) ‘strong is stronger than weak’ proposal (Liberman & Prince1977: 259). In other words, then we can define the relative prominence of unstressed and extrapods in terms of head constituency: unstressed syllables, by virtue of being in the head of the word are more prominent, and correlated with stronger acoustic characteristics than extrapods, which are not in the head of the word. On the other hand, if extrapods are acoustically stronger than unstressed vowels, we have confirmation of the pilot results and support that the minimisation of non-head vowels seen in Section 3.5.1 is a strategy that is reflected across the language. Furthermore, if they are distinct, but not in a consistent direction, this would be evidence of a weak phonetics-phonology mapping. In this case, presence of contrast, rather than a motivated direction of contrast, would be crucial.  136  Finally, if unstressed syllables and extrapods turn out to be acoustically nondistinct, this would support hypotheses that predict that non-heads are not distinct. It would raise two questions in terms of the theory. First, if prosodic constituents are the domains of phonological processes, and distinct domains are marked by distinct acoustic characteristics, how can we explain the direct mapping between acoustics and phonological domains for all domains other than extrapods? Second, if there is no converging acoustic evidence to support non-exhaustively parsed structures, is phonological evidence enough to accept the violability of Strict Layering?  6.3  Methodology  The methodology, including tokens, is the same as that presented in Chapter 5 and is not discussed further here.  6.4  Results  The results of the comparisons between unstressed vowels and extrapods are given in the table below. Recall the prediction was that acoustic correlates to extrapods would be stronger than unstressed vowels (based on the results of the pilot study).  137  Table 6.2 Extrapod > unstressed (In this and subsequent tables, - indicates a significant difference in the opposite direction of the prediction. Shaded cells indicate no significant difference.)  Speaker  Vowel  F0 t-value  AP  CS  HD  LT  RW  a i u a i u a i u a i u a i u  -11.6  Duration t-value p-value  p-value  <.001  -4.3 -2.9 -6.2  <.001 .0051 <.001  -3.5  .001  3.2 3.0  -5.1 -4.7  <.001  .002 .0041  Intensity t-value p-value 2.8 -3.1  .0081 .0041  2.7  .0091  -3.4  .001  -3.3  .002  -4.4 -4.0  <.001 <.001  .001  This table shows that in spite of both being non-heads, and being located in identical segmental, morpho-syntactic and intonational contexts, four of the five speakers make some distinction between unstressed vowels and extrapods. The fifth speaker, CS, makes a marginal distinction only in intensity for /i/. The shaded cells indicate that the number of distinctions is sparse, but when speakers do make a distinction, it is generally strong. The number of distinctions more closely resembles the differences in Chapter 5 between heads than the differences between heads and non-heads. This is consistent with the model, given that neither unstressed vowels nor extrapods are phonologically-specified as more prominent than the other. Results are vowel specific, in that only HD makes distinctions for /a/. This pattern of distinctions for /i/ and /u/ but not /a/ are puzzling, given that the stimuli for /a/ and /i/ are more similar than that of /i/ and /u/. In terms of the direction of difference, the results are summarised in the table below:  138  Table 6.3 Direction of difference Speaker Vowel F0 AP a i u us > ex CS a i u HD a us > ex i us > ex u us > ex LT a i us > ex u RW a i us > ex u  Duration ex > us ex > us  Intensity  General  ex > us us > ex  mixed  ex > us  ex > us  us > ex us > ex us > ex us > ex us >ex us > ex us > ex  us > ex  The table above shows that, for four of the five speakers extrapods have significantly lower F0 than unstressed vowels. AP has significantly longer extrapods than unstressed vowels. For HD and RW, unstressed vowels have significantly higher intensity than extrapods. CS has marginally significantly higher intensity extrapods than unstressed vowels whereas AP has mixed intensity results. The results indicate three of the five speakers (HD, LT and RW), when they make distinctions, have consistently stronger prominence correlates for unstressed vowels than extrapod. AP and CS do not.  6.5  Discussion  The first question addressed in this experiment was whether unstressed vowels and extrapods differed in terms of traditional prominence cues. Based on the hypothesis that phonologically distinct domains should be acoustically distinct, the prediction was that they would be different. This prediction was borne out. Four of the five speakers made significant distinctions between unstressed vowels and extrapods in identical segmental, morpho-syntactic and intonation contexts. The remaining speaker made a marginal distinction for one correlate in one vowel. The distinctions were not as numerous as those between stressed vowels and non-stressed vowels seen in the preceding chapter but were strong when they were made.  139  The answer to the second question, regarding the direction of difference, is more difficult to evaluate. Table 6.3 shows mixed results. HD, LT and RW have stronger acoustic correlates for unstressed syllables than extrapods for each contrast, supporting the hypothesis that footed segments are generally stronger than unfooted segments. CS’s extrapod /i/ is marginally stronger than unstressed syllables whereas AP shows mixed results. The variation between speakers cannot be explained by dialect, since the three Northern speakers (AP, CS, RW) show the complete range of variability, and only HD showed any significant effects for /a/. Three of the five speakers had stronger prominence correlates for unstressed vowels than extrapods, supporting the FSN model. However, given AP’s intra-speaker variation, it seems that the Variable hypothesis is best supported by the data. This hypothesis represents an interpretation of relative prominence defined by the model under sisterhood or motherhood. Since no sisterhood relation exists between unstressed syllables and extrapods, no set direction of difference is predicted. Under this hypothesis, contrast but not direction of contrast, is the crucial factor. In the absence of a direct mapping, conflicting pressures arise, the weighting of which results in speaker variation. HD, LT and RW appear to maximise the distinction between the vowel that is footed, and the one that is not while AP shows possible conflict between foot-internal reduction and a word-final lengthening effect. The results presented here are different from those found in the pilot. In the pilot, LT and RW had significantly greater duration and/or intensity extrapods than unstressed /a/. Neither LT nor RW made any significant distinctions for /a/ in this experiment. In the pilot, only AP made a distinction between unstressed and extrapod /i/, with extrapods having significantly higher F0. This contrasts with the above results in which AP had only greater duration and intensity extrapods, while LT and RW made intensity or F0 distinctions. One possible explanation for the inter-experiment differences could be the different conditions of elicitation and the fact that the pilot study did not take intonation or utterance position into account. A second possibility is that speakers are making use of a trade-off relation between correlates as seen in Lieberman (1960). Since this experiment did not measure all of the possible acoustic correlates to stress, it is possible that the variation reflects  140  speakers making use of other correlates, such as Sluijter and van Heuven’s (1996a, b) glottal parameters, in conjunction with F0, duration and intensity. This sort of correlate interaction and speaker variation in general requires further research. As discussed above, AP appears to have conflicting correlates: unstressed vowels are stronger than extrapods in terms of F0 but weaker in terms of duration. As we saw in Chapter 5, AP had significantly longer extrapods than stressed vowels. This suggests domain-final lengthening (Klatt 1976; Edwards et al. 1991; Wightman et al. 1992). From the across-word comparisons shown here, it is not possible to determine whether the results derive from a conflict between stress and larger domain prominence or intonation contours. One of the limitations of the study is that within-word comparisons that might show the interaction between domain-final effects and stress, are not easily made. To partly address this limitation, and since prominence is generally considered to be a relative phenomenon, the next section presents the within-word comparisons of one speaker.  6.6  Within-word versus Across-word Comparisons: A Case Study  This section examines the prominence correlates of extrapods in relation to surrounding syllables within the same word. Given that the across-word results (particularly AP’s duration) seem to indicate the presence of domain-final effects, examining the relative differences between vowels in the same word should give us a sense of how those results are affected by higher level prosodic effects.97 This section examines the within-word comparisons of one speaker, HD. Recall that HD had significantly higher pitch and intensity unstressed vowels than extrapods, but made no significant durational differences between the two vowel types. First, let us compare the relative differences in /a/ in the tokens /kwánenskan/ and /kwànenskánkelh/ in Table 6.4. The vowels in these two cases permit two comparisons: i) root /a/ has primary stress and affix /a/ is an extrapod, and ii) root /a/ has secondary stress and affix /a/ has primary stress:  97  Recall that within-word measurements are not the focus of this study due to the unavailability of withinword minimal pairs. Interpretation of results should take inherent baseline differences between vowels into account.  141  Table 6.4 Mean relative differences between root and affix /a/ (In this and subsequent tables, examples are given in the orthography. (=foot boundary, [=word boundary, nips=non-initial primary stress, ss=secondary stress, `marks secondary stress, ´ marks primary stress.)  ∆F0 (Hz) [(kwán-ens)=kan] [(kwàn-ens)(=kán=kelh)]  ps-ex ss-nips  9 -4  ∆ Duration (ms) ∆ Intensity (dB) 3.3 2 -11 1.2  From the above table, we can see that when the affix is parsed as an extrapod, the initial /a/ has higher pitch, slightly longer duration and higher intensity. When the affix is parsed as primary stress, it has higher F0, somewhat lower intensity, and is considerably longer than initial /a/. These results seem to indicate that intensity is a positional effect rather than a stress-related effect. The initial vowel has higher intensity, regardless of whether it is primary or secondary stress. The above results also support the classification of extrapods as distinct from non-initial primary stress and secondary stress. The distinction between primary and secondary stress appears to be duration based rather than F0 or intensity based. The primary stressed vowel is 11 ms longer than the secondary stress vowel but only 4 Hz higher in pitch. In spite of being only 4 Hz higher in pitch, the primary stressed vowel might be perceived as significantly higher in pitch. Pierrehumbert (1979) found that speakers in English were predisposed to judging rightmost vowels as higher pitch, even when they were relatively lower in pitch. This was attributed to English having a rightmost nuclear accent. While Pierrehumbert’s measurements were made over IPs rather than smaller domains such as those we have here, it is possible that a perception test might reveal similar results in St’át’imcets. The next table examines the relative differences between primary stress and secondary stress in /u/ tokens: /tsùlhun’túmulh/ and /pàqu7míntumulh/. Table 6.5 Mean relative differences between primary stress and secondary stress ∆F0 (Hz) ∆ Duration (ms) ∆ Intensity (dB) [(tsùlhun’)(túmulh)] ss-nips -1 -7 2 [(pàqu7)(míntu)mulh] ss-nips -6 -27 -4 This table shows that, while non-initial primary stress is higher in pitch than initial secondary stress, the difference is 6 Hz or less. These results are similar to the results for /a/ above. We can also see that the non-initial primary stressed vowel in the second example is considerably longer than the secondary stressed vowel compared to the first 142  example. The intensity differences contrast in opposing directions. In the first example, the vowel with non-initial primary stress has less intensity than the preceding vowel with secondary stress, and in the second example, the opposite presents. The distinction in F0 between primary and secondary stress does not appear to be that great, If, following Sluijter and van Heuven’s (1996a, b) results for English, F0 is a diagnostic of phrase-level prominence, then small F0 differences between primary and secondary stress do not appear to support positing a phrase-level versus word-level distinction between primary and secondary stress in longer words. Duration appears to be the main difference, an effect we might expect for lexical stress. However, if we consider the relative differences between heads and non-heads, we see a clear difference between heads of words/phrases and heads of feet. Table 6.6 presents mean relative changes