UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Postvelar harmony : an examination of its bases and crosslinguistic variation Shahin, Kimary N. 1997

You don't seem to have a PDF reader installed, try download the pdf

Item Metadata

Download

Media
831-ubc_1997-251578.pdf [ 23.48MB ]
Metadata
JSON: 831-1.0088253.json
JSON-LD: 831-1.0088253-ld.json
RDF/XML (Pretty): 831-1.0088253-rdf.xml
RDF/JSON: 831-1.0088253-rdf.json
Turtle: 831-1.0088253-turtle.txt
N-Triples: 831-1.0088253-rdf-ntriples.txt
Original Record: 831-1.0088253-source.json
Full Text
831-1.0088253-fulltext.txt
Citation
831-1.0088253.ris

Full Text

POSTVELAR HARMONY: A N EXAMINATION OF ITS BASES AND CROSSLINGUISTIC VARIATION by KJMARY NOELLE SHAHIN B.A. (Hons.), University of Victoria, 1986 M.A., University of Victoria, 1989  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (Department of Linguistics)  We accept this thesis as conforming to the required standard  THE UNIVERSITY OF BRITISH COLUMBIA October 1997 © Kimary Noelle Shahin, 1997  In  presenting  this  degree at the  thesis  in  University of  partial  fulfilment  of  the  requirements  British Columbia, I agree that the  for  an advanced  Library shall make it  freely available for reference and study. I further agree that permission for extensive copying of  this thesis for  department  or  by  his  or  scholarly purposes may be granted her  representatives.  It  is  by the  understood  that  head of  my  copying  or  publication of this thesis for financial gain shall not be allowed without my written permission.  Department The University of British Columbia Vancouver, Canada  Date  DE-6 (2/88)  ^Jr.iCh  Abstract  This dissertation examines postvelar harmony in two unrelated languages: Palestinian Arabic and St'at'imcets Salish. In contrast to previous studies, it identifies two such harmonies for each language: pharyngealisation (tongue root retraction) harmony and uvularisation (tongue back retraction) harmony. The properties of the two harmonies in each language are detailed.  Acoustic data are provided as support for the proposed  analyses and for the grounded phonological accounts which are subsequently developed. The harmonic feature of pharyngealisation harmony in both Palestinian and St'a'timcets is identified as [RTR] (unspecified for primary or secondary status). The anchor for [RTR] in both languages is the NUC.  Co-occurring secondary-[DOR] and secondary-[RTR] are  identified as the harmonic features of uvularisation harmony. In both languages, the anchor for these co-occurring features is the root node. An Optimality Theory account of the two harmonies in each language is developed, in which Correspondence, Alignment, and Grounded Constraints have central roles. Constraint reranking is shown to yield the observed crosslinguistic variation in the harmonies. In the course of this examination, issues regarding the consonantal and vocalic inventories of Palestinian and St'at'imcets are adressed.  It is argued that each has a more elaborate vocalic system than previously  recognised, and that St'at'imcets, like Palestinian, has a set of underlying emphatic consonants.  ii  Table of Contents  Abstract Table of Contents List of Tables List of Figures Acknowledgments C h a p t e r 1:  . ii iii vii ix xviii  Introduction  1.1. Aims  1  '  1  1.2. Overview of the Thesis  4  1.3. Representational Assumptions 1.3.1. Articulator Theory 1.3.2. The Articulator Feature [TONGUE ROOT] ' 1.3.2.1. The Assumed Problem with Laryngeal Gutturals 1.3.2.2. Another Look at Laryngeal Gutturals 1.3.3. Other Representational Assumptions 1.3.3.1. Laryngeal Specifications 1.3.3.2. The Representation of Secondary Articulation 1.3.3.3. Prosodic Representations 1.4. Postvelars 1.4.1. The Articulation of Gutturals 1.4.2. The Articulation of Emphatics 1.4.3. Postvelar Acoustics 1.4.4. Summary  6 7 10 10 18 23 24 29 33 35 37 47 56 75  1.5. A Harmony Typology  76  1.6. Optimality Theory 1.6.1. Correspondence Constraints 1.6.2. Alignment Constraints 1.6.3. Grounded Constraints  77 79 .83 84  1.7. Phonetics and Phonology 1.7.1. The Distinction Between Phonetics and Phonology 1.7.2. The Use of Phonetics in Phonology  .87 87 94  C h a p t e r 2:  Pharyngealisation H a r m o n y and Uvularisation H a r m o n y in  Palestinian  96  Arabic  2.1. The Language and the Data  96  2.2. Phonemic Inventory 2.2.1. Consonantal Inventory 2.2.1.1. The Palestinian Underlying Consonantal Inventory 2.2.1.2. The Palestinian Surface Consonantal Inventory 2.2.1.3. Postvelars . 2.2.1.3.1. Guttural Approximants  iii  98 98 98 101 .103 103  2.2.1.3.2. No Underlying Non-emphatic Irl 2.2.1.3.3. High-frequency vs. Low-frequency Emphatics 2.2.1.4. Epenthetic Word-initiaH?!• 2.2.2. Vocalic Inventory 2.2.2.1. Previous Analyses of the Palestinian/Levantine Vocalic System 2.2.2.1.1. Non-generative Analyses 2.2.2.1.2. Generative Analyses 2.2.2.1.3. Summary 2.2.2.2. Underlying Length 2.2.2.3. Three Underlying Degrees of Height for the Long Vowels 2.2.2.4. Three Underlying Degrees of Height for the Short Vowels 2.2.2.5. Reduction of IMl 2.2.2.6. Pharyngealised Short Vowels. 2.2.2.7. Uvularised Low Vowels . 2.2.2.8. Summary  104 109 Ill 115 117 117 121 126 127 129 130 135 139 140 141  2.3. Preliminary Issues 2.3.1. Underlying Pharyngealisation, Underlying Uvularisation 2.3.1.1. Acoustic Support 2.3.2. The Derivation of the Palestinian Underlying Postvelar Inventory 2.3.3. Prosodically Conditioned (Closed Syllable) Pharyngealisation 2.3.3.1. Analysis 2.3.3.2. Acoustic Support 2.3.3.3. A Theoretical Account  142 142 148 155 157 157 159 167  2.4. Palestinian Pharyngealisation Harmony 2.4.1. Harmony Under Adjacency to a Postvelar 2.4.1.1. Analysis 2.4.1.2. Acoustic Support 2.4.2. A Theoretical Account: Part I 2.4.3. Non-local Harmony 2.4.3.1. Analysis 2.4.3.2. Acoustic Support 2.4.4. A Theoretical Account: Part II 2.4.5. Opaque Stem-final Vowels and Long Vowels 2.4.5.1. Analysis 2.4.5.2. Acoustic Support 2.4.6. A Theoretical Account: Part III  183 183 183 184 187 196 196 202 211 219 219 229 241  2.5. Palestinian Uvularisation Harmony 2.5.1. Harmony With an Emphatic 2.5.1.1. Analysis 2.5.1.2. Acoustic Support 2.5.2. A Theoretical Account: Parti 2.5.3. Opaque Post-alveolar Obstruents 2.5.3.1. Analysis 2.5.3.2. Acoustic Support 2.5.4. A Theoretical Account: Part II 2.5.5. Transparent Non-low Vowels 2.5.5.1. Analysis  251 252 252 254 268 286 286 287 289 293 293  iv  2.5.5.2. Acoustic Support 2.5.6. A Theoretical Account: Part III 2.5.7. Uvularisation Harmony in Other Palestinian Dialects  294 . 2 9 9 304  r  2.6. Summary and a Final Issue C h a p t e r 3:  306  Pharyngealisation H a r m o n y and Uvularisation St'at'imcets (Lillooet)  Harmony  in  Saiish  317  3.1. The Language and the Data  317  3.2. Phonemic Inventory 320 3.2.1. Consonantal Inventory 320 3.2.1.1. The St'at'imcets Underlying Consonantal Inventory 320 3.2.1.2. The St'at'imcets Surface Consonantal Inventory 326 3.2.1.3. Previous Analyses of the St'at'imcets Consonantal System 327 3.2.1.4. Guttural Postvelars " 335 3.2.1.5. Emphatic Postvelars 337 3.2.1.5.1. Dialectal Variation in Dental Approximant Retraction 338 3.2.1.5.2. Underlying Retracted /cs 1 17 340 3.2.1.5.3. Perceptual Support for the Claim that the Retracted Consonants and Uvular Obstruents are Emphatics 350 3.2.1.5.4. Phonological Evidence 353 3.2.1.5.5. Ii J 7 De-emphaticisation in Lower St'at'imcets 363 3.2.1.5.6. Forms with a Floating Emphasis Feature 367 3.2.1.5.7. Acoustic Support 370 3.2.2. Vocalic Inventory .379 3.2.2.1. Previous Analyses of the St'at'imcets Vocalic System 381 3.2.2.2. The Epenthetic Vowel 382 3.2.2.3. Phonetic Mid Height 389 3.2.2.4. Phonetic Epiglottalisation 395 3.2.2.4.1. Acoustic Support 395 3.2.2.5. Pharyngealised Vowels 400 3.2.2.6. Uvularised Non-high Vowels 401 3.2.2.7. Summary 403 f  3.3. Preliminary Issues 3.3.1. Underlying Pharyngealisation, Underlying Uvularisation 3.3.1.1. Acoustic Support 3.3.2. The Derivation of the St'at'imcets Underlying Postvelar Inventory .  403 403 405 407  3.4. St'at'imcets Pharyngealisation Harmony 3.4.1. Pharyngealisation Under Adjacency to a Postvelar 3.4.1.1. Analysis 3.4.1.2. Acoustic Support 3.4.2. A Theoretical Account  408 408 408 416 456  3.5. St'at'imcets Uvularisation Harmony 3.5.1. Harmony With an Emphatic 3.5.1.1. Analysis 3.5.1.2. Acoustic Support 3.5.2. A Theoretical Account: Part I  471 471 . 471 ' .481 498  v  3.5.3. Neutral High Vowels 3.5.3.1. Analysis 3.5.3.2. Acoustic Support 3.5.4. A Theoretical Account: Part II 3.5.5. 'Transparent Laryngeals' 3.5.5.1. Analysis 3.5.5.2. Acoustic Support 3.5.6. A Theoretical Account: Part III  . •• . -  ,  3.6. Summary Chapter  4:  509 509 510 512 514 514 515 518 523  Conclusion  527  4.1. Summary of the Thesis  527  4.2. A Residual Issue  531  Appendix I: Abbreviations and Symbols  534  Appendix II: Original Location of the Abu Shusha and Jafa Palestinian Dialects  538  Appendix III: Details of the Palestinian Acoustic Study 1. Segmentation  539 539  2. The Palestinian Carrier Forms For Vowel Tokens  545  Appendix IV: Salish Language Classification  560  Appendix V: Salish Language Map  562  Appendix VI: The St'at'imcets Carrier Forms For Vowel Tokens  563  Appendix VII: St'at'imcets Word List  571  Glossary  579  Bibliography  583  vi  List of Tables  Table 1:1 Values of r , d, and All for vocal tract configurations with primary 0  pharyngeal and uvular articulations  60  Table 1:2 Predicted acoustic effects of primary pharyngeal and uvular articulations  64  Table 1:3 Fi and F ranges predicted over all r , d, and AAby the model of Stevens 2  0  • and House (1955)  ,64  Table 1:4 Fi and F ranges of Peterson and Barney data  65  Table 1:5 Low, medium, and high ranges for Fi and F  65  2  2  Table 1:6 Secondary postvelar articulations of uvular gutturals and emphatics  66  Table 1:7 Values of r , d, and All for vocal configurations with pharyngealisation 0  and uvularisation articulations  67  Table 1:8 Predicted acoustic effects of pharyngealisation and uvularisation  70  Table 1:9 Maximum Fi and F changes predicted by the model of Stevens and 2  House (1955)  71  Table 1:10 Ranges of degrees of change for F and F t  2  Table 1:11 Predicted cumulative acoustic effects of the postvelar articulations of gutturals and emphatics Table 1:12 Necessary criteria for speech-phonetic vs. language-phonetic vs. phonological status Table2:lFiandF oftokensofPalestinian/ T/and/jj7 Table 2:2 Fi and F of tokens of Palestinian IXI, IXI, ir\, andM <  2  2  Table 2:3 Judgments of Palestinian /ml and III in a non-pharyngealisation vs. closed-syllable pharyngealisation context  71 72 91 153 154 . 180  Table 2:4 The distinct properties of (Abu Shusha) Palestinian's two postvelar harmonies  315  Table 3:1 Correspondence between JJPA and N A transcription: obstruents  323  Table 3:2 Correspondence between JJPA and N A transcription: resonants  324  Table 3:3 Arabic judgments of St'at'imcets retracted consonants, and uvular and velar obstruents  352  Table 3:4 Fi and F of tokens of St'at'imcets Ikl, Ikl, III, l\l, \fi (= de-emphatic2  ised If I), and / J 7  372  vii  Table 3:5 F i , F , andF of tokens of St'at'imcets vowels not immediately preced2  3  ing /B(')( )/ vs. immediately preceding /B(')( )/ w  397  w  Table 3:6 Fi and F of tokens of St'at'imcets /j7 and lu l w  2  406  Table 3:7 Judgments of St'at'imcets /I/ and III in a non-pharyngealisation vs. pharyngealisation context  469  Table 3:8 The distinct properties of St'at'imcets' two postvelar harmonies  525  Table 4:1 Postvelar constraints proposed in this thesis  529  viii  List of Figures  Figure 1:1 Diagram of the vocal tract (from Ladefoged (1993:4))  36  Figure 1:2 X-ray tracings of[S] and [h] (from Al-Ani (1970:74))  38  Figure 1:3 Sagittal sections based on x-ray tracings of [?], [h], [B], and [%] (from Delattre (1971:130))  39  Figure 1:4 X-ray tracings of [T], [h], [B], and [%] (from Ghazeli (1977:50,57))  41  Figure 1:5 Ghazeli's identification template showing the position of his subject's tongue at rest (from Ghazeli (1977:30))  42  Figure 1.6 X-ray tracings of [t] and [t] (from Al-Ani (1970:57))  48  Figure 1:7 X-ray tracings of [t] and [t] (from Bonnot (1977:85))  49  Figure 1:8 X-ray tracings of [t], [t], [s] and [s] (from Ghazeli (1977:69,70))  50  Figure 1:9 Sagittal section based on an x-ray tracing of [k] (from Delattre (1971:130))  53  Figure 1:10 Conversion of area at constriction to area within a circle  57  Figure 1:11 Schematisation of vocal tract configurations for four sets of values of T , d, and All (from Stevens and House (1955:486))  58  Figure 1:12 Ghazeli's (1977:30) identification template with cm measurements  61  Figure 1:13 F and F predicted by Stevens and House (1955) for vocal tracts of d = 4, All = 6.7, and varying r (points for primary pharyngeal articulation plotted)  62.  Figure 1:14 Fi and F predicted by Stevens and House (1955) for vocal tracts of d = 7, All = 6.7, and varying r (points for primary uvular articulation plotted)  63  Figure 1:15 Fi and F predicted by Stevens and House (1955) for vocal tracts of d = 4, All = 6.7, and varying r (points for pharyngealisation plotted)  68  Figure 1:16 Fi and F predicted by Stevens and House (1955) for vocal tracts of d = 4, Alt = 1.1, and varying r (points for pharyngealisation plotted)  69  Figure 1:17 Fi and F predicted by Stevens and House (1955) for vocal tracts of d = 7, All = 1.1, and varying r (points for uvularisation plotted)  69  0  x  2  0  2  0  2  0  2  0  2  0  Figure 2:1 Fi - F plot of tokens of Palestinian short IV  161  Figure 2:2 Fi - F plot of tokens of Palestinian short IE/  162  Figure 2:3 Fi - F plot of tokens of Palestinian short 101  162  Figure 2:4 Fi - F plot of tokens of Palestinian short/U/  163  Figure 2:5 Fi - F plot of tokens of Palestinian short III in the contexts: (i) open syllable, no trigger; (ii) closed syllable, no trigger  ,164  Figure 2:6 Fi - F plot of tokens of Palestinian short IE/ in the contexts: (i) open syllable, no trigger; (ii) closed syllable, no trigger  165  Figure 2:7 Fi - F plot of tokens of Palestinian short 101 in the context: (i) open syllable, no trigger  165  Figure 2:8 Fi - F plot of tokens of Palestinian short IUI in the contexts: (i) open syllable, no trigger; (ii) closed syllable, no trigger  166  Figure 2:9 Fi - F plot of open syllable, adjacent postvelar tokens of Palestinian short IV • •  185  Figure 2:10 Fi - F plot of open syllable, adjacent postvelar tokens of Palestinian short IUI  185  2  2  2  2  2  2  2  2  2  2  Figure 2:11 Fi, - F plot of closed syllable, adjacent postvelar tokens of Palestinian 2  short IV  ^  186  Figure 2:12 Fi, - F plot of closed syllable, adjacent postvelar tokens of Palestinian 2  short IUI  187  Figure 2:13 Fi - F plot of open syllable, non-local harmony tokens of Palestinian 2  short IV  203  Figure 2:14 Fi - F plot of open syllable, non-local harmony tokens of Palestinian 2  short/E/  204  Figure 2:15 Fi - F plot of open syllable, non-local harmony tokens of Palestinian 2  short IUI  204  Figure 2:16 Fi - F plot of closed syllable, non-local harmony tokens of Palestin2  ian short IV  205  Figure 2:17 Fi - F plot of closed syllable, non-local harmony tokens of Palestin2  ian short IUI  206  X  Figure 2:18 Fi - F plot of tokens of Palestinian short /Ml  207  Figure 2:19 Fi - F plot of tokens of Palestinian short IMl in the contexts: (i) open syllable, no trigger; (ii) closed syllable, no trigger; (iii) open syllable, adajcent guttural; (iv) closed syllable, adjacent emphatic  208  Figure 2:20 Fi - F plot of non-rtr vs. rtr tokens of Palestinian Short IMl  209  2  2  2  Figure 2:21 Fi - F plot of stem-final tokens of Palestinian short III (Abu Shusha 2  and Jafa data)  230  Figure 2:22 Fi - F plot of stem-final tokens of Palestinian short IE/ (Abu Shusha 2  data)  230  Figure 2:23 Fi - F plot of stem-final tokens of Palestinian short IOI (Abu Shusha 2  data)  231  Figure 2:24 Fi - F plot of stem-final tokens of Palestinian short IUI (Abu Shusha 2  Shusha and Jafa data)  231  Figure 2:25 Fi - F plot of stem-final tokens of Jafa Palestinian short IE/  233  Figure 2:26 Fi- F plot of stem-final tokens of Jafa Palestinian short IOI  234  Figure 2:27 F! - F plot of non-pharyngealisation context vs. pharyngealisation context tokens of Palestinian/I:/  236  2  2  2  s  Figure 2:28 Fi - F plot of non-pharyngealisation context vs. pharyngealisation 2  context tokens of Palestinian IE: I  23 7  Figure 2:29 Fi - F plot of non-pharyngealisation context vs. pharyngealisation 2  context tokens of Palestinian IM: / > -I ae: \  23 8  Figure 2:30 Fi - F plot of non-pharyngealisation context vs. pharyngealisation 2  context tokens of Palestinian 10:1  239  Figure 2:31 Fi - F plot of non-pharyngealisation context vs. pharyngealisation 2  context tokens of Palestinian lUil  240  xi  Figure 2:32 Fi - F plot of tokens of Palestinian short IRI in the contexts: (i) no 2  emphatic; (ii) blocked; (iii) emphatic + open syllable; (iv) emphatic + closed syllable  255  Figure 2:33 Fi - F plot of tokens of Palestinian /Mi/ m the contexts: (i) no 2  emphatic; (ii) blocked; (iii) emphatic + open syllable; (iv) emphatic + closed syllable  257  Figure 2:34 Wideband spectrogram of one token each of i tirae:m} 'blanket' and ITtaVrBimr-'shame' (N)  260  Figure 2:35 Wideband spectrogram showing one token each of l\l> \ t \ and l\l>{\\  261  Figure 2:36 Wideband spectrogram showing one token each of \h\ and \\\\  263  Figure 2:37 Wideband spectrogram showing one token each of \ \\ and {S\  264  c  Figure 2:38 Wideband spectrogram showing one token each of i t\} and {h} Figure 2:39 Wideband spectrogram showing one token each of ^ Br and -IBIFigure 2:40 Wideband spectrogram showing one token each of l%X^ d "iX(" a n  265 266 267  Figure 2:41 Wideband spectrogram showing one token each of -it \ in a word containing no emphatic, and blocked U r  288  Figure 2:42 Wideband spectrogram of one token each of -I ti:} and i ti:}  297  Figure 2:43 Wideband spectrogram of one token each of i te: 1- and itei\  298  Figure 3:1 Wideband spectrogram of bursts of St'at'imcets [k] and [k] Figure 3:2 Wideband spectrogram of St'at'imcets [I] and \\]  375 -  376  Figure 3:3 Wideband spectrogram of a token of Lower St'at'imcets •Ixnij'-cu'i'gooseberry bush'  377  Figure 3:4 Wideband spectrogram of a token of Upper St'at'imcets -ixnij'-aej'} 'gooseberry bush'  379  xii  Figure 3:5 Wideband spectrogram showing a lowered F for St'at'imcets III 3  immediately preceding a uvular resonant  399  Figure 3:6 Wideband spectrogram showing a lowered F for the St'at'imcets 3  epenthetic vowel immediately preceding a uvular resonant  400  Figure 3:7 Wideband spectrogram of a token of -I B uj't|- 'to sleep'  407  Figure 3:8Fi-F plot of tokens of St'at'imcets/I/. Speaker: LC  417  Figure 3:9 Fi - F plot of tokens of St'at'imcets III. Speaker: L N  418  Figure 3:10 Fi - F plot of tokens of St'at'imcets IUI. Speaker: LC  419  Figure 3:11 Fi - F plot of tokens of St'at'imcets IUI. Speaker: L N  420  Figure 3:12 F - F plot of tokens of St'at'imcets IMl. Speaker: LC  421  Figure 3:13 F i - F plot of tokens of St'at'imcets/^/. Speaker: L N  422  Figure 3:14 Fi - F plot of tokens of the St'at'imcets epenthetic vowel. Speaker: LC  423  w  2  2  2  2  x  2  2  2  Figure 3:15 Fi - F plot of tokens of the St'at'imcets epenthetic vowel. 2  Speaker: L N  424  Figure 3:16 Wideband spectrogram of St'at'imcets [ o] and [A]  426  Figure 3:17 Fi - F plot of tokens of St'at'imcets III in the context: (i) with no postvelar in the word. Speaker: LC  428  2  Figure 3:18 Fi - F plot of tokens of St'at'imcets III in the context: (i) with no 2  postvelar in the word. Speaker: L N  429  Figure 3:19 Fi - F plot of tokens of St'at'imcets IUI in the context: (i) with no 2  postvelar in the word. Speaker: LC  430  Figure 3:20 Fi - F plot of tokens of St'at'imcets IUI in the context: (i) with no 2  postvelar in the word. Speaker: LN  xiii  431  Figure 3:21 Fi - F plot of tokens of St'at'imcets IMl in the context: (i) with no postvelar in the word. Speaker: LC 2  432  Figure 3:22 Fi - F plot of tokens of St'at'imcets IMl in the context: (i) with no 2  postvelar in the word. Speaker: LN  433  Figure 3:23 Fi - F plot of tokens of the St'at'imcets epenthetic vowel in the 2  context: (i) with no postvelar in the word. Speaker: LC  434  Figure 3:24 Fi - F plot of tokens of the St'at'imcets epenthetic vowel in the 2  context: (i) with no postvelar in the word. Speaker: LN  435  Figure 3:25 Fi - F plot of tokens of St'at'imcets III in the contexts: (ii) immedi2  ately preceding a guttural; (iii) immediately following a guttural. Speaker: LC  436  Figure 3:26 Fi - F plot of tokens of St'at'imcets III in the contexts: (ii) immedi2  ately preceding a guttural; (iii) immediately following a guttural. Speaker: LN  437  Figure 3:27 Fi - F plot of tokens of St'at'imcets IUI in the contexts: (ii) immedi2  ately preceding a guttural; (iii) immediately following a guttural. Speaker: LC  438  Figure 3:28 Fi - F plot of tokens of St'at'imcets IUI in the context: (iii) immedi2  ately following a guttural. Speaker: LN Figure 3:29 Fi - F plot of tokens of St'at'imcets l/EI in the context: (iii) immediately following a guttural. Speaker: LC  439  2  440  Figure 3:30 Fi - F plot of tokens of St'at'imcets IMl in the context: (iii) immedi2  ately following a guttural. Speaker: LN  441  Figure 3:31 Fi - F plot of tokens of the St'at'imcets epenthetic vowel in the 2  contexts: (ii) immediately preceding a guttural; (iii) immediately following a guttural. Speaker: LC  442  Figure 3:32 Fi - F plot of tokens of the St'at'imcets epenthetic vowel in the 2  contexts: (ii) immediately preceding a guttural; (iii) immediately following a guttural. Speaker: LN  443  XIV  Figure 3:33 Fi - F plot of tokens of St'at'imcets III in the contexts: (iv) immedi2  ately preceding an emphatic; (v) immediately following an emphatic. Speaker: LC  445  Figure 3:34 F! - F plot of tokens of St'at'imcets III in the contexts: (iv) immedi2  ately preceding an emphatic; (v) immediately following an emphatic. Speaker: L N  446  Figure 3:35 Fi - F plot of tokens of St'at'imcets IUI in the context: (v) immedi2  ately following an emphatic. Speaker: LC  447  Figure 3:36 Fi - F plot of tokens of St'at'imcets IUI in the context: (v) immedi2  ately following an emphatic. Speaker: L N  448  Figure 3:37 Fi - F plot of tokens of St'at'imcets IV in the context: immediately 2  preceding a laryngeal. Speaker: LC  449  Figure 3:38 Fi - F plot of tokens of St'at'imcets IV in the contexts: immediately 2  preceding a laryngeal or immediately following a laryngeal. Speaker: L N  450  Figure 3:39 Fi - F plot of tokens of St'at'imcets IUI in the (simultaneous) con2  texts: immediately preceding a laryngeal and immediately following a laryngeal. Speaker: LC  451  Figure 3:40 Fi - F plot of tokens of St'at'imcets IUI in the (simultaneous) con2  texts: immediately preceding a laryngeal and immediately following a laryngeal. Speaker: L N  452  Figure 3:41 F - F plot, of tokens of St'at'imcets IMl in the contexts: immediately x  2  preceding a laryngeal or immediately following a laryngeal. Speaker: LC  453  Figure 3:42 Fi - F plot of tokens of St'at'imcets IMl in the contexts: immediately 2  preceding a laryngeal or immediately following a laryngeal. Speaker: L N  454  Figure 3:43 Fi - F plot of tokens of the St'at'imcets epenthetic vowel in the 2  context: immediately following a laryngeal. Speaker: LC  XV  455  Figure 3:44 Fi - F plot of tokens of St'at'imcets IMl in the contexts: immedi2  ately preceding an emphatic or preceding an emphatic with an intervening phonetic laryngeal vs. all other contexts. Speaker: LC Figure 3:45 Fi - F plot of tokens of St'at'imcets IMl in the contexts: immedi2  ately preceding an emphatic or preceding an emphatic with an intervening phonetic laryngeal vs. all other contexts. Speaker: LN Figure 3:46 Fi - F plot of tokens of the St'at'imcets epenthetic vowel in the 2  contexts: immediately preceding an emphatic vs. all other contexts. Speaker: LC Figure 3:47 Fi - F plot of tokens of the St'at'imcets epenthetic vowel in the 2  contexts: immediately preceding an emphatic vs. all other contexts. Speaker: LN Figure 3:48 Fi - F plot of tokens of the St'at'imcets epenthetic vowel in the 2  contexts: preceding an emphatic with intervening non-laryngeal. Speaker LC Figure 3:49 Fi - F plot of tokens of the St'at'imcets epenthetic vowel in the 2  contexts: preceding an emphatic with intervening non-laryngeal. Speaker LN Figure 3:50 Wideband spectrogram of tokens of St'at'imcets /tj/ in non-uvularisation contexts Figure 3:51 Fi - F plot of tokens of St'at'imcets IMl in the contexts: immedi2  ately preceding an emphatic vs. immediately following an emphatic. Speaker: LC Figure 3:52 Fi - F plot of tokens of St'at'imcets /Ml in the contexts: immedi2  ately preceding an emphatic vs. immediately following an emphatic. Speaker: LN Figure 3:53 Fi - F plot of tokens of the St'at'imcets epenthetic vowel in the 2  contexts: immediately preceding an emphatic vs. immediately following an emphatic. Speaker: LC  xvi  Figure 3:54 Fi - F plot of tokens of the St'at'imcets epenthetic vowel in the 2  contexts: immediately preceding an emphatic vs. immediately following an emphatic. Speaker: LN  493  Figure 3:55 Wideband spectrogram of tokens of St'at'imcets IMl in a non-uvularisation vs. uvularisation context  495  Figure 3:56 Wideband spectrogram of tokens of the St'at'imcets epenthetic vowel in a non-uvularisation vs. uvularisation context  496  Figure 3:57 Wideband spectrogram of tokens of St'at'imcets /tj/ in a non-uvularisation vs. uvularisation context  -  497  Figure 3:58 Wideband spectrogram of tokens of St'at'imcets III in a non-uvularisation vs. uvularisation context  511  Figure 3:59 Fi - F plot of tokens of St'at'imcets IMl in the context: preceding an 2  emphatic with intervening phonetic laryngeal. Speaker: LC  516  Figure 3:60 Fi - F plot of tokens of St'at'imcets IMl in the context: preceding an 2  emphatic with intervening phonetic laryngeal. Speaker: LN  xvii  517  Acknowledgments I thank my co-chairs Doug Pulleyblank and Pat Shaw for teaching me phonology and my phonetics supervisor Andre-Pierre Benguerel for teaching me phonetics.  