Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

A comparison of glottalized resonants in Sänčatän and St’át’imcets Caldecott, Marion Gerda 1999

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

Item Metadata


831-ubc_1999-0474.pdf [ 6.89MB ]
JSON: 831-1.0089105.json
JSON-LD: 831-1.0089105-ld.json
RDF/XML (Pretty): 831-1.0089105-rdf.xml
RDF/JSON: 831-1.0089105-rdf.json
Turtle: 831-1.0089105-turtle.txt
N-Triples: 831-1.0089105-rdf-ntriples.txt
Original Record: 831-1.0089105-source.json
Full Text

Full Text

A C O M P A R I S O N O F G L O T T A L I Z E D R E S O N A N T S I N S 3 N C A 9 9 N A N D S T ' A T T M C E T S by M A R I O N G E R D A C A L D E C O T T B . A . , The University o f British Columbia, 1997 A T H E S I S S U B M I T T E D I N P A R T I A L F U L F I L M E N T O F T H E R E Q U I R E M E N T S F O R T H E D E G R E E O F M A S T E R O F A R T S in T H E F A C U L T Y O F G R A D U A T E S T U D I E S (Department o f Linguistics) W e accept this thesis as conforming to the required standard T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A August 1999 ©Marion Gerda Caldecott, 1999 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. Department of /^ V-c^o ._»Wc.s The University of British Columbia Vancouver, Canada Date Av<v S O / ? * ? DE-6 (2/88) Abstract This thesis is a comparison of the glottalized resonants in Sancaean and St'at'imcets, two Salish languages. The Licensing by Cue hypothesis as proposed by Steriade (1997) accounts for the distribution of glottalized resonants based on their phonetic cues. The goal of this thesis is to apply the Licensing by Cue hypothesis to the glottalized resonants in Sancaean and St'at'imcets, and evaluate its success in accounting for these two languages. Sancaean is a North Straits, Coast Salish language which does not permit glottalized resonants word-initially. St'at'imcets is an Interior Salish language which allows glottalized resonants word-initially but only in a particular morphological context. Licensing by Cue suggests that glottalized resonants do not occur word initially because of a lack of a supportive context for cues. The distribution of resonants glottalized as a part of a morphological process, namely the actual in Sancaean and the inchoative in St'at'imcets, should also be governed by the same phonetic factors. This thesis first examines the glottal timing of glottalized resonants in both languages. Preliminary phonetic evidence is given for glottalized resonants in Sancaean, which confirm that glottalization is attracted to stress. In contrast, in St'at'imcets, it is perceived that glottalization is repulsed by stress. Modifications are proposed, which enable the hypothesis to account for the timing of glottal events. It is argued, however, that even after such modification, the Licensing by Cue hypothesis is not sufficient to account for the distribution of non-derived glottalized resonants. The same is shown to be true for derived glottalized resonants. The distribution of glottalized resonants is governed by the interaction of three levels of constraints: phonetic constraints, which determine glottal timing, and phonological and morphological constraints, which govern the distribution of glottalised resonants. Also briefly discussed in this thesis are issues related to the relationship between 111 and glottalized resonant, whether [eg] or [creak] should be used to characterise glottalised resonants, and the Proto-Salish morpheme for the imperfective. Based on the research presented in this thesis, it is concluded that Sancaean and St'at'imcets glottalised resonants do not show strong support for a hypothesis which argues for a strong phonetic presence in phonology. While a cue-based approach can account for the phonetic timing of glottal events for glottalized resonants, segment distribution is determined by phonological and morphological constraints. i i T A B L E OF C O N T E N T S ABSTRACT ii LIST OF FIGURES iv ACKNOWLEDGEMENTS v CHAPTER 1-- INTRODUCTION 1 1.1 Phonetic Background 3 1.2 Previous phonological studies 5 CHAPTER 2 T H E LANGUAGES 6 2.1 Ssncaesn 6 2.2 St'at'imcets 7 2.3 Auditory descriptions 10 2.3.1 Sancaean 10 2.3.2 St'at'imcets 10 CHAPTER 3 PHONETIC EVIDENCE OF GLOTTAL TIMING IN SaNCAeaN 12 CHAPTER 4 -- LICENSING BY C U E AND ITS APPLICATION 16 4.1 Licensing by Cue 16 4.2 Applying Licensing by Cue to Sancaesn 19 4.2.1 POSSIBLE SOLUTION: SILVERMAN (1997) / PLAUCHE, DE A Z C O N A , ROENGPITYA, WEIGEL (1999) : 21 4.3 ST'AT'IMCETS 25 4.4 Chapter Summary 32 CHAPTER 5. GLOTTALIZATION AND MORPHOLOGICAL PROCESSES 34 5.1 Licensing by Cue 35 5.2 Phonetic evidence in Ssncaesn 36 5.3 Background 38 5.4 Ssncaean 39 5.4.1 Montler(1986, 1989) 40 5.4.2 Stonham(1994) 41 5.4.3 Glottalization of resonants 42 5.4.4 Featural Affixation 43 5.4.5 Montler revisited 47 5.5 St'at'imcets 4 7 5.6 [Creak] vs. [eg] 49 5.7 Chapter Summary 51 CHAPTER 6 CONCLUSION 53 References: 57 Appendix 59 iii List of Figures Figure 3.1 60 Figure 3.2 61 Figure 3.3 62 Figure 3.4 63 Figure 5.1 64 Figure 5.2 65 Figure 5.3 66 Figure 5.4 67 IV Acknowledgements I am indebted to so many people who have helped me along the way to completing my thesis. First and foremost, I must thank my wonderful consultant, Mrs. Wright, for her patience and willingness to teach me Sancaean. Not only was she the perfect consultant, but she became my pal, and I can honestly say that my life is richer knowing her. Hay' sxwq'a? si?em'! I'd also like to thank Alice Adolph for the sessions we did together. Hopefully we'll get to do more work in the future. I obviously couldn't have written this thesis without the help of many linguistics folks. Many thanks go to my supervisor, Doug Pulleyblank, whom I see as a great role model, and whose comments never turned out to be as bad as they first appeared. I am indebted to Henry Davis, who despite his abhorrence of phonology agreed to be on my committee, and whose grant and support allowed me to do such interesting research. Aside from being a great teacher, I appreciate his unflagging dedication to students. And of course, I could never have managed without the insightful comments and most especially the encouragement of Pat Shaw. She always made me feel like I could do anything, and do it well. I'd like to thank Laura Downing, who got me into the Master's program in the first place. Her belief in me and her friendship and counselling are much appreciated. Without her who knows where I'd be! I'd also like to thank Elizabeth Currie, who was the first to tweak my interest in both linguistics and academia. I'd like to thank Su Urbanczyk for helping me understand, for her support, her editor instincts and for being encouraging. I need to thank Ian Maddieson for spending so much of his time trying to teach me the fundamentals of phonetics and glottalized resonants. Also in the phonetics department, thanks to Guy Carden for his patience and dedication, and resourcefulness when I was disorganised, and to Edward Flemming for his helpful comments. Special thanks also to Tim Montler for his comments and suggestions. I couldn't have written this without the support and friendship of my classmates. Thanks to Tanya, the only other U B C alumnus in our group. Remember Room with a View! To Eunssssk, my walking partner and spiritual guide. To Sun Young, who brought me food. To Ikuyo, who WILL get done! To Yumiko, who liked my songs. To Uri, who I ' l l always remember in my green beret. To Tomio, my 'good' neighbour. To Mimi , who still seems to be missing around the department. To Crazy Lee and Linda, for their support. And to Syntax Matt, who put up with my panic and got me through this sanely. Last but not least, my family. Absolutely none of this would have been possible without the support and love and belief of my family. I need to thank J. DeCuhna and A . Decuhna for making my research in Victoria possible, and for putting me up (and putting up with me). I would like to dedicate my thesis to my Elders: Oma, Opa and Foss, who couldn't be here, but who I know are watching out for me. For Grammie, for everything. And for Mrs. Wright, who I am honoured to call my pal. Chapter 1— Introduction Glottalized resonants are rare segments cross-linguistically, occurring in only 20 of the 317 languages in a balanced study by Maddieson (1984). Because of the limited number of sources that serve as a basis for the study of these segments, they are not well understood. It is also the case that theories of glottalized resonants will be based on a narrow subset of possible data. Thus, it is important to demonstrate that current theories can accurately account for data outside this limited set. This thesis will test one such theory, namely Steriade's (1997) Licensing by Cue, which proposes that the distribution of glottalized resonants is dictated by the timing of their glottal events. The two languages that will serve as data for the application of the hypothesis are Sancaean (Saanich), a North Straits, Coast Salish language and St'at'imcets, an Interior Salish language. Neither of these languages has been studied with respect to glottalized resonants phonetically or phonologically. This thesis will lay the groundwork for further research on glottalized resonants in both languages. It will consist of a phonological study accented by a preliminary phonetic data from Sancaean. As mentioned above, glottalized resonants occur in only twenty of the 317 languages sampled by Maddieson (1984), while ejectives, their obstruent counterparts, occur in 52. Interestingly, of the twenty languages which had glottalized resonants, nineteen also had ejectives as members of their consonant inventories. Thus, two important questions are raised: i) why are glottalized resonants so uncommon? Is there a physiological or perceptual reason for this? and ii) why is their existence in a language is correlated with ejectives? Since glottalized resonants occur in so few languages, they are not well studied and little is known about them. Thus, it is difficult to answer the above questions without first truly understanding the phonetics and phonology of these segments. The small corpus of languages from which to begin research on glottalized resonants is indeed a hindrance. Fortunately, the languages of the Northwest Coast are rich in glottalized resonants, and thus provide an excellent resource. Unfortunately, these languages are becoming extinct at an alarming rate, with many of the remaining native speakers belonging to the older generation. As a result, the already small corpus of languages which serve as a basis for the study of glottalized resonants will soon become smaller. This is detrimental most importantly to the First Nations communities, but also to the field of linguistics. To fully understand the human mind and how language and speech production and perception interact, it is vital that all of the evidence be considered. It is difficult to extrapolate cross-linguistic generalisations, which are said to represent part of U G , i f we do not understand the full extent of variety of segments and patterns. 1 Sancaean is a North Straits, Coast Salish language, spoken north of Victoria on Vancouver Island. Most linguistic research published on Sancaean (Saanich) comes from Montler (1986) Morphology and Phonology of Saanich, and Montler (1991) Saanich Classified Word list. In 1986, Montler put the number of speakers at twenty, but this has certainly decreased some. M y consultant, Mrs. Wright, is an 83-year-old native speaker, who was taught by her grandmother. Our sessions were conducted in her home, in two-hour increments, and recorded using a Marantz tape recorder. Sancaean is taught in the native school, and the Saanich Indian School Board has recently developed a curriculum for the near-by school district. St'at'imcets is a Northern Interior Salish language spoken in and around the Mount Currie area, north of Vancouver. The main grammatical treatment of Lillooet is van Eijk 1997 The Lillooet language. I rely mostly on this grammar as the source of my material, but have also worked on occasion with a consultant. One of the ways in which we may begin to form an understanding of glottalized resonants is by making hypotheses or theories and in this way we may direct and focus research. Examining the data with a particular hypothesis in mind allows us to frame the sort of questions that will lead to productive answers. Thus, this thesis examines glottalized resonants with the framework of Steriade's (1997) Licensing by Cue hypothesis, which seeks to explain the distribution of glottalized resonants as a reflex of their phonetic timing. Any theory that currently exists to explain glottalized resonants will be based on a very narrow subset of cross-linguistic data. Thus, in order to confirm, or modify a hypothesis such as the one above, it must be applied to as many languages as possible. Both Sancaean and St'at'imcets have distributional asymmetries between glottalized resonants and ejectives, but in varying degrees. If adequate as a theory, Licensing by Cue should be able to account for both languages, and explain their variation as a result of reranking of constraints. A theory which seeks to explain the underlying distribution of glottalized resonants should also be able to govern resonants glottalized as a result of morphologically governed processes, and indeed, Licensing by Cue claims to do just this. This thesis also applies Licensing by Cue to a process which involves glottalization in both languages: the actual aspect in Sancaean and the inchoative in St'at'imcets. It will be shown in this thesis that while the concepts that underlie the Licensing by Cue hypothesis are appealing, in its basic form it cannot account for Sancaean or St'at'imcets glottalized resonants. Phonetic timing cannot explain the distribution of glottalized resonants in the way Licensing by Cue predicts. Cue-based constraints determine the timing of glottalized resonants, but phonological and morphological constraints determine the distribution of the segments. Thus, an analysis of glottalized resonants must appeal to the interaction of all three levels of contraints: phonetic, phonological and morphological. 2 Since very few hypotheses based on limited data can account for the whole range of natural language (or many hypotheses at all), part of this thesis will be an attempt at modifying Licensing by Cue in order to account for the data. The structure of the thesis then will be as follows: This chapter will include a background of the phonetics of glottalized resonants, and also a synopsis of what research has been conducted on glottalized resonants in Salish. Chapter 2) will introduce the two languages, the consonant inventories, the distribution asymmetries, and the auditory descriptions of the timing of glottalized resonants. Chapter 3) will give Sancaean phonetic evidence of timing. Chapter 4) will introduce the Licensing by Cue hypothesis and apply it to Sancaean and St'at'imcets distribution asymmetries, with modification. Chapter 5) will look at the morphologically governed processes of glottalization of resonants in the formation of the Sancaean actual and St'at'imcets inchoative and apply the theory to these as well. 1.1 Phonetic Background As previously mentioned, very little research phonetic or phonological has been collected on glottalized resonants. Before examining such research, it is best to understand what glottalized resonants are. Glottalized resonants are complex segments, consisting of oral constriction with accompanying glottalic constriction. Modal voicing, used in the production of plain voiced segments, is achieved through uniform vibration of the vocal folds, with the arytenoid cartilages in neutral position (Ladefoged and Maddieson 1996:50). The glottal constriction accompanying glottalized resonants is often referred to as 'creaky voice'. Creaky voice occurs when "the arytenoid cartilages are much closer together than in modal voice. Creaky voice also involves a great deal of tension in the intrinsic laryngeal musculature, so that the vocal folds no longer vibrate as a whole."(53) This tension can result in complete glottal closure or irregular vibration. The timing of this glottal constriction with respect to the rest of the segment gives us the descriptive terms pre- or post-glottalized. Common acoustic correlates to what is perceived as creakiness or glottalization on resonants are decreased amplitude, irregular pitch pulses, which can surface as irregularities in the wave form, slower pitch pulses, more energy per pitch pulse, more energy in the higher frequencies, or even full glottal closure'. While ejectives are considered to be the obstruent counterparts of glottalized resonants, phonetically, they are very different. Kingston (1985,1990) claims this difference is a result of the timing of glottal events and the different natures of the oral closures of resonants and obstruents. Ejectives, he claims, will be consistently post-glottalized i.e. glottalic articulation will occur at the ' Maddieson p.c, Silverman (1995), Guy Carden (p.c). See also Ladefoged and Maddieson (1996: 317) 3 release of the stop. Plain stops are formed with complete occlusion in the oral cavity, which causes a build up of intraoral pressure. When the occlusion is released, a particularly salient burst containing cues to place occurs. In this sense, the change in intraoral pressure is the articulatory goal of stops. In ejectives, a second closure is added, namely that of the glottis. The larynx raises and compresses the air above the glottal closure, sometimes up to doubling the amount of pressure. (Ladefoged and Maddieson 1996:78). Since any glottalic movement will impact the articulatory goal i.e. increase or decrease the saliency of the release burst (pressure change), glottalic closure must be co-ordinated with the latter half of the ejective, meaning that ejectives are consistently post-glottalized. Continuants (resonants and fricatives) never have a complete build up of intraoral oral pressure, since there is always a flow of air. Thus, the rapid change in intraoral pressure (burst) that is the articulatory goal in stops, is not an articulatory goal for continuants, and any glottal articulations need not be co-ordinated. As a result, glottal closure accompanying these segments will be have more variable timing, and will modify voice quality and the fundamental frequency of the sonorant and neighbouring vowels instead. (Kingston 1985:247). Silverman (1997) explains why languages have a tendency to have pre-glottalized resonants. He claims that 'heavy glottal constriction may result in sufficient aperiodicity, or jitter, to disrupt transmission of a salient nasal formant structure'. (97) As a result 'contrastive laryngeal gestures are optimally phased such that the laryngeal gesture is truncated with respect to the supralaryngeal gesture, and sequenced with respect to voicing: non-modal phonation is followed by modal phonation. In this fashion, cues are optimally transmitted to the listener." (83) . In other words, the transition formants that are cues to place and manner for sonorants could be obscured by clues to glottalization i f they were to overlap. As for studies focusing specifically on phonetic studies of glottalized resonants in Salish languages, there is only one that the author is aware of to date: Flemming, Ladefoged, and Thomason (1994). Phonetic Structures of Montana Salish. They found that resonants are consistently pre-glottalized, supporting Silverman's theory. "In almost all positions, glottalized sonorants are typically realised as glottal constriction followed by a sonorant. i.e. they are pre-glottalized." (16). Glottal constriction was often realised as complete closure, and other times, creaky voice. They also found that the modally voiced portion of glottalized nasals was substantially shorter then in plain nasals. Since Sancaean and St'at'imcets and Montana Salish are members of the same language family, any phonetic evidence that either is the same as these findings or different will have a bearing on our understanding of the amount of articulatory variation that exists within the same language family. 