Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Towards a phonetic and phonological typology of post-velar articulation Bessell, Nicola J. 1993

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

Item Metadata


831-ubc_1993_spring_phd_bessell_nicola.pdf [ 18.18MB ]
JSON: 831-1.0086353.json
JSON-LD: 831-1.0086353-ld.json
RDF/XML (Pretty): 831-1.0086353-rdf.xml
RDF/JSON: 831-1.0086353-rdf.json
Turtle: 831-1.0086353-turtle.txt
N-Triples: 831-1.0086353-rdf-ntriples.txt
Original Record: 831-1.0086353-source.json
Full Text

Full Text

TOWARDS A PHONETIC AND PHONOLOGICAL TYPOLOGY OF POST-VELAR ARTICULATION  by NICOLA JANE BESSELL B.A. Hons., Memorial University of Newfoundland, 1981 B.A. Hons., University of Oxford, 1983 M.St., University of Oxford, 1984 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 December 1992  © Nicola Jane Bessell, 1992  In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission.  (Signature)  Department of  ^wa-c3  The University of Britis F Columbia Vancouver, Canada '  Date ^  DE-6 (2/88)  Abstract This dissertation develops a typology of post-velar articulation from the point of view of available inventory, phonetic and phonological studies. The database on which such typologies can draw is expanded by the examination of data from the Interior Salish languages of the Pacific Northwest. The post-velar inventory of Interior Salish is examined acoustically in order to place it within the phonetic typology of post-velars as understood from work on Semitic and Caucasian. Pharyngeals from six Interior Salish languages are examined to determine the range of variation. The basic finding of this acoustic work is that the Interior Salish post-velars are commensurate with what is known about post-velars based on Semitic data and articulatory modelling. Interior Salish phonological data support the extension of articulator-based feature geometry to a fourth node, here termed Tongue Root. It is shown that the fourth node is required to class Interior Salish faucals and accommodate their participation in harmony processes. Furthermore, constraints on the phonology of the fourth node in Interior Salish suggest that we are dealing with an Advanced Tongue Root phenomenon such as found in some African vowel harmonies. This is an encouraging result in the sense that it confirms the existence of Tongue Root consonants and does not confine the fourth node to vowels. The analysis of Interior Salish laryngeals without the fourth node that characterizes their Semitic counterparts corroborates our understanding of laryngeals as lacking Place specifications in the default case. Furthermore, it is argued that the descriptivist and Dependency Phonology view of /2, h/ as minimal stop and fricative is phonologically appropriate. Evidence from epenthesis, laryngeal transparency and debuccalization support the analysis of ii, h/ as (+consonantal, -sonorant, +/-continuant]. Debuccalization and epenthesis processes also suggest that h/ do not necessarily bear Laryngeal Node features. It is argued that unless phonemic phonation features are present in an inventory, there is no need for /2, h/ to bear [constricted glottis, spread glottis]. It is noted that the phonology of post-velars in Interior Salish contrasts with their patterning in Semitic (McCarthy 1991) and Nisgha (Shaw 1991b), specifically with respect to the representation of /2, h/. The presence of a fourth node in Interior Salish does not require that laryngeals be dependents of it. The same finding is reported by McCarthy (1991) for Semitic. Given that there is no acoustic evidence at present to suggest that we are dealing with distinct phonetic entities, it appears that the representation of laryngeals in languages with a fourth node must be stipulated.  Table of Contents Abstract^ List of Tables^ List of Figures ^ Acknowledgments^  xii xv  Chapter One: Introduction 1.0 Introduction^  1  1.1 Aims of dissertation^  4  1.2 Database^  6  1.3 Outline of dissertation^  9  Chapter Two: Fourth node history 2.0 Introduction^  12  2.1 Early ATR assumptions ^  12  2.2 Traditional representations ^  13  2.3 Current proposals^  15  2.3.1 Fourth node in Semitic^  16  2.3.2 Fourth node in other languages ^  23  2.3.3 Fourth node for secondary articulations ^ 25 2.4 Conclusions^  24  Chapter Three: Systemic typology of post-velars 3.0 Introduction ^  27  3.1 Uvulars^  28  3.2 Pharyngeals ^  29  iv 3.2.1 Pharyngealized consonants ^  34  3.2.2 Pharyngealized vowels ^  35  3.3 Glottals ^  35  3.3.1 Stop and fricative?^  40  3.3.2 Secondary articulations on glottals ^  43 Rounding on glottals ^  43 Prenasalization on glottals ^  44 Palatalization on glottals ^  45  3.4 Aspiration and ejection ^  47  3.5 Conclusions^  49  Chapter Four: Phonetic investigations into post-velar articulation 4.0 Introduction^  50  4.1 The anatomy of post-velars ^  50  4.1.1 The major cavities ^  51  4.1.2 Places of articulation ^  51  4.1.3 Articulators^  52 The tongue^  52 The pharynx ^  55 The larynx^  57  4.2 Articulatory descriptions of post-velars ^  59  4.2.1 Uvular stops^  59  4.2.2 Uvular fricatives ^  60  4.2.3 Uvular trill^  60  4.2.4 Pharyngeal stop ^  60  4.2.5 Pharyngeal fricatives/approximants ^  60  4.2.6 Epiglottal stop^  61  4.2.7 Epiglottal fricatives^  62  4.2.8 Glottal stop^  64  4.2.9 Glottal fricatives^  64  4.2.10 Summary^  65  4.3 Natural language data ^ 4.3.1 Semitic^  67 67 Dorsals^  69 Pharyngeals^  71 Emphatics ^  75 Glottals^  81 Vowel effects^  81  4.3.2 Caucasian^  85 Dorsals^  85 Pharyngeals^  85 Pharyngealized consonants and vowels ^ 86 Glottals^  88 Vowel effects^  89  4.3.3 Interior Salish ^  91 Dorsals^  91 Pharyngeals ^  93 Retracted coronals^  97 Glottals^  99  4.3.4 Khoisan^  99  4.3.5 Nootka ^  100  4.3.6 Hthda^  102  4.4 Acoustic consequences of post-velar articulation ^ 4.4.1 Uvulars and pharyngeals^  102 103  vi  4.4.2 Emphatics 4.4.3 Glottals 4.5 Summary and conclusions  ^ ^ ^  109 114 115  Chapter Five: Interior Salish Phonetics 5.0 From acoustics to place of articulation ^  118  5.1 The acoustic theory of speech production ^  119  5.2 Moses-Columbia Salish: A case-study in post-velar articulation^ 123 5.2.1 Methods^  124  5.2.2 Moses-Columbia: an overview ^  126  5.2.3 Stressed vowels before coronals ^  132  5.2 4 Stressed vowels before retracted coronals ^ 133 5.2.5 Stressed vowels before velars ^  134  5.2.6 Stressed vowels before uvulars ^  135  5.2.7 Stressed vowels before pharyngeals ^  138  5.2.8 Stressed vowels before glottals ^  140  5.2.9 Bilabials and Summary^  142  5.3 Pharyngeal Articulation^  145  5.3.1 Arabic pharyngeals^  147  5.3.2 Interior Salish pharyngeal articulation ^  153 Kalispel-Spokane^  153 Colville-Okanagan^  159 Coeur d'Alene ^  162 Shuswap^  167 Me?kepmxcin ^  169 Moses-Columbian ^  173 Voiced pharyngeals^  174  vii Voiceless pharyngeals ^ 177 5.4. Conclusion^  187  Chapter Six: Post-velar phonology 6.0 Introduction and overview^  190  6.1 Semitic^  194  6.1.1 Semitic gutturals ^  194  6.1.2 Semitic emphatics ^  199  6.2 Interior Salish^ 6.2.1 Evidence for uvulars as Dorsals^  204 205  6.2.2 Evidence for uvulars and pharyngeals as Tongue Root ^ 209 6.2.3 Evidence for pharyngeals as Tongue Root resonants ^ 212 6.2.4 Coeur d'Alene^  223 Morpheme structure constraints^ 229 Root patterns ^  230 Regressive Faucal Harmony ^  232 Regressive Faucal Harmony formalized ^ 239 6.2.5 Regressive Faucal Harmony in other Interior Salish languages ^ 245 Spokane Regressive Faucal Harmony^ 247 Kalispel Regressive Faucal Harmony ^ 249 6.2.6 Progressive Pharyngeal Harmony ^  250 Type-i: Colville Progressive Pharyngeal Harmony ^ 253 Type-ii: Spokane and Kalispel Progressive Pharyngeal Harmony ^  256 Type-iii: Coeur d'Alene, Shuswap, Lillooet ^ 261 Coeur d'Alene ^  263 Shuswap^  274  viii Lillooet^  277 Type-iv: 1•11e2kepmxcin and Moses-Columbian ^ 278 Type-v: Retraction of consonants ^  291 Summary of Salish^  294  6.3 African ATR systems ^  300  6.4 An alternative: [-high, +back] ^  305  6.5 Conclusions^  312  Chapter Seven: Glottal phonology 7.0 Introduction^  316  7.1 The Sound Pattern of English^  320  7.1.1 Use of [+low] ^  324  7.1.2 Glottals as [-consonantal, +sonorant] glides ^ 332 7.2 Malay^  338  7.2.1 Hiatus^  339  7.3.2 Transparency^  341  7.3 Debuccalization^  347  7.3.1 Stop > [2]; Fricative > [h] ^  348  7.3.2 Stop > [h] ^  354  7.4 fl, hi, Aspiration and Ejection^ 7.5 Glottals and continuancy^  360 363  7.6 Underspecification and licensing ^  367  7.7 Conclusions^  371  Chapter Eight: Conclusions and future research ^ 374 Bibliography and references ^  376  ix  Appendices Appendix A: Interior Salish inventories ^  405  Appendix B: Interior Salish spectrograms  409  List of Tables Chapter Three Table 3.1: Distribution of uvulars^  29  Table 3.2: Distribution of pharyngeals with other post-velars ^ 30 Table 3.3: Distribution of pharyngeal fricatives ^  32  Table 3.4: Distribution of glottals^  36  Table 3.5: Distribution of rounding ^  44  Table 3.6: Distribution of palatalization ^  46  Table 3.7: Distribution of aspiration and ejection ^  48  Chapter Four Table 4.8: Arabic dorsals^  70  Table 4.9: Arabic pharyngeals^  73  Table 4.10: Arabic emphatics ^  80  Table 4.11: Arabic vowels effects ^  83  Table 4.12: Interior Salish uvular effects^  92  Table 4.13: Interior Salish pharyngeal effects ^  96  Table 4.14: Interior Salish retracted coronal effects ^  98  Table 4.15: Kyoquot vowel allophony (Rose 1981) ^  101  Table 4.16: Klatt and Steven (1969): Constriction area 0.05 and 0.1 cm2 ^ 104 Table 4.17: Alwan (1986): Constriction area 0.15 cm2, length of constriction 1 cm. 105 Table 4.18: Arabic post-velars (Alwan 1986) ^  107  Table 4.19: Arabic post-velars (Ghazeli 1977) ^  107  Table 4.20: Iraqi Arabic pharyngeals and glottals (Butcher and Ahmad 1987) ^ 108 Table 4.21: Sudanese Arabic pharyngeals and reference vowels (Adamson 1981) 109 Table 4.22: Alexandrian emphatic effects (al-Ani and el-Dalee 1984) ^ 110  xi Table 4.23: Egyptian Arabic emphatic effects (Norlin 1987) ^ 111 Table 4.24: Consonantal loci in Iraqi Arabic (Giannini and Pettorino 1982) ^ 112  Chapter Five Table 5.25: Interior Salish Database ^  126  Table 5.26: Summary of Nxa'amxcin articulatory effects: MM and JM^ 128 Table 5.27: Plain, rounded, voiced and voiceless pharyngeal effects: MM ^ 139 Table 5.28: Ms-Cm place of articulation effects summarized: offset values ^ 145 Table 5.29: Formant values of pharyngeals examined in text ^ 146  Chapter Six Table 6.30: Guttural and faucal classes ^  191  Table 6.31: Effects of post-velars on vowels (Kinkade and Thompson 1974) ^ 220 Table 6.32: Interior Salish Regressive Harmony onto roots or suffixes ^ 246 Table 6.33: Interior Salish Progressive Pharyngeal Harmony ^ 252  xi i  List of Figures Chapter Four Figure 4.1: Tongue musculature (Laver 1980) Figure 4.2: Velopharyngeal muscles (Laver 1980) Figure 4.3: Iraqi Arabic It!, It/ and /q/ (Giannini and Pettorino 1982) Figure 4.4: Tunisian Arabic: /g,x/ and /t,t/ (Ghazeli (1977)  ^ ^ ^ ^  55 57  77  78  Figure 4.5: Tunisian Arabic: /S/ before initiation (solid line) and during production (dotted line) in [Tx:li]. (Ghazeli (1977)^ Figure 4.6: Udi V1 (Catford !983) ^  79 88  Figure 4.7: Palestinian Arabic short vowels: plain and emphatic ^ 113 Figure 4.8: Palestinian Arabic long vowels: plain and emphatic^ 113  Chapter Five Figure 5.9: Vocal tract with standing waves^  121  Figure 5.10: Ms-Cm Vowels at all POA, Stressed and Unstressed ^ 130 Figure 5.11: Ms-Cm Stressed Vowels^  131  Figure 5.12: Ms-Cm Vowels before Coronals ^  132  Figure 5.13: Ms-Cm Vowels before Retracted Coronals ^ 133 Figure 5.14: Ms-Cm Vowels before Velars ^  134  Figure 5.15: Ms-Cm Vowels before Uvulars^  135  Figure 5.16: Ms-Cm Vowels before Pharyngeals ^  138  Figure 5.17: Ms-Cm Vowels before Glottals ^  140  Figure 5.18: Uvular effects: ciqn 'digging' (Ms-Cm: MM) ^ 141 Figure 5.19: Glottal transparency: (kn) p'i?q 'I'm cooked, burnt' (Ms-Cm: MM) ^ 142 Figure 5.20: Ms-Cm labial effects: yaT'p(qin) 'many, lots'.: Speaker: ED ^ 143 Figure 5.21: Cadu 'enemy' (Colloquial Egyptian Arabic: SW) ^ 148  Figure 5.22: ?aiwa 'yes' (Colloquial Egyptian Arabic: SW) ^ 150 Figure 5.23: sali:d 'happy' (Colloquial Egyptian Arabic: SW) ^ 151 Figure 5.24: halab 'he milked' (Colloquial Egyptian Arabic: SW) ^ 152 Figure 5.25: liast 'char' (Chewelah/Kalispel: AS) ^  154  Figure 5.26: ?ay kwast 'tomorrow' (Moses-Columbian:MM) ^ 155 Figure 5.27: Ciast 'char' (Spokane: MS)^  156  Figure 5.28: Simt 'get angry' (Spokane: PF) ^  157  Figure 5.29: p'aS 'burn' (Spokane: PF) ^  158  Figure 5.30: hec yall'mi 'people are gathering' (Spokane: PF)^ 159 Figure 5.31: Sitmn 'teeth' (Colville: CQ) ^  160  Figure 5.32: pal 'grey' (Colville: CQ) ^  161  Figure 5.33: xwuyt ciyalp 'they have arrived' (Colville: COD ^ 162 Figure 5.34: ?ar 'much, plenty' (Coeur d'Alene: LN) ^ 164 Figure 5.35: cic lay' 'I'm angry' (Coeur d'Alene: LN) ^ 165 Figure 5.36: mel'wns 'he broke it' (Coeur d'Alene: LN)^ 166 Figure 5.37: Tipkn 'I'm angry' (Shuswap:BD) ^  167  Figure 5.38: plpelt 'grey' (Shuswap:BD) ^  168  Figure 5.39:^'dragging things around' (Shuswap:BD) ^ 169 Figure 5.40: Sis 'shrink' (Nie?kepmxcin: DS) ^  170  Figure 5.41: calp 'ripped' (Nle?kepmxcin: DS)^  171  Figure 5.42: pal 'bleached by the sun; grey' (Nte?epmxcin: DS) ^ 173 Figure 5.43: l'1' 'brighe (Moses-Columbian: AB) ^  175  Figure 5.44: nxwlank 'cave; hole in a hill' (Moses-Columbian: AB) ^ 176 Figure 5.45: scyaMmix 'they're gathering' (Moses-Columbian: AB) ^ 177 Figure 5.46: himt 'angry' (Moses-Columbian: AB) ^  178  Figure 5.47: ph 'grey' (Moses-Columbian: AB) ^  179  Figure 5.48: kihana 'teenage girl' (Moses-Columbian: AB) ^ 180  xiv  Figure 5.49: ?dhwa? 'cough' (Moses-Columbian: AB) ^ 181 Figure 5.50: ha2lcw61"Are you warm enough?' (Moses-Columbian: AB) ^ 183 Figure 5.51: pax(paxt) 'wise' (Moses-Columbian: AB) ^ 184 Figure 5.52: CEA /h/ spectra at initiation (top) and midpoint (bottom). ^ 185 Figure 5.53: Moses-Columbian [hi spectra at initiation (top) and midpoint(bottom) 186 Figure 5.54: Moses-Columbian [h]: spectra at midpoint ^ 187 Figure 5.55: Moses-Columbian [x]: spectra at midpoint.^ 187  Chapter Six Figure 6.56: Coeur d'Alene vowels : stressed and unstressed Figure 6.57: Coeur d'Alene Stressed Vowels Figure 6.58: Coeur d'Alene vowels:means  ^ ^ ^  226 227 299  Chapter Seven Figure 7.59: [skixzeh] 'mother' (1•1101cepmxcin: DS) ^ 336  XV  Acknowledgments I thank my committee members, Patricia A. Shaw (Chair and Supervisor), Ewa Czaykowska-Higgins, M.Dale Kinkade and Mark Y. Liberman. This dissertation owes a particular debt to Patricia Shaw, who has always insisted that I think carefully and has not tolerated cavalier phonology. Her example as a fine phonologist with unusual respect for fieldwork has obviously influenced the orientation and execution of this dissertation. Ewa Czaykowska-Higgins supplied countless hours of official and unofficial consultation on Salish and general linguistic issues, consultation and company on fieldwork, and a critical eye not deterred by friendship. Her generous, disciplined and detailed attention to this dissertation has improved it immeasurably. I also thank M. Dale Kinkade who (along with Mandy Na?zinek Jimmie) first introduced me to the initially alarming wonders of Interior Salish and encouraged phonetic work on the languages. This dissertation has clearly benefitted from the depth of his unique scholarship and his willingness to share it. I am grateful to Mark Liberman for generously supporting my research in innumerable practical and psychological ways, not the least of which was granting me access to the Phonetics Lab at the University of Pennsylvania and welcoming me to an exciting and exacting research community. I owe a major debt to the Salish speakers whose data are represented here. Obviously this would be a very different dissertation without their input and we would all know less about Interior Salish phonetics and phonology as a consequence. I especially thank Agatha Bart, Elizabeth Davis, Mandy Nahinek Jimmie and Mary Marchand for friendship and encouragement as well as consultation. I also thank Lavinia Clark, Basile DeNeau, Pauline Flett, Nora Jimmie, Blanche LaSartre, Lawrence Nicodemus, Charlie Quintasket, David Shooter, Margaret Stensgar and Jimmie C'al' Toodlican. I thank Barry Carlson for providing recordings of Margaret Sherwood and Alex Sherwood, M.D. Kinkade for recordings of Jerome Miller, Linda Walsh for recordings of Bertha Azak (Nisgha), Ivy Doak for assistance with Coeur d'Alene and Dwight Gardener for assistance with Shuswap. I also thank Andre-Pierre Benguerel for commentary on phonetic work, advice, and general interest in my work. Bruce Bagemihl and Keren Rice also commented on early versions of some chapters. For encouragement and professional advice I thank Ivy Doak, Douglas Pulleyblank, Edwin Pulleyblank and Mike Rochemont, as well as friends and colleagues in Vancouver, Philadelphia and elsewhere: Helmi Braches, Gene Buckley, Henry Davis, lain Higgins, Katherine Hunt, Mandy Nahinek Jimmie, Bill Reynolds, Mark Steeciman, Yan Feng Qu, Bonnie Webber, Christine Zeller and Katya Zubritskaya. I especially thank Bruce Bagemihl and Mary Peckham for their steady friendship. I am grateful to Carmen da Silva for administration at UBC. The fieldwork on which much of this dissertation is based was made possible by grants from the Phillips Fund to Nicola Besse11, and Jacobs Fund and SSHRCC grants to Ewa Czaykowska-Higgins. I gratefully acknowledge all sources of funding. My final acknowledgement goes to my family, and to Gregory Provan for patience and wisdom that exceed my own.  1  Chapter One: Introduction 1.0 Introduction There has been considerable research activity since Clements' (1985) article "The geometry of phonological features" on the restructuring of phonological feature theory from flat or linear feature representations to hierarchical ones. A major advance is Sagey (1986), who develops a model of feature geometry grouping some of the terminal features of Clements (1985) into articulator node constituents. Sagey (1986) argues for three such articulator nodes: Labial, Coronal and Dorsal, represented here under an organizing node, termed Place. (1) Sagey (1986) ROOT Laryngeal^ [consonantal] [?tr. gl]^ cont] spr. gl.] stiff v.c.]^Supralaryngeal [slack v.c] Place  Soft Palate [nalall^Labial / [rd]  1 Coronal  [ant] [lat]  Dorsal 1^ [distr]^[low]^/ \ [high] [back]  Ladefoged and Maddieson (1988) and Ladefoged and Halle (1988) propose the addition of a fourth articulator node 'Tongue Root' dominating a feature [ATR] (or [Advanced Tongue Root]). The resulting geometry under Place is as in (2).  2 (2) Cole (1987), Sagey (1988) Place Labial  I [round]  Coronal  [distributed] Dorsal [anterior] [lateral] [high] [back] [low]  Tongue Root  It was originally suggested that the fourth node is needed to account for tongue root harmonies as found in the vowel systems of some Nilo-Saharan and NigerKordofanian languages. Its restriction to vowel representation is odd though, particularly in light of recent work examining the phonological interaction between consonant and vowel features under all articulator nodes (Gorecka 1989, Clements 1991, E. Pulleyblank 1989, 1992, Shaw 1991a,b, Hume 1992). However, Cole (1987) uses a fourth node dominating a feature [±ATR] for her account of the Coeur d'Alene post-velars and Hayward and Hayward (1989) and McCarthy (1991) propose that the Semitic 'guttural' class which is composed of uvulars, pharyngeals and glottals, also requires a fourth node. McCarthy (1991) departs from the model in (2) in several interesting ways. He proposes that articulator geometry be complemented by place-of-articulation geometry so that the nodes under Place now designate places of articulation rather than articulators. This reflects a phonological division of the vocal tract into Oral and Pharyngeal places of articulation. The original three articulators of Sagey (1986) fall under the Oral place, the fourth node Pharyngeal falls under the Pharyngeal place.  3 (3) McCarthy (1991) Place 0 Oral  Pharyngeal [pharynge ]^\ [radical]  Labial / [rnd]^ Coronal Dorsal  [ant] [lat]  [distr [back]  Our most developed understanding of the phonetics and phonology of post-velar consonantal articulation comes from Semitic, which plays a dominant role in extensions of feature geometry to a fourth node.1 This limitation invites several obvious questions: What happens in other languages with similar post-velar inventories (i.e. with uvulars, pharyngeals and glottals); Do these languages support the proposed modifications to feature geometry and/or the universality of these proposals; What is the cross-linguistic typology of post-velar phonetics and phonology? Here I must clarify some terminology. The class which I term post-velars consists of uvulars, pharyngeals and glottals. Uvulars are generally made at the very rear of the soft palate, where the oral and pharyngeal cavities intersect. Pharyngeals seem to vary somewhat in precise location within the pharynx, but constriction narrows the pharyngeal cavity rather than the oral one. Glottals are made by vocal fold gestures within the larynx. I use the term 'post-velar' as a neutral descriptive term for this class of segments, referring to their general place of articulation. The term 'pre-uvular' will designate lAlthough Cole (1987) proposes a fourth node for Coeur d'Alene, she does not attempt to justify her innovation beyond Coeur d'Alene harmonies.  4 articulation anterior to the intersection of the oral and pharyngeal cavities. Neither of these terms is common, but their designations are fairly apparent. A neutral descriptive term is necessary since it is not the case that all post-velars are gutturals in the sense established by Semitic phonology. 1.1 Aims of dissertation This dissertation addresses the general issues raised by current proposals for a fourth node and evaluates data relevant to an inquiry into the phonetics and phonology of post-velar articulations. It does so in part by laying the groundwork for typologies of post-velar articulation in three areas: gross systemic, phonetic and phonological. It is argued that typological work in each of these domains contributes to our understanding of post-velar activity in a larger context. A gross systemic typology of post-velar articulation is developed on the basis of the distribution of post-velar consonants in the 693 inventories contained in Ruhien (1975). This dissertation also expands the post-velar database by examining data from the Interior Salish languages of the Pacific Northwest. These languages have extensive post-velar inventories which have not been examined acoustically. Given the rarity of pharyngeal articulation in general, it is important to establish the phonetic composition of the Salish post-velar class. Developing a phonetic typology enables the Salish data to be compared to Semitic and Caucasian post-velars. The Interior Salish post-velars are also phonologically active as a class. Their patterning, however, explicitly excludes laryngeals. This contrasts with our understanding of the Semitic guttural class, which includes laryngeals and so motivates the Pharyngeal node in (3) as a place of articulation node. Given the disparity between Salish and SemiticlNisgha post-velar phonology, the proper phonetic identification of these sounds in each language group is important. Are we in fact dealing with the 'same'  5 segments? These issues are critical to the proper representation of post-velars and the general development of an adequate feature geometry. Within the overall research paradigm pursued here, the major questions addressed are the following: (i) What exactly are the phonetic facts of post-velar articulation? Phonetic descriptions of post-velar articulation are bewildering in their variety. More specifically, considerable confusion arises over: • The involvement (or lack thereof) of the tongue root in uvular, pharyngeal, glottal and emphatic articulation. Descriptions vary as to the contribution of tongue root activity to some and/Or all of these articulations. • The interaction of tongue root activity with constriction of the pharyngeal walls and/or faucal pillars to achieve post-velar articulation or emphatic/retraction effects. • The role of the larynx in pharyngeal articulation and the role of the pharynx in laryngeal articulation. Does the frequent laryngealization of pharyngeals, for example, suggest a relationship between these two segments which is beyond the phonetic? Are there in fact two types of laryngeals—those which pattern with pharyngeals (as in Semitic) and those which do not (as in Salish)? (ii) How do the phonetics and phonology of post-velar articulation in Salish and Semitic compare with what is known about African ATR systems? This question is relevant since the fourth node was originally proposed for ATR systems.  6 (iii) What kind of feature geometry is necessary to capture the phonological patterning of post-velar articulation in Semitic and Salish, while still accommodating the facts of African ATR systems? 1.2 Database The answers to these questions are not immediately apparent for several reasons. First of all, both uvulars and pharyngeals are quite rare cross-linguistically. While labial, coronal and velar consonants are found in virtually every language, uvulars occur in 21% of Ruhlen's (1975) sample of 693 languages and 14.8 % of Maddieson's (1984) sample of 317 languages. Pharyngeals are rarer still. They occur in 7% of Ruhlen's (1975) sample, and 4% of Maddieson's (1984) sample. It turns out that the effect of the putative pharyngeal node/site for consonants can be explored only in three languages groups. Apart from the Afro-Asiatic data examined by Hayward and Hayward (1989) and McCarthy (1991), post-velar inventories which include pharyngeals are known from languages spoken in two other areas of the world: the Caucasus mountains of the former Soviet Union, and the Northwest Coast of the North American Continent. Many of the NW and NE Caucasian languages have inventories of post-velars that surpass those found in Semitic (see Colarusso 1988). Of the wealth of indigenous languages of NW North America, the combination of uvulars, pharyngeals, glottals and retracted coronals appears exclusively within the Interior branch of Salish. Outside Salish, phonemic pharyngeals are found in the Wakashan language Nootka, some Northern dialects of Haida (a language isolate), and in the Stoney dialect of Dakota (Siouan). In both Nootka (Jacobsen 1969) and Haida (Krauss 1979, Levine 1981) the pharyngeals are derived historically from uvulars, but the Stoney pharyngeals (h, C) are derived from what are analysed as phonemic velar fricatives (Shaw 1980). A second problem is that the rich tradition of phonetic and phonological investigation in Semitic is not always matched in other language groups of interest. As  7 McCarthy (1991) comments, very little is known about the phonology of the Caucasian languages, despite their obvious importance. A considerable body of writing in Russian and other Soviet languages on Caucasian exists but much of this material is not easily accessible to readers in the West. The most notable exception to this generalization is Colarusso's thesis (1975, 1988), written in English and detailing a vast spectrum of primarily phonetic and comparative information on the Northwest Caucasian languages. As for the languages of the Pacific Northwest, Krauss (1979:895) remarks that the level of description required for valid phonological analysis is not yet achieved in work on Haida. For Nootka, there is the work of Sapir (1911, 1938) and Sapir and Swadesh (1939). The phonologies of the Wakashan languages in general are intriguing to say the least (see Lincoln and Rath 1986 for Haisla, Rose 1981, Stonham 1990 for Nootka). However, very little work has focussed specifically on the phonological behaviour of Nootkan pharyngeals. There is an obvious need for fieldwork to concentrate on this aspect of Nootka. This leaves Salish. Within the Interior branch of Salish readily accessible research material is scanty, with the fruits of many years' laborious fieldwork often confined to theses or regionally distributed grammars and dictionaries. There are, however, numerous articles on aspects of these languages to be found, particularly in the International Journal of American Linguistics (IJAL). Within the compass of what is available on Interior Salish, much can  be learned about post-velar patterning from these languages. Reichard's (1938) grammar of Coeur d'Alene identifies a class of segments with post-velar articulation which conditions a rule of regressive vowel-lowering across morpheme boundaries. Of immediate interest is the fact that 12, hi are not members of this class, although uvulars and pharyngeals are. This finding is corroborated by Sloat (1966, 1975, 1980) and Doak (1992). Kinkade (1967) identifies the presence of pharyngeal resonants throughout Interior Salish and includes them in a class of 'back' consonants which historically conditioned lowering of preceding vowel quality (Kinkade and Sloat 1972, Kinkade and  8 Thompson 1974). Mattina (1979) isolates a rule of Pharyngeal Movement in certain roots which subsequently lowers stressed vowels in suffixes. Although Pharyngeal Movement proper is proposed for Colville, the presence of cognate 'retracting roots' containing vowels of a lowered/dark/retracted timbre has been an issue in Salish studies for some time. Clearly, Interior Salish critically affects issues of post-velar phonology. Unfortunately, almost no acoustic work exists on the large and very interesting Salish inventories. There is a general lack of systematic phonetic information on most of the indigenous languages of the region, although there are a number of comprehensive phonological and morphological accounts available. This dissertation focusses on Interior Salish data, given its obvious relevance to the theoretical issues sketched above and the acute need for continued documentation of these languages. Salish is spoken on the Pacific northwest coast of the North American continent from the north Oregon coast as far north as Bella Coola in British Columbia, though not in an unbroken line, as Bella Coola is suffounded by Wakashan languages to the north and south and by Athabaskan languages to the west. From this coast, Salish is spoken as far east as western Montana. Much of what is now Washington State was peopled by Salish speakers at the time of contact with Europeans, their presence extending across the Canada-United States border to encompass vast areas of southern British Columbia. Thompson (1979) divides the family into three divisions: Coastal, Tsamosan and Interior. 2 The Interior languages with which this dissertation is primarily concerned are divided into Northern and Southern groups, the constituency of which is given in (4).  2Bella Coola is an isolate belonging to none of these three groups.  9 (4) Interior Salish3 Academic name^Native name  Northern^Lillooet^S^'imxcin Lil'wat Thompson^Niekepmxcin Shuswap^Sxwepmxcin Southern^Moses-Columbian^Nxahmxcin Colville-Okanagan^Sxwy2ifpx SnSickstx  Colville Lakes  Kalispel-Spokane-Flathead Nqlispelikn Npoqinikn Coeur d'Alene^Sn6icu?nikn 1.3 Outline of dissertation This dissertation is organized as follows. Chapter 2 outlines the research context in which proposals for a fourth node have been made and documents current applications of the fourth node. It is noted that most fourth node proposals are based on either African ATR or Semitic data. This raises the question: what other languages motivate a fourth node and how does the phonetic quality and phonological behaviour of post-velars in such languages reflect on what is known from Semitic and African systems. As one 31 note here that in many cases the academic term for a given language often covers several dialects. The parent constituency as constructed by linguists is not necessarily recognized by speakers themselves. As far as I know, for example, there is no term for the dialect continuum termed Colville-Okanagan. Frequently the native term for a given dialect centres on the name of the people for themselves. Thus, the Nie2kepmx are the Thompson people, so named in English after the Thompson River gorge, the lower portion of which is central to their territory (Thompson and Thompson 1992). Their language is Nle?kepmxcin, the language of the N4e2kepmx people. As a consequence of this practice, there can be as many language names as groups of people identified, even though the differentiation may be minimal from a linguistic point of view. As a consequence, I have retained the academic terms for the various languages as defined by linguists. However, I make one exception and refer to Thompson by the name 1\14e2kepmxcin. For Nte?kepmxciui the major written sources (Thompson and Thompson 1990, Thompson and Thompson 1992) are based on data from the central part of Thompson territory which the term N4e2kepmx properly designates. Furthermore, since Mandy Na2zinek Jimmie, a linguist, friend and speaker of N4e7kepmxcin prefers that the language be referred to as Nle?kepmxcin, I choose to respect her wishes. I realize that this decision is in some senses arbitrary.  10 response to these questions, Chapter 3 develops a gross systemic typology of post-velar phonology based on data from Ruhlen (1975). This survey documents the occurrence of post-velar articulation in a fairly large sample of languages and establishes some sense of the statistical likelihood of co-occurrence within the post-velar class. The general findings are that uvulars are not dependent on the presence of pharyngeals or glottals in an inventory; inventories with pharyngeals are highly likely to have uvulars and glottals; the presence of glottals alone does not imply the presence of uvulars or pharyngeals. These findings, while suggestive, are not presented as definitive given the need for a close phonological investigation of post-velar activity in individual languages. Chapter 4 examines the phonetic aspects of post-velar articulation based on vocal tract anatomy, descriptions of post-velars in natural languages and the acoustic consequences of postvelar articulation. This lays the basis for evaluation of the Salish phonetic data presented in chapter 5. The first part of chapter 5 undertakes a detailed acoustic investigation of post-velar segments and their effects on adjacent vowels in Moses-Columbia Salish. This enables an estimation of place of articulation for the Interior Salish post-velars. The second part of chapter 5 examines the acoustic characteristics of pharyngeals as manifested throughout Interior Salish. This investigation is based primarily on data gathered in my own fieldwork but includes analysis of some data which has kindly been made accessible to me by other researchers.4 The information from the acoustic investigation of Interior Salish pharyngeals is compared with the Semitic data and what is known about Caucasian. The acoustic profile of Salish post-velars indicates that the impressionistic accounts of them are entirely appropriate: uvulars are uvulars, pharyngeals are pharyngeals and glottals are glottals. As a consequence, the Salish postvelars are comparable to their Semitic counterparts, bearing in mind that no two languages have exactly the same articulatory settings. What is more interesting is the phonological divergence of the two language groups. Chapter 6 presents a phonological 4In particular I thank Barry Carlson for making Spokane data available to me, and M.D. Kinkade for additional data on Moses-Columbia Salish.  11 typology of post-velar behaviour as it is manifested in the seven Interior Salish languages and compares the results with the outcome of McCarthy's (1991) work on Semitic and the consensus of work on African ATR systems. The major theoretical findings here are (i) Interior Salish motivates a fourth node to characterize the behaviour of its post-velars; (ii) laryngeals are not characterized by this fourth node; (iii) the phonology of Salish postvelars appears to be sensitive to constraints also found in African ATR systems. Since the representation of glottals divides Salish and Semitic, Chapter 7 closely examines glottal phonology. It is argued that 12, h/ are Placeless obstruents bearing continuancy values, and that in the default case, they do not bear Laryngeal Node features. Based on this, it is concluded that the Semitic and Nisgha paradigm is typologically unusual.  