between the head and non-head of a secondary stress foot on the one hand and between the head and non-head of the primary stress foot on the other: Table 6.6 Relative within-foot differences between secondary stress versus unstressed and non-initial primary stress versus unstressed in /u/ tokens ∆F0 (Hz) ∆ Duration (ms) ∆ Intensity (dB) [(tsùlhun’)(túmulh)] ss-us 3 20 2 nips-us 18 14 3 [(pàqu7)(míntu)mulh] ss-us 3 8 2 nips-us 14 52 4 Here we can see that the mean change in pitch between the primary stressed vowel and its unstressed vowel is exponentially higher than that between the secondary stressed vowel and its unstressed vowel for both tokens. If pitch change rather than absolute pitch is a correlate of phrase-level rather than word-level prominence, then we could interpret the large difference in pitch between the head and non-head as indicative that /u/ stimuli are PPhrases rather than PWords. The intensity differences between the primary-stressed vowel and its unstressed vowel are not as dramatic as the others but the magnitude of contrast is still the same: a greater contrast in the head foot of the PWord. The difference in magnitude of relative duration between primary stressed head and non-head and secondary stressed head and non-head in the first example are not  143  great. However, there is a considerable difference between the two in the second example. This large difference can be attributed to two factors: i) the longer primary stressed vowel in the second example and ii) the word position of the unstressed vowel. These effects can be seen from the following figure, which gives the mean duration of the vowels in the /u/ stimuli given above. Figure 6.1 Mean duration values for /u/  tsù  pà  lhun’  qu7  tú  mín  mulh  tu  mulh  Unstressed vowels are considerably shorter when they are not domain-final. The penultimate unstressed /u/ in /pàqu7míntumulh/ (the unstressed vowel in the head foot of the second example) is approximately 15 ms shorter than the domain-final /u/ in /tsùlhun’túmulh/ (the unstressed vowel in the head foot of the first example). Duration results show a greater contrast between the stressed vowel and unstressed vowel in the head foot than in the non-head foot, except in the case where the unstressed vowel is final. This illustrates an interaction of domain-final lengthening and prominence correlate contrast. From Figure 6.1, we also can see that the final non-stressed vowels have a mean duration of roughly 65 ms, regardless of whether they are parsed as unstressed vowels or extrapods. This suggests that domain-final lengthening applies to the syllable and affects both unstressed vowels and extrapods. If we consider the mean differences between unstressed vowels and extrapods within the same word, we see much larger differences than those observed across words.  144  Table 6.7 Relative differences between unstressed vowels and extrapods for /u/ ∆F0 (Hz) ∆ Duration (ms) ∆ Intensity (dB) [(pàqu7)(míntu)mulh] us-ex 12 -12 1 The differences between unstressed vowels and extrapods show the conflicting directionality results between F0 and duration. The unstressed vowel in the example above is 12 Hz higher in pitch than the extrapod. This rate of difference is similar to the difference between primary stress and extrapod in Table 6.4. In contrast, extrapods are longer than unstressed vowels, to a degree roughly equivalent to the primary stress versus unstressed relation in Table 6.6 or the difference between head and non-head in the head foot of the word in Table 6.5. These differences indicate a contrast in both F0 and duration between unstressed vowels and extrapods, but in conflicting directions. The results in the tables and figures above show the complex interaction between stress and intonation in the language. Foot-level stress (i.e. secondary stress-unstressed) is indicated by slightly higher F0 and intensity but mainly by increased duration. This follows Sluijter and van Heuven’s (1996a, b) prediction about lexical stress. Word/phrase level stress (i.e. primary stress-secondary stress) is indicated also by a slightly higher F0 and a difference in duration. Head feet of words are marked by a greater magnitude of contrast between head and non-head. Word/phrase final boundaries are marked by a large fall in F0 and greater duration of the final vowel.  6.7  Discussion of Within-Word Results  One of the goals of this experiment is to examine the acoustic correlates of the relative prominence of unstressed vowels and extrapods. This turns out to be more complicated than expected. For one speaker, extrapods have lower pitch than the preceding unstressed vowel within words, suggesting that they are acoustically weaker. This difference was considerable, equivalent to roughly the distinction between the head and extrapod in Table 4. In contrast, for the same speaker within words (and across words for some speakers), extrapods are significantly longer than preceding unstressed vowels. As a result, extrapods could be interpreted as acoustically stronger than unstressed vowels.  145  The duration differences are again roughly equivalent to some head-non-head distinctions. The magnitude of these contrasts is somewhat surprising given that neither is a head.  98  Conceivably, the pitch drop between unstressed vowels and extrapods is due to  natural F0 declination. However, declination is usually measured within IPs rather than PWords or PPhrases and is usually measured between like syllables some distance apart (e.g. Pierrehumbert 1979). Even if we consider the domains in question to be PPhrases, and unstressed vowels and extrapods to be like syllables (i.e. non-stressed), the rate of declination does not strongly reflect previous descriptions. Pierrehumbert (1979) showed a declination over multiple stressed vowels in a phrase of 6.8-11.1 Hz/ syllable. Koch’s (2008a) results for N¬e÷kepmxcin showed a 3 and 4 Hz/syllable declination between initial and final stressed vowels. Even if we consider that these comparisons are made between stressed syllables and not between non-stressed syllables, the rate of declination between unstressed vowels and the following extrapods represent a relatively large (controlled) drop in pitch. It suggests that, at least for HD, the strength relation is heavily influenced by its position as domain-final. It may not make sense to discuss relative prominence between unstressed vowels and extrapods because of the conflicting prominence correlate directions. What is clear, however, is that a contrast exists. These results support the conclusion based on the across-word comparisons that the ‘Variable’ hypothesis, which predicted contrast but no consistent direction of difference, most accurately accounts for the results.  6.8  Conclusion  This chapter reveals how speakers of St’át’imcets make acoustic distinctions between different non-stressed syllables, and confirms just how complex the concept and realisation of prominence is. Speakers distinguish between unstressed and extrapod vowels using a combination of F0, duration and intensity in identical segmental, morpho-  98  If magnitude of contrast, rather than direction of contrast, are correlates of the contrast between heads and non-heads, this result would follow from van Eijk’s (1997) and Davis’ (forthcoming) analysis of extrapods as head-like elements. Such a proposal would still face the task of disentangling domain-final effects from headedness effects.  146  syntactic and intonational contexts. The direction of difference was variable, supporting a weak interpretation of the phonetics-phonology interface. Three interwoven factors affecting prominence correlates seem to emerge.99 First, while extrapods and unstressed syllables have significantly different acoustic characteristics, the direction of contrast is not consistent. This differs from the directionally constrained contrasts between heads and non-heads seen in Chapter 5. These results suggest that two types of contrast exist in St’át’imcets: direct mapping and indirect mapping. Direct mapping is evident in the correlation between phonologicallyspecified relationships and directionally constrained contrasts in head versus non-head comparisons. Such a direct mapping is not evident in the extrapod—unstressed vowel comparisons. This indirect mapping is consistent with a model in which extrapods are residual elements rather than referenceable domains. The absence of a definable phonological relationship between unstressed vowels and extrapods is correlated with an indirect mapping between the phonetics and the phonology. As such, the Variable hypothesis correctly predicts that the acoustic nature of extrapods and their relation to unstressed vowels is determined by other demands. Conflicts between utterance position and boundary/head enhancement emerged as examples of such demands. Second, morphological and syntactic constituents show different prominence correlates. The across-word distinctions of speakers and the within-word comparisons of HD showed a difference between verb+subject clitic and verb+object clitic constructions. Due to the nature of the stimuli considered here, it is not possible to tell whether it is the presence of the overt object or the lack of overt subject morphology that is affecting the overall intonation pattern. The results did not present any clear evidence to support one analysis over the other. In order to test whether morpho-syntactic constituents have different prominence correlates, comparisons between bare roots and more complex constructions must be considered. Comparisons of full phrases, as compared to phrases with non-overt morphemes should shed light on the mapping with morpho-syntactic domains in  99  It should be pointed out, that there is a distinction between production and perception, and that significant distinctions produced by speakers, may not in fact be what hearers use to perceive differences. Perception tests are left for future research.  147  St’át’imcets. The difference between person enclitics and second position enclitics and their effects on intonation would also be an interesting avenue of future research. Finally, the other conflicting pressure that emerged from the results was head and boundary maximisation. Three of the five speakers showed consistently stronger prominence cues for unstressed vowels than for extrapods across words. AP showed a conflict between foot non-head vowel reduction (hypothesised to maximise the head) and apparent word-final lengthening. When HD’s within-word results were examined, he too showed a directionality conflict between word boundary and head correlates. The increased strength associated with word-final lengthening conflicted with the decreased strength associated with the falling F0. The directionally unconstrained variation in the unstressed-extrapod contrast supports a weaker interpretation of the phonetics-phonology mappings than the directionally constrained type evidenced in head versus non-head comparisons. In addition to the above complications, heads of different prosodic levels are cued differently. Heads of feet seem to be cued by increased duration and small increases in F0 and amplitude. Heads of larger domains seem to be correlated with greater magnitude of change in F0, duration and intensity rather than greater absolute values. The question of whether secondary stress is in fact phrase-level prominence could be examined by comparing feet in longer words, such as those with three or more feet, and bare roots versus root+suffix/enclitics. The definition of strength as a relation of contrast magnitude rather than a relation between absolute values is left for future research. Another line of research to pursue would be to examine the acoustic correlates of extrapods in the absence of domain-final effects. However, it is difficult to construct examples in St’át’imcets that would allow us to explore the relative strength of prominence correlates between unstressed vowels and extrapods free of domain-final effects. St’át’imcets PWords/PPhrases are right-headed. As such, it is not possible to distinguish between domain head versus non-head prominence correlates and the effects of boundary prominence. Since the head foot is often also the final one, this is difficult to tease apart.  148  In order to test the effects of stress without implicating larger domain effects, word-internal extrapods and unstressed vowels could be considered. 100 Such contexts are possible in St’át’imcets but are difficult to construct.101 There are two potential sources for comparisons between word-internal unstressed vowels and extrapods: schwa-only roots and roots with lexical suffixes. The first potential source of non-final extrapods is in constructions of schwa-only roots+person enclitics. In these constructions, primary stress falls on the full vowel of the enclitic rather than on the root. Comparing a disyllabic root+enclitic with a trisyllabic root+enclitic would allow the comparison of a pre-stress unstressed vowel with a pre-stress extrapod. The difficulty in constructing such contexts lies in the dearth of trisyllabic roots in the language. Of the few that exist, most are actually fused disyllabic roots with a lexical suffix. Another possibility would be to consider reduplicated roots but this makes assumptions about the status of reduplicants within the PWord that have yet to be tested. The second potential source of non-final extrapods would be root+lexical suffix compounds. There is a set of ‘strong’ lexical suffixes in the language that attracts stress away from the root, meaning that it is possible to construct examples with pre-stress extrapods.102 The challenge again is finding suitable roots. As mentioned above, most of the roots in the language are monosyllabic and finding a root with suitable foot structure and consonant context would be challenging, if not impossible. Data from a language like Dutch, as presented in de Lacy (2002), could potentially be used to examine the effects of stress independent of domain-final effects. In the Dutch informal register data discussed in Chapter 3.4.6, extrapods occur in nonfinal position. As such, they are presumably free of the word-final complications seen in St’át’imcets.  