2 Since Silverman offers no specific data in support of this claim, I must assume it is theoretically and not empirically motivated. Sancaean and St'at'imcets seem to be counterexamples to this generalisation (see Chapter 3). 4 1.2 Previous phonological studies There is little previous phonological research done on glottalized resonants in Salish which is related to the present research. Most grammars of Salish languages have some treatment of glottalized resonants in them. The reader is invited to examine specific research on glottalized resonants in two Salish languages, namely Blake (1992, 1995) on Sliammon and Taylor (1994) on Shuswap. These papers will not be referred to in this thesis, but provide comparative sources. The next chapter will introduce the two languages examined in this thesis Sancaean and St'at'imcets, give their consonant inventories, distribution asymmetries and timing descriptions. 5 Chapter 2 The languages Both Sancaean and St'at'imcets oppose plain vs. glottalized segments in their resonants and obstruents (glottalized resonants and ejectives). In both languages, glottalized resonants are more restricted in their distribution than ejectives. In Sancaean glottalized resonants occurs only post-vocalically, i.e. not word initially or post-consonantally. In St'at'imcets, glottalized resonants occur post-vocalically, post-consonantally, and do occur word initially, but only as a result of some reduplicative processes. Ejectives have distribution parallel to non-laryngealized obstruents i.e. they occur in the same positions as plain obstruents. Two questions are brought to mind: i) how can one explain the lack of glottalized resonants in word initial position and ii) can both languages be accounted for under the same theory (i.e. can the theory account for why only post-vocalic glottalized resonants occur in Sancaean but St'at'imcets allows post-consonantal as well). In addition to this, we have the question of glottal timing. Recall that Silverman predicted that pre-glottalized resonants are the unmarked form. Wil l this hold true of our data as well? In this chapter the two languages will be introduced in the form of their consonant inventories, distribution asymmetries and glottal timing descriptions. 2.1 Sancaean Sancaean is a North Straits, Coast Salish language, spoken on the East Coast of Vancouver Island, near Victoria. It is more commonly referred to as Saanich in the literature, but my consultant strongly prefers the name Sancaean. I will refer to the language using the phonetic transcription of its name. The two major treatments of Sancaean are Montler (1986) An Outline of the Morphology and Phonology of Saanich, North Straits Salish and Montler (199U Saanich, North Straits Classified Word List. Below is the consonant inventory: (1) Sancaean Consonant Inventory Lab Dent Alv Lat Alveo-palat Lab-vel Uvular Lab-uvular Laryngeal p t c (k) kw q w q p ' t 6 ' t' t l ' c k w , q ' q W ' e s i S w X X x w m n 1 y w g m' n' 1' y' w' 3 Throughout the thesis I will use 'R to refer to pre-glottalized resonants and R' for post-glottalized resonants. 6 As can be seen above, both obstruents and resonants oppose plain vs. glottalized segments (ejectives and glottalized resonants)4. Ejectives contrast with plain stops in all positions, while glottalized resonants are restricted to post-vocalic positions, as shown in the following chart: (2) Glottalised resonant distribution ejectives glottalized resonants # t'63rj'3i 'chest' unattested C V sp'atl'arj 'cigarette' unattested V V t'st 'il 'am' someone singing sk'way'ech3n 'grizzly bear' kw'el'3s 'it's hot' C tl'cas 'island' ?an'xstsxw 'you are groaning' # t'am'iq' w kw9n'3n' 'dolphin' ? is considered to be a resonant, but stands apart from other glottalized resonants in that it may occur word initially: /?ay'ankw9s/ 'you are afraid'. This is not an uncommon occurrence cross-linguistically, but raises the question of the relationship between ? and other glottalized resonants. This issue will be considered in Chapter 5. The next section looks at St'at'imcets. 2.2 St'at'imcets St'at'imcets is an Interior Salish language spoken in the Mount Currie area north of Vancouver. It is also known as Lillooet. The major source of Lillooet is van Eijk (1997) The Lillooet Language. van Eijk notes that few people younger than 50 are fluent speakers, but that a renewed interest in the language will change this. Below is the consonant inventory of St'at'imcets. 4 Also note the presence of rj, a relatively unusual segment in Salish languages. 7 (3) Phoneme inventory Lab. Dent. Lat. Dent Palatal Velar Uvular Laryngeal p t c,c k k w q q w p' tr c' k' k ' w q' q w ' i s,s x x w X X w m n m' n' M r , i ; z y Y ? S w h w z' y' Y ' I ' c-w 2 w' van Eijk considers the dental, velar and uvular glides (z, z', Y,Y , , <T, <I w ') resonants because a) they oppose plain vs. glottalized members and b) like other resonants, they do not occur in the position C C and C_#.5 However, their distribution is more limited than other R' as they cannot occur i n suffixes. He considers ? and h resonants because they do not occur in C C, C # environments either. However, he does note that they are distinguished from other resonants by the fact that both are unvoiced and that ? has a freer distribution, frequently occurring word initially. The most obvious difference between Sancaean and St'at'imcets is the absence of rj in St'at'imcets, and the presence of the glide series. The St'at'imcets distribution is also different from Sancaean. As noted above, R ' does occur word-initially, but only in the cases of reduplication. Glottalized resonants occur post-vocalically, post-consonantally but never inter-consonantally. Ejectives contrast with plain stops in all positions. 5 Conceivably, other phonetic factors, such as common cues, could account for the similar distribution of these segments. This however, will not affect the account presented in this thesis in a meaningful way. 8 (4) eiectives p'ustan' 'rye grass' El only in some reduplicative forms: w'aw'plilc'a? 'caterpillar' m'am's-mas 'calf 'na'natx w -am c__v p'alk 'aq w 'to turn around' k l ' ax w 'muskrat' q ' w a X - q ' w i X 'black' s-qwl 'ip 'black tree moss' n-q'am'q'm'ap'10 people' n-salTac 'to drool, slobber' v_v s-mik'il 'fish oi l ' ti-sqlaw'a 'the beaver, the money' al'ap, -kal'ap 2p obj 'you folks' qwal'utsut 'to talk nonsense' 'niam'an 'to put s.t. into s.t.' _c n-c'qaXa? 'my horse' sk wistq w?am 'waterfall' tew'p to foam lil'wat-emx-ec 'to speak Mount Currie' n- aq'- q'-am-kst'6 people' -q-al'qw penis c_c generally disallowed, but possible; n-c'qaXa? 'my horse' s i q't-amx 'person from a place close to Lillooet unattested c_# ? unattested _# c'lip' 'to pinch' qiqal' 'weak' k w ' u l ' 'to make' p s i l ' daylight pUpan' to find something unexpectedly To summarise, these two languages although they are part of the same language family have different consonant inventories and different distribution asymmetries. Sancaean glottalized resonants 9 occur only post-vocalically, while this generalisation does not seem to apply to St'at'imcets. Word initial position in both languages resists glottalized resonants, but where Sancaean has an absolute ban, St'at'imcets permits them in some morphological contexts. One question that must be asked at this point is why is there a distribution asymmetry between glottalized resonants and ejectives? Is there something inherent in their phonetic nature that causes this? According to Steriade (1997) Licensing by Cue hypothesis, this is indeed the case. The timing of glottal events with respect to oral events dictates where laryngeal segments will be neutralised. The first step in applying this theory will be to examine the glottal timing. 2.3 Auditory descriptions 2.3.1 Sancaean Auditory description of Sancaean glottalized resonants is given in Montler (1986: 13). "The glottalized resonants are usually realised phonetically as voiced resonants with accompanying laryngeal constriction, creaky voice". He claims with respect to glottal timing, that glottalization is attracted to the stressed vowel (i.e. post-glottalized before stress, pre-glottalized following stress). This author supports this description with the modification that glottalized resonants in coda position, followed by a C (i.e., not word finally) are post-glottalized.6 To facilitate between the representation of pre- and post-glottalized resonants, pre-glottalized will be written 'R and post-glottalized R' . Examples in the text will be in broad transcription (written post-glottalization for reasons of convention). Thus, a more accurate representation of glottalized resonants in San could be given as follows: (5) word initial post-C pre-stress intervocalic post-stress intervocalic post-stress pre-C word-final N / A N / A V R ' V V ' R V V R ' C V'R# To summarise, resonants are post-glottalized immediately preceding the stressed vowel (R'V) and immediately following the stressed vowel in coda position when followed by a C (VR'C). They are pre-glottalized following the stressed vowel (V 'R$) and in word-final position. (V'R#). 2.3.2 St'at'imcets van Eijk's description of the timing of glottalized resonants in Lillooet: ".. .glottal stricture is strongest near the onset of the resonant before a stressed vowel, but near the outset in other positions."(van Eijk: 11). (my emphasis) 6 Montler (p.c.) confirms this. 10 This description poses a serious problem for Steriade's analysis. Before we consider Steriade's hypothesis, along with others, let us define exactly, our assumptions given van Eijk's timing. His description is in two parts, the fist part dealing with pre-stress resonants ".. .glottal stricture is strongest near the onset of the resonant before a stressed vowel. This is ambiguous between only the resonant immediately preceding the stressed vowel, and all resonants preceding the stressed vowel. Without explicit data, this is difficult to resolve, but let us assume for the moment that it applies only to resonants immediately preceding the stressed vowel: #'RV, 'ma'ms-mas C ' R V k ' lax w ' , V ' R V 'na'natx w-am. The second part is as follows: ".. .but near the outset in other positions" Again, without knowing explicitly what those other positions are, it is difficult to make any predictions. However, let us assume that the other positions are as follows: V R ' V , ti-sqlaw'a V R ' C qam'ts VR'# psil ' daylight (6) word initial * post-C pre-stress intervocalic f post-stress intervocalic post-stress pre-C word-final #'RV C ' R V V ' R V V R ' V V R ' C VR'# As we can see from the timing descriptions given above, stress seems to have exactly the opposite effect on glottal stricture in Sancaean than it does in St'at'imcets. Could this difference in timing explain the difference in their distributions? The first step this thesis takes to solve this problem is to examine the phonetic evidence of timing relations in Sancaean. This is a very preliminary look, and a more in-depth study, along with a study of St'at'imcets is unfortunately outside the scope of this thesis. 11 Chapter 3 Phonetic evidence of glottal timing in Sancaean The first issue that must be addressed by acoustic analysis is i f the perceptions of glottal timing are indeed accurate. While the St'at'imcets data is unavailable, the Sancaean glottal timing is examined. Glottal timing is perceived to be variable. This predicts that acoustic analysis should show glottal stricture in two positions relative to the resonant: towards the beginning of resonants in post-stress and word-final position, and at the end of resonants in pre-stress and pre-consonantal resonants in coda position. The auditory descriptions also predict that glottalized resonants in coda positions will have two distinct timing relations, dependent on whether they are followed by a consonant or word final. This is unusual given that one might expect codas to function uniformly. In order to confirm both predictions, phonetic evidence must be examined. Data was collected from one 83-year-old female native speaker over a period of nine months in 2-hour sessions in the consultant's home. Approximately 208 words/sentences were collected and recorded on a Marantz PMD430 tape recorder and analysed using Signalize 3.127. Before discussing the findings, two problems arose during the analysis of the collected data. The first is that the consultant has naturally creaky voice due to her age. As discussed above, "phonological" or phonemic creakiness (where creakiness is a feature, which contrasts with plain segments) is caused by tightening and compressing the vocal folds in such a way so that they do not vibrate uniformly. As we age, our membranes become thicker, and our vocal folds less flexible, often causing them to vibrate irregularly, resulting in 'accidental' or phonetic creakiness, playing no role in the phonology. Thus, it is sometimes difficult to distinguish between the phonetic creakiness in my consultant's speech and the feature [creak] associated with particular segments or processes. The second difficulty is that several cues are used in the production of phonological creaky or glottalized resonants, but not consistently, and often in conjunction with other cues. It is difficult to define a distinct acoustic correlate for glottalized resonants in Sancaean, since there is as yet no generalisation governing which cues are used by which resonants, in which contexts or even i f cues are used consistently across tokens. A much more in-depth phonetic study, outside the realm of this paper, is required. However, it will be shown that it is possible to use aperiodicity, slower pitch pulses, more energy per pitch pulse, more energy in the higher frequencies and full glottal closure as cues to distinguish plain vs. phonologically glottalized resonants. 7 Three microphones, Audio-technica AT831b unidirectional electret lavalier, Radioshack 33-3003 omni-directional electret lavalier and a stand mounted dynamic mic, were used to record with. Speech was sampled at 16 bit at 22050 Hz, digitized using the Mac internal audio on a G3-233 MHz desk top, and low-pass filtered using Mac internal filters. 12 Given that the timing description of glottalized resonants varies dependent on stress, this paper examines four contexts in which glottalized resonants occur: intervocalic immediately preceding the stressed vowel, intervocalic immediately following the stressed vowel, pre-consonantal immediately following the stressed vowel, and word finally following an unstressed vowel. Word final glottalized resonants in monosyllabic roots are not considered due to a lack of appropriate data. Contexts that are not relevant (i.e. not involving a stressed vowel) are also not considered. Only a very small amount of data was analysed for this thesis. Approximately 5 pairs of either two or three tokens each were considered for each context, unless otherwise specified. While there was some variation between tokens, the timings given represent strong tendencies found throughout. Obviously, this a very preliminary study, and more research in the form of both more tokens and more speakers needs to be pursued. In order to confirm the timing relations perceived by both Montler and this author, consider Figure 3.1, a comparison of the words /k'weTas/ 'warm' and /can q w alax w / 'the dogsalmon month'. The waveform of IM in a) shows a marked decrease in amplitude compared to the one in b), suggestive of glottalization. Very narrow band spectral cross-section of glottalized and plain IM are given in e) and f) respectively. The large energy peak at 1300 Hz present in e) but not f) is also indicative of glottalization. However, neither determines whether the resonant is pre- or post- glottalized. To establish this, consider the wide band spectrograms in c) and d). In d), the plain IM shows regular pitch pulses with a continuous first formant. The glottalized IM on the other hand, shows an interrupted FI with pitch pulses that are spread very far apart in the transition from vowel to resonant, i.e. the beginning of the resonant. The vertical striations associated with these pitch pulses are another indication of glottalization. It is clear from this example that glottalization occurs at the beginning of the resonant, even at the end of the preceding vowel. Note the space between glottal pulses, the vertical striations and the greater amount of energy in the higher frequencies during the latter half of the preceding vowel, into the transition. As Figure 3.1 represents, post-stress glottalized resonants are indeed pre-glottalized. In fact, glottalization is often characterised on the vowel. The cases in Figure 1 are representative of approximately twelve pairs. Since Salish languages have such a large consonant inventory, exact minimal pairs are difficult to get. Figure 1 represented the clearest example of the set. Turning to Figure 3.2, consider post-stress, pre-consonantal coda resonants. Recall that perceived timings predicted that these resonants should be post-glottalized. a) and b) are expanded versions of the waveforms of c) and d), the words /?an'xstsxw/ 'you are groaning' and /santi/ 'from Sunday to Sunday'. In a) the irregularity and marked decrease in amplitude of the waveform of the resonant again indicates glottalization. Notice the regular shape of the waveform throughout the plain resonant in b). 