I am  very appreciative of their painstaking guidance, help, and encouragement in this project. They have taught me much beyond linguistics and speech science, for which I will always be grateful. I am indebted to my consultants for providing their language data — for Palestinian, the Bilbesy, Masri, and Shahin families of Ramallah and New Jersey, also Khaled Gaber and Maher Nakhala; for St'at'imcets, Alice Adolph, Robert Charley, Les Bucky Ned, Gertrude Ned, and Laura Thevarge. I thank Ibrahim Kassas, Adam Khayu, M.D. Kinkade, John McCarthy, Dil Parkinson, and Munther Younes for providing additional Salish and Semitic data. I thank Fahmi Al-Aboudi for intellectual support during the Ramallah fieldwork, and Henry Davis, Lisa Matthewson, and Norm Pogemuller for facilitating thefieldworkin Lillooet and Mission, B.C. The funding of SSHRC grant #410-92-1629 to Pat Shaw and SSHRC grant #410-94-0035 and a UBC HSS grant, both to Doug Pulleyblank, is gratefully acknowledged. I thank the UBC School of Audiology and Speech Sciences for the use of their Kinsmen phonetics lab and John Nicol for company over some of the late hours there. I also thank Munther Younes for reviewing and providing comments on an earlier draft of chapter 2, and Henry Davis, Jan van Eijk, and M.D. Kinkade for doing the same for chapter 3. I thank M.D. Kinkade for lending his ear to some key Palestinian forms. For helpful feedback and discussion I am grateful to many people.  Deserving  special mention are Stuart Davis, Bushra Zawaydeh, Donca Steriade, Ewa CzaykowskaHiggins, Nicola Bessell, Bert Vaux, Mark Hewitt, and Su Urbanczyk, also David Ingram, Barbara Bernhardt, and Mohamed Elmedlauoi. The encouragement of many people,  xviii  including all the members of UBC Linguistics, made "this task lighter. Particular thanks to Hamida Demirdache, Guy Carden, Laura Downing, Monica Sanchez, Anne-Marie Pulleyblank, Michael Kenstowicz, and Edit Doron. For encouragement in my earlier training, I thank Joe Kess. I thank my fellow graduate students in UBC Linguistics for their comeraderie and friendship, especially my classmates Myles Leitch, Olanike Ola, Ping Jiang-King, Aki Uechi, Sue Blake, Bill Turkel, and Taylor Roberts.  I am grateful to  Carmen da Silva for administrative help. I thank my parents and brothers and sisters for always caring. Finally, and most of all, I thank my husband Khaled for his great support and help and Alaa', Salah, Hosam, and Sarah for always providing happy relief. This thesis is written in loving memory of my mother-in-law, Sarah Shahin.  xix  Soli Dei Gloriae  XX  Chapter 1: Introduction  1.1. Aims  This thesis investigates postvelar harmony in two typologically, genetically, and geographically unrelated languages: Palestinian Arabic, a Semitic language, and St'at'imcets (Lillooet), a Salish language. Thefirstaim is to present evidence that both these languages have two distinct postvelar harmonies, which are articulatorily implemented as pharyngealisation harmony and uvularisation harmony. Acoustic support for the distinction between the two harmonies in each language will be presented. In identifying these two harmonies, the present analysis differs from previous analyses of both Palestinian and St'at'imcets. Previous studies have described only one postvelar harmony for Palestinian: 'emphasis spread'.  1  With the exception of  Elmedlaoui (1995) on Moroccan Arabic, the same is true for Arabic in general.  2  Likewise,  previous  studies have  described  only  one  postvelar  harmony for  'See, e.g., Younes (1982), Card (1983), Herzallah (1990), Younes (1993), and Davis (1995). See, e.g., Lehn (1963), Maamouri (1967), Ghazeli (1977), Goad (1991), Bessell (1992),  2  and Goad (1993).  However, Elmedlaoui (1995) analyses Moroccan 'emphasis' as  consisting of more than one type of harmony.  As I have unfortunately learned of  Elmedlaoui's study only after completing this work, it is not discussed in this thesis.  1  1.1. Aims  St'at'imcets: 'retraction'. The same is true for Salish in general. I will argue that there 3  4  is more complexity to Arabic and Salish postvelar phonology than has been previously recognised. The second aim is to show that pharyngealisation harmony and uvularisation harmony are harmony of the feature (primary- or secondary-) [RTR], and of secondary[DOR] + secondary-[RTR], respectively.  Pharyngealisation harmony is triggered by  postvelar consonants, that is, by both gutturals and emphatics.  The emphatic class  includes segments, such as Isl, that are postvelar-articulated counterparts of non-postvelar consonants (e.g., Isl is the postvelar-articulated counterpart of non-emphatic Isf). Gutturals include the pharyngeals and uvulars /T fi B %l and, in Arabic, the laryngeals /? hi. Uvularisation harmony is triggered only by emphatics. The representations I will propose for gutturals are seen in (1). Those I will propose for emphatics are seen in (2). (In (1) and (2), only the specifications relevant to the two harmonies are shown.) It is assumed that tongue root retraction is represented by [RTR] under [TR], not by [TR] by itself, based on crosslinguistic evidence that [TR] dominates both [RTR] and [ATR], the latter representing tongue root advancement; this issue will be discussed further in §1.3.2. The stacked place specifications for the uvular gutturals and emphatics follow Selkirk's  See, e.g., van Eijk (1985), Czaykowska-Higgins (1987), Remnant (1990), and Bessell  3  (1992). See, e.g., Mattina (1979), Kuipers (1981), Cole (1987), Doak (1987, 1989), Kuipers  4  (1990), Bessell and Czaykowska-Higgins (1991), Bessell (1992), and Doak (1992).  2  1.1. Aims  (1988) representation of primary vs. secondary place. This type of representation will be explained in §1.3.3.2.  (1) The Representations of Gutturals a. pharyngeal and laryngeal gutturals: primary [TR]/[RTR]  b. uvular gutturals: primary [DOR] and secondary [TR]/[RTR] 0 Place  oPlace  1 [DOR]  [TR]  I  [TR]  I  [RTR]  [RTR]  (2) The Representations of Emphatics: primary [COR], [DOR], or [LAB], secondary [DOR] and secondary [TR]/[RTR] a. coronal emphatics  b. dorsal emphatics  0 Place  o Place  1 [COR]  I [DOR]  POR]  poSf 1  [RTR]  R  ]  [RTR]  c. labial emphatics 0 Place 1 [LAB]  IPO^T  ^ .  -^[TR] [RTR]  3  1.2. Overview of the Thesis  The third aim is to present an Optimality Theory (Prince and Smolensky (1993), McCarthy and Prince (1993a)) account of the two harmonies in Palestinian and St'at'imcets.  First, it will be argued that ranked constraints on input/output  Correspondence (McCarthy and Prince (1995)), featural Alignment (McCarthy and Prince (1993b)), and phonetic grounding (Archangeli and Pulleyblank (1994a)), both syntagmatic (Shahin (1993), Jiang-King (1996), Pulleyblank (1997)) and paradigmatic (Archangeli and Pulleyblank (1994a)), are primarily responsible for the properties of both harmonies in each language. Second, the crosslinguistic variation in the two harmonies will be argued to be due to simple constraint reranking.  1.2. Overview of the Thesis  In the remainder of this chapter, the representational assumptions of this study are first presented. The articulatory and acoustic properties of postvelar segments are then discussed. Next, a typology of phonological harmony is proposed. Optimality Theory is then introduced and the general nature of Correspondence, Alignment, and Grounded constraints is explained. Finally, the distinction between phonetics and phonology, and the role of phonetics in phonology, are discussed. Chapter 2 first introduces Palestinian Arabic, clarifying the identity of the Palestinian phonemic inventory, with particular focus on the vocalic system. presents phonological  evidence for Palestinian's  4  distinct  It then  pharyngealisation and  1.2. Overview of the Thesis uvularisation harmonies, and acousticfindingsthat support that distinction. In chapters 2 and 3, to set the strictly acoustic portions of the thesis off from the rest of the text, the sylised spectrogram seen here:  will appear above the header of each section that presents acoustic findings.  Chapter 2 develops a theoretical account of the two harmonies in Palestinian. The OT constraints that impose the properties of each harmony are proposed. Their ranking is identified and the constraint interaction producing the two postvelar harmonies is shown. Chapter 3 first introduces St'at'imcets Salish.  It clarifies the St'at'imcets  phonemic inventory, with focus on both the consonants and the vowels. It then presents phonological evidence for pharyngealisation harmony and uvularisation harmony in the language, and supporting acoustic findings. The specific properties of the two harmonies are shown to differ from those found in Palestinian. However, the fundamental nature of the two harmonies across the two languages is argued to be the same. The OT constraints responsible for St'at'imcets' postvelar harmonies are identified. Most are the same as those identified for Palestinian in chapter 2, but a reranking of the constraints yields St'at'imcets' own version of the two harmonies.  5  1.3. Representational Assumptions  The final chapter summarises the thesis. In conclusion, it will be suggested that Niger Congo and Nilotic '[-ATR]' harmony is pharyngealisation harmony, but with a non5  consonantal source. The thesis will propose that languages in which pharyngealisation harmony has a non-consonantal source are predicted by the optimally vocalic realisation of the pharyngealisation harmony feature, [RTR]; see §2.4.2. Typological work investigating possible systemic differences between consonantal-source vs. non-consonantal-source pharyngealisation harmony, that is, work investigating the phonological consequences of a consonantal vs. vocalic source, is beyond the scope of this thesis.  1.3. Representational Assumptions  The feature geometry assumed here is seen in (3).  6  (See Appendix I for  explanation of the abbreviations and symbols used in this thesis.)  Note that the  representations in (1) and (2) differ from this geometry by showing [COR], [DOR], and [LAB] to dominate [DOR] and/or [TR]/[RTR]. This apparent mismatch will be cleared up in §1.3.3.2.  See, e.g., Stewart (1967), Clements (1985), Ka (1988), Archangeli and Pulleyblank  5  (1989), Clements (1991), Odden (1991), and Archangeli and Pulleyblank (1994a). Tonal features and [SUCTION] will not be discussed in this thesis, as neither is directly  6  relevant to the phonological phenomena under investigation.  See Archangeli and  Pulleyblank (1994a) and Jiang-King (1996) for arguments that tonal features link directly to the mora. See Halle (1995) for evidence for [SUCTION], specified on clicks, and its direct linking to the root node.  6  1.3.1. Articulator Theory (3)  [POST] [DISTR] [STRID]  [HI]  [LOW] [FRONT]  [ A T R ] [RTR]  Important assumptions that bear on this geometry will now be laid out.  1.3.1. Articulator Theory  Articulator Theory, developed by McCarthy (1985), Sagey (1986), and Halle (1988), will be assumed.  Articulator Theory assumes, as do other feature geometric  theories (see Clements (1985) and Mester (1986)) that distinctive features do not occur as unordered bundles, but in a hierarchical arrangement. The evidence for this is two-fold, as summarised by Kenstowicz (1994:146): first, certain features "introduc[e] a subdistinction within the category defined by another feature" (e.g., [DISTR], a finer distinction of [COR]); second, certain features "form recurrent groupings in phonological rules and  1.3.1. Articulator Theory  constraints" (e.g., vowels show assimilation for height features like fHTJ and [LOW], but not for [NAS]). Articulator Theory further assumes that phonological features can be defined in either articulatory or acoustic terms (see McCarthy (1988:99)) and that their hierarchical ordering "directly reflects aspects of the human anatomy used in the production of speech" (Halle and Vaux (1994:1)).  (Features which are defined in acoustic terms are here  assumed to have an articulatory basis.) McCarthy (1988:105) traces the roots of this theory to "Jakobson's fundamental insight in the late 1930s that the classification of speech sounds exploited in phonology has a universal phonetic basis." Articulator Theory is widely adopted in phonological theory. For further discussion of its justification, see 7  Sagey (1986), McCarthy (1988), and Halle and Vaux (1994). A second theory is Vowel Place Theory, so-named by Halle and Vaux (1994) to 8  highlight its central difference from Articulator Theory. Unlike Articulator Theory, Vowel Place Theory assumes that the feature geometry is primarily structured according to function. Phonetic basis is assigned a secondary role. The result is the proposal of distinct 'C-Place' and ' V-Place' nodes, motivated primarily to account for the observed non-effect of intervening consonants in cases of vowel assimilation. Halle and Vaux (1994) reject  7  For examples of works which adopt Articulator Theory, see Halle (1989, 1992), Selkirk  (1993) , Archangeli and Pulleyblank (1994a), Kenstowicz (1994), McCarthy (1994), Vaux (1994) , and Halle (1995). 8  For examples of works which adopt Vowel Place Theory, see Steriade (1987), Clements  (1989), Herzallah (1990), Clements (1991), Odden (1991), Ni Chiosain and Padgett (1993), Clements and Hume (1995), and Padgett (1995).  8  1.3.1. Articulator Theory  Vowel Place Theory because its V-Place node "has no clear anatomical status" [p.4]. They readdress the data presumed by Vowel Place theorists to indicate the V-Place node and show that they can be straightforwardly handled within Articulator Theory.  (See  Selkirk (1993) and Vaux (1994) for further critique of Vowel Place theory.) It is assumed here that 'articulation' refers to constriction by the lips or tongue. Under this definition, laryngeal gesture is not an articulation. It is considered instead to be an airstream mechanism; see §1.3.3.1 for further discussion. The four articulators used in producing speech sounds are the lips, tongue blade/tip, tongue dorsum, and tongue root.  9  They are represented in (3) as the features [LAB], [COR], [DOR] and [TR]. The possibility of an Oral node under Place, as proposed by McCarthy (1994), will not be discussed, as 10  the presence or absence of this node does not impinge on the analysis of postvelar harmony to be presented in this thesis.  ^or examples of previous works which assume these four articulators, see Cole (1987), McCarthy (1988), Selkirk (1993), and McCarthy (1994). 10  For further discussion of a possible Oral node, see Clements and Hume (1995:272) and  Rose (1996).  9  1.3.2. The Articulator Feature [TONGUE ROOT]  1.3.2. The Articulator Feature [TONGUE ROOT]  The feature [TR] is assumed in the present work to represent the active articulator, tongue root. It is assumed to dominate the features [RTR] and [ATR], based on the evidence for distinct retraction and advancement features in Kiparsky (1985), Ringen (1989), Goad (1991), Casali (1993), Archangeli and Pulleyblank (1994a:200-245), and Steriade (1995a:149-151).  This thesis investigates tongue root phenomena involving  [RTR]. [TR] is not assumed to be a feature defining an 'orosensory region', as proposed by McCarthy (1994) for the feature [PHARYNGEAL].  McCarthy (1994) proposes  [PHARYNGEAL] because of a classificatory problem relating to laryngeal gutturals. The discussion that follows first addresses what laryngeal gutturals are, then summarises the problem they have posed. It then summarises newfindingsof this thesis which suggest that laryngeal gutturals do not pose a problem after all.  1.3.2.1. The Assumed Problem with Laryngeal Gutturals  A laryngeal guttural is a /?/ or /h/ that patterns with pharyngeal and uvular consonants (e.g., /T n ts %[) as a natural class.  Such laryngeals have been shown to exist  in Arabic and Hebrew (both Semitic) by McCarthy (1994), in Nisga (Tsimshianic) by Shaw (1991a), and in Iraqw (Cushitic) by, e.g., van der Hulst and Mous (1992) and Mous  10  1.3.2. The Articulator Feature [TONGUE ROOT]  (1993). Languages in which laryngeals do not pattern with pharyngeals and uvulars include Interior Salish languages (see Bessell and Czaykowska-Higgins (1991) and Bessell (1992)) and Tigre (Ethiopic Semitic; see McCarthy (1994) and Rose (1996)). That is, laryngeals do not necessarily pattern with pharyngeals and uvulars; rather, their patterning depends on the language. The natural class consisting of pharyngeals and uvulars, and sometimes laryngeals, is termed the 'guttural' class by McCarthy (1994). McCarthy (1994) presents four types of evidence for the guttural class: (i) root co-occurrence restrictions; (ii) vowel lowering; (iii) avoidance of syllable-final gutturals; (iv) degemination.  Examples of (i) - (iv) from  McCarthy (1994) will be reviewed below. As will be shown, he provides evidence for the patterning of laryngeals with pharyngeals and uvulars in Arabic, which shows that the Arabic guttural class consists of /? h T ft u xl  n  (See Cole (1987) on Coeur d'Alene  (Southern Interior Salish), Shaw (1991a) on Nisga, and Hayward and Hayward (1989) and Rose (1996) on Cushitic for further evidence for the guttural class.) McCarthy draws his first set of evidence from Arabic root co-occurrence restrictions. Arabic roots "rarely or never contain adjacent homorganic consonants" [p.203]. Of interest here is the fact that "[r]oots combining two gutturals are significantly infrequent" [p.205]. As McCarthy points out, the co-occurrence restriction with respect to gutturals 12  u  The guttural class of a particular variety of Arabic might include other gutturals, e.g.,  uvular / G / , documented for Gazan Palestinian by Abu Shark (1997). See McCarthy's (1994:204) table showing tliis statistical infrequency.  12  11  1.3.2. The Articulator Feature [TONGUE ROOT]  requires reference to gutturals as a class, as a significant generalisation is lost if that class is not recognised. The second type of evidence is from vowel lowering. McCarthy presents data from Arabic and Tiberian Hebrew (a pronunciation tradition of Biblical Hebrew) which show such lowering. Examples are presented in (4) and (5). In the transcriptions below, the braces Mr' are used to denote the fact that the data are surface phonological forms — distinct from underlying phonological forms, which are standardly enclosed by slashes ('//'), and phonetic forms, which are standardly enclosed by square brackets ('[ ]'). This notation will be used throughout this thesis. The distinction between underlying phonological form, surface phonological form, and phonetic form will be discussed in §7.1.1. In the present section, the surface phonological status of forms enclosed by Mr' is inferred from McCarthy's discussion. In all other respects, the data cited in this section follow McCarthy's transcription; 'j' denotes a palatal approximant; a capital letter denotes an emphatic consonant (e.g., 'D' denotes emphatic d). In Modern Standard Arabic, the last vowel of the imperfect verb stem is lowered to -I ar when adjacent to a guttural. This is illustrated by the Modern Standard Arabic verbal forms in (5), in comparison with those in (4). These data are from McCarthy (1994:207). (I have added the ungrammatical imperfect forms for clarity.) The relevant generalisation with respect to (4) and (5) is: whereas thefinalvowel of the imperfect verb stem surfaces unpredictably as non-low Vi\ or {u\ when it is not adjacent to a guttural, it surfaces predictably as low *! ar when it is adjacent to a guttural.  12  1.3.2. The Articulator Feature [TONGUE ROOT] (4) Modern Standard Arabic Non-lowered Imperfects - perfect imperfect a: -Ikatabf Ijaktubr (^jaktaM) b. ^Darabr (jaDriM (^jaDrabr)  'to write' 'to beat'  (5) Modern Standard Arabic Lowered Imperfects perfect imperfect a. ifaTaU ijafialr (^jafTiU, ^jafTulr) b. •jradaTj-IjardaTr ("MjardiTK 'MjarduT^)  'to do' 'to prevent'  McCarthy provides no pairs involving laryngeals or uvulars. However, the further data in (6), which illustrate the lowering in forms in which the guttural is a laryngeal or a 13  uvular, support McCarthy's generalisation.  (6) Modern Standard Arabic Lowered Imperfects with Laryngeal and Uvular Gutturals perfect imperfect a. <!5ahaU •ijabhaU (*<!ja5hiU, *-!ja5huU) 'to forget, overlook' b. -lo^ahaz^ ^jao^hazl(*-|jad3hiz^, ^jacfehuzr) 'to finish off (e.g., a wounded person)' c •jjamaxl' -ijajmaxl('Mjajmixr-,'MjaJmuXr) 'to be high, tall, lofty' d. •itaxaml' -Ijatxaml(^atxim!-, *-!jatxumf) 'to suffer from indigestion'  A second, related type of phenomenon is post-guttural blocking of vowel raising in 'Anaiza (Saudi) Bedouin. (McCarthy (1994:212) refers to it as a phenomenon in which raising "is systematically blocked or undone".)  This is illustrated by the data in (8),  compared to those in (7). The data in (7) and (8) are from McCarthy (1994:213) and (1994:212), respectively.  (I have added the ungrammatical forms. McCarthy provides  only surface forms for the data in (8); I have added the underlying forms as implied by his discussion.) The relevant generalisation from (7) and (8) is: whereas the initial vowel of  13  I thank Dil Parkinson for help with the data in (6). 13  1.3.2. The Articulator Feature [TONGUE ROOT]  the verb or noun stem surfaces predictably as raised il\ when it does not immediately follow a guttural, it surfaces predictably as low -i a\ when it immediately follows a guttural.  (7) 'Anaiza (Saudi) Bedouin Raised Forms a. /katab/ ^ kitab r ("MkataM) b. /djamal/ <! dermal <• (*-|d5amalr) c. /bagar/ •{ bigar )• (*-jbagar|-)  'he wrote' 'camel' 'cows'  (8) 'Anaiza (Saudi) Bedouin Forms in which Raising is Blocked a. /?akal/ i?aka\\ (*i?ikalr) 'he ate' b. /hadjim/ H hacfeiin 1- (**!hid3iinr) (type of camel) c. /Tazam/ -ITazaml(*<i Tizamr) 'he invited' d. /fiasuud/ -jftasuud!(*{ tiisuudr) 'envious' e. /BaSab/ ^BaSaM (^BiSaM) 'he forced' f. /xazan/ \x&zan\ (*-!xizanr) 'he stored'  Another example of lowering comes from Tiberian Hebrew: the Tiberian epenthetic vowel is lowered immediately preceding a guttural.  14  This is illustrated by the forms in  (10), compared to those in (9). These data are from McCarthy (1994:210). (I have added the ungrammatical forms.) The relevant generalisation is: whereas the epenthetic vowel in a CVCC noun stem surfaces predictably as non-low {e} when it does not immediately follow a guttural, it surfaces predictably as low -I a\ when it does.  (9) Tiberian Hebrew Non-Lowered Epenthetic Vowel a. /malk/ -{melekf(* |melak|') b. /sipr/ -)se:peri(*lse:par(-) c. /qudf/ -!qo:de^ (*<!qo:daj~f-) <  <  'king/my king' 'book' 'holiness'  McCarthy (1994:210) shows that the Tiberian epenthetic vowel is realised as a full vowel in closed syllables, as shown also by Prince (1975) and Garr (1989). 14  1.3.2. The Articulator Feature [TONGUE ROOT] (10) Tiberian Hebrew Lowered Epenthetic Vowel a. /tu?r/ -|to?ar^ (*{to?er}) a. /lahb/ UahaM (*UaheM) b. /baTl/ ^baTaU (^baTeU) c. /kanj/ ^kanaU ("Hkanejr)  'form/his form' 'flame' 'master' 'lying'  McCarthy (1994) does not provide examples of this lowering which involve Tiberian uvulars. However, the further forms in (11) show that the epenthetic vowel does not lower immediately following uvular / q / .  15  (11) Tiberian Hebrew Epenthetic Vowel in CVCC Nouns with /q/ as C a. /Jeqr/ -lyeqerl(^Jeqari-) 'deception, falsehood' a. /boqr/ <ibo:qer^ (*<!bo:qarr) 'morning' 2  The fact that Tiberian / q / does not trigger this lowering suggests it is not a guttural. As will be evident from the discussion in §1.4.2, the non-guttural status of a / q / such as that in Tiberian is explicable if it is recognised as emphatic [k] instead of a primary uvular like Itil and /%/. Alternatively, Tiberian /q/ might be a guttural, and the lowering just discussed might be triggered only by gutturals that are not also stops. The point about the Semitic lowering cases discussed above is that they require reference to the guttural natural class. The third set of evidence is from the avoidance of syllable-final gutturals in Negev Bedouin Arabic, Tiberian Hebrew, and Tigre.  This is illustrated by the data in (13),  compared to those in (12), which are from McCarthy (1994:214). (I have added the  I thank John McCarthy for providing the grammatical surface forms in (11). 15  1.3.2. The Articulator Feature [TONGUE ROOT]  ungrammatical forms, and periods to denote the first syllable break in each word.  16  See  McCarthy (1994:215-216) for data showing avoidance of syllable-final gutturals in Tiberian Hebrew and Tigre.) The generalisation with respect to (12) and (13) is: Negev Bedouin /CVCC.../ surfaces as iCVC.C...