University of Southern California Dissertations and Theses

The effect of stress and foot structure on consonantal processes

(USC Thesis Other)

The effect of stress and foot structure on consonantal processes

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content THE EFFECT OF STRESS AND FOOT STRUCTURE ON CONSONANTAL PROCESSES by Carolina Gonzalez A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (LINGUISTICS) May 2003 Copyright 2003 Carolina Gonzalez Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 3103894 UMI UMI Microform 3103894 Copyright 2003 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UNIVERSITY OF SOUTHERN CALIFORNIA The Graduate School University Park LOS ANGELES, CALIFORNIA 90089-1695 This dissertation, w ritte n b y C&RPUN ix 6 y 3 A l i - ' 2 Under th e direction o f h.& t. D issertation Com m ittee, an d approved b y a ll its m em bers, has been p resen ted to and accepted b y The Graduate School, in p a rtia l fulfillm ent o f requirem ents fo r th e degree o f DOCTOR OF PHILOSOPHY D ate May 16, 2003 DISSERT A T IO N CO M M ITTEE Chairperson / Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ii DEDICATION I dedicate this thesis to my parents Manuel y Maria Julia, and to my siblings Veronica, Iratxe, Gemma, Julia Maria and Imanol. Dedico esta tesis a mis padres Manuel y Maria Julia, y a mis hermanos Veronica, Iratxe, Gemma, Julia Maria e Imanol. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ACKNOWLEDGMENTS I am profoundly indebted to all of the members of my committee-—Rachel Walker, Kie Zuraw, Dani Byrd, Mario Saltarelli and Abigail Kaun—for their comments, help and encouragement while I was researching and writing this dissertation. Rachel Walker inspired me to become a phonologist and has been an excellent advisor. I am extremely thankful for her discussion and patient revisions of the material in this dissertation. I was very lucky to have Kie Zuraw in my committee. Both Rachel Walker and Kie Zuraw’s insightful comments have challenged and clarified the arguments and presentation of the data in this dissertation. Dany Byrd is responsible for my interest in phonetics and phonetic analysis. She always supported me in these endeavors, for which I am extremely thankful. Mario Saltarelli always saw the bright side of this dissertation and was unfailingly supportive. And Abigail Kaun always made time to talk about the matters discussed here. To all of you my greatest thanks. Outside my committee members I want to thank Jon Franco, who directed me to linguistics and to USC. Even if I am not a syntactician he has remained a great friend who has always continued to believe in me. I was very lucky to have such great friends and colleagues in the Linguistics Department and the Spanish and Portuguese Department at USC. I want to specially thank Hatim Hassan, who from the start has been an unconditional friend and an Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. excellent resource in everything electronic and bibliographical. I also want to thank Asier Alcazar, Monica Cabrera, Susana Cid-Hazard, Suzanne Curtin, Guanjun (Bella) Feng, Teruhiko Fukaya, Shadi Ganjavi, Narineh Hacopian, Fetiye Karabay, Nihan Ketrez, Jelena Krikovap, Uffe Larsen, Agnieszka Lazorczyk, Miae Lee, Mina Lee, YoussefNouhi, Omar Obeas, Isabelle Roy and Milena Petrova. Thanks also to the rest of the faculty and to the staff at the Linguistics Department. Laura Reiter, Linda Culver and Winona Colinco were always ready to help in anything they could, and I really missed them during my last year at USC. Pursuing graduate studies in LA has been enjoyable thanks to a number of people not necessarily related to linguistics. Thanks to Leonel Casasola for everything. Thanks to Zurine Lekuona for her tolerance for (occasional) disorder and for her continued support. Thanks also to my friends Cecilia Choi, Keya Koul, Alvaro Molina and Manuel Sanchez-de-Cima. I also want to thank Erik Bacovic, Beto Elias, Sharon Inkelas and Caro Struijke, and audiences at the ASA, LSA, NELS, USC Phon Lunch and SWOT for discussion on my work at different stages and/or pointing me to relevant languages. Special thanks go to Eugene Loos for discussion on Capanahua, and to all the field workers and linguists who made possible the description and documentation of the languages that I discuss here. This dissertation was completed with the generous support of the College of Letters, Arts and Sciences at USC, and with a fellowship from the Basque Government during 2000-2002. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. V TABLE OF CONTENTS Dedication ii Acknowledgements iii List of tables xiii LIST O F FIGURES XV ABSTRACT xvi Chapter 1: Introduction 1 1. Scope of the survey 2 2. Main findings 4 3. Main theoretical proposals 9 3. I Factors underlying stress- and foot-sensitive consonantal alternations 9 3. 2 Stress and foot structure 13 3. 3 Analysis of stress- and foot-sensitive consonantal phenomena 20 4. Case studies 22 5. Theoretical framework 23 5. 1 Optimality Theory 23 5. 2 Stress and metrical theory 26 5.3 The relationship between phonetics and phonology 30 6. Organization 34 Chapter 2: Prosodically-influenced consonantal processes: A 36 SURVEY 1. Introduction 36 2. Sources for the survey 37 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. vi 2.1 Lenition/fortition 37 2.2 Metathesis 38 2.3 Laryngeal phenomena and attraction phenomena 39 2.4 This survey 39 3. Survey overview 41 4. Duration/magnitude differences 45 4.1 Durational differences 45 Spanish 47 English 47 Senoufo 47 Maori 47 Greek 47 Turkish 47 Urubu-Kaapor 48 Copala-Trique 48 Guelavia Zapotec 49 Popoloca 51 Alutiik Yupik 52 Norton Sound Yupik 53 Gundidj 53 Narinari 54 Wergaia 54 Italian 55 Somali 57 Summary 58 4.2 Flapping 59 American and Canadian English 59 Djabugai 61 Senoufo 62 Kupia 62 Summary 62 4.3 Fortition 63 Norton Sound Yupik 64 Alutiik Yupik 65 Guayabero 67 Maori 70 West Tarangan 71 Farsi 72 Squamish 72 Yuman 72 Summary 74 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. vii 4.4 Lenition 74 English 75 Copala Trique 76 Guayabero 77 Senoufo 77 Somali 78 Spanish 79 Kupia 80 Nganasan 81 Paamese 83 Silacayoapan Mixteco 85 Summary 86 4.5 Secondary articulation 86 Guayabero 86 Senoufo 87 Summary 87 5. Featural timing 87 5.1 Voicing alternations 88 Senoufo 89 Popoloca 89 Djabugai 90 W embawemba 91 Wergaia 91 Farsi 92 Copala Trique 93 Silacayoapan Mixteco 93 Wasco-Wisram 94 Urubu-Kaapor 94 Proto-Germanic 95 Middle English 96 Paamese 96 English 97 Summary 97 5.2 Aspiration alternations 98 5.2.1 Post-aspiration 98 English 98 Maori 99 German 99 Pattani 99 Farsi 100 Chali (Tati) 100 Silacayoapan Mixteco 100 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. viii Squamish 100 Basque Spanish 101 Summary 101 5.2.2 Pre-aspiration 101 Icelandic 102 Standard Faroese 103 Northern F aroese 103 Scots Gaelic 104 Irish Gaelic 105 Ingush 105 Toreva Hopi 105 Ojibwa 106 Tarascan 106 Summary 106 5.3 Glottalization alternations 107 Coast Tsimshian 108 Gitksan 108 Saanich 109 Lilloet 109 Summary 109 6. Attraction 110 6.1 Consonantal attraction 110 Shuswap 111 Twana 112 Thompson River Salish 112 Coast T shimshian 112 Danish 113 Colville 113 Bagneres-de-Luchon French 113 Summary 114 6.2 Stress attraction 114 Aranda 115 Alyawarra 115 Piraha 116 Banawa 116 Iowa-Oto 117 Summary 117 7. Deletion and epenthesis 117 7.1 Deletion 117 English 118 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ix Chali (Tati) 119 Squamish 120 Oneida 121 Popoloca 122 Capanahua 122 Faroese 123 Chilean Spanish 124 Old-English 124 Romance 124 Summary 125 7.2 Epenthesis 125 English 125 German 126 Dutch 127 Paipai 128 Walapai 128 Silacayoapan Mixteco 128 Huariapano 129 Popoloca 129 Summary 129 8. Metathesis 130 Cayuga 130 Mohawk 132 Oneida 133 Thompson River Salish 133 Capanahua 134 Faroese 135 Le Havre French 136 Summary 137 9. Dissimilation 137 Gothic 138 Romance 138 Summary 139 Chapter 3: Typological Generalizations and their Underpinnings 140 1. Introduction 140 2. Typological generalizations (I): stress and foot structure 142 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. X 3. Typological generalizations (II): syllabic, word and morphemic position; 145 features and segments affected 3.1 Syllabic positions 145 3.2 Features and segments affected 146 3.3 Word position, phrase and utterance position, and morphemic 148 structure 3.3.1 Word position and stress 149 3.3.2 Phrase/utterance position and stress 152 3.3.3 Morphemic structure and stress 153 4. The role of perception, aerodynamics, duration and prominence 154 4.1 The role of perception 157 4.2 The role of aerodynamics 161 4.3 The role of duration 164 4.4 The role of prominence 165 4.5 Remaining consonantal alternations 170 4.6 Summary 171 5. Predictions: consonantal alternations that will be unattested 173 6. Phonetics and phonology 174 7. The separation between stress and footing 176 7.1 Strictly stress-sensitive consonantal phenomena 177 7.2 Stress/foot-structure cases 178 7.3 Strictly foot-sensitive consonantal phenomena 178 7.4 Summary 186 8. Rhythmicity 187 9. Accounting for prominence-based consonantal alternations 189 9.1 Previous analyses 190 9.1.1 Augmentation in strong positions 190 9.1.2 Prominence Alignment 192 9.2 Prominence Alignment and stress-sensitive consonantal 194 prominence 9.3 Prominence Alignment and foot-sensitive consonantal 197 prominence 9.4 Summary 202 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. xi 10. Accounting for duration-based consonantal alternations 204 11. Accounting for aerodynamic-based consonantal alternations 206 12. Accounting for perception-based consonantal alternations 208 13. Conclusion 212 Chapter 4: A erodynamics in N orth-Central Peninsular Spanish 214 1. Introduction 214 2. Spirantization in Peninsular Spanish 218 3. An Optimality Theory analysis 220 3.1 Allophonic variation between [b, d, g] and [g, 6, y] 221 3.2 Phonemic distribution of /b, d, g/ and /p, t, k/ 224 3.3 The problem of [Id] 227 4. Frication and devoicing of coda lb, d, g/ in NC Spanish 231 4.1 Coda variation in Spanish dialects 232 4.2 Acoustic study: the influence of stress in coda /b, d, g/ in NC 233 Spanish 4.3 Phonetic and phonological implications 245 5. An OT analysis of coda frication of Po, d, g/ in Basque Spanish 247 6. Conclusion 251 Chapter 5: Rhythmicity in Panoan 254 1. Introduction 254 2. The phonology of Reconstructed Panoan 258 3. Capanahua 260 3.1 The phonology of Capanahua 260 3.2 The metrical system of Capanahua 263 3.3 The distribution of /?/ 282 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. xii 3.4 [?] as an allophone of /tJ7 294 3.5 Exceptional cases 299 3.6 Conclusion 303 4. Huariapano 304 4.1 Huariapano phonology 306 4.2 Coda [h] epenthesis 314 4.3 Proposal 318 4.4 The relationship between secondary stress and [h] epenthesis 325 5. Shipibo 330 5.1 The metrical system of Shipibo 330 5.2 Rhythmic alternations 333 6. Conclusion 336 Chapter 6: Concluding remarks 339 1. Stress and footing in consonantal processes 339 2. Further issues 344 2.1 Are foot heads prominent categories? 344 2.2 Consonantal vs. vocalic alternations 345 2.3 The asymmetry between trochaic and iambic systems revisited 346 2.4 The ambiguity of stress and word position 347 2.5 Unfooted syllables 347 2.6 The ‘after’ pattern’ 348 References 350 Appendix 371 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. xiii LIST OF TABLES Table 2.1. Durational differences 46 Table 2.2. Flapping 59 Table 2.3. Fortition 64 Table 2.4. Lenition 75 Table 2.5. Secondary articulation 86 Table 2.6. Voicing variation 88 Table 2.7. Post-aspiration 98 Table 2.8. Pre-aspiration 102 Table 2.9. Glottalization 108 Table 2.10. Consonantal attraction 110 Table 2.11. Stress attraction 115 Table 2.12. Deletion 118 Table 2.13. Epenthesis 125 Table 2.14. Metathesis 130 Table 2.15. Dissimilation 137 Table 3.1. Consonantal alternations involving codas 146 Table 3.2. Factors in stress- and foot-sensitive consonantal phenomena 172 Table 4.1. List of nonce words (underline corresponds to target consonant) 235 Table 4.2. Stress: main values for dependent variables (by speaker) 239 Table 4.3. Realization of /b, d, g/ by speaker 240 Table 4.4. Mean values for frication (speakers pooled) 241 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. XIV Table 4.5. Mean values for voicing (speakers pooled) 242 Table 4.6. Mean values for voicing duration (speakers pooled) 242 Table 4.7. Mean values for percent voicing (speakers pooled) 242 Table 5.1. Consonant chart of Reconstructed Panoan (IPA; from Shell 259 1975) Table 5.2 Capanahua consonant chart (IPA; based on Loos 1969) 261 Table 5.3 Huariapano consonant chart (IPA: from Parker 1994:96) 306 Table 5.4 Metrical system of Huariapano 325 Table 5.5 Metrical system in Panoan 326 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. X V LIST OF FIGURES Figure 1.1 Stress and footing in onsets of trochaic and iambic systems 18 Figure 2.1 Stress and footing in onsets of trochaic and iambic systems 66 (repeated from Figure 1.1) Figure 3.1 Phonetics and phonology in consonantal phenomena 175 Figure 3.2 Iambic footing and stress 179 Figure 3.3 Predictions for foot-based phenomena 203 Figure 4.1 Spectrogram. Voiced fricative /d/. Sequence [10] from phrase 237 ‘Lambid Bote’ Figure 4.2 Spectrogram. Voiced approximant /d/. Sequence [lu5ba] from 237 phrase ‘LudbaBote’ Figure 4.3 Spectrogram 3. Voiced approximant /d/. Sequence [la5|3an] from 238 phrase ‘Ladban Botella’ Figure 4.4 Hypothesized interaction between stress, frication and 243 constriction degree Figure 5.1 Allophones of /tjV and /?/ in Capanahua 299 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. xvi ABSTRACT A cross-linguistic investigation of stress- and foot-sensitive consonantal processes shows that stress and foot structure are separate entities, and that both are needed to characterize the range of consonantal phenomena attested. This study isolates some cases where consonantal alternations are purely stress-sensitive and others where it is purely foot-sensitive. Among the former are consonantal phenomena where foot structure is either absent or irrelevant. These are usually aerodynamically- or durationally-based. One example is Northern Peninsular Spanish, where frication of coda /b, d, g/ is more likely in stressed syllables. Strictly foot-sensitive consonantal processes where stress is absent or irrelevant fall into three groups: (i) Systems where not all feet are stressed, but where consonantal alternations occur in a rhythmic fashion, as in Capanahua; (ii) Iambic systems, as in Norton Sound Yupik, where fortition coincides with foot-initial syllables, which are unstressed; and (iii) Systems with a mismatch between stress and footing, as in Huariapano, where foot- sensitive [h] epenthesis conflicts with stress in various cases. Strictly foot-sensitive consonantal processes are mostly rhythmic. Perceptual processes can be factors in both strictly stress-sensitive and strictly foot-sensitive consonantal phenomena. In a third group of consonantal phenomena stress and foot structure coincide, causing an ambiguity as to whether stress and/or footing condition the process. This typically occurs in trochaic systems with persistent footing. One example is flapping and aspiration in American English. The cross-linguistic survey is complemented by Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. xvii two case studies, one on a stress-driven phenomenon in North-Central Peninsular Spanish, and one on foot-driven consonantal phenomena in Panoan. The fact that strictly stress-sensitive and strictly foot-sensitive consonantal processes are attested suggests that stress and footing are separately needed in order to correctly characterize the range of attested consonantal processes. It is proposed that consonantal alternations are a possible way to organize syllables into feet, and that prominence alternations need not be stress-based only. The independence of stress-sensitive and foot-sensitive consonantal processes argues for the inclusion of foot-sensitive constraints conditioning consonantal alternations in the grammar. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1 Chapter 1: Introduction Stress and foot structure condition a number of vowel phenomena. Vowels tend to reduce in unstressed syllables (Delattre (1969), Fourakis (1990), Koopmans-van Beinum (1980), Crosswhite (to appear)) and lengthen in stressed syllables (Goldsmith (1990), Hayes (1995), Smith (2002)). Vowel assimilation, metathesis and dissimilation can also be foot sensitive (Flemming (1994), Blevins and Garrett (1998), Suzuki (1998)). A growing body of research shows that stress and foot structure condition a number of consonantal phenomena. Some examples include foot-sensitive fortition in Yupik (Van de Vijver (1998)) and foot-sensitive coda fhJ epenthesis in Huariapano (Parker (1994, 1998)). The influence of stress and foot structure has been considered for specific types of consonantal processes, most notably lenition (Lavoie (1996, 2001)), perceptually-sensitive timing of glottalization (Steriade (1997)) and metathesis (Blevins and Garrett (1998)). However, an extensive cross-linguistic investigation of stress- and foot-sensitive consonantal processes is lacking. This dissertation aims to fill this gap. It focuses on consonantal phenomena influenced by stress and foot structure. It examines the types of consonantal phenomena that are influenced by stress and foot structure and explores the possible factors that underlie them. It makes a number of generalizations concerning these processes, including the specific stress and footing environments in which consonantal phenomena are conditioned cross-linguistically, the types of segments Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2 and syllabic positions that are preferentially targeted, and the relevance of positional factors for specific phenomena. By doing so, this study contributes to the understanding of the possible manners in which stress and footing interact with segmental processes, and, more generally, to the relationship between stress and foot structure. This chapter describes the scope of the cross-linguistic survey of stress- and foot-sensitive consonantal phenomena that is reported in chapter 2 (section 1). It also previews the main findings that emerge from this survey (section 2). The main theoretical proposals of this study are outlined in section 3. Section 4 describes in some detail the case studies included in this dissertation. Finally, section 5 lays out the theoretical framework, and section 6 the organization of the dissertation. 1. Scope of the survey The backbone of this dissertation is a survey of over 74 languages from 36 language families which show stress- and foot-sensitive consonantal alternations. This survey includes both segmental and featural consonantal processes and phonetic (variable) and phonological (categorical) instances of consonantal alternation. Excluded from this survey are contrast-based processes whereby phonological contrasts are maintained in stressed syllables and neutralized in unstressed syllables. These cases have been examined in Beckman (1997, 1998) and analyzed as instances of positional faithfulness, e.g., feature identity to the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. underlying featural specification of strong positions for perceptual and psycholinguistic reasons (Beckman (1997, 1998)). Beckman (1997, 1998) suggests that prominent positions keep the underlying specifications o f the segments more often than non-prominent positions. This would explain that prominent positions have a larger number o f contrasts than non- prominent positions, and that they trigger and resist more phonological processes. In this approach, faithfulness to segmental specification is respected preferentially in prominent positions. This is formalized by means of IDENT/Position (F) constraints (henceforth (Ident/P). Ident/P constraints are faithfulness constraints relativized to prominent positions. Faithfulness to prominent positions is captured through the ranking o f Ident/P constraints over markedness and general faithfulness constraints (Beckman (1998)). This dissertation focuses on consonantal alternations that are sensitive to stress and/or foot structure and cannot be explained through the notion o f contrast. The range of consonantal phenomena examined in this dissertation is shown in (1). (1) Consonantal phenomena surveyed - Consonant lengthening - Consonant reduction - Gemination - Lenition - Fortition - Voicing alternations - Variable laryngeal timing - Consonantal attraction - Stress attraction - Deletion - Epenthesis - Secondary articulation - Metathesis - Dissimilation Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Details on the background and development of the survey are discussed in chapter 2. The next section previews the main findings obtained from this survey. 2. Main findings The main findings concerning the consonantal phenomena in (1) and their relationship to stress and/or footing are summarized in (2). These findings are organized according to the relevant position: stressed or strong-footed position, unstressed/weak-footed position, and all positions. (2) Findings (I) Stressed/strong-footed position - Fortition occurs in stressed syllables and in foot-initial syllables. - Consonantal features and segments are attracted to stressed syllables. Stress is attracted by onsetful syllables and by low-sonority onsets. - Consonantal epenthesis occurs in stressed syllables and in strong-footed syllables. - In laryngealized consonants, the realization of glottalization and aspiration is timed according to stress: glottalization and aspiration will be realized as close to a stressed vowel as possible. In stressed syllables, voiceless consonants become voiced, but voiced consonants do not become voiceless. - Gemination, pre-aspiration and pre-glottalization occur after stressed syllables. Unstressed/weak-footed position - Lenition occurs in unstressed syllables and in foot-medial syllables. - Flapping occurs word-medially in unstressed syllables. - Consonantal deletion occurs in unstressed syllables and in weak-footed syllables. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5 In unstressed syllables, voiced consonants are devoiced and voiceless consonants are voiced. All positions Consonants are longer in stressed syllables than in unstressed syllables. - VOT for obstruents is longer in stressed syllables than in unstressed syllables. - Metathesis and dissimilation can occur in stressed syllables or in unstressed syllables. An important finding is that there are three types of prosodically-influenced consonantal processes, according to whether stress and/or foot structure influence them (3). (3) Findings (II): Types of consonantal phenomena - Stress only Consonantal processes can be strictly influenced by stress in the absence of foot structure (as in frication in North-Central Peninsular Spanish). Stress and/or foot structure Consonantal processes where it is ambiguous whether stress and/or footing are conditioners since both coincide (as in flapping in American English). - Foot structure only Consonantal processes can be strictly influenced by foot structure in the absence of stress (as in deletion in Capanahua). Note that some languages are inconclusive as to whether a consonantal process is strictly stress-motivated, or whether stress and footing coincide in motivating it, since foot structure has not been studied in all of the languages gathered in the survey. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 6 A third important finding is that there are four factors underlying the stress- and foot-sensitive consonantal alternations examined in this survey (4). These factors are defined and discussed in some detail in section (3) in this chapter. (4) Findings (III): Factors - Perception Duration - Aerodynamics - Prominence Furthermore, prosodically-conditioned consonantal processes can be optional or categorical (5), and they frequently target onsets and laryngeal features or segments (6). It will be argued that the predominant targeting of onsets in stress- and foot- sensitive consonantal alternations is probably not significant due to the fact that many languages have restrictions on codas or no codas at all. (5) Findings (IV) Stress and foot structure can condition consonantal processes optionally (Example: voicing in Urubu-Kaapor). Stress and foot structure can condition consonantal processes categorically (Example: fortition in Norton Sound Yupik). (6) Findings (V) - Most consonantal phenomena influenced by stress or foot structure are reported to target onsets. - A large number of the consonantal phenomena surveyed involve laryngeal features and consonants. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 7 The majority of the consonantal alternations in this dissertation involve laryngeal features and consonants. Following work by Lombardi (1991) and others I take laryngeal features to include glottalization, aspiration and voicing. From the consonantal phenomena surveyed, laryngeal features are involved in variable glottalization and aspiration timing, long-distance attraction, metathesis, and voicing alternations. Laryngeal consonants (/h/, /?/) are involved in deletion, epenthesis, metathesis and some types of lenition. It is expected that deletion, epenthesis and relocation target (/h/, /?/) more often than segments with supra-laryngeal articulations. The reason is that laryngeals have no place of articulation—unless they co-occur with pharyngeal or uvular consonants in a language, in which case they are pharyngeal (Rose (1996)). Evidence for lack of place of articulation is provided by their cross-linguistic phonological behavior. Laryngeals often behave phonologically as if they were placeless in terms of vowel copying, debuccalization, epenthesis, reduplication and other processes (e.g. Steriade (1987), Yip (1991), Stemberger (1993), McCarthy and Prince (1994), Halle (1995), Rose (1996), Ladefoged (1997), Parker (2001)). Another consideration is the close connection between laryngeal features and segments and stress. Variations of suprasegmental features such as pitch and stress are partly due to changes in laryngeal behavior (Ladefoged (1993)); it is reasonable that laryngeals are mostly influenced by stress. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Additionally, laryngeal features are expected to be connected to stress since changes in laryngeal behavior are responsible for variations in pitch due to stress (Ladefoged (1967)). Finally, it is found that stress and footing are reported to interact with word position in some cases. In a few occasions, stress- and footing are reported to interact with morphemic structure, phrase position and utterance position. The main findings with respect to the interaction of stress and footing with word, morphemic, phrase and utterance position are summarized in (7). (7) Findings (VI) Word position - Consonantal phenomena are more likely or obligatory word-initially under stress {example: epenthesis in Paipai). - Consonantal phenomena tend to occur both in stressed syllables and word- initially regardless of stress (example: post-glottalization in Gitksan). Consonantal alternations occur in word-final post-tonic syllables {example: epenthesis in Mohawk). - Consonantal alternations tend to occur both in unstressed syllables and word- finally regardless of stress {example: English flapping). - Consonantal alternations in word-medial position occur in unstressed syllables or after stressed syllables {example: flapping in Djabugai). Phrase and utterance position - Phrase and utterance position are rarely reported to interact with stress and foot structure. - Weakening and epenthesis phenomena sensitive to stress are occasionally blocked utterance or phrase-initially {example: lenition in Spanish). Morphemic structure - Morphemic structure is rarely reported to interact with stress and foot structure. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 9 Sometimes stress- or foot-sensitive consonantal processes occur in specific morphemes in the word (e.g., stressed roots in Shuswap attract glottalization). Sometimes a process might occur both in a specific morpheme position and in stressed syllables (example: epenthesis in German). 3. Main theoretical proposals This dissertation puts forward three main theoretical proposals relating to the relationship between stress/footing and consonantal phenomena. These theoretical proposals concern the factors behind the consonantal alternations surveyed (3.1), the nature of the relationship between stress and foot structure (3.2), and the analysis of both stress- and foot-sensitive consonantal phenomena in Optimality Theoretic terms (3.3). 3.1 Factor underlying stress- and foot-sensitive consonantal alternations I propose that four factors are at work in stress- and foot-sensitive consonantal phenomena: perception, duration, aerodynamics, and prominence. These are defined below. • Perception Perceptual factors can drive stress- and foot-sensitive consonantal phenomena for functional purposes (note that contrast-based alternations were excluded from this survey). A process can occur so that features or consonants are better perceived. Two examples are metathesis in Faroese and feature attraction in Shuswap. In Faroese, the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 10 sequence /skt/ is realized as /kst/ after a stressed vowel, with /k/ and /s/ undergoing metathesis (Hume (2000) and references therein, Hume and Seo (2001)). In Shuswap, glottalization is attracted to the final sonorant of a stressed syllable (Steriade (1997) and references therein). In both cases, the realization of a consonant or a feature closer to a stressed vowel improves the perception of the consonant or feature. Another set of perceptually-motivated processes includes cases where perceptually salient features or consonants are avoided in ‘weak’ positions. This is exemplified by certain instances of deletion. For example, in Faroese, fkl in the sequence /skt/ is deleted after unstressed vowels (Hume (2000) and references therein; Hume and Seo (2001)). /k/ deletes because its cues will not be perceptually salient in a cluster in unstressed position. Finally, perceptual ambiguity in unstressed and/or weak syllables can also drive certain cases of metathesis both synchronically (as in Thompson River Salish) and diachronically (as in Cayuga) (Blevins and Garrett (1998)). For example, in Cayuga /h/ and /?/ underwent onset-to-coda metathesis in weak syllables. Since vowels are at least partially globalized or devoiced next to laryngeal consonants, there is certain ambiguity as to the origin of the laryngeal consonant, which might lead to the reinterpretation of the original position of the laryngeal consonant to the coda (Blevins and Garrett (1998) and references therein). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 11 • Duration Durational considerations resulting from stress might cause consonantal alternations. Since stressed syllables generally involve increased duration, the available extra duration of the stressed syllable might lead to a longer consonant, with or without a longer vowel in this position. One example is onset lengthening in stressed syllables in Urubu-Kaapor (Kakamasu (1986)). In the same way, unstressed syllables involve reduced duration, which might translate into reduced consonant duration in various languages. • Aerodynamics The aerodynamics of stress can drive consonantal alternations. The pronunciation of stressed syllables involves increased airflow, which can in turn lead to an increase of aspiration, frication, or affrication. This is seen in English, where aspiration of voiceless stops is longer and stronger in stressed syllables than in unstressed syllables. Another example is Northern Peninsular Spanish, where stress increases the likelihood of frication of coda /b, d, g/ (Gonzalez (2002b)). Voicing variation, which tends to occur in unstressed syllables, is also considered an aerodynamic effect of stress. The exact, complex requirements for voicing are more difficult to achieve in unstressed syllables because there is less time to achieve them. Aerodynamic processes are mostly phonetic (as in frication in Northern Peninsular Spanish) but they can also be phonologized (as in aspiration of Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 12 voiceless obstruents in English; see, among others, Brownian and Goldstein (1992), Davis and Cho (2003)). • Prominence Prominence refers to the relative phonological weight of a syllable. A syllable will be prominent if it is heavy, and non prominent if weak. Typically, prominent syllables have long vowels or coda consonants. Prominent syllables tend to be stressed, and stress tends to be assigned to prominent syllables (Smith (2000, 2002)). Prominence plays a role in stress-sensitive consonantal processes. One example is gemination in Popoloca. In this language, a voiceless obstruent onset is geminated immediately after a stressed vowel. In this case neither perception, duration or aerodynamics can explain gemination. Gemination in this case increases the phonological ‘weight’ of the stressed syllable, making stressed syllables ‘heavier’ than unstressed syllables. Prominence can also cause the reduction of the phonological weight of unstressed syllables, making unstressed syllables Tighter’ than unstressed syllables. Prominence can create rhythmic contrasts in languages between strong and weak footed syllables, as in Capanahua, where coda 1 1 1 is deleted in weak footed syllables (Loos (1969), Safir (1979), Gonzalez (2002a)). When a prominence contrast is created in the foot structure of a language, the process is referred to as rhythmic. Rhythmic phenomena are not limited to stress, and when realized through consonantal alternations, they might make up for the absence of rhythmic stress. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. It is proposed that from these factors, aerodynamics, duration and perceptual ambiguity are primarily phonetic (but can be phonologized in some cases), while other perceptual factors and prominence are primarily phonological. In some cases, various factors can be at work in a single consonantal phenomenon. In other occasions, a consonant alternation might be ambiguous as to the factors that condition it. These cases are discussed in detail in chapter 3. 3.2 Stress and foot structure A second theoretical proposal of this dissertation is that stress and foot structure are separately needed to characterize the range of conditioning contexts for the consonantal phenomena surveyed. The survey evidences three different types of consonantal phenomena: stress-only cases, stress/foot structure cases, and foot-only cases. • Stress-only cases Crucially, strictly stress-sensitive consonantal processes are not foot-based. Stress- sensitive consonant processes occur in two cases: (i) Languages with only one stress and no evidence for foot structure. This is probably the case of Senoufo. Senoufo consonants lengthen in stressed syllables (Mills (1984)). Since Senoufo has only one stress per word and no evidence of Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 14 persistent footing, and since stress can increase consonantal duration, consonant lengthening seems to be strictly stress-sensitive in this language. (ii) Languages where, even if footing is persistent, consonantal processes occur only in a subset of stressed syllables—namely, only in main-stressed syllables, or only in secondary-stressed syllables. One plausible candidate is Urubu-Kaapor, where onsets are lengthened in main stressed syllables and /p, t/ optionally become voiced in ‘non primary stressed syllables’ (Kakumasu (1986)). Onset lengthening in main-stressed syllables could be interpreted as an alternation occurring in the main foot of the prosodic word. However, the fact that optional onset voicing occurs in non-primary stressed syllables suggests that both phenomena are stress-sensitive; main stress increases consonant duration, while non-primary stress might involve aerodynamic instability in consonantal voicing, or reduced duration that for voiceless consonants gives the percept of voicing. Mostly, purely stress-conditioned consonantal phenomena processes are expected to be aerodynamically or durationally based, since duration and aerodynamic properties are readily conditioned by stress. In principle, perceptual and prominence factors might also be involved in strictly stress-sensitive consonantal alternations. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 15 ® Stress/foot structure cases Stress and footing may coincide in conditioning a consonantal process. In such cases, it is ambiguous whether stress and/or footing are conditioners of the process. One example is flapping of N in American English. Omitting some details, N is typically flapped in medial unstressed syllables and aspirated in stressed syllables (Kahn (1980), Borowsky (1986), Turk (1992, 1993), among others). Since English is mainly a trochaic (head-first) system, flapping has been argued to occur foot- medially and aspiration foot-initially (Kiparsky (1979), Hayes (1995), Davis and Cho (2003)). The coincidence of stress and foot structure in determining the flapping of /t/ is exemplified in (8) (stress is marked with (') and foot structure is in parentheses). (8) American English flapping III is flapped in unstressed syllables (8a, b) but aspirated in stressed syllables (8b, d). Flapping coincides with the weak (non-head) syllable in a trochaic foot (8a, b). /t/ is aspirated in strong footed syllables (8b, d). For processes involving onsets, the coincidence between stress and foot structure in the environment conditioning a consonantal process is typical of trochaic systems, where the head is foot-initial and stressed and the non-head is unstressed. In iambic systems, where the foot-initial syllable is unstressed, it is typical for stress and foot structure to be at odds in respect (a) rider [('lai.roi)] (b) potato [p3.('thei.r9u)] (c) take [('t^eik)] (d) attack [a.C^aek)] Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 16 to the conditioning of consonantal processes involving onsets (see footing-only cases below). Foot structure has not been studied in all of the languages gathered in the survey. Thus, some languages are inconclusive as to whether a consonantal process is strictly stress-motivated, or whether stress and footing coincide in motivating it. • Footing-only cases Finally, there are cases where foot structure, but not stress, is the determining factor in conditioning a consonantal alternation. From the languages surveyed in this dissertation this situation arises in three different situations: (i) languages with persistent footing that do not stress all feet in the prosodic word; (ii) iambic systems, and, (iii) languages where, even if the rhythm is trochaic, directionality or alignment between stress and foot structure conflict. These are discussed in turn. (i) Languages with persistent footing that do not stress all feet in the prosodic word. Capanahua (Panoan: Peru) exemplifies this case. Only one stress per word is reported in Capanahua (Loos (1969)) (9a). Coda /?/ is deleted in weak footed syllables without regard to stress (9d, e). Deletion of coda /?/ does not occur in strong syllables (9d, e) or in unfooted syllables (9f). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 17 (9) Capanahua (from Loos (1969); IP A) (a) [('?o.§a.)(ka.?i.)(ni?.ki)] (b) /ta?/ (c) /ra?/ (d) [(‘ ?o .tfi) (ti .ra) {ta? .ki)] (e) [('?o .tji) (ti .ma) (ra?. ta) ki] ‘he falls asleep’ declarative modal ‘probably’ ‘it is probably a dog’ ‘it is probably not a dog’ (f) /?i?sap/ [?i?.('sa)] ‘bird’ Deletion of coda /?/ takes place in a foot-sensitive fashion; deletion occurs in even- numbered syllables from the beginning of the word. Capanahua is a trochaic system, since stress is assigned either to the first or second syllable of the word (9a, f) and /?/ deletion proceeds left-to-right (Safir (1979), Gonzalez (2002a)); /?/ deletion correlates with weak footed syllables and not with stress. (ii) Iambic systems. In trochaic systems the head syllable is left-headed, while in iambic systems it is right-headed. This has potential implications for consonantal processes occurring in both types of systems. Focusing on onsets, in trochaic systems stress and foot position coincide; heads are both stressed and foot-initial, while complements are unstressed and foot medials. For onsets in iambic systems, stress and foot position do not coincide; heads are stressed but foot-medial, and complements unstressed but foot-initial (Figure 1.1). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 18 Figure 1.1. Stress and footing in onsets of trochaic and iambic systems Trochaic Iambic Foot-initial Foot-medial Foot-initial Foot-medial ('0 . a) ( a . 'a) Stressed Unstressed Unstressed Stressed Norton Sound Yupik exemplifies an iambic system where stress and foot position do not coincide in determining a consonantal process. In Norton Sound Yupik main stress occurs in the leftmost foot; the final syllable of a word does not carry stress (10a). Norton Sound Yupik has an alternation between [w, j, 1 ] and [v, z, tj]. In traditional terms this is an instance of fortition, whereby underlying voiced approximants are fricativized foot-initially (first column in lOb-d), but otherwise realized as voiced approximants (second column in lOb-d). Since Norton Sound Yupik is iambic, this means that fortition occurs in unstressed foot-initial syllables. This distribution conflicts with the cross-linguistic tendency for fortition to take place in stressed syllables (2). Additionally, the approximant variants cannot be correlated with stressed syllables, since not all feet carry secondary stress in Norton Sound (10b). Fortition in this language is foot-sensitive rather than stress-sensitive and seems to mark foot boundaries (Leer (1985) and subsequent literature). (10) Norton Sound (from Van der Vijver (1998)) (a) qajapigkani [(qa.‘ jaa.) (pix.,kaa.) ni] ‘his own future authentic kayak’ (b) (ma.'jur) (vik) ('at.) (xar.wik) ‘place to go up/down’ (c) (ma.'jur) (zux.tuq) ('at.) (xar.jux.) tuq ‘he wants to go up/down’ (d) (ma.'jur) (feu.ni) ('at.) (xar.,lu.) ni ‘(he) going up/down’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 19 (iii) Languages with a conflict between directionality or alignment between stress and foot structure. In some consonantal processes footing and stress are potentially antagonistic, with consonantal alternations respecting footing but not stress. One relevant case is Huariapano (Panoan: Peru). In this language, main and secondary stress have opposite directionalities; further, main stress shows quantity-sensitivity, while secondary stress does not. Main stress is assigned via a moraic trochee at the right edge of the word (11a). Default secondary stress is assigned via syllabic trochees starting from the left edge of the word (1 lb). Huariapano has a process of coda [h] epenthesis before voiceless consonants in strong syllables of syllabic trochees constructed left-to-right (11c). [h] epenthesis generally coincides with default (left-to-right) secondary stress assignment; the syllables where [h] is epenthesized are generally stressed trochaic heads (11c). However (and omitting some details), coda [h] epenthesis occurs in other contexts, including syllables that without epenthesis would be unfooted (lid ), and weak, unstressed syllables of trochaic feet in words where secondary stress is assigned right-to-left (lie). This means that stress is not a conditioning factor for [h] epenthesis (Parker (1998)). (11) Huariapano (from Parker (1998:2-5); IP A) (a) [ka.('no.ti)] ‘bow (weapon)’ (b) [(jo.mra.)(,raj.Jpa.)kan.(lgi.ki)] ‘they hunted’ (c) [(l jo.miru.)(1 rah.ka.)(ltih.ksej)] ‘(they) hunted’ (d) [(ja.na.) pah.('kwir))] ‘(I) will help’ (e) [jlis.^ma.noh.X^o.no.X'gi.ki)] ‘I forgot’ (R-to-L footing) (f) [(ja.na.) (pah.'kwtp)] ‘(I) will help’ (L-to-R footing) (g) [(JMs^ma.Xnoh^ko.Xno. 'gi.) ki] T forgot’ (L-to-R footing) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 20 However, foot structure does condition [h] epenthesis. [h] epenthesis regularly occurs in all cases in heads of syllabic trochees built from the left edge (cf. lid , e with Ilf, g). This corresponds to default secondary stress. Cases where [h] epenthesis appear not to correspond to footing are due to non-default secondary stress assignment or cases of extrametrical or irregular main stress. The finding that consonantal processes can be influenced strictly by stress and strictly by footing leads me to propose the separation between stress and foot structure. Even if stress and footing might coincide, they are independent, as suggested by cases where foot structure is independent of stress (as in Panoan languages) and where stress is independent of footing (as in Senoufo; North-Central Peninsular Spanish is another example that will be discussed in detail in a later section). This contradicts the traditional view of foot structure organization as signaled necessarily by stress-based prominence. While stress can be one of the possible ways in which languages organize syllables into groups, there are other ways which are available in order to achieve this aim. Among these are consonantal alternations. Other possible ways in which languages foot syllables include vowel alternations and tonal phenomena. 3.3 Analysis of stress- and foot-sensitive consonantal phenomena Perceptually-based phenomena have recently received wide attention in the literature, with approaches like Licensing-by-cue (Steriade (1997), Hume and Seo (2001), among others). These approaches can easily account for consonantal Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 21 phenomena that are conditioned by the tendency to avoide salient features or consonants in weak positions and to enhance salient features or consonants in strong positions. Prominence-based phenomena have also been considered in recent literature, with approaches like Prominence enhancement (Smith (2000, 2002)) and Prominence Alignment (Prince and Smolensky (1993), Kenstowicz (1996), Crosswhite (to appear), among others). Prominence enhancement exclusively accounts for processes where prominence is increased in strong positions, or where strong positions augment in order to be more prominent. Prominence alignment, which is based on the combination of relative prominence scales, has been shown in previous studies to account for stress- and foot- sensitive consonantal phenomena involving both prominence reduction and augmentation for vocalic and tonal phenomena (Kenstowicz (1996), De Lacy (1999), Crosswhite (to appear)). This dissertation proposes that Prominence Alignment can be successfully deployed to capture prominence reduction and augmentation for consonantal phenomena in both stress- and foot-sensitive contexts. While it is generally recognized that durational and aerodynamic constraints play a role in phonological analyses, these constraints are left unformulated. This dissertation also proposes the extension of Prominence Alignment to account for phonological consonantal phenomena with durational and aerodynamic bases. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 22 4. Case studies Two chapters of this dissertation are dedicated to in-depth case studies of consonantal phenomena from the survey in chapter 2. They involve aerodynamic and rhythmic processes in coda position where laryngeal features or segments are affected and exemplify, respectively, phonetic and phonological examples of consonantal alternations conditioned by stress and footing. The first case study investigates the effect of stress on frication and devoicing of coda /b, d, g/ in North-Central Peninsular Spanish (Gonzalez (2002b)). In this dialect, /b, d, g/ are usually realized as voiceless fricatives in coda position. The acoustic study in Gonzalez (2002b) shows that there is an effect of stress on frication; frication is more likely in stressed syllables than in unstressed syllables. This effect is variable and it is argued that it should not be incorporated in the grammar. This case study argues against previous analyses of this process as phonological coda devoicing (Hualde (1989), Martinez-Gil (1991), Morris (in press)). The second case study investigates rhythmic consonantal phenomena in Capanahua and Huariapano, two Panoan languages from Peru. In both of these languages there are rhythmic phenomena in coda position involving laryngeal consonants (Loos (1969, 1999), Safir (1979), Parker (1994, 1998), Gonzalez (2002a)). In Capanahua, /?/ deletes in weak footed syllables. There is only one stress Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 23 per prosodic word, so this is a case of a foot-only sensitive process. In Huariapano, [h] is epenthesized in strong footed syllables, sometimes even in opposition to stress. 5. Theoretical framework 5.1 Optimality Theory This thesis is couched within Optimality Theory (OT) (Prince and Smolensky (1993)). One of the central tenets of OT is that phonological processes are not obtained by rules, but through universal constraints that operate over output forms in languages. Constraints are potentially opposing forces that may stand in tension with each other; cross-linguistic differences between languages are due to different ways of resolving these tensions. In Optimality Theory this is expressed by different rankings of these constraints in a hierarchy. A claim of this theory is that there are just two levels: input (which roughly corresponds to the underlying form) and output (surface form); there are no intermediate forms or serial derivations. From the input form, a set of possible outputs or candidates are evaluated in parallel with respect to the constraint hierarchy in a language. The output that best satisfies the particular ranking in a language is chosen as the optimal form. OT has three main components: Generator (G e n ), Constraints (C o n ), and Evaluator ( E v a l) . G e n is a function mapping an input to a set of possible outputs. Constraints on GEN determine the set of possible inputs. C o n is a set of universal violable constraints. E v a l is a function evaluating the output candidates generated Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 24 by G e n with resp ect to a particular ranking of C o n . There is no serial derivation; candidate evaluation proceeds in parallel with respect to the whole constraint hierarchy. A tableau shows the interaction of these three components with respect to a particular form (12). (12) Interaction between Gen, Con, and Eval in a tableau. Eval /input/ C o n s t r a i n t 1 C o n s t r a i n t 2 ar Candidate a Candidate b V ' : " v. .. ;V " i ; A ;.- L a . ; ; : v: f ;.V': & t } C o n G e n E v a l The tableau above represents a typical case of conflict between two constraints. Only a subset of candidates has been shown here; usually, only the most competitive or relevant candidates are shown in a tableau. The optimal candidate chosen from the set of generated candidates for a given input is indicated with t3 P . The particular ranking among the constraints supplied by CON is shown by a left to right ordering of the constraints. In this particular tableau Constraint 1 is ranked higher than Constraint 2. This means that the satisfaction of this constraint has preference over the satisfaction of Constraint 2. An asterisk (*) indicates a constraint violation. Candidate (a) in the tableau violates Constraint 2, and candidate (b) violates Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 25 Constraint 1. The exclamation mark (!) indicates that the violation o f Constraint 1 is fatal, and that the candidate being evaluated (b) loses at this point. Even if candidate (a) has two violations of Constraint 2, it is selected over candidate (b) because Constraint 1 is ranked higher in the hierarchy. The two basic types of constraints within OT are faithfulness and markedness constraints (Prince and Smolensky (1993)). Faithfulness constraints conspire to ‘pronounce everything as it is’ in the input form (Hammond (1997)). In this work I assume the Correspondence Theory of Faithfulness (McCarthy and Prince (1995)), whereby the input and the output are in a correspondence relation with each other. Examples of faithfulness constraints are Max-IO, Dep-IO and Ident-IO [voice] (13). (13) Examples of faithfulness constraints (McCarthy & Prince (1995)) Max-IO Every element of the input has a correspondent in the output (Phonological deletion is not permitted). Dep-IO Every element of the output has a correspondent in the input (Phonological epenthesis is not permitted). Ident-IO [voice] Correspondent segments in the input and output have identical values for the feature [voice]. If x is a segment in the input, and y is a segment in the output, and if xRy and x is [yvoice], then y is [yvoice]. Markedness constraints (or ‘phono-constraints ’) capture preferences in phonological well-formedness and in segment inventories. The interaction between faithfulness and markedness constraints captures phonemic and allophonic distributions in languages (Prince and Smolensky (1993), Kager (1999)). Examples of markedness Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 26 constraints are N oCoda (Prince and Smolensky (1993)) and *Voiced Fricative (14). (14) Examples o f markedness constraints N oCoda Avoid codas * V oiced Fricative Voiced fricatives are prohibited. A fundamental principle of OT is Richness of the Base (Prince and Smolensky (1993)). According to Richness of the Base, there are no language-particular constraints on inputs. Hence, inputs are universally available; they do not depend on the language being considered or the underlying form. The characteristics of the set of acceptable outputs depend on the constraint hierarchy of the specific language considered, and not on the characteristics of the set of inputs. 5.2 Stress and Metrical Theory Stress can be defined as prominence produced by means of respiratory effort (Lehiste (1970:119)). Differences in stress are caused by differences in physical effort, which have a direct reflection in the activity of the respiratory muscles. The force exerted by the muscles involved in respiration is transmitted to the air in the lungs and is then reflected in subglottal pressure (Lehiste (1970) and references therein). While there is no one-to-one correspondence between stress and any phonetic feature, stressed syllables are typically louder, have pitch extrema, and last longer Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 27 than unstressed syllables (Hayes (1995)). They are also perceptually and articulatorily prominent (Beckman (1998) and references therein). Across languages, stress has an organizing function; it divides speech into units (Lehiste (1970:147)). The smallest unit carrying stress is the syllable, and the minimal unit of contrastive stress placement is a sequence of two syllables (Lehiste (1970:147)). This idea is behind the Metrical Theory of Stress (Liberman (1975), Prince (1976), Liberman and Prince (1977), Hayes (1980, 1995), Halle and Vergnaud (1987)). The main assumption of this theory is that stress is a rhythmic phenomenon, encoded by strong-weak relations between syllables. Another important assumption is that constituents are grouped into higher units as in the prosodic hierarchy (15) (Selkirk (1980), McCarthy & Prince (1986)). (15) Prosodic hierarchy PrWd Prosodic Word Ft Foot a Syllable 1 1 Mora At the bottom of the hierarchy is the mora (represented with ‘p’). The mora is a weight-bearing unit in a syllable. According to the moraic theory (Hulst (1984), Hyman (1985), Hayes (1989)), the quantity of a syllable depends on its number of moras. In most languages, vowels are always moraic; short vowels have one mora Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 28 and long vowels two. Diphthongs and coda consonants can be moraic or not depending on the language. Light syllables have one mora; heavy syllables, two. Moras are grouped into syllables, which minimally consist of a mora. Syllables are grouped into feet. A foot is formed minimally by a stressed syllable, but it typically has both a stressed and an unstressed syllable. Two basic types of feet exist: iambic (light-headed), and trochaic (left-headed) (Hayes (1995)). An iambic foot can be formed of a stressed syllable, or of an unstressed syllable followed by a stressed syllable. Typically, in the second case the stressed syllable is heavy. A trochaic foot can be moraic (quantity-sensitive) or syllabic (quantity-insensitive). Moraic trochees are formed of a heavy stressed syllable. Syllabic trochees are formed of a stressed syllable followed by an unstressed syllable (Hayes (1995)). Examples of trochaic and iambic footing are shown in (16). (16) Examples of trochaic and iambic feet (a) Trochaic feet: Huariapano (from Parker (1998:2-10); IP A) Moraic ('H) [ja.('wif)] ‘opossum’ ('L.L) f('hi.wi)] ‘branch, stick’ Syllabic ('g o ) |~ ('ha.J3om.) ('J3i.j3i)] ‘they’ [mi.('Pom.|3i.) ('ra.ma)] ‘you (plural)’ (b) Iambic feet: Central Siberian Yupik (from Van de Viver (1998:116, 174); IPA) (Canonical) (L.'H) [(qa.'jaa.) ni] ‘his own kayak’ (L.'L) [(a.'to.l pik] ‘real name’ ('H) [su.'ruu.) ('jaa).ni] ‘in his (another’s) drum’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 29 The canonical iamb has an unstressed syllable followed by a heavy stressed syllable; the two syllables in an iamb contrast in duration. The canonical trochee is syllabic; syllables in a trochee contrast in intensity (Hayes (1995)). This is stated in the Iambic/Trochaic Law (Hayes (1995)) (17). (17) Iambic/Trochaic Law (Hayes (1995:80)) 1. Elements contrasting in intensity naturally form groupings with initial prominence. 2. Elements contrasting in duration naturally form groupings with final prominence. At the top of the prosodic hierarchy is the prosodic word. Typically, prosodic words have a single strongest syllable bearing the main stress. This is the culminativity property of stress. Other cross-linguistic properties are the demarcative property (stress tends to occur at or close to constituent edges), quantity-sensitivity (stress is attracted to long vowels, diphthongs, and closed syllables), and the rhythmic property (stressed and unstressed syllables alternate, and stress clashes or lapses are avoided) (Kager (1999); see also Hayes (1995)).1 These properties of stress are often in conflict. As discussed previously, conflict is solved in Optimality Theory by the language-particular ranking of universal constraints. Some metrical constraints are included in (18). Additional metrical constraints are discussed in chapters 3-5. 1 Note that the rhythmic property and rhythmicity refer to different concepts; the rhythmic property refers to an alternation between stressed and unstressed syllables, while rhythmicity refers to a prominence contrast in the foot structure of a language. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 30 (18) Metrical constraints Culminativity G r W d = P r W d Quantity-sensitivity W e ig h t-b y -P o sit io n S t r e s s-t o -W eig h t Demarcative property L e f t m o s t /R ig h t m o s t Rhythmic property FT-BlN A grammatical word must be a prosodic word (Prince and Smolensky (1993)) Coda consonants are moraic (Hayes (1989), Sherer (1994)) If stressed, then heavy (Myers (1987), Riad (1992)) Align (Hd-Ft, Left/Right, PrWd, Left/Right) ‘The head foot is leftmost/rightmost in PrWd’ (Prince and Smolensky (1993)) Feet are binary under moraic or syllabic analysis (Prince and Smolensky (1993)) A proposal of this dissertation is that footing can be persistent even if not all feet in the prosodic word have stress. Persistent footing can be realized not only through stress, but also through consonantal and vocalic alternations, and possibly by tonal alternations too. This is explored in more detail in chapters 3 and 5. 5.3 The relationship between phonetics and phonology Phonetic realization is mostly subject to language-specific variation, as shown by Keating (1985). For example, while both English and Spanish have coronal stops, the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 31 phonetic realization of these consonants varies; consonant stops are alveolar in English, but dental in Spanish. There are two options if we want to represent this fine-grained detail of realization. One is to include fine phonetic detail in the phonological representation; this will be referred to as the ‘integrated’ approach. Alternatively, if fine detail is excluded from the phonological representation, then a phonetic component should be added to the grammar; this will be referred to as the ‘modular’ approach. Advocates of the integrated approach of phonetics and phonology include Byrd (1996), Kirchner (1997), Steriade (1997), and Flemming (2001). Advocates of the modular approach include Keating (1984, 1990), Archangeli and Pulleyblank (1994), and Howe and Pulleyblank (2001). One argument for the separation of the phonological and the phonetic modules or components is given by Howe and Pulleyblank (2001), who investigate the patterning and timing of consonant glottalization. Their finding is that there is a mismatch between the phonetic predictions of glottalization timing and the attested patterns of glottalization. The phonetic prediction is that the cues of glottalization will be optimal in vocalic environments; thus, consonant pre-glottalization will tend to occur post-vocalically, and post-glottalization pre-vocalically (Sapir (1938), Kingston (1985), Silverman (1997), Steriade (1997), among others). However, in many languages glottal timing correlates with syllabic position—which is a purely phonological construct in many languages. For example, consonants in coda positions are post-glottalized rather than pre-glottalized in Yowlumne and Kashaya, and consonants in onsets are pre-glottalized rather than post-glottalized in Nuu-chah- Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 32 nulth and Yowlumne. Howe and Pulleyblank (2001) conclude that the phonetic factors o f timing and perception of glottalization have no direct effect on the phonological component of the grammar; therefore, the phonetic and the phonological components are separate. In this dissertation I assume a strict separation between phonetics and phonology, as in the modular approach. The grammar will have both a phonological and a phonetic component. The phonological component will comprise phenomena which are category-neutralizing, categorical, or obligatory, while the phonetic component will comprise partial, variable, or gradient phenomena. For example, vowel reduction in Italian, which is category-neutralizing, will be a part of the phonological component; in contrast, non category-neutralizing vowel reduction will be a part of the phonetic component (see discussion in Flemming (2001)). The definition of which phenomena should be included in the phonological or in the phonetic component allows for cases that do not clearly belong to the phonetic or to the phonological component. For example, there could be cases of phenomena which are category-neutralizing but variable, or obligatory but partial. Further investigation might elucidate the extent to which such processes belong to the phonetic or the phonological component. The separation between the phonetic and phonological modules does not make any assumptions about the types of explanations that can influence phonological processes. As is well known, many phonological processes have phonetic bases; articulatory and perceptual factors in many cases condition Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 33 phonological phenomena (Browman and Goldstein (1990, 1992), Hayes (1997), Steriade (1997), among others). Within Optimality Theory, this is captured through phonetic grounding: a constraint can be motivated if it has a phonetic basis, that is, if it expresses a principle of ease of articulation or of maximizing perceptibility (Archangeli and Pulleyblank (1994)). I assume that phonetic explanations or motivations can be encoded in the phonology, provided the process is category- neutralizing, categorical or obligatory. Furthermore, the assumption in this dissertation that phonological and phonetic phenomena are distinct and fall into different components does not make any claims as to the nature of the phonetic component. The phonological component will be represented in optimality theoretic terms. Gradient, partial and optional phenomena will be considered to be phonetic, with no specific claims as to the nature of the phonetic component itself. Variation in a process will be allowed in the phonological component provided that variation occurs in at least half of the speaker population, or half of the time for a set of speakers. Phonological variation will be expressed in the grammar as the free ranking or crucial unranking of relevant constraints. As a contrast, phonetic variation will be noted, but it will not be encoded in the grammar. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 34 6. Organization This dissertation is organized as follows. Chapter 2 reports on an extensive cross- linguistic survey of consonantal processes where stress and/or foot structure are influential. This survey is organized by type of process. Chapter 3 discusses the typological generalizations that arise from this survey, including the specific stress and footing environments in which consonantal phenomena are conditioned cross- linguistically, the types of segments and syllabic positions that are preferentially targeted in these consonantal phenomena, and the relevance of positional factors. It proposes that four different factors can be seen at work in stress- and foot-sensitive consonantal processes from the survey: perception, aerodynamics, duration and prominence. It also proposes that both stress and foot structure are independently needed to account for these phenomena, and that prominence constraints that make explicit reference to foot positions (head and complement) in addition to stress should be included in the grammar. Chapters 4 and 5 are two case studies of coda processes where laryngeal features or segments are influenced by stress and foot structure. Chapter 4 investigates the role of aerodynamics in stress-sensitive frication and devoicing of coda /b, d, g/ in North-Peninsular Spanish. It shows that frication rather than devoicing is the main factor behind this process, and that stress affects frication; frication is more likely in stressed syllables. This exemplifies an aerodynamic-caused Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 35 stress-related consonantal process, and also a phonetic process; the influence of stress on frication is variable rather than categorical. Chapter 5 examines prominence/rhythmic processes in Panoan languages Capanahua and Huariapano. Capanahua has /?/ deletion in weak footed syllables, and optional /?/ onset to coda metathesis in strong footed syllables. Chapter 5 argues that in the absence of secondary stress these processes achieve rhythmicity in the language. Huariapano has a process of /h/ epenthesis in strong footed syllables. It shows how a mismatch between main and secondary stress affects this process. Both of these processes are analyzed though foot-sensitive prominence constraints. The case studies on these two Panoan languages contrast with the case study of North- Central Peninsular Spanish in two ways; Capanahua and Huariapano have foot- sensitive consonantal processes in the absence of or in opposition to stress; their factor is rhythmic, and they are phonological (categorical) processes. In contrast, North-Central Peninsular Spanish exemplifies a purely stress-sensitive consonantal process with an aerodynamic motivation, and also a phonetic (variable) process. Chapter 6, the conclusion, summarizes the main findings of this research, points to issues that have yet to be resolved, and suggests directions for further research. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 36 Chapter 2: S t r e s s - A n d F o o t - S e n s it iv e C o n s o n a n t a l P h e n o m e n a : A S u r v e y 1. Introduction The influence of stress and foot structure has been well studied for a few consonantal phenomena. For instance, stress influences the length and strength of aspiration in onset voiceless stops in English (Turk (1993)), and flapping of /t, d/ occurs in unstressed syllables in American and Canadian English (Kahn (1980), Turk (1992)). In Norton Sound Yupik, fortition appears to be foot-sensitive rather than stress- sensitive (Van de Vijver (1998)), as is Huariapano epenthesis of fhl (Parker (1994, 1998)). Nevertheless, the influence of stress and foot structure on consonantal phenomena has not been studied in broad-based terms, and there is no exhaustive survey or classification of the types of consonantal alternations that are conditioned by these metrical considerations. This chapter aims to fill this gap by presenting an extensive survey of languages with stress- and foot-sensitive consonantal alternations. It also lays down the descriptive groundwork for the generalizations and theoretical proposals in chapter 3. The organization of this chapter is as follows. Section (2) discusses the sources for the survey. Section (3) presents an overview of the survey. Sections (4-9) discuss the types of consonantal alternations that are reportedly influenced by stress or foot structure. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 37 2. Sources for the survey Some recent cross-linguistic studies of consonantal phenomena, including lenition, fortition and metathesis, report that stress and/or foot structure are conditioning factors for some of these processes. Examples of such cross-linguistic studies include Lavoie (1996, 2001) and Kirchner (1998) on lenition/fortition processes, Blevins and Garrett (1998) and Hume (2000) for metathesis, Kehrein (2001) on laryngeal phenomena, and Smith (2002) for stress-attraction processes. These are discussed in sections 2.1-2.3. Section 2.4 describes the sources for the cross-linguistic survey of stress- and foot-sensitive consonantal phenomena that is reported in sections 4-9. 