13 Considering the broad band spectrograms in e) and f), the irregularities in the waveform are apparent as irregular pitch pulses at the end of the glottalized resonant in e). In comparison, the plain resonant in f) shows a smooth gradual transition into the following consonant. These diagrams confirm the perceived timing relations: post-stress, pre-consonantal resonants are indeed post-glottalized. But what about word final resonants? Wil l they be pre-glottalized as predicted? Consider Figure 3.3, the comparison of the words /kwan'an7 'dolphin' and /kwasan/ 'star' given in a) and b). As seen in the broad band spectrograms in c) and d), the glottal pulses of the word final /n'/ slow down considerably, appearing quite spread apart. It is also the case that the speaker has an audible release following most word-final glottalized resonants, whereas this never occurs following plain resonants word finally. This is suggestive to the author of post-glottalization, but due to the lack of appropriate data, it is difficult to make judgements at this time. It appears that reference to pre- or post- glottalized may not be relevant in this case, since the resonant may be both, or better said, glottalized throughout. The implications of this may mean that discussion of cues to glottalization at vowel/resonant transitions may not be relevant. In e) and f), spectral cross-section shows more energy in the higher frequencies in the glottalized word-final resonant (e)) than the non-glottal one in f). However, these implications cannot be meaningfully discussed without better understanding of what is actually occurring. For now, it will be assumed that word-final glottalized resonants are glottalized throughout, and cannot be said to be either pre- or post-glottalized. In Figure 3.4 c) and d) are the waveforms for two words containing pre-stress glottalized resonants: /skway'acan/ 'grizzly bear' and /scuw'aean/ 'Tsawassen'. Glottalized resonants are not 8 common in this position for the following reason. Based on Montler's description of stress , as well as the author's own experience, I will suggest that Sancaean has a mainly trochaic stress system (i.e. stress usually occurs on the first vowel of the word). Since word initial glottalized resonants are not licensed in this language, pre-stress R' are difficult to find. However, here are the only two examples in the collected data, with the expanded waveforms given in a) and b). a) shows that the vowel preceding the resonant is very regular. In b) we can see that the same is true. In both cases, the vowel following the resonant changes dramatically, showing more energy, indicative of heavy glottalization, and suggestive of a post-glottalized resonant. Considering e) and f), the broad band spectrograms clearly show that the glottalization takes place at the end of the resonant. In e) the transition from a to a high front position has already occurred before the glottal stricture takes place. Until this point, there has been a continuous F l , but there is a break in it at this point, marking the glottal stricture. The following vowel shows the now familiar vertical striations and greater energy in the higher frequencies, f) clearly shows the irregular pitch 8 ".. .the first li, lei, or /a/ (i.e., non-schwa) takes the main stress, and if there are only schwas, then the penultimate takes the main stress." (1991:viii) 14 pulses occurring after the steady state vowel/resonant. Unfortunately, there are no corresponding plain examples for either, but the spectrograms show correlations to glottalization present in the diagrams already seen, supporting the perceived timing of post-glottalization on resonants preceding stress. Figures 3.1-3.4 confirm the auditory predictions of both Montler and this author: pre-stress and pre-consonantal resonants are post-glottalized R' : (_V, _C), and post-stress resonants are pre-glottalized ('R: V_). Word-final glottalized resonants seem to be glottalized throughout. From this evidence, stress seems to be a major conditioning factor on the timing of glottal events in Sancaean. This is not unreasonable, given that stressed vowels are often more prominent, and thus amplify any acoustic cues. Silverman (1997) confirms this, ".. .stress plays the functional role of increasing acoustic energy through increased aerodynamic force, as well as overall lengthening and sometimes hyperarticulation.. ."(97). Recall that stress was also a conditioning factor in the timing description of glottal events in St'at'imcets. In that case, though, it seemed that stress repulses glottalization, the opposite of Sancaean. Resonants are pre-glottalized preceding stress and post-glottalized following the stressed vowel, van Eijk's description of the timing of glottal events has not been confirmed, but based on the his experience with the language, and the author's perception of a difference between St'at'imcets glottalized resonants and those in Sancaean, van Eijk's timing descriptions will be used in this thesis. Now that the facts have been presented, it is time to introduce the theory. The next chapter than will introduce the Licensing by Cue hypothesis and apply it to the timing distribution and the facts given above. 15 Chapter 4 ~ Licensing by Cue and its application I n th is chapter the L i c e n s i n g b y C u e hypo thes i s w i l l be i n t r o d u c e d and then a p p l i e d to the g lo t ta l t i m i n g and d i s t r i bu t i on o f g l o t t a l i z e d resonants i n b o t h languages . It w i l l be s h o w n that the v e r s i o n o f the hypo thes i s p r o p o s e d b y Ster iade (1997 ) cannot a ccoun t f o r the d i s t r i b u t i o n fo r g l o t t a l i z e d resonants i n e i ther S a n c a e a n or S t ' a t ' i m c e t s . T h i s chapter w i l l s h o w that the theore t i ca l f r a m e w o r k u n d e r l y i n g the L i c e n s i n g b y C u e hypo thes i s is v a l i d , but i n r e l a t i on to these t w o l anguages , m o d i f i c a t i o n s are necessary . T h e g o a l o f L i c e n s i n g b y C u e is to e x p l a i n the d i s t r i b u t i o n o f segments ba sed o n phone t i c factors . H o w e v e r , it appears i n S a n c a e a n a n d S t ' a t ' imce t s that L i c e n s i n g b y C u e c a n accoun t o n l y f o r the d i s t r i bu t i on o f the t i m i n g o f g lo t t a l events , rather than the d i s t r i bu t i on o f w h o l e segments . It w i l l a l so be s h o w n that stress is a m a j o r f ac to r i n d e t e r m i n i n g the g lo t ta l t i m i n g o f resonants i n b o t h l anguages , but to oppos i t e results . Stress is an enhance r o f g lo t t a l cues i n S a n c a e a n , but an antagon is t f o r g lo t t a l cues i n S t ' a t ' i m c e t s . 4.1 Licensing by Cue Steriade's (1997) Licensing by Cue hypothesis seeks to explain the distribution of glottalized segments as a result of 'phonetic implementation factors'. She argues that perceptual cues, and their salience in particular contexts are what license features in certain positions. Absence of a supportive context for cues results in neutralisation. Thus, she proposed Licensing by Cue, as opposed to Licensing by Prosody, in which laryngeal features are licensed due to their position in the syllable (i.e. onset vs. coda). "The general idea pursued here is that phonological grammars incorporate knowledge of the conditions under which feature contrasts are physically implemented" (1997:1). Thus, phonetics, in the guise of 'implementational constraints' play a vital role in the grammar. Under the Licensing by Cue hypothesis featural contrasts will preferentially surface in environments supportive of the 'cues' to that particular feature. Acoustic cues in the speech signal to what we perceive as voicing, for example, might be voice onset time, burst and closure duration. Steriade distinguishes between two kinds of cues, internal (cues present during the oral closure) and transitional (onset or offset). Voiced stops then would contrast with voiceless stops only in contexts where these cues are salient or perceptible enough, as judged on a relative scale of the feature [voice]. What governs which contexts are acceptable are phonetic implementation factors. "Phonetic implementation factors involve gestural timing, gestural magnitude and contrast perceptibility, i.e. the nature and relative duration of cues available in a given context for the identification of a specific contrast" (1997:61). According to Steriade, speakers are aware that certain contrasts are more salient in certain contexts (i.e. k/g contrast is easier intervocalically than between obstruents) and this awareness of phonetic implementation factors is active in the phonology as 'implementational 16 constraints'. These constraints represent the knowledge of physical implementation of contrast. Steriade frames her hypothesis in Optimality Theory (OT) (Prince and Smolensky 1993). In OT it is the interaction of universal and violable constraints on phonological well formedness that select the surface forms we see. The more highly ranked the constraint, the more important it is in the language. Implementational constraints, then, interact with more familiar constraints on identity (faithfulness) and the different ranking of these constraints with respect to each other should give us cross-linguistic variation. The Licensing by Cue hypothesis is used by Steriade to explain laryngeal neutralization and the distribution asymmetries between ejectives and glottalized resonants. Recall that glottal timing between ejectives and R' varied, with ejectives having the glottal closure aligned with the release burst. This means that the cues to glottalization for obstruents will be at the right edge. The opposite is true for resonants. Glottal constriction, while thought to be more variable, was found to be timed to the onset of oral stricture, in other words, glottalized resonants will be pre-glottalized (Silverman 1997). Since the differences in distribution between ejectives and R' may depend on auditory or articulatory properties or this difference in glottal timing, Steriade suggests that the relevant features that should be used in this analysis are two auditory features [ejective release] and [creak].9 Steriade's evidence that two different features are needed comes from Yokuts, where certain morphologically governed processes glottalize sonorants, but not obstruents. The main implementational constraint used is Context Cues. While not explicitly stated, this could be generally characterised as: Context Cues [F]: *[F] / in positions lacking contextual cues where [F] is the feature in question. To analyse glottalized resonants in particular, Steriade posits Context Cues [creak] (1) Context Cues [creak] * [creak] in positions where context cues to [creak] are absent This constraint represents a perceptibility scale for the cues to [creak]. Vowel-resonant transitions represent the optimal environment for the support of glottal cues, which is why glottalized resonants occur post-vocalically. For pre-glottalized resonants (with cues at the onset of the segment), the optimal environment is after a sonorant ([+son]_J and non-optimal elsewhere (#_,[-son]_J. Translated into constraints, *[creak]/[-son]_,*[creak]/#_ will outrank *[creak]/[+son]_, Thus, Context Cues is a way of ensuring that pre-glottalized resonants will be post-vocalic. 9 The articulatory feature [constricted glottis], normally applied to both, (cf. Howe and Pulleyblank 1999) does not play an active role in this analysis. The implication here is a distinction between production and perception. Use of [eg] supposes that it is articulation or production features that are relevant in an analysis. Steriade, while still suggesting that glottalized segments have the feature [eg], theorises that it is the perceptual or auditory features that play the active role here. Please see chapter 5 for a more complete discussion. 17 Constraints formalising the timing relations described previously are abbreviated SonTiming (Obst Timing for ejectives) (2) Son Timing: The onset of glottal constriction must precede the onset of oral closure1 0 The peak of glottal constriction must precede the oral release This constraint rules out variable timing, ensuring that glottalized resonants will be consistently pre-glottalized. A third constraint, Preserve [creak] penalises loss of the feature [creak]. One of the case studies for Licensing by Cue is Yokuts, where glottalized resonants are assumed to be pre-glottalized. They neutralised anywhere but post-vocalically, so we have the following ranking: (3) SonTiming, Context Cues »Prese rve [creak] SonTiming is an unalterable characteristic of the language and therefore must be very highly ranked. Context Cues [creak] must outrank Preserve [creak] in order to insure that an initial or post-consonantal R' will never surface. Either cues are present for that particular timing in that context, or neutralisation occurs. An example of how the constraints work can be seen in example (40), rewritten here as (4) (4) /lihm/ creak SonTiming Context Cues [creak] » Preserve [creak] lihm- * 'lihm- (pre-glott) *! l'ihm- (post-glott) *! This ranking ensures that i) only post-vocalic glottalized resonants will surface and ii) timing is invariable. Neutralization is preferable to a shift in the timing relations of the glottal events. If we consider briefly again ejectives, we can see the different distribution of ejectives and glottalized resonants as a result of different ranking of constraints. For ejectives, Preserve [ejective] and Obst timing, outrank Context Cues [ejective] so that ejectives will surface regardless of their distribution. Since the focus of this paper is on glottalized resonants, no more will be said about ejectives. In the next section Licensing by Cue will be applied to the Sancaean glottalized resonants we saw above. 1 0 Perhaps constriction would be a better term. 18 4.2 Applying Licensing by Cue to Sancaean The application of the version of Licensing by Cue proposed by Steriade (1997) is not very successful at accounting for the distribution of Sancaean glottalized resonants. Recall that glottalized resonants in Sancaean are also only found post-vocalically. In order for Sancaean to fit Licensing by Cue, all glottalized resonants would have to be pre-glottalized. As seen in chapter 3, this is not the case. To summarise the matches and mismatches between the predictions of Licensing by Cue and Sancaean consider the table below: (5) Matches: -Post-stress -Word finally Mismatches -pre-stress -pre-C LbC V ' R V V'R# V ' R V 'RC Sancaean V ' R V n/a V R ' V V R ' C k w 'e ' las k w an'a 'n skw 'ay'echa ?an'xstsxw As shown in the above table, the version of Licensing by Cue proposed by Steriade (1997) predicts only fifty percent of actual glottal timings that surface in Sancaean. The Sancaean data poses a number of challenges for this version of the Licensing by Cue hypothesis. First, Sancaean seems to have variable timing. It is clear from both perceived and acoustic evidence that stress governs the variation in timing. Without reference to stress, it is impossible to predict when a resonant will surface as pre- or post- glottalized. While not discussed in Steriade (1997), it is possible under the proposed hypothesis to achieve variable timing by subordinating SonTiming to another constraint which will dictate when each timing variation will occur, or by altering Context Cues to be sensitive to stress, since it is stress that seems to dictate the timing of glottal events. In an account of Shuswap Steriade does latter. In Shuswap (Interior Salish), a glottalization feature of suffixes is always attracted to the resonant following the stressed vowel. Steriade explains this as "Stress is an enhancer of glottalization cues. I attribute this to the fact that the stressed vowel is longer, louder and thus better able to carry the contextual cues for creak" (86). She formalises this by introducing a new level to the perceptual scale of [creak]: a stressed vowel ( V ) is a more optimal environment than an unstressed vowel (V), and by adding a new constraint to Context Cues: (6) *creak/V[-stress] »*creak/V[+stress], (renamed here Context Cues2). Thus, after adding the stress constraints to Context Cues2, glottalization not adjacent to the stressed vowel will violate Context cues. This thesis will adopt the constraints proposed by Steriade (1997) 1 9 except SonTiming. This constraint is designed in the original analysis to assure that all resonants will be pre-glottalized. It is a markedness constraint which makes the following predictions: i) languages with only pre-glottalized resonants exist ii) no languages will have just post-glottalized resonants iii) Languages with post-glottalized resonants must also have pre-glottalized resonants. It is not clear that these generalisations hold for Sancaean and St'at'imcets. In fact, timing relations of glottal constriction with regard to oral closure is predicted to be variable by Kingston (1985,1990). This raises questions about the validity of such a constraint. In light of these questions, SonTiming will not be used in this analysis. In any case, i f this constraint is valid, it is lowly ranked in these languages (and possibly others), and thus is not considered in the tableaux to follow. By introducing stress cues into this analysis, I am implicitly assuming comparable constraints relating to stress: Context Cues stress and Preserve Stress cues. This is relevant both here and in the analysis of St'at'imcets glottal timing. However, I make no claims about the phonetic nature of stress, and thus will to include these constraints in the tableaux, other than to assume that they are highly ranked. These tableaux represent the timing relations and environments in Sancaean. Context Cues 2 (including the new optimal context V ) outranks Preserve Creak penalising loss of [creak]. Tableau 1 V R V creak Context Cues 2 » Preserve Creak 1. V ' R V (pre-glott) 2. V R ' V *! 