12  Chapter Two: Fourth node history 2.0 Introduction This chapter reviews the research motivating a fourth node and documents the various proposals for its instantiation in Semitic, Salish and Nisgha. There is considerable variation across proposals, much of it having to do with the phonological representation of segments involving more that one articulator node. 2.1 Early ATR assumptions It has been noted that the fourth node was originally introduced to integrate the description of African ATR harmonies into a feature geometric framework. The assumption that ATR harmonies require a fourth node reflects the disparity between the traditional vowel features [high, back, low] and our understanding of ATR systems. This of course raises the question: what defines an ATR harmony system? ATR harmonies are found in many Sub-Saharan African languages. They may not only be morpheme-controlled (e.g. the root or suffix vowel determines the vowel quality of other morphemes) but may also have dominant and recessive vowel sets. In the latter case, dominant vowels trigger harmony regardless of what morpheme they are found in, thereby erasing the root-affix distinction discussed above. The result of these harmonies is that specified domains usually have vowels from one of the two sets only. Typically, the vowels of one set are transcribed as higher in vowel space than corresponding vowels in the other set. The higher set of vowels is often described as 'breathy, hollow, muffled', whereas the lower set is often recorded as sounding 'hard, creaky, brassy' (Stewart 1971). Such vowel systems are often (but by no means exclusively) composed of ten vowels, five in each set, of the sort /i, e,  3, 0,  u/ versus  It, E, a, o, u/. The first set, often labelled 'close' is the relatively higher and  13 breathy/muffled/hollow set, while the second set, often labelled 'open' is the relatively lower and hard/brassy/creaky set. The articulatory basis of this distinction has been argued to be tongue root position rather than tongue body height such as traditional transcriptions would suggest. Ladefoged (1964) demonstrates from X-rays of Igbo (Kwa) that tongue root displacement is the most salient articulatory distinction between the two sets of vowels. Crucially, he shows that tongue root position is to some extent independent of tongue height. For instance, two vowels may have the same tongue height in the mouth but differ in tongue root position. This and other research led to the distinction being referred to as 'advanced/unadvanced tongue root', or ATR as we now know it. This physiological basis for African ATR systems has been confirmed by the work of Stewart (1967), Painter (1973) and Jacobson (1978). Subsequent work has pointed to the reduction in pharyngeal area as a consequence of retracting the tongue root for the unadvanced vowel set (Delattre 1971, Lindau 1979). Despite the sometimes complex and variable articulation of ATR contrast, it has been assumed that a single phonological feature can account for the vast majority of the data. Ten vowel systems of the sort /i, e, 3, 0, u/ versus It, e, a, o, u/ are of course extremely difficult to describe in terms of the three heights afforded by the various combinations of [+/- high] and [+/- low]. Such systems call for another feature, independent of phonetic arguments that tongue root retraction, as opposed to tongue height, is the articulatory instantiation of harmony. 2.2 Traditional representations An independent source of evidence motivating the fourth node can be found by considering the difficulties which conventional feature analyses have had with characterizing post-velar segments.  14 As noted in Chomsky and Halle (1968), Kenstowicz and Kisseberth (1979) and McCarthy (1991) among others, distinctive feature theory must at least be able to describe all the distinctions made in any of the world's languages, and it must be able to represent the set of natural classes in the world's languages. On the first count, the features used in Chomsky and Halle (1968) are problematic. Chomsky and Halle (1968) characterize velars, uvulars and pharyngeals as follows: (1)^Velars^Uvulars^Pharyngeals high low back + However, both Colorusso (1988) and Catford (1983) confirm that the Byzb dialect of Abkhaz (a Caucasian language) contains plain velars, uvulars, pharyngeals and pharyngealized uvulars in phonemic contrast. The feature array in (1) cannot describe such segments without simultaneously eradicating the distinctions between them. Problems with the use of the Chomsky and Halle (1968) features for post-velar consonants are further discussed in Kenstowicz and Kisseberth (1979), CzaykowskaHiggins (1987), Keating (1988) and McCarthy (1991). Czaykowska-Higgins (1987) notes that Arabic emphasis is not well described in terms of the features in (1). Whether emphasis is uvularization or pharyngealization, the spread of [+back] predicts that emphatic vowels become [+back]. This is empirically false: emphatic vowels may well back slightly but they do not become [+back]. McCarthy (1989), echoing current theoretical concerns with the articulatory integrity of features, notes that [high, back, low] are tongue body features, and that neither uvulars nor pharyngeals are correctly classified using these features. Uvulars are deemed [-high] in (1), whereas their articulation involves raising the dorsum. Pharyngeals are [+low ,+back] in tongue position, but also  15 involve articulators (pharynx wall, epiglottis, tongue root) which bear no straightforward relation to tongue body activity. Thus, reports of so-called 'fronting' from pharyngeals may be explained if pharyngeals lack any dorsal representation. Columbian Salish, for instance, is recorded with slightly fronted low vowels in the environment of voiceless pharyngeals (but never uvulars: Kinkade 1967:232), as are some dialects of Arabic (Harrell 1957), demonstrating some degree of independence between the tongue body and the tongue root. 2.3 Current proposals Early proposals for the fourth node do not attempt to examine the interaction of the Radical/Tongue Root node with other articulator nodes or with other segments in the inventories of languages with tongue root activity.1 This is not surprising since the languages for which the fourth node was originally proposed do not seem to have tongue root consonants2 and, almost without exception, ATR vowel harmonies proceed without affecting consonants at all. Czaykowska-Higgins (1987) explicitly compares the qualities of ATR-type systems with languages such as Arabic and Salish which have post-velar consonants. She argues for a distinction between the features [retracted tongue root] and [advanced tongue root]. Languages which have uvular, pharyngeal, glottal and retracted coronal segments in their inventories are characterized by the presence of a fourth node Tongue Root which dominates two features: [upper pharynx] and [lower pharynx]. Activation of the fourth node implies retraction of the tongue root (and is [RTR] in this sense) but the feature [ATR] is independent of the fourth node, being active instead in languages with voicequality distinctions on its vowels. The resulting feature combinations under the Tongue 1Cole (1987) is something of an exception to this statement in the sense that she uses the fourth node for post-velars. She does not, however, defend this analysis outside Coeur d'Alene. 2For example, of the 51 Niger-Kordofanian languages in Ruhlen (1975) two have some post-velars: Ngemba and Ewe have /CA but no uvulars. Of the twenty-five Nilo-Saharan languages, Tama has /11/ and no uvulars.  16 Root node capture the necessary distinctions between uvulars and pharyngeals on the one hand and emphatic (uvularized) pharyngeals (such as found in some Arabic dialects) or pharyngealized uvulars (Bzyb) on the other hand. (2) Czaykowska-Higgins (1987) q^ciT^CGI^Xr (=rhotacization) Upper Pharynx^+^+ Lower Pharynx^+^+^+ Czaykowska-Higgins' proposal does not rely on the double articulation of uvulars as Dorsal-Pharyngeal, although this type of configuration is suggested as a way to capture velar blocking of Tongue Root features in West Greenlandic and Chilcotin. 2.3.1 Fourth node in Semitic The most recent, fully documented discussion of relevant, mostly Semitic material is that of Hayward and Hayward (1989) and McCarthy (1991), both of whom formalize what has often been remarked on in descriptive studies of Semitic languages—that uvulars, pharyngeals and glottals function as a class for the purposes of a number of phonological phenomena. This class is traditionally referred to as the 'gutturals'. Hayward and Hayward (1989) suggest a 'zone' is required to characterize the gutturals. One of McCarthy's innovations based on Semitic data is to propose that nodes under Place in the feature geometry tree are place of articulation nodes, since it is only on this assumption that 12, h/ can be grouped with /x, i h, S/ in Semitic phonology. Proposals ,  prior to McCarthy (1989, 1991) follow Sagey (1986) in assuming that all nodes under Place, including the fourth node, are articulator nodes. (3) below summarizes the formal details of McCarthy's proposal:  17 (3) McCarthy (1991)  r sonorant 1  L consonantal _I Larynge [cstr. gl] [spr. gl .]  nasal] cont]  Place 0  Oral  haryngeal [dorsal]^[radical] [pharyngeal]  Labial / [rnd]  Coronal Dorsal  [ant] [lat]  [clistr]  [low]7 [high] [back]  Features and nodes under Oral are comparable to those found in Sagey (1986), but the Place node Pharyngeal may dominate three features: [pharyngeal], [radical] and optionally [dorsal]. The feature [pharyngeal] exists to make the distinction between class nodes (such as Oral and Pharyngeal) which 'specify featural subgroupings' (McCarthy 1991:53) and terminal features which can code phonological distinctions. The feature [radical] refers to the tongue root as an active articulator. The feature [dorsal] functions under either the Oral node or the Pharyngeal node and indicates active articulation of the tongue body. Based on this feature geometry, McCarthy's representation of Semitic postvelars is thus:  18 (4) Semitic gutturals, emphatics, and /q/: McCarthy 1991  Guttural Uvulars  Glottals Phar  Phar o [dorsal] [pharyngeal]  [pharyngeal]  Pharyngeals  Phar [radical] [pharyngeal]  A  Coronal Emphatics Oral o^o Phar [coronal] [dorsal] [pharyngeal]  [dorsal]^[pharyngeal]  The feature [radical] which appears on /1, h/ represents the activity of the tongue root in articulating pharyngeals. The combination [dorsal] [pharyngeal] dominated by the Pharyngeal Place Node on /x, is/ indicates activity of the dorsum in the pharynx. The laryngeals /2, h/ are articulator-less, but not Placeless, being represented with a Pharyngeal Place node. Gutturals all have primary articulation in the pharyngeal cavity, with the emphatics and the uvular stop /q/ (which is not a guttural in Contemporary Standard Arabic) considered to be complex segments since they have an additional Place specification under the Oral node. With respect to the identification of primary as opposed to secondary articulation, McCarthy proposes that phonological rules can specify which, if any, place or articulator features are primary or secondary. As an alternative McCarthy suggests that any segment with only an Oral place has a primary Oral articulator and any segment with only a Pharyngeal place has a primary Pharyngeal articulator. If so, then presumably OraVPharyngeal segments will be complex unless otherwise stipulated.  19 The division of the traditional Place node into Oral and Pharyngeal places (with the features dominated by those nodes representing articulators) formalizes some aspects of the traditional phonetic notions of articulator and place of articulation as primitives in the description of segments. Gorecka (1989) represents an attempt to justify this division phonologically across all places of articulation. The use of two nodes (Dorsal and Pharyngeal) in (4) to characterize uvulars is found in Czaykowska-Higgins (1987), Cole (1987), Besse11 and Remnant (1989), Besse11 (1990), Remnant (1990). From a cross-linguistic point of view, Elorrieta (1991) argues that uvulars are of two types: those that are purely dorsal and may be characterized by values for [high] and [back] as opposed to those which are dorsal-pharyngeal and may bear not only features for [high] and [back] but also Tongue Root and its dependent [RTR]. A further contribution to feature geometry research is Gorecka's (1989) dissertation. Gorecka defends and formalizes a view of articulation that defines segments according to their site (i.e. the passive articulator or place of articulation) and articulator. A segment then, is represented by a tree with the traditional root node, but also with a phonological constituent (a Constriction Node) that dominates a Site and Articulator node, both of which are directly relatable to the articulatory properties of that segment. The general model is thus: (5) Gorecka (1989) R ot Laryngeal features^eatures (Constriction)^Conction Site  ^Articulator  ^  20 The possible values for Site are Labial, Anterior, Palatal, Velar and Pharyngeal. Site values do not dominate any features. Articulator values are Lower Lip (dominating [round]), Tongue Blade (dominating [lateral], [distributed]), Tongue Body (dominating [high], [low]), Tongue Root, Pharyngeal and Glottis. Gorecka's representations of the class of Semitic gutturals is such that they all share a Pharyngeal Site, while having different Articulators. Gorecka's work is unusual in formalizing the Glottis as an articulator. (6) Uvulars, Pharyngeals and Glottals (Gorecka 1989) Uvular Con triction Site^Artillator Velar^Tge Body S,h Constriction Site^Artilulator Pharyngeal^Tge Root  Constriction Site^Artilulator Pharyngeal^Tge Root ?,h Constriction Site^Artilulator Pharyngeal^Glottis  Herzallah (1990) develops a feature geometry for Palestinian Arabic post-velars based on Clements (1989, 1990) and McCarthy (1989). The primary distinction between this and McCarthy's model has to do with the representation of secondary arcticulation.  21 (7) Herzallah (1990) X t  Supr  sonorant] consonantal] [continuant]  C- lace^Va1ic [labial] [coronal] [dorsal] [pharyngeal]  Aprure^V-place  [open]^[labial] [coronal] [dorsal] [radical] [pharyngeal]  In Clements' model secondary articulation on consonants is represented by a Vowel-place node. Thus, Herzallah's representation of coronal emphatics considers their primary place of articulation to be coronal, but with [dorsal] and [pharyngeal] (the representation for uvulars) under the Vowel-place node denoting secondary uvular articulation. The Palestinian pharyngeals and glottals have primary [pharyngeal] articulation under the Consonantal node, with /h, S/ involving a radical articulator that is absent for /?, h/. Uvulars are complex [dorsal] [pharyngeal] articulations. These representations parallel McCarthy's (1991) in that the guttural class is comprised of segments without secondary articulation per se. Herzallah's representation of /a/ formally characterizes it as the vocalic version of /?, h/. SL in this diagram refers to the Supralaryngeal node; C-place refers to Consonant Place.  22 (8) Herzallah (1990): Palestinian Arabic Emphatic C Uvulars  Pharyngeals  IL  1_,  IL  C-place  C- ace  C- lace  [dorsal]  [corona'] V- ace  [pharyngeal]  [pharyngeal] [ralical]  [dorsal] [pharyngeal]  Glottals^/a/ SL^SL I C-Jace^V-place I^I [pharyngeal]^[pharyngeal] The claim that /a/ is the vocalic version of /?, h/ is refuted by E. Pulleyblank (1992), who argues for a pharyngeal glide [a] as the non-vocalic counterpart of the low vowel /a/. Pursuing this hypothesis, Pulleyblank suggests that Standard Arabic /q/ is distinguished from the guttural class by having [a] as a secondary glide articulation provided by the radical articulator. Uvular gutturals are then complex segments with dorsal and radical primary articulations. The pharyngeals /h, 1/ are simply obstruent versions of the glide [A]. Under this analysis there is an intimate relationship between vocalic [a], glide [A] and fricative [h, 1] rather than [a] and [2, h] as glides. Finally, a quite different proposal for post-velars is made by Halle (1989), who suggests a node Laryngeal, dominating two articulator nodes, Tongue Root and Glottis. The Tongue Root node dominates two features, [advanced tongue root] and [constricted pharynx]. [Advanced tongue root] is to accommodate African ATR harmonies,  23 [constricted pharynx] reflects retraction of the tongue root towards the rear wall of the pharynx and is used for Arabic uvulars and pharyngeals. The class of Semitic gutturals is thus accessed by reference to the Laryngeal node. (9) Halle (1989, 1992)  Supralaryngeal Soft^Place Palate Labial Coronal  Laryngeal Gkitis  Tongue Root  [stiff vc] [ATR]  ^  [CP]  Dorsal  McCarthy (1991) argues against the grouping of pharyngeal features under the Laryngeal node on the basis of Obligatory Contour Principle effects which reference the Place node and include pharyngeal articulations but not laryngeal ones. A second argument in favour of the fourth node grouping in some manner with the three traditional articulator nodes comes from the complications with stating guttural lowering under Halle's assumptions: whereas Laryngeal features seem to spread without reference to or affect on Place features, guttural features interact with the height specifications of targetted vowels. 2.3.2 Fourth node in other languages The use of the fourth node is not limited to Semitic and ATR systems. Doak (1989) proposes a node Pharyngeal dominating a feature [retracted tongue root] to account for Coeur d'Alene vowel harmonies. In her 1992 paper this feature is abandoned and only the node Pharyngeal is used. Bessell and Remnant (1989) and Bessell (1990)  24 use a fourth node Tongue Root for Coeur d'Alene; Remnant (1990) proposes a feature [epiglottis] under a fourth node Tongue Root to characterize the effects of Lillooet pharyngeals. In none of these cases is there evidence for glottals being characterised under the fourth node, and therefore none of the analyses mentioned attempt to include them there. (10) Remnant (1990): Lillooet uvulars, phaiyngeals and /z/ Uvulars^Pharyngeals^/z/ Place^o Place^Place Dorsal Tongue Root ([epiglottis])  1 Tongue Root Coronal 1^ [+epiglottis]^Dorsal Tongue Root ([epiglottis])  Shaw (1991b) uses a fourth node Pharyngeal to capture alternations between glottals and uvulars in Nisgha, a Tsimshianic language spoken in the Nass River valley of British Columbia (Tarpent 1983, 1987, Shaw 1987). Nisgha lacks pharyngeal resonants of the sort found in Interior Salish, but provides evidence for the characterisation of glottals and uvulars by a fourth node. /?/ in this analysis is distinct from /q7 by bearing a Laryngeal Node that does not dominate the feature [+glottalized]. Structurally, /ca can be viewed as a glottalized version of /2/. Also of note in Shaw's analysis is the representation of PI as [-continuant], on a par with the other stops in the language, but contrary to assumptions in generative phonology that glottals do not bear continuancy values (Chomsky and Halle 1965, McCarthy 1988, Lombardi 1990, Padgett 1991).  25 (11) Shaw (1991b): Nisgha uvulars and glottals /?/^/11/^/q/ [-son]^[-s n]^[-son] L R]^[-cont] [L P1 0  P1 0I^P1 7  [-son] [-cont]^[L R]^[-cant] l P1 7 [+gl ot]  [PHAR]^[PAAR]^[P AR]  [PHAR]  Goad (1989, 1990, 1991) uses a fourth node Pharyngeal dominating a feature [retracted tongue root] with a separate feature [advanced tongue root] under a Vowel place node. Goad represents co-occurrence restrictions between the features [low] and [atr] directly in the feature geometry. She argues that this type of geometry is necessary to account for so-called 'Flattening' processes in the Athabaskan language Chilcotin (Cook 1983, 1987). (12) Goad (1991) Labial Coronal Dorsal Pharyngeal  Place Vo hi h low/atr  2.3.3. Fourth node for secondary articulations The fourth node is also used to capture the notion of tongue root retraction as a secondary articulation in Arabic emphatics (Herzallah 1990, McCarthy 1991), by Bessell and Remnant (1989), Bessell (1990) and Remnant (1990) to characterize Coeur d'Alene /r, r'/, and Lillooet /z, z'/. Doak (1989, 1992) also uses the fourth node for Coeur d'Alene /r, r'/.  26 2.4 Conclusions From this survey it can be seen that the fourth node has had a short and somewhat turbulent history. The precise relationship between the fourth node in African ATR systems as opposed to Semitic and Salish is not apparent and most authors have not attempted to deal with the issue. The inclusion of Semitic data in the debate has led to considerable changes in feature geometry based primarily on the patterning of glottals in Semitic. Shaw's (1991b) analysis of Nisgha compounds the evidence that glottals can pattern as if Pharyngeal in place even in an inventory without /h, T/. Furthermore, a crucial question is raised by the extension of the fourth node to characterise Salish and Semitic pharyngeals. Namely, are the segments we are dealing with phonetically comparable? Are the Salish segments transcribed /q, x,L, h/ uvulars and pharyngeals in the same sense that the Semitic ones are? Chapters 4 and 5 consider these questions directly.  27  Chapter Three: Systemic typology of post-velars 3.0 Introduction This chapter examines the systemic distribution of uvulars, pharyngeals, glottals and pharyngealized segments as they occur in the 693 inventories in Ruhlen (1975).1 This survey establishes the broad distributional properties of post-velars from a crosslinguistic perspective. Section 3.1 examines the distribution of uvulars and section 3.2 considers pharyngeals and pharyngealized segments. In general uvulars are not dependent on the presence of pharyngeals or glottals in an inventory, but inventories with pharyngeals are highly likely to have uvulars and glottals. Section 3.3 considers the distribution of f2, h/ in some detail since their status as stop and fricative respectively is challenged in the generative tradition. The presence of glottals alone does not imply the presence of uvulars or pharyngeals. Inventory evidence also points to /?/ as a voiceless stop and /h/ as a voiceless fricative. Glottals are also highly unlikely to accept secondary articulations such as palatalization and rounding, which, while they have clear Place preferences, can occur on segments at all places of articulation. Furthermore, while the presence of ejectives or aspirates in an inventory is likely to co-occur with /2/ or /h/ respectively, the occurrence of /2, h/ does not necessarily imply the presence of /C', Ch/. A full discussion of glottal representation is deferred to Chapter 7. Before embarking on this survey, I note that there are several unavoidable shortcomings of this type of gross systemic work. One is its reliance on existing analyses. Fieldworkers are neither phonetically nor phonemically infallible, obviously. Of particular concern with respect to post-velars is that the transcription of glottal fricatives can be problematic. It can be difficult to distinguish between [x, x, h, II], especially if they are lightly articulated. For example, Maddieson (1984) considers /h/ so problematic that he does not include it in his discussion of fricatives. Referring to Pike 1I have used Ruhlen (1975) as opposed to Maddieson (1984) for the simple reason that there are 376 more inventories in Ruhlen.  28 (1943) and Merlingen (1977) he considers what is transcribed as /h/ to be variable in place of articulation and therefore difficult to assess in terms of place of articulation categories (see section 4.2.9). The distinction between velar and uvular articulation can also present transcriptional difficulties. A particularly acute problem in the case of 1?, h/ is the decision to accord them phonemic rather than phonetic status. It is well known that /2, h/ are very common as epenthetic consonants, filling onset and sometimes coda positions. Unfortunately, not all analyses are clear with respect to the status of glottals within the total phonology of the language. Although I have checked Ruhlen's original sources in some cases in order to verify the transcription and distribution of glottals, the sources themselves do not always provide sufficient information to determine whether h/ are unquestionably phonemic or not. However, for the purposes of calculating the systemic typological patterning of glottals, I consider /h/ to be a phonemic glottal fricative and IV a phonemic glottal stop. As a consequence of these shortcomings, typological work of this sort needs to be balanced by close investigation of individual languages. It is my purpose here simply to document rhe general distributional properties of post-velars, without making strong claims about patterning based on overt phonological evidence. 3.1 Uvulars Uvulars occur in 146/693 languages in Ruhlen (1975). This amounts to 21% of the total number of languages surveyed. Velars are present in all 146 of these languages (100%); pharyngeals are present in 41(28%); glottals are present in 129 (88%). These figures reflect the almost universal occurrence of velars, the scarcity of pharyngeals and the relatively high occurrence of glottals (see section 3.2 and 3.3 below). However, given that pharyngeals occur in only 7% of Ruhlen's sample, their co-occurrence with uvulars is four times their average distribution throughout the database.  29 The distribution of rounding as a secondary articulation points to a correlation between velars and uvulars. Rounding tends to be associated with Dorsal consonants. Of 41 languages with rounded uvulars, 40 have rounded velars (98%), 12 have rounded Coronals ((29%) and one has a rounded labial (2%). There are 6 languages with rounded velars but no rounded uvulars (Buang, Coos, Diegueno, Eyak, Hopi and Tsimshian). Table 3.1: Distribution of uvulars  Ii  L's with uvulars K also C also 2 also  Number  Percent  146/693 146/146 41/146 128/146  21% 100% 28% 88%  3.2 Pharyngeals Pharyngeals occur in only 48/693 (7%) of the languages sampled by Ruhlen (1975).2 Of the 48 languages with pharyngeals, 39 have both lb. 11(81%); 3 have voiceless /h/ only (6%) and 6 have /1/ only (13%). 41/48 languages with pharyngeals have uvulars as well (85%); 45/48 also have laryngeals (94%). From these figures it can be seen that there is a high co-occurrence of uvulars with pharyngeals (85% as compared with 21% of the entire database). The co-occurrence of laryngeals with pharyngeals is likewise higher than the general occurrence of glottals in the database (94% as opposed to 75%), but nowhere near the same extent as the co-occurrence of pharyngeals with uvulars.  2Pharyngeals occur in 4% of Maddieson's (1984) sample.  30 Table 3.2: Distribution of pharyngeals with otherpost-velars II Number L's with pharyngeals 48/693  I Percent^I 7%  thi and n/  39/48  81%  /h/ only ni only  3/48 6/48  6% 13%  Q also  41/48  85%  ? also  45/48  94%  There are some cases where the appearance of a single pharyngeal can be viewed as filling in holes in the uvular inventory to make it symmetrical with either the velar inventory (Koryak) or the fricative inventory as a whole (Achumawi). (1) Maverick pharyngeals as structural uvulars Koryak (Paleosibetian) P^t^6^k^q^2 V^Y^r m^n^yi^0 1^A Achumawi (Hokan) P^t^ts^k^q^? s^5^x^h^h m n 1 C An intriguing case is Ngemba, where it looks as if the voiced pharyngeal may be the voiced fricative counterpart to N.  31 (2) Ngemba (Niger-Kordofanian) t^k^2 b^d^g f^s f V^Z 3 y^S m^n ji g 1, r Secondary articulations on pharyngeals Rounding and glottalization are the only attested secondary articulations on pharyngeals, and both are exclusive to Interior Salish in Ruhlen's database, which includes 2 of the seven Interior Salish languages. Pharyngeals and glottalized resonants are attested in all seven of the Interior Salish languages, so the appearance of /S', T'Iv/ is expected, given that pharyngeals in these languages are phonologically [+sonorant] (section 6.2.3). The appearance of rounded pharyngeals in Interior Salish is matched by rounded velars and uvulars. The analysis of Agul 'deep' pharyngeals may include a laryngeal feature (see sections and 4.5) as may the Nootka and Haida ones. Since pharyngeals are central to this dissertation, Table 3.3 lists all of the languages in Ruhlen (1975) with a pharyngeal and whatever other post-velars there are in the inventory. Ruhlen (1975) uses square brackets [Ito indicate the phonetic presence of a sound where presumably its phonemic status is questionable. The figure under each language group indicates the number of languages in that group which contain a pharyngeal. For example, Afro-Asiatic (11/29) indicates that of the 29 Afro-Asiatic languages in Ruhlen's database, 11 contain pharyngeals.  32 Table 3.3: Distribution of pharyngeal fricatives: (Ruhlen 1975) Language  Language  Pharyngeals  Uvulars  Glot- C', CT etc. tals  Shilha  h,T  none  [h]  ti,d1,0,sT,zf, nl,lr,rT  Tamazight  h,T  clw,Xw,gw  ?,h  tS,d,si,zi,ni, 11,r1  Bilin  h,1  cbqw  ?,h  t', k','é'  Somali  h,1  G  2,h  cf  1raqw  h,T  ci,(4W  ?,h  [6, cf]  Egyptian  h,T  cl,X,ff  2,h  ts,dc,sc,zc  Iraqi Arabic  h,T  q,x,y  2,h  tS,sl,e,fC,zi, niC,11, [pC,bC]  Moroccan  h,1  (1,X,ff  2,h  tC,c1S,s,cl,[bC,  group Afro-Asiatic (11/29)  Arabic  zT,rnT,1C]  Arabic h,T  cl,X,Is  2,h  bT,tS,c1S,sT,zT, nf,1C, m1  Tigre  h,1  none  ?  p',t',k',s','é'  Geez  h,T  none  2,h  pt,t1,1e,dt,s'  T  none  none  T  none  ?  h  none  h  Tajik Persian h  cl,X,g  none  2,h 2,h  lul, [s19 t',ts',61,1c1, {X]  2,h  f,ts',61,k',X'  Syrian Arabic  NigerEwe Kordofanian (2/51) ,Ngemba Nilo-Saharan Tama (1/25) Indo-  6, cc  European (2/73) Caucasian  Kurdish  h,T  Andi  h,T  cl,X,ff qh,q,,qhw,cfw  Avar  h,T  q, q' ,q'w, x,  (23/37) Xw, If  33 Axvax  ti,f  cl,q'  2,h  pl,t1,ts',6",k', X'  Bagvali  [ti,S]  ?,h  e,c',6',1c1, s',f'  Bezhita  [ti,C]  q h^, ,c1 qh , q , , X, g  2,h  pf,e,ts',6',k', X'  Botlix  [11,C] [h],1  q, q'  ?,h  tI,ts',,k',X'  q'  2,h  f,ts',1c1,X1,  Chamalal  f',[x] Dido  ti,T  q, cf, X, IS  ?,h  p',e,ts',6',1e, XI  Ginux  h,S  q, q', qw,X, Xvv,  7,h  p',f,ts',6', k',X '  g, NW  2,h  e,ts',6',k',X'  11,C  q, q' qh^, ,q  2,h  ti,ts',,k',X'  h,C  q, qw, q', q'w,  ?,h  p',t',61,k',X1  Dargwa  h,f  X, g qh , cf  ?,h  p',e,ts',6',k'  Lak  11,1  q, qw, q', x  2,h  p',e,ts',61,k'  Agul  h,1  ?,h  pe,e,ts',61,k'  Archi  ti,S  qh, q', qw, q'w qh^, , q , X, g  ?,h  p',f,ts',,k', X'  Budux  ti,T  q, q', x, g  ?,h  p',e,tsk'  Kryts  ti,S  qh, q', G  ?,h  p',e,ts',k'  Xinalug  h,C  CI, q',  II  ?,h  p',e,ts',61,1c1  Bats  h,S  q, qX, x, IS  2,h  p',e,ts,k'  Chechen  ti,f  q, qX, X, is  ?,h  p',e,ts',k'  Ingush  h,S  q', qx, X, IS  2,h  p',e,ts',6',k'  Karata  ti,S  Tindi Xvarshi  E. Circassian ti,[S]  siberian (1/8) _  x  G  ,  q, qw, x, xw, is, ?,h isw  r  q, x, IS, N  ?  Buryat  h  none  none  Koryak  r  q, q:  2  Altaic (2/39) Tatar Pa1eo-  ,  p',V,ts',1e,sC,  fl  34 Austro-Tai Atayal (1/67) Salish (2/10) Columbian  Okanagan  h  q  ?  1, Sw, 5', 1'w, q, qw, q', q'w, h, hw X, Xw  ?,h  1, Cw, 5', S'w q, qw, q', q'w,  ?,h  X, Xw Wakashan Nootka (1/2) Hokan (1/19) Achumawi Ge-PanoCapanahua Carib (1/24) _  h,1  q, qw, q', x, xw 2,h  h h  q none  p'41c1,ts',X',1' ,m ' ,n ' 91 ' ,r ' ,y ' , WI p',f,k',ts',X',1' ,m',n1,1',e,y', WI p',f,le,ts',6',  2,h ?  3.2.1 Pharyngealized consonants Pharyngealized consonants occur in 7/693 (1%) languages in Ruhlen (1975). Four of these are Arabic, two are Caucasian and one is a Berber language. As can be seen from (3), in all cases except Shilha, uvulars are present; in all cases except Ubyx pharyngeals are present; in all cases except Shilha, glottals are present. (3) Post-velar inventories in languages with pharyngealized consonants Egyptian Arabic: AT, d1, sf, zC , q, x, Is , h, 1,? , h/ Iraqi Arabic: / t1, sC, 15T, q, h, 1, h, 2/ Moroccan Arabic: AT, dT, s1, IS, q, x, u , h, 1, 2, h/ Syrian Arabic: /1§, tS, (IT, s1, 414 , nC , li , q, x, is , h, 1, ? , h/ Circassian: 15S, 3T, q', q'w, qx, qxw, X, xw, 15, 15W, h, [1], 2 , 2w, h/ Ubyx: 101, p'S ,b5, fC, m5, qh, qhw, q', X, xw, h/ Shilha: AT, dC, kC, s5, z5 ,l5 , ri, h, 1, [h]/ In those Interior Salish languages for which retracted coronals are derived or underlying3, pharyngeals, uvulars and glottals are present in the inventory. Chilcotin, an 3The phonology of Interior Salish retraction is considered in Chapter 6.  35 Athabaskan language which neighbours Lillooet and from which it may have borrowed its retraction phenomena, has retracted coronals, uvulars and glottals (Cook 1983, 1987).  3.2.2 Pharyngealized vowels Pharyngealized vowels are not recorded in any language in Ruhlen (1975). He does notate creaky vowels, which in some cases are otherwise described as pharyngealized. An example in Ruhlen is Archi (Caucasian), which has pharyngealized /e, m/ and a pharyngealized rounded /a/ vowel, as well as uvulars, pharyngeals and glottals. Dido, also Caucasian, has plain and pharyngealized /i, e, a, o, u/ as well as uvulars, pharyngeals and glottals. From this it would appear that pharyngealized vowels do not appear in inventories without primary pharyngeals, uvulars and glottals. Of the Khoisan languages, Ruhlen lists !Kung with 'pressed' vowels /x, o/ with pharyngeal friction. In !X66 the pharyngealized vowels are /a,o,u/ (Traill 1985). !X(56 has two 'uvular-pharyngeal' consonants transcribed by Traill as /x, q/. See section 4.3.4 for further discussion of Khoisan.  3.3 Glottals A glottal of some sort occurs in 520/693 of the languages in Ruhlen (1975). This amounts to 75% of all the languages in his survey. The breakdown between 121 and /h/ is as follows: /2/ occurs in 337/693 languages (49%); /h/ occurs in 442/693 (64%) and is therefore more common that IV; 264/693 (38%) of the languages in Ruhlen have both 12, h/; 78/693 (11%) have /2/ without /h/; 178/693 (27%) have /h/ without /?/. The occurrence of glottals with pharyngeals and uvulars respectively reflects the general distribution of these segments in the database as a whole, and shows that once you have a pharyngeal or uvular the liklihood of a glottal in the inventory is very high.  36 Table 3.4: Distribution of glottals4 Il No. of Ls  1 Percent  12/  337/693  49%  /h/  442/693  64%  Pi or /h/  520/693  75%  12/ and /h/  264/693  38%  /2/ only  78/693  11%  /h/ only  178/693  27%  2 and T  45/520  8.6%  ? and Q  129/520  25%  The fricative /h/, along with coronal Is/, is the most commonly found fricative in inventories. Maddieson (1984) reports that 63% of his sample contains something transcribed as /h/. This is very close to the figure (64%) derived from Ruhlen's sample. The presence of /h/ does not necessarily imply /2/ (178 cases in Ruhlen, 27% of all languages, 34% of all languages with a glottal). It might be argued that the presence of /h/ without /2/ represents an asymmetry since labial, coronal and dorsal fricatives do not usually appear without a homorganic stop counterpart. It is suggested in Chapter 7 that if laryngeals require only the root value [consonantal] or [-sonorant] then the presence of any obstruent provides the features [consonantal] and [-sonorant], which is all that is needed for /?/ if [-continuant] is the unmarked value for continuancy. /h/ then requires the presence of a fricative to provide the inventory with [+continuant]. As it happens, /h/ almost never appears in an inventory without /s/ or some other fricative. While the vast majority of languages have fricatives at some place of articulation, 36/693 (5%) do not. Ruhlen (1975:141-2) notes 'Except in Oceania, almost all of the world's languages have one kind of fricative or another. Only 36 of the 693 languages in our sample lack fricatives altogether. Of these 36 languages, one is found in Africa, three 4These calculations collapse /fi/ with /h/. I have counted 5 cases of /fi/ in Ruhlen's database. Instances of [h] or [7] are not included since this notation indicates some question about the phonemic status of the segments.  37 in South America, and the remaining 32 in Oceania (22 in Australia, seven in New Guinea, two in Bougainville, and one in the Gilbert Islands.'