2.1 Lenition and fortition Lavoie (1996, 2001) presents an extensive survey of languages with lenition and fortition in syllable onsets. She focuses on the interaction between word position and stress on the one hand and lenition/fortition on the other. Lavoie finds that there are many reported instances of word position conditioning lenition and fortition, but very few reported cases where stress conditions consonant alternations. Among the languages that have stress-sensitive lenition are West Tarangan and Somali (approximantization), Old English and Senoufo (voicing), Senoufo (fricativization), and American and Canadian English (flapping). Reported stress-sensitive alternations involving fortition include Yuman (occlusivization and devoicing), Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 38 Guayabero (fricativization), Urubu-Kaapor (onset lengthening) and Maori (affrication). Lavoie (2001) finds that in her two phonetic studies of American English and Mexican Spanish, stress correlates with differences in consonantal realization (such as length or lenition) more often than word position. She concludes that stress affects the realization of consonants phonetically, and that this effect is not reported in descriptions of languages because it is not usually phonologized. Kirchner (1998) builds his lenition survey from Lavoie (1996). He identifies Djabugai (flapping), Farsi and Middle Italian (voicing), Gojri (gemination), Kupia (flapping and lenition), Pattani (pre-aspiration) and Southern Tati (deletion) as further cases of stress-sensitive lenition/fortition. 2.2 Metathesis Cross-linguistic studies of metathesis report the influence of stress/foot structure for some cases. Blevins and Garrett (1998) distinguish between compensatory and perceptual metathesis. Compensatory metathesis affects vowels especially and consists of the weakening of vowel quality or duration with compensation via transfer of the vowel quality or duration to the stressed position. Perceptual metathesis involves the change in the linear order of a consonant due to perceptual ambiguity about its position; ambiguity occurs especially in unstressed syllables. Blevins and Garrett (1998) identify Cayuga, Mohawk, Oneida, Thompson River Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 39 Salish, Colville, Shuswap, and Twana as languages with stress-conditioned perceptual metathesis. Hume’s (2000) metathesis website (www.ling.ohio-state.edu/~ehume/ metathesis) identifies Faroese as a case of stress-sensitive consonantal metathesis (Faroese has also stress-sensitive consonantal deletion). 2.3 Laryngeal phenomena and attraction phenomena Kehrein’s (2001) survey of laryngeal phenomena identifies several examples of laryngeal processes where stress/foot structure is relevant. These include Icelandic, Faroese, Scots Gaelic, Ingush, Ojibwa (pre-aspiration); Shuswap, Twana, Thompson River Salish, Squamish (laryngeal attraction); and Gitksan and Coast Tsimshian (pre/post glottalization). Smith (2002) provides examples of stress attraction by specific consonantal properties—including onsetful syllables or syllables with low sonority onsets, as in Alyawarra, Arandic and Piraha. 2.4 This survey The present survey includes stress- and foot-sensitive consonantal phenomena identified in the previous cross-linguistic surveys described above. In addition, it includes further examples gathered from independent grammars and theoretical studies. Any mention of stress or foot structure as conditioning a process was checked in the original source where possible. If the original source did not clearly Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 40 mention stress or foot-structure as conditioning factors the language was not included in the survey. In some cases, checking the original sources brought in additional exam p les that had not been reported in previous studies. The resulting survey includes all languages found that reportedly and clearly possess consonantal phenomena influenced to some extent by stress and foot structure. The result is 74 languages from all over the word, belonging to 36 different language families. In the survey, languages and dialects have b een entered as separate entries in order to be clear in regard to exactly what the process is and in which dialect it occurs (e.g. American English vs. Cardiff English vs. London English). In a few cases, languages with properties not directly relevant to this survey have been included for comparison; for instance, cases where stressed nuclei attract a glottal feature from a consonant. These cases are included because at least the original position of this glottal feature is consonantal, and because this process is conditioned by stress. This survey focuses on synchronic consonantal alternations. This study does not aim to provide an extensive relation of historical changes; however, the survey occasionally includes instances of diachronic processes that were found to be stress or foot-sensitive. Diachronic alternations are marked with * in the summary tables describing the process. Section (3) presents an overview of the organization of the survey. The abbreviations used in this chapter are listed under ( 1). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 41 List of Abreviations C Coda; consonant sg. Singular C.C Gemination V Vowel C: Lengthening vd. Voiced C/0 From coda to onset vless. Voiceless esp. Especially Syllable boundary gen. Genitive # Word final(ly) inf. Infinitive ## Utterance-final(ly) N Nucleus # Word initial(ly) NM Not mentioned ##_ Utterance-initial(ly) nom. Nominative Low tone 0 Onset High tone O/C From onset to coda - Mid tone opt. Optional a Syllable PIE pi. R Proto Indo European P lu r a l a Unstressed syllable I llUai Sonorant 'a Stressed syllable S Stop * diachronic process 3. Survey overview The survey presented here is the first step towards identifying how stress and foot structure condition consonantal processes. The languages included are identified according to the following parameters: (i) type of process; (ii) prosodic position, and (iii) syllable position. i. Type of process (a) Duration/Magnitude: Duration and magnitude differences in a consonantal feature or segment that pattern by stress or foot structure. It includes: - Durational differences: Consonant duration, Voice Onset Time duration. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 42 - Magnitude differences: Flapping (total closure replaced by a quick, sometimes incomplete, closure); Fortition (sonority decrease that makes a consonant more consonantal); Lenition (sonority increase that makes a consonant more vocalic); Secondary articulation (palatalization, velarization or labialization on the main consonant articulation). (b) Featural timing: Distributional variants of a phoneme patterning by stress or foot structure. It includes: - Pre-aspiration: The onset of aspiration begins before the onset of the closure of the segment. - Post-aspiration: The onset of aspiration begins after or is simultaneous with the release of the consonant. In some cases, stress-sensitive pre-aspiration and post-aspiration are allophonic. - Pre-glottalization: The onset of glottalization begins before the onset of the closure of the segment. - Post-glottalization: The offset of glottalization begins after or is simultaneous with the release of the consonant. In many instances, stress-sensitive pre- glottalization and post-glottalization are allophonic. - Voicing: Variation in the voicing specification of a consonant. (c) Deletion and epenthesis. Segmental content disappears or is introduced for stress or foot structure considerations. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 43 - Deletion: Non-pronunciation of an underlying segment. - Epenthesis: Pronunciation of a segment not present underlyingly. (d) Attraction: Movement of a segment or feature to a non-adjacent position. (e) Metathesis: Reversal in the linear order of adjacent segments or features. (f) Dissimilation: A segment becoming less similar to another segment in a specific domain. ii) Prosodic position: (a) Stress: Stressed syllable or unstressed syllable (b) Foot structure: Strong syllable within a foot or weak syllable within a foot iii) Syllable position: Coda, onset, or both. Cases where the nucleus is relevant are also discussed. ‘Transfer’ and ‘attraction’ are grouped within ‘attraction’ because in both cases, a feature or segment moves from its original position to another position. ‘Fortition’, ‘lenition’ and ‘flapping’ are grouped within ‘durational/magnitude effects’ because they involve both durational and magnitude differences. Gemination processes, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 44 usually classified under ‘fortition’, are included under consonant duration processes, because duration seems to be the main correlate. In some cases, only stress, or on ly foot structure are reported to condition a consonantal alternation. Cases where both stress and foot structure coincide in conditioning an alternation are mentioned in the survey. This survey is organized by the type of the process as outlined above. Each section presents an overview in table form of the languages that have that process. Each table has information on which languages undergo the process, the language family to which they belong, a brief description of the process, and the syllable position involved. Languages in the summary tables are organized thematically rather than alphabetically. Genetic language information, if not mentioned or clear in the source, is from the Ethnologue (Grimes (2000)). Most of the languages considered have synchronic processes. Those which show diachronic processes are marked with an asterisk (*) in the summary tables. Summary tables are followed by as much detail as possible on how the process operates in each language. The process descriptions for each language include cross-references to other sections where the same language is discussed. After the description of each language, each section closes with a brief summary on the processes discussed. An appendix at the end of the dissertation includes all of the languages in the survey with more detail genetic language information and a summary of all stress- and foot-sensitive consonantal phenomena for each language. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 45 The focus of this survey is to gather examples of consonantal variation. Thus, this survey does not include examples of languages where consonantal contrasts occur specifically in stressed syllables, or languages where consonants in stressed syllables trigger or block phonological processes (see chapter 1). For a discussion of these phenomena see Beckman (1998). However, if a language showing consonantal variation additionally shows consonantal contrasts in stressed and unstressed syllables, this is noted in the language description in the survey. The survey is presented in sections (4-9). The main sources for each process are identified, and relevant examples included where available. 4. Duration/magnitude differences 4.1 Durational differences Stress-conditioned durational differences have been reported for consonants in onset and coda position and for voice onset time (VOT) in obstruents. Table 2.1 includes cases of consonant lengthening and VOT for Spanish, English, Senoufo, Maori, Greek, Turkish, Copala Trique and Urubu-Kaapor. Table 2.1 also includes gemination in Guelavia Zapotec, Popoloca, Alutiik Yupik, Norton Sound, Gundidj, Narinari, Wergaia, Italian, and Somali. Throughout this chapter, lengthening will be indicated by [:] after the relevant segment, and gemination by two timing slots [C.C], Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 46 Table 2.1. Durational differences1 L a n g u a g e cr POSITION D e s c r ip t io n L a n g u a g e FAMILY Spanish o ,c ICI~¥ [C!]/[+stress] Romance English 0 /C/-> [Cl]/[+stress]; VOT longer in 'o Germanic Senoufo 0 /C/-» [Cl]/[+stress] Niger-Congo Maori 0 I t / , nasals longer in 'a Austronesian Greek 0 /si longer #_ in 'a Greek Turkish 0 Longer VOT in ‘a for vless. S Altaic Urubu-Kaapor 0 /C/-> [C:]/[+stress] Tupi-Guarani Copala Trique 0 Fortis C, resonants->[Cl]/[+stress] Otomanguean Guelavia Zapotec 0/C Fortis C-»[C:]/'_C and_# Otomanguean Popoloca 0/C Gemination after stress Otomanguean Alutiik Yupik 0/C Gemination in open initial o Yupik Norton Sound Yupik 0/C Gemination in open syllables with [a] Gemination in stems to get stress Yupik Gundidj 0/C Gemination of /m/ after stress Victorian Narinari 0/C Gemination of /l/, In/ after stress Victorian Wergaia 0/C Gemination of IV after stress Victorian Italian 0/C C/0 Gemination in 'V#_'V context Gemination in '_#V context Romance Somali 0/C Gemination of vd. S # after stress Cushitic Spanish: Borzone de Manrique and Signorini (1983) report longer duration of both onsets and codas in stressed syllables in Argentinian Spanish. Lavoie’s (2001) study of onsets reports the same result for Mexican Spanish except for It!. For Peninsular Spanish, Quilis (1981:276, 281) reports that /l, fJ are longer in stressed syllables than in unstressed syllables. While /r/ is also slightly longer in stressed 1 Kirchner (1998:306) reports—citing Sharm a (1979)—that Gojri (India; Indo-Iranian) has geminate stops and fortis singletons except in onset of stressed syllables. However, Sharma (1979, 2002) mentions tone, not stress, in his description of the language. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 47 syllables, It / is longer in unstressed syllables than in stressed syllables (Quilis (1981:292)). Quilis (1981) does not specify syllable positions for these results. English'. Most consonants in English are longer in onsets of stressed syllables than in unstressed syllables (Umeda (1977), Turk (1992), Lavoie (2001)). Voiceless stops have longer VOT in stressed syllables (Lisker and Abramson (1967), Turk (1992), Crystal and House (1988), Lavoie (2001)). Senoufo: Onset consonants are longer in stressed syllables than in unstressed syllables (Mills (1984:119)). In contrast to English and Spanish, vowels in Senoufo are not lengthened under conditions of word stress. Maori: M is reported to have longer contact, and nasals to lengthen in onsets of stressed syllables (Bauer (1993:555)). Other stress-related consonantal processes in Maori are discussed in sections 4.3 and 5.2.1. Greek: Word-initial Is/ is reported to be longer when the syllable is stressed rather than unstressed (Botinis, Fourakis and Bannert (2001)). Turkish: Aspiration is longer in stressed onsets in Turkish, according to Jonathan Barnes (p.c.). 2 Thanks to Sharon Inkelas for bringing this case into my attention. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 48 Urubu-Kaapor: Oral stops /p, t, k, kw, ?/ lengthen in the onsets of primary- stressed syllables (2a-c).3 Lengthening does not occur in secondary-stressed syllables (2c) or in primary-stressed syllables for nasals, approximants or frica tiv es (2d -f). (2) Urubu-Kaapor lengthening (from Kakumasu (1986:399, 401); IP A)4 (a) /k atu / [ka.'tiu] ‘it is a good’ (d) /uruma/ [u.ru.'ma] ‘duck’ (b) /k a?a/ [ka.'?:a] ‘forest’ (e) /w aru w a/ [t wa.ru.'wa] ‘glass’ (c) /nupata/ [nu^pa.'tia] ‘he will hit’ (f) /ixa/ [i.'ja] ‘it is a fact’ The consonants of Urubu-Kaapor are /p, t, k, kw , ?, m , n , q, s, J, h, r, w, j /. Codas are allowed (Kakumasu (1986:399)). Primary stress occurs in word-final syllables. Secondary stress is usually (but not always) assigned to every second syllable counting back (Kakumasu (1986:401)). For related processes see section 5.1. Copala Trique: Resonants and fortis (voiceless) consonants are reported to lengthen before a short, stressed vowel (Hollenbach (1977:37)). Copala Trique has fin al stress. The stressed syllable has more tonal, consonantal and vowel contrasts than unstressed syllables (Hollenbach (1977), Beckman (1998:130)). Fortis consonants and affricates only occur in stressed syllables. This syllable is also the only syllable that can have a coda (/?/ or /h/). Long vowels are extra-long in this position, and short vowels shorter. If the vowel is short, a fortis or resonant onset of the stressed syllable is lengthened. 3 Lavoie (2001:45) mentions that this also occurs utterance-fmally; cf. Kirchner (1998:313). 4 Lenghening in Kakumasu (1986) is represented by consonant doubling; here I use the symbol [:]. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 49 The consonant inventory of Copala Trique is /p, b, t, d, k, g, ?, s, z, j 5 3, g, r, h, ts, tf, m, n, 1, j, w/ (Hollenbach (1977:36)). Stops and sibilants can be fortis (voiceless) or lenis (voiced). Fortis consonants are unaspirated and slightly lengthened. Lenis consonants can vary between voiced and voiceless in unstressed syllables (see section 5.1). /r/ is grouped with the lenis fricatives, probably because it shares some allophonic properties with them; additionally, /§/ has an allophonic retroflex trill. See section 4.3 and 5.1 for related processes. Gualavia Zapotec: Gualavia Zapotec has a contrast between fortis and lenis consonants. Fortis consonants—/p, t, k, ts, tf, tf, s, s, Q , m, n, 1/—are generally longer and more tense than lenis consonants.5 Lenis consonants—/b, d, g, dz, d3, z, ?, j, m, n, 1, r/—make a preceding oral vowel lengthen and have a voiceless release before a pause—except for /m, n/. All consonants can occur in codas except for /(J/. Guelavia Zapotec has both stress and tone. There is one stress per word, usually in the penultimate syllable. Stressed syllables usually have higher pitch. Fortis consonants are reported to ‘lengthen’ before a pause and intervocalically after stressed vowels. Fortis stops and nasals also ‘lengthen’ after stressed vowels before /j, w/ or before a voiced consonant (Jones and Knudson (1977:166)). ‘Lengthening’ does not occur word-initially or in onset of stressed syllables. In this dissertation, gemination and lengthening are distinguished; gemination refers to cases of 5 Jones and Knudson (1977) do not discuss the exact nature of fortis /m n 1/ vs. lenis /m, n, V . Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. lengthening that make a consonant tautosyliabic, while lengthening refers to cases where the consonant is not tautosyliabic. The fact that the context for ‘lengthening’ in Guelavia Zapotec is after stress rather than in stressed syllables suggests that this is gemination rather than lengthening; accordingly, the examples in (3) show consonant gemination rather than lengthening (relevant segments in bold; gemination in underline). (3) Guelavia Zapotec (from Jones and Knudson (1977:164-175)) Gemination after stress No gemination in stressed syllables (a) ['rap.pa7 ] ‘I have’ [ca.'pin.na 1 ‘my pine tree’ (b) ['e ittja 7 ] ‘my onion’ [we.'tip^i] ‘wasp’ (c) ['elam .m a7 ] ‘mv boss’ ['dat.mam] ‘grandfather’ (d) pcpak.ka7 ] ‘my tadpole’ ['kaildt] ‘broth’ (e) ['detjtsa7 ] ‘my back’ ['tsihj] ‘there are ten (objects)’ (f) ['nac.ciq] ‘it is sweet’ t'pa.ga7 ] ‘my tree’ (g) [ca'pin.na7 ] ‘my pine tree’ ['nan.na7 ] T know’ Additional durational processes occur in stressed vowels. Stressed oral vowels lengthen before lenis consonants, as in /'ra.go?/ [Va:.YO? ] ‘you bite’. Stressed laryngeal vowels become interrupted, as in /'ti?.sj/ ['ti? i.si] ‘body’. In the last case the fortis consonant is not lengthened in spite of being intervocalic and occurring after stress. All of these phenomena suggest that stressed syllables in Guelavia Zapotec are heavy: they either have a long vowel or a coda. Vowel lengthening occurs before lenis consonants while gemination occurs for fortis consonants immediately after stress. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 51 Popoloca: In Los Reyes Metzontla Popoloca the main correlate of stress is duration, especially for consonants (Veerman-Leichsenring (1984:21); cf. Stark and Machin (1977:79)). The consonant inventory of Popoloca is shown in (4). Stress generally falls on the penultimate syllable of the word. Stressed syllables have a coda or a long vowel or diphthong. (4) Popoloca consonants (from Veerman-Leichsenring (1984:34); IP A))6 Stops Affricates Fricatives Sonorants pbtd kg ts tf f s x l r m n m n w j th kh tsh t f t f m b nd °g nz % “r Onset fricatives, Hquids, approximants and voiced nasals are geminated after a stressed syllable (Veerman-Leichsenring (1984)). Consonantal clusters and complex consonants, such as pre-nasalized, aspirated, and voiceless nasal consonants are ‘disintegrated’; they are decomposed into a coda and an onset element (5).7 Unaspirated stops and affricates are pre-glottalized in the same position (see section 5.3.2). See also sections 5.1 and 7.1. 6 /ts, t§ y are dental and lamino-palatal respectively, /b, d, g, f/ only occur in non-assimilated Spanish loans (Veerman-Leichsenring (1984:34ff)). 7 For voiceless nasals, the coda element can be /h, ?, xJ (Veerman-Leichsenring (1984:48)). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 52 (5) Popoloca durational processes8 Gemination (from V eerman-Leichsenring (1984:45-7); IP A) (a) Fricatives ['kaf.fe] ‘coffee’ [ln zes.se] ‘sand’ (b) Liquids [ku.'pdLlo] ‘butterfly’ [,n ddr.rd] ‘thin’ (c) Approximants ['tiJ4e l ‘night’ [l9giw.wd] ‘popoloca’ (d) Voiced nasals ['tum.me] ‘money’ ['tun.ni] ‘hail’ ‘Disintegration’ (from Veerman-Leichsenring (1984:48-9, 54); IP A) (a) Pre-nasal stops [tu'rum.ba] ‘spinning top’ ['kdn.da] ‘nopal’ (b) Pre-nasal fricatives [ku.'ten.zo] ‘goat’ ‘six’ (c) Pre-nasal liquids ['in.rx] ‘young single woman’ (d) Aspirated stops [ln ddt.ho] ‘weapon’ ['tuk.hd] ‘potato’ (e) Aspirated affricates ['kats.ha] ‘green bean’ [mtg.he] ‘clothes’ (f) Vless. nasals [ku.'tu?h.nu] ‘turkey’ Alutiiq Yupik: Stress-sensitive gemination occurs in Alutiiq Yupik, a left-to-right iambic system with main stress on the leftmost foot (Van de Vijver (1998:145-6)). Closed initial syllables are always stressed (as in f'an.ci.,qua] ‘I’ll go out’), unlike heavy syllables in the rest of the word (as in ['ap.gar.Jaa.qa] ‘I always ask her’). If the first syllable is light and the second heavy, the onset of the second syllable geminates. This closes the first syllable and allows it to be stressed (6). Related processes are discussed in section 4.3. (6) Alutiiq Yupik (from Van de Vijver (1998:145)) (a) atii [('at.) (,tii)] ‘his father’ (b) ataa [('at.) (,taa)] ‘he put it on’ 8 Here and throughout, middle tone on Popoloca vowels will not be m arked. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 53 Norton Sound Yupik. Norton Sound Yupik is also a left-to-right iambic system. Vowels in open stressed syllables are lengthened (7a). However, if the vowel is [o], vowel lengthening does not occur; rather, the onset of the following syllable is geminated and thus the syllable is closed (7b). (7) Norton Sound Yupik (I) (from Van de Vijver (1998:133, 140)) (a) qayapigkani [(qa.'jaa.) (pix.,kaa) ni] ‘his own future authentic kayak’ (b) atopik ((a.'top.) pik] ‘real name’ In a number of Yupik languages, including Norton Sound Yupik, monosyllabic stems followed by suffixes are closed in order to be stressed. If the stem syllable is open, the onset of the following syllable is geminated (Van de Vijver (1998:114-5, 130-1)) (8). For related processes see section 4.3. (8) Norton Sound Yupik (II) (Van de Vijver (1998:130-1)) Stem Stem plus suffix Gloss (a) kuva 'kuv.vuq ‘it spills’ (b) 'kuv.lu.ni ‘(it) spilling’ Gundidj: Data about this extinct Victorian language is scanty, but /ml is long and tense intervocalically after a stressed vowel. Hercus (1986:159-160) interprets this as gemination since a syllabic break occurs in this context. Gemination does not occur after unstressed vowels. Examples are /dameri/ ['tam.me.ri]‘sheep’; cf. with /gurem ug/ ['ku.ro.mok] ‘possum’. No indication is given about the regular location Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 54 of stress, but in most Victorian languages it falls on the first syllable, and that might also be the case in Gundidj. Narinari: As in Gundidj, there is not much data available for this extinct language. Ill, Ini appear to gemmate intervocalically after a stressed vowel (9) (Hercus (1986:154)). Narinari has initial stress (Hercus (1986:155)).9 (9) Gemination and lengthening in Narinari (Hercus (1986:153-154)) (a) /gali/ ['gaUi] ‘dog’ /lajurg/ ['lajurg] ‘woman’ (b) /dinag/ ['din.nag] ‘foot’ /nari/ ['na:ri] ‘no’ No examples occur with intervocalic III, Ini preceding a stressed vowel, but cf. the realization of intervocalic 1 1 1 , Ini following unstressed vowels, where gemination occurs, with the realization of word-initial III, Ini, where gemination does not occur. Wergaia: Intervocalic III is long in Wergaia, and it geminates in this position after the main stressed vowel (10a, b). Stress is initial and vowels in main stressed syllables are long. Gemination of III does not occur after lul (10c) (Hercus (1986:77, 78)). See section 5.1 for related processes. 9 Gemination of IV and Inf appears to have occurred in the same context in Wadiwadi and Ledjiledji. Evidence comes mainly through spelling, as Wadiwadi chellingoo ‘tongue’, liannoo ‘teeth’; and Ledjiledji nginna ‘you’, and jennagi, chinnangi ‘foot’ (Hercus (1986:158)). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 55 (10) Gemination of III in Wergaia (from Hercus (1986:77; IP A) (a) /w ile/ ['w ille] ‘possum’ (b) /gelalag/ [’geLLv.l'ak] ‘Major Mitchell cockatoo’ (c) /buledj/ ['bu.letj] ‘two’ Italian:1 0 Italian has a contrast between single and geminate consonants word- medially (1 la). Single consonants geminate across word boundaries provided the last syllable of the first word is open and stressed (Nespor and Vogel (1979), Chierchia (1982), Salterelli (in press), among others).1 1 This is known as raddoppiamento (fono-) sintattico (llb-d) (geminates underlined; relevant segments in bold). (11) Italian (from Nespor and Vogel (1979) and Saltarelli (in press) and p.c.) ‘shovel’ (a) /palla/ ‘ball’ vs. /pala/ (b) meta torta [me^at. 'tortal ‘half a cake’ (c) sara cotto [sa,rak. 'kotto] ‘it will be cooked’ (d) citta vecchia [tjit^av. Vekkja] ‘old city’ (e) amor mio [a,mor 'mio] ‘my love’ (f) citta sporca [tjit^tas.'porka] ‘dirty city’ (g) meta strada [me^as.'trada] ‘half way’ If the previous word ends in a consonant, gemination does not occur since this consonant is a proper coda (lie). Gemination does not occur either when the second word begins with an /s/+consonant cluster, since the /s/ resyllabifies as coda of the preceding syllable (Ilf, g). 101 am grateful to M ario Saltarelli for his thorough explanation of the Italian data. 1 1 Nespor and Vogel (1979) report a syntactic restriction for this process whereby the first word has to m-command the second word. This is empirically unconfirmed. For discussion, see Loporcaro (1999:20— 3) and Saltarelli (in press). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 56 Italian has a second type of gemination, known as ‘backwards’ or ‘inverse’ gemination. It occurs across word boundaries in loanwords, neologisms and acronysm that end in a consonant (Chierchia (1982); Saltarelli (1970, in press)). In inverse gemination the coda of a stressed syllable doubles to provide an onset for an immediately following onsetless syllable ( 12) (geminates underlined; relevant segments in bold). (12) Inverse gemination (from Saltarelli (in press); IP A) (a) tram elettrico [^ram.me.'let.tri.ko] (b) Hotel Este [o.,teLles.te] (c) Milan-Inter [lmi:.la.'nin.ter] (d) Hotel Patria [o.,tel.'pa.tija] (e) sport estivo [^por.tes.'ti.vo] ‘electric tramway’ name o f a hotel name o f a football match name o f a hotel ‘summer sport’ Inverse gemination occurs after stressed syllables (12a, b). When an unstressed syllable has a coda and an onsetless syllable immediately follows, the coda resyllabifies as onset and no gemination occurs (12c). Gemination fails to occur in two more cases: when the following syllable has an onset ( 12d) and when the last syllable of the first word has a complex coda (12e). In this last case, the second consonant of the coda resyllabifies as onset of the following syllable. Stress-sensitive gemination in Italian is related to vowel lengthening, which occurs in codaless stressed syllables: /kasa/ ['ka:.sa] ‘house’ vs. /kassa/ ['kas.sa] ‘box’ (Saltarelli (in press)). Both gemination and vowel lengthening appear to be the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 57 result o f a requirement for heavy stressed syllables in Italian (Chierchia (1986), Saltarelli (in press)). Somali: Somali also appears to have stress-sensitive gemination across words (Armstrong (1964)). Somali has four main tones: high, mid, low, and falling. In words with more than one syllable, stress falls on the syllable with high or falling pitch. If syllables have the same tone, stress is similar in all syllables. The consonant inventory of Somali is given in (13). (13) Somali consonant inventory (from Armstrong (1964))1 2 Stops Fricatives Sonorants b, t, d, 4 k, g, g , ?, tj f, s, J, x, h, S, h m, n, 1, r, j, w Somali has phonemic geminates (or double consonants) word-medially for /l, r, n, d, 4) and word-finally for /l, r, n/ (14a, b). In connected speech, a word- initial consonant (especially a voiced stop) geminates after a stressed syllable ending in a short vowel and pronounced usually with high tone (14).1 3 1 2 /t, d 1 are dental. ! < \I is more palato-alveolar than retroflex (Armstrong (1964)). 1 3 Additionally, Ini and III are reported to be weak to the point of almost eliding after long vowels—as in /saan/ ‘skin, hide (of camel, cow)’—and long and strong after short stressed vowels— -as in /sag/ ‘leather’. It is unclear from the source that this phenomenon is related to stress. It is plausible that this pattern is related to moraicity instead: when the vowel is long, it counts as two moras, and the coda consonant is weak. If the vowel is shorter, the coda is moraic, thus being pronounced as strong and long. Armstrong (1964) also reports that after long vowels, It I has fewer taps (about three) than after short vowels (four or five taps). Stress is not mentioned in this case, but it is the same environment for alternations witnessed with Ini or H I. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 58 (14) Somali (from Armstrong (1964; IP A))1 4 (a) [ke.li] ‘sin g le ’ [k e lli] ‘k id n e y ’ (b) ['baa.cjaj] ‘I searched’ ['baacjcjaj] ‘she searched’ (c) ['haawl baan ?uc[.£jllma.najjaa] ‘I am dying of hard work’ (d) ['hag.galo] (e) ['kaG .G aadj ‘Let him go in’ ‘Take from ’ Somali also has lenition of voiced stops, especially after a stressed syllable (Armstrong (1964), Lavoie (2001:36)); see section 4.4 for more details. Summary: This section discussed two types of durational processes: longer consonantal duration (and VOT) and gemination. VOT is longer in stressed syllables than in unstressed syllables. Consonants are longer in stressed syllables than in unstressed syllables. In most cases, this applies to onset consonants; but this might be because codas are restricted or missing in some languages (as in Copala Trique). Onsets immediately following stressed syllables geminate in order to provide a coda for the preceding stressed syllable. This occurs within words or across word boundaries. In the case of Italian, coda gemination across word boundaries is also attested in stressed syllables to provide an onset for the following syllable. No cases of durational processes affecting codas exclusively have been found; it is unclear that this is a significant gap. Most of the languages with consonant or VOT lengthening and gemination show additional stress-related durational effects, as shown in the following sections. 1 4 Armstrong (1964) is not paginated, so the page numbers are not included here. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 59 4.2 Flapping Flapping involves the shortening of the occlusion of a segment and possibly voicing. It can occur without sensitivity to stress, as in Tagalog (Schachter and Qtares (1972)). Flapping occurs in various languages in unstressed syllables, as in American and Canadian English (discussed together), Senoufo, Kupia, and Djabugai. Table 2.2. Flapping 15 American English 0 It, d, n/-^[r]/'V_V (obligatory); V_V (opt.) Germanic Canadian English o It, d, n/->[f]/'V_V (opt.) Germanic Djabugay 0 /r/-»[r]/'V_V Pama-Nyungan Senoufo 0 /d/-»[r]/._V Niger-Congo Kupia 0 /{/ [f]/V._V (opt.) Indo-Iranian American and Canadian English: Flapping affects /t/, /d/, and Ini in onset of medial unstressed syllables (Kiparsky (1979), Kahn (1980), Borowsky (1986), Hammond (1999), among others). Lavoie (2001) reports that flapping of Ini is more ambiguous than flapping for It, d/, since the tongue-tip gesture is reduced but not the velum gesture. Other segments are reported to be flapped; Turk (1992, 1993) and references therein report shortening of all oral stops except /g/ in flapping environments. Turk (1992) interprets these results as probably meaning that flapping in English is caused by an ‘underlying timing mechanism’ (Turk (1992:128)). 1 5 Another possible instance of stress-sensitive flapping is Pattani (Sino-Tibetan), where lc\l is realized as l\j in medial and final position. Sarm a (1982) does not discuss stress in this process, but stress generally falls on the first syllable in Pattani. As with other languages in Table 2, Pattani shows stronger aspiration in stressed syllables (see section 7). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 60 Flapping typically occurs intervocalically, but it can also be found after a sonorant and before a vowel (forty, tenting). It typically occurs word-medially after a stressed vowel and before an unstressed vowel (metal, po'tato, 'rider, 'winter)}6 Flapping is optional between two unstressed vowels (as in pro'vocative, farody). Flapping is also optional word-finally between two vowels regardless of stress (not at 'all, 'private 'airplane). Flapping does not occur in stressed syllables (a'ttack), in non-final codas (atkins, 'atlas), or word-initially (take, to ‘ morrow) regardless of stress (Kahn (1980), Turk (1992)).1 7 Flapping in English is not a categorical process; the probability of its occurrence depends both on stress and phonetic context. Zue and Laferriere (1979) found that flapping of N in a 'V_V context occurs 99% of the time. The probability of flapping for Ixl between two unstressed or reduced vowels lowers to 33%; aspiration of /t/ is more common (66%) in this context. In a context like 'Vn._V, III has a probability of flapping of 14%; this is the same probability that the III deletes in this context. A realization with aspiration is far more common (67% with /n/, and 5% with nasalization of the preceding vowel and nasal deletion). Results for /d/ are less clear, since unstressed /d/ has longer closure than a flap but no release, or both a short duration and a strong release (see Zue and Laferriere (1979) for details). 1 6 The N in winter deletes concomitant with the flapping of Ini. 1 7 Flapping of III is reported to occur in Cardiff English in medial position between voiced sounds, as in better ['bs.ra] and in hospital f'as.pi.rl] (Collins and Mees (1990:90)). In this position /d/ lenites to fricative, as Pol and /g/ do. It is possible that these patterns have the same stress or foot conditioning as flapping in American and Canadian English. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 61 Some authors suggest that flapping in English is foot-sensitive (Kiparsky (1979), among others). English is a trochaic language, and flapping would occur foot-medially since the second syllable of a binary trochaic foot is unstressed. Alternatively, flapping would be the compensatory shortening for the occurrence of a long stressed vowel occurring earlier in the foot (Turk (1992)). Flapping in Canadian English is similar to American English, but it is considered to be more optional (de Wolf and Hasebe-Ludt (1987)). Djabugai: /r/ is pronounced as [r] between a stressed and unstressed vowel; the word /garaj/ ‘to come’ is pronounced ['ga.raj] (Patz (1991:253)). Patz (1991:253) notes that reduction to [r] is more obvious between two low /a/s, less obvious between identical high vowels (as in /biri/ ‘near’) and least obvious between different vowels (as in /dira/ ‘teeth’).1 8 In this last position [r] and [r] are in free variation. It is plausible that in Djabugai flapping actually consists in the reduction of number of taps in unstressed position, as in Mexican Spanish, where trills have fewer taps in unstressed positions than in stressed positions (Lavoie (2001:143)). See 5.1 for related processes in Djabugai. 1 8 Djabugai has a three-way vowel contrast /a, i, u/. For more details on the phonological system of Djabugai, see section 5.1. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 62 Senoufo: Idl is flapped word-medially in the onset of unstressed syllables (Mills (1984)). For related processes see sections 4.1, 4.4, and 5.1. Kupia: The retroflex coronal !\j is optionally flapped intervocalically in onsets of unstressed syllables; e.g. /ca:tu/ ‘leg’ can be both ['tsai.^u] or ['tsa:.£u]. In contrast, /c[/ is always flapped intervocalically (Christmas and Christinas (1975:18)).1 9 For related processes see section 4.4. Summary: Stress-sensitive flapping occurs word-medially in onsets of unstressed syllables. In some languages the preceding vowel has to be stressed. In the case of English, where flapping has been best studied, flapping occurs more often between vowels, especially when the first vowel is stressed; but it is also possible in onset of unstressed syllables, after a coda sonorant in a stressed syllable. Languages with stress-sensitive flapping frequently have additional stress-related consonantal processes such as spirantization, longer duration of consonants or VOT, and voicing differences. Typical segments undergoing stress-sensitive flapping are III (in English varieties), Id/ (in English and Senoufo), /nJ (in English), /r/ (in Djabugai), and /(/ (in Kupia). All of these segments have a tongue-tip articulator in common. 1 9 Kirchner (1998:308) reports the environment of /cf/ as obligatory in onset of unstressed syllables, but C hristm as and Christmas (1975:18) do not mention stress as being relevant in this case. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 63 4.3 Fortition In this dissertation the term fortition will be used to refer to a decrease in consonant sonority, and lenition to an increase in consonant sonority. Fortition makes a consonant more consonantal and thus ‘stronger’, and lenition makes a consonant more vocalic and thus ‘weaker’ (Vennemann (1988)). Usually the identification of a process as fortition or lenition is based on what the analysis posits as the basic or underlying form. If the basic form is more sonorous than the non-basic form, the process falls under ‘fortition’. If the basic form is less sonorous than the non-basic form, then the process involves ‘lenition’. Thus, which allophones are ‘stronger’ and ‘weaker’ is relative. Fricativization of approximants, occlusivization of fricatives, affrication of stops, gemination of singletons, and devoicing of voiced segments are considered instances of fortition because they involve a decrease in sonority (Lavoie (2001), Kirchner (1998)). In this survey, gemination is discussed in section 4.1 under durational differences, and devoicing is discussed in section 5 under featural timing. Fortition may be morphologically conditioned, as in Irish. A number of fortition cases are reported to be conditioned by stress and foot structure. These include fricativization in Norton Sound, Alutiiq Yupik and Guayabero; affrication in Maori and West Tarangan; occlusivization in Yuman and West Tarangan; and strengthening of 11 1 in Farsi and Squamish. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 64 Table 2.3. Fortition Norton Sound Yupik 0 /w, j, l/-> [v, z, 11/foot-initially Yupik Alutiiq Yupik 0 Consonants fortify # and foot-initially Yupik Guayabero 0 [w] fricativizes in' 0 Guahiban Maori 0 /p, t, k/-> affficate/'_V Austronesian West Tarangan 0 /w,j/-»[g, d3]/#_, '_V Austronesian Farsi 0 /?/ stronger plosion/'jV Indo-Iranian Squamish 0 /R? / stronger [? ] when stressed/post-stressed Central Salish * Yunnan 0 /v/->/p/, /j/->/d/ in a Hokan Norton Sound Yupik: This dialect of Central Alaskan Yupik shows foot-sensitive fortition of fricatives. Underlying /w, j, 1/ are fortified to [v, z, lg] in light syllables after a stressed closed syllable (Jacobson (1985:32), Van de Vijver (1998:43)). [v, z, I5] are ‘fortis’ fricatives and [w, j, 1 ] ‘lenis’ fricatives; here I will refer to them as fricatives and approximants respectively.2 0 Norton Sound Yupik is a left-to-right iambic system with main stress in the leftmost foot and no final stress (Van de Vijver (1998:125-132)). Closed syllables are light except word-initially; syllables with long vowels are heavy. The stress and foot pattern of Norton Sound is also discussed in Jacobson (1985), Leer (1985), Kager (1993), Hayes (1995), and Bakovic (1996). Van de Vijver (1998) analyzes the context for occurrence of the fricative variants as foot-initial position.2 1 Some examples are given in (15). 2 0 According to Jacobson (1985) [v, z, I 5 ] have a stronge fricative nature while [w, j, 1 ] are more sonorant. 2 1 But cf. Van de Vijver (1998:132), where it is reported that lenis consonants occur foot medially and fortis consonants elsewhere [i.e., fortis can occur foot-fmally in addition to foot-initially]. If this is so, this pattern would be an instance of foot-medial lenition rather than foot-initial fortition. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 65 (15) Norton Sound /w, 1, j/ (from Van de Vijver (1998:43-4, 132)) Fricatives Approximants (a) (ma.'juv) (vik) ('at.) (xar. wik) ‘place to go up/ down’ (b) (m a .'ju Y ) (z u x .tu q ) ('at.) ( x a y .j u x .) tu q ‘h e w an ts to g o u p /d o w n ’ (c) (m a .'ju Y ) (fe u .n i) ('at.) (x a Y .,iu .) n i ‘(h e) g o in g u p /d o w n ’22 Since approximants can occur medially in onsets of unstressed syllables (15d), and fricatives can occur in unstressed syllables (15a-c), fortition is not stress-sensitive, but foot-sensitive. Fricative allophones occur foot-initially and approximant allophones foot-medially. Since Norton Sound is an iambic system, there is a mismatch in the distribution of fortition for stress and foot structure; fortition occurs in foot-initial position, which is always unstressed; fortition does not occur foot medially, which is usually—but not always— a stressed position. This mismatch is expected in iambic systems, since foot boundaries and stress do not correspond for processes involving onsets; there are two competing strong positions: stressed syllables, and foot-initial syllables. In trochaic systems there is no mismatch, since stress falls on foot-initial syllables. This is exemplified schematically in Figure 2.1. Norton Sound Yupik is also discussed in more detail in chapter 3. For related processes, see section 4.1. 2 2 According to Van de Vijver (1998:126), the fact that the fricative in (15d) is lenis indicates that the final syllable is footed. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 66 Figure 2.1. Stress and footing in onsets of trochaic and iambic systems (repeated _____________from Figure 1.1)______________________________________________ Tr o c h a ic Ia m b ic Foot-initial Foot-medial Foot-initial Foot-medial ('a . c) ( a . 'o) Stressed Unstressed Unstressed Stressed Alutiiq Yupik: Foot-initial fortition is also found in the Alutiiq dialects o f Yupik (Koniag and Chugach). Alutiiq Yupik is a left-to-right iambic system with main stress in the leftmost foot. As in Norton Sound, closed syllables are heavy only when word-initial (Cf. 16b, d). Syllables with long vowels or diphthongs are heavy and stressed (16b) (Van de Vijver (1998:141-152)). Fortition occurs for all onset consonants and involves ‘pre-closure’ and complete lack of voicing for voiceless consonants. Fortition makes consonants ‘more audible’ (Van de Vijver (1998:143), based on Leer (1985, 1989)). According to Van de Vijver (1998) and references therein, fortition occurs both foot-initially and word- initially (16) (fortis consonants are in bold). (16) Alutiiq Yupik (from Van de Vijver (1998:143-4, 176); based on Leer (1985)) (a) taqumaluni (ta.'qu.) ma. (lu.,ni) ‘apparently getting it done’ (b) anciqua [('an.) ci.(,qua)] ‘I’ll go out’ (c) akutaq [(a.'ku.) taq] A food (d) iqllunirtuq [('iq).'hi.(nix.'tuq)] ‘He stopped lying’ (e) apqarlaaqa [('ap). qar.(laa).qa] T always ask her’ (f) apqarlaanka [(ap).qar.(lan).ka] ‘I always ask them’ (g) tannorlirsuqutaquni [(tan).nox.(lix.su.) qu. (ta.qu.) ni] ‘If he (refl.) is going to hunt bear’ (h) kumlaciwilijaqutaquniki ‘if he (refl.) is going to undertake constructing a freezer’ Koniag speakers [('kum.) (la.,ci.) (wi.,li.) ja. (qu.,ta.) qu. (ni^ki)] Chugagh speakers [('kum.) la.(ci.,wi).li.(ja. ,qu). ta. (qu. ,ni.) ki] Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 67 Fortition occurs foot-initially in disyllabic feet with light syllables (16a), in monosyllabic heavy feet (16b), in feet formed by two closed syllables (16d) and in monosyllabic (but bimoraic) feet even if the vowel is phonetically shortened (16f). Fortition does not occur in unfooted closed syllables (16c). (16h) compares footing and fortition in Koniag and Chugag. Even is stress assignment and footing differ in both dialects, fortis consonants correlate with foot-initial position in each dialect. Fortition does not correlate with stress, since it can occur in stressed and unstressed syllables alike (16a, b). For Van de Vijver (1998:143), stressed heavy syllables (as in 16b) are not footed but form a stress unit where fortition can apply. I assume stressed heavy syllables are footed under a binary foot analysis and thus, fortition occurs in foot- initial position in all cases. Van de Vijver (1998) also assumes that word-initial closed syllables are footed with a catalectic syllable. This would make the heavy syllable foot-medial; fortition still applies in this case. Under a footing where this syllable forms its own foot ‘word initial’ fortition falls under ‘foot initial’ fortition. Summarizing, fortition in Alutiik Yupik, as in Norton Sound Yupik, correlates with foot structure rather than stress. For related processes, see section 4.1. Guayabero: Stress-sensitive fricativization of /w/ is reported in Guayabero (Keels (1985); see also Lavoie (2001), Kirchner (1998)). The consonant inventory of this language includes /p, b, t, d, k, ?, < J > , s, x, h, tf, m, n, r, 1, w, j/ (Keels (1985:58)). All consonants except lb, j/ can occur word-fmally and syllable-finally in clusters. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 68 Clusters occur in stressed syllables only. Stress is not predictable and falls on the last or penultimate syllable of the stem. /w/ fficativizes to [J 3 ] before a stressed front vowel, and after a front or high central vowel (Keels (1985:75)). Keels (1985:61, 64) reports that /w/ has a bilabial fricative allophone ‘contiguous to stressed front vowels’. Additionally, /w/ is realized as a high back rounded offglide [u ] ‘in a syllable-final margin’ after a back or low vowel (Keels (1985:75)). In other contexts, /w/ is pronounced as [w] (17). (17) Fricativization. of/w / to [J J ] (from Keels (1985:64, 73-5); IPA) (a) wiam I 'r m ] ‘whip-worms’ (b) nawel [nA.'Jtel] ‘the Guaviare River’ (c) wat ['wat] ‘older brother’ (d) nehxaehw-as [neh.'xaeh.wAs] ‘he is saddened’ (e) babeh-tjiw [bA .'beh.tjiJJ] ‘evil men’ (f) puewa ['phV §A ] ‘all’ (g) sasi?w-it [sA.'tJi?.glt] ‘a hummingbird’ (h) now-x ['nwo~x] ‘I cry’ These examples show that in onset preceding a front vowel /w/ is fricativized (17a, b); in onsets before a non-front vowel or onsets of unstressed vowels fricativization does not occur (17c, d). /w/ is also fricativized in coda and in onset after a front vowel (17e, f). In coda after a round vowel, /w/ is realized as [u ] (17f, h). In (17g), fricativization occurs in onset of an unstressed syllable preceding a front vowel; however, there is a front vowel in the preceding syllable, which is probably causing Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 69 fricativization. Unfortunately, no examples are provided of onset /w/ in unstressed syllables before a front vowel where a preceding front or central vowel does not influence the choice of allophone. Guayabero has also stress-sensitive allophones of /?, h/, which ‘may be realized with voiced or voiceless partial rearticulation of the preceding vowel’ in coda of stressed syllables (Keels (1985:72, 74)) (18). (18) Guayabero: allophonic laryngeal consonants (Keels (1985:72, 74)) (a) b ia?-t [ ' b i T t ] , ['b i3?t] ‘a to o th ’ (b) co ? -t [ ' t r o 2 ° t ] , [ ' t r o ? t ] ‘an ear’ (c) k o?tiya [k ho?.'thi.jA ] ‘fin g e rs’ (d) p a ? iw la x [p hA .'?l.p iA x] ‘je a lo u s y ’ (e) m o ih t ['m o .'h t], ['m o .’h't] ‘h e sle e p s ’ Additionally, in unstressed syllables, /h/ is ‘partially or completely coarticulated with the unstressed final vowel of a preceding vowel cluster if /h/ occurs syllable finally or when the vowel has the same quality as the ensuing vowel if /h/ occurs syllable initially’ (Keels (1985:74)). In (19), coarticulation is indicated with a tiebar ‘ ’. Intervocalic /h/ can have light voicing (Keels (1985:74)). (19) Guayabero: Coarticulation of /h/ (Keels (1985:74)). (a) biaha [ ’b i O i A ] , ['b iA .f iA ] ‘he swallows them’ (b) b iah im ['b iA .h im ], ['b iA .fiim ] ‘s w a llo w th e m !’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 70 Finally, before stressed high central vowels followed by [?], /hi freely varies with a bilabial implosive (20) (Keels (1985:72)). For related processes, see section 4.4. (20) Guayabero: Allophonic variation of /b/ (Keels (1985:72) (a) bia?a-t ['bi3.?At], ['6i3.?At] ‘a shelter’ (b) tabit [t hA .'bit] ‘an e g g ’ Maori: Bauer (1993:530) reports that /p, t, k/ in the onset of a stressed syllable may have stronger aspiration and even be affficated (21) (cf. Lavoie (2001:45)). Affrication is more common ‘in the environment of a high front vowel’ for /p, t/, and before /a/ for /k/ (Bauer (1993:530)). Affrication is also possible for III before devoiced final H I and lul, which are unstressed (Bauer (1993:530)). (21) Maori: affrication in onset of stressed syllable (Bauer (1993:531)) (a) piu ['pg(i)u] ‘swing’ (b) karanga ['kxn.rn.nu] ‘call’2 3 (c) iti fi.tsi], ['i.tgi] ‘small’ Maori has ten consonants—/p, t, k, m, n, r), f, h, r,w/—and five vowels /i, e, a, u, o/.24 Codas and consonant clusters are not permitted. For stress in mono- morphemic words syllables with long vowels take precedence over diphthongs, and 2 3 ‘The canonical realization of short /a/ is a slightly retracted low central unrounded vowel, i.e. [n].' (Bauer (1993:540)). 2 4 There is a controversy about whether vowel length is phonemic in M aori; see Bauer (1993:429 ff). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 71 these over short vowels. If there is a tie, the syllable to the left is stressed. For prefixed words, the prefix usually has main stress (Bauer (1993:555 ff.) and references therein). Maori has additional stress-related consonantal processes: The most constant correlates of primary word stress in Maori are pitch fall and length of vocalic element (the first element in a diphthong). These features are often accompanied by emphatic onset, which consists of a slight pause preceding the stressed syllable, and the appropriate one of the following: aspiration or affrication of stops; pre-glottalization of vowels; stronger friction for fricatives; longer contact for /r/; closer approximantion for /w/; lengthening for nasals. These features of stress may also be accompanied by increased loudness (Bauer (1993:555-6)). West Tarangan: Word-initially and in medial stressed syllables /j/ affricates and /w/ occlusivizes (Nivens (1992)) (22).2 5 The consonantal inventory of this language is /b, t, d, k, s, m, n, p, r, 1 , j, w /.2 6 West Tarangan has right-to-left moraic trochees. (22) Fortition in West Tarangan (from Nivens (1992:139-140; IP A)) (a) (b) (c) (d) (e) (f) (g) /w ow a/ /w aym oj+na/ /jabin/ /jirua/ /suwakan/ /bijarum / /kaw ar/ ['go.wo] [gaj.'moj.no] ['d3a.bin] [dji.'m o] [su.'ga.kon] [bi.'dga.rum ] ['ka.w or] ‘blossom’ ‘slow-3s’ ‘many’ ‘tree (sp)’ ‘elephant tusk’ ‘whale’ ‘fish (sp)’ 2 5 Lavoie (2001:36) reports this process as approximantization of /g, dz/. I follow Nivens’ (1992) original classification of this process as fortition and /j, w/ as underlying. 2 6 All coronal segments are alveolar except /t/ (dental) and /j/ (palatal). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 72 (h) /raj an/ (i) /i+jow/ (j) /roraw / [’ ra.jan] [i'dzow] f'ro.raw] ‘boat shelter’ ‘3s-see’ ‘heat’ In word-initial onsets /w, j/ are pronounced as [g, dg] regardless of stress (22a-d). /w, j/ are also pronounced as [g, dg] in onsets of medial stressed syllables (22e, f). Fortition does not occur in onsets of medial unstressed syllables and in coda position regardless of stress or word position (22g— j).2 7 Farsi: /?/ has ‘emphatic and vigorous’ plosion in onsets of initial stressed syllables (Samareh (1977:17)). /?/ can also occur as coda. See also sections 5.1 and 5.2.1. Squamish: Globalized resonants are reported to have stronger globalization in stressed and post-stressed syllables than in pre-stressed syllables. Globalization can involve post or pre-globalization or a brief interruption of the resonant (Kuipers (1967:25). See also section 7.1. Yuman: Historically, Proto-Yuman /v/ became /p/ or /w/, and /j/ became /d/ in stress-sensitive environments in a number of Yuman languages (Wares (1968); see also Lavoie (2001)).2 8 Consider the examples below for /v/: 2 7 No examples of words with coda /w/ word-initially are reported in the source. 2 8 It is unclear from the source what /d/ stands for. Since /j/ is an approximant and this process involves fortition, it is plausible that 1 4 1 stands for a voiced fricative. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 73 (23) Fortition in Yuman (Wares (1968:59)) Dialect with no change Dialect with change Gloss (a) ?a.'va (Mohave) ?u.'wa* (Diegueno) ‘house’ (b) vi.'ce (Paipai) pi.'ci (Kiliwa) ‘daughter’ (c) ra'.vi (Ya) rap (Taipai) ‘to hurt (intr.)’ /v/ gave rise to /w/ ‘immediately before the stressed vowel’ in Cocopa, Diegueno, Taipai, Paipai, North Yuman and Kiliwa (23a). In Cocopa, Taipai, Diegueno and Kiliwa /v/ turned into /p/ ‘before stressed vowel, but not immediately before’ (23b) and ‘immediately following stressed syllable’ (23c). The relevant contrast seems to be that /v/ became /w/ in onset of stressed syllables, but /p/ in onset of unstressed syllables (and possibly in codas; but codas probably arose from the loss of unstressed final vowels (23 c)). Proto-Yuman /j/ became 1 41 in onset of stressed syllables in Yuman, Mohave and Maricopa, and it remained /j/ in onset after stressed syllables. In Diegueno and Taipai /j/ gave rise to /?/ in onset of stressed syllables, and in Kiliwa to Ihl (24).2 9 (24) Development of Proto-Yuman /j/ (Wares (1968:67)) Dialect with no change Dialect with change Gloss (a) i.'ja (Cocopa) i’.'do’ (Mohave) ‘tooth’ (b) ^ ja (Paipai) '?a (Taipai); 'ha? (Kiliwa) ‘mouth’ (c) ma.sa.'ha.ja (Yavapai) ma.sa.'haj (Mohave) ‘girl, virgin' 2 9 Additionally, 1 1 1 became a retroflex l\l in Yuma in onset after stressed syllables (Wares (1968:54)). One example is ‘backbone’, realized as [ax.tafl in Yuma but as [ax.tat] in Maricopa. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 74 Summary: Fortition includes fricativization of approximants, occlusivization and affrication of fricatives, and stronger plosion. In most cases, fortition occurs in onset of stressed syllables. However, the absence of coda fortition in some languages might be the result of the lack or restriction of codas. Fortition occurs in foot-initial position in some languages. In iambic languages, fortition occurs foot-initially in unstressed syllables to mark foot boundaries. Diachronic fortition in Yuman deserves more investigation. 4 .4 Lenition Lenition refers to an increase in consonant sonority which makes a consonant more vocalic and thus ‘weaker’ (Vennemann (1988)). Phenomena considered under the rubric of lenition include spirantization, by which a stop is pronounced as a fricative or approximant; approximantization of fricatives, and debuccalization, by which supra-laryngeal features are lost. Voicing of underlying voiceless segments in weak positions is also considered to be lenition, since voicing increases the sonority of a consonant. In this study voicing variation is discussed under featural timing because the examination of languages with stress-sensitive voicing in this survey shows that both voicing and devoicing might occur in weak positions (section 5.1). Stress- and foot-sensitive lenition processes discussed here include debuccalization in English, fricativization in Copala Trique, Guayabero and Senoufo; approximantization in Somali and in Spanish, and general weakening in Kupia, Nganasan, Paamese and Silacayoapan Mixteco. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 75 Table 2.4. Lenition English 0 /t/->[?]/ 'V_V; /t/-»[h]/'V.CV_# Germanic Copala Trique 0 Id, g/->[5, r]/V ‘ _V Otomanguean Guayabero c /d/-> [6]/'V_ Guahiban Senoufo 0 Medial weakest vd. stops fricativize Niger-Congo Somali 0 lb, d, g, g/->[|3, 5, r , k]/ V_V, (esp. 'V_V) Cushitic Spanish 0 lb, d, g/-» [p, 3, v]/d Romance Kupia 0 /p/-> p/ V._V in o (opt.) Indo-Iranian Nganasan 0 /h, t, k, s, 9/-> [b, d, g, j, j]/foot-irdtially Uralic Paamese 0 /p/->[f], /t/-» [r]/_'CV; /f/-» [p]/ CV (red.) Austronesian Silacayoapan Mixteco 0 /t/->[d], /3/-> [j] in weak o Otomanguean English: In some dialects of English III is flapped intervocalically in the onset of an unstressed syllable (section 4.2.1). In some other dialects, especially in Britain, III is weakened to a glottal stop in the same context (Roca and Johnson (1999)). One specific instance where this occurs is London English (25) (Carr (1999:152)).3 0 (25) English glottalization of III (from Roca and Johnson (1999:315-6)) (a) 'ten ta'basco 'mastery 'after (b) 'cutting [kA.?irj] 'w aitin g ['w e i.? ip ] (c) a'ttain re'tort Lenition to [?] does not occur word-initially or immediately following a consonant (25a). Lenition to [?] occurs intervocalically in unstressed syllables (cf. 25b, c). 3 0 Ladefoged (1993:53) mentions that in Cockney English, words like butter, kitty and fatter are pronounced as ['bA ?3, 'ki?i, 'fae?3]. He does not refer specifically to stress, but in these words III is realized as [?] intervocalically in unstressed position. He also mentions that the realization of III as [?] is extended in both American and British dialects in words like beaten, kitten, fatten ['bi?n, 'ki?n, 'fo?n]. These particular syllabic nasals are unstressed. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 76 Some dialects of English have a related stress-sensitive process by which /p, t, k/ are pre-glottalized (section 5.3.2). In Liverpool English, where III glottalizing is not common, /t/ is reported to be debuccalized to [h] utterance-finally. Debuccalization occurs most frequently in monosyllabic words, especially function words such as it [ih], or that [5ah], In multi-syllabic words, debuccalization usually occurs in unstressed syllables (Honeybone and Watson (2002)). Some examples are climate ['klai.mih], market ['mai.kih], permit (noun) ['ps'.mih], and permit (verb) [po.'mit]. In the last case debuccalization does not occur because the last syllable is stressed.3 1 In all cases of debuccalization, the preceding vowel must be ‘short’ (i.e., glottalization can occur in words like it [ih] or get [get], but not in words like wait *[weih], out *[auh] or eat *[i:h]) (Honeybone and Watson (2002)). Copala Trique: /d, g/ are reported to lenite to fricatives intervocalically in onsets of stressed syllables (Hollenbach (1977:36)). /b/ is never found intervocalically in this position. Initially in monosyllabic words /d, g/ are pronounced as voiced lenis stops. In other positions they can be voiced or voiceless (section 5.1). See sections 4.1 and 5.1 for related processes in this language. 3 1 But stress does not always block debuccalization, as in forget [fo.'geh] and forgot [fo.'got] (Honeybone and W atson (2002)). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 77 Guayabero: In coda of stressed syllables, /d/ is pronounced as [0], In onset of both stressed and unstressed syllables, /d/ is pronounced as [d] (26a— c); in coda of stressed syllable /d/ is fricativized (Keels (1985:72)). (26) Guayabero: Fricativization of /d/ (from Keels (1985:72); IP A) (a) dal ['dal] ‘fat’ (b) nehtfadit ^neh.tja.'dit] ‘he remembers’ (c) nan-de ['nan.de] ‘stand up!’ (d) mid ['nw u0] ‘an ocelot, a jaguar’ (e) bend-hit ['benO.hit] ‘wax’ Keels (1985) mentions that /d/ never surfaces in coda of unstressed syllables, and suggests that the distribution of [d/0] is position-sensitive rather than stress-sensitive. However, this distribution might be similar to frication of coda /b, d, g/ in Basque Spanish. In this dialect of Spanish, most words with /b, d, g/ in coda are stressed in the syllable containing lb, d, g/; however, acoustic data shows that voiceless fricative realizations are more common in stressed syllables. See chapter 4 for details. Senoufo: Onset consonants are lenited in weak syllables. The consonant inventory is /p, b, t, d, k, g, kp, gb, f, v, s, z, m , (n), (p), (q), 1, j, w/ (Mills (1984:92)). Stress falls on the first syllable of the word and there are no secondary stresses. Stress is realized as long and fortis articulation of the onset, and as intense Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 78 articulation of that syllable. Vowels are not lengthened because of word stress. Mills (1984) distinguishes among three degrees of length/strength for consonants: fortis, lenis and extra-lenis. Fortis consonants occur word-initially in stressed syllables. Lenis consonants occur initially after a pre-final semi-open juncture in unstressed syllables. Extra-lenis consonants occur in medial unstressed syllables in polysyllabic words. Extra-lenis consonants have the shortest duration and undergo different reduction processes, /b, d, g/ are spirantized in medial unstressed syllables (cf. Lavoie (2001:35)). /d/ is flapped in this position (section 4.2), and nasals are ‘slurred over’ (Mills (1984:131)). Additionally, voiceless stops have voiced allophones in this position; see section 5.1. Somali'. Somali has the stop series fb, t, d, 4 k, g, g, ?/. Voiced and voiceless stops are mainly distinguished by aspiration rather than voicing, much as in English (Armstrong (1964)). Voiced stops /b, d, 4 g, g/ frequently devoice word- finally; /b, d, g/ have either not much voicing or none word-initially. Intervocalically, fb, d, g, g/ lenite to [J 3 , 5, r , k] when single between vowels, especially after stressed syllables {fb, d, g/ often; /g/ sometimes) (Armstrong (1964)). Thus, in a word like ['c}a.ban] ‘cheek’, where Pol occurs in an unstressed syllable, lenition is more common than in a word like /da.'bool/ ‘cover’, where lb/ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 79 occurs in a stressed syllable. Although lenition is not restricted to unstressed syllables, it is more common there.3 2 For related processes see section 4.1. 33 Spanish: In Colombian Spanish /b, d, g/ tend to be pronounced as voiced stops rather as approximants in onsets (cf. chapter 4). This is especially the case in stressed syllables (27d-e) (Amastae (1986), Kim (2002)). In unstressed syllables fb, d, g/ tend to be pronounced as voiced approximants (27a-c) even after nasals, which typically block lenition in Spanish (27a). (27) Colombian Spanish: lenition (from Kim (2002)) Onset o f unstressed syllables Onset of stressed syllables (a) donde ['don.de] ‘where’ (d) andar [an.'dar] ‘to walk-inf.’ (b) habla ['a.JM a] ‘he talks’ (e) hablar [a.'blar] ‘to talk-inf (c) fiebre ['fje.J3re] ‘fever’ (f) quebrar [ke.'brar] ‘to break-inf.’ In a study investigating the effect of prosodic and segmental context for /g/ in Castilian Spanish, Cole, Hualde and Iskarous (1998) found that onset /g/ in intervocalic position had a more complete closure and was more resistant to 3 2 Lavoie (2001:36) reports flapping of /c|/ in Somali in similar contexts as spirantization of fb, d, g/. However, Armstrong (1964) does not mention stress-sensitivity for flapping of this phoneme; her examples include flapping in both stressed and unstressed syllables, as in /hacjuno/ ‘dinner’ and /ticji/ ‘she said’, where the underlined segments can in the two cases be realized as [{j. 3 3 Thanks to Kie Zuraw for making available Kim (2002), and to Jose Ignacio Hualde and Khalil Iskarous for bringing to my attention and making available their work. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 80 spirantization in stressed syllables than in unstressed syllables, where /g/ had more energy. For further discussion on lenition in Spanish see chapter 4. Kupia: According to Christmas and Christmas (1975:18) intervocalic consonants in Kupia are generally more lenis in non-prominent syllables than word- initial consonants, /p, cj/ are singled out as having distinctive variants in this position, /p/ is optionally realized as lenis intervocalically in onset of unstressed syllables (28). This source does not mention if the preceding vowel has to be stressed or not. The consonantal inventory of Kupia includes /p, b, t, d, 4 k, 9, m, n, g, r, 1, j, w, s, ts, dz/ (Christmas and Christmas (1975:5)). (28) Kupia: Optional lenition of /p/ (from Christmas and Christmas (1975:18)) (a) /ku:pija/ ['kui.pFa], ['kui.pFa] ‘Kupia’ (b) /su:pu/ ['su:.pu], [’su:.pu] ‘winnowing basket’ The realization of lenis /p/—which is represented with [p] in the source—is unclear. Kirchner (1998) represents this segment as [$]. It is possible that the lenis allophone presents a change in voicing. Kupia also has optional flapping in this very context for /(/; see section 4.2 for more details. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Nganasan:3 4 Nganasan rhythmic consonantal gradation appears to be foot- sensitive. Rhythmic consonantal gradation consists in the alternation of specific consonants word-medially. It is clearly seen in suffixes, but historically it is also attested in roots (Helimsky (1998)). The consonant inventory of Nganasan is fb, t, d, c, j, k, g, ?, s, g, x, m, n, ji, g, 1 , A, r/ (Helimsky (1998:483); IP A).3 5 /j/ and id/ have [j] and [8] as allophones, respectively, [p] is an allophone of either ih/ or /hi (in Nganasan, *p>h). The consonants that undergo rhythmic gradation are given in (29). Note that strong grade consonants are voiceless, and weak grade consonants are voiced. This detail will be crucial in the discussion of Nganasan in chapter 3. Examples of rhythmic gradation for the suffixes {tui, 6i} 3rd sg. possessive and {ra.kxu, ro.gui} Similative are given in (30). (29) Consonant gradation in Nganasan (He imsky (1998), Vaysman (2002))3 6 Strong grade h t k s 9 Weak grade b d/8 g 3 J (30) Rhythmic gradation (from Helimsky (1998:490); IP A) (a) (b) (c) (d) Strong grade 'nui.-tui ‘his wife’ huia.'js.-tm ‘his thumb’ Weak grade bm.'nm.-di ‘his rope’ ko.rra.ga.Xi-Si ‘his march’ bui.nui.- ra.kui ‘similar to a rope’ nui.'-ro.gui ‘similar to a woman’ k3.rui. ga.Ai.-'ra.kui ‘his march’ hina.js.-'ra.gm ‘similar to a thumb’ 3 4 Thanks for Sharon Inkelas to bringing this language to my attention. 3 5 It, d, s, n, 1 , r/ are dental. 36 , ’ Clusters of these consonants with nasals also undergo rhythmic gradation; see Helimsky (1998). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 82 The strong grade occurs after an odd number of vowels (first column in 30) and the weak grade after an even number of vowels (second column in 30) (Helimsky (1998)). In Nganasan, main stress falls ‘on the penultimate vowel or on the vocalic sequence which includes the penultimate vowel’ (Helimsky (1998:486)) (31, b). There is a strong preference for canonical trochaic words constructed with left-to- right syllabic trochees (31c-e). Words with five or more syllables are divided into rhythmic groups with two vowels each; the last group can have two or three syllables (Helimsky (1998:487)). Stems and suffixes tend to be disyllabic. The boundaries between rhythmic groups are very distinct, and sometimes ‘the break in articulation creates the acoustic impression of a glottal stop’ (Helimsky (1998:487)). (31) Nganasan stress and foot structure (from Helimsky (1998:486-7); IPA, syllabification mine) (a) ['ko.ru?] ‘house’ [ko.'ru.So?] ‘houses’ (b) [kjy.'ma.a] ‘knife’3 8 (c) (.ko.rui.) (,ga.Ai.) (,ti.ni) ‘in marches’ (d) (jko.rui.) (,ga.Ai.) (ti. 'ni.no) ‘in my marches’ (e) (,ko.rui.) (,ga.Ai.) (,nn.a.) (ti. 'ni.no) ‘only in my marches’ The description of the environment for rhythmic gradation coincides with the foot structure of the language: the strong grade occurs foot-medially, and the weak grade 3 7 Stress is optionally retracted from a high vowel or [s] to the preceding syllable, which is usually an open syllable: [ba.'ru.sji]~['ba.m.sji] ‘devil’ (Helimsky (1998:486); see also de Lacy (2002)). 3 8 Helimsky (1998) considers long vowels and dipththongs as vocalic sequences (Helimsky (1998:485 ff)). The foot structure for (3 le) suggests that two contiguous vowels are syllabified heterosyllabically; the syllabification in (31b) models this example. For a different approach in which long vowels are tautosyllabic in Nganasan, see de Lacy (2002). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 83 occurs foot-initially, in left-to-right syllabic trochees. For a more detailed discussion see chapter 3.3 9 Paamese:4 0 A puzzling case of stress-sensitive lenition appears to take place in a form of reduplication in Paamese.4 1 In one pattern of reduplication in this language, the initial consonant and vowel of a word are doubled, as shown below: (32) Paamese reduplication (from Crowley (1982:47); IP A, syllabification mine) Root Reduplication Gloss Root Reduplication Gloss (a) 'si.ta.li si.'si.ta.li ‘emerge’ (d) 'ka.a ka.'ka.a ‘fly’ (b) 'me.sa.i me.'me.sa.i ‘sick’ (e) 'to.ko.li to.'to.ko.li ‘touch’ (c) ta.'ho.si ta. ta.'ho.si ‘good’ (f) 'su.a.i su.'su.a.i ‘disappear’ The consonant inventory of Paamese includes /p, t, k, m b, nd, qg, f, s, h, m, n, g, 1 , r, j, w/ (Crowley (1982:12); IP A).4 2 /p, t, k/ have minimal aspiration, if any.4 3 Iad/ is sometimes pronounced with a ‘slight r off-glide’ [ndr ] (Crowley (1982:15)). Paamese appears to have right-to-left moraic trochees (Hayes (1995:178-179)). 3 9 Nganasan has also a process of syllabic gradation whereby the strong grade occurs word-medially in closed syllables and the weak grade word-medially in open syllables. For example, ['ku.hu] ‘skin (nom. sing.)’, but [ku.bu?] ‘skin (nom. pi.)’. For details, see Helimsky (1998). 4 0 Thanks to Caro Struijke for pointing out Paamese as a possible example of stress-sensitive lenition, and to Peter Alengra for providing me with information about reduplication in this language. 4 1 Reduplication in Paamese derives verbal items from roots. For details, see Crowley (1982:47— 51). 4 2 Crowley (1982) takes the labiodental fricative to be voiced, but since in most cases it is pronounced as voiceless, and because all other fricatives are voiceless, I represent it as If/ (Crowley (1982:14)). 4 3 Pre-nasalized voiced stops are represented as /b, d, g/ in Crowley (1982); here they are represented as /“b, " d , °g/. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 84 Lenition/fortition in reduplication applies to /p, nd, ft. When the root begins with /p/ or /n d/ the reduplicated form has word-initial [f] or [r] respectively. When the root begins in /f/ the correspondent consonant in the root occlusivizes to [p] (33). (33) Paamese lenition/fortition (Crowley (1982:48); IP A, syllabification mine) Root Reduplication Gloss (a) 'pi.li.tu ft.'pi.li.tu ‘stick’ (b) 'pu.si fu.'pu.si ‘kick’ (c) 'po.le fo.'po.le ‘burnt’ (d) 'fo.ra.a fo.'po.ra.a ‘noisy’ (e) m de.mi re.'nde.mi ‘think’ There is only one stress per word in Paamese. Stress is penultimate in two-syllable words, as in the roots in (33b, c, e). Stress falls on the antepenult of words with three or more syllables, as in the roots in (33a, d), unless this syllable is lexically unstressed or morpheme-final.44 In this form of reduplication, the first syllable is morpheme-final, so the reduplicated form keeps the stress on the original syllable from the base. This means that the initial syllable of the reduplicated forms is unstressed, and the second syllable is stressed. Thus, lenition of /p, nd/ correlates with unstressed syllables, and fortition of /f/ with stressed syllables 4 5 Note that /p, f7 4 4 Hayes (1995:178— 9) analyzes this stress system as involving right-to-left moraic trochees, with final extrametrical syllables and some exceptional forms. See also Crowley (1982:28— 31). 4 5 /fu.'e.li/, reduplicated to /fu.pu.'e.li/ ‘lost’ does not conform to the stress-based generalization proposed here. However, this is an exceptional form that has an ‘unstressable’ second syllable (Crowley (1982)). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 85 neutralize syllable- and word-fmally. Word-fmally they are in free variation, and syllable-fmally only [f] occurs. Silacayoapan Mixteco: /t/ and / j / lenite in weak positions (North and Shields (1977:22, 23). The consonant inventory of Silacayoapan is /p, t, c, k, kw, ?, m b, nd, 5g, |3, s, J, 3, h, m, n, ji, 1 , r, j/ (North and Shields (1977:21); IP A). North and Shields (1977:21) report that the base of the phonological word is the ‘couplet’, a unit formed of two syllables which shows allophonic conditioning and restriction. The couplet is also the main unit of tone contrast. The first syllable of this unit carries the word stress, and can have a /?/ coda. In foot terms, this corresponds to a syllabic trochee. In the discussion of Silacayoapan, I consider the couplet to be equivalent to a syllabic trochee foot. Voiceless stops are unaspirated (34a) except for foot-initial It/, which has some aspiration. Lenition of /t/ occurs in onsets in unfooted syllables (34c). The nature of lenited It/ is not clear; it is described as ‘softened’ and represented with [d]. 1 ^ 1 alternates with [j] in onset position foot-medially in rapid speech (34d). I consider this as an instance of optional lenition because [j] occurs in weak position. The distribution of /j/ is limited in Silacayoapan; it only occurs alone in two words, and it is more frequent as the second member of a cluster foot-initially (North and Shields (1977:24)). For related processes in Silacayoapan, see sections 5.1, 5.2.1 and 7.2. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 86 (34) Silacayoapan Mixteco lenition (from North and Shields (1977:22, 23)) (a) ('ta.a) ‘m an’ (c) kf?vl-ta [('kf?.vi).da] ‘he is sick’ (b) [('3a.?a)] ‘chile’ (d) [('ka.3i)], [('ka.jf)] ‘cough’ Summary: Lenition correlates with unstressed syllables, weak footed syllables and unfooted syllables. It is found mainly in onsets, but in some languages this might be the case because no codas exist or they are restricted. Types of lenition include debuccalization, fficativization of stops, and approximantization of fricatives. 4.5 Secondary articulation In some languages secondary articulation— as labialization or palatalization—occurs exclusively in stressed syllables. Two reported cases are Senoufo and Guayabero. Table 2.5. Secondary articulation Guayabero 0 ,C C->Cw /'_/u, o/_ Guahiban Senoufo 0 Secondary release in 'a Niger-Congo Guayabero: Guayabero has a stress-sensitive process by which all consonants in onset and coda position are labialized in stressed syllables with /o, u/ (35). (35) Guayabero labialization (from Keels (1985:71,72; IP A) (a) puka [l£ L w h u .k hAj ‘a la g o o n ’ (b) tja p o -t (tfA .'p who tw1 ‘a m irror’ (c ) taknuk fthA k.'n wu k w] ‘he stands watching’ (d) taka ['tha.kha] ‘a shoulder’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 87 (35a, b) show that labialization occurs in stressed syllables only. Codas as well as onsets are labialized (35b, c). (35d) shows that consonants in stressed syllables are not labialized with vowels other than /o, u/. Onset voiceless obstruents are aspirated in Guayabero (Keels (1985:71), Lavoie (2001:44)). See also section 4.3. Senoufo: Except for /n, ji, rg 1, j, w/, consonants in Senoufo are realized with secondary articulation—including palatalization and labialization— in stressed syllables. See sections 4.1, 4.2, 4.4, and 5.1 for related processes. Summary: Secondary articulation is found in codas or in onsets o f stressed syllables. The nature of the immediately preceding or following vowels influences secondary articulation. 5. Featural timing This section considers languages with stress-sensitive variation in featural timing for aspiration, glottalization and voicing. I distinguish between pre-aspiration and post aspiration, and pre-glottalization and post-glottalization. Pre-aspiration or pre- glottalization means that the onset of [spread glottis] or [constricted glottis] begins before the closure of the consonant. Post-aspiration or post-glottalization means that the offset of [spread glottis] or [constricted glottis] comes after the release of the consonant. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5.1 V oicing alternations In languages with stress-sensitive voicing alternations different patterns are observed. In pattern 1, underlyingly voiceless segments voice in unstressed syllables, as in Senoufo and Popoloca. In pattern 2 , underlying voiced segments devoice in unstressed syllables, as in Djabugai, Wembawemba, Wergaia, Farsi and Copala Trique. Silacayoapan Mixteco has patterns 1, 2 for different segments. In pattern 3, underlyingly voiceless segments become voiced in stressed positions, as in Wasco- Wisram and possibly Urubu-Kaapor. In pattern 4, the stress of the preceding syllable influences consonantal voicing, as in Paamese, Middle English and Proto-Germanic. The pattern of voicing for ‘x’ in English is not clear.4 6 Table 2.6. Voicing variation Senoufo (1) 0 /p, t, k/->[b, d, gl/V V Niger-Congo Popoloca (1) 0 pvless.l obstruents-> [voil/f-stressl Otomanguan Djabugai (2) 0 /b, d, g/-> [p, t, k]/[-stress] Pama-Nyungan Wembawemba (2) 0 Oral stops devoice/[-stress] Victorian Wergaia (2) 0 Oral stops devoice/[-stress] Victorian Farsi (2) 0 Vd. obstruents devoice/[-stress] # Indo-Iranian Copala Trique (2) 0 /d, g, z, 3, r/ devoice/[-stress] (opt.) Otomanguean Silacayoapan Mixteco (1, 2) 0 /k/-> [g]/unstressed syl. outside foot Pre-nasal stops devoice foot medially Otomanguean Wasco-Wisram (3) 0 Vd. stops devoice/|"-stressl Penutian Urubu-Kaapor (3) o Ip, t/-> fb, d]/[-stress] (optional) Tupi-Guarani *Proto-Germanic (4) o ,c Vless. fricatives vd./f-stressl Proto-Germanic *Middle English (4) 0 /0/ -> 16/ after weakly 'o Germanic Paamese (4) 0 /nt/-> [n d] medially, esp. after '0 Austronesian English (?) c/o ‘x’ [k.s]/ 'V_._V; -> [g.z]/V_.'_V Germanic 4 6 Grammont (1933:163— 4) reports that in Central Italian /p, t, k/ became voiced in onsets of stressed syllables except word-initially; e.g. coperta>coverta ‘covered’, betulla>bidolla ‘birch tree’ and securu>siguro ‘sure’. Since the data reported by Grammont is not consistent with any variety of Italian (Mario Saltarelli, p.c.) Central Italian is not included in this section. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 89 Senoufo: /p, t, k/ are voiced in onsets of medial unstressed syllables (Mills (1984:94). In other positions /p, t, k/ show light aspiration. Since Senoufo has durational differences for consonants in stressed and unstressed syllables (section 4.4), it is possible that voicing of /p, t, k/ in unstressed syllables is perceptual. Voiceless segments are universally longer than voiced ones (Catford (1977)). In unstressed syllables /p, t, k/ are shorter, which gives the percept of voicing (see Lavoie (2001)). Under this hypothesis, no glottal (i.e., articulatory) voicing actually occurs in this position at least synchronically with the reinterpretation. Popoloca: In Los Reyes Metzontla Popoloca, voiceless stops and affricates become voiced in onsets of ‘reduced’ syllables (Veerman-Leichsenring (1984:43-5). Sibilants become voiced optionally in this position. Additionally, both affricates and sibilants might become voiced after a long vowel (36). (36) Popoloca voicing (from V eerman-Leichsenring (1984:43-5); IP A) (a) /ju + ka/ ['ij.jiu.ga] ‘two leaves’ (b) /ju + su?/ ['lj.jui.zo?] ‘two beds’ (c) /°gu + tu/ ['irj.gu.du] ‘fruit pit’ (d) /n ?da.t§i7 [Vdi.cfzfl / ['n?d:.tgf| ‘old woman’ Stressed syllables in Popoloca have either long vowels or diphthongs, or—more commonly—coda consonants. Gemination occurs after stressed syllables (4.1). /i/ is epenthesized in monosyllabic words so that stressed syllables have a coda from the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 90 following onset. This occurs in /jit/ ‘two’ and /5gu/ ‘fruit’ in (36a-c) (Veerman- Leichsenring (1984:22, 23)). Stress is not realized in ‘/ka/ ‘leaves’, /su?/ ‘beds’, and /tu/ ‘pit’ for morphological or syntactic reasons (36a-c) (Veerman-Leichsenring (1984:22)). This causes these words to be reduced phonetically; their onsets are voiced. In these cases, the ‘reduced’ syllables follow unstressed syllables. (36d) shows that affricates are optionally voiced in reduced position and after long vowels. See also sections 4.1, 5.3.2, and 7.1. Djabugai: Djabugai exemplifies voicing pattern 2. Voiced stops are devoiced intervocalically or medially after /j, 1 , r/ (37). Devoicing occurs in allegro speech; no devoicing occurs in citation form (Patz (1991:252-3)). (37) Djabugai: devoicing in unstressed positions (from Patz (1991:252); IP A). (a) /w a b a r / ['w a.pA r] ‘Hunt’ (b) /w ajgal/ ['wai.kAl] ‘To search’ (c) /bulbal/ ['b u l.p A l] ‘To rub’ (d) /murba/ ['mur.gAl] ‘Under water’ The consonantal inventory for this language is /b, d, j, g, m, n, ji, g, 1, r, j, w/. Only /m, n, g, 1 , r, ^ j/ can be codas (Patz (1991:250-1, 255)). Stress is generally assigned to the first syllable of the word. One exception is three-syllable words with a long vowel in non-final position; in this case, main stress falls on the long vowel. Secondary stress falls on the penultimate syllable in words with more than three Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 91 syllables (Patz (1991:258-60)). This suggests that devoicing occurs in unstressed syllables. However, no examples are provided of the realization of /b, d, g/ in unstressed syllables outside the second syllable, or of the realization of /b, d, g/ in stressed second syllables or in syllables with secondary stress. See also section 4.2. Wembawemba: Wembawemba also exemplifies voicing pattern 2. It has only voiced consonants: /d, d, 4 h 9, tn, n, ji, g, r, ^ 1 , w, j/. Main stress is initial; secondary stress occurs on the third syllable of the word. Oral stops are totally devoiced intervocalically and word-fmally (38d, f), i.e. in syllables that do not receive main stress. Word-initially, devoicing is often partial and gradient; different speakers show different amounts of voicing. Initial voicing is facilitated by /u/ for /b, g/ and by /i, d for /}/ (38a-e). Oral stops are voiced after nasals (38a, b). Nasals can be devoiced syllable-fmally, but never in monosyllabic accented words (Hercus (1986:10)). (38) Wembawemba voicing (from Hercus (1986:6); IP A) (a) bunda ['bun.ck] ‘to bite’ (d) jilega [(ji.le.kA] ‘sick’ (b) gumba ['cpr m.bA] ‘to sleep’ (e) juj;a ['cu.jWJ ‘to gossip’ (c) bana ['ga.n A ] ‘ringtail p o ssu m ’ (f) wujub ['wu-cup1 ] ‘sto m a c h ’ Wergaia: Wergaia has a similar voicing pattern to Wembawemba. The consonant inventory is /b, d, 4 h m, n, g, ji, g, r, ^ 1 , w, j/ (Hercus (1986:73)). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 92 Main stress occurs on the first syllable and secondary stress generally on the third syllable (for details see Hercus (1986:81)). Oral stops are devoiced except word- initially, where they are always voiced (39). This suggests devoicing is sensitive to stress. Intervocalically, there is a tendency towards voicing before certain vowels (39a, c) (Hercus (1986:75)). See section 4.1 for related processes. (39) Wergaia devoicing (Hercus (1986:75, 198)) (a) bubug ['bu.buk] ‘baby’ (c) mijiin ['mi.jiAn] ‘moon’ (b) b abi ['b a '.p i] ‘fe m a le ’ (d) mijag ['m i.cA k ] ‘rain ’ Farsi: Farsi also exemplifies pattern 2. ‘Lenis’ obstruents /b, d, g, j, v, z, 3/ are fully voiced intervocalically and in stressed initial positions, but only partially voiced in unstressed initial syllables. Since there is no preceding voiced sound, it is possible that consonantal voicing is more difficult to maintain word-initially, especially if the syllable is unstressed, /b, d, g, j, v, z, 3/ are totally or partially devoiced in codas (Samareh (1977:24-5)). /b, d, g, 3/ are also reported to be devoiced in contact with voiceless sounds. The consonantal inventory of Farsi is /p, t, k, b, d, g, q, ?, tj, d3, f, s, J, % , h, v, z, 3, m, n, 1, r, j/. All consonants can be codas (Samareh (1977:8-9)). Related processes are discussed in sections 4.3 and 5.2.1. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 93 Copala Trique: Copala Trique distinguishes between ‘fortis’ and ‘lenis’ stops and sibilants. Fortis consonants are voiceless and lenis consonants are voiced (Hollenbach (1977:36)). Lenis consonants /d, g, z, 3, r/ optionally devoice in onset of unstressed syllables and in clusters (Hollenbach (1977:37); only /?, fa/ can be codas). /% / is realized as [d3] word-initially in unstressed syllables, and /r/ is realized as [|;] intervocalically, in clusters with stops, and word-initially following a word final /?/ and preceding a vowel; otherwise it is a sibilant. Fortis consonants and affricates are restricted to stressed positions. It is plausible that devoicing occurs in unstressed positions since the voicing contrast is not relevant there. Cf. related San Andres Chicahuaxtla Trique, where fortis (voiceless) and lenis (voiced) consonants have similar distributional restrictions; the phonemic inventory is slightly different (see Hollenbach (1977:50 ff) for details). For more details on Copala Trique see section 4.3. Silacayoapan Mixteco: Silacayoapan Mixteco shows two apparently opposed tendencies regarding voicing variation. Silacayoapan has both voiceless unaspirated stops /p, t, c, k, kw, ?/ and pre-nasalized voiced stops /m b, nd, ■ ’ ’ j, 5g/ (North and Shields (1977)). Pre-nasalized stops are voiced foot-initially (40a) and optionally devoiced foot-medially (40b, c). Additionally, Dd is optionally realized as [g] in unstressed unfooted syllables (40d). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 94 (40) Silacayoapan Mixteco (from North and Shields (1977:22-3); IP A) (a) [(lm ba.a)] ‘compadre’ [('Me.eO] ‘thick’ (b) la“ba [('M.%a)], [('la.^pa)] ‘bladder’ (c) uMo [(’ ?u.MoO],[('?u.nd)] ‘animal fat’ (d) ak«J«-kandl [aT'ku.Judga-Mi], [a.('ku.ju.)ka.adi] ‘we (excl.) won’t eat more’ /m b, J 1 j, 5g/ are rare; lpj, 9g/ are found foot-medially only. There is no mention of optional devoicing of pre-nasalized voiced stops in onset of unfooted syllables; it is possible that these segments do not occur in this position. Silacayoapan Mixteco has syllabic trochees (4.4). For related processes see also sections 5.2.1 and 7.2. Wasco- Wisram:4 7 Wasco-Wisram (Upper Chinook) exemplifies voicing pattern 3. /pi is realized as [b] in onsets of stressed syllables and as [p] in onsets of unstressed syllables. Coda III has a voiced variant in final position before words beginning with stressed vowels (41). (41) Wasco-Wisram (from Sapir (1925:44)) (a) inat ‘across’ ina'dV ‘across’ (b) 'wa.pul ‘night’ wa.'bul.max ‘nights’ Urubu-Kaapor: Kakumasu (1986:399) reports optional voicing of /p, t/ in non-primary-stressed syllables, which would mean that Urubu-Kaapor exemplifies pattern 3. However, secondary stresses are not marked in the source (42). 471 thank Joaquim Brandao de Carvalho for bringing to my attention Wasco-Wisram and Vemer’s Law. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 95 Aditionally, note that in (42b) onset /p, t/ can be voiced in any of the two last syllables in any combination. fkJ is optionally voiced at word boundaries, but this is not reported to be sensitive to stress (Kakumasu (1986:399)). Oral stops lengthen in primary-stressed syllables (section 4.2). 4 8 (42) Urubu-Kaapor optional voicing of /p, t/ (from Kakumasu (1986:399)). (a) /arapuha/ [arapu'ha] ~ [arabu'ha] ‘deer’ (b) /heta tipe/ [he'ta tibe] ~ [he'ta dibe] ‘there are many, but in vain’ [he'ta tipe] ~ [he'ta dipe] Proto-Germanic: In the development from PIE, /p, t, k, kw / changed to voiceless fricatives /f, 0, x, xw / (Grimm’s Law). A subsequent change voiced these voiceless fricatives if they were surrounded by voiced sounds and the immediately preceding syllable did not have the main accent of the word (43) (Vemer’s Law). This change was obscured by a later change in the position of stress to the first syllable of the word (Wright (1957), Collinge (1985)). (43) V emer’s Law 4 9 Grimm’s Law Vemer’s Law Gloss (a) pa 'ter (PIE) fa 'Oer (Early Germanic) fa 'der father ’ (b) sap ia (Sanscrit) seofon (Old English) sifkn (Old Saxon) ‘seven’ 4 8 In (45b) the onset of the stressed syllable should be lengthened (see section 4.1). It is not clear if this is a typo in the source or if there is some other process going on. 4 9 From W right (1957) and http://homepage.mac.com/branscombcourses/HELWemer.html. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 96 Middle English: Kabell and Lauridsen (1984:80) report that in the stage between Middle and Modem English, /0/ among other consonants began to be voiced after a weakly stressed syllable, or initially, as in the, thou, thee, that and though. No examples are provided for the stress-sensitive change; it is not clear if only onsets are involved, and no other consonants are mentioned.5 0 It is plausible that since this process occurred in function words, ‘weakly stressed syllables’ correspond to unstressed syllables and that this process takes place word-initially. Around the 14t h century the /f7 of weakly stressed o/became /v/ in opposition to strongly stressed off (Kabell and Lauridsen (1984:50)). This seems to be an isolated case, but in Modem English, the /0/ of with becomes voiced when followed by a voiced segment in the next word. A possibility is that in both of these cases the voicing of the following segment causes coda devoicing. Paamese: Medial Inti sequences created by vowel deletion become [n d] obligatorily when the preceding syllable is stressed (44a, b), and optionally if the preceding syllable is unstressed (44c) (Crowley (1982:39)). See also section 4.2.3. (44) Paamese (from Crowley (1982:39; IPA, syllabification mine)5 1 (a) ma+'ani+tei [lm a1 .ndej] ‘I will eat some’ (b) feni+tasi ['fe. ndas] ‘knowing the sea’ (c) manu+taai [ma.1 nda:j] ‘perched flying fox’ 5 0 Cf. Lavoie (2001:32). 5 1 Crowley’s (1982:20— 1) distinction between ‘loose’ and ‘tight’ morpheme boundaries is ignored in this data. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 97 English: In some varieties of English, voicing in the pronunciation of the grapheme ‘x ’ correlates with stress. For instance, in British English ‘x ’ is pronounced as [gz] when ‘x’ overlaps the coda of an unstressed syllable and the onset of a stressed syllable, as in exists [ig.'zists], exert [iq.'zsri]. ‘x ’ is pronounced as [ks] when it overlaps the coda of a stressed syllable and the onset of a stressed syllable, as in exit ['ek.sit] (Collinge (1985:211). For many speakers ['ek.sit] is in free variation with ['eg.zit]. In onset position, ‘x’ is pronounced as [z], as in xylophone ['zai.la.fsun] and xerox ['zio.raks].5 2 Syllable-finally ‘x’ is pronounced as [ks], as in hex [heks]. According to Collinge (1985) the same effect of stress is seen in pairs like possess [pa.'zes] and possible ['pa.ss.bl]; in this case [z] occurs in stressed syllables and [s] in unstressed syllables. This suggests that the onset in [g.z]/[k.s] is sensitive to stress. Summary: Four consonantal voicing patterns are attested. In pattern 1, voiceless consonants are voiced in unstressed syllables. In pattern 2, voiced consonants are voiced in unstressed syllables. In pattern 3, voiceless consonants are voiced in stressed syllables. In pattern 4, the stress or lack of stress of the preceding syllable conditions consonantal voicing. The voicing alternation in the pronunciation of ‘x’ is 5 2 In all words beginning with V , ‘x’ is pronounced [z] in all cases except in the proper name Xhosa (Oxford Advanced Learner’s Encyclopedic Dictionary). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 98 unclear. The fact that a pattern in which voiced consonants would become voiceless in stressed syllables is missing seems to be significant and is discussed in chapter 3. 5. 2 Aspiration alternations 5.2.1 Post-aspiration This section discusses cases where stressed syllables correlate with stronger aspiration, friction or affrication from the release of a consonant. Stronger aspiration is distinguished from longer aspiration; the first refers to the amount of the aspiration, and the second to its duration (see section 4.1). Table 2.1. Post-aspiration English 0 /p, t, k, tj-/->[ph , t* k * 1 , tf]/#_/'_V Germanic Maori 0 Stronger aspiration/friction/'_V Austronesian German 0/C Stronger #_/'_V/'V_ Germanic Pattani NM Stronger # _ _ /'_V Sino-Tibetan Farsi NM Stronger #_ /'_V [ndo-Iranian Chali (Tati) 0 lrl aspirated if geminated in 'o Indo-Iranian Silacayopan Mixteco 0 /t/-> [tV'JV Otomanguean Squamish o Non-glottalic S aspirated/'jV. (opt.) Central Salish Basque Spanish c /b,d,g/^ [$,0,x]/'V _ Romance English: Aspiration of voiceless stops and /tf/ in American English is reported to be longer and stronger word-initially and in stressed syllables (Lisker and Abramson (1967), Kahn (1980), Crystal and House (1988), among others). It has been proposed that aspiration is stronger and longer foot-initially (Davis (2002) among others). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 99 Stronger aspiration in stressed syllables is also reported for Cardiff English and London English (Collins and Mees (1990), Carr (1999)). Affrication of voiceless stops—especially for III—rather than strong aspiration is reported for New York City English, Liverpool English, and London English (45). (45) Affrication/strong aspiration (Carr (1999:153, 156)) (a) cup of tea [ lk hA ?po'tshii] (London English) (b) tin ['tsin] (New York City English) Maori: Aspiration of oral stops increases with loudness and stress. In onsets of stressed syllables either aspiration or affrication might occur (see section 4.2.2). Fricatives also have stronger friction in this position (Bauer (1993:530-1, 555)). German: Stronger aspiration for both onset and coda voiceless stops occurs in Standard German word-initially and in stressed syllables (Kohler (1977), Giegerich (1989), Iverson and Salmons (1995), Alber (to appear)). Alber (to appear) suggests that aspiration occurs foot-initially. Pattani:5 3 Sarma (1982) reports a tendency for stronger aspiration in initial and accented syllables and for reduction of aspiration in pre-accentual (‘weak’) syllables (Sarma (1982:48, 67), cf. Kirchner (1998)). Pattani has phonemic aspirated and unaspirated consonants and initial main stress. 5 3 Pattani (W est-India, Sino-Tibetan) is different from Pattani Malay (Thailand, Austronesian). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 100 Farsi: Aspiration of Ip, t, k, tf/ is reported to be stronger word-initially and in stressed syllables than in other positions (Samareh (1977:24)). See also sections 4.3 and 5.1. Chali Tati: In the Chali dialect of Southern Tati, /r/, a ‘voiceless or partially voiced weak alveolar trill’, becomes ‘somewhat’ aspirated when geminated and followed by a stressed vowel (Yar-Shater (1969:35)). /r/ tends to become a flap in final position. Voiceless stops in Chali, which are slightly aspirated, are more strongly aspirated when the stops are geminated; the source does not mention stress in this context. Silacayoapan Mixteco: III is reported to be slightly aspirated ‘couplet initially’ i.e., foot-initially (section 4.4) (North and Shields (1977)). Word stress in Silacayoapan Mixteco falls on the first syllable of the couplet, so lil is aspirated in stressed syllables. In other positions III has no aspiration or is lenited (North and Shields (1977:22). See also sections 4.4, 5.1 and 7.2. Squamish: Aspiration of non-glottalized plosives occurs optionally in onsets of stressed syllables (Kuipers (1967:25)). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 101 Basque Spanish: In Peninsular Spanish /b, d, g/ are usually pronounced as approximants in most environments. In the dialect of Spanish spoken in the Basque Country coda /b, d, g/ usually fricate and devoice, as in /pared/ [pa.'reB] ‘wall’. For some speakers, frication of coda Po, d, g/ occurs mostly in stressed syllables, indirectly causing a loss of voicing (Gonzalez (2002b)). For details see chapter 4. Summary: Unaspirated segments may be post-aspirated in stressed syllables or foot-initially. Aspirated segments might have stronger post-aspiration in stressed syllables and foot-initially. It is common for post-aspiration to occur both in stressed positions and word-initially. In most cases onsets are more prone to be post aspirated, but lack of codas or restrictions thereon might play a part in this patterning. In some cases, strong aspiration might lead to affrication. In some other cases, strong friction occurs for fricatives in stressed syllables. 5.2.2 Pre-aspiration Pre-aspiration is reported to occur in onsets after stressed vowels in a range of languages (Kehrein (2001)). In some cases pre-aspiration is restricted to preceding short or non-high stressed vowels. In all languages with stress-sensitive pre aspiration oral stops are affected; in Ojibwa fricatives are affected as well.5 4 5 4 Pre-aspiration in Lule Sami (Uralic) might also be related to prosodic structure. According to Ladefoged and Maddieson (1996) it has similar pre-aspiration to Icelandic. Kehrein (2001) mentions the possible role of strengthening due to, among other factors, shortening of preceding stressed vowels. Engstrand’s (1987) phonetic study does not mention any stress differences for pre-aspiration, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 102 Table 2.8. Pre-aspiration Icelandic 0 After short 'V West Scandinavian Standard Faroese 0 After short 'V West Scandinavian Northern Faroese 0 C(C) h- » hC(C) /'V_ (-high) West Scandinavian Scots Gaelic 0 ■ v_ Celtic Irish Gaelic 'v_ Celtic Ingush 0 After short 'V North-East Caucasian Toreva Hopi 0 Stops/ 'V_V Uto-Aztecan Ojibwa 0 Stops, fricatives/ 'V_ Algonquian Tarascan 0 'V_ Tarascan Icelandic: Icelandic has both contrastive and alternating pre-aspirated stops (Silverman (1997)). Phonemically, pre-aspirated stops contrast with voiceless unaspirated, voiceless geminates and voiceless post-aspirated stops. Voiceless unaspirated and voiceless post-aspirated stops occur word-initially (46a, b). Word- medially and word-fmally voiceless unaspirated, pre-aspirated and geminate stops occur (46a-c).5 5 (46) Icelandic stops (from Ringen (1999:138) and references therein, and Ladefoged and Maddieson (1996:71))5 6 (a) (b) (c) Initially [’ pair] ['p V ] ‘bar’ ‘pair’ Medially ['kh o:.por] ['kh oh.par] r'kh op.par] ‘copper’ ‘small pot (n. pi)’ ‘young seal (n. pi)’ Finally ['keep) ‘zeal’ r'kh ahp] ‘zeal’ ['kappl ‘hoax’ but Lule Sami has lexical stress in the first syllable. It is possible that pre-aspiration occurs word- medially or in non-primary stressed syllables. Engstrand (1987) also mentions that pre-aspiration frequently takes the form of frication rather than aspiration, especially in palatal and velar contexts. 5 5 The same contrast exists for coronals and velars. See Ladefoged and Maddieson (1996:71). 5 6 Pre-aspiration is marked as /hC/ rather than /h C/ to reflect the fact that pre-aspiration in Icelandic is longer than post-aspiration (Ladefoged and Maddieson (1996:70) and references therein). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 103 In alternations, pre-aspirated stops are realized ‘coordinated with a stressed syllable’ and ‘away from unstressed syllables’ (47) (Silverman (1997:78)). Main stress in simplex words in Icelandic falls on the initial syllable. Icelandic has left-to-right syllabic trochees with secondary stress on alternate syllables (Hayes (1995:188-198) and references therein). (47) Icelandic alternations (Silverman (1997:77-78), citing Thrainsson (1978))5 7 (a) /kh akh +a/ ['kh a:.kb a] ‘cake (nom. sg.)’ /kh akh +na/ ['kh ah.ka] ‘cake (gen. pi.)’ (b) /nith +a/ ['ni:.th a] ‘utilize (inf.)’ /nith + th+a/ ['nih.ta] ‘utilize (past)’ Standard Faroese: Standard Faroese is reported to have the same stress-sensitive pre-aspiration pattern as Icelandic (Kehrein (2001)). Faroese has initial stress (Hume and Seo (2001)). Pre-aspirated stops occur medially and finally as realizations of geminate voiceless aspirated stops. Initially, there is a contrast between voiceless unaspirated and voiceless aspirated (Ladefoged and Maddieson (1996:70)). Northern Faroese: Northern Faroese (including Vagar) contrasts among aspirated and non-aspirated singleton and geminate oral stops It, tt, th, tth /. Aspirated single stops and geminates are pre-aspirated after non-high stressed vowels (48a-d). Post aspirated stops are found in onsets of stressed initial syllables (48a, d) (Kehrein (2001), Petersen et alia (1998); cf. Ladefoged and Maddieson (1996)). 5 7 Silverman (1997) assumes that pre-aspirated stops are phonological; cf. Thrainsson (1978), Ladefoged and Maddieson (1996), Kehrein (2001), among others, who consider they derive from underlying geminates. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 104 (48) Northern Faroese (cited in Kehrein (2001); from Petersen et alia (1998:126)) (a) papi [lph oa.h pi] ‘dad’ (c) rekkja r'eh tf.tfa] ‘bed’ (b) lappi |~'la.hppi] ‘rag’ (d) taklca F 'tha.h kka] ‘thank’ Scots Gaelic: Scots Gaelic contrasts post-aspirated and non-aspirated stops initially. Post-aspirated stops are realized as pre-aspirated medially and finally after a stressed vowel (Ni Chasaide and Dochartaigh (1984), Ladefoged and Maddieson (1996:70)). Pre-aspirated stops derive from voiceless geminate stops in this language; one example is Proto-Celtic *kattos>['kaht] ‘cat’ (Ni Chasaide and Dochartaigh (1984:155)). Pre-aspirated stops have the same origin in Icelandic and in Irish. Ni Chasaide and Dochartaigh (1984) suggest that the contrast between voiced and voiceless geminate stops in previous stages of Scots Gaelic was lost due to the difficulty in maintaining voicing in a geminate stop. As voicing was reduced and less perceptible, leftward devoicing started to take place before the voiceless geminate, with glottal/cavity friction at the same time, and when it could be perceived it became the relevant contrast. Ni Chasaide and Dochartaigh (1984) suggest that pre aspiration evolves in the following way: kk>h k>hk>xk. The last stage would be pre- affirication. This would explain the variable realization of aspiration in languages like Tarascan (see below) or the differences in pre-aspiration duration in Icelandic, where pre-aspiration is very long. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 105 Irish Gaelic: Irish has weak pre-aspiration, defined as ‘no audible friction’, ‘silence’ or ‘weak glottal fricative’ (Ni Chasaide and Dochartaigh (1984:142); also Kehrein (2001)). It is reported to occur after stressed vowels, but there is no mention o f the absence of pre-aspiration in other contexts. Ingush: Ingush has fixed initial stress (Nichols (1994)). Single stops are post aspirated initially (49a). Geminated stops are pre-aspirated after underlying long, stressed vowels (49b-d). (49) Ingush (Kehrein (2001), citing Nichols (1994); IPA) (a) /tuxa/ ['th u.xa] ‘hit (infin.)’ (b) /la:tta/ ['lah t.(t)(o)] ‘stand (infin., imp.)’ (c) /beatstsara/ ['bea h ts (ts) or(o)] ‘green’ (d) /doaqqa/ ['doah q (q)(o)] ‘take(s)’ Toreva Hopi: Pre-aspirates in the Toreva dialect of Hopi occur intervocalically after stressed vowels (50a). Pre-aspirates contrast with /hC/ clusters (Whorf (1946:160), cited in Steriade (1997:73)). A change in stress makes pre-aspiration disappear (50b). (50) Toreva Hopi (from Whorf (1946:182), cited in Steriade (1997:73)) (a) tal-wfh pi ‘a lightning flash’ (approx.) (b) tal-wipi-ki ‘a lightning-flash design’ (approx.) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 106 Steriade (1997:73) suggests that the restriction to following a stressed vowel may be related to the vowel’s longer duration: ‘a longer vowel will reflect preaspiration without becoming completely aspirated itself (cf. Gordon (1996))’. Ojibwa: Underlying unaspirated voiceless stops and fricatives are pre-aspirated after stressed vowels; elsewhere they are unaspirated (Kehrein (2001), based on Bloomfield (1956). Geminate consonants—/pp, tt, cc, kk/—are pre-aspirated word- medially (Ladefoged and Maddieson (1996), citing Bloomsfield (1956:8)). Tarascan: Pre-aspirates are reported to occur word-medially after stressed syllables (Silverman (2002), citing Foster (1969)). Pre-aspirates vary with pre- spirantized stops; in the case of /t/, it varies to presibilantization after voiced vowels word-medially or after with intervening non-pausal juncture. After [i], pre-aspirates vary freely with vowel length (51). (51) Tarascan (from Silverman (2002), citing Foster (1969)) (a) fum.'bah.pa.ni] ‘to heap things on the floor of the room’ (b) [a.'Aa’ Lku.ni] ‘to cut oneself on the hand’ (c) [lphaf Vta.ni] ~ ['phasfta.ni] ‘to touch the metate’ (d) [ka.'tfub.tfa.nil ~ [ka.'tjiA t f a.ni] ‘to cut off one’s braid’ (e) ['tsih.ku.ni] ~ ['tsk.ku.ni] ‘to drop from one’s hand’ Summary: Pre-aspiration occurs after stressed vowels; plausibly, pre-aspiration is better perceived in this context (Steriade (1997)). In some languages, pre Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 107 aspiration occurs after stressed short vowels that arise from underlying long vowels. In languages where pre-aspirated stops derive diachronically from geminates the reason might be that gemination had the effect of shortening the preceding vowel. In the case of Northern Faroese, pre-aspiration is only found after stressed, non-high vowels. It is possible that the restriction of pre-aspiration to after non-high vowels is related to sonority considerations (see discussion in chapter 3). Pre-aspiration has two possible sources; post-aspirated stops and geminates. Pre-aspiration derives from post-aspirated stops synchronically. A possible explanation is perceptual; aspiration ‘migrates’ from its original position in the release of the onset of an unstressed syllable to the pre-closure position after a stressed vowel so that it is better perceived (following Steriade (1997)). Pre aspiration can also derive from voiceless geminates synchronically or diachronically. Debuccalization of the coda portion of the geminate might result in pre-aspiration. In Icelandic, this would be consistent with the extra-long duration for pre-aspiration; the timing of the underlying geminate is maintained, but the place features of the coda portion have been lost. 5.3 Glottalization alternations Post-glottalization refers to the occurrence of a glottal constriction at or after the release of a consonant. In pre-glottalization, a glottal constriction is timed before the oral constriction of a consonant (cf. Howe and Pulleyblank (2001), Steriade (1997)). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 108 Underlying glottalized consonants are in some cases pre-glottalized after stressed vowels and post-glottalized before them (Coast Tsimshian, Gitksan, Saanich; cf. with Lilloet). Pre-glottalization of underlying non-glottalized consonants is considered in section 7.2 under epenthesis. Table 2.9. Glottalization Coast Tshimshian O /S? /-> [? S]/V_#/'V._V; -»[S? ]/V_'V Penutian Gitksan 0/C /S? /-> [? S]/'V_; _#;->[S? ]/_'V;#_ Penutian Saanich (SoncaOan) 0 /R? /-»[? R]/'V_V; -> [R? ]/V_'V Salishan Lilloet (St’& t’imcet) 0 /R/-»[? R]/ '_V; [R? ] elsewhere Salishan Coast Tsimhian: In Coast Tsimshian there is a contrast between plain and glottalized stops and sonorants (Kehrein (2001)). Glottalized stops are post- glottalized before vowels and pre-glottalized after vowels. Intervocalically, glottalized stops are pre-glottalized if the preceding vowel is stressed and post- glottalized if the following vowel is stressed (Dunn (1995)). Glottalized sonorants are always pre-glottalized (Howe and Pulleyblank (2001:50)). Gitksan: Kehrein (2001) reports that pre- and post-glottalization of stops in Gitksan are conditioned by stress and syllable position. Pre-glottalization occurs word-finally and after stressed vowels; and post-glottalization word-initially and before stressed vowels. Rigsby and Ingram (1990:251, 262) report that glottalized obstruents have pre-glottalized voiceless allophones pre-consonantally following a stressed vowel. This probably means pre-glottalization occurs in codas of stressed Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 109 syllables. Rigsby and Ingram (1990) also report that lax glottalized stops occur in pretonic and syllable-final environments. Saanich: Salish languages also have a plain vs. glottalized contrast for stops and resonants (Kehrein (2001)). In Saanich (SoncaOan), a dialect of Straits Salish, intervocalic glottalized resonants are post-glottalized before a stressed vowel, and pre-glottalized after a stressed vowel (Howe and Pulleyblank (2001:50, 70) and c o references therein). Lilloet: Glottalized sonorants are only pre-glottalized before stressed vowels. In all other contexts, glottalized sonorants are post-glottalized (Howe and Pulleyblank (2001:50-1) and references therein). This pattern does not conform to the preferential perceptual pattern for resonant glottalization—unlike Saanich— because glottalized sonorants tend to be pre-glottalized cross-linguistically. Post- glottalization in resonants would be expected just if there were no preceding vowel, or intervocalically before stressed vowels. See further discussion in chapter 3. Summary: Glottalization in laryngealized consonants can be realized before closure or after release. In some languages, this varies according to stress. After a stressed syllable, pre-glottalization occurs. Before a stressed syllable, post- glottalization occurs. 5 8 This pattern is referred to in this source as ‘attraction of glottalization by the stressed syllable’ and compared to Tsimshian in this respect. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 110 6. Attraction Laryngeal and non-laryngeal features alike can be attracted to onsets or codas of stressed syllables. In some instances glottalization is attracted to stressed nuclei. Stress in some languages is attracted by specific consonantal properties such as voicelessness. 6. 1 Consonantal attraction Table 2.10 indicates languages where glottalization is attracted to the coda or nucleus of a stressed syllable. Table 2.10. Consonantal attraction59,6 0 Shuswap C/N [e.g.] to 'S Interior Salish Twana C [e.g.] to 'o Central Salish Thompson River Salish N [c.g] to 'N Interior Salish Coast Tsimshian N [c.g] to 'N Penutian Danish N [c.g] to 'N West Scandinavian Colville O Pharyngeal R to stressed suffix Interior Salish *Bagneres-de-Luchon French oo Liquid to 1 a Romance 5 9 In Chum ash glottalization is attracted to codas (Applegate (1972), cited by Kehrein (2001)). The conditions for this process are not well-known (Sharon Inkelas, p.c.). 6 0 In Gascon (Occitan) liquids were historically attracted from a complex onset of a non-initial syllable to the preceding syllable if this was heavier than the syllable that the liquid belonged to (Hume (2000), Dumenil (1987)). In both its original and final position, the liquid remains part of a complex onset. Some examples include Latin ca’ pra> Gascon era’ ba ‘goat’ and Latin ca’ stra > Gascon era 'ste ‘ditch’ (Hume (2000)) (the symbol <’> after a vowel means the vowel has an acute accent). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. I l l Shuswap: Shuswap has a contrast between pre-glottalized sonorants and post- glottalized obstruents (Steriade (1997) and references therein).6 1 Glottalized obstruents are unrestricted. Sonorants can be syllabic or non-syllabic. Non-syllabic glottalized sonorants occur only after vowels, but syllabic sonorants can be glottalized in any context. Steriade (1997) suggests that syllabic sonorants are longer and thus better able to carry the cues for glottalization. Glottalization in suffixes like /-il7 ap/ ‘foundation’, /-k?/ ‘implement’ or /- sxn? / ‘rock’ is attracted to the final sonorant of a stressed root (52) (Kuipers (1974:30), Steriade (1997), Kehrein (2001)). The final sonorant might be a non- syllabic coda position (52a, c) or a syllabic nucleus (52b, d).6 2 (52) Shuswap: Laryngeal attraction (Kehrein (2001), citing Kuipers (1974:30)) (a) /'xul-k?/ fxu fk ] ‘firedrill’ (b) /'tw-k?/ ['tw? k] ‘stg. for sale’ (c) /-'q’iw-il29p/ ['q’iw?.kp] ‘chair’ (d) /'t% -sxn2-m-kn/ [tx.'_?-sx.nm.kn] ‘I heat stones’ Steriade (1997) suggests that glottalization attraction to the sonorant in the stressed syllable is a durational effect. Glottalization is realized as pre-glottalization of the sonorant, where the cues for glottalization are carried by the preceding vowel. Since stressed vowels are longer, louder and better able to carry the contextual cues for 6 1 Citing Kuipers (1974:24), Kehrein (2001) reports that Shuswap sonorants are pre-glottalized if they are onsets and post-glottalized if they are codas. 6 2 In Salish languages consonant clusters arise from the deletion of unstressed vowels (Kehrein (2001) and references therein). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 112 creak, glottalization is attracted to the stressed syllable (Steriade (1997:86)). Glottalization can also be attracted to a stressed syllabic sonorant, since syllabic sonorants are long enough to carry the cues for glottalization. Twana: As in Shuswap and other Salishan languages, Twana has stress- triggered glottalization transfer (Thompson (1979), Blevins and Garrett (1998:525)).6 3 Thompson River Salish: Historically, laryngealized resonants in Salishan lost their glottalization to a stressed vowel to the left or to the right (Dunn and Hays (1983:56, fn. 15), based on Laurence Thompson (1980 p.c.)). Thompson River Salish is a clear example of this pattern according to the same authors. This pattern gives rise to glottally interrupted vowels [a?a] (Kehrein (2001)). See also section 8. Coast Tsimshian: Laryngealized codas tend to lose their glottalization to the preceding nucleus, which is laryngealized or glottally interrupted; glottalization can also move to the preceding onset. According to Dunn and Hays (1983:56, fn. 15), it is possible that this process is stress-related, as in Thompson River Salish. However, they mention that the data is not as clear; they suggest that glottal attraction in these dialects is based on the greater strength of nuclei over codas. 6 3 Blevins and G arrett (1998:525), citing Thompson (1979), report that in Squamish, Nle?kepmxcin and other Salishan languages, some roots glottalize any sonorants in an immediately following suffix, even if no glottalic element appears in the root; a glottalizing element appears in the roots in the proto-language. It is unclear if stress is relevant in these cases. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 113 Danish: Stressed vowels in Danish are interrupted (by glottalization); this is usually referred to as stod (Zee (1988) and references therein, Fischer-Jorgensen (1987), Ladefoged and Maddieson (1996)).6 4 Colville: Pharyngeal resonants in the root are attracted to the stressed suffix in some words in Colville (Mattina (1979), cited by Blevins and Garrett (1998:525)). (53) Colville attraction (Blevins and Garrett (1998:525)) (a) p£as ‘scared’ c-ps-Tdya? ‘senseless’ (b) p£aw ‘he ran down’ pw-an-ciat-olx ‘they make noise running down’ (c) q? w £ay ‘black’ q? oy-lsc£at ‘his clothes are dirty’ (d) s-t-q? w £ay-xon-x ‘Blackfeet’ q? w £ay-lqs ‘black robe, priest’ Attraction of /?/ from the root to the stressed suffix is exemplified in (53a-c). (53d) shows that in words where the root is stressed and the suffix unstressed, /£/ remains in the root. According to Bessell (1992) (cited by Blevins and Garrett (1998:525)), this process might be foot-sensitive. Bagneres-de-Luchon French: In Bagneres-de-Luchon French, a liquid in a complex onset of an unstressed syllable is diachronically attracted to a stressed syllable (Blevins and Garrett (1998:526), Grammont (1905-06)). Some examples are included in (54). Grammont (1933) mentions the liquid was attracted to word-initial syllables, but this does not explain (54f). 6 4 Thanks to Uffe Larsen for bringing this data into my attention. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 114 (54) Liquid attraction in Bagneres-de-Luchon French (Grammont (1933:339), Blevins and Garrett (1998:526)) Stage 1 : Latin Stage 2 Stage 3 (a) 'kapra > 'cra.bo ‘goaf (b) 'wespera:s > *wespras > 'bres.pes ‘vespers’ (c) 'pawperum > *pawpru > 'praw.be ‘poor’ (d) 'kamera > *kambra > 'cram.bo ‘room’ (e) 'tenerum > *tendro > 'tren.de ‘tender’ (f) 'spinula > espingla > es.'plin.go ‘safety pin’ Summary: Glottalization is attracted codas and nuclei in stressed syllables. Aspiration can migrate in a non-stress-sensitive fashion, as in Ancient Greek, where aspiration was attracted to the word-initial position (Grassman Law; Collinge (1985) and references therein). Examples of stress-sensitive migration of aspiration might include Icelandic (see Silverman (1997)). Stress-sensitive voicing attraction, or voicing attraction in other contexts has not been reported to the best of my knowledge. Glottalization and pharyngeals are attracted by stress in a similar manner. Liquids are attracted from a complex onset in an unstressed syllable to the complex onset of a stressed syllable. 6.2 Stress attraction Stress can be attracted to syllables with specific consonantal properties. This is the mirror image of stress-sensitive consonantal attraction. Stress is attracted by syllables with onsets, especially if these are of low sonority. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 115 Table 2.11. Stress attraction Aranda Stress attracted to onsetful a Arandic Alyawarra Stress attracted to o. with (low-sonority ) onset Arandic Pirafaa Stress attracted to o. with (low-sonority) onset Mura Banawa Stress attracted to onsetful a Arauan Iowa-Oto Stress attracted to onsetful o Siouan Aranda: Stress is attracted to onsetful syllables— or, alternatively, repelled by onsetless syllables (Smith (2002) and references therein; Breen and Pensalfini (1999)). Aranda has initial stress if the initial syllable has an onset (55a, b). If the initial syllable is onsetless stress moves to the second syllable (55c). Stress cannot be final even if the word is disyllabic and the first syllable lacks an onset (55d). (55) Aranda stress attraction (Smith (2002)) (a) par.pa ‘quickly’ rar.ta.ma ‘to emerge’ (b) ku.tun.gu.la ‘ceremonial assistant’ wo.ra.ta.ra place name (c) er.gu.ma ‘to seize’ u.lur.ba ‘cold; cold wind’ (d) a:twa ‘man’ il.ba ‘ear’ Alyawarra: Stress is attracted to syllables with low-sonority onsets (de Lacy (2001) and references therein). Examples are given in (56). (56) Alyawarra stress attraction (de Lacy (2001:9), citing Yallop (1977:43)) (a) rin.ha 3rd person pronoun (b) i.li.pa ‘axe’ *f.li.pa (c) ju.kun.tJ a ‘ashes’ *ju.kun.tia (d) wa.lfjm.pa.rra ‘pelican’ *wa.lijm.pa.rra Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 116 Stress is initial by default (56a). When the initial syllable is onsetless stress falls on the second syllable (56b). When the first syllable has a glide as onset, stress falls on the second syllable (56c, d). Compare Niaofou (Polynesian) where glides are avoided in stressed syllables, being realized as high vowel nuclei of a separate syllable (Smith (2002), de Lacy (2001)). Piraha: Piraha has the same pattern as Alyawarra; stress is attracted by syllables with (low-sonority) onsets (Everett (1988), Everett and Everett (1984ab), Smith (2002)). Stress in Piraha is final by default (57a) but is attracted by long vowels and diphthongs elsewhere in the word (57b). If there is a tie stress falls on the syllable with a voiceless onset in preference to a voiced onset (57c), or by an onsetful syllable rather than an onsetless one (57d). If all potentially stress-attracting syllables have voiceless onsets, stress falls on the rightmost of them (57d). (57) Piraha (from Smith (2002), citing Everett (1988). Underline shows high Banawa: . This Arauan language spoken in Brazil has left-to-right moraic trochees with main stress on the initial syllable. In words with more than two syllables with initial onsetless syllables, main stress falls on the second syllable (58) (Buller, Buller and Everett (1993), Everett (1996)). tone)) (a) pao.hoa.hai ‘anaconda’ gi.go.gf ‘what about you?’ gii.so.gi ‘turtle’ bii.sai ‘red’ poo.hoi.hi.ai ‘fish’ (b) ?a.ho.ao.gi proper name (c) kaa.gai ‘word’ (d) gao.ii proper name Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 117 (58) Banawa (from Everett (1996:22)) (a) wanakari (b) uwaria spider one (c) uwi (d) uwia cry go out (as a fire) Iowa-0to: Stress is also attracted to onsetful syllables in Iowa-Oto (Downing (1998) and references therein). Summary: Stress is attracted to onsetful syllables or to syllables with voiceless onsets. It has been suggested that non-sonorous onsets in stressed syllables provide a better perceptual consonant-vowel contour, improving sonority in the stressed syllable as a whole and making stress more perceptible (Smith (2002) and references therein). 7. Deletion and epenthesis This section considers prosodically-determined deletion and insertion. Both laryngeal and non-laryngeal features and consonants delete in weak or unstressed positions. Laryngeal consonants are inserted in strong or stressed positions. 7.1 Deletion Various languages are reported to delete laryngeal and non-laryngeal features and consonants in weak and unstressed positions, both in synchronic and diachronic processes. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 118 Table 2.12. Deletion6 5 English O /h/ind Germanic Chali (Tati) O /h/ in d Indo-Iranian Squamish 0/C /?, R? / in d Central Salish Oneida c /h, ?/ in post-tonic d Iroquoian Popoloca 0 /?,h , 7 in d Otomanguean Capanahua c /?/ in weak footed d Panoan Faroese c Ik/ in unstressed /skt/ cluster West Scandinavian Chilean Spanish c te/ind Romance Old English 0/C [1 ] in coda of d; [w] in onset of d Germanic ‘Romance 0/C Liquid in d Indo-European English: /h/ is not pronounced in unstressed, non-initial syllables (Borowsky (1986), Hammond (1999), Davis (1999, 2002), Davis and Cho (2003)). In (59), the notation <h> represents that /h/ is spelled but not pronounced. (59) Deletion of English /h/ (from Borowsky (1986:271-4), Davis (2002)) (a) Ha'waii 'Hilary (e) 'Le<h>iste Le'histe (b) a<h>! a'ha! (f) 'alco,hol 'Ida,ho (c) 've<h>icle ve'hicular (g) .Tarahu'mara (d) pro<h>i'bition pro'hibit (h) ex<h>i'bition ex'<h>ibit /h/ is pronounced word-initially regardless of stress (59a). /h/ is never pronounced in codas (59b). As a word-medial onset, /h/ is only pronounced in stressed syllables (59c, d). Changing the stress in a word has repercussions for the pronunciation of /hI 6 5 Kehrein (2001) reports that in Straits Salish /?/ deletes in unstressed syllables. This is not clear from the description in Thompson, Thompson and Efrat (1974), which mentions a tendency for /?/ to be deleted in coda. The only mention of stress is that some Northern Straits speakers tend to delete all coda /?/, even in stressed syllables (Thompson, Thompson and Efrat (1974:187)). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 119 (59e). Since D \1 is pronounced in stressed syllables only, speakers will or will not pronounce [h] depending on where they are told stress falls, fhl is pronounced in onset of syllables with secondary stress (59f), and in unstressed syllables between an unstressed and a stressed syllable (59g). Finally, IhJ is not pronounced as a possible second member of an onset (59h). According to Davis (2002), the distribution of fhl in English is parallel to the distribution of aspiration. Both aspiration and /hi have in common the feature [spread glottis]; this feature occurs foot-initially. Davis (2002) proposes that in cases like (59g), where [h] occurs in unstressed syllables, the creation of a superfoot explains the occurrence of [h], A word like fTarahu'mara would be footed [(^a.ra.^hu.Cma.ra)}], with [h] being foot-initial (curly brackets indicate a superfoot). For more details see Davis (2002). Chali Tati: In the Chali dialect of Southern Tati (and probably in other related dialects) /h/ is reported to be ‘most apparent’ initially or as the onset of a stressed syllable; some examples are [how.vae] ‘co-wife’ and [do.'hoi] ‘drum’ (Yar-Shater (1969:34); IP A). From this description, it is ambiguous whether deletion or simply weakening of aspiration occurs in unstressed, non-initial positions. IhJ is reported to usually delete in this dialect in Farsi loanwords in ‘final and preconsonantal positions Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 120 and in secondary contact after a consonant when unstressed’. One example is ['psenj (h)aezar] ‘five thousand’ (Yar-Shater (1969:34)). Additionally, preconsonantal coda fhl deletes but lengthens a preceding short vowel; this process also occurs with coda /?/ and /j/ (Yar-Shater (1969:34, 52)). Stress in Southern Tati is morphophonemic (Yar-Shater (1969:57-8)). Squamish: Morpheme-initial /?/ in unstressed syllables sometimes deletes after a consonant. One example is /s-(?)3q’°i'?tl/ ‘siblins and cousins’; parentheses around /?/ indicate it is usually or always deleted (Kuipers (1967:40)). Additionally, morpheme-final /?/ is deleted in some morphological formations, as in unstressed syllables of plural reduplication (60) (Kuipers (1967:52)). These processes exemplify a tendency to delete /?/ in unstressed syllables, as suggested in Kuipers (1967:55). (60) Plural reduplication: deletion of unstressed /?/ (from Kuipers (1967:52); stress is marked after the relevant vowel) (a) lam? ‘house’ lm-.la'm? ‘houses’ (b) man? ‘child’ ma'n?.-mn ‘children’ The structure of plural reduplication words in Squamish is reduplicant+base. In plural reduplication, stress maintains its position as in the non-reduplicated stem (60a). Reduplicated monosyllabic stems with /a/ have stress on the reduplicated syllable (60b) (Kuipers (1967:100)). For related processes see section 4.3. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 121 Oneida\ Coda /h, ?/ delete in stressed syllables, and in post-tonic syllables before a resonant (Lounsbury (1942), Michelson (1988)). In stressed syllables, coda /?, h1 lower the pitch of the syllable; their deletion in this position causes concomitant lengthening of the preceding vowel (Chafe (1977), Michelson (1988:71)). Accent in Oneida usually falls on the penultimate syllable (see Michelson (1988) for details). (61) Oneida deletion of/h, ?/ in post-tonic syllables (from Michelson (1988:76))6 6 (a) /k-ehyahl-?/ ['ke’.ya.le?] ‘I remember’6 7 (b) /k -? s k o 2 -s / ['ke'S.ko2s] ‘I keep drowning’ (c) /hlaw -ehyahl-a-?-u/ [la.weh.yah.'la’ u] ‘He did remember (d) /wa2-k-?sko?-n?/ [wa2.'ke’ s .k o .n e ? ] ‘I’m drowning, I fell in the water’ Deletion of coda /h, 2/ in stressed syllables and concomitant vowel lengthening is shown in (61a, b, d) (single underline). Bold Ihl and /?/ in (61a, d) delete because they occur in coda of post-tonic syllables. In (61b), double-underlined /?/ remains because it is followed by an obstruent. Pretonic laryngeal codas do not delete ((61c); underlined /?/ in (6Id)). Word-finally, /?/ remains (61a, d). Oneida has an additional stress-sensitive phenomenon concerning 111. Post tonic 1 1 1 is pronounced as IhJ before a single onset obstruent utterance-medially 6 6 Syllabification in (61, 62) is m ine. 6 7 [e] is epenthesized into word-final C? clusters in (61a, d) and (62a). See Michelson (1988:144— 6) for details. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 122 (62a), and utterance-fmally regardless of the number of obstruents (62c, d) (Chafe (1977), Michelson (1988)). (62) Oneida post-tonic /?/-> [h] (from Michelson (1988:76-7, 89) (a) /wa?-k-atkAhla2t-?/ [wa?.kat.'kA’.lah.te?] ‘I stopped’ (b) /wak-atkAhla?t-u-?ne?/ [wa.kat.kAh.la?.'tu\ne?] ‘I had stopped’ (c) /tekatja?ks/ [te.'ka.tjahks] ‘I’m tired’ (d) /lohsu?/ [’loh.suh] ‘he finished’ Underlined /?/ is pronounced as [?] in pre-tonic coda (62a, b). In post-tonic syllables before a single onset /?/ is pronounced as [h] (62a, c) (double underline). Popoloca: In Los Reyes Mazatlan Popoloca, various consonants or features are reported to delete in unstressed syllables, including /?/, pre-aspiration, and pre nasalization (Veerman-Leichsenring (1984:52), cf. Stark and Machin (1977:79)). Post-aspirated /th, tf1 , t§h / lose their pre-aspiration, and /kh, tsh / are pronounced as /x, s/. Additionally, in unstressed syllables there is neutralization between /ts/ and /s/, and between Ini and /n/. See also sections 4.1, 5.1 and 5.3.2. Capanahua: 11 1 is reported to delete in codas of even-numbered syllables starting from the beginning of the word (Loos (1969)). In metrical terms, this corresponds to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 123 weak footed syllables; Capanahua has morale trochees (Safir (1979), Gonzalez (2002a)). (63) Capanahua: deletion of /?/ (from Loos (1969)). (a) /ta?/ declarative modal (b) /ra?/ ‘probably’ (c) [(?o .tji) (ti .ra) (ta? .ki)] ‘It is probably a dog’ (d) [(?o .tji) (ti .ma) (ra?. ta) ki] ‘It is probably not a dog’ The declarative modal /ta?/ maintains the coda /?/ in strong position within a foot (63c). In weak position /?/ deletes (63d). The same occurs with the adverb /ra?/ ‘probably’. See chapter 5 for an extensive analysis of this process. Faroese: /k/ deletes in a /skt/ cluster in unstressed syllables (64) (Hume (2000)); Hume and Seo (2001)). In stressed syllables, /s/ and fkJ metathesize to [kst]. (64) Faroese deletion (Hume (2000)) (a) 'fesk 'fekst ‘fresh-fem. sg./neuter singular’ (b) 'ro.shs.kor 'ro.shs-t ‘Russian-stem/neuter singular’ In (64a), the addition of the neuter singular suffix III to a stressed syllable ending in /sk/ causes the change in order of /sk/ to [ks]; the addition of the same suffix to an unstressed position causes the deletion of fkJ. For more details, see section 8.2. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 124 Chilean Spanish: In Chilean Spanish and many other Spanish dialects, /s/ tends to be aspirated or even deleted in syllable-final position. For example, word-final Is/ in a word like /despues/ ‘later ’ will be pronounced [des.'pweh], [des.'pwe], or, less frequently, [des.'pwes]. Deletion of Is/ syllable-finally in Chilean Spanish is more likely in unstressed syllables, and aspiration and to a lesser extent /s/ retention are more likely in stressed syllables (Cid-Hazard (in preparation)). Old English: Deletion of [1 ] occurred for the weakly stressed variants of should and would in Old English (Kabell and Lauridsen (1984:50)). It is also thought that /w/ deleted in onsets of weakly stressed syllables, as in answer (Kabell and Lauridsen (1984:83) and references therein). Romance: In the development of Romance languages, coda liquids were deleted i in unstressed syllables (as in French pa.te. ho.tre from Vulgar French pa. ter. hos.ter) or in onset positions (as in Spanish a. 'ra.do from a. 'ra.tru) (Grammont (1933:289- 303)). Deferrari (1954:141) reports that in the development from Vulgar Latin to Italian final consonants generally deleted, especially after unstressed vowels. Some examples are 'mar.mo from 'mar.mor ‘marble’, and 'cor.po from 'corpus ‘body’ (Deferrari (1954:184, 186)). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 125 Summary: Featural and segmental consonantal deletion takes place in unstressed syllables and can affect both onsets and codas. In many cases, the features or segments deleted are laryngeal. 7.2 Epenthesis Laryngeal features and consonants are inserted in stressed or strong positions in various languages. Epenthesis of [t] is found in some dialects of English between a sequence [ns] preceding an unstressed vowel. T able 2.13. Epenthesis English 0 [?] phrase-initially and in 'o /p, t, k/->[? p, 7 t, ? k]/ 'V._/foot medially [t]/n_sv Germanic German 0 [?] morpheme-initially and in 'a Germanic Dutch 0 [?] in 'o Germanic Paipai 0 [?] in word-initial 'o Hokan Walapai 0 [?] in 'o Hokan Silacayoapan Mixteco 0 [?] word-initially and foot-initially Otomangean Huariapano c [hi in strong open a Panoan (extinct) Popoloca 0 Stops, affricates-^ [? C]/'V_ Otomanguean English: Dilley and Shattuck-Hufnagel (1996) report a greater likelihood of glottalization of onsetless syllables phrase-initially than phrase medially (beginning of an intermediate or full intonational phrase in the ToBi system), and in accented onsetless syllables than in unaccented onsetless syllables, for all speakers and conditions. See also Pierrehumbert and Talkin (1992) and Pierrehumbert (1995). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 126 In some dialects, as in Tyneside and London English, intervocalic voiceless stops pre-glottalize between vowels, especially if the first vowel has primary or secondary stress (Carr (1999)).6 8 Some examples are clipper ['kli.TpA], fitter ['f i.? tA ] , and hacker ['ha.?kA]. This pattern also occurs when a sonorant immediately precedes the stop, as in grumpy ['gium.?pi], auntie ['a:n.?ti], hankie ['hag.Tki]. Carr (1999) suggests that the relevant context is foot-intemal position, parallel to flapping in American English. In many dialects of English a ‘brief transitional stop’ is optionally epenthesized between the sequence [ns] followed by an unstressed vowel (Clements (1987), Hayes (1995)). One example is ‘mensa’ [W ntso] vs. insane [in'sein] (from Hayes (1995:12,13); IPA). German: In Standard German, [?] is epenthesized as onset morpheme-initially regardless of stress (65a, b) and in primary-stressed syllables (65c, d; cf. with 65 e, f) (Wiese (1996), Alber (to appear), among others). Some speakers also epenthesize /?/ in onsetless syllables with secondary stress (65g, h). (65) Epenthesis of [?] in German (from Alber (to appear:4-6); IPA) (a) Oase [?o.'?a:.z9] ‘oasis’ (b) anerkennen ['?an.-?er.-'ken.nen] ‘to acknowledge’ (c) Poet [po.'?e:t] ‘poet’ 6 8 Tyneside English is spoken around the area of Newcastle-upon-Tyne (Northern England). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 127 (d) naiv [na.'?i:f] ‘naive’ (e) kreativ [jkre.a.'tuf] ‘creative’ (f) Joachim ['jo.a.chim] proper name (g) Canaan ['ka.na.,?an] ‘Canaan’ (h) Michael ['mi.xa.^el] proper name Additionally, a phonetic study conducted by Kohler (1994) and cited by Alber (to appear) finds that stress influences epenthesis morpheme-initially. Epenthesis occurred in morpheme-initial, stressed syllables about 95% of the time, but only about 72% of the time in morpheme-initial unstressed syllables. Dutch: In Dutch, a sequence of two vowels is broken by a glide or a glottal stop (Smith (2002), citing Booij (1995)). Which glide is inserted depends on the nature of the first vowel (66). (66) Epenthesis in Dutch (from Smith (2002), citing Booij (1995:22, 23)) (a) Januari [ja.ny.'qa.ri] ‘January’ duo ['dy.qo] ‘duo’ (b) Diet [di.'jetj ‘diet’ Gea ['re.ja] name (c) Ruanda [ru.1 pan. da] ‘Ruanda’ Houen ['hou.uon] ‘to hold’ (d) Chaos ['xa. os] ‘chaos’ farao ['fa.ra.o] ‘pharaoh’ (e) Paella [pa.'?e.lja] ‘paella’ aorta [a.'?or.ta] ‘aorta’ If the first vowel is front rounded, [q] is inserted (66a). If the first vowel is front unrounded, [j] is inserted (66b). [u] is inserted after non-low back vowel (66c). Glide insertion is insensitive to stress, unlike /?/ insertion. After /a/, no insertion takes Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 128 place if the second vowel is unstressed (66d); if the second vowel is stressed, [?] is inserted (66e).6 9 Paipai: Epenthesis of [?] appears to occur in various Yuman languages under stress. In Paipai, [?] is reported to occur word-initially before consonants and vowels. It is obligatory before initial stressed vowels, as in ['?o%] ‘cough’, and optional before initial consonants followed by a stressed vowel, as in ['pa], ['?pa] ‘man, person’ (Wares (1968:43)).7 0 Stress in Yuman languages falls on the last syllable of the stem (Wares (1968:29)). Walapai: Stressed vowels in unextended root forms are always preceded by /?/ or another consonant, while unstressed vowels may occur without a preceding or following glottal stop. Some examples are uci? ‘coals’ and wasPami ‘doorway’ (Wares (1968:28-29)). Silacayoapan Mixteco: [?] is inserted in onset position word-initially and before stressed vowels, as in /i?.m7 [('?i?.nf)] ‘hot’ (North and Shields (1977:21)). See section 4.4 for related processes. 6 9 Stress usually falls on the penultimate syllable in Dutch, but there are many exceptional forms with final or antepenultimate stress (Van der Hulst (1984), Kager (1989), Booij (1995), among others). 7 0 [x] corresponds to a ‘back velar’ voiceless fricative (Wares (1968:43): cf. 27, 28). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 129 Huariapano: [h] is epenthesized in codas in odd-numbered syllables from the beginning of the word, provided that the following consonant is voiceless: [,yo.mra ,rah.ka. 'tih. ktey] ‘(They) hunted’ (Parker (1994, 1998)). This distribution coincides with default secondary stress in this language, which is assigned in syllabic trochees constructed left-to-right. See chapters 3, 5 fora detailed discussion. Popoloca: Unaspirated stops and affricates are pre-glottalized in onsets after stressed short vowels (67). (67) Popoloca pre-glottalization (from Veerman-Leichsenring (1984:43-4)) (a) [ku:.'tja?.pi] ‘mocking bird’ (d) ['nu?.tse] ‘star’ (b) ['ku?.tu] ‘road-runner’ (e) [ln 3i?.tja] ‘candle’ (c) ['tjd?.ku:] ‘face’ (f) ['kd?.(£u:] ‘maguey (pita)’ Fricatives, liquids, approximants and voiced nasals geminate in onsets after stressed vowels. In the same position, pre-nasalized, aspirated, and voiceless nasal consonants are decomposed into a coda and an onset element (section 4.1). Pre- glottalization of unaspirated stops and affricates is a related process; pre- glottalization resyllabifies as the coda of the preceding syllable. Summary: Epenthesis of laryngeal features and consonants is common in onsets of stressed syllables. Epenthesis of [?] is more common that epenthesis o f [h]. For some languages, this distribution is coherent with a foot analysis in which epenthesis Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 130 takes place foot-initially. Epenthesis of brief transitional elements, as in English, can also be stress-sensitive. 8. Metathesis Metathesis is a reversal in the linear order of adjacent segments or features. It can occur in stressed syllables, as in Faroese, in strong footed syllables, as in Capanahua, or in unstressed syllables (Cayuga, Mohawk, Oneida, Thompson River Salish, Le Havre French). Table 2.14. Metathesis7 1 *Cayuga /CVh, CV?/ to /ChV, C?V/ in non-final weak a North Iroquoian *Mohawk /C?V/ to /CV?/ in post-tonic (weak) o North Iroquoian Oneida /C?V/ to /CV?/ in post-tonic (weak) o North Iroquoian Thompson River Salish /h, ?/ onset/coda in pre-tonic a depending on following segment Interior Salish Capanahua Onset /?/ to coda in strong footed o Panoan Faroese /skt/ to /kst/ in 1 a West Scandinavian *Le Havre French *ra> or before non-labial C in d Romance Cayuga: /h/ and /?