3. V R V *! The first candidate is the winning candidate, because glottalization immediately adjacent to the stressed vowel is the most context for cues to creak. A post-glottalized resonant (candidate 2) violates Context Cues, and the third candidate which is neutralised, violates Preserve Creak. A similar tableau, but with the resonant in pre-stress position can be seen below: 20 Tableau 2 / V R V / creak Context C u e s z » Preserve Creak 1. VR'V (post-glott) 2. V R V (pre-glott) *! 3. V R V *! If the glottalized resonant precedes the stressed vowel, the optimal context for cues to glottalization will be adjacent to that stressed vowel i.e. post-glottalized. Candidate 1 satisfies all constraints because it is post-glottalized. Candidate 2. violates Context Cues because [creak] is not adjacent to the stressed vowel. The third candidate violates Preserve Creak. The next tableau shows a problem for this analysis. Tableau 3 / V R C / Context Cues l » Preserve Creak creak 1. VR 'C (post-glott) *! ^ 2 . V ' R C (pre-glott) 3. V R C *! Candidate 1, is the intended winner, but Candidate 2 is designated as the actual winner. Candidate 1, which is post-glottalized, has the cues for creak in a less optimal environment (not adjacent to the stressed vowel) than it should, and thus it violates Context Cues. Candidate 2, with its cues adjacent to the stressed vowel, incurs no violations. As we can see, even with the addition of stress to Context Cues, one case, that of post-glottalized pre-consonantal resonants cannot be accounted for. Clearly, these two constraints alone are incapable of accounting for everything. Another explanation must exist. 4.2.1 Possible Solution: Silverman (1997) / Plauche, de Azcona, Roengpitya, Weigel (1999): Recall that Silverman (1997) claimed there was a tendency for resonants to be pre-glottalized, also based on cues. He hypothesis that sonorants will be preglottalized, to keep from obscuring the place/manner cues in the consonant-vowel transition. Under this hypothesis it seems that there is competition among cues. 21 Plauche et al (1999) notes that Silverman fails to consider resonants in coda position. The authors extend Silverman's hypothesis to glottalized resonants in codas, and by the same logic, reason that glottalized resonants in coda position will be post glottalized. If the vowel-resonant transition is the primary locus for cues to place and manner, and if cues to glottalization risk rendering these 'unrecoverable', then coda glottalized resonants should be post-glottalized. The authors come to the conclusion that "glottalized resonant that rely mainly on creaky voice, full glottal stop and amplitude as phonetic cues to glottalization.. .will surface as pre-glottalized in onset and post-glottalized in coda" (381). Silverman/Plauche et al.'s hypothesis, while also based on cues, comes to slightly different conclusions from the version of Licensing by Cue proposed by Steriade (1997), and makes different predictions. If cues to glottalization work antagonistically against cues to place/manner, resonants should be pre-glottalized in onset position", and post-glottalized in coda-position. Below is a chart comparing the predictions made by Silverman/Plauche et al. and my acoustic findings: (7) matches: S/P Sancaean -pre-C V R ' C V R ' C -post-stress V ' R V V ' R V mismatches ( , -pre-stress V ' R V V R ' V -word final VR'# n/a This version of a cue-based theory also only accounts for fifty percent of the contexts. It is insufficient because it makes generalisations based on onsets and codas without making a distinction based on stress, and also fails to take into account whether the resonant is followed by a consonant or not. However, the basic theoretical framework is valid given the appropriate set of constraints. Assuming that glottal cues obscure place/manner cues, and also the idea that stress is what motivates the timing of glottal events in Sancaean, it is possible to modify Licensing by Cue constraints somewhat and propose a comprehensive account. Let's hypothesize for the moment that there is competition between place/manner and glottalization cues for the vowel-resonant (coda) or resonant-vowel (onset) transitions. In inter-vocalic positions, which have two such transitions, competition will not be that fierce. The cues to glottalization will surface adjacent to the stressed vowel, possibly because they are the marked cues, and the stressed vowel is acoustically strong enough to support both cues. However, in a case where the glottalized resonant is post-vocalic, but preceding another consonant, there is only one vowel-resonant transition, and two competing sets of cues. In this case, faithfulness to the place/manner cues is more important, leading to a shift in glottal timing. ' depending on the relative ranking of place cue constraints and creak cue constraints 22 Importantly, this does not lead to neutralization, so Preserve [creak] must be ranked higher than Context Cues [creak], since [creak] surfaces in a less optimal environment. Since place cues play a role in this analysis, they must be represented by a constraint which outranks Preserve [creak] (8) Preserve [place/manner] Place/manner cues present in the input must be present in the output12 Independent evidence for this can be seen in the fact that Sancaean does not appear to have place assimilation, and possibly also by the fact that in the official orthography of Sancaean, resonants and glottalized resonants are not distinguished. Deglottalization is also common among younger speakers of many Salish languages. The constraints can be seen in the tableaux below. Compare the intervocalic post-stress glottalized resonant in Tableau 4 , with the pre-consonantal glottalized resonant in Tableau 5. Tableau 4 k wan'an' Preserve [place] » Preserve C r e a k » Context Cues creak 2 k w a'na'n (pre-glott) 2. k w an'a 'n (post-glott) *! 3. kwa'nan The first candidate violates no constraints, and thus surfaces as the winner. The second candidate is post-glottalized and thus violates Context Cues2. The third candidate violates Preserve Creak because no creak feature surfaces. This does not show any crucial rankings however. The low ranking of Context Cues2 is shown in the next tableau: 1 2 In a d d i t i o n to th is f a i th fu lness const ra in t , a C o n t e x t C u e s [place] cons t ra in t is a l so c o n c e i v a b l y necessary . H o w e v e r , as this thesis m a k e s n o rea l c l a i m s r ega rd i ng p l ace a s s i m i l a t i o n , a n d the contexts i n w h i c h it m a y or m a y not o c cu r , the const ra in t w i l l no t be e x p l i c i t l y s h o w n i n the tab leaux g i v e n . 23 Tableau 5 ?an'xstsxw Preserve [place] » Preserve C r e a k » Context Cues creak 2 ?an'xstsxw (place cues unobscured) * ?a'Nxstsxw (place cues obscured) *! 3. ?anxstsxw *! The winning candidate is the first candidate, which salvages place/manner at the expense of violating Context Cues. Following Silverman, pre-glottalized resonants in this position would obscure place/manner cues, thus violating Preserve Place. However, since faithfulness to place / manner cues is ranked higher, this will allow the correct candidate to win. The third candidate, that eliminates [creak] if it cannot be in its optimal environment, is ruled out due to a higher ranking of Preserve Creak over Context Cues. Unfortunately, while we are now able to account for the distribution of glottal timing, we are unable to explain the distribution asymmetries which Steriade set out to explain. If one imagines a word with an underlying word-initial resonant, with stress on the immediately following vowel, our constraints cannot secure that this glottalized resonant not surface: Tableau 6 / R ' V C / Preserve p l a c e » Preserve C r e a k » Context Cues creak'1 R ' V C (post-glott) *! ^ ' R V C (pre-glott) * R V C *! The third candidate is the intended winner, but loses due to the Preserve Creak violation. A pre-glottalized resonant violates Context Cues Creak, and a post-glottalized resonant violates Preserve Place. Recall that Preserve Creak had to outrank Context Cues Creak, as we saw in Tableau 5, while it seems the opposite ranking is necessary here. Post-consonantal glottalized resonants cannot be ruled out either. Thus, while we can account for the timing of glottalized resonants in Sancaean, we cannot explain the neutralization that occurs word initially or post-consonantally. Part of the answer lies in a language specific phonotactic constraint against CR initial roots ("there are no root initial obstruent-resonant sequences in Saanich" (Montler 1986: 127)), which ipso facto rules out C R clusters. This can be formalised as a constraint: *CR. Since this constraint is never violated, it must be very highly 24 ranked. Why no word initial glottalized resonants occur remains unsolved. To conclude this section, Licensing by Cue seeks to base the distribution of glottalized segments on their phonetic implementation factors, i.e. types of cues and glottal timing. Word initial and post-consonantal glottalized resonants are predicted not to surface due to a lack of cues. However, in Sancaean it was demonstrated that these phonetic implementation constraints could not rule out word initial or post-consonantal glottalized resonants on their own, and that they were not enough to explain the variable timing of glottalization, even after taking stress into account. It is also worth noting that stress is not a factor in determining whether a glottalized resonant is glottalised or not, but does determine the timing of the glottal constriction with respect to the rest of the segment. The distribution of segments is not directly related to phonetic constraints in the same way that glottal timing seems to be. While Licensing by Cue certainly does not rule out an analysis which involves the interplay between phonological and phonetic constraints, it is clear that the distribution of segments cannot be simply a result of phonetic cues. Phonological constraints such as *CR must also play a role. In the next section we will apply the stress sensitive Licensing by Cue hypothesis to the St'at'imcets data and see i f it can account for this data any better. 4.3 St'at'imcets In the previous section it was demonstrated that Licensing by Cue could not account for the distribution of Sancaean glottalized resonants based on glottal timing relations. This section applies the Licensing by Cue hypothesis to St'at'imcets and shows that with a reranking of the perceptual scale of Context Cues, Licensing by Cue can account for St'at'imcets glottalized resonants and their distribution, but cannot explain the rarity of word-initial glottalized resonants. Just as in Sancaean, stress seems to play a major role in determining whether a resonant is pre-or post-glottalized, rather than syllable position or a context supportive of cues. Recall that the timing of glottalized resonants in St'at'imcets is exactly opposite to Sancaean: pre-glottalized resonants occur preceding the stressed vowel, and post-glottalized resonants occur following the stressed vowel. Given the Sancaean data, and Silverman and Steriade's account that stress is the optimal place for cues to glottalization, St'at'imcets seems to do the opposite of what is phonetically 'logical'. Before applying Licensing by Cue, consider the predictions it makes. Licensing by Cue predicted that cues to glottalization would only occur in environments that supported them, and if no such environment existed, the laryngeal feature would be neutralised. In the Sancaean analysis, stress was a conditioning factor, and so Context Cues, representing a perceptual scale of which contexts best supported cues to creak, was amended to include stress. This version of Licensing by Cue predicted 25 that glottalization would be attracted to stress, so pre-stress resonants would be post-glottalized and post-stress resonants would be pre-glottalized. If we summarise the matches and mismatches between the predictions of Licensing by Cue and the actual descriptions van Eijk gives, we see the following chart: (9) word initial post-C pre-stress post-stress pre-C word final LbC VanEijk #R'V #'RV m'am's-mas 'calf C R ' V C ' R V k'lax 'muskrat' V R ' V V ' R V niam'an 'to put s.t. into s.t.' V ' R V V R ' V ti-sqlaw'a 'beaver' V ' R C VJR'C qam'ts 'to cause to get hit' V'R# VR'# psil ' 'daylight' As we can see, the proposed Licensing by Cue analysis fails to account for even one context. Pre-glottalized resonants appear in post-consonantal position, and post-glottalized resonants occur preceding consonants, precisely where the cues are non-optimal! Consider the tableaux below, which illustrate two of the mismatches between Licensing by Cue and the timing descriptions given by van Eijk. Context Cues will be violated i f glottalization is not adjacent to the stressed vowel. Tableau 7 /#RV/ Preserve Creak » Context Cues creak 1. #R'V (post-glott) 2. #'RV (pre-glott) *! 3.#RV *! Candidate 2. is the intended winner, but Candidate 1. is the actual winner because glottalization is in its optimal position: adjacent to the stressed vowel. The third Candidate loses because it violates Preserve Creak. Considering word-final glottalized resonants, the proposed Licensing by Cue analysis fails again for the same reasons: Tableau 8 /VR#/ Preserve Creak » Context Cues creak 1. VR'# (post-glott) *! -*2. V'R# (pre-glott) 3. VR# *! 26 In this tableau, Candidate 1. is the intended winner and fails because it violates Context Cues (glottalization not adjacent to stressed vowel), while the second Candidate wins with no violations. However, given the explanations that both Silverman and Steriade gave about stress, that it is the optimal cue enhancer, the proposed analysis does not seem to be the right explanation. There must be another motivating factor involved. Perhaps the Silverman/Plauche analysis will work better. Recall that these two papers predicted that resonants will be pre-glottalized in onset position, and post-glottalized in coda position, in order to avoid obscuring place cues. At first glance, this seems to provide a reasonable explanation for St'at'imcets glottalized resonants, which are pre-glottalized preceding the stressed vowel and post-glottalized following it. This would suggest that preservation of place cues be very highly ranked. Consider the chart below: (10) matches S/P VanEijk word initial #'R^ #'RV, post-C C ' R V C ' R V pre-stress V ' R V VfRV pre-C V R ' C V R ' C word final VR'# VR'# mismatch post-stress V ' R V V R ' V This version of a cue-based theory has more matches than the version of the Licensing by Cue hypothesis proposed by Steriade (1997), but this appears to be more coincidence than method. Silverman/Plauche et al. predict that all onset resonants will be pre-glottalized, and all coda resonants will be post-glottalized. The crucial example then is the post-stress, but onset-position resonant, which Silverman et al predict to be pre-glottalized, but is in fact post-glottalized. Thus, syllable position in and of itself is not an accurate determiner of pre- vs. post-glottalization of resonants. The two cue-based approaches considered until now do not appear to be able to account for the data. Absence of cues cannot explain the distribution of glottalized resonants in St'at'imcets, because glottalized resonants have their cues to creak in exactly the contexts predicted to be least optimal, consistently avoiding the most optimal. Obscuration of place cues cannot explain it either, since post-stress R in onset position is post-glottalized not pre-glottalized. Either we must say that St'at'imcets maximally preserves place cues adjacent to the stressed vowel, and that this requires glottal cues to shift to the opposite end of the segment (thus leaving us no way to distinguish St'at'imcets from Sancaean) or we must seek a different solution to this puzzling question: why are the cues to glottalization not in the transition to the stressed vowel, a context which would render them optimally salient? 2 7 Consider again Silverman's theory of competing cues-that glottal cues could render place cues unrecoverable should they overlap. 