Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The interaction of metrical structure, tone, and phonation types in Quiaviní Zapotec Chávez Peón , Mario E. 2010

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

Item Metadata

Download

Media
24-ubc_2010_fall_chavezpeon_mario.pdf [ 9.6MB ]
Metadata
JSON: 24-1.0071253.json
JSON-LD: 24-1.0071253-ld.json
RDF/XML (Pretty): 24-1.0071253-rdf.xml
RDF/JSON: 24-1.0071253-rdf.json
Turtle: 24-1.0071253-turtle.txt
N-Triples: 24-1.0071253-rdf-ntriples.txt
Original Record: 24-1.0071253-source.json
Full Text
24-1.0071253-fulltext.txt
Citation
24-1.0071253.ris

Full Text

THE INTERACTION OF METRICAL STRUCTURE, TONE, AND PHONATION TYPES IN QUIAVINÍ ZAPOTEC by Mario E. Chávez Peón B.A. Hispanic Language and Literatures, UNAM, 2001  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) August, 2010 © Mario E. Chávez Peón, 2010  Abstract This thesis investigates the interaction between different prosodic patterns in Quiaviní Zapotec (Otomanguean), and accounts for them both at the phonetic and the phonological level. In it, I examine an array of complex patterns along multiple dimensions, including metrical structure, tone, and phonation types; as well as how these patterns interact with the fortis/lenis distinction, and syllable structure. Within the framework of Optimality Theory, my analysis sheds light on the phonetics-phonology interface and emphasizes the need for a theory with moraic structure. This dissertation presents the first thorough phonetic documentation of the prosody of Quiaviní Zapotec. It makes a significant empirical contribution by providing descriptive generalizations of vowel and consonant length, a reanalysis of tone as contrastive in Quiaviní Zapotec, and a new approach to the study of the four-way phonation contrast in this language — modal /a/, breathy /a̤/, creaky /a̰/ and interrupted /aʔ/ vowels — (cf. Munro & Lopez, 1999). In addition, this research makes significant contributions to phonological theory, with regards to both segmental and prosodic phenomena. Within an emergent feature approach, I revisit the fortis/lenis distinction, which crosscuts the obstruent-sonorant contrast in Quiaviní Zapotec. I analyze it as a composite of language-specific phonological and phonetic properties, encoded with the feature [+/-fortis]. Adding to the typology of syllable weight, fortis consonants are analyzed as moraic in coda position, but among them, only fortis sonorants may bear tone alongside vowels (i.e. *[-SON][TONE] ‘No tones on obstruents’). Furthermore, I show specific timing patterns for the phonetic implementation of tonal and laryngeal features. Quiaviní Zapotec exhibits compatibility of contrasts; compromise of phonological features (e.g. tonal contrasts are cued during modal phonation, followed by breathiness or laryngealization); or complete incompatibility, which translates into phonemic gaps. This distribution is formalized in terms of markedness interaction and grounded constraints (e.g. ‘If [+spread glottis], then Low tone’, accounting for the absence of high tone with breathy vowels). Overall, the thesis analyzes the minimal prosodic word in Quiaviní Zapotec (a bimoraic foot) as the domain where the full array of tonal and phonation type contrasts takes place, and illustrates particular mechanisms by which phonetic factors shape phonology.  ii  Table of contents Abstract........................................................................................................................................................... ii
 Table of contents ...........................................................................................................................................iii
 List of tables.................................................................................................................................................. vii
 List of figures.................................................................................................................................................. x
 List of abbreviations ...................................................................................................................................xiii
 Acknowledgements ..................................................................................................................................... xiv
 Dedication ................................................................................................................................................... xvii
 Preface (summary and committee members).........................................................................................xviii
 Chapter 1: The thesis and the language...................................................................................................... 1
 1.1
 The thesis.......................................................................................................................................... 1
 1.1.1
 General methodology ................................................................................................................ 3
 1.2
 The language .................................................................................................................................... 4
 1.2.1
 Genetic and geographic background ......................................................................................... 4
 1.2.2
 Previous work ............................................................................................................................ 5
 1.2.3
 Phonology.................................................................................................................................. 5
 1.2.4
 Phonotactics............................................................................................................................. 12
 1.2.5
 Morphosyntax.......................................................................................................................... 17
 Chapter 2: Vowel length and the fortis/lenis distinction in Quiaviní Zapotec...................................... 23
 2.1
  Introduction .................................................................................................................................... 23
  2.2
  Quiaviní Zapotec vowel length....................................................................................................... 25
  2.3
  Fortis and lenis consonants in Quiaviní Zapotec........................................................................... 35
  2.4
  The emergence of the [+/-fortis] feature ....................................................................................... 45
  2.5
  Conclusions .................................................................................................................................... 47
  Chapter 3: Metrical structure of Quiaviní Zapotec................................................................................. 48
 3.1
  Introduction .................................................................................................................................... 48
  3.2
 Moraicity and minimality ............................................................................................................... 49
 3.2.1
 Phonetic experiment: Syllable weight and the fortis/lenis distinction .................................... 55
 3.2.2
 Fortis consonants are not geminates........................................................................................ 66
 3.2.3
 Formal analysis........................................................................................................................ 69
 3.3
 Morphology: Root prominence ...................................................................................................... 73
 3.3.1
 Foot type: Trochaic rhythm ..................................................................................................... 79
 3.3.2
 Prefixes, compounds and complex roots: Foot alignment....................................................... 84
 3.3.3
 Diminutive suffix and clitics: Faithfulness to the base ........................................................... 87
  iii  3.4
  Loanword phonology...................................................................................................................... 94
  3.5
  Summary and conclusions ............................................................................................................ 101
  Chapter 4: Tone in Quiaviní Zapotec ..................................................................................................... 102
 4.1
 Introduction .................................................................................................................................. 102
 4.1.1
 Phonetic properties associated with phonation types ............................................................ 104
 4.2
 Experiment 1: Low tone with modal voice ................................................................................... 106
 4.2.1
 Acoustic description: Modal-L.............................................................................................. 108
 4.2.2
 Phonetic experiment: Modal-L.............................................................................................. 111
 4.2.2.1
 Methods ......................................................................................................................... 112
 4.2.2.2
 Results............................................................................................................................ 116
 4.2.2.3
 Discussion...................................................................................................................... 122
 4.3
 Experiment 2: Rising tone with modal voice................................................................................ 125
 4.3.1
 Acoustic description: Modal-R.............................................................................................. 126
 4.3.2
 Phonetic experiment: Modal-R.............................................................................................. 128
 4.3.2.1
 Methods ......................................................................................................................... 129
 4.3.2.2
 Results............................................................................................................................ 129
 4.3.2.3 Discussion.......................................................................................................................... 135
 4.4
 Experiment 3: Falling tone with modal voice .............................................................................. 135
 4.4.1
 Acoustic description: Modal-F .............................................................................................. 136
 4.4.2
 Phonetic evaluation: Modal-F ............................................................................................... 137
 4.5
  Conclusions: Quiaviní Zapotec tonal inventory with modal voice .............................................. 139
  Chapter 5: The tone-bearing unit in Quiaviní Zapotec: Moraicity and tone...................................... 141
 5.1
  Introduction .................................................................................................................................. 141
  5.2
 Tone-bearing segments in Quiaviní Zapotec ............................................................................... 143
 5.2.1
 Obstruents.............................................................................................................................. 143
 5.2.2
 Sonorants ............................................................................................................................... 146
 5.2.2.1
 High tone (sonorants) .................................................................................................... 148
 5.2.2.2
 Low tone (sonorants) ..................................................................................................... 150
 5.2.2.3
 Rising tone (sonorants) .................................................................................................. 152
 5.2.3
 Discussion.............................................................................................................................. 155
 5.2.4
 Conclusion ............................................................................................................................. 156
 5.3
 Tone representation in Quiaviní Zapotec and formal account .................................................... 158
 5.3.1
 Level tones............................................................................................................................. 162
 5.3.2
 Contour tones......................................................................................................................... 167
 5.4
  Conclusions .................................................................................................................................. 173
  Chapter 6: Non-modal phonation in Quiaviní Zapotec ........................................................................ 174
 6.1
  Introduction .................................................................................................................................. 174
  6.2
  Brief typology of phonation types................................................................................................. 175
  6.3
 Breathy vowels ............................................................................................................................. 178
 6.3.1
 Introduction ........................................................................................................................... 178
 6.3.2
 Breathy-L............................................................................................................................... 179
 6.3.3
 Breathy-F ............................................................................................................................... 182
 6.3.4
 Munro and Lopez (1999): Breathy vowels............................................................................ 184
 6.3.5
 Interim summary: Breathy vowels ........................................................................................ 186
 6.4
  Creaky vowels .............................................................................................................................. 187
  iv  6.4.1
 Introduction ........................................................................................................................... 187
 6.4.2
 Creaky-H ............................................................................................................................... 188
 6.4.3
 Creaky-L................................................................................................................................ 190
 6.4.4
 Creaky-F ................................................................................................................................ 192
 6.4.5
 Munro and Lopez (1999): Creaky vowels............................................................................. 196
 6.4.6
 Acoustic experiment: Creaky vowels .................................................................................... 201
 6.4.6.1
 Introduction.................................................................................................................... 201
 6.4.6.2
 Methods ......................................................................................................................... 201
 6.4.6.3
 Results............................................................................................................................ 204
 6.4.6.4
 Discussion...................................................................................................................... 209
 6.4.7
 Interim summary: Creaky vowels ......................................................................................... 210
 6.5
 Interrupted vowels........................................................................................................................ 211
 6.5.1
 Introduction ........................................................................................................................... 211
 6.5.2
 Interrupted-H ......................................................................................................................... 214
 6.5.3
 Interrupted-L.......................................................................................................................... 220
 6.5.4
 Interrupted-F.......................................................................................................................... 223
 6.5.5
 Munro and Lopez (1999): Interrupted vowels....................................................................... 225
 6.5.6
 Acoustic experiment: Interrupted vowels.............................................................................. 227
 6.5.6.1
 Introduction.................................................................................................................... 227
 6.5.6.2
 Methods ......................................................................................................................... 228
 6.5.6.3
 Results............................................................................................................................ 231
 6.5.6.4
 Discussion...................................................................................................................... 235
 6.5.7
 Interim summary: Interrupted vowels ................................................................................... 237
 6.6
 General discussion ....................................................................................................................... 238
 6.6.1
 Laryngeal vowels (creaky and interrupted): side-by-side comparison.................................. 238
 6.6.2
 Timing in non-modal vowels................................................................................................. 246
 6.7
  Conclusions .................................................................................................................................. 247
  Chapter 7: Laryngeal complexity in Quiaviní Zapotec......................................................................... 249
 7.1
  Introduction .................................................................................................................................. 249
  7.2
 Laryngeal accounts: Literature review & explanations .............................................................. 250
 7.2.1
 The special status of tone ...................................................................................................... 250
 7.2.2
 Laryngeal features ................................................................................................................. 252
 7.3
 Quiaviní Zapotec emergent laryngeal specification .................................................................... 257
 7.3.1
 Quiaviní Zapotec tonal specification..................................................................................... 257
 7.3.2
 Quiaviní Zapotec laryngeal features...................................................................................... 257
 7.4
  Quiaviní Zapotec comprehensive phonological representation................................................... 264
  7.5
  Formal account: Quiaviní Zapotec non-modal vowels................................................................ 268
  7.6
  Conclusions .................................................................................................................................. 274
  Chapter 8: Conclusions ............................................................................................................................ 275
 8.1
  Contribution ................................................................................................................................. 275
  8.2
  Comprehensive comparison with Munro and Lopez (1999) ........................................................ 279
  8.3
  Further research .......................................................................................................................... 286
  Bibliography ............................................................................................................................................... 290
 Appendices.................................................................................................................................................. 300
  v  Appendix A: Phonetic experiment (Chapter 3): Syllable weight and the fortis/lenis distinction (results by consonant type) ....................................................................................................................................... 300
 Appendix B. UBC Certificate of Ethics approval ................................................................................... 307
  vi  List of tables Table 1. Quiaviní Zapotec Consonant inventory ................................................................ 6
 Table 2. Quiaviní Zapotec vowels ...................................................................................... 7
 Table 3. Munro and Lopez (1999, p. 4) Quiaviní Zapotec vowel patterns......................... 9
 Table 4. Tone and phonation co-occurrence in Quiaviní Zapotec.................................... 12
 Table 5. Phonotactics of onset consonant clusters by sonority (mode of articulation)..... 15
 Table 6. Quiaviní Zapotec aspectual prefixes (adapted from Lee, 2006, p. 11)............... 20
 Table 7. Quiaviní Zapotec pronouns and clitics ............................................................... 21
 Table 8. Vowel patterns a’ and aa reanalyzed as phonemic short modal vowels ............ 33
 Table 9. Examples of vowel patterns a’ and aa with reanalysis....................................... 33
 Table 10. 1s and 2s Quiaviní Zapotec clitics. ................................................................... 34
 Table 11. Fortis/lenis characteristics (adapted from Jaeger, 1983) .................................. 35
 Table 12. The four-way contrast in Quiaviní Zapotec (fortis/lenis-obstruent/sonorant) .. 36
 Table 13. Quiaviní Zapotec fortis/lenis consonant pairs .................................................. 36
 Table 14. Quiaviní Zapotec four-way consonant contrasts .............................................. 55
 Table 15. Stimuli by ONSET (4 items for each consonant: / t, d, s, z, nˑ, n /) ................. 58
 Table 16. Stimuli by CODA (4 items for each consonant: / t, d, s, z, nˑ, n /) ................... 59
 Table 17. Results of phonetic experiment (duration of fortis and lenis consonants)........ 61
 Table 18. t-test results of phonetic experiment (duration of fortis and lenis consonants) 61
 Table 19. Vowel and consonant duration (ms): roots and clitisized forms (TiuC)........... 80
 Table 20. Munro and Lopez (1999: 4) low tone vowel patterns..................................... 107
 Table 21. Stimuli: low tone experiment.......................................................................... 113
 Table 22. Periodicity (jitter): Mean and standard deviation (LiaL & TiuC) .................. 117
 Table 23. Jitter results: Probability values from t-test (LiaL & TiuC) ........................... 117
 Table 24. H1-H2 results: Mean and standard deviation (LiaL) ...................................... 118
 Table 25. H1-H2 results: Probability values from t-test (LiaL)...................................... 119
 Table 26. H1-H2 results: Mean and standard deviation (TiuC)...................................... 119
 Table 27. H1-H2 results: Probability values from t-test (TiuC) ..................................... 119
 Table 28. H1-A1 results: Mean and standard deviation (LiaL) ...................................... 120
 Table 29. H1-A1 results: Probability values from t-test (LiaL)...................................... 120
 Table 30. H1-A1 results: Mean and standard deviation (TiuC)...................................... 120
 Table 31. H1-A1 results: Probability values from t-test (TiuC) ..................................... 121
 Table 32. Duration and intensity results: Mean and standard deviation (LiaL) ............. 121
 Table 33. Duration and intensity results: Probability values from t-test (LiaL)............. 121
 Table 34. Duration and intensity results: Mean and standard deviation (TiuC)............. 122
 Table 35. Duration and intensity results: Probability values from t-test (TiuC) ............ 122
 Table 36. Munro and Lopez (1999, p. 4) rising tone vowel patterns.............................. 125
 Table 37. Stimuli (partial): rising-tone experiment ........................................................ 129
 Table 38. Periodicity (jitter): Mean and standard deviation (LiaL and TiuC)................ 131
 Table 39. Jitter results: Probability values from t-test (LiaL and TiuC)......................... 131
 Table 40. H1-H2 results: Mean and standard deviation (LiaL) ...................................... 132
  vii  Table 41. H1-H2 results: Probability values from t-test (LiaL)...................................... 132
 Table 42. H1-H2 results: Mean and standard deviation (TiuC)...................................... 133
 Table 43. H1-H2 results: Probability values from t-test (TiuC) ..................................... 133
 Table 44. H1-A1 results: Mean and standard deviation (LiaL) ...................................... 134
 Table 45. H1-A1 results: Probability values from t-test (LiaL)...................................... 134
 Table 46. H1-A1 results: Mean and standard deviation (TiuC)...................................... 134
 Table 47. H1-A1 results: Probability values from t-test (TiuC) ..................................... 135
 Table 48. Stimuli: falling-tone evaluation ...................................................................... 137
 Table 50. H1-H2 results: Mean and standard deviation (LiaL) ...................................... 138
 Table 51. H1-A1 results: Mean and standard deviation (LiaL) ...................................... 138
 Table 52. Tone in modal voice: vowel pattern reanalysis .............................................. 139
 Table 53. Tone in modal voice: reanalysis with examples ............................................. 139
 Table 54. Quiaviní Zapotec Tone and modal voice........................................................ 140
 Table 55. Words with lenis stops / b, d, ɡ / and fricatives / z, ʒ /. ................................. 144
 Table 56. Words with high tone (sonorants)................................................................... 147
 Table 57. Words with low tone (sonorants).................................................................... 147
 Table 58. Words with rising tone (sonorants)................................................................. 148
 Table 59. Phonetic characteristics of coda segments comparison in Quiaviní Zapotec . 156
 Table 60. Tone-bearing segments in coda in Quiaviní Zapotec ..................................... 157
 Table 61. Breathy vowels and tone interaction............................................................... 179
 Table 62. Munro and Lopez (1999) patterns for what I analyzed here as breathy vowels. ................................................................................................................................. 184
 Table 63. Tone and phonation: modal and breathy vowels ............................................ 186
 Table 64. Creaky vowels and tone interaction................................................................ 187
 Table 65. Munro and Lopez (1999) patterns for what are analyzed here as creaky vowels. ................................................................................................................................. 196
 Table 66. Stimuli: creaky vowels.................................................................................... 202
 Table 67. Control stimuli: modal vowels........................................................................ 202
 Table 68. Creaky vowels: pitch (LiaL) ........................................................................... 204
 Table 69. Probability values from t-test for creaky vowels: pitch (LiaL) ...................... 205
 Table 70. Modal vowels: pitch (LiaL) ............................................................................ 205
 Table 71. Creaky vowels: pitch (TiuC)........................................................................... 206
 Table 72. Probability values from t-test for creaky vowels: pitch (TiuC) ...................... 206
 Table 73. Modal vowels: pitch (TiuC)............................................................................ 207
 Table 74. Creaky-H versus creaky-F: Jitter ppq5 (voice quality)................................... 208
 Table 75. Probability values from t-test for creaky vowels: jitter (TiuC) ...................... 208
 Table 76. Creaky-H vs. L durational patterns (VCfortis) ............................................... 208
 Table 77. Creaky-H vs. L durational patterns (VClenis) ................................................ 209
 Table 78. Laryngeal constriction variation ..................................................................... 210
 Table 79. Interrupted (Checked/rearticulated) vowels and tone interaction................... 211
 Table 80. Quiaviní Zapotec Pronouns and clitics: interrupted (checked)-H .................. 217
 Table 81. Munro and Lopez (1999) patterns for what is analyzed here as interrupted vowels ..................................................................................................................... 226
 Table 82. Stimuli interrupted vowels.............................................................................. 229
 Table 83. Complementary stimuli: modal vowels .......................................................... 229
 Table 84. Pitch results for interrupted and modal vowels — female speaker, LiaL....... 231
  viii  Table 85. Pitch results for interrupted vowels — male speaker, TiuC ........................... 231
 Table 86. Intensity results for interrupted vowels (LiaL and TiuC) ............................... 232
 Table 87. Duration results for interrupted vowels (LiaL and TiuC)............................... 232
 Table 88. Probability values from t-test for duration (LiaL and TiuC) .......................... 233
 Table 89. Voicing results of the second vowel portion (interrupted vowels) (LiaL) ..... 233
 Table 90. Probability values from t-test for voicing (second vowel portion; LiaL) ....... 234
 Table 91. Voicing results of the second vowel portion (interrupted vowels) (TiuC) ..... 234
 Table 92. Probability values from t-test for voicing (second vowel portion; TiuC)....... 234
 Table 93. Interrupted vowels .......................................................................................... 237
 Table 94. Tone & phonation distribution: modal and laryngealized vowels.................. 238
 Table 95. Tone & phonation: modal vs. laryngeal vowels ............................................. 245
 Table 96. Laryngeal constriction variation ..................................................................... 246
 Table 97. Tone & phonation distribution (phonetic implementation) ............................ 246
 Table 98. Tone & phonation distribution in Quiaviní Zapotec....................................... 247
 Table 100. Timing patterns for laryngealized vowels (Quiaviní Zapotec)..................... 255
 Table 101. Laryngeal Feature Mapping (Halle & Stevens, 1971, p. 203)...................... 256
 Table 102. Voice quality feature specification in Quiaviní Zapotec ............................. 264
 Table 103. Voice quality feature specification in Quiaviní Zapotec ............................. 268
 Table 104. Tone & phonation distribution in Quiaviní Zapotec..................................... 268
 Table 105. Munro and Lopez (1999, p. 4) vowel patterns (comparative table) ............. 280
 Table 106. Tone & phonation distribution...................................................................... 281
 Table 107. Tone & phonation distribution...................................................................... 281
 Table 109. Subset of Munro and Lopez (1999, p. 4) vowel patterns, simplified for what is proposed in this dissertation (tone & phonation distribution). ............................... 285
  ix  List of figures Figure 1. Waveform and spectrogram of cha’t [ ʧaʔth ] ‘kiss’ by male speaker TiuL (sound file from Munro et al., 2008)......................................................................... 29
 Figure 2. Waveform and spectrogram of la’t [ lath ] ‘tin can’ by female speaker LiaT. .. 30
 Figure 3. Waveform and spectrogram of naba’j [ naˈβax ] ‘razor’ by male speaker TiuR. ................................................................................................................................... 31
 Figure 4. Waveform and spectrogram of a’s [ as ] ‘hi’ by male speaker TiuL (sound file from Munro et al., 2008)........................................................................................... 32
 Figure 5. Waveform and spectrogram of r-càa’z=a’ [ ɾkha̰ːzaa̰ʔ ] ‘I want…’ by male speaker TiuT. ............................................................................................................ 34
 Figure 6. Box plots and t-test p-values for LiaB: fortis coda vs. lenis coda; fortis coda vs. fortis onset................................................................................................................. 63
 Figure 7. Box plots and t-test p-values for LiaB: fortis onset vs. lenis onset; lenis onset vs. lenis coda............................................................................................................. 63
 Figure 8. Box plots and t-test p-values for TiuC: fortis coda vs. lenis coda; fortis coda vs. fortis onset................................................................................................................. 64
 Figure 9. Box plots and t-test p-values for TiuC: fortis onset vs. lenis onset; lenis onset vs. lenis coda............................................................................................................. 64
 Figure 10. Continuum of phonation types (Ladefoged, 1971)........................................ 104
 Figure 11. Waveforms of voice qualities: modal, breathy and creaky voices. ............... 105
 Figure 12. Waveform and spectrogram of / danj / ˥ ‘harm’ (daany) on the left, and / danj / ˩ (dàany) ‘mountain’ on the right, by male speaker TiuR. ..................................... 109
  Figure 13. Waveform and spectrogram of / ʒi / ˥ ‘tomorrow’ (zhii), on the left, and / ʒi / ˩  (zhìi) ‘quite’, on the right, by male speaker TiuC................................................... 110
 Figure 14. Spectral tilt measurements were taken at five evenly spaced intervals distributed from the onset to the offset of the vowel (Solid lines in the extremes indicate onset and offset of the vowel; dashed lines divide the intervals; and the arrows indicate the points were the measurements were taken). ............................ 115
 Figure 15. Jitter (ppq5 & ddp) mean results (TiuC). ...................................................... 116
 Figure 16. Jitter (ppq5 & ddp) mean results (LiaL)........................................................ 116
 Figure 17. H1-H2 plot for mean results of both speakers............................................... 118
 Figure 18. H1-A1 plot for mean results of both speakers............................................... 120
 Figure 20. Jitter (ppq5 and ddp) mean results (TiuC)..................................................... 130
 Figure 21. Jitter (ppq5 and ddp) mean results (LiaL). .................................................... 130
 Figure 22. H1-H2 plot for mean results of both speakers............................................... 132
 Figure 23. H1-A1 plot for mean results of both speakers............................................... 134
 Figure 24 shows two examples of falling tone in Quiaviní Zapotec. ............................. 136
 Figure 25. Pitch average contours for modal vowels (TiuC).......................................... 140
 Figure 26. Waveform and spectrogram of / dad / Ë ‘father’, by male speaker TiuR. ... 145
  x  Figure 27. Waveform and spectrogram of / ɡiʒ / Ë ‘city person’, by male speaker TiuR. ................................................................................................................................. 146
 Figure 28. Waveform and spectrogram of / n-sualˑ ~ n-sulˑ / ˥ ‘blue’, by male speaker TiuR. ....................................................................................................................... 148
 Figure 29. Waveform and spectrogram of / danʲ / ˥ ‘harm’ and / bal / ˥ ‘bullet’, by male speaker TiuR. .......................................................................................................... 149
 Figure 30. Waveform and spectrogram of / nda / ˩ ‘sensitive’, by male speaker TiuL. The first one shows the word on its own, and the second example includes the 3s clitic (child) / =ɨmˑ /. ....................................................................................................... 150
 Figure 31. Waveform and spectrogram of / bdan / ˩ ‘soot’, by male speaker TiuR. ..... 151
 Figure 32. Waveform and spectrogram of / damˑ / Ë ‘owl’, by male speaker TiuL. And  waveform and spectrogram of / tʃinˑʒ / Ë ‘bedbug’, by male speaker TiuR........... 153
 Figure 33. Waveform and spectrogram of / mes / Ë ‘table’, by male speaker TiuR. ..... 154
 Figure 34. Waveform and spectrogram of / manj / Ë ‘animal’, by male speaker TiuL, and  waveform and spectrogram of / ʐub / Ë ‘dried corn kernel’, by male speaker TiuL. ................................................................................................................................. 154
 Figure 35. Continuum of phonation types (Ladefoged 1971)......................................... 176
 Figure 36. Waveform and spectrogram of / be̤ts / ˩ → [ bè̤tsː ~ βè͡e̤tsː] ‘(man’s) brother’ by male speaker TiuN. ............................................................................................ 180
 Figure 37. Waveform and spectrogram of / be̤ / ˩ → [ bè̤ː ~ βe͡e̤ː] 
 ‘mold (growth)’ by female speaker LiaL. 181
 Figure 38. Waveform and spectrogram of / kṳb / Ü ‘tejate’ by male speaker TiuR....... 183
 Figure 39. Waveform and spectrogram of / na̤ʒj / Ü ‘wet’ by male speaker TiuL (Munro et al., 2008; Sound file: L3-3B) .............................................................................. 183
 Figure 40. Pitch and intensity contours of / na̤ʒj / Ü ‘wet’ by male speaker TiuL ......... 186
 Figure 41. VCfortis example: Waveform and spectrogram of / bḛlˑ / ˥ ‘(woman’s) sister’, by male speaker TiuL (arrows indicate the tense voice portion). ........................... 189
 Figure 42. VClenis example: Waveform and spectrogram of / ɾɡḭbj / ˥ ‘washes’, by male speaker TiuL (arrows indicate the tense voice portion).......................................... 189
 Figure 43. Waveform and spectrogram of / bḛkw / ˩ ‘dog’, by male speaker TiuL (Munro et al., 2008, sound file L3-3C)................................................................... 191
 Figure 44. Waveform and spectrogram of / bdo̰ / ˩ ‘baby’, by male speaker TiuL (Munro et al., 2008, sound file L3-3C). ............................................................................... 191
 Figure 45. Waveform and spectrogram of / mḭʒ / Ü ‘Mixe’, by male speaker TiuL (Munro et al., 2008, sound file L3-3D)................................................................... 193
 Figure 46. Waveform and spectrogram of / ja̰ / Ü ‘up’, by male speaker TiuL (Munro et al., 2008, sound file L3-3D).................................................................................... 194
 Figure 47. Spectrograms and pitch of / bdo̰ / ˩ ‘baby’ and / mḭʒ / Ü ‘Mixe’, by male speaker TiuL. .......................................................................................................... 195
 xi  Figure 48. Waveform and spectrogram of / bdo̰ / ˩ by male speaker TiuT (from personal fieldwork) and by TiuL (Munro et al., 2008, Unida 1; sound file L3-3C) ............. 198
 Figure 49. Waveform and spectrogram of / dḭʒ / ˩ ‘word’ by female speaker LiaCh. ... 199
 Figure 50. Waveform and spectrogram of / bḛl / Ü (bèèe'll) ‘snake’ by male speaker TiuL (Munro et al., 2008, sound file L3-3C)................................................................... 200
 Figure 51. Pitch average contours for creaky vowels (LiaL).......................................... 204
 Figure 52. Pitch average contours for modal vowels (LiaL). ......................................... 205
 Figure 53. Pitch average contours for creaky vowels (TiuC). ........................................ 206
 Figure 54. Pitch average contours for modal vowels (TiuC).......................................... 207
 Figure 55. Waveform and spectrogram of / ɾɡaʔ / ˥ →
 [ ɾɡáʔḁ ] ‘gets green again’ by male speaker TiuC. 215
 ʔ Figure 56. Waveform and spectrogram of / ʒi / ˥ → [ ʒíʔi̥ ] ‘cold’ by male speaker TiuC. ................................................................................................................................. 215
 Figure 57. Waveform and spectrogram of r-càa’z=a’ / ɾka̰zaʔ / ˩ ˥ ‘I want…’ by male speaker TiuT. .......................................................................................................... 218
 Figure 58. Waveform and spectrogram of nàa’ r-àa’p=a’ / na̰ ɾa̰paʔ / ˩ ˩ ˥ ‘I have…’ by female speaker LiaB. .............................................................................................. 218
 Figure 59. Waveform and spectrogram of zhìi’iny=a’ / ʒiʔnjaʔ / Ü ˥ ‘my son’ by female speaker LiaB. .......................................................................................................... 219
 Figure 60. Waveform and spectrogram of / ɾɡjaʔ / ˩ →
 [ ɾɡjàʔà ] ‘dances’ by male speaker TiuC. 221
 ʔ Figure 61. Waveform and spectrogram of / btja / ˩ → [ btjàʔà ] ‘epazote’ by male speaker TiuC. ....................................................................................................................... 221
 Figure 62. Waveform and spectrogram of / ʒiʔ / Ü → [ ʒí ʔì ] ‘nose’ by male speaker TiuC. ....................................................................................................................... 224
 Figure 63. Waveform and spectrogram of / btjaʔ / Ü → [ btjáʔà ] ‘scrapped’ by male speaker TiuC. .......................................................................................................... 224
 Figure 64. Waveform and spectrogram of / ʒi / ˥ → [ ʒíː ] ‘tomorrow’; / ʒ ḭmˑj / 
 ˥ → [ʒí̬mjː] ‘basket’; and / ʒiʔ / ˥ → [ ʒíʔi̥ ] ‘cold’ by male speaker TiuC.  240
  Figure 65. Waveform and spectrogram of / ɾbanj / ˩ → [ ɾbàːɲ ] ‘survives’; / ba̰ / ˩ → [ bà̰ː ] ‘tomb’; / baʔ / ˩ → [ bàʔà ] ‘eyeball’ by male speaker TiuC. ......................... 242
  Figure 66. Waveform and spectrogram of / ʒilj / Ü → [ ʒîːlj ] ‘sheep’; / ʒil̰j / Ü → [ ʒḭ̂ːlj̥ ]  ‘cotton’; / ʒiʔ / Ü → [ ʒíʔì ] ‘cold’ by male speaker TiuC. ..................................... 244
 Figure 67. Continuum of phonation types (Ladefoged, 1971)........................................ 260
 Figure 68. Box plots and Wilcoxon tests of female results for stops. ............................ 301
 Figure 69. Box plots and Wilcoxon tests of female results for fricatives....................... 302
 Figure 70. Box plots and Wilcoxon tests of female results for nasals............................ 303
 Figure 71. Box plots and Wilcoxon tests of male results for stops................................. 304
 Figure 72. Box plots and Wilcoxon tests of male results for fricatives.......................... 305