. Of interest here is that none of the Australian languages has a glottal fricative, nor does any other language mentioned by Ruhlen as lacking fricatives altogether. This might lead one to think that /h/ never occurs without a fricative in the inventory. This is not completely true, since Ruhlen (1975) makes his calculations treating laryngeal articulations as a separate class. There are four languages in Ruhlen's sample which are recorded with an /h/ but no other phonemic fricative. Given the 442 languages with /h/, these four exceptions amount to 1%. The consonantal inventories of the four exceptions are given in (4). Each language contains consonantal phonemes that are phonetic continuants. (4) Occurrence of /h/ without a fricative Hawaiian: /p, k, 2, m, n, 1, h/ Marshallese: /t, U, k, kw, b, hi, bw, m, mi, mw, n, g, gw, 1, lw, r, rw, rJ, h / Yagua: /p, t, k, ?, ts, m, n, r, h/ Tairora: /p, t, k, ?, mp, '1t,^b, m, n, r, h/ Of these exceptions, Bender (1971:450-451) remarks on the Marshallese inventory, "The status of h—velarized (or perhaps pharyngealized) counterpart of y and w—is most tenuous, since it is the onset, transition, or coda one finds in the absence of y,w, or another consonant, and is in this sense predictable. However, its use greatly simplifies canonical forms, eliminating initial and final vowels and vowel clusters or geminate vowels...'. This description points to /h/ as an epenthetic, possibly pharyngeal consonant in Marshallese. It may not be a laryngeal fricative. In the Hawaiian inventory, /11/ is the reflex of Proto-Polynesian */s/• Proto-Polynesian *Th/ has reduced to [0] in most languages, including Hawaiian (Biggs 1971). I have not been able to determine  38 from the sources for Tairora and Yagua if the analysis of /h/ as phonemic is appropriate or not. It would appear then that there is a dependency between the presence of /h/ and the presence of other fricatives, but this is difficult to prove conclusively given the high occurrence of /s/ anyway. However, the distribution of /h/ is almost completely in accordance with the hypothesis that its presence is made possible by the presence of fricatives. The obvious question is: Is the distribution of /2/ likewise predicated upon the appearance of voiceless stops? It would appear so. In those (few) languages which lack a voiceless series of any sort there is no /2/ (or /11/ for that matter, as mentioned above).5 (5) Yanyula (Australian) b^ci^d^di^4^g m^n^nj^11 1  Ngarndji (Australian) b^ddiclj^g m^n^ni 1^li^1 e^r It is difficult to find many true test cases, since almost all languages have either plain voiceless or voiceless aspirated stops. Either of these series would provide the features [+consonantal, -sonorant] for IV, but in inventories with Ch only (and no plain voiceless stops) we might expect /2/ to be /2h/. Chapter 7 remarks on the poor documentation of [2h], despite its occurrence. In line with the undertranscription of [2h], there is no /2h/ in 51f the use of /b, d, g/ in such languages is meant to indicate unreleased or unaspirated stops with variable/secondary voicing, then we do not predict the necessary absence of N. But if I'll appears in such languages one would expect it to be unaspirated at least.  39 the inventories reviewed by Ruhlen. However, the source for Rawang (Morse 1963), one of the inventories entered with only a voiceless aspirated stop series, notes that /7/ is unreleased in the same contexts as the other voiceless stops of the inventory. Morse (1963) analyzes word-initial [7] as epenthetic and says it is less fortis than its phonemic word-final counterpart. It would appear that this is a case where if the laryngeal distinction really is voiceless aspirated, then ill participates, as predicted. Three other languages merit investigation to test the prediction that their /?/ should be Ph/ if (i) the analysis of their stop series as voiceless aspirated is correct and (ii) /?/ is phonemic. These three languages are Motilon, Nengone and Tehran Persian. I note here that Tajik Persian is not transcribed with phonemic Finally, there are cases of /?/ in inventories without a plain voiceless stop series, but with a voiceless aspirated series and either ejectives or implosive stops. Both ejectives and implosives would be a source of the feature [A-constricted glottis]. /?/ is the most common stop after /p/, /t/ and /k/. It is more frequent in occurrence than palatal and uvular stops.6 /2/ occurs in 342 of the 693 languages in Ruhlen (49%). All languages have stops, and the vast majority of languages have voiceless stops. Given the suggestion that every language with voiceless stops licenses /2/ and every language with continuants licenses /h/ one might expect glottals to turn up more often. So why are they so much less frequent than /p,t,k/ or /s/ ? First of all, even if they are not analysed as phonemic, /?/ and /h/ are often phonologically relevant in languages precisely because they are so readily available for syllabification processes in supplying onsets, etc. English is a good example. /7/ is not presented as phonemic in anlayses of English but it is argued to provide an onset for vowel-initial syllables and many dialects have reduction of /t/ in particular to , though it can occur from every place of articulation (Harris  6That is, assuming that alveo-palatal affricates are not better thought of as palatal stops, which may not be quite right. The occurrence of stops by place of articulation in Maddieson (1984) is bilabial 99.1%; dental or alveolar 99.7%; velar 99.4%; palatal/palato-alveolar 18.6%; uvular 14.8%.  40 1990). Malay (see chapter 7) is another example of a language without a phonemic /?/, but [2] nonetheless figures prominently in the phonology of the language.  3.3.1 Stop and fricative? The assumption of non-generative inventory architecture is that /2/ is a voiceless laryngeal stop and /h/ is a voiceless laryngeal fricative. This is really the default descriptivist assumption, though it is not the most common generative assumption. The standard generative tradition argues that /?, h/ are glides (see chapter 7). Given the divergence between descriptivist and generative assumptions, the issue bears some investigation from the point of view of systemic typology. Traditionally, inventories are structured with /?/ as part of the voiceless stop series, /h/ in the voiceless fricative series, and both /2, h/ as a 'final' place of articulation in a continuum from the lips to the glottis. These assumptions are reflected in the presentation of inventories in both Ruhlen (1975) and Maddieson (1984). 264 of the 693 languages in Ruhlen structure /h/ as the fricative counterpart of ill, given the presence of /2/ (38% of the total number of languages). No inventory places /?, h/ with the glides /w, j/. Seri is a typical example. (6) Seri (Hokan, NW Mexico) p^t^k^kw ? sf x^xw h m^n^0 1 If /2/ is present without /h/ (78 languages, 11%) it is always presented as the final place of articulation in the voiceless stop series, as in (7).  41 (7) Angataha (Indo-Pacific, SW New Guinea) p^t^k^? 0  However, some inventories present /h/ as a kind of 'gap-filler', in that it patterns wherever there is a space for it in the voiceless fricative series (typically in the velar or uvular region) and usually if there is no /?/ for it to pair up with. This type of arrangement is presented for more than 170 of the languages in Ruhlen (178 Ls have /h/ without 121; 27%). Moru is an interesting example. (8) Morn (Nilo-Saharan) P^t^W^k b^dj^jwg 6^cf S^sw^h^hw z^zw m^n^0 r1r There are also examples of /h/ as fricative counterpart to /4, gbi (Bini, Ngbandi). The point here is simply that /h/ is considered part of the voiceless fricative series and often has secondary articulations otherwise restricted to obstruents. Even in inventories with /21, 1W can still take secondary articulations which are not present on /7/ but are on velar or uvular articulations. The implication that /h/ patterns as if it were a velar or uvular fricative is of course made possible by the absence of such segments in an inventory. Rawang is a typical example of this sort of inventory. It restricts rounding and palatalization to dorsals (in this case velar stops and nasals), lacks a voiceless velar fricative series, has glottal stop and a series of h's with exactly the secondary articulations otherwise restricted to velar place of articulation.  42 (9) Rawang (Sino-Tibetan, N. Burma) ph^th^6^kh khw^2 b^d^j^g gw s^5^h hw z m n^a ev 1 Similar patterns are found in 14 other inventories in Ruhlen (1975). Of course an inventory on the page of a book says nothing about the phonological evidence for proposing rounded This and /k/'s as fricative and stop counterparts of one another. It turns out that constraints on the content of syllable onsets is the dominant factor in the analysis of /hw/. Thus, for example, C+w sequences in Chipaya are limited to /kw, qw, ?w, hw/, and the syllable structure of the language is maximally (s)CCVCC(t) if we analyze the limited C+w sequences as unitary Cw. In Lao, Rawang and Siona only single consonant onsets are allowed, providing we analyse Cw (and Ci) sequences as unitary. The one language which has both /2w/ and /hw/ is Yay, a Tai language spoken in N. Vietnam, which likewise restricts CC onsets to velars or glottals plus /w/. The two Yuman languages with /hw/ are Mohave and Kiliwa, and there is clear comparative evidence that they are historically derived from voiceless velar fricatives. Thus, for example, Mohave has 2ahwat 'red' where Maricopa has xwetam, Diegueno and Tipai have xwat 'blood'. Kiliwa has sah where Tipai has ?ax, Diegueno 2u2ax and so on.  The relevance of these facts, whether invested in inventory arrangements or statements of phonotactics is of course that /h/ and /2/ are understood to pattern phonologically with fricatives and stops respectively. In more specific terms, /7/ patterns within the class of voiceless stops and /h/ within the class of voiceless fricatives.  43 3.3.2 Secondary articulations on glottals One very striking distributional property of n, h/ is that they tend not to take secondary articulations which otherwise occur fairly freely with other places of articulation. Rounding as a secondary articulation provides the clearest evidence of this constraint. Rounding on glottals Rounded consonants (Cw) occur in 20% of Ruhlen's sample (136 languages). It is rare to find rounding on segments at every place of articulation. In the Circassian language Adyge consonants at all places of articulation are phonemically rounded, including /?/ (but not /h/).7 Instead of being evenly distributed across place of articulation, the preferred locus for labialization is dorsal consonants. 131 of the 136 languages with Cw include rounded dorsals (96%) and 63% restrict rounding to dorsals only.8 The occurrence of rounding by place is tabled in (3.5). These calculations exclude inventories where rounded dorsal stops lack a homorganic fricative counterpart which /hw/ may be argued to take the place of (such as Rawang). In this table, 'P' stands for a labial consonant, 'T' for a coronal, 'K' for a dorsal, 'V for a pharyngeal and '?' for a glottal.  7A very similar inventory is found in Kabardian (East Circassian). The only other language in Ruhlen's collection to labialize consonants across place of articulation is Juat (pw, tw, kw), but it does not have any glottals. This amounts to three languages which labialize across place of articulation, two of which are closely related (2.2% of the 136 languages with Cw). 8 If both velars and uvulars appear in an inventory, both are rounded, with three exceptions where only velars are rounded: Buang, Hopi, Tsimshian.  44 Table 3.5: Distribution of roundin Occurrence of plain segment in languages with Cw Seg't I^No. of L's I^Percent 130/136 136/136 136/136 13/136 111/136  P  T K r 2  96% 100% 100% 9.5% 82%  Occurrence of Cw by Place of Articulation Cw I^No. of L's I^Percent Pw Tw Kw Tw 2w  16/136 20/136 131/136 2/136 7/136  11.7% 14% 96% 1.4% 5%  The resulting hierarchy of preference for rounding is: Dorsal > Coronal > Labial > Glotta1.9 The place of pharyngeals in this hierarchy is a little more complicated to evaluate given their restricted occurrence in general. Rounded pharyngeals are exclusive to Interior Salish. Ruhlen's database includes Moses-Columbian and Okanagan—these are the 2/136 in Table 3.5. Given that there are only 13/136 Cw languages with pharyngeals at all, the presence of Tw represents 15% of the possibilities. This is obviously a higher occurrence than the 7/111 possibilities for 2w (6%). Including pharyngeals, the hierarchy is: Dorsal > Coronal > Labial > Pharyngeal > Glottal. Prenasalization on glottals Sixty languages in Ruhlen have prenasalized segments (9%). The overwhelming preference is for prenasalization to co-occur with voiced stops (86% of prenasalized segments are restricted to the voiced stops of an inventory). In almost all cases consonants at all available places of articulation are prenasalized; of course /2/ as a voiceless stop is not prenasalized, and /h/ as a fricative is not either. In all cases,  9These hierarchies are based on raw figures only—the statistical significance of these claims varies somewhat.  45 prenasalized stops do not occur unless they have a plain counterpart: an unadorned stop must exist before its modified counterpart can. However, there are 8 languages with prenasalized voiceless consonants. Three do not have /2/ so cannot speak to the possibility of prenasalized glottals (Malay, TemTein, Berta). Of the remaining five, four prohibit the prenasalization of /2/ (Wolio, Margi Tairora, Amuzgo). It would appear then that even when voiceless stops can be prenasalized /?/ is not included. However, there is one exception to this. Adzera is transcribed with a prenasalized glottal stop. Adzera prenasalizes every non-continuant non-sonorant consonant in its inventory where pre-nasalization agrees with the place of articulation of the stop: /?/ turns up with engma before it. If the constraint on prenasalization operates off a [+consonantal, -continuant] root node in Adzera, the correct results are obtained. Note that /2/ must be [consonantal] and [-continuant] for it to fall into this class. (10) Adzera (Oceanic, NE New Guinea) P mP b mb  f 111  t^'it d nd ts nts dz ndz s n r  k pk g Og  2^02  h 0 Palatalization on glottals 46/693 languages have palatalization (7%). Palatalization tends to favour the coronal place of articulation, with 39/46 languages having palatalized coronals (85%). 20/46 have palatalized velars (43%), 13/46 have palatalized labials (28%), 5/46 have palatalized /h/ (11%). There are no cases of palatalized pharyngeals in Ruhlen (1975). Clearly palatalized glottals are disfavoured. The hierarchy then is: Coronals > Dorsals >  46 Labials > Glottals > Pharyngeals. In this table, 'C' stands for any consonant, 'P' for a labial, T' for a coronal, 'K' for a dorsal, 'V for a pharyngeal and '2' for a glottal. Table 3.6: Distribution of palatalization I^Segment^II  No. of Ls^Il  Percent^I  Ci Pi Ti Ki  46/693 13/46 39/46 20146  7% 28% 85% 43%  si  0  0  2.1  5/46  11%  On closer insepection, the five cases of 2i (which are all in fact /hi/) are suspicious.10 Lapp has Ci for every segment in its inventory, as does Vepsian. This would argue for a syllable structure constraint. For Igbo, there is likely a syllable structure constraint which better explains the distribution of palatalized segments. Dunstan (1969) argues for C1C2V as the canonical syllable structure, where C2 is limited to /w, j/. Lakkia /hi, hw/ can be argued to pattern as velar fricatives, since it has /ki, kw/ but no velar fricatives. This reduces the isolation of /h/ for palatalization to one case, Nenema. (11) Nenema (Austro-Tai, NW New Caledonia) h^k kh PPhP wP hw ^t t^c mb mbw^nd ilj^Og v^S^V^h Ili hw m rri mw rqw^n 11 .ri^u 0 1r Nenema has aspiration at all places of articulation (turning up as voicelessness on nasals), and rounding on labials only, except for this odd case of /h/. /h.j/ is likewise odd in being the only palatalized segment in the inventory. Nenema also has /w, j/. If /Ili/ and /hw/ are parallel to nasals then they can be regarded as the aspirated/devoiced version of 10Flowever, Catford (1983) reports 2i in Abkhaz Adyghe.  47 /j/ and /w/ respectively, thus substantially reducing the idiosyncracies of this inventory: /hi/ is /j/ and /hw/ is /n/. This reduces cases of palatalized glottals to two languages where every segment in the inventory can take palatalization (Lapp and Vepsian). In summary, it is rare to find secondary articulations on glottals. There are 6 relatively clear cases of a labialized glottal where phonotactic constraints argue for a single segment analysis, one of prenasalization (glottals tend instead to be transparent), and three of palatalization where syllable structure constraints are relevant. Of 242 languages with some sort of secondary articulation then, a maximum of 10 appear to include glottals in the class of segments which can take secondary articulation (4%).  3.4 Aspiration and ejection Phonetics has long made the distinction between /2, h/ as segments with a laryngeal place of articulation, as opposed to adjustments of the vocal folds to generate phonation types, such as voicing, aspiration and ejection. Catford (1988) is very clear about distinguishing the glottis in its initiatory function from the glottis as a place of articulation. Ladefoged and Maddieson (1988) suggest that phonology should recognize the glottis as a place of articulation separate from its initiatory function. The distribution of glottals in the traditional organization of inventories recognizes this distinction by structuring 121 as a stop and /h/ as a fricative, although it is not always clear what the evidence might be for so doing. Here I consider the bald distributional facts concerning the presence of f?, h/ on the one hand and /C', Ch/ on the other.  48 Table 3.7: Distribution of aspiration and ejection No. of L's  Percent^Segment  No. of L's  Percent  337/693  49%  442/693  64%  C'  117/693  17%  h Ch  148/693  21%  ? without C'  220/337  65%  h without Ch 294/442  67%  C' and ?  100/117  85%  Ch and h  84%  C' without ?  17/117  15%  Ch without h 23/148  Segment  125/148  16%  Aspirated consonants occur in 148 of 693 languages (21%). 125 of these languages have /h/ also (85%). Ejective consonants occur in 117 of 693 languages (17%). 100 of these languages have /?/ also (84%). One might consider this as evidence that given the presence of /C'/, the liklihood of /2/ being phonemic is higher than usual; the same is true for /Ch/.11 The distribution of glottals in general confirms this, since  ril  appears 49% of the time, /h/ 64% of the time. However, it is also possible (approximately 15% of the time) to get aspirated or ejective consonants without the presence of /2/ or /h/, respectively. Does the presence of glottal segments imply glottalized and aspirated segments? Of the 337 languages with /2/, 100 have C' also (34%). Of the 442 languages with /h/, 125 have Ch also (35%). Put the other way round, /?/ occurs without C' 65% of the time, and /h/ occurs without Ch 67% of the time. Thus, it is more likely for a language not to have both /C'/ and /V, /Ch/ and /h/ than to have both of them. However, this must be contrasted with the fact that C' occurs in 17% of all languages in Ruhlen, Ch occurs 21% of the time. The co-occurrence of C' with /?/ is thus double that of the distribution of /2/ throughout the database, and the co-occurrence of Ch with /h/ is more than half again. Furthermore, the presence of /C', Ch/ implies the presence of a plain, voiceless counterpart almost without exception. In this respect, /C', Ch/ have a clear Place 11The 148 languages with Ch include 60 in which there is not a three-way distinction between voiceless C, voiced C and aspirated C. As a consequence, aspiration in these languages may well be a phonetic accompaniment to lack of voicing. The individual phonologies must be looked at to determine this. However, abstracting away from these cases, there remain 88 languages with voiceless unaspirated C, voiceless aspirated C and voiced C in contrast. Of these, 78 have /h/ also (89%). 47 have /2/ also (53%).  49 dependency. While there are preferences for the affiliation of ejection with velar or coronal place (Maddieson 1984), stops at all places of articulation can be ejective or aspirated. 3.5 Conclusions From a systemic point of view the following statements can be made about the distribution of post-velars. First, given pharyngeals in an inventory, uvulars are four times more likely to be present than their distribution throughout the database would indicate; glottals are 1.3 times as likely to be present. The connection between pharyngeals and uvulars is thus stronger than that between pharyngeals and glottals. Second, given uvulars in an inventory, pharyngeals are four times more likely to be present than their general distribution would indicate; glottals are 1.2 times more likely to occur. Again, the connection between uvulars and pharyngeals is stronger than that between uvulars and glottals. Third, given the presence of glottals in an inventory, the occurrence of uvulars is 1.2 times that of their usual distribution. The same figure applies for the likelihood of pharyngeals given glottals. While pharyngeals and uvulars strongly imply one another and glottals are 20-30% more likely to occur with uvulars and pharyngeals than their general distribution would indicate, there is a considerable difference in the strength of these implications.  50  Chapter Four: Phonetic investigations into post-velar articulation 4.0 Introduction This chapter develops a phonetic typology of post-velar articulation based on what is known about (i) the physiology of the vocal tract involved in post-velar articulation, (ii) language independent descriptions of post-velars, (iii) descriptions of post-velar articulation in Semitic, Caucasian and Interior Salish and (iv) the theoretical and attested acoustic consequences of post-velar articulation. These issues are considered in Sections 4.1 to 4.4 respectively. In brief, the phonetic facts discussed provide evidence for a distinction between velar and uvular places of articulation with uvulars sharing properties with both velars and pharyngeals as a consequence of their position at the intersection of the oral cavity with the pharyngeal cavity. Pharyngeals fall into two articulatory classes: those made without consistent laryngeal involvement and with the option of constriction relatively high in the pharynx as opposed to those made with a laryngeal component and constriction very low in the pharynx. Both types of pharyngeal articulation can contrast with fl,h/. Finally, Semitic, Caucasian and Interior Salish present 'pharyngealized' segments. The phonetics (and phonology) of these segments is discouragingly unclear, though best understood and documented in Arabic studies. 4.1 The anatomy of post-velars In order to assess and discuss the general articulation of post-velars, as well as language-specific descriptions of them, it is necessary to have some understanding of the vocal tract anatomy involved in their production. While much of this information is trivial in the sense that it can be found in any anatomy (and some phonetic) texts, it remains the case that the details of post-velar articulation are poorly understood and little discussed in the standard phonological literature. In particular, the relationship between  51 tongue root activity and pharyngeal constriction is central to deciphering the barrage of variant phonetic detail in descriptions of pharyngeal articulation. The extent of this variation is discussed in section 4.3. This section divides the anatomy of post-velars in terms of the relevant major cavities (4.1.1), places of articulation (4.1.2) and articulators (4.1.3).  4.1.1 The major cavities For the purposes of describing the articulation of speech sounds, the human vocal tract can conveniently and naturally be divided into three major cavities: the oral cavity (from the lips to the uvula, roughly speaking), the nasal cavity and the pharyngeal cavity. The pharyngeal cavity is usually divided into three parts, the nasopharynx, oropharynx and laryngo-pharynx (Daniloff 1973, Catford 1977,1988). The nasopharynx is the area above the velum, where the very upper reaches of the pharyngeal wall intrude into the nasal cavity. The oropharynx extends from just behind the mouth down to the hyoid bone (Daniloff 1973) or just above the larynx (Catford 1988). This leaves the laryngopharynx, which is usually defined as terminating with the laryngeal and esophageal openings and rising to the level of the hyoid bone. The larynx itself constitutes both a place of articulation (/2,h/) and a source of excitation for many speech sounds but it is not, naturally enough, considered to be a cavity per se.  4.1.2 Places of articulation Within the oral and pharyngeal cavities, and including the larynx this time, the International Phonetic Association in its Alphabet (revised to 1989) labels eleven primary places of articulation: bilabial, labiodental, dental, alveolar, postalveolar, retroflex, palatal, velar, uvular, pharyngeal and glottal. It does not chart, but provides symbols for three epiglottal segments: a voiced and a voiceless fricative C, H and a plosive ? . These latter segments and their place among post-velars are discussed more fully in sections  52 4.2.6-7. The eleven IPA divisions are intended to represent phonologically contrastive sites of articulation, and the symbols contained within the chart are intended to be sufficient to represent 'all the possible sounds of the world's languages' (International Phonetic Association (1989)). Various combinations of place of articulation complete the inventory of possible segments. 4.1.3 Articulators The active articulators of the human speech system are usually considered to be the lips, the tongue, and the velum. However, the muscles in the pharyngeal walls which play a role in swallowing can function to constrict and expand the pharyngeal cavity for the purposes of speech. The larynx of course contains musculature which adjusts the vocal folds for the articulation of glottal segments as well as phonatory gestures such as voicing. In this respect, both the pharynx and the larynx are potential places of articulation and articulators. This section deals with the tongue, pharynx and larynx musculature as each plays a role in post-velar articulationl. The tongue Within the (phonetic) tradition adopted by phonology, the tongue is divided into several zones (Ladefoged 1982, Catford 1988). The tip (apex) is the most anterior part of the tongue, followed by the blade (lamina), the portion of the tongue which lies under the alveolar ridge when the tongue is at rest2. The rest of the tongue is divided into the body (dorsum) and the root (radix). The dorsum in turn is usually divided into anterior and posterior sections, or front and back. The front of the tongue, therefore, is behind the apex iLabial activity is not directly related to post-velar articulation, but the contribution of lip gestures to emphasis is remarked on in some descriptions of Semitic emphatics (Mitchel 1956, McCarus and Yacoub 1962, Blanc 1953, Maamouri 1967). 2Keating (1991) defines the blade of the tongue as the tip plus the first 2-3 cm beyond the tip. The exact length of the blade is not relevant for the discussion here.  53 and blade, and normally lies beneath the hard palate, while the back or posterodorsum articulates with least effort in the velar and uvular zones. The root is the most posterior portion of the tongue, forming the anterior wall of the pharynx. It is the least mobile part of the tongue, having some direct attachments to surrounding structure (Daniloff 1973) Crucially, tongue muscles are connected to the epiglottis, the hyoid bone, the jaw, parts of the soft palate and so on. Thus, tongue movement can affect the dimensions of both the oral and pharyngeal cavities either discretely or simultaneously. A consideration of the basic musculature of the tongue reveals the complex interaction of tongue movement and general vocal tract configuration. Useful treatments of this topic can be found in Perkell (1971), Laver (1980), Borden and Harris (1984). I review the major issues here, since the terminology is useful and an understanding of muscle interaction is crucial to an accurate description of articulation in general. Tongue musculature The major intrinsic (i.e., with no external attachment to another structure) tongue muscles are those that flatten the tongue (vertical muscle), those that narrow and elongate the tongue (transverse lingual muscle), those that curl the tip and edges of the tongue and possibly shorten it (superior longitudinal muscle) and those that pull the tongue tip down or shorten the tongue (inferior longitudinal). The latter muscle has its point of origin in the tongue root. Basically the longitudinal muscles bunch the tongue from front to back when contracted. The major extrinsic (i.e., with an external attachment) tongue muscles are the genioglossus, the styloglossus, the hyoglossus and palatoglossus muscles. The genioglossus is the largest tongue muscle. It is described by Laver (1980:53) as 'the big, fan-shaped vertical muscle that makes up the bulk of the body of the tongue, running from the inner surface of the jaw, at the chin, backwards to the hyoid bone and upwards to the dorsum of the tongue. Its action pulls the body of the tongue forward,  54 when the jaw is in a fixed position'. Daniloff (1973) reports that it can also pull the tongue downward into the jaw, making the tongue concave. The styloglossus is externally attached to the styloid process of the temporal bone in the skull, and runs down to insert into the sides of the tongue where it spreads out along the tongue almost as far as the tip. Contracting the styloglossus lifts the tongue upwards and backwards, and may also lift the edges of the tongue. The hyoglossus, as its name indicates, is externally attached to the hyoid bone. It then runs upwards and enters the tissue at the tongue root and mixes with the palatoglossus muscle. The hyoglossus acts to lower and retract the body of the tongue. The palatoglossus muscle originates in the soft palate and runs down the back and sides of the mouth and into the sides of the tongue, forming what are commonly referred to as the front faucal pillars. This muscle then blends in with the styloglossus and hyoglossus muscles. Contracting the palatoglossus either raises the tongue body or lowers the velum. In general, the styloglossus and palatoglossus muscles lift the tongue upwards and backwards, while the hyoglossus and genioglossus retract and lower the tongue.  55 The basic tongue muscles and their external connections are diagrammed in Figure 4.1, taken from Laver (1980:52).  Schematic diagram of the action and location of the lingual muscles 7. Hyoglossus muscle i . Styloid process 8. Middle pharyngeal z. Styloglossus muscle constrictor muscle 3. Palatoglossus muscle 9. Genioglossus muscle 4. Superior longitudinal muscle 5. Inferior longitudinal muscle ^i o. Geniohyoid muscle 6. Transverse lingual muscle^i i . Hyoid bone  Figure 4.1: Tongue musculature (Laver 1980:52) The pharynx Although the pharynx is not traditionally considered an articulator, it does possess a musculature which can assist in the articulation of sounds that rely on altering the dimensions of the pharyngeal cavity. In general, the oropharynx can be viewed as composed of an inner and an outer layer of muscles. The outer layer of muscles includes the superior, middle and inferior constrictors. The superior constrictor, at the top end of the pharynx, may assist in elevating the velum. Contraction of this muscle also tends to lift the tongue root upwards and backwards since at least some fibres of this muscle originate in the lower sides of the tongue. The middle pharyngeal constrictor originates from the hyoid bone and runs up  56 and backward to encircle the pharynx. Its contraction narrows the oropharynx, lifting the hyoid bone slightly in the process. Finally, the inferior constrictor muscle originates from the cricoid and thyroid cartilages. It is the strongest of the three pharyngeal constrictors. Its contraction can narrow the lower pharynx and upper larynx. It can also pull the larynx upwards if it is not fixed by the action of other muscles (e.g. the infrahyoids). The inner muscles of the pharynx include the palatopharyngeus and the stylopharyngeus. The rear faucal pillars, or palatopharyngeus muscles, are considered to mark the beginning of the pharynx proper. This pair of muscles originates in the velum and runs down the walls of the pharynx until it reaches the back end of the thyroid cartilage which houses the larynx. The palatopharyngeus acts in conjunction with the palatoglossus (the front faucal pillars) to pull the velum downwards. These muscles may also contract like a sphincter, thereby pulling the faucal pillars together and reducing the dimensions of the oropharynx. Contraction of the palatopharyngeus can also assist in raising the larynx, thereby shortening the pharynx and overall vocal tract length, as well as reducing the volume of the pharyngeal cavity. This latter effect is commented on by Pike (1943:123-124). He suggests that faucal approximation is accompanied by constriction in the lower pharynx as well as glottal tension and raising of the larynx. The second set of inner pharyngeal muscles is the stylopharyngeus, which is paired (like the faucal pillars) and runs from the temporal bone in the skull into three structures: the constrictor muscles mentioned above, the palatopharyngeus (front faucal pillars) and the back of the thyroid cartilage. Contraction of the stylopharyngeus pulls the pharyngeal walls and the larynx upwards (via the connection with the thyroid cartilage) or else it can widen the pharynx if the larynx movement is countered by the infrahyoids.  57 Figure 4.2, from Laver (1980:71) shows the front (palatoglossus) and rear (palatopharyngeus) faucal pillars, with the latter anchored to the thyroid cartilage. 10  6  i. 2. 3. 4. 5.  Schematic diagram of the action and location of the velopharyngeal muscles, seen from behind (adapted from Van Riper and Irwin 1958) Soft palate^ 6. Palatal tensor muscle Tongue^ 7. Palatal levator muscle Thyroid cartilage^8. Azygos uvulae muscle Palatogiossus muscle^6. Hamular process of the Palatopharyngeus muscle ^pterygoid bone lo. Skull  Figure 4.2: Velopharyngeal muscles (Laver 1980) The larynx Finally, there are aspects of laryngeal anatomy which are pertinent to a discussion and understanding of pharyngeal articulation (see Sawashima and Hirose (1983) for a technical discussion). The larynx is surrounded and protected by three cartilages. The thyroid cartilage (Adam's apple) is the largest of the three. Both the true and the false (ventricular) vocal folds are attached to the inside front surface of this cartilage. Beneath the thyroid is the cricoid cartilage. It too serves to protect the larynx. The two cartilages are linked by the cricothyroid muscles. At the back of the cricoid cartilage, the arytenoid cartilages are  58 located. The posterior ends of the true vocal folds are attached to the lower ends of the arytenoids, while the posterior ends of the ventricular folds are attached to the upper portion of the arytenoids. The large thyroid cartilage is also connected to the hyoid bone by the thyrohyoid muscles and ligaments. Recall that the thyroid cartilage also anchors tongue muscles (hyoglossus) and pharyngeal muscle (middle and interior constrictor) as well as the stylopharyngeus and palatopharyngeus (rear faucal pillars). The muscular coordination involved in positioning the vocal folds for the various phonation types is complex and delicate. A lengthy and detailed discussion can be found in Laver (1980:99 ff.). The major point of interest here is the possible contribution of the ventricular or false vocal folds to pharyngeal(ized) articulation. Both the true and false vocal folds are part of the thyroarytenoid muscle- so named because it anchors in both the thyroid and arytenoid cartilages as described above, with the true vocal folds connected to a lower portion of the arytenoids than the false vocal folds. There is, therefore, a certain "depth" to the vocal folds, with the false vocal folds in a superior position to the true ones. The upper portion of the thyroarytenoid muscles, these so-called false or ventricular vocal folds also have a slightly different composition from the lower, true vocal folds. The ventricular folds contain less muscle fibre, being described by Kaplan (1960, quoted by Laver 1980) as soft and flaccid, and they are covered by mucous tissue. In fact, Kaplan describes them as thick rounded folds of mucous membrane. Because they contain few muscles fibres, the contribution to phonation made by contacting the thyroarytenoid muscle proper is inefficient in the case of the ventricular folds. An extreme setting of the thyroarytenoid is needed to get the ventricular folds to vibrate regularly. Possible ventricular articulations are discussed in sections 4.2.6 and 4.2.7.  