1 3 Along this same line of reasoning, i f cues to stress were somehow incompatible with cues to glottalization, the same sort of competition would occur. Stress is a perceptual phenomenon which has different acoustic cues in different languages. For example, Ladefoged (1993) defines stress in the following way: " A stressed syllable is often, but not always, louder than an unstressed syllable. It is usually, but not always, on a higher pitch." (113). This carefully non-committed description contains a piece of alarming information—stressed segments may have higher pitch. ".. .an increase in the flow of air out of the lungs will also cause an increase in pitch, so that stressed sounds will usually have a higher pitch" (251). The worrisome part of this is that creakiness is often cued by a decrease in pitch. "Thus creaky voice usually has a low pitch as well as a particular voice quality." (251). Recall that one of the acoustic correlates to glottalization was slowing down of the vibration of the folds, and thus a decrease in pitch. If in St'at'imcets one of the correlates of stress is higher pitch, and one of the correlates of glottalization is lower pitch, then the two are not compatible. This would result in a perception conflict for the listener. If it is the case that the two may not overlap, then the cues to stress would render cues to glottalization less recoverable. This hypothesis would predict that glottalization of resonants should never occur on the transitions to or from a stressed vowel. This is exactly the distribution of glottal timing seen in St'at'imcets, as we can see below in the chart comparing the predictions of the Stress Theory (ST) and van Eijk's timing descriptions. (11) matches ST , van Eijk word initial #'RV #'RV post-C C ' R V C ' R V pre-stress V ' R V V ' R V post-stress V R ' V V R ' V pre-C V R ' C V R ' C word final VR'# VR'# The only way to test this hypothesis is to conduct phonetic research, first determining whether the timing descriptions are accurate. Then, the acoustic correlates to both stress and glottalization must be documented. Finally, i f the above prove to be as predicted, an analysis of glottal timing when the resonant is adjacent to an unstressed vowel must be done, (on data such as : C ' R V C V q?al'utsut V ' R C V C qa'mtap?l V C V R ' # , qiq?l' V C R' V nsalT?c) The prediction made by the theory would be that glottalization could occur adjacent to the vowel in these contexts, because no antagonistic stress cues would overlap. We have some evidence of this from intervocalic glottalized resonants already. However, it would be important to consider any 1 3 Thanks to Su Urbanczyk, Edward Flemming and Alan Yu for suggesting this. 28 effects from secondary stress. What now must be done is to translate this theory into the Licensing by Cue framework and formalise it. Since the stressed vowel is the worst context for cues to glottalization in St'at'imcets, does this force something as radical as inverting the Context Cues scale? Clearly, this would not be advisable, since inverting the perceptual scale for creak would mean that *creak/[-son] would be lowly ranked, suggesting that creaky obstruents are less marked than creaky sonorants. Since Steriade predicts that glottal obstruents never have the feature creak, but rather the feature ejective, this is not a favourable solution1 4. However, some change needs to occur. Steriade notes that Context Cues is relative to a specific set of cues, and recall that there are numerous cues to glottalization. Plauche et al very carefully claim that "for those languages that rely mainly on creaky voice, full glottal stop and amplitude as phonetic cues for glottalization, these segments will surface as preglottalized in onset and post-glottalized in coda". (1999:381). The authors examined a language (Lai) and discovered it did not fit the timing they predicted. As a result, they concluded that other cues to glottalization, such as sonorant length or vowel length may not be forced into the same distribution as a result of their cues. This demonstrates the fact that Context Cues must be sensitive to the particular cues that are involved. Thus, it is not unreasonable, and in fact is consistent with Licensing by Cue that Context Cues be altered in general, and specifically for St'at'imcets. Perhaps, then, Context Cues Creak is too broad a constraint, and needs to be dissembled into its individual cues, for each language. As seen in chapter 3), glottalization in Sancaean has several different cues. Presumably, each has its own perceptibility scale and faithfulness constraints which apply to it. These constraints would also have to be ranked with respect to one another, and it may turn out that glottalization in one language is cued most importantly by one thing, thus its highly ranked, and less so in other languages. Obviously, before any of this can be done, phonetic research on the exact nature of which cues are present in which contexts needs to be done. The alternative to such a decomposition, adopted here, is to account for the stress effects by reranking Context Cues. If adjacent to the stressed vowel is the least optimal place for cues to glottalization, owing to the fact that they will be neutralised due to the perceptibility difficulties then Context Cues creak for St'at'imcets must rank *V7[creak] very highly. Recall Steriade's definition of the Context Cues constraint in (1) Context Cues [creak] * [creak] in positions where context cues to [creak] are absent 1 4 Please see Chapter 5. 29 Sharing a transition to or from a stressed vowel with cues to stress would represent the ultimate absence cues, and thus violate Context Cues. The perceptual scale that this constraint represents would be *V7[creak]»* [creak] elsewhere, or Context Cues 3. Given this, glottalization adjacent to the stressed vowel will violate Context Cues [creak] 3. Since no neutralization of glottalized resonants occurs in St'at'imcets, Preserve Creak must be highly ranked. Word initial glottalized resonants are allowed in St'at'imcets, and this analysis will secure for us that they are possible. Applying these constraints to the St'at'imcets data we see the following tableaux: Tableau 9 / R V C / creak . Preserve Creak Context Cues [creak] 3 l . R ' V C (post-glott) *! 2.-»'Rv"C (pre-glott) 3. R V C *! The first candidate, post-glottalized, with its cues being obscured by the cues of the stressed vowel, violates Context Cues. The third candidate violates Preserve Creak, and the second candidate is our winner with no violations. Post-C glottalized resonants are also allowed and accounted for, as in Tableau 10: Tableau 10 C R V creak Preserve C r e a k » Context Cues creak 3 1. C R ' V (post-glott) *! 2. - ^ C ' R V (pre-glott) The first candidate again has its cues overlapping with the stress cues of the vowel, violating Context Cues creak, while the winning candidate incurs no violations. The same pattern can be seen in Tableau 11, where the pre-glottalized candidate 2 violates Context Cues. Tableau 11 V R C creak Preserve C r e a k » Context Cues creak 3 l .-»VR'C (post-glott) 2. V ' R C (pre-glott) *! 30 Tableau 12 and 13 show intervocalic glottalized resonants, in 12 the resonant is pre-stress and in 13 it is post-stress: Tableau 12 V R V creak Preserve C r e a k » Context Cues creak 3 l . V R ' V (post-glott) *! 2. - » V ' R V (pre-glott) Tableau 13 V R V creak Preserve C r e a k » Context Cues creak 3 1. - » V R ' V (post-glott) 2. V ' R V (pre-glott) *! In Tableau 12, the post-glottalized resonant (candidate 1) loses because glottalization is adjacent to the stressed vowel. The mirror image is seen in Tableau 13, where the post-glottalized resonant (candidate 1) is the winner because its cues are not adjacent to the stressed vowel. In conclusion, just as we saw in the Sancaean analysis, stress needs to be taken into consideration in St'at'imcets. Pure Licensing by Cue theory, which predicted that all glottalized resonants are pre-glottalized, and therefore neutralised in word-initial and post-C positions as a result of lack of cues does not apply to St'at'imcets. First, the timing description given predicts variable timing, dictated by stress, and second, glottalized resonants occur in positions which under Steriade's hypothesis they should be neutralised. Applying a modified version of the Licensing by Cue hypothesis which reasons that a stressed vowel is the optimal context for glottal cues fails because the opposite is true in St'at'imcets. It was hypothesised that i n fact, stress has cues which work antagonistically against cues to glottalization, thus resulting in the pre-stress/pre-glottalized post-stress/post-glottalized pattern evident in St'at'imcets. Unfortunately, confirmation of this hypothesis is outside the realm of this thesis and must remain another topic for further research. The question of why glottalized resonants do not occur regularly in root-initial position also remains unanswered. 31 4.4 Chapter Summary This chapter introduced the Licensing by Cue hypothesis, which claims that the distribution of glottalized segments is governed by their glottal timing and the perceptual scale which represents where the cues to glottalization are most salient. Glottalized resonants in neither Sancaean nor St'at'imcets could be accounted for in the way this theory predicted. Since Sancaean had the same distribution as Yokuts, one of the languages Steriade uses to confirm her hypothesis, in order for Sancaean R' to be accounted for under this hypothesis, they would have to be consistently pre-glottalized. From the evidence in chapter 3 this is obviously not the case. However, with the addition of stress to Context Cues [creak], and the further addition of a constraint based on Silverman/Plauche et al.'s idea of cue obscuration, the distribution of glottal event timing was accounted for. The distribution of glottalized resonants, however, could not be accounted for using these constraints. St'at'imcets glottalized resonant distribution could not be accounted for the original Licensing by Cue, the version applied to Sancaean or the Silverman/Plauche analysis. This was because, in direct contrast to what Licensing by Cue predicts, glottalization occurs in exactly the contexts where it should be neutralised. The solution proposed was that cues to stress are antagonistic to glottal cues, and instead of rendering them unrecoverable (i.e. neutralization), glottal stricture occurs at the opposite end of the segment. Thus while it first appeared that Licensing by Cue could not account for St'at'imcets at all, it turns out to provide a much neater account than for Sancaean. In both languages stress is the major determiner of glottal timing. It has opposite effects in the two languages, suggesting that the cues to stress are different. More phonetic research must be conducted in order to define exactly which cues to glottalization are present in which language and in . which contexts, before the proposed analysis can be evaluated. It was suggested in this chapter that perhaps Context Cues was too broad a constraint, and that each individual cue to glottalization in a particular language might need its own context cues perceptibility scale. This suggests that different languages could have their cues ranked differently to one another. For example, some languages may rank faithfulness cues such as pitch drop higher than more energy in the higher frequencies. If these two cues are salient in different contexts, then maybe distribution of glottalized resonants in different languages is dependent on how the language rates the importance of each cue. Perhaps this might explain the lack of word-initial glottalized resonants in both Sancaean and St'at'imcets. It would be interesting to see i f glottalized resonants in this position had all of the same cues as glottalized resonants in other positions. As we will see in the next chapter, in Sancaean, derived glottalized resonants do not necessarily have the same cues as non-derived 32 glottalized resonants, lending some support to this notion of 'unpacking' Context Cues into the individual cues. The conclusion drawn from this chapter is that while an analysis within the same general framework, and with the same underlying concepts as Licensing by Cue is possible, it does not occur as Steriade (1997) predicted. Phonetic constraints determine the timing of glottalized resonants, but phonological and morphological constraints determine the distribution of segments. The fact that stress has two very different effects on glottal timing in these two languages suggests that phonetic cues to both stress and glottalized resonants vary between languages. While this is not terribly surprising, the implications are noteworthy. Just how much variation in the production of glottalized resonants exists? It is possible that languages may use any or all of the cues demonstrated in Sancaean in any number of combinations. The interaction of these cues with different cues to stress or place or manner suggests an almost infinite number of possible timing, context and distribution relations. If two languages in the same family vary so much, how much variation occurs cross-linguistically? While the author doubts that there is truly an infinite number of possibilities, much more research needs to be done in order to understand the implications of the variation possibilities predicted by the proposed analysis. 33 C h a p t e r 5. G l o t t a l i z a t i o n a n d m o r p h o l o g i c a l p r o c e s s e s The previous chapters of this thesis have examined the nature of underlying glottalized resonants in Sancaean and St'at'imcets, and have applied to them a theory which seeks to account for their distribution based on their phonetic implementation factors. Any comprehensive theory of glottalized resonants must account not only for the distribution of underlying segments, but also for segments derived by morphological processes. Derived glottal resonants will also have phonetic cues to glottalization, and therefore, i f the theory is correct, their distribution will be governed by the same principles: phonetic implementation factors, timing, and cues. Parallel distribution, then, should depend on whether or not the cues are the same. The Licensing by Cue analysis adopted by Steriade for Yokuts can indeed account for the distribution of both derived and non-derived glottalized resonants in that language. However, as we saw in the previous chapter, when applied to Sancaean and St'at'imcets, Licensing by Cue could not explain the distribution of segments, but with some modifications, could account for the distribution of the relative timing of glottal events. This chapter further confirms this pattern. It will be shown that the timing patterns accounted for by Licensing by Cue in Chapter 3) for Sancaean are also exhibited by derived glottalized resonants in post-stress position, but not clearly so for word final glottalized resonants. However, this chapter will show that an analysis of featural alignment, rather than a cue based analysis best explains the phonological distribution of resonants glottalized as a part of a certain morphological process, namely the a c t u a l in Sancaean and the i n c h o a t i v e in St'at'imcets. In a sense, then, this chapter will be a further test of the capabilities of OT. OT claims to account for variation in related languages as a result of simply reranking constraints. As demonstrated in the previous chapter, this is not straightforward. Both reranking and the addition of other constraints are required. Fortunately, OT fares better when accounting for the comparison of the phonological process of glottalization than it does for the distribution of timing of underlying glottalized resonants in the two languages. This chapter will show that the same set of constraints accounts for the glottalization of resonants in both languages. While historical linguistics is not the focus of this thesis, any comparison of two languages in the same family must consider the precursor to both. In this case, the proposed account of the glottalization of resonants casts some new light on the Proto-Salish form for the i m p e r f e c t i v e , and may clarify the relationship between Sancaean a c t u a l and St'at'imcets i n c h o a t i v e . 34 The research conducted here also raises some tangential issues. Regarding phonetics, there is the issue of cues. Are the cues of a derived segment the same as non-derived segment? This research also raises the issue of the relationship between glottal stop and glottalized resonants. Should they share the same phonological feature? If so, should it be the perceptive feature [eg] shared by glottal stop, ejectives and glottalized resonants, or [creak] potentially shared by only /?/ and glottalized resonants. Morphologically speaking, this kind of research provides linguists with valuable data about how languages use glottalization as a morphological strategy, and provides further support for the theory of featural affixation. Historically, it may provide some insight into the nature of glottalization in Proto-Salish, the kind of changes that have occurred and the variation that exists in the language family today. This first section will examine the predictions of Licensing by Cue with regards to derived glottalized resonants. 5.1 Licensing by Cue The Licensing by Cue hypothesis makes the claim that glottalized resonants i) have neutralization/distribution based on cues and the contexts in which they are most realised and ii) are characterised by the articulatory feature [creak]. A reflex of the hypothesis proposed by Steriade is "that it determines the outcome of morphologically governed processes of glottal association." (my emphasis)(1997:83). This is demonstrated in Yokuts, which has two suffixes which have a floating glottal feature associated with them. The glottalization surfaces on a post-vocalic sonorant in the root (t'oyx->t'oy'x-o? 'give medicine'). If no post-vocalic sonorant is available, glottalization surfaces as a /?/ (wis->wis?a? 'straighten') or not at all (picw->picwa? 'catch'Xp. 83). The failure of obstruents to participate in this process confirms for Steriade that [creak] rather than [eg] is the relevant feature. Second, it shows that resonants glottalized as the result of a morphological process are governed by the same cue-based constraints as underlying glottalized resonants. Resonants in post-consonantal position are not glottalized, just as no underlying glottalized resonant surfaces in this position. Thus, the distribution of derived glottalized resonants is not a result of a morphological constraint, but rather is actively 'determined' by a phonetic cue-based constraint. This certainly accounts well for the Yokuts data, but how will it fare with Sancaean and St'at'imcets? As we saw in these two languages, Licensing by Cue did not really 'determine' segment distribution, but rather the timing of glottal events. The saliency of cues does not explain the lack of word initial glottalized resonants in Sancaean and why they only occasionally occur in St'at'imcets. 35 Licensing by Cue was however, successful in accounting for the timing distributions. The predictions made by the analysis so far with respect to derived glottalized resonants then are as follows. Based on the previous chapter, it will not be surprising i f an analysis based on cues, while compatible with the analysis proposed, is not the dominating factor in determining which resonants are glottalized, as in Yokuts. If derived glottalized resonants make use of the same phonetic cues as underlying glottalized resonants, the timing distribution will be the same. Although the St'at'imcets data to verify this is unavailable, the Sancaean data is considered. Since only post-stress resonants are glottalized as a result of this morphological process, the version of Licensing by Cue developed thus far predicts that glottalized resonants immediately following the stressed vowel are pre-glottalized, and word-final glottalized resonants should be pre-glottalized. As it turns out, while some cues are consistent, others are not, and the timing status of word-fmal glottalized resonants is ambiguous. In general Licensing by Cue does seem to be able to predict the timing of derived glottalized resonants. The next section shows the phonetic timing of derived glottalized resonants. 