  xii  List of abbreviations ‘-’ = affix boundary ‘=’ = clitic boundary ( ) = foot boundary √ = root 1p = first person plural 1s = first person singular 2p = second person plural 2s = second person singular 3p = third person plural 3s = third person singular C = consonant dB = decibels F1 = first formant F2 = second formant F3 = third formant Ft = foot Hz = hertz IP = intonation Phrase MA = manner of articulation ms = milliseconds N = nucleus O = obstruent obj = object OT = Optimality Theory PA = point of articulation Pos = possessive PPhrase = Prosodic Phrase PWord = Prosodic Word R = resonant (sonorant) s = singular S = stress SPE = The Sound Pattern of English (Chomsky & Halle, 1968) subj = subject SVO = Subject-Verb-Object word order TBU = tone-bearing unit us = unstressed V = full vowel V = vocal VOS = Verb-Object-Subject word order VOT = Voice Onset Time VSO = Verb-Subject-Object word order  xiii  Acknowledgements First and foremost, I thank the Zapotec community of San Lucas Quiaviní and Zapotec residents of the Los Angeles area. I am especially grateful for the friendship and kindness of two families: the López and the Cruz Aguilar. Francisco López has been my main consultant throughout my research. I thank him for his patience, intuitions, as well as for walking me through San Lucas Quiaviní and through his language. Thanks also to Felipe Lopez, the beginner of the Quiaviní Zapotec linguistic documentation, as well as to their sister Avelina López and their father Alfredo López. The Cruz Aguilar family has always treated me as part of their family. My sincere thanks to my uncle Rogelio Cruz, my aunt Isabel and my cousins Naú, Lupe, Julia, Alberto and Andrés: Great consultants and beautiful people. I owe thanks to many other Zapotec speakers who have helped me in my research, including Alberto Curiel, Brígida Núñez, Don Cayetano and family, Efraín, Don Emiliano, the Martínez family (Don Felipe, Doña Magdalena, Esther, Misael), the Olivera Aguilar family (Doña Victoria, Graciela, Celina, Virginia), the extended López family (Don Gerardo, esposa e hijas) for their hospitality, Don Hipólito López, Irma, Don Leonardo, Marcelino Antonio Nuñez, y Víctor Cruz, as well as many children who taught me tons of Zapotec vocabulary and shared with me their toys and smiles. This Zapotec community has generously shared their incredible language and culture with me. In return, I hope this thesis will contribute towards the documentation of the language and to maintain it as well. Xtyozën yuad. This thesis could not have been written without the unwavering help and support from my co-supervisors, Joseph Stemberger and Douglas Pulleyblank, and my committee members, Bryan Gick and Gunnar Hansson. Their constant encouragement, advice and invaluable patience were essential throughout all the stages of my dissertation writing. Joe was my mentor since my arrival at the University of British Columbia (UBC). He supported my Ph.D. from the beginning and shared with me an immense amount of knowledge from all kind of topics in thoughtful discussions. I sincerely thank him for his generosity, guidance and for sharing the ups and downs of fieldwork, either drinking a delicious tejate, running away from the rain, or awaiting for a car repair. Doug’s ability to challenge every thought, formulate the keen questions, and read my work over and over are the hallmarks of a great supervisor. Thank you for your constant encouragement and advice in both academic and personal matters. I appreciate Bryan’s global vision of scholarship: his constant reminder that its important to see the big picture. I thank him for the experimental support and thoughts on general research, and I will always remember his ability to “exasperate” my ideas and his willingness to fight for explanations. I admire Gunnar for his genuine passion for linguistics, and I am deeply grateful for his extensive and thought provoking comments, constant support, and his challenge to me to come up with the best analysis. To my committee: ¡Gracias!  xiv  I was very fortunate to have had Moira Yip as my external examiner, and Carla Hudson Kam and and Valter Ciocca as my university examiners. I appreciate the careful reading they provided to my thesis, with plenty of questions and thoughtful comments. Apart from my committee, a number of professors helped me on my Ph.D. journey. Thank you to Pat Shaw, for an immense amount of insights into phonology. Thanks to Martina Wiltschko and Rose-Marie Déchaine for most of the morpho-syntax I know, and thanks to Lisa Matthewson and Hotze Rullmann for sharing their enthusiasm for semantics. In addition to those I have just named, Patricia Bonaventura, Emily Curtis, Henry Davis, Erik Bateson-Vatikiotis, Molly Babel gave me invaluable teachings and support. Special thanks to Edna Dharmaratne, our department secretary and administrator, but in fact, she is much more than that. In terms of peer support, words are not enough to thank my cohorts for their help, particularly during the first two years of my Ph.D. These are Calisto Mudzingwa, Maria Amelia Reis Silva, Seok Koon Chin (Shujun), Donald Derrick and Ryan Waldie. Thank you all. In particular, I have a long-standing friendship with a determined and resilient buddy who never gives up, Calisto Mudzingwa, as well as with an inspiring Brazilian lady, Maria Amelia Reis Silva. You guys made my rainy Vancouver days surprisingly sunnier. Mazvijita! Obrigado! Thanks are also due to Jason Brown, Raphael Girard, Jeff Muehlbauer, Fusheini Hudu for their friendship and help at different stages of my academic journey at UBC. I would like to also thank my other peers at UBC in the Linguistics department: Solveiga Armoskaite, Leszek Barczak, Clare Cook, Chenhao Chiu, Joel Dunham, Analía Gutiérrez, John Lyon, Masaki Noguchi, Martin Oberg, Tyler Peterson, Gessiani Picanço, Beth Rogers, Murray Schellenberg, Mark Scott, Anita Szakay, Sonja Thoma, James Thompson, and Adriano Vilela Barbosa. Outside UBC, I owe a lot of gratitude to Francisco Arellanes, old friend and incomparable linguist. Gracias Paco, por tu sincera amistad, por las charlas lingüísticas y no lingüísticas, y por compartir tu pasión por la fonología conmigo. I also want to thank Verónica Vázquez, my B.A. supervisor and the linguist who showed me the beauty of linguistic typology, but specially the love for (Mexican) indigenous languages. Gracias Vero. I owe Pam Munro a special place in these acknowledgements. She introduced me to Quiaviní Zapotec through her work and she has always been incredibly generous with me, willing to listen and discuss my stubborn ideas and willing to give me her invaluable advice. I sincerely admire all the work she has done for this language. This thesis has also benefited from comments and suggestions received from many other scholars: José Elías Ulloa, John Esling, Matthew Gordon, John Harris, Esther Herrera, Larry Hyman, Yuwen Lai, Felicia Lee, Brook Lillehaugen, Stephen Marlett, Scott Moisik, Kenneth Stevens, Jie Zhang, and particularly to our dear professor Thomas Smith Stark †. Thank you all. The friendship of many people gave me motivation during all these long and short 6 years. Beginning with (tantos y tan chidos lengweros): Paco, Samuel, Lucero, Maribel, Rochi, Ruy Rodrigo Romero, Vero Muñoz-Ledo, Vero mala, el Rogu Rodrigo Gutiérrez, Anita, Elenita, Carlitos, Blanca, Isra, Michelín, Rafa, Esaú, Julio, Maribel, Toño, Violeta; and Alejandro de la Mora, Vero Vázquez, Rebeca Barriga, Pedro Hernández, Roberto Zavala, y B’alam.  xv  Also thanks to companions that have been there from younger stages: Jaznum, mi carnal Toño, el compa; my dear guayabo friends (TPZ5); and finally, my oldest friends, Fernando Carreño, Jorge Luis Herrera, Julio Ornelas and Rubén Vargas. Gracias hermanos, por las chelas, los tequilas, los emilios, los saludos, los abrazos, y los recuerdos. Mil gracias. Saving the best for the last, I am deeply grateful to my family and their unconditional help: my parents and siblings (Jefa, muchísimas gracias, jefe, carnal, pequeña, gracias por todo el amor que me han brindado en estos años y en toda la vida). I am grateful for your wonderful nourishment. Thanks also to my mother and sisters in law; and thanks to my father in law and younger siblings in law. The distance was always easier to bear with your close warm. I have mentioned a lot of wonderful people here, but there is one person who supported me daily, with lots of patience, love, and sacrifice during my studies in Canada: my wife, Paola. Gracias morenita blanca, you are the best. All my achievements are yours. And a special thank you to my wonderful daughter Lía Inés for enduring the last stages of my work; long nights of work, and long nights of crying, but days full of smiles and discovery. For the most beautiful, to you Lía, I dedicate my work.  Sincere thanks are due to all. Vancouver, August 11, 2010  During my first year, my Ph.D. was funded by a Graduate Entrance Scholarship from the University of British Columbia, and a Graduate Complementary Scholarship from SEP (Secretaría de Educación Pública). For the rest of my program, I had a Doctoral Scholarship granted by CONACYT (Consejo Nacional de Ciencia y Tecnología) as my main funding source. Complementarily, I was funded by a SSHRC (Social Sciences and Humanities Research Council of Canada) project granted to Dr. Joseph Stemberger. During my last year, I also received a Complementary Scholarship from CIESAS (Centro de Investigaciones y Estudios Superiores en Antropología Social). I thank all these funding sources. My fieldwork research was also funded by the SSRCH project mentioned above, an AMS Initiative fund (UBC) and by a Jacobs Research Fund grant. I thank each of these sources for making my work possible. Mexico City, August 12, 2010  xvi  Dedication  To my daughter, Lía Inés  xvii  Preface (summary and committee members)  Name:  Mario E. Chávez-Peón  Degree:  Ph.D.  Title of Thesis:  The Interaction of Metrical Structure, Tone, and Phonation types in Quiaviní Zapotec  Committee:  Dr. Douglas G. Pulleyblank Co-Supervisor Dr Joseph P. Stemberger Co-Supervisor Dr. Bryan Gick Committee member Dr. Gunnar O. Hansson Committee member  xviii  Chapter 1:  The thesis and the language 1.1  The thesis  This dissertation investigates the phonetics and phonology of San Lucas Quiaviní Zapotec (henceforth, Quiaviní Zapotec), an Otomanguean language spoken in southern Mexico, in the state of Oaxaca. Specifically, I examine the interaction of metrical structure, tone, and phonation types. This study proposes a unified account for these patterns, explaining their individual characteristics and how their interaction is constrained. Two topics are discussed in detail: the role of the mora as the link for different patterns in the phonology of this language and the mapping between phonology and phonetics in the expression of laryngeal features. The goals of this dissertation are twofold. First, the description and analysis of these phenomena will improve our understanding of the phonology of Quiaviní Zapotec. Second, this research will explore the implications of tone-phonation interactions in phonological theory and contribute to the growing literature on this subject (Silverman, 1997a, 1997b; Herrera, 2000; Blankenship, 2002; DiCanio, 2008, among others), as well as the role of metrical structure in such interactions. This chapter aims to provide a basic overview of Quiaviní Zapotec and its speakers as well as the basic features of Quiaviní Zapotec phonology and morpho-syntax.  1  Chapter 2 shows that vowel length in Quiaviní Zapotec is dependent on the type of syllable and on the type of coda consonant: stressed vowels are short before fortis consonants (both obstruents and sonorants), and long before lenis consonants or in open syllables; as such, the categorization of this vowel length pattern relies on the fortis/lenis contrast, which entails a complex set of phonetic properties encoded with the feature [+/-fortis] under an emergent feature approach (Mielke 2008 [2004]; Pulleyblank, 2006). The third chapter describes and analyzes Quiaviní Zapotec metrical structure. The goal is to account for word stress in this language; thus, the domain of analysis is the prosodic word. According to Munro and Lopez (1999), the last syllable of uninflected words is stressed in Quiaviní Zapotec (referred to as the key syllable by the authors, p. 3), but no subsequent study has examined more details of the prosodic system of this language. I discuss prosodic issues like moraicity and minimality, as well as foot structure, building from monosyllables, up to morphologically complex disyllabic and longer words. The analysis of prominence provides a foundation for the other two central topics of this dissertation: tone and phonation types. Chapters 4 and 5 examine the tonal system of Quiaviní Zapotec. Munro and Lopez (1999) argue that tone is predictable from phonation types in Quiaviní Zapotec. I put this claim to question in Chapter 4, analyzing instrumentally the voice quality of lexical items with low, rising and falling tones that appear to have modal voice. Results show that tone is contrastive within modal voice; consequently, a new categorization is presented for particular lexical items. These findings are then taken into account in the analysis of non-modal phonation (Chapter 6). Chapter 5 establishes the association between moraicity and tonal patterns in Quiaviní Zapotec. This chapter relates the metrical structure proposed in Chapter 3 to the tone findings of Chapter 4, where tone is established as a contrastive feature in this language. I analyze Quiaviní Zapotec non-modal phonation in Chapters 6 and 7. The goal of Chapter 6 is to provide descriptive generalizations governing breathy (/ a̤ /), creaky (/ a̰ /) and interrupted (/ aʔ /) vowels in Quiaviní Zapotec. I present a detailed description of each type of vowel and clarify their underlying representations along with their phonetic  2  realizations. In light of the controversial distinction between creaky and interrupted vowels, acoustic comparisons are presented, supporting the contrast between two degrees of laryngealization in Quiaviní Zapotec. The laryngeal complexity of Quiaviní Zapotec, a language with both contrastive tone and contrastive phonation, is accounted for in an integrated fashion in Chapter 7. I make a proposal for the laryngeal specifications in Quiaviní Zapotec and provide a comprehensive phonological representation for vowels, in terms of featural and prosodic information. Finally, the chapter examines the phonetic implementation of phonological features and seeks to test the hypothesis that the surface complexity of this language derives from a simpler phonological representation.  1.1.1 General methodology The two sources of data for this study are first-hand data, collected in the town of San Lucas Quiaviní and in the Los Angeles area, from fluent native speakers of the language, and Munro and Lopez’ (1999) dictionary of Quiaviní Zapotec. My field research was conducted in Mexico in the summers of 2005, 2006, 2007 and 2008, as well as in Los Angeles in May and September 2009. Throughout these periods, I recorded individual elicitation sessions with different speakers. Recordings were made with a Marantz 660 solid-state recorder and a lapel Countryman microphone (phantom power), and digital files were stored on the computer and burned onto CDs. Acoustic analysis included the use of Praat for Mac (version 5.1.07; Boersma & Weenink, 2009) and statistical programs. In the following chapters, I explain in detail the specific methods, structure of the tasks, and stimuli for the phonetic experiments. The dictionary of Munro and Lopez (1999) is a seminal and groundbreaking study of Quiaviní Zapotec, and has been an essential source at all stages of my research. Many generalizations, minimal pairs, elicitation plans, etc, were facilitated by this study. The Quiaviní Zapotec second-language course Cali Chiu? (Munro, Lillehaugen, & Lopez, 2008) has also been a constant reference guide.  3  1.2  The language This section provides a genetic and geographical background of Quiaviní  Zapotec, as well as discussion of the previous work on the language, followed by an overview of the phonological and morpho-syntactic properties of Quiaviní Zapotec. In these latter sections, a large proportion of the basic information described here, as well as much of the terminology I adopt, was first observed, documented and proposed in the Quiaviní Zapotec dictionary (Munro & Lopez, 1999). However, my proposal regarding the phonology of Quiaviní Zapotec presents a considerable reanalysis that is developed in detail in later chapters of this study. Quiaviní Zapotec data come from my own fieldwork unless otherwise indicated.  1.2.1 Genetic and geographic background Quiaviní Zapotec is spoken in southern Mexico, in the state of Oaxaca. It belongs to the Zapotec language family, which is part of the Otomanguean stock. Zapotec languages are divided into three subgroups (Kaufman, 1994): Northern, Central, and Southern Zapotec. Central Zapotec includes the variants of the Valley, where Quiaviní Zapotec is spoken, and the Isthmus. The exact number of Zapotec languages is under debate; mutual intelligibility declines rapidly within relatively short distances. The SIL Ethnologue (Grimes, 2005) currently lists 58 Zapotec languages, but other scholars believe there are only 15 (T. Kaufman, personal communication, October 2007). Quiaviní Zapotec is spoken in the town of San Lucas Quiaviní, in the Central Valley of Oaxaca state. The town has a population of close to 2000 people, most of whom speak Zapotec as their first language; nevertheless, Spanish is encroaching on the Zapotec community because of the matrix culture, media, schooling, jobs, etc. In addition, many families have re-located to the United States, into the greater Los Angeles area (probably around 2000 people). As a result the language is considered threatened by both Spanish and English. 4  1.2.2 Previous work Munro and Lopez’ (1999) dictionary of Quiaviní Zapotec constitutes the first comprehensive study of this language. Since then, more studies have investigated morpho-syntactic aspects of the language, including two M.A. theses (Méndez, 2000; Lillehaugen, 2003), three Ph.D. dissertations (Galant, 1998; Lee, 2006 [1999]; Lillehaugen, 2006), research articles (Munro, 1996, 2003, 2006, among others) and a second-language course (Munro et al., 2008). An ongoing project on First Language Acquisition in San Lucas Quiaviní Zapotec directed by Dr. Joseph Stemberger at UBC has focused on phonological development in this language (Stemberger & Lee, 2007; Chávez-Peón, Stemberger, & Lee, in press). I have carried out continuous fieldwork since 2005 with children and adults, analyzing phonological (Chávez-Peón, 2006, 2008a, 2008b) and morphosyntactic (Chávez-Peón & Mudzingwa, 2008) aspects of this language. A number of closely related languages spoken in the region surrounding San Lucas have been also documented: San Juan Guelavía Zapotec (Jones & Knudson, 1977), Santa Ana del Valle Zapotec (Broadwell, 1991; Esposito, 2003; Rojas, 2010), San Pablo Güilá Zapotec (López Cruz, 1997; Arellanes, 2009), and Mitla Zapotec (Briggs, 1961; Stubblefield & Stubblefield, 1991).  1.2.3 Phonology This section provides an overview of Quiaviní Zapotec phonology. The purpose is to present the segmental inventory and the topics that are analyzed in detail in subsequent chapters. In addition, two sections are background for the rest of the dissertation: Quiaviní Zapotec phonotactics and general morphosyntactic characteristics. These sections are mainly based on Munro and Lopez (1999). Quiaviní Zapotec has a complex phonetic and phonological system, which includes a pervasive contrast between fortis and lenis consonants, phonemic distinctions  5  among four phonation types (voice qualities), tone and stress patterns and a complex syllable structure. First, following Munro and Lopez (1999), I present the phonemic inventory. Consonants The inventory of consonants in Quiaviní Zapotec is presented in Table 1 (Munro & Lopez, 1999), with standard orthography (where different) in parentheses. Table 1. Quiaviní Zapotec Consonant inventory  Plosive  fortis lenis  Nasal  fortis lenis  Bilabial Lab-dent Dental/ Prepalatal Retroflex Palatal Velar Alveolar p t k (c/qu) b d ɡ mˑ (mm) m  nˑ (nn) n  Trill Tap Fricatives fortis Lateral  lenis fortis lenis  Affricate fortis  f  r (rr) ɾ (r)  ŋˑ (nng) ŋ (ng)  s  ʃ (x)  ʂ (x:)  z  ʒ (zh)  ʐ (zh:)  lˑ (ll) l ts  Glides  x (j)  ʧ (ch) j (y) w  Similar to other Zapotec languages, Quiaviní Zapotec has a fortis/lenis contrast in its consonant pairs, rather than a strict voiced/voiceless opposition. Fortis obstruents are voiceless, never fricated (in the case of stops), and relatively long. Lenis obstruents are often (but not always) voiced, variably fricated, and relatively short. For sonorants, the main difference between fortis and lenis is duration, with fortis being longer. Chapter 2 provides more details on the fortis/lenis issue. There is no consensus on how to represent the fortis/lenis contrast. Among obstruents, voicing is normally a salient difference, thus, most studies rely on voicing to represent the fortis/lenis contrast; voiceless symbols are used for fortis consonants (e.g. / p t k … /) versus voiced for lenis consonants (e.g. / b d g … /). This convention is  6  adopted in this study. Sonorants embody a greater challenge with respect to the representation of the fortis/lenis distinction, since they basically rely on duration. For this study, in order to maintain the harmonization of phonetic and phonological representation, I represent fortis sonorants with the semi-long IPA symbol (e.g. / nˑ /), and lenis sonorants as plain ones (e.g. /n/).  Vowels Quiaviní Zapotec has the following six monophthongal vowels: / i, ɨ, u, e, o, a /, distributed as shown in Table 2. (Diphthongs are presented in the phonotactics section below.) Table 2. Quiaviní Zapotec vowels high  front i  mid  e  central ɨ (ë)  low  a  back u o  Some variation in Quiaviní Zapotec vowels include tense-lax allophones [ i ~ ɪ, e ~ ɛ, ɨ ~ ʌ ] and to a lesser degree [ u ~ ʊ, o ~ ɔ ] (Stemberger, Chávez-Peón, & Lee, 2007). The high central unrounded vowel, / ɨ /, appears less frequently than other vowels; some speakers use it only rarely, replacing it with [e] in most contexts. The low vowel seems to be used as the default in epenthetic contexts. Phonation and tone Munro and Lopez (1999) recognize modal, breathy, creaky and checked vowels in Quiaviní Zapotec; Gordon and Ladefoged (2001) describe the phonetic properties of the first three.  7  (1) Quiaviní Zapotec phonation types a. b.  Modal Breathy  /a/ / a̤ /  c.  Creaky  / a̰ /  d.  Checked  / aʔ / 1  In the orthography, “Diacritic symbols indicate phonation type: Vh represents a breathy vowel (ah, eh, ëh, ih, oh, uh), and V’ a checked (interrupted) vowel (a’, e’, ë’, i’, o’, u’). A creaky vowel is indicated with a grave accent (à, è, ì, ò, ù), except for the vowel ë, for which creakiness is indicated with a circumflex accent (ê). Vowels without one of these diacritics have plain (modal) phonation” (Munro & Lopez, 1999, p. 4). In addition, these scholars claim that syllable nuclei “may contain up to three individual vowels, each with its own phonation” (p. 3). (See vowel patterns in Table 3.) Munro and Lopez (1999) recognize Quiaviní Zapotec as a tone language; however, the authors state that “tone melodies on Quiaviní Zapotec vowel complexes [syllable nuclei] are derived from the number and phonation type of the vowels in the complex and its phonological environment rather than representing primary contrasts” (Munro & Lopez, 1999, p. 3). According to Munro and Lopez (1999), the chart below presents the major vowel patterns (syllable nuclei) in Quiaviní Zapotec. These vowels are represented with the vowel a (and with ia for patterns that occur only with diphthongs). Each vowel pattern includes one example, its derived tone, and its combination form.2  1  The glottal stop is analyzed as a property of the vowel rather than as an independent consonant. I discuss this issue in detail in Chapter 6. 2 According to Munro et al (2008, Unida 1, p. 32) “many Valley Zapotec words shorten to simpler COMBINATION FORMS when endings are added to them, or when they occur with other words following them.”  8  Table 3. Munro and Lopez (1999, p. 4) Quiaviní Zapotec vowel patterns Combination aa (same) ia a' (same) ah (same) ah àa (same) a'a (same) a'a a'a a'a a' (final), a'ah (same; non-final) 12 a'ahah a'ah 13 a'aah a'ah 14 a'aha a'ah 15 aa'ah aa' (final), a'ah (same; non-final) 16 a'aa' aa' 17 aa' aa' (same) 18 a'àa àa 19 ààa' àa' 20 a'àa' àa' 21 àa'ah àa' 22 ààa'ah àa' 23 àa' àa' (same) 24 àa'a+n àa'a (same) 25 aàa'ah aàa' 26 aàa' aàa' (same) 27 aahah aah 28 iiah aah 29 aah aah (same) 30 àah àah (same) 31 ahaha aha 32 aaha' aha' 33 aha' aha' (same) 1 2 3 4 5 6 7 8 9 10 11  Pattern aa iia a’ ah ahah àa a'a a'aa àaa àaa' a'ah  Examples rdaa badiia tyo'p zah bihih bòo gyi'izh chi'iinnzh nnàaan rsìii’lly zhi'ih  ‘gets bitter’ ‘roadrunner’ ‘two’ ‘grease’ ‘air’ ‘charcoal’ ‘city person’ ‘bedbug’ ‘mother’ ‘morning’ ‘nose’  Tone high high high low low low rising rising rising rising falling  gahll gui'ihihzh ‘sickness’ be'euh ‘turtle’ re'ehiny ‘blood’ baa'ah ‘earlier today’  falling falling falling falling  bi'ii'by ‘pipe (plant)’ bax:aa't ‘toad’ zhi'ìilly ‘sheep’ bèèe'll ‘snake’ zhi'ìi'zh ‘pineapple’ bàa’ah ‘eyeball’ rcwààa'ah ‘throws’ bèe’ll ‘sister’ zhìi'iny ‘son’ rloòo'oh ‘floods’ zhiìi'lly ‘cotton’ iihahz ‘year’ cu'liiahd ‘altar boy’ baahlly ‘flame’ rzùahz ‘gets drunk’ curehehizh ‘cabbage’ barcwiaha'cw ‘bwitch’ nsehe's ‘fast’  falling falling falling falling falling falling falling falling falling falling falling falling falling falling falling falling falling falling  In Munro et al. (2008), these 33 vowel patterns are reduced to 20. The 13 vowel patterns that were not included in this work are underlined in the table above. In this dissertation, I refer most of the time to the original Zapotec dictionary vowel patterns (Munro & Lopez,  9  1999), but I also commonly cross-reference the simplification in Munro et al. (2008), and both analyses are considered in the concluding chapter. Clearly, tone and phonation represent the most challenging issues in the phonology of Quiaviní Zapotec. Munro and Lopez’ (1999) analysis is the first comprehensive account for these issues, with particular focus on the orthographic representation as part of the Quiaviní Zapotec dictionary. These authors, nonetheless, acknowledge that “our analysis of San Lucas Quiaviní Zapotec tone and phonation is ongoing” (Munro & Lopez, 1999, p. 5). Based on their previous work, this dissertation seeks to continue this analysis. In addition, the present work acknowledges and adopts many aspects of Munro and Lopez (1999), including the consonant and vowel inventory, the fortis/lenis distinction among both obstruents and sonorants, the tone melodies (high, low, rising, falling), the four-way phonation contrast, the stress and loanword description, and basically all the morphosyntactic analysis. In what follows, I present a synopsis of my analysis of tone and phonation types in Quiaviní Zapotec, developed in subsequent chapters. I argue that tone is contrastive. The analysis is presented in detail in Chapter 3. Here, I illustrate the tone melodies with the following minimal and near-minimal sets. (2)  a. High tone  / ʒi /  ˥ → [ ʒíː ]  b. Low tone  / ʒi /  c. Rising tone  / ʒilj / Ë → [ ʒǐːl ]  ‘tomorrow’  ˩ → [ ʒìː ] j  ‘quite’ 3  ‘saddle’  d. Falling tone  / ʒilj / Ü → [ ʒîːlj ]  ‘sheep’  e. High tone  / nda / ˥ → [ ndáː ]  ‘bitter’  f. Low tone  / nda / ˩ → [ ndàː ]  ‘sensitive’  g. Rising tone  / dad / Ë → [ dǎːð ]  ‘father’  h. Falling tone  / nda̰ / Ü → [ ndâ̰ː ]  ‘hot’4  3  Underlyingly, glides are represented as / j, w /, which basically correspond to a vocalic segment without a mora. On the surface, they may have different realizations, such as secondary articulation of a consonant (e.g. / ʒilj / Ë → [ ʒǐːlj ] ‘saddle’), as a palatal nasal (e.g. / ʒiʔnj / Ü → [ ʒíʔìɲ ] ‘son’), or as part of the onset (e.g. /njet/ ˥ → [njét] ‘Anita’). 4 As we will see in Chapter 3, falling tone is mostly found with non-modal voice. In order to keep this contrastive set as similar as possible, the voice quality of the last example is creaky.  10  As the examples above illustrate, in the underlying representation (UR), tone is transcribed with the tone letters ˥ ˩ Ë Ü, whereas in the surface form it is indicated with the accent symbols: [ é è ê ě ]. Both are equivalent IPA symbols to represent tone; nonetheless, the accent marks allow for a more precise phonetic transcription, necessary, for example, in diphthongs (e.g. / beu / Ü → [ béù ] ‘moon’) and non-modal vowels (e.g. /n-ɡaʔ/ Ü → [ ŋɡáʔà ] ‘green’). This convention is adopted throughout the dissertation. Vowel length is not lexically contrastive, but is prosodically relevant (Chapters 2 & 3), and is therefore only indicated in phonetic transcriptions of surface forms. With respect to Quiaviní Zapotec phonation types, different acoustic analyses and phonetic experiments in the subsequent chapters support the four-way contrast in Quiaviní Zapotec proposed by Munro and Lopez (1999). However, I reanalyze some of the Munro and Lopez (1999) vowel patterns and advocate explaining some of the surface complexity as phonetic implementation of phonological features (Chapters 6 & 7). These contrasts are illustrated by the following contrastive sets.5 (3)  (4)  a. Modal  / be / → [ beː ]  ‘mesquite bean’  b. Breathy  / be̤ /  ‘mold (growth)’  c. Creaky  / bḛ / → [ bḛː ]  → [ be̤ː ]  ‘notch made in a sheep's ear’  d. Interrupted  / be / → [ be e ]  ‘mushroom’  a. Modal  / lat / → [ lat ]  ‘(tin) can’  b. Breathy  / la̤t / → [ la̤t ]  ‘place’  c. Creaky d. Interrupted  ʔ  ʔ  / la̰ts / → [ la̰ts ] ʔ  ̰ʔ  / na / → [ na a ]  ‘flat area’ ‘heavy’  Phonation types refer to the manner in which vocal folds vibrate. Quiaviní Zapotec includes modal voice, which consists of regular vibration of the vocal folds (the standard vibration type), breathy phonation (or murmur), where the folds are held partly apart while the vibration continues, creaky voice, where folds are held stiffly and vibration is partially inhibited, and interrupted vowels, represented as modal voice followed by a 5  In order to reduce the amount of information and to focus on phonation types, I have left out tone from the transcriptions.  11  glottal closure. Both creaky and interrupted vowels are referred to as laryngealized vowels. Interrupted, which can also be referred to as glottalized voice, is controversial as a unified phonation type. However, there is solid evidence for analyzing the glottal feature as part of the vowel, and not as an independent segment in Quiaviní Zapotec (Chapter 6). Cross-linguistically, the literature on voice qualities supports glottalized voice as a possible laryngeal setting (Gordon & Ladefoged, 2001; Edmondson & Esling, 2006, among others). In my analysis of Quiaviní Zapotec, interrupted vowels / aʔ / may be realized as either checked, [ aʔ ] (with high tone), or rearticulated, [ aʔa ] (with low and falling tones). This terminology will be used throughout this work. As anticipated, tone and phonation interact closely in Quiaviní Zapotec. Table 4 illustrates this interaction. Table 4. Tone and phonation co-occurrence in Quiaviní Zapotec High Modal √ Breathy X Creaky √ Interrupted √  Low √ √ √ √  Falling √ √ √ √  Rising √ X X X  Modal vowels may have all four tones. Within non-modal phonation, breathy vowels appear with low and falling tones, whereas laryngealized vowels, both creaky and interrupted, appear with high, low and falling tones. This distribution will be exemplified and analyzed in detail in Chapters 6 and 7.  1.2.4 Phonotactics The phonotactics of a language are concerned with restrictions on the permissible combinations of phonemes. They define permissible syllable structure, consonant clusters, and vowel sequences by means of phonotactic constraints. These conditions and constraints will be important in the following chapters, in defining characteristics of prominent syllables and tone-bearing segments, among other things. This section is based on the Quiaviní Zapotec dictionary (Munro & Lopez, 1999). 12  Languages of the world differ in their syllable phonotactics. Some languages are extremely restrictive and allow only CV sequences; others allow more complex structures both in the margins and nuclei. Across languages, segments are organized into wellformed sequences according to universal principles of segment sequencing. The organization of segments within the syllable (and possibly across syllables) is traditionally assumed to be driven by principles of sonority, a property that ranks segments along a hierarchy from most sonorous to least sonorous. A number of strong cross-linguistic tendencies on the distribution and sequencing of segments are explained with reference to the sonority hierarchy, where obstruents (subdivided into stops and fricatives) have the lowest sonority and vowels are the most sonorous. (5) Sonority Hierarchy (SH): O(bstruent) < N(asal) < L(iquid) < G(lide) < V(owel) Principles such as the Sonority Sequencing Principle, introduced as early as the 19th century by Sievers (1881), and later by Jespersen (1904), explains, for instance, the tendency, within a syllable, for more sonorous segments to stand closer to the syllable peak than less sonorous ones. (6) Sonority Sequencing Principle (SSP) Sonority increases towards the syllable peak and decreases towards the syllable margins With respect to Quiaviní Zapotec, all consonants may appear in singleton onsets and singleton codas. As for consonant clusters, this language has a wide variety of sequences. Below, I present all licit clusters in onset position in terms of sonority (manner of articulation). As the purpose of these examples is to illustrate consonant sequences, a phonemic transcription is sufficient. Examples in this study are always presented with morphological boundaries; verbs are shown in the habitual form.  13  (7) CC Rising sonority stop + fricative stop + nasal stop + liquid stop + glide  ‘José’  (< Sp. José) 7  ‘jackrabbit ‘balloon’ ‘tree’  (< Sp. globo)  fricative + nasal  / ʃnia /  ‘red’  fricative + liquid  / fɾuat /  ‘fruit’  fricative + glide  / baɾʃje̤k /  ‘mountain turkey’  nasal + liquid nasal + glide liquid + glide  / nɾḛḭnjdṵa̰t / ‘soft and tender’ / nja / ‘clean’ / ljḛʐ / ‘misfortune’  (8) CC Equal sonority stop + stop fricative + fricative (+ glide) nasal + nasal  / bdo̰ / / fsjuan /  ‘baby’ ‘coral snake’  / mna̰ /  ‘woman’  / ɾ-lo̰ / None  ‘floods’  / ʃtḛ /  ‘of, about’  nasal + stop nasal + fricative liquid + stop  / n-dṵa̰ʃ / / n-sual / / ɾ-ɡḛz /  ‘powerful’ ‘blue’ ‘hugs’  liquid + fricative  / ɾsillj /  ‘early morning’  liquid + liquid glide + glide (9) CC Reversed sonority fricative + stop  6  / bse / 6 None / bli̤a̤n / / glob / / ɡja̤x /  liquid + nasal  / ɾmudj /  ‘medicine’  glide + stop (+ glide) glide + fricative (+ glide) glide + nasal (+glide) glide + liquid  / wbwi̤ʒ /  ‘sun’  / wʒjar /  ‘spoon’  / wnja̰ /  ‘traditional healer’  / wli̤a̤z /  ‘daughter-in-law’  (< Sp. fruta)  (Sp. azul)  (< Sp. remedio)  (< Sp. cuchara)  In sequences of lenis stop plus another segment, the initial consonant may be fricated, e.g. /bse/ → [bse ~  βse] ‘José’; or /bdo̰/ → [bdo ~ βdo̰] ‘baby’ below (the latter creates a reversed sonority cluster). 7 Following the convention of the Quiaviní Zapotec dictionary (Munro & Lopez, 1999), after borrowings I include in parentheses the symbol < Sp. and the Spanish spelling of the source word.  14  One could question whether the reversed sonority clusters in (9) are tautosyllabic, since they go against the SSP. The question of their syllabicity, however, falls outside of the scope of this dissertation; I will assume these sequences form complex onsets and, therefore, that the SSP plays only a restricted role in determining the phonotactics of Quiaviní Zapotec. The table below summarizes the possible consonant sequences in onset position in Quiaviní Zapotec. The only sequences not attested are stop + nasal and glide + glide. Table 5. Phonotactics of onset consonant clusters by sonority (mode of articulation) C1↓ C2→ Stop Fricative Nasal Liquid Glide Stop √ √ * √ √ Fricative √ √ √ √ √ Nasal √ √ √ √ √ Liquid √ √ √ √ √ Glide √ √ √ √ * These gaps seem to be systematic ones. The cluster stop + nasal is banned in many languages (e.g. English), and even more so is the glide + glide sequence (Greenberg, 1965, 1978). Consonant clusters in coda position are less common than in onset. In the native lexicon, practically the only native underlying sequence seems to be a consonant + glide,8 which surface as a complex segment in the form of stops with secondary articulation, either labialization (for dorsals, e.g. [kw, ɡ w, xw]) or palatalization (for coronals, e.g. [dj, ɲ, lj]). (10) Native words a. / bḛkw / → [ bḛʔkw ] ‘dog’ b. / bṳdj / → [ bu͡ṳðj̥ ] ‘chicken’  8  The native coda cluster /lˑ+d/ appears in the QZ dictionary, but all these cases seem to be phonetic  alternations derived from a simple fortis /lˑ/. E.g. rzàa'll, rzàa'lld ‘drops’, behll, behlld ‘fish’.  15  The detailed analysis of these segments is beyond the scope of this dissertation. Following Munro and Lopez (1999), I will assume they are separate segments underlyingly. Other sequences are found in loanwords. Apart from single coda consonants (examples above), loanwords present coda clusters of two and, rarely three consonants. (11) Loanwords a. / liebɾ / b. / alt / c. / mandaɾjenˑd / d. / njespɾ /  ‘book’ (< Sp. libro ) ‘tall’ (< Sp. alto) 'tangerine' (< Sp. mandarina) ‘loquat’  (< Sp. níspero )  With respect to the syllable nucleus, this constituent has to be occupied by a vowel; there are no syllabic consonants, although more research on the topic is necessary, especially with respect to consonant clusters that go against the Sonority Sequencing principle. Below, I present examples of all six Quiaviní Zapotec vowel qualities, / a e o i u ɨ /, in both monosyllabic and disyllabic words.  (12) Quiaviní Zapotec Vowels Monosyllabic words a. / ɡaz /  → [ ɡaːz ]  ‘seven’  b. / ɡe̤s /  → [ ɡe̤sː ]  ‘clay pot, earthenware pot’  c. / ʂop /  → [ ʂopː ]  ‘six’  d. / ɡiʒ /  → [ ɡiːʒ ]  ‘city person’  e. / ɾ-dṵb /  → [ ɾ-dṵːb ]  ‘sweeps’  f. / tsɨ̰ ~ tsɨ̰a /  → [ tsɨ̰ː ~ tsɨ̰ːa ]  ‘ten’  → [ sju.daː ]  ‘city’  Disyllabic words g. / sjuda /  h. / juhkwḛl / → [ juh.kwḛːl ]  ‘type of yellowish clay’  i. / teʔblo̤ /  → [ teʔ.blo̤ː ]  ‘flat’  j. / ɡiʒillj /  → [ ɡi.ʒill ]  ‘chair’  j  (< Sp. ciudad)  16  k. / candṵb /  → [ can.dṵːb ]  l. / baɡeiʒ ~ baɡɨiʒ /  → [ ba.ɡeiːʒ ~ ba.ɡɨiːʒ ]  ‘is sweeping’ ‘fly’  Quiaviní Zapotec vowels may be combined to form a number of diphthongs (Munro & Lopez, 1999, p. 3): / ai, au, ei, eu, ia, ie, iu, ua, ue, ɨi /, as well as other diphthongs that may appear in certain Spanish loanwords. Consider the following examples (both rising and falling diphthongs). Vowel duration on diphthongs is comparable to with monophthongs, and is addressed in Chapter 7. (13) Quiaviní Zapotec Diphthongs a. / ɾ-a̰ḭ / b. / kau / b. / ɡei̤ʒ /  → [ ɾa̰ˑḭ ] → [ kau ] → [ ɡèi̤ʒ ] → [ ɡḛˑṵ ]  ‘river’  d. / ɡjia̰ /  → [ ɡ ḭˑa̰ ]  ‘flower’  c. / ɡḛṵ /  j  ‘gets cooked’ ‘Claudia’ ‘townʼ  e. / njienj /  → [ ɲieˑɲ ]  ‘is audible’  f. / bien /  → [ biˑen ]  ‘wine’  g. / bjiṳ /  → [ bjiṳ ]  ‘ground up’  (< Sp. vino)  h. / banɡual / → [ ban.ɡuˑal ]  ‘elder’  i. / ɾ-dṵa̰ʒ / j. / luas / k. / rued /  ‘finishes’ ‘light’ ‘wheel’  (< Sp. luz) (< Sp. rueda)  ‘new’  (< Sp. nuevo)  → [ ɾdṵˑa̰ʒ ] → [ luas ] → [ rueˑd ]  l. / n-kwɨibj / → [ nkwɨibj ]  1.2.5 Morphosyntax The goal of this section is to provide an overview of the basic morphosyntactic properties of Quiaviní Zapotec. Many of these properties will be considered when presenting the data in the following chapters, particularly in the metrical structure chapter, where morphologically complex words are analyzed.  17  This section largely draws from the Quiaviní Zapotec dictionary (Munro & Lopez, 1999), as well as from Lee (2006); the orthography used in this section is from Munro and Lopez (1999). The basic word order in Quiaviní Zapotec is VSO. This can seen in the following examples: (14)  R-gwèe' Chie'cw Dìi'zhsah HAB-speak Chico Zapotec ‘Chico speaks Zapotec’  (15)  B-guhty bèe'll bzihihny PERF-kill snake mouse ‘The snake killed the mouse’  Quiaviní Zapotec also allows SVO and OVS word orders when the fronted argument is interpreted with contrastive focus: (16)  Bèe'll b-guhty bzihihny snake PERF-kill mouse ‘The snake killed the mouse’ / ‘The mouse killed the snake’  (17)  Bzihihny b-guhty Bèe'll mouse PERF-kill snake ‘The mouse killed the snake’ / ‘The snake killed the mouse’ As seen in the previous examples, Quiaviní Zapotec lacks overt case marking.  When arguments are fronted, the thematic role of arguments is ambiguous. Embedded clauses generally appear without complementizers or other markers of subordination and their word order is identical to that of matrix clauses. As Lee (2006, p. 7) points out, Quiaviní Zapotec “shows the canonical features of most VSO languages: it has prepositions rather than postpositions, adjectives generally follow nouns, relative clauses are head-initial, and possessive constructions are possessor final”. Quiaviní Zapotec uses body part words as prepositions (grammaticalized nouns), for instance, lohoh ‘face’ is used as a preposition meaning ‘at’ or ‘on’; laa’iny ‘stomach’ means ‘inside’ in its prepositional use; and dehts ‘back’ conveys the meaning of ‘back  18  side, behind’. The grammatical analysis of these types of words is provided in detail in Lillehaugen (2003, 2005). (18) loh yu'uh face/PREP house  ‘in front of the house’  (19) dehts yu'uh back/PREP house  ‘in the back of (behind) the house’  (20) laa’iny yu'uh ‘inside the house’ stomach/PREP house Nouns without determiners or quantifiers can be interpreted as either definite or indefinite entities, and either singular or plural. The use of the plural marker ra is optional. Possessive constructions are possessor-final. The possessed nominal is preceded by the possessive marker / ʂ-, ʃ- / (x:-, x-) and followed by the possessor (Lee, 2006, p. 9): (21)  x:-ca’rr Wsee POSS-car Joseph  ‘Joe’s car’  An alternate possessive construction, which apparently does not differ in usage or meaning from the one shown above, is formed with the possessed nominal x:tèe’ (or x:tèe’n): (22)  x:-me’s-a’ POSS-teacher-1S  ‘My teacher’  (23)  me’s x:-tee’n-a’ teacher POSS-1S  ‘My teacher’  Verbal morphology Quiaviní Zapotec verbs can take complex forms. Besides carrying standard inflectional features (tense and agreement), they may also carry additional morphological material encoding direction, causation, manner, and modality, among other things.  19  Following a long-standing tradition in Zapotec linguistics, Quiaviní Zapotec aspectual and mood prefixes are classified under the broad category of aspectual prefixes (Munro 2006). Further, Lee (2006) argues that these markers can also express tense covertly. Table 6 illustrates the seven inflectional prefixes of Quiaviní Zapotec, along with the verbal paradigm of rtàa’az 'beats'. Table 6. Quiaviní Zapotec aspectual prefixes (adapted from Lee, 2006, p. 11) Terminology from Munro and Lopez (1999) Aspect Habitual Progressive Perfective  Prefix rca-  -tàa’z ( beat) rtàa’aza’ catàa’aza’  Translation ‘I beat (regularly)’ ‘I am/was beating’  btàa’aza’  ‘I beat’  — ytàa’aza’  ‘I will beat’  Definite  b-, w-, gu-, mn-, ∅y-, chi-, g-, ls-, z-  stàa’aza’  ‘I will surely beat’  Subjunctive  n-, ny-  ntàa’aza’  ‘I was going to beat’  Neutral9 Mood Irrealis  Quiaviní Zapotec verbs obligatorily appear with aspect markers, but no more than one is permitted (no stacking). Furthermore, there are neither bare nor infinitive forms. Adapted from Lee (2006: 27), (24) schematizes the internal structure of Quiaviní Zapotec verbal morphology. (24)  Quiaviní Zapotec verbal morphology (based on Lee 2006: 27) ASP (DIR/CAUS) ROOT (APPL/INT)(ADV)(SUBJECT CLITIC)(OBJECT CLITIC)  Verbal morphology is illustrated with the verb rda'uh ‘eats’ in perfective form in the following examples. Samples are presented on the left-hand side, whereas morphemeclass labels appear on the right-hand side.  9  The neutral prefix appears on a small number of mostly stative or locational verbs. It also has been analyzed as an affix used to derive adjectives (R. Rojas & T. Smith-Stark, personal communication, April 2008).  20  (25)  B-da’uh PERF-eat ‘I ate’  nàa’ 1s  ASP-ROOT  (26)  B-t-a’uh nàa’ PERF-DIR-eat 1s ‘I went to eat / I went and ate’  ASP -DIR-ROOT  (27)  B-z-a’uh nàa’ PERF-CAUS-eat 1s ‘I made Mike eat’  ASP -CAUS-ROOT  (28)  B-da’uh=a’ PERF-eat=1s ‘I ate’  Gye’eihlly Mike  ASP -ROOT=SUBJECT CLITIC  Pronouns and Pronominal clitics Quiaviní Zapotec has no subject agreement morphology; pronominal subjects appear as clitics that follow the verb stem. Table 7. Quiaviní Zapotec pronouns and clitics (adapted from Lee 2006; and Munro & Lopez, 1999) 1s 2s informal 2s formal 3s proximate 3s distal 3s formal 3s animal 1p 2p informal 2p formal 3p proximate 3p distal 3p formal 3p animal  Pronoun nàa’ liu’ làa’yuu la’anng la’ai làa’b làa’mm dannoohnn làa’d làa’yuad làa’rëng làa’rih làa’rëb làa’rëmm  Clitic -a’ -u’ -yuu’ -ëng -ih -ëb -ëmm -ënn -ad -yùad -rëng -rih -rëb -rëmm  Gloss ‘I’ ‘you (informal)’ ‘you (formal)’ ‘he/she/it (nearby)’ ‘he/she (out of sight)’ ‘he/she (formal)’ ‘he/she/it(animal/child)’ ‘we’ ‘you (plural, informal)’ ‘you (plural, formal)’ ‘they (nearby)’ ‘they (out of sight)’ ‘they (formal)’ ‘they (animals/children)’  Quiaviní Zapotec pronouns and clitics are semantically rich. Munro and Lopez (1999) observe four distinct levels of reference to living beings in third-person pronouns,  21  depending on age and social status. Two other forms of reference, the distal and proximate, are determined by the proximity of the referent to the speaker (See also Munro, 2001). Suffixes Quiaviní Zapotec makes use of different types of suffixes, including adverbial suffixes (Munro, 2006) and the diminutive suffix. The latter is one of the major types of derived nominal forms and its frequency is high. (See Munro et al. (2008) for details on the diminutive suffix analysis, and variation.) (29)  a. bra'au-e’eh lizard-DIM  ‘little lizard’  b. zhyàa'p-e’eh girl-DIM  ‘little girl’  In conclusion, this chapter has provided an overview of the phonological and morphosyntactic characteristics of Quiaviní Zapotec. This serves as a background for the rest of the dissertation, where the metrical structure, tone and phonation types of this language are analyzed in detail.  