}.  By contrast, /CVGC.../ surfaces as  •I CV.GVC... I-, where G is a guttural. That is, the guttural surfaces in syllable-initial rather than syllable-final position.  (12) Negev Bedouin Arabic Plain Roots: {(C)VC.C... \ a. ijaf.rabr (*-ija.J...r) 'he drinks' b. UJ.raM (*la.J...r) 'I drink' c. -Ibnaf.raM (*-!bna.y...f-) 'we drink' d. Uaf.raM (*^ta.J...r) 'you drink' (13) Negev Bedouin Arabic Guttural Roots: *<! (C)VG.C... a. <!ja.hard3f(*-ijah.r.. r) 'he speaks' b. U.Tarft (^aT.r.r) 'I know' c. -ia.rialaml(*-{an.l...r) 'I dream' d. -ibna.BaziU (*-jbnaB.z... \) 'we spin' e. •{ta.xabar}' (*-!tax.b...}) 'you know' f  The point is the need to refer to the guttural class in the statement of the coda condition responsible for the syllabification in (13). The final set of evidence is from guttural degemination.  This is illustrated by the  Tiberian Hebrew examples in (15), compared to those in (14), from McCarthy (1994:217). (I have added the ungrammatical forms in (15).  17  See McCarthy (1994:217) for forms  showing guttural degemination in Tigre.) The generalisation here is: geminate gutturals  McCarthy (1994) provides no underlying forms for these data, so the underlying/non-  16  underlying status of the vowels in (12) and (13) remains unclarified here. "McCarthy (1994) provides no underlying forms for (14a-b) and (15a-b). 16  1.3.2. The Articulator Feature [TONGUE ROOT]  are not observed in contexts where they are expected, based on the gemination observed for non-guttural consonants in parallel forms.  This indicates that non-gutturals can  geminate in Tiberian, but gutturals cannot.  (14) Tiberian Hebrew Non-Guttural Roots a. -IdibbeTib. -i dallirmr c. /jinte:/ -!jitte:nr  'he said' 'weak ones' 'he gives'  (15) Tiberian Hebrew Guttural Roots a. -jme:?e:n|(*-!me??e:nr) b. ira:?i:mr(^raTTi^r) c. /jinnat/ -!je:nat|(*<!jefiriatr)  'he refused' 'evil ones' 'he marches down'  Guttural degemination also requires reference to gutturals as a class. In sum, Semitic consonant co-occurrence restrictions, vowel lowering, avoidance of syllable-final gutturals, and guttural degemination all require reference to a guttural natural class, as without some formal way of clarifying this group of segments uniquely, a significant set of linguistic generalisations would not be captured. McCarthy's facts establish that laryngeals are part of the guttural class in both Arabic and Hebrew. This is seen from the forms with laryngeals in (6a-b), (8a-b), (lOa-b), (13a) and (15a) as well as from McCarthy's (1994:204) table of roots combining adjacent consonants. The problem he identifies for laryngeal gutturals is that no phonetic basis for their guttural status can be found in the articulatory or acoustic literature. In particular, there is "no evidence for a pharyngeal or uvular constriction accompanying the glottal gesture" (McCarthy (1994:193)).  This is troublesome within Articulator Theory,  17  1.3.2. The Articulator Feature [TONGUE ROOT]  since features and feature groupings are supposed to be articulatorily based.  He  concludes: "the difference between Arabic and Tigre laryngeals, phonologically important but phonetically invisible, may have shown us a limit in our understanding of the relation between phonetic events and phonological features" (p.225). As a result, McCarthy (1994) proposes that the feature capturing the guttural class is not an articulator feature like [TR], but an orosensory feature, [PHARYNGEAL]. He defines [PHARYNGEAL] as "the orosensory pattern of constriction anywhere in the broad region of the pharynx" (p. 199), after Perkell (1971). His proposed geometry is seen in (16). (As noted in §1.3.1, his proposal for the Oral node under Place will not be discussed or adopted here.)  (16) McCarthy (1994) Geometry With Orosensory [PHARYNGEAL]  [lab] [cor] [dor]  [pharyngeal]  1.3.2.2. Another Look at Laryngeal Gutturals  This section first provides a critical examination of the bases for McCarthy's (1994) conclusion that there is no evidence that Arabic laryngeals are tongue root articulated. It will be suggested that the studies on which McCarthy's conclusion is based do not provide  18  1.3.2. The Articulator Feature [TONGUE ROOT]  conclusive evidence for that conclusion.  Finally and most significantly, this section  summarises new acoustic findings of this thesis, to be presented in chapter 2, which support an assumption that laryngeal gutturals are tongue root articulated. McCarthy (1994:193), noting first that there is no articulatory (viz., x-ray, fibreoptic, EMG, etc.) data on Arabic laryngeals, bases his conclusion on the fact that no acoustic data have shown that Arabic laryngeals have "formant transition or other effects on adjacent vowels". He cites thefindingsof Klatt and Stevens (1969) on Arabic laryngeals and Younes's (1982) use of Palestinian laryngeals as the 'neutral context' as evidence for this. Klatt and Stevens (1969) examined tokens of Arabic laryngeals, pharyngeals, and uvulars followed by tokens of the short vowels HI, IMl, and IUI, with each consonantvowel sequence forming a single open syllable. They describe their subjects as speakers of various Arabic dialects, and identify one of the dialects as Lebanese. They report: "[t]he /hi and the glottal stop /?/... differ from the pharyngeals in that there are no formant transitions at the vowel onsets" (p.211, underlining in the original). However, I suggest that no general conclusion with respect to Arabic laryngeals can be based on Klatt and Stevens' finding. This is because the vowel in a token of /hi/, IhMI, IhUI, l?V, I?Ml, or l?UI occurs in stem-final position. In chapter 2 it will be shown that Palestinian laryngeals trigger tongue root retraction harmony on short vowels, resulting in a raised Fx and lowered F for the vowels (except for tokens of the low short vowel, I Ml, 2  19  1.3.2. The Articulator Feature [TONGUE ROOT]  which show no Fi and F effects). However, this harmony is not observed if the vowels 2  are stem-final because stem-final short vowels do not undergo the harmony. In the (Beirut) Lebanese Arabic data in (17), short vowels occur in stem-final position.  (My transcriptions in this thesis follow the notational conventions of the  International Phonetic Alphabet (IPA), which is reproduced in Appendix I.  As an  exception, ' ', which denotes tongue root retraction in the IP A, will be used to denote 'emphasis', that is, simultaneous pharyngealisation and uvularisation; see §1.4.2 for further discussion of these secondary articulations.  The symbol '-' will denote a morpheme  boundary; '#' will denote a word boundary.)  (17) (Beirut) Lebanese Arabic Forms a. /b-j-lBlI/  ^ 'b-j-ibt.lir  b. /tillU/  Vni.lu^  c. / s ^ m m - U / Vsam.m-u!-  n'b-j'-iB.lir)  'he boils (something)'  OH'ni.lur)  'pretty (masc. sg.)'  (*-j'sam.m-ur)  'they (masc.) named (someone, something)'; 'they (masc.) said the words: bi-sm-illah ir-rahlm  ir-rahman  'in the name of God,  the Gracious, the Merciful' '  The stem-final short vowels in (17) surface non-tongue root retracted ('non-rtr'). They do not surface rtr, as would be expected if they harmonised with the guttural (17a-b) and/or rtr vowel (17a-c) in the word. (The expected harmony in the two contexts just mentioned is based the behaviour of short vowels in Palestinian, which will be detailed in §2.3 and §2.4. In the surface form in (17c), ia\ is rtr.) This indicates that stem-final  20  1.3.2. The Articulator Feature [TONGUE ROOT]  vowels in Lebanese, like stem-final vowels in Palestinian, do not undergo tongue root retraction harmony. The foregoing observations with respect to Palestinian and Lebanese suggest that Klatt and Stevens searched for acoustic effects from laryngeals in a context where none would be found for independent reasons. An examination of Younes' (1982) use of laryngeals as the 'neutral context' shows that his usage is based on thefindingsof Stevens and House (1963).  18  Thefindingof  Stevens and House (1963:116) with respect to laryngeals was that "comparison of the average data for the /h-d/ and the /#-#/ [isolation] environments shows that there is no significant difference between the values of Fi and F for these two environments." Based 2  on thisfinding,they conclude (ibid.) that "the vowel in the context /h-d/ is generated with essentially the same articulatory configuration as the vowel in isolation" and that  /h-d/  and /#-#/ are both 'null' contexts for vowels. However, Stevens and House analysed English laryngeals — they describe their speech material [p. 112] as "common vowels of American English and... consonants that can appear both initially and finally in" American English" — and English laryngeals are not gutturals. As stated by Klatt and Stevens (1969:207), "[i]n English all of the consonants are produced with a constriction in the oral cavity between the velum and the lips"; thus, English has no guttural class. The point here is: given McCarthy's (1994) phonological evidence that Arabic laryngeals do systematically behave as gutturals, Arabic laryngeals  See Younes' (1982:2) description of the procedures for his acoustic experiment.  18  21  1.3.2. The Articulator Feature [TONGUE ROOT]  cannot be considered a 'neutral' context like English laryngeals unless they are established as such. Based on the foregoing review of Klatt and Stevens (1969) and Younes (1982), there is as yet no clear evidence that Arabic laryngeals are not tongue root articulated. The investigation of Palestinian postvelar harmony to be presented in chapter 2 will show that Palestinian laryngeals, like the other gutturals, trigger tongue root retraction harmony, which will be labelled 'pharyngealisation harmony'. Acoustic data showing the effect of this harmony on vowel formants will be presented.  The finding is that  pharyngealisation harmony results in a raised Fi and lowered F for tokens of the 2  harmonising vowel. Important to the point at hand, the formant effects are observed even when the harmony trigger is a laryngeal. This means that Palestinian laryngeals have acoustic effects on vowel formants. Since a raised Fi and lowered F are expected for 2  segments which are produced with pharyngealisation articulation, as will be discussed in §1.4.3, the effects are consistent with an assumption that Palestinian laryngeals are tongue-root articulated. An alternative suggestion, that the raised Fi effect from Palestinian laryngeals might result from some completely non-tongue root articulation, is rejected here.  This is  because Palestinian laryngeals, pharyngeals, and uvulars constitute a guttural class and in Articulatory Theory, phonological classes are assumed to be based on some shared articulatory implementation.  22  1.3.3. Other Representational Assumptions  Note that the present assumption that Palestinian laryngeals are tongue-root articulated predicts, e.g., that tokens of Palestinian / h / will have some first formant resonance that is not observed for tokens non-guttural Ihl in a language such as English. Experimental investigation of this prediction remains for work outside the context of this thesis. In summary, this section has presented arguments for assuming that gutturals are all produced with a single articulator: the tongue root. On the basis of those arguments, I propose that all gutturals are specified for [TR] For this reason, [PHARYNGEAL], representing an orosensory region, does not appear in (3), but is replaced by the articulator feature [TR].  1.3.3. Other Representational Assumptions  Privative features, which are traced to Trubetskoy's (1939) privative oppositions between sounds, as discussed by Hyman (1975:27), are assumed.  19  It is assumed that  [CONS], and [SON] characterise the root node, as proposed by Schein and Steriade (1986) and Halle (1988); see McCarthy (1988) for further discussion. Given Shaw's (1991b) evidence that the phonological continuancy feature is one that represents no continuancy, the feature [STOP] is assumed. Following Shaw (1989, 1991b) and La Charite (1993),  19  For examples of previous works which assume privitive features, see Goldsmith (1985),  Rice and Avery (1989), and Steriade (1995). See also Schane (1984), Anderson, Ewen, and Staun (1985), van der Hulst and Smith (1985), Kaye, Lowenstamm, and Vernaud (1985), and Schane (1987) on similar unary-value assumptions. 23  1.3.3. Other Representational Assumptions  [STRID] is assumed to be a daughter of [COR], rather than of the root node. Given the substantial evidence for phonological fronting rather than backing (e.g., umlaut, palatalisation), the feature [FRONT], as used by McCarthy (1997), rather than [BACK], is assumed. Given the crosslinguistic evidence that coronals are subclassified for a feature representing non-anteriority,  20  the feature [POST], as the privative expression of  [-ANTERIOR], is assumed. Finally, following Selkirk (1993), it is assumed that the feature [ROUND] does not exist.  This is based on Selkirk's argument that a rounded  interpretation of secondary-[LAB] is always predictable and thus belongs in the phonetics.  1.3.3.1. Laryngeal Specifications  Following several previous works, it is assumed that the Laryngeal node, dominating the features [VOICE], [SG], and [CG] is directly dominated by the root node, not deeply 21  embedded in the geometry, as proposed by Halle (1995) and assumed by Halle and Vaux (1994) and Vaux (1994.) Halle's (1995) geometry is seen below:  See, e.g.,  Sagey's (1986:132-135) discussion of n-retroflexion in Sanskrit, as  documented by Steriade (1986), and Hirose (to appear) on palatalisation in Plains Cree. 21  For examples of previous works which assume the Laryngeal node to be directly  dominated by the root node, see Sagey (1986), Cole (1987), Archangeli and Pulleyblank (1994a), McCarthy (1994), and Clements and Hume (1995).  24  1.3.3. Other Representational Assumptions (18) The Halle (1995) Feature Geometry [suction] -— [c ontinuant] - — _ ^ ^ [strident][lateral]— [nasal] L  —  Soft Palate  [consonantal] [sonorant]  Halle (1995:18) bases the grouping of laryngeal and tongue root features on the finding of several previous studies that laryngeal characteristics frequently correlate with tongue root activity. E.g., Czaykowska-Higgins (1987:2), based on data in Gregerson (1976), reports that Mon Khmer rtr vowels are systematically accompanied by breathy voice, higher pitch, and voiceless initial consonants; the atr vowels are accompanied by creaky voice, lower pitch, and voiced initial consonants.  22  In the model in (18), this  correlation is assumed to have a structural basis: Tongue Root and Larynx are dominated by the same node, Guttural. I suggest, however, that the relation between laryngeal and tongue root features can be understood as a grounded relation! Under this view, the f /tongue root correlations 0  22  For further discussion of larynx-tongue root correlations, see Hayward and Hayward  (1989), Trigo (1991), Meechan (1992), and Vaux (1994). 25  1.3.3. Other Representational Assumptions  observed in Mon Khmer indicate paradigmatic grounded constraints like RTR/SG ('If [RTR], then [SG]') and ATR/CG ('If [ATR], then [CG]'). I propose that such constraints are paradigmatically grounded in the anatomical interconnectedness of the lower vocal tract structures: the tongue root, suprahyoid muscles, hyoid bone, thyroid and cricoid cartilages, and the laryngeal muscles — especially the cricothyroid (primary controller of longitudinal tension in the vocal folds, which determines pitch) and the lateral cricoarytenoid (primary controller of medial compression, which determines breathy or creaky voice); see Saunders (1964) and Zemlin (1988).  23  The breathy voice which is  sytematically observed for rtr vowels in Mon Khmer indicates that RTR/SG is highly ranked in that language. Further development of this grounded account is deferred for work elsewhere. The position of laryngeal features under the root node, structurally distant from all articulator features (that is, laryngeal features are not sisters to any articulator feature) follows from the fundamental source/filter distinction in the acoustic signal. After Fant (1960), the speech signal is produced when the glottis provides a source wave and the vocal tractfiltersthat wave. Four articulators can affect the transfer (filter) function: the 24  lips, tongue tip/blade, tongue dorsum (sometimes referred to as the 'tongue body'), and  2 3  See Laufer and Condax (1979), Denning (1989), Meechan (1992), Hirai et al. (1993),  Honda et al. (1993), and McCarthy (1994) for relevant articulatory findings and discussion. Fant (1960) also considers the 'radiation' function, which is thefinalmodification of the  24  signal caused by outward radiation from the lips. This will not be discussed here because it applies after the source andfilterfunctions. 26  1.3.3. Other Representational Assumptions  tongue root. These four can articulate at specific places. Although there is convincing argumentation that it is not 'place of articulation', but 'articulator' that is phonologically real, as discussed by McCarthy (1988), a node labelled 'Place' is retained in the tree. The Place node corresponds to the representation of a filter. By definition, Place dominates features that implement the four articulators: [LAB], [COR], [DOR], and [TR]. The larynx, by contrast, does notfilterthe speech signal. Rather, it provides a source wave and passes it on for potential articulatory modification by one or more of the four articulators. For this reason, (non-guttural) laryngeals, which are produced with only vocal fold adjustment, are placeless. Their placelessness has been expressed in several phonological analyses.  25  Vocal fold adjustments that effect voicing, glottalisation,  implosion, or aspiration are adjustments of the source, not of the filter. Consider Ladefoged's (1993) classsification of glottalisation under 'airstream meachanisms', not 26  under 'secondary articulation'.  His classification is consistent with the source/filter  distinction. The definition of'articulation' assumed in this thesis (see §1.3.1) also reflects the source/filter distinction. The position of the laryngeal node in (3) encodes the physical distinctions just discussed.  See, e.g., Clements (1985), Sagey (1986), Steriade (1987), Bessell and Czaykowska-  25  Higgins (1991), and Bessell (1992). After Pike (1943) 'airstream mechanism' refers to the movement of a body of air for  26  speech, corresponding to Catford's (1977) 'initiation' function. Airstream mechanisms are characterised according to whether the body of air is pulmonic, glottalic, or velaric, and whether the air is moved upward/outward or downward/inward. 27  1.3.3. Other Representational Assumptions  Cases of debuccalisation, in which all Place specifications of a segment are lost, and laryngeal or nasal specifications remain (see Clements (1985), McCarthy (1988), Trigo (1991)) are relevant to the issue at hand, in a manner to be explained shortly. Forms illustrating word-final obstruent debuccalisation in Kelantan Malay, from Trigo (1991:124), are presented in (19). (The surface form status of these data is inferred from Trigo's discussion.)  After Halle (1995:16-17), it is assumed that Malay stops are  redundantly assigned [CG] and that the continuants are redundantly assigned [SG], presumably enhancing the continuancy distinctions. The examples below show retention of these laryngeal specifications.  (19) Kelantan Malay Debuccalisation Standard Malay  Kelantan  a. -{Tasap!-  *! ?asa?r  'smoke'  b. ikilaU  i\dla?[  c. -jbalasi-  H balah r  'finish'  d. <!negatii>  -Inegat+M  'negative'  'lightning'  The final Kelantan consonants in (19) have all the feature specifications of the corresponding Standard consonants, minus all Place specifications. This indicates that, for the final Kelantan consonants, the phonological representation of a filter has been removed. What is left is the representation of a source, viz., the laryngeal node. A final discussion concerns the glottal stop. Ladefoged and Maddieson (1996:38) state:  28  1.3.3. Other Representational Assumptions  The larynx, among its many other functions, can also serve as place of articulation for stops.  Glottal stops occur in many languages.  They  frequently pattern with other consonants... making it clear that glottal gestures must be taken into consideration when discussing places of articulation that are possible for stop consonants.  Based on this, they conclude [p. 11] that "the glottis has to be recognized as an articulator in some circumstances, forming Glottal articulations." While the existence of glottal stops is clear, their existence is here not considered to indicate that the larynx is an articulator but, rather, that the phonological representation of stop manner, viz., the feature [STOP], is implemented in association with features which implement either laryngeal or articulator gesture (producing laryngeal or supralaryngeal constrictions, respectively). This is encoded by the geometry in (3), in which [STOP] is immediately dominated by the root node.  1.3.3.2. The Representation of S econdary Articulation  Selkirk's (1988) representation of secondary articulation is assumed, in which a primary articulation feature is immediately dominated by the Place node. A secondary articulation feature is one that is dominated by a primary articulation feature, assuming Mester's (1986, 1988) head- vs. dependent-feature distinction. representation of the labialised labial / p / is as seen in (20). w  29  For example, the  1.3.3. Other Representational Assumptions (20) [CONS]  I o Place [LAB] [LAB]  Sagey's (1986) pointer system for the representation of secondary articulation is not adopted. The reason for this is that it cannot represent a primary and a secondary instance of the same feature on a single segment.  As discussed by Selkirk (1993), this  representational capacity is necessary for segments, like a labialised labial, that have both primary and secondary specification for one feature. Another such case is a uvularised velar (i.e., a dorsal emphatic), which will be argued in this thesis to be specified for both primary- and secondary-[DOR].  In Sagey's system, a pointer points to a primary  articulation feature; no pointer points to a secondary articulation feature. A segment with dual specification for one feature requires that a pointer both point and not point to the feature, something that is impossible in Sagey's one-level representation. (Once a twolevel representation is adopted, the primary vs. secondary distinction is captured and the additional pointer device is unnecessary.) Togo's (1991) lPlace/2Place representation of secondary articulation is not adopted because it posits two new nodes in the geometry: a 'lPlace' node, which dominates  30  1.3.3. Other Representational Assumptions  primary articulation features, and a '2Place' node, which dominates secondary articulation features. This enriches the geometry, an undesirable result. A Padgett (1991,  1994)  (see  also Walli-Sagey 1986) stricture-under-Place  representation of secondary articulation is also not adopted.  Padgett proposes this  representation on the basis of several cases of place harmony with contingent stricture harmony. Stricture-under-Place is not adopted here because, like Sagey's pointer system, it cannot represent a primary and secondary specification for a single feature on one segment: in Padgett's system, a primary articulation feature dominates stricture features; a segment with dual specification for one feature would require, impossibly, that stricture features both appear and not appear under one articulator feature. Stricture-under-Place is generally rejected on three other grounds. First, in Padgett's geometry, multiple instances of the stricture features occur, one set under each articulator feature. This presents a redundancy which has no parallel elsewhere in the geometry. That is, no other set of features displays such structural properties. Second, Strictureunder-Place does not capture all place harmony facts: cases of non-stricture-contingent place harmony exist; e.g., in Sanskrit (Steriade (1986)) and Tahltan (Shaw (1991b)). Third, cases of non-stricture-contingent place harmony involving [COR] force Padgett (1991) to propose a distinct node called 'Site' under [COR], a type of node unparalleled elsewhere in the geometry. Selkirk (1993) proposes a 'feature-node' theory of primary and secondary specification.  Feature-node theory presumes the representation of Selkirk (1988) as  31  1.3.3. Other Representational Assumptions  illustrated in (20) but, in addition, claims that each articulator feature is associated with a node. The features are understood as labelling the nodes. Selkirk (1993) proposes this system primarily to permit representation of null primary place. She cites [p. 82] the Irish consonant lenition described by Ni Chiosain (1991) as indicating a need for such a representation.  Feature-node theory is not adopted here for two reasons.  