100  Thanks to Pat Shaw for suggesting this. Thanks to Henry Davis for suggesting this pair: n’ásaka7xit ‘to send something to someone’ and n’asaká7em ‘to send (something)’ Lit. ‘to make something go by hand’. This pair will provide an interesting future comparison. 102 Davis (forthcoming: Chapt. 46) lists the ‘strong’ lexical suffixes as: –álhcw “location, venue”, -al’íkst “leaf, page”, -áw’s “together”, -áwlh “conveyance”, -áy’lh/-á7ilh “child, person”, -ám’ “row”, -sqáxa7 “domestic animal”, -ásq’et “day”, -íw’s “day”, ’-aszánucw “year”, -úlh “step-relative”, -alhníw’t “side of body”, -áliw’s “body” and some others. 101  149  This chapter has tested and confirmed the hypothesis that speakers of St’át’imcets would distinguish unstressed vowels and extrapods using prominence correlates. The next two chapters examine the effects of boundary strength on segments in the language. Chapter 7 tests the hypothesis that the Prosodic Strengthening effects seen in languages like English, French and Korean also apply in St’át’imcets. Chapter 8 compares the boundary strength results of unstressed syllables and extrapods to determine whether such effects also distinguish exhaustively parsed from non-exhaustively parsed vowels.  150  Chapter 7 Prosodic Strengthening in St’át’imcets  7.1  Introduction  This chapter presents the Prosodic Strengthening literature and the results of the first part of experiment 2. This part of the experiment tests the hypothesis that Prosodic Strengthening applies cross-linguistically and that speakers of St’át’imcets produce boundary strength effects in line with the results of Cho (2005). Cho’s results showed that English speakers produce more peripheral vowels on the F1/F2 plane at the boundaries of higher levels (i.e. larger domains) of the Prosodic Hierarchy. The hypotheses tested are i) that St’át’imcets domain-final vowels are more peripheral at higher boundaries, in the same way that English vowels are, and ii) that extrapods in St’át’imcets show parallel (though not necessarily identical) behaviour to unstressed vowels. In other words, just as unstressed vowels at higher boundaries are predicted to be more peripheral than those at lower boundaries, extrapods at higher boundaries are predicted to be more peripheral than those at lower boundaries. The results of this experiment are used as background for the second part of Experiment 2, which examines whether and how speakers make distinctions between unstressed vowels and extrapods using boundary strength effects. In addition to examining boundary strength effects in St’át’imcets, this chapter also reveals more about the acoustic characteristics of vowel quality in St’át’imcets. Previous research has focused primarily on the contrast between retracted and nonretracted vowels and the acoustic characteristics of schwa. The quality of unstressed vowels and extrapods has not been previously examined. The results, like those found in other languages, are vowel-specific. While speakers make foot-internal/foot-final and foot/word distinctions for /a, i, u/, speakers only make a word/phrase distinction for /u/. Results also indicate syllabification plays a  151  role in vowel quality. St’át’imcets vowels in open syllables are less peripheral than those in closed syllables. The next section discusses the Prosodic Strengthening literature.  7.2  Prosodic Strengthening Background  Much research has shown that segments at higher boundaries in the Prosodic Hierarchy are articulated more strongly than those at lower boundaries. Domain-initial and domainfinal prosodic strengthening effects have been demonstrated in several languages, including English (Pierrehumbert & Talkin 1992; Fougeron & Keating 1997; Byrd & Saltzman 1998; Byrd 2000; Cho 2002, 2004, 2005; Keating & Cho 2005), French (Fougeron 1998, 2001; Tabain 2003a, b), Korean (Jun 1995; Cho 1998; Cho & Jun 2000; Cho & Keating 2001), Japanese (Onaka 2003; Onaka et al. 2004), Dutch (Cho & McQueen 2005), Tamil (Byrd et al. 2000) and a four language comparison between English, French, Korean and Taiwanese (Keating et al. 2003). Much of the research on Prosodic Strengthening has focused on articulation, including the seminal electro-palatograph (EPG) study by Fougeron and Keating (1997), which pioneered this line of research. Another type of articulatory methodology used is electromagnetic articulography (EMA) (Byrd & Saltzman 1998; Byrd 2000; Byrd et al. 2000; Tabain 2003b). Considerably less research has examined the acoustic correlates of prosodic strengthening. Several potential acoustic cues to prosodic strengthening have been examined, including VOT (Jun 1995; Cho & Keating 2001), aspiration (Pierrehumbert & Talkin 1992) acoustic nasal energy (Cho & Keating 2001), glottalisation of vowels (Byrd & Saltzman 1998), formant rate of change (Tabain, 2003a; Wouton & Macon 2002), spectral tilt rate of change (Tabain 2003a), and static F1, F2 (Cho 2005). In general, Prosodic Strengthening results support a weak interpretation of the phonetics-phonology mapping rather than a strong one. Results are speaker- and/or boundary- and/or segment-specific. In Fougeron and Keating’s (1997) results, no boundary was distinguished by all speakers, and no speaker distinguished all boundaries. Keating (2006) cites Onaka (2003), Onaka et al. (2004) and Kim (2001) as studies which found speaker- and vowel-specific results. In the case of Japanese, Onaka (2003) and  152  Onaka et al (2004) found differences in the distinctions made by speakers between /n/ and /t/. Kim (2001) found differences between two Korean sibilants (/s/ and /s*/). Fougeron (1998, 2001) found speakers made fewer distinctions for /s/ and /t/ than for /n, k, l/ in French. These results suggest that Prosodic Strengthening is implemented by language-specific phonetics rather than a universal system. In spite of the variable results, the direction of contrast is consistent: segments at the higher boundary are always articulated more strongly than those at the lower boundary, not vice versa. This leads Keating (2006) to refer to Prosodic Strengthening as a trend rather than an absolute pattern, and as such, the results reflect a weak rather than strong mapping between phonetics and phonology. Cho (2005) investigates the connection between phonological features and phonetic realisation at domain boundaries and under accent. He measures both articulation and acoustics, assuming that a change in articulation would be mirrored in the acoustics. In particular, he examined the effect on a vowel’s position on the F1/F2 plane. This experiment tested the hypothesis that vowels (in this case /a/ and /i/) under accent would maximise sonority (both should lower), while vowels at higher boundaries (word versus intermediate phrase versus IP) would maximise phonemic contrasts (/a/ should lower and /i/ should raise). Cho’s domain-final boundary results confirm that the English speakers’ vowel position on the F1/F2 plane became more peripheral in both dimensions (i.e. height and frontness) the higher their adjacent boundary was in the Prosodic Hierarchy. Results indicated that /i/ was consistently higher and fronter at higher boundaries and /a/ was consistently lower and backer. These results were interpreted as evidence of positional prominence. A vowel’s phonological features are maximised at higher boundaries, i.e. in positions of increased prosodic prominence. Cho (2005) did not consider /u/ but cites de Jong (1995), in which it was shown that /u/ under accent is backer than its non-accented (i.e. less prosodically strong) counterpart. If we also assume that /u/ is characterised by [+bk], we can extend Cho’s hypothesis to predict that /u/ will be further back at higher boundaries. One point which must be considered before specific predictions for St’át’imcets are laid out is that Cho tested English vowels, for which position on the F1/F2 plane and 153  phonological features are generally well understood (Chomsky & Halle 1968; Hillenbrand et al. 1995, Kent & Read 2002; Peterson & Barney 1952). While it is relatively clear that /a/ in English is a low back vowel, its position in St’át’imcets is not. Van Eijk (1997: 3) says that /a/ is realised ‘broadly’ as /ε/, whereas Shahin (2002) argues that the underlying vowel is in fact /æ/. The production of /a/ also varies by dialect, with Southern speakers tending to produce it backer than Northern speakers (H. Davis p.c.).103 For the purposes of this experiment, the vowel features assumed are Shahin’s (2002) categorisation of surface vowels: /i/ is [+high] and [+front], /u/ is [+high] [+back], and /a/ is [+low]. In terms of where vowels are located on the vowel plane, only two previous acoustic studies on vowel quality have been conducted: Bessell (1997) and Shahin (2002). Bessell (1997) examined the formant values of stressed vowels from four Northern St’át’imcets speakers (three women, one man). Approximate F1 and F2 ranges are given below. Table 7.1 Bessell (1997) approximate formant values Female  Male  F1 (Hz)  F2 (Hz)  F1 (Hz)  F2 (Hz)  a  750-1100  1500-2000  700-900  1100-1700  i  250-450  2500-3000  250-350  2100-2400  u  250-600  750-1500  250-500  700-1200  Shahin (2002) measured values F1/F2 values for two male speakers—one Northern and one Southern. She found the following means for full vowels:  103  In this chapter and the next, I use the terms ‘fronter’ and ‘backer’ following Cho (2005) rather than ‘further front’ and ‘further back’.  154  Table 7.2 Shahin (2002: 224-226) formant value means Speaker 1  Speaker 2  F1 (Hz)  F2 (Hz)  F1 (Hz)  F2 (Hz)  æ  641  1562  653  1658  i  315  1910  286  2329  u  381  1053  361  991  The results of the two experiments above show a range of variation in formant values across speakers. It is likely that such variation will also emerge from the results of this experiment. If we apply Cho’s results to the extrapod model, we predict that these speakers should show similar results for both unstressed vowels and extrapods. Unstressed vowels should follow a peripherality hierarchy as follows: Utt>IP>PPh>PWd>Ft. Unstressed vowels at the IP level should be more peripheral than those at PPh, etc. Extrapod vowels should show results that follow the peripherality hierarchy as well. In other words, the fact extrapods lack a foot boundary should not affect the direction of difference. As to the magnitude of the effect, conceivably extrapods will show a weaker effect than unstressed vowels due to the lack of a foot boundary. Given the feature specifications above, peripherality is defined along two dimensions: height (F1) and frontness (F2). In this experiment, peripherality is defined as being more extreme in height or frontness, so that /i/ is considered more peripheral if F1 is lower and/or F2 is higher; /a/ is more peripheral if F1 is higher, and /u/ is more peripheral if F1 and/or F2 is lower. As such, /i/ should be higher and fronter on the F1/F2 plane in IPs than in PWords and so on; /a/ should be lower; and /u/ should be higher and backer. Following previous Prosodic Strengthening research, it is expected that results will support a weak rather than a strong phonetics-phonology mapping. Predictions are summarised in the tables below:  155  Table 7.3 Unstressed vowel predictions (fi=foot internal, ff=foot final, wd=word final, ph= phrase final, > indicates the value on the left is predicted to be greater than that on the right, < indicates that the value on the left is predicted to be less than that on the right)  a i  u  F1  fi < ff  ff < wd  wd < ph  F1  fi > ff  ff > wd  wd > ph  F2  fi < ff  ff < wd  wd < ph  F1  fi > ff  ff > wd  wd > ph  F2  fi > ff  ff > wd  wd > ph  Table 7.4 Extrapod predictions a i  u  F1  wd < ph  F1  wd > ph  F2  wd < ph  F1  wd > ph  F2  wd > ph  If these predictions are borne out, we have increased the empirical basis for Prosodic Strengthening and have shown that St’át’imcets speakers implement boundary effects in the same way English speakers do. If these predictions are not borne out, we have further evidence that Prosodic Strengthening is implemented through languagespecific phonetics.  