22  Chapter 2:  Vowel length and the fortis/lenis distinction in Quiaviní Zapotec  2.1  Introduction Vowel-length in Zapotec languages has been a matter of some contention in terms  of whether it is lexically specified, prosodically, or segmentally determined. In Quiaviní Zapotec, this issue closely interacts with the fortis/lenis distinction, which is pervasive in the consonantal system of this language (Munro & Lopez, 1999). This chapter demonstrates that vowel length in Quiaviní Zapotec is dependent on the type of syllable and on the type of coda consonant: stressed vowels are short before fortis consonants (both obstruents and sonorants), and long before lenis consonants or in open syllables. As such, the categorization of this vowel length pattern relies on the fortis/lenis contrast, which in turn involves a complex set of phonetic properties of unclear phonological status.  23  Phonological research from a variety of language stocks (Indo-European, Austronesian, Niger-Congo, Afro-Asiatic, Otomanguean, Mixe-Zoque, Athabaskan, Pama-Nyungan, North Caucasian (see Kohler, 1984; and DiCanio, 2008, for an overview)) describes consonants with a fortis/lenis contrast. These terms are generally considered to capture a contrast in articulatory strength, where articulations are produced with greater versus lesser muscular or pulmonic force. Despite similar phonetic correlates in languages with a fortis/lenis distinction (such as length, voicing, intensity (Jaeger, 1983; Avelino, 2001; DiCanio, 2008, among others)), the precise manifestation of the fortis/lenis distinction seems to be language-specific, with no universal phonetic property. Such variation in phonetic correlates can be found in a range of phonological phenomena and features. Among others, the tense/lax categories in vowels are vague in terms of their specific phonetic correlates (Jakobson, Fant, & Halle, 1951, p. 38, see definition below), but their phonological status is crucial in several languages (e.g. for English high vowels). Stress, like other prosodic phenomena, is another relevant example for this discussion (see Kenstowicz, 1994; Hayes, 1995). Acoustic correlates of stress include pitch, duration and intensity, among others, but none of these can be unambiguously and universally associated with prominent syllables. In other words, what distinguishes one category (e.g. fortis, tense, stressed, etc.) in a particular language may differ from what distinguishes it in another language. What unifies all these phenomena is that they can be encoded by a composite of properties, including both language-specific phonetic and phonological characteristics of the attested distinctions. Based on this, it is possible to postulate categories that most accurately correspond to those sets of properties. Mutatis mutandis, this is the proposal of the emergent feature approach, as represented by Mielke (2008 [2004]; see also Pulleyblank, 2006). With this background, the goal of this chapter is to explain vowel length in Quiaviní Zapotec and establish the characteristics of the fortis/lenis distinction in this language, foundational issues for the prosodic analyses presented in subsequent chapters. The hypothesis is that several gradient properties interact to create the fortis/lenis contrast. These properties include voicing, degree of constriction, sonority, and duration, as well as phonological distribution, markedness, and prosodic prominence. As shown  24  below, some of these characteristics crosscut both obstruent and sonorant categories. Under an emergent feature approach (Mielke, 2008 [2004]), this composite of properties can be encoded with the feature [+/-fortis] (Kohler, 1984; Pulleyblank, 2006). This chapter is organized as follows: §2.2 describes vowel length in Quiaviní Zapotec in light of the orthography of Munro and Lopez (1999). Having established this distribution, §2.3 presents a detailed description of the full range of realizations of fortis and lenis consonants in Quiaviní Zapotec, determining the distinctive characteristics involved in their contrast. Based on the described sound patterns, §2.4 validates the use of a feature [+/-fortis] within Quiaviní Zapotec grammar.  2.2  Quiaviní Zapotec vowel length  This section shows that the type of coda consonant determines vowel length in this language. In doing so, I analyze the vowel patterns a’ (single checked vowel /Vʔ/)10 and aa (long vowel /VV/) in the orthography of Munro and Lopez (1999) as underlying short modal vowels. As reflected in the Quiaviní Zapotec dictionary (Munro & Lopez, 1999), a salient feature in this language is vowel length, for which the following distribution can be drawn: In prominent syllables11, checked vowels (a’) appear before fortis consonants, whereas long vowels (aa) are followed by lenis consonants or occur in open syllables. (1) Short vowels (a’) before fortis coda consonants a. yuhdye'p b. Mihste'c c. a’s d. yze'nny e. rcah gye'rr  ‘uncultivated land’ ‘Mixtec’ ‘hi’ ‘will arrive’ ‘gets branded’  10  Other vowel patterns from Munro and Lopez (1999) containing checked vowels are analyzed in subsequent chapters, especially in Chapter 6. 11 In this study, I will use interchangeably the terms stressed or prominent syllable, to refer to the most salient syllable in a word based on the prosodic properties assumed by metrical theory (to be presented in the next chapter).  25  (2) Long vowels (aa) before lenis coda consonants a. teeby b. rrueeg c. wyaazh d. x:eeny e. ma'anyseer  ‘only, alone’ ‘basil’ ‘rented’ ‘stupid’ ‘bee’  (3) Long vowels (aa) in open syllables a. bdaa b. maa c. ndii d. canoo e. zuu  ‘shadow’ ‘girlie, little girl’ ‘right’ ‘than’ ‘is standing’  The Munro and Lopez (1999) orthography is extremely consistent with this pattern, particularly with obstruents. In terms of loanwords (analyzed in more detail in Chapter 3), the pattern above is also very clear. (4) Short vowels (a’) before fortis coda consonants 12 a. la’t b. Be’t c. Lu’c d. naba'j e. cla’s f. Ba'll  ‘can tin’ ‘Alberto’ ‘Lucas’ ‘razor’ ‘class’ ‘Valeriano’  (< Sp. lata) (< Sp. Beto < Alberto) (< Sp. Lucas) (< Sp. navaja) (< Sp. clase) (< Sp. Vale < Valeriano)  (5) Long vowels (aa) before lenis coda consonants a. laad b. Beed c. juug d. nabaazh e. laaz f. baal  ‘side’ ‘Pedro’ ‘juice’ ‘pocket knife’ ‘twine’ ‘bullet’  (< Sp. lado) (< Sp. Pedro) (< Sp. jugo) (< Sp. navaja) 13 (< Sp. laso) (< Sp. bala)  12  Short vowels (a’) also appear before coda consonant clusters in Spanish borrowings (there are no native complex codas). This issue is analyzed in the next chapter within the loanword phonology section (§3.5). 13 The "lexical split" between (4d) and (5d) is particularly illustrative of Quiaviní Zapotec vowel length. Even if it is not completely predictable whether the final consonant gets borrowed as fortis or lenis, the length of the vowel follows automatically from that "choice" (short vowel before fortis and long vowel before lenis).  26  (6) Long vowels (aa) in open syllables a. Lia Daa b. Nabidaa c. Wsee d. tee e. rreloo f. Chuu  ‘Soledad’ ‘Christmas’ ‘José’ ‘tea’ ‘watch’ ‘Chuy, Jesus’  (< Sp. Soledad) (< Sp. Navidad) (< Sp. José) (< Sp. té) (< Sp. reloj) (< Sp. Chuy)  Despite some exceptions to this distribution,14 this pattern clearly resembles what has been reported for several Zapotec languages. The first work reporting vowel length in Zapotec languages is an unpublished manuscript by Swadesh quoted in Pike (1948, p. 167). Swadesh describes vowel length in several variants of Zapotec as being non-phonemic, but predictable from consonant environment. “La vocal tiende a ser corta ante el saltillo [ʔ] y ante las consonantes fuertes […], mientras que ante las demás consonantes y, en menor grado, al final de las palabras generalmente es larga.” 15 The pattern of having short vowels before fortis consonants and long vowels before lenis consonants has been described for a number of Zapotec languages, including Cajonos Zapotec (Nellis & Barbara E. Hollenbach, 1980), Chichicapan Zapotec (SmithStark, 2003), Yalálag Zapotec (Avelino, 2004), San Francisco Ozolotepec Zapotec  14  Within obstruents this pattern is found for the majority of entries in the dictionary. Some potential exceptions are bax:aa’t ‘toad’, zh:aa’cw ‘cockroach’, see’st ‘sixth’, mbii’sy ‘stingy’, among others. I compared the vowel duration of 20 of these potential exceptions (recorded by a Quiaviní Zapotec speaker), with the duration of 20 items with short (checked) vowels. Results show the similarity of these items: vowels with the pattern aa’ average a duration of 82 ms versus 81 ms for a’ items (the difference was not significant). As we will see in the next chapter, this duration corresponds to that of short vowels. Accordingly, it seems appropriate to reanalyze these aa’ vowel patterns as short vowels. Among sonorants, differences were also non-significant for nasals when comparing apparent exceptions like Juu’nny ‘June’ versus items with the vowel pattern a’. For liquids, apparent exceptions include long vowels followed by fortis liquids, such as bchiilly ‘knife’ or ganiilly ‘ring’. The results for these words were in the opposite direction, as the vowels were in fact long (~150 ms), but the coda consonants were too short to be considered fortis (below 100 ms). The fortis/lenis distinction among sonorants is challenging because the difference relies only on duration. This asymmetry has also been found with other vowel patterns, including non-modal phonation as in rguììi’lly ‘waters’, which seems to have a long vowel followed by a lenis sonorant (see fortis/lenis duration differences in coda in the acoustic comparison for the creaky vowels section of Chapter 6). 15 “Vowels tend to be short before glottal stop and fortis consonants, whereas before the rest of the consonants and, to a lesser degree, in utterance final position they are generally long.” [Translation mine]  27  (Leander 2008), Quioquitani Zapotec (Ward, Sánchez, & Marlett, 2008), and San Pablo Güilá Zapotec (Arellanes, 2009), among others. In what follows, I would like to draw the reader’s attention to some of these analyses, where a clear relationship between prosody and the vowel and consonant duration has been reported. For Cajonos Zapotec, Nellis & Hollenbach (1980, p. 93) state the following: “All fortis consonants are lengthened following a vowel with primary stress, whereas stressed vowels are themselves lengthened preceding a lenis consonant. This lengthening serves to maintain a fairly constant length for stressed syllables […] and provides an additional distinction between the two series [fortis versus lenis].” Smith-Stark (2003, p. 124) describes a similar relation between vowel and consonant length in Chichicapan Zapotec: “Las raíces simples de dos sílabas varían en la duración de la vocal tónica [referring to the first syllable]. Si la consonante intermedia es débil […], la vocal tónica se alarga; si es fuerte […], la vocal tónica es breve y la consonante intermedia se alarga”.16 In sum, Smith-Stark reports that a stressed vowel followed by a lenis consonant is lengthened, whereas if followed by a fortis consonant the vowel is shortened and the fortis consonant is lengthened. For Quiaviní Zapotec, Munro and Lopez (1999, p. 2) state: “Some phonological rules refer to the classes of fortis and lenis consonants. The main such rule lengthens otherwise identical vowels or vowel sequences before lenis (but not fortis) consonants, which is generally comparable to the behavior of fortis versus le  nis  consonants  in  other Zapotec languages, as described, for example, by Nellis and Hollenbach (1980) and Jones and Knudsen (1977)”. Despite noting the predictability of vowel duration differences, Munro and Lopez (1999) nonetheless encode these vowel-length differences in the orthography. Phonologically, however, since this length is predictable (i.e. not contrastive), it is possible to analyze the vowel pattern aa (long surface vowel) as an  16  “Simple disyllabic roots vary with respect to the duration of the stressed vowel [in the first syllable]. If the intermediate consonant is lenis […], the stressed vowel is lengthened; if it is fortis […], the stressed vowel is shortened and the intermediate consonant is lengthened.” [Translation mine]  28  underlying short modal vowel /a/, which in prominent positions lengthens in open syllables and before lenis coda consonants: /a/ → aː / _ (Clenis)]σ .17 Related to this short/long vowel distribution, I argue that checked vowels in content words should be reanalyzed as short modal vowels; specifically I refer to vowels followed by fortis (voiceless) stops. Let us examine this issue in more detail by reviewing some examples to investigate the voice quality of this vowel pattern.  0.798  0  -0.9314  0  0.7253  5000  0  Time (s)  0  0.7253 Time (s)  [  ʧ  a  ʔ  t  h  ]  Figure 1. Waveform and spectrogram of cha’t [ ʧaʔth ] ‘kiss’ by male speaker TiuL18 (sound file from Munro et al., 2008).  17  Another conceivable analysis would be that long vowels shorten before fortis consonants. This is rejected on the basis that vowel length is not lexically contrastive (as noted by Munro & Lopez, 1999, p. 2 and confirmed in my fieldwork research). Moreover, phonetic long vowels only appear in prominent syllables. 18 Throughout the dissertation, I will use the title Tiu —a respectful title used before a man's name— and the first letter of my consultant’s name to refer to male speakers; likewise, I will refer to female speakers with the first letter my consultant’s name, preceded by the title Lia — the title used before a woman's given name.  29  0.6928  0  -0.8986  0  0.8341  5000  0  Time (s)  0  0.8341 Time (s)  h [ l a t ] Figure 2. Waveform and spectrogram of la’t [ lath ] ‘tin can’ by female speaker LiaT.  The presence of creakiness at the end of the vowel (see §4.1.1 and §6.5) and the abrupt cessation of vocal fold vibration may signal the presence of a glottal stop (the second characteristic is clearly found in Figure 1). However, these phonetic characteristics are not consistent in all the sequences of “checked” vowel plus oral stop that I have analyzed. Figure 2 illustrates a case with no glottal stop. Both the waveform and the spectrogram show that the vibration of the vocal folds does not cease immediately at the end of the vowel. In fact, it seems that the voicing bar and formant structure (echo) continue throughout the closure and (less noticeably) at the release of the stop. This vibration clearly indicates the absence of the glottal stop (which implies a complete cessation of the vocal fold vibration). I have not found a phonetic or phonological factor to determine the presence or absence of the glottal stop; it simply seems to be variable. It is possible that the type of speech (careful/emphatic versus colloquial), or extra-linguistic factors (such as gender and age) play a role with respect to the presence or absence of the glottal stop. In addition, fortis consonants are considerably longer in coda position compared to syllable-initially (see phonetic experiment in Chapter 3); thus, this “unusual” fortis  30  stop length (with closure sometimes lasting more than 200ms) may be perceived as a glottal stop — whether it is articulated with the oral stop or not. (It is worth mentioning that fortis stops are long in coda regardless of the voice quality of the vowel.) The presence of an allophonic glottal stop in the context of vowel + oral stop has also been found in other languages and dialects. Notably, this phenomenon is well documented for the British English "Received Pronunciation" (Christophersen, 1952; Roach, 2004), where coda voiceless plosives /p/, /t/, /tʃ/, and /k/ may be preceded by a glottal stop. This phenomenon is also well known for Japanese geminate voiceless stops (Sawashima & Miyazaki, 1973) and coda voiceless stops in some Chinese languages (Haudricourt, 1954; Hombert, Ohala, & Ewan, 1979). The phenomenon is called preglottalization or glottal reinforcement, and, as in Quiaviní Zapotec, it takes place with consonants in coda position and to a certain extent it is variable. As its name indicates, the glottal stop reinforces the oral closure, ensuring the stoppage of airflow during the closure. I turn now to alleged cases of checked vowels followed by fortis (voiceless) fricatives. Figure 3 shows the spectrogram of naba’j ‘razor’, whereas Figure 4 provides the spectrogram of a’s ‘hello’. 0.6255  0  -0.6862  0  7000  0  0.872 Time (s)  0  0.872 Time (s)  [ n a β a x ] Figure 3. Waveform and spectrogram of naba’j [ naˈβax ] ‘razor’ by male speaker TiuR. 31  0.5789  0  -0.7502 1.1·104  0  0  0.7569 Time (s)  0  0.7569 Time (s)  [  a  s  ]  Figure 4. Waveform and spectrogram of a’s [ as ] ‘hi’ by male speaker TiuL (sound file from Munro et al., 2008). All the sequences I have analyzed of putative checked vowels plus fricatives show no glottalization. As the waveforms and the spectrograms in Figures 3 and 4 show, there is neither a glottal stop nor any indication of creaky voice, but rather a smooth transition from the vowel into the fricative. Other properties such as intensity, periodicity, and pitch are stable and similar to those of (prototypical) modal vowels. Based on the above findings, I propose to reanalyze short checked vowels in prominent syllables as short modal vowels (/a/ instead of /aʔ/). In Munro and Lopez’s (1999) analysis, single modal vowels were not included as part of the vowel inventory within prominent syllables. In summary, Table 8 includes the vowel patterns a’ (single checked vowel) and aa (long vowel) from the Munro and Lopez (1999) orthography, the corresponding phonemic transcription and tone, as well as the present reanalysis, where I reexamine these vowel patterns as phonemic short modal vowels.  32  Table 8. Vowel patterns a’ and aa reanalyzed as phonemic short modal vowels Munro and Lopez (1999)  Reanalysis  Orthography phonemic19 Tone phonemic a’ / aʔ / ˥ /a/ aa / aa / ˥ /a/  Surface realization [a] before fortis C [aː] before lenis C  Table 9 presents examples of the vowel patterns under consideration. On the left side of the table, I include Munro and Lopez’s (1999) orthography, a phonemic transcription and the gloss. The reanalysis, on the right, shows the proposed phonemic transcription and its phonetic realization. Table 9. Examples of vowel patterns a’ and aa with reanalysis Munro and Lopez (1999)  Reanalysis  (7)  orthography tyo’p  phonemic / tjoʔp /  Gloss ‘two’  phonemic / tjop / ˥  Surface realization [ tjoph ] ~ [ tjoʔph ]  (8)  cha’t  / ʧaʔt /  ‘kiss’  / ʧat / ˥  [ ʧath ] ~ [ ʧaʔth ]  (9)  naba’j  / nabaʔx /  ‘razor’  / nabax / ˥  (10) a’s  /as/  ‘hi’  / as / ˥  [ as ]  (11) teeby  / tebj /  ‘alone’  / tebj / ˥  [ teːβj ]  ʔ  (12) laad  / laad /  ‘side’  (13) laaz  / laaz /  ‘twine’ / laz / ˥ ‘bullet’ / bal / ˥  (14) baal  / baal /  / lad / ˥  [ naβax ]  [ laːd ] [ laːz ] [ baːl ]  As a final note, in contrast to single checked vowels in content words, an exception to this reanalysis is clitics. First and second person singular clitics are described in the dictionary (Munro & Lopez, 1999) as containing checked vowels / aʔ /, and I agree with this specification on the basis of my own research.  19  Munro and Lopez (1999) do not provide a phonemic transcription of the entries in the dictionary; the orthography, however, is phonologically goal-oriented, thus, the phonemic transcription presented here is my interpretation of their orthography.  33  Table 10. 1s and 2s Quiaviní Zapotec clitics. Munro and Lopez (1999) orthography  Reanalysis  phonemic  Gloss  phonemic  Surface realization  (15) =a’  / aʔ /  1s  2s (informal) / uʔ / ˥  [ aʔ ] ~ [ aa̰ʔ ] ~ [ aʔḁ ]  (16) =u’  / uʔ /  / aʔ / ˥  (17) =yuu’  / juuʔ /  2s (formal)  / juʔ / ˥  [juʔ ] ~ [juṵʔ ] ~ [juʔu̥ ]  [ uʔ ] ~ [ uṵʔ ] ~ [ uʔu̥ ]  As with previous tables, Table 10 shows orthography, phonemic transcription and gloss for clitics according to Munro and Lopez (1999), followed by my reanalysis, along with the phonetic transcription. In these clitics, the glottal stop may be fully realized, it may be short, or it may consist of a period of creakiness. As an illustration, consider the following example. r-càa’z=a’ / ɾka̰zaʔ / HAB-wants-1s  (18)  ‘I want…’  0.4785  0  -0.5202  0  0.8545  5000  0  Time (s)  0  0.8545 Time (s)  [  ɾ  kh  a̰  z a a̰  ʔ  ]  Figure 5. Waveform and spectrogram of r-càa’z=a’ [ ɾkha̰ːzaa̰ʔ ] ‘I want…’ by male speaker TiuT. In the spectrogram above, the last vowel, the clitic / =aʔ /, starts with a very short period of modal phonation, followed by creaky voice (two or three pulses); after that, we observe a glottal stop. As mentioned before, short creakiness and the abrupt cessation of  34  vocal fold vibration indicate the presence of a glottal stop; both are present in this sample. I return to the analysis of clitics in Chapter 6, §6.5. To conclude, this section established that prominent vowels are short before fortis consonants (both obstruents and sonorants), and long in open syllable and before lenis consonants. Although the data presented here corresponds to modal voice, the prediction is that this pattern also applies for non-modal phonation (see Chapters 6 & 7). Finally, vowel length is one of several components that contribute to the fortis/lenis contrast, to which I now turn.  2.3  Fortis and lenis consonants in Quiaviní Zapotec  Fortis and lenis are controversial terms. Linguists disagree about both their definition and their validity. The terms are used to characterize a basic phonological contrast in consonant systems, which cannot be explained in terms of a simple voicing distinction. The basic claim is that one member of a contrasting pair of phonemes is produced with greater “force of articulation” than the other (Jakobson et al., 1951; Malécot, 1966; Fischer-Jørgensen, 1968; Catford, 1977, pp. 199-208; Jakobson & Waugh, 1979, pp. 135-9).20 However, “force of articulation” refers to different phonetic aspects, hence, there is no consensus on a phonological feature that refers to a specific phonetic characteristic. Most descriptions of systems with a fortis/lenis distinction have focused on obstruents, including the following characteristics for each class: Table 11. Fortis/lenis characteristics (adapted from Jaeger, 1983) 21 Fortis long (contextually) voiceless high intensity noise closure (stops)  Lenis short fluctuate in voicing (e.g. [ b, b̥, p ]) lower intensity noise stop closure varies with low-amplitude frication (“stops” only)  20  A wide range of phonetic phenomena have been included in this “force” including: pulmonic, articulatory, timing and glottal factors (see Jaeger, 1983, p. 178). 21 Based on Yaté Zapotec and Jawoñ (an Australian language).  35  Jaeger (1983, p. 184) states that “the prototypical fortis obstruent is long and voiceless, with no variation in closure type, and higher amplitude noise. The prototypical lenis consonant is short, usually voiced but often voiceless, has much variation in closure type, and lower amplitude noise.” Furthermore, she mentions that the terms fortis and lenis may be considered phonological categories, which are associated with a set of phonetic cues. In what follows, this consideration is evaluated in light of the Quiaviní Zapotec data. Munro and Lopez (1999) propose the fortis/lenis distinction as the most comprehensive and persistent contrast for consonants in this language. They maintain that “the distinctive characteristic of fortis obstruents is articulatory tension; that of fortis sonorants is increased duration” (Munro & Lopez, 1999, p. 2). This four-way contrast is illustrated in Table 12. Table 12. The four-way contrast in Quiaviní Zapotec (fortis/lenis-obstruent/sonorant) Fortis Lenis Obstruents √ √ Sonorants √ √ The fortis/lenis contrast mostly occurs in pairs, as illustrated in Table 13. Table 13. Quiaviní Zapotec fortis/lenis consonant pairs stops affricates fricative nasals liquids fortis p t k ts ʧ s ʃ ʂ f x mˑ nˑ ŋˑ lˑ r lenis b d ɡ zʒʐ m n ŋ l ɾ The fricative phonemes / f, x / appear only in Spanish borrowings, and along with affricates they pattern with fortis consonants (see properties below) and they do not have lenis counterparts. Within liquids, we could arguably analyze the trill and tap phonemes / r, ɾ / as a fortis/lenis pair, although apart from loanwords [r] only appears as a result of morpheme concatenation / ɾ-ɾ / → [r].  36  The fortis/lenis contrast is well attested in Quiaviní Zapotec, motivated by numerous minimal pairs in the Munro & Lopez (1999) dictionary. In addition, there are two morphosyntactic cases of fortition in Quiaviní Zapotec that illustrate the fortis/lenis distinction: the possessive and the causative (Munro & Lopez, 1999, p. 2). Possessive constructions are possessor-final. The possessed nominal is preceded by the possessive marker ʂ- /ʃ- and followed by the possessor (Lee, 2006, p. 9): (19) ʃ-kaɾ ɡje̤lˑj POSS-car Mike  ‘Mike’s car’  (20) ʃ-tiu-aʔ POSS-uncle-1s  ‘my uncle’  When the possessed noun underlyingly begins with a lenis consonant, the initial consonant of the noun surfaces as its fortis counterpart, showing fortition. (21) / ʃ-dad-aʔ / → [ ʃtaːdaʔ ] POSS-father-1s  ‘my father’  With respect to the morphological causative, a subset of verbs shows fortition of root-initial lenis consonants (Lee, 2006, p. 24), as illustrated below. (22) ɾ-ɡaʔ ‘gets caught’ HAB-gets caught (23) ɾ-kaʔ ‘takes, gets’ HAB-take In what follows, I present in more detail the contextual realization of Quiaviní Zapotec consonants, in order to shed light on their phonological properties (detailed descriptions of the fortis/lenis contrast in Zapotec languages include, among others, Nellis & Hollenbach, 1980; Avelino, 2001; Antonio Ramos, 2007; and Arellanes, 2009, from which I adopt the format to present the Quiaviní Zapotec data).  37  Fortis stops, / p t k /, “are voiceless and often aspirated” (Munro & Lopez, 1999, p. 3), particularly word-finally. In addition, these consonants are never weakened to fricatives, and they are long in coda position. This is illustrated with the following examples word-initially, in intervocalic position and word-finally. Fortis stops: / p t k / (24) Word-initially ( #_ ): voiceless stops a. / pes /  ˥  → [ pésː ]  ‘peso’  b. / tiu /  Ë  → [ tìú ]  ‘Mr. / uncle’  → [ kû͡ṳɸ ]22  ‘tejate (traditional beverage)’  c. / kṳb / Ü  (25) Intervocalic ( V_V ): voiceless stops a. / ʃ-tjop=uʔ / ˥ ˥ → [ʃtjó.púʔ ]  ‘your two’  b. / ʃ-tʃat=uʔ / ˥ ˥ → [ ʃtʃá.túʔ ]  ‘your kiss’  c. / ʃ-luk=u / Ë ˥ → [ ʃlˑǔ.kúʔ ] ʔ  ‘your Lucas’  (26) Word-finally ( _# ): long voiceless stops a. / tjop / ˥  → [ tjópːʰ ]  ‘two’  b. / tʃat / ˥  → [ tʃátːʰ ]  ‘kiss’  c. / luk /  Ë  → [ lǔkːʰ ]  ‘Lucas’  Lenis stops, / b d ɡ /, “range in most positions from voiced stops to very lenited voiced fricatives” (Munro & Lopez, 1999, p. 2). More specifically, and according to my data, lenis stops tend to be fricated and voiced intervocalically, and fricated and devoiced word-finally. Word-initially, they are in free variation, fluctuating in both voicing ([voice]) and in closure width ([continuant]).  22  The transcription of non-modal vowels in some of these examples implies a surface sequence of modal plus non-modal phonation. These realizations are explained in detail in Chapter 6.  38  Lenis stops: / b d ɡ / (27) Word-initially ( #_ ): voiced stops or voiced fricatives 23 a. / ba /  → [ bǎː ~ βǎ ]  Ë  b. / danj / ˩ c. / ɡe̤t /  ‘already’  → [ dàːɲ ~ ðàːjɲ ] ‘mountain’  ˩  → [ ɡè̤tː ~ ɣè̤tː ]  ‘tortilla’  (28) Intervocalic ( V_V ): voiced fricatives a. / ʃ-dub=uʔ / ˩ ˥ → [ ʃtùː.βúʔ ]  ‘your maguey’  b. / ʃ-ɡi̤a̤d=u / ˩ ˥ → [ ʃkì̤à̤ː.ðúʔ ]  ‘your century plant’  ʔ  c. / ʃ-neɡ=u / ˥ ˥ → [ ʃnˑéː.ɣúʔ ] ʔ  ‘your fanega (large sack)’  (29) Word-finally ( _# ): voiceless fricatives (most of the time) 24 a. / dub / ˩  → [ dùːɸ ]  ‘agave, maguey’  c. / xuɡ / ˥  → [ xúːx ]  ‘juice’  b. / ɡi̤a̤d / ˩  → [ ɡì̤à̤θ ]  ‘century plant’  Clearly, in terms of voice and manner of articulation, fortis stops (specified as [-voice] and [-continuant]) are stable regardless of the context, whereas lenis stops (presumably specified as [+voice] and [-continuant]) vary according to the phonological context. Since lenis stops are the most variable of all lenis consonants, I list their allophones in (30). (30) Lenis stop allophones Phonemes Allophones a. / b / → [ b, b̥, β, ɸ] b. / d / c. / ɡ /  23  → [ d, d̥, ð, θ ] → [ ɡ, ɡ̊, ɣ, x ]  Word-initially and intervocalically, lenis stops may also surface as devoiced segments [b̥ d̥ ɡ̊], but these  examples show the most common realizations. 24  Occasionally, word-final lenis “stops” appear as stops.  39  Note that these fricative allophones of lenis stops cannot neutralize with other lenis fricatives in the language (because there are no / β ɸ ð θ ɣ / phonemes), except for / x /; nonetheless, the phoneme / x / is relatively restricted as it only occurs in Spanish loanwords. Furthermore, /ɡ/ is rarely devoiced in onsets (the onset alternation is mainly between [ɡ] and [ɣ]), whereas in coda position the fricative phoneme /x/, being fortis, is always longer than the [x] allophone of /ɡ/. The affricates / ts / and / tʃ / are parallel with fortis obstruents, as they show the same invariant contextual characteristics in terms of voicing (always voiceless) and manner of articulation. They are also long in coda position. In accordance with LaCharité (1995), Clements (1999), among others, affricates can be grouped with stops as [continuant] segments (strident stops). As described by Munro and Lopez (1999), there are no lenis affricates. Fortis fricatives, / s ʃ ʂ /, are always voiceless ([-voice]) and long in coda position; whereas lenis fricatives, / z ʒ ʐ /, “are devoiced in final position” (Munro & Lopez, 1999, p. 2). However, fortis and lenis fricatives do not neutralize in coda position, as the former are always longer (see phonetic experiment in Chapter 3). This is an exact parallel to the /g/ → [x] (word-finally) vs. /x/ case discussed above. The retroflex characteristic of /ʂ ʐ / is “a feature that varies in salience from speaker to speaker” (Munro & Lopez, 1999, p. 2). In my experience, it is quite common for retroflex segments to neutralize with their corresponding prepalatal fricatives / ʃ ʒ /. Below, I illustrate Quiaviní Zapotec fricatives. As with stops, initial, intervocalic and word-final positions are presented.  40  Fortis fricatives: / s ʃ ʂ / (31) Word-initially ( #_ ): voiceless fricatives a. / silʲ /  ˥  → [ síːlʲ ]  b. / ʃabdi̤a̤ / ˥ ˩ → [ ʃáb.dì̤à̤ ]  ‘Basilio’ ‘locust’  (32) Intervocalic ( V_V ): voiceless fricatives a. / ʃ-mes=uʔ /  ˥ ˥ → [ ʃmˑé.súʔ ]  b. / ʃ- nˑa̤ʃ=u /  ‘your professor’  ˩ ˥ → [ ʃnˑà̤.ʃúʔ ]  ʔ  ‘your chocolate’  (33) Word-finally ( _# ): long voiceless fricatives a. / mes / ˥  → [ mésː ]  ‘professor’  b. / nˑa̤ʃ / Ü  → [ nˑà̤ʃː ]  ‘much, a lot of’  Lenis fricatives: / z ʒ ʐ / (34) Word-initially ( #_ ): voiced fricatives a. / za̤ /  ˩  b. / ʒ ḭʒ / Ü  → [ zà̤ː ]  ‘grease, fat’  → [ ʒî͡ḭʒ̊ ]  ‘pineapple’  (35) Intervocalic ( V_V ): voiced fricatives a. / ʃ-bɡa̰z=uʔ /  ˩ ˥ → [ ʃab.ɡà̰.zúʔ ]  b. / ʃ-wbwi̤ʒ=u / Ü ˥ → [ ʃaw.bwî͡i̤.ʒúʔ ] ʔ  ‘your place in the mountains’ ‘your sun’  (36) Word-finally ( _# ): voiceless fricatives a. / bɡa̰z /  b. / wbwi̤ʒ /  ˩ Ü  → [ bɡà̰ːz̥ ~ bɡà̰ːs ]  → [ wbwî͡i̤ʒ̊ ~wbwî͡i̤ʃ ]  ‘name of a place in the mountains’ ‘sun’  Finally, as reported by Munro and Lopez (1999), the fricatives / f / and / x / appear primarily in Spanish loanwords. These sounds pattern with the other fortis fricatives.  41  As a final remark with regard to obstruents, we observe a clear difference with respect to fortis versus lenis, in that the former are more stable in the way they are produced. Fortis stops are invariant in terms of voice and manner, whereas lenis stops vary depending on the context. For fricatives, fortis are invariant, and lenis vary in their voicing. Cross-linguistically, obstruents form a complementary class to sonorants. Obstruents are produced by a narrowing or complete closure of the vocal tract, and the lack of voicing is the default setting for this type of segment, i.e. the existence of voiced obstruents implies voiceless ones. In summary, fortis obstruents are not only invariant in voicing and manner, but also the ones that manifest the prototypical, or typologically unmarked, properties of obstruents. I now turn to the analysis of fortis and lenis sonorants. Quiaviní Zapotec fortis sonorants are the nasals / mˑ nˑ ŋˑ /, and the liquid / lˑ /. They have similar characteristics, as they are voiced, and long in coda (although fortis sonorants may be partially devoiced following breathy vowels, especially / lˑ /, Munro & Lopez, 1999, p. 2). The lenis sonorants / m n ŋ / and / l / are shorter than their fortis counterparts and may devoice word-finally, particularly after interrupted vowels. Fortis sonorants (37) Word-initially ( #_ ): voiced sonorants a. / mˑuʒ / ˥  → [ mˑúːʒ̊ ]  ‘blond’  b. / nˑan / Ë  → [ nˑǎːn ]  ‘mother’  → [ lˑâ̤nː ]  ‘smelling of eggs’  c. / lˑa̤nˑ / Ü  (38) Intervocalic ( V_V ): voiced sonorants a. / ʃ-damˑ=uʔ / b. / danˑo̤nˑɨŋ / c. / nsualˑ-eʔ /  Ë˥ → [ ʃtǎmːúʔ ]  ‘your owl’  ˥ ˩ → [ nsuálːèʔ ]  ‘little blue’  ˥ Ü → [ dánˑô͡o̤nːɨŋ ] ‘It's us’  42  (39) Word-finally ( _# ): voiced sonorants a. / damˑ /  → [ dǎmː ]  ‘owl’  b. / danˑo̤nˑ / ˥Ü  → [ dánˑôo̤nː ]  ‘we’  c. / nsualˑ / ˥  → [ nsuálː ]  ‘blue’  Ë  Lenis Sonorants (40) Word-initially ( #_ ): voiced sonorants a. / nan / ˩  b. / laŋˑ / Ë  → [ nàːn ] → [ lǎŋː ]  ‘thick’  ‘s/he/it (nearby)’  (41) Intervocalic ( V_V ): voiced sonorants a. / ʃ-ɡuʔan=uʔ /  Ü ˥ → [ ʃkúʔànúʔ ]  ‘your bull’  b. / ʃ-lua̰n=uʔ /  Ü ˥ → [ ʃlˑúà̰ːnúʔ ]  ‘your sleeping platform’  (42) Word-finally ( _# ): voiced or voiceless sonorants a. / ɡuʔan / Ü  → [ ɡúʔàn ~ ɡúʔàn̥]  ‘bull’  b. / lua̰n / Ü  → [ lúà̰ːn ~ lúà̰ːn̥]  ‘sleeping platform’  According to Munro and Lopez (1999, p. 2), the trill / r / “appears in Spanish loans or over a morpheme boundary in non-loans, where it functions phonologically as a cluster […]; only the tap r occurs internal to native morphemes.” As mentioned above, arguably, trill and tap act as fortis/lenis counterparts. The feature [+sonorant] characterizes sounds that are produced in such a way that the vocal cords vibrate spontaneously (i.e. vowels, glides, liquids and nasals), thus voicing is the default property of sonorants, as is the case for fortis sonorants in Quiaviní Zapotec. However, voicing variation between fortis and lenis sonorants is not as salient as in the case of obstruents. Munro and Lopez (1999, p. 2) propose that duration is the most important cue to differentiate fortis versus lenis sonorants.  43  Similar characteristics to the ones presented above are described for Güilá Zapotec by Arellanes (2009). He makes use of the markedness concept in the analysis of the fortis/lenis contrast. “The concept of markedness, in its most general characterization is concerned with the distinction between what is neutral, natural, or most expected (unmarked), and what departs from the neutral (marked) along some designated parameter” (Kean, 1992, p. 390). Along these lines, Arellanes (2009, p. 176) establishes for Güilá Zapotec that “…las fortis son los elementos más básicos del sistema tanto porque constituyen un mayor número de en el inventario consonántico, como porque tienen una distribución más amplia en los distintos contextos fonológicos básicos”.25 In Quiaviní Zapotec, fortis consonants manifest the unmarked features of the class they belong to (stable in all contexts), whereas lenis consonants may have the marked features of the class and their realization fluctuates depending on the context. What unifies fortis consonants as a natural class in Zapotec, then, is the fact that they express the unmarked features of the sub-class they belong to, and their production is phonetically constant. Fortis obstruents are always voiceless (and stops always [-continuant]); whereas fortis sonorants are always voiced. In turn, these characteristics make fortis segments “stronger” than lenis ones in terms of duration, tension and intensity. In addition, the length of fortis vs. lenis consonants is also worth remarking on. All fortis consonants are particularly long in coda position; this phonetic duration is taken to be prosodically relevant in that fortis consonants are moraic in coda position, as proposed and explained in detail in Chapter 3. This fact provides additional evidence for these segments as a natural classes. (A singleton-geminate alternative analysis is also discussed in Chapter 3.) Accordingly, the fortis/lenis contrast (or phonological strength) is not something we can characterize with the heretofore-standard features or properties, although it clearly encodes a phonological contrast.  25  “Fortis consonants are the most basic elements of the system as they are more numerous in the consonant inventory, and they have a wider distribution in basic phonological contexts.” [Translation mine]  44  2.4  The emergence of the [+/-fortis] feature  In emergent feature theory (Mielke, 2008 [2004]), features are abstract categories based on generalizations that emerge from phonological patterns. In other words, different phonetic properties can be relevant for defining sound patterns, and as such, we would expect some degree of variation cross-linguistically (contra the nativist approach of Universal Grammar of a single set of features present in all languages (Chomsky & Halle, 1968, etc.)). The argument for emergence partially depends on distinctions like fortis/lenis, tense/lax, stressed/unstressed being vague composites of properties that vary across languages but that clearly combine to create some overall distinction (strength, loudness, prominence, etc.). Along these lines, phonological strength may be encoded differently in the languages of the world. In Quiaviní Zapotec, the sound pattern that arises from the description of the previous section is that fortis obstruents and fortis sonorants form a natural class based on the following two facts: (i) fortis consonants are the unmarked segments of the class (determined by different phonetic characteristics); and (ii) fortis consonants are long in coda position (playing a crucial role in the prosodic system of this language, in terms of moraicity (Chapters 3 & 5) and tone (Chapter 5)). Based on these phonetic and phonological properties, there must be a way in which the grammar classifies these subsets of consonants (across obstruents and sonorants). A feature that emerges from these language-specific patterns is a legitimate approximation. What is the best feature then for the fortis/lenis contrast? Hollenbach (1984) adopts the feature [+/-tense] to account for the fortis/lenis contrast in Copala Trique. She defines lenis obstruents as [+voice] but [-tense] while fortis ones are [-voice] and [+tense]. The original definition of this term, may certainly encode the fortis/lenis observed properties. As defined in (Jakobson et al., 1951, p. 38): Tense phonemes are articulated with greater distinctness and pressure than the corresponding lax phonemes. The muscular strain affects the tongue, the walls of the vocal tract and the glottis. The higher tension is associated with a greater deformation of the entire vocal tract from its neutral position. This is in agreement with the fact that tense phonemes have a longer duration than their lax  45  counterparts. The acoustic effects due to the greater and less rigidity of the walls remain open to question. The feature [+/-tense], however, has been used for a wide variety of phenomena and it is most often associated with vowel contrasts (e.g. / i, u / vs. / ɪ, ʊ /). It seems that we can simply refer to the feature [+/-fortis], employed in a diverse and compelling literature (Debrock, 1978; Gerhardt, 1980; Kohler, 1984; Pulleyblank, 2006; among others). 26 To conclude, the composite properties from which the fortis/lenis distinction arises in Quiaviní Zapotec, across both obstruents and sonorants, is encoded here with the feature [+/-fortis].27 An alternative analysis is presented by Arellanes (2009) for Güilá Zapotec. The author explains the variation of lenis consonants on the basis of feature underspecification. The described generalizations, however, do not hold for the Quiaviní Zapotec data presented here. As described above, I assume that Quiaviní Zapotec consonants are specified for the features [+/-voice], [+/-continuant], [+/-sonorant], as these features maintain particular contrasts (e.g. voicing is clearly distinctive for obstruents in initial and intervocalic positions) or predict specific patterns (see the role of [+/-sonorant] in the expression of tone in Chapter 5). However, in order to account for the full range of properties of the fortis/lenis contrast in Quiaviní Zapotec, and to encode a natural class across obstruents and sonorants, an additional specification is necessary, that of the [+/-fortis] feature.  26  Kohler (1984), for instance, presents evidence for the importance of a [+/-fortis] feature in the description of phonological segment systems in the world’s languages, particularly for Germanic languages. In his proposal, the feature is associated with articulatory timing (“power in the supraglottal movements and in the air stream” (p. 168)) and with laryngeal tension. Both features [tense] and [fortis] refer to phonological and phonetic strength. More recently and within an emergent feature approach, Pulleyblank (2006) proposes the use of the feature [+fortis] in Luo, to group oral stops and pre-nasalized stops. 27 Under Optimality Theory (Prince & Smolensky, 2004 [1993]), the markedness distinction between fortis and lenis consonants in Quiaviní Zapotec, may be conceptually analyzed by the following harmonic scale: FAITH[fortis] >> MARKEDNESS >> FAITH[lenis] (see Howe & Pulleyblank, 2004 for an analysis of harmony as faithfulness; cf. de Lacy, 2006). Accordingly, the feature specification of fortis consonants (e.g. voicing, manner) requires a faithful input-output correspondence, whereas that of lenis consonants is subject to markedness constraints (e.g. an intervocalic context demanding voicing versus a final utterance position that favors devoicing; cf. Arellanes, 2009, Chapter 4). This is in fact what we observed in the adaptability of lenis consonants. The details of such an analysis, however, are beyond the scope of this study.  46  2.5  Conclusions  In this chapter, based on Munro and Lopez (1999), I showed that vowel length is predictable from the consonant type in stressed syllables (short before fortis, and long before lenis). The relevance of the fortis/lenis distinction in determining this vowel pattern represents an important characteristic of Quiaviní Zapotec consonants, as both fortis obstruents and sonorants are long in coda position. In addition, fortis segments present phonetic characteristics that make them the unmarked elements of their consonantal class: fortis stops are the extreme of “strong” articulation, being always voiceless, and invariant in their constriction. Fortis fricatives are also always voiceless and, consequently, of higher amplitude compared to their lenis counterparts (cf. Jaeger, 1983). Finally, fortis sonorants are always voiced. These language-particular properties support the hypothesis that a number of phonetic and phonological characteristics contribute to the fortis/lenis contrast in Quiaviní Zapotec, and that the grammar of this language needs to refer to these patterns. This is in accordance with emergent feature theory (Mielke, 2008 [2004]; Pulleyblank, 2006), for which features emerge from phonological patterns rather than the other way around (as in a nativist approach with a set of universal features). Accordingly, I adopt the feature [+/-fortis] (Kohler, 1984; Pulleyblank, 2006) to account for Quiaviní Zapotec consonant contrasts. The importance of this chapter derives from the characterization of the fortis/lenis distinction in Quiaviní Zapotec, as a pervasive contrast in the consonants of this language and of particular relevance for its metrical structure and tone. The generalizations arrived at in this chapter form the basis and preamble for the analysis of different prosodic patterns of Quiaviní Zapotec, presented in subsequent chapters.  