5.2 Phonetic evidence in Sancaean The corpus of derived glottalized resonants is also small—approximately eight actual/non-actual pairs were considered. The clearest examples are given in Figures 5.1-5.4 Recall that in Chapter 2) it was claimed that common cues to glottalization were aperiodicity (irregular glottal pulses), drop in pitch (slower pitch pulses), drop in amplitude, and more energy in the higher frequencies. If we consider Figure 5.1, a comparison of a) the non-actual /t'ilam/ and e) the actual /t'at'il'am'/. The broad band spectrograms in c) and g) show resonants are pre-glottalized as expected. It is clear from the vertical striations, slowing down in pitch pulses and interrupted F l at the transition from vowel to glottalized resonant in g) that pre-glottalization is occurring. In fact, there is considerable glottalization on the vowel, perhaps more so than on the resonant itself. In i) and j) we see an expanded form of a) and e) and we can clearly see the much more periodic transition from the vowel into the first resonant in i) then in the j). This supports the prediction by Licensing by Cue developed so far: post-stress glottalized resonants are pre-glottalized. We can also see that the /a/ of the actual form is more glottalized than the nonactual from the fact that the pitch pulses are considerably farther apart. The final /m/ of the actual has slower pitch pulses than the one occurring in the nonactual, but just as in non-derived resonants, it is not clearly pre-or post-glottalized.. In b) is a spectral cross-section of the 40 ms. before the l\l closure. Comparing this with f), the lil preceding the glottalized resonant, we do not see the comparison we would expect. In chapter 3), more energy in the higher 36 frequencies was a cue to glottalization. From these two examples, we can see that this does not appear to be a cue to derived glottalized resonants. The energy peaks around 3000 Hz, while considerably different in appearance, are not larger in f) as we'd expect. Further evidence of this can be seen in d) and h), both taken in the middle of the IV closure, where it seems that the non-glottalized IV has more energy than the glottalized one. This would seem to lend more evidence to the notion that context Cues needs to be broken down into its component parts. Clearly glottalization is still perceived in this case, suggesting that some other cue may be more salient or important to the perception of glottalized resonants in Sancaean. Figure 5.2 is a comparison of the nonactual / actual pair a) /?aitanan/ and b) /?aiten'arj7. This is an example of what Montler (1986) claims is 'merger' between an inserted /?/ and a following resonant which become a glottalized resonant. Comparing the nonactual in a) with the actual in b) a marked irregularity and compression of the waveform can be seen to occur at the onset of the glottalized resonants, indicative of glottal constriction. This can further be seen by comparing c) (glottalized) with d) (non-glottalized). By comparing the transition from the vowel to the first resonant, we can see that the glottal pulses at the end of the vowel in c) are markedly slower than those in d), suggestive of glottalization. In addition of this, the now familiar interruption of fundamental frequency occurs at the onset of the resonant. The /a/ between the two glottalized resonants also exhibits much slower pitch pulses than in the non-glottalized example. In e) and f), it is shown that higher energy in the higher frequencies is not a cue to post-stress glottalized resonants, as seen in Figure 5.1 The word-final glottalized resonant, however, does not show any obvious signs of being pre-glottalized. As we saw in chapter 2 non-derived glottalized resonants, the resonant seems to be glottalized throughout. To further complicate matters, a narrow band spectral cross-sections in e) and f) shows the opposite of what one might expect, based on the cues for glottalization found in non-derived glottalized resonants. Previously, an energy spike in the higher frequencies was shown to be a cue to glottalization. However, given this example, the opposite seems to be true, thus calling into question the role this particular cue plays in signalling the glottalization of derived glottalized resonants. With a small corpus of data it is obviously impossible to make any generalisations, but it seems that while some cues are consistent not all are. Further evidence of this can be seen in Figure 5.3 In figure 5.3 we see the comparison of the nonactual / actual pair a) /yayasarj/ and b) /yayasarj'. Here again the word-final glottalized resonant is glottalized throughout. This is more clearly seen in the broad band spectrograms in c) (non-glottalized) and d) (glottalized). There is a gap in the formant structure at the onset of the nasal in d) not seen in non-glottalized c). While this may be a sign of pre-glottalization, it does not occur in the second token of this word, leading to doubt as to its 37 marking glottalization. As seen in e) and f), more energy in the higher frequencies is not a cue to glottalization in word-final derived glottalized resonants either. Thus, just as in Chapter 3), the timing of word-final glottalized resonants remains unclear. Figure 5.4 shows the word /q w alq w aliw'an7 'I 'm thinking'. From the broadband spectrogram we can see the onset of glottalization at approximately 700 ms. in a) marked by the dramatic decrease in amplitude. In b) the much fainter fundamental frequency can be seen at the end of the vowel and much slower pitch pulses carry on through to the end of the word. Notice that the /a/ becomes more heavily glottalized towards the end of the vowel. Comparing this with the pre-stress IV shows that only post-stress resonants are glottalized. While the /w/ is clearly pre-glottalized, as expected, the timing of the word-final resonant is unclear. To conclude this section, it seems that some cues to glottalization for non-derived glottalized resonants, namely, slower pitch pulses, irregular wave form and decrease in amplitude are cues for derived glottalized resonants as well. More energy in the higher frequencies does not seem to be. Post-stress glottalized resonants, are as predicted, pre-glottalized, while word final glottalized resonants appear to be glottalized throughout. Given this phonetic information, the rest of the chapter focuses on a phonological account of the formation of the actual in Sancaean and the inchoative in St'at'imcets. The next section gives some background on the morphology and history of the Salish imperfective. 5.3 Background The Sancaean 'actual' and St'at'imcets 'inchoative' make up part of the class of aspects referred to by Kinkade (1997) as the 'imperfective'. The actual tense in Sancaean is described as the 'action, state or other reference of the predicate is occurring at an indicated time" (Montler: 111). The St'at'imcets ingressive or inchoative 'generally refers to an incipient change or a change in progress." (van Eijk: 67). The formation of this aspect in both languages involves glottal insertion and the glottalization of resonants; however, it is not completely clear that the two are reflexes of the Proto-Salish imperfective morpheme. Kinkade (1997) claims that "the basic aspectual opposition throughout Salish is one of 'perfective' vs. 'imperfective"'(2) with the imperfective generally being the marked form. Included in the imperfective category is, among others,' actual' and inchoative, where the St'at'imcets inchoative is considered a tertiary aspect. Kinkade (1997) attempts to reconstruct the Proto-Salish imperfective, concluding that there are two competing candidates for the role: ?-insertion and C V reduplication. Kinkade suggests "that either the Interior Salish 'inchoative' infix or the Interior Salish 'receptive' infix is cognate with this 38 Coast Salish 'imperfective' infix, and that this allows us to reconstruct an infixed *-?- as the Proto-Salishan morpheme for 'imperfective'. The change from 'imperfective' to 'inchoative' is not a great one, although the resulting odd distribution of the two 'inchoative' forms in Interior Salish is unexplained, as is the origin of the -p variant there."(17). From this statement it is clear that while the exact nature of the relationship between the Sancaean actual and St'at'imcets inchoative is not known, the possibility that they are related is a strong one. This being the case, one would expect OT to be able to account for the process in both languages using the same constraints. This is indeed what the data suggests. The next section will present the formation of the actual in Sancaean, give a brief account of previous analyses which do not focus on glottalization, and will develop an analysis based on featural alignment. The following section will do the same for the St'at'imcets inchoative using the same constraints. 5.4 Sancaean As mentioned above, the Sancaean actual is an aspect which conveys the meaning of a change in progress. Montler (1986) claims that it is the most frequently occurring aspect after the 'nonactual'. The formation of the 'actual' in Sancaean has three major allomorphs: ?-insertion (the most common form, occurring in half of the three hundred tokens recorded by Montler), C V reduplication15 and metathesis. A residual form, C reduplication, is lexically determined by the root, and occurred in only nine forms recorded by Montler. Allomorphy is otherwise determined by root shape. Each of these forms is accompanied by glottalization of post-stress resonants. Two previous analyses of the formation of the actual tense are Montler (1986, 1989) and Stonham (1994), both of which only briefly mention the glottalization of resonants. Since glottalization is the focus of this thesis, I will only briefly introduce the allomorphy conditions, followed by summaries of the two previous analyses. I will then examine the data relating to glottalization of resonants. As mentioned above, the Sancaean actual has three surface forms, and the resultant allomorph is dependent on root type. Canonical root types of Sancaean are: C V , C C , C V C , C V C C , C C V C , and C V C V . It is not completely clear what the generalisations regarding root type are (there is some disagreement between Montler and Stonham about which root types pattern together), but the following are broad observations: (1) i . Underlyingly vowelless roots (CC) form the actual using metathesis (1) ii.Roots with the stressed vowel in a syllable closed with only one consonant use C V reduplication. Attested roots are: C V C C V C , C C V C , C V C (2) 1 5 In Montler (1989) he renames this C l reduplication. 39 iii.?-insertion is used elsewhere, i.e. C V , C V V C , C V C C , C V C ? V , C V C V C , and if the stressed vowel occurs in a suffix. (3) (2) Non-actual Actual Gloss t'sat fast break something (intentionally) 2 q w al q w aq w a l ' talk/say 3 weq?s we'-?-qas 'yawning' (Montler 1986) There is much discussion in Salish linguistics as to the nature of /a/, ( c f Blake (1999), Czaykowska-Higgings (1993), Kinkade 1997b) Shaw (1996). Urbanczyk (1999), It may be underlying, phonetically excrescent or epenthetic for purposes of stress. CC roots (as in 1.) are considered by Montler to be underlyingly vowelless, and surface with stress on a suffix which never otherwise stressed. Roots as in 2. are considered to have /a/ underlyingly (they do surface with stress). While this issue of underlying vs. epenthetic schwa is interesting, it is somewhat outside the purview of this thesis, and so I will follow Montler's assumptions. Given these generalisations, I will now briefly summarise the two existing analyses of the actual allomorphy. The reader is encouraged to consult these sources for a more in-depth account. 5.4.1 Montler (1986,1989) Montler (1986) proposes a rule-based analysis of the actual, which is altered to become a templatic analysis in Montler (1989). Montler (1989) links all of the allomorphs of the actual to an underlying form that "can be thought of as an abstract C_CC frame for the stressed vowel of the stem." (p.93). The goal of the actual, then is to " achieve a C V C C structure. If the stressed syllable of the stem is open, close it with a glottal stop. If it is already closed, rearrange whatever is there into C V C C . So i f all that is there is C V C , reduplicate the first C, and if there are three consonants, C C V C , metathesize the vowel and the second consonant"(1989:101). With respect to glottalization, Montler makes claims that glottalization of resonants accompanies all forms of the actual. In reduplicated and metathesized forms, all non-initial resonants are glottalized and in infixed forms, all post-stress resonants are glottalized. He suggests that the floating feature [eg] is always associated with the actual. With glottal infixation, [eg] is realised as /?/ which closes the open syllable. It is anchored there and spreads rightward. In reduplicated and metathesized forms, [eg] is not anchored, and thus spreads to all non-initial resonants. 40 5.4.2 Stonham (1994) 3 Stonham 1994 accounts for the different formations of the actual by proposing a mora-based hypothesis. Rather than using a template, he proposes that the actual morpheme is a mora. Each allomorph surfaces based on individual root shapes adding a mora according to a hierarchy of strategies for root change, based on structure preservation: 4 (3) i . Lengthen the root vowel 5 i i . metathesize C and V or root to close the syllable 6 iii .Add a consonant to the coda 7 iv. Reduplicate first mora (P-175) The first, he argues, is not possible in Sancaean due to a lack of length distinction for vowels. The second occurs in the roots Montler considers vowelless and which Stonham considers underlyingly CCa. He gives no reason for selecting CCa rather than CC as the underlying representation. The simplest way to increase the mora count of a CCa root with minimal change is simply to metathesize the vowel and C2, thus forming a bimoraic root from a previous monomoraic root. The third option is selected when the stressed vowel occurs in open syllables. Since metathesis in these cases is impossible (no extra C to move to coda position for extra weight), weight must be added by a different strategy. The next least 'destructive' strategy would be to add a placeless consonant to the coda. For roots with the stressed vowel in an open syllable, a -?- is inserted following the stressed vowel. Reduplication, then is the least favourable strategy, occurring only when weight cannot be added by any other means, i.e. CVC# or CVC+C roots. In these cases, the bimoraic root already has the maximum allowed number of moras, and the only option left, according to Stonham, is to reduplicate the stressed vowel. For some inexplicable reason, C l is also reduplicated. Both of these analyses focus on finding a comprehensive phonological solution that will derive each of the three main allomorphs from one phonological or morphological source. However, as noted in Kinkade (1997), C V reduplication and ?-insertion are considered competing candidates for the Proto-Salish morpheme for the imperfective, "Both create imperfectives in Halkomelem and Straits, but have different functions elsewhere; it is therefore likely that they had different functions in Proto-Salish, but merged as conditioned variants of 'imperfective' in Halkomelem and Straits" (p.17-18). Given this statement, there is some question as to whether one should try and link the various allomorphs phonologically at all. The one characteristic all allomorphs of the actual share is resonant 41 glottalization. Since even the allomorphs which do not have ?-insertion use resonant glottalization as part of the actual, this strongly suggesting that the proto-morph may well have been the feature [creak] or [eg]. Based on the fact that all allomorphs use resonant glottalization, I propose that it is possible to separate resonant glottalization from the other allomorphs of the actual, and to consider it as a phenomenon on its own. This separation allows us to form a unified analysis of resonant glottalization which governs all allomorphs. 5.4.3 Glottalization of resonants Each of the three 'major' allomorphs of the actual is accompanied by glottalization of only and all post-stress glottalized resonants. The pattern of glottalization is the same for each allomorph, including lexical C reduplication. In the data below, examples 1-3 are metathesis, 4-5 are C V reduplication, 6 is C reduplication and 7-11 are ?-insertion: (4) Nonactual 1. t's-at 2. st-an t'am'-at 4. q w al 5. kwin-tol 6. t'ilam Actual t'as-t sat-an' walk t'am'-t qw3qw9l' Gloss break something (intentionally) hit (as in the table) talk/say kwiw'9n'-t3l' fight (note,kw~w) t'at'fl'am' sing 16 17 7. weq?s 8. x w i t 9 n 9. ?fi9n 10. ?9it9n9n 11. yoyassn we-?-qos yawning (Montler 1986) xwi-?-t9rj' jump ? i - ? - ion ' eat ?3iten' orj' berry picking y3ya-?-son' play 1 6 Example 1 shows an underlyingly vowelless stem with no resonants. According to Montler, the underlying form is/t's/with the transitivising suffix/-at/. The actual form has C2 and the suffix hi metathesized. Examples 2 and 3 are also underlyingly vowelless roots, but have different surface forms. This is the result of the *CR phonotactic constraint. The underlying forms of the nonactual for 2 and 3 are t'm' with the transitivising suffix /-at/, and sht with the 'middle' suffix, /an/. If the form of the nonactual is taken to be CC-aC, ex. 2 /t'm'-9t/ would violate *CR and thus a /a/is inserted between Cl and C2. If the form of the actual is taken to be CaC-C, then ex. 3 /sat-nV would also violate *CR, and thus a /a/ is inserted between C2 and C3. 1 7 An artefact of my elicited set of data makes it appear that CC roots which metathesize end in obstruents, while the CV reduplication cases (4-5) which also have hi as their vowel, tend to end in resonants. This is merely a result of the examples I have elicited. Montler (1986) lists nine obstruent final CC roots and six resonant final CC roots. 42 Example 1 shows that obstruents are not glottalized, while ex. 2 shows that post-stress resonants, even in suffixes, are. This suggests that the domain of glottalization is the word rather then the root and more evidence of this can be seen throughout the data. Example 3 shows that an already glottalized resonant absorbs glottalization. Examples 4. and 5. are C V reduplication. In ex. 5., we can see that all post-stress resonants are glottalized, not just a final one. Example 6. shows that all post-stress resonants are glottalized. In ex. 7-9 we see that a glottal stop is inserted following the stressed vowel and preceding the obstruent, which does not become glottalized. In ex.10, however, no glottal stop surfaces. This is because glottal stop merges with a following resonant, suggests Montler (1986) . In ex. 10. we see a change from a to full vowel preceding the glottalized resonant. Ex. 11. shows that only post-stress resonants are glottalized. Assuming a separation of ?