7.3  Methodology  The methodology presented in this section is the same as that used for the part of Experiment 2 presented in Chapter 8. Table 7.5 presents the speakers and the total number of tokens used for the results reported in this chapter and Chapter 8 as well as in Appendices E and F.  156  7.3.1 Subjects Four fluent speakers of St’át’imcets participated in this experiment: AP, CA, HD and LT. Their information is presented in detail in Chapter 4. A summary of their dialects and tokens included and excluded are presented in the table below: Table 7.5 Speaker and token information Speaker Dialect AP CA HD LT  Number of tokens used (excluded) 104  Northern Northern Lower Southern Southern  423 (14) 447 (42) 434 (39) 405 (6)  Data were excluded for several reasons: measurement error (as determined in SPSS as being 3 standard deviations away from the mean in boxplots), misspeaking (including complete deletion of the token), or being preceded or followed by a pause. 7.3.2 Stimuli Tokens were selected from van Eijk (1987, 1997) and Davis (forthcoming) and constructed into target sentences with the help of H. Davis (p.c.). Cho (2005) showed that results in English were stronger and more consistent domain-finally than domaininitially and that only the segment immediately adjacent to the boundary showed any effects. Since this is the first experiment examining boundary effects in St’át’imcets, domain-final tokens only were tested. This is based on the assumption that, parallel to English, they would be most likely to show possible effects. The number of vowel-final roots in the language is extremely limited, so root+affix/clitic constructions were constructed for comparisons above the foot level.105 Specifically, tokens from the possessive paradigm were used. Tokens for /i/ are /-i/, the  104  Numbers include all tokens used for comparisons in this chapter, Chapter 8 and in Appendices E and F. A number of vowel-final roots do exist, though it could be argued that the vowel is in fact a glide or /h/. For example áma, alternates with amh- (Davis p.c.; Mudzingwa 2007). 105  157  3pl. possessive. Tokens for /u/ are /-su/, the 2sg. possessive106. The existential enclitic /=a/ following the 3sg. possessive suffix /-s/ is used for the /a/ comparison. It is assumed that suffixes and enclitics reflect Prosodic Strengthening in the same way. It is also assumed that suffixes and enclitics are part of the same domain but it is possible that results in which /a/ patterns differently from /i/ and /u/ might arise. Given that /=a/ is a second-position enclitic rather than a person suffix/clitic, such a difference in results would be interesting in terms of the mapping between morpho-syntax and prosody.107 As discussed in Chapters 1 and 5, whether root+affix/clitic constructions are PWords or PPhrases is not clear. This distinction is not crucial for this experiment. The prediction is that IP-final vowels will be more peripheral than those in smaller domains, so whether that domain is a PWord or PPhrase is not crucial. For simplicity’s sake, IPfinal tokens are referred to as ‘phrase-final’ and the non-IP-final tokens as ‘word-final’. The positions of the unstressed vowels compared are foot-internal (fi), foot-final (ff) word-final (wd) and phrase-final (ph). The structures used are given below: Table 7.6 Unstressed syllables across boundaries (In this and subsequent tables α=stressed syllable and β=unstressed syllable)  Foot-internal  Foot-final  Word-final  Phrase-final  Ph / | \ W | Ft /\ α βC  Ph / | \ W /\ Ft \ /\ \ α β Cv  Ph / | \ W | Ft /\ α β  Ph / | W | Ft /\ α β  As mentioned above, domain-final extrapods are predicted to show parallel effects to unstressed syllables: phrase-final extrapods will be more peripheral than word-final extrapods. The extrapod structures are presented in Table 7.7 below:  106  Matthewson (1994) proposes /sw-/ as the underlying representation. She argues that the surface realisation is [sәw], though vocalisation of the glide to /u/ is the other possibility (H. Davis, p.c.). 107 Recall that person suffixes/enclitics and second position enlicitcs are hypothesised to be in different subPWord constituents based on stress facts (van Eijk, 1997; Davis, forthcoming).  158  Table 7.7 Final extrapods across boundaries (α= stressed vowel, β= unstressed vowel, γ=extrapod)  Word-final Ph / | \ W /\ Ft \ /\ \ α β γ  Phrase-final Ph / | W /\ Ft \ /\ \ α β γ  Vocabulary, word order and meaning were confirmed with speakers before the first session. Due to lexical differences between dialects, I could not find the relevant lexical items for a complete vowel paradigm, so only /i/ and /u/ were elicited from Southern dialect speakers while /a/ and /u/ were elicited from Northern speakers. As with the experiment reported in Chapter 5, one of the limitations of this study is the lack of minimal pair segmental contexts for all comparisons. Every effort was made to find as similar segment, morpho-syntactic and intonation contexts as possible, as well as targets that were easy to represent graphically. As a result, foot-internal and foot-final tokens are the same noun root, with the foot-final context being constructed by adding the existential enclitic /=a/. For example, the token used to elicit foot-internal unstressed /i/ was [(káliæ)] and the token used to elicit foot-final /i/ was [(káli)æa]. In the first case, the /i/ is in a closed syllable while in the second it is at the foot boundary in an open syllable. Word-final and Phrase-final comparisons are made between the same morphemes in different phrasal positions. Sample tokens are given in Table 7.8:  159  Table 7.8 Tokens108 (=foot boundary, [] = NPA transcription  /a/ twan [twan] ‘salmonberry’ smúlhats [(ßmú¬aæ)]  /i/ tsitcw [æitx∑] ‘house’ kálits [(káliæ)]  woman  ‘carrot’ i kálitsa [i (káli)æa]  foot-final extrapod wordfinal/phrase-final unstressed  ta smúlhatsa [ta (ßmú¬a)æa] ‘the/a woman’ N/A qmút.sa [(qmútßa)]  qmúti [(qmúti)]  ‘his/her hat’  ‘their hat’  wordfinal/phrase-final extrapod  músmustsa [(múßmuß)æa] ‘his/her cow’  músmusi [(múßmu)ßi]  stressed vowel foot-internal unstressed foot-final unstressed  ‘the carrots’ N/A  ‘their cow’  /u/ sútik [ßútik] ‘winter’ xúsum [(⋲∑úßum)] ‘soapberry ti/ta xúsuma [ti/ta (⋲∑úßu)ma] the soapberry’ N/A qmút.su [(qmútßu)] ‘your hat’ músmustsu [(múßmuß)æu] ‘your cow’  In addition to the tokens in Table 7.8, unstressed schwas were elicited in a second stage of the experiment. The word elicited from all speakers was “sécsec” (silly/foolish/crazy). These were recorded in order to provide a central landmark for relative peripherality, along with the stressed vowels. 7.3.3 Equipment All tokens were recorded using a Marantz solid-state pmd660 recorder, on mono channel, at pcm 44.1 k. and resampled at 16000 Hz. The full vowels (/a/, /i/, /u/) were recorded from all speakers using an Audio-technica ATM75 head-mounted cardioid condenser microphone. Schwas were recorded later using a Shure headworn condenser microphone WH30XLR.  108  The phonetic transcriptions in this paper are all broad. /a/ is often pronounced /å/ or /´/, and /u/ sometimes /o/. See van Eijk (1997) and Namdaran(2006) for further information on the narrow transcriptions of the vowels. See also Appendix A for orthography-NAPA conversion chart.  160  7.3.4 Recording All recordings were conducted in private homes. Tokens were put into sentences that were optimised for phrase-level intonation and segmental context. It was not possible to construct minimal pairs across all boundaries. As with the segmental contexts for tokens mentioned above, target sentences were selected to maximise similarities in the place and manner of token-adjacent segments. Target sentences also took into account dialect differences of word order and vocabulary. As a result, the speakers of the two dialects produced in slightly different sentences. Sample target sentences are given below.  161  Table 7.9 Northern dialect target sentences109 /a/ stressed vowel Tsut sLisa “twan” inátcwas. ‘Lisa said “salmonberry” yesterday.’ foot-internal us  foot-final  word-final us  Tsut sLisa “smúlhats” inátcwas. ‘Lisa said “woman” yesterday.’ Ít’em ta smúlhatsa inátcwas. ‘The woman sang yesterday.’ N’áscit ta qmút.sa (kw)s110 Henry. ‘Give his/her hat to Henry.’  /u/ Wá7lhkalh ít’em lhas sútik múta7 lhas pipántsek. ‘We sing in the winter and in the summer.’ Tsut sLisa “xúsum” inátcwas. ‘Lisa said “xusum” yesterday.’ T’ec ti xúsuma múta7 i q’welápa. ‘The xusum and the strawberries are sweet’. N’áscit ku qmút.su (kw)s Henry. ‘Give one of your hats to Henry.’  word-final ex  N’áscit ta músmustsa (kw)s Henry. ‘Give his/her cow to Henry.’  phrase-final us  Síma7cits ta qmút.sa. Síma7cits ku qmút.su. Cw7it I Ámhasan’. qmút.swa. ‘Give me his/her hat. It looks ‘Give me one of your hats (which nice’. I think you might have in your bag). You have lots of hats.’ Síma7cits ta músmustsa. Síma7cits ku músmustsu. Cw7it i Ámhasan’. músmus.tswa. ‘Give me his/her cow. It ‘Give me one of your cows (which looks nice.’ I think might be behind the tree). You have lots of cows.’ Tayt múta7 sécsec ta sqáycwa. The man is hungry and a little crazy/foolish.  phrase-final ex  Schwa  N’áscit ku músmustsu (kw)s Henry. ‘Give one of your cows to Henry.’  109  All characters in this experiment are fictitious, and any resemblance to actual professors in the linguistics department is purely coincidental. Sentences are given in the orthography. 110 Northern speakers most often abbreviated /kws/ to /s/, while Southern speakers abbreviated it to /k/ or /ks/.  162  Table 7.10 Southern dialect target sentences /i/ stressed syllable Tsut sLisa “tsitcw” inatcwas. ‘Lisa said ‘house’ yesterday.’ foot-internal us foot-final us  word-final us  word-final ex  phrase-final us  phrase-final ex  Schwa  Tsut sLisa “kálits” inátcwas. Lisa said ‘carrot’ yesterday.’ Xzum i kálitsa múta7 i qáwtsa. ‘The carrots and the potatoes are big.’ Úm’en ku pála7 qmúti k(ws) Henry. ‘Give one of their hats to Henry.’ Úm’en ku pépla7 músmusi k(ws) Henry. ‘Give one of their cows to Henry.’ Úm’ents ku pála7 qmúti. Cw7it i qmútiha. ‘Give me one of their hats. They have lots of hats.’ Úm’ents ku pépla7 músmusi. Cw7it i musmusíha. ‘Give me one of their cows. They have lots of cows.’  /u/ Wá7lhkalh ít’em lhas sútik múta7 lhas pipántsek. ‘We sing in the winter and in the summer.’ Tsut sLisa “xúsum” inátcwas. ‘Lisa said ‘xusum’ yesterday.’111 T’ec ti xúsuma múta7 i q’welápa. ‘The xusum and the strawberries are sweet.’ Úm’en ku pála7 qmút.su k(ws) Henry. ‘Give one of your hats to Henry.’ Úm’en ku pépla7 músmustsu k(ws) Henry. ‘Give one of your cows to Henry.’ Úm’ents ku pála7 qmút.su. Cw7it i qmút.swa. ‘Give me one of your hats. You have lots of hats.’ Úm’ents ku pépla7 músmustsu. Cw7it i músmustswa. ‘Give me one of your cows. You have lots of cows.’  Q’7ál’men múta7 sécsec ti sqáycwa. ‘The man is hungry and a little crazy’  Twenty-two repetitions of each target sentence were randomised and presented to speakers. Recording sessions were no longer than 1.5 or 2 hours with breaks every 10-20 slides and longer breaks every half hour. Most speakers (except LT) did not finish the  111  Zánucw~zánucwem were also elected from speakers. Zánucwem was not an acceptable token for Northern dialect speakers, and Southern speakers produced it with a closed second syllable (the secondary labialisation acting as onset for the following syllable i.e. (zánuc-wem). As a result, the /u/ was no longer foot final, and this comparison was omitted.  163  experiment in one session. In these cases, tokens and target sentences were balanced across sessions, so that every comparison was made on every day112. The next section lays out how formant measurements were made. 7.3.5 Measurements and analysis Measurements were taken at the vowel midpoint, even for the vowel /u/, which often showed a steady state later in the vowel. The decision to measure consistently at midpoint was taken in order that comparisons with previous experiments could be made and also because measurements later in the vowel were not possible due to the short duration of most tokens. Praat requires a window length of .025 on either side of the measurement, which proved too long for the majority of vowels to be measured anywhere but midpoint.113 Given that vowels followed a palatal fricative, the measurements may have been taken from a lower point than the ultimate trajectory of the /u/. More research on tokens that have longer vowels may shed more light on this. More information on measurement and analysis is given in Chapter 4. Formant values were obtained by a Praat script. As in the experiment reported in Chapters 5 and 6, no inter-vowel or inter-speaker statistical comparisons were made. Results were analysed in SPSS using independent sample t-tests, following Koch (2008) and the reasoning laid out in Chapter 4. For unstressed vowels, F1 and F2 were measured for three comparisons: •  foot-internal/foot-final  •  foot/word-final  •  word/phrase-final For extrapods, only word/phrase comparisons were made. Bonferroni corrections  for multiple comparisons were applied resulting in p≤.008 for the unstressed data and p≤ .0125 for the extrapod data, marked **. A marginally significant p≤.05 is marked *. Two further notes on speaker measurements bear mentioning here. First, CA and to some extent also AP and HD, used breathy voicing in phrase-final position. Vowel 112  With the exception of schwa since those tokens were recorded later. Marking vowel onset earlier and offset later would have increased the window over which calculations could have been made (E.Vatikiotis-Bateson p.c.). 113  164  length was measured to include this breathy voicing. Second, I noticed a pattern of very low F3 for these speakers. Since neither of these is the focus of the study, these two interesting generalisations are mentioned here but left for future research.  