First, by  positing new nodes, one for each articulator feature, it entails an enrichment of the geometry. Second, its theoretical implications are unclear: for example, is a Place node dominating only an unlabelled node a well-formed representation? Finally, under Articulator Theory, a V-Place representation of secondary articulation, after Clements (1989, 1991), is not an option. This is because the elements in a V-Place representation are not necessarily anatomically grounded, that is, based on some anatomical characteristic of the vocal tract. Articulator Theory holds that such a basis is necessary. Thus, Articulator Theory cannot be maintained where a V-Place representation of secondary articulation is adopted. It is assumed here that primary articulation features are in the universally-fixed hierarchical arrangement of the feature geometry, but that the head-dependent relations of primary and secondary articulation features are not universally-fixed.  This is implicitly  assumed by Selkirk (1993), but differs from Mester's (1986) original proposal, that the head/dependent relation characterises all features.  Reserving this relation for the  representation of secondary articulation captures the fact that secondary articulation features are dependent on primary articulation features for realising their secondary status.  32  1.3.3. Other Representational Assumptions  The head/dependent relation means, for example, that secondary-[LAB] can occur under primary-[LAB] (as for / p / ) or primary-[DOR] (as for /k /), but that, by stipulation, the w  w  lack of a common head has no ramifications for the ability of the two instances of secondary-[LAB] to function together as a class. For complex representations, such as primary-[DOR]/[FRONT] + secondary-[LAB], it is assumed that the secondary articulation feature is immediately dominated by the primary articulation feature. This is illustrated in (21).  (21) [CONS]  0 Place 1 [DOR] [FRONT] [ L A B ]  1.3.3.3. Prosodic Representations  A final body of background assumptions concerns the representation of prosodic structure. Shaw's Nuclear/Moraic theory of prosodic representations will be assumed. This theory was first motivated on the basis of templatic facts; see Shaw (1992, 1993). It has been shown to also provide an explanatory account of syllabification, including the intricate syllabification of languages like Mon Khmer, St'at'imcets Salish, and Berber; see  33  1.3.3. Other Representational Assumptions  Shaw (1994, 1996a, 1996b). Further support for Nuclear/Moraic theory has been found in Niger-Congo prosodic minimality (Ola (1995)), Spanish segmental alternations (Valerga (1995)), and Chinese tone-vowel interaction (Jiang-King (1996)). In this theory, a segment is either moraic or non-moraic, nuclear or non-nuclear, where the mora ('p') is a unit of prosodic weight and the nucleus ('N') is the head of a syllable ('a'). A segment can be both nuclear and moraic, but nuclear status does not entail moraic status. The reverse is also true. The Nuclear/Moraic prosodic hierarchy is seen in (22). In (22), 'L' abbreviates 'light syllable' and 'H' abbreviates 'heavy syllable'.  (22) The Prosodic Flierarchy in Nuclear/Moraic Theory ((a-f) are from Shaw (1992, 1993); (g) is from Shaw (1996a, 1996b)) a. open L  b. open H  c. closed L  d. closed H  e. super H  f. non-nuclear  a  g. weightless a  0  a  (C)  34  1.4. Postvelars 1.4. Postvelars  Postvelars are of central interest in this thesis because, it will be argued, they are the segmental source of postvelar harmony in Palestinian Arabic and St'at'imcets Salish. Bessell and Czaykowska-Higgins (1991:1) define postvelars as "sounds articulated wholly or partly in the postvelar region of the vocal tract." This definition identifies gutturals and emphatics. Gutturals include the consonants /(? h ) ? f i G q B yj, where parentheses around ? and h indicate that laryngeals can be, but are not necessarily, gutturals; see §1.3.2.1. Emphatics are consonants like / 5 s r k/. Gutturals are articulated wholly in the postvelar region of the vocal tract (McCarthy (1994)). Non-guttural laryngeals, such as the laryngeals of Interior Salish, Tigre, and English, as discussed in §1.3.2.1, §1.3.2.2, are excluded from the postvelar class because (having no articulation) they lack postvelar articulation. Emphatics are postvelars because they are partly articulated in the postvelar region of the vocal tract. This section first details the articulation of gutturals and emphatics, as shown by the articulatory literature.  'Articulation' is used here in a static sense to refer to the  position of some articulator, in an overall vocal tract configuration, by which some articulatory constriction is produced. The aim is to identify just how gutturals are wholly articulated, and emphatics partly articulated, in the postvelar region of the vocal tract. The focus will be on data from Arabic since, as noted by Bessell (1992), there are no 27  27  0n the historical treatment of Arabic gutturals and emphatics, see, e.g., Mattsson  (1911), Lehn (1963), Bonnot (1977), Ghazeli (1977), and Card (1983). 35  1.4. Postvelars  articulatory (viz., x-ray,fibreoptic,EMG, etc.) data on Salish. The predicted acoustic effects of guttural and emphatic articulation will then be discussed.  The acoustic  discussion will be primarily in terms of the model of Stevens and House (1955); the models of Fant (1960) and Lindblom and Sundberg (1971) will also be discussed. The predictions of those models, as applied to postvelar articulations, will provide a means of interpreting the acoustic findings on Palestinian and St'at'imcets to be reported in the next two chapters. The diagram of the vocal tract in Figure 1:1, which is from Ladefoged (1993:4), will serve as reference for the discussion to follow.  Figure 1:1 Diagram of the vocal tract (from Ladefoged (1993:4))  36  1.4.1. The Articulation of Gutturals 1.4.1. The Articulation of Gutturals  X-ray data on gutturals from three articulatory studies will be presented below. This excludes data on laryngeals: as noted in §1.3.2.2, there are no articulatory data on laryngeal gutturals. The first set of data, seen in Figure 1:2, are x-ray tracings of tokens of the pharyngeals /S tV in Modern Standard Arabic as spoken by an Iraqi speaker,fromAl-Ani (1970). (Al-Ani provides no tracings of tokens of uvular gutturals. He uses the symbol 'e' to denote ?.) As seen, these tracings do not show the position of the epiglottis nor the very base of the tongue root; these structures are also lackingfromAl-Ani's identification template [p.27]. I have added the label 'pharyngeal constriction' to the tracings. In other figures to be presented in this section, I have added the label 'uvular constriction'. These labels will be discussed shortly.  37  1.4.1. The Articulation of Gutturals  pharyngeal constriction  K  T r a c i n g N o . 18: / h / in / ha / / e / in / ea /  Figure 1:2 X-ray tracings of [T] and [Ti] (from Al-Ani (1970:74))  Figure 1:3 shows sagittal sections based on x-ray tracings of tokens of pharyngeal and uvular gutturals in Lebanese Arabic, from Delattre (1971). (Delattre denotes ti as 'h', B as 'R '. In the figure, his shading, which indicates the area of the lower pharyngeal cavity, is retained.)  38  1.4.1. The Articulation of Gutturals  a. Pharyngeals /?/ and /TV  b. Uvulars lul and lyj  Figure  1:3  Sagittal sections based on x-ray tracings of [?],  Delattre (1971:130))  39  [n],  [B], and [%] (from  1.4.1. The Articulation of Gutturals  Figure 1:4 shows tracings of tokens of the same four consonants in Tunisian Araabic, from Ghazeli (1977). Ghazeli (1977:30) notes that his subject exhibits a lingual tonsil. The location of the tonsil is seen from Ghazeli's identification template, which is presented in Figure 1:5. (Figure 1:5 will be properly introduced below.) Because of the tonsil, the epiglottis and tongue root are not independent structures for his subject. This means that the position of the tongue root must be inferred in each of Ghazeli's tracings.  40  1.4.1. The Articulation of Gutturals  a. Pharyngeals ISI and /TV  pharyngeal constriction  If]  b. Uvulars lul and /%/  Cx3  Figure 1:4 X-ray tracings of [T], [fi], [B], and [%] (from Ghazeli (1977:40,57))  41  1.4.1. The Articulation of Gutturals  From Figures 1:2 - 1:4, the articulation of guttural consonants can be identified. To aid in this, Ghazeli's identification template, which he describes [p.29] as showing a "rest position", is presented in Figure 1:5. Compared to this rest position, the tracings in Figures 1:2 - 1:4 indicate (i) tongue root retraction for both the pharyngeal and uvular gutturals and (ii) tongue back retraction for the uvulars. The first articulation produces a constriction between the tongue root and the rear pharyngeal wall in the lower pharynx. The second articulation produces a constriction between the tongue back and the uvula. The labels 'pharyngeal constriction' and 'uvular constriction' added to the tracings in Figures 1:2 - 1:4 identify these constrictions.  Pig. 2-1.  Diagram of the template used for measurements of the cinefluorograms. Notice that the subject used for this experiment exhibits a lingual tonsil, therefore, there is little space visible between the epiglottis and the root of the tongue. The shaded area anterior to the epiglottis, and which will not be indicated in later tracings, represents a laryngologist's estimate of the location of the lingual tonsil. Figure 1:5 Ghazeli's (1977:30) identification template showing the position of his subj ect' s tongue at rest  42  1.4.1. The Articulation of Gutturals  Other structures besides the tongue root and tongue back have been shown to be involved in the production of gutturals. Ladefoged and Maddieson (1996:168-169) find evidence for involvement of the epiglottis in the Al-Ani and Ghazeli tracings, also in Arabic data from Boff-Dkhissi (1983), Bukshaisha (1985), and Laufer (cited as p.c), and Hebrew data in Laufer and Condax (1981).  Ghazeli (1977:37) reports inward  displacement of the pharyngeal wall for the pharyngeals. Lee (1995:355-356) reports a low and retracted jaw position for the pharyngeals and uvulars. It is assumed here that epiglottal, pharyngeal wall, and jaw gesture, to the extent that they occur for a given guttural token, co-operate with the tongue, but that gutturals can be defined in terms of two main articulations: tongue root retraction and tongue back retraction. For tokens of Arabic /? TV that might have only epiglottal articulation, such as those described by Ladefoged and Maddieson (1996:169), it is assumed here that the epiglottal articulation of such tokens simply shows the range of phonetic implementation of the phonological specification [TR]/[RTR]. Catford (1983) documents contrasting epiglottals and pharyngeals in Caucasian languages, e.g., epiglottal M vs. pharyngeal /TV in Agul; see also Ladefoged and Maddieson (1996) for discussion. This is evidence that, besides tongue root retraction and tongue back retraction, gutturals can also be defined in terms of epiglottal articulation. The featural basis for a phonological distinction between epiglottals and pharyngeals will not be discussed in this thesis. For discussion of phonetic epiglottalisation of vowels in St'at'imcets, see §3.2.2.  43  1.4.1. The Articulation of Gutturals  Finally, McCarthy (1994), citing several studies, notes that a raised larynx has been documented for pharyngeal and uvular gutturals, and creaky voice for /T/.  28  With  McCarthy (1994:195), it is here assumed that laryngeal involvement in the production of gutturals is a "superficial mechanical effect", where the phrase just quoted is interpreted to mean 'phonetic effect'.  This is because, to my knowledge, there are no phonological  phenomena involving gutturals that require reference to some feature representing larynx height or laryngeal gesture.  (This excludes glottalised gutturals such as / B ' B ' / in w  St'at'imcets Salish, for which the laryngeal gesture is due to their phonemic contrast with non-glottalised /u B / ; see §3.2.1.) W  The articulation of pharyngeal and uvular gutturals will now be more closely defined. To enable this, a distinction between primary and secondary articulation will first be drawn, based on the assumption in (23) and the criteria in (24).  (23) A segment has a maximum of one primary articulation (Anderson (1976), Selkirk (1993)).  On this laryngeal involvement, McCarthy (1994) cites Delattre (1971), Ladefoged (1975), Ghazeli (1977), Laufer and Condax (1979), Sasse (1979), Hayward (1989), and Hayward and Hayward (1989).  44  1.4.1. The Articulation of Gutturals (24) a. Primary Articulation: the only articulation (Selkirk 1993:7) or the articulation with the tightest constriction (Selkirk 1993) at at which the stricture features of the segment are realised (Sagey 1986). b. Secondary Articulation: the articulation made by two organs of speech that are not involved in the primary articulation (Ladefoged (1993:296)).  A procedure for determining the 'tightest constriction' will now be outlined. 'Tightest constriction', for a given vocal tract configuration, is defined here as the point along the vocal tract with the smallest cross-sectional area.  However, no radius of  constriction from which the cross-sectional area at a specific point can be determined, is apparent from an x-ray tracing because an x-ray tracing shows only the midsagittal plane. From the two dimensions of an x-ray tracing, the tightest constriction can only be guessed. Additional information regarding the cross-sectional area function of the vocal tract is required for an exact determination.  In the absence of such information, it can be  "guesstimated" from a tracing as the point along the vocal tract at which the upper and lower vocal tract surfaces connect, otherwise as the point at which they are closest together. (For the upper surface, the connection will be at some point along the inferior edge of the upper lip, upper incisors, alveolar ridge, soft palate, velum, uvula, or along the pharyngeal wall. For the lower surface, the connection will be at some point along the superior edge of the lower lip, lower incisors, tongue tip, tongue blade, tongue dorsum,  45  1.4.1. The Articulation of Gutturals  tongue back, tongue root, or epiglottis.)  The points just described can usually be  identified by eyeballing the tracing. The assumption that the primary articulation is the one with the tightest constriction results in problematic cases like Ngbaka /kp/. Treatment of such cases will not be undertaken here; see Selkirk (1993:7) and Halle (1995:8) for further discussion. Based on (23) and (24) and on the procedure for identifying the tightest constriction shown by an x-ray tracing, described above, the data in Figures 1:2 - 1:4 show that [? "h] have primary pharyngeal articulation. Their pharyngeal articulation is primary because it is their only articulation. Because the tightest constriction for [B X] is at the uvula, and their frication is achieved at the uvular constriction, [B X] have primary uvular articulation. They also have secondary pharyngeal articulation, which will here be referred to as 'pharyngealisation'. (The suffix '-isation' encodes the secondary nature of the articulation.) Primary uvular articulation and pharyngealisation are ascribed to [B X] also by Trigo (1991:122). Since neither pharyngeal(-isation) or uvular articulation produces a constriction between the velum and the lips, gutturals are wholly articulated in the postvelar region of the vocal tract. As a final point, the articulatory data examined in this section indicate that uvular articulation does not occur without pharyngealisation. (See also Bothorel et al. (1986) for x-ray tracings showing uvular articulation with pharyngealisation for French /R/.) The pharyngealisation of uvular segments is considered here to be an automatic consequence  46  1.4.2. The Articulation of Emphatics  of the primary uvular articulation.  In the terminology of Fujimura (1990), it is a  'resultant' gesture of the uvular 'control' gesture. This is assumed to follow from the anatomical proximity of the tongue back and root.  1.4.2. The Articulation of Emphatics  In this section, x-ray tracings of emphatic consonants from three sources will be presented. A first tracing, from Al-Ani (1970), is seen in Figure 1:6. Thisfigureshows a token of emphatic l\l and a token of non-emphatic 1X1. (Al-Ani denotes emphatic t as 't\ He notes [p.21] that his emphatic data were provided by both Iraqi and Jordanian consultants.)  47  1.4.2. The Articulation of Emphatics  Tracing N o . 11: / t / in / ti /  l\ I in lul  Figure 1:6 X-ray tracings of [t] and [t] (from Al-Ani 1970:57)  Figure 1:7 shows a tracing of a token of emphatic IM in Classical Arabic as spoken by a Saudi speaker, from Bonnot (1977). (Bonnot denotes emphatic t as't'.) The figure also shows a tracing of a token of non-emphatic IM.  48  1.4.2. The Articulation of Emphatics  Ct] dans [ a t a ]  (50/8)  rt]  (56/12)  dans [ a t a "  Figure 1:7 X-ray tracings of [t] and [t] (from Bonnot (1977:85))  Further tracings, from Ghazeli (1977), are presented in Figure 1:8: a token of emphatic lil is seen alongside a token of non-emphatic IXI in (a), a token of emphatic Isl alongside a token of non-emphatic Isl in (b). (Ghazeli denotes emphatic t and s as't' and 's', respectively.)  49  1.4.2. The Articulation of Emphatics  a. Emphatic IXI  b. Emphatic I si  Figure 1:8 X-ray tracings of [t], [t], [s] and [s] (from Ghazeli (1977:69,70))  50  1.4.2. The Articulation of Emphatics  The tracings in Figures 1:6- 1:8 show that emphatics are produced with (i) a nonpostvelar articulation, (ii) tongue back retraction, and (iii) tongue root retraction. Articulation (i) produces a constriction between the velum and the lips; (ii) produces a constriction between the tongue back and the uvula; (iii) produces a constriction between the tongue root and the rear pharyngeal wall in the lower pharynx. I have added labels to the tracings in Figures 1:6 - 1:8 to identify the constrictions resulting from articulations (ii) and (iii). Emphatic Ikl, however, is produced with only articulations (ii) and (iii). The exceptional articulation of Ikl will be discussed shortly. Articulation (i) is uncontroversial. The non-postvelar articulation identifies the emphatic as the emphatic counterpart of some non-emphatic segment (e.g., IXJ, emphatic counterpart of IXI, etc.).  As seen from segments like /5 s r kl, the non-postvelar  articulation produces a constriction which varies both in place and manner. Emphatics differ from their non-emphatic counterparts by having the postvelar articulations (ii) and (iii). Both these articulations have been identified in previous studies. Articulation (ii) is described by Herzallah (1990:2), who states that emphatic articulation involves "the back of the tongue body". Similarily, Ghazeli (1977:72) refers to "rearward movement of the back of the tongue".  Younes (1982:216) describes emphasis as "a  secondary articulation involving the back of the tongue".  In earlier studies, this  articulation was labelled 'velarisation' (see, e.g., Obrecht 1968) or described as similar to velarisation (see Lehn 1963:20). But it is identified more precisely as uvularisation by Dolgopolsky (1977:1), who states: "In Arabic, the 'emphatics' are pronounced as  51  1.4.2. The Articulation of Emphatics  uvularized consonants.  Uvularization is the modification of consonants or vowels by  moving back the rear part of the tongue towards the uvula and the back wall of the pharynx." In a footnote on the same page, he continues: "I prefer to call it 'uvularization' and not 'velarization' (which is obviously wrong, because even the velar k, when emphasized, becomes uvular)." Articulation (ii) is identified as uvularisation also by Czaykowska-Higgins (1987:2), who uses 'emphatic' and 'uvularised' synonymously, and by McCarthy (1994:219), who states: "The so-called pharyngealized consonants of Arabic should really be called uvularized." Previous studies have also identified articulation (iii). E.g., Obrecht (1968:20) refers to emphatics as "velarized or pharyngealized"; Al-Ani (1970) refers to them as 'pharyngealised'. Lehn (1963:31) lists "faucal and pharyngeal constriction (pharyngealization)" as an articulatory feature of emphatics, distinct from their 'velarisation'. Neither Obrecht nor Lehn explicitly refer to the tongue root.  However, the tongue root is  explicitly identified as an articulator for emphatics by Ali and Daniloff (1972:98), who state, "the tongue dorsum and/or tongue root... is the primary articulator for emphatic sound production", and by Woldu (1981:117), who descries emphatic production as involving "retracting of the tongue root towards the pharyngeal wall". This tongue root retraction is supported by the EMG findings of Kuriyagawa et al. (1986), who report no posterior genioglossus (GGP) activity for Jordanian Arabic Isl and IV, but GGP activity for non-emphatic Isl and /XI. They interpret this [p.25] as showing that "activity of the  52  1.4.2. The Articulation of Emphatics  GGP is supressed when the retraction of the tongue occurs and the root of tongue [sic] is brought back toward the wall of the pharynx." Emphatic /k/ differs from all other emphatic in that it is produced with only tongue back retraction and tongue root retraction. It lacks non-postvelar articulation. A sagittal section showing this, from Delattre (1971), is seen in Figure 1:9. (Delattre denotes this segment as 'q'.)  F i g u r e 1:9  Sagittal section based on an x-ray tracing of [k] (from Delattre 1971:130)  53  1.4.2. The Articulation of Emphatics  There are two points of view on this uvular stop. Thefirstconsiders it to be an emphatic Ikl, the emphatic counterpart of non-emphatic velar /k/. it to be simply a uvular stop, that is, not an emphatic /k/.  30  29  The second considers  In studies with the first  opinion, the segment under discussion is usually usually denoted as ' q ' (It is denoted as ' q ' in studies with the second opinion.) In the present study, following Harrell (1957), it is denoted as 'k' in order to be explicit that, for Palestinian Arabic, it is analysed as an emphatic, the emphatic counterpart of non-emphatic Ikl. The primary argument for considering the uvular stop to be simply a uvular stop is that it has no non-postvelar articulation; see discussion in Alioua (1993). However, after Dolgopolsky (1977:1) it is assumed here that, for Palestinian Arabic, the segment lacks non-postvelar articulation because its velar and uvular articulations are phonetically realised together to produce a single uvular articulation. That is, its articulations (i) and (ii) are phonetically fused. The main argument for recognising the uvular stop as emphatic Ikl in Palestinian Arabic is its phonological behaviour. In Palestinian, this segment functions phonologically as an emphatic: it patterns with the other emphatics in triggering uvularisation harmony, a.k.a 'emphasis harmony'. This will be shown in chapter 2. However, crosslinguistic evidence indicates that the uvular stop is not an emphatic in all Arabic dialects: Ghazeli  See, e.g.,  29  Harrell (1957), Lehn (1963), Trubetzkoy (1969), Delattre (1971),  Dolgopolsky (1977), and Jakobson (1978). 30  For the second point of view, see, e.g., Bonnot (1977), Ghazeli (1977), Giannini and  Pettorino (1982), and Alioua (1993).  54  1.4.2. The Articulation of Emphatics  (1977:59-65) presents data which support recognising the uvular stop as non-emphatic in Tunisian Arabic. Previous studies have also mentioned involvement of the lips in the production of non-velar emphatics, for either protrusion or rounding; see Mitchell (1960), Lehn (1963), and Uldall (1992). This labialisation receives partial support from the EMG study of Kurijagawa et al. (1986:27), who report orbicular oris activity for activity for /5/, but not for Isl. Lee (1995:356) reports a retracted and lowered jaw position for the uvular stop. A higher degree of muscular tension in the mouth and throat has been mentioned for emphatics (see Lehn (1963), Bonnot (1977)), although Woldu (1981:116) observes that there is no EMG data to support this. Finally, Ghazeli (1977:73) notes a raised larynx for Isl but not for the other emphatics. He ascribes the raised larynx of Isl to its oral cavity articulation because he found a raised larynx also for non-emphatic Isl, a finding consistent with those of Perkell (1969). After Ali and Daniloff (1972), emphatic articulation is here ascribed primarily to the tongue. The foregoing discussion indicates that emphatics can be defined in terms of two main articulations (besides their non-postvelar articulation): tongue back retraction and tongue root retraction. The additional articulatory properties discussed above, to the extent that they might characterise a given emphatic token, are assumed only to cooperated with the tongue in the production of emphatics. The articulation of emphatics is more closely defined as follows: based on (23) and (24), the data in Figures 1:6 - 1:8 show that the non-postvelar articulation of emphatics is  55  1.4.3. Postvelar Acoustics  their primary articulation. This is because it produces the tightest constriction and is the constriction at which the segment's stricture features are realised. pharyngeal articulations are both secondary. pharyngealised.  The uvular and  Emphatics, thus, are uvularised and  Although emphatics have non-postvelar articulation, they qualify as  postvelars because they are partly articulated in the postvelar region of the vocal tract. The exception to this is /k/. The data in Figure 1:9 indicate that /k/ has no nonpostvelar  articulation but  pharyngealisation.  