47  Chapter 3:  Metrical structure of Quiaviní Zapotec  3.1  Introduction  Metrical structure refers to the organization of segments in terms of prosodic units (e.g. mora, syllable, foot). This organization, or rhythmic structure, may be reflected as the stress or prominence pattern of a language. According to Munro and Lopez (1999), the last syllable of uninflected words is stressed in Quiaviní Zapotec. Nonetheless, no further study has accounted for more details of the prosodic system of this language, such as the moraicity of its segments, the properties of foot structure, or minimality effects. The goal of this chapter is to account for the metrical structure in this language (up to the Prosodic Word (PrWd)), establishing the foundations needed for two central topics of this dissertation: tone and phonation type.  48  All the examples used in this chapter are content words (nouns, verbs, adjectives); most function words (articles, some adverbs, clitics, etc.28) do not need to be stressed, and are prosodically dependent on content words. In order to account for the constituency of the PrWd in this language, the chapter begins with the analysis of the smallest prosodic domain: monosyllables, where I establish the prosodic minimality and moraicity of Quiaviní Zapotec (§3.2). Section 3.3 continues to the next morphological level, analyzing prefixed, suffixed and clitisized words and compounds. These disyllabic and longer words will provide evidence for a trochaic (foot type) rhythm in Quiaviní Zapotec as well as its demarcative characteristic, with the root consistently carrying prominence. Finally, loanword phonology, §3.4, concludes the analysis of word stress in Quiaviní Zapotec. Focusing on modal phonation, all these sections present a formal analysis within the approach of Optimality Theory (OT; Prince & Smolensky 2004 [1993]).  3.2  Moraicity and minimality This section analyzes monosyllables in Quiaviní Zapotec. In this language, most  noun roots are monosyllables and coextensive with prosodic words. Verbs require the presence of an aspectual prefix, but many of these prefixes (see Chapter 1, §1.4.5) are single consonants; thus, verbs may also surface as monosyllables. These words constitute the minimal words in Quiaviní Zapotec and are the focus of this section. The goal is to account for the prosodic requirements of monosyllabic native words in Quiaviní Zapotec. As explained in the preceding chapter, vowel length is a perceptually salient feature of Quiaviní Zapotec; however, it is not contrastive, but rather conditioned by prominence and by the consonant type in coda position. In stressed syllables (all monosyllabic nouns and verbs being stressed), short vowels appear before fortis  28  As presented in Chapter 1, most prepositions originally come from grammaticalized nouns (Lillehaugen, 2003, 2006), for instance làa’iny, listed in the dictionary of Munro and Lopez (1999) as ‘stomach’, but also as the preposition ‘inside, in, into’. The precise prosodic status of these “prepositions” is unclear and beyond the scope of this study.  49  consonants and long vowels before lenis consonants or in open syllables. Consider the following examples. (1) Short vowels before fortis coda consonants a. / bak / → [ bakː ] ‘person from Tlacolula’ b. / nanˑ /  → [ nanː ]  ‘knows’  c. / damˑ / → [ damː ]  ‘owl’  d. / mes /  ‘professor’  (< Sp. maestro)  ‘knife’  (< Sp. cuchillo)  → [ mesː ]  e. / bʧilˑj / → [ bʧilː ] j  (2) Long vowels before lenis coda consonants a. / baɡ / → [ baːx ] 29 ‘cow’  (< Sp. vaca)  b. / nan /  → [ naːn ]  ‘thick’  d. / bal /  → [ baːl ]  ‘bullet’  (< Sp. bala)  e. / sɨby /  → [ sɨːɸj ]  ‘Eusebio’  (< Sp. Eusebio)  (3) Long vowels in open syllables a. / la / → [ laː ] ‘is named’ b. / n-ɡi / → [ ŋɡiː ]  ‘sour’  c. / wi /  → [ wiː ]  ‘guava’  d. / tu /  → [ tuː ]  ‘who’  e. / ʃnia /  → [ ʃniaː ]  ‘red’  Cross-linguistically, it is well known that vowels are shorter before voiceless consonants and longer before voiced ones. The magnitude of this effect in most languages without contrastive vowel length may vary between 10 to 20 ms (e.g. Mendoza et al., 2003 report a difference of 16 ms for Spanish, with means of 126 ms vs. 142 ms). However, the magnitude ratio in Quiaviní Zapotec exceeds this phonetic universal and resembles that of languages with contrastive vowel length (e.g., in Tamil (Dravidian), short vowels average 93ms, long vowels 152ms (Maddieson, 1984)).30 Nevertheless, in Quiaviní Zapotec, voicing is clearly not the determining factor for this vowel-duration difference, but the fortis/lenis contrast, as established in the previous Chapter. On the one hand, 29  As presented in Chapter 1, lenis stops are frequently fricated, devoiced and short word-finally. English seems to be a language with middle range values, reporting differences of 30 to 40 ms, or more before pause (e.g. Chen, 1970; Keating, 1984; Erickson, 2000 among others). 30  50  obstruent fortis consonants are always voiceless and so the realization of vowels as short is expected, but lenis obstruents also tend to be voiceless in word-final position, and preceding vowels are nonetheless long. On the other hand, fortis sonorants are always voiced, whereas lenis may devoice word-finally. Despite this difference with respect to obstruents, the vowel-lengthening pattern is the same: short before fortis sonorants, and long before lenis ones. Previously, I reviewed how this vowel and consonant length has been reported for different Zapotec languages (Swadesh in Pike 1948: 167; Nellis & Hollenbach 1980, Smith-Stark 2003; Avelino 2004; Leander, 2008; Ward et al. 2008, Arellanes 2009, among others). According to Arellanes (2009), vowel length in San Pablo Güilá Zapotec is adequately explained in terms of minimality and moraicity. Arellanes proposes that the minimal prosodic word in San Pablo Güilá Zapotec consists of a bimoraic foot (Prince & Smolensky, 2004 [1993]). This condition, in conjunction with the predominance of monosyllables in the language, forces content words to form bimoraic syllables. In words of the CVCfortis type, the vowel contributes an underlying mora, and the fortis coda consonant gets a mora by virtue of Weight by Position (applicable to fortis consonants only, see discussion below). The claim that lenis consonants are not moraic follows from the fact that in CVːClenis words, the vowel lengthens to satisfy the minimality requirement of bimoraicity. (The same is observed in CV words.) Further, Arellanes and Chávez-Peón (2009) extend this analysis to Valley Zapotec (including the variants of Güilá and Quiaviní Zapotec, which are mutually intelligible and spoken in neighboring towns). Quiaviní Zapotec word types are presented again in (4-5), with their moraic analysis in (8). (4) CVCfortis a. / bak /  → [ bakː ]  b. / nanˑ / → [ nanː ]  ‘person from Tlacolula’ ‘knows’  (5) CVːClenis a. / baɡ /  → [ baːx ]  b. / nan / → [ naːn ]  ‘cow’ ‘thick’  51  (6) *CVClenis (7) *CVːCfortis (8) Moraic representation of Quiaviní Zapotec words a. Foot | σ | µ µ | | b a k ‘person from Tlacolula’ c. Foot | σ | µ µ | | n a n ‘knows’  b. Foot | σ µ µ b  a x ‘cow’  d. Foot | σ µ µ n  a n ‘thick’  Fortis coda consonants contribute a mora to the formation of the foot, and the preceding vowel is short (monomoraic) (8a&c). Lenis coda consonants, in contrast, do not contribute a mora (but link directly to the syllable), and the preceding vowel must consequently become bimoraic (8b&d). As a result, both types of rhymes satisfy Quiaviní Zapotec minimality. Minimality and Weight-by-Position, the two crucial aspects of this analysis, merit discussion. The notion “minimal word” builds on earlier work by Prince (1980), Broselow (1982), and, particularly, McCarthy and Prince (1986). In many languages, there is a minimum placed on the prosodic size of a word. Some languages require every content word to have at least two syllables (e.g. Mohawk, Michelson 1988); in other languages, every word must contain at least two moras (e.g. Fijian, Hayes 1995); that is, it must consist of at least one heavy syllable or two light ones. Within metrical theory (e.g. Hayes 1995), these requirements can be stated as the requirement that every Prosodic Word by definition (e.g. universally) contains at least one foot (in the same way as a foot requires at least a syllable in it) and that minimality is just a restriction that feet must be binary. 52  In Quiaviní Zapotec, minimality refers to the requirement that a freestanding, stressable (nonclitic) word has a specific minimal weight: bimoraic. Accordingly, vowels lengthen in open syllables and when followed by lenis consonants, whereas fortis consonants get a mora in coda position (as illustrated in (8)). The moraic status of fortis consonants is achieved via the principle Weight-by-Position, to which I now turn. Cross-linguistically, closed syllables vary with respect to their contribution to syllable weight. Coda consonants are moraic in some languages, whereas in others they have no prosodic role. (9) Typology of moraicity for consonants (cf. Zec 1988; Morén 2003; Gordon 2006) a. Every coda consonant is moraic CVV, CVC > CV (e.g. Latin; Finnish (Kiparsky 1968); Japanese (Vance 1987); Arabic (Broselow 1995)) b. No coda consonant is moraic CVV > CVC, CV (e.g. Khalkha Mongolian (Bosson 1964, Walker 1997); Lardil, Huasteco (Broselow 1995: 189)) It is assumed that the difference between these languages is the role of the principle Weight-by-Position (Hayes, 1989: 258), schematized below. (10) Weight by Position (Hayes 1989: 258) σ | µ | α β  →  σ |\ µ µ | | α β  Elements of type β get a mora in the derivation by virtue of being in coda position (i.e. when they belong to the rhyme). Languages of the type in (9a) apply this principle thoroughly, whereas languages of the type in (9b) do not. In addition, the β element in the configuration of (10) may be defined, not only in terms of syllabic position (e.g. coda), but also in terms of a specific type of segment, e.g.  53  sonorant. This is the case in some languages, where only a subset of consonants are moraic in coda position. (11) Extended typology of moraicity for consonants c. Some types of consonants are moraic, others not CVV, CVC1 > CVC2, CV (Kwakw’ala (Boas 1947, Bach 1975); Lithuanian (Zec 1988); Ponapean (Goodman 1995); Yawelmani (Broselow 1995: 201)) Languages in this category, where only a subset of consonants contribute to syllable weight, normally base this distinction on sonority: sonorant consonants are moraic, whereas obstruent consonants are not. Arellanes and Chávez-Peón (2009) propose a new distinction among consonants to determine their prosodic status. In Valley Zapotec, including Güilá and Quiaviní Zapotec variants, fortis coda consonants (both sonorants and obstruents) are moraic, whereas lenis coda consonants are not (neither sonorants nor obstruents). The proposal then is that this language also belongs to this third type of languages; however, the distinction between moraic versus non-moraic consonants is based on duration (not sonority), encoded by the class of fortis vs. lenis segments. Consequently, in Quiaviní Zapotec it is possible to group, on the one hand, fortis obstruents and sonorants, / p t k s ʃ mˑ nˑ ŋˑ lˑ … /, and, on the other, lenis obstruents and sonorants, / b d g z ʒ m n ŋ l … /. Following the schema in (10), in Quiaviní Zapotec, the β type of elements are fortis consonants (i.e. segments specified as [+fortis]). (12) Weight by Position in Quiaviní Zapotec σ | µ → | C V Cfortis C  σ |\ µ µ | | V Cfortis  Onset consonants are non-moraic in Quiaviní Zapotec, regardless of the type of consonant, but crucially, fortis consonants are longer in coda than in onset position. In 54  quantity-sensitive languages, there is a correspondence between duration and quantity: moraic consonants are longer than their non-moraic counterparts, more or less in the same way that bimoraic vowels are longer than monomoraic ones. Duration is still significant in onset position, differentiating fortis versus lenis, but to a much lesser degree (see §3.2.1 below); other features (e.g. voicing), nevertheless, are the main cues to maintain the contrast (see previous Chapter). To sum up, in this section I showed that short vowels appear before fortis consonants and long vowels before lenis consonants (or in open syllables). This pattern was adequately explained in terms of minimality and moraicity. Prosodic words are required to minimally form a bimoraic foot. Fortis consonants get a mora in coda position by virtue of Weight-by-Position, so that the mora of the short vowel plus the mora of the fortis consonant satisfy minimality. Lenis consonants cannot bear a mora; consequently, vowels followed by a lenis consonant lengthen to become bimoraic and form a bimoraic foot (an OT account of this pattern is presented below in §3.2.2). In order to confirm this analysis, the following section consists of a phonetic experiment where I test the hypothesis that fortis consonants are moraic in coda position. Considering duration as one phonetic expression of moraicity, I will show that the differences in vowel and consonant length are not simply by-product effects of differences in intrinsic duration between voiceless vs. voiced consonants, but rather enhanced characteristics that must be considered overt prosodic bimoraicity.  3.2.1 Phonetic experiment: Syllable weight and the fortis/lenis distinction Quiaviní Zapotec presents an uncommon four-way contrast within its consonantal system, which includes the obstruent/sonorant contrast as well as the fortis/lenis distinction. Table 14. Quiaviní Zapotec four-way consonant contrasts Fortis Lenis Obstruents √ √ Sonorants √ √  55  The fortis/lenis distinction crosscuts the contrast between obstruents and sonorants. In the previous section, I argued that fortis coda consonants —both sonorants and obstruents— are moraic, whereas lenis consonants are not —neither sonorants nor obstruents. Onset consonants do not contribute to syllable weight. This section tests this analysis acoustically. This experiment assumes consonant duration as one phonetic expression of moraicity resulting from Weight-by-Position, supported by numerous studies (e.g. Hyman, 1985; Hayes, 1989; and more recently Cohn, 2003; Gordon, 2006; de Lacy, 2007, p. 293). Moreover, onset versus coda differences are considered. It has been shown that these syllabic positions have phonetic differences in gestures and timing (e.g. Gick & Wilson, 2006). Phonological differences include the well-known observation that only coda consonants may be moraic, along with the fact that moraic consonants are longer than their non-moraic counterparts (Hayes, 1989; Perlmutter, 1995). From this background, the two main predictions of this study are, first, that fortis coda consonants are significantly longer than lenis coda consonants, and, second, that fortis consonants are longer in coda than in onset position. The second prediction also follows from the fact that fortis consonants are singleton segments (Munro & Lopez, 1999), moraic in coda (as argued here), but not underlyingly (see discussion below in §3.2.2). Based on previous studies in Zapotec languages (e.g. Jaeger 1983, Avelino 2004), the fortis/lenis distinction is expected to show duration differences regardless of syllable position, so we also expect a significant difference between fortis/lenis in onset, but to a lesser magnitude than that of coda position. Finally, no theory predicts much difference for non-moraic lenis consonants in onset vs. coda position. In summary, the hypothesis of this study is that fortis consonants are moraic, from which the predictions listed in (13) follow.  56  (13) Predictions 1. Fortis coda consonants are longer than lenis coda consonants (expected ratio ≈2:131) 2. Fortis consonants are longer in coda than in onset position 3. The duration difference between fortis and lenis in onsets is smaller 4. The duration difference between onset and coda lenis consonants is not significant  Methodology Subjects: Two native speakers of Quiaviní Zapotec participated in the study: female Speaker Lia L (30 years old), and male speaker Tiu C (46). Stimuli: Obstruent and sonorant consonants were included in the stimuli (including both stops and fricatives for obstruents).32 Because place of articulation plays no major role distinguishing sonority or moraicity, all segments in the stimuli were coronals.  (14) Segments considered in the stimuli Obstruents / \ Stops Fricatives / \ / \ Fortis Lenis Fortis Lenis | | | | t d s z  Sonorants | Nasals / \ Fortis Lenis | | nˑ n  31  The salient difference between fortis vs. lenis consonants in coda position allows us to predict an approximate ratio of 2:1 (similar to that found in languages with a singleton/geminate contrast). No specific ratios are expected for the other predictions. 32 Because of the dissimilar manner of articulation, both stops and fricatives were included in this experiment. In addition, future comparisons between obstruents and sonorants may include intensity as a phonetic parameter (impossible to obtain from stops, but recoverable from fricatives). For sonorants, the assumption was that the difference between nasals and liquids would be minimal; only nasals were included.  57  The words used as stimuli, listed in Table 15 and Table 16, were chosen so that each of the consonants in (14) appears four times in onset position and four times in coda position. Table 15. Stimuli by ONSET (4 items for each consonant: / t, d, s, z, nˑ, n /) 1 2 3 4  /tan/ ˥  /tuat/ ˥ /tas/ ˥  /ta̤p/ ˩  ‘Cayetana’ ‘marrow bone’ ‘cup’ ‘four’  5  /dad/ ˥  ‘dice’  6  /dad/ Ë  ‘father’  7  /damˑ/ Ë  ‘owl’  8  /danj/ ˩  ‘mountain’  9  /sanˑ/ Ë  ‘Santos’  10 /sanˑʒ/ Ë ‘pet sheep’ 11 /sja̰b/ Ü ‘atole’ 12 /sualˑ/ ˥  13 /zuas/ Ü  ‘blue’ ‘type of plant’  14 /zak/ ˥  ‘good’  15 /ze / Ü  ‘corn on the cob’  ʔ  16 /zua̤z/ Ü  ‘drunk’  17 /nˑad/ ˩  ‘hard-headed’  18 /nˑan/ Ë  ‘mother’  19 /nˑje̤s/ Ü  ‘water’  20 /nˑuan/ ˥ ‘chirimoya’ 21 /nan/ ˩ ‘thick’ 22 /nanˑ/ ˥  ‘woman's nickname’  23 /njanˑ/ ˥  ‘Marcelo’  24 /njan/ Ë  ‘spicy’  58  Table 16. Stimuli by CODA (4 items for each consonant: / t, d, s, z, nˑ, n /) 1 2 3 4  /tʃat/ ˥ /lat/ ˥  /tuat/ ˥  /ʒyet/ Ë  ‘kiss’ ‘tin can’ ‘marrow bone’ ‘cat’  5  /lad/ ˥  ‘side’  6  /dad/ ˥  ‘dice’  7  /dad/ Ë  ‘father’  8  /nˑad/ ˩  ‘hard-headed’  9  /na̤s/ ˩  ‘the day before yesterday’  10 /tas/ ˥  11 /zuas/ Ü  12 /nˑje̤s/ Ü  ‘cup’ ‘type of plant’ ‘water’  13 /ni̤a̤z/ Ü  ‘corn field’  14 /ɡaz/ ˩  ‘seven’  15 /klaaz/ ˥ ‘Nicolasa’ 16 /zua̤z/ Ü ‘drunk’ 17 /tʃonˑ/ ˩ 18 /sanˑ/ Ë  ‘three’ ‘Santos’  19 /nanˑ/ ˥  ‘woman's nickname’  20 /njanˑ/ ˥  ‘Marcelo’  21 /tan/ ˥  22 /nˑan/ Ë  ‘Cayetana’ ‘mother’  23 /nˑuan/ ˥ ‘chirimoya’ 24 /nan/ ˩ ‘thick’ These words were recorded within the following carrier phrases. (15) Carrier phrases: a. Stops:  [ ɾnḭŋ  __________ kuan dḭʒ sa̤]  ‘He says _______ in Zapotec’  b. Fricatives: [ ɾnḭaʔ ɾa __________ kuan dḭʒ sa̤ ]  ‘I say _______ in Zapotec’  c. Nasals:  ‘I say _______ again’  [ ɾnḭaʔ ɾa __________ steːbj ]  59  The use of slightly different carrier phrases is due to the different types of consonants that are considered. The main issue is that lenis stops are produced as fricatives intervocalically; thus, in order to maintain the same category in the comparison (lenis stop vs. fortis stop), the first part of the carrier phrase includes a nasal (after which lenis stops are realized as voiced stops). The carrier phrases were intended to obtain the best pronunciation of each of the consonant types for a proper comparison as well as to look for the easiest environments in which to measure these segments. Four repetitions of each word within its carrier phrase were collected based on a randomized list. In the cases where the speaker was unable to read, the phrase was given in Spanish by the facilitator (the author). Recordings were made with a Marantz 660 solid-state recorder and a Countryman lapel microphone (phantom power). Each consonant was measured for duration (based on both the waveform and the spectrogram): total constriction in obstruents (closure and release for stops, and frication period for fricatives) and total duration of nasals, cued by change in amplitude and formant transition. In total, 384 consonants were measured by hand in Praat for Mac (version 5.1.07; Boersma & Weenink, 2009) (4 /t/ onset + 4 /t/ coda + 4 /d/ onset + 4 /d/ coda + 4 /s/ onset + 4 /s/ coda + 4 /z/ onset + 4 /z/ coda + 4 /nˑ/ onset + 4 /nˑ/ coda + 4 /n/ onset + 4 /n/ coda = 48 x 4 repetitions = 192 tokens/speaker x 2 consultants = 384 total duration measurements). Results were compiled in Excel 2004 for Mac and the statistics were run in R (version 2.8.1, R Development Core Team, 2009). Data was statistically evaluated with two-tailed t-tests.  60  Results The results of the experiment are presented (in ms.) in Table 17. Table 17. Results of phonetic experiment (duration of fortis and lenis consonants) Female - LiaB Mean SD ------ Male - TiuC t_coda 142.87 43.93 t_coda t_onset 101.93 7.75 t_onset d_coda 61.05 10.31 d_coda d_onset 58.01 7.89 d_onset  Mean SD 131.68 22.35 99.75 9.11 55.01 8.43 52.68 11.81  s_coda s_onset z_coda z_onset  169.87 108.31 86.05 80.12  52.76 10.65 20.61 25.16  s_coda s_onset z_coda z_onset  145.75 105.87 79.42 72.56  nn_coda nn_onset n_coda n_onset  134.56 70.13 85.64 78.05  25.01 22.09 24.81 25.16  nn_coda nn_onset n_coda n_onset  121.56 13.79 69.68 17.41 87.43 8.57 60.31 15.74  V before lenis V before fortis  158.51 26.59 82.61 14.75  35.71 18.94 17.42 15.91  V before lenis 155.51 16.96 V before fortis 77.91 14.74  Table 18. t-test results of phonetic experiment (duration of fortis and lenis consonants) 33 Prediction 1 Prediction 2 Prediction 3 Prediction 4  Female speaker (LiaL)  Male speaker (TiuC)  t (79.018) = 10.1644, p < 0.001 t (79.089) = 7.8135, p < 0.001 t (99.723) = 4.4029, p < 0.001 t (101.135) = -0.7861, p = 0.43  t (91.721) = 12.1018, p < 0.001 t (97.641) = 7.6297, p < 0.001 t (86.027) = 7.3037, p < 0.001 t (95.132) = -2.8854, p = 0.013  The results in Tables 17 and 18 show the same trends for both speakers. Fortis coda stops, fricatives and nasals in coda are significantly longer than their lenis correspondents, with a ratio of close to 2:1, which confirms the first prediction. The difference between fortis versus lenis consonants in onset position (prediction 2) is also significant with a small ratio, 1.3:1, and with more overlap. Significant differences also 33  According to standard conventions, results above 0.12 are considered not significant (ns.); results between 0.12 and 0.05 are marginally significant; finally, any value below 0.05 is statistically significant.  61  arose with respect to the distinction between fortis segments in coda and fortis segments in onset position (prediction 3), with an approximate ratio of 1.5:1. With respect to the last prediction, lenis consonants in coda are only slightly longer than in onset (1.1:1). The difference was not significant for the female speaker, as expected, but significant for the male speaker. In addition, the durational difference between vowels before lenis consonants (long) and vowels before fortis consonants (short) is also large and significant (ratio= 1.9:1; p <0.001). The following figures illustrate the results of the experiment with box-plot diagrams along with the t-test p-values. Results for stops, fricatives and nasals are presented together.  62  Figure 6. Box plots and t-test p-values for LiaB: fortis coda vs. lenis coda; fortis coda vs. fortis onset.  Figure 7. Box plots and t-test p-values for LiaB: fortis onset vs. lenis onset; lenis onset vs. lenis coda.  63  Figure 8. Box plots and t-test p-values for TiuC: fortis coda vs. lenis coda; fortis coda vs. fortis onset.  Figure 9. Box plots and t-test p-values for TiuC: fortis onset vs. lenis onset; lenis onset vs. lenis coda.  64  Discussion In order to support the hypothesis that fortis coda consonants are moraic, four predictions were tested in this phonetic experiment. The first and most important stated that fortis coda consonants should be longer than lenis coda consonants. This was clearly confirmed in the experiment with a ratio of 2:1 between fortis versus lenis consonants in that position. These results strongly suggest that the contrast relies on a prosodic distinction (i.e. moraicity). Fortis consonants are only moraic in coda position, not underlyingly, and the second prediction refers to this, as fortis consonants are expected to be longer in coda than in onset. Results confirm this difference with a ratio of 1.5:1. This is in agreement with the assumption that moraic consonants are longer than their non-moraic counterparts (Hayes, 1989; Perlmutter, 1995).34 Since the fortis/lenis distinction is expected to show duration differences regardless of syllable position (e.g. Jaeger, 1983; Avelino, 2004), we also expect a significant difference between fortis/lenis in onset, as was the case. Since consonants are not moraic in onset position, we also expect this difference to be considerably less in comparison to fortis consonants versus lenis consonants in coda position. Coda differences are in a ratio of 2:1 (fortis:lenis), whereas in onset we found only 1.3:1. Differences in onset are comparable to those found, for example, between voiceless versus voiced segments in English (e.g. Baum & Blumstein, 1987). All the comparisons were statistically significant when grouping together stops, fricatives and nasals (above Figures), and importantly, differences were also significant when compared separately (see Figures 69, 70, 72 & 73 in Appendix A). The only exception to this was the comparison between fortis and lenis nasals in onset position (p = 0.4). It is possible that the fortis/lenis distinction is neutralized for /n/ (and possibly other sonorants) in onset position in Quiaviní Zapotec (see §2.3 for a description of fortis and lenis sonorants). Lee (1996) also reports contentious numbers with respect to the 34  The fortis vs. lenis consonant difference in codas is viewed as moraic vs. nonmoraic, as well as the fortis coda vs. fortis onset difference. The ratios are 2:1 and 1.5:1, respectively. This difference in ratios relies on the fact that duration relates to moraicity, but it is also a phonetic correlate for the fortis/lenis distinction, both in onset and coda positions.  65  duration of fortis/lenis sonorants in onset position for Quiaviní Zapotec. Moreover, Arellanes (2009) claims that in Güilá Zapotec only fortis nasals appear in onset position. This issue, however, is beyond the scope of this study. Finally, the difference between lenis consonants in onset vs. coda position was predicted to be small. Results were only significant for the male speaker, but the difference is too small to posit any prosodic difference between lenis onsets vs. codas: the means were 75 ms vs. 66 ms, respectively, with a small ratio of 1:1.1. This distinction is simply attributed to the phonetic differences in gestures and timing in terms of syllable position (see Maddieson, 1984; Gick & Wilson, 2006). In conclusion, the phonological and phonetic evidence shown in this study supports the claim that fortis consonants contribute to prosodic weight in Quiaviní Zapotec, establishing a new distinction—that of fortis/lenis—among the feature contrasts to which Weight-by-Position can be sensitive. These findings emphasize the relationship between syllable weight and syllabic duration as a clear place where phonology and phonetics interact.  3.2.2 Fortis consonants are not geminates Up to this point, I have proposed that fortis consonants in Quiaviní Zapotec lengthen in the coda and become moraic. An alternative to this analysis is that fortis consonants are geminates, underlyingly moraic, that shorten in the onset. Both approaches are based on the analysis that length may be the expression of syllable weight, by means of moras. This section discusses this alternative, and shows that, although a singleton-geminate distinction has been proposed for Proto-Zapotec (Fernández de Miranda, 1995; cf. Swadesh, 1947), this analysis is not appropriate for the synchronic language.35 With respect to the segmental distribution, fortis consonants occur initially and finally in Quiaviní Zapotec, two cross-linguistically unusual positions for geminates. 35  Swadesh (1947) proposes that lenis consonants diachronically derived from single consonants, whereas fortis consonants derived from consonant clusters.  66  Moreover, following the Munro and Lopez (1999) consonant inventory, Quiaviní Zapotec has some unpaired obstruents, the affricates / ʧ / and / ts /, not to mention the borrowed sounds / f / and / x / from Spanish. These four sounds pattern with fortis obstruents; they are short in onset and long in coda. That would mean these were underlying geminates with no singleton counterpart, which seems typologically unusual (see Ham, 2001; Curtis, 2003). In contrast, from a feature-based perspective, it is extremely common cross-linguistically to find languages with voiceless obstruents [-voice] and no voiced counterparts (e.g. / s, ʧ / in Spanish). The following two arguments are structurally and theoretically based, and they are supported by Quiaviní Zapotec phonetic data. Assuming an account that encodes geminates by lexical specification of mora structure (Hyman, 1985; Hayes, 1986; Schein & Steriade, 1989; cf. Curtis, 2003),36 two testable predictions arise. First, the singleton/geminate contrast should be neutralized in onset (word initially) (cf. Kraehenmann, 2001, 2003). Second, fortis consonants should be long both intervocalically and in coda position. The data indicates that these predictions do not hold in Quiaviní Zapotec. The fortis/lenis contrast is maintained in onset position in Quiaviní Zapotec, where the main cue for obstruents is voicing or manner (stops vs. low-amplitude fricative). Consider the following examples.  (16)  a. / palˑ / ʔ  b. / te /  ‘shovel’  vs.  / balˑ /  ‘Valeriano’  ‘one’  vs.  / de̤ /  ‘dust’  The realization of fortis consonant as short in onset vs. long in coda position was confirmed in the preceding phonetic experiment, with a significant average magnitude ratio of 1:1.5. Fortis consonants in onset were always shorter than those in coda, even though they were preceded by a vowel in the carrier phrase (in the case of fricatives and nasals in the phonetic experiment). In turn, results for the durational difference between fortis and lenis consonants in onset were not very different from those reported for other 36  Curtis (2003) proposes to analyze geminates as underlyingly moraic two-root node segments, both facts being necessary to account for all the typological properties of geminates.  67  languages with a voicing distinction (see Baum & Blumstein, 1987, for English fricatives). The second prediction under a geminate analysis is that fortis consonants are long both intervocalically and in coda position; however, this is not the case for fortis obstruents word-internally, as illustrated in (17). a. / tʃat /  (17)  → [ tʃatː ]  b. / tʃat-e / can-DIM c. ʔ  ‘kiss’  → [ ˈtʃa.teʔ ] ‘little kiss’ → *[ ˈtʃatːeʔ ]  Fortis coda obstruents are long word finally, but when a clitic or the diminutive suffix is added, the fortis obstruent resyllabifies as a short (singleton) onset segment. This observation is based originally on data from by-ear transcription from 4 different speakers with several words, followed by phonetic measurements. As shown in more detail below (§3.3.1), duration measurements from one speaker showed that fortis obstruents averaged 230 ms in coda position, whereas the mean of resyllabified fortis obstruents was 113 ms. Similar length differences have been found in languages with singleton/geminate contrast (e.g. Swedish, see Thorén, 2005), but crucially, it is in intervocalic position where we expect these segments to be long if they are geminates. (The different behavior of sonorants will be discussed in subsequent sections.) Consequently, the geminate configuration in (18) is rejected for Quiaviní Zapotec. (18) Geminate representation *  σ  tʃ  σ  µ |  µ |  µ |  a  t  eʔ  In cases like (17), the fortis consonant alternates from something that could be called a geminate, (17a), to something that appears to be a singleton, (17b), but crucially  68  not a lenis obstruent.37 In other words, if the fortis/lenis contrast is a geminate/singleton one, then this change in duration should reflect a fortis/lenis neutralization, which is not the case. When obstruent fortis codas resyllabify as onsets, the contrast between fortis and lenis is still maintained. As shown in Chapter 2, fortis stops are voiceless and short in intervocalic position, whereas lenis stops are voiced, normally fricated, and short. These cases then clearly illustrate the fortis/lenis pattern in the phonetics. Accordingly, positing underlying geminates seems to be an indirect and therefore less illuminating way of capturing the distribution of long and short vowels in a situation where consonant and vowel length clearly are prosodically conditioned. All in all, there is nothing about fortis-lenis obstruents in onsets that would lead anyone to even suspect that it is a short-geminate distinction, partly because of differences in voicing and continuance, and partly because the durations are simply not in the region of short-geminate consonants of languages that have them (see, for instance, Swiss German in Kraehenmann, 2001, 2003). Based on phonetic and phonological evidence, ranging from segmental to prosodic issues, the singleton/geminate analysis fails to account for the full range of facts in Quiaviní Zapotec. Instead the fortis/lenis distinction is the most adequate analysis, where fortis consonants become moraic segments in coda position.  3.2.3 Formal analysis The phonetic experiment described in §3.2.2 supports the claim that fortis consonants are moraic in codas in Quiaviní Zapotec, whereas lenis consonants are not. As presented in §3.3, this pattern satisfies the prosodic requirement for words to minimally form a bimoraic foot. As a result, two types of syllable rhymes emerge in Quiaviní Zapotec monosyllables: monomoraic vowels followed by moraic fortis coda consonants (19), and bimoraic vowels followed by non-moraic lenis consonants (20).  37  The duration of this segment is quite similar to that of short stops in Swedish (Thorén, 2005), a language with geminate counterparts that contrast in intervocalic position.  69  (19) CVCfortis  baµkµ  ‘person from Tlacolula’  (20) CVːClenis  baµµx  ‘cow’  The goal of this section is to formally account for the minimality and moraic characteristics of Quiaviní Zapotec within Optimality Theory (Prince & Smolensky 2004 [1993]). The section focuses on monosyllabic words. Starting with minimality, languages may require content words to have some minimum size. This minimal word typically equals a single foot, consisting of two syllables or two moras. As already established, in Quiaviní Zapotec, where monosyllables are the majority of words, monomorphemic monosyllabic words form bimoraic feet (21). (21) Min PrWd in Quiaviní Zapotec = Bimoraic foot Following the prosodic hierarchy (22), this prosodic requirement is encoded with the constraint FT-BIN, as defined in (23). (22) Prosodic Hierarchy (Selkirk, 1980, McCarthy & Prince, 1986) PrWd | Ft | σ | µ  Prosodic Word Foot Syllable Mora  (23) FT-BIN (Kager, 1999, p. 156; cf. Bernhardt & Stemberger, 1998) Feet are binary under moraic or syllabic analysis Metrical theory assumes a set of universal prosodic categories in a hierarchical relation, as every prosodic category in the hierarchy has as its head an element of the next lower level category. In other words, every PrWd contains a foot, every foot contains a  70  (stressed) syllable, while every syllable contains a mora.38 Accordingly, in Quiaviní Zapotec every phonological word must contain a bimoraic foot (FT-BIN). In Quiaviní Zapotec, in order to satisfy FT-BIN, monosyllables insert a mora, either for the fortis coda consonant or for the vowel to become bimoraic. As such, violations to DEP-µ will be incurred.39 (24) DEP-µ Output moras have input correspondents (No insertion of moras) (25) Minimality: FT-BIN >> DEP-µ / baµk / ‘Tlacolula’ FT-BIN DEP-µ a. baµk *! b.  baµµk * c.  baµkµ * / baµɡ / ‘cow’ d. baµx  *!  e.  baµµx  *  f.  baµxµ  *  / lo̤ / ‘face’ g. lo̤µ h. lo̤ːµµ  *! *  Tableau (25) shows three types of monosyllables, the first one with a fortis consonant in coda, the second with a lenis consonant in coda, and the last word without a coda. In this tableau, all the faithful monomoraic candidates (a), (d), and (g) are eliminated, in fatal violation of the high ranked constraint FT-BIN. For the open syllable / lo̤ / ‘face’, candidate (h) wins because it only violates the low-ranked DEP-µ. With respect to 38  It is also commonly assumed that a ‘grammatical word must be a prosodic word’: GW = PW (Kager, 1999, p. 152). I assume this constraint to be undominated in Quiaviní Zapotec, and thus it indirectly follows that grammatical words must have minimally one foot. 39 If every token of a morpheme has two moras, assuming one mora in the input may appear to violate lexicon optimization; however, as claimed here, bimoraicity in monosyllables corresponds to prosodic requirements rather than underlying parameters. I show below (§3.3) that in compounds and some suffixed words, roots surface as light syllables, which suggests their underlying monomoraicity. In the case of Quiaviní Zapotec, it simply seems more problematic and complex to propose an analysis in which bimoraic roots lose a mora in cases where the root is not prominent.  71  candidates with a coda, the ranking in (25) is insufficient to decide between candidates with long vowels (b. & e.) and candidates with moraic coda consonants (c. & f.). Both types of candidates satisfy FT-BIN by violating DEP-µ. In order to capture the fact that fortis consonants are moraic in coda, it is necessary to include the concept of Weight-by-Position as a constraint, which is ranked below FT-BIN (this crucial ranking will be illustrated in subsequent sections). (26) WEIGHT BY POSITION (WBYP)  (Hayes, 1989)  Coda consonants are moraic (27) Fortis coda consonants: FT-BIN >> WBYP, DEP-µ / baµk / FT-BIN WBYP DEP-µ ‘Tlacolula’ a. baµk *! * b. baµµk *! * c.  baµkµ * / baµɡ / ‘cow’ d. baµx e.  baµµx  *!  * *!  f.  baµxµ  * *  In (27), candidate (b) satisfies minimality but violates WBYP. The winning candidate (c) satisfies both minimality and WBYP, violating only DEP-µ. This ranking is still insufficient to account for words with lenis coda consonants. In tableau (25) above there is a tie between candidates (e) and (f). Vowels are always long before lenis consonants in prominent syllables. It follows then that these consonants are not able to bear a mora on their own, and vowels lengthen to satisfy minimality. In order to capture this, Arellanes (2009, p. 348) proposes a constraint to ban lenis consonants from being moraic, adapted in (28).40 This constraint is highly ranked along with FT-BIN, accounting correctly for candidates with lenis consonants in coda.  40  *L⇔µ (Arellanes 2009: 348) ‘Los segmentos lenis no pueden constituir moras de modo autónomo’ (“Lenis consonants cannot be moraic autonomously’ [Translation mine].)  72  (28) *Lenis-µ If lenis then non-moraic  (adapted from Arellanes, 2009, p. 348)  (29) Lenis coda consonants: FT-BIN, *Lenis-µ >> WBYP, DEP-µ FT-BIN *Lenis-µ WBYP DEP-µ / baµɡ / ‘cow’ a. baµx  *!  *  b.  baµµx c. baµxµ  * *!  * *  Candidate (a), the faithful one, violates minimality. Candidate (c) satisfies minimality, but the moraic lenis consonant incurs a fatal violation of *Lenis-µ. Candidate (c) wins because it only violates the low-ranked WBYP and DEP-µ. To conclude, I repeat below the statements of (30), which summarize the prosodic analysis for Quiaviní Zapotec monosyllables. (30) Quiaviní Zapotec minimality and moraicity (from above) a. Minimal Prosodic Word = bimoraic foot b. Lenis consonants are non-moraic c. Vowels lengthen before lenis C d. Fortis consonants are moraic in coda  3.3  (FT-BIN) (*Lenis-µ) (WBYP)  Morphology: Root prominence Prominence or stress in a language is based on the syntagmatic comparison  among syllables at the word (and phrase) level; i.e. a syllable is prominent in relation to non-prominent ones. In Quiaviní Zapotec, the majority of native roots are monosyllabic, but the addition of affixes as well as disyllabic or longer loanword roots (~20% of the lexicon) allows us to make such syntagmatic comparison, which is the focus of the Chapter. This section in particular shows the root (final) syllable prominence in this  73  language, instantiated by increased duration41, and the ability to bear all phonological contrasts. In Quiaviní Zapotec, neither prefixes nor suffixes are ever stressed. In addition, affixes are never as complex, prosodically and segmentally, as prominent (root) syllables. Starting with examples with prefixes, the most common forms are inflected verbs and derived nouns forms.  (31)  / ka-[ʐṵnˑj]42 / → [ ka.ʐ ṵ ɲː ] PROG-run  ‘(someone) is running’  (32)  / ka-[kwanˑj] / → [ ka.kwa̰ ɲː ] PROG-run  ‘(someone) wakes up (somebody)’  (33)  / ba-[ɡidj] / ANIM-skin  ‘butterfly’  → [ baɡiːdj ]  Examples above show that in all words formed by prefix+root, the root is a heavy syllable (by means of a bimoraic vowel or a moraic fortis coda consonant). (34)  / ka-[ba̤b] / PROG-itch  → [ ka.ba̤ːɸ ] ‘(it) is itching’ *[ ka.ba̤ɸ ]  (35)  ‘(it) is itching’  In order to show that Quiaviní Zapotec prominence pays attention to morphological domains, examples with the diminutive suffix and clitics demonstrate that stress is not simply word-final; in these examples, stress is still located on the root syllable (vowel length in relationship to the diminutive suffix and clitics will be discussed in the next section).  (36)  / [ba̰t]-eʔ /  → [ ba̰.teʔ ]  ‘little skunk’  skunk-DIM 41  Chávez-Peón (2008) investigates the phonetic cues to stress in Quiaviní Zapotec. The study compares pitch, intensity and duration in prominent versus non-prominent syllables; results show that duration is the main acoustic correlate of Quiaviní Zapotec stress. 42 Roots are marked with square brackets.  74  (37)  / ɾ-[ka̰z]=aʔ / → [ ɾka̰ː.zaʔ ] HAB-to  (38)  want-1S  / w-[ba̤nˑj]=i̤ / → [ wba̤ ɲː i̤ ] PERF-to  (39)  ‘I want’  ‘He (that one) wakes up’  wake up-3S.DISTAL  / ka-[ɡje̤t]=rɨŋ / → [ kaɡje̤trɨŋ ] PROG-to  ‘They are playing’  play-3P.PROX  All in all, the data above illustrate that affixes have no effect on stress location. Compounds illustrate the final root (morphological) prominence. When two roots are attached to form a compound, prominence is located on the second root. (40)  / tsɨ /  → [ tsɨː ]  ‘ten’  (41)  / tjop /  → [ tjopː ]  ‘two’  (42)  / tsɨ(b)-[tjop] / → [ tsɨb.tjopː ]  (43)  *[ tsɨːbtjopː ]  ‘twelve’  The word / tsɨ / ‘ten’ is stressed on its own; however, when it forms the first part of the compound / tsɨ(b)-tjop / ‘twelve’, it is no longer stressed and, therefore, the vowel does not lengthen. This can be shown with compound verbs as well.  (44) (45)  (46) (47)  / ɾ-[ɡwḛ] / → [ ɾɡwḛː ] HAB-speak / ɾ-[ɡwḛ]-[zak] / → [ ɾɡwḛzakː ] HAB-speak-good  ‘speaks (a language)’ ‘speaks (a language) well’  *[ ɾɡwḛːzakː ] *[ ɾɡwḛːzak ]  The vowel of the verb / ɾ-ɡwḛ / ‘speaks (a language)’ is long when it is stressed, but when it forms the compound / ɾ-[ɡwḛ]-[zak] / ‘speaks (a language) well’, the stress is on the last syllable / zak /; thus, the vowel / ḛ / is short. (The vowel in the syllable / zak /  75  is not long because it is followed by a fortis consonant.) Compound verbs are very common in Quiaviní Zapotec (there are many entries in the dictionary of Munro and Lopez (1999)).  (48)  / ɾ-[inj]-[dja̤ɡ] / HAB-to go-ear  → [ɾin.dja̤ːɡ]  ‘hears’  (49)  / ɾ-[za-lˑo̤] / HAB-?-face  → [ ɾza.lo̤ː ]  ‘starts, begins’  Finally, although all native roots seem to be monosyllabic, for several words it is not possible to establish the etymology (indicated by the question marks). These words, whether monomorphemic or not, show prominence in the last syllable.  (50)  / ba-[nˑṵa]̰ / ANIM-?  → [ ba.nṵˑa̰ ]  ‘scorpion’  (51)  / b-[ud]-[ɡeɾ] / ?ANIM-?-?  → [ bud.ɡeːɾ ]  ‘segment, section of fruit’  (52)  / ɡi-[tseinj] / ?skin-?  → [ ɡi.tseˑiɲ ]  ‘cricket’  (53)  / [damɡḛs] /  → [ dam.ɡḛs ]  ‘type of black and white grasshopper with orange spots’  (54)  / [laba̰] /  → [ la.ba̰ː ]  ‘root’  Finally, some loanwords are clear examples of polysyllabic roots with final syllable prominence, as shown in (55). (See loanword phonology below, §3.4, for more details.)  76  (55) Polysyllabic loanwords Spanish Quiaviní Zapotec a. [ naβaxa ] → [ naβaxː ] ‘pocket knife’  (<Sp. navaja)  b. [ konxuan ] → [ kon.xuanː ] ‘musical group’  (< Sp. conjunto)  c. [ koɾason ] → [ ko.