7.4  Results  This section presents the results of the experiment, reporting first on unstressed vowels and then on extrapods. A discussion directly follows the results in both sections. 7.4.1 Domain-final unstressed vowels This section presents the results for the comparisons of unstressed syllables across Prosodic Hierarchy boundaries. The first four tables present means and standard deviation. Individual speakers and their significant differences are discussed separately below: Table 7.11 AP’s unstressed vowel means (In this and subsequent tables, numbers in parentheses represent standard deviation.)  a N foot-internal foot-final word-final phrase-final  12 13 21 18  F1 Mean (Hz) 627 (175) 473 (76) 594 (31) 614 (52)  Table 7.12 CA’s unstressed vowel means a N F1 Mean (Hz) foot-internal 12 370 (20) foot-final 17 286 (23) word-final 17 326 (49) phrase-final 22 330 (62)  u F2 Mean (Hz) 1673 (381) 1650 (87) 1653 (37) 1639 (47)  N 17 21 32 23  F1 Mean (Hz) 426 (41) 440 (32) 358 (21) 363 (24)  F2 Mean (Hz) 1181 (130) 1267 (106) 1029 (82) 1054 (77)  u F2 Mean (Hz) 1822 (31) 1858 65) 1807 (55) 1824 (76)  N 14 19 19 21  F1 Mean (Hz) 296 (12) 295 (22) 282 (27) 266 (27)  F2 Mean (Hz) 1032 (83) 1217 (196) 1193 (166) 1038 (134)  165  Table 7.13 HD’s unstressed vowel means i N F1 F2 Mean (Hz) Mean (Hz) foot-internal 16 338 (31) 2091 (183) foot-final 19 388 (34) 1839 (193) word-final phrase-final  21 13  326 (27) 323 (36)  Table 7.14 LT’s unstressed vowel means i N F1 Mean (Hz) foot-internal 13 380 (22) foot-final 11 411 (28) word-final 19 334 (16) phrase-final 17 348 (19)  1957 (113) 1931 (303)  u N 19 6  F1 Mean (Hz) 374 (47) 292 (29)  F2 Mean (Hz) 1365 (141) 1174 (71)  13 25  409 (34) 423 (33)  1390 (121) 1273 (114)  114  u F2 Mean (Hz)  N  F1 Mean (Hz)  F2 Mean (Hz)  2414 (48) 2192 (119) 2542 (115) 2502 (148)  14 19 20 20  504 (55) 453 (35) 410 (28) 401 (37)  1578 (58) 1405 (142) 1703 (153) 1514 (147)  For a clear representation of the significant differences based on the results above, each speaker’s significant results are presented in tabular and scattergram form below. There are two scattergrams per vowel. In order to situate results, in the first scattergram, stressed vowels and schwas are compared to unstressed vowels as a whole. In the second scattergram, unstressed vowels across boundaries are compared. The first speaker’s results discussed are AP’s. Table 7.15 AP’s unstressed vowel significant differences (In this and subsequent tables, - indicates a significant difference in the opposite direction of the prediction. Shaded cells indicate no significant difference. 1 indicates a marginal difference.)  a  fi-ff ff-wd wd-ph  u  F1 F2 F1 F2 t-value p-value t-value p-value t-value p-value t-value p-value -6.5 <.001 -2.3 .031 11 <.001 10 <.001 9 <.001  This table shows that AP makes significant foot-internal/foot-final and foot-final/wordfinal distinctions but not word-final/phrase-final distinctions. It also shows the predicted 114  HD tended to delete the vowel entirely in foot-final position, resulting in lower than normal numbers.  166  vowel-specific effects, in that F2 distinctions are made for /u/ not /a/. These comparisons are demonstrated in the following scattergrams. The first figure shows the differences in position between stressed /a/, unstressed /a/ and schwa: Figure 7.1 AP’s schwa, stressed /a/ and unstressed /a/  The figure above shows that AP’s stressed /a/ has greater distribution variation than schwa and unstressed /a/. Unstressed /a/ lies lower than schwa, whereas stressed /a/ appears backer. The unstressed /a/ tokens represented here are considerably higher than Bessell’s (1997) results. The next figure illustrates unstressed vowels across boundaries: unstressed footinternal /a/ is **lower than foot-final /a/ and foot-final /a/ is ** higher than word-final /a/.  167  Figure 7.2 AP’s unstressed /a/  In the figure above, foot-internal /a/ is more peripheral and shows wider distribution than foot-final /a/, which is quite centralised. Word-final /a/ is more peripheral than foot-final /a/. The next figure shows AP’s schwa and stressed and unstressed /u/. Unstressed /u/ is in the F1 range of Bessell’s findings while the stressed vowel is considerably fronter. The position of the stressed vowel between the more peripheral unstressed vowel and schwa is unexpected.  168  Figure 7.3 AP’s schwa, stressed /u/ and unstressed /u/  Figure 7.4 shows that for AP, foot-internal /u/ is marginally backer than foot-final /u/ and foot-final /u/ is **lower and **fronter than word-final /u/: Figure 7.4 AP’s unstressed /u/  The figure above again demonstrates an overlap of word and phrase-final tokens. It also shows that foot-internal /u /is marginally more peripheral than foot-final. Foot-internal and foot-final tokens are distinctly lower and fronter than word- and phrase-final tokens.  169  Turning to CA’s results, we see he does make a distinction between word-final and phrase-final tokens but only for /u/. Table 7.16 shows CA’s results: Table 7.16 CA’s unstressed vowel significant differences a u F1 F2 F1 F2 t-value p-value t-value p-value t-value p-value t-value p-value fi--ff -10 <.001 -4 .001 1 ff--wd 3 .006 2.5 .018 wd--ph 3.3 .002 Table 7.16 shows that CA makes a distinction between all boundary types: footinternal/foot-final and foot-final/word-final for /a/ and foot-internal/foot-final and word/phrase-final for /u/. He makes only a frontness distinction for /u/. Figure 7.5 compares CA’s schwa and stressed and unstressed /a/. Figure 7.5 CA’s schwa, stressed /a/ and unstressed /a/  Unlike AP’s /a/, CA’s vowels show considerable overlap. Given its position, the stressed /a/ also appears to be produced more as an /ε/ than /a/. Both stressed and unstressed /a/ show a highly variable distribution. CA’s /a/ results are considerably higher and fronter than both Bessell’s (1997) and Shahin’s (2002) results. This may be attributed to the different consonantal contexts in which the vowels were elicited, or to dialectal or speaker differences.  170  The next figure shows that CA’s unstressed foot-internal /a/ is **lower than footfinal /a/ and foot-final /a/ is **higher and marginally fronter than word-final /a/. Figure 7.6 CA’s unstressed /a/  Figure 7.6 shows a wide range of token distribution and the overlap of word- and phrasefinal tokens. Foot-internal /a/ is more peripheral than foot-final, and foot-final /a/ is less peripheral than word-final /a/. Interestingly, word- and phrase-final tokens are produced lower than both the stressed vowel and schwa. Figure 7.7 illustrates CA’s schwa, stressed /u/ and unstressed /u/:  171  Figure 7.7 CA’s schwa, stressed /u/ and unstressed /u/  This figure shows a clear demarcation between stressed /u/ and schwa on the one hand and unstressed /u/ on the other. As with AP’s /u/, the stressed vowel is much closer to schwa than the unstressed ones. The stressed vowel results are more in line with Bessell’s findings than Shahin’s in terms of height but considerably fronter than both. The figure below illustrates that CA’s foot-internal /u / is **backer than foot-final /u/ and wordfinal/u/ is **fronter than phrase-final /u/. Figure 7.8 CA’s unstressed /u/  172  Figure 7.8 shows once again that foot-final /u/ is less peripheral than foot-internal /u/. Word-final /u/ is also significantly fronter (i.e. less peripheral) than phrase-final. Turning now to HD’s results, Table 7.17 shows the significant differences for unstressed vowels across boundaries: Table 7.17 HD’s unstressed vowel significant differences i u F1 F2 F1 F2 t-value p-value t-value p-value t-value p-value t-value fi-ff -4.5 <.001 -4 <.001 4 .001 4 1 ff-wd 6.5 <.001 2 .022 -7.2 <.001 -4 wd-ph 2.9  p-value <.001 .001 .006  The results in Table 7.17 show that HD makes strong distinctions between foot-internal and foot-final vowels as well as between foot-final and word-final vowels but an F2 distinction only for word- and phrase-final /u/. The figure below shows HD’s schwa and stressed and unstressed /i/: Figure 7.9 HD’s schwa, stressed /i/ and unstresed /i/  This figure illustrates the overlap of stressed and unstressed /i/ but shows that both are generally higher than schwa. Unstressed /i/ shows a much wider distribution than either the stressed vowel or schwa. Results are in line with Shahin’s results (though  173  backer than Bessell’s male Upper speaker). The distribution of unstressed /i/ tokens is illustrated in Figure 7.10 below. Figure 7.10 shows that HD’s unstressed foot-internal /i/ is **higher and **fronter than foot-final /i/ and foot-final /i/ is **lower and marginally backer than word-final /i/. Figure 7.10 HD’s unstressed /i/  Figure 7.10 shows again that foot-final tokens are less peripheral than foot-internal tokens. It also shows that foot-final tokens are less peripheral than word-final tokens but no difference between word and phrase-final tokens was made. The next figure shows HD’s schwa and stressed and unstressed /u/.  174  Figure 7.11 HD’s schwa, stressed /u/ and unstressed /u/  The figure above shows that HD’s /u/ patterns differently from both AP’s and CA’s. While stressed and unstressed /u/ overlap, there is separation between the full vowels and schwa that the Northern dialect speakers did not demonstrate. The range of distribution of the full vowels is greater than schwa. The /u/ tokens are considerably backer, and in some cases higher than schwa but fronter than in Bessell’s and Shahin’s results. Figure 7.12 shows HD’s foot-internal unstressed /u/ is **lower and **fronter than foot-final /u/, foot-final is **higher and **backer than word-final /u/, and word-final /u/ is **fronter than phrase-final /u/.  175  Figure 7.12 HD’s unstressed /u/  Figure 7.12 shows that HD’s foot-final /u/, most often deleted, appears very high and back and so more peripheral than both foot-internal and word-final. Word-final /u/ is fronter (i.e. less peripheral) than phrase-final /u/. Finally, let us consider the results of the other Southern dialect speaker, LT. Table 7.18 shows LT’s results for unstressed vowels across boundaries. Table 7.18 LT’s unstressed vowel significant differences i u F1 F2 F1 F2 t-value p-value t-value p-value t-value p-value t-value p-value fi-ff -3 .007 -5.8 <.001 2.8 .011 4.7 <.001 ff-wd 8.2 <.001 8 <.001 4.3 <.001 -6.3 <.001 wd-ph -2.5 .0181 4 <.001 In Table 7.18 we can see that, like HD, LT makes distinctions for vowels across all boundaries. Particularly strong are the F2 results for /u/. Comparisons of schwa, unstressed and stressed /i/ can be seen in Figure 7.13 below:  176  Figure 7.13 LT’s schwa, stressed /i/ and unstressed /i/  Figure 7.13 illustrates LT’s high front position of the full vowels with respect to schwa. As we see in the figure below, the tokens closest to the schwa are the foot-final unstressed vowels. There is overlap between the stressed and unstressed vowels. In comparison with Bessell’s results, LT’s /i/ is slightly backer but in the same height range. Figure 7.14 shows LT’s unstressed foot-internal /i/ is **higher and **fronter than foot-final, foot-final is **lower and **backer than word-final, word-final is **higher than phrase-final.  177  Figure 7.14 LT’s unstressed /i/  Again, foot-final tokens are considerably less peripheral than foot-internal. Foot-final /i/ is less peripheral than word-final /i/ but word-final /i/ is more peripheral than phrase-final /i/. The next figure shows LT’s schwa and stressed and unstressed /u/: Figure 7.15 LT’s schwa, stressed /u/ and unstressed /u/  178  From this figure we see overlap of stressed and unstressed vowels, and a less clear distinction between schwa and the full vowels. We can see that stressed /u/ is backer but not higher than schwa. LT’s results are considerably fronter than Bessell’s. Figure 7.16 shows LT’s unstressed foot-internal /u/ is marginally lower and **fronter than foot-final, foot-final is **lower and **backer than word-final, and wordfinal is **fronter than phrase-final. Figure 7.16 LT’s unstressed /u/  This figure again shows that foot-final tokens are backer (more peripheral) than footinternal, which are also quite low. Word-final /u/ is significantly less peripheral than foot-final and phrase-final /u/, overlapping to some extent with the stressed vowel and schwa seen in the figure above. In order to understand the global results of all speakers, results across speakers are summarised in the table below:  179  Table 7.19 Summary of results Northern dialect a  ft internal < ft final ft final < wd final wd final < ph final  i  ft internal < ft final ft final < wd final wd final < ph final  u  ft internal < ft final ft final < wd final wd final < ph final  Different? How? Different How? Different? How?  AP  CA  Yes fi>ff Yes ff<wd No N/A  Yes fi>ff Yes ff<wd No N/A  Different How? Different? How? different? How? Different? How? Different? How? Different? How?  Yes fi>ff Yes ff<wd No N/A  Yes fi>ff No N/A Yes wd<ph  Southern dialect HD  LT  Yes fi>ff Yes ff<wd No N/A  Yes fi>ff Yes ff<wd Yes wd>ph  Yes ff>fi Yes ff>wd Yes wd<ph  Yes ff>fi Yes ff<wd Yes wd<ph  From the table above we make three generalisations: i) speakers make a distinction in vowels at each level, except /a/ at the word/phrase level; ii) foot-internal vowels were more peripheral than foot-final ones, contrary to prediction; and iii) in most other cases, speakers made a distinction that was consistent with Prosodic Strengthening results. In other words, segments at higher boundaries are more peripheral than those at lower boundaries. These results are discussed in detail in the following section. 