59 4.2 Articulatory descriptions of post-velars In so far as it is rational to describe the articulation of segments independent of languages in which they occur, the following articulatory generalizations can be offered for post-velar segments. 4.2.1 Uvular stops /q, According to Ladefoged (1982:148), 'Uvular sounds are made by raising the back of the tongue towards the uvula'. A uvular stop then involves contact between the posterior part of the tongue dorsum and the uvula, or extreme end of the soft palate, as is made explicit in Catford (1988:98): 'The extreme back of the tongue is in contact with the uvula and the extreme back of the velum' (see also Catford 1977:160, where he says that the dorso-radical surface of the tongue may be involved). Judging from this description, the tongue is the active articulator and the uvula the passive one. While this is generally the case, the uvula itself contains the azygos uvulae muscle, the action of which may be involved in sealing off the velar-pharyngeal port. It is not inconceivable that the action of this muscle may also lower the uvula towards the retracting tongue dorsum in the formation of uvular segments. Constriction of the faucal pillars can also bring the uvula closer to the tongue dorsum, but the primary gesture in the production of uvular stops is the raising and backing of the tongue dorsum to contact the extreme back region of the soft palate. Catford (1988) describes uvular stops as having a more 'sloppy' release than velars, since the tongue dorsum must pull away from the soft and irregular surface of the back velum and uvula rather than the relatively firm and concave surface of the velum. This tendency is often manifested by phonetic fricative release of uvular stops. This effect is certainly evident in spectrograms of Interior Salish, and is reported by Matisoff (1973) for the aspirated uvular stop of Lahu also. There are cases of allophonic variation between uvular stops and fricatives in a number of languages. Elorietta (1991) develops  60 these phonetic facts in support of an argument that uvulars are halfway between velars (which show stable continuancy values) and pharyngeals (which do not bear distinctive continuancy values) in showing a ready tendency to lose distinctive values for continuancy. 4.2.2 Uvular fricatives /x, g/ Uvular fricatives are also produced by moving the posterior tongue dorsum upwards and backwards towards the uvula, but with a 'small central channel' (Catford 1988) between the tongue and the uvula which allows the airstream to pass through. Widening this channel produces an uvular approximant. Uvular fricatives may tend towards a trill because the small uvula is easily vibrated by the fricative airstream. 4.2.3 Uvular trill /R/ The uvula vibrates 'in a deep longitudinal groove ... formed in the back of the tongue'. The tongue shape is thus modified from the 'relatively flat or convex' configurations for the corresponding fricatives (Catford 1988:99). 4.2.4 Pharyngeal stop Said to be next to impossible to produce by Ladefoged (1975:143)3, but see the discussion of epiglottal stop in section 4.2.6. 4.2.5 Pharyngeal fricatives/approximants /h, S/ Pharyngeals are described by Ladefoged (1975:143) as 'pulling the root of the tongue back towards the back wall of the pharynx', with /S/ often accompanied 'with a great deal of laryngealization, perhaps because the necessary constriction in the pharynx also causes a constriction in the larynx'. Ladefoged (1982:149) rephrases this to state that 3Ladefoged (1982:149) rephrases this to say 'many people cannot make a stop in this position'.  61 the epiglottis is pulled back towards the rear wall of the pharynx. The distinction rests on whether or not the epiglottis can be considered an articulator independent of the tongue root. Laufer and Condax (1979) argue that the epiglottis is an independent articulator, Laufer and Baer (1988) retract this and say that it is not. Catford (1977, 1983) distinguishes two types of pharyngeal articulation: faucal or transverse pharyngeal as opposed to linguo-pharyngeal. The former is produced by lateral compression of the faucal pillars high in the oro-pharynx, possibly with concomitant raising of the larynx. Catford (1977, 1988) suggests trying to touch the uvula, thereby triggering the 'gag' reflex, as an approximation of the gesture for faucal or transverse pharyngeals. Catford (1977) considers this the most common articulation of Arabic /h,V. Linguo-pharyngeals are described as articulation where 'the root of the tongue, carrying with it the epiglottis, moves backwards to narrow the pharynx in a frontback dimension' (Catford 1977:163). It is considered a lower articulation than that produced by transverse pharyngeal constriction. Catford's terminology obviously reflects an attempt to distinguish two types of pharyngeal articulation- one relatively high in the pharynx, the other relatively low. While linguo-pharyngeals may better be described as epiglottal, as Catford himself later concedes, the IPA (1989) does not fully recognize the possibility of two distinctive sites for pharyngeal articulation. The IPA provides symbols for epiglottal segments, but outside the bounds of the chart proper. 4.2.6 Epiglottal stop I? / What is termed an epiglottal stop here (following revised IPA terminology), Catford (1988:101) describes as complete closure in the pharynx 'formed by folding the epiglottis back, as in the act of swallowing'. Catford (1988) remarks that some dialects of Arabic as well as what he calls Oriental Hebrew have a pronunciation of the 'amn (ICI) as  62 an epiglottal stop. He further comments that it is found in several Caucasian languages, where in Chechen, Ingush and Batsbiy it contrasts with fit. In earlier work Catford (1977) considered this type of epiglottal articulation to be laryngeal in the sense that it involves a ventricular (false vocal folds) place of articulation, albeit distinct from glottal (involving the true vocal folds). Thus Catford (1977:105) suggests that a ventricular stop can be formed by intense constriction of the upper larynx, though it likely cannot be made without simultaneous glottal stop. Plain glottal stop is distinguished from ventricular (plus glottal) stop in that 'the latter involves a considerable feeling of upper-larynx constriction, which is quite absent from the former' (Catford 1977:105). That ventricular/epiglottal stop (also termed 'strong glottal stop') cannot be dismissed as an allophone of glottal stop is argued for by the contrast between glottal stop and ventricular stop in some of the Caucasian languages, mentioned above. Its existence is referred to in the Russian literature as 'pharyngealized glottal stop' (Gaprindashvili 1966, quoted by Catford 1977). It is possible that such a segment is better regarded as glottal stop with secondary pharyngealization or epiglottalization. The fact that in later work Catford does not use the term 'ventricular' and instead substitutes the term 'epiglottal', in line with other descriptions (such as the 1989 IPA chart) would indicate that a consensus has developed which considers these articulations as primarily epiglottal, although they may involve some ventricular activity. 4.2.7 Epiglottal fricatives /H, c / Such sounds, referred to as 'deep' or 'emphatic' pharyngeals are reported distinct from the higher pharyngeals AN in the Burkikhan dialect of Agul (a Northeast Caucasian language) by Catford (1983). Colarusso (1988) mentions a Bzhedukh (Northwest Caucasian) pharyngeal with breathy voice which he notes is made by approximating the epiglottis with the opening of the larynx. Catford (1977) also hypothesizes that the Classical Arabic pronunciation of 'amn (ICI) was likewise a ventricular/epiglottal  63 approximant, as evinced by some phonetic descriptions of medieval Arabic (see Semaan's (1963) translation of Ibn Sina's Risalah for example). Using the term ventricular (as discussed in section 2.2.6) Catford (1977) also reports a ventricular fricative trill in Abaza , and in Adyghe and Kabardian (all Northwest Caucasian languages) where the sound is used in Arabic loan words containing 'am. Finally, what Catford (1977:103) calls ventricular articulation in the Caucasian languages is said to impart a '"strangulated" nuance of pharyngealization' to adjacent vowels, whereas glottal stop and /h/ do not4. Catford's description links the Caucasian articulation with the strident, pressed epiglottal vowels in !X66. Ladefoged (1983a/b) discusses epiglottalization in !X6s5 (based on work with Tony Traill), saying that it functions like a phonation type, forming strident or "pressed" vowels. Such vowels are described as produced "by constricting the upper part of the larynx so that the posterior end portion of the aryepiglottic muscle nearly contacts the root of the epiglottis" (Ladefoged 1983b:183). !X68 speakers (and Traill) are reported to have enlarged muscular pads just above the larynx, presumably as a result of making this sort of sound. Ladefoged (1983b) refers to the !Xda phonation type as "ventricular" while at the same time describing their articulation as involving the epiglottis. It should be noted that while some studies of Semitic pharyngeals (Laufer and Condax 1979) have claimed that the epiglottis is an independent articulator, later work (Laufer and Baer 1988) suggests that it is almost impossible to separate tongue root activity from epiglottal activity, and it is not clear that the epiglottis is an independent articulator. However, based on the descriptions noted above, it may be that epiglottal activity can be a component of primary laryngeal activity (as in !!X(56 and possibly the 4Catford (1977) also mentions the contribution of ventricular voice to diplophonia, a condition where 'two voices' can be heard, one with higher pitch relative to the other. While this type of vocal ability is generally regarded as pathological in Western culture, 'double voice' is reportedly used by Tibetan monks when chanting certain mantras. Laver (1980) discusses 'ventricular dysphonia' or 'ventricular voice' as a pathological or paralinguistic phenomenon (expressing anger) but not as a phonological phenomenon. It is described in the voice pathology literature as the result of excessive laryngeal tension so extreme that the true vocal folds are no longer well approximated but the ventricular folds are. Ventricular voice is described by Plotkin (1964, also quoted by Laver) as 'characteristic deep hoarse voice'.  64 Agul 'deep' pharyngeals) as well as primary tongue root activity (as in the case of pharyngeals lacking a distinctive laryngeal component). One further variation on the epiglottal theme must be mentioned. Hess (1990), based on observations on glottal articulation made by Lindqvist (1969), suggests that the patterning of Arabic glottals with pharyngeals may be explained if their articulation involves the aryepiglottic folds (muscles which link the arytenoids to the epiglottis) as well as a purely glottal gesture. This effectively links glottal articulation with pharyngeal/tongue root anatomy, and so treats all guttural glottals with a phonetic (and potentially phonological) pharyngeal component. This view is formalized in Halle's feature geometry (Halle 1989), see section 2.3.1. 4.2.8 Glottal stop /2/ Catford (1988) and O'Connor (1973) define /?/ as complete closure of the vocal folds with a subsequent build-up of pressure below the glottis and explosive release. Ladefoged (1982) simply defines it as complete closure of the vocal folds. The supralaryngeal vocal tract may be in a configuration determined by adjacent segments. 4.2.9 Glottal fricatives /h/ Approximation of the vocal folds with the supra-laryngeal tract usually in the configuration of the following vowel (Pike 1943, Jones 1957, Abercrombie 1967, O'Connor 1973, Catford 1988). Shuken (1984( collects the various descriptions of /h/ as a voiceless glottal fricative (Malmberg 1963, IPA), voiceless or whispered vowel (Jones 1957, Abercrombie 1967, Pike 1943), glottal friction plus oral cavity friction (Ladefoged 1975, Pike 1943). Catford (1977) says that /It/ has cavity friction if supralaryngeal stricture is greater that glottal stricture, if the other way round, it has glottal friction.  65 If/ The voiced glottal fricative is described by Catford (1988) as a brief period of breathy or whispery voice (i.e. a reduced glottal opening compared to voiceless gestures, but not so close along the whole length of the vocal cords that the whole glottis vibrates, as in the case of fully-voiced sounds). 4.2.10 Summary In the abstract then uvular articulation is achieved by rearward movement of the tongue dorsum to approximate the extreme rear of the velum. This is fairly clear from Xrays of uvular articulation in Arabic (Ghazeli (1977), Giannini and Pettorino (1982) and also Caucasian (Colarusso 1988), but it must be remembered that such data are languagespecific and the precise site of uvular articulation will vary somewhat from language to language, as with other consonantal sites and depending to some extent on whether the language has contrasting velars of the same manner of articulation. Depending on how far back uvular articulation is, it will affect the dimensions of the upper pharynx, simply by virtue of the fact that the uvular site is located at the juncture between the oral cavity and the upper pharyngeal cavity. Phonologically this is reflected in the fact that phonetic uvulars can pattern as dorsal articulations on a par with velars, or as pharyngeal articulations on a par with primary pharyngeals. This latter grouping is attested by the Semitic guttural class, which includes uvular fricatives and /h,S/. The patterning of uvulars with velars as a class of dorsals is also attested in a number of languages. Both patternings are discussed in Chapter 6. From a phonetic point of view descriptions of ventricular/epiglottal as opposed to faucal/transverse articulation suggest that pharyngeals can be achieved by (i) contraction of the upper pharynx where the palatoglossus and palatopharyngeus muscles (front and rear faucal pillars) intersect with the oropharynx and/or (ii) rearward movement of the tongue root, carrying with it the epiglottis and thereby achieving a constriction lower in  66 the pharynx. The distinction between the two types of pharyngeals boils down to pharyngeal constriction achieved by internal and external pharyngeal musculature as opposed to that primarily achieved by tongue activity (retraction into the pharyngeal cavity). Noting that the rear faucal pillars, while they originate in the velum also extend down the walls of the pharynx until they reach the thyroid cartilage, one might expect upper pharyngeal constriction to bring with it a certain amount of general pharyngeal constriction. On the other hand a purely tongue root/epiglottal pharyngeal may not be so prone to upper pharyngeal constriction. In support of this, Ghazeli (1977) notes that the configuration for Tunisian A/ has a relatively large upper pharygneal area and maximal constriction low in the pharynx. Further, although Caucasian data make it clear that languages can make phonological use of the distinction between two types of pharyngeal constriction, what remains unclear is whether the laryngeal aspect of 'lower' pharyngeals in the Caucasian languages is secondary or primary. Given Caucasian contrasts such as found in the Burkikhan dialect of Agul /h, T, H, 1, ?, h, ?/ (Catford 1983) it would appear that we must distinguish between pure glottals (h,?) pure pharyngeals (h,S) and deep/emphatic/epiglottal/ventricular pharyngeals (i„ ?). As will be discussed in the next section, there is evidence from Arabic that its pharyngeals have a phonetic laryngeal component, but the phonological status of this aspect of pharyngeal articulation is disputed. McCarthy (1991) claims it is phonologically irrelevant, whereas Halle (1989) invests the laryngeal nature of Arabic pharyngeals into his feature geometry.  67 4.3 Natural language data In this section I consider how the language independent articulatory and acoustic descriptions of uvular and pharyngeal articulations correspond to real language data, specifically dialects of Arabic (with some reference to Hebrew and other Semitic languages where relevant), Caucasian and Interior Salish. This section collates available descriptive information on post-velar articulation from the various language groups mentioned. Much of the information contained here is provided for purposes of reference and comparison rather than argumentation. 4.3.1 Semitic There is so much dialectal variation among Arabic speakers, that it makes little sense to discuss consonant and vowel articulation as if there existed a general "Arabic". What is commonly called "Contemporary Standard Arabic" is a modern form of Classical Arabic (i.e. the language in which the Koran is written and formally read). It is the language of learning, education and public address (radio, television, theatre etc.). In this sense a general Arabic does exist, with the caveat that it is spoken slightly differently in the various Arabic speaking countries, and by the educated portion of the population. The vast majority of Arabic-speaking people use the vernacular of their region and here the dialectal variation is considerable. Dialects described as "contemporary" refer to that region's version of Contemporary Standard Arabic, the classically based language. Dialects described as "colloquial" refer to the vernacular of the area or specific religious groups. Arabic dialect studies identify about eight major dialectal regions: Syrian (Syria, Lebanon, Palestine/Israel, Jordan); Egyptian (the colloquial of the sedentary population of the lower Nile valley); Iraqi (ancient Mesopotamia); North African (Morocco, Algeria, Tunisia, Libya etc.); Maltese; Arabian Peninsula (Oman,Yeman,Kuwait, Saudi Arabia etc.); Sudanese and Central Asian. The material surveyed here is limited to examples  68 from the Syrian, Egyptian, North African and Iraqi regions5. Where I have been able to find information (in a European language) on Arabian Peninsula dialects (Saudi, for example) it is included. Maltese is a unique dialect both historically and synchronically. Modern Maltese lacks some of the phonemes of other Arabic dialects and I will not discuss it further. The inventory of Contemporary Standard Arabic is as in (1). Although dialectal inventories vary, they are almost always considered to vary with respect to (1), and for this reason it is useful to refer to it. (1) Contemporary Standard Arabic (McCarthy 1991) Lab Coronal^Emphatic^Velar Uvular^Phary Glottal b^d f^0, s^  X^h^h  z^z  1,r m n  Tables 4.8, 4.9 and 4.10 detail articulatory descriptions of dorsal, pharyngeal and emphatic articulation from at least one language in each dialect group. These tables and the ensuing discussion do not claim to be exhaustive: there remains a considerable degree 5The major sources I have consulted for the various dialect regions are: Syrian: Lebanese:Delattre (1971), Obrecht (1968) Palestinian: Blanc (1953), Card (1983), Herzallah (1990) Egyptian: Mitchell (1956), Harrell (1957), Erwin (1963,1969), McCarthy and Raffouli (1964), Khalafallah (1969), Gary and Gamal-Eldin (1982) Iraqi: al-Ani (1970), All and Daniloff (1972), Giannini and Pettorino (1982), Butcher and Ahmad (1987). N. African: Moroccan: Harrell (1965), Heath (1987), Abdel-Massih (1974) Tunisian: Ghazeli (1977), Talmoudi (1980) Arabian Peninsula: Saudi: Lehn (1967b) Classical Arabic: Gairdner (1925) Medieval Arabic: Semaan (1963)  69 of extremely interesting variation that would doubtless reward further investigation. It is my purpose here merely to give some indication of the type of variability in post-velar articulation that is attested cross-dialectally in the Arabic-speaking world. Dorsals The following general points can be made about the data in Table 4.8. First of all, there is some variability in the description of velar and uvular segments. The plosives /k/ and /q/, where dialects have both, seem to be distinguished by aspiration in the case of /k/, with /k/ also palatalizing in the environment of high front articulations. Some Gulf dialects (Holes 1990) show allophonic variation between such palatalized velars (/k,g/) and the (post-alveolar) affricates [tf, d3]. This would further distinguish velar plosives from uvular plosives. Many dialects have lost the /q/ of Classical Arabic altogether, replacing it with glottal stop in many cases, but also with a voiced uvular plosive [G] (Gulf dialects) or a velar, as in some Iraqi dialects. The fricatives symbolized /x/ and /g, y, g/ also vary between velar and uvular place of articulation. In some dialects they are clearly and unambiguously described as uvular in place, and pattern with the guttural class (uvulars, pharyngeals, glottals). In other dialects such as Palestinian, /x/ is described as velar Blanc (1953). According to Blanc (1953), velar /x/ takes emphasis, but all segments in Palestinian are reported to take emphasis, so this will not distinguish it from uvulars. Herzallah (1990) considers the Palestinian guttural class to consist of /K, x, y, h, 2, h/ where /K, x, y/ are considered to be back velars. This latter classification seems to be phonologically rather than phonetically based.  70 Table 4.8: Arabic dorsals 1 Seg't 1 Region^IDialect^li Articulatory description^II Alternations I k  Syrian  Lebanese  no info  Palestinian  post-palatal stop  kc /i,a,y k, / C  q  Egyptian  velar stop  k, /front V  Iraqi  vls. aspirated velar stop  ki /i(:)  N. Africa Tunisian  velar stop  Moroccan  vls. aspirated. velar stop  ArabP  Saudi  velar stop  Syrian  Lebanese  reduced pharyngeal area; tongue  k  dorsum at uvula  26  vls. post-velar stop  q, /emphasis  Egyptian  vls. unaspirated. uvular stop  2  Iraqi  vls. unaspirated. uvular stop  q,g,y,k  N. Africa Tunisian  reduced phar. area, tongue dorsum  Palestinian  pressed against uvula, narrowing of oropharynx, narrowest at level of epiglottis, slight larynx raising but no contraction of laryngo-pharynx. Moroccan  vls. unaspirated uvular stop, may be  2  glottalized ,  ArabP g,y„g` Syrian  Saudi  has lost /q/  Lebanese  constriction high in pharynx formed by a circular motion of tongue root, some trilling. vd. back-velar spirant  g, /emphasis  Egyptian  vd. uvular fricative  g, /emphasis  Iraqi  vd. uvular fricative  velar /i(:) q (nomadic)  N. Africa Tunisian  vd. uvular fricative  Palestinian  ArabP 6ICfasrsghab dialect.  Moroccan  uvular, like Fr. 'r'  Saudi  vd. uvular fricative  q (Hufaf)  71 x  Syrian  Lebanese Palestinian  Egyptian Iraqi N. Africa Tunisian  ArabP  Morrocan Saudi  vls. /g/; stricture narrower than for /g/ vls. velar fricative vls. (dorso-)uvular fricative vls. velar fricative vls. post-velar/uvular fricative like /g/ but not as far back uvular approximant vls. uvular fricative  x, /emphasis x, /emphasis Pharyngeals Within the class of Arabic pharyngeals, two major articulatory questions arise, both of which were raised in section 1.1. The first is whether tongue root retraction is the primary gesture, with concomitant general pharyngeal contraction or whether the pharynx is the primary articulator with secondary tongue root retraction. The second queries the role of the larynx in the production of pharyngeals. With respect to the first question, Erwin (1963, 1969, Iraqi Arabic) and el-Dalee (1984, Lower Egyptian Arabic) explicitly insist that pharyngeals are produced by posterior pharyngeal musculature to the exclusion of tongue-root retraction. At the very least, if the tongue root is involved, it plays a secondary role. Of interest is that Egyptian Arabic is described as permitting the spread of emphasis to all segments. It may therefore be of considerable significance that the primary pharyngeals are described as pharyngeal rather than tongue root articulations, thereby permitting the tongue-rooting (i.e. "emphasis") of pharyngeals, so to speak. It may be that dialects in which the primary pharyngeals are described in terms of tongue root activity do not have this option available, and so do not evince emphatic pharyngeals. Palestinian Arabic, which is also described as having emphatic pharyngeals would be another case in point, although there is no explicit mention in the literature that I have reviewed that Palestinian /S,h/ are only pharyngeal in articulation, and not tongue-root. That there may be two types of  72 pharyngeals (those which can be emphatic, those which cannot) which are phonetically distinct, has not been overtly discussed, to my knowledge. This matter requires further research. With respect to the second issue, the role of the larynx in pharyngeal articulation, it can be seen from Table 4.9 that alternation between the voiced pharyngeal and glottal stop, or creaky, allophones is fairly common. This alternation is attested historically also. The synchronic allophonic variation is non-distinctive, but suggests that pharyngeals are probably being articulated with some laryngeal component, either raising of the larynx or compression of the larynx due to the action of the middle and inferior pharyngeal constrictor muscles. The laryngeal component may be expected whether the relevant articulator is the tongue root or the pharynx since both the hyoglossus and genioglossus are attached to the hyoid bone, which is also the origin of the middle pharyngeal constrictor. Further, the strongest external pharyngeal constrictor muscle (the inferior constrictor) originates in the cricoid and thyroid cartilages. Thus even general pharyngeal constriction may simply bring with it some laryngeal side-effects. There is particular phonetic variability noted in the manner of Arabic voiced and voiceless pharyngeals. For example, Al-Ani (1970) reports that the most common allophone of /3/ in Iraqi Arabic is a voiceless stop, whereas Ladefoged (1982) doubts the possibility of a pharyngeal stop. Ghazeli (1977) refers to /V and /h/ as fricatives, but as noted by Butcher and Ahmad (1987), Ghazeli's spectrograms do not support this assumption for M. Klatt and Stevens (1969) suggest that /f/ is a sonorant along with /is/ in Lebanese Arabic at least. Butcher and Ahmad (1987), in an effort to make sense of the variant descriptions of pharyngeals as glides, approximants, stops and fricatives, investigated Iraqi Arabic pharyngeals both aerodynamically and spectrographically. Their findings are that /h/ is a voiceless continuant with high rates of air flow, high intensity noise in the waveform and a marked formant structure. /h/ did not have sufficient airflow to match the voiceless fricative profile that accompanied most tokens of  73 /h/. Based on airflow measurements /C/ is classified as a voiced approximant with glottal release in final position and a glottal burst in initial position some 60% of the time, accompanied by creak in almost all cases and with no evidence of friction in any tokens. In other words, the interpretation of A/ as a stop in Iraqi is better interpreted as a glottal gesture accompanying the pharyngeal constriction, possibly due to the extreme lowness of the articulation or the fact that some of the muscles involved in pharyngeal constriction may raise the hyoid bone and thus affect the orientation of the larynx in the throat. Table 4.9: Arabic pharyngeals 1 Seg't Region^Dialect Syrian  Lebanese  Palestinian Egyptian Iraqi  N.Africa  Tunisian  Articulatory Description  Alternations  C, /emphasis v. low stricture between tongue and pharynx wall; small pharyngeal cavity area, tongue dorsum higher and more fronted than for /a/; some larnyx lowering, dorsal bulge. Some creak. vd. pharyngeal fricative with some stop characteristics c, /emphasis (radico-)pharyngeal vd. phar. approximant, creaky with glottal release. Some tokens may be made at glottis. Constriction of muscles of upper throat, walls of pharynx contract. ?, creak vd. phar. fricative, contraction of laryngo-pharynx by tongue root retraction and pharyngeal wall movement; larynx raised and tense. Tongue shape like a pyramid with blade and dorsum straight and depressed. Large upper pharynx area. Most constriction 3.5 cm from glottis, at level of epiglottis. Fl raised, F2 low.  74 Moroccan  ArabP Saudi Medieval  h  Syrian:  Lebanese Palestinian  Egyptian  Iraqi  N.Africa  Tunis  Moroccan  ArabP Saudi Medieval  vd. phar glide with some glottal stop characteristics. Simultaneous raising of larynx and backing of tongue root (Harrell). Constriction lower and narrower than for /h/, tongue dorsum raised to velum (Boff Dkhissi). 214 vd. pharyngeal deep in the throat, where the air in vomiting is located, at the opening of the larynx. stricture lower and narrower than 1, pharyngeal cavity smaller than I. h, /emphasis vls phar spirant made in throat at point larynx rises to in swallowing, suggests retching, with root of tongue filling throat. produced entirely with throat action, tongue should not curl up or back (Erwin 1969); tongue root against pharyngeal wall (al-Ani 1970); pharyngeal approximant, larynx raised and maybe constricted (B&A 1987). as with /1/ but constriction narrower, slight depression of tongue root —1cm above epiglottis. heavy breath through light constriction; fricative with higher constriction than I (B off Dhkissi). vls pharyngeal where the air involved in clearing the throat is located (i.e. higher in the throat).  75 Emphatics The use of the terms "emphatic" and "emphasis" to refer to a subset of the coronal obstruents and their spreading effects in Arabic dialects is fairly recent. Traditionally the emphatic segments (in Classical Arabic) fall into two classes referred to by a number of terms. The term Wtbaq or mutbaqa is used by Sibiwayh (a prominent eighth century Arab linguist) and translates as "covered, lidded" (see Ghazeli 1977, Giannini and Pettorino 1983). Used only of A, 4, s, 0/, the term refers to the 'trapping' of these sounds between two places of articulation, one of which is a raising of the tongue toward the upper palate. While this description does not clarify whether the secondary articulation of z`baq sounds is velar or uvular, it is useful to bear in mind the following. The term 2isrlal ("raised") encompasses A, cl, $, 0/ and the dorsals /x, y, q/. It is presumably a reference to the fact that the tongue dorsum is raised towards the roof of the mouth in their articulation. The grouping together of dorsals/uvulars with retracted coronals in an articulatory description may suggest they shared a uvular site in the Classical pronunciation which Sibiwayh describes. While it can be seen (Table 1) that there is frequent confusion within the class of dorsals, it is noteworthy that the phoneme /k/ is never included in the class of 2isilal consonants. Tafxim or mufaxxama ("thickening", "heaviness") is a more recent term  introduced in 19th century and applied to the emphatics. It has been used to refer to the fortis nature of A, 4, s, 0/. It is sometimes used to encompass the four 2itbaq consonants, the uvulars and possibly /1,r,y/, but not, according to Ghazeli, the primary pharyngeals. Ghazeli (1977) interprets tafxim as an auditory term, referring to the thick or muffled quality of such sounds. The discussion of emphasis by modern scholars tends to be inconsistent with the traditional divisions discussed above. In some cases all post-velars (except glottals) are regarded as emphatic, in others, the classical divisions are referred to, with only the 2itbaq  76 consonants being emphatic. Both phonetic and phonological factors seem to have contributed to the confusion, quite apart from the fact that both factors can vary from dialect to dialect. It seems useful to reserve the term 'emphatic/emphatics' for the four 2itb aq consonants. These are sometimes referred to as the primary emphatics,  particularly in analyses which claim that the feature associated with the phonological spread of emphasis is underlying only on A, d, s, 0/ or some limited set of consonants. The terms pharyngeal and pharyngealized can then be reserved for /h,S/ and whatever their effects may be. This is contrary to some usage, such as Heath (1987), who refers to effects from ?itbaq consonants under the rubric 'pharyngealized'. In terms of production, there has been tremendous controversy about the proper articulatory and acoustic description of emphatics, with the lips, tongue dorsum, tongue root, epiglottis and pharyngeal walls all claimed to be relevant articulators. The acoustic argument over whether emphasis is velarization, uvularization or pharyngealization has generally resolved itself into a claim that emphasis is better described as acoustically similar to pharyngealization, meaning with a constriction somewhere in the pharynx (see Giannini and Pettorino (1982) for a discussion of Iraqi Arabic). Whether emphatics can be described as uvularized as opposed to pharyngealized remains unclear. Both phonetic claims exist in the literature. Instrumental investigations, while interesting and helpful, do not entirely solve the problem.  77 Figure 4.3 reproduces X-ray tracings for plain /t,s /, emphatic /t, s/ and /q/. (Giannini and Pettorino (1982) : Iraqi Arabic). It is clear that reduced pharyngeal cavity volume is common to the emphatics and /q/, and based on this one might describe the emphatics as uvularized. Giannini and Pettorino (1982) argue against this interpretation of their data however, preferring to consider the emphatics 'pharyngealized' on the basis of observed tongue root retraction which is claimed not to be present in uvular articulation. Giannini and Pettorino (1982) do not provide examples of pharyngeal articulation to compare with emphatic articulation.  Figure 4.3: Iraqi Arabic: /t/, /t/ and /q/ (Giannini and Pettorino 1982) Laufer and Baer (1988) in a fibrescopic study of emphatics and pharyngeals in Hebrew and a number of Arabic dialects, also claim that emphatics involve lower pharynx constriction. They found the greatest degree of constriction at the level of the epiglottis, with the tongue root the active articulator. They argue that emphatics are pharyngealized, that is, with less constriction than /h,f/, but definitely pharyngeal rather than uvular. Further, they report a variety of Muslim Arabic in which the tongue dorsum is pulled down in the velar region, presumably ruling out the possibility of uvular coarticulation for emphatics. Ghazeli (1977) provides tracings of his own production of Tunisian emphatics, which he interprets as having a place of articulation midway between uvulars and  78 pharyngeals, that is to say, with a general constriction midway in the pharynx. He does provide tracings of pharyngeal as well as uvular and emphatic articulation. When tracings of emphatic articulation are laid over tracings for uvular fricatives and pharyngeals, the closest match is between emphatics and uvulars. Pharyngeals have a larger upper pharyngeal volume than either emphatic or uvular articulation. Pharyngeals also have a more reduced laryngo-pharyngo volume than either uvulars or emphatics. This can be seen in Figures 4.4 and 4.5. In Figure 4.4 the leftmost tracing is of [x] in the word [xali] (solid line) and fg] in the word [ffali] (dotted line). The rightmost tracing is of [t] in [biSa] s(olid line) and [t] in [tbi:SEe] (dotted line). These tracings can then be compared with that of pharyngeal articulation in Figure 4.5.  