/ in non-final odd-numbered syllables underwent coda-to- onset laryngeal metathesis (68a, b) (Chafe (1977:175-178), Foster (1982), Michelson (1988:104), Dougherty (1993), Blevins and Garrett (1998)). This had the effect of ‘weakening’ odd-numbered syllables (Michelson (1988:104)). 7 1 Zoque (Mixe-Zoquean) has onset-to-coda metathesis that achieves a reduction in the number of syllables (Blevins and Garrett (1998)). It is unclear if this case is prosodic. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 131 (68) Cayuga: laryngeal metathesis (from Chafe (1977:177-178); IPA) Proto-Northem Iroquoian Cayuga Gloss (a) *o.cih.'s5?.ta? (b) *keh.nya?.'sa?.keh (c) o.'cih.s?o.ta? ‘star’7 2 ke.'nya?.s?a.keh ‘On my neck’ ke.'ka.ha?.keh ‘my eye’ Metathesis of /?/ did not lead to onset glottalization; /?/ is articulated separately (Chafe (1977:177)). Metathesis of /h/ had the effect of devoicing the preceding vowel (68b). No metathesis of IhJ occurred in even-numbered syllables (68a, b). Metathesis was blocked when the onset is a laryngeal (68c). According to Chafe (1977), odd-numbered syllables in Cayuga are phonologically weaker and never eligible for main stress, which is usually assigned to the rightmost even-numbered, non-final syllable, unless the original accent in Proto-Northem-Iroquoian fell on a long vowel in an open syllable. Footing is iambic; laryngeal metathesis can be described as occurring in the weak element of the iambic foot. Foster (1982:61) reports secondary stress for even-numbered syllables that occur before the primary stress.7 3 For Blevins and Garrett (1998) the origin of metathesis in Cayuga is phonetic; they analyze this process as an instance of perceptual metathesis, by which a consonant in an ambiguous phonetic string is reinterpreted as originating from a non-historical position. They suggest that nuclei of weak syllables in Cayuga were 7 2 Penultimate stress from Proto-Northem-Iroquian is maintained if the penult is long; otherwise stress is placed on the rightmost nonfinal even syllable (Michelson (1988:98— 100) and references therein). 7 3 Dougherty (1993:162— 263), as cited by Blevins and G arrett (1998), did not find any acoustic correlates of secondary stress in Cayuga in spite of the impression of iambic rhythm . Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 132 phonetically shortened. Coda fhl and /?/ overlapped the nuclei of weak syllables; in the case of fhJ, the adjacent vowel was devoiced. CV was interpreted as standing for ChV rather than CVh. The same occurs for coda /?/, which overlapped the preceding vowel with creaky voice. Evidence for their analysis comes from conflicting evidence from orthography as to the position the laryngeal is associated with, and from a phonetic experiment by Dougherty (1993) where ‘laryngeal spreading’ failed to occur when words were read with pauses before each syllable. Mohawk: Some /C?V/ sequences appear to have metathesized to [CV?] in post tonic final position in Mohawk (69) (Blevins and Garrett (1998)). (69) Mohawk metathesis (from Michelson (1988:56); IP A, syllabification mine) (a) /k-at-it?a-s/ [ka.'ti.ta?s] ‘I get into something’7 4 (b) /hs-rihw-is?a-s/ [se.rih.'wi.sa?s] ‘You always promise’ (c) /k-oj?ak-s/ ['ko.ja?ks] T throw’ (d) /wak-nuhs-is?(a)-u-hatje/ [wakenuhsisu?'hatje] ‘I finish the house a little bit at a time’ This case of apparent diachronic metathesis is frequently considered to involve two separate processes of vowel reduplication and subsequent vowel deletion (Lounsbury (1953:87), Postal (1969), Michelson (1988:56-8)). For discussion see Blevins and Garrett (1998:524). 7 4 Stress in Mohawk falls on the penultimate syllable in inflected words (all verbs and most native nouns) and in some uninflected particles. Other uninflected particles and most borrowings have final stress (Michelson (1988:53-4)). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 133 Oneida: Diachronic metathesis of /C?V/ sequences to /CV?/ is reported in post-tonic position, as in Mohawk (Blevins and Garrett (1998:524-5)). Thompson River Salish: Regardless of the underlying position of /h, ?/, these segments surface as codas in Thompson River Salish (Nie?kepmxcin) before tonic syllables starting with a consonant or a cluster beginning with an obstruent (70a, c, e; cf. with 70 b, d). They surface as onsets when preceded by tonic syllables whose onset is null or a cluster beginning with a resonant (70) (Thompson and Thompson (1992), Thompson, Thompson and Edgesdal (1996), Blevins and Garrett (1998)).7 5 (70) Thompson River Salish metathesis (from Blevins and Garrett (1998:513)) (a) (b) (c) (d) (e) (f) (g) (h) (i) Underlvinglv prevocalic /m9kw ?u-t-ejV [mkw u?.tefl /m9kw ?u-en?ih-n-t-ej7 [m9kw ?en? inj] /m?a?-uj9m? xw / [ma?.9ujm? xw ] /m?a?-mm-t-Jej-me/ [m?a?mm.tji.me] /tj ?o-tJ ?6z/ [tr 0 ? .t r ? 6 z ] Underlvinglv postvocalic /tjeh-'ame/ [tjhem] /tjeh- t-ejV [tje.tej] /tfeY-'ame/ [tj? ?'9m] /J-naYz- itj? e?/ [fna?.zi.tJ7 e?] ‘she wraps it’ ‘she wraps, covers it up’ ‘it grows light’ ‘it grows light on me’ ‘it is rather dark’ ‘she puts (things) away’ ‘he fixes it’ ‘she lays out (s.t. to sit on)’ ‘mountain-goat-hair blanket’ In (70g, i) the laryngeals are underlyingly postvocalic and surface as codas because they precede a single consonant (/h/ is not pronounced in (70g) because non-pre- 7 5 Resonants in Thompson River Salish are /m, n, 1 , z, j, y, w , 9, ?*/ (Thompson, Thompson and Edgesdal (1996:615)). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 134 vocalic /h/ is deleted). In (70f, h) the laryngeals are intervocalic but pre-morpheme- boundary; they surface as onsets because they precede an onsetless vowel. In (79f, h) the vowels immediately preceding the tonic syllable delete (Thompson and Thompson (1992:31, 39)). Blevins and Garrett (1998:513) propose that the origin for metathesis in Thompson River Salish is related to coarticulation between the vowel and a following laryngeal. Before a single consonant or a cluster beginning with an obstruent, a laryngeal will make a preceding vowel fully laryngealized. Coda /h, ?/ before a cluster beginning with a resonant will only make the preceding vowel partially laryngealized. Fully laryngealized vowels are interpreted as originating from /V?/, while partially laryngealized vowels are interpreted as originating from /?V/. See Blevins and Garrett (1998) for details. Capanahua: Loos and Loos (1998:21-2) report onset-to-coda III metathesis in odd-numbered syllables, especially in fast speech. If this syllable already has a coda /?/, the onset I'll deletes. (71) Capanahua metathesis of/?/ (Loos and Loos (1998:21, 22); IP A) (a) /ketsin-?ino/ [ke.(‘ tsl.)(l?.no)] ‘painted tiger’ (b) /toa?i?ka/ [('to.a.)(i?.ka)] ‘mattress’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 135 Since Capanahua has moraic trochees (Safir (1979), Gonzalez (2002a)), onset-to- coda metathesis of /?/ occurs in strong footed syllables. In this language, metathesis ‘deletes’ the onset to produce a coda, a position which is cross-linguistically dispreferred. In chapter (4) I propose that metathesis creates a rhythmic contrast between strong and weak footed syllables. See also section 7.1 in this chapter. Faroese: The sequence /skt/ after a vowel or nasal metathesizes to [kst] in stressed syllables (72). In unstressed syllables, Dd deletes from the sequence (section 7.1). These sequences arise from suffocation of III. (72) Faroese metathesis (from Hume (2000)) (a) 'dansk ‘danks-t ‘Danish-fem. sg./neuter singular’ (b) 'svensk 'svsnks-t ‘Swedish-fern. sg./neuter singular’ (c) ‘fesk 'feks-t ‘fresh-fem. sg./neuter singular’ (d) ro.sbs.kor lr0 .s1 is-t ‘ Russian-stem/neuter singular’ (e) 'falsk 'fals-t ‘false, stem/neuter singular’ (f) 'ir.kir 'ir.ti ‘write poetry, 2n d /3r d p. pres./past sg’ Metathesis occurs in stressed syllables (72a-c); /k/ deletes from the /skt/ cluster in unstressed syllables (72d). Dd also deletes from a /skt/ (72e) or /kt/ (72f) cluster in a stressed syllable if preceded by a liquid. Hume (2000) and Hume and Seo (2001) suggest that Faroese metathesis is triggered by perceptual optimization. Moving Dd to postvocalic or post-sonorant Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 136 position in a cluster enhances the perceptibility of /k/ due to the availability of a vowel or sonorant transition. Le Havre French: Diachronically, the sequence /*ra/ metathesized to [or] in unstressed syllables, except when the sequence was immediately followed by If, v, m/. Furthermore, in unstressed syllables before /f, v, ml, /*or/ became [ra] (73) (Grammont (1909), cited by Blevins and Garrett (1998:516)). (73) Le Havre French metathesis (Blevins and Garrett (1998:517); syllabification mine) Unstressed /*ra/ to fori before consonants other than If. v. m/ (a) bretelle [bor.'del] ‘suspenders’ (b) brebis [bor.'bi] ‘ewe’ (c) grenier [gor.'nje] ‘granady’ (d) gresil [gor.'zi] ‘sleet’ (e) ecrevisse [e.kra.'vif] ‘crayfish’ (f) crevasse [kra.'vaj] ‘crack’ Stressed /*or/ to Iral before If, v, m/ (g) epervier [e.pro.'vie] ‘sparrow-hawk’ (h) ferme [fro.'me] ‘closed’ (i) fourmi [fra.'mi] ‘ant’ (j) gargouiller [gar.du.'je] ‘gurgle’ In (73a-d), /ra/ metathesizes to /or/ in an unstressed syllable. Metathesis does not occur before /v/, as shown in (73e, f). In (73g-i) /or/ metathesizes to /ro/ before Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 137 /m, v/; metathesis does not occur before other consonants (73j). One question is why If, m, v/ are key in this process. Summary: Metathesis tends to occur in unstressed syllables. It tends to affect features and segments ‘whose acoustic or perceptual cues have relatively long durations’ (Blevins and Garrett (1998:511)), including laryngeals and rhotics. Metathesis can also occur in stressed syllables. In both cases, the motivation for this process appears to be perceptual. In unstressed syllables, metathesis appears to take place because of perceptual ambiguity as to the original position of the consonant; in stressed syllables, in order to improve the perceptibility of a consonant. See discussion in chapter 3. 9. Dissimilation Dissimilation refers to featural change that makes a segment less like an adjacent segment. Reported cases of stress-sensitive dissimilation are diachronic and include Gothic and other Romance languages. Table 2.15. Dissimilation *Gothic O/C Voice dissimilates in a East Germanic *Romance o,c Liquids dissimilate in < 5 Indo-European Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 138 Gothic: Thumey sen’s Law in Gothic refers to the dissimilation in voicing of a consonant from that of a preceding obstruent after an unstressed syllable (Chomsky and Halle (1968), Mosse (1956), Collinge (1985)). Dissimilation seems to occur only in three-syllable words (74). Stress in Gothic falls on the first syllable (Wright (1957)). (74) Gothic dissimilation (Chomsky and Halle (1968:351), Mosse (1956:68); Collinge (1985:183); IP A, syllabification mine). (a) ['ha.tis] ‘hatred’ ['ha.ti.zis] ‘hatred (gen. sg.) (b) fa.gis] ‘fright’ [’a.gi.sis] ‘fright (gen. sg.)’ (c) ['<J>as.tuJ3.ni] ‘position’ ['$rais.tuJ3.ni] ‘temptation’ (d) ['wal.du^.ni] ‘force’ ['wun.du^.ni] ‘wound’ In (74a, c), there is dissimilation from a voiceless obstruent. In (74b, d) there is dissimilation from a voiced obstruent. In (74c, d) dissimilation is between coda and onset within the same syllable. Romance: Stress-sensitive dissimilation is common in Romance languages according to Grammont (1933). These changes have to do mostly with liquid dissimilation, and follow the pattern r-'r— >l-'r and 'r-r-Vr-l (Grammont (1933:276)). That is, in a word with two rhotics in adjacent syllables, the rhotic in the unstressed syllable dissimilates to a liquid (or a nasal in some instances) from a stressed syllable. Examples include Vulgar Latin al berga from *arberga, Spanish 'manual Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 139 from Latin marmor, and Italian pette 'grinu from Latin peregrinum (Grammont (1933:276-286)). Posner (1961) suggests that changes among liquids are relatively frequent in Romance languages since [1, r] are very similar phonetically and are equally important in the phonemic structure of these languages (Posner (1961:101)). She also mentions that in many Romance languages and dialects, [r] deleted or dissimilated in unstressed syllables because of the influence of another [r]. However, dissimilation in stressed syllables is also attested; one example would be ta ' Iadro from Latin taratru (Posner (1961:61, 108)). Summary: Stress-sensitive dissimilation is reported diachronically. In the case of Romance, the dissimilation change tends to occur in unstressed syllables, with the segment triggering dissimilation belonging to a stressed syllable. In the case of Gothic, dissimilation occurs after unstressed syllables. See discussion in chapter 3. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 140 C h a p te r 3: Typological Generalizations and Their Underpinnings 1. Introduction This chapter describes the typological generalizations that arise from the survey presented in chapter 2. These comprise generalizations about the consonantal processes and whether they occur in stressed, unstressed, strong or weak syllables; whether stress or foot structure (or both) are relevant, and whether word-position and morphological structure are related. Additionally, this chapter examines whether there is a relationship between syllabic position (onset/coda) and these processes, whether these processes are variable or categorical, and which segments and features are more likely to be affected. This chapter also examines the factors that underlie stress- and foot-sensitive consonantal phenomena: aerodynamics, duration, perception, and rhythmic prominence. It classifies the phenomena surveyed according to them, and it proposes that these are the factors that are minimally needed in order to provide an adequate characterization of these phenomena. This chapter also discusses in detail the role of stress and foot structure in consonantal phenomena. Three main cases are found: processes where stress strictly influences consonantal processes (as in frication in North-Central Peninsular Spanish); processes where stress and foot structure converge in conditioning these processes (as in flapping in American English); and processes where foot structure Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 141 alone is a conditioning factor (as in deletion in Capanahua). Where each of these processes is found is examined in detail. Importantly, it is shown that strictly foot- sensitive consonantal phenomena are driven by rhythmic considerations, that is, the creation of a prominence contrast between strong and weak syllables within a foot. The creation of a rhythmic contrast can compensate for the lack of persistent stress, or it can go against stress assignment in some cases. This chapter reviews previous approaches that can account for certain stress- sensitive consonantal alternations, including Licensing-by-Cue (Steriade (1997)) and contrast reduction (Crosswhite (to appear)) for perceptually-based phenomena, and prominence augmentation (Smith (2000, 2002)) for prominence-based phenomena. It is shown that the mechanism of Prominence Alignment (Prince and Smolensky (1993)) can account for stress- and foot-sensitive consonantal phenomena involving prominence reduction and augmentation. I also propose that Prominence Alignment can also be extended to account for phenomena where duration and aerodynamics are relevant. The organization of this chapter is as follows. Section 2 states the typological generalizations concerning stress and foot structure in consonantal alternations. Section 3 discusses the typological generalizations related to syllabic, word and morphemic position, and the type of features and segments affected. Section 4 discusses the factors underlying consonantal processes: perception, aerodynamics, duration and prominence. Sections 5 and 6 describe the consonantal alternations that are predicted not to occur, and what processes are in the grammar. Sections 7 and 8 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 142 deal with the separation between stress and footing and the notion of rhythmicity. Sections 9 through 12 discuss how prominence-based, durational, aerodynamic and perceptual phenomena can be analyzed in the grammar. Finally, section 13 is the conclusion. 2. Typological generalizations (I): stress and foot structure This section states the generalizations that arise from the survey of stress- and foot- sensitive consonantal phenomena discussed in chapter 2. (1) lists the generalizations regarding the relationship between stress/footing and consonantal processes. The factors behind the processes in (1) fall into four groups: aerodynamics, duration, perception and prominence. These are discussed in detail in section 3 in this chapter. For more details about specific consonantal phenomena see chapter 2. (1) Relationship between stress/footing and consonantal alternations Stressed/strong-footed position Fortition occurs in stressed syllables and in foot-initial syllables. Consonantal features and segments are attracted to stressed syllables. Stress is attracted by onsetful syllables and by low-sonority onsets. Consonantal epenthesis occurs in stressed syllables and in strong-footed syllables. In laryngealized consonants, the realization of glottalization and aspiration is timed according to stress: glottalization and aspiration will be realized as close to a stressed vowel as possible. In stressed syllables, voiceless consonants become voiced, but voiced consonants do not become voiceless. Gemination, pre-aspiration and pre-glottalization occur after stressed syllables. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 143 Unstressed/weak-footed position Lenition occurs in unstressed syllables and in foot-medial syllables. Flapping occurs word-medially in unstressed syllables. Consonantal deletion occurs in unstressed syllables and in weak-footed syllables. In unstressed syllables, voiced consonants are devoiced and voiceless consonants are voiced. All positions Consonants are longer in stressed syllables than in unstressed syllables. VOT for obstruents is longer in stressed syllables than in unstressed syllables. Metathesis and dissimilation can occur in stressed syllables or in unstressed syllables. An important finding of the survey is that consonantal processes fall into three groups according to whether stress or foot structure condition them (2). (2) Types of consonantal processes regarding stress and foot structure (i) Stress only Consonantal processes can be strictly influenced by stress in the absence of foot structure (as in frication in North-Central Peninsular Spanish). (ii) Stress and/or foot structure Consonantal processes can be ambiguous as to whether stress and/or footing condition them since both coincide (as in flapping in American English). (iii) Foot structure only Consonantal processes can be strictly influenced by foot structure in the absence of stress (as in deletion in Capanahua)._________________________ Strictly stress-sensitive consonantal processes are predicted to potentially occur in languages with only one stress per prosodic word and no evidence for persistent footing (as in Senoufo; see section 7.1) and in languages where, even if footing is persistent, a consonantal alternation occurs only in a subset of the stressed syllables Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 144 in a word, namely, only in main-stressed syllables, or only in secondary-stressed syllables (as in Urubu-Kaapor; see section 7.1). For processes conditioned to some extent by stress, there are three possibilities. The stress of the syllable where the consonant occurs might be relevant, as in epenthesis in German. Alternatively, the lack of stress of the syllable where the consonant occurs might influence its realization, as in flapping in English. Finally, the stress of the immediately preceding syllable might be affecting consonant realization. Gemination in Popoloca exemplifies this last possibility. A summary of the consonantal processes that are attested in each of these contexts is given in (3). (3) Consonantal processes by stress contexts Stressed syllables Unstressed syllables After stressed syllables Longer duration Longer VOT Fortition Post-aspiration Post-glottalization Voicing Metathesis Dissimilation Epenthesis Flapping Lenition Voicing and devoicing Metathesis Dissimilation Deletion Gemination Pre-aspiration Pre-glottalization In consonantal processes where both stress and foot structure are possible conditioners, consonantal alternations occur in stressed/strong-footed positions or in unstressed/weak-footed positions. The coincidence of stress and footing is typical of trochaic systems, where the head is foot-initial and stressed and the non-head foot- final and unstressed. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 145 Consonantal processes where only foot structure is relevant occur in three cases: (i) systems where not all feet are assigned stress, as in Capanahua; (ii) iambic systems (as Norton Sound Yupik): and (iii) systems with mismatches between stress and foot structure (as in Huariapano). Each of these cases is explored in section 7.3. Finally, some metrically-conditioned consonantal processes are tendencies or optional (frication in North-Central Peninsular Spanish), while others are fully regularized (deletion in Capanahua). The division between these two processes will be connected to the distinction between phonetics and phonology in section 6. Sections 4— 8 discuss the factors behind stress- and foot-sensitive consonantal phenomena, the distinction between strictly stress-sensitive, strictly foot-sensitive and foot/stress-sensitive consonantal processes and the fact that some processes are optional or categorical. Section 3 discusses other findings from the survey, including the interaction of stress- and foot-sensitive consonantal alternations with syllabic positions, laryngeal features, and word and morphological position. 3. Typological generalizations (II): syllabic, word and morphemic position; features and segments affected 3.1 Syllabic positions In most cases consonantal processes that are influenced by stress or foot structure occur in onset position (as in epenthesis in German) or depend on onsets for their occurrence (as in gemination in Popoloca). However, there are a large number of Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 146 languages in the survey where codas are absent in the language (as in Maori; Bauer (1993)) or are seriously restricted as to segmental content (as in Silacayoapan Mixteco; North and Shields (1977)). This leaves uncertain whether the correlation between onset position and stress or foot structure is significant. Table 3.1 provides examples of processes that do not exclusively occur in onset position. Processes where it is not clear whether onsets and codas are involved include fortition in Yuman and voicing variation in Wasco-Wisram. Note that glottalization is frequently attracted by coda and nuclei. Table 3.1. Consonantal alternations involving codas Coda only Guayabero Allophonic variation of /?/and /d/ Northern Peninsular Spanish Frication and devoicing of /b, d, g/ Capanahua Deletion of /?/ Huariapano Epenthesis of /h/ Coda and onset Spanish Longer consonant duration Guayabero Allophonic variation IhJ, fb! Secondary articulation German Post-aspiration English Voicing o f ‘x’ Gemination Onset-to-coda (all cases) Coda-to-onset (inverse gemination in Italian) Pre- aspiration All cases Pre- glottalization All-cases 3.2 Features and segments affected The majority of the consonantal processes considered in the survey have alternations involving laryngeal segments (/?, h/) or laryngeal features (aspiration, glottalization, and voicing). Stress-sensitive aspiration in the form of longer or stronger VOT, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 147 frication, or affrication of voiceless stops occurs in a range of languages for segments that are underlyingly unaspirated. For segments with underlying aspiration, pre-aspiration occurs after stressed syllables, and post-aspiration in stressed syllables. Stress-sensitive deletion and epenthesis of /h/, deletion of aspiration, and stress-sensitive variation between [h] and [?] are also attested. Debuccalization can also produce [h]; metathesis of fbJ is also common. Glottal stops are frequently deleted, epenthesized, or metathesized in a stress- sensitive fashion. Debuccalization can give rise to [?]. Glottalization is frequently attracted to codas or nuclei of stressed syllables. Non-glottalized consonants are pre- glottalized after stress. Globalized consonants have variable timing of glottalization according to where stress is assigned. Stronger plosion of 111 and stronger glottalization in sonorants is also attested in stressed syllables. Glottalization can also delete in unstressed syllables. Finally, voicing features are involved in various processes, including voicing and devoicing in unstressed syllables and voicing in stressed syllables. Voicing is also involved in flapping; regardless of the underlying voicing of the consonant undergoing flapping, the resulting consonant is voiced. Voicing is also involved in Nganasan rhythmic gradation. Non-laryngeal processes generally involve sonority and durational properties of segments (as in gemination, fortition, lenition and flapping). Deletion affects a range of consonants and segments, including liquids, /w/, Dd, and nasalization. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 148 Stress-sensitive attraction can affect liquids and pharyngeals; metathesis can affect liquids and /k/; dissimilation affects liquids. Pointing to the pervasiveness of laryngeal features and consonants in stress- and foot-sensitive phenomena is the fact that laryngeals lack supra-laryngeal features. Unless laryngeals co-occur with pharyngeal or uvular consonants in a language, in which case they are pharyngeal, laryngeals have no place of articulation (Rose (1996)). Evidence for their lack of place is provided by their phonological behavior cross-linguistically; laryngeals often behave phonologically as if they were placeless in terms of vowel copying, debuccalization, epenthesis, reduplication and other processes (e.g. Steriade (1987), Yip (1991), Stemberger (1993), McCarthy and Prince (1994), Halle (1995), Rose (1996), Ladefoged (1997), Parker (2001)). Since there are no supra-laryngeal gestures involved in laryngeal consonants, they are plausibly easier to delete, insert, and relocate. Another consideration is the close connection between laryngeal features and segments and stress. Variations of suprasegmental features such as pitch and stress are partly due to changes in laryngeal behavior (Ladefoged (1993)); it is reasonable that laryngeals are mostly influenced by stress. 3.3 Word-position, phrase and utterance position, and morphemic structure This section discusses the interaction between stress and word position (3.3.1), stress and phrase and utterance position (3.3.2) and morphemic structure and stress in consonantal alternations (3.3.3). Roots, root-initial, and word-initial and final Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 149 syllables are psychologically prominent; they are important for processing and word recognition. This suggests that these positions would be relevant for any type of segmental process taking place. Previous discussion of the interplay between segmental processes and stress and/or word/morpheme position is found in Beckman (1998), Lavoie (2001), Walker (2001) and Alber (2002). 3.3.1 Word position and stress This section examines the interaction of consonantal alternations with word-initial, word-final, word-medial, and intervocalic positions. • Word-initial syllables are relevant in stress- and foot-sensitive consonantal alternations in five cases: (i) stressed word-initial syllables; (ii) unstressed word- initial syllables; (iii) stressed and word-initial syllables; (iv) foot-initial and word- initial syllables. In some cases it is ambiguous whether stress or word-initial position is relevant (v). (i) Stressed word-initial: Some consonantal alternations occur in stressed word- initial syllables, as epenthesis in Paipai and German, which is obligatory or more likely if the syllable is stressed. In German, epenthesis also occurs in stressed syllables elsewhere. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 150 (ii) Unstressed word-initial: Some consonantal alternations occur in unstressed word-initial syllables, as in Farsi, where underlying voiced (lenis) consonants are partially devoiced in onsets in this position but fully voiced in onsets of stressed word-initial syllables or intervocalically (in codas, there is total or partial devoicing). (iii) Word-initial and stressed syllables: A process might occur both word- initially and in stressed syllables. Examples include post-glottalization in Gitksan, epenthesis in Silacayoapan Mixteco, deletion blocking in English and Chali Tati, and strong aspiration in English, Pattani and Farsi. (iv) Word-initial and foot-initial syllables: Fortition is reported to occur word- initially and foot-initially in Aluttiik Yupik (Van de Vijver (1998)). Chapter 2 argues that word-initial positions in Alutiik Yupik can be collapsed with foot-initial positions. (v) Ambiguity between word initial and stressed position: In languages where word-initial syllables are stressed, it is ambiguous whether stress or word position is relevant for a consonantal alternation. This occurs in Wembawemba and Wergaia, where word-initial syllables are resistant to total devoicing. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 151 • Word-final position is relevant for stress- and foot-sensitive consonantal alternations in two cases: (i) processes occurring in word-final, post-tonic syllables, and (ii) processes occurring in word-final syllables and in unstressed syllables. (i) Word-final post-tonic syllables: In Mohawk epenthesis occurs in word-final post-tonic (unstressed) syllables. Stress usually falls on the penultimate syllable. (ii) Unstressed syllables and word final syllables: Certain phenomena occur in unstressed syllables and in word-final syllables regardless of stress, as in English flapping or in pre-glottalization in Gitksan. • Word-medial position is relevant for stress- and foot-sensitive alternations in two cases: (i) processes occurring in word-medial unstressed syllables, and (ii) consonantal alternations occurring word-medially after stress. (i) Word-medial unstressed: Flapping occurs in unstressed word-medial syllables (stress-insensitive flapping generally occurs word-medially). Voicing alternations frequently occur word-medially in unstressed syllables, as in Popoloca. (ii) Word-medial after stress: Stress-sensitive pre-aspiration and gemination typically occur word-medially after stress. English pre-glottalization occurs in the same context. In Paamese, voicing of word-medial /nt/ is more common after stress. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 152 • Intervocalic positions are relevant for stress- and foot-sensitive consonantal alternations in two cases: (i) in unstressed syllables, and (ii) after stress. (i) Intervocalic unstressed: Typically, lenition occurs intervocalically in unstressed syllables (English, Senoufo, Somali, Spanish, Kupia, and Silacayoapan Mixteco). Voicing occurs in this same position in Senoufo. (ii) Intervocalic after stress: Pre-aspiration occurs in this context in Toreva Hopi. Summarizing, word position interacts with stress often. Word-initially, and leaving aside ambiguous cases where it is not clear whether word position or stress are relevant, consonantal processes might be more likely (or obligatory) if the word- initial syllable is stressed, or they can occur both in stressed syllables and word initially. Word-finally, consonantal processes might occur in word-final post-tonic syllables or both in unstressed syllables and word-finally regardless of stress. Word- medially and intervocalically consonantal alternations occur in unstressed syllables or after stress. Farsi and Mohawk do not conform to these generalizations. 3.3.2 Phrase/utterance position and stress Phrase- or utterance-initial: Lenition in /b, d, g/ in Spanish—which is stress sensitive in some dialects—is blocked in utterance-initial positions. In English, epenthesis of/?/ occurs phrase-initially, especially under stress. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 153 Utterance-final: In Oneida, /?/ was replaced by [h] utterance-medially in post tonic syllables preceding a single obstruent. In post-tonic utterance-final position, this process occurred regardless of the number of consonants after /?/. Summarizing, blocking of lenition and epenthesis occurs utterance or phrase- initially. The case of the influence of utterance-final position is Oneida is not clear. 3.3.3 Morphemic structure and stress Stem/root: In Norton Sound, gemination occurs after a monosyllabic stem so that the stem is closed and can be stressed. In Shuswap, a stressed root attracts glottalization. In Paamese, alternations are found in reduplication; the onset of the reduplicated syllable—which is unstressed—has the weakest consonant, and the syllable from the base— which is generally stressed—the strongest consonant. Suffix: Contrary to related Shuswap, in Colville a stressed suffix attracts glottalization away from the root. Morpheme-initially: Epenthesis of [?] occurs morpheme-initially in German, especially under stress, and also in stressed syllables regardless of morphemic structure (Kohler (1994), cited by Alber (to appear)). In Squamish, deletion of /?/ is reported to occur morpheme-initially in unstressed syllables after a consonant. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 154 Morpheme-finally: Deletion of 11 1 is reported to occur morpheme-finally in unstressed syllables in plural reduplication in Squamish (Kuipers (1967)). Summarizing, there are not many cases in the survey where morphological structure interacts with stress or foot structure. Sometimes, stress- or foot-sensitive consonantal processes occur in specific morphemes in the word, as in gemination in Norton Sound, attraction in Shuswap and Colville, fortition in Paamese, and deletion in Squamish. In other cases, a process might occur in a specific morpheme position and in stressed syllables, as in epenthesis of /?/ in German. 4. The role of perception, aerodynamics, duration, and prominence This section discusses the factors that can be seen at work behind stress and/or foot- sensitive consonantal phenomena: aerodynamics, duration, perception, and prominence. These are defined below. In some cases, different factors might be at play at the same time; in other cases, it is unclear what factors are relevant. Sections (4.1-4.4) discuss in detail the consonantal alternations falling under each factor. • Perception: A process can occur to improve the perceptibility of features or consonants. One example is Shuswap, where glottalization is attracted to the final sonorant of a stressed syllable. The realization of the glottal feature next to a stressed Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 155 vowel improves the perceptibility of this feature (Steriade (1997) and references therein). Alternatively, a process can arise in order to avoid features or consonants in perceptually-weak positions. This occurs in Faroese, where fkJ is deleted in the sequence /skt/ in unstressed syllables, /k/ deletes because the unstressed syllable cannot support its cues in this cluster (Hume (2000) and references therein; Hume and Seo (2001)). Last but not least, perceptual ambiguity can motivate consonantal processes. This occurs in certain cases of metathesis, especially in unstressed syllables (Blevins and Garrett (1998)). Features or segments with long duration overlap a vowel and possibly other segments, and because of perceptual difficulties in unstressed syllables, speakers might reinterpret the original position of the feature or segment, causing metathesis. One instance is Cayuga, where Pal and /?/ metathesize from onset to coda in unstressed syllables. • Aerodynamics: The aerodynamics of stress can drive consonantal alternations. Stressed syllables involve increased airflow (Lehiste (1970)), which can lead to an increase of aspiration, frication, or affrication. Increased airflow in stressed syllables can also lead to higher subglottal pressure, which in turn would induce more likelihood of voicing. The aerodynamic effect of stress can also affect voicing in unstressed syllables. Voicing has exact and complex physiological requirements: typically, the vocal cords have to be lightly adducted and there must be enough air passing through them Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 156 (Ohala (1983)). In unstressed syllables, consonantal voicing might be unstable since there is arguably less time to achieve the necessary configuration for voicing. One relevant example from the survey in Silacayoapan Mixteco. • Duration: Stressed syllables are longer than unstressed syllables, and stress can increase the duration of vowels and consonants (Lehiste (1970), Hayes (1995)). In unstressed syllables the duration of vowels and consonants is typically reduced. Which segments are affected by stress in this respect depends on the specific language. Both consonants and vowels might be longer in stressed syllables, as in English (Lavoie (2001)); only vowels might be longer in this position, as in Swedish (Botinis, Bannert, Fourakis and Pagoni-Tetlow (2002)); or only consonants might be longer in stressed syllables, as in Senoufo (Mills (1984)). Apart from being a correlate of stress, duration can also serve to increase prominence. However, in spite of this possible ambiguity, I will refer to durational processes as processes where stress or lack thereof affect the duration of a consonant (or a vowel) independently of prominent considerations. • Prominence: The term prominence is used frequently in connection to certain positions, including stressed syllables, word-initial syllables and others, which are phonetically or psycholinguistically ‘salient’, that is, they play a role in word- recognition, are longer, and so on (Beckman (1998) and references therein). In this work I will refer to prominence as the presence of a property that makes a syllable Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 157 phonologically stronger or weaker. When a prominence contrast is created in the foot structure of a language, I refer to the specific prominence process as rhythmic. Prominence goes beyond perceptual, durational or aerodynamic factors and can potentially create rhythmic contrasts in a language. Prominence causes strengthening of stressed syllables and reduction of unstressed syllables. Sometimes, both of these go hand in hand. Strengthening and reduction have been well-noted for vowels; in unstressed syllables, vowels tend to reduce to [s] and even delete, while in stressed syllables, vowels tend to lengthen. Consonants can also be affected by prominence. Stress is usually present in prominence-sensitive phenomena, but it does not have to be, as exemplified by strictly foot-sensitive processes like Capanahua (Loos (1969), Safir (1979), Gonzalez (2002a)). 4.1 The role of perception Perception-driven processes cause improved perceptibility of features and segments in stressed or strong syllables and the avoidance of features and segments in positions where they are only weakly perceptible. It can also drive alternations through ambiguity in perceptually-challenged positions. Perception is important for some types of attraction, metathesis, variable timing of aspiration and glottalization in laryngealized consonants, and deletion. Perception-driven processes might also cause the voicing of voiceless segments in unstressed syllables and aspiration of voiceless obstruents in stressed syllables. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 158 Perception and attraction: Stress-sensitive featura! attraction seems to be driven by perceptual motivations. In Shuswap, [constricted glottis] in an unstressed syllable is attracted to the stressed syllable. Feature attraction from unstressed to stressed syllables improves the perceptibility of a feature or the chance that a feature will be perceived. This is because stressed syllables are typically longer, louder and better able to carry perceptual cues (Steriade (1997)). For stress-sensitive segmental attraction both perception and prominence might be underlying. See section 4.4 for more details. Perception and metathesis: Perceptually-based metathesis is attested within stressed and unstressed syllables. In unstressed syllables it occurs because of perceptual ambiguity. This type of metathesis is known as perceptual metathesis (Blevins and Garrett (1998)). In stressed syllables, metathesis occurs in order to improve the perceptibility of a feature or segment. Perceptual metathesis occurs with features or segments with relatively long acoustic of perceptual duration cues, including aspiration, glottalization, rhoticity, and pharyngealization (Blevins and Garrett (1998)). Such features or segments are long enough to overlap vowels. In some cases, they overlap vowels fully, and there is ambiguity as to what the original position of the consonant is. This is more likely in unstressed or weak syllables because vowels are shorter and the consonant can overlap the vowel to a greater extent. Full consonantal overlap in the vowel potentially causes metathesis, since the speaker might reinterpret the consonantal Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 159 feature or segment as originating from a different position (Blevins and Garrett (1998)). One example is laryngeal metathesis in Cayuga. Faroese is the only language in the survey where metathesis occurs within stressed syllables. In /skt/ clusters in stressed syllables, IkJ and Is/ metathesize, with /k/ being realized immediately after the stressed vowel. This makes the cues for IkJ to be better perceived (Hume (2000), Hume and Seo (2001)). In unstressed syllables, IkJ in a /skt/ cluster deletes. At first sight, the case of Faroese might seem to contradict the preferential targeting of unstressed syllables by metathesis. However, the generalization is that IkJ is not supported in a /skt/ cluster in Faroese. Only a stressed syllable can retain the segment at all, which it accomplishes by means of metathesis. The deletion that takes place in unstressed syllables effectively removes the potential for metathesis to occur. While perceptually-based metathesis occurs both within stressed and unstressed syllables, the specific motivation differs for both cases. Perceptual metathesis in unstressed syllables does not make a segment more easily perceived; this type of metathesis is symmetrical; CV>VC cases coexist with VC>CV cases for all segments undergoing perceptual metathesis. One example is Le Havre French, where both *ro>or and *ar>ro metathesis co-occur (Blevins and Garrett (1998) and references therein). This type of metathesis arises though perceptual ambiguity of a phonetic string, which might lead to phonological reinterpretation of the original position of a feature or segment (Blevins and Garrett (1998)). On the contrary, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 160 metathesis within stressed syllables improves the perceptibility of salient features or consonants—that is, salient features or consonants occur as close to the stressed syllable as possible. It is up for discussion whether in the last case metathesis also serves to improve ease of articulation. Perception and variable timing o f glottalization and aspiration: Perception-driven processes cause variable timing of glottalization or aspiration in laryngealized segments: glottalization and aspiration surface where they can be best perceived (Steriade (1997); cf. Howe and Pulleyblank (2001)).1 For example, intervocalic globalized stops in Coast Tsimshian are pre-glottalized if the preceding vowel is stressed, and post-glottalized if the following vowel is stressed. Stress conditions where glottalization occurs, overriding the general preference for glottalization to occur after the release of obstruents, that is, as post-glottalization (Sapir (1938), Kingston (1985)). The reason is that stressed syllables are longer, louder and better able to carry perceptual cues (Steriade (1997)). Perception and deletion: Consonantal deletion frequently occurs in unstressed syllables, which are shorter and less loud and consequently not ideal for the perception of features. For example, in Faroese /skt/ clusters in unstressed syllables /k/ deletes because the cues for /k/ are not well perceived in this context (Hume 1 Howe and Pulleyblank (2001) examine cases where variable timing of glottalization is not perceptually-based. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 161 (2000), Hume and Seo (2001)). In other instances of deletion, durational and prominence factors might play a role (sections 4.3,4.4). Perception and VOT: The occurrence of longer or stronger aspiration in voiceless segments in stressed syllables can be partly attributed to perceptual reasons. Aspiration can be better perceived in perceptually salient syllables, including word- initial and stressed syllables. One example is English, where aspiration is more audible in these more perceptually salient positions. Aerodynamic and durational factors might also be relevant for longer and stronger aspiration for obstruents in stressed syllables. Perception and voicing: Voicing pattern (1)—voicing o f voiceless segments in unstressed syllables—might be perceptually-motivated. Reduced duration of voiceless segments in unstressed syllables might give the percept of voicing, since voiced obstruents are shorter than voiceless ones (Lavoie (2001)). One possible example is Senoufo. Alternatively, voicing contrasts might neutralize in unstressed syllables because this position is perceptually weak (Beckman (1998)). Voicing alternations might also be caused by aerodynamic factors, as discussed in 4.2. 4.2 The role of aerodynamics Aerodynamics as a factor behind stress-sensitive consonantal processes refers to the increase in airflow that occurs in stressed syllables and that might lead to increased Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 162 aspiration or increased subglottal pressure. It also refers to the complex aerodynamic requirements for voicing that are plausibly more difficult to achieve in unstressed syllables because of their reduced duration. Aerodynamics and increased airflow. Stressed syllables have increased airflow (Ladefoged (1982)), which, depending on the configuration of the supraglottal cavity, and provided the glottal folds are apart, can lead to (post) aspiration, ffication, or afffication, or a greater strength or likelihood for their occurrence in this position. Longer and stronger aspiration o f voiceless stops and affricates in English, increased likelihood of ffication of /b, d, g/ in coda in Northern Peninsular Spanish and afffication of voiceless stops in Maori are examples of aerodynamic processes. For English, it has been suggested that /h/ deletion is connected to the lack of aspiration of voiceless stops and affricates in unstressed syllables (Davis (2002)). Increased airflow might also increase subglottal pressure and consequently lead to more likelihood of voicing in stressed syllables, as discussed below. Aerodynamics and voicing: In unstressed syllables both voicing of voiceless segments (voicing pattern 1, as in Senoufo (Mills (1984)) and devoicing o f voiced segments (voicing pattern 2, as in Djabugai (Patz (1991)) are attested. However, in stressed syllables only voicing of voiceless consonants is attested (Voicing pattern 3, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 163 as in Wasco-Wisram (Sapir (1925)); devoicing of voiced consonants is not attested in stressed syllables. A number of properties of these patterns are explained through the aerodynamics of stress. First, the occurrence of voicing but not devoicing in stressed syllables can be explained through the fact that stressed syllables involve higher airflow (Lehiste (1970)). Higher airflow might induce higher subglottal pressure, which—provided the glottal cords are adducted (Ohala (1983))—would increase the likelihood of consonantal voicing. Second, the occurrence of both voicing and devoicing processes in unstressed syllables also has an aerodynamic motivation. Voicing tends to be unstable, since it is a particularly complex aerodynamic phenomenon in which a precise supraglottal and subglottal configuration has to be achieved. This configuration is particularly difficult to achieve in unstressed syllables, since there is less time available to do so. This would explain the attested processes of voicing and devoicing in unstressed syllables, even within the same language, as in Silacayoapan Mixteco. This does not exclude that perceptual forces might be allied with aerodynamics, causing voicing to occur more readily in unstressed syllables. Third, some of these processes are optional or occur in free variation, as in Copala Trique, or partial, as in Farsi. Both of these facts are consistent with the phonetic nature of aerodynamics. This will be considered in more detail in section 6 in this chapter. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 164 4.3 The role of duration Stressed syllables are longer than unstressed syllables, and stress can increase the duration of vowels and consonants (Lehiste (1970), Hayes (1995)). In some languages longer duration in stressed syllables (or shorter duration in unstressed syllables) might affect consonants, with or without affecting vowels. Consonantal processes motivated by duration include segmental or featural lengthening, flapping, lenition, deletion, and plausibly fortition and epenthesis. Lengthening-. Since stressed syllables are longer than unstressed syllables, the articulation of consonants can be prolonged. The way in which consonant duration is manifested depends on the specific consonant. It might include longer occlusion (for stops), longer frication (for fricatives), or more taps (for rhotics). Consonantal lengthening and increased aspiration also occur in stressed syllables because of the longer duration of stressed syllables. Ensuring optimal consonantal perception could also influence lengthening and increased aspiration (section 4.1). Duration andflapping: Flapping consists of the reduction of a gesture, whereby a complete occlusion is aimed for but not achieved. Instead, a fast and/or partial occlusion occurs. The cause for this gestural reduction appears to be that unstressed syllables have reduced duration; in this context, a total occlusion is hard to achieve because there is less time to complete it (Kirchner (1998)). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 165 Duration, lenition and fortition: Lenition (including approximantization of stops and fricatives, fricativization of oral stops, debuccalization, and flapping) and fortition (including occlusivization of fricatives and approximants, fricativization of approximants and others) have been proposed to have a durational basis (Lavoie (2001)). Reduced duration in unstressed syllables provokes gestural ‘undershoot’, and segments are not realized with complete closure, supraglottal information, appropriate voicing, and so on. In stressed syllables, it has been suggested that there is less gestural overlap in stressed syllables due to hyperarticulation in this position (De Jong, Beckman and Edwards (1993)). Duration, deletion and epenthesis: Since unstressed syllables are shorter than stressed syllables, consonants might delete in this position (but see also 4.1 and 4.4). Epenthesis might take place because of durational factors too, but it is more probably due to prominence considerations, as discussed below. 4.4 The role of prominence Prominence makes a syllable ‘stronger’ and non-prominence makes a syllable ‘weaker’. Rhythmicity, by which I refer to foot-based prominence, achieves a contrast between strong and weak syllables by way of strengthening the strong syllable, reducing the weak syllable, or both. This might be done in connection with durational, aerodynamic and perceptional factors. Prominence is a factor in aspiration, attraction, gemination, pre-aspiration of non-aspirates, pre-glottalization Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 166 of non-glottalized consonants, fortition, lenition and flapping, epenthesis, and deletion. Prominence and aspiration: Aspiration or stronger aspiration occurs in stressed syllables. Aspiration makes a consonant less sonorous. Onsets and specifically low- sonority onsets enhance the perceptual prominence of the syllable (Smith (2000)). This is supported by the auditory effect of adaptation; with continued exposure to a stimulus, auditory sentitivity gradually drops, and adaptation occurs (Delgutte (1997)). Intermingling syllable onsets between vowels provides the auditory system with the opportunity to recover from adaptation to vowels, thus enhancing prominence. Since low-sonority onsets are maximally distinct from vowels, recovery from adaptation is optimal, and thus prominence is maximally enhanced (Smith (2000); see also discussion in Gordon (2002)). The absence of aspiration or the occurrence of weaker aspiration in unstressed syllables decreases the prominence of these positions. Prominence and attraction: While featural attraction might be strictly perceptually- motivated (a feature is realized in the optimal perceptual position), segmental attraction could in principle be motivated by prominence. One relevant example is attraction of a pharyngeal resonant to stressed suffixes in Colville. It could be argued in this case that attraction increases the prominence of stressed syllables and decreases the prominence of unstressed syllables at the same time. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 167 Prominence and gemination: Gemination generally involves the lengthening of an onset immediately following a stressed syllable so that it is heterosyllabic. Gemination is typically blocked if the stressed syllable has a long vowel or a coda (as in Italian or Guelavia Zapotec). As such, gemination is a way to create a prominence contrast between stressed and unstressed syllables—provided that the coda adds to the overall duration and/or relative strength of the syllable. However, it is plausible that in some languages the coda might make the preceding vowel shorter and the overall duration of the syllable is not altered. In Italian, an inverse type of gemination is attested whereby a coda in a stressed syllable lengthens and becomes heterosyllabic before an onsetless vowel. Codas in unstressed syllables immediately preceding onsetless syllables resyllabify as onsets and gemination does not apply (Loporcaro (1999), Saltarelli (in press)). Prominence also plays a role in this type of gemination, since stressed syllables maintain their codas, unlike unstressed syllables. Prominence, pre-glottalization and pre-aspiration: As in gemination, pre-aspiration and pre-glottalization split an onset immediately following a stressed syllable into a coda and an onset. This contributes to the prominence of the stressed syllable, especially in segmental cases like Popoloca. In this language, fricatives, liquids, approximants and voiced nasals geminate after a stressed syllable, and complex segments (pre-nasalized, aspirated, and voiceless nasals) are decomposed into a coda and an onset. Additionally, stops and affricates are pre-glottalized in this position. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 168 Since all other consonants undergo gemination or ‘desintegration’ into a coda and onset, pre-glottalization also seems to be segmental and have the same motivation as gemination and ‘disintegration’ (Veerman-Leichsenring (1984)): the creation of a prominence contrast between stressed and unstressed syllables. The factors behind featural pre-glottalization and pre-aspiration are not clear. One example of apparent featural pre-glottalization are certain English dialects where obstruents are pre-glottalized immediately after stressed syllables, even if a sonorant intervenes between the obstruent and the stressed syllable, as in grumjyy. The fact that there is already a coda in the preceding syllable suggests that pre- glottalization is featural. Even so, in this and similar cases glottalization might lengthen the stressed syllable or increase its prominence. In contrast, featural pre-aspiration might overlap with the preceding vowel, potentially causing devoicing and plausibly shortening. This might decrease the duration of the stressed syllable or its prominence, since devoiced vowels are not very sonorous and thus not prominent. In Faroese pre-aspiration is blocked in the context of high vowels. High vowels are cross-linguistically less sonorous than non-high ones (Prince and Smolensky (1993)). It is possible that the blocking of pre-aspiration in this case is related to the potential devoicing of the stressed vowel, which would achieve a non- sonorous nucleus in a stressed syllable. It would be worse to have high devoiced vowels than non-high devoiced vowels, especially in stressed syllables, since the former are not veiy sonorous. It is plausible that pre-aspiration might contribute to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 169 prominence in some languages, but that in some others it might interfere with prominence by devoicing the preceding vowel. Prominence andfortition: Less-sonorous consonants are more ‘consonantal’ than more sonorous consonants. Less sonorant consonants in onsets increase the prominence o f a syllable; the less sonorant the onset consonant is, the more distinct it is from an adjacent vowel, and the more opportunities there are for recovery from adaptation to the vocalic stimulus, thus causing prominence (Smith (2000), Gordon (2002)). Since fortition involves a decrease in consonant sonority, in onset position fortition will help to enhance the prominence of the stressed syllable. One example would be West Tarangan (Nivens (1992)). Note that in some cases, as in Norton Sound Yupik and in Alutiik Yupik, fortition serves to distinguish foot boundaries (Leer (1985)). For afffication the situation is slightly different. The relative strength of affricates is not clear (Lavoie (2001)). In same languages with stress-sensitive affrication, affrication seems to be a stronger form of aspiration. English and Maori are relevant examples. It is clear that debuccalization consists of the reduction of supra-glottal features, which fail to be pronounced in weak position. Stronger ‘articulation’ or ‘emphatic and vigorous plosion’ of segments, as in Farsi (Samareh (1977)), would also fall in this group. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 170 Prominence, lenition andflapping-. Lenition and flapping are also candidates for a prominence analysis. Lenition (and flapping) involves an increase in consonantal sonority, which makes a consonant more vocalic. If less sonorant consonants in onset positions increase prominence, as explained in the previous paragraph, then more sonorant consonants in onsets would decrease syllabic prominence. Durational factors also seem to be behind these two phenomena (section 4.3). Prominence and epenthesis: Epenthesis can contribute to the creation of prominence contrasts between strong and weak syllables. One example is Huariapano, where [h] is epenthesized in coda of strong footed syllables. Prominence and deletion: Deletion potentially contributes to the creation of prominence contrasts between strong and weak syllables. One example is Capanahua, where 11 1 deletes in coda of weak footed syllables. 4.5 Remaining consonantal alternations The motivation for some types of consonantal alternations remains unclear. These include voicing pattern 4 (voicing alternations conditioned by the stress of the preceding syllable), secondary articulation and voicing dissimilation. Voicing pattern 4: It has been suggested that aerodynamics and—to a lesser extent—perception are relevant for voicing patterns 1-3. Other voicing alternations Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 171 are reported where the stress of the preceding syllable is relevant. These include Paamese-— where an immediately preceding stressed syllable is conducive to voicing of word-medial h J (Crowley (1982))—and diachronic alternations in Proto- Germanic and Middle English, where ‘non main stressed’ and ‘weakly stressed’ syllables respectively conditioned voicing of voiceless consonants (Wright (1957), Kabell and Lauridsen (1984)). It is unclear whether these environments just include unstressed syllablesor both unstressed and secondary-stressed syllables, and how the stress or lack thereof in the preceding syllable influences consonantal voicing. Secondary articulation: Stress-sensitive secondary articulation as in Guayabero has an unclear motivation. Secondary articulations might add to the overall duration of stressed syllables, or create a prominence contrast between stressed and unstressed syllables. Alternatively, if stressed vowels cause consonant place assimilation, this would be a case where a vowel in a prominent position triggers a phonological process rather than a consonantal process per se (Beckman (1998)). Voicing dissimilation: Finally, the position of stress and the motivation for dissimilation in Gothic is unclear (Collinge (1985)). 4.6 Summary Section 3 has considered the main factors underlying stress-sensitive consonantal processes. It has been argued that perception, aerodynamics, duration and Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 172 prominence are responsible for the majority of stress-sensitive consonantal phenomena from the survey in chapter 2. It has been shown that for some processes various factors might be relevant, and problematic cases where the factors behind them are far from clear have been identified. To close this section, I include a table with the stress- and foot-sensitive consonantal phenomena discussed for each factor. An asterisk * appears under the relevant column for the motivation. Parentheses () are placed around the asterisk when more than one motivation is plausible. Table 3.2. Factors in stress- and foot-sensitive consonantal phenomena A l t e r n a t io n P e r c e p t io n A e r o d y n a m ic s D u r a t io n P r o m in e n c e Attraction (*) (*) Stronger/longer aspiration (*) (*) (*) Variable aspiration * Variable glottalization * Metathesis * Voicing in a (Pattern 1) (*) (*) Voicing in 'a (Pattern 3) * Devoicing in a (Pattern 2) * Post-aspiration * Frication * Affrication Consonant lengthening * Flapping (*) (*) Gemination * Pre-aspiration * Pre-glottalization * Lenition n (*) Fortition c*) (*) Deletion (*) (*) (*) Epenthesis (*> (*) Stress attraction * Secondary articulation Unclear Voicing dissimilation Voicing after 'a/a (Pattern 4) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 173 5. Predictions: consonantal alternations that will be unattested The stress- and foot-sensitive consonantal phenomena attested and the effect of aerodynamics, perception, duration and prominence predict a series o f phenomena which should not occur in stress- or foot-conditioned environments (4). (4) Predictions: unattested stress- and foot-sensitive consonantal alternations Durational - Longer duration of consonants and VOT in unstressed syllables. Consonant reduction in stressed syllables. - Flapping in word-medial stressed syllables. Prominence-related - Onset fortition (sonority decrease) in unstressed or in foot-medial syllables in trochees. - Onset lenition (sonority increase) in stressed or in foot-initial syllables in trochees. - Flapping in word-medial stressed syllables. - Attraction of prominence enhancing features to unstressed syllables. Stress attraction to syllables without onsets or with high-sonority onsets. - Feature or consonant deletion in stressed syllables but not in unstressed syllables. - Gemination, pre-aspiration and pre-glottalization after unstressed syllables. - Prominence-enhancement in unstressed or weak syllables. - Prominence-reduction in stressed syllables. - Alternations that work together to enhance unstressed syllables and reduce stressed syllables. Aerodynamic - Longer or stronger frication, aspiration or affrication in unstressed syllables. - Aerodynamic devoicing of voiced consonants in stressed syllables. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 174 Perceptual Perceptual devoicing of voiced consonants in unstressed syllables. Feature or consonant deletion in stressed syllables but not in unstressed syllables. Salient features in unstressed syllables together with non-salient features in stressed syllables. In laryngealized consonants, glottalization and aspiration realized as close to the unstressed syllable as possible. Perceptual ambiguity in stressed syllables but not in unstressed syllables. A case that goes against the predictions in (4) is Lilloet. Globalized resonants in Lilloet are reported to be post-glottalized in all positions except before stressed vowels, where they are pre-glottalized (Howe and Pulleyblank (2001) and references therein). Glottalization is realized closer to the unstressed vowel in this case, and goes against the general tendency for post-glottalized resonants in this language. To further understand this case it would be necessary to know if sonorants in Lilloet are restricted to post-vocalic or pre-vocalic positions, or whether syllabic structure determines their distribution. 6. Phonetics and phonology An assumption in this dissertation is that phonetics and phonology are separate modules; phonetics will include processes which are optional, gradient or partial, while phonology will include processes which are category-neutralizing, categorical or obligatory (see section 5.3 in chapter 1 for more details). Discussion on the factors that underlie stress-sensitive consonantal phenomena begs the question of which Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 175 factors are purely phonetic and which purely phonological. I propose that consonantal phenomena with aerodynamic and durational bases are in principle phonetic. This does not exclude that in some cases, aerodynamically- and durationally-based phenomena can be phonologized. Consonantal alternations where perceptual ambiguity is at stake are phonetic, while other perceptually-based consonantal alternations are phonological. Prominence-based consonantal phenomena are always phonological (Figure 3.1). Figure 3.1. Phonetics and phonology in consonantal phenomena P h o n e t ic s Aerodynamics Perceptual ambiguity Duration Prominence Perception P h o n o l o g y Aerodynamic and durational factors follow from the articulatory and aerodynamic properties of stress: stress involves longer duration and higher airflow (Lehiste (1970)). Aerodynamic and durational consonantal phenomena are frequently variable, as in frication of coda /b, d, g/ in Basque Spanish (Gonzalez (2002b)), and optional, as voicing variation in Silacayoapan Mixteco (North and Shields (1977)). Nevertheless, durational and aerodynamically-based consonantal phenomena might Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 176 be phonologized; one example is flapping and aspiration in voiceless obstruents in English (Davis (2002), Davis and Cho (2003)). Consonantal phenomena where perceptual ambiguity is at stake belong to the phonetics. Phonetic devoicing or glottalization of a vowel by an adjacent laryngealized consonant might be reinterpreted by the speaker in a linguistic community as a different position of the consonant in the phonetic string (see Blevins and Garrett (1998)). By contrast, perceptually-based phenomena whereby salient features or segments are avoided in unstressed syllables and preferred in stressed syllables is phonological. Last but not least, prominence-based phenomena belong to the phonology. Prominence-based phenomena are not caused by stress, unlike aerodynamically- and durationally-based phenomena. Prominence primarily involves phonological weight. Rhythmicity, a type of prominence found in foot-sensitive contexts, creates a contrast between strong and weak syllables, with or without stress. 7. The separation between stress and footing The survey evidences three different types of stress-/foot-sensitive consonantal phenomena: (i) strictly stress-sensitive cases; (ii) phenomena where stress and footing coincide; and (iii) strictly foot-sensitive cases. The discussion below makes explicit in which contexts each occurs and the types of phenomena (aerodynamic, durational, perceptual or prominence-related) involved in each. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 177 7.1 Strictly stress-sensitive consonantal phenomena Some consonantal phenomena are influenced only by stress. These include alternations based on aerodynamic and durational factors, especially if optional or variable—like optional voicing alternations in Silacayoapan Mixteco, or variable ffication of coda /b, d, g/ in Basque Spanish. The reason is that aerodynamics and duration are consequences of stress, not of abstract foot structure (see previous section). Perceptual ambiguity seems to be related to stress only; metathesis and plausibly other consonantal alternations caused by perceptual ambiguity are more likely to occur in unstressed syllables due to reduced duration in this position (Blevins and Garrett (1998)). Stress-sensitive consonantal alternations are expected to occur in two specific stress systems: in languages with only one stress per word and no evidence for persistent footing, and in languages with persistent stress and consonantal alternations in a subset of stressed syllables. A plausible candidate for the first case is Senoufo, which has a number of reported stress-sensitive consonantal alternations including lengthening in stressed syllables, lenition in unstressed syllables and second articulation in stressed syllables (Mills (1984)). Urubu-Kaapor is plausibly a candidate for the second case, if, as reported in Kakumasu (1986), optional onset devoicing occurs in non-primary stressed syllables (i.e. secondary-stressed syllables) (see discussion in chapter 2). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 178 7.2 Stress/foot structure cases Most commonly foot structure and stress go hand in hand. Most languages have trochaic rhythms, with stresses falling on the initial syllable of every foot. In such cases it is ambiguous whether stress or foot structure is deterministic since both coincide. This is exemplified by some dialects of English, where III is flapped or debuccalized in unstressed syllables but aspirated in stressed syllables, and where /h/ is not pronounced in unstressed syllables. Since stress and foot structure coincide, it has been proposed that aspiration occurs foot-initially and flapping and debuccalization foot-medially (Hayes (1985), Davis (2002)). Since foot structure has not been studied in all of the languages gathered in the survey some languages are inconclusive as to whether a consonantal process is strictly stress-motivated, or whether stress and footing coincide in motivating it. 7.3 Strictly foot-sensitive consonantal phenomena Finally, there are cases where foot structure but not stress is deterministic in a consonantal process. This situation arises at least in three occasions: (i) iambic systems; (ii) languages with persistent footing that do not stress all feet in the prosodic word; and (iii) languages where, even if the rhythm is trochaic, directionality or alignment between stress and foot structure conflict. (i) Iambic systems: Iambic systems are right-headed, which might have potential implications for the relationship between stress and foot structure in consonantal Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 179 processes involving onsets. Onset phenomena in stressed syllables would occur foot- medially in disyllabic feet, but foot-initially in monosyllabic feet. Onset alternations in unstressed syllables would occur foot-initially in disyllabic feet; they would also occur in unfooted syllables (Figure 3.2). Figure 3.2: Iambic footing ant stress for onsets Stressed Unstressed Monosyllabic Foot Disyllabic foot Disyllabic Foot Unfooted Foot-initial Foot-medial Foot-initial ('a) ( a . 'a) ( a . 'a) a Not many cases of iambic systems are attested (Van de Vijver (1998)). In the survey, the only iambic systems examined are Norton Sound Yupik, Alutiik Yupik and Cayuga. Cayuga exemplifies an instance of a synchronic alternation caused by perceptual metathesis; it is phonetic. Norton Sound Yupik and Alutiik Yupik have foot-sensitive fortition which applies foot-initially. Norton Sound Yupik has left-to- right iambs and secondary stresses except on the final syllable of the word (5c). Initial closed syllables are footed on their own (5c). Foot-final syllables are lengthened if open (5a, b). (5) Norton Sound Yupik stress and foot structure (from Van de Vijver (1998)) (a) qajani [(qa.'jaa) ni] ‘his own kayak’ (b) qajapigkani [(qa.'jaa.) (pix.,kaa.) ni] ‘his own future authentic kayak’ (c) angjarpangjugtuq [('aq.) (jax^pap.) (jux.tuq)] ‘he wants to get a big boat’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 180 Norton Sound Yupik has fortition of /w, j, 1/ to fricatives after stressed closed syllables (first column in (6)). Elsewhere /w, j, 1/ are realized as approximants (second column in (6)).2 (6) Norton Sound Yupik /w, 1, j/ (from Van de Vijver (1998)) Fricatives Approximants Gloss (a) (ma.'jur) (vik) ('at.) (xar. wik) ‘place to go up/ down’ (b) (ma.'jur) (zux.tuq) ('at.) (xar. jux.) tuq ‘he wants to go up/down’ (c) (ma.'jur) (feu.ni) ('at.) (xar.,Iu.) ni ‘(he) going up/down’ Fortition of /w, j, 1/ is unrelated to stress. In (6a) both [w] and [v] occur in unstressed syllables. Approximants might be found in both stressed and unstressed syllables (second column of (6a-c)). It is not likely either that the stress of the preceding syllable conditions fortition, especially since the immediately preceding stressed syllable has a coda and no ambisyllabicity can apply. Additionally, fortition cross-linguistically affects stressed syllables, not unstressed syllables. Leer (1985) and subsequent literature proposes that fortition generally marks foot boundaries in Yupik. Fricative allophones of [w, j, 1 ] occur foot-initially and approximants elsewhere, namely, foot-medially and in unfooted syllables.3 2 In (5c), the second and the fourth syllables of [('ag.) (jax^pag.) (jux.tuq)] ‘he wants to get a big boat’ are transcribed with [j] in the source. This [j] should be fortis [z] because it is following a stressed closed syllable. Plausibly, since these examples illustrate the stress pattern of Norton Sound Yupik rather than the allophonic variants of approximants and fricatives, these are not m arked. 3 Unfortunately, I could not find examples of approximants occurring in unfooted positions. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 181 (ii) Feet without stress. Consonantal processes can be conditioned by foot structure even if stress is not realized in all feet of the prosodic word. One example is Capanahua, a left-to-right, quantity-sensitive trochaic system where only one stress per word has been reported (Loos (1969)) (7a). In Capanahua, coda /?/ deletes in the weak syllable of a trochaic foot, regardless of whether stress is realized or not (7b-f). (7d, e) show that I'll is deleted from the morphemes /ta?/, /ra?/ in the weak syllable of a foot. No deletion occurs in unfooted syllables (7f). In chapter 5 I argue that I'll deletion is a way to create a prominence contrast between strong and weak syllables in the foot; this is supported by the lack of deletion in unfooted syllables, and that I'll deletion achieves persistent footing in the language and compensates for the lack of secondary stresses. (iii) Conflicting metrical tendencies: In some consonantal processes footing and stress are antagonistic, with consonantal alternations respecting footing but not stress. Two relevant cases from the survey are Huariapano and Nganasan. (7) Capanahua (from Loos (1969); IPA) (a) [('?o.ga.)(ka.?i.)(ni?.ki)] (b) /ta?/ (c) /ra?/ (d) [('?o .tji) (ti .ra) (ta? .ki)] (e) [('?o .tji) (ti ,ma) (ra?. ta) ki] ‘he falls asleep’ declarative modal ‘probably’ ‘it is probably a dog’ ‘it is probably not a dog’ ‘bird’ (f) /?i?sap/ [?i?.('sa)] Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 182 • In Huariapano, main and secondary stress have opposite directionalities and quantity-sensitivity (Parker (1998)). Main stress is assigned via a moraic trochee at the right edge of the word (8a, b). Default secondary stress is assigned via syllabic trochees starting from the left edge of the word (8c-e). (8) Huariapano main and secondary stress (from Parker (1998:2-5); IP A) (a) ['(hi.wi)] ‘branch, stick’ (b) [ka.('no.ti)] ‘bow (weapon)’ (c) [(.ku.pjaj.) J3a.('§i.ki)] ‘I cooked’ (d) [('wa.nra.) ki.('rag.) ki] ‘they have returned’ (e) [(jo.mui.) (,raj.j3a.) kan.('gi.ki)] ‘they hunted’ Omitting some details, coda [h] is epenthesized in the strong syllable o f syllabic trochees constructed left-to-right provided the following onset is voiceless (9a). This distribution generally coincides with default (left-to-right) secondary stress assignment. However, coda [h] epenthesis also occurs in the main stressed foot (9a, c). In other cases, quantity-sensitivity for main stress assignment might lead to [h] epenthesis in unfooted syllables (9b, left column). Last but not least, in words with secondary stress assigned right-to-left from the main stressed foot, coda [h] epenthesis occurs in the weak syllables of trochaic feet (9d, left column), [h] epenthesis always coincides with the strong syllable in a syllabic trochee built from the left edge (9, right column), even if it clashes with main stress and (non-default) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 183 secondary stress assignment and the apparent footing for them. A detailed analysis of Huariapano is given in chapter 5. (9) Huariapano: footing for main and secondary stress (from Parker (1994)) Main/secondarv stress footing fhl epenthesis footing Gloss (a) [(jo.mui.) (,rah.ka.) ('tih.kay)] [(jo.mui.) (,rah.ka.) ('tih.kaej)] ‘(they) hunted’ (b) [(ja.na.) pah.('kwiij)] [(ja.na.) (pah.'kwirj)] ‘(I) will help’ (c) [(,J3o.no.) ('gih.kaej)] [(^o.no.) (‘gih.kaej)] ‘(they) will take’ (d) [§is.(,ma.noh.) (,ko.no.) ('gi.ki)] [(J3is.,ma.) (noh.,ko.) (no.'gi.) ki)] ‘I forgot’ • Nganasan has a process of ‘rhythmic gradation’, by which intervocalic [h, t, k, s, c] lenite to [b, d/5, g, z, j] in certain contexts (Vaysman (2002)).4 Nganasan builds left-to-right syllabic trochees (Helimsky (1998), de Lacy (2002)). Main stress tends to fall on the penultimate syllable.5 Secondary stresses are reported to occur in words with five or more syllables (Helimsky (1998)).6 (10) provides examples of stress and footing in this language; the foot structure in (10c) will be questioned later. (10) Nganasan stress and footing (from Helimsky (1998:486-7); IP A) (a) (,k3.riu) (,ga.Xi) ('ti.ni) ‘in marches’ (b) (,kin.to) (job.tui) ('ku.cig) ‘you are smoking’ (c) ^ka.rtu) (,ga.Ai) (ti.'ni.no) ‘in my marches’ (d) bo.('lou.ko) ‘a kind of movable dwelling on runners’ 4 These same segments undergo a different type of gradation related to syllabic weight (chapter 2). 5 Stress is optionally assigned to the antepenultimate syllable if this has a mid or low vowel [a, e, o] and the penultimate has a high or central vowel [i, y, u, o, i] (Helimsky (1998), de Lacy (2002)). 6 Helimsky (1998) does not mention secondary stress in words of less than five syllables long. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 184 According to Helimsky (1998), rhythmic gradation is conditioned by foot structure, with [fa, t, k, s, c] occurring foot medially— as in [('ro.kux)] ‘similar’— and [b, d /6 , g, z, j] foot-initially—as in ^ks.rra.) (,g3.Ai.) ('ra.kui) ‘similar to a march’. In words with an even number of syllables the distribution of [h, t, k, s, c] and [b, d /6 , g, z, j] is straightforward: voiceless allophones occur foot-medially (11a). However, in words with an odd number of syllables, voiced allophones are also found foot-medially (lib , c). (11) Rhythmic gradation: All syllabic trochees (from Helimsky (1998:490)) (a) ('nui.tra) ‘his wife’ (hui.a).('j3.tui) ‘his thumb’ (b) bui.('nm.6i) ‘his rope’ (ko.rui.) go.('Xi.6i) ‘his march’ (c) nxu.('r9. giui) ‘similar to a woman’ ^bra.niu.) (Va.kui) ‘similar to a rope’ I propose an alternative analysis of the stress system of Nganasan based on a prohibition against stressed final syllables. In odd-numbered syllables the last syllable is unparsed to avoid stress. This means that the fin al foot is iambic and that there are no medial unparsed syllables or feet with more than two syllables. Under this analysis, main penultimate stress is consistent with left-to-right footing, and the distribution of [h, t, k, s, c] and [b, d/6, g, z, j] is captured. Voiceless allophones correlate with foot-medial position (12a, c), and voiced allophones with either foot- initial or unfooted syllables (12b, c). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 185 (12) Rhythmic gradation: Final iamb (Helimsky (1998: 490) (a) ('nui.tra) ‘his wife’ (hui.a).(‘ j3.tm) ‘his thumb’ (b) (bui. 'nui).5i ‘his rope’ (ko.rui.) (go.’ XL) 5i ‘his march’ (c) (ntu.'ro.) gin ‘similar to a woman’ ^btu.nra.) ('ro.kru) ‘similar to a rope’ A final iambic foot in an otherwise trochaic system is motivated by the avoidance of both final stress and unfooted medial syllables (see Kager (1999), Van de Vijver (1998)). It allows the distribution of ‘rhythmic gradation’ to be equal in all feet in the word. An analysis where all feet are syllabic trochees does not capture these facts. Rhythmic gradation in Nganasan relates to foot structure rather than stress; voiced allophones would occur both in stressed and unstressed syllables. Regardless of the type of feet in the prosodic word, voiceless allophones are kept in foot-medial position and voiced allophones in foot-initial and unfooted positions. If rhythmic gradation in Nganasan is an instance of lenition, then this language contradicts one of the predictions in (4). The allophones [b, d/3, g, z, j] are more sonorous than [h, t, k, s, c]. While the former are found foot-initially and in unfooted positions, the later are found foot-medially. It would be expected that foot-initially the less sonorous allophones would be found, since less sonorous consonants make better onsets and provide the auditory system with recovery from the adaptation to the vocalic stimulus, hence enhancing prominence (Delgutte (1997), Smith (2000)). However, the opposite is found. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 186 I speculate that gradation in Nganasan is voicing-based. Recall from chapter 2 that four different voicing patterns are attested; in voicing pattern 1, underlying voiceless consonants became voiced in unstressed syllables; in voicing pattern 2, underlying voiced consonants become voiceless in unstressed syllables; in voicing pattern 3, underlying voiceless consonants become voiced in stressed syllables; in voicing pattern 4, the stress of the preceding syllable influences onset voicing. Depending of whether [b, d/5, g, z, j] or [h, t, k, s, c] are considered to be underlying, Nganasan would exemplify voicing pattern 2 or 3. The fact that the voiced allophones are found in unfooted syllables suggests that these are underlying, and that consequently, Nganasan exemplifies voicing pattern 2: underlying voiced consonants become voiceless in weak syllables. If consonant gradation in Nganasan is voicing-based, this pattern does not contradict the predictions made in (4). Further research on the nature of gradation in this language would help to dilucidate this case. 7.4 Summary Summarizing this section, neither stress nor foot structure alone are sufficient in order to characterize the range of attested consonantal alternations. Both can coincide in conditioning consonantal phenomena, but they do not have to, as evidenced by strictly stress-sensitive and strictly foot-sensitive phenomena. More research on the foot structure of most languages is needed in order to determine whether the coincidence of stress-footing is a universal tendency, or whether the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 187 three possibilities (strictly stress, strictly footing, both) are equally salient in the languages of the world. 8. Rhythmicity Foot-sensitive phenomena seem to be motivated by prominence considerations, in the sense that a prominence contrast is created between strong and weak footed syllables in the absence of or in opposition to stress. This prominence contrast might increase the prominence of the foot head (as in Huariapano) or decrease the prominence of the foot non-head (as in Capanahua). I would like to use the term rhythmicity to refer to the occurrence of a prominence contrast within a foot. Rhythmicity typically involves an alternation among stressed and unstressed syllables. It is often observed in stress placement (Hayes (1995)). But stress is not the only way that rhythmicity manifests itself; it can also be expressed through consonantal alternations. These can make strong footed syllables stronger or more prominent through processes like gemination, pre aspiration, pre-glottalization, fortition and epenthesis. They can make weak footed syllables weaker or less prominent through lenition, flapping and deletion. Rhythmicity can make strong footed syllables more prominent and weak syllables less prominent at the same time. Some attraction processes and metathesis (as in Capanahua) may achieve this effect. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 188 I suggest that rhythmicity achieves a balance between the drives of maximizing perceptibility and minimizing effort. Rhythmicity does not go to infinite; strong syllables are not infinitely strengthened and weak syllables are not infinitely reduced. If they did, reduced syllables would be totally deleted and strong syllables strengthened to a great extent. Consider the following two scenarios. In the first, all syllables in a word or sentence are equally strong. This would mean that perception is maximized, but so is effort. In the second, all syllables in a word are equally reduced. This would mean that effort is minimized, but so is perception. Thus, a language where there is a contrast between strong and weak syllables does the best for minimizing effort and maximizing perception at the same time. This does not exclude that certain other tendencies or pressures in the system might conflict with rhythmicity and obscure it, as discussed for stress systems in Hayes (1995). Rhythmic contrasts are observed in stress patterns, consonantal processes, vocalic processes and tonal phenomena (13) (see also Hayes (1995)). All of these processes can work together to achieve rhythmicity. However, some languages might ‘select’ some of these patterns over others as better suited to express prominence contrasts. For instance, while stress and foot structure typically correspond to each other, they do not in certain languages, either because of the absence of secondary stress (as in Capanahua), or other considerations, including diachronic change (this plausibly occurs in Huariapano; see chapter 4). Alternatively, some languages might have foot-sensitive vocalic phenomena but not consonantal phenomena, or the other way around. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (13) Examples of rhythmic contrasts 189 S tress Stress in strong syllables but no stress in weak syllables. Avoidance of clashes and lapses in stress patterns. C on son an ts Fortition in strong syllables and lenition in weak syllables. Epenthesis in strong syllables and deletion in weak syllables. Codas/onsets in stressed syllables and no codas/onsets in weak syllables. Vowels Long vowels in strong syllables and short vowels in weak syllables. Deletion of vowels in weak syllables. Sonorous vowels in strong syllables and non-sonorous vowels in weak syllables. Vocalic feature attraction from weak syllables into strong syllables. Tone Tone in stressed syllables. High tone in strong syllables and low tone in weak syllables.________________ Foot-sensitive consonantal alternations achieve rhythmicity and with it the organization of syllables into groups of strong and weak syllables. These processes are typically categorical and obligatory for all speakers. However, even in cases where they are optional, as in metathesis of /?/ in Capanahua, the fact that these processes achieve the organization of syllables into higher constituents makes them a part of the grammatical knowledge that speakers have. The following section puts forward a proposal to account for foot-sensitive consonantal prominence cases. 9. Accounting for prominence-based consonantal alternations This section proposes an analysis of prominence-based consonantal phenomena for stress-sensitive and foot-sensitive phenomena. Section 9.1 reviews previous Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 190 prominence-type analyses, including prominence augmentation (Smith (2000, 2002)) and prominence reduction through prominence alignment (Prince and Smolensky (1993), Crosswhite (to appear), among others). Sections 9.2 and 9.3 show that Prominence Alignment can capture prominence-based foot-sensitive consonantal phenomena. Section 9.4 is the conclusion. 9.1 Previous analyses 9.1.1 Augmentation in strong positions Prominent positions— onsets, long vowels, stressed syllables, initial syllables, and roots—are augmented in many languages, achieving enhanced prominence (Smith (2000, 2002)). For instance, stressed syllables might become heavy (as in Mohawk), acquire tone (as in Slave), have an epenthetic onset (as in Dutch) or reject high- sonority onsets (as in Niaufo’ou). On the other hand, prominent elements may attract stress; stress can be attracted to heavy syllables (as in Aguacatec), to high-toned syllables (as in Golin), to syllables with onsets (Arremte) or to syllables with low- sonority onsets (Piraha) (Smith (2000, 2002)). Both augmentation of stressed syllables and stress attraction phenomena are attested in the survey from the previous chapter. Phenomena where augmentation is at stake included gemination, pre-aspiration, pre-glottalization, epenthesis and segmental attraction—the last one possibly hand in hand with prominence reduction—and possibly longer or stronger aspiration. Various instances of stress attraction are also gathered in the survey; some examples are Arremte and Piraha. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 191 One possible approach that captures augmentation phenomena is given in Smith (2000, 2002). Smith (2000, 2002) proposes to capture augmentation phenomena through the ranking of M/Str(ong) (henceforth M/Str) constraints, that is, markedness constraints that make specific reference to strong positions. Only markedness constraints that enhance prominence can be relativized to a strong position. This is known as the ‘prominence condition’ (Smith (2000, 2002)). For example, heaviness enhances the prominence of a stressed syllable. This is expressed by the constraint H e a v y o /'g (14). If M/Str and stress location constraints outrank faithfulness, augmentation will occur. If M/Str and faithfulness constraints outrank stress location, stress attraction will occur (Smith (2000, 2002)). (14) Heavyg/'c For all syllables x, if x is a 'o, then x is heavy (bimoraic) ‘Stressed syllables are heavy’ (Smith (2000)) A problem with this approach is that strictly foot-sensitive consonantal processes cannot be captured, since foot heads (independent of stress) are not part of the prominent categories. Additionally, this approach targets augmentation only, not reduction. The following section discusses how Prominence Alignment (Prince and Smolensky (1993)) can capture both prominence reduction and augmentation in stress- and foot-related contexts. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 192 9. 1. 2 Prominence Alignment Prominence-alignment is a mechanism first proposed to account for the preferential alignment of sonorous segments with syllabic nuclei and for the preferential alignment of non-sonorous segments with syllabic margins (Prince and Smolensky (1993)). The basic idea behind Prominence Alignment is that two prominence scales can be crossed in order to encode the preference of prominence reduction in non- prominent positions and higher prominence in prominent positions. To give one example, Crosswhite (to appear) accounts for stress-sensitive prominence reduction in vowels by means of crossing the accentual prominence scale with the vocalic prominence scale (15) (prominence-reduction applies in cases where loud and long vowel qualities are reduced in unstressed positions, while prominent positions show augmentation or faithfulness to underlying qualities). (15) Prominence reduction and augmentation for vowels (Crosswhite (to appear)) Scale 1: Accentual prominence stressed p r o m >unstressed Scale 2: Vocalic prominence 3- prom >^'i 3 pronP’&i O pronr>^t U prom '> 3 Prominence reduction for vowels: Constraint Hierarchy *unstressed/a»*unstressed/e,o»*unstressed/e,o»*unstressed/i,u»*unstressed/3 Prominence faithfulness/augmentation for vowels: Constraint Hierarchy *stressed/o»*stressed/i, u»*stressed/e,o»*stressed/e,o»*stressed/a Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 193 The accentual prominence scale expresses the fact that stressed syllables are more prominent than unstressed syllables. The vocalic prominence scale encodes the fact that sonorous vowels are more prominent than non-sonorous vowels. Crossing the unstressed member of the accentual prominence scale with the members of the vocalic prominence scale achieves a hierarchy of constraints prohibiting sonorous (i.e. prominent) vowels in unstressed positions. The ranking among the resulting constraints is fixed and captures the implicational relationship among them. Namely, if a language permits, say, /a/ in unstressed syllables, it also allows for all other vowels in this position. If the stressed member of the accentual scale is combined with the vocalic sonority scale, the resulting hierarchy of constraints expresses the fact that non- sonorous (i.e. non-prominent) vowels are dispreferred in stressed syllables. Depending on the relative ranking with markedness and faithfulness constraints, this can account for faithfulness of underlying vowel sonority in this position and/or ‘augmentation’ in the sonority in stressed syllables. Thus, the mechanism of prominence alignment offers a way to capture both prominence reduction and augmentation. Prominence Alignment has also been employed to account for syllabic prominence (Prince and Smolensky (1993)), stress prominence (Kenstowicz (1996)), moraic prominence (Crosswhite (1999)), and tonal prominence (de Lacy (1999)). The following sections show that Prominence Alignment can also be extended to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 194 stress-sensitive and foot-sensitive consonantal phenomena involving prominence reduction or augmentation. 9.2 Prominence Alignment and stress-sensitive consonantal prominence Prominence-based consonantal phenomena where stress is relevant include augmentation cases (gemination, pre-aspiration, pre-glottalization, fortition, epenthesis, and possibly attraction and longer aspiration) and reduction cases (lenition, flapping, and deletion and possibly attraction). All of these can be analyzed with Prominence Alignment by means of combining the accentual and segmental scales in (16). (16) Accentual and segmental scales Accentual prominence ® prom ' (Crosswhite (to appear)) Coda prominence C]a prom - > V]0 Moraic prominence PPprom ^P (Crosswhite (1999)) Vocalic nrominence 3- prom '> ! '-\ 0 prom ^ (Prince and Smolensky (1993)) Consonantal sonoritv w >f> p (based on Venneman (1988)) Here I will discuss the combination of the accentual scale with the coda prominence and consonant sonority scale. Different instantiations have been proposed for the sonority hierarchy for consonants (see discussion in Crosswhite (to appear)). Here I am simplifying over the different proposals for this hierarchy, focusing instead on the fact that stops are less sonorous than fricatives, and fricatives less sonorous than glides. Onset consonants enhance syllable prominence because they provide the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 195 auditory system with recovery from adaptation to a constant vocalic stimulus (Delgutte (1997), Smith (2000)). Since less sonorous consonants are maximally different from vowels, they will also help to enhance syllable prominence. This will mean than in stressed syllables, less sonorous onsets will be preferred to more sonorous onsets in order to express prominence (17). (17) Stress prominence and consonant sonority in onsets: constraint hierarchies *la/0[w » * 'a /a[f » * 'a /0[p ‘Stressed syllables don’t have sonorous consonants in onsets’ *a/a[p » * o /0[f » * 6 / a[w ‘Unstressed syllables don’t have non-sonorous ___________________________ consonants in onsets’ __________________________ The constraints in (17) account for onset lenition in unstressed syllables and onset fortition in stressed syllables. For example, West Tarangan has fortition of /w, j/ in onsets of stressed syllables (Nivens (1992)). High-ranked *'a/0[w outranks the constraint enforcing sonority identity from input to output. As a result, /w, j/ fortify to [g, dz] in stressed syllables (Tableau 1). Tableau 1: Fortition in West Tarangan /s u w a k a n / ‘elep han t tu sk ’ * V a[w IDENT (SONORITY) ^ a. s u .'g a .k o n * ' b. su .'w a .k o n * 1 ■ - A partial typology of prominence-based consonantal phenomena in onsets where stress and sonority interact is given in (18). Fortition will take place in onsets if any Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 196 member of the hierarchy *'a/a[w » * ' o / 0[f » * 'a / c[p outranks Id en t (sonority). Lenition or flapping will take place in onsets if any member of the hierarchy *d/o[ p » * 0 /o[f » * a / 0[w outranks Ident (sonority). (18) Partial typology: Consonant sonority in onsets Augmentation in 'a *'a/a[w » Ident (son) Fortition: West Tarangan Reduction in a * d/a[p » Ident (son) Lenition: Senoufo Faithfulness in d Ident (son) » * d/a[p Non-sonorous C: ? Faithfulness in 'o Ident (son) » * V ^ p Sonorous: C ? The accentual scale can also be combined with the coda prominence scale. The resulting constraints are given in (19). (19) Coda prominence: constraint hierarchies *'a/V](j » * 'a /C ]a ‘Stressed syllables are not open’ *a/C]a » * a /V ]0 ‘Unstressed syllables are not closed’ The universal ranking *'a/V]a» * la/C]( j expresses the fact that there is a tendency for stressed syllables to achieve prominence by means of having a coda. If *'o/V is sufficiently high-ranked, it can drive consonant attraction to the stressed syllable and gemination, pre-aspiration and pre-glottalization immediately following the stressed syllable. The universal ranking * a/C]0 » * 6 /V ]0 captures the fact that there is a tendency for unstressed syllables to avoid extra prominence from a coda. If *d/ C]C T is Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 197 sufficiently high-ranked, it can drive coda deletion. A partial typology of coda prominence related to stress is given in (20). (20) Partial typology: coda prominence and stress Coda augmentation *'o/V]<j » I n t e g r it y Popoloca: desintegration Coda reduction *d/C]a » M ax Oneida: [?, h] deletion Faithfulness to underlying coda F a ith » * 'ct /V]0 Senoufo: CV in 'o FA ITH »*a/C]a Guayabero: CYC in a Other stress-related prominence consonantal phenomena that can be analyzed through Prominence Alignment include onset epenthesis in stressed syllables and segmental attraction to complex onset. The next section discusses the extension of Prominence alignment to foot-sensitive prominence-based consonantal phenomena. 9.3 Prominence Alignment and foot-sensitive consonantal prominence Previous research on the interaction between stress and vocalic sonority and stress and tone has shown that reference to heads in constraints does not capture the full range of cross-linguistic data for certain processes that apply in non-heads (Kenstowicz (1996), de Lacy (1999), Crosswhite (to appear)). In other words, reference to both heads and non-heads is necessary to account for the attested cross- linguistic patterns. This is also evident from the survey in chapter 2, where some prominence phenomena affect foot heads (as in Huariapano), but others affect foot non-heads (as in Capanahua). The mechanism of Prominence Alignment makes it Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 198 possible to refer both to foot heads and foot complements for prominence-based phenomena. Prominence alignment for foot positions is discussed below for coda prominence, consonant prominence and accentual prominence. The scales relevant for this discussion are given in (21). (21) Prominence and sonority scales Foot prominence Headp r o m >Weak (Kenstowicz (1996)) Accentual prominence ® prom (Crosswhite (to appear)) Coda prominence C]oprom " > C]0 Consonantal sonoritv w > f >p (based on Venneman (1988)) Note that Headprom >Weak is assumed here rather than Head p r o m >non head. Head Headprom >Weak means that strong footed syllables are more prominent than weak footed syllables. Head pr0m> non head means that strong footed syllables are more prominent that both weak footed and unfooted syllables. The choice of Headpro m >Weak is grounded in the aim of contrast within the prosodic constituent. However, the selection of one scale over the other, or the fact of whether a ternary scale that includes heads, weak syllables and unfooted syllables is needed is an empirical question and interacts with the accentual prominence scale. For further discussion, see chapter 6. The foot prominence scale can be combined with other prominence scales to capture the full range of prominence phenomena in both foot heads and foot Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 199 complements. The constraints that result from combining foot and coda prominence scales are given in (22). (22) Coda prominence: constraint hierarchies *Head/ V]0 » *Head/ C]0 ‘Foot heads are not open’ *WEAK/ C]g » * W eak/ V]g_______ ‘Foot complements are not closed’ The ranking *Head/V]0» *Head/C]0 expresses the fact that foot heads want to be prominent by means of having a coda. If *Head/V]0 is sufficiently high-ranked, it can drive coda epenthesis and onset-to-coda metathesis in foot heads, as attested in Huariapano and Capanahua. Prominence reduction occurs when *Weak/C]0 » * W eak/V]c is visibly active. High-ranked *Weak/C]<, can result in coda deletion in weak footed syllables, as in Capanahua (Tableau 2). A partial typology of foot prominence related to codas is given in (23). Tableau 2: Coda /?/ deletion in Capanahua: /bitji?/ ‘I grab’ *Weak/C]0 Max e r a. ('bi. tji) I ' ’ " *• b. ('bi. tji?) *! j - . • ' (23) Partial typology o f foot prominence/codas CVC head *Head/V]<j » Dep Huariapano CV head N o Coda» * H ead/V]0 Maori CV weak *Weak/C]0 » Max Capanahua CVC weak Max » * W eak/ C]c Norton Sound Yupik Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 200 Foot prominence can also be combined with consonantal sonority. I suggest that, similarly to the case of stressed syllables, less sonorous onsets enhance prominence in foot heads since onsets and especially low-sonority onsets achieve recovery from adaptation to the vocalic stimulus (Delgutte (1997), Smith (2000)). For this reason non-sonorous consonants are preferred in onsets of foot heads while sonorous consonants are preferred in onsets of weak footed syllables. The relevant constraints are in (24). From the survey, cases involving onset consonantal sonority relate to stress rather than footing or involve foot boundary marking, as in Norton Sound and Alutiik Yupik. However, a predicted partial typology of foot-related prominence phenomena involving consonantal sonority is given in (25) (for vocalic sonority and foot prominence, see Kenstowicz (1996) and chapter 5 on this dissertation on Shipibo). (24) Foot prominence and consonant sonority in onsets *HEAD/<j[w » * H E A D /0[ f » * H E A D /0[p *WEAK/a[p » *WEAK/a[ f » * WEAK/0[w ‘F o o t h ead s d o n ’t h a v e son orou s c o n so n a n ts’ ‘F o o t co m p lem en ts d o n ’t h ave n o n -so n o ro u s co n so n a n ts’ (25) P red icted partial ty p o lo g y o f fo o t-se n sitiv e con son an tal son ority in o n sets H ead s h a v e n on -son orou s on sets H ead s h a v e son orou s on sets W ea k p o sitio n s h a v e n on -son orou s on sets W eak p o sitio n s h a v e son orou s on sets * H ea d /a[w » lD E N T (C ) Id e n t ( C ) » * H e a d / a[w *Weak/ 0[p » Id e n t (C ) Id e n t (C )»*W eak/ 0[p Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 201 Foot prominence can also be realized through stress. There is a cross-linguistic preference for foot heads to have stress and for foot non-heads to be unstressed. The relevant constraints are given in (26). (26) Footing and stress prominence *H e a d / o ‘F oot h ead s are n ot u n stressed ’ * W e a k / 'o ‘Weak footed syllables are not stressed’ T o the b est o f m y k n o w le d g e , constraints su ch *H e a d / o and *W EAK/‘a h a v e n ot been proposed before. These constraints express the universal tendency for stress to occur in foot heads. It could be argued that the interaction between GRWD=PRWD, P a r s e and A l l - F e e t ( l e f t / r i g h t ) suffices to account for stress systems. If P a r s e » A l l - F e e t (L/R), only one foot will be constructed in a prosodic word; where A l l - F e e t ( L / R ) » P a r s e , feet will be persistent in a prosodic word. However, the interaction among G r w d = P r w d , P a r s e and A l l - F e e t (L/R) does not capture cases where footing is persistent but stress is not (as in Capanahua). Attested foot- sensitive consonantal alternations where stress is not relevant argues for the inclusion of constraints that derive footing independently of stress. Making *Head/g violable has interesting repercussions for stress systems. If *H e a d / g is und om in ated or su ffic ien tly h igh -ran k ed stress w ill su rface in a ll fo o t heads. If * H e a d /6 is low-ranked or not active not all feet will have stress. One example is Capanahua, where only the initial foot is stressed. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 202 *W eak / 'a expresses the fact that weak syllables tend not to be stressed. This is apparently always the case cross-linguistically, with the possible exception of Huariapano (see chapter 5). Plausibly this is because, unlike for *Head/o, few or no constraints at all dominating *W e a k / 'a have the effect of driving stress in weak footed syllables. Note that (26) does not include *H e a d /'o or *W e a k / 6 as possible rankable constraints. The reason for this is that these constraints might not be well- motivated in any grammar. A partial typology of footing and stress is given in (27). M a x s t r e s s enforces one and only one stress per prosodic word (Gonzalez (2002a); see also chapter 5). N o n -f in a l it y prohibits final stress (Kager (1999:117)), and * C l a s h prohibits two stresses being realized in adjacent syllables (Kager (1999:130) and references therein). If any of these constraints outranks *H e a d / o , not all foot heads in the prosodic word will be stressed. (27) Partial typology of footing and stress All heads stressed Undominated * H e a d / o Icelandic One head stressed M a x s t r e s s» * H e a d /S Capanahua Most heads stressed N o n -f in a l it Y » * H e a d /a Norton Sound Yupik Weak a unstressed Undominated C l a s h , *W e a k /'o Most languages Weak o stressed *H e a d /C V , Id e n t (S tress)»*W E A K /'a Huariapano 9.4 Summary It has been shown that Prominence Alignment can be extended not only for stress- sensitive consonantal phenomena where prominence is at stake, but also for foot- Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 203 sensitive consonantal processes. Additionally, Prominence Alignment can make reference to stressed and unstressed positions and to strong and weak footed positions for both augmentation and reduction phenomena. This mechanism is independently needed to account for other phenomena (Prince and Smolensky (1993), Kenstowicz (1996), Crosswhite (1999, to appear), de Lacy (1999)). Deploying this mechanism is just extending logical properties of the theory. Languages might enforce prominence through sonority, coda and stress alternations. Other possibilities, like onset prominence, have not been shown in detail to avoid redundancy, and also because other motivations (such as duration) might be involved. Combining foot prominence with stress and segmental prominence predicts three types of foot-sensitive, prominence-related consonantal phenomena (29). Prominence enhancement and reduction can take place in two different phenomena in the same language (as in Capanahua); additionally, both can be achieved by the same phenomenon (as in Bagneres-de-Luchon French attraction). (28) Figure 3.3. Predictions for foot-based prominence + Prominence (heads) Prominence (weak footed c) + Prominence (heads) & - Prominence (weak footed o) Same process Same language E pen th esis (Huariapano) D eletio n (Capanahua) A ttra ctio n (B-de-L French) M eta th esis & deletio n (Capanahua) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 204 10. Accounting for duration-based consonantal alternations Stress-sensitive consonantal processes where duration is a factor include consonantal lengthening and shortening, lenition, flapping, and plausibly fortition. For consonantal lengthening and shortening the relationship between stress and duration is straightforward: a consonant will be longer in a stressed syllable than in an unstressed syllable. Stress-sensitive lenition and flapping involve both durational and magnitude reduction in unstressed syllables (Browman and Goldstein (1990, 1992)). Stress-sensitive fortition has been shown to occur in stressed syllables through a decrease in articulatory overlap in this position due to hyperarticulation (De Jong, Beckman and Edwards (1993)). Previous formal analyses (Kirchner (1998), Flemming (2001)) allow for the incorporation of stress in durational phenomena. Lenition and flapping are captured by the effort-type approach (Kirchner (1998)), where perceptual motivations and the drive to maintain contrasts conflict with a general drive to minimize effort. The effort approach does not explicitly discuss stress-sensitive lenition, but it can be extended to capture it. Since unstressed syllables have reduced duration, the target duration and constriction of a segment cannot be as easily achieved as in stressed syllables. This will cause lenition, deletion, and general shortening. Kirchner (1998) suggests that processes where insertion of gestures occurs might be driven by perceptual considerations, i.e., by making clear the relevant contrasts. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 205 In Flemming’s (2001) approach, both categorical and scalar phenomena are formalized through the interaction of (mostly) scalar constraints, with overall constraint weighting rather than strict domination. Flemming (2001) acknowledges the need to incorporate stress-sensitive constraints in this model, including stress- sensitive durational constraints or rhythmic constraints of the Iambic-Trochaic law type (Hayes (1995)). However, he does not develop an analysis for stress-sensitive durational phenomena. Both durational and prominence can be factors for lenition, flapping and fortition phenomena. The previous section proposed an analysis for phonological, prominence-related phenomena that includes these. Phonological consonantal lengthening and shortening also need to be analyzed in the grammar. Prominence Alignment can be extended to account for stress-sensitive durational processes with are obligatory and categorical. The relevant scales and constraints that obtain are given below: (29) Prominence alignment for duration Scales Accentual prominence Consonantal length C « p r o n ? *C prom ^C Vocalic length v > v >v v * p r o m v p r o m v Constraints *d/C:»*d/C'»*6/C ‘Unstressed syllables don’t have long consonants’ *'o/C »*'c/C -»*'o/C : ‘Stressed syllables don’t have short consonants’ *,o/V »*'aA r ' » * ,a/V: ‘Stressed syllables don’t have short vowels’ *d/V: » * a/V ' »*o/V ‘Unstressed syllables don’t have long vowels’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 206 Here I will focus on consonantal phenomena. Cross-linguistically, shorter consonants tend to occur in unstressed syllables and longer consonants in stressed syllables. One example is Urubu-Kaapor, where onsets are lengthened (but not geminated) in main stressed syllables (Kakamasu (1986)) (Tableau 3). A partial typology of stress- related durational phenomena is given in (30). Tableau 3: Urubu-Kaapor onset lengthening /katu/ ‘it is a good’ ^ a. ka. t:u b. ka. tu *'o/C Ident (length) (30) Partial typology: stress-related durational phenomena Shortening in 6 *6/C :»*o/C ’» * 6 /C Lengthening in 'a *'a/C »*'a/C '»*'a/C : No length difference Ident-(length) » *'<j/C Ident-(length) » *a/C: American English flapping Urubu-Kaapor Swedish It is predicted that lengthening can also be foot-sensitive. In such cases, the foot prominence scale should be combined with the durational scale. 11. Accounting for aerodynamic-based consonantal alternations Few analyses have been proposed to capture aerodynamic processes (see Davis (2002) for English aspiration and flapping). The reason might be that aerodynamic processes are usually variable and gradient and thus not included in the grammar. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 207 However, in some cases aerodynamic phenomena are part of the grammar (as in English aspiration). Prominence Alignment can be extended to capture aerodynamic processes. Focusing on aspiration, the relevant scales and resulting constraints are given below. (31) Aerodynamic phenomena: scales and constraints Scales Accentual prominence G p ro m '>£ > Aspiration Aspiration p r o m >No aspiration Constraints *[asp]/d»*No [asp]/d ‘unstressed syllables have no aspiration’ *No [asp]/'a »*[asp]/'a ‘stressed syllables don’t lack aspiration’ This family of constraints predicts that aspiration will be greater or more likely in stressed syllables than in unstressed syllables. One example involving is English, where Ihl is not pronounced in unstressed syllables (tableau 4). For aspiration, a partial typology of possible cases is given in (32). Tableau 4: /h/ in English /vihikl/ ‘vehicle’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 208 (32) Partial typology (aspiration) Aspiration admissible everywhere IDENT(asp)»*[asp]/c, *No [aspj/'o Thai Deletion in 6 *a/[asp]» Max English Epenthesis in 'o *'or/n o [asp] »D ep English No aspiration anywhere *[asp], *[asp]/d » *No [asp]/'c Spanish 12. Accounting for perception-based consonantal alternations Perceptual processes can achieve a better perception of features or segments, avoid features or consonants in positions where they cannot be perceived; or cause perceptual ambiguity in perceptually-challenged positions. Perceptual approaches that take into consideration the role of stress include Licensing-by-cue (Steriade (1997)) and contrast reduction (Crosswhite (to appear)). Steriade (1997) examines laryngeal neutralization and concludes that neutralization arises due to absent or diminished perceptual cues in certain environments. Perceptual cues include quantity, quality, and duration. For laryngealized consonants, the timing of aspiration and glottalization depends on whether the consonant is a stop or a sonorant. If a stop, aspiration and glottalization will tend to fall on the following vowel; the consonant will be post-aspirated or post- glottalized. Laryngealized sonorants will tend be pre-aspirated or pre-glottalized. In intervocalic position, the timing of aspiration or glottalization will depend on stress. If the preceding vowel is stressed, the consonant will be pre-aspirated or pre- glottalized. If the following vowel is stressed, the consonant will be post-aspirated or Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 209 post-glottalized. The reason for this is that cues are better perceived in stressed vowels. Steriade (1997) suggests fixed perceptibility scales which depend on the environment where a consonant is realized. These perceptibility scales include pre- or post-vocalic environments, before or after a resonant, and so on. These perceptibility scales are formalized as segmental context constraints. These constraints are ranked with faithfulness constraints in a language-specific basis. If faithfulness (‘preserve’) constraints outrank the whole hierarchy of segmental context constraints, features are preserved in all contexts. If the perceptual hierarchy outranks the relevant faithfulness constraint a given feature or segment will not surface in the language. However, in many languages features occur in certain positions where cues are relatively perceivable. This is captured through the ranking of the specific perceptual constraint over the relevant faithfulness constraint. Crucially, Steriade claims that syllable positions do not play a role.7 This and similar approaches account for variable timing of glottalization and aspiration (Steriade (1997)), metathesis in stressed syllables (Hume (2001)), and perceptually-based consonant attraction. Licensing-by-Cue cannot obtain perceptually-motivated consonantal processes where perceptual ambiguity is at stake. Since in perceptual metathesis a change arises because of perceptual ambiguity, and since both CV>VC and VC>CV changes are attested for all features or segments undergoing metathesis, it is not possible to invoke the interaction 7 Cf. Howe and Pulleyblank (2001). They argue that in some cases prominent prosodic structure (onsets or moras) licenses features like glottalization regardless of perceptual considerations. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 210 between maxkedness and improved perception as a driving force (Blevins and Garrett (1998)). Moreover, perceptual metathesis is in many cases diachronic (Blevins and Garrett (1998)). Crosswhite (to appear) examines vowel reduction processes and distinguishes between prominence-reduction and contrast enhancement cases. In contrast enhancement, perceptually-challenging vowel qualities are limited to stressed positions, while these contrasts are eliminated in unstressed positions. Contrast enhancement processes are formalized through licensing constraints in which marked features are licensed in contexts enhancing correct perception. The range of processes that this approach captures is similar to Licensing-by-Cue. As an example, let us examine how speech perception-based accounts capture metathesis and deletion in Faroese. The analysis is taken from Hume and Seo (2001). Only the basic analysis distinguishing metathesis and deletion will be presented. For details, see Hume and Seo (2001). Recall that in Faroese, the cluster [skt] metathesizes to [kst] in stressed syllables, while in unstressed syllables [k] deletes. Hume and Seo (2001) consider that metathesis enhances the perceptibility of [k] in Faroese. The cues for [k] are not well perceived between two consonants, so [k] metathesizes to a position right after the stressed vowel to enhance its cues. In unstressed syllables the cues for [k] are not well perceived, so [k] deletes. The perceptual constraints proposed by Hume and Seo (2001) are given below: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 211 (33) Perceptual constraints in Faroese (from Hume and Seo (2001:9, 12)) Avoid X/Y (*x/y) Avoid a segment type X in context Y *stop/obst___stop Avoid a stop between an obstruent and a consonant. *stop/v c Avoid a pre-consonantal stop after an unstressed sonorant *stop/'v__c Avoid a pre-consonantal stop after a stressed sonorant Avoid X/Y are contextual constraints that capture the perceptual factors behind cross- linguistic patterns and that reflect the scale perceptibility of the segments involved. For example, *stop/v c»*stop/'v c captures the fact that it is worse for a stop to be pronounced in the vicinity of an unstressed sonorant than after a stressed sonorant, because the cues for the stop will be better enhanced after stress (since stress has increased duration and/or amplitude of cues) (Hume and Seo (2001)). The perceptual constraints proposed by Hume and Seo (2001) interact with Max-C and Linearity in the following way: *stop/obst stop, *stop/v c»M ax-C . This ranking will trigger [k] deletion in a [skt] cluster: it is better to delete [k] than to have it before a consonant after an unstressed sonorant or between two other consonants (Tableau 5). Since *stop/'v c outranks Linearity, [k] will metathesize in stressed syllables (Tableau 6). Tableau 5: Deletion of [k] is [skt] clusters (from Hume and Seo (2001:12)) /foirisk-t/ ‘Faroese’ *stop/ obst__stop *stop/v c a. 'foi.riskt b. 'foi.rikst ® ” c. 'foi.rist Max-C Linear stop/ v c Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 212 Tableau 6: Metathesis in [skt] clusters in Faroese (from Hume and Seo (2001:12)) /fesk-t/ ‘fresh’ *stop/ obst stop *stop/v c Max-C - s t o p / W | Linear a. 'fsskt ■ = , ! ■ ‘ b. 'fekst c. 'fest *! I " | ■ ' ' I 13. Conclusion This chapter summarizes the main findings of the survey from chapter 2. It discusses the main generalizations by process and by position and how these relate. Additionally, it explores the factors behind stress- and foot-sensitive consonantal phenomena: perception, aerodynamics, duration, and prominence/rhythmicity, and classifies the phenomena in chapter 2 according to these. It proposes that alternations with durational, aerodynamic or perceptual ambiguity bases are phonetic (but can be phonologized) and prominence-based and perceptually-based alternations (excluding perceptual ambiguity) are phonological. An important finding is that three different types of consonantal phenomena are attested: strictly stress-sensitive, strictly foot-sensitive, and stress/foot sensitive. Rhythmicity is distinguished from prominence as the drive for the rhythmic organization of syllables within a word. Strictly foot-sensitive consonantal processes are mainly rhythmic (prominence) oriented. Prominence-related approaches to foot structure are not very common; exceptions include Kenstowicz (1996) for vowel sonority and de Lacy (1999) for tonal phenomena. This chapter shows in detail how prominence-related, durational and aerodynamic phenomena can be analyzed by Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 213 means o f Prominence Alignment (Prince and Smolensky (1993)). Stress-sensitive perceptual-based phenomena are straightforwardly captured in the grammar via approaches like licensing-by-cue and others, which formalize the notion that cues are avoided in perceptually-challenged positions (Steriade (1997), Hume and Seo (2001), among others). The next two chapters consist of case studies. The first is an aerodynamically motivated process in Basque Spanish by which frication of codas is more often found in stressed syllables. The second considers rhythmically determined alternations in related Capanahua and Huariapano. These demonstrate in more detail how Prominence Alignment works in relation to foot structure and stress. For detailed analyses of stress-related perceptual and durational phenomena I refer the reader to Steriade (1997), Hume and Seo (2001), and Crosswhite (1999). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 214 Chapter 4: Aerodynamics in N orth-Central Peninsular Spanish 1. Introduction This chapter consists of a case study on the role of stress in the pronunciation of coda /b, d, g/ in North-Central Peninsular Spanish (henceforth NC Spanish). In this dialect /b, d, g/ are often fficated and devoiced in coda position. This process has been previously analyzed as phonological coda devoicing (Gonzalez (1999), Morris (in press)). Experimental research has shown that frication and devoicing of coda /b, d, g/ is sensitive to stress, with frication being more likely in stressed syllables (Gonzalez (2002b)). Based on these results, this chapter argues that frication and devoicing of coda /b, d, g/ is caused by increased aspiration in stressed syllables and that this process is an example of an aerodynamically-motivated consonantal process. Additionally, this chapter shows that the effect of stress in this process is variable across and within speakers, thus exemplifying a phonetic rather than a phonological consonantal process. This case study contrasts with that in chapter 5 on Panoan foot-conditioned consonantal alternations, which will be shown to be rhythmic and phonological. Stress-sensitive frication and devoicing in NC Spanish is discussed within the background of Spanish spirantization. Spirantization is defined as a type of lenition whereby fricatives (spirants) are formed by way of weakening from a stop Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 215 (Kenstowicz (1994:35)). It is widely assumed that Spanish /b, d, g/ have fricative allophones, i.e., produced with turbulent airflow (Harris (1969), Lozano (1979), Mascaro (1984), Amastae (1986, 1995)). However, acoustic studies show that the continuant allophones of /b, d, g/ are generally approximants, i.e., produced without turbulence (Martinez Celdran (1984), Romero (1995); see also Ladefoged (1982), Bakovic (1995)). This is my take here; nonetheless, I keep the term spirantization as a practical cover term for this process in Spanish. To briefly discuss how NC Spanish departs from spirantization, consider the following. Typically, /b, d, g/ have continuant allophones after vowels regardless of syllable position; non-continuant allophones occur utterance-initially (1). (1) Utterance-initially: [b, d, g] After a vowel: [J 3 , 5, r] bota ['bo.ta] ‘boot’ la bota [la 'J3o.ta] ‘the boot’ dolor ['do.lor] ‘pain’ hado ['a.5o] ‘destiny’ gata ['ga.ta] ‘cat’ la gata [la 'ra.ta] ‘the cat’ pared [pa.'reS] ‘wall’ In NC Spanish, a word like pared ‘wall’ is pronounced [pa.'reG], with a fricated and devoiced /d/ rather than with a voiced approximant. The majority of Spanish words that have a coda /b, d, g/ are stressed in the syllable containing /b, d, g/. In some dialects of Spanish like Colombian Spanish, stress influences the realization of onset /b, d, g/ (Amastae (1986), Kim (2002); see chapter 2). An obvious question is Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 216 whether stress is a factor in the devoicing and frication of coda /b, d, g/ in NC Spanish. This chapter reports an acoustic experiment that tested the influence of stress in this consonantal process (Gonzalez (2002b)). One of the main findings of this study is that stress affects frication, but not voicing. Frication is more likely in stressed syllables. This is explained through the aerodynamics of stress; stressed syllables produce more airflow (Ladefoged (1967, 1993)) and are more conducive to aspiration, frication, or even affrication (chapter 3). In the case of NC Spanish, more airflow leads to more likelihood of frication, which directly translates into less voicing because of the conflicting aerodynamic requirements between the two (Ohala (1983, 1997), Smith (1997)). This argues against previous approaches to this process as phonological coda devoicing (Gonzalez (1999), Morris (in press)). A second finding is that NC Spanish can be divided into two different dialects: the Northern dialect, or Basque Spanish (the dialect of Spanish spoken in the Basque Country) and the Central dialect (the dialect of Spanish spoken in a large part of Castile). The distinction between these two dialects is based on the realization of coda /b, d, g/. In Basque Spanish coda /b, d, g/ are pronounced as voiceless fricatives, while in Central Spanish coda /b, d, g/ are mostly realized as stops or voiced fricatives. The geographical location of both dialects is shown in (2). Other languages and dialects that are mentioned in this chapter are also shown in (2). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 217 (2) Languages and dialects spoken in Spain Basque Spanish 'Northern Dialect1 North-Central Peninsular Spanish Catalan 'Central Dialect' S P A I N Andalusian Spanish A third and final finding is that the effect of stress on the realization of coda /b, d, g/ in Basque Spanish is variable. Most speakers pronounce /b, d, g/ as fricatives in coda position, but only a subset of them have more frication in stressed syllables than in unstressed syllables. Together with the aerodynamic nature of this process, this suggests that stress-sensitive frication of coda /b, d, g/ is a variable phonetic effect. Because of this, it is proposed in this chapter that the effect of stress on frication should not be encoded in the grammar in Basque Spanish. This chapter is organized as follows. Section (2) provides background on spirantization in Peninsular Spanish, and section (3) provides an analysis of spirantization in Optimality Theoretical terms. Section (4) focuses on frication and devoicing of coda /b, d, g/ in NC Spanish. This section shows that stress has a variable effect on frication, and that devoicing is a by-product of frication. It also Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 218 shows that NC Spanish is divided into two dialects regarding the pronunciation of coda fh, d, g/: Basque Spanish, where coda /b, d, g/ are realized as voiceless fricatives, and Central Spanish, where coda /b, d, g/ are realized as voiceless stops or voiced fricatives. Section (5) provides an Optimality Theoretic analysis of the categorical frication process in the Northern dialect. Finally, section (6) is the conclusion. 2. Spirantization in Peninsular Spanish Spirantization is extremely variable across Spanish dialects (Amastae (1986, 1995), Carreira (1998), among others). I focus here on Castilian Spanish, the standard dialect spoken in Spain. A phonemic inventory for Castilian Spanish is given in (3) (based on D ’lntrono, Del Teso and Weston (1995)). All consonants in (3) are phonemic except for [J 3 , 5, y] and [ < j > , r)], included in the chart for the purposes of the discussion. (3) Castilian Spanish: phonemic inventory Labial Labio dental Inter Dental Dental Alveolar Palatal Velar Stops p b t d k g Fricatives $ f 0 s X Approximants § 5 T Affricates tf Nasals m n J 1 g Liquids r r 1 A Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 219 Castilian Spanish has both voiceless and voiced stops (4, 5). Voiced stops are realized as such in utterance-initial position and after nasals (6). They are realized categorically as voiced approximants after vowels, liquids, approximants, and fricatives (7). The data below and in the remainder of this chapter (unless noted otherwise) comes from Harris (1969), Jannedy, Poletto and Weldon (1994), Kenstowicz (1994), Harris-Northhall (1990), and from the author’s fleldnotes from work with native speakers from Spain. (4) Minimal pairs: voiceless and voiced stops (a) /bino/ ‘wine’ /gino/ ‘pine’ (b) /m anda/ ‘he sends’ /manta/ ‘blanket’ (c) /manga/ ‘sleeve’ /m anka/ ‘one-handed’ (5) Spanish stops Bilabial Dental Velar Voiceless P t k Voiced b /J3 d/a g/r (6) Distribution of [b, d, g] (a) (b) (c) Initially ['bo.ta] [do.'lor] fga.to] After nasal ‘boot’ [im.'bjer.no] ‘winter’ ‘pain’ ['an.da] ‘beat it’ ‘cat’ ['ten.go] ‘I have’ After rn ['kal.do] ‘broth’ (7 ) (a) (b) Distribution of [J 3 , 5, r] After vowel [sufkte.'ra.neo] [0iu.'5a5] ‘subterranean’ ‘city’ After fricative/approximant [ad.'[3ien.to] ‘advent’ [aj3.'5i.ka] ‘he abdicates’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 220 (c) [0iy.'0ay] ‘zig-zag’ [dis.'yus.to] ‘trouble’ (d) [la 'J3o.8a] ‘the wedding’ [m a.ra.'kaj.§o] ‘Maracaibo’ (e) ['ka.5a] ‘each’ ['dew .da] ‘debt’ (f) ['a.Yo] ‘I do’ ['kaj.yo] T fall’ After liquids (g) ['kal.j3o] ‘bald’ [0er.'§e.0a] ‘beer’ (h) _ [’ber.Se] ‘green’ (i) ['sal.yo] ‘I go out’ [kar.'yar] ‘to load’ The distribution of /d/ stands apart in one respect, /d/ is realized as a stop after /l/ but as an approximant after /[/; /b/ and /g/ are realized as approximants in both cases. Leaving aside this case for the moment, the generalization is that [J 3 , 5, y] occur after [+continuant] segments, and [b, d, g] elsewhere. 3. An Optimality Theory analysis This section provides an Optimality Theory analysis of the spirantization facts in Castilian Spanish. It captures the allophonic distribution of [b, d, g] and [§, 8, y] and the phonemic distinction between [b, d, g] and [p, t, k]. This analysis provides the basic ranking for spirantization and voicing in Spanish, which will be relevant for the analysis of frication of /b, d, g/ in Northern Spanish (section 5). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 221 3.1 Allophonic variation between [b, d, g] and [J 3 , 5, r] Allophonic variation is captured in OT through the ranking of markedness constraints above faithfulness constraints. For [b, d, g] and [§, 5, r] the relevant faithfulness constraint is Ident-IO [continuant] and the relevant markedness constraints are * V oiced stop and *[+son, +cont, -lat, -rhotic] (8). (8) Ident-IO [continuant] Correspondent segments have identical values for the feature [continuant]. Let x be a segment in the input and y a segment in the output. IfxRy and x is [ycontinuant], then y is [ycontinuant] (McCarthy and Prince 1995) * V oiced stop Voiced obstruent stops are prohibited. *[+son, +cont, -lat, -rhotic] [+son, +cont, -lat, -rhotic] segments are prohibited. Voiced stops and consonantal approximants other than glides and liquids are dispreferred cross-linguistically (Maddieson (1984), Ladefoged and Maddieson (1996)). Voiced stops in inventories imply voiceless stops (Maddieson (1984)) and they impose harder aerodynamic requirements than voiceless stops (Ohala (1983)). Consonantal approximants other than glides or liquids are rare cross-linguistically (Maddieson (1984)). In the absence of any preceding segment (i.e., utterance- initially), a voiced stop is preferred to an approximant. This indicates that *[+son, +cont, -lat, -rhotic] outranks *Voiced stop (Tableau 1). Tableau 2 shows that *[+son, +cont, -lat, -rhotic] outranks Ident-IO [continuant]. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 222 Tableau 1: *[+son, +cont, -lat, -rh o tic]» *Voiced STOP +son, +cont, -lat, -rhotic V oiced stop you go a. bas Tableau 2: *[+son, +cont, -lat, -rhoticl»lDENT-IO [continuant] /J3as/ ‘you go’ *[+son, +cont, -lat, -rhotic] Ident-IO [continuant] a. bas * b. J3as *, Candidate (a) is selected as optimal in Tableau (2). It is better to change the [continuant] specification from the input than to have an approximant in initial position. Tableaux (1, 2) also show that even if the input is changed, the right candidate is selected (Richness of the Base; Prince and Smolensky (1993)). Previous analyses of spirantization differ in the choice of [b, d, g] or [J 3 , 5, y] as basic (Harris (1969), Lozano (1979), Mascara (1984), among others). With the notion of Richness of the Base a basic form is not necessary. Both [b, d, g] and [$, 5, y] are possible inputs; which one is selected depends on the particular constraint ranking. Voiced stops cannot occur in Spanish after [+continuant] segments. Context- sensitive [A 0 ]/*voiced stop and [A f]/*voiced stop capture this fact (9). (9) [A 0]/*voiced stop A voiced stop is prohibited after an open segment (vowels and approximants are open; Steriade (1993, 1994)). [A f]/*voiced stop A voiced stop is prohibited after a fricative. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 223 The constraints in (9) are based on Aperture Theory (Steriade (1993, 1994)). Their motivation is articulatory; it is hard to achieve the closure necessary for a voiced stop if the surrounding segments are not closed (Kirchner (1998)). In Spanish, voiced stops are allowed utterance-initially and after segments with closure (nasals, other stops). Tableau 3 shows that it is better to prevent a voiced stop after an open segment than to avoid an approximant. Tableau 3: [A 0 ]/*voiced stop, [Af]/*voiced stop»*[+son, +cont, -lat, -rhotic] /odio/ ‘hatred’ [A0 ]/*voi stop [A f]/*voi stop *[+son, +cont, -lat, -rhot] a. odio • b. odio * It might be desirable to generalize [A 0 ]/*voiced stop to all stops. However, voiceless stops are unrestricted in Spanish, and Ident-IO [continuant] is ranked low to allow for spirantization of voiced stops. A constraint like [A 0 ]/*stop would be problematic in that it would always choose a voiceless fricative rather than a voiceless stop after an open segment. This is shown in Tableaux 4, 5. Tableau 4: [A o]/*stop » IDENT-IO [continuant] (I) /rata/ ‘rat’ [A 0]/*stop IDENT-IO [continuant] ! a. ra.ta *! r • ■ ^ b.ra.Ba Tableau 5: [A 0 ]/*stop » Ident-IO [continuant] (II) /ra0a/ ‘race’ [A0]/*stop IDENT-IO [continuant] a. ra.ta *! * b. ra.Ba Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 224 For this reason I formalize the constraints as [A 0 ]/*voiced stop and [A f]/*voiced stop. If [A 0]/*voiced stop, [A f]/*voiced stop»*[+son, +cont, -lat, -rhotic], and if *[+son, +cont, -lat, -rh o tic ]» * Voiced Stop, Ident-IO [continuant], then, by transitivity, [A 0 ]/*voiced stop, [A f]/*voiced stop »*[+ son , +cont, -lat, -rhotic]»* Voiced Stop, Ident-IO [continuant]. Tableau 6 summarizes this ranking. Candidate (a) is selected as optimal since all other candidates violate high- ranked [A 0 ]/*voiced stop or [A f]/*voiced stop. Tableau 6: /abdikar/ ‘to abdicate’ /abdikar/ ‘to abdicate’ [A 0 ]/ *vd stop [Ad/ *vd stop T c S i s a. agdikar ** , | ** b. ajMikar * ■ * * . c. abdikar *! . * • ■ * ' * d. abdikar *! 1 **• - [ 3.2 Phonemic distribution of /b, d, g/ and /p, t, k/ Phonemic distribution is captured in OT through the ranking of faithfulness constraints over markedness constraints (Prince and Smolensky (1993)). In this case, Ident-IO [voice] (10) outranks * Voiced stop; it is more important to keep the voicing specification from the input than to avoid voiced stops. This ranking also predicts that there is a voicing contrast between stops (Tableaux 7, 8). (10) Ident-IO [voice] Correspondent segments have identical values for the feature [voice]. Let x be a segment in the input, and y a segment in the output. IfxRy andx is [yvoice], theny is [yvoice]. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 225 Tableau 7: I d e n t -IQ [v oice]» *VoiCED STOP /b in o / ‘w in e ’ Id e n t -IO [voice] * V o ic e d St o p a. bino b. pino *! 1 Tableau 8: Id e n t -1 0 [ v o i c e ] » * V o ic e d S to p /pino/ ‘wine’ Id e n t -IO [voice] * V o ic e d St o p a. bino * b. pino Spanish has voiced approximants but not voiceless ones, and voiceless fricatives, but (usually) not voiced fricatives. This is expressed through markedness constraints against these dispreferred segments (11). (11) * V o ic e l e ss A p p r o x im a n t [+son, +cont, -voi] segments are prohibited * V o ic e d fr ic a t iv e Voiced fricatives are prohibited. Voiceless approximants are cross-linguistically dispreferred, and voiced fricatives are more marked than voiceless fricatives cross-linguistically (Maddieson (1984)). Voiced fricatives also have conflicting aerodynamic requirements for voicing and frication. A pressure drop across the glottis is needed for voicing to occur, and at the same time high pressure is needed for frication (Ohala (1983, 1997), Smith (1997)). Since voiceless approximants never occur in Spanish, I assume that ^V o ic e l e ss A p p r o x im a n t is undominated. Voiced fricatives are rarer than voiced approximants in Spanish; thus, *V o ic e d f r ic a t iv e » * [ + s o n , +cont, -lat, -rhotic] (Tableau 9). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 226 Since voiced fricatives can only be allophones of voiced segments, I assume that * V o ic e d f r ic a t iv e is outranked by Id e n t [voice] (see section 5 for discussion). Tableau 9: * V o ic e d f r ic a t iv e »*[+ son , + con t, -lat, -rhotic] /ada/ ‘fairy’ * V o ic e d f r ic a t iv e *[+son, +cont, -lat, -rhotic] a. a5a • ^ b.ada * Id e n t -IO [voice] outranks * [+ so n , + co n t, -lat, -rhotic]; it is more important to keep the input voice specification than to avoid a voiced approximant. This ranking does not produce a phonemic contrast between voiced and voiceless approximants in Spanish since ♦ V o ic e l e ss A p p r o x im a n t is undominated (Tableau 10). Tableau 10: Id e n t [voice]» *[+son, +cont, -lat, -rhotic] /redes/ ♦V o ic e l e ss Id e n t ♦ V o ic e d ♦[+son, +cont, ‘nets’ A p p r o x im a n t [voice] F r ic a t iv e -lat, -rhotic] a. redes ♦ b. re0es c. re0es *! * After vowels /b, d, g/ are realized as voiced approximants, not as voiced fricatives (Tableau 11). Note that in Tableau 11 * V o ic e d f r ic a t iv e and [A o ]/* v o iced stop are not crucially ranked. Tableau 12 shows that /p, t, k/ are optimally realized as voiceless stops. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 227 Tableau 11: N o voiceless fricatives /redes/ ‘nets’ *Voi FR IC ! [Ao]/*voiced stop *[+son, +cont, -lat, -rhotic] a. redes 1 .* b .redes *( 1 c. redes Tableau 12: /ata/ ‘he ties’ /ata/ ‘he ties’ * V l e ss A p p r o x Id e n t [voi] [A 0]/ *Vd Stop *[+son, +cont, -lat, -rhotic] a. ata b. ada *(0 *0) c. a0a *! * d. ada * In conclusion, Id e n t -IO [voice]»*VoiCED S t o p captures the phonemic contrast between voiced and voiceless stops and the universal preference for voiceless over voiced stops. [A 0 ]/*voiced stop, [A f]/*voiced stop»*[+son, +cont, -lat, -rh o tic]» lD E N T -IO [continuant] accounts for allophonic variation between [b, d, g] and [§, 5, y]. Undominated *Voiceless Approximant and high-ranked *Voiced f r ic a t iv e capture the universal dispreference for voiceless approximants and voiced fricatives and their absence or rarity in Spanish. 3.3 The problem of [Id] It was noted in section 2 that /d/ is pronounced as a voiced stop after III. This does not occur for /b, g/, which are pronounced as voiced approximants (Mascaro (1984), Palmada (1997), Kenstowicz (1994)). This section explores two different solutions Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 228 for this problem. The first relates the failure of spirantization in /Id/ to the assimilation of III to Id! in place features (Harris (1969, 1984); Mascara (1984); Padgett (1991); Palmada (1997); Kirchner (1998)). Since III triggers spirantization in most contexts, I assume III is [+continuant] (but see Harris (1969:39), Kirchner (1998) on a different approach, and Chomsky and Halle (1968:318) for the dual patterning of III as [+/-continuant] across languages). Laterals place-assimilate to a following dental or palatal consonant (12a) (Harris (1969); cf. Alarcos Llorach (1950), Quilis (1981)). Elsewhere III is alveolar (12b, c). The diacritic [ _ ] is used for palatal articulations, and [ „ ] for dental articulations. (12) Place assimilation of III in Spanish (a) ['alto] ‘tali’ ['koj.tja] ‘quilt’ (b) fa.la] ‘wing’ ['bal.sa] ‘boat’ (c) ['al.6a] ‘raise’ Laterals do not assimilate to interdental segments (12c). L a t e r a l P l a c e A ssim il a t io n and * In t e r d e n t a l l a t e r a l capture this fact (13). (13) L a t e r a l P l a c e A s sim il a t io n A lateral agrees in tongue tip orientation with an immediately following coronal (based on Gafos (1996) * In t e r d e n t a l l a t e r a l Interdental laterals are prohibited. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 229 * In t e r d e n t a l l a t e r a l blocks L a t e r a l P l a c e A s s im il a t io n (henceforth LPA) (Tableau 13). The diacritic [ J indicates an interdental articulation. Tableau 13: In t e r d e n t a l L a t e r a l» L P A /al0a/ ‘raise’ * In t e r d e n t a l L a t e r a l LPA a. al0a * b. al0a , LPA outranks [A 0]/*voiced stop. In the sequence [Id], it is better to assimilate to the dental articulation of the stop than to avoid a voiced stop after a [+continuant] segment (Tableau 14). Tableau 14: /kaldo/ ‘broth’ /kaldo/ ‘broth’ ■ " I n t e r d e n t a l L a t e r a l LPA [A 0]/*voiced stop t * * a. kaldo * ■ : b. kaldo V * " c. kaldo d. kaldo ■ , In tableau 14 candidate (b) violates LPA because it has an alveolar [1] followed by a dental [d]. Candidate (c) violates LPA because it has an alveolar [1] followed by an interdental [5]. Since candidate (d) violates undominated *In t e r d e n t a l L a t e r a l , the optimal candidate is (a), where III is dental and /d/ realized as a voiced stop. Before Pol and /g/ spirantization occurs after III because neither * In t e r d e n t a l L a t e r a l or LPA are relevant (Tableau 15). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 230 Tableau 15: /kalbo/ ‘bald’ /kalbo/ ‘bald’ * Interdental Lateral LPA [A0 ]/*voiced stop a. kal]3o b. kalbo Previous p h o n o lo g ic a l analyses of spirantization have analyzed spirantization blocking in /Id/ as a homorganicity effect: /d/ fails to spirantize after III because both share the same place of articulation (see, for instance, Carreira (1998), Kirchner (1998)). An important assumption for this analysis is that III is [-continuant] and that continuancy depends on place (Padgett (1994)). This analysis predicts that spirantization will also fail to apply in /d l/ sequences—as in the word /adlatere/ ‘inseparable companion’—because the sequence /dl/ is also homorganic (Kirchner (1998)). However, spirantization applies in /d l/ sequences, and /adlatere/ is realized as [adlatere]. Under the approach proposed here [adlatere] is not problematic; the choice of [d] versus [d] is captured by high-ranked A 0 /*voiced stop and the lack of interaction of * Interdental lateral and LPA for /dl/.1 A second possibility to account for [Id] is that [d] does not violate [A 0 ]/ *voiced stop in [Id] since [1] counts as a closed segment for [d]. This insight goes back to Mascaro (1984), who states that /d/ is a stop following III because there is no airflow in the central dental region during the articulation of the lateral. Since airflow 1 Kirchner (1998:148) suggests that paradigmatic faithfulness to the prefix [a3-] in Spanish underlies spirantization for /dl/ cases. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 231 is essential for approximants and this is lacking when [1] precedes [d], the lateral acts as a stop with respect to [d]. In other words, spirantization occurs to make a segment more similar to the preceding segment. To spirantize /d/ after IV makes no sense, since the tongue tip is already in stop position. Spirantization would make the second consonant less similar to the first. How to precisely formalize this is a matter for further research. A ranking lattice for all constraints considered so far is given in (14). No crucial ranking has yet been established for Id e n t -IO [continuant] and * V o ic e d St o p or for * V o ic e d F r ic a t iv e and A 0 /* voiced stop, A f/*voiced stop. (14) Ranking lattice for Castilian Spanish * [+ so n , + con t, -v o i] * in t e r d e n t a l L a t e r a l *[+son, +cont, -lat, -rhotic] | \ Id e n t -IO [continuant] *V o ic e d S t o p 4. Frication and devoicing of coda /b, d, g/ in NC Spanish This section focuses on the realization of coda /b, d, g/ in NC Spanish. Section 4.1 provides background on this process. Section 4.2 reports the acoustic study in LPA Id e n t -IO [voice] * V o ic e d f r ic a t iv e [A 0 ]/*voiced stop [A f]/*voiced stop Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 232 Gonzalez (2002b) on the effect of stress ffication and voicing in coda coda /b, d, g/ in this dialect. 4.3 discusses the phonetic and phonological implications of the findings from Gonzalez (2002b). 4.1 Coda variation in Spanish dialects The realization of coda /b, d, g/ in Spanish varies highly from dialect to dialect. Typically coda /b, d, g/ are not pronounced in dialects of American Spanish or in Andalusian Spanish. In Peninsular Spanish coda lb, d, g/ are mostly pronounced as voiced approximants. Coda lb, d, g/ are realized as voiceless fricatives [ < j > , 0, x] in NC Spanish (Hualde (1989), Martinez-Gil (1991), Gonzalez (1999, 2002b), Morris (in press)). A comparison of the pronunciation of coda lb, d, g/ in three dialects of Peninsular Spanish is given in (15). (15) Realization of coda lb, d, g/ Peninsular Spanish Andalusian Spanish NC Spanish Gloss [pa.'red] [pa.'re] [pa.'reG] ‘wall’ ['klu§] ['klu] ['klu$] ‘club’ [maY.da.'le.na] [ma.da.'le.na] [max.da.'le.na] ‘muffin’ Morris (in press) assumes that spirantization yields fricatives rather than approximants, and considers that coda lb, d, g/ in NC undergo devoicing. Thus, NC Spanish would be similar to other languages like Dutch and Catalan, where obstruent voicing is neutralized in syllable-coda position and obstruents become uniformly Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 233 voiceless (Mascaro (1984), Kager (1999)). However, continuant allophones of coda fb, d, g/ are more often than not approximants (Martinez Celdran (1984), Romero (1995)). It would be expected that coda devoicing in NC Spanish would yield voiceless approximants, or even voiceless stops; but not voiceless fricatives. Another consideration is that most words with coda fb, d, g/ in Spanish are stressed in the syllable containing these segments. As mentioned in chapter 2 stress plays a role in the realization of /b, d, g/ in various Spanish dialects, including Colombian Spanish, fb, d, g/ tend to be realized as stops in onsets of stressed syllables, but as approximants in onsets of unstressed syllables (in Colombian Spanish coda /b, d, g/ are not pronounced) (Amastae (1986, 1995); Cole, Hualde and Iskarous (1998); Kim (2002)). The question arises whether stress has an effect on the outcome of coda fb, d, g/ in NC Spanish. 4.2 Acoustic study: the influence of stress in coda fb, d, g/ in NC Spanish This acoustic study reported in Gonzalez (2002b) was designed to test whether stress had any effect on frication and devoicing in coda fb, d, g/ in NC Spanish. The hypothesis is that coda fb, d, g/ in stressed syllables are more likely to be devoiced and fricated than coda fb, d, g/ in unstressed syllables. For example, in a word like pa'red ‘wall’, where /d/ occurs in a stressed syllable, /d/ is expected to be pronounced as [0] (a voiceless fricative). However, in a word like 'huesped ‘guest’, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 234 where /d/ occurs in an unstressed syllable, /d/ is expected to be pronounced as [5] (a voiced approximant). The reason for these expectations is that stress occurs with an increase in the activity of the respiratory muscles, causing a greater amount of air to be pushed out of the lungs (Ladefoged (1967, 1993)). Assuming the same particular glottal and supraglottal configurations occur (Romero (1995); see discussion in section 4.3 in this chapter), this will result in greater airflow. A greater amount of airflow is conducive to frication and inhibits voicing due to the creation o f higher oral pressure, which conflicts with the requirement of a pressure drop across the glottis needed to sustain vocal fold vibration (Ohala (1983, 1997), Smith (1997)). The experiment used nonce words for two reasons. First, while many Spanish words have /d/ in coda—either word-medially or word-finally— only a few have Pol or /g/ in this position. Secondly, most words with Id/ in coda are stressed in that syllable, so comparison between stressed/non-stressed pairs of words is limited. The nonce words in this study were disyllabic words in an alternating stress pattern with either Pol, /d/ or Igl in coda position. In half of the words, the syllable containing the coda /b, d, g/ was stressed, and in the other half of the words it was unstressed. The independent variables in this experiment were: (i) Target consonant (coda Po, d, g/); (ii) Target syllable stress (stressed/unstressed); (iii) Word position (medial/final); and (iv) Preceding vowel (/a, i, u/). A total of thirty-six nonce words were constructed. The context for the pronunciation of fb, d, g/ is post-vocalic (so Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 235 that spirantization is triggered) and before a /b/ (for coda /d, g/) or before a /g/ (for coda /b/). These segments are always voiced so that possible regressive devoicing is not a factor (cf. Morris (in press)). The words were recorded in a sentence frame of the form Se llama ‘His name is )’. The frame for these words was controlled to avoid the influence of stress from neighboring words. Table 4.1 shows a complete list of the words used in this experiment. Stress is marked in this list according to the conventional orthography of Spanish; stress marks are not directly relevant to the target syllable stress of fb, d, g/. Table 4.1: List of nonce words (underline corresponds to target consonant) Vowel Word Position Target Stress Target Consonant fbl Id/ /g/ /a/ Medial Stressed Labga Gota Ladba Bote Lagba Bote Unstressed Labgan Gotera Ladban Botella Lagban Botella Final Stressed Labab Gotera Labad Botella Labag Botella Unstressed Lambab Gota Lambad Bote Lambag Bote M Medial Stressed Libga Gota Lidba Bote Ligba Bote Unstressed Libgan Gotera Lidban Botella Ligban Botella Final Stressed Labib Gotera Labid Botella Labig Botella Unstressed Lambib Gota Lambid Bote Lambig Bote /u/_ Medial Stressed Lubga Gota Ludba Bote Lugba Bote Unstressed Lubgan Gotera Ludban Botella Lugban Botella Final Stressed Labub Gotera Labud Botella Labug Botella Unstressed Lambub Gota Lambud Bote Lambug Bote This list of words was repeated seven times in order to obtain seven tokens for each sentence. This yielded a total of 252 tokens. The order of the sentences was Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 236 randomized for each block and speaker. Three filler sentences were included in the list, one at the beginning and two at the end of the list. Ten native speakers from Peninsular Spanish were recorded. Only university students or college graduates were selected due to the difficulty of the reading task, which included unfamiliar words with unusual stress patterns. The data from one speaker was thrown out due to a number of stress misplacement errors, so the analysis includes data from 9 subjects. Of these, six subjects are from the Basque Country (indicated as N or ‘Northern Spanish’ in the result tables), and three from Madrid (indicated as Cen ‘Central Spanish’). None of the speakers from the Basque Country are bilingual in Basque. A Marantz PMD 201 tape recorder and a Shure SM10 head-mounted microphone were used for the voice recordings. Subjects were asked to use an informal style of pronunciation. The data were digitized at a 22,000 Hz sampling rate using the Macquirer system. The following dependent variables were measured: (16) Dependent variables (i) Vowel duration (preceding coda consonant): First to last glottal pulse with evident formant structure. (ii) Consonant duration (coda): - Voiced glide: Formant interval (low amplitude). - Voiceless fricative: Frication interval. - Voiced fricative: From sudden amplitude drop (constriction formation) to end of frication and/or voicing (which ever is later). (iii) Frication (presence/absence): Aperiodic waveform and random noise in spectrogram. (iv) Voicing (presence/absence): Periodic waveform and vertical striations in spectrogram. (v) Voicing duration: Duration of voicing in milliseconds. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 237 (vi) Percent voicing: Percent of voicing duration relative to total consonant duration. Waveforms, spectrograms and power spectra were used as displays. Figures (4.1- 4.3) show sample waveforms and spectrograms for voiceless fricative [0], voiced fricative [5], and voiced approximant [6]. Figure 4. 1. Spectrogram of voiced fricative /d/. Sequence [lu5ba] from ‘Ludba Bote’ II 8200 Figure 4.2. Spectrogram of voiceless fricative /d/. Sequence [i0] from ‘Lambid Bote’ . . . .W i l l - . . . . . . ■ I .. ...... . . . . . . . . . . . Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 238 Figure 4.3. Spectrogram of voiced approximant /d/. Sequence [ladban] from ‘Ladban Botella’ A four-factor full interaction ANOVA was run on the data using StatView (SAS (1999)). The factors were preceding vowel, target consonant, target syllable stress and word position. The result was considered significant if p<0.05. Significant results are marked with * in the results tables. For speakers 6 and 8, a three-factor full interaction ANOVA was run instead due to the amount of missing values (see Table 4.3), with target consonant, target syllable stress and word position as factors. The following discussion concentrates on two main findings: the effect of stress on each dependent variable (vowel duration, consonant duration, frication and voicing), and the dialectal differences found for frication and voicing. For other results including word position see Gonzalez (2002b). Table 4.2 shows the results obtained for stress on each dependent variable for each speaker. Voicing duration is considered for all speakers pooled later rather than by speaker due to the insufficient number of voiced tokens for each speaker in this fully crossed model. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 4.2. S 1 xess: Mean values for dependent variables (by speaker) Table 2 Stressed Unstressed p - VALUE *V. duration 109 ms 92 ms <.0001 Sp.1 C. duration 100 ms 97 ms .2358 (NOR) *% Fricated 99.1% 95.4% .0137 % Voiced 23.7% 27.8% .5435 *V. Duration 93 ms 83 ms <.0001 Sp.2 C. Duration 72 ms 71 ms .4253 (NOR) *% Fricated 93.1% 78.4% <.0001 *% Voiced 0.9% 5.9% .0325 *V. Duration 102 ms 83 ms <.0001 Sp.3 C. Duration 54 ms 54 ms .8202 (NOR) *% Fricated 88.1% 76.7% .0262 *% Voiced 39.4% 26.2% .0191 *V. Duration 83 ms 68 ms <.0001 Sp.4 C. Duration 69 ms 63 ms .3715 (nor) % Fricated 95.9% 90.1% .572 % Voiced 23.7% 18.8% .8761 *V. Duration 102 ms 83 ms <.0001 Sp. 5 C. Duration 76 ms 80 ms .1575 (NOR) % Fricated 94.7% 90.9% .2654 % Voiced 14.2% 13.6% .9613 *V. Duration 117 ms 95 ms <.0001 Sp. 6 C. Duration 111 ms 102 ms .0717 (NOR) % Fricated 65.4% 64.3% .1071 % Voiced 37.2% 35.7% .5720 V. Duration 69 ms 68 ms .7475 Sp. 7 *C. Duration 55 ms 64 ms .0026 (CEN) * % Fricated 75.2% 56% .0094 % Voiced 65.3% 64% .9338 V. Duration 114 ms 100 ms .0672 Sp. 8 C. Duration 106 ms 102 ms .8431 (Cen) % Fricated 30.4% 34.4 % .6657 % Voiced 19.6% 26.2% .5433 *V. Duration 97 ms 80 ms <.0001 Sp.9 C. Duration 84 ms 79 ms .1004 (Cen) % Fricated 12.1% 14.3 % .8831 * % Voiced 29.3% 11% .0002 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 240 Vowels were significantly longer in stressed syllables than in unstressed syllables (p<.0001) for all speakers excepting 7 and 8; speaker 8 follows the same trend. This result is consistent with previous observations that stress in Spanish increases vowel duration (Martinez Celdran (1984:246)). No significant effect was found for consonant length and stress overall; coda consonants are about the same length in stressed and unstressed syllables. Speaker 7 has a different pattern; codas are longer in unstressed syllables (p-.0026). For speakers 1, 2, 3 and 7, frication was significantly more likely to occur in stressed syllables than stops or approximants (Sp.l: F (1, 186)=6.189,/?=0137; Sp.2: F (1, 182)=20.610, p<.0001; Sp.3: F (1, 176)=5.027, p=.0262; Sp. 7: F (1, 165)=6.901, /?=.0094). Speakers 4 and 5 follow the same pattern but a ceiling effect may be preventing the difference from being significant. Table 3 shows the number o f fricatives, glides and stops produced by each speaker. Table 4.3. Realization of fb, d, g/ by speaker Fricative s Glides Stops Missing Sp. 1 (nor) 216 6 0 30 Sp. 2 (nor) 188 1 29 34 Sp. 3 (nor) 176 22 15 39 Sp. 4 (nor) 184 2 12 54 Sp. 5 (nor) 207 9 7 29 Sp. 6 (nor) 96 11 41 104 Sp. 7 (cen) 132 24 45 51 Sp. 8 (cen) 49 5 99 99 Sp. 9 (cen) 25 0 165 62 Total count 1273 80 413 502 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 241 The effect of stress on voicing is speaker-dependent. Speakers 3 and 9 show more instances of voiced tokens of /b, d, g/ in stressed position (Sp.3: F (1, 176)=4.644, p=.0191; Sp. 9: F (1, 154)=14.385, p=.0002). For speaker 2 voicing is more likely in unstressed position (F (1, 182)=4.644, />=.0325). No other speakers show significant differences in the number of voiced tokens as a function of stress. Summarizing, stress is a factor in vowel duration across speakers (with vowels being longer in stressed syllables) but not in consonant duration. Frication is more likely in stressed syllables for six out of the nine subjects, but there is no consistent effect of stress on voicing. A post-hoc examination of the speaker patterns suggests that speaker dialect played an important role in determining the patterns of frication and voicing. For this reason, the speakers were grouped into two dialects—North and Central—based on their geographical origins. Using Unixtat (Perlman & Horan (1986)), a three factor full-interaction ANOVA with the independent variables consonant, stress, and dialect group was conducted for frication, voicing, voicing duration and percent voicing. Speaker was treated as a between variable for dialect and a within variable for stress and consonant. The results are included in tables (4.4-4.7). Table 4.4. Mean values for frication (speakers pooled) Dialect North 70.6% Central 27.2% Stress Stressed 60.3% Unstressed 51.9% C onsonant Pol 47.2% /d/ 52.4% /g/ 68.7% Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 242 As Table 4.4 shows, dialect has a robust effect on frication (F (1, 7)=10.8Q8, /?=. 013). Most instances of coda /b, d, g/ are fricated in the Northern dialect (70.6%) while only less than a third of tokens are fricated in the Central dialect (27.2%). Frication is also more likely in stressed syllables than in unstressed syllables (F (1, 7) =10.750, p= .014). Frication was also more likely for /g/ than for fb/ and Id/ (F (2, 14)=3.858, j?=.046). Neither dialect, stress or consonant had any effect on voicing, voicing duration or percent voicing (Tables 4.5-4.7). Significant interactions were found among the three factors (dialect, stress and consonant) and all voicing measures, but due to the complexity of the interaction it is not be discussed here (Voicing: F (2,14)=4.085, jP=.040; Voicing duration: F(2,14)=4.590, p=.029; Percent voicing: F(2,14)=5.426,j p=.018). Table 4.5. Mean values for voicing (speakers pooled) Dialect North 17.2% Central 26.9% Stress Stressed 22.2% Unstressed 18.7% Consonant fb/ 21.5% /d/ 22% /g/ 17.8% Table 4.6. Mean values for voicing duration (speakers pooled) Dialect North 7 ms. Central 14 ms. Stress Stressed 9 ms. Unstressed 9 ms. Consonant /b/ 11 ms. /d/ 11 ms /g/ 6 ms. Table 4.7. Mean values for percent voicing (speakers pooled) Dialect North 9.5% Central 20.9% Stress Stressed 14.1% Unstressed 12.5% Consonant /b/ 14.5% /d/ 17% /g/ 8.5% Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 243 Summarizing, frication of coda lb, d, g/ is more likely in the Northern dialect than in the Central dialect. Frication is also more likely in stressed syllables than in unstressed syllables. Finally, frication is also more likely for /g/ than for /b/ or /d/. This study aimed at testing the significance of stress in the frication and/or devoicing of coda lb, d, g/ in NC Spanish. The results show that stress affects frication of coda fb, d, g/; frication is more likely in stressed syllables. Stress does not affect voicing of coda lb, d, g/. Additionally, frication is more common in the Northern dialect than in the Central dialect, and for /g/ than for Ibl or Id/. I suggest the following explanation for the first finding. Stress causes an increase in respiratory activity, which causes more airflow (Ladefoged (1967, 1993)). A higher volume of airflow is conducive to frication. The supraglottal constriction has to be narrow enough to allow for the possibility of frication; if the constriction is wide, no pressure will build up, and frication will not occur, even in a stressed syllable (Figure 4.4). Figure 4.4 Hypothesized interaction between stress, frication, and constriction degree Wide Constriction Stressed syllable Unstressed syllable i Less airflow Higher airflow i Frication not likely Frication not likely Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 244 Figure 4.4 (continued) Hypothesized interaction between stress, frication, and constriction degree Narrow Constriction Stressed syllable Higher airflow Frication likely Unstressed syllable | Less airflow i Frication not likely Fricatives and approximants are traditionally described as having different constriction degrees. Fricatives would have a narrower constriction, producing the turbulence required for frication. Approximants, however, have a wide constriction with which turbulence would not be possible, and frication non-existent. However, Romero’s (1995) study on fricatives and approximants in Spanish shows that there is no constriction degree difference between these two types of continuants. Romero (1995) conducted an articulatory study using movement tracking data for fricatives and approximants in Western Southern Spanish where fb, d, g/ has two allophonic sets: approximants in spirantized contexts and fricatives in contexts where spirantization interacted with debuccalization of /$/. His data show that fricatives and approximants have a similar constriction and that the difference between the two is the duration of the constriction, not the constriction degree. In an attempt to understand a simple scenario to explain these results, I assume, in accordance with Romero’s findings, that the supraglottal configuration is the same for all continuant allophones of fb, d, g/. An initial simplifying assumption Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 245 is that a similar glottal configuration also exists for the two. This yields a possible scenario, that shown in the bottom half of Figure 4.4. The results showed that stress has an effect on frication, but not on voicing. This suggests that more airflow in stressed syllables produces more frication. 4.3 Phonetic and phonological implications This section discusses the phonetic and phonological implications of the articulatory study reported in section 4.2. It focuses on the effect of stress on frication, the variability of this effect, and the dialectal differences between the Northern and Central dialects of NC Spanish. Frication is more pervasive in the Northern dialect that in the Central Dialect; it occurs about 70% of the time in the Northern dialect but less than 30% of the time in the Central dialect. This suggests that frication is close to categorical in the Northern Dialect and variable in the Central dialect. This fact supports the analysis of frication in coda fb, d, g/ in the grammar for the Northern Dialect, but not for the Central Dialect. Stress affects frication; frication occurs about 60% of the time in stressed syllables, and only about 52% of the time in unstressed syllables. The difference is significant, but the difference between frication in stressed and unstressed syllables is not so large. This suggests that the effect of stress should not be incorporated into the grammar. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 246 Voicing of coda lb, d, g/ in the Northern dialect occurs in 17% of the cases; in the cases where /b, d, g/ are realized as voiced, voicing is sustained for 9.5 % of the total duration of the consonant. For the Central dialect, voicing occurs in 27% of cases, and it is sustained for about 21% of the total duration of the consonant. This suggests that voicing is gradient in both dialects, and slightly longer and more likely to occur in the Central Dialect. Summarizing, the grammar should capture the pervasiveness of frication of coda lb, d, g/ in the Northern dialect since it is close to categorical, but not in the Central dialect. The effect of stress on frication is aerodynamic and far from categorical and should not be included in the grammar. In the Northern dialect frication is pervasive and it is aerodynamically influenced by stress and devoicing is a by-product of frication; this argues against an analysis of coda devoicing in this dialect. In Central Spanish frication is not so common, but devoicing still occurs. Only three of the subjects recorded belonged to this dialect, so the interpretation of the results for this dialect is tentative. Two patterns can be distinguished. Speaker 7 has a large number of fricative realizations of coda lb, d, g/ and considerable voicing. Speakers 8 and 9 show a large number of stops and few voiced tokens of coda fb, d, g/ (table 4.2). Additionally, many tokens of coda lb, d, g/ are missing for these speakers (table 4.3). The pattern of speakers 8, 9 suggests that coda devoicing Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 247 might take place in Central Spanish. More data from more speakers from this dialect could help to elucidate whether this is the case. 5. An OT analysis of coda frication of fb, d, g/ in Basque Spanish This section provides an OT analysis for the only (near) categorical result that arises from the experiment described in section 4.2: the frication of coda fb, d, g/ in Basque Spanish. I propose to analyze this process through a dispreference for coda approximants and stops in this dialect (17). (17) *[+son, +cont, -lat, -rhotic]/coda Avoid [+son, +cont, -lat, -rhotic] in coda *[-son, -cont]/coda Avoid [-son, -cont] in coda Perceptual considerations partially motivate the constraints *[-son, -cont]/coda and *[+son, +cont, -lat, -rhotic]/coda (henceforth *voi approx]0). Stops are less perceptible in coda position; their burst and transitions are better perceived before vowels (Steriade (1997)). Non-liquid approximants are not very salient in coda position; they frequently delete or turn into stops. These two constraints are unranked. *[-son, -cont]/coda and *voi approx]0 outrank Ident [voice] (tableaux 16, 17). Tableau 16: *[+son, +cont, -lat, -rhotic]/coda » I dent [voice] /red/ ‘net’ *voi approx]a Ident [voice] a. red 7 t3 P b. re0 * Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 248 Tableau 17: *[-son, -contl/coda » I d e n t [voice] /red/ ‘net’ *[-son, -cont]/coda Ident [voice] ^ a. red : b. re0 .............. .......— , : , r......................... ' Markedness constraints against dispreferred voiceless approximants and voiced fricatives, and context-sensitive markedness constraints penalizing voiced approximants and stops in coda position result in the selection of voiceless fricatives in this position (Tableau 18). Coda /d/ surfaces as [0], Candidate (d) violates low- ranked Ident [voice], but the rest of the candidates violate higher-ranked constraints, so (d) is selected as optimal. For space reasons, the constraint *[+son, +cont, -lat, -rhotic] will be abbreviated in tableaux 18-22 to *5, and *voi approx]a will be abbreviated to *5] a Tableau 18: Realization of coda /d/ in Northern Spanish *5]a | *[-son, : -cont]]a /red/ ‘net’ a. red b. ret c. red ®"d.re0 e. red Tree- * Vless APPROX *1 *VD FRIC * 1 * 1 *! ~ * T Ident [voice] [A0 ]/*vd *5 Ident stop [cont] Tableaux (19-21) show how the interaction among [+son, +cont]/coda and [-son, -cont]/coda with the rest of the constraints proposed for Peninsular Spanish in section (3) captures the pronunciation of /d/, /t/, /0/ and /nd/ outside coda position. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 249 Tableau (19) shows how intervocalic /d/ surfaces as [5]. Candidate (e) is optimal, since, unlike its competitors, it violates only low-ranked constraints. Tableau 19: Intervocalic /d/ /ad a/ *V less *VD *0]c *[-son, Id e n t [A0]/*vo *6 Ident ‘fairy’ APPROX FRIC -cont]]„ [voice] i stop [cont] a. ada *! b. ata c. ada d. a0a *; i i : — — ®"e. ada f. a9a __ * ---------------- - J - - J - The analysis proposed also captures the pronunciation of III as [t] intervocalically (Tableau 20). Candidate (b), which does not violate any constraints, is selected as optimal. Tableau 20: Intervocalic III /rata/ ‘rat’ ♦VLESS FRIC *VD FRIC *S]„ i ‘ [-son, ! -cont]n Id en t [voice] [A0 ]/ *Vd stop *5 Ident [cont] a. rada *! * '" ’b.rata c. rada .....* d. raOa e. rada * . * f. ra0a *! I .......-............» .................- * * The same is valid for /0/, which is always pronounced as a fricative. In tableau 21, candidate (d), which does not violate any constraint, is selected as optimal. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 250 Tableau 21: Intervocalic /0/ /raGa/ ‘race’ a. rada *VLESS APPROX *VOI FRIC *5]„ : *[-son, : -cont]c Id e n t [voice] #5 [A0]/*V d Stop *5 Ident [con] * b. rata *! c. rada % ' ®"d. ra0a e. rada *! * f. raGa *1 ' : After a nasal, Id/ is pronounced as a voiced stop (Tableau 22). Candidate (a), which violates low-ranked *V o ic e d S t o p, is selected as optimal. Tableau 22: /nd/ /anda/ ‘w alk!’ *VLESS APPROX *VD FRIC * 6 ]0 I *[-son, I -cont]„ Ide n t [voi] [Ao]/*Vd stop *5 VD Stop ®"a. anda ; * b. anta ; *! ■ c. anda *! \ d. anGa i ■ . e. anda ; *1 f. anGa ■ * * This analysis captures the phonemic distinction between lb, d, g/ and /p, t, k/, and between /p, t, k/ and /f, 0, x/. It also captures spirantization after open segments. Last but not least, this account captures the preference of coda frication in Basque Spanish, through constraints penalizing voiced approximants and voiceless obstruents in coda. To conclude this section, a ranking lattice for Basque Spanish is given below. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 251 (18) Ranking lattice for Basque Spanish * V oiceless a ppr o x im a n t * in terdental L a teral *[-son, -cont]/Coda *[+son, +cont, -lat, -rhotic]/Coda LPA Ident-IO [voice] * Voiced fricative [A 0 ]/*voiced stop [+son, +cont, -lat, -rhotic] \ Ident-IO [continuant] *Voiced Stop Future detailed investigation of coda realization in the Central dialect will determine whether coda devoicing applies in this dialect or not. If the pervasive pattern is that of speakers 8 and 9, where voiceless stops are frequent realizations of coda fb, d, g/, a dispreference against coda voiced approximants and against voiced fricatives will capture it. If the pervasive pattern is that of speaker 7, where coda /b, d, g/ are predominantly realized as voiced fricatives, Ident-IO [voice] will outrank the constraint penalizing voiced fricatives; additionally, there is also a dispreference against coda voiced approximants. Future investigation will determine which analysis is correct. 6. Conclusion This chapter has examined the realization of coda lb, d, g/ in two dialects of NC Spanish: the ‘Northern dialect’ or Basque Spanish, and the Central dialect. Frication Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 252 is pervasive and close to categorical in the Northern dialect, but not very common in the Central dialect. Stress has an aerodynamic effect on coda /b, d, g/; it affects the likelihood of frication of these codas through the increase in airflow that accompanies stressed syllables. The effect of stress on frication is variable; it varies within and across speakers. Stress has no effect on voicing. The phonetic and phonological implications of these findings have also been considered. Since the effect of stress on frication of /b, d, g/ is variable, it has been argued that it should not be encoded in the grammar. By contrast, since frication in the Northern dialect is pervasive and close to categorical, it has been argued to be included in the grammar. Frication in coda position has been captured through constraints militating against coda approximants and stops that override the tendency for postvocalic underlying voiced stops to be realized as voiced approximants (or stops). Devoicing in this position follows from the contradictory requirements for voicing and frication, expressed as a high-ranked constraint against voiced fricatives. This consonantal process, which is stress-conditioned and phonetic in nature, contrasts with the foot-sensitive, phonological consonantal processes that occur in Panoan, and that are analyzed in the following chapter. This chapter has also argued against a phonological coda devoicing analysis for the process yielding [$, 0, x] as allophones of coda fb, d, g/ in the Northern dialect. Devoicing is treated as a by-product of frication, i.e., as an aerodynamic effect, not a phonological process. For the Central dialect, the results are inconclusive. Two patterns are found: voiceless stops (speakers 8 and 9) and voiced Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 253 fricatives (speaker 7). The first pattern suggests coda devoicing, and the second faithfulness to the underlying voicing specification. Future experiments with further subjects from the Central dialect will help to clarify the role of coda devoicing in the Central dialect. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 254 C hapter 5: R hythm icity in Panoan 1. Introduction In this chapter I investigate consonantal processes involving in Panoan languages. As in North-Central Spanish, discussed in the previous chapter, Panoan languages show consonantal phenomena affecting laryngeal properties in coda position. In contrast with North-Central Spanish, Panoan shows consonantal phenomena that are rhythmic, foot-conditioned, and categorical. This chapter focuses on glottal stop deletion and metathesis in Capanahua, Huariapano [h] epenthesis, and alternate mora timing in Shipibo. All of these languages belong to the Panoan family and are or were spoken in Peru. According to the Ethnologue (Grimes (2000)), Capanahua and Shipibo belong to the North-Central branch of Panoan; Huariapano, also known as Wariapano or Panobo, is unclassified. Of these languages, Shipibo is the most widely spoken; it has about 30,000 speakers (Grimes (2000)). Capanahua is endangered; there were about 400 speakers in 1969 but only about thirty per cent of these remained by 1998 (Loos (1969), Loos and Loos (1998)). According to Jose Alberto Elias (p.c.), who did fieldwork on this language, in 2001 there were only thirty people who could speak it and some additional people who could understand it. Huariapano has been extinct since 1991 (Parker (1994), Grimes (2000)). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 255 The rhythmic alternations in Capanahua and Huariapano were previewed in chapter 3 as cases of strictly foot-sensitive consonantal processes; Capanahua exemplifies a system where not all feet are stressed, and Huariapano a system where stress and foot structure do not line up. The alternations in Capanahua and Huariapano have been previously described as involving odd/even alternations. According to Loos (1969), a coda 111 in Capanahua is pronounced in odd-numbered syllables counting from the beginning of the word (la), but it deletes in even- numbered syllables (lb). (1) Capanahua /?/ alternation (from Shell (1975:39)) (a) /hono-ta?-ki/ ['ho.no. ta?. ki] ‘It is a wild pig’ (b) /hono-ma-ta?-ki/ ['ho.no.ma. ta. ki] ‘It isn’t a wild pig’ [h] epenthesis in Huariapano occurs in the coda of open odd-numbered syllables only (Parker (1994)). One example is found in the progressive suffix /-ini/. When the second syllable of this suffix occurs in even-numbered syllables epenthesis does not occur (2a). If it occurs in an odd-numbered syllable, [h] is epenthesized (2b). (2) Huariapano [h] alternation (from Parker (1998:29)) (a) [pi.'m.kaj] ‘they are eating’ (b) [jjpm.naj.'nih.kaj] ‘they are looking, searching’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 256 Other Panoan languages are reported to have phonological processes conditioned by odd/even alternations, including Shipibo vowel harmony and suffix alternations; Amahuaca plosive nasal release in even-numbered syllables; and Sharanawa, where /s/ verb affixes are lost in odd-numbered syllables (Lauriault (1948); Loos (1973, 1999)). In all of these cases, the odd/even alternation is true of the data. However, the question arises why a phonological process should be based on ‘mere’ counting, and what counting reflects. As Parker (1998:31) puts it, generalizations relying on odd/even number counting have no explanatory force and go against the cross- linguistic evidence that languages cannot count beyond two. Another question is why so many of these processes should occur in the Panoan family. Recent investigations in Panoan languages propose that the apparent odd/even conditioning for the processes mentioned above is actually foot-based or rhythmic-based. These include Safir (1979) and Gonzalez (2002a) for Capanahua, Elias (2000) for Shipibo, and Parker (1994, 1998) for Huariapano. In this chapter I explore in detail the foot and rhythmic bases for these alternations. I argue that glottal stop deletion in Capanahua is foot-related, following the proposals in Safir (1979) and Gonzalez (2002a). My analysis takes into consideration stress opacity and related rhythmic processes in the language. I fully investigate foot structure in Capanahua and show how the distribution of coda /?/ correlates with weak and strong positions in a foot. Capanahua contrasts with Huariapano, which also has a rhythmically conditioned consonantal process, [h] epenthesis proceeds left-to-right in the strong Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 257 syllables of syllabic trochees. In some cases [h] epenthesis and stress assignment coincide; in other cases they do not, since [h] epenthesis and main stress are assigned from different edges, and secondary stress, while typically assigned left-to-right, can also be right-to-left. This causes ‘mismatches’ in the patterns of [h] epenthesis and stress assignment and apparent irregularity in the distribution of coda [h] epenthesis regarding metrical structure results. One possible solution to this problematic distribution is that Huariapano has two metrical tiers; one for stress assignment, and another for [h] epenthesis. This is the solution proposed by Parker (1998). I propose to view the phenomenon in a different light. My account centers on the observation that [h] epenthesis mimics the default pattern for secondary stress, i.e., left-to-right syllabic trochees. I propose that, as in Capanahua, rhythmicity is preferentially achieved by the occurrence of [h] in strong footed syllables, not by stress. In general secondary stress and [h] epenthesis will coincide; the cases where they do not coincide are explained by the preference for rhythmicity to be segmentally realized rather than realized by stress. The chapter is organized into four parts. Section 2 discusses the basic phonology and metrical system of Reconstructed Panoan, since the stress and tonal system of Reconstructed Panoan sheds light on the metrical systems of the Panoan languages under investigation. Section 3 discusses in detail /?/ deletion in Capanahua. Section 4 discusses [h] epenthesis in Huariapano. Section 5 discusses the metrical system and suffix alternations in Shipibo. Finally, section 6 is the conclusion. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2. The Phonology of Reconstructed Panoan 258 Shell (1975) presents a reconstruction of early Panoan phonology based on the comparison of the phoneme inventory and cognate words of seven Panoan languages spoken in Peru: Capanahua, Shipibo-Conibo, Amahuaca, Chacobo, Cashibo, Cashinawa, and Marinahua. Her results are contrasted with data from additional Panoan languages. This reconstruction is useful to understand the present-day differences among Panoan languages. A consonant chart for Reconstructed Panoan is provided in table 5.1. Note that 1*71 was a phoneme and there was no /*h/. This detail will be important in the discussion of the metrical system of Huariapano. Table 5.1. Consonant chart of Reconstructed: Panoan (IPA; from She 1 1 (1975)) Bilabial Alveolar Post- Alveolar Retroflex Velar Glottal Stop *p *t *k *kw *? Fricative * J 3 *s *g Affricate *ts *tj Flap Nasal *m *n Approximant *w *j Syllables were of the type (C) V. There was only one vowel per nucleus. Consonant clusters—involving 1*71, /*s/, /*j7 and /*§/—occurred exclusively word-medially. In such cases, syllable boundaries might have been between the consonants in the cluster, or they could derive from the loss of a vowel among them from an earlier Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 259 stage in the language. This second possibility is in accordance with the development of some present day Panoan languages, where consonant clusters have developed out of the loss of a vowel in two-morpheme words (Shell (1975:93)). Most words in Reconstructed Panoan had two or three syllables. Four- syllable words existed but they were probably formed of more than one morpheme. There was a phonemic tonal contrast between high and low tone; most probably high tone correlated with strong stress. Two-syllable words fell into two different tonal patterns. In the first pattern, the second syllable had high tone in all cases. In the second pattern, both syllables had low tone if the word was pronounced in isolation, and high tone in the first syllable and low tone in the second syllable otherwise (Shell (1975:94-5)). This contrast was lost in Capanahua, Shipibo-Conibo and Cashinawa, where disyllabic words have high tone or strong stress in the first syllable and contrasting tone or stress in the second syllable. Reconstructed Panoan also had three-syllable words. In the development of Panoan languages, mono-morphemic three-syllable words were generally reduced to two syllables.1 In Capanahua, Shipibo-Conibo and Cashinawa such words have high tone or strong stress in the second syllable (Shell (1975:97)). As shown in (3), the last vowel was lost, causing the onset to re-syllabify as the coda o f the preceding syllable (3a, b) or to delete (3c). As will be shown later, this development explains some apparent exceptions to the stress pattern in Capanahua and Shipibo. 1 Except in Chacobo and, in some cases, Marinahua. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 260 (3) Panoan: from three-syllable to two syllable words (from Shell (1975:96)) Reconstructed Panoan Capanahua Gloss (a) * ?a.wi.m [?a.'win] ‘wife’ (b) *[?]o?.pos.grn [?o?.'pog] ‘chigoe (tropical flea)’ (c) * ra.bui.ta [ra.'btu] ‘two’ 3. Capanahua This section discusses /?/ deletion in Capanahua. (3.1) presents an overview of Capanahuan phonology including its basic stress pattern. (3.2) provides a detailed analysis of the metrical system of the language in Optimality Theoretic terms. This is essential to show how /?/ deletion coincides with the metrical structure of the language. Section (3.3) discusses the distribution of /?/ and rhythmic processes in coda, including deletion and metathesis, and analyzes these in Optimality Theoretical terms. Section (3.4) considers a separate process in which [?] occurs as an allophone of /tj7, and section (3.5) discusses apparent counterexamples to a metrical account of /?/ coda deletion. Finally, (3.6) is the conclusion. 3.1 The phonology of Capanahua Capanahua has 16 consonants and 4 vowels: /a/, III, lol, ImJ. There is no contrastive vowel length (Loos (1969)). A consonant chart is included below; note that /?/ is a phoneme in the language. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 261 Table 5.2. Capanahua consonant c lart (IPA, based on Loos (1969)) Bilabial Alveolar Post- Alveolar Retroflex Velar Glottal Stop P t k 1 Fricative § s J § h Affricate ts tf Flap r Nasal m n Approximant w j Syllables are generally of the type (C)V(C). The structure CCV is found only in the adverb /ska/ ‘already’. The future modal l§a?n/ exemplifies the only case of the syllable structure CVCC (Shell (1975:40)); in the surface, this string is realized as CVC (see section 3.5 for discussion). In the surface forms of the language codas are restricted to the sibilants /s/, /J V , /§ » /, to the glottal stop, and to nasals (Shell (1975:41)). Other consonants which would be syllabified syllable-fmally are deleted in the surface form. Nasals only surface as codas when they precede a non-continuant, as in [tsi.'pon.ki] ‘downriver’. Nasal codas delete word-finally or before a continuant, causing concomitant nasalization of preceding and following strings of vowels and laryngeals. Some examples are /J3imanan/ ['J3i.ma.na] ‘face’ and /?ueran-?ue/ ['?u.e.ra.?u.e] ‘push it’ (Loos (1969:177-81)). Finally, coda 111 is special in that it only surfaces in odd-numbered syllables and deletes in even-numbered syllables. Thus, the word /ta?no/ ‘grub’ is pronounced ['ta?.no], but the word /J3itfi?/ ‘I grab’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 262 is pronounced ['j3i.tji]. The distribution of coda 1 11 in Capanahua is discussed in section (3.3). Capanahua has both a tonal and stress contrast. Both tone and stress coincide in citation forms (Loos (1969:187)). High and low tones are phonemic; stressed syllables have high tone. High tones are realized as mid word-finally. Since tone normally correlates with stress I abstract away from it; for more details see Loos (1969:187fi£). Only one stress per word is reported, in either the first or second syllable (Loos (1969)). Thus, there is a two-syllable window for stress assignment. Stress is assigned to a second heavy syllable, or else to the first syllable (4). Heavy syllables have underlyingly a post-vocalic word-final or a post-vocalic, pre-consonantal consonant. The only exception is /?/, which never makes a syllable heavy. (4) Main stress in Capanahua (a) [ruis] ‘ just’ /man/ [ma] ‘you (plural)’ (b) ['ma.po] ‘head’ ['kin.tfa] ‘bare’ (c) ['tsis.ti] ‘ashes’ ['mef.pi] ‘clumsy’ (d) ['tji.tji.ka] ‘knife’ [']3i.ma.na] ‘face’ (e) ['son.ta.ko] ‘girl’ ['ni?.]3o.na.wu] ‘walk along’ (f) [hi.'sis] ‘ant’ []3a.'kof] ‘foam’ (g) [tsi.'ponJd] ‘downriver’ [?o.'nan.na.wu] ‘learn’ (h) /rajiitjV [ra.']3i] ‘two’ /mapop/ [ma.'po] ‘clay’ (i) /waran/ [wa.'ra] ‘squash’ (j) [his.'mis] ‘looker’ /pijTcap/ [pif.'ka] ‘small’ (k) [nuig.'ning] ‘brownbird’ (1 ) /pit]!?/ ['JBi.tji] ‘I grab’ /raka?ti/ ['ra.ka.ti] ‘he lies down’ (m) /pi-tfa?tfikin/ ['pi.tla.tfi.lu] ‘to poke him in the ribs’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 263 Monosyllables, which are pronouns and clitics, never carry stress (4a). In words with two syllables or more, stress falls on the initial syllable whenever the peninitial syllable lacks a coda (4b-e). When the peninitial syllable has a coda, this syllable is stressed (4f-g). Stress is sensitive to the occurrence of underlying postvocalic consonants even if these are not pronounced in the surface (4h-i). If both the initial and peninitial syllable have a coda, stress falls on the peninitial syllable (4j-k). Finally, /?/ is exceptional; even if it is postvocalic underlyingly, the syllable containing it does not attract stress (41-m). 3.2 The metrical system of Capanahua In this section I propose an analysis of the metrical system of Capanahua. This analysis is formalized with Optimality Theoretic constraints (Prince and Smolensky (1993)). A preview of this analysis appears in (5). (5) Preview: The metrical system of Capanahua (a) One and only one stress per word. (b) Left-to-right footing. (c) Exhaustive footing. (d) Syllables can be unparsed. (e) Trochaic rhythm. (f) Quantity-sensitivity. (g) Stress opacity. (h) Initial syllable light. (a) One and only stress per word: The words in (4) show that regardless of the number of syllables within a word, there is always one and only stress per word. The Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 264 only exception is monosyllables, which are always unstressed. The constraints needed to capture these facts are defined below: (6) Grwd=Prwd A grammatical word must a prosodic word (Kager (1999)) Stress Prom Maximize stress prominence ‘Only one stress per word’ *Head/g Foot heads are not unstressed Leftmost Align (Hd-Ft, Left, PrWd, Left) ‘The left edge of the head foot coincides with the left edge of some prosodic word’ (McCarthy & Prince (1993)) Monosyllables in Capanahua are clitics and pronouns (Loos (1969:193)). Function words are usually exempt from having a stress, presumably because they cliticize to a phonological word (Hayes (1995)). For this reason, Grwd=Prwd will be assumed to apply to content words only, at least in Capanahua. I propose the constraint Stress Prom to capture the fact that in certain languages there is only one stress per prosodic word (Halle and Vergnaud (1987), Hewitt (1992), Van de Vijver (1998)). Some examples are Creek and Modem Standard Arabic, where only the main foot is stressed. In these two languages, stress can only be derived if binary feet are built from the left edge and the rightmost foot is stressed; otherwise, the wrong syllable is stressed (Van de Vijver (1998:36-42) and references therein). Stress Prom conflicts with * Head/6, which disallows unstressed foot heads (Chapter 3). Stress Prom is undominated in Capanahua, since no word has more Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 265 than one stress. Stress Prom outranks *Head/6. Grwd=Prwd is also undominated. The ranking of L eftm ost is established later. (b) Left-to-Right footing: Two facts point to left-to-right footing in Capanahua. First, stress is assigned either to the initial or the peninitial syllable in a word, no matter how many syllables the word has: [hi.'sis] ‘ant’, fson.ta.ko] ‘girl’, ['ni?.J3o.na.wu] ‘walk along’. If footing was right-to-left, in a three-syllable word with all light syllables—like ftji.tji.ka] ‘knife’—the two final syllables would be footed together and stress would fall on the second syllable. The result would be unattested *[tji.('tji.ka)]. This shows footing is left-to-right rather than right-to-left. Secondly, 111 deletion occurs in alternate even-numbered syllables from the beginning of the word. Consider the word /?otjiti- ma- ra?- ta?- ki/ ‘It is probably not a dog’. This word has two underlying syllable-final glottal stops, of which only one surfaces: ['?o .tji. ti.ma. ra?. ta.ki]. Assuming glottal stop deletion is foot- sensitive, there are two possible directions of parsing; left-to-right, and right-to-left If feet are constructed left-to-right, deletion of the glottal stop in the morpheme /ta?/ corresponds to the second syllable of a foot: [(?o .tji.) (ti.ma.) (ra?.ta.) ki]. If feet are constructed right-to-left, deletion corresponds to the first syllable of a foot: [?o. (tji.ti.) (ma.ra?.) (ta.ki)]. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 266 Two pieces of evidence argue against right-to-left footing. First, under right- to-left parsing the initial syllable of the word remains unparsed: [?o.(tji.ti.) (ma.ra?.) (ta.ki)]. Since the initial syllable is stressed in this word, this means that it cannot be unparsed. Second, right-to-left footing does not provide a unique generalization for glottal stop deletion. Under right-to-left footing, glottal stop deletion occurs both in the first syllable of a foot [?o.(tJi.ti.) (ma.ra?.) (ta.ki)] and in the second syllable of a foot: /fhtfi?/ [(’ jpi.tjl)] ‘I grab’. Left-to-right footing captures the stress pattern of the language and the correct generalization for 11 1 deletion; it occurs only in the second syllable of a foot: [(?o .tji.) (ti.ma.) (ra?.ta.) ki] ‘It is probably not a dog’, /fiitfi?/ |~('j3i.tfi)] ‘I grab’. Assuming footing is exhaustive in Capanahua (see (c) below), these two facts are easily captured through the interaction of the following constraints: (7) A ll-Ft-Left Align (Ft, Left, PrWd, Left) ‘Every foot stands at the left edge of the prosodic word’ (McCarthy and Prince (1993); Kager (1999)) All-Ft-Right Align (Ft, Right, PrWd, Right) ‘Every foot stands at the right edge of the prosodic word’ (McCarthy and Prince (1993); Kager (1999)) I assume that footing can be assigned in both directions in the same language; however, in this case, the preferred direction is left-to-right. This means the constraint All-Ft-Left has priority over All-Ft-Right (Tableau 1). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 267 Tableau 1: A l l-Ft-Left» A ll-Ft-Right /Pana-ipi-ki/ ‘planted it yesterday’ All-Ft-Left All-Ft-Right a. ('J3a.na)(?i.pi) ki s } : * * *** b. J3a.('na.?i.)(pi.ki) * * * * f 9 ** (c) Exhaustive footing: Despite the reported lack of secondary stresses, I claim that Capanahua has exhaustive footing. One piece of evidence is the rhythmic distribution of coda /?/. Coda /?/ deletes in even-numbered syllables beyond the first two syllables. This situation is parallel to other languages such as Norton Sound Yupik, where consonantal phenomena occur in a foot-sensitive fashion in spite of the fact that not all feet have stress. Thus, the fact that /?/ deletion occurs in even- numbered syllables is compatible with an analysis where syllables are footed in binary groups and deletion occurs on the second syllable of each foot. Further, it has been shown that languages can only count up to two. An analysis where syllables are grouped into feet fares better conceptually than an analysis where deletion occurs in the second, fourth, and sixth positions, because it limits counting to two. Elaborating on the last point, I assume that syllables are organized into binary feet in Capanahua. Binarity can be achieved in a syllabic or in moraic sense; feet must have either two syllables, or two moras. The undominated constraint Foot Binarity (henceforth FtBin) expresses this fact (Prince (1980), Kager (1989), Prince & Smolensky (1993)). (8) FtB in Feet are binary under moraic or syllabic analysis Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 268 (d) Syllables might be left unparsed'. Only when syllables cannot be part of a binary foot might they be left unparsed. Specific instances include light monosyllabic words, initial syllables followed by a heavy syllable—as in /hismis/ [his.('mis)] ‘looker’, /hisis/ [h|.('sis)] ‘ant’—and final stranded syllables— as in /sontako/ [('son.ta).ko] ‘girl’. Parse Syllable captures the tendency to parse syllables into feet (9) (Prince and Smolensky (1993)). Tableau 2 shows that FtBin » P arse Syllable; syllables are parsed into feet as long as FtBin is respected. (9) Parse Syllable Syllables are parsed by feet Tableau 2: Foot B inarity» Parse Syllable /tjitfika/ ‘knife’ Foot Binarity Parse Syllable a. ('tfi.tfi.) ka * b. ('tfi.tfi.) (ka) *! Parse Syllable outranks All-Ft-Left. It is better to leave syllables unparsed than to have all feet aligned with the left edge of the prosodic word (Tableau 3). Tableau 3: Parse-Syllable » A ll-Ft-Left /ni?J3onawu/ ‘walk along’ Parse-Syllable A ll-Ft-Left a. fni?.J3o) (na.wu) ** b. ('ni?.j3o) na.wu - ‘ In candidate (a), the rightmost foot is misaligned from the left edge by two syllables. Candidate (b) is perfect as regards All-Ft-Left, but two syllables are left unparsed, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 269 so this candidate is dispreferred. So far, GrWd=PrWd, Stress Prom, Ft- Bin» P arse Syllable» A ll-Ft-Left» All-Ft-Right . (e) Trochaic rhythm. Rhythm in Capanahua is trochaic for the following two reasons. First, stress is assigned to the initial syllable unless the peninitial syllable is heavy. Second, /?/ deletion occurs in the second syllable within a foot starting from the left edge of the word, and it persists to the end of the word. This means that binary feet are formed where the first syllable of the foot is the head. Trochaic footing in Capanahua explains both main stress assignment and /?/ distribution. Stress falls on the head syllable of the main foot, as in [('ma.po)] ‘head’, [('tji.tji.) ka] ‘knife’, and [('tsis.ti)] ‘ashes’. /?/ deletion occurs in the ‘complement’ or weak syllable of a foot, as in /JMtJi?/ [('J3i.tJi)] T grab’. Undominated Rhtype=Trochaic captures trochaic footing (10). (10) Rhtype=T Feet have initial prominence (Kager (1999)) (f) Quantity-sensitivity. It has been observed that the coda content of the second syllable is relevant for main stress assignment. A second syllable with an underlying postvocalic consonant is assigned stress even if the consonant is not pronounced. This suggests that at least for main stress assignment codas are moraic. The only exceptional consonant is 111, which is never moraic. If it were, a word like /|3itji?/ ['jlitji] would be stressed on the second syllable, which has a Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 270 postvocalic 111 underlyingly. Plausibly, /?/ is the only non-moraic consonant in Capanahua because it is among the least sonorous consonants. Moraicity is related to sonority; vowels, the most sonorous segments, are moraic in most languages. Moraicity of coda consonants is captured through W e ig h t-B y -P o s itio n (Hayes (1989), Sherer (1994)). The fact that heavy syllables attract stress is captured through Weight-to-Stress (Prince (1983), Prince and Smolensky (1993)). The constraint *Moraic /?/ expresses the fact that /?/ is never moraic (Gonzalez (2002a)). (11) Weight-By-Position (WBP) Coda consonants are moraic Weight-to-Stress (WSP) Heavy syllables are stressed *Moraic [?] The glottal stop is non-moraic Motivation for *Moraic [?] comes from the fact that [?] is the lowest sonority segment; and the lower the sonority of a segment, the less able it is to support a mora (Zee (1994, 1995); Blevins (to appear)). The fact that /?/ is non-moraic is probably related to the deletability and insertability of /?/ in the language (see section 3.3 for details). *Moraic /?/ is undominated and outranks WBP, since /?/ is non-moraic. WSP is dominated either because not all heavy syllables are stressed. WSP outranks Parse-syllable (Tableau 4); it is better to leave a syllable unparsed (candidate (a)) than not to assign stress to a heavy syllable (candidate (c)). WSP also outranks Leftmost (candidates (a, c) in Tableau 5). The rankings between Parse and Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 271 Leftmost and between WSP and WBP have not been established yet. The moraic content of the initial syllable is irrelevant for stress assignment, since the initial syllable is always light. This is discussed in more detail in (h) below. Tableau 4: WSP » Parse /hisis/ ‘ant’ WSP WBP Parse a. hi. ('sis) 1 F * b. ('hi.sis) *(0 *(!) c. (' hi. sis) I F Tableau 5: W S P » Leftmost /hisis/ ‘ant’ j WSP WBP Leftmost a. hi.('sis) 1 F _ , , b. ('hi.sis) * 0 ) * (0 c. (' hi. sis) F ; (g) Stress Opacity: Connected with quantity-sensitivity is the fact that only some syllable-final consonants are pronounced in the surface. These include /s, j, §/, which are unrestricted; /m, n/, which occur only before non-continuants, and /?/, which is found in unfooted or strong syllables. All other coda consonants— Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 272 /w, j, ts, tj, p, t, k/—delete or are never found in coda position.2 (12) exemplifies cases where syllable-final /w/ and /tj/ delete. Examples of other coda consonants are found in (4). Deletion of coda /tj/ is analyzed in detail in section (3.4). (12) Coda condition (Loos (1969:167)) (a) /haw+jama/ [ho.ja.ma] ‘not come’ (b) /rajpitj/ [ra.'jM] ‘two’ In spite of deletion, underlying postvocalic consonants make a syllable heavy and attract stress (12a, b). Previous descriptions of the language state that stress is assigned before the deletion of coda consonants (Loos (1969)). I will refer to this as the rule-based approach. In derivational terms, the fact that a coda consonant attracts stress even if deleted in the surface produces opacity. (13) provides a rule-based derivation for the minimal pair [ma.'po] ‘clay’ and ['ma.po] ‘head’, which differ only in the location of stress. The derived word ['ma.po.pan] ‘clay (subject)’ is also included since it provides evidence for a final underlying /p/ in [ma.'po] ‘clay’. (13) Derivation: [ma.'po] ‘clay’, ['ma.po.pan] ‘clay’, and ['ma.po] ‘head’ UR /mapop/ /mapopan/ /mapo/ Stress Assignment ma('pop) ('mapo) pan ('mapo) Coda Condition ma('po) ('mapo) pan ____ SR ma('po) ('mapo) pan ('mapo) 2 [h] is sometimes found in coda since sibilants optionally debuccalize before [r]: /hisis- ri?bi/ ‘ant+also’ [hi.'sih.ri?.J3i] (Loos (1969:185)). Thus, I consider [h] as a possible coda in Capanahua. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 273 Since stress can be predicted from the (underlying) coda consonant, stress is not lexical. Within Optimality Theory, there are various possibilities to account for opaque cases, including Sympathy Theory (McCarthy (1998, 1999)), Output-Output correspondence, and Turbidity (Goldrick (2000) and Goldrick and Smolensky (2000)). I propose to account for opacity in stress assignment in Capanahua with a Turbid approach. According to the theory of turbid representations, outputs can be complex, which m ean s that it can contains both pronounced and unpronounced material. For example, an input like /mapop/ in Capanahua maps to the output [m a 'p o ], where the last syllable is footed in apparent violation of footing structure. However, this output has a covert or ‘turbid’ structure, where unpronounced material influences the pronunciation of other parts of the output (Goldrick (2000)). The idea is that there is a single but complex output representation rather than intermediate stages in a derivation. Turbid Representations include two types of output associations: projection, and pronunciation. Projection is a structural or abstract relation between a segment/feature in the output with respect to its input. Pronunciation is an output relation describing the surface realization of the structure. Segments not pronounced but present in the output are enclosed within angle brackets <>— as in containment theory (Prince & Smolensky (1993)).3 Pronounced output segments lack angle 3 The representation of projection and pronunciation in this chapter differs from Goldrick (2000), where projection is represented with an upward-pointing arrow 7 1 and pronunciation with a downward-pointing arrow X I. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 274 brackets. Additionally, moraic consonants are indicated with the symbol ‘p.’.4 Opaque cases such as (14) need to be accounted for. The relevant constraints needed to account for such opaque cases are given in (15). (14) Opaque case: is not pronounced but projects a mora /ma'pop/ -> [ma. 'po)] _______________ P (15) Pronounce-ji All moras must be Pronounced (Goldrick (2000)) *[-cont, -nasal]-p [-continuant, -nasal] moras are not Pronounced *[+cont, +son]-p [+continuant, +sonorant] moras are not Pronounced Project WBP: All codas must Project their own mora (based on Goldrick (2000), Hayes (1989), Sherer (1984)) Reciprocity xY {R xY) If Y projects to X, then X must pronounce Y (Goldrick (2000)) Project WBP replaces WBP in this discussion, since they are equivalent. Project WBP enforces coda moraicity. *Moraic /?/ outranks P roject WBP since coda /?/ is never moraic. Pronounce-p enforces the pronunciation of all moras. Reciprocity X Y aims towards the pronunciation of projected material. Since some moraic consonants are not pronounced in the surface, both of these constraints are dominated. 4 I assume that vowels project moras in Capanahua, but I won’t represent them in the outputs for simplicity. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 275 The constraints *[-cont, -nasal]-p and *[+cont, +son]-p account for the coda restriction in Capanahua. *[-cont, -nasal]-p penalizes coda /p, t, k, ts, tjV , and *[+cont, +son]-p, coda /J 3 , r, j, w/. The consonants that can surface in coda position—nasals, sibilants, and /?/—are all moraic except for /?/. The pronunciation of /?/ in the surface is not an issue even if /?/ is [-cont, -nasal]; /?/ is non moraic and has its own set of constraints regulating its distribution. The coda restriction in Capanahua is reminiscent of the allophonic alternation of coda /b, d, g/ in Basque Spanish, where both stops and consonantal approximant allophones are disallowed in coda position in favor of voiceless fricatives (see chapter 4). *[+cont, +son]-p and *[-cont, -nasal]-p are undominated. These constraints outrank Pronounce-p since only a subset of moras is pronounced. Tableaux for /hisis/ ‘ant’ and /mapop/ ‘clay’ are included below. In Tableau 6, coda /s/ is pronounced in /hisis/ since it is not affected by *[+cont, +son]-p or *[-cont, -nasal] - p. In Tableau 7, the final /p/ in /mapop/ is not pronounced because it violates *[-cont, -nasal]-p. However, /p/ projects a mora and attracts stress. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 276 Tableau 6: /hisis/ ‘ant’ /hisis/ ‘ant’ FT B IN *[-cont, - nasal]-p *[+cont, +son]-p Max-Io Project WBP Pron- F S x Y ^ a . hi.('sis) F b. hi.('si<s>) F *(!) *(0 c. hi.('si<s>) d. hi.('sis) e. hi.('si) f. ('hi.sis) * f * f *! *i - * * g. ('hi.si<s>) F *0) *(0 h. ('hi.si<s>) i lllll i. ('hi.si) ||; |1 A comparison among candidates (a, c, f, g, i) in Tableau 7 shows that *[-cont, -nas]-p M ax-IO, WSP and Project WBP dominate Pronounce-p and i^ Y . So far there is no direct evidence about the relative ranking of M ax-Io, WSP and P ro ject WBP, so I assume that they are not ranked between them. Undominated FtB in has been included for comparison. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 277 Tableau 7: /mapop/ ‘clay’ /mapop/ ‘clay’ Ft- B IN * [-cont -nas]p *[+cont, +son] p M ax- Io WSP Proj Wbp P ro n- M - BXY ®"a.ma.('po) 1 u % * b.ma.('po) * c. ma.('pop) 1 p d. ma.('pop) & e. ma.('po) f. ('ma.po) *! g.('ma.po) * 1 ..... J h.