7.4.2 Unstressed vowel discussion If we consider the results above on a dialectal basis, we see that for Northern speakers, foot-final vowels were consistently less peripheral than foot-internal ones. This suggests a closed versus open-syllable effect whereby vowels in open syllables reduce to (almost) schwa in terms of vowel quality. Both CA and AP show significant differences between foot-final and word-final /a/, fulfilling the directional prediction in terms of height. Neither speaker shows any significant differences between word- and phrase-final /a/. Both speakers made foot180  internal/foot-final distinctions for /u/ but only one other distinction, which was not the same for both speakers. In terms of vowel quality, both Northern speakers showed results for /a/ in low central position (as compared to schwa), supporting Shahin’s (2002) characterisation of /a/ as a low central vowel. Southern speakers showed the same foot-internal/foot-final effects for /i/ that Northern speakers showed for /a/, namely that foot-internal vowels were more peripheral than foot-final vowels. The opposite effect was observed for /u/. Both Southern speakers made the same word/phrase-final distinctions for /u/, confirming the prediction that segments at higher boundaries (i.e. IP) would be more peripheral than those at lower boundaries (i.e. word). Both had significant differences in /u/ at all levels. The prediction that St’át’imcets speakers would show similar Prosodic Strengthening results to those found in Cho (2005) was not strictly supported. The results were both speaker and vowel-specific, paralleling other cross-linguistic Prosodic Strengthening results. Taken as a group, three generalisations regarding the results can be made: i) speakers made contrasts that are consistent with Prosodic Strengthening, as predicted; ii) in the one exceptional case foot-internal vowels were consistently more peripheral than foot-final ones; and iii) an asymmetry between /u/ on the one hand and /a/ and /i/ on the other emerged at the word/phrase level. First, all speakers made significant distinctions for at least two of the three boundaries for each vowel tested. The fact that not all speakers made distinctions at all boundaries is consistent with previous Prosodic Strengthening research. Also consistent with that research is that when speakers did make a distinction, the direction of contrast was generally consistent with Prosodic Strengthening. That is, segments at higher boundaries were more peripheral than those at lower boundaries. Second, all speakers showed a consistent exception to the first generalisation in a peripherality effect associated with syllable structure. Speakers demonstrated more peripheral foot-internal (closed syllable) /a/ and /i/ than foot-final (open syllable) /a/ and /i/, and Southern speakers showed this distinction for /u/ as well. This exception to the directional Prosodic Strengthening effects could be the result of head enhancement (or foot non-head reduction) in open syllables (e.g. Crosswhite 1999; de Lacy 2002, 2006;  181  Bye 2007a,b; Bye & de Lacy 2008). Bye (2007a) analyses Finno-Saamic as a language in which open syllables are host to vowel reduction. He proposes foot-head enhancement as a possible underlying motivation. This strategy would parallel the phonological foot non-head reduction discussed in Chapter 3.5.1. Finally, word/phrase contrasts showed an asymmetry between /u/ on the one hand and /a/ and /i/ on the other. It is unclear why only /u/ showed the expected Prosodic Strengthening effects at the word/phrase level. One explanation might lie in the methodology chosen. Cho (2005) found an imperfect mapping between his articulatory results and accompanying acoustic results, so it may be the case that an articulation study would show clearer results. One complicating factor for /u /is that changes in F2 can also be an indication of roundness. The other possible explanation is the position of measurement. As previously mentioned, measurements were taken at the midpoint of the vowel, which means that formants were possibly still in transition. It may be that wordand phrase-final segments have different rates of formant change and that this is reflected in the observed differences. Future research using video recordings could shed light on whether and how rounding is used by speakers to signal boundary strength. Examination of formant rate of change would also be helpful. While not the focus of this thesis, vowel quality, particularly for the unstressed vowels, is of interest. In most cases, unstressed tokens of the vowel were considerably different than stressed tokens, and apart from the foot-final tokens, considerably different from schwa. This differs from Czaykowska-Higgins & Kinkade’s (1998) generalisation about unstressed vowels in Salish languages. Recall from Chapter 1 that in other Salish languages, unstressed vowels reduce to schwa. The results presented here show that that is not the case in St’át’imcets, at least in terms of vowel quality. These results also seem to support Bessell’s (1997) phonetic categorisation of St’át’imcets /a/ as a central vowel but showed that its position was highly variable. Results also confirm previous representations of /i/ and /a/ but not /u/. Three possible reasons for the difference in results to both Bessell (1997) and Shahin (2002) could be elicitation method, where the vowel was measured, and the specific consonant contexts.  182  7.4.3 Domain-final extrapods The prediction laid out in Section 7.1 was that extrapods would show similar Prosodic Strengthening effects to St'át'imcets unstressed syllables, parallelling Cho’s (2005) results for English. Given that there are, by definition, no foot-internal or foot-final extrapods, the peripherality hierarchy for the extrapods tested is: PPhrase-final > PWord-final. Peripherality is defined in the same terms as for unstressed vowels: /i/ should be higher and/or fronter, /a/ should be lower, and /u/ should be higher and/or backer at PPhraselevel than PWord-level. In the following table, mean results and standard deviation are presented. Significant differences are presented in the second table and further illustrated by two scattergrams for each speaker’s vowels. As with the unstressed vowel results presented above, the first scattergram shows schwa and stressed vowel tokens in order to situate the combined extrapod tokens. The second scattergram distinguishes extrapods at different boundaries. Table 7.20 Extrapod means across boundaries a N F1 F2 Mean (Hz) Mean (Hz) AP word-final 18 610 (28) 1663 (36) CA  phrase-final word-final phrase-final  16 21 19  word-final  N 24  F1 Mean (Hz) 366 (34)  F2 Mean (Hz) 992 (55) 982 (78) 1351 (183) 1137 (186)  1638 (45) 1804 (41) 1797 (85)  16 21 21  18  648 (46) 360 (27) 351 (88) i 332 (32)  1796 (236)  16  345 (27) 271 (36) 255 (16) u 386 (41)  phrase-final word-final  15 23  338 (43) 351 (16)  1828 (105) 2365 (91)  18 19  437 (31) 411 (45)  1346 88) 1722 (197)  phrase-final  24  353 (20)  2305 (141)  21  415 (42)  1587 (157)  HD  LT  u  1442 (121)  Table 7.21 presents the significant differences by speaker.  183  Table 7.21 Extrapod significant differences (A negative t-value indicates that the direction of difference is opposite to what is predicted (phrase-final extrapods are more peripheral than word-final extrapods). Shaded cells indicate no significant difference.)  a  u  F1 t-value AP  3  F2  p-value  t-value  p-value  .006  F1 t-value 2.1  F2  p-value  t-value  p-value  3.8  .001  2.0  .0501  2.4  .0211  1  .041  CA i HD  u -4.2  LT  <.001  From Table 7.21, we see that AP makes an F1 distinction between word and phrase-final vowels for both /a/ and /u/, though a marginal one in the case of /u/. Conversely, CA makes only one significant distinction between word- and phrase-final extrapods: a strong distinction in the F2 values of /u/. HD makes no word/phrase distinction for /i/ but makes a strong distinction for /u/ in height and a marginal distinction in frontness. LT also makes no distinctions between word-final and phrase-final extrapod /i/. She makes a marginal distinction in frontness for /u/. The following figures are scattergrams comparing the positions of speakers’ extrapods and stressed vowels, followed by scattergrams comparing only the extrapod tokens across boundaries. Figure 7.17 shows that AP’s extrapod /a/s are generally lower than both stressed /a/ and schwa. Extrapods show less varied distribution than the stressed vowel.  184  Figure 7.17 AP’s extrapod /a/, stressed /a/ and schwa  Stressed /a/ shows more distribution variation in height and frontness than any other token, and is generally higher and backer than extrapods. Figure 7.18 shows that AP’s word-final extrapod /a/ is **higher than phrase-final /a/. Figure 7.18 AP’s extrapod /a/  Figure 7.18 demonstrates that while word and phrase-final extrapods overlap to a large extent, phrase-final vowels are lower (more peripheral).  185  Figure 7.19 shows that AP’s extrapod /u/s are considerably backer than both the stressed vowel and schwa. Figure 7.19 AP’s extrapod /u/, stressed /u/ and schwa  Figure 7.20 shows that AP’s word-final extrapod /u/ is marginally higher than phrase-final /u/. Figure 7.20 AP’s extrapod /u/  186  From Figure 7.20, we see that AP’s word- and phrase-final extrapods overlap to a large extent, and that word-final /u/ is only marginally higher. In terms of backness, they are not significantly different. Moving to CA’s results, Figure 7.21 illustrates more distributional variation for extrapods than the stressed vowels: Figure 7.21 CA’s extrapod /a/, stressed /a/ and schwa  Unlike AP, CA shows no discrete token distribution. However, the majority of the extrapod tokens are lower than both stressed vowels. Looking at Figure 7.22, we can see that CA has highly variable phrase-final values for /a/ in contrast with the relatively invariant word-final /a/. There is significant overlap between the two sets of tokens.  187  Figure 7.22 CA’s extrapod /a/  Figure 7.23 compares the tokens of CA’s extrapod /u/ with stressed /u/ and schwa: Figure 7.23 CA’s extrapod /u/, stressed /u/ and schwa  From the figure above, we see that extrapods show more distributional variation than stressed /u/ and schwa, and are generally backer than both. Finally, Figure 7.24 shows that CA’s word-final extrapod /u/ is **fronter than phrase-final /u/:  188  Figure 7.24 CA’s extrapod /u/  From Figure 7.24, we see that word-final /u/ is fronter, though more variable in height than phrase-final extrapod /u/. Turning to HD’s results, Figure 7.25 shows the position of HD’s extrapod /i/ in relation to the stressed vowels. Extrapod /i/ tokens are produced backer than the stressed vowel and higher than schwa. Again, extrapods show a large distribution range relative to schwa and the stressed vowel.  189  Figure 7.25 HD’s extrapod /i/, stressed /i/ and schwa  Figure 7.26 illustrates the overlap of HD’s word- and phrase-final extrapods. Figure 7.26 HD’s extrapod /i/  190  Figure 7.27 shows HD’s extrapod /u/ in relation to stressed /u/ and schwa: Figure 7.27 HD’s extrapod /u/, stressed /u/ and schwa  This figure above demonstrates that schwa is distinct and shows a considerably more compact distribution than the full vowels, both of which are considerably backer. Figure 7.28 shows that HD’s word-final extrapod /u/ is **higher and marginally fronter than phrase-final: Figure 7.28 HD’s extrapod /u/  191  Finally, let us consider LT’s extrapod results. Figure 7.29 illustrates LT’s extrapod /i/ in relation to stressed /i/ and schwa:. Figure 7.29 LT’s extrpod /i/, stressed /i/ and schwa  The figure above shows that that extrapod /i/ tokens overlap with stressed /i/, and are considerably fronter and higher than schwa. Figure 7.30 below demonstrates that LT makes no significant distinctions between word and phrase-final extrapod /i/. Figure 7.30 LT’s extrapod /i/  192  Unlike the relatively discrete full vowel and schwa tokens in LT’s /i/ results, Figure 7.31 shows the overlap between LT’s extrapod /u/ in relation to stressed /u/ and schwa: Figure 7.31 LT’s extrapod /u/, stressed /u/ and schwa  While stressed /u/ can be described as backer than schwa, extrapod /u/ overlaps both vowels. Finally, Figure 7.32 shows LT’s word-final extrapod /u/ is ** fronter than phrasefinal /u/. Figure 7.32 LT’s extrapod /u/  193  In Figure 7.32, we see that both LT’s word- and phrase-final extrapod /u/ show a high range of variation but that word-final /u/ is marginally fronter than phrase-final /u/. In order to compare the global Prosodic Strengthening effects on extrapods, results across speakers are summarised in the table below: Table 7.22 Results for predictions of peripherality in domain-final extrapods Prediction AP CA HD a Different? Yes No How? i u  ph > wd  LT  N/A  Different?  No  No  How?  N/A  N/A  Different? How?  