Figure 4.4: Tunisian Arabic: /g, x/ and it, t/ (Ghazeli 1977)  79  Figure 4.5: Tunisian Arabic: /C/ before initiation (solid line) and during production (dotted line) in [Tx:li]. (Ghazeli 1977) An interesting approach to the difficulties of reconciling what appears to be contradictory data is Hess' (1990) factor analysis of X-ray data, which includes Ghazeli's Tunisian and some Damascan data. Her results may lend some credibility to the term 'pharyngealized' for emphatics in the sense that emphatics and pharyngeals were found to share positive amounts of her Factor A (retraction of tongue root at the epiglottis, raised larynx and rearward movement of the upper pharyngeal wall). The pharyngeals AN are separated from the emphatics by Factor B (lowering of tongue blade and dorsum, retraction of tongue root, raising of larynx) which achieves a low tongue body position. As coronals, the emphatics do not have a low tongue body position. As emphatics, they have reduced lower pharyngeal volume. Unfortunately, Hess does not include tracings of uvulars in her analysis, so it is unclear what emphatics might share with uvulars under a factor analysis. Table 4.10 charts the description of emphatic articulation in the same dialect regions for which dorsal and pharyngeal articulation have been presented. Bracketed  80 phonemes indicate emphasis as a derived feature, though sources are often very unclear as to what segments are underlyingly emphatic, and what are derived. Table 4.10: Arabic Em atics [Language i Dialect Syrian  Emphatic If Articulation  Lebanese b, d, t, z, secondary pharyngeal constriction. $, Ill, 4,1/ L. Palest'n t, p, (f), retraction of tongue, raising of dorsum to velum, advancing of lips, possible glottal constriction. m, 0, t, (Z), $, 1,  (r) Egyptian  t, d,^ z, (r)  Iraqi  t. 4, 0, (a), (1)  N.A.  ,  Tunis  t, $, 0  Vowels: pharyngeal constriction, lowering of tongue, retraction of tge, protraction of tge. Consonants: velarization, alveolarization of dentals glottalization of AA post-palatals with tge broadening at closure point. Lip protrusion. centre of tge raised,tge tense and retracted, some constriction of pharyngeal walls. Tge tip may be retracted. al-Ani's X-ray tracings show a flattened tongue body, with considerable narrowing in the mid-pharynx. Giannini & Pettorino report that the tge root is retracted at the level of the 3rd and 4th vertebrae, the hyoid bone is slightly raised, the larynx is not raised. Ali and Daniloff report tongue root retraction, with overall tongue shape distinguishing between plain and emphatic consonants. back of tge moves to pharyngeal wall at level of 2nd vertebrae midway between place of articulation for /q/ and /h/, palatine dorsum of tge depresses, rearward movement of epiglottis, no laryngeal displacement, no lip-rounding, no contraction of laryngo-pharynx.  81 Mor  ArabP Saudi Medieval  t, ci, s, (r) /b,m,z,l/ may become emphatic  lower pitch,greater muscular tension, raising and backing of tge in roof of mouth.  t, d, s, z  larger surface of tge in length and breadth, hollow in tongue surface, lateralization Glottals In all the dialect regions surveyed here, /2/ is described as a voiceless glottal stop or 'catch in the throat' (Ferguson et al. 1961). /h/ is referred to as a voiceless glottal spirant. Care is taken in some descriptions to contrast glottalic with pharyngeal constriction. Thus Erwin (1963, 1969) states that /h/ in Iraqi Arabic has no closure or stricture elsewhere in the vocal tract and that the throat is 'relaxed'. Gairdner (1925) makes the same kind of comment for the pronunciation of Classical Arabic. None of these sources record lowering effects on vowels from glottals. Alternation of PI with zero is noted in a number of dialects, for example by Blanc (1964) for Iraqi Arabic and by Garbell (1958) for Eastern Mediterranean dialects. Blanc (1953) reports /h/-->0/-# for North Palestinian. Vowel effects Accounts of vowel quality in Arabic reflect a variation parallel to that found in consonantal articulation. This is not surprising since vowel quality is clearly affected by consonantal environment. The basic variation is tabulated in Table 4.11. Gaps in the table indicate lack of specific information in the sources consulted. The effects of adjacent uvular and pharyngeal consonants are always local and affect only vowels, whereas emphatic effects are long-distance and include consonants  82 and vowels. In general, the effect of uvulars and emphatics is considered roughly comparable in phonetic terms, though it is always stressed that emphatic effects are the most striking. Effects from pharyngeals seem to vary from on/off glides to centralization of vowels. It is noteworthy that Harrell (1957) describes the pharyngeals of Egyptian Arabic as having no backing effect on the /a/ vowel. The phonetic realization of this vowel is recorded as roughly [m] in pharyngeal environments as opposed to [a] in uvular and emphatic environments. This is sometimes interpreted as a 'fronting' effect, though it is not entirely clear that the vowel is being fronted by the pharyngeal. This may be a consequence of two factors: (i) the pharyngeal is simply not backing the vowel, and so what emerges is a quality otherwise associated with pre-uvular contexts (ii) it may be that the pharyngeal in question is so low in the context of /a/ that F2 is raised, creating the fronted quality relative to uvulars and emphatics. The mechanics of this possibility are discussed further in Chapter 5, but basically the explanation is this: a constriction less than 2 cm from the glottis is close to an F2 velocity minimum, constriction in this location will raise F2, creating a fronting effect. Trubetzkoy (1969:124) makes similar comments about palatalization effects from pharyngealized consonants in Caucasian, as will be discussed in section In general I would comment though, that until descriptions of subtle variation in vowel quality are investigated acoustically, they will remain difficult to assess. This is not to say that researchers have transcribed incorrectly, but there can be no question that (i) transcriptions are filtered through the phonetics of the transcriber's native language and (ii) the sound in question may contain a number of phonetic cues with the transcription reflecting a decision based on only one of them. While spectrography does not solve all of these problems, it can frequently provide more accurate representations of vowel quality. I wonder, for instance, if the [x] in the environment of Egyptian pharyngeals (Harrell 1957) is as high and front as that found in the context of pre-uvulars.  83 Table 4.11: Arabic Vowel Effects [Region^II Dialect Syrian  Leb  I UR i i:  NPal  Pal  Egypt  'Plain II Emphatic^1 Uvular^II Phary. t i:  Obrecht  centralized centralized and lowered backed backed u u: centralized  a m: u u: i,t u,u x,a i: i e: e m: s  a i: i e:, e a:, a: a  0:  0:5  0  o  u: U i: e: u:  u: u i: e: u:  defective distribution u: u, as /q  o: a: i  o: m:, mr I,i  i  e  U  0,u  u  a  x,x'"^_  a  o a,x  a a: U u: i U a i: i e: e a: a o:  0 Source^I  a o/O a  Blanc  Card  I e:'5 u:o:  i: + glide Harrell glide u:- + glide o: + glide  a:  84 Iraqi  N. Africa Susa7  Mor8  Class9  i  t.,i  i  L,E  i: e: a  i: e,c  a i,ia c,e  c,e  E,M  a,o  a  a:  x:,a:  a,o  o: U u:  o: u,u u:  0 ir tr  i a U i U a i: i u: U a: a  i c u i u x i t u u x:  ii a o e o a/a, glide glide u u a  Erwin  u  ii a o e o a  ii a o t u a  Talmoudi Heath  Gairdner  x  7A Tunisian dialect included here because Talmoudi gives more precise phonetic notations for allophonic effects that does Ghazeli (1977). Ghazeli (1977) also says that for his dialect the true pharyngeals have only a minor coarticulatory effect on adjacent vowels which does not change steady-state vowel quality. He reports all vowels backed adjacent to emphatic coronals. 8Harrell (1962) also describing Moroccan Arabic, gives the following alternations: Ii/^fel/emphatic C, uvulars and pharyngeals^/a/^[a]/ uvulars, pharyngeals,# [i]] elsewhere^ [a]/ emphatic C [m] elsewhere /u/^[ti )141,x,R,tiS^ [01/emphatic C, q [u] elsewhere 9McCarus and Yacoub (1962) describing the literary Arabic used in Syria (i.e. educated speech based on Classical Arabic) note similar allophonic variation, with both long and short /u/ seemingly impervious to the presence of back consonants or emphatics. They do note however that the "backed" allophone of all vowels is not so back in the environment of uvulars and pharyngeals as it is in the environment of emphatic s .  85 4.3.2 Caucasian Dorsals Velars and uvulars in Caucasian seem generally to be velar and uvular in place, although there is some interaction between velars and palatals. The rounded velars found in all Northwest Caucasian languages are argued by Colarusso (1988) to be velar in place, with the distinction with rounded uvulars made by place, and not the affricated release which occurs with plain uvular plosives. X-rays of Ubykh /x/ show greatest tongue approximation to the uvula and reduced volume in the upper pharynx. This aspect of uvular articulation appear to be crucial in the distinction between palatalized uvulars, plain uvulars and pharyngealized uvulars such as found in Ubykh. Palatalized uvulars show no reduction of pharyngeal volume. Colarusso describes the tongue root as advancing for this articulation, so that the tongue bunches in the mouth. Pharyngealized uvulars on the other hand, show evidence of tongue root retraction to the extent of pushing the epiglottis back against the rear wall of the pharynx. Pharyngeals As mentioned above (section 4.2.4 ff), the Burkikhan contrasts /h, 1, H, C', ? , h, 2/ (Catford 1983) argue that languages can distinguish among two types of pharyngeals as well as glottals. The Caucasian languages seem to be unique in using this degree of differentiation in the pharynx-larynx region. Arabic and Interior Salish both contrast pharyngeal with glottal articulation, but not two types of primary pharyngeals. The terms 'strong glottal stop' and `pharyngealized glottal stop' (Catford 1977) both suggest the participation of the larynx in the low, epiglottal pharyngeals. Catford's early description of them as ventricular also emphasizes their laryngeal component.  86 Pharyngealized consonants and vowels Pharyngealized consonants do not occur in all Caucasian languages. The coreinventory, as represented by Kartvelian for instance, does not contain CS (Catford 1977b) However, pharyngealization occurs on labials and uvulars in Ubykh. Two other Northwest Caucasian languages, Abaza and Ashkarwa Abkhaz have pharyngealized uvulars. Such segments are relatively common in the North East Caucasian languages, being reported in the Lakk-Dargwa subgroup, Kubachi and some dialects of Dargwa, the Lezgi subgroup and the Avaro-Ancli subgroup (Colarusso 1988). As mentioned in the discussion of Caucasian uvular articulation, X-ray data of Ubykh pharyngealized /x/ shows backing of the tongue into the pharynx with the tongue tip curled upwards. Bzyb pharyngealized /x/ shows no such tongue tip action, but similar retraction of the tongue root with the epiglottis pushed against the pharyngeal wall. Chechen (Northeast Caucasian) pharyngealizes its labials, dentals, alveolars and palatals. Ingush and Bats (Northeast Caucasian) also pharyngealize labials, as do some dialects of Lakk and possibly Udi (Colarusso 1988). Pharyngealization is also reported on vowels in some of the Lezgian languages. I deal here with the effects of pharyngealized consonants first, excluding pharyngealized uvulars since they have already been discussed. Acoustic investigation of the pharyngealized consonants of Chechen by Kingston and Nichols (1986) reveals that Fl rises consistently, with lowering of F2 and F3 dependent on the primary place of articulation of the CS. F2 does not lower with pharyngealized bilabials, a result which is explained by the already low F2 conditioned by a labial place of articulation. F2 is higher for alveolars, and so shows the effect of pharyngealization by lowering. Based on such an acoustic profile, Kingston and Nichols (1986) argue that pharyngealization in Chechen is accomplished by constriction in the lower pharynx, with a resultant compact (F1 close to F2 and a lowered F3) as opposed to  87 flat (all formants lowered) spectrum. In support of this, they note that the acoustic effects of CC and the pharyngeal consonants of Chechen are identical. Colarusso (1988:132) reports that pharyngealization of Ubykh (Northwest Caucasian) labials (/pC, bl, phS, vl, wS, mS/) results in consonants of a longer duration, with an added formant centering around 500 Hz and 'the occasional addition of a pharyngeal formant around 1100 Hz'. Colarusso further notes (footnote 4, p174) that 'These formants coalesce with the first formant of an adjacent vowel to produce a very wide and loud first formant. Some workers have interpreted this as a raising of the vowel, since high vowels have low first formants'. Under this interpretation pharyngealization is called 'emphatic softening' or 'emphatic palatalization' (Trubetzkoy 1969: 131). Trubetzkoy assumes that the tongue body actually fronts during the articulation of Caucasian pharyngealized consonants, thus explaining what he calls an opening effect on vowels: i-->e, a-->m, u-->ä. This effect is basically a centralizing one, as can be seen very clearly in Figure 4.6, which plots Fl and F2 values given in Catford (1983) for VC in Udi, one of the languages specifically mentioned by Trubetzkoy as having emphatic palatalization effects. The Udi vowels are /i, e, a, o, u, il, eS, aC, oS, uC/ with pharyngealized variants represented as capital letters in this figure.  88  U  §A  a -2000^-1500  -1000  Figure 4.6: Udi VC (Catford 1983) Catford does not specify at what point in the vowel these readings are taken or whether the vowels themselves are derived or phonemic, but Colarusso (1988:174) notes that analyses of these segments (in the Russian literature) argue that the vowels bear pharyngealization rather than deriving it from consonantal environments. X-ray tracings of VI in both Udi and Tsalchur (Gaprindashvili 1966, reproduced in Catford 1983) show tongue root retraction at the level of the epiglottis, a depression in the tongue dorsum opposite the uvula and a bulge further forward in the tongue. This configuration is very close to that of American English In (Delattre and Freeman 1968, Ladefoged 1979). Glottals Caucasian /2,h/ are considered laryngeal articulations along the lines of those found in other languages, with the exception of the epiglottal/ventricular articulations  89 already mentioned. /h/ can be voiced intervocalically. None of the glottals are recorded as having coarticulatory effects on vowels. Vowel effects Descriptions of Caucasian vowel allophony are strikingly reminiscent of those found for Interior Salish. Kuipers (1960:22-24) charts front, central, back, higher and lower variants of the short vowels /a,a/ in Kabardian, with choice of variant dependent on consonantal environment. Thus: 'Front variants (i,e) are found after lateral, palatalized palatovelars, uvulars and laryngals (after the latter two consonants central vowels are also heard), back rounded variants (u,o) after labialized palatovelars, uvulars and laryngals, central variants (a,ce) after other consonants. Before labialized consonants halfrounded vowels are found, central (0, 0) or back (u,o) depending on what precedes.. .Before the pharyngeal h there is no distinction between a higher and a lower vowel....(a:) in most cases produced as a front a; back variants are found in the neighbourhood of uvulars and pharyngeals'. Colarusso (1988) reports that all tautosyllabic consonants colour vowels, with considerable variability from speaker to speaker and with some variation in default vowel height from language to language. Thus the vowels in Circassian tend to be higher than those in Ubykh or Abaza. However, Colarusso's description of coarticulatory effects from consonants to vowels in Ubykh is mostly very similar to Kuipers' description of Kabardian, given above. That is to say, anterior consonants condition front vowels, high consonants condition high vowels, back consonants produce back vowels. Colarusso reports the pharyngeal behaves as a back consonant for purposes of vowel allophony10. With respect to laryngeal effects, Colarusso has the following telling comment: 'To determine the independent articulatory positions of /a/ and /a/ one may elicit them either formant values, Colarusso gives [a] in /qa/ sequence with Fl 800 F2 1400, but I know of no place in the text where he gives formant values for pre-uvular articulation. With respect to the 3 vowels which he transcribes as /a,a,a:/ Colarusso shows a formant plot for one Ubykh speaker which shows Fl for /a/ higher than Fl for /a/, and /a:/ with a slightly higher Fl than /a/. There is considerable overlap, however.  90 in isolation, in the environment of a laryngeal, or next to a labial, providing in the last case that the informant utters the word in an isolated form and not in connected speech so that the formant transitions due to the labial consonant can be ignored. In such environments other factors affecting the position of the tongue are absent so that it may assume the positions inherent in /a/ and /a/ themselves' (p. 299). Thus, laryngeals have no coarticulatory effect on vowels. There is a complication however. Contrary to Kuiper's statement that the distinction between high and low vowels is neutralized in the environment of pharyngeals (and the tendency for fieldworkers to report lowering of vowels in the environment of pharyngeals), Colarusso (1988:336ff) claims that Caucasian pharyngeals absolutely do not lower vowel quality, and may even raise vowels. He cites Fl lowering from the pharyngealized uvular of Ubykh as an example and attributes this affect to the fact that the Caucasian pharyngeals are not articulated with dorsal gestures, and are in this sense similar to laryngeals and labials. This claim might be compared with Choi (1990) on the coarticulatory effects of Kabardian consonants on vowels. Analysing Kabardian with /i, a, a:/ Choi's plot of the steady-state values of /a/ shows it lowest in the environment of pharyngeals so that it approaches the value of /a:/ in the environment of velars and palatals. Uvulars lower all vowel qualities, palatals raise and front, labials raise and back, velars raise the /a:/ vowel and back /i, a/. In general Choi's findings support Kuiper's impressionistic account of Kabardian allophony and contradicts Colarusso's claim that pharyngeals do not lower vowels. However, Colarusso's investigations are based on Ubykh, not Kabardian, and Choi shows pharyngeal effects only on the /a/ vowel. Clearly, further instrumental work is needed on these languages to assess exactly what range of coarticulatory effects consonants may have on vowels, and what the acoustic cues are for the extensive post-velar inventory.  91 4.3.3 Interior Salish As discussed in chapter 1, some of the languages of Northwest North America present inventories with uvular and pharyngeal consonants. The most extensive postvelar inventory is found in Moses-Columbia Salish. The other Interior Salish languages do not appear to have both voiced and voiceless pharyngeals, but otherwise have comparable inventories with respect to post-velar places of articulation. As with the Semitic material, descriptions of Salish pharyngeals vary somewhat from language to language. In particular, the degree of constriction seems to vary, resulting in fairly fricative-like sounds to very approximant-like sounds. Tables 4.124.14 present the available descriptions of Salish uvulars, pharyngeals and retracted coronals along with the quality of contingent vowels. Glottals are not consistently noted as having an effect on vowel quality. It is sometimes noted that vowels followed by glottals are not as high as they are before pre-uvulars. This might be interpreted as a lowering effect, but this can be argued against on both phonetic and phonological grounds (see section 5.2 and chapter 6) For now, suffice it to note that glottals are never reported to have effects comparable to those found in the context of uvulars and pharyngeals. Likewise, glottals in Caucasian and Semitic are not included in the class of post-velars for the purposes of vowel effects. Dorsals The following generalizations may be drawn from Table 4.12.11 /q/ is generally agreed to be uvular in place of articulation but the potential for non-native confusion between uvulars and velars is noted. Uvulars in all Interior Salish languages are 11The major sources for Interior Salish as represented in this table are: Thompson and Thompson (1992); Kinkade (1967) N4e7kepmxcin (Thompson) van Eijk (1985); Kinkade (1967) Lillooet Kuipers (1974, 1989); Gibson (1973) Shuswap Mattina (1973); Kinkade (1967) Colville-Okanagan Carlson (1972); Kinkade (1967) Spokane Reichard (1938); Johnson (1977); Kinkade (1967) Coeur d'Alene Kinkade (1967) Moses-Columbia  92 consistently reported to condition vowels of a lower quality than found in pre-uvular environments. Here, allophonic effects are reported from following consonants because these produce the strongest and most predictable effects. Table 4.12: Interior Salish uvular effects Lang Articulatory description  /V/  /-preI uvulars  lit /e/ /u/ /a/ /i/ /e/  [i,i'] [2] [u,u' ] [i,a,t,u]  [m] [o] [A]  [e] [x]  [E] [a]  [0] [a] [i]  [o] [A]  [ NI  Lt  Sh  Cr  MsCm  CoOk  Sp  uvular place, but may be fronter depending on speech community  uvular place, but close to velars; /q',q'w/ with affricated release that velar counter-parts lack  /a/ /a/ uvulars described as "post-velar" in Ii/ place /e/ /u/ A/ q-series called 'velar' ,but uvular in place, k-series called 'palatal'. uvular  post-velar  unrounded q-series produced 'v. far back in the mouth' but rounded velars and uvulars easily confused by non-native speakers  /-uvulars [e,e1,  [E,xl [u, o] [t, a]  [2,11 [a] [o, A] [A]  /i/  [i,e]  [eta]  /a/ /u/ lit /a/ /u/ /a/  [x, 8] [u,c]  [a] [a]  Rel [x,a] [u,u,o] [a ,u,i]  [e',2]  [A]  /i/ /a/ /u/  [i,e] [x,a] [u,c]  [9i, ia] [a] [o,au, 0]  /a/  [OA  Ii/ le/ /u/ /a /  [i] [x,e] [u] [a, DAC]  [a] [e,ia] [a] [o ,o] [A,o]  [a] [ov,o]  93  Mayes (1979) conducted a spectrographic investigation of velar and uvular stop articulation of one female speaker of Niekepmxcin.12 Formant transitions from consonants to vowels were not measured, so there is no indication of allophonic variation due to a change in place of articulation from velar to uvular. Instead Mayes measured closure duration, burst duration and amplitude. Her findings are that /k/ has greater burst duration and frequency than /q/, but that /q/ has greater burst amplitude than /k/. It is shown in Chapter 5 that formant transitions also serve to distinguish velar from uvular articulation, as indicated by the vowel variation in Table 4.12. Pharyngeals Pharyngeal segments (/1, I', Cw, S'w/) are found throughout Interior Salish, but only Moses-Columbian has phonemic voiceless pharyngeals (/ti, hw/); all other languages report voiced pharyngeals only. Pharyngeals may have some upper pharyngeal component. The degree of friction in the voiced pharyngeal varies with some languages presenting articulations so glide-like they are sometimes heard as vowels. Pharyngeals are reported to lower all adjacent vowels. Usually the similarity between pharyngeal and uvular effects is noted, as in Arabic. Kinkade (1967) notes fronting effects from MosesColumbian voiceless pharyngeals. An example of the considerable difficulty experienced in the transcription of these sounds can be found in Vogt's work on Kalispel (1940). Vogt proposes stressable /i,e,a,o,u/ for Kalispel, noting that /o/ is usually a "morphophonemic variant of /u/, but cases as sanos 'snot', oost 'he gets lost', contrasted with mus 'four', ust 'he dives' etc. clearly establish /o/ as a distinct phoneme" (p.14). Subsequent work has shown that there are two issues here. One has to do with the proper identification of pharyngeals, for  12Mayes's consultant was Mabel Joe from the Shulus Reserve in the Lower Nicola Valley, B.C., Canada.  94 which Vogt transcribes long vowels13, the other with a class of retracted roots found throughout IS. sanos belongs to the class of retracted roots sometimes characterised by the presence of one of a set of retracted coronals and/or a vowel of particular low/retracted quality. Historically these roots are thought to have contained a pharyngeal which frequently appears in Colville cognate roots (Mattina 1979). Thus, the Colville word for 'snot' is transcribed by Mattina as seas (Mattina 1987). The root for Kalispel oost 'he gets lost' is transcribed in Spokane as Awos (Carlson and Flett 1989). Vogt also  transcribes yesaacim 'I am tying it', with the root Vaac cognate with Spokane Vac(i), Colville Vfac, Nfakapmxcin vac. These correspondences suggest that what Vogt heard and transcribed as long vowels are actually pharyngeals. Gladys Reichard, whose grammar of neighbouring Coeur d'Alene (1938) was read by Vogt, transcribed what she terms 'faucals', with the symbols /R, R'/ and /rw, r'w/. These segments, plus an apical /r/, form what Reichard terms the 'r-series', described as follows:14 The most distinctive phonetic characteristic of Coeur d'Alene is the r-series. The sound written r is close to the apical slightly trilled initial r as it is spoken in most parts of the United States, although the tongue is pulled farther back and the r is anticipated by drawing the preceding vowels correspondingly farther back. The faucal trill which may be surd (R) or labialized (rw) is difficult to make and describe.. .in order to achieve it, place the tongue in position for a, draw the larynx down and back ... and trill the faucal region. For the labialized rw place the lips in the position for labialization ... at the same time following directions for R. The implication of the term laucal' and the use of variants of r to transcribe them, suggests that Reichard thought of the r-series as uvular. Kinkade (1967) comments on the mismatch between Reichard's symbols and terminology (r-sounds, trilling in faucal region) as opposed to the symbols and terminology used by Vogt (1940) in describing the closely-related Kalispel. Vogt 13Interestingly, some Arabic course books warn students that they may have trouble hearing the pharyngeals at first, and will likely mistake them for vowels, e.g. Erwin (1963). 14It is not quite clear what Reichard means by terming R and rw as 'surds' if 'surd' means voiceless. As far as I know the Coeur d'Alene pharyngeals are not described as voiceless elsewhere in the literature, and the tokens examined in section are not voiceless.  95 neither transcribes nor mentions an r-series. In fact, he interprets the Coeur d'Alene faucals as voiced counterparts to the voiceless uvular fricative series, appealing to the fact that Coeur d'Alene has voiced counterparts to its coronal, alveo-palatal and velar stop series, so that the faucals can be seen as filling in a structural gap among the uvular segments. For structural reasons this analysis disintegrates when applied to ColvilleOkanagan and Nxa'amxcin (Moses-Columbia Salish), since neither of these two languages has a voiced stop series, but both have pharyngeals (see Kinkade 1967). Kinkade's fieldwork in Nxa'amxcin, which has unmistakable /C, Cw, h, hw/, clarified the description of faucals/long vowels/r-sounds in Interior Salish. Then, gathering his own data and collating material from other sources, Kinkade demonstrated the presence of pharyngeals throughout Interior Salish. Kinkade (1967) reports that the phonetic manifestation of these sounds varies somewhat from language to language and this is confirmed by the range of descriptions in the various grammars. Kalispel pharyngeals are reported to be very hard to hear, while the series in Moses-Columbian is not only supposedly the most pharyngeal, but has split into a voiced and voiceless series. The Nfekepmxcin pharyngeals are considered to involve some uvular frication/trill as well as tongue retraction and general pharyngeal narrowing (Thompson and Thompson 1992). van Eijk (1985) describes Lillooet pharyngeals as further back than the French or German uvular /R/ and more lax than Arabic IS/. The various descriptions of pharyngeals in Interior Salish are tabulated in Table 4.13 along with the reported coarticulatory effects on adjacent vowels. While I have not asked speakers of all the languages how to make pharyngeals, Agatha Bart offered the following description for Moses-Columbian pharyngeals: 'pull the tongue all the way back in your throat, as far as you can go'. Mrs. Bart also reports general tension in the throat in making pharyngeals. These instructions can be compared to Mrs. Bart's description of /q/, which she reports is made as if something is caught in  96 the throat or nose. This would suggest a very back articulation at the velar-pharyngeal port. Table 4.13: Interior Salish pharyngeal effects /-pre-uvulars l-pharyngeal  Lang. Articulatorx description  /V/  Nf  as Table 4.12 as /q, see as Table Table 4.12 4.12  Lt  /T,Sw/ involve some uvular friction or trill, some creak. Retraction of tongue and general narrowing of pharynx. pharyngeal glide with wide aperture; further back than uvular stops. More lax than Arabic 1.; some uvular  Ii/ /a/ /u/ /a/  quality reported  Sh  close to a voiced uvular fricative or trill, with wide aperture. Sounds like a  as Table 4.12  as /q, see Table 4.12  as Table 4.12  as /q, see Table 4.12  /a/  [x]/h,f  pharyngealized back /a/.  faucal trills formed with throat drawn back, larynx drawn down and back; voiced and uvular /V-V, -# but #S- is a voiceless [a] with friction; Avv- is a voiceless [D] with friction. /1/ vd. pharyngeal fricative with little friction; /ti/ strong vls. pharyngeal fricative  other vowels as /q, see Table 4.12  97 Sp^uvular-pharyngeal resonant as series formed by narrowing Table the pharyngeal cavity with 4.12 frequent concomitant narrowing in the post-velar area ^ Co- Vowel plus pharyngeal as ^ ^ Ok constriction, with greatest Table ^ constriction initially; all 4.12 four pharyngeals voiced, with NS- sometimes a uvular flap in Ok, sometimes voiceless in Cv.  as /q, see Table 4.12  as /q, see Table 4.12 Retracted coronals The occurrence of retracted coronals is relatively rare, but consistent within Nxa'amxcin, Nlakapmxcin and Lillooet. Where they occur, retracted coronals are always apical articulations, as opposed to laminal. They also lower adjacent vowel quality and are frequently referred to as imparting a 'darkened' timbre (e.g. Kuipers 1989). The phonological patterning of retraction is discussed in Chapter 6.  98 Table 4.14: Interior Salish retracted coronal effects (Lang. Articulatory description Nf  Lt  /V/  -/pre-uvular  k, c',^, z, z'/ tongue tip  as  as Table 4.12 as /q  with hollowing of tongue behind occlusion. /1,17 are 'dark'.  Table 4.12  k,S,1,17 tongue-tip articulation with retraction of tongue root considered as  I-C  as /q  velarization. /z, z'/ are lax denti-laterals with local retracting effects. Sh  k,s/ rare  as /q  Cr  In apical trill with tonguepulled back; backed  as /q  apical trill Nx  k,^ 1, 1%(13)/  CoOk  None  Sp  ,  as /q  , None  To summarize: with the exception of Lillooet, vowel quality in the environment of retracted coronals, pharyngeals and uvulars is reported to be very similar. van Eijk (1985) remarks on the pharyngealized quality of Lillooet vowels adjacent to pharyngeal phonemes, and Kinkade (1967) notes that the low vowel /a/ of Nxa'amxcin is often fronted in the environment of voiceless pharyngeals. Other than this single case, vowel quality in the environment of post-velars is transcribed as lower or more back than that found in pre-velar environments.  99 Glottals Interior Salish glottals are referred to as glottal in place of articulation, with aspirated release noted for /7/, usually word-finally and in some cases post-tonically (Vogt 1940, Carlson 1972, Thompson and Thompson 1992). Thompson and Thomspon (1992) note occasional phonetic confusion between /h/ and dorsal spirants in Me?kepmxcin. /h/ is generally not found word-finally and is not common. Both /h/ and /2/ alternate with zero in rapid speech (Mattina 1973 for Colville-Okanagan). /?/ is recorded as transparent to lowering from immediately right-adjacent uvulars in Lillooet (van Eijk 1985) and, as will be shown in Chapter 5, Moses-Columbian /2/ is transparent to coarticulatory effects from consonants onto preceding vowels. It is sometimes the case that vowels do not achieve as high a quality in the environment of glottals as they do with pre-uvulars. There is variation on this: I have transcribed relatively high and relatively low allophones in glottal environments, but vowel quality in the environment of glottals is never equated with that found in the context of other post-velar segments. This is clear in all sources. The Nxatamxcin data presented in Chapter 5 show that all consonants have an effect on vowel quality, with pre-velars tending to raise vowels, post-velars lowering and glottals doing neither--hence the transcription of less extreme vowel qualities in glottal environments. 4.3.4 Khoisan It has already been mentioned that some Caucasian language have pharyngealized vowels. Some of the Khoisan languages also have VS (Traill 1985). The acoustic cues to Khoisan /aS/ reported in Ladefoged and Maddieson (1990) are raised Fl and F2, lowered F3 and diminution of energy in the 400-700 Hz region. The raised F2 of the pharyngealized /a/ vowel, which has the effect of fronting (or not backing) the vowel recalls reports of fronting from pharyngeals in Nxa'amxcin, Caucasian and some Arabic dialects.  100 Khoisan strident vowels, which are distinct from pharyngealized vowels and are mentioned in section 4.2.7, have a laryngeal component created by a constriction 'between the part of the tongue below the epiglottis and the tips of the arytenoid cartilages in the upper part of the larynx' (Ladefoged and Maddieson 1990:114). These vowels have even more radical Fl and F2 raising, as well as F3 lowering on the /a/ vowel at least. The acoustic profile of both types of Khoisan articulation suggests an extremely low place of articulation. 4.3.5 Nootka Dialects of the Wakashan language Nootka include phonemic pharyngeals derived from uvulars. /1/ is derived from /q', q'w/; /h/ from /x, xw/. Jacobsen (1969) argues that this is a very recent innovation which does not go back to Proto-Wakashan or ProtoNootkan. There are no /q', q'w, xw/ in the modern dialects of Nootka. /x/ occurs in one root and one suffix in the Kyoquot dialect of Nootka described by Rose (1981). /h/ sometimes alternates with [x], reflecting its historical origin as a uvular fricative. /hw/ occurs before /u/. In Nitinaht /q', q'w/ have become /V, but /x, xw/ have not become /h/ as in Nootka (Jakobsen 1969). All descriptions of the Nootkan sounds reflect a strong laryngeal element. Sapir and Swadesh (1939, quoted by Jacobsen 1969) describe the Nootka IT/ as a 'glottal stop pronounced with the pharyngeal passage narrowed by the retraction of the back of the tongue toward the back of the pharyngeal wall', while /h/ is described as 711/ pronounced with the pharyngeal passage thus constricted'. The sense of a glottal element is so strong that Sapir and Swadesh (1939) label the sounds as flaryngealized glottals'. Sapir (1911) earlier had described Nootka /1/ as "a peculiarly harsh and choky glottal stop.resembling Arabic 'am' "; in a later article he described M as 'a peculiar glottal stop of strangulated articulation and velar resonance'; Swadesh (1939) referred to M as a 'glottal stop with pharyngeal constriction'. Nootkan /h/ also is likened to Semitic /h/ by  101 Sapir (1911). Rose (1981:14-15) describes pharyngeals in the Kyoquot dilaect of Nootka as follows: The Nootka pharyngeals are not characterized by associated labialization or nasalization. The h is like a fricative in being composed of aperiodic noise, but it is like a resonant in having large formant transitions which are perceived as offglides and onglides in adjacent vowels. The I consists of a pharyngealized glottal closure which, like the h, is accompanied by a raised larynx and a retracted tongue root. I is like a resonant in having no burst (i.e. a stop release). However, associated laryngealization, perceived as a series of 'cracks', gives the impression of a series of stop bursts. Impressionistically, the Nootka I sounds much more stop-like and crisp than the Salish sound written with the same symbol. The lowering effect of both pharyngeals on high vowels is noted in early descriptions (see Jacobsen 1969). Rose (1981) transcribes local vowel alternation which is very similar to the Salish and Semitic effects already discussed. Table 4.15: K o uot Vowel allo hon (Rose 1981:16) /V/  Basic  Alveopalatal  Velar  Velar Uvular _ Round ,  Uvular Pharyn- LarynRound geal geal .  