-insertion and resonant glottalization, we are free to consider the glottalization accompanying all allomorphs of the actual, separately from these allomorphs themselves. The following section will introduce the basis of the proposed analysis, namely Featural Affixation, before proceeding to the analysis itself. 5.4.4 Featural Affixation I suggest that the morpheme representing the actual is the floating feature [creak]. This feature suffixes to the stressed vowel and then aligns itself with the right edge of the word. Obstruents are not affected by this process due to a highly ranked feature co-occurrence constraint *[-son]/creak. The idea that features can function as grammatical morphemes is not a new one, with the basis for an analysis within OT being Akinlabi (1996). In Featural Affixation, Akinlabi (1996) shows that featural morphemes are subject to the same constraints as segmental morphemes. Just as it has been argued that segmental morphemes are governed by well-formedness constraints on alignment (McCarthy and Prince 1993), so are featural morphemes. Akinlabi bases his theory on McCarthy and Prince (1994) Generalized Alignment. Under this theory, categories are aligned with respect to their edges, as in the now familiar Alignment Constraint: (5) Align (Catl,Edgel,Cat2,Edge2) (p.338-3) Edges are left or right and categories can be prosodic or morphological like Pword or foot or affix. Misalignment (like the famous Tagalog um-infix) is derived when a constraint on phonological well-formedness (like ONS) outranks the alignment constraint. Featural Alignment, argues Akinlabi, is an 1 8 Please see section 5.6 for a discussion of this. 43 extension of this theory. He proposes the following well-formedness constraint on alignment of phonological features: (6) Featural Alignment Align (Pfeat,Gcat) A prosodic feature is aligned with some grammatical category.(p. 4) The difference between featural and segmental morphemes is that while segments stand alone, features are often realised as a part of a segment. Thus, a featural morpheme will surface only if the correct 'licensor' or target is available. This also redefines the Generalized Alignment concept of edges to refer to moras and root nodes for featural alignment. Misalignment of featural morphemes occurs when a phonotactic or feature co-occurrence constraint outranks the align constraint. Taking Featural Alignment as a basis, then, I propose the following analysis. The featural morpheme of the actual is [creak]. It is suffixed to the stressed vowel and then aligned with the right edge of the word. Although Akinlabi first proposes that PFeat align with grammatical categories, he also states that 'it does not however exclude alignment to prosodic categories"(p.4). This, coupled with the fact that attraction to the stressed vowel is not uncommon, gives us the following alignment constraint: (7) Align actual L Align (actual L,V[+stress]R) The left edge of the actual must be aligned with the right edge of the stressed vowel. Phonetically, it was shown that glottalization surfaces on the vowel-resonant transition, or the last part of the vowel and the beginning of the resonant. Alignment to the left edge of the stressed vowel, then means glottalization must surface on the last part of the vowel or the transition from vowel to resonant. Alignment to any other position (i.e. the beginning of the vowel) wil l violate Align Left. As seen from the data, not only the first resonant following the stressed vowel is glottalized. A l l post-stress resonants are glottalized, while no pre-stress resonants are glottalized. This suggests a right alignment constraint: (8) Align actual R Align (actual R,word R) The right edge of the actual must be aligned with the right edge of the word. This constraint suggests that glottalization is continuous from the stressed vowel/resonant transition to the end off the word. The vowels between glottalized resonants do surface as glottalized, as seen from Figures 5.1-5.4, suggesting that this is indeed the correct constraint. More phonetic research is needed to show whether the vowels can be considered glottalized of their own accord, or only parasitically 44 from neighbouring resonants. Obstruents are not glottalized, and must therefore be excluded from glottalization. This is achieved through a phonotactic constraint against creaky obstruents. (9) *[-son]/[creak] If [-son] then not [creak] This constraint must outrank the two align constraints in order to guarantee correct outputs. An align violation will occur when a potential target, in this case root node, is skipped. Since obstruents are never glottalized as a part of the actual morpheme, any obstruent intervening between the stressed vowel and the end of the word will violate align right. However, the higher ranking of *[-son]/creak, will correctly rule out glottalized obstruents. In order to assure that [creak] surfaces in the output, we have Akinlabi's Parse-Morph constraint (10) Parse-actual the actual must be realised in the output. This constraint is also highly ranked, since the actual must always surface. The following tableau shows the interaction of these constraints, and shows that the same constraints can account for the glottalization across the 4 types of allomorphs. Tableau 1 /starj/ *[-son]/[creak] Parse a c t u a l » Align actual L Align actual [creak] R 1. stirj' 2. sterj *! Tableau 1. shows an example of a CC root (st). In this tableau, the second candidate violates Parse actual, and thus the winner is candidate 1. The next tableau shows an example of glottal insertion preceding an obstruent. Candidate 1. is the winning candidate, violating low ranked align R because the obstruent is not glottalized. Candidate 2., lacking resonant glottalization violates Parse actual. The third candidate, in which the obstruent has been glottalized, violates the highly ranked feature co-occurrence constraint. Tableau 2 /?i-?-isn/ [creak] *[-son]/[creak]» Parse a c t u a l » Align actual L Align actual R 1. ?i?ion' * 2. ?f?fcm *! 3. ?i?i'an' *! 45 In the next tableau, we see an example of C reduplication. Candidate 1 violates no constraints, while candidate 2 violates Align left, because [creak] is not aligned to the stressed vowel. Candidate 3 violates align right because [creak] is not aligned all the way to the right edge of the word. The fourth candidate violates Parse actual Tableau 3 /t 'at'ibm/ [creak] *[-son]/[creak] Parse a c t u a l » Align actual L Align actual R 1. t'at'il'am' 2. t 'at ' ibm' *! 3. t'at'il'am *! 4. t 'et ' ibm *! In tableau 4 we see the constraints acting to rule out pre-stress glottalized resonants. The winning candidate violates align right, because the Is/ is not glottalized. The second candidate is ruled out by a Parse actual violation, while the third candidate violates the highly ranked * [-son]/[creak] constraint. Candidate 4 violates Align L because [creak] is aligned past the stressed vowel. Tableau 4 /yaya?san/ [creak] *[-son]/[creak] Parse a c t u a l » Align actual L Align actual R 1. yaya?san' * 2. yaya?sarj *! 3. y ay as'an' *! 4. y'ay'asarj' * * i * Having considered the above tableaux, it is clear that featural affixation is capable of accounting for Sancaean actual resonant glottalization. An analysis which suggests that cues are the motivation factor behind the distribution of derived glottalized resonants fails in this case to explain why no pre-stress resonants are glottalized. For example, in tableau 4. there is no phonetically motivated reason that the intervocalic lyl should not be glottalized. Thus, it would seem that while a cue-based analysis certainly does not rule out a featural affixation based analysis, the claim that cues fully 'determine' the distribution of glottalized resonants is clearly inaccurate. 46 5.4.5 Montler revisited Recall that Montler (1989) claimed that all allomorphs of the actual have a [eg] feature associated with them, but that in cases of ?-insertion, [eg] is anchored to closing the stressed vowel, and it is this fact that distinguishes the two generalisations regarding resonant glottalization. In glottal insertion cases, post-stress resonants are glottalized, while in metathesis and C reduplication, all non-initial resonants are glottalized. By proposing that all allomorphs have the same distribution of resonant glottalization, and are governed by the same constraints, we no longer have a disjunct generalisation, and can account for all of the data under one generalisation Another advantage to positing that the Proto-Salish allomorph of the actual is the feature [creak] is shown in the next section, where the same analysis as for Sancaean also accounts for the St'at'imcets inchoative. This provides some support for the OT theory that related languages can be accounted for by using the same constraints. 5.5 St'at'imcets The St'at'imcets inchoative conveys the general meaning of a change in progress as well. There are two allomorphs for the inchoative, one of which it shares with Sancaean, namely, glottal insertion. This allomorph is used only in roots with full vowels. Roots which surface with reduced vowels use the suffix -p. Resonant glottalization accompanies only ?-insertion. Consider the data below: (11) non-inchoative l . l a k w - l a k w 2. tim 3. a i l 4. nus 5. me? inchoative gloss la?k w get loose, untied ti?am' snow melts everywhere cu?al' stretch nu?s damp ma?ai' breaking daylight Just as in Sancaean, ? is inserted immediately following the stressed vowel (1). If a resonants follows the stressed vowel, ? is inserted, but does not merge with it (2,3). Instead, a /a/ is inserted between the /?/ and the resonant. This is due to a phonotactic constraint *CR#. Examples 4 and 5 demonstrate that just as in Sancaean, only post-stress resonants are glottalized. Since word initial glottalized resonants are possible in St'at'imcets, this is strong evidence in favour of a featural analysis. 4 7 Below are two examples of roots that surface with /a/. Notice that -p is used rather than glottal insertion, and that resonants are not glottalized. 6. naq w naqw-p warming up 7. k w am kwamp dull As Kinkade (1997) notes, the origin of the -p suffix is unknown, and clearly not a possible reflex of Proto-Salish imperfective. This would explain the lack of resonant glottalization associated with it. If we now apply the same constraints as we saw in Sancaean, we will see that the same constraints will derive for us the correct data. The Align actual constraints have been renamed Align imp. (imperfective) and *CR# has been added to rule out ?R'#. In Tableau 6, the first candidate is the winner, because it incurs no violations. The second candidate violates *CR#, and the third candidate, which has glottal insertion, but not resonant glottalization, violates Parse imp.: Tableau 6 Aim/ [creak] *[-son]/[creak]» Align imp. L Align imp. R Parse imp. *CR# tfram' ti?m' i ti?am *! Tableau 7 shows that the constraint interaction rules out pre-stress glottalized resonants. Since word initial glottalized resonants occur in St'at'imcets, this distribution is clearly morphologically and phonologically, rather than phonetically, determined. Tableau 7 /lakw/ [creak] *[-son]/[creak]» Align imp. L Align imp. R Parse imp. *CR# la?k w la?k w ' *! l 'a?k w *! 48 Thus, having shown that the featural alignment analysis proposed for Sancaean can also account for St'at'imcets, I will briefly discuss the lack of merger in St'at'imcets. The explanation for the lack of merger may be grounded in phonetics. Recall that immediately following the stressed vowel in Sancaean, glottalized resonants are preglottalized. Thus, inserting a glottal stop immediately preceding a resonant might be homophonous to a pre-glottalized resonant. There seems to be some variety in how merged glottalized resonants are produced. What Montler (1986) calls merged glottalized resonants, Montler (1989) says have phonetic realisations of [?R'] [?R] or [R']. A l l three are written /RV because there is no phonemic contrast between the three. Recall that in St'at'imcets, glottalized resonants are described as being post-glottalized immediately following the stressed vowel. Thus, a glottal stop inserted immediately preceding a resonant will not be homophonous with a glottalized resonant in this position. Furthermore, sequences of ?R' are allowed in St'at'imcets, and so merging the inserted /?/ with the following glottalized resonant will not be perceived as a glottalized resonant. This discussion of merger brings us to the issue of features, the decoupling of [eg] into [creak] and [ejective] as proposed by Licensing by Cue. 5.6 [Creak] vs. [eg] Licensing by Cue as proposed by Steriade refers to the articulatory feature [creak], rather than the more traditional productive feature [eg]. Ejectives are taken to be characterised by the feature [ejective], while glottalized resonants have the feature [creak]. It is unclear what feature characterises glottal stop under this analysis. Under the traditional [eg] analysis, all three segments are characterised by [eg]. Throughout this thesis I have been following Steriade, but at this point, I'd like to consider the issue of which feature is preferable. The two major reasons [creak] is considered the active feature in Licensing by Cue is that ejectives and glottalized resonants seem to behave differently with respect to timing (ejectives are consistently post-glottalized, while resonants vary) and also with respect to morphologically processes. Obstruents are not glottalized while resonants are. Sancaean and St'at'imcets seem to lend support to the claim that [creak] is the active feature, since obstruents are not glottalized as a part of the imperfective, while resonants are. In Licensing by Cue as it is presented here, this is accomplished via a *[-son]/[creak] constraint. However, there is some question about the legitimacy of such a constraint. Under OT, all constraints are considered violable. The *[-son]/[creak] constraint on the other hand is never violated in Sancaean and St'at'imcets. Thus, the feasibility of using such a constraint depends upon the existence of a creaky obstruent in some language. Obviously, this is outside the scope of this thesis, but for the sake of 49 argument, assume for the moment that this is an unrealistic constraint. A counter analysis to this division of the traditional [eg] feature into [creak] and [ejective] is Howe and Pulleyblank (1999). Based on data from Wakashan, they propose an analysis using only the feature [eg]. In Wakashan, there is a class of glottalizing suffixes which targets any stem final consonant. Stem final obstruents become ejectives and resonants become glottalized resonants. Clearly having two features in this case is not as effective as suggesting that only the feature [eg] be used. In order to account for both the Wakashan and the Yokuts pattern, Howe and Pulleyblank (1999) propose an analysis based on markedness as faithfulness. (43) Core idea 1 glottalization is optimally realised on obstruent stops, less optimally realised on sonorants, and still less optimally realised on fricatives Core idea 2 Faithfulness maximises the retention of glottalization in optimal environments: MAXCG/stop»MaxCG/son»MaxCG/Fric Core idea 3 Faithfulness penalises more salient changes than less salient changes DepCG/Stop»DepCG/Son»DeptCG/Fnc In other words, it is easier to change the glottalization value of resonants than it is obstruents. The differing patterns of Wakashan and Yokuts are accounted for by the position of the align constraint with respect to stop faithfulness constraints. In the Yokuts (and Sancaean and St'at'imcets) patterns, stop faithfulness constraints outrank align, while in Wakashan patterns, the inverse is true. Since adding the feature [eg] to obstruents violates a higher ranked constraint than adding [eg] to resonants, a candidate which glottalizes only resonants, as we see in Sancaean and St'at'imcets, will win. Thus, it seems that an analysis which makes use of the feature [eg] is just as capable of accounting for the glottalization patterns as one which refers to [creak]. Another question that arises when considering this topic is the relationship of ? to glottalized resonants. Under the [eg] theory, both are characterised by the same feature. Under the [creak] theory, it is unclear what the relationship should be. Montler (1989), using [eg] suggests that the inserted glottal stop shares the same features as the glottalized resonants. However, glottalized resonants created as a result of merger behave different phonologically than those which are glottalized as a part of the glottalization process. In Montler (1986) he notes that a stressed /a/ in the nonactual becomes a stressed Id in the actual, when followed by a glottal stop. I suggest that this is due to a pan-Salish constraint against a? sequences (Shaw 1987,1997) *a?. In roots with glottal insertion, inserting a /?/ following a /a/ would violate *a?. Thus, the /a/ becomes e, as in ex. 10. /?aitanan/->/?aiten'arj V 'berry picking'. However, i f we consider the following example of a CC root, which uses the metathesis strategy rather than glottal insertion, we can see that the stressed a does not become e 50 preceding the glottalized resonant: 12. kw 'anat kw 'a'n't see This seems to indicate that no ? is present to cause the shift to a full vowel. If Licensing by Cue is correct in suggesting that phonological distribution is governed by phonetic factors, then the cues to a resonant glottalized as part of alignment must be different to a resonant derived as a result of merger (or a ?R sequence), despite the fact that both are perceived as glottalized resonants. If features are used in phonology to represent phonetic distinctions, and the phonetic cues of a ?R sequence are different to those of a derived glottalized resonant, then a distinction between ? and R' must be made. One way of accounting for this would be to use different features. Under Licensing by Cue, ejectives and glottalized resonants have different features because their cues and timing are different. The Sancaean data appears to suggest that /?