is  produced with  primary uvular articulation and  This was explained above as due to the fact that its primary velar and  uvularisation articulations are phonetically fused, yielding a phonetically primary uvular articulation.  1.4.3. Postvelar Acoustics  The aim of this section is to identify the expected acoustic effects of primary pharyngeal and uvular articulation, and pharyngealisation and uvularisation, as identified by the articulation-to-acoustics model of Stevens and House (1955). Similar models by Fant (1960) and by Lindblom and Sundberg (1971) will also be discussed. All three models view the vocal tract as a sequence of cylindrical tubes of unit length, excited by an acoustic source at one end and open at the other. The variable crosssectional area of the tract is small enough for assuming plane wave propagation. Stevens and House proposed that the resonances of such a model can be predicted if the values of  56  1.4.3. Postvelar Acoustics  three parameters are known. The parameters are (i) the radius ro of the rube at the point of the narrowest constriction, (ii) the distance d of the narrowest constriction from the glottis, and (iii) the ratio All of the cross-sectional area of the front tube of the vocal tract divided by its length.  The value for r corresponds to the cross-sectional area at the 0  constriction in the manner schematised in Figure 1:10. Thefronttube of the vocal tract is that portion infrontof the teeth, for which the cross-sectional area is controlled primarily by the lips and the mandible.  area at constriction shown within a circle; radius = r  area at constriction  0  Figure 1:10 Conversion of area at constriction to area within a circle  57  1.4.3. Postvelar Acoustics  Stevens and House varied the values of the three parameters between 0.3 and 1.2 cm for r , between 4 and 13 cm for d, and between 0.1 and 20 cm for All,  31  0  thus  corresponding to 306 distinct configurations. Their schematisation [p.486] of the vocal tract configuration for four sets of values of r , d, and All is presented in Figure 1:11. I 0  have added on the right transcriptions of the sounds, all of them vowels, which the configurations in thefigurewould produce.  DISTANCE  F R O M  GLOTTIS  IN C M  Figure 1:11 Schematisation of vocal tract configurations for four sets of values of ro, d, and All (from Stevens and House 1955:486)  31  See Stevens and House (1955) for justification of these values and further explication of  the model and its calculations. 58  1.4.3. Postvelar Acoustics  Stevens and House implemented their model using an electrical analog, i.e., each tube element was represented by its analog electrical circuit. Each one of the 306 vocal tract configurations yielded a spectrum from which three formant frequencies were obtained and plotted in several ways, as a function of the three articulatory parameters, r , 0  d, All. They found the output signal obtained from their model to match the expected output, both acoustically and perceptually. The model of Fant (1960) predicts the resonances of the vocal tractfromvalues of parameters similar to those of Stevens and House. By result, the two models are very similar. Lindblom and Sundberg's (1971) model predicts the resonances of the vocal tract fromfiveparameters: state of the lip muscles, position of the jaw, shape of the tongue body, position of the tongue body, and larynx height.  Since a change in tongue body  shape and/or in jaw position results in a change in degree of constriction and a change in degree of constriction results in a change in r , their model ultimately considers the effect 0  of a change in r . The acoustic predictions of the Fant (1960) and Lindblom and Sundberg 0  (1971) models are in general agreement with the predictions to be identified in this section based on the model of Stevens and House (1955). It is important to note that the Stevens and House (1955) model, and the other two models, consider only vocal tract configurations with a single constriction. They do not consider configurations with a second constriction. Additional constrictions are of interest here because some postvelars are presumed to be produced with simultaneous primary and secondary articulations, resulting in two constrictions in the vocal tract; see  59  1.4.3. Postvelar Acoustics  Figures 1:3 - 1:4 and Figures 1:6-1:8. A means of circumventing this limitation will be suggested below.  But first, the Stevens and House model will be applied to primary  postvelar articulations. The values in Table 1:1 will be assumed for r , d, and All for vocal tract 0  configurations with primary pharyngeal and uvular articulations.  Table 1:1  Values of r , d, and All for vocal tract configurations with primary pharyngeal and uvular articulations G  Articulation  r (cm)  d (cm)  All (cm)  primary pharyngeal  0.4  4  7  primary uvular  0.3  7  7  0  The r values in Table 1:1 are estimates based on the x-ray tracings of gutturals in 0  Figures 1:3 - 1:4.  Those tracings indicate that r is smaller for a primary uvular 0  constriction than for a primary pharyngeal constriction. It is assumed that a range of 3 - 7 cm defines d for postvelar articulations. This range is assumed by Klatt and Stevens (1969) for the constriction sites of pharyngeal and uvular gutturals. A 3 - 7 cm range is also indicated by Ghazeli's (1977) x-ray template, presented earlier in Figure 1:5 and again in Figure 1:12. I have added cm measurements to the template, assuming a vocal tract length of 17 cm. Based on Figure 1:12, d = 4 for a pharyngeal constriction and d = 7 for a uvular constriction.  60  1.4.3. Postvelar Acoustics  Figure 1:12 Ghazeli's (1977) identification template with cm measurements  The All values in Table 1:1 are estimates based on the articulatory findings pertaining to the cases of retracted and lowered jaw position for gutturals, as discussed in §1.4.1. The resonances predicted by the Stevens and House model for the vocal tract configurations defined by the values in Table 1:1 will now be discussed. The predicted resonances are given in Fig. 3 of their paper. Their Fig. 3 presents six graphs in which formant frequency is plotted vs. d for different Alt; r is held constant. These graphs can 0  be thought of as a stack of graphs, each horizontal layer corresponding to the graph of a given r value. Figures 1:13 - 1:17 are derivedfromthat Fig. 3 {ibid.) and correspond to 0  vertical slices in the stack at specific values of d, the abscissa in the six graphs of the stack.  61  1.4.3. Postvelar Acoustics  In the figures presented here, the ordinate represents formant frequency (in Hz) and the abscissa the radius r (in cm) at the constriction point. 0  frequency  Figure 1:13 Fi and F predicted by Stevens and House (1955) for vocal tracts of d = 4, All-6.1, and varying r (points for primary pharyngeal articulation plotted) 2  0  Figure 1:13 shows formant frequency vs. r for primary pharyngeal articulation, 0  i.e., d = 4, All = 7.  j2  Figure 1:14 shows formant frequency vs. r for primary uvular 0  articulation, i.e., d = 7, All = 7. Each offigures1:13 - 1:14 shows two curved lines, the bottom one for Fi, the top one for F .  JJ  2  3  Along dashed vertical lines in Figure 1:13, point  Fig. 3 of Stevens and House (1955) does not provide data for configurations of All = 7  cm. Figures 1:13 and 1:4 show the data for their All value closest to 7 cm. j3  The curves in Figures 1:13 - 1:14 (and Figures 1:15 - 1:17) are fitted. As seen, each  curvefitsclosely with the points along it. 62  1.4.3. Postvelar Acoustics  frequency (Hz) BB  1750  F?  1500 1250 1000 750  Fi  B-f  500 250  0 .2  0.4  0 .6  0 .8  1  1.2  ro(cm)  Figure 1:14 Fi and F predicted by Stevens and House (1955) for vocal tracts of d = 7, PJt = 6.7, and varying r (points for primary uvular articulation plotted) 2  0  A marks Fj and AA marks F for primary pharyngeal articulation. In Figure 1:14, point B 2  marks Fi and BB marks F for primary uvular articulation. 2  Given the degree of approximation reflected by the values in Table 1:1, precise acoustic predictions will be avoided here. Rather, the general Fi and F effects predicted 2  for the two primary articulations under discussion will be identified. summarised in Table 1:2.  63  The effects are  1.4.3. Postvelar Acoustics Table  1:2  Predicted acoustic effects of primary pharyngeal and uvular articulations Articulation  F  Fi  2  primary pharyngeal  A  high  AA  low  primary uvular  B  medium  BB  medium  The descriptions of Fi and F as 'high' and 'medium' above are based on 2  comparison of the frequency at each point A(A) and B(B) in Figures 1:13 - 1:14 with the Fi and F ranges predicted by the Stevens and House model, over all r , d, and All. Those 2  0  ranges, delimited by their maxima and minima, are given in Table 1:3.  Table  1:3  Fi and F ranges predicted over all r , d, and All by the model of Stevens and House (1955) 2  0  Fx (Hz)  F (Hz) 2  min  max  min  max  200  950  700  2500  Peterson and Barney (1952) studied the formant frequencies of English vowels produced by 76 speakers of which 33 were adult males, 28 adult females, and 15 children of unspecified age. The ranges in Table 1:3 closely match the ranges of Fi and F found by 2  Peterson and Barney for their 33 male speakers, as shown in Table 1:4.  64  1.4.3. Postvelar Acoustics Table  1:4  Fx and F ranges of Peterson and Barney data 2  F (Hz)  Fi(Hz)  2  min  max  min  max  ranges of Peterson and Barney (1952) average values  270  730  840  2290  ranges of Peterson and Barney individual values  190  840  650  2700  34  Based on Table 1:3, Fi ranges between 200 and 950 Hz. F ranges between 700 2  and 2500 Hz. Dividing each of these ranges by 3 gives the sub-ranges seen in Table 1:5. The acoustic effects identified in Table 1:2 refer to the ranges in Table 1:5.  Table  1:5  Low, medium, and high ranges for Fi and F  2  Fi (Hz)  F (Hz)  low  200 - 450  700 - 1300  medium  450 - 700  1300 - 1900  high  700 - 950  1900 - 2500  2  Finally, the articulatory data discussed in §1.4.2 indicate that velar emphatics, like uvular gutturals, are produced with primary uvular articulation.  The acoustic effects  identified for primary uvular articulation in Table 1:2 thus apply to both uvular gutturals and velar emphatics.  The individual value ranges in Table 1:4 were obtained from Peterson and Barney's raw data, which are not available in their paper, but are available on the Internet. 65  1.4.3. Postvelar Acoustics  The expected acoustic effects of secondary postvelar articulations will now be addressed.  The x-ray tracings in Figures 1:3 - 1:4 and Figures 1:6 - 1:8 indicate that  uvular gutturals and emphatics are produced with the secondary postvelar articulations listed in Table 1:6.  Table 1:6 Secondary postvelar articulations of uvular gutturals and emphatics Segment uvular gutturals  Secondary Postvelar Articulation(s) pharyngealisation  velar emphatics  pharyngealisation  non-velar emphatics  uvularisation, pharyngealisation  The values in Table 1:7 will be assumed for r , d, and All for vocal tract 0  configurations with secondary pharyngeal and uvular articulations. As seen, two vocal tracts are defined for pharyngealisation. The first has a large AIL, i.e., no rounding, and is appropriate for uvular gutturals and velar emphatics. The second one has a small All, i.e., rounding, and is appropriate for non-velar emphatics. This will be explained shortly. The two vocal tract configurations for pharyngealisation are distinguished because the Fi and F; effects predicted for pharyngealisation potentially vary depending on the value of All.  66  1.4.3. Postvelar Acoustics Table 1:7 Values of r , d, and All for vocal tract configurations with pharyngealisation and uvularisation articulations 0  r (cm)  d (cm)  All (cm)  pharyngealisation (appropriate for uvular gutturals and velar emphatics)  0.6  4  7  pharyngealisation (appropriate for non-velar emphatics)  0.6  4  2  uvularisation  0.6  7  2  Articulation  G  The r values in Table 1:7 are estimates based on the x-ray tracings in Figures 1:30  1:4 and Figures 1:6 - 1:9, which indicate: r is larger for secondary pharyngeal and uvular 0  articulations than it is for the corresponding primary articulations; the secondary vs. primary difference is greater for uvular constrictions than it is for pharyngeal constrictions. The values for d for pharyngeal and uvular articulation, and the large All for uvular gutturals and velar emphatics, were explained earlier.  The smaller All for non-velar  emphatics reflects their lip protrusion and/or rounding. A dynamic view of pharyngealisation and uvularisation is assumed, that is, these secondary articulations result from a decrease in r from1.2 (reflecting no constriction) to 0  0.6. This corresponds to the change in r at d = 4 (for pharyngealisation) and d = 7 (for 0  uvularisation) for All = l (for uvular gutturals and velar emphatics) and All = 2 (for nonvelar emphatics), which occurs when a secondary pharyngeal or uvular gesture is added to a segment. In this manner, the Stevens and House model, which considers only vocal tract configurations with a single constriction, is here applied heuristically to  67  1.4.3. Postvelar Acoustics  configurations with a second constriction. The relevant data derived from Stevens and House (1955:487 Fig.3) are presented in Figures 1:15 - 1:17.  35  frequency  Figure 1:15 Fi and F predicted by Stevens and House (1955) for vocal tracts of d = 4, M = 6.7, and varying r (points for pharyngealisation plotted) 2  0  Figures 1:16- 1:17 show the data from Stevens and House (1955:487) for their All value closest to 2 cm.  68  1.4.3. Postvelar Acoustics  frequency  Figure 1:16 Fi and F predicted by Stevens and House (1955) for vocal tracts of d = 4, A/l = 1.1, and varying r<> (points for pharyngealisation plotted) 2  frequency (Hz)  1000  aoo• 600 •  0.2  0.4  Figure 1:17 Fi and F predicted by Stevens and House (1955) for vocal tracts of d = 7, AJt- 1.1, and varying r (points for uvularisation plotted) 2  0  69  1.4.3. Postvelar Acoustics  Under the dynamic view just explained, the Fi and F effects predicted for 2  pharyngealisation and uvularisation are identified as the formant changes seen in Table 1:8. The direction and size of each change were determined from the nature of the curved lines in each graph for the r interval starting at 1.2 and ending at 0.6. For example, the bottom 0  curved line in Figure 1:17, which represents Fi for uvularisation, rises 70 Hz over the r  0  interval starting at 1.2 and ending at 0.6 (that is, ending at point E). This is recorded in Table 1:8 as a small rise. The procedure by which each change was identified as small, medium, or large will be explained next.  Table 1:8 Predicted acoustic effects of pharyngealisation and uvularisation Articulation  F_2  Fj  pharyngealisation of uvular gutturals and velar emphatics  large rise  pharyngealisation of non-velar emphatics  medium rise  DD medium drop  small rise  EE medium drop  uvularisation  E  C  C  medium drop  The size of each formant change was identified by first determining the difference between the formant frequency at ro = 1.2 and r = 0.6 in each of Figures 1:15 - 1:17. 0  (For example, in Figure 1:17, the difference between Fi at r = 1.2 cm and FT at r = 0.6 0  0  cm is 70 Hz.) Each difference was then compared with the maximum difference predicted for the formant by the Stevens and House model, between ro = 1.2 and some other r  70  0  1.4.3. Postvelar Acoustics  value, over all d and All  Those maximum differences are presented in Table 1:9.  They  represent the maximum changes predicted for Fi and F . 2  Table  1:9  Maximum Fi and F changes predicted by the model of Stevens and House (1955) 2  Fi (Hz)  F (Hz)  300  600  2  Dividing each of these maximums by 3 gives the three ranges for Fi and F change seen in 2  Table 1:10. The degrees of change identified in Table 1:8 refer to the ranges in Table 1:10.  Table  1:10  Ranges of degrees of change for Fi and F  2  Fi (Hz)  F (Hz)  < 100  <200  medium  100 - 200  200 - 400  large  200 - 300  400 - 600  small  2  Based on the effects in Tables 1:2 and 1:8, the predicted cumulative effects of the postvelar articulations of gutturals and emphatics are identified as those in Table 1:11.  Based on Fig. 5 of Stevens and House (1955:487), the maximum Fi difference predicted by their model is between r = 1.2 and r = 0.4 at d = 4 and All - 20. The maximum F 0  0  difference is between r = 1.2 and r = 0.3 for d = 6 and'All= 0.11. 0  0  71  2  1.4.3. Postvelar Acoustics Table  1:11  Predicted cumulative acoustic effects of the postvelar articulations of gutturals and emphatics Postvelar  Prediction  Articulation  GUTTURALS Pharyngeal Gutturals: primary pharyngeal Uvular Gutturals:  EMPHATICS Velar Emphatics:  F,  F  high  low  2  primary uvular, pharyngealisation  medium or high  low or medium  primary uvular, pharyngealisation  medium or high  low or medium  medium or large rise  large drop  Non-velar Emphatics: uvularisation, pharyngealisation  The cumulative effects were identified as follows: as pharyngeal gutturals have no secondary postvelar articulation, the A(A) effects in Table 1:2 transfer unmodified to the first line of Table 1:11. For uvular gutturals and velar emphatics, the B(B) effects in Table 1:2 and the C(C) effects in Table 1:8 combine: the medium Fi for B + the large rise for C results in afinalprediction in Table 1:11 of either a medium or high Fi. The medium F  2  for BB + the medium drop for CC results in afinalprediction in Table 1:11 of either a low or a medium F . (Whether thefinalFi is medium or high, and the final F low or medium, 2  2  will depend on whether Fi and F for the primary articulation are at the low or high end of 2  the Fi and F medium ranges in Table 1:5.) 2  For non-velar emphatics, the D(D) and E(E) effects in Table 1:8 combine in Table 1:11: for Fi, the medium rise for D + the small rise for E results in afinalprediction of a  72  1.4.3. Postvelar Acoustics  medium or large rise. For F , the medium drop for DD + the medium drop for EE yield a 2  final prediction of a large drop. NofinalFi or F range (low, high, or medium as defined 2  in Table 1:5) is predicted here for non-velar emphatics because theirfinalFi and F ranges 2  will depend on the Fi and F ranges predicted for their primary (non-postvelar) 2  articulation, which were not discussed in this section. In sum, the predictions are: 1) pharyngeal gutturals will have a high Fi and a low F ; 2) uvular gutturals will have a medium or high Fi and a low or medium F ; 3) velar 2  2  emphatics will have a medium or high Fi and a low or medium F ; 4) non-velar emphatics 2  will a medium or large rise in Fi and large drop in F . 2  The effects just summarised generally agree with thefindingsof previous acoustic studies of gutturals and emphatics. For example, a high Fi and a medium or low F for 2  pharyngeal gutturals is reported by Butcher and Ahmad (1987), cited by McCarthy (1994). A medium Fi and low F for uvular gutturals is reported by Ghazeli (1977). 2  ('High', 'medium', and 'low' are used here as defined in Table 1:5.) A rise in Fi for nonvelar, emphatics (or for vowels in contact with non-velar emphatics) is reported by Bonnot (1977, 1979), Woldu (1981), and Younes (1982). A drop in F for non-velar emphatics is 2  a standard finding.  37  In this thesis, tokens of gutturals and emphatics in Palestinian and St'at'imcets will be examined to see if their formant frequencies show the expected Fi and F effects 2  "See, e.g., Obrecht (1968), Ghazeli (1977), Woldu (1981), Younes (1982), and Card (1983). 73  1.4.3. Postvelar Acoustics  identified in this section. Should they match, they will be taken as acoustic support for assuming that the analysed tokens were produced with the corresponding postvelar articulation(s) in Table 1:11. Tokens of consonants and vowels which undergo phonological postvelar harmony will also be examined. Where the harmony is presumably implemented with secondary pharyngeal articulation, the tokens will be examined to see if their formant frequencies show the general effects for pharyngealisation, viz., C(C) and D(D) in Table 1:8: a medium or large rise in Fi and a medium drop in F . Where the harmony is presumably 2  implemented with both secondary uvular and secondary pharyngeal articulation, the tokens will be examined for the effects predicted for both those articulations, viz., the effects for non-velar emphatics in Table 1:11: a medium or large rise in Fi and a large drop in F . If 2  the formant frequencies of the tokens of harmonising segments are found to match these predictions, they will be taken as acoustic support for assuming that the tokens were produced with the corresponding secondary pharyngeal and uvular articulations. As afinalnote, thefindingswill be interpreted as support under a hypothesis that the tokens were not produced with, instead of the postvelar articulation(s), some completely non-postvelar articulation which might result in the Fi and F effects identified 2  in this section, e.g., a lowered jaw and concomitant wide mouth opening, which has been shown to raise Fi, or lip rounding, which has been shown to lower F . This hypothesis is 2  stated here because it is important to note that, while the articulation-to-acoustics mapping is one-to-one, the reverse mapping is one-to-many. That is, a specific vocal tract  74  1.4.4. Summary  configuration will always have the same and unique set of formant frequencies, but a given set of formant frequencies will not in general correspond to a unique vocal tract configuration. This was shown by Stevens and House (1955), and later and in more detail by Atal et al. (1978). See Figures 13 - 15 and 21 - 25 in Atal et al. (1978:1548-1554), which show multiple area functions (configurations) with identical values for thefirstthree formant frequencies for IM, lal, and lul. While it is true that many possible articulations can be excluded for a given segment token, the point here is that particular acoustic effects cannot be considered actual evidence for a particular articulation.  1.4.4. Summary  This section has examined the articulatory nature of guttural and emphatic postvelars.  The pharyngeal gutturals /T TV are produced with only primary pharyngeal  articulation. Tokens of /? fV are predicted to have a high Fi and a low F . Uvular 2  gutturals such as lu xA and the emphatic velar Ikl, are produced with primary uvular articulation and pharyngealisation.  Tokens of uvular gutturals and emphatic velars are  predicted to have medium or a high Fi and a low or medium F . Non-velar emphatics, 2  such as IX s r/, are produced with a primary non-postvelar articulation and uvularisation and pharyngealisation. Tokens of non-velar emphatics are predicted to have a medium or large rise in Fi and large drop in F . 2  75  1.5. A Harmony Typology 1.5. A Harmony Typology  The harmony typology in (25) will be assumed. In (25), V denotes the feature for which harmony is observed. After Shaw (1991b), 'harmony' is used here to refer to both assimilatory and dissimilatory phenomena.  (25) Harmony Typology a. Primary Articulation (AP) Harmony: only primary instances of x are involved. b. Secondary Articulation (AS) Harmony: only secondary instances of x are involved. c. Articulation (A) Harmony: both primary and secondary instances of x are involved.  This typology is based on Selkirk's (1993:6) identification of three natural classes relevant to multiply-articulated segments: (i) the class of segments for which feature* is a primary specification; (ii) the class of segments for which feature* is a secondary specification; (iii) the class of segments for which feature* is either a primary or a secondary specification. See Selkirk (1993:5-20) for discussion of these classes in Ngbaka and Berber. This dissertation will argue that Palestinian pharyngealisation harmony is A harmony, involving Selkirk's natural class (iii). St'at'imcets pharyngealisation harmony and uvularisation harmony in both Palestinian and St'at'imcets will be argued to be AS harmony, involving Selkirk's natural class (ii). The data to be examined happen not to exemplify AP harmony, involving Selkirk's natural class (i). See Selkirk (1993:8-29) for an example of AP Dissimilation in Tashlhyt Berber.  76  1.6. Optimality Theory 1.6. Optimality Theory  In the next two chapters, a theoretical account of pharyngealisation and uvularisation harmonies in Palestinian and St'at'imcets will be developed within the framework of Optimality Theory (OT) (McCarthy and Prince 1993a, Prince and Smolensky 1993). In OT, phonological alternations are standardly assumed not to result from serially ordered rules, as assumed in generative phonology since Chomsky and Halle (1968). Rather, they are the surface effect of the interaction of ranked constraints. The ranked constraints are violable. They are universal, but their ranking is language-specific. A function GEN is assumed to generate candidate surface ('output') forms from a single underlying ('input') form. A function EVAL then evaluates the candidate outputs with respect to the constraint ranking of the language. Candidates are evaluated simultaneously and in parallel, in a one-step mapping from input to output. The candidate with the least serious constraint violations is the 'winner' and emerges as the actual output form.  38  This parallel grammar is illustrated by the 'tableau' in (26), from McCarthy and Prince (1993a: 6).  (26) A  B  *  1®° candidate 1 candidate 2  1  *!  [lllilllllllllllllllllllllllliiroillj  Note that a theory of ranked and violable constraints need not be implemented in parallel. See Prince and Smolensky (1993) for discussion of a non-parallel version of OT.  77  1.6. Optimality Theory  The tableau in (26) shows two constraints, A and B, ranked A » B, where ' » ' denotes 'is more highly ranked than'. In the tableau, domination order of the constraints is reflected by the left-to-right column order. There are two candidate outputs: candidate 1, which violates Constraint B; candidate 2, which violates Constraint A. Constraint violation is marked by '*'. The higher the ranking of a constraint, the more serious is a violation of it. Thus, given the ranking A » B, candidate 1, marked by '«*'', is the winner and actual output form because its violation of Constraint B is less serious than the violation of Constraint A by candidate 2.  Candidate 2's fatal violation of the higher-  ranked Constraint A is marked by '!'. Satisfaction of a constraint is indicated by a blank cell. Shading means the constraint does not choose between viable candidates. In OT, cross-linguistic variation is the result of constraint reranking. This is illustrated by the tableau in (27), which should be compared with the tableau in (26).  (27) B candidate 1  A  1 1  *!  *  candidate 2  1  candidate 3  In (27), the constraints are the same as those in (26), but the ranking of the constraints is reversed. As seen, candidate 2 is the winner under this reranking. This is because it violates only Constraint A which is more lowly ranked than Constraint B in this new grammar. In (27), an additional candidate 3 shows that constraint violations can accrue:  78  1.6.1. Correspondence Constraints  candidate 3 violates Constraint A twice while candidate 2 violates it only once. The more a single constraint is violated, the more serious is the violation. Thus, candidate 2 emerges the winner because it incurs fewer violations of Constraint A than does candidate 3. This thesis will argue that three types of constraints figure crucially in Palestinian and  St'at'imcets postvelar harmony: Correspondence, Alignment, and Grounded  constraints. These constraint types are explained below.  1.6.1. Correspondence Constraints  A fundamental issue in OT is the faithfulness between related representations. In its original formulation (McCarthy and Prince 1993a, Prince and Smolensky 1993), OT approached faithfulness between input and output forms in terms of a class of Faithfulness constraints which consisted of PARSE and FILL constraint families. Under this approach, deleted elements (PARSE violations) or inserted elements (FILL violations) were the primary focus. McCarthy and Prince (1995) propose that the PARSE/FILL system be replaced with Correspondence Theory. This newer theory considers input and output forms to be in a relation of correspondence. Where the structure of the input and output is viewed as a string, Correspondence is defined by McCarthy and Prince (1995:262) as:  (28) Correspondence (McCarthy and Prince 1995:262) Given two strings Si and S2, correspondence is a relation (2fromthe elements of Si to those of S . Elements of ct9S and PaS are referred to as correspondents of one another when a$p\ 2  2  79  2  1.6.1. Correspondence Constraints  In this thesis, it will be argued that the DEP and MAX families of Correspondence constraints, proposed by McCarthy and Prince (1995), play a central role in the surface realisation of the features responsible for postvelar harmony in Palestinian and St'at'imcets.  MAX subsumes the former PARSE.  DEP subsumes the former FILL.  MAX and DEP are defined by McCarthy and Prince (1995:264) as:  39  (29) The MAX Constraint Family (McCarthy and Prince 1995:264) General Schema  Every segment of Si has a correspondent in S . 2  Specific  Instantiation  MAX-IO Every segment of the input has a correspondent in the output. (No phonological deletion.) (30) The DEP Constraint Family (McCarthy and Prince 1995:264) General Schema  Every segment of S has a correspondent in Si. 2  Specific  Instantiation  DEP-IO Every segment of the output has a correspondent in the input. (Prohibits phonological epenthesis.)  MAX and DEP are here assumed to govern features as well as segments. They are also assumed to govern the associations ('links') between features and segments. That is, besides MAX-IO and DEP-IO, the further instantiations in (31) are also assumed.  McCarthy and Prince (1995) provide base-reduplicant MAX and DEP correspondence as further instantiations of MAX and DEP. Because reduplicative correspondence will not be discussed in this thesis, those further instantiations are omitted in (29) and (30). 80  1.6.1. Correspondence Constraints (31) a. MAX-F Every feature in the input corresponds to a feature in the output. (No feature is deleted.) b. MAX-LINK Every association in the input corresponds to an association in the output. (No link is deleted.) c. DEP-F Every feature in the output corresponds to a feature in the input. (No feature is added.) d. DEP-LINK Every association in the output corresponds to an association in the input. (No link is added.)  The tableau in (32) illustrates the satisfaction and violation of these constraints. (In (32), they are equally ranked. Constraints of equal rank are separated by a dotted column border.)  (32) input: [DOR] a J3 (a) [DOR]  a  MAX-F  MAXLINK  DEP-F  DEPLINK  *  **  3  (b) a 3 (c) [DOR] | \  a 3 — [RTR]  81  1.6.1. Correspondence Constraints  The candidate outputs in (32) are evaluated as follows: candidate (a) has all the features and links that the input has, so it satisfies MAX-F and MAX-LINK. It has no feature or link that the input does not have, so it also satisfies DEP-F and DEP-LINK. Candidate (b) does not have all the features and links that the input has: the underlying [DOR] is absent, as is its underlying link to a. The missing [DOR] violates MAX-F; the missing link violates MAX-LINK. However, candidate (b) has no feature or link that the input does not have, so it satisfies both DEP-F and DEP-LINK. Candidate (c) has all the features and links that the input has, so it satisfies MAX-F and MAX-LINK. However, it has an added feature, [RTR], and two added links, viz., the link of [DOR] with P and the link of [RTR] with p. This means that candidate (c) violates DEP-F once and DEP-LINK twice. In chapters 2 and 3, the MAX and DEP constraints in (31) will be argued to have a crucial role in Palestinian and St'at'imcets postvelar harmony. For example, it will be argued that MAX-RTR ('Every [RTR] in the input corresponds to an [RTR] in the output') and MAX-LINK are highly ranked in the grammar of both languages.  The  observed effect is that [RTR] 'stays' on the segment to which it is underlyingly linked, in order to serve as a source of [RTR] harmony in the string. It will be further argued that DEP-LINK is more lowly ranked than MAX-RTR and MAX-LINK in both languges. The observed effect is that new [RTR] associations, the manifestations of the harmony, occur in the output with little penalty. The role to be identified for DEP-RTR ('Every [RTR] in the output corresponds to an [RTR] in the input') mirrors that argued by Pulleyblank  82  1.6.2. Alignment Constraints  (1994a) for a functionally equivalent constraint, REC-F ('RECOVER-F'), in Yoruba and Wolof. Pulleyblank argues that the ranking of REC-F with respect to certain Alignment and Grounded constraints derives the effects of opacity and transparency in Yoruba and Wolof '[-ATR]' harmony. E.g., it will be argued that in Palestinian, DEP-RTR is ranked above certain Grounded constraints but below certain others, where some of the Grounded constraints are Alignment constraints.  This ranking derives the observed  pattern of opacity in Palestinian pharyngealisation harmony.  1.6.2. Alignment Constraints  Generalized Alignment, proposed by McCarthy and Prince (1993b), is a constraint schema governing the coincidence of categories at constituent edges. It is formulated by McCarthy and Prince (1993b:2) as:  (33) Generalized Alignment (McCarthy and Prince 1993b:2) Align(Catl, Edgel, Cat2, Edge2) = d f e  V Catl 3 Cat2 such that Edgel of Catl and Edge2 of Cat2 coincide. Where Catl, Cat2 e PCat u Gcat Edgel, Edge2 e {Right, Left)  PCat and GCat refer to prosodic and grammatical (morphological or syntactic) categories, respectively.  Generalized Alignment provides a uniform expression for constraints that  83  1.6.3. Grounded Constraints  reference a constituent edge; see McCarthy and Prince (1993b) for several examples of such constraints. An example of a particular parameterisation of Generalized Alignment, from McCarthy (1993b:2) is:  (34) English Stress Align(PrWd, L, Ft, L) This requirement is satisfied in [(Tata)ma(gouchee)], since the left edge of the Prosodic Word coincides with the left edge of a foot.  In chapters 2 and 3 it will be argued that Alignment constraints on the features [RTR] and [DOR] are the basic force driving postvelar, harmony in Palestinian and St'at'imcets.  It will be further argued that certain of those Alignment constraints are  syntagmatically grounded, as discussed below.  1.6.3. Grounded Constraints  Grounded constraints provide phonological evidence for the Grounding Hypothesis of Archangeli and Pulleyblank (1994a). The Grounding Hypothesis says that featural relations are "rooted in the physical properties of the vocal tract or speech signal" (Archangeli and Pulleyblank 1994a: 172). That is, featural constraints must observe, and not contradict, the phonetic basis of speech.  84  1.6.3. Grounded Constraints  Grounding can be either paradigmatic or syntagmatic.  Paradigmatic grounding  relations, proposed by Archangeli and Pulleyblank (1994a), hold of featural relations within a segment.  An example of a paradigmatic grounding relation is that holding  between the features [HI] and [RTR]: in several languages, a segment specified for [HT] cannot also be specified for [RTR]; see Archangeli and Pulleyblank (1994a) for further discussion. (Here it is assumed, after Pulleyblank (1994a: 3), that [RTR] as just referenced corresponds to Archangeli and Pulleyblank's '[-ATR]'. The reasons for assuming this are discussed in chapter 4.) This prohibition is grounded in the anatomical proximity of the tongue body and tongue root. Given the proximity of those structures, tongue body and tongue root interaction are such that raising the body makes it difficult to simultaneously retract the root.  It is also based on acoustic effects: raising the tongue dorsum and  retracting the tongue root have been shown to have contradictory Fi effects.  See  Archangeli and Pulleyblank (1994a) and Pulleyblank (1994a) for discussion.  This  grounded relationship is evidence for the paradigmatic grounded constraint HI/*RTR (Tf a segment is specified for [HI], it is not specified for [RTR]'), as argued by Archangeli and Pulleyblank (1994a) and Pulleyblank (1994a).  (The two studies just cited phrase the  constraint as HI/ATR (Tf [+high], then [+ATR], not [-ATR]').) In chapters 2 and 3 it will be argued that HI/*RTR is lowly ranked in both Palestinian and St'at'imcets. Syntagmatic grounding relations hold not within segments, but between them. Shahin (1993) suggested that the physical properties of the tongue root are the phonetic basis for tongue root retraction harmony. Specifically, given the relative large mass and  85  1.6.3. Grounded Constraints  resultant sluggishness of the tongue root, tongue root articulation has a natural tendency to span more than one segment in a word.  This was identified as an instance of  syntagmatic grounding. Syntagmatic grounded constraints are formulated by Jiang-King (1996) and Pulleyblank (1997). Jiang-King proposes a syntagmatic constraint against a sequence of [HI]-[LOW]. Pulleyblank proposes that place assimilation, observed in several languages, is the effect of highly ranked syntagmatic grounding constraints on ClusterIdentity. Two syntagmatic grounded constraints to be proposed in this thesis are Align([RTR], L; Wd, L) and Align([RTR], R; Wd, R), which say, respectively, 'The left edge of the word is aligned with the left edge of any [RTR]' and 'The right edge of the word is aligned with the right edge of any [RTR]'. These will be proposed as constraints that are syntagmatically grounded in the sluggishness of the tongue root. It will be argued that they are more lowly ranked in St'at'imcets than in Palestinian. In Palestinian, they interact with a more lowly ranked DEP-LINK to force an expanded distribution of [RTR] in the output. In St'at'imcets, the reverse ranking, DEP-LINK »  Align([RTR], L; Wd,  L), Align([RTR], R; Wd, R), results in more restricted postvelar harmony for that language.  86  1.7.1. The Distinction between Phonetics and Phonology  1.7. Phonetics and Phonology  This section addresses the distinction between phonetics and phonology, and the use of phonetics in phonology. A phonetics-phonology distinction will first be drawn, based on the various characteristics of sound properties. Building on the conclusions of previous studies, the distinction will then be refined to recognise distinct speech-phonetics vs. language-phonetics, which are both distinct from the phonology. The use of phonetics in phonology will then be discussed.  1.7.1. The Distinction between Phonetics and Phonology  It is assumed here that a sound property can be identified as either phonetic or phonological according to the criteria in (35). (Unless otherwise noted, 'phonological' is used in this thesis to mean part of either the lexical or postlexical phonology. On the distinction between lexical and postlexical phonology, see Mohanan (1982), Kiparsky (1985), and Kaisse & Shaw (1985). For further discussion of criteria 2 - 4 in (35), see Pulleyblank (1986:7-8).)  87  1.7.1. The Distinction between Phonetics and Phonology (35) Necessary Criteria for Phonetic vs. Phonological Status 1. Phonological visibility, after Mohanan (1982) and Pulleyblank (1986) This means that the property has a phonological effect and is referenced in a constraint. Phonetic properties are not phonologically visible, but phonological properties are.  If there is no evidence that a given sound property has  phonological visibility, then, based on economy considerations, the property is considered phonetic. 2) Sensitivity to word-internal structure, from Mohanan (1982) and Kiparsky (1985) This means that the principles governing the distribution of the sound property refer to word-internal structure. Phonetic properties are not sensitive to wordinternal structure, but (lexical) phonological properties can be.  3) Non-discreteness, from Henke (1966), Ohman (1966), Browman and Goldstein (1990), and Keating (1990) Phonetic properties are non-discrete, that is, continuous. Because of this, they are necessarily realised with a temporal dimension so that the span of presence of one phonetic property can partially overlap with that of another.  By result, it is  frequently difficult to group a set of phonetic properties into a larger unit that can be defined by the presence of those properties. From the non-discreteness of phonetic properties, it follows that phonetic properties are gradient, that is, showing change corresponding to distance from a source. Phonological properties, by contrast, are discrete. Because they are discrete, they are not gradient. By result, segmentation, e.g., of a consonant or vowel defined by a set of phonological properties, is straightforward (although, as in cases of harmony, distinct phonological segments may share a single instance of a phonological property).  88  1.7.1. The Distinction between Phonetics and Phonology  4) Lexical exceptions, from Liberman (1983) and Kiparsky (1985) Phonetic properties cannot have lexical exceptions, but phonological properties can.  Liberman (1983) suggests another criteria, viz.: number of properties.  With  Liberman, it is assumed that phonological properties are bounded in number. This is based here on the assumption that the human brain has a limited storage capacity. However, in contrast to Liberman, the number of phonetic properties is assumed to be bounded as well, based on the assumption that the properties of the physical world, as determined by forces like pressure, magnetism, viscosity, etc., are bounded, that is, that they are in principle enumerable. The criteria in (35) imply that there is only one type of phonetics. However, there is evidence, presented, e.g., by Pierrehumbert (1980), Liberman and Pierrehumbert (1982), and Liberman (1983), that some properties which would be classified as phonetic by criteria 1-4 nevertheless vary cross-linguistically. That is, some phonetic properties are language-specific.  They must, therefore, be cognitive and part of language.  Furthermore, Steriade (1995b, 1997) presents evidence that certain language-specific sound phenomena previously assumed to be phonological cannot be given a tenable phonological account. phonetic.  As Steriade argues, such phenomena must be recognised as  On the basis of these two types of evidence for language-specific phonetic  properties, it is here assumed, following several previous works, that language has a 40  See, e.g., Liberman and Pierrehumbert (1982), Liberman (1983), Pulleyblank (1986),  40  Steriade (1995b), Flemming (1995a, 1995b), and Steriade (1997).  89  1.7.1. The Distinction between Phonetics and Phonology  phonetic component. This means there are two types of phonetics: speech-phonetics and language-phonetics. Recognition of distinct speech- vs. language-phonetics leads to the following phonetics vs. phonology distinctions: speech phonetics is purely physical, that is, it is defined in terms of the physical (anatomical, acoustic, and aerodynamic) properties of speech and lies outside cognition. Language-phonetics, while still defined in terms of the physical properties of speech, is cognitive, that is, it is part of the cognitive representation of the sound structure of language. Finally, all the phonology is cognitive — although it is constrained by the physics of speech, as discussed in §1.6.3. Table 1:12 summarises the necessary criteria for identifying a sound property as belonging to the speech-phonetics vs. language-phonetics vs. phonology.  90  1.7.1. The Distinction between Phonetics and Phonology Table 1:12 Necessary criteria for speech-phonetic vs. language-phonetic vs. phonological status Property Characteristics  SPEECHPHONETICS  LANGUAGEPHONETICS  PHONOLOGY  Nature  • • • • •  not language-specific no phonological visibility no reference to word-internal structure non-discrete no lexical exceptions  physical  • • • • •  language-specific no phonological visibility no reference to word-internal structure non-discrete no lexical exceptions  cognitive  • • •  language-specific phonological visibility possible reference to word-internal structure (for the lexical phonology) discrete possible lexical exceptions  • •  cognitive  The model of language vs. speech in (36) is assumed. The modular schematisation of language follows Archangeli and Pulleyblank (1994a:5), except that (36) makes the distinction between language- and speech-phonetics explicit. The placement of languagephonetics as feeding into phonology follows the suggestion of Donca Steriade (p.c). On the interaction between the syntax, semantics, morphology, and phonology modules, see Archangeli and Pulleyblank (1994a:4-5,433).  91  1.7.1. The Distinction between Phonetics and Phonology (36)  LANGUAGE SPEECH PHONETICS  \  \  passed to speech-plruonetics  A final discussion concerns the notational representation of phonological and phonetic data. In this thesis, three types of representations will be used in transcriptions: 1) Underlying phonological form, presented between slashes ('//').  2) Surface  phonological form, presented between the braces 'i}'. 3) Phonetic form, presented between square brackets. The three types of representations are explained as follows: the underlying form is the phonological form stripped of its predictable properties. The surface ('output') form is the underlying form + all properties added in the phonology. The phonetic form is the phonological output + additional phonetic properties, but minus any word-internal morphological boundaries. (The absence of word-internal morpheme boundaries follows from criterion 2 in (35).) Selected phonetic properties are included in the transcription of the phonetic form, depending on the point of discussion. This means  92  1.7.1. The Distinction between Phonetics and Phonology  that a single phonetic form might be represented in various ways, since certain phonetic properties might be included in its transcription on one occasion but not on another. For example, the Palestinian form meaning 'perfume', represented these three ways, is:  (37)  underlying:  /TUtr/  surface:  VTu.turl  phonetic:  ['To.torl or ['Tu.turl, etc.  where '  ' represents phonetic voicing.  The three-way distinction between underlying, surface, and phonetic form is adopted here because it permits transcription in which the claims of the transcriber, with respect to the sound system of the language being transcribed, are made explicit. For example, the representations in (37) make the claims listed below:  (38) In Palestinian, a. The properties that combine to form the sequence "Tu.tur are phonological. That is, they are language-specific, phonologically visible and discrete; they have possible reference to word-internal structure and possible lexical exceptions. b. The presence of T t r and U in a morpheme is non-predictable. c. Syllabification, stress assignment, vowel epenthesis, uvularisation of ?, and pharyngealisation of U are predictable. d. The form meaning 'perfume' is monomorphemic. e. In the form meaning 'perfume', mid vowel height and devoicing of ? are phonetic properties. That is, they are phonologically invisible and non-discrete, have no reference to word-internal structure, and no lexical exceptions; they are either Palestinian-specific or nonPalestinin-specific.  93  1.7.2. The Use of Phonetics in Phonology  In (38e), mid vowel height and devoicing of T in the example form are described as either Palestinian-specific or non-Palestinian-specific. This is because, to my knowledge, there is no study which has determined whether those properties are due to Palestinian or solely to the physics of speech; such a study will not be undertaken in this thesis. However, if they were the former, they would be language-phonetic.  If they were the  latter, they would be speech-phonetic.  1.7.2. The Use of Phonetics in Phonology  It is assumed that phonetic data can provide crucial support for a phonological account. The support may be articulatory, or both articulatory and acoustic, as follows: let us assume that a phonological analysis claims that a segment is specified for feature F. If articulatory data indicate that the segment is produced with the appropriate articulation(s) (e.g., lip rounding for a segment claimed to be specified for [LAB]), then that articulatory data can be interpreted as phonetic support for the phonological claim that the segment is specified for F. Additionally, the supporting articulatory data can be mapped through a reliable articulation-to-acoustics model, such as the model of Stevens & House (1955), discussed in §1.4.3.  If acoustic data from the segment match the  predictions of the model (e.g., a lowered F for a segment shown to be produced with lip 2  rounding), then that acoustic data can be interpreted as support for the articulation(s) on which the acoustic predictions were based, strengthening the articulatory support for the phonological claim with respect to F.  94  1.7.2. The Use of Phonetics in Phonology  In the absence of articulatory data, it is assumed that acoustic data by itself can strengthen presumed articulatory support, but under a hypothesis that leaves the door open for other articulations that could have produced the same acoustic effects; see §1.4.3 for further discussion.  95  Chapter 2: Pharyngealisation Harmony and Uvularisation Harmony in Palestinian Arabic  2.1. The Language and the Data  Palestinian Arabic belongs to the Palestine-Jordanian variety of Levantine Arabic; see the dialectal classifications of Cantineau (1940/46) and Eisele (1987). Palestinian dialects can be classified as either medini (urban), fellahi (rural) or bedui (bedouin), and as either northern, southern, eastern, western, central, or coastal, according to their location in the former Palestine. On the sociolinguistic classification, see Cadora (1992); on the geographical classification,  see Shahin (1996).  