ɾa.sonː ] ‘heart’  (< Sp. corazón)  d. [ kanela ]  ‘cinnamon’  (<Sp. canela)  ‘window’  (< Sp. ventana)  → [ kaneːl ]  e. [ bentana ] → [ ben.taːn ]  In addition to prosodic characteristics, cross-linguistically, prosodic heads may display segmental and featural contrasts not found in non-prominent positions. In other words, clusters of information tend to occupy salient positions (see Beckman, 1998; Michael J. Kenstowicz, 1994, 1996a; Paul Valiant de Lacy, 2002; Paul de Lacy, 2006; Zoll, 1998, 2004). This is the case in Quiaviní Zapotec, where particular distributional properties differentiate prominent versus non-prominent syllables, as noted by Munro and Lopez (1999). Many of these properties were described in the phonotactics section of Chapter 1; I repeat them here in the context of Quiaviní Zapotec prominence. Beginning with segmental properties, all consonants may appear in singleton onsets and most of them in singleton codas (as outlined in Chapter 1). More importantly, all licit consonant clusters occur in prominent syllables, and rarely in non-prominent ones. With respect to vowels, all six Quiaviní Zapotec vowel types, / a e o i u ɨ /, may bear stress (i.e. constitute prominent syllables). Diphthongs,43 predominantly, and derived long vowels, exclusively, are found in prominent syllables. One of the over-arching themes of this study is voice quality. In terms of metrical structure, all Quiaviní Zapotec phonation types appear in both unstressed and stressed syllables. However, non-modal vowels are considerably more common in prominent positions (see restrictions and combination forms described in Munro & Lopez, 1999). The interaction between stress and tone shows more distributional evidence for prominence in Quiaviní Zapotec. This language has four tones: two levels (high and low) and two contours (rising and falling). All four tones may appear in prominent syllables,  43  The only diphthongs in unstressed syllables may be found in compounds, with roots that appear as the non-head.  77  whereas only level tones seemed to be found in non-prominent syllables.44 In other words, stress constrains the complexity of tone, restricting contours to stressed syllables. In brief, prominent syllables in Quiaviní Zapotec are characterized by the ability to bear all phonological contrasts. (56) and (57) below summarize the segmental, tonal, and voice quality properties that are restricted, or statistically restricted, to prominent syllables (non-prominent syllables have no exclusive properties), which illustrates the crucial fact that Quiaviní Zapotec restricts a considerable amount of phonological complexity to prominent positions. (56) Exclusive properties of prominent syllables i. (Derived) long vowels ii. Contrastive contour tones (rising and falling) (57) Near exclusive properties of prominent syllables i. Non-modal vowels ii. Diphthongs iii. Consonant clusters (both in onset and coda) Based on the observations above, Quiaviní Zapotec prominence patterns illustrate two cross-linguistic properties of stress: culminativity and demarcativity. The former consists of having a single prosodic peak for a morphological or syntactic constituent (stem, word, phrase). The latter concerns how stress tends to be placed near the edges of constituents. Quiaviní Zapotec shows culminativity in that there is only one prominent syllable per word, and demarcativity in that the (final) root syllable is always prominent.  44  The analysis of tone outside stressed syllables (roots) is still inconclusive, but it seems fairly restricted. Surface contour tones appear to be non-contrastive in unstressed syllables.  78  3.3.1 Foot type: Trochaic rhythm In accounting for the metrical structure of Quiaviní Zapotec, a necessary step is to establish its rhythmic type of feet (iambic or trochaic). As shown above, words may be morphologically complex and present different stress patterns, e.g. S, Sw (with a suffix), wS (with or without a prefix).45 I start this section by presenting examples of root plus suffix in more detail, as these types of words may be the only cases of light stressed syllables. Having established the durational patterns of these words, I consider the different word-stress patterns in Quiaviní Zapotec and evaluate different foot-type possibilities, suggesting a trochaic analysis. As demonstrated in §3.2.1, lenis consonants are always short, independent of their syllabic position. In contrast, the analysis of monosyllables showed that fortis consonants are long (moraic) in coda position. This generalization, however, is different when the diminutive suffix or clitics are added to monosyllabic roots (see morphosyntactic section in Chapter 1): Fortis obstruent codas resyllabify as onsets, surfacing as short segments; on the other hand, fortis sonorant segments still surface as long, ambisyllabic consonants (see formal analysis in §3.3.3). Consider the examples in (58-61), monosyllabic nouns and their affixed forms with the diminutive suffix /-eʔ/.  45  S = stressed syllable; w = weak / unstressed syllable.  79  (58) Fortis obstruent coda a. / tʃat / → [ tʃaµtµ ] b. / tʃat-eʔ /  → [ ˈtʃaµ.teʔµ ]  ‘kiss’  Bimoraic root syllable  ‘little kiss’  Monomoraic root syllable  ‘father’  Bimoraic root syllable  ‘daddy’  Bimoraic root syllable  ‘snake’  Bimoraic root syllable  ‘little snake’  Bimoraic root syllable  ‘mother’ ‘mommy’  Bimoraic root syllable  ʔ  c.  *tʃaµtµe µ  d.  *tʃaµµ.teʔµ  (59) Lenis obstruent coda a. / dad / → [ daµµð ] b. / dad-e / → [ ˈdaːµµ.ðe µ ] ʔ  ʔ  (60) Fortis sonorant coda a. / bḛlˑ / → [ bḛ̀µlµ ]  b. / bḛlˑ-e / → [ ˈbḛ̀µlµe µ ] ʔ  ʔ  (61) Lenis sonorant coda a. / nan / → [ naµµn ] ʔ b. / nan-e / → [ ˈnaµµ.neʔµ ]  Bimoraic root syllable  These duration patterns were initially detected using data from by-ear transcription from 4 different speakers with several words, followed by phonetic measurements. One male speaker, TiuC, produced five words with fortis obstruent in coda position, combined with the same words in their clitisized forms. Each word was recorded three times in isolation in careful speech. Table 19. Vowel and consonant duration (ms): roots and clitisized forms (TiuC) Rhyme type VOfortis plus clitic VOlenis plus clitic VRfortis plus clitic VRlenis plus clitic  V  C 103 98 209 148 91 84 184 133  Prediction 230 VCː 113 VCV 80 VːC 61 VːCV 146 VCː 126 VCːV 72 VːC 54 VːCV  80  From the values above, short vowels average 94 ms and long vowels 168 ms, whereas the mean for short consonants is 75 ms and 167 for long consonants.46 (Similar ratios have been observed in languages with contrastive vowel and consonant length, e.g. Tamil (Maddieson, 1984); Swedish (Thorén, 2005).) In fortis obstruent-final roots, vowels are short with or without a clitic, whereas fortis obstruent consonants are long in coda, but short in the clitisized form. In lenis-final roots (both obstruents and sonorants), vowels are long and consonants short with or without a clitic. Finally, in fortis sonorantfinal roots vowels are short and consonants long regardless of the type of prosodic word. These results support the analysis above (58-61) in terms of vowel and consonant duration for monosyllables and clitisized forms; Sw words may be formed by two light syllables, LL, or heavy and light, HL. Having established the duration of vowels and consonants in monosyllables and clitisized words, let us move on to the discussion of foot type in Quiaviní Zapotec. All words in this language contain only one prominent syllable (culminativity property), and thus I will assume one foot per word. Consider the word types in Thorén (2005), which are by far the most common in Quiaviní Zapotec. (The previous section illustrated actual examples of these types of words.) (62) Word stress patterns and syllable weight type Stress pattern a. S b. wS c. Sw  Syllable weight type H LH HL, LL  A theory of rhythmic units or feet assumes the following universal inventory (McCarthy and Prince 1986, Hayes 1987, 1995, Kager 1993, 1999, p. 147):  46  Results for the same speaker, TiuC, from the fortis/lenis experiment (§3.2.1) show smaller values but similar ratios (the difference possibly derives from the use of carrier phrase and stimuli vs. words produced in isolation, at a lower speech rate).  81  (63) Foot inventory 47 a. Syllabic trochee (quantity-insensitive): b. Moraic trochee (quantity-sensitive): c. Iamb (quantity-sensitive)  (σσ) (LL) (H) (LL) (H) (LH)  In light of this foot inventory, there are three possible analyses to account for rhythmic type of feet in Quiaviní Zapotec: monosyllabic feet (heavy syllables, regardless of the specific rhythm type), iambs or trochees. Monosyllabic feet, where only the stressed syllable (root) is part of the foot (regardless of the word type), accounts for all the data except for Sw words with light stressed syllable (e.g. [ ˈtʃaµ.teʔµ ] ‘little kiss’). These words would be parsed as (L)L and they would not satisfy Quiaviní Zapotec minimality (FT-BIN), crucial in accounting for monosyllables and the vowel lengthening pattern. In addition, this account would leave us with a considerable number of unparsed syllables. The iambic analysis accounts for all wS type words, which surface as LH syllables, forming the cross-linguistic preferred iamb: (LH) (see Kager, 1993; Bruce Hayes, 1995 among others). Lengthening, commonly found in iambic systems (iambic/trochaic law, Bolton, 1894; Hayes, 1995; cf. Kager, 1993), might also support this analysis. In addition to length, the segmental complexity described in final syllables of non-compound uninflected native words (Munro & Lopez, 1999, p. 3) suggests that wS is basic, leading most directly to the idea that the basic foot is iambic. However, the iambic rhythm fails to account for words of the syllable weight type LL. A degenerate iamb like (L)L violates minimality, and parsing these words as (LL) would imply the clitic is stressed, which is clearly not the case. (Other problems arise with the formal analysis of iambs as we will see below.) Finally, the trochaic rhythm as moraic trochees accounts for monosyllables and wS words, leaving the initial unstressed syllable unparsed; it is the most fitting analysis for Sw words, parsing them either as moraic trochees, (H)L and (LL), or as syllabic ones (HL). The most crucial data then, are cases of Sw words with LL syllables, for which both the monosyllabic feet and iambic approaches are inadequate. 47  A trochaic rhythm entails left foot prominence (either syllabic or moraic), whereas an iambic rhythm demands right foot prominence (always moraic). In this notation, ‘L’ stands for a light syllable (one mora) and ‘H’ for a heavy syllable (two moras). The head of the foot, the stressed syllable, is marked in boldface.  82  Additional internal evidence supports a trochaic rhythm for Quiaviní Zapotec. In the previous section I showed that stressed syllables in Quiaviní Zapotec (roots) display more segmental and featural contrasts than unstressed syllables. We also find differences comparing final unstressed syllables (diminutive suffix and clitics) with initial unstressed syllables. The former bears tone (probably restricted to level tones) and all non-modal phonations, whereas the latter may not be specified for tone,48 and shows extremely reduced contrasts with respect to non-modal phonation. Caldecott (2009) shows phonological and phonetic differences between parsed versus unparsed syllables in St’át’imcets (Lillooet Salish), with the former being longer and with higher pitch values, as well as having phonological properties absent in unparsed syllables (e.g. glottalization). These differences parallel those found in Quiaviní Zapotec, in favor of trochaic foot parsing. An interesting issue is that of the acquisition of foot type in Quiaviní Zapotec. If the child is faced with data where there are not overwhelming reasons for choosing one analysis over another, which does the child identify as the correct pattern? In response to this question Stemberger and Lee (2007) and Stemberger, Chávez-Peón and Lee (2008) show that Quiaviní Zapotec children acquire Sw outputs before wS outputs. The high frequency of the diminutive and pronominal clitics, along with the arguments above seem to facilitate the child choosing a trochaic pattern over an iambic one. In summary, three arguments suggest a trochaic analysis in Quiaviní Zapotec: First, the ability to account for all types of words, particularly Sw words with stressed light syllable (LL); second, the phonological properties carried by the diminutive and clitics over initial unstressed syllables (tone and phonation contrasts); and third, acquisition data where Sw is favored over wS. The following two subsections account formally for the prominence patterns outlined in this and the previous section.  48  From my preliminary observations, it seems that initial unstressed syllables in dysillabic roots and prefixes have a phonetic mid tone. In terms of acquisition of Quiaviní Zapotec, J. Stemberger (personal communication, March 15, 2010) has observed that the pitch of these syllables is highly variable. The status of tone outside the root requires further investigation.  83  3.3.2 Prefixes, compounds and complex roots: Foot alignment The previous section presented the morphology of Quiaviní Zapotec in relationship to the prominence pattern. This subsection analyzes all root-final words, including prefixed roots, compounds and disyllabic roots, leaving the formal account of the diminutive and suffixes for the next section. The goal is to formalize the stress pattern in this language, integrating the proposed Quiaviní Zapotec foot structure. Disyllabic words present additional problems for the formal analysis of section 3.2.3. The constraints discussed so far are not sufficient to lead to the observed pronunciation. Consider the tableaus (64) and (65), showing a prefixed root and a compound, respectively. (64) Prefix + root: /ba-ɡidj/ → [ baɡiːdj ] ‘butterfly’ FT-BIN *Lenis WBYP DEP-µ / baµ-[ɡiµdj] / -µ * a.  (baµˈɡiµdj) b.  (ˈbaµɡiµdj)  *  c.  baµ(ˈɡiµµdj)  *  *!  d. (baµˈɡiµµdj)  *  *!  e. baµ(ˈɡiµdjµ) f. baµ(ˈɡiµdj)  *! *!  * *  (65) Compound: / ɾ-ɡwḛ-zak / → [ ɾɡwḛ.zakː ] ‘speaks (a language) well’ FT-BIN *Lenis WBYP DEP-µ /ɾ-[ɡwḛµ]-[zaµk] / -µ *! a. (ɾɡwḛµzaµk) b.  ɾɡwḛµ(zaµkµ)  *  c.  (ɾɡwḛµzaµkµ)  *  d.  (ɾɡwḛµzaµkµ)  *  e. ɾɡwḛµ(zaµµk)  *!  *  Minimality is no longer an issue for this type of word; except for (64f), all candidates satisfy FT-BIN under moraic or syllabic analysis. The problem is the location of stress, and the syllable weight of the stressed syllable. This constraint-based grammar is  84  insufficient to decide whether stress is word-initial or located in the root-final syllable (64b & 65d). In addition, no constraint regulates the parsing into feet (64a&b vs. the actual output 64c; and 65b vs. 65c&d). In short, this account is not enough to explain the prominence pattern of Quiaviní Zapotec. (From these examples, however, we do see now that WBYP >> DEP-µ (65a vs. 65b)). This ranking will be equally insufficient for inflected verbs (e.g. / ka-ʐṵnˑj / → [ka.ʐṵɲː] ‘(someone) is running’) or disyllabic or longer roots (e.g. / ɡji-tseinj /→ [ ɡji.tseˑiɲ ] ‘cricket’). In all these cases, the final syllable is heavy. The Stress-to-Weight Principle (SWP), which states that every stressed syllable is heavy, seems a likely way to account for Quiaviní Zapotec prominence pattern. However, the previous section showed that root-final fortis obstruents surface with a light prominent syllable once a suffix or clitic is added (see formal analysis in the next section). (66)  / lat-eʔ / → [ ˈlaµ.teʔµ ] can-DIM  ‘little can’  Since minimality is not enough and SWP by itself cannot account for morphologically complex words in Quiaviní Zapotec, the explanation seems to need to rely on the alignment of the root and stress. In metrical structure, this can be attained by means of alignment of the head of the foot with the right edge of the root. Following the format of ‘Generalized Alignment’ (McCarthy & Prince, 1993), the constraint is formulated in (67). (I will show that this constraint is also adequate for capturing loanword phonology in §3.5.) (67) ALIGN (Hd(Ft), R, Root, R) (ALIGN-R) ‘For every stressed syllable (= head of a foot) there must be some root such that the right edge of that syllable matches the right edge of the root’ The “head of the foot” automatically means the stressed syllable (not the stressed mora), because syllables are the next level down from feet in the prosodic hierarchy. As we will see in the next section, to refer to the head of the foot instead of the foot itself (i.e. the right edge of the foot) is crucial in accounting for cases like (66) above, where the foot  85  includes the suffix: (ˈlaµ.teʔµ) ‘little can’, as shown in the previous section. The undominated constraint ALIGN-R will eliminate candidates with initial stress (64b & 65d).49 In arguing that OT constraints are categorical, McCarthy (2003) proposes different ways of assessing alignment constraints, such as Align-by-segment, Align-bysyllable, Align-by-foot, among others (see also Horwood, 2008, p. 8). Along these lines, the alignment constraint proposed in (67) evaluates candidates in the form of Align-bysyllable, that is, the head of the foot must be the rightmost syllable of the root, and not necessarily the precise segmental edge of the root. This is important as some root-final coda consonants resyllabify when a clitic is added (e.g. /dad/ → [daːθ], but /dad+eʔ/ → [daː.ðeʔ]); in these cases the consonant would not be part of the stressed syllable, but no violation of ALIGN-R would be incurred as the rightmost syllable of the root is stressed (these cases are evaluated in the next section). This subtle issue of determining prosodic boundaries is not exclusive to Quiaviní Zapotec, but common cross-linguistically, particularly in languages where a morphological domain (e.g. the root) is prosodically salient. In addition to the alignment of stress, the previous section established trochaic rhythm as the most appropriate for Quiaviní Zapotec, formalized in (68). (68) RHTYPE=T (TROCHEE) Feet have initial prominence  (Kager, 1999, p. 172)  As a result of the undominated constraints ALIGN-R and TROCHEE some syllables of the output candidates will be left unparsed, violating the low ranked constraint PARSE-σ. (69) PARSE-σ Every syllable must belong to some foot. (No syllable may be left unparsed)  49  With respect to example (65d), although both syllables are roots the Align-R constraint refers to the first root in the derivation.  86  With these revisions, the summary tableaus below demonstrate how the interaction of ALIGN-R and the foot type constraint TROCHEE correctly derives the Quiaviní Zapotec prominence pattern. (70) Complex words: lenis coda (ALIGN-R & TROCHEE) FT-BIN ALIGN-R TROCHEE *Lenis-µ WBYP DEP-µ PARSE-σ /baµ-[ɡiµdj] / a. (baµˈɡiµdj)  *!  b. (ˈbaµɡiµdj)  *  *!  *  c. baµ(ˈɡiµµdj) d. (baµˈɡiµµdj)  *!  e. baµ(ˈɡiµdjµ) f. baµ(ˈɡiµdj)  *  *  *  *  *! *!  * *  * * *  (71) Complex words: fortis coda (ALIGN-R & TROCHEE) /ɾ-[ɡwḛµ]-[zaµk] / FT-BIN ALIGN-R TROCHEE *Lenis WBYP DEP-µ PARSE-σ -µ *! * a. (ɾɡwḛµzaµk) b. ɾɡwḛµ(zaµkµ)  *  c. (ɾɡwḛµzaµkµ) d. (ɾɡwḛµzaµk)  *! *!  e. ɾɡwḛµ(zaµµk)  *  * * *!  *  *  This analysis adequately explains all words where the root is word-final. I turn now to the analysis of words with final unstressed syllables.  3.3.3 Diminutive suffix and clitics: Faithfulness to the base This section continues the morphological analysis of prominence in Quiaviní Zapotec, focusing on the diminutive suffix and clitics. The goal is to account for the additional prosodic phenomena found with elements within the constraint-based Quiaviní Zapotec grammar.  87  Section 3.3.1 illustrated the special behavior of root final fortis segments once the diminutive suffix or a clitic is added. Fortis obstruent codas resyllabify as onsets, surfacing as short segments, whereas fortis sonorant segments still surface as long, ambisyllabic consonants. In contrast, roots with lenis consonants in the coda, both obstruents and sonorants, always surface as bimoraic in their suffixed or clitisized forms. The examples below, repeated from §3.3.1, illustrate these patterns. (72) Fortis obstruent coda a. / tʃat / → [ tʃaµtµ ] ʔ  b. / tʃat-e /  ‘kiss’  → [ ˈtʃaµ.te µ ] ʔ  (73) Fortis sonorant coda a. / bḛlˑ / → [ bḛ̀µlµ ]  ‘snake’  b. / bḛlˑ-e / → [ ˈbḛ̀µlµe µ ] ʔ  ʔ  (74) Lenis obstruent/sonorant coda a. / dad / → [ daµµð ] b. / dad-e / → [ ˈdaːµµ.ðe µ ] c. / nan / → [ naµµn ] ʔ d. / nan-e / → [ ˈnaµµ.neʔµ ] ʔ  ‘little kiss’  ʔ  ‘little snake’  ‘father’ ‘daddy’ ‘mother’ ‘mommy’  Under this analysis, the generalization is that all the roots maintain their bimoraicity in their suffixed form, except roots with fortis obstruent in coda. Considering the formal account of the previous section, the differences between the base and the suffixed or clitized forms are not reflected in the current ranking, illustrated below. (Since parsing is not decisive in selecting the optimal output, candidates are left unparsed. I will come back to this issue at the end of the section.) (75) Root with fortis obstruent coda + suffix (or clitic) FT-BIN ALIGN-R TROCHEE *Lenis-µ WBYP DEP-µ PARSE-σ / [tʃaµt]-eʔµ/ a.  (tʃaµ.teʔµ) b. (tʃaµtµeʔµ)  *!  c. (tʃaµµ.teʔµ)  *!  (76) Root with fortis sonorant coda + suffix (or clitic) 88  /[bḛµlˑ]-eʔµ/  FT-BIN ALIGN-R TROCHEE *Lenis-µ WBYP DEP-µ PARSE-σ  a.  (bḛµleʔµ) b.  (bḛµlµeʔµ)  *!  c. (bḛµµleʔµ)  *!  (77) Root with lenis sonorant (or obstruent) coda + suffix (or clitic) FT-BIN ALIGN-R TROCHEE *Lenis-µ WBYP DEP-µ PARSE-σ /[naµn]-eʔµ / a.  (naµ.neʔµ) *!  b. (naµ.nµeʔµ)  * *!  c.  (naµµ.neʔµ)  The formal analysis of this behavior does not rely on purely phonological facts; it is necessary to refer to prosodic morphology. One possible explanation relies on the moraic correspondence between the stand-alone root form (“the base”) and its derived forms (“affixed form”), in combination with the likelihood that segments bear syllable weight. This is a case of paradigm uniformity (Kurylowicz, 1945), formally treated as an Outputto-Output (OO) correspondence within OT (Benua, 1995; Kenstowicz, 1996). The notion of OO-correspondence corresponds to the maximization of phonological identity between morphologically related output forms, as portrayed in the following diagram (Benua, 1995; Kager, 1999, pp. 263, 275). (78) Basic Model of stem-based affixation BA-Identity IO Faithfulness  Base ⇔ Affixed form ⇕ Input  The base is a freestanding output form of the language, compositionally related to its derived counterpart (the affixed form). That is, “the base contains a proper subset of the grammatical (semantic, morphological) features of the derived form.” (Kager, 1999, p. 281). (On the extension of Correspondence Theory (McCarthy & Prince, 1995) to relations between surface forms within a paradigm: the Base (B) and the Affixed form (A), see Benua (1995) and Urbanczyk (1996).)  89  As a starting point, I adopt the analysis of Arellanes (2009) accounting for similar data in San Pablo Güilá Zapotec. He proposes an interaction between the universal hierarchy of moraic elements in (79) and the OO-correspondence constraint in (80), which forces the Base-Affixed form correspondents to have the same moraic content. (79) Universal hierarchy of moraic elements (cf. Morén 1997, 2003) *µ/O >> *µ/R >> *µ/V This universal hierarchy penalizes moraic segments based on their sonority, preferring moraic vowels over moraic sonorants, and moraic sonorants over moraic obstruents. (80) MAX-µ-BA50 (Arellanes, 2009, p. 365) preliminary ‘Every mora in the base (B) has a correspondent in the affixed form (A)’ For this constraint, the base in Quiaviní Zapotec would be the unsuffixed or unclitisized content words (in the case of verbs, it implies the presence of an aspectual prefix, see §1.4.5). The crucial ranking for these constraints is in (81). (81) MAX-µ-BA and the moraic hierarchy ranking *µ/O >> MAX-µ-BA >> *µ/R >> *µ/V Within the global current ranking, the ranking in (81) is located between the contraints WBYP and DEP-µ. On the one hand, moraic faithfulness to the base (MAX-µ-BA) outranks the penalty against inserting moras (DEP-µ); on the other hand, the moraic status of fortis obstruents in coda position (e.g. as in monosyllables like [ zaµkµ ] ‘good’) implies that WBYP >> *µ/O.  50  Arellanes (2009: 365): MAX-µ-BCLI ‘Las moras de una base (B) tienen un correspondiente en su forma clitizada (CLI)’  90  (82) Root with fortis coda obstruent + suffix (MAX-µ-BA) /[tʃaµt] -eʔµ/ Base: tʃaµtµ a.  (ˈtʃaµ.teʔµ) b. (ˈtʃaµtµeʔµ)  F TBIN  ALIGN TRO -R CHEE  *Lenis -µ  WBYP  *µ/O  MAX -µBA  *µ/R *µ/V  *  DEP -µ  PARSE -σ  **  *!  **  *  (83) Root fortis coda sonorant + suffix (MAX-µ-BA) /[bḛµlˑ] -eʔµ/ Base: (bḛµlµ) a. (ˈbḛ̀µ.leʔµ) b.  (ˈbḛ̀µlµeʔµ)  F TBIN  ALIGN TRO -R CHEE  *Lenis -µ  WBYP  *µ/O  MAX -µBA  *µ/R *µ/V  *!  DEP -µ  PARSE -σ  ** *  **  *  (84) Root lenis coda (same for obstruents & sonorants) + suffix (MAX-µ-BA) /[daµd] -eʔµ/ Base: daµµd a.(ˈdaµ.ðeʔµ) b. (ˈdaµµ.ðeʔµ)  F TBIN  ALIGN -R  TRO CHEE  *Lenis -µ  WBYP  *µ/O  MAX -µBA *!  *µ/R *µ/V  DEP -µ  PARSE -σ  ** ***  *  This correctly accounts for the length of vowels and sonorants, which is not due to minimality anymore, but to the base correspondence in the affixed form. However, there is another candidate we must consider within suffixed or clitized forms with root-final fortis obstruents: a candidate with a long vowel. The importance of this candidate derives from the significant preference for heavy syllables in Quiaviní Zapotec prominent positions.  91  (85) Root with fortis coda obstruent + suffix (MAX-µ-BA) /[tʃaµt]-eʔµ/ Base:_tʃaµtµ a. (ˈtʃaµ.teʔµ) b. (ˈtʃaµtµeʔµ) c.  (ˈtʃaµµ.teʔµ)  F TBIN  ALIGN TRO -R CHEE  *Lenis  WBYP  *µ/O  -µ  MAXµ-BA  *µ/R *µ/V  *!  DEPµ  PARSE -σ  **  *  *  ***  *  *  **  *!  Candidate c. is faithful to the moras of the base (although now both moras are with the vowel) and follows the tendency of prominent syllables to be heavy. Nonetheless, it is the incorrect output. In order to account for this fact, the moraic faithfulness to the base must be encoded as IDENTITY instead of MAXIMALITY, as formalized in (86). (86) WEIGHT-IDENT-BA (WT-IDENT-BA) (Kager, 1999, pp. 269, 271; Benua, 1995) ‘Base-Affixed form correspondent segments have the same moraic content.’ This constraint establishes correspondence relations between Base and the Affixed form with regard to the moraic content associated with segments. Based on this revision, I consider more candidates in the tableaus below (particularly in terms of foot possibilities), conclusive for the suffix analysis. (87) Root with fortis coda obstruent + suffix (WT-IDENT-BA) /[tʃaµt]-eʔµ/ Base: tʃaµtːµ  F TBIN  ALIGN -R  TRO CHEE  *Lenis -µ  WBYP  *µ/O  a.  (ˈtʃaµ.teʔµ) *!  b.(ˈtʃaµtːµ)eʔµ  *  c.(ˈtʃaµµ).teʔµ d.(ˈtʃaµ).teʔµ  W TIDENT -BA *  *!  *µ/R *µ/V  DEP -µ  PARSE -σ  **  *  *  ***!  *  *  **  **  e.(ˈtʃaµtµeʔµ)  *!  f.(ˈtʃaµµ)tµeʔµ  *!  *  *  **  *  ***  **  *  The moraicity of fortis obstruents in coda position responds to minimality and the constraint WBYP, which outranks *µ/O. In turn, *µ/O outranks WT-IDENT-BA, as illustrated in (87), thus the moraicity of the obstruent does not carry over to the suffixed 92  form. In contrast, WT-IDENT-BA outranks *µ/R and *µ/V, accounting for the paradigm uniformity between the base and the affixed form for roots with lenis and fortis sonorant codas. (88) Root fortis coda sonorant + suffix (WT-IDENT-BA) /[bḛµlˑ]-eʔµ/ ‘snake’ Base: bḛµlµ  F TBIN  ALIGN -R  TRO CHEE  *Lenis -µ  WBYP  *µ/O  W TIDENT -BA  DEP -µ  PARSE -σ  **  *  *!  *!  ***  *  *  *  **  *!  a.(ˈbḛµ.leʔµ)  ** *  b.(ˈbḛµlµ)eʔµ c.(ˈbḛµµ).leʔµ d.(ˈbḛµ).leʔµ e.  (ˈbḛµlµeʔµ)  *µ/R *µ/V  *!  *!  f.(ˈbḛµµ)lµeʔµ  *  *  **  *  *  ***  **  *  (89) Root lenis coda (same for obstruents & sonorants) + suffix (WT-IDENT-BA) /[daµd]-eʔµ/ Base: daµµd  F TBIN  ALIGN -R  TRO CHEE  *Lenis -µ  WBYP  *µ/O  a.(ˈdaµ.ðeʔµ) *!  b.(ˈdaµðµ)eʔµ  *  W TIDENT -BA *! *  c.(ˈdaµµ).ðeʔµ d.  (ˈdaµµ.ðeʔµ) e.(ˈdaµ).ðeʔµ  *!  f.(ˈdaµðµeʔµ)  *!  *  g.(ˈdaµµ)ðµeʔµ  *!  *  *µ/R  *µ/V  DEP -µ  PARSE -σ  **  *  *  ***  *  *!  ***  *  **  *  **  *  *  **  *  ***  **  *  The final issue worth noting is parsing in relation to the trochaic rhythm, exemplifying the emergence of the unmarked. The issue is relevant for root-final fortis sonorant and root-final lenis obstruent suffixed forms. Compare, in particular, candidates (88b) (ˈbḛµlµ)eʔµ, a moraic trochee with the final syllable unparsed, vs. (88e) (ˈbḛµlµeʔµ), an uneven syllabic trochee (HL). The low ranked constraint PARSE-σ becomes visibly active, favoring the optimal candidate (88e) over (88b). The syllabic trochee (88e), unmarked with respect to PARSE-σ, emerges as optimal, even though the presence of  93  PARSE-σ in the grammar is generally hidden. The same condition is observed in (89), where candidate (89d) wins over (89c). This shows that this constraint-based grammar favors parsing over the presence of uneven trochees (cf. GROUPING HARMONY, Elias Ulloa, p. 85; Kager, 1993; Hayes, 1995). To sum up, this section added suffixes to the prominence analysis of Quiaviní Zapotec, demonstrating the correspondence between the base and its affixed form, where a division among segments and their likehood to be moraic is found.  3.4  Loanword phonology  In adapting a non-native word, the challenge for a speaker is to try to be faithful to the source while obeying her/his own language-specific restrictions. Several conflicts may emerge in this process due to the segmental inventory, phonotactics, prosodic domains, and so forth. Quiaviní Zapotec has been in continuous contact with Spanish for over 400 years; as a result, the language has borrowed heavily from Spanish. These loanwords provide valuable evidence with respect to Quiaviní Zapotec prosodic prominence. As such, the goal of this section is to apply the prosodic and formal analysis of native words to loanword phonology. The examples and description are based on Munro and Lopez (1999), Munro et al. (2008) and Chávez-Peón (2006). (See also Stemberger & Lee, 2008, with respect to the acquisition of loanwords.)  94  (90) Spanish loanwords Spanish a. [ lata ] b. [ beto ]  →  Quiaviní Zapotec [ latː ] ‘tin can’  (< Sp. lata)  →  [ betː ]  (< Sp. Beto < Alberto)  ‘Alberto’  c. [ koɾason ] →  [ ko.ɾa.sonː ] ‘heart’  (< Sp. corazón)  d. [ lado ]  →  [ laːd ]  ‘side’  (< Sp. lado)  e. [ pedɾo ]  →  [ beːd ]  ‘Pedro’  (< Sp. Pedro)  [ ben.taːn ]  ‘window’  (< Sp. ventana)  f. [ bentana ] →  The borrowing process, exemplified with the words above, has the following characteristics (first described by Munro & Lopez, 1999): (91)  Loanword adaptation  a. Unstressed Spanish final vowels in open syllables are consistently deleted. b. Stressed Spanish vowels are always maintained and retain their quality. c. Stressed syllables of Spanish words are borrowed into Zapotec as the prominent syllable of the word. These generalizations are observed in the examples above and apply to all loanwords in Quiaviní Zapotec without exception. The fact that unstressed final vowels in open syllables are routinely dropped follows the prominence pattern of Quiaviní Zapotec, as the prosodic head of a word, that is, the prominent or stressed syllable, must be the last one within the root. In the previous sections about root prominence, this pattern was attained by the foot-root alignment constraint ALIGN-R. This constraint, in combination with the trochaic rhythm (RHTYPE=TROCHEE), is essential in the analysis of loanwords. See (94) and (95) below. With respect to the Zapotec segmental assimilation of the consonants in loanword phonology, Pamela Munro (personal communication, March 2005) notices that lenis coda consonants are preceded by prominent (stressed) long vowels, whereas fortis consonants by prominent short vowels.51 More examples of this pattern are provided below and its prosodic relevance has already been discussed in the section on moraicity and 51  Recall from §2.2 that in Munro and Lopez (1999), these short vowels are analyzed as checked vowels, but reanalyzed here as modal short vowels instead.  95  minimality. In order to form bimoraic feet, fortis consonants are moraic in coda and contribute to syllable weight, whereas lenis consonants are not moraic and, thus the vowel lengthens to become bimoraic. (92) and (93) illustrate these patterns. (92) Short V + fortis C Spanish a. [ bloke ]  →  Quiaviní Zapotec [ blokː ] ‘cement block’  (<Sp. bloque)  →  [ alˑt ]  (<Sp. alto)  Spanish a. [ xuɡo ]  →  Quiaviní Zapotec [ xuːɣ̊ ] ‘juice’  (<Sp. jugo)  b. [ kanela ]  →  [ kaneːl ]  (<Sp. canela)  b. [ alto ]  ‘tall’  (93) Long V + lenis C  ‘cinnamon’  The adaptation of Spanish obstruents is based on voicing: Spanish voiceless obstruents are adapted as fortis consonants, whereas voiced obstruents are adapted as lenis consonants. The adaptation of sonorants into the fortis or lenis classes is less clear, since there is no “preliminary” distinction in Spanish among sonorants. The adaptation seems to rely more heavily on Spanish phonetic vowel duration (see Chávez-Peón 2006 for more details). The following tableaus show the formal analysis of loanwords.  96  (94) Loanwords (polysyllabic): fortis consonants52 /maµ'tɾaµkaµ/53 'bull roarer'  F TBIN  a. maµ('tɾaµkaµ) b. maµ(tɾaµ'kaµ) c. maµ('tɾaµk)  TRO CHEE  ALIGN -R  *Lenis  WBYP  *µ/O  *µ/R  -µ  *! *! *!  d. (maµ'tɾaµk)  ***  *  **  *!  *  **  *  ** *  **  *!  g. maµ('tɾaµµk) *!  PARSE -σ *  * *!  DEP -µ  ***  *!  e. ('maµtɾaµk) f.  maµ('tɾaµkµ) h. (maµ'tɾaµkµ)  *µ/V  *** *  **  (95) Loanwords (polysyllabic): lenis consonants /beµn'taµnaµ/ ‘window’ a.beµn(ˈtaµnaµ) b.beµn(taµˈnaµ) c. beµn(ˈtaµn) d. (beµnˈtaµn) e. (ˈbeµntaµn) f. beµn(ˈtaµnµ) g.  beµn(ˈtaµµn) h. (beµnˈtaµµn)  F TBIN  TRO CHEE  ALIGN *Lenis -R -µ *!  *! *! *! *! *!  *!  *!  WBYP * * ** ** ** * ** **  *µ/O  *µ/R  *  *µ/V *** *** ** ** ** ** ***  DEP -µ  PARSE -σ * *  ***  Quiaviní Zapotec preserves the original stressed Spanish vowel as the prominent syllable and deletes any potential syllabic nucleus that follows (but see below). This deletion, however, applies only in final open syllables. If the final unstressed vowel is in  52  I leave out from these tableaus the constraint WT-IDENT-BA, since it is vacuously satisfied for all loanwords as they are analyzed as an Input-Output correspondence, not an Output-Output one (i.e. there is no Base or Affixed form to evaluate). 53 Presumably the UR is not synchronically the Spanish form, but the aim here is to show the process for the first step in adapting a loanword. Diachronically one could think of the Spanish output form as the input for Zapotec speakers, this input is evaluated by the grammar and an optimal candidate surfaces. This new “incorporated” output could presumably become the new stored input for this word. The details of this implementation, however, fall beyond the scope of this study.  97  a closed syllable (as is the case for a minority of words in Spanish), the vowel and the coda are maintained (96a). When the Spanish word has antepenultimate stress, the penultimate unstressed vowel is also maintained (96b). (96) Loanwords: Non-final prominent roots Spanish a. [ fasil ]  b. [ baskula ]  → →  Quiaviní Zapotec [ fasi̤lj ] ‘easy’ [ baskwalː ]  ‘scale’  (<Sp. fácil) (<Sp. báscula)  These types of words show that it is more important to be faithful to the original prosodic head than to shift the stress to the final syllable (i.e. Quiaviní Zapotec grammar is faithful to the original prosodic head of the Spanish word). Faithfulness to the location of stress between one string and another (be it input-output or, output-output) can be obtained via IDENT-HEAD, as defined below.54 (97) IDENT-HEAD (Plag, 1998, p. 203) The prosodic head of the input is the prosodic head of the output (= no stress shift). As a consequence, the type of words in (96) is the only instance that violate the alignment of the head of the foot with the right edge of the root, thus the small change in the ranking IDENT-HEAD >> ALIGN-R. Moreover, Quiaviní Zapotec grammar shows that it is more important to preserve consonants than vowels. While final unstressed vowels are always deleted in open syllables, consonants in unstressed (final and penultimate) syllables are preserved (see examples in 96). Formally, this consonant-retention is obtained by ranking MAX-C over MAX-V.55 (98) MAX-C Input consonants must have output correspondents (‘No consonant deletion’) 54  Variations of this constraint include, e.g. MAX/IDENT-Stress (M. Kenstowicz, 2007). On the relative importance of faithfulness to C versus V in loanwords, this is true in other cases also, such as Cantonese (Yip, 2006), where MIMIC-TONE(STRESS), MIMIC-CONS >> MIMIC-VOWEL >> MIMIC-LENGTH. 55  98  (99) MAX-V Input vowels must have output correspondents (‘No vowel deletion’) (100) Non-final prominent loanwords (IDENT-HEAD, MAX-C >> ALIGN-R >> MAX-V) F TBIN  /faµsiµl/ ‘easy’  TRO CHEE  IDENTHEAD  MAX -C  *Le nis  ALIGN -R  -µ a.  (faµsiµl) *!  BY  W TIDENT -BA  *µ/R  DEP -µ  PAR SE  MAX -V  -σ **  *  *** *  *  *µ/V  *  *!  d.(faµµ)siµl  *µ/O  P *  b.faµ(siµµl) c. (faµsµ)  W  *  *  *  *  ***!  * *  *  /baµs kuµlˑaµ/ ‘scale’ a.  (baµsµ) (kwaµlµ) b. (baµsµ) (kwaµlµ) c. (baµsµ)  *  *! *!*  *  *  **  *  *  *  **  *  *  **  *  In turn, the faithfulness to the input’s prosodic head (IDENT-HEAD) also rejects the possibility of shifting stress in loanwords, for instance, a hypothetical output /ben'tana/ → [benta'naː] ‘window’. Such candidates would satisfy both ALIGN-R and TROCHEE, but would violate IDENT-HEAD. Finally, the last descriptive fact that impacts the theoretical analysis involves loanwords with a complex coda formed by lenis consonants. As the examples below illustrate, vowels are short in these words. (101) Loanwords: lenis complex coda Spanish a. ['kable] b. ['kwadɾa]  → →  Quiaviní Zapotec [ kabl ] ‘insulated wire’ [ kwadɾ ] ‘(city) block’  (< Sp. cable) (< Sp. cuadra)  c. ['sjempɾe]  →  [ sjemɾ]  (< Sp. siempre)  ‘always’  99  I have proposed (§3.3) that the minimal word in this language is a bimoraic foot, and that lenis consonants are not moraic. Accordingly, words with complex lenis codas seem monomoraic at first glance. Nonetheless, following Arellanes (2009), a possible analysis is that although a lenis consonant cannot be moraic on its own, it can share a mora; therefore, in order to satisfy minimality lenis coda clusters contribute a mora. Crosslinguistically, it is common to ban long vowels before coda clusters (e.g. in Scandinavian languages, Kristoffersen, 2000). This is also the case in Quiaviní Zapotec. The moraic representation is below. (102) Moraic representation of Quiaviní Zapotec words a. Foot | σ  k  µ  µ  a  b  l  ‘wire’  This explanation seems a more adequate solution than to assume that these words are sub-minimal prosodic words in Quiaviní Zapotec, and correlates with the duration of these  segments.56  Formally,  then,  we  need  a  slight  rectification  on  the  *Lenis−µ constraint. As originally proposed in Arellanes (2009, p. 348), the word “autonomously” reflects the fact that a single lenis consonant cannot be moraic on its own, but as a cluster it can share a mora. (103) *Lenis−µ If lenis then non-autonomously moraic  (adapted from Arellanes, 2009, p. 348)  This final modification neither affects the analyses of previous cases nor changes the proposed constraint ranking. 56  Vowels are short followed by coda clusters; in turn, each of these lenis consonants is short.  100  3.5  Summary and conclusions  This Chapter analyzed the metrical constituency of Quiaviní Zapotec in terms of the prosodic hierarchy (PrWd – Foot – Syll – Mora), accounting for the prominence pattern in this language. I have argued that the minimal prosodic word consists of a bimoraic foot. In monosyllables, this is satisfied in one of two ways. First, if the syllable is open, or closed by a lenis consonant, the vowel is lengthened, and becomes bimoraic. Second, if the coda consonant is fortis, it contributes a mora. Valley Zapotec is unique in that both fortis coda sonorants and obstruents are moraic. This claim was tested acoustically in a production study with significant results that clearly suggest that differences between lenis and fortis consonants in codas reflect prosodic contrasts in terms of moraicity, thus enriching the typology of syllable weight. In disyllabic and longer words, Quiaviní Zapotec displays a trochaic metrical pattern at the moraic and syllabic level. Further, in accordance with Munro and Lopez (1999), stress is demarcative, with the root-final syllable consistently carrying prominence.57 (104) Metrical properties of Quiaviní Zapotec a. Culminative b. Demarcative c. Rhythmic d. Quantity-sensitive  one prominent syllable per word root-final syllables are prominent trochaic (moraic) trochees (LL) (H)  The metrical structure of Quiaviní Zapotec presented in this chapter was particularly illustrated with items with modal voice (mostly with high tone). Nonetheless, the principles outlined here hold for all the phonation types and tones in the language. This chapter sets the basis for the prominence pattern in Quiaviní Zapotec and will be taken as foundational to understand subsequent phonological patterns in the language: tone and phonation type.  57  Only a few exceptions (<10 dictionary entries) are found in loanwords.  101  Chapter 4:  Tone in Quiaviní Zapotec  4.1  Introduction Tone (the use of pitch to distinguish lexical or grammatical meaning) occurs in  many languages in of the world; according to Yip (2002, p. 1), 60~70 percent of the world’s languages are tonal. Hyman (2006, p. 229) defines a tonal language as follows: “A language with tone is one in which an indication of pitch enters into the lexical realisation of at least some morphemes.” With respect to Otomanguean languages, which include Amuzgo, Chatino, Chinantec, Mazatec, Mixtec, Zapotec, among other linguistic groups,58 the contrastive use of tone is so consistent that it has been considered to be a genetic feature (Rensch, 1976; Suárez, 1983). Nonetheless, the phonological patterns and tonal inventories are very diverse across languages in the family. Also within the Otomanguean stock, practically all Zapotec languages have been analyzed as tonal.59 Valley Zapotec variants spoken in communities neighboring on San Lucas Quiaviní, such as Santa del Valle Zapotec (Rojas, 2010), San Pablo Güilá Zapotec (López Cruz, 1997, Arellanes, 2003), and San Juan Guelavía Zapotec (Jones & Knudson, 58  See INALI’s catalogue (National Institute of Indigenous Languages in Mexico): http://www.inali.gob.mx/catalogo2007/mapa.html#5. 59 According to Jaeger and Van Valin (1982, p. 127) “all Zapotecan languages are tone languages”.  102  1977), have been reported to use tone contrastively.60 In Quiaviní Zapotec, Munro and Lopez (1999) recognize four different tone melodies, including two level tones, high and low, and two contour tones, rising and falling. Based on these facts, we would expect tone to be contrastive in Quiaviní Zapotec. However, in the most complete work to date on this language, Munro and Lopez (1999) make the controversial claim that tone is predictable from phonation types. They state that “tone melodies on Quiaviní Zapotec vowel complexes [syllable nuclei] are derived from the number and phonation type of the vowels in the complex and its phonological environment rather than representing primary contrasts” (Munro & Lopez, 1999, p. 3). Their proposal implies that for there to be a pitch difference, there must be a phonation type difference. This is in contrast to the natural tendency in tonal languages of carrying lexical contrasts within modal voice. This is a testable prediction that rests upon particular items in the Munro and Lopez (1999) description of Quiaviní Zapotec. This chapter instrumentally evaluates the categorization of some words in Munro and Lopez’ (1999) analysis, that are claimed to have non-modal voice. The prediction is that if there is a phonologically distinctive four-way tonal contrast in Quiaviní Zapotec, it ought to appear with modal voice. The ultimate goal of the chapter is to establish the phonological status of tone in Quiaviní Zapotec. Section §4.2 presents an overview of the phonation type mechanisms found in the languages of the world. The phonetic properties considered in subsequent sections are presented here. Sections §4.3, §4.4, and §4.5 analyze potential cases of modal low, rising and falling tone items, respectively. The chapter concludes with a summary of the findings and a reanalysis of the Quiaviní Zapotec tonal inventory, arguing that all four tones occur in modal voice. The implications of tone as contrastive in Quiaviní Zapotec, including an analysis of their tone-bearing units and the phonological representation of tone, are investigated in the next chapter.  60  It is important to mention that, despite the differences among these variants and Quiaviní Zapotec, there is a high degree of intelligibility among them.  103  4.1.1 Phonetic properties associated with phonation types  In order to establish what modal voice is and is not, this section provides a brief overview of the phonetic properties associated with phonation types. The phonetic properties described in this section serve as background to the acoustic descriptions and phonetic experiments of all the following sections. Phonation types refer to the manner in which vocal folds vibrate. Modal voice is the standard vibration type. The vocal folds are adducted along their full length and with a suitable degree of tension to allow vibration in a rhythmic manner, opening and closing at regular intervals of time. Breathy voice or murmur is where the folds are held partly apart while the vibration continues, and creaky voice or laryngealization is where the folds are held stiffly and vibration is partially inhibited. The different ways the vocal cords vibrate, or do not vibrate at all, create a variety of phonation types (Ladefoged, 1971; Catford, 1977; Laver, 1980). As suggested by Ladefoged (1971; see also Catford, 1964), these various glottal states may be represented in the form of a phonation continuum, “[…] defined in terms of the aperture between the arytenoid cartilages, ranging from voiceless (furthest apart), through breathy voiced, to regular, modal voicing, and then through creaky voice to glottal closure (closest together).” (Gordon & Ladefoged, 2001, p. 384). This is schematically represented in the following figure. Most open Phonation type  Voiceless  Most closed Breathy  Modal  Creaky  Glottal closure  Figure 10. Continuum of phonation types (Ladefoged, 1971)  The following unambiguous examples of breathy, modal and creaky vowels in Quiaviní Zapotec exemplify some of the phonation types mentioned above.  