Yes  Yes  Yes  Yes  ph > wd  ph > wd  wd > ph  ph > wd  From the table above, we can make two generalisations: i) /u/ is again the only vowel for which speakers consistently make word versus phrase distinction, and ii) where speakers do make a distinction, three out of the four speakers demonstrate the predicted peripherality hierarchy. HD is the exception. Southern speakers and CA made the predicted distinction for /u/ and no distinction in the other vowels. AP made only a significant height distinction for both /a/ and /u/ in the predicted direction. 7.4.4 Extrapod discussion The prediction that domain-final extrapods would pattern like unstressed syllables in St’át’imcets was borne out. The results were vowel- and speaker-dependent, with /u/ showing the only consistent Prosodic Strengthening effects. Both domain-final unstressed vowels and extrapods showed the same patterns of peripherality at word/phrase levels for /u/ and lack of distinction for /a/ and /i/. The prediction that they would pattern like the English speakers in Cho (2005) is less clear. If speakers made a distinction, it was generally in the direction predicted by Cho (2005) and the Prosodic Strengthening literature. However, only /u/ was distinguished by all speakers. Such speaker- and segment-specific results are in line  194  with previous Prosodic Strengthening research. It should also be mentioned that the limited number of speakers examined in this study is also consistent with other Prosodic Strengthening experiments. The seminal experiment by Fougeron and Keating (1997) had only three participants and Cho (2005) had six. The four speakers who participated in this experiment fall well within the usual number of participants for this type of study.  7.5  Conclusion  The results of this experiment confirm that Prosodic Strengthening does apply in St’át’imcets but not necessarily in the same way as in other languages. In keeping with previous Prosodic Strengthening results, there was no perfect correspondence between boundaries and significant distinctions, and results were vowel-specific. In other words, these results show that Prosodic Strengthening is implemented as language (or dialect) specific phonetics. These results support a weak rather than a strong interpretation of the phonetics-phonology mapping. Why /u/ and not /a/ or /i/ showed word/phrase distinctions is puzzling. As previously mentioned, it may have to do with the measurement of values before the steadier state of the vowel has been achieved. The other proposal was that rounding of the vowel cues Prosodic Strengthening more directly than the position on the F1/F2 plane. Articulatory and/or video investigation might shed some light on this issue. Interestingly, one unexpected result of the experiment was the reduction of full vowels in foot-final position rather than foot-internal position. This was argued to be a syllabically based stressed vowel enhancement strategy. It may be that the difference in vowel quality could shed some light on the ambisyllabic status of glottalised resonants assumed in Chapter 4. An experiment examining vowels before inter-vocalic plain and glottalised resonants could potentially favour one analysis over the other. One final conclusion we can draw from these results is that Prosodic Strengthening applies to unstressed syllables and extrapods in the same manner. The effect of a lack of boundary on Prosodic Strengthening effects is examined in the next chapter, which compares the boundary strength effects of unstressed vowels and extrapods.  195  Chapter 8 Boundary Strength: Unstressed Vowels versus Extrapods  8.1  Introduction  This chapter presents the results for the second half of Experiment 2: the comparison of domain-final boundary effects between unstressed vowels and extrapods. This chapter uses the same methodology and tokens as reported in the previous chapter to test the hypothesis that St’át’imcets unstressed vowels and extrapods should be acoustically distinct. The predicted result is that of the Footed-is Stronger-than-Not (or FSN) model presented in Chapter 3. Given a finely cumulative interpretation of Prosodic Strengthening, this difference would be attributable to the extrapod’s lack of boundary. Specifically, extrapods, which lack a foot boundary, will be weaker (i.e. less peripheral on the F1/F2 plane) than unstressed vowels, which have a foot boundary. The acoustic effects of a lack of parsing have hitherto not been investigated. The results indicate that speakers make a distinction between unstressed vowels and extrapods in identical segmental, syntactic, morphological and intonation contexts and in the absence of a stress distinction, supporting the hypothesis that parsing is reflected in the acoustic signal, at least in this language. These results confirm the prediction of the current model that phonological domain distinctions are reflected by distinct acoustic characteristics. The direction of difference was speaker-/dialectspecific, suggesting that the presence of contrast rather than direction, is crucial. These results parallel the distinction between the directly and indirectly mapped contrasts seen in Chapter 6. The hierarchical relations between domains in the Hierarchy are correlated with contrast constrained in a particular direction. That is, segments at higher boundaries are stronger than those at lower boundaries. The absence of a direct mapping for the contrast between unstressed vowels and extrapods means that Prosodic Strengthening effects can extend to that ‘embedded’ contrast but need not. Overall, the results support a weak interpretation of the phonetics-phonology mapping. 196  The first section of this chapter reviews the predictions of the model with respect to boundary strength. Section 8.3 briefly reviews the methodology; Section 8.4 presents the results, Section 8.5 discusses those results and Section 8.6 concludes the chapter.  8.2  Background  Chapter 3 presents evidence that there is a phonological distinction between unstressed syllables and extrapods in St’át’imcets. In other words, the gammas in the examples below show different phonological characteristics. (1a)  (b) W | Ft /\ αγ  W /\ Ft \ /\ \ α β γ  [(α γ)]  [(α β) γ]115  In the first example, γ is both foot and word-final whereas in the second, γ is word-final (extrapod). This thesis argues that all else being equal, distinct phonological domains should show distinct acoustic characteristics and that one dimension of difference might be boundary strength effects. As we saw in the previous chapter, the Prosodic Strengthening literature shows that the articulation of segments is magnified at higher domain boundaries of the Prosodic Hierarchy. In other words, the boundary a segment is adjacent to affects how it is articulated. Following Cho (2005), this chapter focuses again on the acoustic effects of the differing articulations. If constituency in prosodic domains is reflected in the fine phonetic details of articulation, then it follows that a lack of constituency might also be reflected in different articulatory characteristics. In this way, a constituent that violates the Strict Layer Hypothesis by not being fully parsed, and therefore having fewer boundaries, might be weaker than one that is fully parsed. Following Cho (2005) we assume that the difference in articulation will be reflected in a difference in the acoustics.  115  Recall from the previous chapter that γ can be the same suffix/enclitic in both cases.  197  Research on Prosodic Strengthening, including Cho (2005), has considered only fully parsed prosodic domains. As a result, it is unclear whether Prosodic Strengthening reflects only the final boundary (i.e. the highest level) in the Prosodic Hierarchy or if it is more finely cumulative, reflecting each individual domain the segment is in. These hypotheses are impossible to distinguish based only on fully parsed domains. In order to test which hypothesis is correct, we need to examine cases that violate the Strict Layering Hypothesis (i.e. in which two constituents have the same final boundary but differ in other boundaries): in other words, a case of domain-final extrapods. There are two general predictions that could be made based on the Prosodic Strengthening literature in terms of the acoustic characteristics of word-final extrapods and word-final unstressed syllables. First, if Prosodic Strengthening reflects the effects of the largest domain only (i.e. the highest boundary), we predict a peripherality relationship as in (2). In this case, Prosodic Strengthening reflects only the boundary of the largest domain (in this case ‘word’) and makes no distinction between the presence and absence of any other boundaries. (2)  Largest domain only:  V)foot]word = V]word  Second, if Prosodic Strengthening were finely cumulative, the fact that the unstressed vowel is adjacent to two boundaries (foot and word) means it will be articulated more strongly than the extrapod, which is adjacent to only one (word). This would predict the peripherality hierarchy in (3). (3)  Finely cumulative:  V)foot]word > V]word  Results consistent with example (2) would reflect the standard version of the theory in which unstressed syllables and extrapods are indistinguishable at the PWd level, represented by the ‘Traditional Hypothesis’ from Chapter 3. If, however, we get results consistent with (3), this can be taken as evidence that lack of boundaries is both phonologically and phonetically relevant. This would strengthen the underlying claim of  198  Prosodic Strengthening that prosodic domains are reflected systematically in fine phonetic details. St’át’imcets presents an ideal case study to test these predictions and to give us further insight into the acoustic characteristics of extrapods. As we saw in the previous chapter, the possessive suffixes /-su/ (2sg.) and /-i/ (3pl), and the 3sg.poss suffix+existential enclitic /-s=a/ (3sg.) occur in both extrapod and unstressed position, as in (4):116 (4a) W | Ft /\ α γ qmút-su [(qmútßu)] hat-your  (b) W /\ Ft \ /\ \ α β γ músmus-(t)su [(múßmuß)(t)ßu] cow-your  The root+suffix tokens allow a comparison of unstressed and extrapod vowels in identical segmental, morpho-syntactic and intonational contexts. One of the limitations of the experiment reported in the previous chapter was that exact minimal pairs across all comparisons were impossible. The fact that the unstressed versus extrapod comparison involves exact minimal pairs means that any significant differences between them will be particularly determinative. Table 8.1 restates the hypotheses proposed in Chapter 3 and their predictions regarding differences between unstressed vowels and extrapods in terms of boundary strength:  116  Both /-su/ and /-sa/ sometimes alternate with /-tsu/ and /-tsa/ following fricatives.  199  Table 8.1 Hypotheses and their predictions Name Hypothesis Traditional  us =ex  Variable  us ~ ex  Maximise Prominence (MaxProm)  ex > us  Footed Stronger than Not (FSN)  us > ex  Acoustic prediction No peripherality distinction between unstressed and extrapod vowels Extrapods and unstressed syllables are distinct in terms of position on the F1/F2 plane but the difference is not predictable or consistent. Extrapods are more peripheral on the F1/F2 plane than unstressed syllables: /a/ will be lower, /u/ will be backer and higher and /i/ will be fronter and higher. Extrapods will be less peripheral than unstressed vowels: /a/ will be higher, /u/ will be fronter and lower and /i/ will be lower and backer.  Under the Traditional model, there should be no difference in peripherality between unstressed syllables and extrapods. In terms of Prosodic Strengthening, this would reflect a non-cumulative interpretation, where only the highest domain boundary affects articulation. The Variable model predicts a difference but not one that is consistent in direction. Recall that this claim was based on theories such as Liberman and Prince (1977) and Pierrehumbert and Beckman (1988), which propose that relative prominence is defined structurally (i.e. under sisterhood or motherhood). Given that unstressed syllables and extrapods do not stand in such a relationship, there is no correlated relative strength relation. Applying this hypothesis to boundary strength effects, we could argue that the lack of a phonologically-specified relationship might also apply to relative boundary strength. In this case, the fact that speakers make a distinction rather than how they make it, is important. The absence of a direct mapping between unstressed vowels and extrapods means that the contrast will be determined by other demands, such as head and/or domain maximisation.  200  MaxProm predicts that extrapods will be more peripheral than unstressed syllables due to a foot-internal reduction constraint. If we find that extrapods are more peripheral than unstressed vowels, we have support for the notion that a consistent prominence relation does hold between unstressed syllables and extrapods, and that the morpheme-specific process of non-head reduction we saw in Chapter 3 is applicable to boundary strength as well. Finally, if we find that unstressed syllables are consistently stronger than extrapods, we have evidence that Prosodic Strengthening is finely cumulative—each boundary contributes to the hyperarticulation of segments, not merely the highest one. This would support the FSN hypothesis. The predicted outcome is that of the FSN model: i) unstressed syllables and extrapods will show different positions on the F1/F2 plane, and ii) unstressed syllables will be more peripheral than extrapods, by virtue of the fact that they are footed while extrapods are not. If these predictions are borne out, we have evidence of a finely cumulative, consistent mapping between boundaries and acoustic characteristics. This would increase the empirical basis for Prosodic Strengthening and would also support the notion that unstressed syllables and extrapods are distinct. The specific acoustic predictions of the FSN model are that unstressed /a/ will be lower (have higher F1) than extrapod /a/, unstressed /i/ will be higher and fronter (lower F1, higher F2) than extrapod /i/ and unstressed /u/ will be higher and backer (lower F1 and F2). These predictions are summarised in Table 8.2: Table 8.2 Unstressed vowel versus extrapod predictions a F1 us > ex i F1 us < ex F2 us > ex u F1 us < ex F2 us < ex  8.3  Methodology Review  The methodology for this part of the experiment is comprehensively discussed in Chapters 4 and 7 but is reviewed here briefly.  