To my knowledge, a complete  sociogeographical classification of Palestinian dialects has not been compiled. Unless otherwise noted, the Palestinian data in this thesis are from the Abu Shusha Palestinian dialect, which is the western central fellahi  spoken in the pre-1948 Palestine  village of Abu Shusha. For the location of this former village, see Appendix II. In Abu Shusha, Old Arabic Iql is variously realised as emphatic Ikl or as velar Ikl. As a fellahi, Abu Shusha's most salient marker is the affrication of Old Arabic Ikl to /tj/; see Fischer and Jastrow (1980) for discussion of this feature. The dialect is also marked as rural by its  96  2.1. The Language and the Data  lack of emphatic Id/: emphatic /5/ occurs in Abu Shusha where Idl occurs in urban Palestinian dialects; this rural marker is also noted by Card (1983:107). The data were gathered by the author, most of it during six months offieldwork in The consultants were fellahln  Ramallah on the West Bank, 1994-1995.  (villagers,  peasants) from Abu Shusha. The primary consultants were a male, aged 58, and a female, aged 45, with whom I lived for the duration of the fieldwork. Frequent daily interaction with these and six other consultants (a male, aged 80, and five females, aged 85, 68, 67, 65, and 28) and periodic interaction with 18 others (eight males, aged approximately 40 85, and ten females of approximately the same age range) yielded a large set of lexical items and phrases, which were then tape-recorded from the female consultant of age 45. Subsequentfieldworkwas conducted in Vancouver with a male native speaker, aged 32. The total corpus is approximately 1500 words and phrases. For the acoustic study of Palestinian from whichfindingswill be reported in this chapter, both Abu Shusha and Jafa (northern coastal medini) tokens were used. (Only one Abu Shusha speaker was available for the acoustic study.) The original location of the Jafa dialect is also shown in Appendix II. Jafa has the salient medini marker of /k/ > /?/. Its lacks fellahi affrication, and has 161 where Abu Shusha has /5/. One dialectal feature directly relevant to postvelar harmony is that in Abu Shusha, stem-final vowels do not pharyngealise, but in Jafa they sometimes do, at least phonetically; this will be discussed in §2.4.5. Data showing this and other dialectal differences appear in Appendix ITJ, which lists the carrier forms for the Palestinian consonant and vowel tokens that were analysed in  97  2.2.1. Consonantal Inventory  the acoustic study. (Appendix JJI will be properly introduced in §2.3.1.) The acoustic data were tape-recorded in Vancouver from the 32-year-old Abu Shusha speaker mentioned above and a male native speaker of Jafa, aged 29.  2.2. Phonemic Inventory 2.2.1. Consonantal Inventory 2.2.1.1. The Palestinian Underlying Consonantal Inventory  The underlying consonantal inventory of (Abu Shusha) Palestinian is presented in (1). The lack of underlying non-emphatic Irl will be discussed in §2.2.1.3.2. The Palestinian underlying and surface vowels are presented in (2) and (3). surface inventory in (3) is the inventory at the output of the phonology.  The  (It is not a  phonetic inventory). The vocalic inventories are presented here to provide a frame of reference for the vowels that occur in the data to be given in this section. They will be thoroughly discussed in §2.2.2. Upper case T E M O U' denote underlying vowels. The featural values represented by T E JE 0 U' and by the IPA symbols in (3) will discussed in §2.3.3. Palestinian has an epenthetic vowel, which surfaces either highfrontor high back. Its featural values will be also be discussed in §2.3.3.  98  2.2.1. Consonantal Inventory (1) The (Abu Shusha) Palestinian Underlying Consonantal Inventory LAB INTERALV POSTVEL UV PHAR DENT ALV OBSTRUENTS STOPS: t b  b  t  k  k  d  TRILLS:  r AFFRICATES:  d5 FRICATIVES:  •f  e  s s  5 5  z  s  RESONANTS  NASALS:  m m  n  APPROXIMANTS:  w  I I  j  X  (2) The (Abu Shusha) Palestinian Underlying Vocalic Inventory a. short vowels BACK FRONT HIGH MID LOW  I E  U 0 M  b. long vowels FRONT  BACK  HIGH  I:  U:  MID  E:  0:  LOW  JE:  99  n  GL  2.2.1. Consonantal Inventory (3) The (Abu Shusha) Palestinian Surface Vocalic Inventory a. short vowels NON-RTR  BACK  CENTRAL  FRONT RTR  NON-RTR  NON-RTR  RTR  NON-RD non-bk  HIGH  i  I  MID  e  e  LOW  ae  non-bk  RTR RD  NON-RD  U 8  > 3  RD  bk  3  O  u  A  0  a  a  b. long vowels FRONT NON-RTR NON-RD  BACK NON-RTR NON-RD  RD  HIGH  i:  u:  MID  e:  o:  LOW  ae:  e:  The IP A vowel chart (revised to 1993) is presented in (4), to show the placement of the Palestinian mid and low surface vowels in the IPA vowel space.  ('Mid' in (3)  correspond to 'close-mid' or 'open-mid' in (4); 'low' corresponds to 'open'. For symbols that are paired in (4), the one on the left denotes a non-rounded vowel, the one on the right denotes a rounded vowel.)  100  2.2.1. Consonantal Inventory (4) The IP A Vowels (revised to 1993)  FRONT i\y  CLOSE  CENTRAL +T « I Y  BACK UU • U  \  U  9 •e  CLOSE-MID e \ 0  a  OPEN-MID  OPEN  \  e • CB  3•o  ee \  e  a • CE  a• D  1  2.2.1.2. The Palestinian Surface Consonantal Inventory  The Palestinian surface consonantal inventory, that is, the consonantal inventory at the output of the phonology, is seen in (5).  The surface inventory differs from the  underlying inventory in (1) by containing several additional emphatics. All underlyingly non-emphatic consonants, except the post-alveolar obstruents l\ tf, &J, have surface emphatic counterparts which arise through uvularisation harmony. E.g., in \ n.'Qiif \ 'clean (masc. sg.)' (Adj), underlyingly non-emphatic In fl surface as emphatics because they undergo uvularisation harmony with the underlying emphatic 161.  The effects of  uvularisation harmony will be detailed in §2.5. Acoustic findings that support recognition of the surface emphatics in (5), including the surface emphatic gutturals, will be presented in §2.5.1.  The theoretical reasons for the lack of surface emphatic  addressed in §2.3.2 and §2.5.4.  101  tf  will be  2.2.1. Consonantal Inventory (5) (Abu Shusha) Palestinian Surface Consonantal Inventory  LAB  INTERDENT  ALV  POSTALV  VEL  TJY  PHAR  GL  OBSTRUENTS STOPS: t  b b  t  k  k  ?  •  ?  dd  TRILLS:  r r AFFRICATES:  d5  FRICATIVES:  f f  8 8 58  s s  J  h h  z z  RESONANTS NASALS:  m m  n n  w vy  II  j j  %X  ft  B B  T T  •  •  •  The surface consonantal inventory also differs from (1) by containing surface nonemphatic U r . The non-underlying status of non-emphatic U r in Palestinian is indicated by the findings of Younes (1993, 1994) for a northern Palestinian dialect. Abu Shusha data support Younes' analysis, as will be shown in §2.2.1.3. Finally, in word-initial position, epenthetic -j ? \ is observed. This will be discussed in §2.2.1.4.  102  2.2.1. Consonantal Inventory  2.2.1.3. Postvelars  Palestinian has 14 postvelars: six gutturals, /? h ? ti u yj, and eight emphatics, /m b I 5 s t r k/. On the articulatory nature of these segments, see §1.4.1 and §1.4.2. This section addresses certain issues regarding their classification.  Theoretical issues  pertaining to gutturals and emphatics — their feature specifications, their roles in Palestinian's two postvelar harmonies, and the formal bases for their roles — will be addressed in §2.3 - §2.5.  2.2.1.3.1. Guttural Approximants  As reflected in (1) and (5), Palestinian pharyngeal and uvular gutturals /? tifcfyj are analysed here as approximants, not fricatives. This manner identification for gutturals was proposed by Catford (1977) and is pursued by McCarthy (1994:194) and Ladefoged and Maddieson (1996:168). Frication is frequently observed for Palestinian /? Ti u %l in voiceless contexts, but the frication is considered here to be a phonetic effect. The change from approximant to fricative manner is explained as a result of the increased rate of airflow in the voiceless context; see Stevens (1971) for a detailed discussion. Observing this, Catford (1977:122) defines 'approximant' as "non-turbulent when voiced; but the flow becomes turbulent when they are made voiceless" (italics in the original); see Ohala and Ohala (1993:232-233) for a similar description.  103  2.2.1. Consonantal Inventory  I have examined some 30 Palestinian guttural tokens in both word-initial and wordinternal, voiced contexts (where 'voiced context' means that the guttural is flanked by vowels or voiced consonants.) Presentation of that study is deferred for work elsewhere. However, preliminary findings are that word-initial IS til are frequently voiceless for most of their duration and are accompanied by frication. In word-medial, voiced context, they are voiced throughout, high amplitude, and drive full formant structure. Further study is necessary to determine the degree of possible frication of IS til in the voiced context. Because the frication of voiceless lf\ yj can be attributed to aerodynamic coincidence, as can the frication of voiced IS til in voiceless contexts, /fi yj are analysed here as voiceless approximants. The usual frication of voiceless approximants is discussed further by Catford (1977:120). The observations of Esling (1996, 1997) suggest that the frication of Arabic pharyngeals might be produced in the region of the epiglottis rather than the tongue root. In §2.3.1, it will be proposed that the approximant manner of Palestinian gutturals results from their specification for [SON], and that their [SON] specification is the basis for the phonological distinction between Palestinian pharyngeal gutturals IS TV and laryngeal gutturals /? hi.  104  2.2.1. Consonantal Inventory 2.2.1.3.2. No Underlying Non-emphatic Irl  Younes (1994) argues that surface non-emphatic M is not underlying in the Dar Younes (northern fellahi) Palestinian dialect. He argues that M is underlyingly emphatic Irl which surfaces 'de-emphaticised'. His evidence for this is that the Dar Younes trill always surfaces emphatic, except in a defined set of contexts. The trill in Abu Shusha displays the contextual alternation described by Younes.  Data showing this will be  presented below. On the basis of these data, it will be concluded that, like Dar Younes, Abu Shusha lacks underlying non-emphatic Irl. The data in (6) are Dar Younes forms, from Younes (1994:218), in which the trill surfaces as emphatic \ r\. (In this section, Younes' transcription of Dar Younes vowel quality will be retained. The surface form status of Younes' data is inferred from his discussion.  I have added the ungrammatical forms.  Finally, Younes does not mark  emphasis for the surface emphatic consonants in his data; their transcription below follows Younes' (1994:218-219) summary of their contexts of occurrence in his dialect.) (6) Dar Younes Forms a. <!na:r^  (*-jnee:r|-)  'fire'  h.\&&\r\  (*-!cfeae:r|-)  'neighbour'  Younes explains that in Palestinian, the occurrence of the back (long or short) low vowel indicates the presence of an emphatic consonant in the word, as does the occurrence of emphatic variants of otherwise non-emphatic consonants; cf. •In d5(- in (6).  and  He analyses back ia:\ and surface emphatic -In cfeMn (6) as arising  105  2.2.1. Consonantal Inventory  through uvularisation harmony triggered by the emphatic M .  (Younes refers to  uvularisation harmony as 'emphasis spread'. For acoustic support for emphatic ir\ in forms such as those in (6), see §2.3.1.1. The properties of uvularisation harmony in the Abu Shusha dialect will be addressed in detail in §2.5.) Younes (1994:220-221) lists three contexts for /rV-de-emphaticisation. He describes 1  the first as "in the neighborhood of a noninflectional, nonepenthetic, nonlowfrontvowel." This context is here rephrased as: in a word containing a root-internal /!(:)/ or /E(:)/. Dar Younes forms showing non-emphatic M in this context, from Younes (1994:220), are presented in (7). (I have added the underlying forms.) (7) Dar Younes Forms a. M r t E : n /  ix\r.'fae:n\  (^xir.'fainh  'lambs'  b. /b^Erml:!/  ibeer.'mi:^  (*ibar.'mi:lh  'barrel'  c. /yEir-iEk/  \ 'ye: .r-aek I-  (*<! 'ye: .r-ak \)  'other than you [masc. sg.]'  As seen the low vowels in (7) surface front, not back. This is evidence for non-emphatic \x\ in these forms. The Abu Shusha cognates of the data in (7) are presented in (8). As seen, the Abu Shusha trill also surfaces non-emphatic in this context. Were the trill emphatic in (8), the low vowels and non-emphatic consonants in each form would surface as seen in the ungrammatical forms provided; this claim will be substantiated in §2.5.  2  ^he three contexts are also identified in Younes (1993). The \ e: \ in (8a) is the backed variant of Abu Shusha long  2  reduced variant of /JE/; this will be shown in §2.2.2.5. 106  Mid \ z\ in (8b-c) is a  2.2.1. Consonantal Inventory (8) Abu Shusha Forms a. /xIrfiE:n/ b. IbJErml'M c. ltiE:r-JEkl  -Ixir.'faeml-  (*-!xir. fe:n!0 l  v  ibsr.'mi:^ -|'bfe:.r-3k|-  y  *  (^baVmiilr-) (* ! Be:.r-3 k^) <  l  'lambs'  h  >  'barrel' 'other than you (masc. sg.)'  The second context Younes identifies for /r/-de-emphaticisation is immediately before one of /8 5 t d s z n f tj cfe j / in the same stem. This set of consonants corresponds to the set of (non-emphatic) non-lateral coronals.  De-emphaticisation in this context is  illustrated by the data in (9), which are from Younes (1994:221).  (9) Dar Younes Forms a. ^baer.'d-eeinl-  (*-!bar.'d-a:n^)  'cold (masc. sg.)'  b. -i 'daer.s-aek !•  (*•! 'dar.s-ak <•)  'your [masc. sg.] lesson'  c. r?aer.naeb^  (*V?ar.nab^)  'rabbit'  The Abu Shusha cognates of the forms in (9) are presented in (10). As seen, the trill is non-emphatic in this context.  3  (10) Abu Shusha Forms a. -!b3r.'d-ae:n^ ("Hba'V.'d-einh  'cold (masc. sg.)'  b. \ 'dar.s-3k \  (*•! dAr.s-3 k \)  'your (masc. sg.) lesson'  c. Har.nsbi-  ^i'TArwb})  'rabbit'  l  >  Potential counterexamples to Younes' second generalisation are presented by forms  3  Abu Shusha \a\ \s the pharyngealised variant of IMl in a closed, that is, CVC, syllable;  { A} is a uvularised-pharyngealised variant of IMl in the same context; see §2.2.2.6. 107  2.2.1. Consonantal Inventory  containing /w/.  An Abu Shusha example is /wJErd-JE/  -j'war.d-8^ 'flower'.  4  In the  surface form of this word, the initial-syllable vowel is more similar to back [A] than front [a].  However, forms such as this are here not considered to counter Younes'  generalisation because in them underlyingly non-emphatic consonants do not surface emphatic. E.g., in i'war.d-ai-,  -Iwi- and-ld^ are non-emphatic. Were the trill in this word  emphatic, surface emphatic -i vy r and {d \ be observed. Because they are not, the non-front quality of the initial-syllable vowel is considered a coarticulatory phonetic effect of the preceding <! vW, yielding the transcriptions: /W/Erd-^E/ \ 'war.d-a^ ['wa* r.d-s], where denotes phonetic backing and raising. The third context Younes identifies is before a velar in the same root. Forms which show non-emphatic Ur in this context,fromYounes (1994:221), are presented in (11).  (11) Dar Younes Forms a. i'teer.reex} (*Vtar.raxh b. -{'faer.raeyl(*Vfar.ray!-) c. \ 'rae:.faek\  'he dated' 'he emptied'  (*i Ya:.faek})  'he befriended'  Abu Shusha has uvular lu %l instead of velar /y x/. However, Abu Shusha data with the trill before velar /k/ in the same root are presented in (12).  As seen, the trill is non-  emphatic in this context.  4  Mid "i a \ is a second reduced variant of IRI (besides the one mentioned in note 2). The  reduced variants of /Ml will be discussed in §2.2.2.5.  108  2.2.1. Consonantal Inventory (12) Abu Shusha forms a. -I'raei.fak}b. \ 'mae.r3k^  ('"•i'rei.fs^h  'hebefriended'  (*i mA.r3 k|>)  'he spoiled (someone/something)'  l  >  On the basis of data such as those in (8), (10), and (12), Abu Shusha \r\ is here analysed as de-emphaticised Irl. For this reason, the underlying inventory in (1) does not include non-emphatic Irl.  2.2.1.3.3. High-frequency vs. Low-frequency Emphatics  In Palestinian, /m b \l occur lessfrequentlythan the other emphatics. Forms in which they do occur are seen in (13). Younes (1982:57) identifies (13d) as a borrowing from Italian, (13e) as a borrowingfromClassical Arabic.  5  'water'  d. Vba.ba^  'daddy'  b. -I'Wri.m-oi-  'paternaluncle'  e. V?Al.l3 h}'  'God'  c. -I'ma.maf-  'momma'  f. <!'jxl.lsJ-  'let'sgo!'  (13)a.  >  The segments i m b I} have been called ' secondary emphatics' while \ 5 s t r k} have been °  h  h  h  J  r  •  •  •  •  •  called 'primary emphatics'; see, e.g., Younes (1994). In this thesis, the first set will instead be referred to as 'low-frequency emphatics', the second set as 'high-frequency  Younes (1982:57) and Herzallah (1990:39) record the initial-syllable vowel in (13c-d) as  5  long in the Dar Younes and Yaibad dialects, respectively. (Yaibad is a northern fellahi very similar to Dar Younes; see note 18.) However, in Abu Shusha, it is short. 109  2.2.1. Consonantal Inventory  emphatics', to avoid confusion with the terms 'primary articulation' and 'secondary articulation'. An issue with respect to the two sets of emphatics is whether or not irn b I} should be recognised as underlying / m b I/; see, e.g., Younes (1994) for discussion. With regard to this issue, the differences between the two sets, which are solely distributional, are usually discussed. Maamouri (1967), Younes (1982, 1994), and Herzallah (1990) observe that there are more restrictions on the occurrence of the low-frequency emphatics than on high-frequency emphatics. Younes (1994) shows that the latter (i) have minimal lexical contrasts with non-emphatic counterparts (e.g., / t l : n / it\ir\ \ 'figs' vs. / t l : n / ^ti:nr- 'mud), (ii) have a high frequency of occurrence, and (iii) occur in all positions and adjacent to all vowels. These properties are not shared by the low-frequency emphatics.  As Younes  discusses, there are no lexical contrasts between emphatic -Irn b \} and non-emphatic {m b 11-; that is, there are no minimal pairs like {'ma.ma } • vs. *•! ' m a s . m a !•). Furthermore, {m b 11- occur only in a handful of forms, several of them borrowings. (This is presumably why there are no lexical contrasts between •! m b I} and -I m b IK) As for vowel contexts, Younes (1994:216) describes -|m b \\ as occurring only with low vowels. Maamouri (1967:49) says they occur "almost exclusively with low vowels." More study is needed to determine whether Abu Shusha data match Younes' or Maamouri's generalisation on this last point.  110  2.2.1. Consonantal Inventory  Regardless of the distributional limitations on i m b \ }, there is a strong argument for recognising them as underlying /m b I/, viz.: the uvularisation of •! m b I} in forms like those in (13) but not in forms like those in (14) is unpredictable.  (14) a.'imaTr b.  rea.tiibr  c. "i'naei.l-ir  (*lmA?r)  'with'  (*r6A.ttabr)  'gold'  (*rtte:.j-ir)  'myself  This unpredictability indicates that their uvularisation is underlying. For this reason, /m b 1/ are included in the underlying inventory in (1). After Card (1983:106-107), the following procedure will be used for identifying underlying emphatic /m b I/: if a high-frequency emphatic occurs in a word containing one of surface emphatic <!rn b U, it will be assumed that im b \\ are underlyingly nonemphatic /m b I/, and that surface {m b I} are derived in such words by uvularisation harmony. Example transcriptions illustrating this procedure are: /bJE\JEit/  ^b3 .'lB:tr >  'tiles', and ImJErJE/ {'ma.ra!- 'woman, wife'.  2.2.1.4. Epenthetic Word-initial {?}  It is assumed here that most word-initial glottal stops in Palestinian are epenthetic. The evidence for this is that word-initial glottal stop is not observed when another consonant is present to serve as onset of the word-initial syllable. This is seen from the  i l l  2.2.1. Consonantal Inventory  data in (15), compared with those in (16). (Palestinian verbs glossed as infinitives are in the colloquial citation form, which is the root + imperfect vocalism.)  (15) a./b-ItJ5Ib/ b./b-Iktlb/  (16) a. /Itjaib/ b. /Iktlb/  rb-itj.5ib^  (^bi.-'TrtJ.Sib^)  'he lies'  Vb-ik.tib^  (*i bi.-'Tik.iibr)  'he writes'  r?rtj.5ib^  (*VilJ.5ib^)  'to lie' ('to tell an untruth')  V?ik.tib^  (*rik.tib^  'to write'  Forms such as those in (17) are an exception to this generalisation. (An epenthetic vowel occurs in the initial syllable of the grammatical surface forms in (17).  The  Palestinian epenthetic vowel will be discussed in §2.3.3.3)  (17) a. /b-?JEbJE:V b./b-?^ExXlf/  {bi.^ <  Jbeit}  (*{b-aVbeitj-  lbi.-'?ax.X i'  'b-aX-X <")  ir  ir  'he hugs' 'he causes (someone, something) to be late'  In (17), {?} is observed despite the -ibl- that is available to serve as onset of the initial syllable. To rephrase the analysis of (15) and (16), epenthetic i ? \ occurs at a left word edge in words which would otherwise contain no word-initial consonant. Since the in (17) do not occur at a left word edge, they are unexpected and so require an explanation. The explanation adopted here is: the -i?^s in (17) are unexpected because they are unpredictable. That is, they are underlying (Ci of the lexical root). On this basis, it is concluded that Palestinian has both an underlying /?/ and an epenthetic i ?}.  112  2.2.1. Consonantal Inventory  Word-internally, however, glottal stop does not occur to provide an onset for a vowelinitial syllable. That is, word-internal vowel hiatus occurs. Forms showing this are:  (18) a. ffJEiQl-JE/  (*i 'fBI.5i.-?9 0  'empty (fern, sg.)' (Adj)  b. /m^JI-I:n/  <| ma.yi.-'iinr  (*lrrtB.Ji.-'?i:rU)  'walking (masc. pi.)' (Adj)  c. tfJEfU-V  <!'fa.ru.-ii-  (*i 'fa.ru.-7rt)  'my fur'  d. /rriiErlUil/  •im9.ri.-'u:l^  (*-im9.ri.-'?u:rt)  'apron'  e. /wltl-^Et/  Vwi.ti.-3 t^  (*< 'wi.ti .-73*10  'lowered (fern, sg.)' (Adj)  •  >  The hiatus forms in the Palestinian corpus of this study involve the vowel sequences \, io:}.  i\.Q\,  V \ M : \ , iei.W,  iu:.\\, and {u.e:K None involve short ie\, short io\, or long  This is presumably a coincidental result of the general lower frequency of the mid  vowels. The lack of hiatus involving two long vowels is expected, since a long vowel is shortened before another long vowel in the word; see Abdo (1969) and Abu-Salim (1986) for discussion. In Abu Shusha hiatus forms, glide formation does not occur. It does occur in the Dar Younes dialect, as seen from the Dar Younes forms: {'fa:5.j-i !• and i maeJ.'j-Kn i-. In the 6  spectrogram of an Abu Shusha hiatus form, acoustic support for a syllable boundary between the two vowels is a brief pause at the point of hiatus. In a form like i me.Ji.-'kn \, f is higher for the stressed, final-syllable Vw} than for the unstressed, penultimate-syllable 0  Vii  6  I thank Munther Younes for these data. 113  2.2.1. Consonantal Inventory  Hiatus does not occur in all contexts. Prefixal vowels are elided before a stem-initial vowel. This is seen from comparing forms such as -im3.-Juf-t.-'hee:-J} T didn't see her' (no elision) and im-a.'d^as:-^  'he didn't come' (elision; cf. "I'Tae.dja^ 'he came'). More  study is needed to determine the full range of contexts in which hiatus is and is not allowed. Important to the issue at hand is that in Abu Shusha a glottal stop is not inserted to break up hiatus, so that surface forms such as *{md.$\.-'?kr\},  *Vfe:.5i.-?8^,  *i wi.ti.-?3 ti\ etc., are unarammatical. Based on thisfinding,and on the evidence that l  >  onsetless syllables are crosslinguistically highly disfavoured (see, e.g., McCarthy and Prince 1993a, Prince and Smolensky 1993), Palestinian word-initial i?} is here analysed not as a phonetic effect but as a default consonant epenthesised in the phonology to provide an onset for a word-intial syllable. The same is assumed for word-initial glottal stop in Classical Arabic by Prince and Smolensky (1993). Palestinian word-initial {?} is assumed to be imposed by the constraint 'ONSET' ('ONS': 'Syllables must have onsets'; McCarthy and Prince, 1993a, Prince and Smolensky 1993). Hiatus data such as those in (18) indicate that ONS has the decompositions 'ONS-wd[cr' ('Word-initial syllables have onsets') and 'ONS-G' ('Syllables have onsets') which are ranked in Palestinian: ONS-wd[o~ »  ONS-rj.  Finally, the data in this section have shown that in Abu Shusha Palestinian, onsetless syllables are illicit word-initially, but licit word-internally.  114  This counters the usual  2.2.2. Vocalic Inventory  assumption, expressed, e.g., by Majdi and Winston (1994:186) and Lee (1995:359), that all Arabic syllables must begin with a consonant.  2.2.2. Vocalic Inventory  Palestinian's underlying vocalic inventory is presented in (19).  It has a length  distinction and three degrees of height. There is no underlying low front vs. low back distinction.  (19) The (Abu Shusha) Palestinian Underlying Vocalic Inventory a. short vowels  HIGH MID LOW  FRONT  BACK  I  u  E  0 JE  b. long vowels  FRONT  BACK  HIGH  I:  U:  MID LOW  E:  0: JEl  The surface vocalic inventory is presented in (20).  