104  (1) Phonation types: modal, breathy and creaky a. Modal:  / be / ˩ → [ bèː ~ βèː ] ‘mesquite bean’  b. Breathy  / be̤ /  c. Creaky  ˩ → [ bè̤ː ~ βe͡e̤ː] ‘mold (growth)’  / bḛ / ˩ → [ bḛ̀ː ~ βe͡ḛ ] ‘Tanivet (X:ta'isy Dàany Bèèe')’  Phonetic properties associated with phonation types include differences in periodicity, fundamental frequency, spectral tilt, duration and intensity. Periodicity among different phonation types is illustrated in the following figure, showing waveforms of Quiaviní Zapotec vowels.  0.03384  Modal 0 –0.01495  0  0.24839 Time (s)  0.01456  Breathy  0  –0.0181  0  0.246077 Time (s)  Creaky  0.04312  0 –0.02649  0  0.254104 Time (s)  Figure 11. Waveforms of voice qualities: modal, breathy and creaky voices. Jitter is an effective calculation for measuring the periodicity of the signal. Jitter corresponds to measurements of the variation in the duration of adjacent pulses. This parameter has been used to establish differences in phonation types (e.g. Gordon & Ladefoged, 2001; Ladefoged, Maddieson, & Jackson, 1988). As shown above, adjacent pulses vary less during modal vowels than during non-modal vowels, especially creaky ones, typically characterized by irregularly spaced pulses. Another reliable way to measure phonation is spectral tilt, defined as “the degree to which intensity drops off as frequency increases” (Gordon & Ladefoged, 2001, p. 15).  105  Subtracting the amplitude of a higher frequency harmonic from the amplitude of the fundamental frequency (also called the first harmonic) yields a largely positive value for breathy vowels, a smaller positive value for modal vowels, and a negative value for creaky vowels. Spectral tilt has been a reliable measure of phonation in numerous languages such as Jalapa Mazatec, Gujarati, Kedang and Hmong (as reviewed by Gordon & Ladefoged, 2001). There are different ways to characterize spectral tilt. Primarily, the difference between the amplitudes of the first and second harmonics (H1-H2), which correlates with the percentage of a glottal vibration cycle during which the glottis is open (i.e. open quotient, Holmberg et al. 1995), has been used to distinguish between modal and breathy phonation. However, other studies have made use of the relationship between H1 (first harmonic) and harmonics exciting higher formants, which correlates with the abruptness of the closure of the vocal folds. These measurements include: H1-F3 (Stevens & Hanson, 1995), H1-F1 or H1-F2 (Ladefoged, 1983; Blankenship, 2002) and the average of H1-H2 compared to F1 (Stevens, 1988). Other studies have used the relationship of higher formants to lower ones such as F2-F3 (Klatt & Klatt, 1990). First and second formants (F1 and F2) are commonly referred to as A1 and A2, as it is the harmonic with the highest amplitude within the formant that is considered. Duration and intensity may also play a role in distinguishing modal versus nonmodal phonation. Non-modal vowels tend to have lower intensity and longer duration compared to modal vowels, e.g. Hupa for intensity (Gordon, 1998), and Jalapa Mazatec for duration (Silverman, Blankenship, Kirk, & Ladefoged, 1995; Silverman, 1997b).  4.2  Experiment 1: Low tone with modal voice Munro and Lopez (1999) recognize Quiaviní Zapotec as a tonal language with  four tones (high, low, falling and rising); however, they state that tones do not represent primary contrasts, but melodies derived from voice qualities. By contrast, a prototypical tonal language would use its tonal inventory distinctively within modal voice. This chapter reconsiders some vowel patterns described in Munro and Lopez (1999) by  106  examining contrasts in Quiaviní Zapotec. In this section I investigate the case of low tone. I argue that the low tone vowel pattern àa has modal voice. Munro and Lopez (1999) present the following Quiaviní Zapotec vowel patterns with low tone: Table 20. Munro and Lopez (1999: 4) low tone vowel patterns Pattern ah ahah àa  Combination61 ah (same) ah àa (same)  Examples zah ‘grease’ bihih ‘air’ bòo ‘charcoal’  Tone low low low  The first two have breathy voice and will be analyzed in chapter 6, which examines nonmodal phonation. The pattern àa is of crucial interest to this chapter. According to the orthography, it appears to represent / a̰a /; however, this is more an orthographic convention rather than a phonological representation. The authors maintain “the vowel complex we write as creaky vowel followed by plain vowel is suspicious. […] We have considered the idea that àa […] should be represented as a sequence of two creaky vowels, but in fact the degree of creakiness of this vowel is (perceptually and instrumentally) considerably less than any other sequences […] that include creaky vowels (p. 5).” The suspicious status of this vowel pattern makes it a clear candidate to look for the expression of low tone within modal voice.  61  Recall from Chapter 1, that according to Munro and Lopez (1999) many Valley Zapotec words shorten to simpler combination forms in some contexts (e.g. when suffixes are added to them).  107  4.2.1 Acoustic description: Modal-L The purpose of this section is to describe the acoustic characteristics of low-tone items with the vowel pattern àa (Munro & Lopez, 1999). In order to clearly see the voice quality and pitch of items with this vowel pattern, I compare them with the unambiguous modal-H pattern. Consider the contrastive sets in (2).62 (2) Modal voice minimal pairs: High vs. Low tones 63 a. / danj /  ˥  ‘harm’  vs.  / danj /  ˩  ‘mountain’  b. / ʒi /  ˥  ‘tomorrow’  vs.  / ʒi /  ˩  ‘bitter’  vs.  / nda /  ˩  ‘quite’  ‘side’  vs.  / lad /  ˩  a. / nda /  ˥  d. / lad /  ˥  ‘sensitive’ ‘between’  Figure 12 shows the waveform and the spectrogram of /danj/ ˥ ‘harm’ (daany) on the left, and /danj/ ˩ (dàany)64 ‘mountain’ on the right, by male speaker TiuR. The spectrogram frequency range is 0-5000 Hz (on the left) and the pitch frequency (blue line) on the range of 50-300 Hz.  62  A contrastive set is defined by Pike (1947, p. 161) as “a group of tone sequence patterns, in some particular position, which differ only by one tone in the same relative place in the sequence”. 63 For these examples, no phonetic transcription is included, as it does not add any information with respect to the issue at hand; all these examples surface with long vowels. 64 Since the voice quality of modal-L is in question I present these items with my hypothetical phonological transcription, followed by the dictionary’s orthography (Munro & Lopez, 1999).  108  d  aː  j  ɲ  d  aː  j  ɲ  Figure 12. Waveform and spectrogram of / danj / ˥ ‘harm’ (daany) on the left, and / danj / ˩ (dàany) ‘mountain’ on the right, by male speaker TiuR.  Beginning with a pitch evaluation, the high-tone word / danj / ˥ ‘harm’ has a pitch of 143 Hz during the vowel, whereas the pitch for the low-tone word / danj / ˩ ‘mountain’ averages 123 Hz. (High-tone items for this speaker average 155 Hz whereas low-tone items average 121 Hz.) In both cases, the pitch is stable and relatively flat throughout the vowel. It starts to lower with the glide and the consonant. Most tokens with high or low tone have a slight pitch lowering (more noticeable for low tone) towards the end of the vowel if the syllable is open, or closed by an obstruent. If followed by a sonorant coda, the pitch is maintained if the sonorant is fortis, but normally drops if it is lenis. The next chapter discusses in detail the type of coda consonant and its relevance with respect to tone. According to Rietveld and Gussenhoven (1985), pitch differences of 1.5 semitones (about 10 Hz) can reliably be interpreted as prominence differences. Mambila (Connell, 2000), for instance, has four level tones and they are spaced an average of 10 Hz apart. In a language like Quiaviní Zapotec with only two level tones, my prediction is to find a more spacious separation between tones. Based on the examples illustrated above, the difference between high and low tone is more than 20 Hz. Quiaviní Zapotec, then, looks like a tonal language in terms of its pitch characteristics. 109  With respect to voice quality, the periodicity of the sounds in Figure 12 (a correlate of modal voice) is clear throughout both examples. In turn, the spectrograms are clear and with no signs of laryngealization in the case of the low tone (e.g. no strong or weak “trillization” (Pike, 1947, p. 21) during the vowel). Although non-modal phonation is normally associated with lowering of the fundamental frequency, the contrastive use of low tone with modal voice within a tonal system is prototypical, and this parameter on its own (pitch at the acoustic level) is not enough to determine voice quality in a tonal language. As reviewed above (§4.2), non-modal vowels may be of longer duration than modal vowels. In Quiaviní Zapotec, however, length plays an important role in the prosody (Chapter 3). Short vowels appear before fortis consonants and long vowels before lenis consonants or in open syllables. Both the high and low tone examples in Figure 12 have long vowels: 323 ms and 360 ms, respectively, including the glide. Finally, intensity levels are very similar: 60 dB for the modal-H token, and 67 dB for the modal-L. Further examples by a different speaker are provided in Figure 13.  ʒ  iː  ʒ  iː  Figure 13. Waveform and spectrogram of / ʒi / ˥ ‘tomorrow’ (zhii), on the left, and / ʒi / ˩ (zhìi) ‘quite’, on the right, by male speaker TiuC.  110  Pitch for / ʒi / ˥ ‘tomorrow’ averages 122 Hz, whereas for / ʒi / ˥ ‘quite’ it is 106 Hz. Both measurements are within the averages of level tones for this speaker. The high tone word shows brief rising that can be taken as a phonetic preparation for the phonological expression of high tone; it goes from 115 Hz to 124 Hz at the highest pitch value. Then, after about 100 ms of flat pitch, it lowers towards the end of the word. The pitch in the low tone word is stable and relatively flat during the first 100 ms, then it starts to lower, a common tendency with low-tone items in this language. In terms of phonation type, the glottal pulses of both sounds are regular and the spectrograms show clear formant frequencies in both examples. Towards the end of the low-tone example, we notice some weakening of the formant frequencies, correlated with a drop in intensity. This may be an utterance-final effect. Overall intensity for the modalH token is 69 dB, and a slightly lower value of 66 dB for the modal-L one. Finally, although the low-tone item has a longer vowel, both are well within the range of long vowels at 238 ms (modal-H) and 276 ms (modal-L). In summary, based on the acoustic description from above (§4.2), items with the vowel pattern àa appear to have modal voice. In order to confirm this analysis, I conducted a phonetic experiment to instrumentally and statistically test the phonation type of items that I anticipatorily called modal-L.  4.2.2 Phonetic experiment: Modal-L This section consists of a phonetic experiment that examines the voice quality of items with the vowel pattern àa, originally analyzed in Munro and Lopez (1999) as having some amount of creakiness (tension in the vocal folds) and compares them with unambiguous cases of modal voice (high-tone items) and unambiguous cases of creaky voice (low-tone items). The hypothesis of this study is that Quiaviní Zapotec uses tone contrastively, with the specific prediction that low tone is used with modal voice. Accordingly, the vowel pattern àa is tentatively called modal-L.  111  In order to test this prediction, the phonetic parameters I considered are periodicity (jitter), spectral tilt, duration and intensity. The first two are considered primary since both have been reliable parameters in distinguishing different voice qualities in several languages (see §4.1.1 above). Specifically, spectral tilt has already been applied successfully to illustrate modal voice (high tone), as well as unambiguous creaky and breathy voice in Quiaviní Zapotec (Gordon & Ladefoged, 2001, pp. 15-17; Ladefoged, 2003, pp. 178-181). Duration and intensity may also play a role in distinguishing modal from non-modal phonation (§4.1.1). Nonetheless, in this study they are considered secondary parameters due to the mixed results from previous studies. Gordon and Ladefoged (2001, p. 18) report no durational differences among breathy, modal and creaky vowels in Quiaviní Zapotec (although the sample analyzed is small and numbers are not reported). On the other hand, Chávez-Peón (2008) found that breathy vowels were longer than modal ones. Also in this study, intensity values were slightly higher for modal vowels versus breathy ones.  4.2.2.1 Methods Subjects: Two native speakers of Quiaviní Zapotec participated in the study: 1 female speaker (LiaL, 35), and 1 male speaker (TiuC, 40). Stimuli: This experiment considered as control cases the unambiguous modal voice of the modal-H tokens, and the unambiguous creaky voice quality of creaky-L tokens.65 These control cases were compared with each other, and with the voice quality of the modal-L tokens. (3) Stimuli groups 1. Modal-H 2. Modal-L 3. Creaky-L  control: modal voice under investigation control: creaky voice  65  I agree with Munro and Lopez (1999) on the voice quality of the control cases considered in this experiment; for creaky vowels, however, the cases considered to have low tone are reported to have falling tone in Munro and Lopez (1999). (See Chapter 6 for the analysis of tone in non-modal vowels.)  112  The actual stimuli consisted of four words for each group. All of these words have long vowels (open syllables or lenis coda), because (a) the longer duration in these environments allows a better comparison of the voice quality, and (b), most of the (near) minimal pairs that I have identified have these syllabic characteristics. Table 21. Stimuli: low tone experiment dictionary Modal -H 1 daany 2 ndaa 3 daad 4 bdaa Modal-L 5 dàany 6 ndàa 7 nàan 8 bdàan Creaky-L 9 gààa' 10 bààa' 11 lààa'z 12 yààa'n  gloss ‘harm’ ‘bitter’ ‘dice’ ‘shadow’ ‘mountain’ ‘sensible’ ‘thick’ ‘soot’ ‘nine’ ‘tomb’ ‘heart, center’ ‘corncob’  All of these words were recorded in the following carrier phrase: (4) Carrier phrase [ ɾiː ɾa _________ ɾuk ] ‘There are _______ here’ (orthography: rii ra ______ ru’c) This particular carrier phrase was used because it contains only modal voice vowels, thus avoiding any possible contextual influence from non-modal voice. Four repetitions of each phrase were collected based on a randomized list, for a total of 96 tokens (4 modalH + 4 modal-L + 4 creaky-L = 12 x 4 repetitions x 2 speakers = 96 tokens). The stimuli were recorded using a Marantz 660 solid-state recorder and a Countryman lapel microphone (phantom power). Measurements were done in Praat for Mac (version 5.1.07; Boersma & Weenink, 2009); results were compiled in Excel 2004 for Mac; and statistics were run in JMP IN 5.1 for Mac (two-tailed unequal variance t-tests).  113  Measurements: Periodicity was calculated by jitter, measuring the variation in duration of glottal cycles. The measures of jitter considered in this study are ppq5 and ddp:66  (5) Jitter (ppq5) (Praat manual: jitter) This is the five-point Period Perturbation Quotient, the average absolute difference between a period and the average of it and its four closest neighbors, divided by the average period.67 (6) Jitter (ddp) (Praat manual: jitter) This is the average absolute difference between consecutive differences between consecutive periods, divided by the average period. Jitter (ppq5) was chosen, as it is the least dependent calculation on pitch. In order to have an additional jitter reference, Jitter (ddp) was also considered. This is Praat’s original ‘Get jitter’ function, and probably the most common calculation in the literature. Since jitter measures the variation in duration of glottal cycles, changes in pitch will show variation in duration of these cycles. In other words, rising and falling contours may influence jitter values. For this reason measurements were not taken for the whole vowel, but during a specific portion: six glottal pulses at the center of the vowel (the minimum required by jitter (ppq5) are 5 pulses). By measuring jitter at the center of the vowel we also avoid effects of the preceding and following consonants, or effects of final lowering at the end of the phrase. Spectral tilt measurements include H1-H2 and H1-A1,68 defined as follows: (7) H1-H2 (open quotient): Difference in dB between the first and second harmonics in the Fourier spectrum. Used to estimate the proportion of a cycle in which the glottis is open (Ni Chasaide & Gobi, 1997). (8) H1-A1 (spectral slope): Difference in dB between the first harmonic and the most prominent harmonic in the F1 region (Kirk et al., 1993). 66  Other jitter calculations include: local, local absolute, and rap. The Multi-Dimensional Voice Program (MDVP), a standard software tool for quantitative acoustic assessment of voice quality, calls this parameter PPQ, and gives 0.840% as a threshold for pathology (that is, in languages without phonemic laryngealization). 68 A1 corresponds to the amplitude of the harmonic within the first formant (F1) that has the greatest amplitude. 67  114  -0.2449  0 0.2844 Measurements were obtained from FFT spectra at specific points during the Time (s)  vowel duration. Since non-modal phonation may be localized to a portion of the vowel (a 0.1667observed in Otomanguean languages, e.g. Jalapa Mazatec in Silverman et al. pattern  1995, Blankenship 1997), the measures H1-H2 and H1-A1 were taken at five evenly spaced 0intervals distributed from the onset to the offset of the vowel.69 Figure 14 illustrates this procedure. 70 -0.2449  0  1  2  3  Time (s)  4  5  0.2844  0.1667  0  -0.2449  0  0.2844 Time (s)  Figure 14. Spectral tilt measurements were taken at five evenly spaced intervals distributed from the onset to the offset of the vowel (Solid lines in the extremes indicate onset and offset of the vowel; dashed lines divide the intervals; and the arrows indicate the points were the measurements were taken). Finally, each vowel was measured for duration (ms; total timing of vowel) and intensity (dB; average within vowel duration).  69  Based on House (1961) and Gordon (2004), the duration of each vowel was measured from the waveform in conjunction with a wide band spectrogram. The onset and offset of the second formant served as the beginning and end points, respectively, of each duration measurement. Duration criteria also included the initiation and cessation of voicing, and F1 and F2 transitions. 70 I thank Christian DiCanio for sharing the Praat script ‘Get_spectral_tilt’ to obtain these measurements (See DiCanio, 2008).  115  4.2.2.2 Results Beginning with jitter results, Figure 15 shows the mean results for both jitter (ppq5) and jitter (ddp) for TiuC and LiaL. Tables following each figure present the means and standard deviations, as well as the statistical analysis results.  Figure 15. Jitter (ppq5 & ddp) mean results (TiuC).  Figure 16. Jitter (ppq5 & ddp) mean results (LiaL). 116  Table 22. Periodicity (jitter): Mean and standard deviation (LiaL & TiuC) LiaL TiuC Jitter (ppq5) Jitter (ddp) Jitter (ppq5) Jitter (ddp) modal-H Mean 0.224% 0.443% 0.217% 0.596% SD 0.111 0.298 0.126 0.423 modal-L Mean 0.172% 0.337% 0.295% 0.624% SD 0.267 0.131 0.208 0.285 creaky-L Mean 0.921% 1.141% 0.706% 1.777% SD 0.639 0.619 0.392 1.500 Table 23. Jitter results: Probability values from t-test (LiaL & TiuC)71 LiaL TiuC Jitter (ppq5) Jitter (ddp) Jitter (ppq5) Jitter (ddp) modal-H vs. creaky-L <0.001 0.003 <0.001 0.007 modal-H vs. modal-L 0.206 0.215 0.212 0.830 modal-L vs. creaky-L <0.001 0.020 0.001 0.008 For both types of jitter, modal-H and modal-L are grouped together. Results for the male speaker show slightly higher jitter in modal-L tokens than modal-H, but the reverse is observe in the results for the female speakers. There are no significant differences between modal-H and modal-L. Creaky-L is statistically different. Spectral tilt: I provide below a figure with the average plot results for spectral tilt H1-H2 for both subjects. Although the male speaker has lower average values, both speakers show the same tendency, and thus it is possible to combine their results in the same graph. The figure is followed by the results of the female speaker (LiaL) and another table with the corresponding t-test results. I then present results and statistics for the male speaker (TiuC). Figure 17 shows that at the first two intervals, all three types of vowels exhibit similar patterns. By the third interval creaky-L tokens start to be noticeably different, and at intervals 4 and 5, all creaky-L numbers are negative, for both subjects (Tables 24 and 27). The modal-L tokens from the female speaker (LiaL) show lower spectral tilt values 71  According to standard conventions, results above 0.12 are considered not significant (ns.); results between 0.12 and 0.05 are marginally significant; finally, any value below 0.05 is statistically significant.  117  than modal-H ones; whereas the male speaker (TiuC) shows more similar values for both modal-L and modal-H tokens. As expected, modal-H versus creaky-L as well as modal-L vs. creaky-L show significant differences for both subjects at intervals 3, 4 and 5. Unexpectedly, differences between modal-H and modal-L were significant for the female speaker in all intervals except the first one. For the male speaker, however, H1-H2 was higher (less creaky-like) for modal-L than for modal-H during the third and fourth intervals, and practically identical during the second and last measurement points.  Figure 17. H1-H2 plot for mean results of both speakers. Table 24. H1-H2 results: Mean and standard deviation (LiaL) 1H1-H2 2H1-H2 3H1-H2 4H1-H2 5H1-H2 modal - H Mean 6.34 7.14 7.36 6.93 5.23 SD 2.23 2.07 2.46 2.60 3.41 modal - L Mean 4.40 4.93 4.12 2.80 1.79 SD 3.17 3.11 3.50 3.66 3.79 creaky - L Mean 6.73 4.83 1.64 -3.66 -3.09 SD 3.04 3.90 2.96 4.64 4.22  118  Table 25. H1-H2 results: Probability values from t-test (LiaL) 1H1-H2 2H1-H2 3H1-H2 4H1-H2 5H1-H2 modal-H vs. creaky-L 0.683 0.048 <0.001 <0.001 <0.001 modal-H vs. modal-L 0.055 0.025 0.005 0.001 0.011 modal-L vs. creaky-L 0.042 0.940 0.038 <0.001 <0.001  Table 26. H1-H2 results: Mean and standard deviation (TiuC) 1H1-H2 2H1-H2 3H1-H2 4H1-H2 5H1-H2 modal - H Mean -0.05 -0.15 -0.08 -0.36 -0.06 SD 0.74 0.95 0.88 0.92 0.75 modal - L Mean -0.49 -0.11 0.02 -0.02 -0.07 SD 0.77 0.46 0.68 0.54 0.61 creaky - L Mean -0.47 -1.05 -2.28 -5.17 -4.59 SD 1.06 1.09 3.35 6.83 5.55 Table 27. H1-H2 results: Probability values from t-test (TiuC) 1H1-H2 2H1-H2 3H1-H2 4H1-H2 5H1-H2 modal-H vs. creaky-L 0.202 0.018 0.021 0.013 0.005 modal-H vs. modal-L 0.108 0.901 0.714 0.201 0.960 modal-L vs. creaky-L 0.956 0.004 0.015 0.008 0.005 With respect to the H1-A1 spectral tilt measure, all results were as predicted for both speakers. Modal-H and modal-L have similar results, i.e., spectral tilt values are not consistent with greater creakiness on modal-L tokens. Results cluster together in comparison with creaky-L, with statistically significant differences at intervals 3, 4 and 5.  119  Figure 18. H1-A1 plot for mean results of both speakers. Table 28. H1-A1 results: Mean and standard deviation (LiaL) 1H1-A1 2H1-A1 3H1-A1 4H1-A1 5H1-A1 modal-H Mean -3.01 -3.51 -3.04 -3.06 -2.91 SD 4.77 3.70 3.62 3.59 3.70 modal-L Mean -0.42 -0.19 -1.06 -0.60 0.33 SD 5.09 5.06 3.08 3.81 5.55 creaky-L Mean -2.14 -3.86 -8.68 -13.67 -9.64 SD 3.59 4.54 3.80 4.95 5.66 Table 29. H1-A1 results: Probability values from t-test (LiaL) 1H1-A1 2H1-A1 3H1-A1 4H1-A1 5H1-A1 modal-H vs. creaky-L 0.563 0.810 <0.001 <0.001 <0.001 modal-H vs. modal-L 0.147 0.043 0.106 0.07 0.062 modal-L vs. creaky-L 0.278 0.038 <0.001 <0.001 <0.001 Table 30. H1-A1 results: Mean and standard deviation (TiuC) 1H1-A1 2H1-A1 3H1-A1 4H1-A1 5H1-A1 modal - H Mean -5.53 -7.27 -7.43 -7.62 -4.68 SD 3.43 4.67 3.58 4.16 5.38 modal - L Mean -7.17 -6.86 -6.88 -5.57 -2.28 SD 3.10 2.45 2.80 2.74 2.27 creaky - L Mean -4.39 -8.74 -11.43 -11.08 -8.77 SD 4.88 3.08 4.57 6.12 3.95 120  Table 31. H1-A1 results: Probability values from t-test (TiuC) 1H1-A1 2H1-A1 3H1-A1 4H1-A1 5H1-A1 modal-H vs. creaky-L 0.451 0.302 0.010 0.073 0.020 modal-H vs. modal-L 0.165 0.762 0.635 0.110 0.115 modal-L vs. creaky-L 0.065 0.067 0.002 0.003 <.001 Duration and intensity results are presented in the following tables. Tables 32 and 34 present averages and standard deviation, Tables 33 and 35 statistical results (t-test). Neither duration nor intensity yields significant differences among the items in consideration. (The only significant result was the difference in intensity between modalL vs. creaky-L for female speaker LiaL; modal-H vs. modal-L was marginally significant.) These parameters were not even reliable between the control cases modal-H and creaky-L. In short, all the vowels in the study have similar duration and intensity values. Table 32. Duration and intensity results: Mean and standard deviation (LiaL) Duration (ms) Intensity (dB) modal-H Mean 235 69.62 SD 35.84 2.55 modal-L Mean 230 67.5 SD 33.29 3.75 creaky-L Mean 225 69.87 SD 32.07 2.70 Table 33. Duration and intensity results: Probability values from t-test (LiaL) Duration Intensity modal-H vs. creaky-L 0.40 0.78 modal-H vs. modal-L 0.67 0.07 modal-L vs. creaky-L 0.67 0.04  121  Table 34. Duration and intensity results: Mean and standard deviation (TiuC) Duration (ms) Intensity (dB) modal-H Mean 195 72.81 SD 22.98 4.02 modal-L Mean 185 71.88 SD 11.83 2.36 creaky-L Mean 194 70.69 SD 19.98 3.03 Table 35. Duration and intensity results: Probability values from t-test (TiuC) Duration Intensity modal-H vs. creaky-L 0.858 0.102 modal-H vs. modal-L 0.133 0.429 modal-L vs. creaky-L 0.147 0.226  4.2.2.3 Discussion Results for both jitter and spectral tilt show that while some of the small nonsignificant differences are consistent with very light laryngealization in modal-L tokens, others suggest the reverse (less laryngealization than modal-H). This is exactly as expected if both vowel types are equally modal. Let us discuss these parameters in more detail. Periodicity (jitter) results clearly confirm the modal voice quality of the modal-L items in question. Measures of jitter (ppq5 and ddp) establish creaky-L items as having clear aperiodicity, as opposed to modal-H and modal-L, which show periodicity in their signal with no statistical difference between them. This experiment demonstrates the effectiveness of jitter as an acoustic parameter in the distinction of phonation types. To my knowledge, this experiment is the first one that uses jitter in the description of Otomanguean languages. Spectral tilt. Beginning with the comparison between modal-H versus creaky-L, spectral tilt results indicate modal voice at the beginning of these vowels. All measurements (H1-H2 and H1-A1) are similar in both subjects at intervals 1 and 2. From interval 3 to 5 (and from interval 2 in H1-H2), the differences between modal-H and creaky-L are statistically significant. As expected, creakiness in creaky-L tokens is found  122  from the middle towards the end of the vowel. Overall, this confirms that the amplitude differences H1-H2 and H1-A1 serve as an indicator of phonation types in the language. More specifically, the second and higher harmonics (A1) have greater energy relative to that of the fundamental (F0) in creaky phonation, whereas the difference is smaller in modal phonation. As for the case under investigation, modal-L versus the modal-H control, results show that it is possible to group them as cases of modal voice. According to the hypothesis, we expect modal-L to have spectral tilt results within a modal phonation range, and this is, in fact, what was obtained. For both subjects, none of the results were statistically different when comparing the prototypical modal phonation with modal-L tokens, with the exception of H1-H2 for the female subject, LiaL. With respect to this difference, modal-L results still show positive numbers, which is expected for modal phonation when comparing H1-H2. Put in other words, within a spectral tilt modal range we may expect differences, and in this case the differences can be attributed to tone and gender. Supporting this reasoning, spectral tilt has also been used as an effective indicator of stressed syllables versus non-stressed (for English, see Laver, 1994, Campbell & Beckman, 1997; for Spanish, see Ortega-Llebaria & Prieto, 2007); since stress is typically associated with high pitch in non-tonal languages, this may also explain the differences between modal-H and modal-L in Quiaviní Zapotec. As for speaker differences, the female speaker produced notably different pitch for each type of tone; thus, this is reflected in the spectral tilt results. Pitch (tone) differences were more subtle in the case of male speaker TiuC. Finally, with respect to the comparison between modal-L (the case under investigation) versus control creaky-L, the tendency for both subjects in all parameters is for modal-L to pattern with creaky-L during the first two intervals, but statistically differ for intervals 3, 4 and 5. This is basically the same pattern found for modal-H versus creaky-L, that is, all three patterns together at first. Recapitulating, I mentioned in the introduction that the pattern analyzed here as modal-L was analyzed in Munro and Lopez (1999) as having some amount of laryngealization, being probably somewhere in between modal and creaky voice, maybe tense voice. Let us consider this in more detail in light of the results of the experiment.  123  With respect to modal-L, results demonstrated that these tokens do not have creaky voice, and, more importantly, the results also rule out the possibility of attributing tense voice to modal-L items. Studies analyzing tense (or pressed) voice (DiCanio, 2009; Tejada, 2009) show that this type of phonation tends to pattern with creaky voice, with slightly less negative numbers for the different spectral tilt measurements, and with considerably different values to those of modal voice. This was not the case in Quiaviní Zapotec modal-L tokens. (See also the analysis of creaky vowels with high tone in Quiaviní Zapotec as cases of tense voice in Chapter 6.) The last parameters considered in this study were duration and intensity. They yield no significant results in comparing the control cases: modal-H vs. creaky-L. For female speaker LiaL, the difference between modal-H vs. modal-L was marginally significant, and that of modal-L vs. creaky-L was significant. The latter difference would be in line with the prediction of the experiment; however, the lack of significant differences between the control cases diminishes the assessment of any other dissimilarity. No significant results were obtained for the male speaker. With respect to duration, I mentioned above that Gordon and Ladefoged (2001) report no duration differences between modal, creaky and breathy vowels in Quiaviní Zapotec. In addition, duration plays an important role in the prosodic pattern in this language (see previous chapter); hence, phonation types seem to be subordinated to prosody. With respect to intensity, which was measured for the overall duration of the vowel, perhaps measurements at specific points throughout the vowel (intervals) could have shown significant variation. All in all, it seems that neither duration nor intensity are useful parameters to distinguish modal vs. laryngealized vowels in Quiaviní Zapotec. In summary, the following conclusions can be drawn from the results of this phonetic experiment. First, jitter and spectral tilt results confirm the modal voice quality of modal-L tokens, as they pattern with modal-H for most of the parameters in both subjects. Whenever the results were significantly different (H1-H2 for LiaL), results are still within the modal phonation range and the differences can be attributed to pitch. Second, the modal voice control (modal-H), as well as the modal voice case under investigation (modal-L), are significantly different from the creaky voice control (creaky-  124  L) at the intervals 3, 4 and 5, i.e. at the middle and second part of the vowel production, which is the part where the creakiness is mainly manifested. Third, tone and non-modal voice are sequenced: based on the above results, we can confirm that laryngealization is found towards the second half of the vowel in Quiaviní Zapotec creaky vowels with low tone (similar phonetic characteristics are found for creaky-F examples; see Chapter 6 for more details). According to Yip (2002, p. 25) “two contrastive surface tones is the minimum necessary to earn the name of ‘tone language’ ”. This section confirms two distinctive tone categories in Quiaviní Zapotec, the level tones high ( ˥ ) and low ( ˩ ), and thus, corroborates the hypothesis of the study, that Quiaviní Zapotec uses tone contrastively. In turn, there is a partial confirmation of the prediction that if there is a four-way tonal contrast in Quiaviní Zapotec, it ought to appear with modal voice. Having established the contrastive use of tone in Quiaviní Zapotec for the two level tones, the next two sections evaluate the possibility of contour tones occurring with modal voice in this language.  4.3  Experiment 2: Rising tone with modal voice  Rising tone in Quiaviní Zapotec is reported in Munro and Lopez (1999) with the vowel patterns in Table 36. Table 36. Munro and Lopez (1999, p. 4) rising tone vowel patterns 1 2 3 4  Pattern a’a a’aa àaa àaa’  Combination a’a (same) a’a a’a a’a  Examples gyi’izh ‘city person’ chi’iinnzh ‘bedbug’ nnàaan ‘mother’ rsìii’lly ‘morning’  Tone rising rising rising rising  According to Pam Munro (p.c.) a’a is an orthographic convention for rising tone items with a certain amount of non-modal phonation. Additionally, Munro and Lopez (1999, p. 32) note that “the brief glottal gesture interrupting a checked vowel preceding another vowel at the beginning of a vowel complex can be difficult to perceive. The glottal stop is 125  clearer in vowel complexes where the checked vowel is flanked by other vowels.”72 Other reasons to define these patterns this way include the native-speaker intuition of one of the authors (Felipe Lopez), as well as the comparison with cognates in other Zapotec languages. Notice that all of the vowel patterns in Table 36 are reduced to the first one, a’a, in their combination forms.73 The vowel pattern a’a is the most frequent in rising tone items. In my fieldwork experience, the voice quality of these tokens varies slightly among speakers, but is predominantly modal. Women always produce them with modal voice, whereas for some male speakers, their low pitch range may cause it to sound as if they were produced with some tension in the vocal folds at the beginning. Acoustically, however, I can detect only modal voice in rising-tone tokens, as shown in the acoustic description below. In the search of the four-way tonal contrast with modal voice in this language, the purpose of this section is to establish the voice quality of items with rising tone in Quiaviní Zapotec. Towards this goal, I follow the same structure as in the previous section. First, I present a preliminary acoustic description of rising-tone items, then a phonetic experiment that instrumentally examines their phonation.  4.3.1 Acoustic description: Modal-R This section describes the acoustic characteristics of rising-tone items with the vowel pattern a’a (Munro & Lopez, 1999), with the purpose of demonstrating the contour shape of these lexical items, as well as evaluating their voice quality. Consider the following (near) minimal pairs.  72  In agreement with these authors, these latter vowel patterns, including for example àa’ah and àa’ah, are analyzed here as interrupted vowels (see Chapter 6, §6.5). 73 Combination forms are shortened realizations of some vowel patterns when endings are added to them, or in compounds (see Munro & Lopez, 1999; and Munro et al, 2008).  126  (9) Modal voice (near) minimal pairs: High vs. Rising tones a. / ʧan / ˥ ‘Feliciano’  vs.  / ʧan /  Ë ‘respectful greeting’  b. / dad / ˥ ‘dice’  vs.  / dad /  c. / ʒjet / ˥ ‘little’  vs.  / ʒjet /  Ë ‘father’  d. / ʒi /  vs.  / ʃi /  ˥ ‘tomorrow’  Ë ‘cat’ Ë ‘what (ellip.)’  (10) Modal voice (near) minimal pairs: Low vs. Rising tones a. / danj / ˩ ‘mountain’  vs.  b. / nan / ˩ ‘thick (liquids)’ vs.  / bdanj /  Ë ‘type of traditional dress’  / nˑan /  Ë ‘mother’  c. / nda /  ˩ ‘sensitive’  vs.  / dad /  Ë ‘father’  d. / nla /  ˩ ‘greedy’  vs.  / nlas /  Ë ‘extremely thin’  The first contour tone to be analyzed within modal vowels is the rising tone. The distribution of this tone is not restricted segmentally; fortis and lenis consonants may appear both in onset and coda position. Rising tone may also appear in open syllables, but the number of lexical items of this type is small. The following figure illustrates the realization of rising tone in Quiaviní Zapotec. 0.08374  0.2289  0  0  -0.07495  0  0.6511  -0.2197  0  0.528 Time (s)  Time (s)  500  500  300  300  200  200  150  150  100  100  70  70  50  0  0.6511  50  0  0.528 Time (s)  Time (s)  x  i˘  Z  ɣ  iˑ  ʒ  Figure 19. Waveform and spectrogram of / ɡiʒ / Ë ‘city person’, by male speaker TiuR and female speaker LiaB.  127  The spectrograms above illustrate rising tone with the word / ɡiʒ / Ë ‘city person’ in both male and female speech. The pitch of the former goes from 121 Hz to 151 Hz, whereas the female’s token starts at 190 and finishes at 245 Hz. Overall results for rising tone for TiuR average a gliding curve of 121-144 Hz, whereas LiaB results show 198-226 Hz. These numbers are very similar to the individual correspondents of low and high tone. Finally, as the figures above show, the contour of the rising tone tends to be located in the second half or towards the end of the vowel. With respect to the voice quality, my analysis of lexical items with the rising tone indicated no laryngealization (either creakiness or a glottal closure). As shown in the figure above, neither pitch nor intensity is interrupted during the vowel duration, as expected with a checked (interrupted) vowel (see §6.5 in Chapter 6).  4.3.2 Phonetic experiment: Modal-R As mentioned above, the vowel pattern a’a was originally analyzed (Munro and Lopez 1999) as having some amount of laryngealization. In contrast, the acoustic description in the preceding section provides evidence for the re-categorization of risingtone items as modal-R. In order to test this hypothesis, rising-tone items are acoustically analyzed. These items were part of the recordings made for the evaluation of the modal-L (àa) items previously presented. As such, the characteristics of the analysis are the same: modal-R tokens are compared with unambiguous cases of modal voice (high-tone items) and unambiguous cases of creaky voice (low-tone items). The hypotheses and predictions are the same for the analysis of the whole chapter. The phonetic parameters considered in this section are periodicity (jitter) and spectral tilt. Since duration and intensity showed no significant results in the evaluation of modal-L in the previous section, they are not included here.  128  4.3.2.1 Methods The methodology of this experiment is the same as that of the previous section for low tone, with the addition of the following rising tone items. Table 37. Stimuli (partial): rising-tone experiment Modal-R 1 2 3 4  da’ad na’an cha’an zhya’ab  ‘father’ ‘mother’ ‘respectful greeting’ ‘bad, evil’  As shown in Table 36, the four rising-tone vowel patterns described in Munro and Lopez (1999) may be reduced to the most common pattern a’a. For this reason all the modal-R items are of this type. As before, four tokens of each item were recorded by Quiaviní Zapotec native speakers LiaL (female) and TiuC (male), in the same carrier sentence and under the same conditions of the previous experiment (§4.2.2).  4.3.2.2 Results Figures 20 and 21 show the average results for jitter (ppq5 & ddp) for modal-H and creaky-L of the previous section, along with the results for the items in question in this section: modal-R. For both speakers, we observe that modal-R is different from creaky-L, and how it patterns with the other two modal items. This is confirmed statistically, presented in Tables 38 and 39.  129  Figure 20. Jitter (ppq5 and ddp) mean results (TiuC).  Figure 21. Jitter (ppq5 and ddp) mean results (LiaL).  130  Table 38. Periodicity (jitter): Mean and standard deviation (LiaL and TiuC) LiaL TiuC Jitter (ppq5) Jitter (ddp) Jitter (ppq5) Jitter (ddp) modal-H Mean 0.224% 0.443% 0.217% 0.596% SD 0.111 0.298 0.126 0.423 modal-R Mean 0.188% 0.332 0.246% 0.808% SD 0.118 0.091 0.129 0.351 creaky-L Mean 0.921% 1.141% 0.706% 1.777% SD 0.639 0.619 0.392 1.500 Table 39. Jitter results: Probability values from t-test (LiaL and TiuC) LiaL TuC Jitter (ppq5) Jitter (ddp) Jitter (ppq5) Jitter (ddp) modal-H vs. creaky-L <0.001 0.003 <0.001 0.007 modal-H vs. modal-R 0.124 0.315 0.520 0.133 modal-R vs. creaky-L <0.001 0.020 <0.001 0.022 As regards spectral tilt results, modal-R is within the range of modal voice (with triangles in yellow in the figure below) reporting positive values by female speaker LiaL, although significantly different from modal-H, and values around zero for male speaker TiuC. Both speakers’ results were statistically different between modal-R and creaky-L.  131  Figure 22. H1-H2 plot for mean results of both speakers. Table 40. H1-H2 results: Mean and standard deviation (LiaL) 1H1-H2 2H1-H2 3H1-H2 4H1-H2 5H1-H2 modal - H Mean 6.34 7.14 7.36 6.93 5.23 SD 2.23 2.07 2.46 2.60 3.41 modal - R Mean 3.75 3.14 2.75 3.06 2.79 SD 3.99 3.53 3.88 3.38 2.82 creaky - L Mean 6.73 4.83 1.64 -3.66 -3.09 SD 3.04 3.90 2.96 4.64 4.22 Table 41. H1-H2 results: Probability values from t-test (LiaL) modal-H vs. creaky-L modal-H vs. modal-R modal-R vs. creaky-L  1H1-H2 2H1-H2 3H1-H2 4H1-H2 5H1-H2 0.683 0.048 <0.001 <0.001 <0.001 0.0325 <0.001 <0.001 0.0011 0.0355 0.0243 0.2091 0.3698 <0.001 <0.001  132  Table 42. H1-H2 results: Mean and standard deviation (TiuC) modal - H Mean SD modal - R Mean SD creaky - L Mean SD  1H1-H2 2H1-H2 3H1-H2 4H1-H2 5H1-H2 -0.05 -0.15 -0.08 -0.36 -0.06 0.74 0.95 0.88 0.92 0.75 -0.27 -0.32 -0.59 -0.34 0.33 0.93 0.92 1.16 1.74 1.43 -0.47 -1.05 -2.28 -5.17 -4.59 1.06 1.09 3.35 6.83 5.55  Table 43. H1-H2 results: Probability values from t-test (TiuC) 1H1-H2 2H1-H2 3H1-H2 4H1-H2 5H1-H2 modal-H vs. creaky-L 0.202 0.018 0.021 0.013 0.005 modal-H vs. modal-R 0.449 0.595 0.176 0.965 0.345 modal-R vs. creaky-L 0.586 0.050 0.071 0.013 0.003 H1-A1 results are parallel to H1-H2. Rising tone items pattern with modal-H and L (no significant differences), and are statistically different from creaky-L from intervals 3 to 5 for both speakers.  133  Figure 23. H1-A1 plot for mean results of both speakers. Table 44. H1-A1 results: Mean and standard deviation (LiaL) 1H1-A1 2H1-A1 3H1-A1 4H1-A1 5H1-A1 modal-H Mean -3.01 -3.51 -3.04 -3.06 -2.91 SD 4.77 3.70 3.62 3.59 3.70 modal-R Mean -0.63 -3.40 -4.35 -3.83 -1.60 SD 5.45 4.16 4.19 4.48 5.55 creaky-L Mean -2.14 -3.86 -8.68 -13.67 -9.64 SD 3.59 4.54 3.80 4.95 5.66 Table 45. H1-A1 results: Probability values from t-test (LiaL) 1H1-A1 2H1-A1 3H1-A1 4H1-A1 5H1-A1 modal-H vs. creaky-L 0.563 0.810 <0.001 <0.001 <0.001 modal-H vs. modal-R 0.199 0.936 0.350 0.592 0.437 modal-R vs. creaky-L 0.365 0.763 0.004 <0.001 <0.001 Table 46. H1-A1 results: Mean and standard deviation (TiuC) 1H1-A1 2H1-A1 3H1-A1 4H1-A1 5H1-A1 modal - H Mean -5.53 -7.27 -7.43 -7.62 -4.68 SD 3.43 4.67 3.58 4.16 5.38 modal - R Mean -3.30 -5.14 -6.49 -6.60 -1.22 SD 3.07 3.20 2.91 3.21 4.17 creaky - L Mean -4.39 -8.74 -11.43 -11.08 -8.77 SD 4.88 3.08 4.57 6.12 3.95  134  Table 47. H1-A1 results: Probability values from t-test (TiuC) 1H1-A1 2H1-A1 3H1-A1 4H1-A1 5H1-A1 modal-H vs. creaky-L 0.451 0.302 0.010 0.073 0.020 modal-H vs. modal-R 0.062 0.145 0.420 0.442 0.051 modal-R vs. creaky-L 0.456 0.002 0.001 0.016 <.001  4.3.2.3 Discussion As in the modal-L experiment, results for modal-R tokens show some inconsistency in the direction of both jitter and spectral tilt. While some of the small nonsignificant differences are consistent with very light laryngealization in modal-R tokens, others suggest the reverse tendency. Once again, this is expected if the voice quality of these vowels is modal. In more detail, jitter results clearly demonstrate the modal voice of rising tone items. Numbers and statistics are according to the expected results in this experiment. As for spectral tilt, the vowel pattern a’a (modal-R) showed no signs of laryngealization. Modal-R results at the middle interval were statistically different from those of creaky-L and similar to modal-H and L. The exception to this similarity was H1H2 for LiaL, where differences can be attributed to pitch; and regardless of the difference both modal-H and R are within the range of modal voice (with positive spectral tilt values).  