i  I  i  i  I'  c  £  £v  U  U  U'  U'  U '.  0'.  S  o'  U  a  a,a  £'  £'  A'  a  A'  a'  a  I'  -  The glottal aspect of Nootkan /I/ is reflected in several processes. For example, some inflectional suffixes in Nootka cause glottalization of a preceding stop, affricate or resonant. /q, qw/ surface as [C] in these cases, suggesting that [5] is both uvular and [+constricted glottis]. Under the same conditions a fricative is changed to a homorganic glide. /h/ alternates with [w] in these cases, reflecting its historical origin in /xw/. Furthermore, Rose (1981) states that ejectives and 'other glottals such as /h, ?, I/ do not occur morpheme-finally, and that vowels are retracted when adjacent to labialized or  102 pharyngealized consonants and are laryngealized adjacent to ejectives, glottalized resonants, I and  ?.  It would appear then that the Nootkan pharyngeals maintain a structural relationship with uvulars. They may be better thought of as phonetically pharyngealized uvulars with a glottal feature. 4.3.6 Haida Haida is a language isolate spoken on the Queen Charlotte Islands in British Columbia as well as Ketchikan and Hydaburg in Alaska. There are two basic dialects, Nothern Haida (spoken by the Alaskan and Massett people) and Southern Haida. Haida pharyngeals, like those in Nootka, are derived from uvulars. Krauss (1979) reports that uvular /G,x/ have become pharyngeals in Northern Haida, but /q,q7 have remained unchanged. /x/ has become a voiceless pharyngeal like Semitic /h/ in Massett, but in Hydaburg, Krauss describes the sound as a 'hoarse pharyngeal trill'. Massett /G/ has become a pharyngeal much like Semitic /C/, but in Hydaburg the /G/ has become 'an affricate, glottal stop followed by a hoarse pharyngeal trill' (p. 840). Levine (1981) notes that this phoneme can reduce to glottal stop then delete, whereas /h/ does not. If [G, x] turn up at all in Northern Haida, they can be traced to borrowings from Southern Haida. Very little is known about the phonology of these sounds.15 Sapir's (1923) paper 'The phonetics of Haida' deals with the Southern SIddigate dialect which has not innovated pharyngeals from uvulars, and so contains no information relevant on this point. 4.4 Acoustic consequences of post-velar articulation The acoustic consequences of constriction in the vocal tract can be predicted theoretically by modelling the vocal tract dimensions and cross-sectional areas resulting 15I thank William Seaburg for sending me examples of Hydaburg pharyngeals. The /7/ is odd indeed and sounds like a cough almost. It is nothing at all like Salish /C/ which is entirely resonant.  103 from various types of articulation. This section discusses the acoustic effects of uvular and pharyngeal articulation as predicted by vocal tract modelling and surveys the results of acoustic studies on Arabic uvular, pharyngeal and emphatic articulation. Since modelling studies do not explore the theoretical consequences of post-velar articulation in combination with some other (primary) place of articulation (such as is found with Arabic emphatics) one can turn only to data-oriented studies to develop an acoustic profile of post-velar constriction as a secondary articulation. It should be noted however that such a profile is specific to the language and segments in question and can be extended to other cases considered to involve some post-velar component only with caution. Modelling studies predict that there is a unique class of articulation made in the pharynx with two major constriction sites: the upper and lower pharynx. Differences in Fl values distinguish the two articulations, with what we transcribe as pharyngeals having a higher Fl than what we transcribe as uvulars. Natural language data from studies on Semitic confirm these predictions. 4.4.1 Uvulars and pharyngeals Klatt and Stevens (1969) calculate the resonances for sounds with primary constriction in the pharynx, given three different constriction sizes. Their predictions are that for a constriction between 3-7cm from the glottis, which they estimate as appropriate for the articulation of pharyngeal and uvular consonants in general, the first formant will be high and the second formant will be low. Although they remark that this area of the vocal tract is relatively insensitive to the exact location of constriction, within this articulatory and acoustic class, a constriction approximately 3-4 cm from the glottis (pharyngeals) besides having high Fl and low F2 is predicted to have F3 lower than a constriction 7 cm from the glottis (uvulars). Klatt and Stevens comment further that F2 for the uvular articulation may also be lower relative to F2 for the pharyngeals. Uvulars are also predicted to exhibit a higher F3 than pharyngeals, with F3 close to F4, leaving a  104 wide gap between F2 and F3. For more constricted configurations, Fl is predicted to be lower for uvulars than for pharyngeals. These general predictions are given in Table 4.16. Table 4.16: Klatt and Stevens (1969) Constriction area 0.05 and 0.1 cm2 1F3 F2 ^—II Constriction Consonant type Fl site velar near F3 low 11cm from glottis lower close to F4 uvular high 7 cm from glottis lower pharyngeal higher low 3-4cm from glottis  Alwan (1986), uses a vocal tract model with tapered junctions between each cavity and the constriction site, thus avoiding sharp discontinuities in area functions that the Klatt and Stevens (1969) model does not. In general her more refined modelling supports the predictions made by the Klatt and Stevens model. Table 9 summarizes Alwan's predictions for uvular and pharyngeal articulation based on a constriction area of 0.15 cm2. Of the three constriction areas for which Alwan makes predictions, 0.15 cm2 is closest to the figures (0.05 and 0.1 cm2) discussed by Klatt and Stevens (1969) in the text of their paper. Thus it is the most appropriate choice for comparisons between the two models. Alwan's results also indicate that Thu s articulated with a tighter constriction than /V, since her model compares more favourably with real data when the area of constriction for /11/ is less than that for /c/ (0.15 vs 0.25 cm2).  105 Table 4.17: Alwan (1986) Constriction area 0.15 cm2, length of constriction lcm. Fl (Hz) F2 (Hz) I^F3 (Hz) Constriction Consonant type site 2263 r 715 1152 3cm from glottis 2344 747 h 1785 1241 2610 483 8cm from is glottis 1241 2737 553 X Alwan (1986) also conducts perceptual tests which confirm the centrality of Fl values as a cue to distinguishing uvular from pharyngeal place of articulation in Arabic. Using native speaker16 judgements of [Sa:] and [ica:] stimuli in which F2 is held constant, it was found that Fl is an essential cue for identifying place of articulation. Alwan (1986:109-110) reports that an onset value of Fl which is 'at least as high as that in the steady state of the vowel...results in the perception of a pharyngeal, and its perception increases with increasing Fl onset value. The uvular on the other hand, is perceived when the onset value of Fl is at least 90 Hz less than that during the steady state of the vowel...'. The results of perceptual experiments reported in el-Halees (1985) are in agreement with the finding that Fl is a crucial cue not only to a post-velar place of articulation, but to distinctions within it. Tokens with higher Fl transitions from consonant to vowel were judged as pharyngeal rather than uvular by el-Halees' Jordanian subjects. In fact, Fl transitions alone proved sufficient to cue the distinction between uvular and pharyngeal place of articulation. Natural language data Acoustic studies of Arabic language data confirm that post-velar articulation raises Fl values, with the Fl for pharyngeals being higher than that for uvulars. The results of several such studies are presented in Tables 4.18-21. Of particular note is the 16mwan's subjects for this particular experiment consist of 3 Iraqis, 1 Kuwaiti, 1 Lebanese and 1 Sudanese.  106 range of formant values reported for both uvular and pharyngeal articulation. Mostly it varies depending on the vowel which follows, but it also varies considerably across studies and according to dialect as well as within and across speakers of the same dialect. Based on four studies alone, the Fl of /I/ ranges from 464-900 Hz, of /h/ from 433-1100 Hz; F2 ranges form 1065-1887 Hz for /S/ and from 1050-1937 Hz for /h/. There is less data on uvulars, but what there is produces an Fl range of 382-600 Hz, and an F2 range of 782-1300 Hz. Presumably the considerable variation reflects both contextual effects and basic variation in location of articulation. The studies below concentrate on uvular and pharyngeal data, but Kuriyagawa (1984) compares Cairene /k/ versus /q/ and finds strong F2 lowering and Fl raising effects on vowels from /q/ which distinguish it from /k/. Kuriyagawa (1984) particularly notes the extreme extent of F2 lowering conditioned by /q/. Alwan (1986) Table 10 is based on data averaged from four speakers (3 Iraqi, 1 Lebanese) in Alwan (1986), with measurements taken at the onset of the vowel in CV: sequences taken from ?V:CV: or CV: nonsense utterances. As a point of comparison, Alwan gives the steady-state values for /i:,a:,u:/ in post-velar contexts averaged across four speakers and based on ten tokens per speaker. In the case of pharyngeal-V: sequences Fl falls consistently regardless of vocalic environment, but for the uvulars, Fl falls in Qi: and Qu: but rises in Qa: sequences. The fall in Fl from offset of the pharyngeal to steady-state of the vowel is greater in the case of pharyngeals than uvulars, thus corroborating the prediction that pharyngeals have a higher Fl than uvulars.  107 Table 4.18: Arabicpost-velars (Alwan 1986) Vowel /i:/  /a:/ I^Seg't^Vowel I 1 h x q Steady -state Vowel,  Fl 728 739 584 587 661  F2 1181 1266 1294 1143 1206  F3 2224 2321 2553 2522 2470  Fl 492 453 382 409 311  F2 1887 1937 1987 1784 2213  Vowel /u:/ F3 2561 2592 2604 2574 2751  Fl 464 453 403 401 350  F2 1065 1181 790 782 786  F3 2022 2084 2472 2562 2414  Ghazeli (1977) Table 4.19 summarizes the figures available from Ghazeli (1977) who recorded two CV tokens of each type (C a uvular or pharyngeal, V one of /i,a,u/) from twelve subjects of various dialects (6 Tunisian, 2 Libyan, 1 Algerian, 1 Cairene, 1 Jordanian, 1 Iraqi). Readings are taken at the boundary between the consonant and the vowel. He notes interdialectal variability of +/- 50 Hz on Fl, +/- 100 Hz on F2 and +/- 250 Hz on F3 and considerable dialectal variation on the value of the low vowel when it is not in a postvelar environment, e.g. [c/x] for Libyan and Cairene speakers and [w/a] for the Jordanian speaker. Despite variation Fl is high for pharyngeals regardless of following vowel, though highest for pharyngeal-a sequences. Ghazeli does not provide figures for uvular articulation other than next to the /a/-vowel. Table 4.19: Arabic ost-velars Ghazeli 1977  IBM^Vowel /a/^Vowel /i/ I h is  Fl 900 1100 500600  F2 1450 1700 12001300  F3 2300 2300 23002600  Fl 700 700  F2 F3 ,^Fl 1700 , 2700 650 1800 2700 550  Vowel /u/ F2 F3 1300 ,^1700 1100 1700  108 Butcher and Ahmad (1987) Butcher and Ahmad (1987) recorded three speakers of Iraqi Arabic with five repetitions of a series of CVC tokens. Their tokens were (mostly nonsense) words of the shape CIVC2 with a pharyngeal in C1 or C2 position, and V ranging over /i:, e:, a:,.o:, u:, i, a, u/. Values for the pharyngeals tabulated here are taken at the pharyngeal-vowel boundary of a T/hVC sequence. For comparison, Butcher and Ahmad present formant values taken at the steady-state portion of each vowel following a glottal consonant. This is intended to serve as an indication of what vowel-quality is like when unaffected by adjacent consonants. Fl values for pharyngeal -a: sequences are not as high as those found by Ghazeli (1977). Fl values for the pharyngeal -i sequence are higher than those reported by Ghazeli. Compared to the values for vowels preceded by glottals, Fl is consistently higher for pharyngeals, by as much as 100% on the Ii:/ vowel for instance. This should be compared to the values reported by Adamson (1981) for Sudanese Arabic, Table 4.21, where Fl is not nearly so high in absolute terms for pharyngeal-i sequences, but Fl doubles in pharyngeal environments, just as it does in the Iraqi data. Table 4.20: Iraqi abicpharygeals and glottals (Butcher and Ahmad 1987)  T h 2 h  Vowel /i:/  Vowel /a:/  [^Seg't Fl 861 855 690 727  F2 1408 1517 1195 1296  F3 2036 1979 2588 2487  Fl 815 761 398 375  F2 1461 1767 2366 2329 _  Vowel /u:/ F3 2386 2789 3130 3070 _  Fl 602 642 412 416  F2 1203 1222 592 680  F3 1566 1778 unclear 2535  Adamson (1981) This study uses real (rather than nonsense) C1VC2 words in a sentence frame, with C1 a pharyngeal and readings taken at the pharyngeal-vowel boundary for /h/ but during the consonant itself for /C/. The vowel values which are provided for reference are  109 taken from the steady-state portion of the same CVC sequences. Readings are taken from four speakers of Sudanese Arabic, 23 words per subject. Table4.21: Sudanese Arabic pharyngeals and reference vowels (Adamson 1981) Vowel /a/^Vowel /i/  I^Seg't 1 h Steady -state Vowel  Fl 580 650 500  F2 1440 , 1450 1500  F3 2266  Fl 500 433 250  F2 1600 1800 2100  Vowel /u/ F3 weak  Fl 475 460 250  F2 1150 1050 900  F3 weak  4.4.2 Emphatics As mentioned, acoustic modelling of post-velar articulation does not deal directly with the effect of post-velar secondary articulation. However, extrapolating the verified prediction that post-velar constriction, such as found in both uvular and pharyngeal articulation, results in high Fl and low F2 values, one would expect to find some effect of this sort even in complex articulations where post-velar constriction is not the only articulatory gesture. The classic example of such an articulation is of course the class of Arabic emphatics, which combine a coronal place of articulation with post-velar constriction. A comprehensive review of the history of scholary inquiry into Arabic emphatics can be found in Giannini and Pettorino (1982). In brief, all studies concur that F2 lowering is a cue to emphatic articulation (Obrecht 1968, al-Ani 1970, Giannini and Pettorino 1982, Card 1983, el-Halee 1985) but the contribution of Fl, which has been shown to be so critical in modelling uvular and pharyngeal articulation, is mostly ignored by both Obrecht17 (1968) and Card (1983). Giannini and Pettorino (1982) note the omission of Fl effects in Obrecht's work and based on their own investigation of an Iraqi 170brecht (1968) justifies the exclusion of Fl effects from his study on the basis of preliminary spectrographic investigation which showed that F2 was the most powerful cue. His own data though, do show Fl effects.  110 Arabic speaker, they find that both Fl and F2 effects cue emphasis, although F2 seems to be the most consistent cue. This general finding is supported by al-Ani (1970), Ghazeli (1977), al-Ani and el-Dalee (1984), el-Dalee (1984), el-Halees (1985) and Laufer and Baer (1988). While F2 effects feature most prominently in descriptions of emphasis elDalee (1984) and el-Halees (1985) found that Fl lowering was particularly striking on the emphatic pharyngeals [C, h]. In contrast, Card (1983) reports a 700 Hz lowering of F2 for  [s] and does not mention Fl effects.18 Table 4.22 charts the results of al-Ani and elDalee's work on 220 [?VCV] utterances by an Alexandrian speaker with C varying between emphatic and non-emphatic articulation. Measurements are taken at the vowel's steady state. Table 4.22: Alexandrian emphatic effects (al-Ani and el-Dalee (1984)) Emphatic  Plain Vowel a i u  Fl 743 314 357  F2 1614 2086 850  Vowel a i u  Fl 736 495 371  F2 1062 1667 867  Norlin (1987) investigates the acoustic properties of emphasis in Egyptian Arabic using real monosyllabic and disyllabic words containing the vowels /i:, e:, a:, o:, u:/ and /i, a, u/ in four phonolocial contexts: CV(:)C; CV(:)C; CV(:)C ; CV(:)Q. All tokens were produced six times in a sentence nine male native speakers from Cairo. The first instance was discarded, formant frequencies were taken from the steady state portion of stressed vowels. Fl and F2 variation across speakers was found to be small, with F3 variation larger. Table 4.23, which presents average formant values for nine speakers (5 tokens each) of /i, a, u/ in plain and emphatic contexts, illustrates the dramatic lowering of F2 triggered by emphatic consonants. Fl raises, but minimally in 18However, from what I can decipher of the spectrograms in my copy of her dissertation, it looks as if Fl raises also.  111 comparison to F2 movement. Based on t-tests on measurements of the difference between long vowels in plain and emphatic contexts, a cline of the order /a:/ > /i:, el > /u:, o:/ is reported. However, Norlin reports that differences in transition onsets increase so as to compensate for the minimal steady state distinction between high back vowels in plain and emphatic contexts. In summary, high vowels are affected less that low vowels while front vowels are affected more than back vowels. The /a, al vowels of Egyptian Arabic are relatively front. Table 4.23: Eg • tian emphatic effects (Norlin 1987) ^ Plain Em hatic Vowel i a u  Fl 435 615 415  F2 1915 1585 1120  Vowel i a u  Fl 450 630 450  F2 1485 1165 955  Norlin also investigates the spectral properties of emphatic and non-emphatic sibilants using data from real words in initial postion of a carrier sentence. No differences could be quantified based on examination of the waveform and spectrograms. The only quantitative measure which differentiates between plain and emphatic consonants is the centre of gravity (a measure of the overall pitch level of the spectrum) in the critical band spectra. 14/ has lower centre of gravity than /s/ except for one speaker, and greater dispersion (i.e. a flatter spectrum). /z/ has a lower gentre of gravity than /z/, and /z, z/ have flatter spectra than /s, 4/. Norlin finds very little affect from emphatics on /u/.  An obvious question raised by the acoustic description of emphasis is how it compares with the acoustics of other segments involving post-velar articulation. Comparing the loci for Fl F2 and F3 for velar, uvular, pharyngeal and emphatic consonants, Giannini and Pettorino (1982) note that emphatic articulation is like post-  112 velar articulation in having (i) a high Fl loci and (ii) a low F2 loci. Table 4.24 summarizes the basis for these observations. The Fl locus for uvulars is reported as 500 Hz, for pharyngeals 1000 Hz and for emphatics 600 Hz. All of these values are dramatically higher than the Fl loci reported for velars (250 Hz) and for the nonemphatic counterparts (250Hz) of the emphatics under consideration. Emphatics also pattern with uvulars and pharyngeals against both velars and non-emphatic coronals in observed F2 loci. Table 4.24: Consonantal loci in Iraqi Arabic (Giannini and Pettorino 1982) Articulation Velar stop Uvular stop Uvular fricatives Pharyngeal fricatives Emphatic s Non-emphatics  Fl loci (Hz)  F2 loci (Hz)  250 500 500 1000  2300 1400 1500 1500  600 250  1000 2000  Despite not discussing the contribution of Fl to emphatic effects, Card (1983) does provide Fl and F2 readings for emphatic and non-emphatic minimal pairs. Figures 4.7 and 4.8 plot the steady-state readings of long and short vowels in an emphatic and non-emphatic environment. Arrows go from the non-emphatic vowel to the emphatic vowel. All four of Card's Palestinian speakers are represented. While some speakers show considerable Fl effects and lower emphatic vowels, all speakers show F2 effects on short vowels. The situation with long vowels is more complicated. /u:/ is minimally effected by emphasis, but that is well-described in the literature. 1o:/ lowers considerably for all four speakers, but these findings are based on only one form for each speaker. /a:/ mostly just backs; /i:/ even raises for two speakers and /e:/ lowers for three and raises for one. For one speaker the difference between emphatic and non-emphatic /e:/ is lowering.  113  0  0  a^  ^, A A  -1600^-1400^-1200^-1000 F2(Hz) Source: Card (1983)  Figure 4.7: Palestinian Arabic Short Vowels: Plain and Emphatic  CI  0 _ 0  o  -2000  -1500^-1000 F2(Hz) Source: Card (1983)  Figure 4.8: Palestinian Arabic Long Vowels: Plain and Emphatic  114 The acoustic effects of Caucasian pharyngealized consonants have already been discussed in section 4.3.2. The VT of Udi (Figure 4.6) show centralizing of a sort not at all reminiscent of the Arabic data plotted in Figures 4.7 and 4.8. There is no acoustic data on Interior Salish retracted coronals other than that reported on in the next chapter of this dissertation (section 5.2.4). 4.4.3 Glottals As discussed in sections 4.2.8 and 4.2.9, glottal articulation is achieved by adduction of the vocal chords. While there may be tongue body movement associated with glottals, it varies according to vocalic environment and so cannot be viewed as intrinsic to the glottal segment itself. Given the lack of an intrinsic supra-laryngeal vocal tract configuration for glottals, the effect of a purely glottal gesture should in itself have no effect on the formant structure of adjacent segments. This can be shown to be true from both a perceptual and acoustic point of view. Proving the perceptual relevance of flat formant transitions for /?/, Alwan (1986:106) synthesizes flaa/ sequences with no formant transitions from the /?/ to the vowel. Such tokens are judged as glottal stop initial by her Arabic-speaking subjects, whereas /Taa/ and /isaa/ sequences crucially require formant movement from the consonant to the vowel in order to be correctly identified. On the acoustic front, Butcher and Ahmad (1987) use /?/ as a touchstone to highlight the formant effects of pharyngeal articulation in Arabic. As phoneticians their understanding is that glottal articulation, since it lacks an intrinsic gesture other than that of vocal fold adduction, is ideally suited as a neutral benchmark against which the supra-laryngeal articulatory effects of other segments can be gauged. In confirmation of this, spectrograms of vowel-glottal sequences in the Iraqi Arabic of Butcher and Ahmad's consultants show no transition from the vowel into the consonant (Butcher and Ahmad 1987). This is in direct comparison to spectrograms of vowel-pharyngeal sequences, which show F2 lowering and Fl rising as the vocal tract assumes the post-velar  115 constriction appropriate for A,h/. Klatt and Stevens' (1969) spectrograms of a Lebanese speaker's /ha-/ and /7a-/ sequences likewise show no formant transitions from the consonant to the vowel onset. That the supra-laryngeal vocal tract assumes the  configuration of adjacent segments while the glottis adjusts appropriately for /h/ or /7/ is illustrated by O'Connor (1973), who shows spectrograms of [#hi], [#ha] and [#11o] sequences in which the intial [h] reflects the formant structure of the following vowel. 4.5 Summary and conclusions It has been demonstrated from language-independent and language-particular facts that there is an articulatory and acoustic class of uvulars distinct from that of velars. This of course has been known for some time, but the various types of evidence for the distinction between the two classes are rarely brought together in a way that predicts the dual nature of uvular articulation given its uniqueness with respect to position in the vocal tract and its connections with both dorsal and pharyngeal musculature. The uvular place of articulation distinguishes itself by being at the juncture between the extreme posterior end of the oral cavity and the top end of the pharyngeal cavity. As a consequence, uvular articulation reduces upper pharyngeal cavity volume in a way that velar articulation does not. This results in high Fl values for uvulars relative to pre-uvular segments. This is verified by the acoustic examination of uvular articulation in Semitic and will be shown to hold for Interior Salish in the following chapter. Furthermore, while the velars and uvulars contrast in many languages, there are also clear phonetic (and phonological) affiliations between the two since both velars and uvulars are made by dorsal gestures to the front (velars) or rear (uvulars) of the soft palate (see Cole 1987, McCarthy 1989, 1991, Elorrieta 1991). With respect to pharyngeals, there appear to be two types: those with a relatively high constriction in the pharynx, and those with a relatively low constriction in the pharynx. No language contrasts an upper pharyngeal, a middle pharyngeal and a low  116 pharyngeal. In those (rare) cases where a phonemic contrast is maintained between two types of pharyngeals, a laryngeal component is evident in the lower one. This is certainly true in the description of Caucasian 'deep' or 'emphatic' pharyngeals. Another example of pharyngeal articulation with a phonemic laryngeal component is found in the Interior Salish contrast between /f/ with /V. In some of the languages the contrast exists with iiw/ and /C'w/ as well. Glottalization is a pervasive feature on all resonants in Interior Salish- /T'S'w/ are simply patterning as resonants with the laryngeal feature [constricted glottis] in these cases. This raises the question of whether the Caucasian segments are structurally to be viewed as glottalized versions of the 'higher' pharyngeals (T'), or as pharyngealized glottal stops (21). The latter interpretation is suggested by Colarusso (1988) given the pervasive secondary pharyngealization in many of the languages which are reported to have 'strong' glottal stop. In the Semitic languages dealt with here there is no segmental phonemic contrast between upper and lower/glottal pharyngeals and no systemic contrast between plain and laryngealized segments, although there is of course the phonological spread of emphasis which is reported to affect /?/ in several dialects, producing [7]. No study that I have been able to review relates the occurrence of phonetic glottal effects on pharyngeals with the spread of emphasis onto /?/—in other words, descriptions of [7] do not indicate similarity with versions of [1] that have a laryngeal component. Articulatory descriptions of pharyngeals in Semitic and in Interior Salish indicate a fairly broad range of possibilities within the confines of pharynx. Semitic pharyngeals nonetheless maintain a phonological affiliation with 17, h/ which is lacking in Interior Salish. This is discussed in greater detail in Chapter 6. The precise nature of Caucasian pharyngeal articulation remains something of an issue given conflicting reports on Kabardian (Choi 1990) and Ubykh (Colarusso 1988) with respect to vowel effects. Glottals in the languages dealt with here are described as laryngeal gestures with little or no affect on adjacent vowels, contrary to both uvulars and pharyngeals.  117 Whatever the phonetic distinctions in post-velar articulation in Salish, Semitic and Caucasian, a very striking fact is the presence of 'pharyngealized' consonants in all three language groups. The primary acoustic effect of Arabic emphatics seems to be backing of targetted vowels, although lowering is also a factor. In must be borne in mind that since the primary emphatics (the Wtbaq consonants) are all coronals, we would not expect the same extent of Fl raising that is found with uvular and pharyngeal articulation. The situation is not clear concerning the effects of Caucasian pharyngealized consonants. One study (Kingston and Nichols 1986) reports Fl raising, whereas Colarusso (1988) reports a relatively low Fl and comments on the impression of vowelraising and slight fronting from pharyngeals and on the pharyngealized vowels reported for Tsakhur. On the other hand the data in Catford (1983), plotted in Figure 4.6, show pharyngealized vowels in Udi are clearly centralized in comparison to their nonpharyngealized counterparts. The same is true for the Tsalchur data in Catford (1983). What is being called fronting seems only to apply to back vowels. It is not clear how the gesture which accomplishes CC is related to other pharyngeal gestures in the Caucasian languages. Interior Salish retracted coronals are impressionistically recorded as lowering and darkening vowel quality. It will be shown in Chapter 5 that retracted coronals in Nxa'amxcin result in lowered vowel quality, but not as low as that found in the environment of uvulars and pharyngeals.  118 Chapter Five: Interior Salish Phonetics 5.0 From acoustics to place of articulation This chapter turns to an acoustic study of consonantal place of articulation in Moses-Columbia Salish (Nxa'amxcin). The aim of the study is to achieve a reasonable estimate of place of articulation for all the Moses-Columbian consonants, but particularly the post-velars in the inventory. The investigation itself is divided into two parts, the first of which is presented in Section 5.2, the second in Section 5.3 of this chapter. The first part of the study establishes estimates of place of articulation for the Moses-Columbian post-velars through the examination of co-articulatory effects on preceding stressed vowels from consonants at all places of articulation in the language's inventory. The resulting data enable (i) a characterization of the nature and extent of -VC- co-articulation from every consonantal place (bilabial, coronal, velar, uvular, pharyngeal, glottal) and (ii) a comparison of the effects of pre-uvular articulation with post-velar articulation as has been done with the Arabic data discussed previously. The second part of the study compares pharyngeal articulation across the various Interior Salish languages and evaluates their acoustic properties by comparison to what is known about Arabic pharyngeals. Results of both parts indicate that Interior Salish uvulars and pharyngeals (i) display the acoustic properties predicted by a post-velar constriction site and (ii) pharyngeals on average condition higher Fl effects on preceding vowels than do uvulars. Further, glottal stop is shown to participate in two effects. It can be transparent to coarticulatory effects from following consonants in the configuration -V?C- , or else it manifests itself as varying degrees of phonetic creak on the preceding vowel, without affecting vowel quality in any observable way. Studies of speech sound acoustics such as the one undertaken here are relevant to speech production in so far as one can establish a relationship between acoustic events and  119 articulatory events, thereby enabling an interpretation of acoustic phenomena in articulatory terms. Section 5.1 makes explicit the general guidelines for the translation of acoustic events into their articulatory counterparts. 5.1 The acoustic theory of speech production The natural resonances of a uniform tube open at one end are in a 1:3:5 ratio. Assuming that the human vocal tract can be likened to a tube open at one end, and given the dimensions of the average male vocal tract (17cm in length) these ratios convert to resonances at 500Hz, 1500 Hz, 2500Hz respectively by the following formula (Chiba and Kajiyama 1941, Fant 1960).1 (1) Resonance frequencyn = (2n-1) c / 4 (vocal tract length) c= speed of sound (343 metres/second) Applied to the dimensions of the female vocal tract (15 cm in length) the resultant resonant frequencies are 572 Hz, 1716 Hz and 2858 Hz.2 Largely as a consequence of the different dimensions of male and female vocal tracts, the formant frequencies of women's speech tend to be 18-20% higher than those found in men's speech (Fant 1960, Nordstrom 1977).3 These general differences in the predicted formant values of female and male vocal tracts must be born in mind when evaluating formant readings and will be of some relevance in the comparision of Interior Salish data with Arabic. To illustrate the relevant aspects of the acoustic theory of speech production, in 'See Chiba and Kajiyama (1941) and Fant (1960) for a detailed explication of the physics behind the acoustic theory of speech production. These three authors are the original references on the topic. 2Female and male vocal tracts also differ in terms of mouth cavity dimensions and pharynx length as well as the size and length of the vocal cords (Chiba and Kajiyama 1941, Fant 1966, Nordstom 1977, Ohala 1983). 3Nordstom (1977) shows that there is more to the discrepancy between female and male formant values than physiology since there is a residue of variation unaccounted for by what is known about the anatomical differences between female and male vocal tracts. Nordstrom (1977) suggests that social, cultural and perceptual factors may motivate manipulation of the vocal tract to exaggerate or enhance physiologically determined differentiation.  120 Figure 5.9 places of articulation in the human vocal tract are transferred onto three representations of the vocal tract as a tube open at one end. Within the tubes are representations of the amplitude envelope for the volume velocity waves of the first, second and third resonances respectively, assuming a resonator closed at one end (the glottis): onequarter of the wavelength of the first resonance fits into the tube (labelled Fl, the topmost tube), 3/4 of the wavelength of the second resonance (labelled F2, the middle tube) and 5/4 of the wavelength of the third resonance (labelled F3, the bottom tube). There is a velocity maximum (or antinode) for all resonances at the lips, and a velocity minimum (or node) for all resonances at the glottis. In Figure 5.9 velocity antinodes correspond to the highest amplitude portion of the resonance waveform. Velocity nodes correspond to the lowest amplitude portion. Assuming a 17cm vocal tract there is an additional velocity antinode for the second resonance 5.67 cm from the glottis, and a velocity node 11.33 cm from the glottis. For the third resonance there are velocity antinodes at 3.4 cm and 10.2 cm from the glottis (and at the lips, of course, as for all resonances). Velocity nodes for F3 are 6.8 cm and 13.6 cm from the glottis as well as at the glottis. Constriction at a velocity antinode lowers the frequency of the standing wave, whereas expansion at the same velocity antinode raises the resonance frequency. The converse is true for constriction and expansion at velocity nodes (Chiba and Kajiyama 1941, Fant 1960, Ohala 1990).  121  PAL  VEL^uv  I  PHAR  Fz  F3  Figure 5.9: Vocal tract and first three resonances (adapted from Giannini and Pettorino (1982).  