201  8.3.1 Subjects Subjects and tokens are given in Chapter 7. 8.3.2 Stimuli Tokens were selected from van Eijk (1997) and Davis (forthcoming) and constructed with the help of Davis (p.c.). The tokens structured as the following: Table 8.3 Unstressed vowel versus extrapod structures (α = stressed V, β= unstressed V, γ= extrapod)  unstressed Word-final Ph | \ W | Ft /\ αγ  extrapod Phrase-final Ph /| W | Ft /\ αγ  Word-final Ph |\ W /\ Ft \ /\ \ α β γ  Phrase-final Ph /| W /\ Ft \ /\ \ α β γ  As in Chapter 6, whether root+affix constructions are PWords or PPhrases is not crucial. The prediction is that IP-final vowels will be more peripheral than those in smaller domains, so whether that domain is a PWord or Phrase should not affect whether speakers make a distinction, nor the direction of the contrast. IP-final tokens are referred to as ‘phrase-final’ and non-IP-final tokens as ‘word-final’. They are given in Table 8.4 below. Table 8.4 Unstressed vowel and extrapod tokens /a/ /i/ wordti qmút.sa [(qmútßa)] qmúti [(qmúti)] final/phrase‘their hat’ final us ‘his/her hat’ wordti músmustsa músmusi final/phrase[(múßmuß)æa] [(múßmu)ßi] final ex ‘his/her cow’ ‘their cow’  /u/ qmút.su [(qmútßu)] ‘your hat’ músmustsu [(múßmuß)æu] ‘your cow’  202  8.3.3 Equipment Equipment details are given in Chapter 7. 8.3.4 Recording As reported in Chapter 7, tokens were put in different target sentences for each dialect. Sample sentences are repeated below: Table 8.5 Northern dialect target sentences117 /a/ word-final us N’áscit ta qmút.sa (kw)s118 Henry. ‘Give his/her hat to Henry.’ word-final ex  N’áscit ta músmustsa (kw)s Henry. ‘Give his/her cow to Henry.’  phrase-final us  Síma7cits ta qmút.sa. Ámhasan’. ‘Give me his/her hat. It looks nice.’  phrase-final ex  Síma7cits ta músmustsa. Ámhasan’. ‘Give me his/her cow. It looks nice.’  /u/ N’áscit ku qmút.su (kw)s Henry. ‘Give one of your hats (which I think is in your bag) to Henry’ N’áscit ku músmustsu (kw)s Henry. ‘Give one of your cows (which I think is behind the tree) to Henry.’ Síma7cits ku qmút.su. Cw7it I qmút.swa. ‘Give me one of your hats (which I think is in your bag). You have lots of hats.’ Síma7cits ku músmustsu. Cw7it i músmus.tswa ‘Give me one of your cows (which I think is behind the tree). You have lots of cows.’  117  Sample sentences are given in the orthography only. See the conversion chart in Appendix A for NAPA and IPA equivalents. 118 Upper speakers most often abbreviated /kws/ to /s/, while lower speakers abbreviated it to /k/ or /ks/.  203  Table 8.6 Southern dialect target sentences /i/ word-final us Úm’en ku pála7 qmúti k(ws) Henry. ‘Give one of their hats to Henry.’ word-final ex Úm’en ku pépla7 músmusi k(ws) Henry.  phrase-final us  phrase-final ex  /u/ Úm’en ku pála7 qmút.su k(ws) Henry. ‘Give one of your hats to Henry.’ Úm’en ku pépla7 músmustsu k(ws) Henry.  ‘Give one of their cows to Henry.’ Úm’ents ku pála7 qmúti. Cw7it i qmútiha. ‘Give me one of their hats. They have lots of hats.’  ‘Give one of your cows to Henry.’ Úm’ents ku pála7 qmút.su. Cw7it i qmút.swa. ‘Give me one of your hats. You have lots of hats.’  Úm’ents ku pépla7 músmusi. Cw7it i musmusíha. ‘Give me one of their cows. They have lots of cows.’  Úm’ents ku pépla7 músmustsu. Cw7it i músmustswa. ‘Give me one of your cows. You have lots of cows.’  8.3.5 Measurements and analysis All vowels were measured as previously stated in Chapters 4 and 7. Statistical analyses in the form of t-tests were performed, with a Bonferroni correction applied. As a result, the p-value for these comparisons is p≤.025 and represented by **. A marginal p-value of p≤.05 is represented *.  8.4  Results  Results are presented first for word-final comparisons and then for phrase-final comparisons. The first table in each section gives the means and standard deviations. The second table presents the significant differences, which are subsequently represented by scattergrams.  204  8.4.1 Word-final unstressed versus extrapods This section presents the results from the comparison of word-final unstressed versus extrapod vowels. Table 8.7 presents the means and standard deviations of F1 and F2 measurements, and significant differences by speaker are presented in Table 8.8. Table 8.7 Word-final means and standard deviation a N F1 F2 N Mean (Hz) Mean (Hz) AP CA  HD LT  us ex us ex  21 18 17 21  us ex us ex  u F1 Mean (Hz)  F2 Mean (Hz) 1029 (82) 992 (52) 1193 (166) 1351 (184)  1653 (37) 1663 (36) 1807 (55) 1804 (41)  32 24 19 21  358 (21) 366 (34) 282 (27) 271 (36)  21 18 19  594 (31) 610 (28) 326 (49) 360 (27) i 325 (27) 332 (31) 334 (16)  1957 (113) 1796 (236) 2542 (115)  13 16 20  409 (34) 386 (41) 410 (28)  1390 (121) 1442 (176) 1703 (153)  23  351 (16)  2365 (91)  19  411 (45)  1722 (197)  u  The next table presents the significant differences speakers make between unstressed and extrapod vowels word-finally. The prediction tested was that unstressed vowels would be more peripheral than extrapod vowels. Table 8.8 Word-final significant differences (– indicates significant difference in direction opposite to prediction. Shaded cells indicate no significant difference.)  a t-value AP CA  -2.7  F1 p-value  u t-value  F2 p-value  .012  3.6  .001  F1 p-value  F2 t-value  p-value  2.9 u  i HD LT  t-value  2.8 5.6  .007  .009 <.001  The results in the table above indicate that AP is the only speaker not to make any distinction at the PWord level. CA makes a significant height distinction between wordfinal unstressed and extrapod /a/ (extrapods had greater mean F1) and a significant  205  frontness distinction for /u/ (extrapods had greater F2). In contrast to CA, the two Southern speakers make distinctions for /i/ only. LT has significantly higher and fronter unstressed vowels than the extrapods. HD makes an F2 distinction at the word-level only for /i/, with fronter unstressed vowels than extrapods. For a clear representation of the results presented in Table 8.8, scattergrams illustrating significant differences between unstressed vowels and extrapods for each vowel are presented below. Recall first that AP made no significant distinctions at the word level. As a result, we will begin with scattergrams of CA’s vowels. First, Figure 8.1 shows that CA’s word-final unstressed /a/ is ** higher but not backer than extrapod /a/: Figure 8.1 CA’s word-final unstressed /a/ versus extrapod /a/  In this figure, we see that CA’s unstressed tokens vary more in height than extrapods. In contrast, Figure 8.2 shows CA’s word-final unstressed /u/ is **backer than extrapod /u/:  206  Figure 8.2 CA’s word-final unstressed /u/ versus extrapod /u/  The figure above shows that CA’s extrapod /u/s are significantly fronter (less peripheral) than unstressed /u/s. Unlike CA, HD showed no significant differences /u/ at the word level, but did have significantly different /i/s. Figure 8.3 shows HD’s word-final unstressed /i/ is **fronter than phrase-final /i/: Figure 8.3 HD’s word-final unstressed /i/ versus extrapod /i/  207  Figure 8.3 above illustrates that HD’s unstressed /i/ is more peripheral in terms of frontness than extrapod /i/. While LT also makes no distinction for /u/, she makes significant distinctions in both height and frontness for /i/ at the word level. Figure 8.4 shows LT’s word-final unstressed /i/ is **higher and **fronter than extrapod /i/: Figure 8.4 LT’s word-final unstressed /i/ versus extrapod /i/  Figure 8.4 demonstrates that LT’s word-final /i/ is significantly more peripheral in terms of both height and frontness than extrapod /i/. The figures above represent the range of speaker variation. AP makes no distinctions, while the Southern speakers make significant distinctions only for /i/. Only CA makes significant distinctions for both. As we see in the next section, however, at the phrase level there is a reversal in the pattern of the Northern speakers. The next section presents the results of comparisons of phrase-final unstressed versus extrapod vowels. 8.4.2 Phrase-final unstressed versus extrapods This section presents the tables and figures of significant differences in the comparison of phrase-final unstressed versus extrapod vowels. As above, means and standard deviations  208  are presented first, followed by speakers’ significant differences presented in tabular and graphic form. Table 8.9 Phrase-final means and standard deviation a N F1 F2 Mean (Hz) Mean (Hz) AP us 18 614 (50) 1639 (49) ex 16 648 (48) 1638(44) CA us 22 330 (62) 1824 (76) ex 19 351 (88) 1797 (85) i HD us 13 323 (36) 1931 (303) ex 15 338 (43) 1828 (105) LT us 17 348 (19) 2502 (148) ex 24 353 (20) 2305 (141)  u F1 Mean (Hz) 363 (24) 345 (27) 266 (27) 255 (16) u 423 (33) 437 (31) 401 (37) 415 (42)  N 23 16 21 21 25 18 20 21  F2 Mean (Hz) 1054 (77) 982 (78) 1038 (134) 1137 (186) 1273 (114) 1346 (88) 1513 (147) 1587 (157)  The next table illustrates the significant differences based on the data presented in Table 8.9. In a mirror image of word-final results, AP makes phrase-final distinctions for both vowels while CA makes none. HD and LT make frontness distinctions but for different vowels. The predicted direction of difference was that unstressed vowels should be more peripheral than extrapods. Table 8.10 Phrase-final significant differences  (- represents a significant difference in the opposite direction of the prediction. 1 indicates a marginally significant difference.)  a  u  F1 t-value AP  -2.1  F2 p-value  t-value  F1  p-value  1  .044  t-value -2.3  F2  p-value  t-value  p-value  -2.9  .007  1.2  .0281  1  .03  CA i HD LT  u 4.3  <.001  The values in Table 8.10 indicate that, once again, speaker- and vowel-specific effects are observed. CA makes no significant distinctions, whereas AP has marginally higher unstressed /a/ than extrapod /a/, and significantly fronter and marginally higher  209  unstressed /u/ than extrapod /u/. Again, Southern speakers make significant distinctions for one vowel only. HD has marginally backer unstressed /u/ than extrapod /u/, and LT has significantly fronter unstressed /i/ than extrapod /i/. As above, the significant differences are illustrated in the scattergrams below. Figure 8.5 shows AP’s phrase-final unstressed /a/ is *higher than extrapod /a/. Figure 8.5 AP’s phrase-final unstressed /a/ versus extrapod /a/  The above figure shows that unstressed syllables are marginally less peripheral in terms of height than extrapod syllables. Figure 8.6 shows AP’s phrase-final unstressed /u/ is **fronter and *higher than extrapod /u/.  210  Figure 8.6 AP’s phrase-final unstressed /u/ versus extrapod /u/  Figure 8.6 shows that while unstressed and extrapods overlap to some extent, extrapod /u/ is significantly more peripheral than unstressed /u/ in terms of frontness and marginally in terms of height. Unlike, AP, HD makes a marginally significant distinction only for /u/. Figure 8.7 illustrates that HD’s phrase-final unstressed /u/ is *backer than extrapod /u/:  211  Figure 8.7 HD’s phrase-final unstressed /u/ versus extrapod /u/  From looking at Figure 8.7, we see that, while both unstressed /u/ and extrapod/ u/ overlap, the unstressed /u/ is marginally backer (more peripheral) than the extrapod. Like HD, LT makes significant distinctions for only one vowel, but she makes the distinction for /i/, not /u/. Figure 8.8 shows LT’s phrase-final unstressed /i/ is **fronter than extrapod: Figure 8.8 LT’s phrase-final unstressed /i/ versus extrapod /i/  212  The overall results for all speakers are summarised in the following tables. Table 8.11 presents the word-final results and Table 8.12 presents the phrase-final results. Table 8.11 Word-final summary results Northern AP CA a Different? No Yes How? N/A ex>us i Different? How? u Different? No Yes How? N/A us>ex  Southern HD  Yes us>ex No N/A  LT  Yes us>ex No N/A  First, Table 8.11 shows that Southern dialect speakers pattern more closely than Northern dialect speakers at the word level. This is somewhat unexpected, given that AP and CA are siblings. AP, unlike the other speakers, shows no significant distinctions at the wordlevel at all, whereas CA makes distinctions in both /a/ and /u/ but only /u/ follows the predicted direction of difference. Both Southern speakers show the predicted peripherality pattern for /i/ but no distinction for /u/. Second, Table 8.12 shows that CA, mirroring AP, makes no distinctions at the phrase level. Table 8.12 Phrase-final summary results Northern AP CA a Different? Yes No How? ex>us N/A i Different? How? u Different? Yes No How? ex>us N/a  So