/ also needs its own feature.19 Since the goal of Licensing by Cue is to bring phonetics into phonology, this interplay between features and cues is central to the overall framework of the theory. If the distribution of /a/ is affected by the cues to /?/, surely this should be represented in the phonology as a distinctive feature. However, given that ?R and R' can be perceived as the same segment would suggest that they do have a feature in common. It is clear from this discussion that more research needs to conducted in order to answer the questions raised. 5.7 Chapter Summary This chapter deals with resonants glottalized as a result of a morphological process. It is clear from the phonetic evidence presented that resonants immediately following the stressed vowel, whether glottalized as a result of ?-infixation or glottalization, are pre-glottalized as predicted. Word-final derived glottalized resonants were glottalised throughoutThus, while the preliminary phonetic evidence conducted here lends support to Licensing by Cue, more research needs to be done in order to verify the findings presented here. As seen in the previous chapter, Licensing by Cue is not sufficient to determine the distribution of segments. The distribution of glottalized resonants is shown here to be determined by morphological constraints and phonological rather then phonetic constraints. And while Licensing by Cue is clearly compatible with such an analysis, it is not directly responsible for the distribution of glottalized segments as predicted by Steriade (1997)Accounting for the distribution of glottalized resonants in Sancaean and St'at'imcets requires the interactions of three levels of constraints: phonetic to determine 1 9 It was pointed out to me that there may be a syllabic explanation for this alternation., rather than a featural one. Presumably, glottalized resonants created as a result of merger stil l have a ? root node, and thus the syllabification of R' and ?R may be different. (Pat Shaw p.c.) 51 glottal event timing, and phonological and morphological constraints to determine segment distribution. This chapter showed support for the OT claim that related languages are governed by the same set of constraints, and that language variation can be accounted for using the same constraints. Exactly the same constraints which accounted for Sancaean actual also accounted for the St'at'imcets inchoative. This offers strong evidence that the St'at'imcets inchoative is indeed the reflex of the Proto-Salish imperfective. The data presented here seems to suggest that a featural morpheme, whether it be [eg] or [creak] may well have been the proto-morpheme. How ?-insertion is related to this feature remains unclear. 52 Chapter 6 Conclusion The research presented in this thesis sought to answer many questions. However, as is common in the field of linguistics (or so I'm told) many more questions were raised than were answered. Glottalized resonants remain mysterious segments, but this thesis has laid the groundwork for future research. While no conclusions can possibly be drawn from such a small set of data, it is nonetheless important that all the small sets of data are eventually merged to form a more complete set of data, from which linguists may one day begin to understand language processing. , r While the study of glottalized resonants may seem an obscure place to try to understand the human mind, this thesis makes an important contribution in two ways. First, it is important to understand the diversity the mind is capable of, and since glottalized resonants are not well represented across the world's languages, this makes them something of a curiosity. They are not well studied, and are thus not well understood. Second, the languages that do have them are disappearing rapidly. Aside from the devastating loss to the community, the loss of these languages impacts the field of linguistics in that the already small subset of data is rapidly becoming smaller. One rich source of glottalized resonants is the languages of the Northwest Coast, and this thesis considers two Salish languages, Sancaean and St'at'imcets. No research, phonetic or phonological has been conducted on the glottalized resonants of these two languages. As demonstrated in this thesis, a surprising range of variation occurs even two members of this family. A comparison of the glottalized resonants of Sancaean and St'at'imcets proved quite challenging to the hypothesis considered, and to the claim that OT can account for the variation between related languages by simply reranking constraints. It was the goal of this thesis to apply a current theory of glottalized resonants, specifically the Licensing by Cue hypothesis, to languages outside the standard corpus. In doing this, we not only have parameters with which to frame our questions, but also are able to test and improve current theories. The Licensing by Cue hypothesis, and OT as well, should be able to account for the variation between Sancaean and St'at'imcets, glottalized resonants with minimal modifications. They should be able to account for derived as well as non-derived segments. When put to the test, these two theories did not perform as predicted. The Licensing by Cue hypothesis seeks to increase the role of phonetics in current phonological theory. The version of Licensing by Cue as proposed by Steriade (1997) claims that featural contrast is dependent on phonetic implementation factors, i.e. gestural timing, and the perceptibility of cues. A contrast will only occur in contexts where the cues to a feature are optimally perceptible. In this way, phonetics directly influences the phonological distribution of features, not only of underlying segments, but of segments derived by morphology as well. Steriade goes so far as 53 to say that cues 'determine' the distribution of featural morphemes. Since the basis of the theory lies in phonetic cues, the first step taken was to examine the phonetic evidence. The following preliminary timing patterns emerged in Sancaean: Non-derived Derived pre-stress post-glottalized n/a post-stress pre-glottalized pre-glottalized post-stress, pre-C post-glottalized n/a word final glottalized throughout glottalized throughout Armed with this phonetic data, and descriptions of glottal timing events in St'at'imcets, Licensing by Cue was applied to the distribution of underlying glottalized resonants. It was shown that in Sancaean Licensing by Cue had to be modified to be sensitive to stress, and also that cues to glottalization were in competition with cues to place/manner. Since stress can act as an enhancer of phonetic cues, it is unsurprising that glottalization is attracted to the stressed vowel, since this is most optimal context for perceptibility. While interaction with place/manner cues was unexpected, it is certainly compatible with a cue-based analysis. Licensing by Cue, slightly modified, was able to account for the timing distribution of Sancaean. With regard to glottalized segments, though, it failed to rule out word initial glottalized resonants, on the basis of cues. The versions of Licensing by Cue proposed by Steriade accounts for glottal timing in St'at'imcets with much less modification. In fact, only a reranking of Context Cues is required. This despite the fact that it seemed to pose such a problem at first. The timing of glottal events in St'at'imcets glottalized resonants is also determined by stress, but in a very different way to Sancaean. In St'at'imcets, transition to or from the stressed vowel is the least optimal context for cues. This was proposed to be a result of competition of cues between glottalization and stress. The proposed analysis simply reranked context cues so that glottalization adjacent to a stressed vowel was the optimal context. This analysis accounted for the timing of glottal events, and also for the possibility of post-consonantal and word-initial glottalized resonants. It could not however, explain why while glottalized resonants are permissible in word-initial position, they occur there only as the result of some reduplication processes. This strongly suggests that it is the interaction of phonetic, phonological and morphological constraints that determines the distribution of glottalized resonants in Sancaean and St'at'imcets; phonetic constraints determine timing, phonological constraints (such as *CR, and *CR#), and morphological constraints (such as the reduplicative constraints in St'at'imcets, and the imperfective constraints in both Sancaean and St'at'imcets) determine distribution. While it may turn out that cue-based constraints do rule out word-initial glottalized resonants, much more research in terms of the roles played by individual cues must be pursued. 54 The analysis proposed in order to account for both the Sancaean and the St'at'imcets patterns raises the question of how much variation in timing and distribution is possible. Licensing by Cue and OT failed to account for the differences between the two languages with as much ease as might be expected of them, and this analysis implies that an almost infinite number of patterns is possible. In answer to how much variation may exist cross-linguistically, consider the timing descriptions for the three Salish languages studied so far. Recall that Flemming, Ladefoged and Maddieson found all glottalized resonants in Montana Salish to be pre-glottalized. Sancaean and St'at'imcets glottal timing is determined by stress, but Sancaean glottalization is attracted to stress, whereas St'at'imcets is repulsed. Given that this much variation occurs in related languages, cross-linguistic variation indeed seems boundless. However, it seems to the author that since there is a finite number of cues, and a finite number of contexts, with enough research patterns will begin to emerge. Further forays into the research of glottalized resonants must first establish what the cues to glottalization in each language are, before linking timing and distribution. In order for Licensing by Cue to be considered a comprehensive theory, it should also be able to account for the distribution of derived glottalized resonants, since they too will have phonetic cues to glottalization. If the cues are the same, the distribution should be the same. Considering the phonetic data from Sancaean again, recall that derived post-stress glottalized resonants were indeed pre-glottalized as expected, However, it appeared that one of the cues to glottalization in non-derived resonants, namely more energy in the higher frequencies, is not a cue to glottalization in derived resonants. It may be that once we are able to break up Context Cues into its individual cues, that we can rank them with respect to 'importance' in a particular language. It may turn out that one particular cue may be more salient as a cue to glottalization than others, and that the unrecoverability of a less important cue will still be perceived as a glottalized resonant by the listener. Perhaps it is the lack of salience of the most salient cues to glottalization that rules glottalized resonants out word-initially. Since they do occur there in some contexts in St'at'imcets, they are obviously physiologically possible to produce in this position. Obviously much more research needs to be done to discover which cues are reliable, and whether or not there really is a difference in cues between derived and non-derived glottalized resonants, and what impact this has on distribution. The vehicle used to test whether Licensing by Cue determines the distribution of segments glottalized morphologically is the Proto-Salish imperfective morpheme. It was shown in this thesis that the morpheme representing the 'actual' in Sancaean and the 'inchoative' in St'at'imcets are indeed related. While featural alignment constraints rather than phonetic constraints were shown to determine the distribution of resonant glottalization in these two languages, Licensing by Cue is not incompatible 55 with such an analysis. It is simply not the motivating factor that Steriade (1997) claims it to be. By positing that the Proto-Salish morpheme for the imperfective is a floating feature, be it [creak] or [eg], which suffixes to the stressed vowel, and spreads rightward, this analysis is able to account for four allomorphs of the actual in Sancaean and one allomorph of the inchoative in St'at'imcets with one generalisation. It also supports the OT claim that related languages are governed by the same constraints. Since OT does so well for phonological variation, but so poorly for phonetic variation, the question that must be asked is OT restricted to phonological variation, or are the constraints relating to phonetics simply not yet well enough defined to give a unified account? Another questions raised by the analysis of imperfective is why glottalization accompanies each allomorph of the Sancaean actual, but only the allomorph used with full vowels in St'at'imcets. This remains a topic for further research. Finally, the issue of the relationship between /?/ and glottalized resonants, and which feature should be used to characterise both segments brings us back to the role of phonetics in phonology. Under Licensing by Cue, ejectives and glottalized resonants are given different features because they are phonetically different. In Sancaean it was shown that /?/ and glottalized resonants also behave differently. Should they also have their phonetic differences characterised by separate features? An alternative analysis based on faithfulness and markedness constraints can account for the distribution of glottalized segments equally well, but begs the same question. The research in this thesis brings up the larger issue of the implications that the Licensing by Cue hypothesis makes for language processing and cognitive functions. By proposing that constraints imposed on us by our physical limitations, both productive and perceptive, are active in the phonology, Licensing by Cue removes some of the control normally attributed to U G . This takes the onus off of abstract principles, and also off of the child to learn them. The Sancaean and St'at'imcets data presented in this thesis do not offer strong support for the above hypothesis. While the timing was shown to be determined by phonetic cue-based constraints, the actual distribution of segments is governed by morphological and phonological constraints. The motivation of the one phonetically grounded constraint, * [-son]/creak, is questionable, and could be replaced by a set of markedness-as-faithfulness constraints. A strong phonetic presence in phonology is not supported by the data here, but a better understanding of the importance of individual cues to glottalization may shed more light on this fascinating issue. 56 References: Akinlabi, A . 1996. Featural Affixation, in Journal of Linguistics 32, 239-289. Archangeli, D. and D. Pulleyblank. 1994.Grounded Phonology. Cambridge: MIT Press Blake, S. 1999. Toward an Analysis of Schwa in Sliammon in Papers for the 34 t h International Conference on Salish and Neighboring Languages. Secwepemc Cultural Education Society/SFU --. 1995. 'Glottalized Resonants in Sliammon (Salish),' paper presented at the Canadian Linguistics Association Meeting, U Q A M , Montreal, Quebec. —.1992. Two Aspects of Sliammon Phonology: Glide/Obstruetn Alternationand Vowel Length. M A Thesis, University of British Columbia, Vancouver. Caldecott, M . 1998a. Applying Licensing by Cue to Saanich. U B C Working Papers in Linguistics 1. Flemming, E., P. Ladefoged, and S. Thomason. 1994. Phonetic Structures of Montana Salish. Fieldwork Studies of Targeted Languages II. Los Angeles: U C L A Working Papers in Phonetics 87. Howe, D. and D. Pulleyblank. 1999. Patterns of Glottalization in Wakashan. Paper presented at Glow 22 Kingston, J. 1985. Phonetics and Phonology of the timing of oral and glottal events. UCBerkeley PhD. dissertation —. 1990. Articulatory Binding, pp. 406-434 in J. Kingston and M . Beckman (eds.) Papers in Laboratory Phonology, Between the Grammar and Physics oj Speech_pp. 406-434, Cambridge University Press. Kinkade, D. (1997) Reconstructing aspect in Salishan languages, in Papers for the 32 n d International Conference on Salish and Neighbouring Languages Ladefoged P. 1993. A Course in Phonetics Harcourt Brace: Fort Worth Ladefoged, P. and Maddieson, I. 1996. Sounds of the World's Languages. Blackwell. McCarthy and Prince, Maddieson. I. 1984. Patterns of Sound. Cambridge: Cambridge University Press Montler, T. 1991. Saanich, North Straits Salish Classified Word List. Canadian Ethnology Service 119. —. 1989. Infixation, reduplication, and metathesis in the Saanich actual aspect, in Southwest Journal of Linguistics 9:1. 92-107. —. 1986. An Outline of the Morphology and Phonology of Saanich, North Straits Salish .University of Montana Occasional Working Papers in Linguistics 4. Prince A . and P. Smolensky. 1993. Optimality Theory. Constraint Interaction in Generative Grammar. Technical Report #2 of the Rutgers Center for Cognitive Science, Rutgers University. 57 Plauche. M . , R. de Azcona, R. Roengpitva, W. Wiegel. (1999). Glottalized sonorants: A phonetic universal?. U C Berkeley, ms. Silverman, D. 1997. Phasing and Recoverability. Garland: New York Shaw, P (1987) The Complex status of complex segments in Dakota. In Gerdts,D and Michelson, K (eds.) Theoretical Perspectives on Native American Languages. State University of New York Press. Shaw. P (1998) Escher effects in Morphology. C L A / A C L handout. Steriade. D 1997. Phonetics in Phonology: The Case of Laryngeal Neutralization. Ms. U C L A --. 1995. Positional neutralization. Ms. U C L A Stonham, J. 1994 Combinatorial Morphology. John Benjamins:amsterdam van Eijk, J. (1997). The Lillooet Language. Vancouver: U B C Press 58 Appendix 5 9 60 61 63 2Sa.fiterr 7902 1523 23-a t'itenvspc 2756.2 23.0 JtT |.iAfC| 7902 |1 1523 H C * t -23s.t' i l*m 7902 11)2 i52s _gu 2756.2 20.3 _T_g L.Wb'1 29.V«t"iT«ft' 10690 347 H C + t -2756.2 43.5 29feVi1>m-10690 347 •< j JIP llfiA.I 29. fe f iT«m' 10426 2431 H C JIT liflft' 29 . f e f il'em 1 1754 -623 JIT 2 2 0 5 0 H ; mwv^  d 0.00 ItT 434.1 FIGURE 5.1 700 ms A . 4son Hz • « iMwwimtmmnttni i <nirTinwwi>^'^.ui)''11 * * t *» •if WwWHU A 500 1500 2000 2500 3000 3500 4500 Hz 600 900 ms 500 1 500 2000 2500 3000 I. I--6160 400 ms J 5360 500 ms 64 65 66 67 


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items