4.4  Experiment 3: Falling tone with modal voice  At this point, we have reanalyzed Quiaviní Zapotec as a tonal language that contrasts two level tones, high and low, and one contour tone, rising, within modal voice. I now turn to falling tone. In the 33 vowel patterns in Munro and Lopez’ (1999) description 23 correspond to falling tone. There are two cases of vowel patterns with falling tone that seem to have modal voice on the basis of my fieldwork and preliminary  135  acoustic evidence: a’àa and a’aa’. The next section offers an acoustic description of some of these items, followed by an acoustic evaluation.  4.4.1 Acoustic description: Modal-F The following (near) minimal pairs include comparisons between falling tone items versus the other three lexical tones in Quiaviní Zapotec. (11) Modal-Falling (near) minimal pairs a. / aʒ /  ‘s/he’  vs.  / n-ʒaʒ /  ˩ ‘greedy’  b. / nkai / Ü  ‘dark’  vs.  / kai /  ˥ ‘street’  c. / -ɡelˑ / Ü  ‘by chance’  vs.  / ɡwelˑ /  Ë ‘chance, turn’  d. / ʒilj / Ü  ‘sheep’  vs.  / ʒilj /  Ë ‘saddle’  e. / bibj / Ü  ‘pipe (plant)’  vs.  / n-ʒibj /  ˥ ‘scared’  Ü  Figure 24 shows two examples of falling tone in Quiaviní Zapotec. 0.1575  0.1472  0  0  -0.1135  -0.1378 0  0.5631  0  0.405 Time (s)  Time (s)  500  500  300  300  200  200  150  150  100  100  70  70  50  0  0.5631  50  0  0.405  Time (s)  ʒ  i ˘  Time (s)  l  ʲ̥  300 500 30 10 200 150 200 100 150 70 50 100 30 20 70 15 -12 50 10 0  ʒ̊  i˘  lʲ  Figure 24. Waveform and spectrogram of / ʒilj / Ü ‘sheep’, by male speaker TiuR and female speaker LiaB.  0 Time (s)  136  0.405  In this example, TiuR’s pitch falls from 143 Hz to 117 Hz, whereas LiaB’s pitch is 218170 Hz. Overall results for falling tone for TiuR average a gliding curve of 146-116 Hz, whereas LiaB results show 220-181 Hz. The falling contour shape is distributed either along the whole vowel/rhyme or towards the second half. Additionally, the pitch of creaky-L tokens is generally lower than that of modal-F tokens (see Chapter 6, §6.4). In terms of voice quality, the periodicity of the sounds in Figure 24 is clear throughout both examples. Likewise, the spectrograms are clear and with no signs of laryngealization, particularly compared with prototypical creaky voice (see Chapter 6, §6.4).  4.4.2 Phonetic evaluation: Modal-F This is a post-experiment evaluation. Fewer tokens of hypothetical lexical items with modal voice and falling tone were included in the recordings for modal-L and -R tokens. Consequently, instead of conclusive experimental results, in what follows, I present a preliminary evaluation. The following lexical items were analyzed: Table 48. Stimuli: falling-tone evaluation Modal-F 1 2 3 4  a'àazh: gue'èell nca'ài zhi'ìilly  ‘s/he’ ‘by chance’ ‘dark’ ‘sheep’  Each of these items was recorded twice by female speaker LiaL, under the same conditions as the previous experiments. The jitter and spectral tilt results are presented in the following tables, in comparison with the control cases, modal-H and creaky-L tokens.  137  Table 49. Periodicity (jitter): Mean and standard deviation (LiaL) LiaL Jitter (ppq5) Jitter (ddp) modal-H Mean 0.224% 0.443% SD 0.111 0.298 modal-F Mean 0.325% 0.502% SD 0.111 0.298 creaky-L Mean 0.921% 1.141% SD 0.639 0.619 As with modal-L and modal-R, modal vowels with falling tone have low jitter values, similar to those of the control case modal-H. Likewise, modal-F tokens pattern with modal voice in terms of spectral tilt throughout the five different intervals considered for H1-H2 and H1-A1. As with previous cases under investigation, the nonmodal voice control case, creaky-L, departs from the positive values of modal-F from the third interval onwards. Table 50. H1-H2 results: Mean and standard deviation (LiaL) modal - H Mean SD modal - F Mean SD creaky - L Mean SD  1H1-H2 2H1-H2 3H1-H2 4H1-H2 5H1-H2 6.34 7.14 7.36 6.93 5.23 2.23 2.07 2.46 2.60 3.41 5.40 5.76 4.82 4.34 3.92 3.17 3.51 2.50 3.16 3.49 6.73 4.83 1.64 -3.66 -3.09 3.04 3.90 2.96 4.64 4.22  Table 51. H1-A1 results: Mean and standard deviation (LiaL) 1H1-A1 2H1-A1 3H1-A1 4H1-A1 5H1-A1 modal-H Mean -3.01 -3.51 -3.04 -3.06 -2.91 SD 4.77 3.70 3.62 3.59 3.70 modal-F Mean -1.42 -1.19 -1.06 -0.50 1.33 SD 4.09 3.86 3.57 4.81 5.75 creaky-L Mean -2.14 -3.86 -8.68 -13.67 -9.64 SD 3.59 4.54 3.80 4.95 5.66 The above results suggest that modal-F tokens have modal voice, and therefore, this completes the tonal inventory of modal voice in Quiaviní Zapotec.  138  4.5  Conclusions: Quiaviní Zapotec tonal inventory with modal voice  This final section concludes the chapter providing a complete picture of the reanalysis of tone with modal voice in Quiaviní Zapotec. The section includes comparisons of the Munro and Lopez (1999) vowel patterns, along with a final comprehensive illustration of Quiaviní Zapotec tone pitch contours for vowels with modal voice. Table 52 summarizes the vowel patterns from the Quiaviní Zapotec dictionary considered in this chapter, in parallel with my reanalysis of these vowels. On the left, I present Munro and Lopez’ (1999) orthography and tone, along with the proposed phonological transcription and tone. Table 52. Tone in modal voice: vowel pattern reanalysis Munro and Lopez (1999) orthography tone 1 aa H 2 àa L 3 a’a  R  Reanalysis phonemic tone /a/ H L /a/ R /a/  4 a’àa, a'aa'  F  /a/  F  Table 53 encodes the same information as Table 52, but with actual examples instead of only with the patterns. Within the reanalysis, another column is added to present the phonetic transcription. Table 53. Tone in modal voice: reanalysis with examples Munro and Lopez (1999) orthography tone gloss 1 daany H ‘harm’  Reanalysis phonemic phonetic / danj / ˥ [dáːɲ] [dàːɲ]  R  ‘mountain’ / danj / ˩ ‘father’ / dad / Ë  F  ‘s/he’  / aʐ / Ü  [ âːʐ ]  2 dàany  L  3 da’ad 4 a'àazh:  [dǎːð]  139  Based on the findings of this chapter, I conclude that modal voice may bear all tones in this language. The four contrastive tone categories in Quiaviní Zapotec are included in the following table. Table 54. Quiaviní Zapotec Tone and modal voice High Low Rising Falling Modal √ √ √ √ Finally, the following figure schematizes the four tone melodies in Quiaviní Zapotec. Means correspond to the production of 10 tokens of each category by male speaker TiuC (§6.4). This is an illustration of the overall shape of Quiaviní Zapotec tones.  Figure 25. Pitch average contours for modal vowels (TiuC). Having established the contrastive use of tone in Quiaviní Zapotec, with examples of all four tones on vowels with modal voice, the next chapter investigates the tonebearing unit in Quiaviní Zapotec, as well as the phonological representation of tone.  140  Chapter 5:  The tone-bearing unit in Quiaviní Zapotec: Moraicity and tone  5.1  Introduction  Under non-linear phonology (e.g. Autosegmental Phonology, Goldsmith, 1976), tone is represented on a separate tier from segmental and other prosodic material. A tone is only realized on the surface if it is associated with some segment or prosodic entity such as the syllable or the mora, on which it is eventually pronounced.74 A large amount of evidence in the literature has established the mora as the prosodic tone-bearing unit (TBU; Hyman, 1985; Pulleyblank, 1994; Jiang-King, 1999, among others). Moreover, there are languages in which the TBU is not just any mora, but those associated with vowels and sonorants only (Yip, 2002, p. 73; see Zec, 1988; and Steriade, 1991 for discussion). Taking into account this theoretical background, I assume that the mora is the TBU in Quiaviní Zapotec. The question remains, however, of how tone is manifested at the segmental level. In the previous chapter, it has been illustrated how vowels express 74  Except in that floating (L) tones, for example, are often taken to be realized in the form of downstep effects on a following (H) tone (Hyman & Schuh, 1974).  141  tone (being the optimal segments to do so), but Quiaviní Zapotec also has a wide variety of syllable rhymes, with the full inventory of consonants allowed in the coda. Of particular interest is the pervasive fortis/lenis distinction in the consonant inventory, a contrast that is found both in obstruents and sonorants. The goal of this section is to determine the segmental tone-bearing units in Quiaviní Zapotec, focusing on syllables with modal vowels only (in case other voice qualities may make it more difficult to isolate what is going on). Consequently, only the level tones (high and low) and rising tone will be considered, since falling tone has a restricted distribution with modal voice (i.e. few lexical items; see Chapter 4, §4.4). Since tone associates with the mora, only moraic segments will bear tone. Among the moraic segments, vowels clearly bear tone in Quiaviní Zapotec. Coda fortis consonants are also moraic and so coda fortis sonorants could in principle bear tone phonetically, but fortis obstruents cannot bear tone phonetically due to their voicelessness. Finally, since the prosodic affiliation of segments determines their tonebearing status, it follows that (non-moraic) lenis consonants (including sonorants) will not bear tone and that onset consonants of all sorts may not bear tone. I thus predict that fortis coda sonorants may be tone-bearing segments in Quiaviní Zapotec, along with vowels (cf. Arellanes, 2003). I now turn to the phonetic and phonological analysis of tone in Quiaviní Zapotec.  142  5.2  Tone-bearing segments in Quiaviní Zapotec  5.2.1 Obstruents The phonetics of tone requires voicing and, as mentioned above, the constriction that characterizes obstruent segments makes it very difficult, and impossible in some cases, for these sounds to bear tone phonetically. Since fortis obstruents are always voiceless, the lack of voicing prevents these segments from manifesting pitch (tone), even though they are moraic in coda position. It remains to be determined whether lenis obstruents are able to bear tone in this language. Lenis obstruents are voiced intervocalically, but may devoice word-initially and word-finally. In addition, these segments are analyzed as non-moraic (Chapter 3), based on the fact that vowels followed by lenis consonants become long in order to satisfy the bimoraic requirement of the minimal word. All in all, the characteristic stricture of lenis obstruents, the inconsistency of their voicing, and their non-moraic prosodic status lead us to predict that lenis obstruents do not bear tone in Quiaviní Zapotec. To investigate the possibility of tone with lenis obstruents, I carried out an informal acoustic investigation using the lexical items in Table 55, which include the lenis stops /b, d, ɡ/ and lenis fricatives /z, ʒ/ in coda position. I looked for two acoustic parameters of these items: (i) voicing; and (ii) consistency with the pitch of the vowel. This is not intended to be a formal acoustic analysis. Rather, examination of the pitch contour is intended as a supplement to by-ear transcription of the tone, to give the reader an idea of what is going on with pitch during the consonants (where no tone is perceived).  143  Table 55. Words with lenis stops / b, d, ɡ / and fricatives / z, ʒ /. 1 2  / ʐub /  3  / ʒjab / / dad /  4  / dad /  5 6 7 8 9  Ë → [ ʐuːβ ] ~ [ ʐuːɸ ] Ë → [ ʒjaːβ ] ~ [ ʒjaːɸ ] ˥ → [ dað ] ~ [ daθ ]  ‘dried corn kernel’ ‘bad’  ˥ → [ klutz ] ~ [ kluts ] ˥ → [ nɾaːʒ ] ~ [ nɾaːʃ ]  ‘Nicolas a’  Ë → [ ɡiːʒ ] ~ [ ɡiːʃ ]  ‘city person’  ‘dice’  Ë → [ dað ] ~ [ daθ ] ‘father’ / nlˑaɡ / ˩ → [ nlˑaːɣ ] ~ [ nlˑaːx ] ‘wide’ Ë → [ luːɣ ] ~ [ luːx ] ‘from San Lucas’ / lug / / ɡaz / ˩ → [ ɡaːz ] ~ [ ɡaːs ] ‘seven’ / klaz /  / nɾaʒ /  10 / ɡiʒ /  ‘orange’  Each word was produced three times in isolation by two male native speakers (TiuR, 50 years old, and TiuL, 35) for a total of 60 tokens (10 words x 3 repetitions x 2 speakers = 60). All lenis obstruents, both stops and fricatives, demonstrated the following patterns: they were produced as voiceless or partially voiceless; when they manifested pitch, it was inconsistent, dropping for the most part, and without continuation of the trajectory of the phonological tone manifested in the vowel. These characteristics held regardless of the type of tone, confirming the prediction that lenis obstruents are not tonebearing in Quiaviní Zapotec. As an illustration, Figure 26 shows a vowel with rising tone before a lenis “stop” realized as a (low-amplitude) fricative, spoken by a male speaker. From the middle to the end of the vowel, the pitch rises from 125 to 144 Hz. As soon as the lenis obstruent begins, the pitch becomes inconsistent. First, it slightly drops (138 Hz), then, it stays flat, and finally it shows a small rise. The lenis obstruent does not continue the shape of the phonological tone manifested in the vowel, nor does it show any different pitch contour of its own. In addition, the characteristic allophony of lenis consonants is particularly salient in coda position; thus, different F0 patterns were obtained with different tokens of a word. Apart from the voiced fricative realizations ([ð]), common allophones for lenis  144  plosives are voiceless fricatives ([ð̥]), where the lack of voicing prevents the expression of tone during the obstruent’s constriction.  0.08035  0  -0.1082  0  1.134 Time (s)  500 300 200 150 100 70 50  0  1.134 Time (s)  [ d  aː  ð ]  Figure 26. Waveform and spectrogram75 of / dad / Ë ‘father’, by male speaker TiuR. Lenis (high-amplitude) fricatives show the same inconsistency; they cannot manifest tone phonetically. In Figure 27, the example of / ɡiʒ / Ë ‘city person’ illustrates the behavior of lenis fricatives in coda position. During the vowel, we observe the pitch rising, but during the transition into the fricative, the pitch drops and disappears, as voicing fades out. The fricative is practically devoiced, thus unable to manifest tone. As voicing is variable for lenis consonants in final utterance position, other examples show a little more voicing in their production. However, the pitch is not sustained, neither consistent with the tone of the vowel nor consistent across different tokens of the same vowel.  75  As in previous sections, the spectrogram frequency is 0-5000 Hz. (except those containing alveolar fricatives which are 0-8000 Hz.), but since this chapter concerns tone, the pitch frequency is superimposed on the range of 50-500 Hz.  145  0.08374  0  -0.07495  0  0.6511 Time (s)  500 300 200 150 100 70 50  0  0.6511 Time (s)  x  i˘  Z̊  Figure 27. Waveform and spectrogram of / ɡiʒ / Ë ‘city person’, by male speaker TiuR.  5.2.2 Sonorants In contrast to obstruents, sonorants are cross-linguistically voiced by default and have an F0 that could in principle be raised or lowered enough to realize contrastive tone. Sonorant consonants may even constitute syllable nuclei in many languages and bear tone on their own (e.g. Bantu languages, Hyman & Schuh, 1974; Nieves Chinantec (Otomanguean), P. Hernández, personal communication, August 2008). Nonetheless, sonorant consonants in Quiaviní Zapotec are never syllabic, and therefore, all syllables must have a vowel bearing tone. The question is whether in addition to the vowels moraic (fortis) sonorants bear tone and whether nonmoraic (lenis) sonorants bear tone. Since the mora is the TBU, the prediction is that only fortis coda sonorants bear tone. In order to corroborate this prediction, I selected several lexical items with level and contour tones with both fortis and lenis sonorants in the coda (see tables below). As in the previous section, I carried out an informal acoustic investigation, examining the data with respect to: (i) voicing; and (ii) consistency with the pitch of the vowel. Once again, the examination of the pitch contour is intended as a supplement to by-ear 146  transcription of the tone. No experimental data is reported; instead, the following sections present the results of the analysis as a phonetically informed description. (The examples with long vowels in open syllables from the previous chapter were considered the control case, as a parameter of comparison for the tonal shapes in Quiaviní Zapotec.) The words I evaluate contain five lexical entries with fortis sonorants in coda and five with lenis sonorants, making a total of 10 words for each of the tones in consideration: high, low and rising. (Falling tone was excluded because it occurs mostly with non-modal vowels.) Within each comparison group, there is at least one item with a low vowel (/a/), and one item with a high vowel (/i/ or /u/). Two male native speakers of Quiaviní Zapotec (TiuR and TiuL) produced every word three times in isolation. In total, the words consisted of 180 tokens (5 words with a fortis coda sonorant + 5 words with a lenis coda sonorant x 3 tones x 3 repetitions x 2 speakers = 180 tokens). Table 56. Words with high tone (sonorants) VCfortis  VClenis  / njanˑ / ˥  →  [ njáńː ]  ‘Marcelo’  / tan / ˥  →  [táːn ]  ‘Cayetana’  / xalˑ / ˥  →  [xáĺː]  ‘job’  / danj / ˥  →  [dáːɲ ]  ‘harm’  / belˑ / ˥  →  [béĺː]  ‘Avelina’  / bal / ˥  →  [báːl]  ‘bullet’  / nˑdenˑ / ˥  →  [ndéńː]  ‘this (one)’  / nuan / ˥  →  [núːán]  ‘chirimoya’  / n-sualˑ / ˥  →  [nsúáĺː]  ‘blue’  / banɡual / ˥  →  [banɡúːál]  ‘oldʼ  Table 57. Words with low tone (sonorants) VCfortis  VClenis  / ɡalˑj / ˩  →  [ ɡàl̀ː ]  ‘twenty’  / danj / ˩  →  [ dàːɲ ]  ‘mountain’  / nalˑ / ˩  →  [ nàl̀ː ]  ‘is hung’  / nan / ˩  →  [ nàːn ]  ‘thick’  / tʃonˑ / ˩  →  [ tʃòǹː ]  / bdan / ˩  →  [ bdàːn ]  / nˑdenˑ / ˩  →  [ ndèǹː]  ‘three’ ‘that (one)’  / bkwel / ˩  →  [ bkwèːl ]  / bunˑj / ˩  →  [ bùɲ̀ː ]  ‘person’  / zinj / ˩  →  [ zìːɲ ]  ‘soot’ ‘corn husk’ (totomoztle) ‘spring (of water)’  j  147  Table 58. Words with rising tone (sonorants) VCfortis  VClenis  / damˑ / Ë  →  [ dàḿː ]  / manj / Ë  →  [ mǎːɲ ]  [ sàńːʒ ]  ‘owl’ ‘tame’  / sanˑʒ / Ë  →  / nan / Ë  →  [ nǎːn ]  ‘animal’ ‘mother’  / kanˑ / Ë  →  [ kàńː ]  ‘Alejandra’  [ tʃǎːn ]  ‘respectful greeting’  →  [ ɡwèĺː ]  ‘turn, chance’  / tʃan / Ë / bjol / Ë  →  / ɡwelˑ / Ë  →  [ bjǒːl ]  ‘agave flower bud’  / tʃinˑʒ / Ë  →  [ tʃìńːʒ ]  ‘bedbug’  / nɡwinj/ Ë  →  [ nɡwǐːɲ ]  ‘sickness’  5.2.2.1 High tone (sonorants) I present first the characteristics of rhymes consisting of vowel plus fortis sonorant (VCfortis). In terms of pitch, vowels expressing high tone may show an initial period of phonetic consonant pitch perturbation (raised pitch after voiceless consonants, lowered pitch after voiced ones), followed by a pitch level that is more stable and relatively flat. The fortis sonorant continues the tonal trajectory initiated by the vowel and maintains it during the majority of its duration. This is illustrated in Figure 28. 0.109  0  -0.1302  0  0.6771 Time (s)  500  0  0  0.6771 Time (s)  [  n  s  u  lˑ  ]  Figure 28. Waveform and spectrogram of / n-sualˑ ~ n-sulˑ / ˥ ‘blue’, by male speaker TiuR. In contrast, in rhymes formed by a vowel plus a lenis sonorant (VClenis), both the duration and the manifestation of pitch are different. Vowels are always long, whereas 148  the lenis consonants are short. For pitch, coda consonants do not show the same continuity with the vowel as their fortis counterparts. The most common pattern is that pitch drops in these cases. 0.2041  0.2648  0  -0.2845  0  0  1.191  -0.2554  Time (s)  0.9242 Time (s)  500  0  0  500  0  1.191 Time (s)  [  d  aː  0  0  0.9242 Time (s)  j  ɲ]  [ b  aː  l ]  Figure 29. Waveform and spectrogram of / danʲ / ˥ ‘harm’ and / bal / ˥ ‘bullet’, by male speaker TiuR. Figure 29 shows two examples with a coda lenis sonorant. In the case of / danʲ / ˥ ‘harm’, after a small initial rise (due to /d/), pitch is steady during the vowel, but begins to fall with the glide and continues to fall through the nasal. Because the phonological tone is manifested during the steady state of the vowel, the nasal does not need to maintain a flat F0, thus, the pitch lowering is the expected trajectory in utterance final position. The case of the liquid in / bal / ˥ ‘bullet’ is even clearer in showing the role of lenis consonants. The pitch is clear and sustained during the vowel duration; the liquid continues the pitch trajectory for a few pitch periods and then it suddenly drops and voicing disappears. In summary, these examples suggest that lenis sonorants do not bear phonological tone whereas fortis ones do.76  76  The possibility of lenis consonants bearing a L tone is rejected below.  149  5.2.2.2 Low tone (sonorants) With respect to low tone, let us start with a particular example. Figure 30 shows an interesting comparison between two types of rhymes in Quiaviní Zapotec, both in terms of duration and pitch. The first one is the word / nda / ˩ ‘sensitive’ on its own, which consists of an open syllable, hence, with a rhyme made up of a single vowel (V). The spectrogram on the right corresponds to the same word plus the 3s clitic (child) / =ɨmˑ /, which forms in this case a rhyme with a vowel and a fortis sonorant (VCfortis).  0.3881  0.09125  0  0  -0.5403  0  0.8086  -0.1402  0  0.548  Time (s)  Time (s)  500  500  300  300  200  200  150  150  100  100  70  70  50  0  0.8086  50  0  0.548  Time (s)  [  a̰  a  n d  Time (s)  aː  ]  [  a  n  d  a  mː  ]  Figure 30. Waveform and spectrogram of / nda / ˩ ‘sensitive’, by male speaker TiuL. The first one shows the word on its own, and the second example includes the 3s clitic (child) / =ɨmˑ /. The vowel in the first spectrogram expresses the low tone throughout its entire duration. Apart from the little phonetic perturbation at the beginning, the pitch is stable, averaging 110 Hz. The second spectrogram suggests that tone is manifested in both the vowel and the consonant. The pitch shape initiated by the vowel continues stably into the consonant for its entire duration. These characteristics exemplify the prosodic bimoraic requirement of the minimal prosodic word. In the first case, the vowel is the only segment in the rhyme, thus, it is the only prosodically active element. It is lengthened in 150  order to satisfy minimality and tone is expressed fully. In the second case, both the vowel and the consonant are moraic and both manifest the phonological tone.77 For lenis sonorants, the case is the same as the one outlined above for high tone; namely, they do not show continuity with the vowel pitch. The pitch expressed in the lenis sonorants is normally irregular and commonly drops. An example is given in Figure 31, which corresponds to the word / bdan / ˩ ‘soot’. The vowel last 177 ms and averages a pitch of 136 Hz, whereas the consonant shows no pitch track and lasts ~70 ms. When I plotted the pitch by hand, the result was a lowering of about 20 Hz compared to the vowel, and with considerable irregularity. 0.2271  0  -0.2787  0  0.7215 Time (s)  500  0  0  0.7215 Time (s)  [  a β ð  aˑ  n  ]  Figure 31. Waveform and spectrogram of / bdan / ˩ ‘soot’, by male speaker TiuR.  77  This set of examples shows the complementary distribution of vowel length: long in open syllables and before lenis consonants and short with fortis coda consonants. In more detail, the difference in duration is noticeable in this example. The vowel lasts 274 ms. in the open syllable, and 89 ms. in the closed one. In the latter case, the coda compensates for the duration of the rhyme, lasting 133 ms. (for a total rhyme duration of 222 ms.).  151  5.2.2.3 Rising tone (sonorants) The last type of tone to consider is the rising contour tone. It shows the same characteristics outlined above for the level tones with respect to sonorants in coda position. In addition, the rising contour tone adds crucial evidence to support the claim that fortis coda sonorants are the only tone-bearing consonants: these consonants continue the pitch trajectory of the preceding vowel, and often it is during the coda consonant that the pitch rise takes place. On the other hand, lenis sonorants normally do not show continuity with the vowel pitch. Figure 32 provides examples of words with a vowel-fortis sonorant sequence in the rhyme. For the word on the left, / damˑ / Ë ‘owl’, the pitch starts to rise only towards the end of the vowel, but the most noticeable rise occurs throughout the fortis nasal. The average pitch during the vowel portion is 108 Hz (very close to the average for low tone tokens for this speaker, 110 Hz). At the mid point of the vowel, the pitch is 106 Hz, and at the end point it has risen only to 112 Hz. From there, the nasal continues rising until 144 Hz. The rise during the vowel portion is too small on its own to be interpreted as a contour; the whole rhyme is used to create the contour tone. We observe the same characteristics for the word on the right, / tʃinˑʒ / Ë ‘bedbug’. The vowel has a quite flat pitch averaging 155 Hz, and only rises slightly at the end. It is during the nasal where we find a salient rise, from 158 Hz to 205 Hz.  152  0.114  0.6468  0  0  -0.1418  0  0.4032  -0.6647  0  0.6966  Time (s)  Time (s)  500  500  300  300  200  200  150  150  100  100  70  70  50  0  0.4032  50  0  0.6966  Time (s)  [ d  aˑ  mˑ  Time (s)  ]  [  tʃ  iˑ  nː  t ʒ ̥ ]  Figure 32. Waveform and spectrogram of / damˑ / Ë ‘owl’, by male speaker TiuL. And waveform and spectrogram of / tʃinˑʒ / Ë ‘bedbug’, by male speaker TiuR.  The hypothesis that fortis sonorants are the only consonants capable of bearing tone in Quiaviní Zapotec entails that in any other syllable without a fortis coda sonorant, only vowels will bear the tone, including contour tones. Having this consideration in mind, it seems important to compare the above case (rising tone with fortis coda sonorant) with a rhyme with a fortis obstruent to confirm that the shape of the tone is realized during the vowel production only. In the word / mes / Ë ‘table’, in Figure 33, we observe that there is no manifestation of pitch during the long (more than 300 ms) obstruent coda. Instead, the realization of tone is entirely located during the vowel production, as predicted. Contrary to the vowel of / damˑ / Ë ‘owl’ in Figure 32, which practically has a flat tone, the vowel in / mes / Ë ‘table’ shows a clear rising contour. At the beginning, there is a 34 ms period of flat pitch of 128 Hz., and then it takes about 115ms to rise to 156 Hz.  153  0.3656  0  -0.4512  0  1.089 Time (s)  500 300 200 150 100 70 50  0  1.089 Time (s)  [  m  eˑ  sː  ]  Figure 33. Waveform and spectrogram of / mes / Ë ‘table’, by male speaker TiuR. As demonstrated for lenis consonants in codas, either obstruents or sonorants, their duration is short and the pitch is not consistent with the vowel. Similar to Figure 27 above, in Figure 34 the vowel is long and the pitch contour takes place during its duration; during the production of the lenis coda, the trajectory of the pitch changes (drops). The change in slope is particularly abrupt in the case of the nasal. 0.09848  0.6391  0  0  -0.1199  0  0.4498  -0.6623  Time (s)  1.162 Time (s)  500  500  300  300  200  200  150  150  100  100  70 50  0  70 0  0.4498 Time (s)  [  m  aːj  50  0  1.162 Time (s)  ɲ]  [  ʐ  uː  β  h  ]  Figure 34. Waveform and spectrogram of / manj / Ë ‘animal’, by male speaker TiuL, and waveform and spectrogram of / ʐub / Ë ‘dried corn kernel’, by male speaker TiuL.  154  The examples in Figure 34, with long vowel plus lenis coda consonant, have a similar pitch pattern to that of vowel plus fortis sonorant sequences (Figure 32), where the pitch is realized in the entire rhyme. Contrastively, the shape of the rising tone is somewhat reduced in sequences of vowel plus fortis obstruent (Figure 33). There were a few tokens in which lenis sonorants continue the pitch contour started in the vowel, but it is precisely this inconsistency that demonstrates that lenis coda consonants do not bear phonological tone in Quiaviní Zapotec. Furthermore, when a H tone follows these lenis consonants, e.g. the 1st person clitic /-aʔ/, as in [ʐǔːβáʔ] ‘my corn’, then the lenis consonant shows continuation with the phonological tone manifesting a high pitch. This is consistent with the fact that lenis coda consonants do not have L tone —despite the tendency for dropping the pitch; rather, they simply show phonetic inertia to their context. Once again, it is clear that fortis sonorants show continuity with the vowel in the expression of rising tone, whereas lenis sonorants do not.  5.2.3 Discussion The evidence presented here supports the hypothesis that fortis coda sonorants bear tone in Quiaviní Zapotec. Nonetheless, it is necessary to discuss some aspects of this issue. In the case of level tones, although the pitch trajectory is continued during fortis sonorants, it could be argued that the vowel on its own expresses the phonological tone, and the pitch found in the coda consonant is simple phonetic inertia. However, this consistency with vowel pitch does not take place with lenis sonorants. Moreover, there are cases in which it is necessary to include the fortis sonorant as a tone-bearing unit together with the vowel. This is the case for rising contour tones, where all or most of the rise takes place during the consonant. On the other hand, the data confirms that it is not necessary to include lenis sonorants for the expression of the phonological tone in Quiaviní Zapotec. In the cases analyzed here, lenis sonorants are short, many of them have low amplitude and weak formant frequencies, and practically all tend to cause the pitch to drop. This pitch lowering is common word-finally (words were recorded in isolation). If another word 155  follows, lenis sonorants may have a different pitch shape. This inconsistency is crucial to support their lack of phonological tone. In the case of the level tones, the lenis sonorants rarely continue the flat or level pitch started in the vowel. In the case of rising tone, this pitch disruption is even more noticeable as the pitch lowering goes against the trajectory of the phonological tone. In brief, the pitch of lenis sonorants is not manipulated to bear tone. Vowels with lenis codas or in open syllables are long, and their duration is sufficient to clearly manifest tone. In terms of syllable structure, the fact that vowels and some coda consonants bear tone indicates that tone may be located in the whole rhyme. The fact that some segments are not able to bear tone in coda is related to their specific articulatory characteristics and prosodic status. Obstruents (fortis and lenis) have a significant constriction and lack of formant structure. Lenis sonorants are normally short, sometimes devoiced and their formant structure is weak. These circumstances make it difficult, or even impossible, to achieve the necessary characteristics to express tone. Prosodically, although fortis obstruents are claimed to be moraic (Chapter 3), they are unable to manifest pitch due to their voicelessness. As for lenis consonants, I described them as non-moraic (Chapter 3), mainly based on their short duration. Their inability to bear phonological tone provides additional evidence for this prosodic characterization.  5.2.4 Conclusion Table 59 summarizes the phonetic characteristics of coda segment comparison in Quiaviní Zapotec; all of them apply to the different tones analyzed here. Table 59. Phonetic characteristics of coda segments comparison in Quiaviní Zapotec Fortis obstruents Voiceless (no pitch)  Lenis obstruents Inconsistent voicing & pitch  Fortis sonorants Long Manipulation of pitch Continue vowel pitch trajectory or carry latter half of pitch contour  Lenis sonorants Short Pitch drops (tendency) Independent of vowel pitch Low amplitude and weak formant frequencies  156  As illustrated, there is a split between fortis and lenis sonorants, with only the former presenting the necessary phonetic characteristics to bear phonological tone. The implication of these findings is that the feature [+sonorant] is not enough for a segment to bear tone in Quiaviní Zapotec; the necessary conditions to do so are to be moraic (fortis) and [+sonorant]. This hierarchy is represented in the following table. Table 60. Tone-bearing segments in coda in Quiaviní Zapotec Coda type  Moraic segments  fortis obstruent → fortis obstruent lenis obstruent fortis sonorant → fortis sonorant → lenis sonorant  Tone-bearing coda  fortis sonorant  The TBU in Quiaviní Zapotec is the mora associated with vowels and fortis sonorants in coda: these segments obligatorily express phonological tone in this language. The formal expression of this pattern in terms of a constraint-based grammar will be presented in the following section. Finally, and as mentioned above, this section has focused on the TBU in Quiaviní Zapotec modal voice. The assumption is that the prosodic and segmental characteristics outlined in here will apply to the expression of tone in non-modal vowels (see Chapters 6 & 7).  157  5.3  Tone representation in Quiaviní Zapotec and formal account  Chapter 4 established the tonal inventory of Quiaviní Zapotec in modal voice, and the preceding sections of this chapter established the segmental distribution of tone, as well as how tone is implemented phonetically. The goal of this section is twofold: first, to map the phonetic characteristics previously defined onto a phonological representation, adopting moraic theory (Hyman, 1985; McCarthy & Prince, 1986; Hayes, 1989); and second, to provide a grammatical account of the patterns observed in this language. The overall analysis is presented within the framework of Optimality Theory (Prince & Smolensky, 2004 [1993]). The analysis is restricted to monosyllabic roots (the majority in the language). However, an important comment regarding larger domains is that tone in Quiaviní Zapotec shows little or no mobility. As long as the syllable is prominent, level and contour tones remain within the root in bigger forms. Consequently, I assume that tone is underlyingly anchored to the root. As shown in Chapter 3, vowels have one mora before fortis consonants and two moras before lenis consonants in monosyllables. A single (level) tone is linked underlyingly to the vowel, as on the left of (1). When a second mora is inserted in the output (due to minimality), tone spreads to it (1b, c & d), unless prevented by feature incompatibility (1a).  158  (1) Level tone configuration (T = H or L) Input  Output  a. T | µ | C V Ofortis  T | µ µ | | → C V Ofortis  b. T | µ | C V Olenis  T |\ µ µ |/ → C V Olenis  c. T | µ | C V Rfortis  T |\ µ µ | | → C V Rfortis  d. T | µ | C V Rlenis  T |\ µ µ |/ → C V Rlenis  Contour tones are standardly analyzed as complex: HL (falling) or LH (rising) (e.g. Akinlabi, 1985; Akinlabi & Liberman, 1995). As mentioned above, based on the fact that roots with rising and falling tones are lexically contrastive, and that tones always remain within the root, I assume that the tones of the contour sequence are linked underlyingly to the only underlying mora (left part of (2)). To a certain extent this is only an assumption for convenience in monosyllables. It could be argued that only the first tone of the sequence is linked underlyingly, or neither, but regardless of this assumption, the constraint ranking presented below correctly accounts for the surface patterns as optimal outputs. Nonetheless, in most languages contour tones are bimoraic, with each tone associating with a different mora (Zhang, 2001). Quiaviní Zapotec illustrates this preference: when a second mora is inserted, the second tone links to it (2b, c & d)— except when the second mora is attached to a fortis obstruent (2a). Before fortis  159  obstruents, Quiaviní Zapotec has a short vowel with a contour tone (two tones linked to one mora), which is typologically unusual. (2) Contour tone configuration (T1T2 = LH or HL) Input  Output  a. T1 T2 |/ µ | C V Ofortis  T1 T2 |/ µ µ | | → C V Ofortis  b. T1 T2 |/ µ | C V Olenis  T1 T2 | | µ µ |/ → C V Olenis  c. T1 T2 |/ µ | C V Rfortis  T1 T2 | | µ µ | | → C V Rfortis  d. T1 T2 |/ µ | C V Rlenis  T1 T2 | | µ µ |/ → C V Rlenis  The above input and output representations are motivated on the basis of the phonetics-phonology mapping. Both the moraicity of segments and how tone is implemented in the phonetics of the language (timing patterns) were taken into account. Specifically, all input forms are monomoraic. It is commonly assumed that single vowels are monomoraic, long vowels bimoraic, and consonants are not moraic underlyingly except for geminates. Under an Optimality theory approach, this is not the only possibility (cf. Richness of the base, Prince & Smolensky, 2004 [1993]), but it is the best assumption in light of lexicon optimization (Prince & Smolensky, 2004 [1993];  160  McCarthy, 2002). As presented in Chapter 3 (§3.3), monosyllabic roots do not always surface as bimoraic; in non-prominent syllables and some suffixed words vowels are monomoraic. This variation then suggests the underlying root monomoraicity, rejecting the apparent violation of lexicon optimization in (1) and (2). In order to formally account for the facts in (1) and (2), I present first some assumptions for the analysis. The representations in (1) and (2) assume that there are no floating tones in Quiaviní Zapotec. In addition, no tones are deleted or inserted in Quiaviní Zapotec roots.78 This is formalized by the constraints below (cf. Pulleyblank, 1997, p. 79; Myers, 1997; Yip, 2002, p. 79). (3) *FLOAT 79 Every tone must be associated to some mora (TBU) (No floating tones) (4) MAX-TONE Input tones have output correspondents (No deletion of tones) (5) DEP-TONE Output tones have input correspondents (No insertion of tones) I assume these constraints are undominated in Quiaviní Zapotec grammar. For simplicity, they are not included in the following sections. Furthermore, as the moraicity of segments is crucial in determining their tone association, the formal account of monosyllables in Quiaviní Zapotec from Chapter 3 (§3.2.3) is included. Below, I repeat the ranking and constraint definitions. (6) FT-BIN, *Lenis-µ >> WBYP >> DEP-µ (7) FT-BIN Feet are binary under moraic or syllabic analysis (8) *Lenis-µ If lenis then non-moraic  78  Different tonal processes take place at the morphological level, as in verb inflection and with person clitics, among others. See Munro et al (2008). 79 This constraint evokes the second well-formedness condition from Goldsmith (1976), which states that ‘Every tone must be associated to some TBU’.  161  (9) WEIGHT BY POSITION (WBYP) Coda consonants are moraic (10) DEP-µ Output moras have input correspondents (No insertion of moras)  5.3.1 Level tones The first type of rhyme to consider for level tones is a short vowel followed by a fortis obstruent in coda (VOfortis). Examples of this type of rhyme are provided below. (11) Rhyme: VOfortis a. / baµk / ˥ b. / meµs / ˥  → [ báµkµ ]  ‘person from Tlacolula’  → [ méµsµ ]  ‘professor’  Fortis obstruents, despite being moraic, are unable to bear tone; hence, their mora is unspecified for tone: (12)  H | µ | C V Ofortis  H | µ µ | | → C V Ofortis  The restriction that obstruents cannot express tone is common cross-linguistically (Yip, 2002) and is encoded by the following markedness constraint (this constraint would also prevent tone on lenis obstruents, which at any rate are non-moraic): (13) *[-SON][TONE]80 No tones on obstruents  (Yip, 2002, p. 80)  This constraint is undominated and outranks the markedness constraint SPECIFY T that penalizes any mora (TBU) that is not associated with a tone. 80  Formally, "Tone" here refers to tonal autosegments in QZ (H & L).  162  (14) SPECIFY T (cf. Myers, 1997, pp. 861-863; Yip 2002, p. 83) 81 A mora must be associated with a tone (15) Fortis obstruent coda – level tone H FT-BIN *Lenis-µ *[-SON][TONE] WBYP SPECIFY T | /baµk/ ‘tlacolula’ a. H *! * | baµk b. H * | baµkµ c. H *! |\ baµkµ d. H *! |\ baµµk  DEP-µ  * * *  Candidate (a), the faithful candidate, violates minimality (FT-BIN) as well as the constraint WBYP that requires coda consonants to have a mora. Candidate (b), the winning candidate, inserts a mora for the /k/ and thus satisfies WBYP. This mora does not associate with the tone, in violation of SPECIFY T, which is low ranked. Candidate (c) incurs a fatal violation of *[-SON][TONE], which penalizes tone on obstruents. Finally, candidate (d) satisfies minimality and SPECIFY T, but at the cost of fatally violating WBYP (which illustrates the crucial ranking WBYP >> SPECIFY T). The second type of rhyme is a vowel followed by a lenis obstruent. (16) Rhyme: VOlenis (level tone) a. / daµd / ˥ → [ dáµµd ]  ‘dice’  b. / nɾaµʒ / ˥ → [ nɾáµµʒ ] ‘orange’  81  This constraint is in the spirit of the first well-formedness condition of Goldsmith (1976), which states that ‘Every TBU must have a tone’.  163  (17)  H | µ | C V Olenis  H |\ µ µ |/ → C V Olenis  As lenis consonants are not moraic in Quiaviní Zapotec, the vowel lengthens to satisfy minimality (FT-BIN). The inserted mora is attached to the vowel and so it is allowed to link to a tone satisfying SPECIFY T. Nonetheless, this new association entails other constraint violations. Tones are preferably associated with only one mora, as stated by *LONGT. (18) *LONGT A tone may be associated with at most one mora In addition, the constraint that penalizes associations that deviate from the input is DEPPATH (Pulleyblank, 1996), here formulated as DEPPATH(T). Conversely, the constraint that prevents loss of tone associations is MAXPATH(T). (19) Tone-mora faithfulness constraints a) DEPPATH(T) Any output path between a tone and an anchor (mora) must have a correspondent path in the input. b) MAXPATH(T) Any input path between a tone and an anchor (mora) must have a correspondent path in the output.  164  (20) Lenis obstruent coda – level tone FT-BIN *Lenis *[-SON] WBYP SPEC T H -µ [T] | / nɾaµʒ/ ‘orange’ a. H *! * | nɾaµʒ b.  * H |\ nɾaµµʒ c. H * *! | nɾaµµʒ d. H *! *! |\ nɾaµʒµ e. H *! * | nɾaµʒµ  DEP -µ  *LONGT  DEP PATH(T)  *  *  *  *  *  MAX PATH(T)  *  *  *  The faithful candidate (a) violates the requirement of a prosodic word to form a bimoraic foot, along with WBYP. The rest of the candidates satisfy minimality by inserting a mora, in violation of DEP-µ. The optimal candidate (b) and candidate (c) violate WBYP, but the latter also violates SPECIFY T. In turn, this violation allows us to rank SPECIFY T over *LONGT and DEPPATH(T) (the latter violated by the optimal candidate). Candidates (d) and (e) are eliminated as they violate *Lenis-µ. This ranking also accounts for the remaining types of rhyme, with fortis and lenis sonorants in coda. The third type of rhyme is a vowel followed by a fortis sonorant: (21) Rhyme: VRfortis (level tone) a. / nˑdeµnˑ / ˥ → [ndéµńµ] ‘this (one)’ b. / n-suaµlˑ / ˥ → [nsúáµĺµ]  ‘blue’  c. / nˑdeµnˑ / ˩ → [ ndèµǹµ] ‘that (one)’ d. / buµnˑj / ˩ → [ bùµ ɲ̀µ ] ‘person’  165  (22)  H | µ | C V Rfortis  H |\ µ µ | | → C V Rfortis  As in the case above, the optimal candidate incurs violations of *LONGT, DEP-µ, and DEPPATH(T), all constraints that are low-ranked, as shown in (23). (Also note that the constraint *[-SON][TONE] plays no role in evaluating these cases, and for that reason it is left out of the tableau.) (23) Fortis sonorant coda – level tone F TL BIN | /buµnˑj/ ‘person’ a. L *! | buµ ɲ b. L |\ buµµ ɲ c. L | buµ µ ɲ d.  L |\ buµ ɲµ e. L | buµ ɲµ  *Lenis WBYP -µ  SPEC T  DEPµ  *LONGT  DEP PATH(T)  *  *  *  *  *  MAX PATH(T)  *  *!  *!  *  *  *  *!  *  166  The final type of rhyme is a vowel followed by a lenis sonorant coda: (24) Rhyme: VRlenis (level tone) a. / nuaµn / ˥ → [núµáµn] b. / banɡuaµl / ˥ → [banɡúµáµl] c. / bkweµl / ˩  → [ bkwèµµl ]  d. / ziµnj /  → [ zìµµ ɲ ]  (25)  ˩  H | µ | C V Rlenis  ‘chirimoya’ ‘oldʼ ‘corn husk’ (totomoztle) ‘spring (of water)’  H |\ µ µ |/ → C V Rlenis  Once again, this candidate incurs the violations of *LONGT, DEP-µ, and DEPPATH(T). The formal account for this type of rhyme is identical to that of lenis obstruents, presented in tableau (20).  