115  2.2.2. Vocalic Inventory (20) The (Abu Shusha) Palestinian Surface Vocalic Inventory a. short vowels  FRONT NON-RTR  RTR  CENTRAL NON-RTR RTR non-bk  HIGH i  I  MID  e  S  LOW  33  BACK NON-RTR NON-RD RD  RTR NON-RD RD  non-bk bk  U 3  >  3  3  O  a  U A  0  a  b. long vowels  FRONT NON-RTR NON-RD  BACK NON-RTR NON-RD RD  HIGH i:  u:  MTD  e:  o:  LOW  ae:  e:  The surface inventory differs from the underlying inventory by containing several more short vowels. The surface short vowels comprise non-retracted-tongue-root ('nonrtr')/retracted-tongue-root ('rtr') pairs: ii} vs. ii}, ie\ vs. is}, last vs. la*t, ia\ vs. i3*t, \ o \ vs. \ o I-, and \ u \ vs. -I u}. In the low back position, however, only rtr i a 1- occurs. The reason for the lack of a non-rtr low back short vowel will be discussed in §2.5.2. Two additional short vowels are mid central backed {3 \ and mid back i AK >  The surface inventory also contains two diphthongs: i ai >• and •! au\, which are here analysed as arising from underlying /y£/-glide sequences because they always occur  116  2.2.2. Vocalic Inventory  followed by V\\ and iw^, respectively; cf., e.g., -I 'saij.jid} 'to hunt', -jkau.'waeim^ 'quickly'. Finally, the surface inventory differs from the underlying inventory by containing two long low vowels instead of one. The claim that the Palestinian vocalic system has three degrees of height is controversial. To my knowledge, there has been no thorough analysis of the surface short system, and no systematic investigation of the vowel reduction. In the discussion that follows, previous treatments of the Palestinian/Levantine vocalic system will first be summarised. Justification of the inventories in (19) and (20) will then be presented.  2.2.2.1. Previous Analyses of the Palestinian/Levantine Vocalic System 2.2.2.1.1. Non-generative Analyses  Non-generative analyses typically define the Levantine vowels in terms of the Classical Arabic system of three long, three short, and no mid height.  7  Although no  underlying vs. surface inventories are typically posited in such studies, an underlying inventory is usually implied. Non-generative studies of Arabic have usually implied that the back low vowels result from emphasis harmony in the context of an emphatic  See, e.g., Schmidt and Kahle (1918/30), Bauer (1926/70), Cantineau (1960), Grotzfeld  7  (1964, 1965), Palva (1988), and Nishio (1992). 117  2.2.2. Vocalic Inventory  consonant. In so doing, they imply the distinct underlying vs. surface inventories seen below:  (21) The Vocalic System of Palestinian/Levantine Arabic Implied by Non-Generative Studies a. underlying inventory FRONT BACK HIGH LOW  I: I  U: U JE: JE  surface inventory FRONT HIGH LOW  i: i as: ae  BACK u: u a: a  The non-generative studies note that long mid [e:] and [o:] occur in forms where Classical Arabic has the diphthongs /ai/ and /au/. (In this section, where previous studies either implicitly or explicitly ascribe underlying, surface or phonetic status to a vowel, the status will be indicated by the vowel transcription presented; e.g., I interpret the studies just mentioned as implying that the long mid vowels are phonetic.)  They record  occurrences of short mid [e] and [o] but analyse them as lowerings of I'll and / u / , respectively, conditioned by gutturals and emphatics. For example, Cantineau (1960:110111) states: Dans les dialectes modernes de l'arabe, les timbres vocaliques semblent a premiere vue nombreux et varies... De fait les sujets parlants, dans la plus grande partie du monde arabe, ne distinguent actuellement comme autrefois que trois timbres phonologiques de voyelles breves, timbres susceptibles de diverses realisations phonetiques suivant la nature des phonemes voisins...  118  2.2.2. Vocalic Inventory  Les pharyngales h et , parfois les velaires h et g attirent vers a le timbre e  des voyelles yoisines...  Les consonnes emphatiques ou mufahhama,  parfois les velaires h, get q, reportent en arriere le point d'articulation des voyelles voisines. [Cantineau denotes '? t\ u %' as ' h g h\ respectively.] £  Bauer (1926/70:11) remarks: "Unter dem Einfluss eines umgebenden h h ' r wird kurzes i — seltener langes 1 —zu e getriibt... u ... [w]ird durch umgebende Kehllaute ziemlich nach o, auch 6." (Bauer denotes '? Ti % r' as '' h h f, respectively.) Rtr short vowels are usually unmentioned.  As an exception, Bauer (1926/70:11)  describes [i] and [u], which he assumes to be the invariant values of hi and lul, respectively. He states: "i = kurzes i in Sinn izm\, z.B. bint Tochter... u = kurzes u in Mutter -I 'mu.ta^, z.B. kutub Bucher" (italics and German transcriptions are mine/KNS). Note that rtr short vowels are recorded for Classical Arabic. Gairdner (1978:194) quotes the traditional Arabic grammarian Ibn As Sarraj as stating: "The tongue sinks lower with the kasra, i (short i), than it does with the ye, long i. Gairdner explains: "In other words, /' was wider than 1, which latter was the narrow or extreme /'. We have here, apparently, the same distinction between English 'bit' and 'beat'; and we may reasonably assume that the same distinction held good in the u family, namely that short u was wider than u, as in the difference between English 'foot' and 'school'." As a variation on (21), Palva (1988) recognises mid long \e\ \ and -lo:!-, stating [p.228] that they are "most often the reflexes of /ay/ and /aw/ respectively". (1992) asserts long mid vowels for Jbali (Sinai) Arabic, but provides no comment.  119  Nishio  2.2.2. Vocalic Inventory  Nishio's short vowel inventory includes -ie\ and io\. He states [p.xiv] that short {e\ and io\  "...can be regarded as having some relevant status, at least at the surface  representation", but concludes (ibid):  "[p]honologically speaking, this dialect has a  familiar functionally triad system of short vowels". By this, he implies the underlying short set: /I JE U/. Reduced vowels are documented by Mattsson (1911), Grotzfeld (1964, 1965) and Nishio (1992). Mattsson (1911) documents short high vowel reduction for Beirut Lebanese. Grotzfeld (1965:12-13) describes it for Damascus Syrian, citing forms like /min/ > {man} 'from', /yiktubu/ > •iyaktbul' 'they write', /kutub/ > -ikatoM 'books'. (Groztfeld's transcription is retained here.) Such high vowel reduction does not occur in Abu Shusha: the Abu Shusha cognates of the Damascas data just cited are "JmirU, {'b-i.krt.b-u \, and i 'ku.tub}. Nishio (1992) describes both high and low short vowel reduction for Jbali, noting that low vowel reduction is the less frequent of the two. For the low vowel, he states [p.xvi]: "in the unstressed syllable, particularity at the beginning of a word, /a/ is reduced to [a] in casual speech as in e.g. [rawwahtu ~ rawwahtu] (= "you (pl.m.) went" Pf. 2 pl.m.)." Short low vowel reduction occurs in Abu Shusha, as will be discussed in §2.2.2.5.  120  2.2.2. Vocalic Inventory  2.2.2.1.2. Generative Analyses  Previous generative studies which explicitly address the Palestinian vowel system include Johnson (1979, 1982), Younes (1982), Card (1983), Herzallah (1990), and Younes (1993). The moderate analysis, represented by Younes (1993), is seen in (22). As seen, Younes (1993) does not recognise mid height for the short vowels. Rtr short vowels and reduced vowels are not mentioned in that study.  (22) Previous Moderate Generative Analysis of the Palestinian Vocalic System (Younes 1993) a. underlying inventory FRONT HIGH I: I MID E : LOW JEl JE  BACK U :  U  0:  surface inventory FRONT HIGH MID LOW  i:  i  e: 331 33  BACK  u: u o:  a: a  Herzallah (1990) argues for a smaller inventory than that in (22). For the YaTbad (northern fellahi) dialect, she does not recognise the underlying long mid vowels, / E : Oil, nor underlying short IUI. For the former, she cites [p. 146] the historical diphthong argument, but does not investigate the issue further.  121  For the latter, she argues that  2.2.2. Vocalic Inventory  historical IUI has merged with III and IMl, and that surface \ u\\s derivable from either HI or IMl by morphological or phonological conditioning. Herzallah's evidence for the lack of underlying short IUI in YaTbad is extensive, and is the focus of chapter 3 of her dissertation. The discussion that follows will review a portion of her evidence. It will be shown that the same evidence is not always found in Abu Shusha. On this basis, it will be concluded that, unlike YaTbad, Abu Shusha has underlying IUI. A first type of evidence which Herzallah discusses is the levelling of historical IUI with HI or IMl in historical CUCV(V)C nouns: YaTbad \ u \ does not occur in such a noun where it did historically.  YaTbad data showing this, from Herzallah (1990:161), are  presented in (23). (Herzallah's transcriptions will be retained in this section. Their surface form status is inferred from her discussion.  A word-final \X\ is included in (23a). In  Arabic in general, a final \X\ is observed for feminine nouns which otherwise end in a vowel. Theoretical issues surrounding this -I t \ will not be examined in this thesis.)  (23) YaTbad Forms a. \ dikkaani(t)\ b. ^zinnaarl-  'shop'(N) 'belt'  (compare Classical Arabic: •{dukkaan^ (compare Classical Arabic: -Izunnaarl-)  The Abu Shusha cognates of the YaTbad forms in (23) are presented in (24).  As  seen, the vowel of interest surfaces as \u\ in Abu Shusha, indicating no levelling of historical IUI in Abu Shusha CUCCV(V)C nouns. (Abu Shusha \u\\s the rtr counterpart  122  2.2.2. Vocalic Inventory  of non-rtr •lul-; this will be shown in §2.2.2.6 and §2.4. Non-rtr vs. rtr variants of the short vowels are not distinguished in Herzallah's transcriptions.)  (24) Abu Shusha Forms a. ^durj.'tjaeinl  (*iditT'tfae:n})  'shop' (N)  b. izun.'na:^  (*izin.'ne:rh  'belt'  A second type of evidence that Herzallah discusses is dorso-pharyngeal phonological conditioning in a/i imperfectives: in Yaibad, stem  occurs in roots with one of  /t s z 5 r x y K7, but not in roots without one of those consonants. (Herzallah describes 'K' as a back velar. It is cognate to Abu Shusha emphatic /k/.) The set of segments just listed is here analysed as the class of segments which are specified for either primary- or secondary-[DOR]; see §2.3.1 for the basis for this analysis. Dorso-pharyngeal conditioning in a/i imperfectives is illustrated by the YaTbad data in (25) , from Herzallah (1990:167,169). In each form in (25), the stem vowel is the finalsyllable vowel.  (25) YaTbad Forms a. •iyinbuYi- --lyunbuyl  'he excels'  b. "lyistubl- --lyustubl-  'he crosses out'  (compare YaTbad: 'lyimlisl- 'he smooths' and -jyiftim}- 'he bewitches', which contain none o f / t s z 5 5 r x  y K / and have a front stem vowel)  123  2.2.2. Vocalic Inventory  In Abu Shusha, V\\lii\ can occur in a/i imperfectives with one of IX. s z 5 5 r %  y  k/,  indicating no general dorso-pharyngeal conditioning in such forms. This is seen from the data in (26). (Uvular 1% ul are the Abu Shusha cognates ofYaTbad fx y/. Surface -jiMs the rtr counterpart of non-rtr i i \.)  (26) Abu Shusha Forms a. i 'b-ib.siU  (*>{ 'b-ib.suU, *i 'b-ub.'sutrt  'he gets happy'  b. Vb-is.nk^  (*Vb-is.rukK *Vb-us.rukK)  'he steals'  A third type of evidence that Herzallah discusses is dorso-pharyngeal phonological conditioning in biliteral roots: in YaTbad, stem \u} occurs in roots with one of IX s z 5 r x y YJ, but not in roots without one of IX s z 5 r x y KA This is seen in the forms in (27), from Herzallah (1990:171).  (27) YaTbad Forms a. -iysuffi-  'he lines up'  b. •iybumml-  'he annexes'  (compare YaTbad: -lybizzr- 'he comes out' and •iybimml' 'he dispraises', which contain none o f / t s z 5 5 r x y K 7 and have a front stem vowel)  In this case, the distribution of Abu Shusha.{u \l\ u \ seems to follow the distribution Herzallah identifies for YaTbad. This is illustrated by the Abu Shusha data (28), which are cognates of the YaTbad forms in (27).  124  2.2.2. Vocalic Inventory  (28) Abu Shusha Forms a. -jbi.-'sum  'he lines up'  b. <!bi.-'5umrrU  'he annexes'  (compare Abu Shusha: •jbi.-'bizzl- 'he squeezes (something) out' and •i bi.-'Simm} 'he dispraises (someone/something)')  Despite thefindingwith respect to biliteral roots, data such as those in (24) and (26) indicate that the contexts of -I u }l\ u} in Abu Shusha are not reducible to those described by Herzallah. That is, {u}/iu} are not always derivable from HI or IMl. Abu Shusha \u\l\u\  On this basis,  are analysed as underlying IUI, and IUI is included in the underlying  vowel inventory in (19). Herzallah's evidence for the lack of IUI in YaTbad indicates that the underlying or non-underlying status of surface { u}/{u I- is a matter of analysis for each Palestinian dialect. Further research should reveal which dialects follow YaTbad in the distribution of -i u \l\ u \ and which do not. Stuart Davis (p.c.) reports that underlying IUI must be recognised for the southern Palestinian dialect of Davis (1993, 1995). The three theoretical studies remaining to be discussed are Johnson (1979, 1982) and Card (1983). Johnson and Card present expansionist variants of (22). They assume three 8  degrees of height for both underlying and surface inventories. For the surface inventory, they recognise some rtr short vowels: Johnson (1982:63) describes IEI and 101 as basically  "The analyses of the vocalic inventory in Johnson (1979) and (1982) are essentially the same.  125  2.2.2. Vocalic Inventory  \i} and {u\, respectively, with "lower allophones next to a pharyngeal or in a final syllable"; Card recognises ii} and iv} as surface variants of underlying /I/ and IUI, respectively. (Both Johnson and Card refer to the rtr vowels as 'lax'. In this thesis, iax' is equated with 'rtr'; see §2.4.2 for discussion.) For the vowels for which they posit rtr surface variants, they do not recognise non-rtr surface variants. That is, Johnson does not recognise non-rtr {e \ and \ o I- as surface variants of IEI and 101, respectively. (Johnson assumes that short HI and IUI always surface, respectively, as •! i} and \ u}.) Card does not recognise non-rtr -i i} and <! u \ as surface variants of HI and IUI, respectively. Finally, Herzallah (1990), Johnson (1979, 1982) and Card (1983) do not mention vowel reduction.  2.2.2.1.3. Summary  In  summary, there has been considerable confusion over just what the  Palestinian/Levantine vowel system is. The uncertainty is not confined to Levantine, as summarised by Norlin (1987:50-51):  A comparison between the studies of Egyptian Arabic and other dialects in the eastern dialect area shows that the phonemic analysis of the short vowel systems is uncertain and surrounded by guarded arguments. It seems as if the short vowel system is in a state of flux and that phonemic oppositions are under development and not yet quite established. Another explanation of the vagueness might be the weakness in many presentations of the various phonological vowel systems in so far that they seldom go into  126  2.2.2. Vocalic Inventory  phonetic details nor present examples of minimal pairs where the contrastive function of the phonemes is obvious.  As a result, the same  dialect can be said to have a different number of short vowels, depending on the author. Many dialects in the neighboring countries seem to have developed more short vowels than the classical three.  Card (1983)  identifies five short vowel phonemes in her investigation of the Palestinian dialect, but does not go into detail. Rice and Said (1960:xx) also recognize five short vowels in the same dialect. None presents minimal pairs.  The following sections (§2.2.2.2 - §2.2.2.7) will present data which support the analysis of the Palestinian vocalic system in (19) and (20). This will include minimal pairs, with the aim of following also the advice of Bouquiaux and Thomas (1992:97): Many studies show only a table summarizing the phonemes of the language with a few supplementary remarks. For us, defining each phoneme is a small problem to be resolved.  Data must be presented and a solution  proposed. This is the only scientifically valid procedure, we feel, since it allows the reader to verify the results. Some arbitrariness is unavoidable, but at least we limit it to the selection of data. A linguist who simply gives a list of phonemes adds the arbitrariness of his interpretations, which are not open to evaluation.  2.2.2.2. Underlying Length  An underlying length distinction in Palestinian is supported by the rninimal/nearminimal pairs in (29). The corpus of this thesis contains no pairs for IE: I vs. / E / and /Oil vs. IOI. However, the mid vowels occur with general lower frequency than the high and  127  2.2.2. Vocalic Inventory  low vowels. It is assumed here that the lack thus far of pairs for IE:I vs. /E/ and IO\l vs. IOI is due to the lower frequency of the mid vowels, and that furtherfieldworkmight yield pairs showing the IE:I vs. IE/ and 10:1 vs. IOI contrasts. Finally, in each pair in (29), it is the underlying vowels that are in contrast, regardless of observed differences in surface vowel quality, e.g., non-rtr W vs. rtr -ill in (29a). This is because those differences are due to the phonology, as will be shown in §2.3 and §2.4.  (29) Data Pairs Showing Underlying Length Distinction a. 11:1 vs. Ill i. Izl:rl  iz\:r\  'large water urn'  ii. / z l r r /  -izirrf  'button'  i. ISJE:\JEml  VTaei.ksrrrt  'world'  ii. nmjEml  VTa.lsrrrt  'flag'  rku:.r^  'forehead'  Vku.r^  'ball'  • •  b. IE:I vs. IE/ (none found)  c. /JE:/vs. IJEI  d. 10:1 vs. IOI (none found)  e. /U:/vs. IUI i.  lkU:r/EI  ii. IkUrJEI  128  2.2.2. Vocalic Inventory  2.2.2.3. Three Underlying Degrees of Height For the Long Vowels  Minimal/near-minimal pairs showing three degrees of height for the underlying long vowels are presented below:  (30) Data Pairs Showing Three-way Height Distinction for the Long Vowels a. fill vs. / E : / i. /tfl:f/  itfhn  'how'  ii. /tfEif/  itfe:n  'mood'  i. /b^ErI:d/  ibs.'mdl  'mail'(N)  ii. IbfiLvM'AI  {b3 .\^\6\  'coolness'  i. /dl:r/  \d\\r\  'to pour'  ii. /dO:r/  -! do:r\  'turn' (as in a game; N)  ibe.'rkd^  'mail'(N)  b. fill vs. IJE\I >  c. /I:/vs. IO\l  d. /I:/vs. /U:/ i. / b i E r I : d /  ii. /bvErU:d/ |b3 .'ru:d|,  >  'gunpowder'  e. IE\I vs. I2E\I i. /dE:r/  •! de:r|-  'Christian parochial school'  ii. /dJEir/  ide:r\  'house'  i. / b E : t /  \be\X\  'home'  ii. /bO:t7  \bo\t\  'shoe'  f. lEil vs. IO\l  129  2.2.2. Vocalic Inventory  g. /E:/vs.  fill/  i. /sE:f/  <se:f\  ii. /sU:f/  isu:f\  summer 'wool'  h. /JE:/ vs. 10:1 lldJEwl  \6v:r\  'house'  ii./dO:r/  ido:ri-  'turn' (as in a game; N)  i. IJE'Jvs.  fU:l  i. / b ^ r ^ E : d / •  ii. / b ^ r U i d /  'coolness'  ^b3 .'re:d^ >  •  •  •  ^b3 .'ru:d^  'gunpowder'  iro:m • • iru:m  'spirit'  >  j./0:/vs. fU:l i. /rO:tV ii. /rU:IV  'to go'  2.2.2.4. Three Underlying Degrees of Height For the Short Vowels  This section presents evidence for underlying short HI vs. IEI vs. IJEI vs. 101 vs. IUI. However, arguments for recognising a phonetic high vowel lowering willfirstbe given. It will be argued that the lowering cases must be discriminated and filtered out before true data pairs showing three degrees of height for the short vowels can be identified. Mid vowels are consistently observed in words that contain a postvelar. illustrated by the forms in (31), in which diphthong.  130  This is  identifies the second half of a phonetic  2.2.2. Vocalic Inventory  (31) a. ['Tie.leim] 'dream' (N)  b. ['Te.cfeeil] 'calf  c.[morr] 'bitter (masc. sg.)'  After Bauer (1926/70:11) and Cantineau (1960:111), the mid height in such words is analysed as the result of lowering in the context of the guttural or emphatic. In lowering forms, the farther the high vowel is from the postvelar, the less the vowel is lowered. This is illustrated by (31a-b): in those forms, while thefirst-syllablevowel is mid [e], the second-syllable vowel is perceptually a short diphthong from mid [e] to high [i]. In other words, in a form containing a postvelar, high vowels are gradiently mid and their degree of mid-ness depends on the degree of proximity between the vowel and the postvelar. In short, the lowering is gradient. Criteria by which a sound property can be identified as either phonetic or phonological were proposed in §1.7.1.  One criterion is non-discreteness: phonetic  properties are non-discrete; phonological properties are discrete. As explained in §1.7.1, gradience follows from non-discreteness. properties, not phonological properties.  Gradience, then, characterises phonetic Because Palestinian high vowel lowering is  gradient, it is here assigned phonetic status. Complete transcriptions of the forms in (31) are presented in (32). These transcriptions encode the phonetic status of the high vowel lowering. (In (32a-b), the Palestinian epenthetic vowel is observed. It is epenthesised for syllabification of the CVCC nouns, as discussed by Abu Salim (1980, 1987a) and Herzallah (1990); see §2.3.3. for further discussion of the epenthetic vowel.)  131  2.2.2. Vocalic Inventory (32) a./hllm/  rfii.lirrrt  [Tie.leim]  'dream'(N)  b. /TIcfel/  VSi.dsiW  ['Te.dssil]  'calf  c. /mUrr/  imurr}  [morr]  'bitter (masc. sg.)',  The lowering in (32) contrasts with the lack of lowering observed in (33).  The lack of  lowering is expected for (33), since neither form contains a postvelar.  (33) a./film/  Vfilim^  ['fi.lim]  (*['fe.lim], *['fi.lem],  'movie'  ['fe.leim], *['fei.lim], etc.) b./flcfel/  Vfi.cfem  ['fi.cfeil]  (*['fe.d5il], *['fi.05el],  'radishes'  ['fe.cfeeil], *['fei.<feil], etc.)  Support for a phonetic analysis of the postvelar-induced lowering comesfromthe fact that, to my knowledge, there is no evidence that the lowered height is phonologically visible, that is, referred to in the phonology.  By the criteria outlined in §1.7.1,  phonological properties are phonologically visible, but phonetic properties are not. Phonetic status for Palestinian high vowel lowering means that certain forms that appear to show three degrees of height for the short vowels must be disregarded. Examples are presented in (34).  132  2.2.2. Vocalic Inventory  (34) Faux Amis for Three-way Three-way Height Distinction for the Short Vowels a. i. /IrUEff/  b.  a'ftaffl  [I.Tiaff]  'heavy cotton cover'  ii./hUff/  \T\uff}  [tioff]  'barefoot'  LfhJEW  \X\a\W  [fiall]  'it (masc.) bled'(as a dye bleeds)  ii./Mil/  \X\i\W  [ftcll]  'to bleed'(as a dye bleeds)  The pairs in (34) are labelled 'faux amis' because in each mid-vowel form, (34a.ii) and (34b.ii), the mid height can be attributed to the phonetic lowering just discussed. Underlying mid height is not impossible in forms like each (ii) datum, above. However, if it were underlying, it could not be established as such because it would be phonetically neutralised with a phonetically lowered height in each form.  Thus, no mid height is  posited as underlying in (34) because there is no clear evidence that it is underlying. Since the mid height can be ascribed to the phonetics, whether or not it is really underlying in each form is untestable. The point here is that mid height cannot be established on the basis of forms that contain a postvelar. However, true contrasts exist. A true contrast is one in which the form with the mid vowel'lacks a postvelar consonant, that is, contains no phonetic source for the mid height. Those provided by the present corpus are seen in (35).  For (35a), the vowels being  contrasted are the initial-syllable vowels. Datum (35j.i) is a loanfromEnglish. As seen, no pairs for /E/ vs. 101 and IEI vs. /U/ were found. This is considered coincidental; it is presumed that further fieldwork would yield minimal pairs for those contrasts also.  2.2.2. Vocalic Inventory  (35) Data Pairs Showing Three-way Height Distinction for the Short Vowels a. HI vs. IE! i. /silk/  rsi.lik^  wire  ii. /sElEk/  rse.lsk^  'boiled (masc. sg.)' (Adj)  b. HJvs./JEI i. /-I/  (1 sg. obj.)  as in, e.g.: /Ism-I/  V?is.m-i^  'my name'  ii. /-JE/  (3 masc. sg. obj.)  as in, e.g.: Hsm-JEI  •I 'Tis.m-al-  'his name'  i. /HbLE/  Hi.bi.^  'Lybia'  ii. /lObLE/  Vlo.bi.9l  (a type of small pea)  c. HI vs. 101  d. IV vs. HJI i. /hi/  'she'  ii. /hU/  'he'  e. IE/ vs. /JE/ i. /sElEk/  rse.lek^  'boiled (masc. sg.)' (Adj)  ii. ZsJElJEk/  Vsa.lsk^  'he boiled'  f./E/vs./O/ (none found)  g. Hit vs. HJI (none found)  134  2.2.2. Vocalic Inventory  h.AE/vs./O/ (3 masc. sg. obj.)  i. I-Ml as in, e.g.:  /XMU-JE/  rxai.-t-s^  'his maternal aunt'  ii. I-OI  (endearment suffix)  as in, e.g.:  /X^:l-t-0/  Vxal.-t-o^  'maternal auntie'  rka.r^  'small boat'  Vku.ra^  'ball'  i. /kOrdin/  Vko.r3rrt  'corners'  ii. /kUrflin/  Vku.r3n|'  'horns'  i.AE/vs./U/ i. / k M / •  ii. /kUr^E/  j. /O/vs. /U/  There are far fewer forms motivating the contrasts for the mid short vowels than for the high and low short vowels. That is, more pairs could be listed for the high and low vowels than for the mid vowels. However, the existence of (35a,c,e,hj) is here considered to indicate that there is underlying mid height for the short vowels, too.  2.2.2.5. Reduction of/JE/  Underlying short IMl surfaces mid and central when not under primary lexical stress. This is illustrated by the data in (36). The alternation of interest is shown by the steminitial low vowel in (36a) compared to the same vowel in (36b). It is also shown by the  135  2.2.2. Vocalic Inventory  stem-final vowel in (36a) compared to the same vowel in (36b). In (36b), the surface length of stem-final /JE/ is the result of lengthening under shifted stress. The low surface quality of the lengthened vowel shows that it is underlyingly /JE/. Finally, reduced /JE/ surfaces as one of three mid central surface vowels: non-rtr ia}, rtr non-back i3}, rtr back i3 t. >[  These variants arise through pharyngealisation and uvularisation harmonies. This  will be shown in §2.2.2.6 and §2.4.  (36) a. /xJEUJE/ b. /mJE-xJEWJE-^/  i 'xal.la*t  'he left (something)'  i m3.-x3l.'lce:-J!  'he didn't leave (something)'  Short /JE/ reduction is also seen in (37).  Specifically, it is shown by the vowel in the  3 fern. sg. obj. suffix, /-hJE/, in (37a) compared to the same vowel in (37b).  (37) a. / [ U i f - t - h ^ / b. / i n E - J U : f - t - h i E - J /  i 'suf-t.-hs}  T saw her'  1 m3.- uf-t.-'hae:-^  T didn'