Using the principles outlined above and armed with formant values one can derive a reasonable estimate of place of articulation for a given segment without the intrusion of physical investigation of the vocal tract during articulation. For example, a rule of thumb for conversion between formant values and articulation is that constriction in the front region of the vocal tract (hence near a velocity antinode for all three resonances) leads to  122 low Fl values. Conversely, constrictions nearer the glottis result in higher Fl values since constriction is now nearer a velocity node. By the same token, F2 is high in the case of palatal constriction since the articulation is closer to the velocity node for the second resonance of the tube than it is to a velocity antinode. Thus, for instance, the /i/-vowel of many varieties of English, since it is articulated with a constriction fairly forward in the vocal tract, has low Fl and high F2. A high back unrounded vowel will have a slightly higher Fl as the constriction moves away from the front palatal region and a significantly lower F2 since the constriction is virtually at the velocity node for the second resonance of the vocal tract. Rounding, depending on the gesture involved, can lengthen the vocal tract by lip protrusion and narrow the lip orifice. Both components of labialization lower all resonances (Fain 1968, Lindblom and Sundberg 1971). As a further example, consider the maximally high Fl of pharyngeals. There are two things one might do to raise Fl from its standing wave frequency of 500 Hz. One is to maximize expansion at F1's velocity antinode (the lips), and the other is to constrict at F1's velocity node, near the glottis. Pharyngeal articulation does both of course, the jaw lowers, the vocal tract opens like a horn and the tongue backs into the pharynx. While it is traditionally assumed that Fl is correlated with phonological features reflecting tongue height, this is a simplistic and not entirely accurate translation between acoustics and tongue configuration. The role of mandibular movement in conditioning Fl values is made clear in Lindblom and Sundberg (1971), who demonstrate the correlation between jaw lowering and rising Fl values given a constant tongue position. As such, mandibular movement plays a role in determining the relative openness of the vocal tract, which in turn is reflected in Fl values. Lindblom and Sundberg's articulatory modelling also shows the extreme sensitivity of F2 to the position (and shape) of the tongue body in the mouth. F2 lowers dramatically as the tongue body moves towards the pharynx. F2 also tends to be affected by the degree of constriction. In the case of palatal constriction, F2 rises as the constricton narrows, for velar and pharyngeal constriction F2 lowers as the  123 constriction narrows. Finally, I note that the translation from acoustics to articulation is not always oneto-one (Stevens and House 1955, Atal et al. 1978) and should be accompanied with descriptive and instrumental articulatory data. The ideal case is for acoustic studies to be complemented with X-ray or electromagnetic resonance data. At present this kind of information simply is not available for the Interior Salish data examined here. 5.2 Moses-Columbian (Nxa?amxcin): A case-study in post-velar articulation To assess the place of articulation of Interior Salish post-velars I concentrate on data elicited from four speakers of Moses-Columbian. This is done for several reasons: (i) Moses-Columbian pharyngeals are the easiest to identify within Interior Salish. Because they are the most fortis4 of the Interior Salish pharyngeals, segmentation procedures are slightly less problematic and co-articulatory effects can be documented more accurately. (ii) Moses-Columbian has the fullest inventory of pharyngeals found in Interior Salish, viz /c, V, Cw, V w, h, hw/. The other Interior Salish languages do not have the phonemic voiceless pharyngeals found in Moses-Columbian.5 Since Arabic has both /S,h/, examining the Moses-Columbian inventory enables one to place both voiced and voiceless Interior Salish pharyngeals within a cross-linguistic context. (iii) I was able to gather data from three speakers of Moses-Columbian (MM, AB, ED) with a fourth (JM) added from M.D. I(inkade's recordings. By considering four speakers one can be reasonably sure that consistencies across speakers reflect effects specific to the language rather than individual speakers. While data from all four speakers were analysed, for purposes of illustration in the text data from a single speaker (MM) are presented. In general, the data presented are representative of coarticulatory effects that are consistent across all four speakers, but there 41 use the term 'fortis' loosely here. Nxa'amxcin pharyngeals are less readily confused with vowels (unless in the context [-aTa-]) than in the other languages. 5/1,w/ occurs in only one morpheme in Moses-Columbian. Furthermore, it may be that Okanagan has some rare occurrences of /11/ (Tony Mattina, personal communication, 1992).  124 is some variation, particularly in the influence of uvulars (section 5.2.6). Preliminary investigation of data from speakers of other Interior Salish languages (Colville-Okanagan, Coeur d'Alene, Shuswap, Nielkepmxcin) indicate that the findings presented here are reasonably representative of co-articulatory effects found throughout Interior Salish, but further comparative study is an obvious need. 5.2.1 Methods All data collected and analysed here are actual phonological words in the given language, elicited in citation form with each token repeated twice. In most cases the recordings were made with a Marantz 430 cassette recorder, an AKG D320B unidirectional microphone and metal cassette tapes. Some early recordings were made with a Uher 4000 Report 1C and Ampex 631 1/4 inch polyester tape. Most recordings were made in the field, but with considerable care to ensure a quiet environment with no appliances running. To this end I have even waited for refrigerators to cut out and have re-recorded items if outside noise interferes. A considerable portion of the Moses-Columbian database and some Niakapmxcin were recorded in a sound booth and so is of optimal quality. In all languages where I was able to record several speakers, the same lexical items were recorded if possible. Recorded data were digitized at 10 or 12 kHz and analyzed with the waves+ program on a Sun or Sparc workstation at the University of Pennsylvania. Some analysis was done using MacSpeech Lab II at the University of British Columbia. All data analysed with waves+ were formant-tracked, with the output hand-corrected if necessary. Correction is necessary when formant tracks clearly do not coincide with the placement of formants in the spectrographic display. Such mis-matches are by no means dominant in the data, but it tended to be a factor with speakers with a relatively high fundamental frequency (typically female), or with speakers whose source characteristics include a high-amplitude first harmonic. Such voice-quality characteristics tend to confuse accurate identification of  125 Fl in particular. As is well-known but little discussed in the literature on spectrography, the wider-spaced harmonics of a voice with a high fundamental simply provide fewer opportunities for the instantiation of well-resolved formant structure and so can complicate the proper identification of all formants. Whenever formant mis-matches arose in the data, both FFT and LPC spectra as well as narrow-band spectrograms were generated to aid in more accurate identification of formant structure. Formant-tracks can be re-drawn on the basis of such information and corrected values read into the data-files. The analysis of data from all speakers in all languages focusses on stressed vowels left-adjacent to the consonant under investigation. This is because coarticulatory effects in Interior Salish tend to be strongest on vowels preceding a given consonant and without stress Interior Salish vowels tend either to delete or reduce to schwa. Two measurements were taken of each vowel, the first one at the midpoint (defined as the local maximum of Fl, or else of F2 if Fl is steady; if both formants are in transition, the temporal midpoint was taken) and the second one at vowel offset. While the inventory of velars, uvulars and pharyngeals in Moses-Columbian contains rounded phonemes, I have excluded these from consideration wherever possible so as to avoid confusing the effects of rounding with the effects of primary place. However, there are some cases, due to shortage of data, where I was obliged to measure vowels in the context of rounded consonants. Where this occurs, it is noted in the text. Table 5.25 outlines the entire Interior Salish database on which the acoustic analysis presented in this dissertation was undertaken, as well as speakers whose data has been transcribed impressionistically but has not been systematically investigated spectrographically. Information relevant to the Moses-Columbian place of articulation study presented in section 5.2 is given in the first five rows of the table.  126 Table 5.25: Interior Salish Database II Speaker (Sex) 1 Age 0 Recorded by II Language I^Total V's II Stressed V's I MM (F)  60's  NJB  ED (F)  70's  NJB  AB (F)  60's  NJB  JM (M)  60's  MDK  MosesColumbian MosesColumbian MosesColumbian MosesColumbian  Total vowels in Moses-Columbian database  PF (F) MS (F) AS (M) CQ (M) LC (F)  60's 60's 60's 70's 60's  NJB BC BC NJB NJB  Other IS Languages in database Spokane Spokane Ch-Kalispel Co-Ok Co-Ok  LN (M) BL (F), MSt(F)  70's 60's  NJB NJB  Coeur d'A Coeur d'A  DS (M) JT (M), NJ (F), MJ (F) BD (M)  60's 40's; 40's; 60's 60's  NJB NJB  Nte?kepmx Nfe2kepmx  NJB  Shuswap  Total analysed vowels in IS database  239  117  506  280  485  266  250  116  1480  0^779^I  133 96 Pharyngeals only Pharyngeals only I^166 242 only impressionistic data available 222 428 only impressionistic data available 202 275^ only impressionistic data available  I^  I  181^I  137  2739^1^1602^I  5.2.2 Moses-Columbian: an overview (2) presents the Moses-Columbian consonantal inventory for reference purposes. Phonemic vowels are /i, a, u, a/.  127 (2) Moses-Columbian (Moses-Columbia Salish: Kinkade 1981) Phar^Glottal Lab Coronal^C^Velar Uvular q qw p^t^c (tf)^(ts) k kw q coy p'^c'^(tf')^k' k'w h hw S^x xw X Xw ççw nrl y^1^w C'w m'^n' r'^y'^1'^w' As noted above, the major effects illustrated here are fairly consistent across all four Moses-Columbian speakers and can be considered representative of co-articulatory effects in the language. Further, both impressionistic and acoustic examination of data from speakers of other Interior Salish languages suggest that the general findings presented here for Moses-Columbian can be extrapolated to Interior Salish in general, although further work is required to gauge the full extent of speaker- and language-specific variation. Spectrograms illustrating place-of-articulation effects for some of the other languages in the database can be found in Appendix B. Table 5.26 summarizes the formant values plotted in the following figures for speaker MM and for comparison, summarizes formant values for the same contexts from JM, who is MM's father, recorded by M.D. Kinkade in 1966. Gaps in the table indicate lack of information. In brief, the results of this investigation are that pre-uvulars (labials, coronals and velars) distinguish themselves from post-velars (uvulars and pharyngeals) by raising vowels in formant space. Furthermore, velars and labials tend to back vowels, coronals tend to front them. Uvulars and pharyngeals pattern together in lowering all vowels. Pharyngeals lower vowels more than uvulars do. Glottals neither lower nor raise vowel quality.  128 Table 5.26: Summary of Moses-Columbian articulatory effects: MM and JM J. Miller (TM)^ I Speaker: M.Marchand (MM) Nucleus^I Offset Vowel II Nucleus^I Offset Bilabials F2 F2 Fl Fl Fl F2 Fl F2 285 2049 245 i 1164 1212 424 528 u 1411 585 1422 579 a 511 1074 497 1257 a Coronals 2059 2281 420 1939 347 427 2245 374 i 1238 346 1240 429 1126 406 435 1271 u 1532 587 1501 540 1966 708 1797 541 a 475^1654^454^1556 1667 569 1490 478 a Retracted coronals 1814 2319 418 1865 428 2170 397 499 i 652 1229 703 457 1081 605 I; 1341 765 1360 777 1641 1500 737 869 a 1353 568 1280 547 1350 1470 550 639 q Velars 2143 357 2100 367 1710 416 2170 319 i 1016 1318 418 905 445 609 1284 431 u 1779 1468 592 618 1693 708 1723 569 a 1763 1785 472 1292 479 514 1486 381 a Uvulars 1760 470 1807 502 1950 i 2088 633 585 1080 1119 415 1213 499 572 1288 681 u 1393 681 1669 661 1400 835 1721 a 862 1453 1468 647 673 1616 1594^_ 735 a^__ 713 Pharyngeals (Nw,tiw/ not excluded: see Table 5.27) 1618 1979 601 613 i U 1325 1301 832 1449 722 899 1729 817 a 1519 1498 652 1569 663 1586 800 743 a Glottals 2008 2059 377 2527 387 317 2473 355 i 1254 1091 377 428 984 996 371 506 u 1519 571 555 1539 789 1668 1681 815 a a  )  To illustrate the coarticulatory effects from which place of articulation can be estimated, a series of Fl vs F2 plots of the measurements from Moses-Columbian speaker MM are presented: Figure 5.10 plots all vowels in speaker MM's database, Figure 5.11 plots all the stressed vowels and Figures 5.12 to 5.17 deconstruct the Moses-Columbian  129 vowel space according to place of articulation of following consonant, starting from the coronals in the inventory and ending with glottals. Bilabial effects are illustrated at the end of the section because there were insufficient data to plot their effects on all vowels. The plots which follow should be interpreted in the context of the acoustic theory of speech production, as discussed in section 5.1. That is to say, the phonological front-back dimension can be viewed as reflected in F2 values, with a high F2 corresponding to a forward articulation. The degree of openness of the vocal tract is reflected in Fl values, with high Fl values indicating an unconstricted, open vocal tract configuration.  130 Figure 5.10 plots the midpoint values of all vowels, stressed and unstressed, in the database for speaker MM. This will give some idea of the total vowel space for this speaker. It is a remarkably triangular system in the sense that /i/, /u/ and /a/ define the three Fl ,F2 extremes of the data in a classic way. The schwa vowel is transcribed as 'E' in these figures. Moses-Columbian also has a set of retracted vowels, occurring in a small subset of roots and cognate with other Interior Salish roots which have lost a historic pharyngeal. The vowels in these roots are transcribed /e,A,o,V/ in this figure, although usually they are represented as 49  III Q/.  The phonological status of these vowels is  discussed in Chapter 6. 0 cJ  1.  ^F  _  i  Euu E E E E^uuu e E^V e i^ EE i^ u uou VE a Et 0 E^2 I^ E E E g „^ E u ^09 ° i^lii^E g-^v E jE E E • E^131E o E^ ti fa E u A EA 21 iva^ V^ V a^ .a e  E^  E^E^ e .e^  0  (9  -a lifi %a_ a z^ AA o a^ aa4?^ EA Ea a^a  E  0  a a E a^VA  4^uE6 asa a aA^A E^.aa A a aaaefa AA o  —  aEA A A A A  V  V  -3000^-2500^-2000^-1500^-1000 F2(Hz) Spkr:MM  Figure 5.10: Moses-Columbian vowels at all places of articulation, stressed and unstressed  As mentioned, Interior Salish vowels tend to delete or reduce in quality unless  131 stressed, although the different languages vary somewhat as to which option they take. To eliminate stress shift as a condition for alternation in vowel quality, the core of my analysis is confined to stressed vowels. Figure 5.11 plots the midpoint values of MM's stressed vowels.  Gi  0 0 _  Euu  E^EE  e .e le^Ii'  EE i^ i^ i  u  , E E Eu  . a^u a oaa a aE A^  0 _  0  V 0 E^00^ 000 E^ V BJEE u^ 'ii^E^ v v ^ E i^  ,  a fa aaE IA E A  o  E  a E  ak^u E^  as^A aa  A  la A A  o—  1,-^A  A  7  0 0 CV —  7  -3000^-2500^-2000^-1500^-1000 F2(Hz) Spkr:MM  Figure 5.11: Moses-Columbian Stressed Vowels The most obvious consequence of eliminating unstressed vowels is to clear out the central portion of the vowel space, indicating that stress loss in Moses-Columbian results in centralization. This figure also illustrates the considerable allophonic variation in vowel quality which is characteristic of all the speakers I have examined. For example, Fl values range from 400 Hz to 700 Hz for the front vowel IV, while F2 values range from 2000 Hz to 2500 Hz for the same vowel. In some languages (such as English) this range would encompass several phonemic vowel qualities, but in Moses-Columbian the variation is  132 predictable based on place of articulation of following consonants. This pervasive coarticulation is also documented for many Caucasian languages with similar inventories in terms of consonants at most places of articulation between the lips and the glottis and a minimal vowel inventory (Kuipers 1963, Colarusso 1988, Choi 1990) as is noted in Chapter 4. 5.2.3 Stressed vowels before coronals 7 Figure 5.12 shows vowel quality in the environment of non-retracted coronals, averaging the midpoint and offset values. This gives an extremely simplified, two-point formant trajectory, illustrating the direction and extent of formant transitions from its midpoint to offset. All vowels preceding non-retracted coronals raise somewhat in formant space, with /a/ raising most and the /i/ vowel minimally affected. Both of these effects are to be expected given the low, back constriction site for /a/ and the high front constriction site for /i/.  8  0  -3000^-2500^-2000^-1500^-1000 F2(Hz) Spkr:MM  Figure 5.12: Moses-Columbian Vowels before Coronals 5.2 4 Stressed vowels before retracted coronals /s, c, 1, 17 Figure 5.13 plots the same data for vowels before the retracted coronal series.  133 Comparison of Figure 5.12 with 5.13 shows that retracted vowels are somewhat lowered in vowel space compared to non-retracted ones, especially /A/, and that retracted vowels raise somewhat at offset, presumably in response to the coronal component of the following consonant. In every case midpoint values for these retracted vowels are lower than their non-retracted counterparts.  8  _  A  0 0  -3000^-2500^-2000^-1500^-1000 F2(Hz) Spkr:MM  Figure 5.13: Moses-Columbian Vowels before Retracted CoronaIs  134 5.2.5 Stressed vowels before velars 1k, le,x/ Turning to velars, the effects found here are entirely consistent with a high, back articulation. Figure 5.14 shows how each vowel is raised and backed in formant space. /a/ is least affected in the F2 (i.e. front-back) dimension.  i 3000  -2000 F2(Hz) Spkr:MM  Figure 5.14: Moses-Columbian Vowels before Velars  135 5.2.6 Stressed vowels before uvulars /q,q1,x/ Figure 5.15 shows the effect of uvular consonants on vowels. Comparing the position of midpoint and offset values of vowels before coronals and velars with those before uvulars, it is clear that a consistent consequence of uvular articulation is that of Fl raising. For example, to take velar (Figure 5.14) as opposed to uvular articulation (Figure 5.15), there are several indications of the lowering effect of uvular articulation. First, at midpoint, vowels before uvulars have higher Fl values than their -VK- counterparts. This is true for all vowel qualities. Second, Fl values at offset remain high for all vowel qualities. Third, in all cases the nucleus and offset Fl values of vowels before uvulars is higher than either the onset or offset Fl value of vowels before velars. The same array of facts hold for a comparison between uvular and coronal effects.  8  NI;  X0 0  .a  -3000  ^  -2500^-2900^-1500^-1000 F2(Hz) Spkr MM  Figure 5.15: Moses-Columbian Vowels before Uvulars F2 effects from uvulars are most pronounced on the /i/ vowel, where F2 lowers by 138 Hz. This is /a/ hardly shifts at all on the F2 dimension, and schwa moves forward a little bit. The /u/ vowel here shows a high Fl and low F2 at the midpoint, but Fl and F2 lowering at its offset. This is because the vowels in the forms analysed here are followed by a rounded uvular. The rounding then has the effect of lowering Fl and F2 at the  136 vowel's offset and thereby counteracting the effect of pure uvular articulation. Considering the raw formant values for vowels in the context of uvulars for both JM and MM, it can be seen that JM's values are in all cases not as high as those recorded for MM (Table 5.26). This is to be expected given the difference in female and male vocal tract dimensions, as already discussed. These differences must be born in mind, since MM's values for uvular articulation as given in Table 5.26 are high compared with those noted in Tables 4:18 and 4:19 for Arabic uvulars. At face value this would indicate that the Columbian uvulars are articulated with a constriction further back in the vocal tract than the Semitic ones. However, JM's data are closer to those recorded for Arabic uvulars, which is to be expected insofar as the Arabic studies analyse male speech. Furthermore, data from two other (female) speakers of Moses-Columbian (AB and ED) suggest that there is some genuine variability in uvular articulation not related to the known artifacts of speaker sex. Uvular data from both AB and ED show greater F2 lowering than Fl raising in the context of uvulars and in this respect are very similar to their Arabic conterparts, for which noticeable F2 effects are often reported. There is also some evidence that variation in uvular articulation is common across Interior Salish. Data from Colville (CQ) show a similar profile to that found for MosesColumbian as spoken by AB and ED, that is to say, greater F2 lowering than Fl raising (see Appendix B). These mixed results suggest that uvular articulation in Interior Salish may be accompanied by variable decrease in pharyngeal volume either because the articulation is sometimes made at the extreme end of the soft palate, or because there is accompanying general pharyngeal constriction. This in turn might be understood to indicate that Salish uvulars sometimes have a 'pharyngealized' variant. However, it was noted in section that Caucasian /cir/ is perceived as raising vowels, that is to say, lowering Fl (Colarusso 1988). This does not appear to be the case with the Salish data examined here, where there is unambiguous Fl raising for uvulars in data from MM and JM, although less so in data from ED and AB. A more detailed understanding of variation  137 in Interior Salish uvular articulation and a close comparison of the Salish data with Caucasian material is obviously desirable, but will have to await further research.6 For now I note that the phonetic variation in Salish uvular articulation reflects the production of very back as well as slightly less back articulations. For the most part the acoustic results reported here confirm the impressionistically derived observation prevalent in studies of Interior Salish that vowels before uvulars lower in quality compared to a pre-uvular articulation (see Table 5 for a summary of these observations). These results are also in line with the predictions made by acoustic modelling studies, in which post-velar articulation is cued by relatively high Fl values (Klatt and Stevens 1969, Alwan 1985).  60ne problem in carrying out such research is the real paucity of data on those Caucasian languages with pharyngealized uvulars.  138 5.2.7 Stressed vowels before pharyngeals /C,C',Sw,C'w,h,hw/ Figure 5.16 shows the effect of pharyngeal articulation on /i,a/ and /a/, collapsing the effects of both rounded and unrounded pharyngeals. There are no cases of /u/ before pharyngeals in the database. Compared to pre-uvular consonantal effects, all vowels before pharyngeals lower dramatically. All tokens of the /i/ vowel represented here are followed by /?w/, the effect of which is to lower and back the vowel considerably. Rounding effects can be seen most dramatically on /a/ in the environment of the voiced rounded pharyngeal, where a very low value is recorded at midpoint, but a higher and backer value is found at offset, thereby obscuring the full extent of lowering.  F2(Hz) Spkr:MM  Figure 5.16: Moses-Columbian Vowels before Pharyngeals In order to separate the effects of rounding and voicing of pharyngeals, Table 5.27 gives formant values for vowels in the various pharyngeal environments. Gaps in the table indicate gaps in the database for this speaker. As has been mentioned, the effect of rounding is to lower formant frequencies as a consequence of constriction at a velocity maximum (the mouth) for all three formants. This can be seen clearly for Nw/ as well as /hw/• The form recorded here, [26hwa?] 'cough, have a cold' is the only morpheme in the  139 language in which itiw/ occurs so I have given formant values for both examples of this form. The high offset Fl reading for the second token (1053 Hz) suggests that rounding is not affecting Fl at the point the reading was taken in this token. Kinkade (1967) reports that the stressed vowel in this form is fronted and fully rounded in anticipation of the following ihw/. The F2 values recorded on /a/ and /a-vowels before unrounded pharyngeals are in the range of those found for MM at the offset of /a/ when followed by a coronal, which would account for the impression of a non-backed vowel, but Fl values are much higher than those found with V-coronal sequences. The combination of relatively high F2 and high Fl gives the effect of low, relatively front vowel. Table 5.27: Plain, rounded, voiced and voiceless pharyngeal effects (MM: MosesColumbian Offset Nucleus Nucleus Offset Fl^1 F2 F!^1F2 Fl^1 F2 Fl^1 F2 Vowel -T -Sw 601 1618 1979 613 i U 1437 714 952 1718 a 592 1275 576 1375 924 1633 955 1690 a _hw  -h  i a a  789 937  1778 1767  812 898  '1876 1804  734 , 731  1549 1551  754 1053  1525 1705  140 5.2.8 Stressed vowels before glottals Finally, Figure 5.17 plots vowel quality in the environment of glottal stop. The database does not contain examples of -V!h- sequences, but they are not reported to have an effect any different from /?/, although of course they will not make a preceding vowel creaky.  8  1 3000^-2500^-2000^-1500^-1000 F2(Hz) Spkr:MM  Figure 5.17: Moses-Columbian Vowels before Glottals As can be seen, coarticulatory effects from glottals are minimal. Vowels followed by glottal stop commonly show one of two effects. Sometimes there is minimal formant movement, as on the /i/ vowels plotted here, and the vowel is simply creaky. Frequently however, the glottal stop is transparent to coarticulatory effects on the vowel from the consonant following it (-WC). In such cases the vowel preceding /?/ is affected as if the glottal were not there. This effect can be seen in Figure 5.17 on the /u/ vowel, which in the tokens plotted here is followed by a coronal and shows therefore a raising and slight fronting effect. There are many cases of both transparency and creak on preceding vowels in the total Interior Salish database, and there are cases which combine such effects. To illustrate glottal transparency, Figure 5.18 shows the effect of uvular articulation on an immediately preceding [i] vowel in the form [ciqn] 'digging'. This can be compared to the  141 form [kn pq?q] 'I'm cooked, done' in Figure 5.19, where the intervening glottal stop permits the uvular to affect the offset of the [i] vowel in a similar way. Both tokens are from MM, the same speaker whose data have been plotted all along. 6000  5000  4000  3000  2000  1000  0.0^ C.^  0.3^ r.4 mm29.d  0.6  Figure 5.18: Uvular effects: ciqn 'digging' (Moses-Columbian:MM)  142  6000  um210.d  Figure 5.19: Glottal transparency: kn p'i2q 'I'm cooked, burnt' (Moses-Columbian:MM) In no context does /2/ lower a preceding vowel. Vowels are sometimes reported with more lax allophones before glottals, but never with the full lowering effect of uvulars and pharyngeals. Of course, glottals do not raise vowels in the way that pre-uvular articulation can either. Given that co-articulation is cleary the norm in Interior Salish, glottals are unique in not imposing a co-articulatory effect on preceding vowels. It must be noted however that there are no cases of [a/2], but only of [all]. It would appear that the contrast between /a/ and /a/ is neutralized in this environment. 5.2.9 Bilabials and Summary Unfortunately, the database for MM contains no tokens of bilabial /p, p'/ rightadjacent to a stressed vowel. There are a number of examples of /-Vm/, but it is difficult to abstract the effect of nasal resonances and anti-resonances from the waveform so as to  143 isolate reliably the effect of labial closure. However, Figure 5.20 shows an example of Moses-Columbian bilabials having the expected effect of lowering all formants, given that closure takes place at a velocity maximum for all resonances of the vocal tract. This particular example (from speaker ED) is all the more striking since the /p/ occurs after /17 with its characteristically high Fl. At midpoint of the vowel-pharyngeal sequence Fl is 885 Hz, F2 1807 Hz, F3 2654 Hz. At offset the values are Fl 687 Hz, F2 1657 Hz. F3 has disappeared. This form also shows the glottal effects discussed above. While the root is phonologically 4yaT', the glottalized pharyngeal is realized as creaky offset, possibly followed by Pt In either case, the glottal aspect of /C7 is clearly transparent to formant transitions triggered by the following /p/. 6000  5000  4000  3000  2000  1000  0 00 13954—  6977-  -6977 00  0.1^0.2  0.3^0.4^0.5^0.6 40a .d  Figure 5.20: Moses-Columbian labial effects: yaVp(qin) 'many, lots': Speaker: ED This investigation of Moses-Columbian has shown that consonants at every place of articulation trigger coarticulatory effects reflecting the place of articulation of the consonant itself. The exception is the class of glottals, which render preceding vowels  144 creaky and do not themselves affect preceding vowel quality, although they may be transparent to effects from following consonants. Bilabial, coronals and velars pattern together in conditioning relatively low Fl values on left-adjacent stressed vowels. Uvular's, pharyngeals and retracted alveolars pattern together in that these and only these segments are preceded by vowels which are consistently higher in Fl (and hence lower in quality) than those found in other consonantal environments. The main acoustic manifestation of this is raising of the first formant although there are clear F2 effects also. Both effects can be linked with reduced pharyngeal cavity volume. Table 5.28 collects the averaged values at offset for vowels at each place of articulation. In effect it is a summary of Tables 5.26 and 5.27. The values recorded in this table for pharyngeals are those found in the environment of unrounded pharyngeals, specifically in the context of/-h/ for the a-vowel, and of /-S/ for the a-vowel. Full details of pharyngeal values are given in Table 5.27. There are no examples of /i/-pharyngeal sequences. This is because there were no cases of stressed /i/ before a pharyngeal in MM's database. There are /i/-pharyngeal sequences in the database for several other MosesColumbian speakers though. For example, [kihana2] 'teenage girl' (Figure 5.48) shows an Fl reading of approximately 800Hz at the vowel-pharyngeal boundary. F2 is in the 2400 Hz region, having lowered from 2800 Hz. The Fl value here is higher than the reading for i/-uvulars given in Table 5.28, reflecting a more open vocal tract configuration. Uvular and pharyngeal effects are obviously distinct for /a/ and /3/: Fl is lower in the environment of urffounded pharyngeals than it is in the environment of unrounded uvulars. F2 values are roughly comparable for the /a/-vowel, but /a/ before pharyngeals has a higher F2 than before uvulars. This would make it more front. Further confirmation of the observation that pharyngeal articulation conditions a higher Fl than uvular articulation is presented at the end of section where the spectral attributes of// and  N are contrasted. Glottals do not lower /V or /u/ in any way. The values found here are comparable to  145 those found in pre-uvular environments. The /a/ vowel is extremely interesting. It has an average Fl of 789 Hz before glottal stop, 924 Hz before pharyngeals and 835 Hz before uvulars. a/-7 is thus not as low as before uvulars or pharyngeals, but it is lower than before pre-velars. This is consistent with the hypothesis that pre-velars tend to raise preceding vowels, post-velars lower, and glottals have little to no effect since they do not have a supralaryngeal gesture other than lack of constriction to impose on the preceding vowel. As a consequence, they neither lower nor raise preceding vowels. Table 5.28: Moses-Columbian ilace of articulation effects summarized: offset values Velars Uvulars Pharyngeals V Labials^Coronals Glottals (JM)^(' OM (MM) (MM) L(MM) F2 Fl F2 F2 IJ Fl F2 Fl F2 Fl F2 Fl I^Fl 633 1950 2527 i 245 2281 319 1710 355 374 371 984 u ' 424 1164 ' 366 1240 319 1710 572 1119 a 579 1411 541 1966 569 1693 835 1669 924 1633 789 1668 a 511 _1074 478 1667 381 1292 735 1616 898 1804  5.3 Pharyngeal Articulation This section seeks to clarify the empirical and phonetic issues of Salish pharyngeal articulation namely, whether the relevant segments are indeed pharyngeal in place (they are reported to have a uvular component in some languages), what their manner of articulation is and how they compare to what is known about pharyngeals in other languages, in particular, Arabic. First, Section 5.3.1 briefly reviews the relevant aspects of Arabic pharyngeals so as to establish points of reference for the Salish data. Then in section 5.3.2 the results of spectrographic and waveform analysis of pharyngeals from six of the Interior Salish languages is presented and compared to the Arabic data. In general the results of acoustic investigation support Kinkade's (1967) analysis of a set of pharyngeals throughout Interior Salish and indicate that Salish pharyngeals are very much like their Arabic counterparts in terms of manner and place of articulation. By way of summarizing the results of this investigation of Interior Salish pharyngeals, Table 5.29 collects formant information on all pharyngeals represented  146 spectrographically in the text. In brief, it is quite clear that high Fl is a consistent cue to Salish pharyngeals as it is with Arabic ones. F2 tends to be quite sensitive to adjacent vowels, as is reported for Arabic also (Alwan 1986). The Fl range in these data are 495 Hz to 1052; F2 from 1138 Hz to 2021 Hz. F2 is highest in the context of following /V. Fl is lowest in the context of a rounded pharyngeal. It is impossible to say, without the benefit of X-ray or comparable data, that the Salish pharyngeals have 'the same place of articulation' as the Arabic ones. However, of direct relevance in the comparison of Salish with Arabic is that the range of Fl values documented here for Salish is directly comparable to that found for Arabic (section 4.4.1). Table 5.29: Formant values of Interior Salish pharyngeals examined in text H Language If Speaker If Segment^if Fl (Hz)^if F2 (Hz)^H F3 (Hz)^I Figure 1393 964 #CaCEArabic SW 21 2241 1633 818 #Sa21 2140 1595 882 -afi23 1630 2750 1052 #ha 24 1180 #si689 Chewelah AS 25 2887 1434 #Si695 MS Spokane 27 2804 1273 #si732 PF 28 1239 2788 875 -S# 29 1281 2908 -aT'C'm807 30 1462 2264 712 Colville #TiCQ 31 1469 2569 -f# 826 32 1449 2463 802 -Sp 33 1246 2351 658 ViaCoeur d'Al -LN 34 2198 1478 616 -cTs35 1642 1440 495 36 -Ef'wn1419 2239 678 #1iShuswap BD 37 1203 2374 760 38 PTP1241 2300 #1S'i894 39 2113 1429 727 #SiNI DS 40 2116 1141 559 -aSp# 41 2422 799 1138 -c# 42 1871 1438 703 AB Ms-Cm 43 #Swa2445 1615 1051 -aTa44 2475 1782 ' 865 -af'S'a45 2021 2522 832 #hi46 1706 2593 1034 -h# 47 2361 2823 811 -iha48 1613 2386 620 49 -ahwa-  147 5.3.1 Arabic pharyngeals The phonetics of Arabic pharyngeals are discussed in section There are several salient details for purposes of the ensuing comparison of Arabic pharyngeals with Salish which I outline here: (i) Arabic pharyngeals are produced with a constricted pharynx although descriptions vary on how this is achieved and the exact location of the maximal constriction; (ii) /5/ is often 'creaky' and/or glottalized on release; further, [2] is sometimes an allophone of IS/; (iii) /5/ is a voiced approximant in Iraqi (Butcher and Ahmad 1987), reported as a spirant in Palestinian (Blanc 1953), as fricative-like in Tunisian (Ghazeli 1977), a sonorant in Lebanese (Klatt and Stevens 1969); (iv) /h/ is a voiceless spirant in Moroccan (Harrell 1957) and Palestinian (Blanc 1953) or approximant in Iraqi (Butcher and Ahmad 1987), produced lower than /5/. Some descriptions suggest the constriction area is less for fh / than /V. (v) Fl is high for both /5/ and /h/, whereas F2 tends to be low (Butcher and Ahmad 1987, Laufer and Baer 1988, Alwan 1986, Ghazeli 1977). The realization of /1,h/ is more readily appreciated by reference to spectrograms, where the manner of articulation is visually observable. The literature contains a number of examples of pharyngeals in various Arabic dialects, as noted in the previous paragraph. Below are several examples of pharyngeal articulation in Colloquial Egyptian Arabic as spoken by a male native speaker in his early thirties (SW). All forms analysed here were elicited in the same way as the Interior Salish data, that is to say, they are citation forms of words existing in the language, with each token repeated twice. Colloquial Egyptian Arabic is reported to have emphatic as well as plain pharyngeals (Mitchel 1956, Khalafallah 1969, but cf Harrell 1957 who says he cannot tell the difference between emphatic and non-emphatic uvulars or pharyngeals). The examples below are of non-emphatic pharyngeal articulation so as not to introduce the complication of emphasis into the discussion. Figure 5.21 is of the word [Sadu] 'enemy'. The uppermost display in this and following figures of this sort, is of course the spectrogram itself. Immediately below the  148 spectrogram is the corresponding waveform, on a matching time-scale. The large-scale expanded waveform at the bottom of the figure is provided to facilitate closer examination of the moment of pharyngeal articulation in the token. The expanded waveform is always of the pharyngeal segment in question with as much adjacent material as can be included. The black line beneath the unexpanded waveform indicates the portion which is expanded in the lower display.  