5.3.2 Contour tones Within contour tones, probably the most interesting case is that of rhymes formed by a short vowel followed by a fortis obstruent, where each segment in the rhyme has a mora, but the contour is fully realized on the vowel. As the obstruent consonant is not able to bear tone, the contour tone must then be associated entirely with the short vowel, on a single mora. (26) Rhyme: VOfortis (contour tone) a. /nɡaµs/ Ë → [nɡǎµsµ] ‘black’ b. / ʒjeµt / Ë  → [ ʒjěµtµ ]  ‘cat’  c. /meµs/  → [měµsµ]  ‘table’  (27)  Ë  LH |/ µ | C V Ofortis  LH |/ µ µ | | → C V Ofortis  167  The mora of the coda consonant is prevented from taking the H tone by *[-SON][TONE], forcing the contour tone to associate entirely with one mora, thus violating SPECIFY TONE (as the mora in the coda is not associated with any tone). This scenario and ranking is very similar to that of a level tone with a VOfortis rhyme; however, in the case of the contour tone there are two tones that require association with a mora (TBU), violating the constraint *CONTOUR, as defined in (28). (28) *CONTOUR A mora may be associated with at most one tone 82 This analysis makes a clear prediction in terms of the phonetics-phonology mapping, already confirmed in §5.2 The temporal profile of the contour is different between the VOfortis and the other type of rhymes. For the former, the shape of the contour tone must be expressed fully in the short vowel, and so the slope is steeper. For the remaining rhymes (with lenis obstruent, and fortis and lenis sonorant codas), the contour tone is always realized as long, with each portion of the tone associated with a mora. Consequently, the rhyme VOfortis provides evidence that contour tones in Quiaviní Zapotec may be realized on short vowels, on a single mora, against the common typological tendency to have contour tones only on long vowels (Zhang, 2001; Zoll, 2004, p. 236).  82  This is equivalent to the following formulation (and representation). *T T ONET/M: One tone per mora (Zhang, 2001, p. 2) \/ µ  168  (29) Fortis obstruent coda – contour tone 83 LH FT- *Lenis *[-SON] WBYP SPEC T BIN -µ [TONE] |/ /nɡaµs/ a. L H *! * |/ nɡaµs b.  * LH |/ nɡaµsµ c. L H *! | | nɡaµsµ d. LH *! || nɡaµµs  DEP -µ  *CON  DEP PATH(T)  MAX PATH(T)  *  *  *  *  *  *  TOUR  *  *  *  In the following three types of rhymes with contour tones, the second tone of the contour — originally associated with the underlying mora — is reassociated with the inserted mora, in violation of both DEPPATH(T) and MAXPATH(T). I present examples of the remaining types of rhymes, followed by its moraic representation and tableaus. (30) Rhyme: VOlenis (contour tone) a. / ʐuµb / Ë → [ ʐǔµµb ]  ‘dried corn kernel’  b. / zhyaµb / Ë → [ zhyǎµµb ]  ‘bad’  c. / daµd /  Ë → [ dǎµµd ]  ‘father’  d. / ɡiµʒ /  Ë → [ ɡǐµµʒ ]  ‘city person’  (31)  LH |/ µ | C V Olenis  LH | | µ µ |/ → C V Olenis  83  Since the constraint *LONGT plays no role evaluating candidates with a contour tone, it is left out of the following tableaus for reasons of space and clarity.  169  (32) Rhyme: VRfortis (contour tone) a. / daµmˑ / Ë → [ dàµḿµ ]  ‘owl’  b. / saµnˑʒ / Ë  →  [ sàµńµʒ ]  ‘tame’  c. / ɡweµlˑ / Ë  →  [ ɡwèµĺµ ]  ‘turn, chance’  d. / tʃiµnˑʒ / Ë  →  [ tʃìµńµʒ ]  ‘bedbug’  (33)  LH |/ µ | C V Rfortis  LH | | µ µ | | → C V Rfortis  (34) Rhyme: VRlenis (contour tone) a. / maµnj / Ë → [ mǎµµ ɲ ] ‘animal’ b. / tʃaµn / Ë → [ tʃǎµµn ] (35)  LH |/ µ | C V Rlenis  ‘respectful greeting’  LH | | µ µ |/ → C V Rlenis  170  (36) Lenis obstruent (and sonorant) coda – contour tone (/ʐub/ Ë ‘dried corn’) L H F TBIN |/ /ʐuµb a.LH *! |/ ʐuµb b.LH |/ ʐuµbµ c. LH || ʐuµµb d.LH |/ ʐuµµb e.LH | | ʐuµbµ  *Lenis -µ  *[-SON] [T]  WBYP  SPEC T DEP -µ  *  *!  *  *!  *!  TOUR  DEP PATH(T)  *  *  *  *CON  *  *  *  *  *  *  *!  MAX PATH(T)  *  *  *  *  *  *  *  The analysis of roots with lenis obstruents and sonorants in coda is almost identical. The only difference is that a hypothetical candidate with a moraic lenis sonorant would not violate *[-SON][T], as candidate c. does in this tableau. Nonetheless, the constraint *Lenis-µ eliminates candidates with moraic lenis consonants, both obstruents and sonorants.  171  (37) Fortis sonorants coda – contour tone (/damˑ/ Ë ‘owl’) FT- *Lenis *[-SON] WBYP SPEC T LH BIN -µ [T] |/ /daµmˑ/ a. L H *! * |/ daµm b. LH *! |/ daµmµ c. LH *! || daµµm d. L H | | daµmµ  DEP -µ  *  *CON DEP TOUR PATH(T)  MAX PATH(T)  *  *  *  *  *  *  *  *  *  To conclude this section, the following diagram shows the final ranking and dominance relationship among the employed constraints. (38) Constraint dominance (TBU) FT-BIN  *L⇔µ  *[-SON][TONE]  WbyP SPECIFY T DEP-µ  *LONGT, *CONTOUR, DEPPATH(T), MAXPATH(T)  172  5.4  Conclusions  Relating the metrical structure analysis of Chapter 3 and the tone findings from Chapter 4, this chapter has established the association between moraicity and the tonal patterns in Quiaviní Zapotec. Assuming the mora as the tone-bearing unit (Hyman, 1985; Pulleyblank, 1994), I showed that only vowels and fortis coda sonorants bear tone in this language. This follows from the phonological analysis of fortis consonants as moraic in coda position (Chapter 3), and the typological tendency for avoiding tone on obstruents (i.e. *[-SON][TONE]), even when moraic (fortis). These segmental restrictions lead to contour tones being associated with only one or two moras depending on the type of rhyme (cf. the typology of contour tones and the statement ‘One tone per mora’ (Zhang, 2001, p. 2)). The proposal was supported by acoustic data and encoded formally into an OT grammar. In this chapter, I analyzed tone at the (monosyllabic) root level only. Polysyllabic forms, including prefixes, suffixes and clitics, raise interesting issues that need to be considered in further work. Preliminaries in this respect are presented in the concluding chapter of this dissertation.  173  Chapter 6:  Non-modal phonation in Quiaviní Zapotec  6.1  Introduction  Quiaviní Zapotec has a cross-linguistically uncommon four-way phonation contrast between modal /a/, breathy /a̤/, creaky /a̰/ and interrupted /aʔ/ vowels (Munro and Lopez 1999). Of particular interest is the distinction between creaky and interrupted voice, a phonetic distinction that is rarely used contrastively cross-linguistically (Ladefoged & Maddieson, 1996). I provide new phonetic and phonological evidence that supports these contrasts, and propose a novel analysis of the tone-phonation interaction in this language. Departing from Munro and Lopez (1999), Chapter 3 demonstrated that tone is used contrastively in Quiaviní Zapotec, showing that modal vowels —the default phonation type— may be associated with all four tones in this language (high, low, rising and falling). Within non-modal vowels, I propose in this chapter that breathy vowels are restricted to syllables with low and falling tones, whereas creaky and interrupted vowels appear with high, low and falling tones. That creaky and interrupted vowels can bear the same tones means that the distinction between them cannot be derived phonologically from tonal differences. The goal of this chapter is to present descriptive generalizations  174  governing tone and non-modal phonation in Quiaviní Zapotec; theoretical consequences, such as the featural specification and phonological representation of these vowels, are addressed in Chapter 7. I begin with a general overview of phonation types in the world’s languages, focusing on typological diversity. I then dedicate a separate section to each non-modal phonation type in Quiaviní Zapotec: breathy, creaky and interrupted vowels. Issues in these analyses include the interaction between phonology and phonetics: how contrastive tone and phonation are manifested phonetically. These sections all follow the same structure. First, I provide a description of these vowels based on the tones they interact with, along with minimal pairs. Second, I present the original analysis of these vowels in Munro and Lopez (1999). Third, I compare and justify the differences between the two approaches. For the creaky and interrupted vowels sections, this comparison is accompanied by an acoustic evaluation to quantitatively validate the proposed classification. Once the properties of these types of vowels are determined, I present a further comparison of the laryngeal vowels, confirming the phonological contrast between creaky and interrupted vowels. This study finishes with a typological discussion of these findings.  6.2  Brief typology of phonation types  This section presents a cross-linguistic overview of how different languages exploit voice quality contrasts. The goal is to contextualize the typological relevance of Quiaviní Zapotec phonation types, as well as to present some preliminaries to the phonetic and phonological description of subsequent sections. Ladefoged (1971) suggested that there might be a continuum of phonation types —the manners in which the vocal folds may vibrate— defined in terms of the aperture between the arytenoid cartilages, ranging from voiceless (furthest apart), through breathy voiced, to regular modal voicing, and then on through creaky voice to glottal closure (closest together).  175  Most open Phonation type  Voiceless  Most closed Breathy  Modal  Creaky  Glottal closure  Figure 35. Continuum of phonation types (Ladefoged 1971) Languages exploit different points on the continuum to manifest linguistic oppositions. The contrastive use of phonation types in vowels include two-, three- and (rarely) four-way contrasts. Two-way contrasts systems are relatively common. For instance, Hmong (Huffman, 1987) and Gujarati (Fisher-Jorgensen, 1967) make a contrast between breathy and modal voice, whereas Totonac (Alarcón, 2008) and Mundurukú (Picanço, 2005) are examples of the contrastive use of creaky and modal voice. The three-way phonemic contrast of modal, breathy and creaky vowels has been reported for Chong (Thongkum, 1991; cf. DiCanio, 2009), Xochistlahuaca Amuzgo (Herrera, 2009) and, within the Otomanguean stock, in Jalapa Mazatec (Kirk et al., 1993), Santa Cruz Tepetotutla Chinantec (Herrera, 2009), and Santa Ana del Valle Zapotec (Esposito, 2003; Rojas, 2010), among others. (For more examples of two- and three-way phonation type contrast see the appendix of Gordon & Ladefoged, 2001.) All of the above phonation types refer to Ladefoged’s continuum, which relies on the manner in which the vocal folds vibrate. Edmonson and Esling (2006) propose the use of supra-glottal mechanisms in the expression of phonation types, which allows adding faucalized (“hollow”), harsh (“pressed”) and strident (“harsh trilled”) voices to the diversity of phonation types. As such, another three-way-contrast example includes Bai (Edmondson & Esling, 2006), with modal, breathy, and harsh voice. Apart from Quiaviní Zapotec (Munro & Lopez, 1999), there exist two more cases of a four-way contrast with respect to voice qualities: in !Xóõ, Traill (1985) describes the contrastive use of modal, breathy (murmur), creaky and strident phonation types; in Dinka, Edmondson and Esling (2006) report modal, breathy, harsh and faucal voices. The contrast between creaky /a̰/ and interrupted /aʔ/ vowels has been reported only for Zapotec languages. This contrast, which implies two degrees or variants of laryngealized vowels, was first reported in Quiaviní Zapotec (Munro & Lopez, 1999),  176  and subsequently for Chichicapan Zapotec (Smith-Stark, 2003) and Güilá Zapotec (Arellanes 2008). Specifically with respect to Quiaviní Zapotec, Munro and Lopez (1999) report modal, breathy, creaky and checked (interrupted) vowels. There are, however, important differences with respect to this study. Chapter 4 shows that the four Quiaviní Zapotec tones (high, low, falling and rising) may be expressed with modal voice. Taking this contrastive feature into account in the language, a revision of the vowel patterns within non-modal phonation is presented here. Moreover, Munro and Lopez (1999) argue for vowel patterns that phonologically combine different voice qualities in the same syllable nucleus. In this study, some of those combinations are claimed to be phonetic implementations of a single phonological specification for phonation. A detailed analysis and comparison is provided throughout this chapter. Previous work on Quiaviní Zapotec phonation types also includes Gordon and Ladefoged (2001) and Ladefoged (2003), who describe the acoustic characteristics of modal, breathy and creaky phonation in this language. In those descriptions, the tonephonation interaction is not analyzed, nor is the fourth phonation type that is described by Munro and Lopez (1999; checked vowels, called interrupted here, §6.5), nor is the possibility that tense voice is a possible variation of creaky vowels (§6.4.2). Gender differences and rate of speech may affect the realization of phonation types. Munro, Lillehaugen and Lopez (2008) report that phonation may vary from speaker to speaker: “when men pronounce creaky vowels they sound more creaky than when women pronounce them” (p. 35); and “the amount of breathiness you hear in a vowel may vary from community to community or even from speaker to speaker. Vowels that are shown as breathy in the pronunciation guide [dictionary’s orthography] will sound a lot breathier in Tlacolula or San Lucas than in San Juan Guelavía or Santa Ana del Valle, for example. You may also notice that when women pronounce breathy vowels they sound more breathy than when men pronounce them” (p. 31). Gordon and Ladefoged (2002, p. 10) also reported noticeably creakier vowels for men and breathier vowels for women in Quiaviní Zapotec, and pointed out that gender dependent differences of this sort, particularly increased breathiness for female speakers, have also been observed in languages with allophonic rather than contrastive non-modal phonation,  177  including English (e.g. Henton & Bladon, 1985, Klatt & Klatt, 1990, Hanson and Chuang 1999). For Mundurukú (Tupí, Brazil), Picanço (2003, p. 37) reports that “the degree of constriction varies according to the rate of speech; in a sentence, speakers tend to produce creaky vowels with less constriction, but if the same words are pronounced in isolation, the vowels may be heavily creaky”. Similar findings have been reported for Otomanguean languages, including, for example, Esposito (2003) for Santa Ana del Valle Zapotec. Throughout my personal fieldwork and phonetic analysis, these characteristics have been noticeable in Quiaviní Zapotec. I will briefly refer to these issues in the following sections; however, as mentioned elsewhere, Quiaviní Zapotec intonational patterns at sentence level are beyond the scope of this dissertation. These observations show that non-modal phonation is relative rather than absolute, similar to the linguistic analysis of tone. In light of the cross-linguistic phonetic and phonological properties of phonation types presented here, the goal of the following sections is to characterize Quiaviní Zapotec voice qualities and understand their phonetic realization in the production of different tones.  6.3  Breathy vowels  6.3.1 Introduction This section presents a phonetic and phonological description of breathy vowels in this language, with the goal of providing a descriptive generalization of this voice quality in Quiaviní Zapotec. Breathy voice is a phonation in which the vocal cords vibrate, as they do in normal (modal) voicing, but are held further apart, so that a larger volume of air escapes between them (see Laver 1980, Ladefoged 1971, Gordon and Ladefoged, 2002 among others). A slightly less open stage of the vocal folds is attained with slack voice, where the vocal folds vibrate more loosely than in modal voice, also with a slightly higher rate of airflow than in modal voice (Maddieson and Ladefoged  178  1996: 48). Breathy and slack voices may freely vary with each other allophonically. I show that breathy vowels in Quiaviní Zapotec may be associated with low and falling tones, as presented in Table 61. (The analytic and theoretical implications are discussed in the next Chapter.) Table 61. Breathy vowels and tone interaction High Low Falling Rising Breathy X √ √ X The following examples illustrate the contrast between breathy-L (low tone) and breathy-F (falling tone). (1) Breathy -L / be̤ / ˩ → [ bè̤ː ~ βè͡e̤ː ]  ‘mold (growth)’  (2) Breathy-F / beṳ / Ü → [ béṳ̀ ~ βéṳ̀ ]  ‘turtle’  As the narrow phonetic transcription shows, the realization of these items normally includes a modal vowel portion followed by breathiness. As explained below, length patterns are the same for breathy-L and breathy-F lexical items. I now turn to the description of each of the breathy vowels.  6.3.2 Breathy-L An interaction between low tone and breathiness is extremely common crosslinguistically (see Gordon and Ladefoged 2002 and references therein), and is found in Quiaviní Zapotec. Examples below include fortis and lenis coda consonants. As demonstrated in Chapter 2 and 3 with modal vowels, the coda type determines the duration of the vowel. Phonetically, fortis consonants are preceded by short vowels, whereas vowels are long before lenis consonants or in open syllables.  179  (3) Breathy-L examples: fortis coda consonant a. / ta̤p /  ˩ → [ tà̤pː ]  ‘four’  b. / ɡje̤t / ˩ → [ ɡjè̤tː ]  ‘squash’  c. / nˑa̤ʃ / ˩ → [ nà̤ʃː ]  ‘chocolate’  d. / na̤s /  ‘the day before yesterday’  ˩ → [ nà̤sː ]  (4) Breathy-L examples: lenis coda consonant or open syllable a. / ɡei̤ʒ / ˩ → [ ɡèi̤ʒ ]  ‘townʼ  b. / na̤ /  ‘now’84  c. / jṳ /  [  ˩ → [ nà̤ː ]  ˩ → [ jṳ̀ː ] ‘soil’  b  è  e̤ e̥  t  s  ]  Figure 36. Waveform and spectrogram of / be̤ts / ˩ → [ bè̤tsː ~ βè͡e̤tsː] ‘(man’s) brother’ by male speaker TiuN.  84  Also / na̤ / ˩ ‘hard’ (nahah ‘hard’ vs. nah ‘now’ in Munro & Lopez, 1999).  180  [  β  è  e̤  e̥  ]  Figure 37. Waveform and spectrogram of / be̤ / ˩ → [ bè̤ː ~ βe͡e̤ː] female speaker LiaL.  ‘mold (growth)’ by  The above examples illustrate the modal-breathy-voiceless phonetic sequence as a common realization of breathy vowels. This is especially clear in the long vowel of /be̤ / ˩ ‘mold (growth)’, where the high amplitude and the periodicity of the waveform decreases as the vowels progresses and fades away. Pitch values during the modal portion are equivalent to those of modal-L items for these speakers. Cross-linguistically, non-modal vowels are commonly accompanied by modal phonation, especially at the beginning of the vowel. This is the case in both tonal and non-tonal languages (Gordon and Ladefoged 2002); however, for tonal languages this laryngeal timing is particularly important as it is during modal phonation that tone is realized, because tone is realized during modal phonation (see Silverman, 1997). As implied by examples in (3) and (4), when underlyingly breathy vowels encode breathiness and tone, modal voice is used to implement phonetically the realization of tone (see also breathy vowels with falling tone below). In order to confirm the contrastive character of breathy vowels in Quiaviní Zapotec, the following minimal (contrast) sets consist of triplets made of modal-H, modal-L and breathy-L items. (5)  a. / ʒi /  ˥ → [ ʒíː ]  ‘tomorrow’  b. / ʒi /  ˩ → [ ʒìː ]  ‘quite’ 181  c. / ʒi ̤ /  ˩ → [ ʒìː̤ ]  a. / ɡjia / ˩ → [ ɡjíˑá ]  (6)  ‘day’ ‘will go home’  b. / ɡjia / ˩ → [ ɡ ìˑà ]  ‘agave root’  c. / ɡji̤a̤ / ˩ → [ ɡjì̤ˑà̤ ]  ‘rock’  j  a. / nʒibj / ˥ → [ nʒíːbj ] ‘scared’  (7)  b. ---  c. / nʒ i̤bj/ ˩ → [ nʒ ì̤bj ] ‘knee’ (8)  a. --b. / ze /  ˩ → [ zèː ]  ‘was going’  c. / ze̤ /  ˩ → [ zè̤ː ]  ‘will go’  a. / belˑ / ˥ → [ bélː ] b. --c. / be̤lˑ / ˩ → [bè̤lː ]  (9)  (10)  (zeheh def. of rihah ‘goes’)  ‘Abel’ ‘fish’  a. --b. / na /  ˩ → [ nàː ]  ‘is (copula)’  c. / na̤ /  ˩ → [ nà̤ː ]  ‘now, hard’  6.3.3 Breathy-F The following examples illustrate breathy vowels with falling tone. (11) Breathy-F: fortis coda consonant a. / nje̤s /  Ü → [ njé͡è̤sː ]  ‘water’85  b. / ba̤lˑj /  Ü → [ bá͡à̤lː̤j ]  ‘fire’  (12) Breathy-F: lenis coda consonant a. / na̤ʒj / 85 86  Ü → [ ná͡à̤ːʃ j ] 86 ‘wet’  One variant of this item contains a diphthong: / nie̤s / Ü → [ ní͡è̤s ] ‘water’. Recall from Chapter 2 that lenis obstruents typically devoice word finally.  182  b. / ɡa̤lˑɡi̤ʒ / Ü → [ ɡa̤lˑɡî͡i̤ːʃ ] ‘sickness’ c. / bṳdj /  Ü → [ bû͡ṳːð̥j ]  ‘chicken’  d. / kṳb /  Ü → [ kû͡ṳːɸ ]  ‘tejate (traditional beverage)’  [  k  û  ṳ  ɸ  ]  Figure 38. Waveform and spectrogram of / kṳb / Ü ‘tejate’ by male speaker TiuR.  [  n  â  à̤ ː  ʃ  ]  Figure 39. Waveform and spectrogram of / na̤ʒj / Ü ‘wet’ by male speaker TiuL (Munro et al., 2008; Sound file: L3-3B) As with breathy-L (low tone) examples, the modal-breathy voice quality sequence is also noticeable in breathy-F (falling tone) examples. During the modal portion of the vowel in Figures 38 and 39 we observe a quick rise (during the vowel in / kṳb / Ü and  183  during the nasal in / na̤ʒj / Ü) and a slow fall, becoming breathy towards the end. The preliminary rise can be analyzed as a phonetic preparation to reach a high pitch level so that the falling tone can be adequately perceived. In Figure 39, for instance, pitch reaches 130 Hz (equivalent to modal-H for this speaker) and falls below 100 Hz during the breathy portion. The minimal pair in (13) contrasts modal-F vs. breathy-F, whereas (14) illustrates the distinction between breathy-L vs. breathy-F. Modal-F vs. Breathy-F: (13)  a. / beu /  Ü → [ béù ]  ‘moon’  b. / beṳ / Ü → [ béṳ̀ ]  ‘turtle’  Breathy-L vs. Breathy- F: (14)  a. / nˑa̤ʃ / ˩ → [ nà̤ʃː ]  ‘chocolate’  b. / na̤ʒj / Ü → [ ná͡à̤ːʃ ] ‘wet’ j  6.3.4 Munro and Lopez (1999): Breathy vowels The previous sections show my analysis of breathy vowels in Quiaviní Zapotec, where I propose that they can be associated with low and falling tones. The purpose of this subsection is to compare this account with the previous analysis of Munro and Lopez (1999), who propose a larger inventory of vowel patterns with breathy voice in Quiaviní Zapotec. These vowel patterns are included within the following table. Table 62. Munro and Lopez (1999) patterns for what I analyzed here as breathy vowels.87 Breathy  High X  Low ah ahah  Falling a’ah+C 88 a’ahah  Rising X  87  As mentioned before, in the Munro and Lopez (1999) orthography a = modal vowel, ah = breathy, à = creaky vowel and a’ = checked vowel. 88 Consider this kind of variation in a dictionary entry, with possible different vowel patterns within this cell (breathy-F): wbwi'ihzh, wbi'ihzh, wwi'ihihzh, wbwihzh ‘sun’.  184  aa’ah+C The Munro and Lopez (1999) vowel patterns ah and ahah are both included within the category breathy-L in my account, and both analyses agree in describing these pattern as having low tone. These durational differences certainly exist between short and long breathy vowels. Regardless of what is the best orthographic representation, the difference is predictable by coda type (short ah before fortis consonant, and long ahah before lenis consonant), and not phonologically contrastive. Similar durational differences seem to be encoded with the vowel patterns a’ah+C, a’ahah and aa’ah+C, for which Munro & Lopez (1999) report falling tone, as presented here. In addition, we have the presence of checked vowels (a’), and my indication of coda consonant (+C). The latter indication refers to the fact that my reanalysis of words with the patterns a’ah and aa’ah is split between those with coda consonants, classified here as breathy vowels with falling tone, and those in open syllable, classified as interrupted vowels (§6.7). In my analysis, lexical items with the patterns a’ah and aa’ah with coda (breathyF) indicated no laryngealization. Certainly, there is always a modal beginning where the falling tone is manifested, but the falling tone passes from modal into breathy, as shown above with Figures 38 and 39. To further illustrate this, Figure 40 shows only the pitch contour (Hz) and amplitude envelope (dB) of the example in Figure 39, / na̤ʒj / Ü ‘wet’. We observe that neither pitch nor intensity is interrupted during the vowel duration (demarcated by the dashed lines), as expected with a checked vowel (compared with the vowel pattern a’ah in open syllable in section §6.5 below). Instead, these acoustic correlates show a quick rising and slow falling pitch contour, passing from modal voice to breathy. As mentioned above, this phonation type sequence, rather than being phonologically specified, results from the phonetic manifestation of breathy-F items. Since breathy voice is unable to express high frequency pitch, the necessary high pitch at the beginning of a falling tone is implemented via modal voice, and then the second portion of the vowel expresses breathiness. The inclusion of a checked vowel for these patterns in Munro and Lopez (1999) could have been an orthographic convention to indicate the described tone contour.  185  100 70 50  0  0.8681 Time (s)  300 200 150  intensity  100 70 50  pitch  0  0.8681 Time (s) [  n  â  à̤ ː  ʃ  ]  Figure 40. Pitch and intensity contours of / na̤ʒj / Ü ‘wet’ by male speaker TiuL Other vowel patterns in Munro and Lopez (1999) with breathy voice include aha’, àah, ahaha, presumably breathy-L, and a’aah, a’aha, aahah, iiah, aah, aaha’, presumably breathy-F. However, these vowel patterns were not considered in Munro, Lillehaugen and Lopez (2008); so the authors themselves simplified the original account of Munro and Lopez (1999).  6.3.5 Interim summary: Breathy vowels Breathy voice is cross-linguistically associated with lowered tone in languages with this voice quality (Hombert et al., 1979). As shown above, this is also the case in Quiaviní Zapotec, where breathy vowels are restricted to low and falling tones, as shown in Table 63. This distribution is discussed and accounted for formally in Chapter 7. Table 63. Tone and phonation: modal and breathy vowels High Low Falling Rising Modal √ √ √ √ Breathy X √ √ X  186  Breathy voice is found in different Otomanguean languages, but this contrast is not as widespread as laryngealized voice. A possible analysis would be to consider breathiness as an enhancement of low and falling tones (along the lines of Enhancement Theory, e.g. Stevens and Keyser 1989) in that breathiness has developed historically as a mechanism for more easily controlling low tone (by itself, or after a high tone, so pitch falls). However, for the current synchronic state of Quiaviní Zapotec, it seems unlikely that this is the source of breathy voice in the language, because low and falling tones are also used contrastively with modal voice, as illustrated above with several minimal pairs (modal-H, L and F vs. breathy-L and F).  6.4  Creaky vowels  6.4.1 Introduction Creaky voice, also called laryngealized voice or vocal fry,89 is produced with the vocal folds vibrating anteriorly, but with the arytenoid cartilages pressed together; this induces a considerably lower rate of airflow than in modal voice (see e.g. Laver, 1980; Ladefoged, 1971; Gordon & Ladefoged, 2001). In Quiaviní Zapotec creaky vowels may be associated with both high and low level tones, as well as the falling contour tone. This is shown in the following table and illustrated with examples (15-17). Table 64. Creaky vowels and tone interaction Creaky  89  High √  Low √  Falling √  Rising X  The term laryngealized voice is used here as a cover term to refer to both creaky and interrupted vowels.  187  (15) Creaky H / bḛlˑ / ˥ → [ bé̬lː ]90  ‘(woman’s) sister’ (bèe’ll)  (16) Creaky L / bḛlˑ/ ˩ → [ bè͡el̰ ː ]  ‘snake’ (bèèe’ll)  (17) Creaky F / bḛl / Ü → [ bé͡ḛ̀ːl ] ‘meat’ (beèe'l) The purpose of this section is to describe in detail the phonetic and phonological properties of creaky vowels, providing a full account of the expression of creaky voice in Quiaviní Zapotec. The Munro and Lopez (1999) analysis (whose orthography is included in parentheses within the above examples) is presented below in §6.6.5.  6.4.2 Creaky-H The first cases I analyze are creaky vowels with high tone. This interaction is uncommon, as creaky voice is cross-linguistically associated with lowering of the fundamental frequency. As we will see below, the actual realization of creaky vowels with high tone is a weak laryngealization, in the form of tense (stiff) voice, which is presented in Ladefoged and Maddieson (1996, p. 48) as an intermediate step between modal and creaky voice, where the vocal folds vibrate more stiffly and with a slightly lower rate of airflow than in modal voice.91 Tense voice, in contrast to prototypical creaky voice, is compatible with the manipulation of pitch. As discussed extensively in Chapters 2 and 3, fortis coda consonants are preceded by short vowels, and lenis consonants by long vowels. Both types of syllables are found with creaky vowels as well.  90  As illustrated here, creaky-H vowels are produced with tense voice, [ e̬ ] (symbol from Ladefoged and Maddieson, 1996, p. 100). 91 This voice is not to be confused with harsh voice, sometimes also called “pressed” voice, which is produced with a different mechanism, with the upper larynx becoming highly constricted with the ventricular folds (see Edmonson and Esling 2006 for more details).  188  (18) Creaky-H examples: fortis coda consonant a. / bḛlˑ /  ˥ → [ be̬lː ]  ‘(woman’s) sister’  b. / ɾɡḭlˑj/ ˥ → [ ɾɡil̬ːj ]  ‘looks for’  c. / zḭlˑj/  ‘a lot of’  ˥ → [ zi̬lːj ]  (19) Creaky-H examples: lenis coda consonant a. / bḛl /  ˥ → [ bé͡é̬ːl̬ ]  ‘naked’  b. / ɾɡḭbj / ˥ → [ ɾɡí͡í̬ːɸ ] ‘washes’ j  [  e  b  e̬  l̬  l  ]  Figure 41. VCfortis example: Waveform and spectrogram of / bḛlˑ / ˥ ‘(woman’s) sister’, by male speaker TiuL (arrows indicate the tense voice portion).  [  ɾ  ɡ  i  i̬  ɸj  ]  Figure 42. VClenis example: Waveform and spectrogram of / ɾɡḭbj / ˥ ‘washes’, by male speaker TiuL (arrows indicate the tense voice portion).  189  In terms of phonation, we observe in Figures 41 and 42 that the first part (or beginning) of the vowel is modal, whereas the second portion of it shows tense voice, mainly characterized here by the lower amplitude envelope.92 In the case of / bḛlˑ / ˥ ‘(woman’s) sister’, the stiff or tense voice is observed at the end of the vowel and beginning of the fortis liquid. In addition, /ɾɡḭbj/ ˥ ‘washes’ (Figure 42) shows aperiodicity of the signal (i.e. some creakiness) at the end of the vowel. Due to the possible co-articulation of tense voice and high tone, there might be instances without modal voice in the realization of short vowels. The degree of laryngealized voice varies by speaker. With respect to tone, we observe a relatively flat pitch all the way through the vowel (and the fortis /lˑ/ in Figure 41), in both the modal and tense portions. It never drops so much that it can no longer be tracked automatically by pitch extraction, which commonly happens with true creaky vowels that have low and falling tones (see below).  6.4.3 Creaky-L As mentioned above, creaky voice is commonly associated with lowering of the fundamental frequency; thus, we would expect to find creaky-L items in Quiaviní Zapotec. Consider the following examples. (20) Creaky-L examples: fortis coda consonant a) / bḛlˑ/  ˩ → [ bè͡ḛl̰ː ]  ‘snake’  b) / bḛkw / ˩ → [ bḛ̀ kː ] ‘dog’ ʔ  w  (21) Creaky-L examples: lenis coda consonant or open syllable a) / ɾɡḭlj/ ˩ → [ ɾɡì͡ḭ̀ːlj ] b) / sḭlj /  ˩ → [ sì͡ḭ̀ːl ] j  ‘waters’ ‘breakfast’  92  As noticed in Chapter 4, this striking change in the amplitude envelope is not observed in modal vowels with low, rising or falling tones, previously analyzed as items with weak laryngealization (Munro & Lopez, 1999).  190  c) / ɾɡḭdj / ˩ → [ ɾɡì͡ḭ̀ːdj]  ‘sticks on’  d) / bdo̰ / ˩ → [ bdò͡òː ̰] ‘baby’  e) / ɾka̰z / ˩ → [ ɾkà͡à̰ːz ] ‘wants’ d) / jdo̰ / ˩ → [ jdò͡ò̰ː ]  [  b  è  ‘church’  ʔ  ḛ̀  k  w̥  ]  Figure 43. Waveform and spectrogram of / bḛkw / ˩ ‘dog’, by male speaker TiuL (Munro et al., 2008, sound file L3-3C).  [  b d  ò  ò̬  o̰  ʔ  ]  Figure 44. Waveform and spectrogram of / bdo̰ / ˩ ‘baby’, by male speaker TiuL (Munro et al., 2008, sound file L3-3C).  191  Figures 43 and 44 exemplify the fact that Creaky-L items normally start with modal phonation, to continue into a creaky voice portion. Due to the degree of variation, it is possible to find some tokens with short vowels with creaky voice only; since creaky voice inherently has low pitch, tone and non-modal voice may phonetically co-occur for these items. A crucial point here in determining the tone of these items is that pitch values during the first portion of the vowel are similar to modal-L values. Concomitantly, the amplitude envelope goes along with pitch: it is sustained (higher) during the less laryngealized vowel portion, and then it drops as the vowels show more laryngeal constriction. Figure 43 illustrates a rhyme formed by a short vowel with fortis coda consonant. The highlighted part corresponds to the less laryngealized portion of the vowel (close to modal). The pitch averages 110 Hz (range 100-114 Hz). Although the pitch is not quite flat, the numbers are in the range of modal-L tokens of this speaker (whose pitch also tends to drop towards the end). Figure 44 illustrates a long vowel in an open syllable. The mean pitch of the first portion (highlighted) is 107 Hz (range 115-98 Hz). Phonetically, these long vowel tokens may seem to have falling pitch. However, two points suggest that these tokens are creaky-L. First, creaky-F items normally have a phonetic rise in pitch at the beginning of the vowel, but no rise is found in creaky-L items. Second, and more important, the values of the first portion of creaky-L items are lower that those found in the first portion of creaky-F vowels, and within the modal-L range.  6.4.4 Creaky-F Creaky vowels also occur with falling tone, as illustrated below. (22) Creaky-F examples: fortis coda consonant a) / n-ɡa̰ts /  Ü → [ ŋɡá͡à̰tsː ]  b) / n-ɡasja̰ts / Ü → [ ŋɡasjá͡à̰tsː ]  ‘yellow’ ‘really black’  192  (23) Creaky-F examples: lenis coda consonant or open syllable a) / mḭʒ / Ü  → [ mí͡ḭ̀ːʒ ]  ‘Mixe’  b) / ja̰ /  → [ já͡à̰ː ]  ‘up’  → [ ndá͡à̰ː ]  ‘hot’  Ü  c) / nda̰ / Ü d) / ʒ ḭʒ / Ü e) / beṵ / Ü  [  → [ ʒí͡ḭ̀ːʒ ]  ‘pineapple’  → [ béṵ̀ ]  ‘coyote’  m  íː  ḭ̀  ʒ  ]  Figure 45. Waveform and spectrogram of / mḭʒ / Ü ‘Mixe’, by male speaker TiuL (Munro et al., 2008, sound file L3-3D).  193  Figure 46. Waveform and spectrogram of / ja̰ / Ü ‘up’, by male speaker TiuL (Munro et al., 2008, sound file L3-3D).93 In terms of phonation, the first part of the vowel in creaky-F tokens is always modal, whereas the second (or last) part is creaky. Both Figures 45 and 46 clearly illustrate this voice sequence. With respect to tone, often there is a small rise at the beginning of the vowel, so that the pitch might be sufficiently high to attain a significant falling contour.94 This initial rise is clear at the beginning of the vowel in Figure 45, which reaches a maximum pitch of 157 Hz (even higher than the high tone pitch average of this speaker). The pitch falls to 101 Hz during the modal portion, continuing to fall even lower during the final creaky vowel portion. These high pitch values at the beginning are similar to those found in modal-H values and this is the crucial difference to distinguish creaky vowels with falling tone versus low tone. Let us consider in parallel an example of each.  93  Minimum pitch: 105 Hz; Maximum pitch: 142 Hz. In Figures 44, we also observe a small rise at the beginning the creaky-L vowel (/ bdo̰ / ˩ ‘baby’), but this seems to be related with the voicing of the consonant (Hombert et al., 1979). In contrast, the rise is clearer with /m/ in Figure 45 for / mḭʒ / Ü ‘Mixe’, and cross-linguistically nasals don't lower F0. 94  194  5000  0  0  2.091 Time (s)  Creaky-L Creaky-F Figure 47. Spectrograms and pitch of / bdo̰ / ˩ ‘baby’ and / mḭʒ / Ü ‘Mixe’, by male speaker TiuL. There are important differences between creaky-L and creaky-F items, illustrated by Figure 47. For the former items there is never a clear pitch rise at the beginning, pitch is relatively flat during the less laryngealized portion, where I claim the phonological tone is expressed; then the pitch drops as the vowel gets creakier, to the point that it becomes difficult (or impossible) to track. On the other hand, creaky-F tokens normally show a rise at the beginning of the vowel and always have higher pitch values during the first modal vowel portion, showing a different pitch contour than that of creaky-L tokens. Strictly speaking both types of creaky vowels phonetically have a falling pitch (creaky-F shows a high-falling contour, whereas creaky-L a low-falling one); however, in the case of creaky-L tokens most of the fall occurs during the laryngealized portion, where the phonological tone is no longer expressed. In contrast, the examples of creaky-F tokens show that the fall is noticeable during the modal portion. Perceptually, this seems like a case where the listener may be abstracting away from effects that are predictable, as in the case of abstracting away from the effects of coarticulation. Perceivers know that creaky voice causes pitch lowering, so the lowering due strictly to such phonation does not cause the tone to be perceived as falling. The phonetic pitch fall characteristic of creaky vowels with low tone has also been described for other languages. According to Picanço (2005), Mundurukú (a Tupí language spoken in the Amazonian basin of Brazil) has both contrastive tones and  195  phonation types: modal voice allows high vs. low tone, whereas creaky vowels only allow low tone. Comparing modal vs. creaky vowels with low tone, the latter is characterized by “lowered fundamental frequency, glottal pulses with longer duration, and variation between adjacent glottal pulses”; on top of that, pitch may lower as the vowel gets creakier, in other words, “Creaky voice is […] manifested as a gradual fall in pitch.” (Picanço, 2005, p. 38).  6.4.5 Munro and Lopez (1999): Creaky vowels The previous sections show my analysis of creaky vowels in Quiaviní Zapotec, where I propose that these vowels can be associated with high, low and falling tone. The previous account of Munro and Lopez (1999) proposes only falling tone for different vowel patterns with creaky vowels, included in Table 65. In this section, I compare the two analyses, followed, in the next section, by an acoustic analysis that quantitatively establishes the phonetic characteristics of creaky vowels in Quiaviní Zapotec. Table 65. Munro and Lopez (1999) patterns for what are analyzed here as creaky vowels.95 Creaky  High Low àa’ (some)96 ààa’  Falling a’àa’ aàa’  Rising X  In all these vowel patterns there is a creaky vowel followed by a checked one at the end (àa’). If my understanding of the Munro and Lopez (1999) orthography is correct,  95  Other vowel patterns with creaky vowels described in Munro and Lopez (1999) include àa'a+C, ààa'ah and aàa'ah. The dictionary entries with the pattern àa'a+C that have a coda consonant other than /n/ (e.g. rtàa'az ‘beats up’ or a variation of bèe'cw / bèe'ecw ‘dog’, are reclassified here as creaky-L tokens; whereas the laryngealization of examples with /n/ in coda, as gùu’an ‘bull’ and zhìi’iny child’, are considered here interrupted vowels (§6.5). The vowel patterns ààa'ah and aàa'ah are found in only a few lexical items, reanalyzed here as creaky-L and creaky-F, respectively. These vowel patterns were not included in the simplified analysis of Munro et al (2008). 96 The pitch analysis of lexical items with the vowel pattern àa’ suggests a split between tokens with high and low tone.  196  it corresponds phonologically to / a̰aʔ /, but probably corresponds phonetically to [ a̰ʔ ], as there is never modal voice between the creakiness and a glottal stop (when one is clearly present). I argue here that the presence of the glottal stop is phonetically predictable and thus not part of the underlying phonological representation. The realization of creaky vowels is directly related to the segment that follows the vowel. If followed by fortis stops ([-continuant, -sonorant, +fortis]) or pause, creaky vowels normally end in a glottal closure (see examples in (25) and Figure 48); whereas if the following segment is a fricative, liquid, glide or a vowel (i.e. [+continuant] segments97) a glottal stop does not occur (examples in (26) and Figures 49 & 50). This may be represented with the (not absolute) phonetic rules in (24).  (24)  a. creaky vowel → a̰ʔ / _ [-continuant, -sonorant, +fortis] or # 98 b. creaky vowel → a̰ / (elsewhere)  (25)  Creaky vowels followed by fortis oral stop or utterance-final  a. / la̰ts /  ˩ → [ là͡aʔ̰ ts ]  (làa'ts)  ‘flat area’  b. / baRɡa̰ / ˩ → [ baɾ.ɡà͡a̰ː ] (bargàa’)  ‘grasshopper’  c. / bdo̰ / ˩ → [ bdò͡o̰ː ]  (bdòo’)  ‘baby’  d. / mna0 / ˩ → [ mnà͡a0˘ ]  (mnnààa’)  ‘woman’  ʔ  ʔ  / 99  (26)  Creaky vowels followed by [+continuant] segment  a. / dḭʒ /  ˩ → [ dì͡ḭːʒ̊ ]  (dìi'zh)  ‘language’  b. / ɾka̰z / ˩ → [ ɾkà͡a̰ːz ]  (r-càa'z)  ‘wants’  c. / ɡḛlˑ /  (guèe'll)  ‘midnight’  ˩ → [ ɡè͡el̰ ː ]  97  Although nasals are orally [-continuant] sonorants, they seem to pattern with the rest of [+cont] segments with respect to the presence or absence of a glottal stop (see Bernhardt & Stemberger, 1998, for a discussion of the feature [continuant] for nasals). So far, in tokens of creaky vowels followed by nasal consonants, no glottal has been detected (the featural implications of this pattern are beyond the scope of this dissertation, but see Mielke (2008 [2004]) on the ambivalence of nasals with respect to [continuant] specification). Lenis stops are normally fricated in coda position, and thus no glottal stop is present in the vowel. 98 The symbol # indicates a pause or end of utterance. 99 Gordon and Ladefoged (2003, p. 9) illustrates the same issue of creaky vowels ending with a glottal stop at the end of utterance, their transcription of mnnààa' ‘woman’ is also [mnaa̰ʔ].  197  d. / bḛlˑ /  ˩ → [ bè͡el̰ ː ]  (bèèe'll)  ‘snake’  The above patterns are illustrated in Figure 48, with some acoustic examples.  0.328  0.4743  0  -0.3877  0  0  0.7065  -0.5882  Time (s)  0.8269 Time (s)  5000  0  0  5000  0  0.7065  0  0  0.8269 Time (s)  Time (s)  [ β d ò̰ˑ  ʔ  ]  [ b d ò  o̰ː  ʔ  ]  Figure 48. Waveform and spectrogram of / bdo̰ / ˩ by male speaker TiuT (from personal fieldwork) and by TiuL (Munro et al., 2008, Unida 1; sound file L3-3C) The possible (expected) ending of creaky voice into a glottal stop may be explained physiologically. A creaky vowel usually starts with modal phonation (although sometimes the vowel is creaky right from the beginning), then, the vibration occurs only anteriorily with the arytenoid cartilages pressed together. As creakiness continues, the glottal pulses become more and more sporadic (this period of creakiness can be considered successive glottal closures). If an oral stop or a pause follows one creaky vowel the natural way to finish the vowel is to simply cease the vibration, i.e. maintain a glottal closure. In addition, the presence of the glottal stop may enhance the glottalization of the vowel and the oral stop closure. I consider this optional (potential) final closure to be phonetic variation rather than phonemic contrast in Quiaviní Zapotec,100 and possibly 100  Morphologically, the presence of the glottal stop in creaky vowels is also predictable in Quiaviní Zapotec. When a vowel-initial clitic/suffix follows a stop final root with creaky vowel, there is no glottal closure, as the stop resyllabifies with the clitic/suffix: / la̰ts / ˩ → [ là͡a̰ʔts ] ‘flat area’, but / la̰ts + eʔ/ ˩ → [là͡a̰.tseʔ] ‘small flat area’.  198  cross-linguistically (see also Picanço, 2005; and Jiang-King, 1999, for similar characteristics in the description of creaky vowels in Munduruku and Chinese, respectively). Other phonetic analyses of creaky vowels in different Zapotec languages have also shown the possibility of creaky vowels ending in a glottal stop (including Jones and Knudson, 1977; Antonio Ramos, 2007; and Arellanes, 2009, among others). According to Jones and Knudson (1977), in San Juan Guelavía Zapotec creaky vowels “are checked before pause” (p. 17); crucially, what they mean by checked here is a creaky vowel that ends with a glottal stop before pause, as an allophonic variant of a creaky vowel, just as presented for Quiaviní Zapotec here. In comparison, creaky vowels followed by continuant segments do not end with a full glottal closure. Consider the following figures.  0.1709  0  -0.1426  0  0.5308 Time (s)  7000  0  0  0.5308 Time (s)  [  d̥  h  ì  ḭ  ʒ̊  ]  Figure 49. Waveform and spectrogram of / dḭʒ / ˩ ‘word’ by female speaker LiaCh.  199  0.2612  0  -0.2982  0  0.5247 Time (s)  5000  0  0  0.5247 Time (s)  [ b  é̬  ḛ̀  lː̰  ]  Figure 50. Waveform and spectrogram of / bḛl / Ü (bèèe'll) ‘snake’ by male speaker TiuL (Munro et al., 2008, sound file L3-3C) The important fact about the figures above is that there is no glottal stop at the end of the vowel, there is just the transition from the creaky vowel into the coda consonant, which in the case of fortis /l/ shows appreciable creakiness. In addition to the issue of the final checked vowel in these vowel patterns, differences between the vowel patterns àa’ and ààa’ imply a duration difference, with the latter being longer. Another inherent difference seems to be the presence of a modal beginning for the vowel patterns a’àa’ and aàa’, which I recategorize as creaky vowels with falling tone. The orthographic convention to represent modal voice may suggest a correlation between the above phonetic description, where I show that creaky vowels with falling tone always start with modal phonation, displaying the phonological tone in this portion of the vowel. In order to properly compare the analysis presented in the previous sections and that of Munro and Lopez (1999), I conducted an acoustic analysis to clarify the phonetic and phonological properties of these vowels.  200  6.4.6 Acoustic experiment: Creaky vowels  6.4.6.1 Intr