0 0  0.2^0.3  0.4^0.5  0.6  07  a 13200  6600  -6600 0.0  0.1^0.2^0.3^0.4^0.5^0.6^0.7 sw112.d  1320C  1r  660C  •  — — — -660C-  C o  •  0.1  0.2  •  3w112.101.g.d  Figure 5.21. Saciu 'enemy' (Colloquial Egyptian Arabic: SW) The expanded waveform in Figure 5.21 shows this word-initial voiced pharyngeal having a period of more than 100ms very low amplitude but fairly regularly voiced  149 initiation, corresponding to low-frequency (1000 Hz and below) energy on the spectrogram. There is no 'burst' portion such as found by Butcher and Ahmad (1987) with some of their Iraqi tokens. Butcher and Ahmad (1987) report an average Fl of 964 Hz, F2 of 1467 Hz at the pharyngeal-vowel boundary of the form [Taft For [Ta!du] 'enemy' values at the pharyngeal-vowel boundary are Fl 921 Hz, F2 1393 Hz and F3 unreliable. There is no appreciable evidence of frication in these or Butcher &Ahmad's (1987) forms. The Egyptian form is thus quite comparable to the Iraqi fonns7.  7With one exception: the duration of the pharyngeals in the two tokens of [cadu] 'enemy provided by SW is amost twice that of the example in Butcher and Ahmad (1987). Too much cannot be made of this difference though, since there are insufficient data from SW to make reasonable statements about the duration of his pharyngeals.  150 Compared to the waveform (shown here expanded) of the word [?aiwa] 'yes' in Figure 5.22, the peculiar properties of initial [f] are obvious: in the Pad sequence there is no comparable period of low-amplitude wave-form and instead the waveform immediately and regularly gains amplitude as it moves from release of the glottal into the vowel. At the glottal stop-vowel boundary of Paiwa] Fl is 818 Hz, F2 1633 Hz. F3 is not visible but the formant-tracker provides 2241 Hz. Thus, Fl is just over 100 Hz higher and F2 240 Hz lower in [cad as opposed to [2a-]. Note also that there is no formant movement in [?aiwa] until the [a] shifts into [i], at which point Fl lowers and F2 raises. Critically, there is no formant transition at the [2-a] boundary, indicating that the supra-laryngeal vocal tract is in a position for [a] as it forms and releases the glottal stop.  sw83.expandedA  Figure 5.22: ?ai(wa) 'yes' (Colloquial Egyptian Arabic: SW) The voiced pharyngeal, when intervocalic as in [sali:d] 'happy', does not have the period of low amplitude initiation seen above in Figure 5.21.  151 Figure 5.23 shows intervocalic [1] to be entirely resonant, although close inspection of the expanded waveform suggests that the [1] is none the less relatively constricted compared to flanking vowels, having a curiously jagged' waveform at a point corresponding to the transcription of a pharyngeal. Isolation and playback of this portion of the waveform confirms that this is indeed the most pharyngeal moment. Finally, this form shows clear raising of F2 and lowering of Fl as the vocal tract moves towards the following [Nvowel. This of course is entirely in line with the documentation of pharyngeals with high Fl and low F2. Formant values at the point of maximal pharyngeal constriction in this token as judged from the period of low amplitude waveform are Fl 882 Hz, F2 1595 Hz, F3 2140  0 0^0.1^0.2^, 0.3^0.4^0.5  0.6^0.7^0.8^0.9^1.0^1.1^1.2^1.3^1.4 sw24.d  VA 00  .^.  0.1  0.2  sw24.expanded.d  Figure 5.23: saTal 'happy' (Colloquial Egyptian Arabic: SW)  152 As a final reference example of pharyngeal articulation, consider Figure 5.24. This shows a word-initial voiceless [h] in the form [halal)] 'he milked'. The expanded waveform in this case is very much like that reported by Butcher and Ahmad (1987) indicating a noisy unvoiced sound with an aperiodic, fricative-like waveform. The higher amplitude portion may indicate the difficulty controlling regular air-flow through the constriction site for [h], as well as the considerable force with which air is forced through the constriction. At the pharyngeal-vowel boundary Fl is 1052 Hz, F2 1630 Hz, F3 2750 Hz.  6842  00^04^0.2^0.3^0.4^0.5^1 0.6^0.7^0.8^0.9^1.0 sw94.d1  II  3421  .**0444#01)****6"1111 -3421 —--00  0.2 sw94.exica1ded.d  .^t^.  0.3^0.4  Figure 5.24: halab 'he milked' (Colloquial Egyptian Arabic: SW)  153 5.3.2 Interior Salish Pharyngeal Articulation As mentioned, the proper identification of what are now termed pharyngeals in Interior Salish historically has been something of a problem. The sounds are often difficult to hear as consonants and descriptions of their place of articulation vary between uvular and pharyngeal, depending on the language. This has been discussed in In order to address the question of what Interior Salish pharyngeals are phonetically and how they might vary across the different languages, this section presents the results of acoustic analysis of samples of pharyngeals from each of the Interior Salish languages. Collection and analysis methods are as documented in Section 5.2.1 of this chapter. Kalispel-Spokane From a phonetic point of view, as we shall see, it was not unreasonable for Vogt to transcribe VV in sequences in Kalispel where in other Interior Salish languages a pharyngeal-vowel sequence is usually transcribed8. Phonologically, the transcription makes less sense, since pharyngeals pattern as full-blooded consonants in terms of CVCand -(V)C reduplication throughout IS and behave as consonantal resonants with respect to svarabhakti vowel insertion and the phonetic manifestation of glottalization. The phonological behaviour of Interior Salish pharyngeals is discussed fully in Chapter 4. By way of examining the sorts of sounds Vogt (1940) transcribed as double -aavowels, Figure 5.25 is a spectrogram of the form [Sieast] 'char' as spoken by Alex Sherwood, a Chewelah9 speaker recorded by Barry Carlson in 196910. Although this is a word-initial pharyngeal as in the Colloquial Egyptian Arabic form [Tactu] 'evening' (Figure 5.21) there seems to be little indication of extreme constriction at onset. One token for AS shows some initial noise before voicing begins. The one given here shows less but the 8Vogt was dealing with a very limited corpus and may not have had forms with pharyngeal reduplication. However, he did know Reichard's Coeur d'Alene material as noted in Section 9Carlson (1972:iv-v) notes that Chewelah is 'virtually identical with Kalispel but is considered distinct by the Indians'. 101 thank Barry Carlson for allowing me to analyse data from Alex Sherwood and Margaret Sherwood.  154 waveform in all cases transposes very quickly into what sounds like a regular, voiced [a] with Fl 689 and F2 1180, indicating a low, open articulation comparable to the BE /a/ of 'father'll. 6000  "T'"'' "I^''"I"^'^I^'^I^'^I^'^I^'^I^'^I^'^I  5000^-E  4000^-F  3000  2000  1000  5^  00  0.1^0.2  1.3  0.3^0.4^0.5^0.6^0.7^0.8^0.9^1.0^1.1^1.2^1.3  ^4^foo22.d  0  0.1  foo22.big.d  0.2  0.3  Figure 5.25: Tiedst 'char' (Chewelah/Kalispel: AS) Figure 5.25 (and 5.27, 5.28) should be compared to Figure 5.26 of the MosesColumbian word [?ay' kw4st] 'tomorrow'. This form shows an [-ay-] sequence, without the influence of a pharyngeal. The contrast is relevant since the form Mast] 'char' could be, and sometimes is, transcribed as [Tayast]. It is not an unreasonable transcription, and limy own formant values for the BE [A] vowel are Fl 900, F2 1100. Catford (1988:161) cites Fl 750, F2 940 as average values for Cardinal Vowel 5 [A].  155 shows how pharyngeals can give the percept of an [a]-vowel.  C C^ C.3^ C.4^ C.5^ C.E^ C.? mrn1  Figure 5.26: ?ay' kw4st 'tomorrow' (Moses-Columbian: MM)  156  Figure 5.27 shows Margaret Sherwood's pronunciation of Fiastrchaf, the same word given by AS in Figure 5.25, but in Spokane, which appears to have retained some more consonant-like constriction for the initial pharyngeal. Fl during the pharyngeal is 695 Hz, F2 1434 Hz, F3 2887. 6000  5000  4000  3000  2000  1000  0 0  t 1.5 .1'  11828 5914  -5914 .......^  .  ..I^..■  0 0^0.1^0.2^0.3^0.4^0.5 10224 ^ .^...  ....  0.6^0.7^0.8^0.9^1.0^1.1^1.2^1.3^1.4^1.5 foo24.d  I  5112 — 0  -51127-1— 0 0  ^  0.1^ ms24.expanded.d  0.2  Figure 5.27: Ficestrchar' (Spokane: MS)  157 Figure 5.28 is of a Spokane speaker in her sixties (PF) pronouncing [Timt] 'get angry'. As with the Kalispel-Chewelah form, word-initial [c] has a period of lowamplitude initiation, but this speaker moves into a full-amplitude vowel more rapidly than AS and far more rapidly than MS. Fl 732 Hz, F2 1273 Hz, F3 2804 Hz. 6000  5000  4000  3000  2000  1000  0 00  1.0  'T  18472  9236--  -9236: 00^0.1  0.2^0.3^0.4^0.5^0.6^0.7^0.8^0.9^1.0 pf45.d  18472  9236  -9236  0 0  0.2  0.1 pf45.big.d  Figure 5.28: Timt 'get angry' (Spokane: PF)  158 Word-finally PF's pharyngeals tend to voiceless expiration and loss of energy between formants, as seen in Figure 5.29. Fl is 815Hz , F2 1239 Hz, F3 2788 Hz as the waveform loses amplitude. F3 in particular decays at offset. 6000  5000  4000  3000  2000  1000  0.1^0.2^0.3^0.5 pf4b.d  ^  0.6  ^  kO\ 0.0  ^  0.1  ^  0.2 pf46.big.d  Figure 5.29: p'aS 'burn' (Spokane: PF)  0.7  159 That this is a word-final phenomenon is clear from the form [hec ya.T'C'mi] 'people are gathering' (Figure 5.30) where the pharyngeal remains fully voiced and fully resonant despite phonemic glottalization realized phonetically as varying periods of glottal closure (and some creak) after the pharyngeal itself. Formant values are Fl 807 Hz, F2 1281 Hz, F3 2908 Hz at the midpoint of the second pharyngeal.  6000  5000  4000  3000  2000  1000  9  19552  ,I„  In  ,,,,,,,^,,,,,,,,^•  ^' 1  9776  -9776  0.0^0.1^0.2^0.3^0.4^0.5^0.6  0.7^0.8^0.9^1.0^1.1^1.2^1.3^1.4 pf 54.d  Figure 5.30: [hec yaMmi] 'people are gathering' (Spokane:PF) Colville-Okanagan Colville pharyngeals have been described as difficult to hear. Mattina (1987:vii) in his dictionary of Colville-Okanagan notes that 'the most likely cases of phonetic uncertainty  160 involve the glottalized resonants and the pharyngeals12. The speaker (CQ) who provided me with the forms shown below is a respected story-teller with considerable knowledge of dialect variation within the Colville-Okanagan continuum. His word-initial pharyngeals (Figure 5.31) start off with what seems to be a characteristic low-amplitude section followed by gradual increase (full amplitude at 130 ms into the utterance). Formant values are Fl 712 Hz, F2 1462 Hz, F3 2264 Hz at voice onset. 6000  s on:-  4000:-  3000-  2000-  00 7450  0. 7n  3725  -3725 0.0  0.4^0.5^0.6^0.7 loc99.d  Figure 5.31: Sitmn 'teeth' (Colville: CQ) 12This problem is not specific to Colville though. Low vowel-pharyngeal sequences are notoriously difficult to transcribe in proper order without the aid of phonological information. The Nxa'amxcin word for 'crow' is transcribed xaixaz,, but one speaker (AB) wrote it variably as 'Oak, nana.  161 Word-finally, [S] tends to voiceless aperiodic release as in the Spokane forms above. In Figure 5.32 Fl is 826 Hz, F2 1469 Hz, F3 2569 Hz just before cessation of voicing. 6000  5000-  4000-  3000-  2000-  1000-  o7  0 p 7930.j.^11  3965-  ^ye,*  0 0^0.1^0.2^0.3^0.4^0.5^0.6^0.7 cq57.d  cq57.biq.d  Figure 5.32: pal 'grey' (Colville: CQ)  162 This amplitude decay in the waveform occurs before a voiceless obstruent also, with corresponding loss of energy in the higher region of the spectrum. This can be seen in Figure 33 where F3 and F4 pretty much disappear. Fl is 809 Hz, F2 1439 Hz, F3 2475 Hz just before amplitude reduction. (F1 is 802 Hz, F2 1449 Hz, F3 2463 Hz during reduced period). 6000  5000:  4000:  3000-  2000-  011411011' T8I14,11:  ion"-  38401  -3840: 0 0  I^I^I^1^..I^I^I^■^I^■^I^I^I 0.1^0.2^0.3^0.4^0.5^0.6^0.7 cq79.d  IH^I^ 1^.■^1.^,.,.^,^. 1^' 0.8^0.9^1.0^1.1^1.2^1.3  oq79.expanded.d  Figure 5.33: [xwuyt ciya^'they have arrived' (Colville: CQ) Coeur d'Alene The speaker (LN) who provided me with the forms shown below also served as  163 Reichard's consultant in 1935-36. The form in Figure 5.32 is of a word-initial IV followed by a low vowel which is backed to [a] by the following /r/. I include this form here since Coeur d'Alene In is a member of the faucal class for this language and plays a role in various phonological harmonies which lower and retract vowels (section  164 The obstruent quality of the glottal gesture for /?/ is apparent from this spectrogram—the form starts off with a classic stop release. Full amplitude on the vowel is reached very quickly (within 4 glottal cycles), which again differentiates the articulation of /?/ from /l/. Fl is high (658Hz) and F2 low (1246 Hz), F3 2351 Hz even at voice onset, illlustrating the well-documented effect of retracted Coeur d'Alene In on preceding vowels. As with American /r/, Coeur d'Alene In lowers F3 dramatically. 6000  5000  4000  3000  2000  1 0 00  000  0.7  10964—  5482 -  istiofft/0004-44-44.0-mw+-4,----00  0.1^0.2^0.3^0.4^0.5^0.6^0.3 1n74.d  10964  5482 — — — —  — — — -5482 — 0. 0  frj 0.1  0.2  1n74.1n1g.d  Figure 5.34: ?ar 'much, plenty' (Coeur d'Alene: LN) Consonant-resonant sequences in Interior Salish are typically phonetically separated by a brief svarabhakti vowel, the quality of which is determined by consonantal  165 environment. This is illustrated in Figure 5.35 with the form [cic C6/'[ 'I'm angry'.  00  0.1  0.2  0.3  0.4  0.5^0.6 1n219.d  12150  a-  00  4/Of'1, 1  0.1 1n219.expan3ed.d  0.7  0.8  0.9  1.0  1.1  J 0.2  Figure 5.35: cic Tgy"I'm angry' (Coeur d'Alene: LN) After release of the [s] of Rs], there is a schwa-like vowel of one glottal cycle, then a 70 ms period of the now-familiar low-amplitude irregular waveform corresponding to the period of maximal constriction. Fl is 616 Hz, F2 1478 Hz , F3 2198 Hz as voicing resumes. Voicing and amplitude become regular as the vocal tract moves into the following letvowel. This particular example shows creak coming out of the pharyngeal and going into the vowel. To my knowledge this form is not transcribed with [V]. Again, as with the Spokane forms above, this phonetic creak is quite different from the realization of  ^  166 phonemically glottalized /V/. As seen in Figure 5.36 [mcf'wns] 'he broke it', phonemic glottalization of pharyngeals can phonetically decompose into a [?5] sequence, leaving in this case a period of voiceless aspiration showing the formant transition from [e] to [S"']. This example also shows the low-amplitude start-up for [VI. F3 lowering in this form is also visible, but may be exaggerated by the rounding on [Tw]. Fl, F2 and F3 at glottal closure are 546 Hz, 1732 Hz, 2384 Hz. At onset of voicing for [Vv] the values are 495 Hz, 1440 Hz and 1642 Hz respectively. These relatively low formant values undoubtedly reflect the rounding on [M.  ^0  6000  "F "—'1""1"1" l'—{1-1" 1'1^'^I^I^'^I^1 '  5000^-E  4000^-E  3000^-E  2000  1000  ^ 0.0  0 0^0.1^0.2  0.3^0.4^0.5^0.6^0.7^0.8^0.9^1.0^1.1^1.2 1n51.d  ilb,^p  ri 4^r  A^/Itl^A =  6  ,AJ^/VI^r\ " '7?  0.2  1n51.expanded.d  Figure 5.36: mET'wns 'he broke it' (Coeur d'Alene: LN)  6  L  167 Shuswap The three Northern Interior Salish languages, Lillooet, Shuswap and Nie?kepmxcin (Thompson) are considered a sub-branch of Interior Salish. I deal here with Shuswap and then Niekepmxcin. The Shuswap pharyngeals are consistent with what has been discussed for the Southern Interior languages. Word-initially [1] shows low-amplitude irregular initiation, with Fl 678 Hz, F2 1419 Hz, F3 2239 Hz (Figure 5.37).  1 0 0^0.1^0.2^0.3  0.4^0.5^0.6^0.7 39 .d  25232  12616  0 0  0.2  0. 1  bd39.expari1ed.d  Figure 5.37: Sipkn 'I'm angry' (Shuswap: BD)  168 In the form [pfpeCt] 'grey' (the form is reduplicated with syncope of the prefix vowel conditioned by stress loss) Fl and F2 clearly raise from their position right-adjacent to initial [p] and there is mirror-image lowering just before the point of bilabial closure (Figure 5.38). Maximal formant values are Fl 760 Hz, F2 1203 Hz, F3 2374 Hz respectively. The pharyngeal is very resonant with most excited Fl, F2 and F3, but clear F4 and F5.  1:d116.d  bd116.big.d  Figure 5.38: pipeft 'grey' (Shuswap:BD) The Shuswap version of [51 can be seen in Figure 5.39. Phonologically the form is^'dragging things around', but phonetically it is [lTi7im] with clear transition to the  ^  169 stressed vowel just before and during the glottal stop portion of [S']. Midpoint values of the pharyngeal portion are Fl 894 Hz, F2 1241 Hz and F3 2300 Hz. F2 and F3 rise notably to following IV. I^"'I""'" 'I"^'^I^'^I^'^I  6000  + +^+ + + +  5000^+  + +^+ + + +  4 000 -  3000 -  2000-  1000  -  ryi  09  8666  -8666 00  0. 1  0.2^0.3  0.4^0.5 bd130.d  0.6^0.7^0.8^0.9  17332  8666  -8666 00  ^  0.1^ bd13 0 . big d  0.2  Figure 5.39: 1Ci2im 'dragging things around' (Shuswap:BD) Nle?kepmxcin (Thompson) Figure 5.40 shows a word-initial pharyngeal, spoken by DS, a male speaker of IsRekepmxcin. Isolation and playback of the portion preceding transition into [i] suggests that this initial consonant has some uvular quality. It sounds much like a uvular trill. It was difficult to read format values for this consonant, but the formant tracker provides Fl  170 446 Hz, F2 1386 Hz with F3 unreliable. Readings taken at the point of transition into the /i/ vowel are Fl 727, F2 1429, F3 2113. The increase in Fl from a relatively low 450 Hz value to a value otherwise co-occurring with pharyngeals suggests that this segment moves from a uvular place of articulation to a pharyngeal one. This speaker also has some creak right at transition into the [i] vowel, a moment clearly corresponding to release of a low back articulation and into a high front one13.  0 0^0.1^0.2^0.3^0.4^0.5^0.6^0.7 s4.d ^  0 8^0.9^1.0^1.1  --w2Aryi1/2 0.1  0.2 ds4.expanded.d  Figure 5.40: Sis 'shrink' (1•14e2kepmxcin: DS) It has been mentioned that [Ta, all sequences are difficult to distinguish 131t should be noted that DS, recorded in the summer of 1991, had had bronchitis that winter. Laryngeal gestures in his tokens cannot be confidently attributed to Nlakapmxcin rather than the left-overs of bronchitis.  171 phonetically, although there are often phonological processes such as reduplication which will serve to order the two segments. It would seem that the diagnostics of low-amplitude and waveform-shape can be used to disambiguate [Ca, as] sequences. For example, the transcription of [cCap] 'ripped' as CCVC rather than CVCC is derivable from the characteristics of the waveform, see Figure 5.41. 6000  5000  -  4000  3000  2000  1000  Co^  c  21674 —  10837  0 0^0.1^0.2^0.3^0.4 ds122.d ^i  0.5^0.6^0.7^0.E.  21674  10837  -10837h-7  Co  0.1^  0.2  Figure 5.41: cCap 'ripped' (Islle?kepmxcin: DS)  The waveform is markedly different immediately following release of the initial affricate compared to the full amplitude vocalic portion preceding the final [p], cf. for example the low amplitude irregular waveform noted for #S- in Figures 5.25, 5.27, 5.31 and 5.37. In confirmation of the analysis of this form as CCVC rather than CVCC, the waveform of /a/  172 does not expire to low-amplitude before final [p] as it seems to in other cases of MI before a consonant, such as in Figure 5.38. Of further note is the fact that the 'pharyngeal' portion of this waveform also sounds much like a uvular trill". However, an FFT spectrum taken at the onset of [S] shows the harmonics of the spectrum decreasing in amplitude from the frist harmonic of 109 Hz (=F0) until a slight amplitude peak at 914 Hz. If this highest harmonic (914 Hz) is interpreted as Fl, then it is very high and suggests a very open vocal tract, which is not consistent with relatively high uvular constriction. On the other hand, at the transition from [f] to [a], Fl is 559 Hz, F2 1141 Hz and F3 2116 Hz. Fl here is low for a pharyngeal and in line with the understanding that while Fl is a cue for 'low' articulations, Fl will be lower for uvulars than for pharyngeals. The vowel's steady state values are 600 Hz, 1179 Hz and 2160 Hz respectively.  14 I isolated and played this portion to two phonetically-oriented linguists, both of whom thought it sounded like a uvular trill.  173 Word-finally the 1■14e2kepmxcin pharyngeal seems very much like the Spokane ones, with voiceless aspirated expiration, as in Figure 5.42. Formant values for the pharyngeal are Fl 799, F2 1138, F3 2422.  0 0  ^  0.1  ^  0.2^0.3^0.4^0.5^0.6^0.7^0.8 ds160.d ^ 4  14714  7357  -7357  0.0  ^  0.1^ ds160.expanded.d  0.2  Figure 5.42: paf 'bleached by the sun; grey' (Nlekepmxcin: DS) Moses-Columbian I turn now to pharyngeal articulation in Moses-Columbian, which is unique in Interior Salish in adding phonemic [h,hw] to a pharyngeal inventory which in the other  174 languages contains only voiced pharyngeals.15 None of the Moses-Columbian pharyngeals I have recorded sound at all uvular-like in the way that some N1eikepmxcin pharyngeals do. Voiced pharyngeals: /C,CY do not occur initially in Moses-Columbian (Kinkade 1967). Instead, Figure 5.43 shows word-initial [SW] in the form [Cw611 'bright, shiny, glisten'. The spectrogram of this word shows no energy above 1000 Hz for about 200 ms, matching the long period of low-amplitude initiation before the vocalic portion begins.  15Although, as noted, /hw/ occurs in only one morpheme and Okanagan may also have /h/.  175  6000  5000  +  + + +  4000  +  + + +  3000  +  + + +  2000  1000  0.3  i  -6839-  ^.1^I•.. I,... ^.I  0 3^0.1^0.2^0.3^0.4^0.5 i^  0.6^0.7^0.8^0.9^1.0^1.1^1.2^1.3^1.4^1.5  ab156.d  8372  4186  -418675-00  0.1  ab156.expanded.d  0.2  Figure 5.43: '1' 'bright' (Moses-Columbian: AB)  176 An example of intervocalic [T] is given in Figure 5.44. This form is transcribed [nxwaTank] but may better be [neSank/. In any case, the pharyngeal is very resonant in this position and there is no question of suppressed initiation. It is like intervocalic pharyngeals in this respect, including the Arabic one in Figure 5.23. (F1 1051 Hz, F2 1615 Hz, F3 2445 Hz). 6000  5000r-  4000-  3000-  2000-  1000-  1111.60HOWww11,.  ,  0'. 0 0 12080  K,^1.4 I^'^I  6040-  -60400 0^0.1^0.2^0.3^0.4^0.5  0.6^0.7^0.8^0.9^1.0^1.1^1.2^1.3^1.4 ab137a.d  1 0.2  0.1  ab137.expanded.d  Figure 5.44: nxw(a)idnk 'cave; hole in a hill' (Moses-Columbian: AB)  177 The form [scyaT'S'inix] 'they're having a gathering' (Figure 5.45) shows a glottalized [V] which is phonetically a little different from the Spokane form given by PF, Figure 5.30. While [C'] is still fully resonant and regularly voiced in this form, glottalization is realized with what may be a source effect along the lines of breathiness. At midpoint of the first pharyngeal, Fl is 865 Hz, F2 1782 Hz, F3 2475 Hz. 6000.....,..„,1^1,^,^I^'^  I^I^  1'111'  0.5^0.6^0.7^0.8 ab146a.d  abl4E.expanded.d  Figure 5.45: scyaT'C'mix 'they're gathering' (Moses-Columbian: AB) Voiceless pharyngeals: Word-initial voiceless [h] is shown in Figure 5.46. [h] has an obvious formant structure with concentrations of energy in the 1000+ Hz, 1800 Hz, 2500 Hz and 4400 Hz  178 regions. The waveform shows a strong aperiodic noise component of varying amplitude, very much like the Arabic [hi in Figure 5.24. Formant measurements taken at the point of transitions to the following [i] vowel are Fl 832 Hz, F2 2021 Hz, F3 2522 Hz. These values are not exactly the same as those discussed for CEA [halabi, but of course the following vowel is different. An -ha- sequence can be seen in Figure 5.48. Fl for MosesColumbian [h] in this form is nonetheless high and in the region of that found with voiced [1].  0.5^0.6 ab164a.d  0.7^0.8^0.9^1.0^1.1  ab164.expanded.d  Figure 5.46: himt 'angry' (Moses-Columbian: AB)  179 Word-finally, as can be seen in Figure 5.47, [h] maintains its formant structure and is similarly noisy, but with reducing amplitude as the constriction is released.  0 0^0.1^0.2^0.3^0.4^0.5^0.6 ab168a.d  ab168.expanded.d  Figure 5.47: ph 'grey' (Moses-Columbian: AB)  180 Intervocalic /h/ does not appear to voice and shows clear movement of formants from the preceding to the following vowel. [h] does not influence the midpoint or steadystate quality of preceding [i] in the form [Icit. nal 'teenage girl' Figure 5.48, although there are certainly cases where [h] has effects intruding more into the nucleus of preceding vowels. [h] in this form still shows obvious co-articulatory effects of Fl rising and F2 lowering as the vocal tract configures for the [h]. 6000  kift1041C. iii1119 ,1099 11.  -  0.0  0.1  0.2  0.3^0.4  0.5^0.6^0.7^0.8^0.9^1.0^1.1^1.2^1.3^1.4 ab171.d  ab171.expanded.d  Figure 5.48: kiha na 'teenage girl' (Moses-Columbian: AB)  ^  181 An example of [hw] is given in Figure 5.49. Fl and F2 are a little lower than the non-labialized [h], as would be expected under the influence of rounding. Again there is no voicing of intervocalic [hw]. At offset, the stressed schwa has an Fl of 620 Hz, F2 1613 Hz and F3 2386 Hz. At the temporal midpoint of [hw] formant values are 931 Hz, 1729 Hz and 2000+ Hz. 6000.^  I^ ' ^ I  I^'^I  -F 4000-  30007^4-  20007  10007^4-  0 ^t.,^I 00 19486  0 0^0.1^0.2^0.3^0.4^0.5^0.6^0.7^0.8 ab181.d  0.0  0. 1  0.9  0.2 ab181.expanded.d  Figure 5.49: ?e hwa? 'cough' (Moses-Columbian: AB) I have not shown that [h] is distinct from laryngeal [h] and uvular [x], the two segments of which one might suggest they are potential variants. Word-initial laryngeal [h]  0.3  182 is shown in Figure 5.50. In this position it contrasts with /h/. For speaker AB, initial laryngeal [h] is markedly and consistently shorter in duration than initial [h]: 143 ms vs. 244 ms respectively, averaged over four tokens for each type. Intervocalically (six tokens) and before a consonant (four tokens), [h] has a duration close to 200 ms. Intervocalic [hw] (two tokens) has an average duration of 202 ms for this speaker. Thus, all /h/ are 200+ ms for AB. For speaker ED the distinction in word-initial duration seems to be less striking: 132 ms vs. 144 ms averaged over four tokens for [#h-] and five tokens of [#h-] Intervocalically for this speaker, duration of [h] ranged from 75 to 180 ms, but was very difficult to measure because of heavy coarticulation with flanking [al-vowels (the form is [saelihaha?u) 'neighing'). From the expanded waveform of Figure 5.50 it can be seen that [h] does not have the amplitude of noisy frication that [h] appears to have. In Arabic, /h/ is commonly referred to as 'soft h', an appellation which seems appropriate here also, since in contrast to [h], laryngeal [h] seems to me to be noticeably muted, at least in these recordings.  183 However, M.D. Kinkade (pc) reports that /h/ and /h/ are often difficult to distinguish in Moses-Columbian and it appears that in some tokens given me by ED, /h/ has lenited to [h]. Unfortunately, the data do not exist to construct diagnostics based on airflow measurements for these two consonants. 6000  5000  :3 -1-' 13  ;  -1-  +  +  +  +  -E  -i-  +  -E  -1-  -E  400  _  +  300  _  -4-  200  .^33  '1^  -E  -1-^•  31  ,14,1:10,1411.  +  100  ,  1292  ''  '^'^'^' I'^'^I  _  3^-  0  +  , 011.:^i .^ev,:!*  k,  -  .  - 1- -  + 3.i.,  ,  k  _  3^ 3  ;33h3;33!3wilyfiktid .^.__,  1 ^0  +  -I-  +  +  .;  +  .  7 i  I  i  i  '  I  '  1  kw  L  1^1  f  '  1 ^1  '  1.4  1I  1111 11111  621  i -621 00  0.1  0.2  0.3^0.4  0.5  0.6  0.7^0. ab179a.d  0.2  ab179.expanded.d  fr  0.3  Figure 5.50: ha? kw61"Are you warm (enough)?' (Moses-Columbian: AB) As for the voiceless uvular [x], this fricative too tends to have formant structure, with bands at 1500+ Hz, 3000+ Hz and 4000 Hz. These are quite different from the formant structure of [h] (as in Figure 5.47) which has formants in the region of 1100 Hz and 1800 Hz. The formant structure of [x] can be seen quite clearly in Figure 5.51.  184 Coarticulatory vowel effects from uvulars are distinct from those conditioned by pharyngeals and glottals as has been seen. 6000  .1.^I,^J 0 0^0.1^0.2^0.3 ab51a.d  i 0 0  0.1  ab51a.d  0.2  Figure 5.51: pax(paxt) 'wise'  185 Finally, to underscore the distinction between uvular, pharyngeal and laryngeal articulation, consider the following four figures. In terms of the spectral attributes of voiceless [h], broad peaks at high frequencies are reported for Arabic /h/ (Alwan 1985). Figure 5.52 shows two spectra of Colloquial Egyptian Arabic [h] (taken from the form shown in Figure 5.24) taken at the beginning and middle of [h]. The window size for all spectra shown here is 50 ms. p fmntilincihomehicolakfbiegyptispeechisw94.d: 759.48n 1445.1Hz^85.6dB^80. 3dB  ,^. " 1170.1 qi i 'W . 31T r 1' '^TO ..rl't II I. 4E I 60 — 1 _ 40 — _ 20  _ —0— 1000  2006^3000^4000^5000  10 fmntilinc/hometnicolatdbiegyptispeethisw94.d: 759.56E  100 —  eo  201. 2Hz^104.1dB^97. 7dB .,^I,^i^i^,^1^1^1^■^1^■  '`)01  .^,`  60 — 40  '6itr'q  '  I  — . -—  20 r-  1  000^Zood^iooci^4006^iooO  Figure 5.52: Colloquial Egyptian Arabic [hi spectra at initiation (top) and midpoint (bottom).  186 From Figure 5.52 it can be seen that Colloquial Egyptian Arabic [h] has slightly broader peaks than the Salish [h] spectra in Figure 5.53, which are taken at the beginning and middle of the first pharyngeal in the form [2ihaha2am] 'neighing' as spoken by AB. 0 ilmnt/lincihomeinicola/db/mosesabispeechiab164.d: 120 1097.6Hz 80 '-,601  '1^1 '  90.4dB^86.3dB^90  'Ill,. ,  III  .^I • i  ! 40 .'. 20  :  O-  6^ioc6^ioo6^3006^4006^5000 0 imntilincihomeinicola/db/mosesab1speechlab164.d: 12IQ 1061.0Hz^106.8dB^105.4dB 100 80  4! 1  60 I^1  , r^1  1I  1  .'  " Ir 1^'1  40 20  ' .1.'16166'''''20I66'''''30l06'.'''40106'''''0'0 '''''''  Figure 5.53: Moses-Columbian [h] spectra at initiation (top) and midpoint(bottom) Further, while the Colloquial Egyptian Arabic [h] shows little decline in energy in the higher energy regions and has a relatively flat spectrum, this is not the case for the MosesColumbian /h/, particularly in the middle portion where there is a noticeable decline in energy in the 2500-4500 Hz region. Compared to the spectrum for Moses-Columbian [h] in Figure 5.54, perhaps the most striking difference is the position of F3 relative to F4. These two formants are much closer together for [h] than [h].  187 [gi finntlinc/homeinicolaidbArnosesabispeechiabl 79.d: 160 1042.7Hz^90.7d8^90.6dB^90  80 60 40 20  -20  2000^3000^4000^5000  Figure 5.54: Moses-Columbian [h] spectra at midpoint Finally, with respect to uvular /x/, F3 and F4 are close together as with [h], but the distance between F2 and F3 is larger than that for both [h] and [h] (Figure 5.55). Fl in the spectra of [x] is in the region of 600 Hz16 and there is a second peak at 1500 Hz. These readings are quite different than those for [hi, viz. 1000+ Hz and 2000 Hz. 0 tmntf1incihomeinicola/dbfrnosesa3ispeechiab51 a.d: 24.E0  237.8Hz^80.6.8^87.5dB^80 1111N1111111111111111.111.11.111111.11N11111111111111111111E  .  1^.  1^.. I  r ilf  I  40 20  -20  0^1000^2000  3000  4000  5000  Figure 5.55: Moses-Columbian [x]: spectra at midpoint. 5.4. Conclusion For the phonological purposes of this dissertation it is important to establish the fact that Interior Salish has segments very much like Semitic  is, h/. Given a clear phonetic  16Note that this is a case where an unusually high harmonic in the region of 200+ Hz has been picked out as as a peak. This is the FO value for this speaker, but in tracking this as a (formant) peak, the LPC algorithm has chosen to treat the peaks in the region of 400-800 Hz as a trough between the higher amplitude peaks at 200 Hz and 1500 Hz.  188 affiliation between Semitic and Salish pharyngeals, their phonological affiliation (or lack thereof) can be explored without the complaint or suspicion that we are dealing with different physiological and acoustic entities. From a phonetic point of view, the investigation I have undertaken replicates analysis techniques that render the results reported easily comparable with the vast majority of investigative phonetic work on Arabic, as well as that reported for Caucasian and Khoisan. There are several directions that further work might concentrate on: first, it is recorded in the literature that fundamental frequency is lowered in the production of Arabic IS/ (Ghazeli 1977) and on /1, is/ (Alwan 1986, et al.). While this study has noted occasional creaky voice and voiceless word-final expiration of /5/, it has not presented structured information on the topic. The situation in Interior Salish is complicated by the phonemic contrast between /1/ and /17, which does not exist in Arabic. One would not expect phonetic laryngeal effects on /1/ to challenge consistently the interpretation of /57, since this would neutralize the contrast. Adequate investigation of the FO characteristics of Interior Salish pharyngeals requires an extensive investigation of the FO characteristics of other fricatives and resonants in the inventory as well as their glottalized counterparts (where they occur), since these are the two classes of segments to which pharyngeals might be argued to belong and from which they must be distinguished. I leave this as a matter for further investigation, and simply note here that the descriptive literature does not report glottal effects on non-glottalized pharyngeals in Interior Salish that parallel the descriptions of glottal effects on Arabic /1/, and my own investigations do not suggest the same extent of laryngeal involvement in Salish pharyngeals as is reported for Arabic. Second, structured information on formant bandwidths and the duration of the various pharyngeals has not been presented. However, Alwan (1986) found that the voicing contrast had more to do with conditioning duration on uvulars and pharyngeals than place of articulation, so one would not expect duration to substitute for place of articulation cues. Third, absolutely  189 no perceptual work has been done17. Alwan (1986) reports perceptual data showing that pharyngeals can be differentiated from uvulars on the basis of Fl bandwidth. It is hoped that these shortcomings will be remedied in due course.  17 A potentially interesting experiment would be to isolate and play Interior Salish pharyngeals to Arabic speakers, and vice versa. I have played Niekepmxcin pharyngeals to a phonetician who is also an Iraqi Arabic speaker. He immediately identified the pharyngeals as pharyngeals, even though NIelkepmxcin is a language in which they are sometimes recorded as somewhat uvular.  190  Chapter Six: Post-velar phonology 6.0 Introduction Chapters 3 and 4 have considered the broad typological and the general phonetic aspects of post-velar articulation. Chapter 5 examined Interior Salish data in some phonetic detail, noting that post-velars (with the exception of /2, h/) form a phonetic class with respect to co-articulatory effects on vowels. This chapter turns to consider the phonological behaviour of post-velars, focussing on data from Interior Salish. The investigation of Interior Salish brings three main results: (i) the phonetic classhood of post-velars as examined in Chapter 4 is phonologized in ways which provide evidence for a fourth articulator node; (ii) there are constraints on post-velar activity analogous to those found in some African ATR systems and some descriptions of Arabic emphasis; (iii) Interior Salish laryngeals are not characterized under the fourth node motivated for uvulars and pharyngeals. In motivating a fourth node Interior Salish parallels Semitic, but the class of Semitic gutturals does not have a correspondent in Interior Salish. While uvulars and pharyngeals are members of natural classes in both language groups, the Semitic grouping of laryngeals with uvulars and pharyngeals is unattested in Interior Salish. This is despite the fact tha