('ma.po) 1 P *( * ■ i. ('ma.pop) In Tableau 7, candidate (a) is optimal; it has an unpronounced mora-projecting coda and violates low-ranked PRONOUNCE-p and Reciprocity. Candidates (b, d, e) fall out under undominated FtBin. Candidate (c) violates *[+cont, +son]-p. Candidate (f) shifts the stress to the first syllable, parsing the two syllables into a foot and deleting the coda consonant. This violates M ax-IO. Finally, candidates (g, i) violate Project-WBP since they have non-moraic coda consonants, and candidate (h) violates WSP because a heavy syllable is not stressed. (h) Initial syllable light. On discussing the best metrical structure for Capanahua, Safir (1979:104) considers one option where the initial syllable is stipulated to be Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 278 always light. I argue that this is the case, as supported by two different pieces of evidence. First, monosyllables are never stressed and lack high pitch, even if they have a coda consonant. Examples include [rras] ‘ just’, [ra?] ‘perhaps’, and /ham/ [ha] ‘he’. Second, in words with both first and second heavy syllables, the second syllable is stressed rather than the first (Loos (1969)). Examples include [mran.'tsis] ‘nail’, [piJ7ka] ‘small’ and [his.'mis] ‘looker’. This means that in words like /sontako/ ‘girl’ and /tsisti/ ‘ashes’ the first syllable is parsed as part of a two- syllable foot: [('son.ta).ko], [('tsis.ti)]. However, in words like [rras] ‘ just’ and /hismis/ ‘looker’, the first syllable remains unparsed: [rras] ‘ just’, [his.('mis)]. If this is correct, for main stress Capanahua builds moraic trochees of the form ('LL) or ('H) (16). (16) Capanahua foot structure (a) [rras] ‘ just’ /man/ [ma] ‘you (plural)’ (b) [('ma.po)] ‘head’ [('kin.tja)] ‘bare’ (c) K'tji-tji.) ka] ‘knife’ [('son.ta.) ko] ‘girl’ (d) [hi.('sis)] ‘ant’ [?o.('nan.)(na.wu)] Team’ (e) /raJpitjV [ra.('pi)] ‘two’ /mapop/ [ma.('po)] ‘clay’ (f) [his.('mis)] ‘looker’ /pijkap/ [pij.('ka)] ‘small’ (g) / m ^ i c m m T grab’ /raka?ti/ [(’ra.ka.) ti] ‘he lies down’ (h) /pi-tja?tjikin/ [('pi.tja.)(tfi.kl)] ‘to poke him in the ribs’ In order to capture the light status of initial syllables in Capanahua I propose the constraint Initial Light. While this constraint seems to be specific to Capanahua, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 279 future research might reveal its application to other languages or its wider motivation. Initial Light and WSP dominate WBP, as tableau (8) shows. (17) Initial Light Initial syllables are light Tableau 8: Initial heavy light » W B P /rras/ ‘ just’ Initial Light WSP WBP ^ a. rras * b. rras 1 M - c. ('rras) i *! i i i i i i a i i s I a ' In tableau 9, candidate (b) is selected because both syllables are parsed together into a disyllabic foot, respecting both FtBin and Parse. Tableau 9: Foot Binarity» Parse /tsisti/ ‘ashes’ Ft Bin Parse a. (’ tsis.) ti *! ^ b. (’ tsis.ti) So far, the ranking among stress and foot structure constraints is: G rW d=PrW d, Stress Prom, Rhtype=T, FtBin, *Moraic /?/, Initial Light » W B P » W S P , Leftmost, Parse» A ll-Ft-Left» A ll-Ft-Right, *Head/o. Below are summary tableaux for cases ('LL), fHL), L('H), H(H), and ('LL)L. For simplicity, the cover constraint ‘Stress’ subsumes undominated Initial Light, Grwd=Prwd, Stress Prom, Rhtype=T and FtB in. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 280 In disyllabic words with two light syllables, both syllables are footed together and the first syllable is stressed (Tableau 10). Candidate (a) does not violate any constraint and thus is selected as optimal. Tableau 10: ('LL) /mapo/ ‘head’ /mapo/ ‘head’ Stress WBP Leftmost Parse AFL ^ a. ('ma.po) b. (ma.'po) c. (’ma.) po d. ma. ('po) e. ma.po *! (Rh=T) *! (FtBin) *! (FtBin) *! (Gw=Pw) Disyllabic words with initial heavy and peninitial light syllables stress the first syllable. Both syllables form a foot, because of undominated Initial Light (Tableau 11). Candidates (a, b, d, e) violate undominated constraints. Candidates (c, f) tie on a violation of WBP; candidate (f) loses on a violation of Parse. Tableau 11: ('HL): /tsisti/ ‘ashes’ /tsisti/ ‘ashes’ a. ( tsis.) ti b. tsis. ('ti) c. ( tsis.ti) d. (tsis.'ti) e. tsis.ti f. ( tsis) ti Stress WBP Leftmost Parse *! (Initial Light) (FtBin) M(Rh=T) *i (Grw^Prw) AFL ate. - :v » ■ SliJeIL&SM Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 281 When a disyllabic word is formed of an initial light and a peninitial heavy, the second syllable is footed on its own and stressed; the first syllable remains unparsed (Tableau 12). Candidate (a) violates Leftmost, but it is optimal since the other candidates violate higher-ranked constraints. Tableau 12: L ('H): /hisis/ ‘ant’ /hisis/ ‘ant’ Stress WSP WBP Leftmost Parse AFL a. hi.('sis) i | ' * * ' .* . I F " b. ('hi.) sis *! (FtB) * • • • ; • * j c. (hi.'sis) *(Rh=T) * J d. hi.sis (Grw=Prw) * 1 l ' ” e. ('hi.sis) F *' ■ ■ f. ('hi.sis) *! [ ■ I ' ■ 1 Tableau (13) exemplifies a three-syllable word formed of three lights. The two first syllables are footed together, and the first syllable is stressed. Candidate (a) violates Parse but is selected over candidates (b-e), which violate higher-ranked constraints. Tableau 13: ('LL) L: /tjitjika/ ‘knife’ /tjitjika/ ‘knife’ Leftmost All Ft left a. ( tfi.tfi.) ka tji.( tji.ka) c. ( tji.) (.tfi.ka) ! (SP, FB) d. ( tji.) ( t j i . k a ) e. (tji.) ( tji.ka) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 282 Finally, tableau 14 shows the four-syllable word /?onannawu/ ‘learn5 . Candidates (b, d-f) violate undominated constraints. Candidate (c) violates high-ranked WBP, and candidate (d) violates high-ranked WSP. Candidate (a), with violates lower- ranked Parse and AFL, is selected as optimal. Tableau 14: /?onannawu/ ‘learn’ /Vonannawu/ ‘learn’ Stress WSP WBP Leftmos T Parse AFL ® *a. ?o.('nan).(na.wu) 1 p * * * ** b. (?o.'nan).(na.wu) *!(Rh=T) * • c. ('?o.nan).(na.wu) *! ** d. ('?o.nan).(na.wu) 1 p *! ** e. ?o.('nan.)(l na.wu) 1 F *! (Stress Prom) * ■ \ I. > f. ?o.nan.na.wu *! (Gr=Pr) * 3.4 The distribution of /?/ /?/ is a special segment in Capanahua. It can occur as an onset or a coda. As an onset it can be epenthesized in two different cases: between vowels at a morpheme juncture (18a-c) and before liquids at a sentence juncture (18d-e). Onset /?/ is also deleted word-medially in some cases (18f-i). /?/ is non-deletable in the remote past morpheme (18f, h), but it can be deleted in the recent past morpheme (18g, i). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 283 (18) Epenthesis and deletion of onset /?/ (from Loos (1969:176-7)) Epenthesis (a) /j3ana-i/ [J3a.na.?i] ‘planting’ (b) /pi-i/ [pi.?i] ‘eating’ (c) /flana-ipi-ki/ [J3a.na.?i.pi.ki] ‘planted it yesterday’ (d) /jkna-wui/ [?J3a.na.wm] ‘plant’ (e) /ra?maJ3i +rxnra-wui/ [?ra?.ma.J3i. rui.ra.wui] ‘chop now’ Deletion (f) /mapuit-?ogki/ [ma.'pui.?o§.ki] ‘he ascended (remote past)’ (g) /mapuit-?a-§-ki/ ['ma.pui.tag.ki] ‘he ascended (recent past)’ (h) /hamak-?ogki/ [ha.'ma.?og.ki] ‘he stepped (remote past)’ (i) /hamak-?a-g-ki/ ['ha.ma.kag.ki] ‘he stepped (recent past)’ /?/ is also deleted in specific coda positions. The description given by Loos (1969) states that coda [?] surfaces in odd-numbered syllables, including monosyllables (19a-f), but that it deletes in even-numbered syllables (19g-h). (19) Distribution of coda /?/ (from Loos (1969)) (a) /ta?/ [ta?] Declarative modal (b) /ra?/ [ra?] ‘Probably’ (c) /ta?no/ r'ta?.no] ‘Grub’ ( d ) /?i?sap/ f?i? 'sa] ‘Bird’ (e) /?onan-i?-ki/ ['?o.na.ni?.ki] 1 2 3 4 5 6 ‘He knows’ (f) /?oga-ka?ina-i?-ki/ [' ?o.§a.ka.?i.ni?.ki] ‘He falls asleep’ (g) /Jpitj-i?/ ['km T grab’ ( h ) /raka?ti/ 1 2 3 fra.ka.ti] ‘he lies down’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 284 12 3 4 (i) /p i-tja?tjik in / ['pi.tfa.tji.kin] ‘to poke him in the ribs’ 1 2 3 4 5 0) /hono-ma-ta?-ki/ ['ho.no.ma.ta.ki] ‘it is not a wild pig’ A number of alternations show that /?/ is underlying. The examples below involve the declarative modal /-ta?-/, and the adverb /-ra?-/. In (20a), /ta?/ occurs in the fifth syllable of a word; /?/ remains. However, when this suffix occurs in the sixth syllable of a word (20b), /?/ is not pronounced. The same thing happens with /ra?/.5 (20) Alternations: /ta?/ declarative modal and /ra?/ ‘probably’ 1 2 3 4 5 6 (a) ['?o .tfi. ti -ra- ta?-.ki] ‘It is probably a dog’ 1 2 3 4 5 6 7 (b) ['?o.tji. ti ,ma- ra? -ta-.ki] ‘It is probably not a dog’ The distribution of coda /?/ cannot be due to epenthesis in odd-numbered syllables since open odd-numbered syllables are tolerated, as in /mapo/ ['ma.po] ‘head’, /tjitjik a / ['tfi.tji.ka] ‘knife’. It was stated above that syllable counting is not desirable since crosslinguistically languages do not show sensitivity to numeric count beyond two and I'll coda deletion proceeds further than the two first syllables of the word. I proposed that syllables are organized into feet in Capanahua, and that the generalization that coda [?] is allowed only in odd-numbered syllables is an 5 An additional alternation is shown in example (1) in this chapter. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 285 epiphenomenon of foot structure. More specifically, 1 11 deletes in the second syllable o f a foot. Since Capanahua has trochees, the position where coda 11 1 deletes corresponds to a weak footed syllable. Coda 1 1 1 is pronounced elsewhere, i.e., in strong footed syllables and in unfooted syllables. Examples (10, 11) are restated below: (21) Distribution of 1 1 1 (a) /ta?/ [ta?] declarative modal (b) /ra?/ [ra?] ‘probably’ (c) /?i?sap/ [21L Csa)] ‘bird’ (d) /ta?no/ [('ta?.no)] ‘grub’ (e) /?onani?ki/ [('?o.na.) (ni?.ki)] ‘he knows’ (0 /?o§aka?inai?ki/ r('?o.sa.)(ka.?i.)(ni?.ki)] ‘He falls asleep’ (g) /jJitji?/ [('Pi.tfi)] ‘I grab’ (h) /raka?ti/ [Y'ra.ka.) ti] ‘he lies down’ (i) [('?o. tfi.) (ti. ra.) (ta?. ki)] ‘It is probably a dog’ (D [('?o. tji.) (ti. ma.) (ra?. ta.) ki] ‘It is probably not a dog’ I propose to treat deletion of coda 111 in weak footed syllables as a prominence- reducing process that aims at maintaining a rhythmic contrast between the two syllabic components of a foot. The head of a foot can have 111 as coda, but its complement cannot. This is an instance of the general tendency of languages to show rhythmicity. Additional examples were presented in chapters 2, 3. My proposal is that coda [?] deletion achieves rhythmicity, making weak syllables weaker in Capanahua. Furthermore, I propose that coda [?] deletion is a Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 286 rhythmically-oriented process which, in the absence o f secondary stresses, helps to organize syllables into strong and weak pairs. As it was observed in chapter 2, prominence- and rhythmic-based consonantal phenomena occur in numerous languages. In some cases they reinforce metrical structure of the language. In this case, /?/ deletion makes up for the lack of secondary stresses in the word. [?] remains in unfooted syllables because the strong/weak contrast is not relevant in that position (lOa-c). A related process in Capanahua that shows the effect of rhythmicity is metathesis of /?/ from onset to coda position. /?/ Metathesis occurs in odd-numbered syllables, especially in fast speech (22a) (Loos and Loos (1998)). If the syllable already has a coda /?/, onset /?/ deletes (22b). (22) Onset-to-coda metathesis (Loos and Loos (1998:21, 22); IP A) (a) /ketsin-?ino/ [ke.('tsl.)( l?.no)] ‘painted tiger’ (b) /toa?i?ka/ [('to.a.)(i?.ka)] ‘mattress’ Interestingly, metathesis ‘deletes’ the onset to produce a coda, a structure which is cross-linguistically dispreferred. However, onset-to-coda metathesis creates a contrast between the two syllables of a foot that would otherwise have equal strength. The occurrence of both /?/ deletion in weak syllables and metathesis in strong syllables strongly suggest that the nature of both processes is rhythmic. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 287 Safir (1979) was the first to propose a foot structure analysis for coda 1 1 1 deletion in Capanahua, whereby 1 11 deletion applied in the weak position of a binary (trochaic) rhythmic group. I take this proposal a step further and claim that 1 1 1 deletion and onset-to-coda metathesis tend towards the same goal: the distinction between strong and weak syllables in a language where only one stress per word is reported. Thus, this process is not only metrically conditioned; it also has a metrical motivation. Furthermore, I claim that metrical structure can be present in a language in the absence of stress with segmental indicators of rhythm. In the case of Capanahua and the languages considered in chapter 2, these indicators are consonantal, and can arise through very different processes. Next I outline an analysis in Optimality Theoretical terms of both 1 1 1 deletion and onset-to-coda movement. For /?/ deletion, I propose that a force aiming towards making the weak syllable weaker takes precedence over deletion o f 111 (23). (23) Max-IO (No deletion) Every element of the input has a correspondent in the Output (McCarthy & Prince (1995)) No Coda (* C ]c t ) Syllables are open (Prince & Smolensky (1993)) *Weak/CVC Weak syllables do not have codas (chapter 3) *Weak/CVC is a near mirror image of Stress-to-Weight ‘If stressed, then heavy’ (Myers (1987), Riad (1992)); in other words, ‘the strong get stronger’. While Stress-to-Weight enforces heaviness of stressed syllables, *Weak/CVC enforces Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 288 weakness of non-stressed syllables, i.e. ‘the weak get weaker’.6 Weak syllables are in this case weak footed syllables. Unfooted syllables are neither weak nor strong because they do not contrast rhythmically with any other syllable. Additionally, the ranking *W eak/CVC» No Coda is a case of a specific constraint ranked over a general constraint. The ranking *Weak/CVC, *Moraic [?] » M a x - io » N o Coda captures /?/ deletion in weak footed syllables (tableaux 15-17). A comparison between candidates (a) and (b) in Tableaux 15, 16 shows that Max-Io » N o Coda. Similarly, a comparison between candidates (a) and (b) and (a) and (c) in Tableau 17 shows, respectively, that *Weak/CVC » M ax-Io, and that * Moraic [ ? ] » Ma x-Io. Note that Max-Io dominates Project-WBP; [?] does not delete unless forced by *Weak/CVC, even if it violates Project-WBP (Tableau 18). Tableau 15: [?] remains in strong footed syllable /ta?no/ ‘grub’ * Weak/C VC * Moraic [?] Ma x-io No Coda ___ a. ('ta?.no) b. ('ta.no) c. (' ta?) no 1 P * Tableau 16: [?] remains in unfooted syllable /ta?/ declarative modal *WEAK/CVC * Moraic [?] Max-io No Coda a. ta? b. ta *! c. (' ta?) I *i * 1 n 61 thank Adam Ussishkin for bringing this fact to my attention. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 289 Tableau 17: [?] deletes in weak footed syllable /jJitji?/ ‘I grab’ *Weak/CVC * M oraic [?] Max-io No Coda a. OjM.tJi) $ b. OJJi.tJi?) * c. J3i.(' tji?) * 1 * Tableau 18: Max-Io» P roject-WBP /ra?/ ‘perhaps’ Max-IO Project-WBP ^ a. ra? * ■ b. ra *! *Weak/CVC and * Moraic [?] are undominated, since they are always respected. The ranking *Weak/CVC » M ax-Io won’t make all codas in the language delete, since apart from /?/, the rest of the codas in the language are moraic and will make their syllable heavy. /?/ is thus deleted because it is the only weightless coda. The distribution of /?/ deletion cannot be attributed to positional faithfulness to stressed syllables (Beckman (1998)), since there is only one main stress per word and /?/ can surface in codas beyond the second syllable in the word. It cannot be attributed to positional faithfulness to the strong or first syllable in a foot, since coda /?/ occurs in unfooted syllables. The constraints just discussed account for the rhythmic deletion of /?/ in weak syllables. In order to derive onset-to-coda /?/ metathesis in strong footed positions *Head/CV is needed to enforce a rhythmic contrast between strong and weak syllables within the same foot (chapter 3). Rhythmic contrasts in Capanahua Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 290 aim at reducing prom inent weak syllables, and at enhancing strong syllables. Ill has to be available inside the foot to be deleted or m etathesized; epenthesis of I’ ll in coda is banned (24). (24) *Head/CV Head syllables are not open (chapter 3) Dep-IO (No Epen th esis) Every element of the output has a correspondent in the input (McCarthy & Prince (1995)) L in e a r it y -IO The output reflects the precedence structure of the input, and vice versa (McCarthy and Prince (1995)) (*[c.g.]...[c.g.])a A sequence of [e.g.] is prohibited within a syllable (Suzuki (1998:126)) *Head/CV enforces a prominence contrast between the strong (head) and weak (complement) syllables of a binary foot by way of the strong syllable having a coda. Since all consonants except for III are moraic and Capanahua is quantity-sensitive, the only consonant that respects both *H e a d /C V and WBP is III. D ep outranks *Head/CV. A contrast between strong and weak footed syllables is created by deletion of an underlying III in coda of a weak syllable, or by ‘moving’ an underlying III in onset to coda position in strong footed syllables, not by epenthesizing III in codas of all foot heads. Thus, D e p» *H e a d / C V » L in e a r it y (Tableaux 19, 20). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 291 Tableau 19: Onset-to-coda metathesis of [?] /k e t s in - ? in o / ‘p ainted tig er’ D ep *H e a d /C V ®°a. k e .('ts i< n > .)(i? .n o ) 1 1 (l u b. k e .('ts i< n > .)(? i.n o ) 1 F In tableau 19, candidate (a) is preferred over candidate (b) because the disyllabic foot formed at the end of the word expresses a rhythmic contrast between its syllables; this contrast is missing in (b). Candidate (b) fares better on L in e a r it y and No C o d a , but in this language, but expressing a rhythmic contrast within a foot has preference. Since 111 occurred in the input, no epenthesis or deletion have occurred, only a change in linearity. Tableau 20 shows that epenthesis of coda 11 1 is not permitted. Candidate (a), which violates *H e a d /C V is selected over candidate (b), which epenthesizes /?/. Tableau 20: Failure of epenthesis in binary feet /m a p o / ‘h ea d ’ D ep *H e a d /C V L in e a r it y M a x *CODA ■ ® ”a. ('m a .p o ) * . b. ('m a ? .p o ) '■ * (*[c.g.]...[c.g.])0 is a dissimilatory constraint proposed by Suzuki (1998) to account for deletion of 11 1 in Seri (Hokan: Mexico). This constraint enforces only one instance of 1 11 in Capanahua per syllable. In strong syllables with two 111, the glottal stop in onset deletes in order to respect *H e a d /C V (Tableau 21). It is predicted that Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 292 when a sequence of two 111 occurs in weak or unfooted syllables, the glottal stop surfaces in coda will delete to satisfy N oC o d a , sin c e * H e AD/C V is not relevant in unfooted positions. Unfortunately, no examples of this case have been found. Tableau 21: Onset-to-coda metathesis of [?] /to a ? i? k a / ‘m attress’ * c g ...c g ] a *H e a d / L in e a M a x *C ]a C V RITY «"a. (to .a .) (i? .k a ) * | * r * b .(to .a .) (? i? .k a ) * , ■ ' . " J;* c .(to .a .) (?i.ka) **i j ' 1 " 1 ;■1 The set of constraints in (24) do not replace the constraints that account for coda 1 1 1 deletion. Consider tableau (22). The set of constraints in (24) would select as optimal the candidate where no deletion applies (candidate b); rather, (a) is selected. Undominated * W e a k /C V C captures this case (Tableau 23). Tableau 22: Deletion of [?] (I) /b itji? / T grab ’ D ep *H e a d /C V L in e a r it y M a x *CODA © a. (’ bi.tji) * ■ * " b. ('b i.tji? ) * ■ * c. ('bi?.tji) ' (*) * * d. ('bi?.tji?) *! ; ■ s j i^sS ill (*) ■ . ** Tableau 23: Deletion of [?] (II) /bitji?/ ‘I grab ’ *W e a k /C V C D e p *H e a d /C V L in e a r it y M a x *C o d a a. ('b i.tji) I * ’ I - ' 1 | ’ b . (’b i.tj i? ) j ;• |‘ * c. ('bi?.tji) • ■ ■ \ (■ *)’ ' j * ■ *' d. ('bi?.tji?) *! 1 J ( * ) I ' . 1 . * * Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 293 A constraint penalizing [J3, r] in onsets outranks D e p, since 17/ is epenthesized in these cases (Tableau 24); this constraint is undominated. Tableau 24: *a [P, f] » D ep /pana-wra/ ‘plant’ M P u ] D e p a. ('?Pa.na.) wui * ■ . b. ('Pa.na.) win *1 • • ■ This section has shown that coda 17 1 achieves rhythmicity through deletion and onset-to-coda metathesis. It has also been shown that 17/ is epenthesized in certain situations. /?/ is unique in various ways. It is the only segment epenthesized before /p, r/ sentence-initially and the only segment that undergoes rhythmic deletion and metathesis. A question is why the 17/ is targeted rather than any other segment. It is plausible that 17 1 is targeted because it has no place features, or because it does not contribute to weight in Capanahua, i.e., because it is non-moraic (Cf. Parker (1998)). The epenthesis, deletion and movement facts where /?/ is involved suggests that both explanations are relevant. Usually, epenthetic consonants in languages are less marked, and tend to be coronal or laryngeal. So far I have provided an OT analysis of the foot and stress system of Capanahua, including opaque effects between stress and coda consonants. This analysis brings new insights to the metrical system of Capanahua and to the connection between foot structure and the distribution of 171. Capanahua exemplifies Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 294 a language where rhythmicity is brought about by consonantal alternations. Specifically, /?/ deletion and onset-to-coda metathesis enforce a contrast between strong and weak syllables within a foot. Furthermore, rhythmicity makes up for the lack of secondary stress in the language. To close this section, I provide a ranking lattice for Capanahua (25). For reasons of space, constraints *[-cont, -NASAL]-p and * [+CONT, +S0N]-p are subsumed under Coda condition, and constraints Rht=T, S tr ess Prom and GrWd=PrWd under S tress. (25) Capanahua: ranking lattice CODA CO *a [J3 , f] *WEAK/CVC * MORAIC [?] FtBIN Stress [*Cg...Cg]a IN LIGHT DEP-IO Linearity WSP MAX-IO PROJECT-WBP xY PRON-g NO CODA Leftm o st Pa rse A ll Ft Left A ll F t Rig h t, *H e a d/ c f 3.5 [?] as an allophone of /tj/ Unlike other non-continuant consonants in Capanahua, /tj/ does not always delete syllable-finally. In some occasions, it is pronounced as [?] (26). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 295 (26) Realization of / t j / (Loos (1969) and p.c.) (a) /nitj-i/ [('ni.tfi)] (b) /pe-nitj-i/ [flJe.ni.) tji] (c) /nitj-w e/ [('ni?.we)] ‘He walks’ ‘He walks on the surface’ ‘Walk!’ (d) /pe-nitj-ritjpi-i/ [Pe.(‘ nL)(fi?.pi)] ‘He walks on the surface again’ The morpheme ‘walk’ has /tj/ in the input, /tj/ is realized as an affricate in onsets (26a, b). (26c) shows that /tj/ has a [?] allophone in coda position; [?] is in strong footed position. Example (26d) is puzzling. The syllable with input /tj/ is stressed, which means that it forms its own foot; however, [?] does not occur. But [?] occurs as the allophonic realization of another input /tj/ in the third syllable of the same word. Under a rule-based approach which depends on an odd/even generalization, the distribution of [?] exemplified in (26c, d) is not problematic. Ill is simply not pronounced in coda of even-numbered syllables. Derivationally, after stress assignment, coda /tj/ turns into [?]. Then, [?] falls to the glottal stop deletion rule (27). (27) [ni?.we] ‘Walk!’, [J3e'ni. ri tj.JBi] ‘He walks on the surface again’ C o d a / t j/- > [?] C o d a [?] d e l e t io n (e v e n ) SR UR S t r e ss A s s ig n m e n t /nitj we/ 'nitj.we 'ni?.we ['ni?.we] /§e nitj fitj|3i / J3e.'nitj.ritj.J3i J3e 'ni?.ri?.pi J3e 'ni.ri?.pi [J3e 'ni.fi?.§i] Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 296 Under a foot structure analysis, the absence of [?] in strong positions is problematic. (26d) appears to support the claim that [?] distribution relates to an odd/even generalization rather than to foot structure (Loos (1969)). However, this only occurs when an input /tj/ is involved. Consider the suffix /ri?jJi/ ‘again’ (28), a lexicalized combination of [ritj] ‘yet’ and emphatic [|3i] (Loos, p.c.). (28) Suffix /ritj]3i/ [fi?.j3i] ‘again’ (a) /his + fitjfSi +we/ [his.(hi.) (Pi.we)] ‘Look again’ (b) /J 3 e +his +ritjJ3i+we/ [Pe. ('is.) (ri?.pi) we] ‘Look him in the face again’ (c) /his+ritjpi+ma+wi/ [his.('ri.)(Pi.ma.)wi] ‘make him take a second look’ (d) /his+ma+ritjpi+wi/ [('his .m a. )(ri?.pi.)wi] ‘show to him again’ (e) /nitj + ritjpi +we/ [ni?. (hi.) (pi.we)] ‘Walk again’ There appears to be a conflict between /tj/ mora projection and its pronunciation, /tj/ either projects a mora and attracts stress or it turns into [?]. Compare (28a, b) and (28c, d). In (28a, c), /tj/ is not pronounced but it projects a mora and attracts stress. In (28b, d), coda /tj/ is realized as non-moraic [?], and the syllable is part of a bigger foot. Finally, in (28e) two adjacent syllables have /tj/ in syllable-final position. The second syllable attracts stress; /tj/ projects a mora but is not pronounced. The first syllable realizes /tj/ as [?] and remains unparsed. I propose that stress and foot structure are responsible for this split, /tj/ is special in the consonant system of Capanahua in that it is the only supra-laryngeal Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 297 consonant that that has the potential to be non-moraic. /tj/ is moraic only when stress is relevant. A moraic /tj/ attracts stress. This happens in the second syllable of the word. As seen before, the second syllable has priority over the first for stress assignment in case the second syllable is heavy. There is a drive to parse all syllables in the word into feet. If /tj/ is moraic, parsing might leave a syllable unfooted. If there is an available syllable next to it, then the mora of /tj/ is not projected, but rather, /tj/ is reduced to [?], creating a contrast between the two syllables of the foot. The distribution of /tj/ is captured through the interaction between stress and foot structure constraints. Some tableaux are included below. Tableau (25) shows a word where /tj/ is pronounced as [?] in coda. Candidate (a) violates Project W bp. Candidates (b, c) violate high-ranked Initial light and *[-cont, -nasal]-p. Tableau 25: /tJ/-» [?] /nitj-we/ tf o 0 1 *Head/ Max Project Parse ‘walk!’ lig h t j -nasal]-p c v Wbp ®"a.(mr.we) b. ( m<tj>) we c. ( mtj.we) Tableau 26 shows a word where two underlying syllable-final /tj/ occur. The first is not pronounced in the surface but projects a mora, and the second is pronounced as [?]. Candidates (b) and (c) violate high-ranked *[-cont, -nasal]-ji and Max-Io, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 298 respectively. Candidate (d) violates *Head/CV. Candidate (a), which violates lower- ranked Project Wbp, is selected as optimal. Tableau 26: /pe-nitj-fitjpi-i/ ‘He walks on the surface again’ /j3e-nitj-fitfl3i-i/ INIT. L IG H T *[-cont, -nasal]-|j, *Head/ CV Max Wbp Par SE ® “a. j3e.(‘ni<tj>.)(ri?.j3i) ji * * b. Jte.('nitf.)(fi?.J3i) p * c. pe.('ni?.ri.) pi *! * ** d. ('pe.ni?.)(ri?.pi) * 1 * * ; ■ These examples suggest that there are three different sources for the glottal stop in Capanahua: underlying, epenthetic, and derived (Figure 5.1). Underlying /?/ is phonemic; it is present in the input, and it is pronounced everywhere except in coda of a weak syllable in a foot. Epenthetic I’ ll occurs before a vowel word-medially in some cases, and sentence initially before /J 3 , r/. It can also occurs through metathesis for rhythmic reasons. Finally, there is a derived [?] which is an allophone of [?]; it is found in coda of certain syllables. Recognizing that there are three sources for [?] and that nevertheless foot structure and stress condition them still captures the fact that [?] is never pronounced in weak footed syllables. A foot-structure analysis is superior to the descriptive generalization that all instances of coda [?] occur in odd- Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 299 numbered syllables; it makes clear the connection between foot structure and stress to coda deletion processes and [?] distribution. Figure 5.1. Sources of [?] in Capanahua /tj/ /?/ ffl [] P] -Onset - Coda - Coda - Onset - Coda Moraic, Non-moraic -Coda (weak footed a) own foot part of foot (strong/ unfooted o) - Onset - Coda (strong o) 3.5 Exceptional cases Sections 3.3, 3.4 have shown that /?/ distribution is consistent with a moraic trochee analysis of foot structure in Capanahua. This section examines potential counterexamples for this distribution: (i) suffixes that always keep coda /?/, and (ii) H(LL) cases with [?] in the third syllable. (i) Some suffixes always keep their glottal stop regardless of their position in the word (Loos (1969) & p.c.). This is problematic for both a foot structure and an odd/even alternation analysis, since /?/ would surface in both strong and weak syllables, or both in odd and even numbered positions. Examples of these suffixes are /§a?nJ future subjunctive and /gi?k/ future indicative. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 300 (29) Suffixes keeping [?]: /ga?n/ future subjunctive and /gi?k/future indicative (a) /§a?n/ [ga?] future subjunctive (b) /ka-tan-ga?n-wra/ [ka ('ta) (ga?.wtu)] ‘Go over there sometime soon’ (c) /wui-ra-ja-ga?n-wi/ [('wxu.ra) (ja.ga?) wl] ‘Push it sometime’ (d) /gi?k/ [gi?] future indicative (e) /bana-gi?k-i/ [('ba.na.) (gi?.ki)] ‘will plant’ (f) /bana-ma-gi?k-i/ [('ba.na.)(ma.gi?.) ki] ‘will cause to plant’ For the future subjunctive suffix, /?/ is kept in unfooted position (29a), in strong position (29b), and in weak position (29c). Coda Ini is not pronounced because the following syllable starts with a non-continuant; however, it nasalizes the preceding /?/ and the vowel. In the future subjunctive suffix, /?/ also occurs in unfooted, strong and weak syllables (29d-f). However, in this case Ik/ is resyllabified as the onset of the next syllable (29e, f). There are two possible ways to account for this distribution. One is that these suffixes are exceptional and form their own prosodic word. This is suggested by Loos (p.c.), who mentions that these suffixes seem to derive from a previous periphrastic construction in which they were auxiliaries. According to Loos, they still remain outside the prosodic word, plausibly because they maintain their stem status and start a new prosodic word. Another instance is /ja?pa/, ‘the one who has’; this suffix always maintains its glottal stop (Loos (1969)). The second alternative is to explain the status of these suffixes through opacity. As seen before, outputs in Capanahua contain covert structure, and the interaction between constraints enforcing deletion of segmental material and Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 301 constraints enforcing projection of moraic consonants generally resolves in mora projection of coda consonants. For the suffix /ga?n/, /n/ projects a mora, provided that /n/ is pre-consonantal or word-final. The output representation for this suffix would be [ga?<n>], As a consequence, /§a?a/ is a foot on its own. No matter what its position in the word is, this suffix is always heavy preceding a consonant-initial morpheme, and /?/ is always kept because /ga?n/ is always in a strong position. For [gi?<k>], /k/ resyllabifies as onset in examples (29d-f) above; before a consonant, Hd deletes and /?/ remains, as in ['has.ka.j3i ska 'ha.no.gi?.?ue] ‘May you be like this forever’ (Loos and Loos (1998:508)). The revised footing and output structure for the future suffixes is provided in (30). (30) Revised footing for /ga?n/ (a) /ga?n/ [ga?<n>] Future subjunctive (b) /ka tan ga?n wui/ fka fta <n>) (ga?<n>) wui] ‘go over there sometime soon’ (c) /wui ra ja ga?n wi/ [('wra.ra) ja (ga?<n>) wi] ‘push it sometime’ (d) /gi?k/ [gi?<k>] future indicative (e) /bana-gi?k-i/ [('ba.na.) (gi?<k>.) ki] ‘will plant’ (f) /bana-ma-gi?k-i/ [('ba.na.) ma.(gi?<k>.) ki] ‘will cause to plant’ Two facts argue against this alternative. First, this would not explain the suffix /ja?pa/, which also keeps its glottal stop. Second, for the future indicative suffix, /k/ has to be posited as moraic in spite of resyllabification with the following vowel (30e, f). Consequently, I assume the suffixes considered in this section are exceptional in that they are prosodically independent. However, the distribution of Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 302 coda /?/ in these cases is not problematic, since /?/ occurs in the first syllable of a foot, which is a strong position; thus, it does not contradict other facts in its distribution. (ii) Cases like the following would be expected to be found in Capanahua: Odd Even Odd Odd Even Odd (a) (H) (LL) (b) (H) (LL) ?]« ?]„ In both cases, a heavy syllable disrupts footing; it interrupts binary syllabic footing and makes footing restart. The moraic trochee analysis and the odd/even alternation analyses make different predictions for both cases. The odd/even alternation analysis predicts that coda /?/ will remain in (a) since it occurs in an odd-numbered syllable; it will delete in (b) since it occurs in an even-numbered syllable. The foot structure analysis predicts that /?/ should not occur in (a) since it appears in a weak syllable; it should appear in (b), since it occurs in a strong syllable. Cases that appear to be like (a) are shown in (31). (31) CVC L L (from Loos p.c. and Loos and Loos (1998))7 (a) his-nika?-bo [('his.ni.)(ka?.bo)] ‘those who see, seers’ (b) ('hog.ko.)(ro?.ti) ‘to fall, to trip’ (c) ('hen.ke.)(tsa?.pa) ‘which kind’ (d) ('han.po.)(ko?.ti) ‘to get full with food’ 7 Thanks to Eugene Loos and Beto Elias for providing these examples. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 303 (e) tana-nika?-bo [('ta.na.) (ni.ka.) bo] ‘those who track’ (f) taran-nika?-bo [ta.('ran.) (ka.bo)] ‘those who roll things’ In (31a-d), a closed syllable occurs followed by two light syllables, the second of which has a coda /?/; [?] surfaces. The derivational approach captures this fact due to the assumption that /?/ occurs in odd-numbered syllables. The moraic trochee account discussed in (3.2) fares as well, since under this account an initial closed syllable is always light. In (3 If), a second heavy is followed by two lights, the first of which loses its coda /?/. This goes against both a derivational and a moraic trochee account, since this is odd-numbered position, and since this is strong footed position. However, compare (3If) with (31e); one syllable has been deleted in (3If). A plausible explanation is that the deleted syllable in (3If) is present and footed in the output; I will leave this matter for further investigation. Further examples including heavy syllables in non-initial syllables would be needed to support the conclusion that the examples in (31) are not counterexamples to the distribution of the glottal stop for the derivational and moraic trochee account. These examples are not reported in the literature. 3.6 Conclusion The preceding sections have shown that 111 deletion in Capanahua is consistent with a foot structure analysis, whereby coda /?/ deletes in weak footed syllables in order Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 304 to make weak syllables weaker. Ill does not delete in unfooted syllables since the strong/weak contrast is not relevant in such positions. The fact that only 11 1 has this function is related to its placeless nature and the fact that it is not moraic. A foot structure analysis of 11 1 deletion in Capanahua has three main advantages over an odd/even alternation analysis. First of all, this analysis obtains both the distribution of coda 11 1 and the stress system of the language. Second, in combination with the turbidity approach, this analysis obtains the distribution of coda /tj- / and apparent counterexamples. Finally, this account brings explanation to this process; namely, glottal deletion has a rhythmic motivation. Ill deletion is fully regular in Capanahua. I propose that in the absence of secondary stresses in the language, 1 1 1 achieves a metrical effect by grouping syllables into weak and strong. I suggest that different Panoan languages achieve this effect through segment structure in different ways. In the remainder o f this chapter I investigate related rhythmic processes in Huariapano and Shipibo. 4. Huariapano Huariapano is a former Panoan language spoken in Peru. It is closely related to Shipibo. Huariapano has a process of coda [h] epenthesis which is in various ways a mirror image of coda 11 1 deletion in Capanahua. While laryngeal fricatives are epenthesized in Huariapano, laryngeal stops are deleted in Capanahua. Both Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 305 processes take place in coda position and are rhythmically conditioned. This section focuses on how exactly [h] epenthesis is rhythmically conditioned and how it contrasts with /?/ deletion in Capanahua. As previewed in chapter 3, it will be shown that [h] epenthesis is strictly conditioned by foot structure, and that footing and stress do not always line up in the language. The main sources for Huariapano data and description are Parker (1994, 1998). Parker (1994) proposes coda [h] epenthesis is a rhythmic process. Parker (1998) argues that, though rhythmic, coda [h] epenthesis does not directly relate to stress, and proposes the existence of two different metrical tiers in Huariapano: one for stress assignment, and another for [h] epenthesis. I argue that the distribution of coda [h] is the same as default secondary stress, and suggest that the apparent irregularity of the distribution of coda [h] regarding stress is due to conflicting ways of assigning secondary stress. The discussion brings the synchronic state of the language, but it also takes into account the historical development from Reconstructed Panoan and comparative evidence from Panoan. Section 4.1 discusses the phonological and stress system of Huariapano and provides an Optimality Theoretic account of the metrical system of the language. 4.2 presents the distribution of coda [h] epenthesis. My proposed analysis of coda [h] epenthesis is described in 4.3. Finally, section 4.4 focuses on historical and comparative evidence to bear from other Panoan languages. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 306 4. 1 Huariapano phonology Table 5.3 shows the consonant chart of Huariapano. The only difference with the consonant phonemes of Capanahua is the lack of 111. There are 4 vowels: /i, ui, a, o/. Table 5.3: Huariapano consonant chart (IPA, from Parker (1994:96)) Bilabial Alveolar Post- Alveolar Retroflex Velar Glottal Stop P t k Fricative P s J § fa Affricate ts tj Flap f Nasal m n Approximant w j The syllable structure is (C)V(C); glides can be the second part of an onset. Is, J, §, n, m, j, w, h/ can be codas. Syllable-final Ini optionally deletes, and can also nasalize a preceding vowel (Parker (1998:3)). The distribution of coda Ihl is predictable (see section 4.2). [?] occurs optionally in word final position after a vowel but it does not contribute to weight; one example is /pino/ [lpi.no]~['pi.no?] ‘hummingbird’ (Parker (1994:97)). Monosyllabic content words are stressed and must have a coda or a long vowel; ['paw] ‘sheTT, ['jan] ‘lake’, ['|3o:] ‘hair’, ['tji:] ‘fire’ (Parker (1998:3)). Long vowels occur only in monosyllabic words. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 307 Main stress generally falls on the penultimate syllable (32a-c). Final syllables with coda consonants are stressed (32d).8 This means that Huariapano forms quantity-sensitive, moraic trochees at the right edge of the word (Parker (1998:5)). This is the opposite edge for main stress as in Shipibo and Capanahua. (32) Huariapano main stress (from Parker (1998:2-5); IPA) (a) ['hi. wi] ‘branch, stick’ ['ta.po?] ‘cot’ (b) [’win.ti] ‘oar, paddle’ ['rtus.JJi] ‘rope’ (c) [ka.'no.ti] ‘bow (weapon)’ [ram.bo.'go.Jpo] ‘knees’ (d) [ja.'wif] ‘opossum’ [§a.'J3in] ‘bee’ There are exceptional forms where stress falls on a light final syllable. Such words form about a quarter of disyllabic nouns and adjectives (Parker (1994:98)). Some examples are [uj.'ta] ‘garbage’ and [jo.'J3ui] ‘witch’. In most cases verbal suffixes are extrametrical; even if their last syllable is heavy, it does not attract stress. For instance in /nika-panan/ [nih.ka.'pa.nan] ‘that you might hear’ main stress is attracted by the light penult rather than by the heavy final syllable. Secondary stress is left-headed and quantity-insensitive. This means that for secondary stress Huariapano forms syllabic trochees. There are opposing directions for secondary stress: left-to-right, and right-to-left (33). 8 Parker (1998:4) reports that final syllables with underlying coda consonants are stressed. However, the only consonant that can delete in the surface is Ini, and this deletion is optional, not obligatory (Parker (1998:3)). Thus, the right generalization appears to be that final codaful syllables are stressed, and that deletion of syllable-final Ini is a phonetic process. One further argument is that [h] epenthesis can occur in a syllable where a coda nasal has deleted: [jah.'ta] ‘afternoon; late’ (Parker (1998:29)) (see also section 5.3). Consequently, unlike in Capanahua, opacity-related constraints are not needed in the account of the metrical system of this language. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 308 (33) Secondary stress (from Parker (1998:6-10); IPA) (a) [(,ma.na.) ('paj.) ri] ‘I will wait’ (b) [(jo.mra.) (,ra.no.) ('gi.ki)] ‘he is going to hunt’ (c) [(,ha.pom.) (‘ bi.JJi)] ‘they’ (d) [(,ku.J3jaj.) pa. (“ gi.ki)] ‘I cooked’ (e) [('wa.nra.) ki. ('rag.) ki] ‘they have returned’ (f) [(jo.mra.) (,raj.[3a.) kan.('gi.ki)] ‘they hunted’ (g) [pis.(,ma.noh.) (,ko.no.) ('gi.ki)J ‘I forgot’ (h) [mi^^om.pi.) ('ra.ma)] ‘you (plural)’ Examples (33ab) are ambiguous as to the directionality and quantity-sensitivity of secondary stress. Examples (33c-d) show that secondary stress is quantity- insensitive. (33c) does not determine directionality, but (33d) does; secondary stress is assigned left-to-right; if it was assigned right-to-left, [J3jaj] would have been stressed. (33e-f) are crucial to establish left-to-right directionality of secondary stress. (33g-h) show that secondary stress can be assigned right-to-left from the main foot. Without counting ambiguous cases, left-to-right parsing of secondary stresses has a statistical frequency of 66% and right-to-left parsing of 34% (Parker (1998:9)). The analysis of the metrical system of Huariapano described below is based on Parker (1998:10-26), but departs in some respects from it. Only the main details of the analysis relevant for the discussion of [h] epenthesis have been included; I refer the reader to Parker (1998) for further details. In some occasions, the constraint names have been changed in order to make comparison easier with Capanahua. Undominated R h=Trochaic and GrWd=PrWd enforce left-oriented feet and stress in all grammatical words, respectively. The ranking Ft B i n » Parse Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 309 ensures binary footing and parsing of as many syllables as possible into feet, provided the foot formed is binary (Tableau 27). Tableau 27: FtB in» P arse /jawif/ ‘opossum’ Ft Bin Parse ^ a. ja.fw ij) : * b. (ja.)('wij) *! , The constraint Rightmost enforces main stress towards the right edge of the word (34).9 R ightm ost dominates A ll-F t-L e ft. This captures the conflicting directionality of main and secondary stress, ensuring that main stress surfaces at the rightmost edge, and that all other feet are built from the left edge (Tableau 28). (34) Rightmost Align (Hd-Ft, Right, PrWd, Right) ‘The right edge of the head foot coincides with the right edge of some prosodic word’ (McCarthy & Prince (1993)) Tableau 28: Rightmost » All-Ft-Left /haJ3ombi|3i/ ‘they’ Rightmost All-Ft-Left a. (,ha.|3om.)('bi.j3i) ** • b. ('ha.j3om.) bi.J3i **! ■ Words with right-to-left secondary stress assignment are specified in the input with the ranking All-Ft-Right» A ll-Ft-Left, overriding default secondary stress assignment from the left (Parker (1998)). Words with stress on the final light syllable, like [us.'ta] ‘garbage’ are also specified for stress in the input. Exceptional 9 M a in R ig h t in Parker (1998). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 310 verbal suffixes, as in /nika-panan/ [( nih.ka.'pa.nan] ‘that you might hear’, with no stress on a final heavy syllable are analyzed as having an extrametrical mora. I will assume that in all of these cases stress is specified in the input; the constraint Stress Identity, adapted from Parker (1998:20), will ensure in these three cases that the specified syllable in the input will keep its stress (or lack of stress). (35) Stress Identity If syllable o l is specified as [astress] in the input, then its Output correspondent a2 must be [astress]. Project WBP outranks Parse and Rightmost (Tableaux 29, 30).1 0 These two last constraints are not ranked with respect to each other. Project WBP is dominated by Stress Identity, since in extrametrical suffixes the coda consonant is not moraic (Tableau 31; the extrametrical mora of the suffix is represented as <n> in the input). Tableau 29: Project WBP » Parse /winti/ ‘oar’ Project WBP Parse a. ('win).ti 1 ji 1 • b. ('win.ti) *! |- Tableau 30: Project WBP » Rightmost /winti/ ‘oar’ Project WBP Rightmost a. ('win).ti P * . b. ('win.ti) 1 0 In Parker (1998), Peak Prominence: 'H >'L (Prince and Smolensky (1993)). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 311 Tableau 31: S tress Id en tity » P roject W BP /nika-pana<n>/ ‘that you might hear’ Stress Identity Project WBP ^ a. (,nih.ka.) ('pa.nan) . . * .• b. (,nih.ka.) pa.('nan) 1 p To capture quantity-sensitivity in the main foot I propose the relativized constraints W eigh t-to-Stress-M ain and R h-C ontour-M ain.1 1 (36) Weight-to-Stress-Main Heavy syllables are stressed in the main foot Rh-Contour-Main The main foot must end in a strong-weak contour at the moraic level While Weight-to-Stress enforces stress in heavy syllables (Prince (1983), Prince and Smolensky (1993)), Weight-to-Stress-Main enforces stress in heavy syllables in the main foot. Rh-Contour penalizes a foot ending in a strong-weak contour at the moraic level, i.e., syllabic trochees with the first syllable heavy, and iambs where both syllables are light (Kager (1993, 1995, 1999)). Together with RhType=T, Rh- Contour-Main enforces main feet composed of a heavy syllable, or of two light syllables. Undominated Weight-to-Stress-Main and Rh-Contour-Main capture the dichotomy between moraic trochees for main stress and syllabic trochees for secondary stress. Additionally, these constraints crucially capture one of the environments where [h] epenthesis does not apply (see section 4.3). 1 1 In order to capture quantity-sensitivity for main stress, Parker (1998) proposes relativizing Peak Prominence for main stress with the constraint Peak Prominence-main. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 312 The ranking so far is S tre ss Id en tity , R h=Trochaic, G rW d=PrW d, F t- Bin, W eight-to-Stress-M ain, Rh-C ontour-M ain » P r o j e c t WBP » R ightm ost, P arse » A ll-F t-L e ft » A ll-F t-R ig h t.1 2 Summary tableaux and a ranking lattice are given below. For simplicity, Ft-Form is used to refer to the undominated constraints. Tableau (32) shows a word where main stress is assigned on a penultimate heavy syllable and secondary stress falls on the leftmost syllable. Candidates (d-g) violate high-ranked Ft-Form or WBP. Candidate (c) loses on multiple violations of R ightm ost and Parse. At this point, candidates (a, b) tie. Candidate (a) is selected since candidate (b) has more violations of A ll-F eet-L eft. Tableau 32 /wanrakiragki/ ‘they have returned’ Ft- Form W BP Right most Parse AFL AFR ^*a. (jW a.nrn.) ki.('rag.)ki 1 * ** b. wa.^nui.ki.) ('rag.) ki F * ** * ***( 5 c. ('wa.nui.) ki. (,rag.)ki 1 F sfcsH* * * = .= d. (jWa.nuiOki.Crag.ki) 1 F * i (Rhcont main) * **y e. (I wa.nui.) ki.('rag.ki) *! * *** *** f. wa.nra.ki.rag.ki (Gw=Pw) g. (wa.,nux.) ki.('rag.)ki 1 .............. . . . . ........ ........................F......... *! (Rh=T) & S i* ** * *** 3 1 2 A n o th er u n d o m in ated co n strain t is A l l -F e e t-R ig h t » A l l-F e e t - L e f t fo r w o rd s th at h a v e n o n d efau lt secondary stress. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 313 Tableau 33 shows a word where the leftmost foot has two syllables with codas. Candidates (d-g) violate undominated constraints. Candidates (b, c) have multiple violations of Rightmost and Parse, so at this point candidate (a) wins. Tableau 33 /ku|3jaj]3agiki/ ‘I cooked’ Ft Form WBP Right most Parse AFL AFR a. (g kujp.jaj.)j3a.('gi.ki) 1 1 |! H * *** *** b. kuJ3.jaj.J3a.('gi.ki) *** *** ■ c. (I kuj3.jaj.)|3a.(l§i.ki) ! 1 a a * *** *** d. (I kujp.jaj.)(j3a.)('gi.ki) 1 1 M - P - (FtBin) ** *** ** *** e. ku§.jaj.§a.§i.ki (Gr=Pw) ***** f. (,ku(3.jaj.)(J3a.'gi.)ki 1 1 a a (Rh=T) * ■ * , ' ** * *** g. (1 kuJ3.jaj.)]3a.('gi.ki) .* *** (37) Ranking lattice for Huariapano (I) Rh=T GrW=PrW WSP-Main Rh-Contour-Main Ft-Bin Stress Ident I Project WBP Parse Rightmost All-Ft-Left A ll-Ft-Right Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 314 4. 2 Coda [h] epenthesis [fa] epenthesis occurs in codas of odd-numbered syllables before voiceless onsets. The quality of Pal varies depending on the nature of the previous vowel. It has the following allophones: [h] after /a/, [§] after /i/, [x] after /ml, and [w] after lot (Parker (1994:115)). Allophones of IhJ are not shown in the Huariapano data discussed in this chapter. Pal is phonemic only in word-initial position (38). P a! never occurs intervocalically or word-finally. (38) Phonemic /h/ (from Parker (1994:96-7)) (a) ['ha.na?] ‘tongue’ ['ka.na?] ‘macaw’ (b) ['a.no?] ‘paca rodent’ Epenthetic [h] occurs word-intemally in coda position before voiceless consonants (39, 40). Epenthesis does not occur initially if the syllable has main stress; cf. (39a, b) with (39c). Epenthesis does not occur if the syllable already has a coda (39d). (39) [h] epenthesis in initial syllables (Parker (1994:100-1, 108-9; 1998:26)) (a) [kmhu'ptun] ‘(I) open’ [pah.'tsa.kra] ‘(we) washed’ (b) [,J3ih.tsa.'kag.ki] ‘(they) laughed’ (c) fna.ka?] ‘flea’ ['pi.kxxi] ‘(he) ate’ (d) [J3og.'ka?] ‘head’ In non-initial syllables, stress or its lack thereof does not determine [h] epenthesis (40). Epenthesis occurs in unstressed syllables (40a), main stressed syllables (40b, c) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 315 and secondary stressed syllables (40c). Epenthesis does not occur before a voiced onset (cf. 40 d, e). [h] is never epenthesized in even-numbered syllables (41). (40) [h] epenthesis in non-initial syllables (from Parker (1994:101, 102)) (a) [ja.na.pah.'kwig] ‘(I) will help’ (b) [,13o.no.'gih.kaej] ‘(they) will take, carry’ (c) [jo.mui^rah.ka.'tih.kaej] ‘(they) hunted’ (d) ^paj.ri.'rah.kaj] ‘still; yet (they)’ (e) [.paj.ri.'ra.naj] ‘still; yet (they)’ (41) No epenthesis in even-numbered syllables (Parker (1994:102)) (a) [pj.to .ki.'rag.ki] ‘(it) came running’ (b) [,kaej.J3a.kan.‘ £i.ki] ‘(they) went’ [h] epenthesis is quite regular. Out of 115 morphemes, only 9 are exceptional and do not undergo epenthesis even if all conditions are met. In other words, [h] epenthesis has a statistical productivity of 93% (Parker (1998:30-1)). (42) shows cases where [h] epenthesis fails in the initial unstressed syllable before a voiceless consonant. (42) Exceptional morphemes (from Parker (1994:116)) (a) [kra.'pu.ki] *[kuxh.'pu.ki] ‘he/it closed’ (b) [tju.'gi.kiu] *[tfuh.'§i.kui] ‘he/it dried (up)’ (c) [hi.'ki.ki] *[hih.'ki.ki] ‘he/it entered, went in’ First I review the account developed by Parker (1994, 1998). Parker (1994, 1998) argues that /h/ is moraic based on the occurrence of [h] epenthesis in light unfooted Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 316 syllables, as in [kuih.('piun)] ‘I open’. If [h] is morale, the light syllable can be footed, and no syllables remain unparsed: [(kuih.)('ptxin)]. Parker (1998) concludes that epenthesis is rhythmic but not directly related to stress, since it can occur in main-stressed, secondary-stressed and unstressed syllables. Parker (1998) proposes the existence of both a rhythm and a stress tier to account for the apparent contradiction of a rhythmically-oriented, stress-insensitive process. The stress tier encodes main and secondary stress placement, while the rhythm tier encodes [h] epenthesis. These two tiers are not required to coincide (Parker (1998:36)). (43) exemplifies both tiers for the word [a,ri.]3ah.lkag.ki] ‘they repeated’. The stress tier is formed respecting the metrical constraints that capture stress and footing. The rhythm tier is formed through quantity-insensitive trochees starting from the left edge of the word. Epenthesis of [h] occurs in the first or strong syllable of every foot in this tier where coda [h] is admissible. (43) Stress and rhythm tiers (adapted from Parker (1998:35)) (a) Stress tier a.(,ri.ba.) ('kan.ki) (b) Rhythm tier (a.ri.) (bah.kan.) ki More specifically for (43), the main foot is created in the stress tier via a moraic trochee aligned with the right edge. Secondary stress in this word is assigned via a syllabic trochee constructed leftwards from the main foot. The rhythmic tier forms left-to-right syllabic trochees, [h] is epenthesized in the third syllable of the word, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 317 since this syllable is the head in this tier and requires a coda. Epenthesis does not apply to the first syllable because of the requirement that a coda [h] precedes a voiceless consonant. To formalize this idea, Parker (1998:34) proposes the constraints A l l -F t- L e f t-R h y t h m and H e a v y fo o t h e a d -R h y t h m (44). (44) A l l -F t -L e f t-R h y t h m Align (Foot, L, PrWd, L) ‘The le ft edge of every rhythmic foot must be aligned with the left edge of some prosodic word’ H e a v y fo o t h e a d -R h y t h m The head syllable of a rhythmic foot must be h ea v y A l l -F t-L e f t-R h y t h m en su res left-to-righ t fo o tin g o f the rh ythm tier and H e a v y fo o t h e a d -R h y t h m a ch ie v es rhythm ic [h] ep en th esis. T ogeth er, th ese con strain ts enforce [h] epenthesis in a left-to-right fashion regardless of stress. The failure of coda [h] epenthesis to apply in initial main-stressed syllables is explained through the constraint [*SG-cq & *SG-'o]. This conjoined constraint penalizes the occurrence of /h/ in initial syllables with main stress. To avoid deletion of [h] onsets in this same position, Parker proposes the constraint [Max-onset (sg) & Max-ai (sg)]; this penalizes deletion of underlying /h/ in initial main-stressed syllables. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 318 4.3 Proposal It is desirable that metrical processes occur in one tier only. One exclusive metrical tier enforces theoretical economy and restrictiveness. Further, as shown in chapter 2, many consonantal processes, including epenthesis and deletion, are sensitive to stress. An analysis in which a rhythmic consonantal process is totally unrelated to stress (or foot structure) misses a connection with consonantal processes in other languages which show similar conditioning contexts. My analysis involves only one metrical tier and focuses on [h] epenthesis as a process where two separate metrical tendencies converge. The first is the creation of a rhythmic contrast between syllables within a foot, as in [(^ih.tsa.) ('kaq.) ki] ‘(they) laughed’. Epenthesis of [h] in strong footed syllables creates a contrast between the strong and weak members of the foot, in the same way as /?/ deletion in Capanahua. [h] epenthesis proceeds left-to-right and coincides with default secondary stress assignment. This shows that foot construction proceeds from the left edge. The second metrical tendency is stress assignment. Main stress is assigned to the opposite edge of the word and it can override left-to-right footing if there is a conflict. Further, some words have non default secondary stress. In these cases, I propose that secondary stresses are faithful to the input, and left-to-right footing still applies. Apparently, in both non-default secondary stress cases and in cases Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 319 involving extrametrical suffixes, [h] epenthesis and stress assignment conflict, with [h] epenthesis appearing to occur in weak footed syllables. I suggest that this conflict should be seen in a different light. Recall from chapter 3 that there is a cross-linguistic tendency for head syllables to be stressed and to have codas, and for weak footed syllables to be unstressed and to lack codas. The universal rankings *Head/o » * Head/'o, and *Head/CV » * H ead/CVC express the tendency for head syllables to be stressed and have codas. The universal rankings *Weak/'o » * W eak/ o, and * Weak/C V C » * W eak/CV capture the tendency for weak syllables to be unstressed and open. The apparent irregularity in the metrical system of Huariapano concerning [h] epenthesis can be viewed as a preference for rhythmicity to be realized segmentally rather than through stress. Together with the irregularity of the metrical pattern of some words in the system, this can lead to the violation of *Head/o and even *Weak/'c j in some cases. The violation of *Weak/'c t , though rare, is typologically predicted (chapter 3). In Huariapano, *Head/CV outranks *Head/c t , since it is more important to create rhythmic contrasts segmentally than through stress. *Head/CV also outranks Dep-IO, since epenthesis occurs in all binary feet where the first syllable lacks a coda and [h] epenthesis is permissible. *Head/CV is outranked by Parse, since epenthesis of [h] occurs in otherwise unfooted open syllables, which are thereby footed. These rankings will be shown in the following tableaux. In the following discussion, undominated *Clash, Weight-to-Stress-Main and Rh-Contour- Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 320 Main are incorporated into the cover constraint FtForm. The cases that need to be accounted for are listed in (45). Each will be illustrated with a tableau. (45) Epenthesis cases to account for Epenthesis Initial o before main stress: Secondary 'a: [kuih. 'puin] T open’ [^ih.tsa.'kan.ki] ‘they laughed’ Main ‘ a (extrametrical suffix): [^o.no.'sih.kajn] ‘they will take’ Non-default secondary stress: [JBis^ma.noh^ko.no.'si.ki] T forgot’ No epenthesis (i) Initial syllable before main stress (Tableau 34). In this case [h] epenthesis applies. Candidate (f) violates Ft-Bin and *Head/CV. Candidate (d) epenthesizes [h] and stresses the resulting foot; this violates *Clash. Candidates (b, c, e) violate WBP (additionally, for candidate (e) [h] epenthesis is not licensed by a [-voi] consonant; this is not shown in the tableau). Candidate (a) violates *HEAD/d since there is a foot head that is not stressed; however, this candidate is selected as optimal because the competitors violate higher-ranked constraints. This tableau shows that Parse » * H ead/g. Main stressed initial a: ['pi.kui] ‘(he) ate’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 321 Tableau 34 /krapttm/ ‘I open’ Ft Form WBP *Head/ c v *Head/ a Dep *Head/ CVC . ** ^ a. (kuih).('puin) P P * * b. kux.('pmn) *! * c. kuih.('puin) *, * d. (I kmh.)('puin) P P (*Clash) * f t e. (krah.)('puinh) p |i *! " *#r ** f. (kin.)('pran) * 1 (Ft-Bin) * * ■ * (ii) No epenthesis in main stressed syllables (tableau 35). Candidates (b, c) show epenthesis of [h] but violate R h-C ontour-M ain and WBP respectively; candidate (d) violates Parse. Candidate (a), which violates *Head/CV, is selected as optimal. This tableau shows that Parse outranks *Head/CV. Tableau 35 /pikui/ ‘(he) ate’ Ft Form WBP Parse *Head /CV *Head /& D ep *H/ CVC ® "a. ('pi.kui) * b. ('pih.ktu) 1 n *! (Rhcon main) * ■ % c. ('pih.kui) *! * • * d. ('pih.) kui 1 P * * Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 322 (iii) Epenthesis in secondary-stressed syllables (Tableau 36). Candidates (d, e) violate undominated constraints. Candidate (b, f) violate high-ranked WBP. Candidates (a, c) tie on a violation of Parse. Candidate (c) epenthesizes [h] in the weak syllable of the foot and violates *H/CV. Candidate (a) is selected as optimal. Tableau 36 /pitsakanki/ ‘they laughed’ Ft Form WBP Par SE *H/ CV *H/ 0 Dep *W CVC ar a. (,pih.tsa.)('kan.)ki 1 1 a a * * #* b. (,pi.tsa.)('kan.)ki 1 a *! ■ * * * c. (,pi.tsah.)('kan.)ki 1 1 a a * *! * ** d. pi.(,tsah).('kan).ki 1 1 a a *! (* Clash) ** * ■ ** e.(l Pih.tsa.)('kanh).ki a a a *! (* Ftbin) * **. ** f. (^ih.tsa.X'kan.ki) 1 a *! * ** (iv) Extrametrical suffixes. When the last syllable of the word is extrametrical, epenthesis applies to the preceding syllable (tableau 37). I assume that extrametrical suffixes are specified for stress in the input. The symbol <> in the input means that the syllable is specified for lack of stress. Candidates (b, c, e) violate undominated constraints. Candidate (b) violates the restriction against [h] epenthesis before voiced consonants. Candidate (c) violates Rh-Contour-M ain, because the main foot is Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 323 (HL). Candidate (e) stresses the final syllable and violates Id en t-str ess. Candidate (a), which has only one violation of WBP, is selected over candidate (d), which has two violations of WBP. Tableau 37 /|Jo.no.sih.<kajn>/ [-stress] ‘they will take, carry’ Ft Form WBP Parse *Head /CV *Head /a Dep a. (,po.no.) ('sih.) kajn 1 * * * - b. (J3oh.no.) ('sih.) kajn H v *! H-VOI * ' * c.(J3o.no.) ('sih. kajn) 1 r1 *! Rhcon- M AIN * * * d.(,J3o.no.) ('si. kajn) * * i * e.(J3o.no.) (sih.) ('kajn) *! Ident STR ESS ** * ; * - * (iv) Non-default secondary stress. The final case is words with right-to-left secondary stress, as [Jpis^ma.noh^ko.no.'si.ki] ‘I forgot’. In this case, secondary stress is realized as right-to-left, but left-to-right footing and epenthesis apply. This causes strong footed syllables to be unstressed and weak footed syllables to be stressed. I assume that secondary stress is specified in the input for this and other exceptional words. Candidate (b), which epenthesizes [h] in the second and third feet counting from the left, violates Rhtype-T, because the main foot has final prominence. Candidate (d) realizes stresses from the left as well as [h] epenthesis but violates Ident [stress], since this word is prespecified for secondary stress. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 324 Candidates (a, c) tie on a violation of P a r s e . In candidate (c) footing is assigned right-to-left, and as a result [h] epenthesis coincides with w ea k footed syllables. Since strong footed syllables have no codas, this violates * H e a d /C V more times than candidate (a), and thus (a) is selected as optimal. Note that candidate (a) violates * W e a k /'o . *W eak /'cj is dominated by *Head/CV; otherwise, candidate (a) would not be optimal. Tableau 37 /{fis^ma.no^ko.no.si.ki/ ‘I forgot’ Ft Form pa r se *H/ CV *w/ 'o *H/ d D ep *H/ CVC ® “a. (JJis^ma.Xnoh^ko.) (no.'si.) ki * *** * b. (pis^ma.Xnoh.jko.) (noh.'si.) ki *! R h t= T * * * c. Jiis.^ma.noh.) (,ko.noh.) ('si. ki) * ■ ■ ■ ** d. (jJpis.maO^noh.ko.) (,no.si.) ki Id e n t st r e s s * * This analysis of Huariapano [h] epenthesis and stress does not rely on the language- specific separation between a stress and a foot structure tier. Rather, it is based on the preference for rhythmicity contrasts to be realized segmentally rather than through stress. This is shown in cases where main stress and non-default secondary stress coincide with [h] epenthesis. An important tendency in the metrical system of Huariapano is parsing as many syllables into feet as possible, even if [h] has to be epenthesized, and even if Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 325 this violates *Head/6. In the following section I will propose a historical explanation for this tendency. A revised ranking lattice for Huariapano closes this section. (46) Ranking lattice for Huariapano (II) R h = T G r W = P r W W S P -M a in R h -C o n t o u r -M a in F t -B in S t r e s s Id e n t P ro ject W B P P a r s e R ig h t m o s t *H e a d /C V A l l -F t -L e f t *W e a k /'o *H e a d /g A l l -F t -R ig h t D e p *H e a d /C V C 4. 4 The relationship between secondary stress and [h] epenthesis This section discusses the connection between secondary stress and [h] epenthesis from a historical and comparative perspective. A summary of the facts of the metrical system of Huariapano is given in table 5.4. Table 5.4. Metrical system of Huariapano Main stress: Exceptions Moraic trochee, right edge of the word Antepenult/ final light stress (nouns) Extrametrical verbal suffixes Secondary stress: Syllabic trochees. Left-to-right: more common. (About 67%) Right-to-left: less common (About 34%) f¥| epenthesis: Exceptions Syllabic trochees, left to right. It does not occur in 12 verbs Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 326 Quantity sensitivity is im portant for main-stress assignment but not for secondary- stress assignment or [h] ep en th esis. Two directionality patterns exist with respect to secondary stress; left-to-right from the beginning of the word, and right-to-left from the main foot in the word. Coda /h/ epenthesis follows exactly the same quantity- insensitivity and directionality as the most widely attested pattern of secondary stress assignment. Finally, there are exceptions to the default assignment of main and secondary stress and [h] epenthesis. Coda [h] epenthesis has exactly the same environments as default secondary stress assignment; it occurs in alternate syllables from the beginning of the word. The differences are that epenthesis is licensed by a following voiceless onset, and that epenthesis does not occur in syllables with coda consonants. I hypothesize the following historical scenario for the development of the current metrical stage of the language. This scenario is supported by historical evidence from the metrical system of Proto-Panoan, and by comparative evidence from related Shipibo and Capanahua (Table 5.5). Table 5.5 Metrical system in Panoan Proto- Panoan Capanahua Shipibo Huariapano Tone Yes Yes No No Main Stress Leftmost? ? Leftmost M. trochee Leftmost M. trochee Rightmost M. trochee Secondary stress ? No Rightmost (Morphol.) Bidirectional Syl. trochee Laryngeal process Directionality Type ? 1 1 1 deletion Left to Right Trochee No [h] epenthesis Left to Right Syl. trochee Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 327 In an early stage of the language main stress was assigned at the leftmost edge (Shell (1975)). When tones were lost from the development of Proto-Panoan, [h] epenthesis developed in Huariapano from the left in order to create a rhythmic contrast between strong and weak syllables. Secondary stress might have developed at the same time or later; the lack of secondary stress in Capanahua, which has a related process of /?/ deletion, and the emergence of secondary stress in Shipibo for certain suffixes suggests that secondary stress developed out of [h] epenthesis. At some point main stress shifted to the right edge of the word in Huariapano. This caused the development of right-to-left secondary stress in a subset of words in the language, triggering the apparent mismatch between [h] epenthesis in words with right-to-left secondary stress. Capanahua, Shipibo and Huariapano have moraic trochees for main stress. However, while Capanahua and Shipibo build the main foot at the leftmost edge, Huariapano builds it at the rightmost edge. Both Shipibo and Huariapano show quantity-insensitive secondary stress at the opposite edge from main stress. In Shipibo, secondary stress is optional in words of four syllables or more with certain suffixes. In Huariapano, it is usually assigned left-to-right from the beginning of the word, but can also be assigned rlght-to-left from the main stressed foot. The distribution of /?/ in Capanahua suggests that footing is persistent in spite of the absence of secondary stresses. Coda /?/ is deleted precisely in weak footed positions; it is pronounced in strong footed syllables and in unfooted syllables. In Shipibo there is some evidence for persistent footing in the alternation of the suffix Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 328 [rij3i/ri|3a] ‘again’ and in morphological secondary stress (see section 5). Footing in Huariapano is clearly persistent since secondary stresses are assigned in all feet (excepting, obviously, main-stressed feet), [h] epenthesis takes place in both strong footed and otherwise unfooted syllables, [h] epenthesis parses syllables into feet and creates a contrast between strong and weak syllables. I hypothesize that the creation of a contrast between strong and weak syllables is the original function of [h] epenthesis; its parsing function derived through the development of main stress at the opposite edge of the word. Both Capanahua and Huariapano have laryngeal processes sensitive to rhythmic structure. In Huariapano this process is rendered obscure because the existence of conflicting stress patterns in the language. The pattern in Huariapano follows left-to-right secondary stress assignment everywhere, conflicting with right- to-left secondary stress and with main stress assignment in some cases; mainly, when there are extrametrical suffixes that cause stress to shift to the penultimate syllable even if the final syllable is heavy. [h] epenthesis scans the entire word, including the main stressed foot. This suggests that epenthesis generally applied from the left, as deletion of /?/ in Capanahua. This would explain two facts in the distribution of [h] epenthesis: the restriction to odd-numbered syllables, and the absence of [h] epenthesis in initial syllables with main stress, [h] epenthesis would not occur because no secondary stress is needed. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 329 Unfooted and strong footed syllables share a common property in both Capanahua and Huariapano. In Capanahua, these two types of positions resist deletion of coda 17 1 ; deletion occurs in weak footed syllables, and onset-to-coda metathesis occurs in strong footed syllables In Huariapano, these two types of syllables undergo [h] epenthesis. In both Capanahua and Huariapano, it is crucial to to create a contrast between strong and weak syllables. In Huariapano, this tendency is found side by side with a tendency for exhaustive parsing. Common to both languages is the resistance of weak syllables to be prominent. Finally, there is the question of whether 1 7 1 deletion and [h] epenthesis is significant or just coincidence. Note that Pal was not a phoneme in Proto-Panoan, but an subsequent development in some of the Panoan languages. It is possible that [h] was selected as an epenthetic segment in Huariapano because it was not phonemic in the language. Relevant to this question is the non-moraicity of 17 1 in Capanahua, and the moraicity of Pal in Huariapano. Since Pal is more sonorous than 1 7 1 it can be explained that Pal is moraic in Huariapano while 17 1 is not in Capanahua. But it is possible that the moraicity of Pal is relevant only in the most recent stage of Huariapano, provided the tendency to parse as many syllable into feet derived after the tendency of creating a contrast between strong and weak footed syllables. Coming back to the first question, it would be necessary to establish precisely how a coda 1 7 1 or /h/ affect the syllable where it occurs, namely, whether the previous vowel is shortened, lengthened, or devoiced; whether the overall length of the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 330 syllable is in creased or decreased; and so on. Phonetic studies might help to elucidate this question. 5. Shipibo This section describes the phonological system of Shipibo, focusing on its metrical system and on the rhythmic alternation of one of its suffixes. It will be shown that Shipibo expresses rhythmicity through vocalic alternations. Shipibo is closely related to both Capanahua and Huariapano. It has about 50% intelligibility with Capanahua in spite of different phonological and grammatical systems. It is not intelligible with Huariapano except for brief conversations (Shell (1975:25)). The data and the description of Shipibo in this section are taken from Elias (1999, 2000, 2001).1 3 Section 5.1 describes the metrical system of Shipibo, and section 5.2 the rhythmic alternation of the suffix [riba/ribi] ‘again’. 5. 1 The metrical system of Shipibo Shipibo and Capanahua have similar consonants and vowels (see Table 5.2). Both languages have the same syllable structure (C)V(C) and the same coda restriction: only nasals and sibilants can be codas. Most words in Shipibo have two or three syllables. Words with four or more syllables are either compounds or suffixed forms. 1 3 I thank Beto Elias for making available his work on Shipibo, and for his helpful comments on Panoan. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 331 As in Huariapano, monosyllabic words in Shipibo always have a long vowel and can also have a coda consonant. Examples include /hi/ [hi:] ‘hair’ and /kin/ [kxn'n] ‘desire’. Long vowels are not found elsewhere in the system (Elias (1999)). Unlike Capanahua, Shipibo lacks tone (Shell (1975:46-52)). Main stress is assigned to moraic trochees as near as possible to the left edge of the prosodic word (47). (47b) crucially shows that stress is left-oriented; /atapa/ ‘hen’ has three light syllables, and stress falls on the leftmost syllable. (47d) shows that stress is quantity- sensitive; if a second syllable is heavy, it is stressed. For main stress, Shipibo is similar to Capanahua. (47) Main stress in Shipibo (from Elias (2000)) (a) [‘ ti.ta] ‘mother’ (b) ['a.ta. pa] ‘hen’ (c) ['pis. bi] kind o f wasp (d) [tja.'ras] C a t a l a n (a bird) A difference with Capanahua is that in Shipibo secondary stress is optional in words of four syllables or longer, depending on the type of suffix added to the word. There are two types of suffixes: prosodic, and non-prosodic (cf. Elias (2000), type L /II classification). Prosodic suffixes—including /-'ju.ku/ diminutive', /-'run.ki/ reportative; and /-nin/ ergative—form a foot of their own and carry their own stress. Non-prosodic suffixes are not footed on their own and do not carry stress. Examples are /a; ki; ai/ finished action (interrogative/indicative/indicative-interrogative); /-J3u/ plural; and /-ra/ evidential. On occasion, non-prosodic suffixes can modify the main Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 332 stress o f the word. One case is the ergative suffix /-an/. When added to a disyllabic syllable, this suffix causes stress to shift from the first to the second syllable. The reason for this shift is that this suffix contains a coda consonant, and stress is attracted by second heavy syllables. Secondary stress appears to be morphologically determined. Elias (2000) suggests that prosodic suffixes were independent words in a previous stage of the language. Below is a paradigm of the word /a.ta.pa/ ‘hen’ with different types of suffixes added. (48a-c) show that certain suffixes do not carry secondary stress even if the word has four or more syllables. (48d-f) show that some suffixes introduce secondary stress; more than one secondary stress is possible, depending on how many prosodic suffixes are added to the word (48g). (48h) shows a word with a non- prosodic suffix and a prosodic one. Secondary stress is only seen in the prosodic suffix. (43i) shows that compounds have secondary stress on the second word. (48) Paradigm: /atapa/ ‘hen’+ suffixes. (a) [’a.ta.pa.bu] (b) ['a.ta.pa.ra] (c) ['a.ta.pa.bu.ra] (d) ['a.ta.pa^nin] (e) ['a.ta.pa^ran.ki] (f) ['a.ta.pa.ju.ku] (g) ['a.ta.pa.ju.ku ^un.ki] (h) ['a.ta.pa.bu^run.ki] (i) ['a.ta.pa.jii.ni] ‘hen (plural)’ ‘hen (evidential)’ ‘hen (plural, evidential)’ ‘hen (ergative)’ ‘hen (reportative)’ ‘hen (diminutive)’ ‘hen (diminutive, reportative)’ ‘hen (plural, reportative)’ ‘rooster (hen+male)’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 333 5. 2 Rhythmic alternations Shipibo has a rhythmic alternation which suggests that footing is exhaustive even if secondary stress is not always realized. The existence of this rhythmic alternation was first noted by Lauriault (1948). According to Lauriault, the emphatic suffix /rib/ has two allomorphs, [ribi] and [riba], depending of the number of moras that precede. After an even number of moras [ba] occurs (49a, c, d); after an odd number of moras [bi] occurs (49b, e). (49) Emphatic suffix /rib/: alternations (from Lauriault 1948:22-3; IP A) 1 2 3 4 (a) [a.'-ri.ba.-kui] 1 2 3 4 5 ‘did it again’ (b) ['a. -ma.-ri.bi.-kui] 1 2 3 4 5 6 ‘made him do it again’ (c) ['a.-ma.-rm. si.-ba.-km] 1 2 3 4 5 6 ‘merely made him do it again’1 4 (d) ['a.-pa.ri.-ri.ba.-kui] 1 2 3 4 5 6 7 ‘did it first again’ (e) ['ja.ka.-pa.ri.-ri.bi.-kui/ ‘he sat down again and immediately’ This distribution is consistent with foot structure, as noted by Elias (2000), who states that [ribi] surfaces when the prosodic word has unfooted syllables, and [riba] when all of the syllables in the prosodic word are footed. The foot structure for the examples in (44) is given in (45). 1 4 The underlying form is /a-ma-ris-riba-ki/ (Lauriault (1948:23)). It appears that assimilation takes place between an adjacent /sr/ sequence with /s/ surfacing. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 334 (50) Emphatic suffix /rib/: footing (e) [a.('ri.) (ba.kui)] ‘did it again’ (f) [fa.ma.) (ri.bi.) kui] ‘made him do it again’ (g) [('a.ma.) (rui.si.) (ba.kui)] ‘merely made him do it again’ (h) [('a.pa.) (ri.ri.) (ba.kui)] ‘did it first again’ (e) [('ja.ka.) (pa.ri.) (ri.bi-) kui] ‘he sat down again and immediately’ I propose this is another instance of a rhythmic contrast encoded as a segmental pattern independently of stress. In this case, it is a vocalic pattern which determines rhythm. The choice of [i] or [a] as the last vowel of the emphatic results from the inherent sonority of the vowel, [a] is more sonorous than [i]; in forms selecting [riba], [ba] is footed as the strong syllable of the foot. In forms selecting [ribi], [bi] is footed as the weak part of the foot. This suggests that the most sonorous vowel is selected in positions favoring maximal sonority.1 5 This rhythmic alternation can be analyzed with the mechanism of Prominence Alignment (Prince and Smolensky (1993)). The combination of the foot prominence and vocalic prominence scales in (46) results in the constraint hierarchies in (47), which express the cross-linguistic tendency for weak syllables to avoid sonorous vowels, and the tendency for head syllables to avoid non-sonorous vowels. 1 5 The em phatic suffix in Shipibo [ribi/riba] is realized as [ri?.bi] or [ri.bi] in Capanahua, depending on rhythm ic considerations (see section 3). This suffix appears to derive from a combination of /ritJ7 + /§i/ (Loos p.c.). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 335 (51) Prominence scales Foot prominence Headpro m >weak (Kenstowicz (1996)) Vocalic prominence a p ro m >e, o p ro m >i, u (Prince and Smolensky (1993)) (52) Constraint hierarchies *HEAD/i, u » * HEAD /e, o » * Head /a ‘Head syllables do not have non-sonorous vowels’ *WEAK/a»*WEAK/eo»*WEAK/i, u ‘Weak syllables do not have sonorous vowels’ This rhythmic alternation only occurs for one suffix. I assume that the last vowel of the suffix ‘again’ is not specified in the input; the choice of this vowel is brought about by the constraint ranking. The ranking of the prominence/sonority constraints in (46) with Max Stress and *Head/d in Shipibo derive this vocalic rhythmic alternation (tableaux 38, 39). In tableau 38, if the second syllable of ‘again’ is footed as head, its nucleus will surface as the most sonorous vowel, /a/. Candidate (a) loses to *Head/i and candidate (c) loses to Max Stress, since it expresses rhythmicity through secondary stresses. Candidate (b) is selected as optimal. Tableau 38 /apari-ribV-ki/ Max *Weak/ *HEAD/i *Head/g ‘he did it again and again’ Stress a (ribV) (ribV) a. ('a.pa.) (ri.ri.) (bi.ki) ** b. ('a.pa.) (ri.ri.) (ba.ki) - ** c. (I a.pa.)(,ri.ri.) (,ba.ki) *! In tableau 39, the second syllable of ‘again’ surfaces as a weak footed syllable, and its nucleus is realized as [I]. Candidate (b) loses because it realizes rhytmicity Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 336 through secondary stresses. Candidate (c) loses because [a] is too sonorous to occur in the weak syllable of the suffix. Candidate (a), which violates *H ead/< j, is selected as optimal. Tableau 39 /jaka-pari-ribV-ki/ ‘he sat down again and immediately’ M ax Stress *Weak/ a (ribV) *HEAD/i (ribV ) *Head/o ^ a. ('ja.k a.) (pa.ri.) (ri.bi) ki ** b. ('ja.ka.) (,pa.ri.) (,ri.bi) ki ** ,, c. ('ja.ka.) (pa.ri.) (ri.ba) ki *! , ■ Similar to related Capanahua and Huariapano, Shipibo has some form of rhythmic alternations. Unlike them, the alternation is vocalic and it only affects one suffix. This alternation is productive and suggests that footing is exhaustive in Shipibo even if secondary stresses are not realized. 6. Conclusion This chapter has explored rhythmic consonantal processes in related Capanahua and Huariapano. In both of them a laryngeal segment in coda position creates a rhythmic contrast within a foot, differentiating between strong and weak footed syllables. In Capanahua, this contrast is achieved through deletion of coda 1 1 1 in a weak syllable and through onset-to-coda metathesis in a strong syllable. It has been argued that these processes achieve persistent footing and make up for the absence of secondary Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 337 stress in the language. In Huariapano, the rhythmic contrast between strong and weak footed syllables is achieved through the epenthesis of [h] in coda position. This epenthesis reinforces default left-to-right secondary stress. On top of this, [h] epenthesis achieves the parsing of otherwise unfooted syllables into feet. Capanahua and Huariapano provide a nice comparison to strictly stress- sensitive consonantal phenomena, of which the case study of North-Central Peninsular Spanish from the previous chapter is an example. Capanahua and Huariapano have strictly foot-conditioned consonantal processes. Capanahua has only one stress per word, but nonetheless, /?/ deletion and onset-to-coda metathesis occur persistently within the word. In Huariapano, [h] epenthesis can be found in both stressed and unstressed syllables. However, persistent footing explains the distribution of [h] epenthesis. It is proposed that the exceptions for [h] epenthesis are also foot-based; [h] epenthesis conflicts with words where secondary stress is assigned right-to-left from the main foot, and in words where the last syllable is extrametrical. In these cases, the preference for creating rhythm through consonantal processes rather than stress explains this conflict. This chapter also considered the case of Shipibo, where the alternation of the suffix [riba/ribi] ‘again’ is also metrically determined. It was shown that [riba] occurred after an unfooted syllable, and that this suffix was footed as [(.. ,ri.)(ba...)]. When this suffix is attached to a word where all syllables are parsed into binary feet, the form selected is [ribi], footed [(ri.bi)]. It was suggested that [riba] was selected Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 338 when [ba] was the first syllable of a trochaic foot for prominence reasons; since led has a higher sonority than /i/, /a/ occurs in strong footed syllables, and lit in weak footed syllables. Thus, a vocalic process is shown to create a rhythmic contrast in another Panoan language. Rhythmic consonantal and vocalic phenomena are commonly found in this language family. Panoan languages evidence the four predictions about rhythmic processes that were made in chapter 3. Rhythmic processes might increase the prominence of the head footed syllable, as [h] epenthesis in Huariapano. Rhythmic phenomena might decrease the prominence of the weak footed syllable, as Capanahua /?/ deletion. They might do both at the same time in the same language, as in Capanahua, where Capanahua /?/ deletion occurs side by side with onset-to-coda metathesis; and finally, a rhythmic process might achieve both rhythmic aims at the same time, as in Shipibo. It remains to be studied how other Panoan languages achieve these means. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 339 Chapter 6: Concluding remarks 1. Stress and footing in consonantal processes This dissertation has examined stress and foot-sensitive consonantal processes in a 74 languages from 36 language families. It has shown that a large number of languages evidence consonantal phenomena which are caused either by stress or by foot structure. The examination of such phenomena suggests four main factors underlying them: aerodynamics, perception, duration, and prominence/rhythmicity. I have proposed that aerodynamic and durational factors are mostly stress- related and can influence consonantal phenomena in a phonetic (i.e. variable, partial and gradient) way. They can also be phonologized. Perceptual factors related to perceptual ambiguity and reinterpretation are mostly phonetic, while other perceptual considerations concerning the avoidance of salient features or segments in unstressed or weak positions are mostly phonological. Finally, prominence/rhythmicity is phonological. One of the most important findings of this study is that stress and footing are separate forces that can govern consonantal processes. Strictly stress-sensitive consonantal phenomena and strictly foot-sensitive phenomena coexist with consonantal phenomena where both stress and footing coincide. This is especially true of trochaic languages, where foot heads are typically initial and stressed and foot complements medial and unstressed. Strictly stress-sensitive consonantal phenomena Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 340 are mostly aerodynamic or durational, as shown by the case study on Basque Spanish. Strictly foot-sensitive consonantal phenomena are found in three main situations: iambic languages, languages with persistent footing but not stress in all feet, and languages where stress and footing seem to diverge. In iambic languages, stress and foot-based processes are not expected to coincide because foot heads are stressed but medial, and foot complements unstressed but foot initial. Iambic systems evidence consonantal processes that are foot-sensitive rather than stress-sensitive; one example is Norton Sound Yupik, where fortition occurs to signal foot boundaries rather than to increase the prominence of stressed syllables or as a consequence of stress. In languages with persistent footing where not all feet are stressed, consonantal alternations can create rhythmicity and thus compensate for the absence of secondary stress; one example is Capanahua. In some other cases, footing for stress and for consonantal processes seems to differ; this is the case of Huariapano. The finding that both strictly foot-sensitive and strictly stress-sensitive phenomena are attested has implications for the current metrical theory, which tends to equate footing with stress. In this study I propose that footing does not necessarily depend on stress, and that rhythmicity can be expressed through stress, consonantal alternations, vocalic alternations, and tone. An important proposal of this dissertation is that prominence- and rhythmic- based consonantal phenomena can be analyzed with Prominence Alignment (Prince (1993)). Prominence Alignment, which is based on the combination of segmental Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 341 and prosodic scales, derives stress- and foot-sensitive consonantal processes where prominence is at stake. Prominence in a stressed syllable or a foot head can be realized through stress, reduced consonantal sonority, or by the realization of codas, while prominence is reduced in unstressed syllables or foot complements through lack of stress, increased consonantal sonority, or lack of codas. Which option is chosen depends on the individual language considered. Prominence Alignment also captures the independent nature of stress and foot structure. Foot prominence and stress prominence can be combined into rankable constraints that express the universal tendencies for foot heads to be stressed and for foot complements to be unstressed. The fact that these constraints are rankable allows for situations where these universal tendencies are overridden, causing foot heads not to be stressed, or even (more rarely) foot complements to be stressed. Prominence alignment is also suggested to capture stress-sensitive, phonological consonantal phenomena with aerodynamic and durational basis. This dissertation also bears on the relationship between phonetics and phonology. It has been proposed that aerodynamic and durational processes are phonetic because they follow from the articulatory and aerodynamic properties of stress: stress involves longer duration and higher airflow (Lehiste (1970)). Additionally, aerodynamic and durational consonantal phenomena are frequently variable and optional. Consonantal phenomena where perceptual ambiguity is at stake belong to the phonetics; by contrast, perceptually-based phenomena whereby salient features or segments are avoided in unstressed syllables and preferred in Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 342 stressed syllables is phonological. Finally, prominence-based phenomena belong to the phonology; prominence is not an effect of stress, and it primarily involves phonological weight. Perceptual and prominence-based phenomena are categorical and obligatory. The fact that durational and aerodynamically-based consonantal phenomena are phonetic does not exclude their possible phonologization. Consonantal processes with a phonetic motivation might be phonologized if they become obligatory and categorical. These cases can be analyzed through the extension of Prominence Alignment to account for categorical or obligatory processes with aerodynamic or durational bases. Additional findings of this study relate to the factors interacting with stress and footing in consonantal processes: word position, morpheme position, and phrase/utterance position. It is found that word position interacts with stress in most cases, while morpheme and phrase/utterance position are rarely reported to be relevant to stress- or foot-sensitive processes. This study also examines the role of syllabic positions, and it finds that most stress- and foot-sensitive consonantal phenomena occur in onsets or depend on onsets for their occurrence. However, the significance of this finding is not clear, especially since a large number of languages lack or have restrictions on codas. Furthermore, it is found that a large number of stress- and foot-sensitive phenomena are related to laryngeal segments and features. This is especially true in epenthesis phenomena. Except in one case, all attested epenthesis cases involve Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 343 laryngeal consonants or features, especially glottalization. It is suggested that laryngeal phenomena are the most sensitive to stress and foot structure because the lack of supra-laryngeal features makes laryngeal segments and features to be deleted, epenthesized and moved around more easily. Also, laryngeal features are related to stress, since variation in laryngeal behavior can produce modifications of stress and pitch. The thesis examined in detail various case studies on stress-sensitive, aerodynamic, phonetic phenomena (North-Central Spanish) and foot-conditioned, rhythmic, phonological phenomena (Panoan). These case studies involve laryngeal features or consonants in coda position. North-Central Spanish shows that increased airflow in stressed syllables induces more likelihood of frication of coda /b, d, g/ in this position, but this effect is variable and thus this process is not analyzed in the grammar. Capanahua and Huariapano show that languages can use a range of consonantal processes to express rhythmicity; in these particular cases, deletion and epenthesis of coda laryngeals, and onset-to-coda metathesis. Shipibo, related to Capanahua and Huariapano, shows that rhythmic alternations through other means than stress are common in Panoan as a whole; this language has vocalic alternations in at least one of its suffixes. In the following pages I discuss some further points that merit further investigation related to the relationship between stress and footing and phonetics and phonology. The first relates to the status of foot heads and the possibility that they are considered prominent categories. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 344 2. Further issues 2.1 Are foot heads prominent categories? On the one hand, stress and footing are separate entities which can coincide but do not need to. On the other hand, foot-influenced consonantal processes are prominence-related. In other words, foot-sensitive consonantal processes frequently create a rhythmic contrast between strong and weak footed syllables and may do so in the absence of or in spite of stress. The prominence distinction between foot heads and complements has been expressed in the foot prominence scale Head p r o m >weak (Kenstowicz (1996)); however, foot heads— independently of stress— are not considered to be in the set of prominent categories. This set has been shown to include stressed syllables, onsets, word-initial syllables, and roots (Beckman (1998), Smith (2000, 2002)). Prominence-enhancing frequently correlates with foot heads, and prominence-reduction with non-heads. This cannot be expressed by means of making reference to other prominent categories; clearly not to onsets and roots, and neither to word-initial syllables, since foot-sensitive consonantal processes are iterative. Stressed syllables cannot make reference to these processes either, since in many occasions, foot heads are not stressed. One example is Capanahua. This suggests that foot heads should be included in the set of prominent categories. Footing can have stress as its rhythmic expression, but in the absence of stress other segmental or prosodic material can express rhythm. The data examined Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 345 in this dissertation suggests this is the case. Including foot heads in the set of prominent categories would achieve more than to refer to the domain of application of consonantal (or other segmental or prosodic) phenomena. It would also make possible to make direct reference to foot positions for different phenomena. This could include positional faithfulness, positional markedness, and others. Additionally, the inclusion of foot heads in the set of prominent categories would mean that M/Str constraints (Smith (2000, 2002)) could be extended to cases of augmentation in foot heads. 2.2 Consonantal vs. vocalic alternations This dissertation concentrates on stress- and foot-conditioned consonantal processes. It shows that many languages undergo these processes. However, in most languages it is vowels that are affected by these prosodic factors. A question to investigate is whether rhythmic alternations of consonants or vowels depend on the correlates of stress that are predominant in the language, or whether there is some other explanation. Since vowels are mostly affected by stress (Hayes (1995)), most phonetic studies concentrate on this effect. If stress has as correlates pitch and duration, and then amplitude, vowels will be more readily affected by stress and better suited to express rhythmicity. More phonetic studies on the influence of stress on consonants could show that actually, many languages evidence stress effects on both vowels and consonants. This is especially true for lesser known languages that have not been Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 346 widely studied. Rhythmicity has been shown to occur with and without the effect of stress; more study of strictly foot-sensitive consonantal and vocalic processes would add to our understanding of these processes. 2.3 The asymmetry between trochaic and iambic systems revisited It is well established that trochaic and iambic systems are asymmetric in many respects; for example, iambic systems express a difference in duration and trochaic in intensity (Iambic Law, Hayes (1995)); iambic systems are also rarer across languages (see, for instance, Van de Vijver (1998)). An asymmetry in iambic and trochaic systems is also evident in this survey; very few of the languages examined are iambic. Iambic systems are special in that stress and foot position are inverse than in trochaic systems, and thus stress and footing can hardly coincide in conditioning a process. Foot-sensitive processes seem to mark foot boundaries. Additional consonantal phenomena in iambic systems would aid to our knowledge of the drives of consonantal patterns in these systems. Related to this is the assumption that foot initial position arises from the interaction of stressed syllables and word-initial positions, and that foot medial positions arise from the interaction between unstressed and medial positions (Lavoie (2001)). From the examination of stress and position, it is shown that different patterns concerning position and stress exist. In one, consonantal alternations occur in word initial stressed syllables. But in others, a process occurs in stressed syllables Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 347 and in word-initial syllables regardless of stress. More investigation of these issues is needed. 2.4 The ambiguity of stress and word position Lavoie’s (2001) survey on lenition finds that only a small number of cases mention stress as a conditioning context, while word position is pervasive. However, her phonetic studies find that stress correlates more often than word position with lenition. Lavoie suggests that there is a bias towards considering word position as the relevant factor for lenition processes, and that this arises from the fact that it is easier to see the relationship between segmental distribution and word position than that of segmental distribution and stress. Also, she notes that word position leads to phonologization more often than stress. As shown in this dissertation, in some cases it is ambiguous whether stress or word position is relevant. Word initial positions, for instance, tend to be stressed, and word-final positions unstressed. 2.5 Unfooted syllables An interesting issue is the patterning of unfooted syllables with respect to foot- sensitive consonantal phenomena. This is decisive in the formulation of prominence scales. For example, there are two possibilities for the prominence scale: Head p r o m > non Head, or Head p r o m 5" weak. Which one is correct, or whether both are needed separately to account for different languages is an empirical issue. Head p r o m > non Head would correspond more closely to stress-sensitive phenomena; a process Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 348 occurs in stressed syllables, or in unstressed syllables. On the contrary, the scale Head pron^weak implies three distinct foot-sensitive elements: head syllables, complements, and unfooted syllables. There are reasons to suggest that Head p r o m >weak is correct. There are consonantal processes that affect just weak footed syllables syllables and not unfooted syllables; one example is Capanahua. Further investigation is needed to confirm this conclusion. It is plausible that unfooted syllables are essential in distinguishing cases of stress sensitive and foot sensitive phenomena. In languages with iterative footing, even in the absence of secondary stress, unfooted syllables should not behave like unstressed, footed syllables. In Capanahua, unfooted syllables do not undergo 11 1 deletion. Ill deletion is a characteristic of weak-footed syllables, not of all unstressed syllables. 2.6 The ‘after’ pattern Several alternations considered in this thesis are reported to have a conditioning environment of ‘after’ stressed or unstressed syllables. Examples are gemination, (pre-glottalization), pre-aspiration, some cases of voicing alternations, dissimilation, and others. In the case of gemination, epenthesis and pre-aspiration the reference to ‘after a stressed syllable’ makes sense; consonants are ‘doubled’ in some sense for Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. prominence reasons. However, in some other cases, the reference to the after context is not clear. Some examples include voicing alternations in Paamese, Middle English, and Proto-Germanic (Vemer’s Law). In Paamese, a medial /nt/ is voiced obligatorily after stress; in Middle English and Proto-Germanic, lack of stress was necessary for voicing alternations to arise. Dissimilation in Gothic is another example. In these cases, it is unclear what the ‘after’ pattern means; whether it means that the stress of the preceding syllable is conditioning the process, or that the consonant was/is syllabified as a coda, or whether the preceding stress makes more airflow available, causing more voicing, fortition, or other phenomena. These issues remain to be investigated. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 350 REFERENCES Alarcos Llorach, Emilio. 1950. Fonologia espahola (segun el metodo de la escuela de Praga). Madrid: Gredos. Alber, Birgit. 2002. Maximizing first positions. In Caroline Fery, Antony Dubach Green, and Ruben van de Vijver (eds.). Proceedings of HILP 5. University of Postdam. 1-19. Alber, Birgit. To appear. Regional variation and edges: glottal stop epenthesis and dissimilation in Standard and Southern varieties of German. Zeitschrift fur Sprachwissenschaft 2001. Amastae, Jon. 1986. A syllable-based analysis of Spanish spirantization. In Osvaldo Jaeggli and Carmen Silva-Corvalan (eds.), Studies in Romance Linguistics. Dordrecht: Foris. 3-21. Amastae, Jon. 1995. Variable spirantization: constraint weighting in three dialects. Hispanic Linguistics 6/7. 265-85. Applegate, John. 1972. Ineseno Chumash Grammar. Ph.D. Dissertation, University of California at Berkeley. Archangeli, Diana, and Douglas Pulleyblank. 1994. Grounded Phonology. Cambridge, Mass.: MIT Press. Armstrong, Lilias E. 1964. The Phonetic Structure o f Somali. Famborough: Gregg Press Limited (First published in 1934 as part of Mitteilungen des Seminars fur Orientalische Sprachen zu Berlin, n. 37). Bakovic, Erik. 1995. Strong onsets and Spanish fortition [ROA 96, http://roa.rutgers.edu]. Bakovic, Erik. 1996. Quantitative Adjustments in Yupik. Proceedings ofWCCFL 15. Standford: CSLI publications. Bauer, Winifred. 1993. Maori. London, New York: Routledge, Beckman, Jill. 1997. Positional faithfulness, positional neutralization and Shona height harmony. Phonology 14.1. 1-46. Beckman, Jill. 1998. Positional Faithfulness. Ph. D. Dissertation, University of Amherst, Massachussets, GLSA. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 351 Bessell, N. 1992. Towards a phonetic and phonological typology of post-velar articulation. Vancouver: University of British Columbia dissertation. Blevins, Juliette. To appear. The independent nature of phonotactic constraints: An alternative to syllable-based accounts. In Caroline Fery and Ruben van de Vijver (eds.). The syllable in Optimality Theory. Cambridge: Cambridge University Press. Blevins, Juliette, and Andrew Garrett. 1998. The origins of consonant-vowel metathesis. Language 74. 508-56. Bloomsfield, Leonard. 1956. Eastern Ojibwa. Ann Arbor: University of Michigan Press. Booij, Geert. 1995. The Phonology o f Dutch. Oxford: Clarendon Press. Borowsky, Toni. 1986. Topics in the Lexical Phonology o f English. Ph.D. Dissertation, University of Massachussetts at Amherst: GLSA Publications. 271-4. Borzone de Manrique, A. and A. Signorini. 1983. Segmental Duration and Rhythm in Spanish. Journal o f Phonetics 11. 117-28. Botinis, Antonis, Marios Fourakis and Robert Bannert. 2001. Prosodic interactions of segmental durations in Greek. Working Papers o f the Department of Linguistics and Phonetics, Lund University 49. 10-3. Botinis, Antonis, R. Bannert, M. Fourakis and S. Pagoni-Tetlow. 2002. Crosslinguistic segmental durations and prosodic typology. Proceedings from Fonetik 2002, The Swedish Phonetics Conference. Speech, Music and Hearing Quarterly Progress and Status Report. Stockholm, Sweden, May 2002, vol. 44. Department of Linguistics, Stockholm University. 77-80. Breen, Gavan, and Rob Pensalfini. 1999. Arremte: a language with no syllable onsets. Linguistic Inquiry 30. 1-26. Browman, Catherine and Louis Goldstein. 1990. Tiers in articulatory phonology, with some implications for causal speech. In John Kingston and Mary Beckman (eds.), Papers in Laboratory Phonology I. Cambridge: Cambridge University Press. 341-76. Browman, Catherine, and Louis Goldstein. 1992. Articulatory phonology: an overview. Phonetica 49. 155-80. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 352 Buller, Barbara, Ernest Duller and Daniel Everett. 1993. Stress placement, syllable structure, and minimality in Banawa. International Journal o f American Linguistics 59. 280-93. Byrd, Dani. 1996. A phase window framework for articulatory timing. Phonology 13. 139-69. Carr, Philip. 1999. English phonetics and phonology. An introduction. Blackwell: Oxford. Carreira, Maria. 1998. A constraint-based approach to Spanish spirantization. In I. Lema and E. Trevino (eds.), Theoretical Analyses of Romance Languages. Amsterdam: John Benjamins. 138-43. Catford, I. 1977. Fundamental Problems in Phonetics. Bloomington, IN: University of Indiana Press. Chafe, William. 1977. Accent and related phenomena in the five nations Iroquois languages. In Larry Hyman (ed.). Studies in Stress and Accent. Southern California Occasional Papers in Linguistics 4. Los Angeles: Dept, of Linguistics, University of Southern California. Chierchia, Gennaro. 1982. An autosegmental theory of raddoppiamento. In James Pustejovsky and Peter Sells (eds). Proceedings o f the North East Linguistics Society (NELS) 12. University of Massachusetts at Amherst: Graduate Linguistic Student Association. 49-62. Chierchia, Gennaro. 1986. Length, syllabification and the phonological cycle in Italian. Journal o f Italian Linguistics 8. 5-33. Chomsky, Noam and Morris Halle. 1968. The Sound Pattern o f English. New York: Harper & Row. Christmas, R. and Christmas, J. 1975. Kupia phonemic summary. Kathmandu: Summer Institute of Linguistics and Institute of Nepal and Asian Studies, Tribhuvan University. Cid-Hazard, Susana. In preparation. Variacion socioestilistica en el espanol de Santiago de Chile. Ph. D. Dissertation, University of Southern California. Clements, George. 1987. Phonological feature representation and the description of epenthetic stops. Parasession on Aoutosegmental and Metrical Phonology, Chicago Linguistics Society. 29-50. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 353 Cole, John, Jose Ignacio Hualde and Khalil Iskarous, Khalil. 1998. Effects of prosodic and segmental context on /g/ lenition in Spanish. In O. Fujimura, B. Joseph and B. Palek. Proceedings of LP ’ 98. Charles University, Prague: Karolinum Press. Collinge, N. E. 1985. The Laws o f Indo-European. Amsterdam: John Benjamins. Collins, Beverley, and Inger Mees. 1990. The Phonetics of Cardiff English. In Nikolas Coupland and Alan Thomas (eds.), English in Wales: Diversity, Conflict and Change. Clevedon: Multilingual matters Ltd.. 87-103. Crosswhite, Katherine. 1999. Vowel reduction in Optimality Theory. UCLA Ph. D. Dissertation. Crosswhite, Katherine. To appear. Vowel Reduction. In Bruce Hayes, Robert Kirchner and Donca Steriade (eds.), Phonetically Based Phonology. Crowley, Terry. 1982. The Paamese language o f Vanuatu. Pacific Linguistics Series B, no. 87. Canberra: Australian national University. Crystal, Thomas and Arthur House. 1988. Segmental durations in connected speech signals: Current results. Journal o f the Acoustic Society o f America 83 (4). 1553-73. Davis, Stuart. 1999. The parallel distribution of aspirated stops and IhJ in American English. Indiana University Working papers in Linguistics 1. 1-10. Davis, Stuart. 2002. The Foot alignment Analysis of the Feature [Spread Glottis] in American English. Handout from the 10t h Manchester Phonology Meeting, May 25, 2002. Davis, Stuart and Mi-hui Cho. 2003. The distribution of aspirated stops and P a J in American English and Korean. An alignment approach with typological implications. Linguistics 41. 607-52. De Jong, Kenneth, Mary Beckman and J. Edwards. 1993. The interplay between prosodic structure and coarticulation. Language and Speech 36 (2, 3). 197- 212 . De Lacy, Paul. 1999. Tone and Prominence [ROA 333, http://roa.rutgers.edu]. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 354 De Lacy, Paul. 2001. Markedness in Prominent Positions. To appear in Adam Szczegielniak and Lance Nathan (eds.). Proceedings o f HUMIT (Harvard- MIT conference). Cambridge, MA: MIT Working papers in Linguistics [ROA 432, http ://roa.rutgers.edu]. De Lacy, Paul. 2002. Conflation and Scales. In Masako Hirotani (ed.), Proceedings o f the North East Linguistics Society (NELS) 32. University of Massachusetts, Amherst: GLSA. 93-112. De Wolf, G. and E. Hasebe-Ludt. 1987. The relationship of speakers’ language attitudes to intervocalic voicing in formal and informal speech. Proceedings o f the 14th International Congress o f Linguists II. Berlin: Akademie-V erlag. 1562-67. Deferrari, Harry. 1954. The phonology o f Italian, Spanish and French. Washington, D.C.: Georgetown University Press. Delattre, P. 1969. An acoustic and articulatory study of vowel reduction in four languages. International Review of Applied Linguistics 7. 295-325. Delgutte, B. 1997. Auditory neural procesing of speech. In W. Hardcastle and J. Laver (eds.), The Handbook o f Phonetic Sciences. Oxford: Blackwell. 507- 38. Dilley, L. and Stephanie Shattuck-Hufnagel. 1996. Glottalization of word-initial vowels as a function of prosodic structure. In Journal o f Phonetics 24. 423- 44. D ’Introno, Francesco, Enrique del Teso and Rosemary Weston. 1995. Fonetica y fonologla del espahol actual. Madrid: Catedra. Dougherty, B. 1993. The acoustic phonetic correlates o f Cayuga word stress. Cambridge, MA: Harvard University dissertations. Downing, Laura. 1998. On the prosodic misalignment of onsetless syllables. Natural Language and Linguistic Theory 16. 1-52. Dumenil, Annie. 1987. A rule-account of metathesis in Gascon. Lingvisticoe Investigationes XI:1. 81-113. Dunn, J. 1995. Sm ’ algyax: a reference dictionary and grammar for the Coast Tsimshian language (Part II, 4). Juneau: Sealaska Heritage Foundation. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 355 Dunn, J. and R. Hays. 1983. Tsimshian uvular syllables. International Journal o f American Linguistics 49. 46-63. Ellas, Jose Alberto. 1999. Sllabas Pesadas ^Monomoraicas?: Reflexionando sobre el Peso Silabico en Shipibo. Comunicacion presentada en el I Congreso de Lenguas Indigenas de Sudamerica, 4-7 de agosto de 1999, Universidad Ricardo Palma. Lima, Peru. Elias, Jose Alberto. 2000. El acento en Shipibo. Tesis para Licenciado en Linguistica, Universidad Nacional Mayor de San Marcos, Lima, Peru. Elias, Jose Alberto. 2001. Ghost segments in Shipibo. Ms, U. Rutgers. Engstrand, Olle. 1987. Preaspiration and the voicing contast in Lule Sami. Phonetica 44. 103-16. Everett, Daniel. 1988. On metrical constituent structure in Piraha phonology. Natural Language and Linguistic Theory 6. 207-46. Everett, Daniel. 1996. Optimality Theory and Arawan Prosodic Systems. Chapter One: Prosodic Levels and Constraints in Banawa and Suruwaha [ROA 121, http://roa.rutgers.edu]. Everett, Daniel, and Keren Everett. 1984a. On the relevance of syllable onsets to stress placement. Linguistic Inquiry 15. 705-11. Everett, Daniel, and Keren Everett. 1984b. Syllable onsets and stress placement in Piraha. Proceedings o f the West Coast Conference on Formal Linguistics 3, 105-16. Fischer-Jorgensen, E. 1987. A phonetic study of the stod in Standard Danish. Annual Report of the Institute o f Phonetics o f the University o f Copenhagen 21. 55- 65. Flemming, Edward. 1994. The role of metrical structure in segmental rules. In M. Gonzalez (ed.), Proceedings o f the North-Eastern Linguistics Society 24 (vol. 1). Amherst, MA: GLSA. 97-110. Flemming, Edward. 2001. Scalar and categorical phenomena in a unified model of phonetics and phonology. Phonology 18. 7-44. Foster, Mary LeCron. 1969. The Taras can Language. University of California Publications in Linguistics 56. Berkeley and Los Angeles. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 356 Foster, Michael. 1982. Alternating weak and strong syllables in Cayuga words. International Journal of American Linguistics 48. 59-72. Fourakis, Marios. 1990. Tempo, stress and vowel reduction in American English. JASA 90. 1816-27. Gafos, Adamantios. 1996. The articulatory Basis o f Locality in Phonology. Ph. D. Dissertation, University of Maryland. Giegerich, Heinz. 1989. Syllable structure and lexical derivation in German. Bloomington: Indiana University Linguistics Club. Goldrick, Matthew. 2000. Turbid Output Representations and the Unity of Opacity. In M. Hirotani, A. Coetzee, N. Hall, J.Y. Kim (eds.), Proceedings o f the North East Linguistics Society (NELS) 30, Rutgers University. Amherst: GLSA. Goldrick, Matthew, and Prince Smolensky. 2000. Phonological Opacity and Turbid Representations in Optimality Theory. Handout of talk presented at UCLA, April 21, 2000. Goldsmith, John. 1990. Autosegmental and metrical phonology. Oxford: Blackwell. Gonzalez, Carolina. 1999. Spirantization and obstruent coda devoicing in Spanish. Unpublished M.A. paper, University of Southern California. Gonzalez, Carolina. 2002a. The effect of prosody on glottal stop deletion in Capanahua. In Masako Hirotani (ed.), Proceedings of the North East Linguistics Society (NELS) 32. University of Massachusetts, Amherst: GLSA. 133-52. Gonzalez, Carolina. 2002b. Phonetic variation in voiced obstruents in North-Central Peninsular Spanish. Journal of the International Phonetic Association 32.1. Cambridge: Cambridge University Press. 17-31. Gordon, Matthew. 1996. Laryngeal phonology of Hupa. Ms., UCLA. Gordon, Matthew. 2002. An auditory basis for onset-sensitive stress. Poster presented at the North East Linguistics Society (NELS) 33. MIT, Cambridge: Massachussets. Grammont, Maurice. 1905-6. La metathese dans le parler de Bagneres-de-Luchon. Memoires de la Societe de linguistique de Paris 13. 73-90. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 357 Grammont, Maurice. 1909. Une loi fonetique generate. Philologie et linguistique: Melanges offerts a Louis Havet par ses anciens eleves et ses amis a I ’ occasion du 60e anniversaire de sa naissance, le 6 janvier 1909. Paris: Hachette. 179-83. Grammont, Maurice. 1933. Traite de Phonetique. Paris: Librairie Delagrave. Grimes, Barbara (ed.). 2000. Ethnologue: Languages o f the World. 14th Edition (CD-ROM Version). [1s t edition 1957]. Dallas, Texas: SIL International. Halle, Morris. 1995. Feature geometry and feature spreading. Linguistic Inquiry 26. 1-46. Halle, Morris, and Jean-Roger Vergnaud. 1987. An Essay on Stress. Cambridge, Mass.: MIT Press. Hammond, Michael. 1997. Optimality Theory and Prosody. In Diana Archangeli and D. T. Langedoen (eds.), Optimality Theory: An Overview. Malden, Mass.: Blackwell. 33-42. Hammond, Michael. 1999. The Phonology o f English. A Prosodic Optimality- Theoretic Approach. Oxford: Oxford University Press. Harris, James. 1969. Spanish Phonology. Cambridge: MIT Press. Harris, James. 1984. La espirantizacion en castellano y la representation fonologica autosegmental. In Joan Mascaro, Bartra, Trucar, Nadal and Rigau (eds.), Estudis Gramaticals 1, Working Papers in Linguistics. Barcelona: Universitat Autonoma de Barcelona. Publicaciones del Departamento de Filologia Hispanica, Universidad Autonoma de Barcelona. Serie Linguistica, Barcelona. 149-67. Harris-Northhall, Raymond. 1990. Weakening processes in the history o f Spanish consonants. London, New York: Routledge. Hayes, Bruce. 1980. A metrical theory of stress rules. Ph.D. dissertation, MIT. [Published 1985, New York: Garland Press.] Hayes, Bruce. 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20. 253-306. Hayes, Bruce. 1995. Metrical stress theory: principles and case studies. Chicago: University of Chicago Press. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 358 Hayes, Bruce. 1997. Phonetically driven phonology: the role of Optimality Theory and Inductive Grounding. Proceedings o f the 1996 Milwaukee Conference of Formalism and Functionalism in Linguistics. Helimsky, Eugene. 1998. Nganasan. In Daniel Abondolo (ed.), The Uralic languages. New York & London: Routledge. 480-515. Hercus, Louise. 1986. Victorian Languages: A Late Survey. Pacific Linguistics B77, Australian National University, Canberra. [Revised and enlarged version of The Languages o f Victoria: A Late Survey (in two volumes), Australian Institute of Aboriginal Studies, Canberra, 1969]. Hewitt, Mark. 1992. Vertical Maximization and Metrical Theory. Doctoral Dissertation, Brandeis University, Waltham, Massachusetts. Hollenbach, B. 1977. Phonetic vs. phonemic correspondence in two Trique dialects. In William Merrifield (ed.), Studies in Otomanguean Phonolog. Dallas: SIL and University of Texas at Arlington. 35-67. Honeybone, Patrick, and Kevin Watson. 2002. Visarga in pausa in Liverpool English. Poster presented at the 10th Manchester Phonology Meeting, 23-25 May 2002. Howe, Darin, and Douglas Pulleyblank. 2001. Patterns and timing of glottalization. Phonology 18. 45-80. Hualde, Jose Ignacio. 1989. Procesos consonanticos y estructuras geometricas en espanol. Linguistica ALFAL 1. 7-44. Hulst, Harry van der. 1984. Syllable Structure and Stress in Dutch. Foris: Dordrecht. Hume, Elizabeth. 2000. Metathesis website [www.ling.ohio-state.edu/~ehume/ metathesis]. Hume, Elizabeth, and Misun Seo. 2001. A Comparative Account of Metathesis in Faroese and Lithuanian. In Elizabeth Hume, Norval Smith and Jeroen van de Weijer. Surface Syllable Structure and Segment Sequencing. HIL Occasional Papers. Leiden, NL: HIL. Hyman, Larry. 1985. A theory o f phonological weight. Dordrecht: Foris. Iverson, G. and J. Salmons. 1995. Aspiration and laryngeal representation in Germanic. Phonology 12. 369-96. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 359 Jacobson, Steven. 1985. Siberian Yupik and Central Yupik prosody. In M. Krauss (ed.), Yupik Eskimo Prosodic systems: Descriptive and Comparative Studies. Alaska Native Language center, Fairbanks, Alaska. 25-46. Jannedy, Stefanie, Robert Poletto and Tracey Weldon. 1994. Language files: materials for an introduction to language and linguistics. Columbus: Ohio State University Press. Jones, Ted and Lyle Knudson. 1977. Guelavla Zapotec phonemes. In William Merrifield (ed.), Studies in Otomanguean Phonology. Dallas: SIL and University of Texas at Arlington. 163-80. Kabell, Inge and Hanne Lauridsen. 1984. On the evidence of the earliest Danish works on English pronunciation. Part I: Consonants. In Inge Kabell, Hanne Lauridsen and Arne Zettersten (eds.), Studies in Early Modern English Pronunciation- A DEMEP Publication. Copenhagen: University of Copenhagen. 25-87. Kager, Rene. 1989. A metrical theory of stress and destressing in English and Dutch. Dordrecht: Foris. Kager, Rene. 1993. Alternatives to the Iambic/Trochaic Law. Natural Language and Linguistic Theory 11. 381-432. Kager, Rene. 1995. Review of Hayes (1995). Phonology 12. 437-64. Kager, Rene. 1999. Optimality Theory: A Textbook. Cambridge: Cambridge University Press. Kahn, Daniel. 1980. Syllable-Based Generalizations in English Phonology. Garland: New York. Kakumasu, James. 1986. Urubu-Kaapor. In D. Derbyshire and G. Pullum (eds.), Handbook o f Amazonian Languages, vol. 1. Berlin: Mouton de Grayer. 326- 403. Keating, Patricia. 1984. Phonetic and phonological representation of stop consonant voicing. Language 60. 286-319. Keating, Patricia. 1985. Universal phonetics and the organization of grammars. In Victoria Fromkin (ed.), Phonetic linguistics: essays in honor o f Peter Ladefoged. New York: Academic Press. 115-32. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 360 Keating, Patricia. 1990. Phonetic representations in a generative grammar. Journal of Phonetics 18. 321-34. Keels, Jack. 1985. Guayabero: Phonology and Morphophonemics. In Ruth Brend (ed.), From Phonology to discourse: Studies in Six Colombian Languages. Dallas, Texas: Summer Institute of Linguistics. Kehrein, Wolfgang. 2001. The Prosodic Phonology o f Laryngeals. PhD dissertation, Philipps University Marburg. Kenstowicz, Michael. 1994. Phonology in Generative Grammar. Cambridge, MA: Blackwell. Kenstowicz, Michael. 1996. Sonority-driven Stress [ROA 33, http://roa.rutgers.edu]. Kim, Christina. 2002. Spirantization of Voiced Obstruents in Spanish. Ms., MIT (term paper for undergraduate phonology course). Kingston, John. 1985. The Phonetics and Phonology of the Timing of Oral and Glottal Events. Ph.D. Dissertation, University of California at Berkeley. Kiparsky, Paul. 1979. Metrical Structure Assignment is Cyclic. Linguistic Lnquiry 10.4. 421— 4 -1 . Kirchner, Robert. 1997. Contrastiveness and faithfulness. Phonology 14. 83-111. Kirchner, Robert. 1998. An Effort-based Approach to Consonant Deletion. Ph. D. dissertation, UCLA. Kohler, K. 1977. Einfuhrung in die Phonetik des Deutschen. Berlin: Schmidt. Kohler, K. 1994. Glottal Stops and Glottalization in German. Phonetica. 38-51. Koopmans-van Beinum, F. 1980. Vowel contrast reduction: an acoustic and perceptual study of Dutch vowels in various speech conditions. Amsterdam: Academische Pres. Kuipers, Aert. 1967. The Squamish Language. Grammar, Texts, Dictionary. The Hague: Mouton. Kuipers, Aert. 1974. The Shuswap Language: Grammar, Texts, Dictionary. Janua Linguarum, Series Practica 225. Ladefoged, Peter. 1967. Three Areas o f Experimental Phonetics. Oxford: Oxford University Press. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 361 Ladefoged, Peter. 1982. A Course in Phonetics (2n d edn.). Chicago: Harcourt Brace [1s t edition 1975], Ladefoged, Peter. 1993. A Course in Phonetics (3r d edn.). Chicago: Harcourt Brace [1s t edition 1975]. Ladefoged, Peter. 1997. Linguistic phonetic descriptions. In William J. Hardcastle and John Laver (eds.). The handbook of phonetic sciences. Oxford and Cambridge, Mass.: Blackwell. 589-618. Ladefoged, Peter and Ian Maddieson. 1996. The Sounds o f the World’ s Languages. Oxford: Blackwell. Lauriault, James. 1948. Altemate-mora timing in Shipibo. International Journal of American Linguistics 14. 22-4. Lavoie, Lisa. 1996. Consonant strength: results of a data base development project. Working Papers o f the Cornell Phonetics Laboratory 11. 269-316. Lavoie, Lisa. 2001. Consonant Strength: phonological patterns and phonetic manifestations. New York: Garland. Leer, Jeff. 1985. Prosody in Alutiiq (The Koniag and Chugach dialects of Alaskan Yupik). In M. Krauss (ed.), Yupik Eskimo Prosodic systems: Descriptive and Comparative Studies. Alaska Native Language center, Fairbanks, Alaska. 77- 134. Leer, Jeff. 1989. Prosody in Chugach Alutiiq. Ms. Alaska native Language Center, Fairbanks, Alaska. Lehiste, Ilse. 1970. Suprasegmentals. Cambridge, MA: MIT Press. Liberman, Mark. 1975. The intonational system o f English. PhD dissertation, MIT. Liberman, Mark, and Alan Prince. 1977. On stress and linguistic rhythm. Linguistic Inquiry 8, 249-336. Lisker, Leigh, and Arthur Abramson. 1967. Some effects of context on Voice Onset Time in English stops. Language and Speech 10, 1-28. Lombardi, Linda. 1991. Laryngeal features and laryngeal neutralization. Ph. D. Dissertation. University of Massachusetts at Amherst: GLSA. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 362 Loporcaro, Michele. 1999. L ’ origine del raddoppiamento fonosintattico. Basel: Francke Verlag. Loos, Eugene. 1969. The Phonology o f Capanahua and its Grammatical Basis. Summer Institute of Linguistics, University of Oklahoma. Loos, Eugene. 1973. La senal de transitividad del sustantivo en los idiomas panos. Estudios Panos I. Serie Lingulstica Peruana n. 10, Instituto Linguistico de Verano, Peru. 133-84. Loos, Eugene. 1999. Pano. In R. M.W. Dixon and Alaxandra Aikhenvald (eds.), The Amazonian Languages. Cambridge Language Surveys. Cambridge: CUP. 227-50. Loos, Eugene and Betty Loos. 1998. Diccionario Capanahua-Castellano. Serie Linguistica Peruana n. 45. Yarinacocha, Peru: Instituto Linguistico de Verano. Lounsbury, Floyd. 1942. Phonology o f the Oneida Language. M.A. thesis, University of Wisconsin. Lounsbury, Floyd. 1953. Oneida Verb Morphology. Yale University Publications in Anthropology, 48. New Haven: Yale University Press. Lozano, Maria Jose. 1979. Stop and Spirant Alternations: Fortition and Spirantization Processes in Spanish Phonology. Ph. D. dissertation, Indiana University Linguistics Club. Maddieson, Ian. 1984. Patterns o f Sounds. Cambridge: Cambridge University Press. Martinez Celdran, Eugenio. 1984. Fonetica. Barcelona: Teide. Martinez-Gil, Fernando. 1991. The insert/delete parameter, redundancy rules, and neutralization processes in Spanish. In Fernando Martinez-Gil and Hector Campos (eds.), Current Studies in Spanish Linguistics. Georgetown: Georgetown University Press. 495-571. Mascara, Joan. 1984. Continuant spreading in Basque, Catalan, and Spanish. In M. Aronoff, M. and R. Oehrle (eds.), Language Sound Structure. Cambridge, MA: MIT Press. 287-98. Mattina, A. 1979. Pharyngeal movement in Colville and related phenomena in the Interior Salish languages. International Journal o f American Linguistics 45. 17-24. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 363 McCarthy, John. 1998. Sympathy and phonological opacity [ROA 252, http ://roa.rutgers. edu]. McCarthy, John. 1999. Sympathy, cumulativity, and the Duke-of-York gambit [ROA 315, http://roa.rutgers.edu]. McCarthy, John, and Alan Prince. 1986. Prosodic morphology. Ms, University of Massachusetts, Amherst and Brandeis University. [Annotated version 1996, issued as Technical report #32, Rutgers Center for Cognitive Science.] McCarthy, John and Alan Prince. 1993. Prosodic morphology I: Constraint interaction and satisfaction. Technical Report #3, Rutgers University Center for Cognitive Science. McCarthy, John and Alan Prince. 1994. The emergence of the unmarked: Optimality in Prosodic Morphology. In M. Gonzalez (ed.), Proceedings o f the North East Linguistic Society 24. 333-79. McCarthy, John and Alan Prince. 1995. Faithfulness and reduplicative identity. In Jill Beckman, Suzanne Urbanczyk and Laura Walsh Dickey (eds.), Papers in Optimality Theory. University of Massachusetts Occasional Papers in Linguistics 18. 249-384 [ROA 216, http://roa.rutgers.edu]. Michelson, Karen. 1988. A comparative study o f Lake-Iroquoian accent. Dordrecht: Kluwer. Mills, Elizabeth. 1984. Senoufo Phonology. Discourse to Syllable (A prosodic approach). Summer Institute of Linguistics and University of Texas at Arlington. Morris, Richard. In press. Coda obstruents and local constraint conjunction in Peninsular Spanish. In D. Cresti, T. Statterfield and C. Tortora (eds.), Current Issues in Linguistic Theory: Selected Papers from the XXIXth Linguistic Symposium on Romance Languages. Amsterdam: John Benjamins [ROA 383, http://roa.rutgers.edu]. Mosse, Ferdinand. 1956. Manuel de la langue gotique: grammaire, texts, notes, glossaire. Paros: Aubier, Editions Montaigne. Myers, Scott. 1987. Vowel shortening in English. Natural Language and Linguistic Theory 5. 485-518. Nespor, Marina and Irene Vogel. 1979. Clash avoidance in Italian. Linguistic Inquiry 10.467-82. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 364 Ni Chasaide, A. and C. O Dochartaigh. 1984. Some durational aspects of preaspiration. In J. Higgs and R. Thelwall (eds.), Topics in Linguistic Phonetics: in Honor o f E.T. Uldall. Occasional Papers in Linguistics and Language Learning n. 9, Department of Linguistics: The New University of Ulster. 141-57. Nichols, Joanna. 1994. Ingush. In R. Smeets (ed.), The Indigenous Languages of the Caucasus, vol. 4. North East Caucasian Languages, Part 2. New York: Caravan. 79— 146. Nivens, R. 1992. A Lexical Phonology of West Tarangan. In Donald Burquest and Wyn Laidig (eds.), Phonological Studies in Four Languages of Maluku. Summer Institute of Linguistics, The University of Texas at Arlington, and Pattimura University. North, Joanne and Jana Shields. 1977. Silacayoapan Mixtec Phonology. In William Merrifield (ed.). Studies in Otomanguean Phonology. Dallas: SIL and University of Texas, Arlington. 21-33. Ohala, John. 1983. The origin of sound patterns in vocal tract constraints. In P.F. MacNeilage (ed.), The Production o f Speech. New York: Springer-Verlag. 189-216. Ohala, John. 1997. The relation between phonetics and phonology. In W. Hardcastle, and J. Laver (eds.), The Handbook o f Phonetic Sciences. Cambridge, MA: Blackwell. 674-94. Padgett, Jaye. 1991. Stricture in Feature Geometry. Ph.D. Dissertation, University of Massachusetts, Amherst. Padgett, Jaye. 1994. Stricture and nasal place assimilation. Natural Language and Linguistic Theory 12, 465-514. Palmada, Blanca. 1997. Continuant spreading and feature organization. In Fernando Martinez-Gil and Alfonso Morales-Front (eds.). Issues in the phonology and morphology o f the major Iberian languages. Washington, D.C.: Georgetown University Press. 151-71. Parker, Steve. 1994. Coda epenthesis in Huariapano. International Journal of American Linguistics 60. 95-119. Parker, Steve. 1998. Disjoint metrical tiers and positional markedness in Huariapano. Generals Paper, UMass. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 365 Parker, Steve. 2001. Non-optimal onsets in Chamicuro: an inventory maximised in coda position. Phonology 18, 361-86. Patz, Elisabeth. 1991. Djabugay. In R. M.W. Dixon and Barry Blake (eds.), The Handbook o f Australian Languages, vol. 4. Oxford: Oxford University Press. 250-60. Perlman, G. and F. Horan. 1986. Report on Stat Release 5.1 data analysis programs for UNIX and MSDOS. Behavior Research Methods, Instruments and Computers 18.2. 168-76. Petersen, Hjalmar, Jogvan i Lon Jacobsen, Zakaris Hansen and Hoskuldur Thrainsson. 1998. Faroese. An Overview for Students and Researchers. Ms., Torshavn & Reykjavik. Pierrehumbert, Janet. 1980. The phonetics and phonology of English intonation. Ph.D. dissertation, MIT. Distributed 1987, Indiana University Linguistics Club. Pierrehumbert, Janet. 1995. Prosodic effects on glottal allophones. In O. Fujimura and M. Hirano, (eds.), Vocal fold physiology: voice quality control. San Diego: Singular Publishing Group. 39-60. Pierrehumbert, Janet and D. Talkin. 1992. Lenition of /h/ and glottal stop. In G. Doherty and D. Ladd (eds.), Papers in Laboratory Phonology II: gesture segment prosody. Cambridge: Cambridge University Press. 90-117. Posner, Rebeca. 1961. Consonantal Dissimilation in the Romance Languages. Oxford: Basil Blackwell. Postal, P. 1969. Mohawk vowel doubling. International Journal of American Linguistics 35. 291-98. Prince, Alan. 1976. Applying stress. Ms., Department of Linguistics, University of Massachusetts, Amherst. Prince, Alan. 1980. A metrical theory for Estonian quantity. Linguistic Inquiry 11. 511-62. Prince, Alan. 1983. Relating to the grid. Linguistic Inquiry 14. 19-100. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 366 Prince, Alan and Paul Smolensky. 1993. Optimality Theory: constraint interaction in generative grammar. Ms, Rutgers University, New Brunswick and University of Colorado, Boulder. [Technical report #2, Rutgers University Center for Cognitive Science. To appear, Cambridge, Mass.: MIT Press.] Quilis, Antonio. 1981. Fonetica Acustica de la Lengua Espahola. Madrid: Editorial Gredos Biblioteca Romanica Hispanica. Rigsby, Bruce and John Ingram. 1990. Obstruent voicing and glottalic obstruents in Gitksan. International Journal o f American Linguistics 56. 251-63. Riad, Thomas. 1992. Structures in Germanic Prosody: A Diachronic Study with Special Reference to the Nordic languages. Ph. D. Dissertation, Stockholm University, Stockholm. Ringen, Catherine. 1999. Aspiration, preaspiration, deaspiration, sonorant devoicing and spirantization in Icelandic. Nordic Journal o f Linguistics 22. 137-56. Roca, Iggy and Wyn Johnson. 1999. A Course in Phonology. Oxford: Blackwell. Romero, Joaquin. 1995. Gestural Organization in Spanish: An Experimental Study o f Spirantization and Aspiration. Ph.D. dissertation, University of Connecticut. Rose, Sharon. 1996. Variable laryngeals and vowel lowering. Phonology 13, 73-117. Safir, Ken. 1979. Metrical Structure in Capanahua. In Ken Safir (ed.), MIT Working papers in Linguistics, vol. 1. Papers on Syllable Structure, Metrical Structure and Harmony Processes ’. Cambridge, Mass: MIT. Saltarelli, Mario. 1970. A phonology o f Italian in a generative grammar. The Hague: Mouton. Saltarelli, Mario. In press. A constraint interaction theory of Italian raddoppiamento. In Rafael Nunez Cedeno (ed.), Proceedings o f LSRL 31. Amsterdam: John Benjamins. Samareh, Yadollah. 1977. Arrangement o f segmental phonemes in Farsi. University o f Tehran, Tehran University Publications. Sapir, Edward. 1925. Sound Patterns in Language. Language 1. 37-51. Sapir, Edward. 1938. Globalized continuants in Navaho, Nootka, and Kwakiutl (with a note on Indo-European). Language 14. 248-74. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 367 Sauna, D. 1982. Studies in Tibeto-Himalayan Linguistics: A descriptive analysis of Pattani, a dialect of Lahaul. Hoshiarpur: Vishveshvaranand Vishva Bandhu Institute of Sanscrit and Indological Studies, Panjab University. SAS Institute Inc. [Statistical Analysis System]. 1999. StatView Reference Manual (3r d edition.). Schachter, Paul, and Fe Otanes. 1972. Tagalog Reference Grammar. University of California Press, Berkeley. Selkirk, Elizabeth. 1980. The role of prosodic categories in English word stress. Linguistic Inquiry 11. 563-605. Sharma, Jagdish Grander. 1979. Gojri Phonetic Reader. CIIL Phonetic Reader Series 19. Mysore: Central Institute of Indian Languages. Sharma, Jagdish Chander. 2002. Gojri and its relationship with Rajasthani, etc. In M. Thirumalai (ed.), Language in India, vol. 2, 2 april 2002 [http://www.languageinindia.com/april2002/jcsharmagojri.html] Shell, Olive A. 1975. Estudios Panos III: Las Lenguas Pano y su Reconstruccion. Yarinacocha: Instituto Linguistico de Verano. Version espanola by Ezequiel Romero Sanchez-Concha. Sherer, Tim. 1994. Prosodic phonotactics. PhD dissertation, University of Massachusetts, Amherst [ROA 54, http://roa.rutgers.edu]. Silverman, Daniel. 1997. Phasing and recoverability. Outstanding Dissertations in Linguistics Series. New York: Garland. Silverman, Daniel. 2002. On the rarity of pre-aspirated stops. Handout presented at the 10t h Manchester Phonology Meeting, May 25, 2002. Smith, Caroline. 1997. The devoicing of /z/ in American English: effects of local and prosodic context. Journal of Phonetics 2 5 ,471-500. Smith, Jennifer. 2000. Prominence, augmentation and neutralization in phonology. In L. Conathan, J. Good, D. Kavitskaya, A. Wulf and A. Yu (eds.), Proceedings o f the Twenty-Sixth Annual Meeting o f the Berkeley Linguistics Society, Feb 18-21, 2000. General Session and Parasession on aspect. Berkeley Linguistics society, Berkeley. Smith, Jennifer. 2002. Phonological Augmentation in Prominent Positions. PhD dissertation, Umass, Amherst [ROA 570, http://roa.rutgers.edu]. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 368 Stark, Sharon, and Polly Machin. 1977. Stress and tone in Tlacoyalco Popoloca. In William Merrifield (ed.), Studies in Otomanguean Phonology. Dallas: SIL and University of Texas at Arlington. 69-91. Stemberger, Joseph Paul. 1993. Glottal transparency. Phonology 10. 107-38. Steriade, Donca. 1987. Locality conditions and feature geometry. Proceedings o f the North East Linguistics Society 17, vol. 2. 595-617. Steriade, Donca. 1993. Closure, release, and nasal contours. In Marie Huffman and Rena Krakow (editors), Phonetics and phonology 5: nasals, nasalization, and the velum. San Diego: Academic Press. 401-70. Steriade, Donca. 1994. Complex onsets as single segments: the Mazateco pattern. In Jennifer Cole and Charles Kisseberth (eds.), Perspectives in Phonology. Standford: CSLI Publications. 203-91. Steriade, Donca. 1997. Phonetics in phonology: the case of laryngeal neutralizations. Unpublished manuscript, UCLA. Suzuki, Keiichiro. 1998. A typological investigation o f dissimilation. Ph.D. Dissertation, University of Arizona. Thrainsson, Hoskuldur. 1978. On the phonology of Icelandic preaspiration. Nordic Journal o f Linguistics 1. 3-54. Thompson, Laurence. 1979. Salishan and the Northwest. In L. Campbell and M. Mithun (eds.), The languages o f native America: Historical and comparative assessment. Austin: University of Texas Press. 692-765. Thompson, Laurence and M. Thompson. 1992. Thompson language. University of Montana Occasional Papers in Linguistics, 8. Missoula: Linguistics Laboratory, University of Montana. Thompson, Laurence, M. Thompson and S. Edgesdal. 1996. Sketch of Thompson, a Salishan language. In I. Goddard (ed.), Handbook o f North American Indians, vol. 17, Languages. Washington: Smithsonian institution. 609-43. Thompson, Laurence, M. Thompson, and B. Efrat. 1974. Some phonological developments in Straits Salish. International Journal o f American Linguistics 40. 182-96. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 369 Turk, Alice. 1992. The American English flapping rule and the effect of stress on stop consonant durations. Working Papers o f the Cornell Phonetics Laboratory 1. 103-33. Turk, Alice. 1993. Effects of Position-in-Syliable and Stress on Consonant Articulation. Cornell Ph. D. Dissertation. Umeda, N. 1977. Consonant Duration in American English. Journal o f the Acoustical Society o f America 61. 846-58. Van de Vijver, Ruben. 1998. The iambic Issue. Iambs as a result of constraint interaction. [HIL dissertations 37], Leiden: Holland Institute of Generative Linguistics. Vaysman, Olga. 2002. Two dialects of Nganasan: what they tell us about prosody. GLOW 25 abstract [http://www.meertens.knaw.nl/glow2002/vaysman.pdf ]. Veerman-Leichsenring, Annette. 1984. El Popoloca de Los Reyes Metzontla. Revue d ’ ethnolinguistique amerindienne, n. 4. Paris: A.E.A. Venneman, Theo. 1988. Preference laws for syllable structure. Berlin: Mouton de Gruyter. Walker, Rachel. 2001. Positional markedness in vowel harmony. In Caroline Fery, Antony Dubach Green and Ruben Van de Vivjer (eds.), Proceedings o f HILP 5. Linguistics in Potsdam. Vol 12. University of Potsdam. 212-32. Wares, Alan Campbell. 1968. A Comparative Study o f Yuman Consonantism. The Hague: Mouton. Whorf, Benjamin. 1946. The Hopi language, Toreva dialect. In Harry Hoijer (ed.), Linguistic Structures o f Native America. Viking Fund publications in Antrhopology 6. New York: Viking Fund. 158-83. Wiese, Richard. 1996. The Phonology o f German. Oxford: Clarendon Press. Wright, Joseph. 1957. Grammar of the Gothic Language. Oxford: Clarendon Press. Yallop, C. 1977. Alyawarra: an Aboriginal Language o f Central Australia. Canberra, Australian Institute of Aboriginal Studies. Yar-Shater, E. 1969. A Grammar o f Southern Tati dialects. Columbia University Publications in Near and Middle Eastern Studies, Series B3. The Hague: Mouton. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 370 Yip, Moira. 1991. Coronals, consonant clusters, and the coda condition. In Carole Paradis and Jean-Franfois Prunet (eds.). The special status o f coronals: internal and external evidence. San Diego: Academic Press. Zee, Draga. 1988. Sonority Constraints on Prosodic Structure. Doctoral dissertation, Stanford University. Zee, Draga. 1994. Footed Tones and Tonal Feet: Rhythmic Constituency in a Pitch Accent Language. Ms., Cornell University. Zee, Draga. 1995. Sonority constraints on syllable structure. Phonology 12, 85-129. Zue, Victor, and Marta Laferriere. 1979. Acoustic Study of medial It, d/ in American English. In Journal of the Acoustic Society of America, 66(4), 1039-50. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 371 APPENDIX Languages mentioned in main text. Language Language Family Alternations References Alutiik Yupik Yupik Gemination (4.1) Fortition (4. 3) Van de Vijver 1998 Alyawarra Arandic Stress-attraction (6.2) Yallop 1977, de Lacy 2001 American English Indo-European: Germanic Flapping (4.2) Lisker and Abranson 1967, Umeda 1977, Kiparsky 1979, Kahn 1980, Borowsky 1986, Crystal and House 1988, Turk 1992, 1996, Hayes 1995, Davis 2002 Aranda Arandic Stress-attraction (6.2) Breen & Pensalfini 1999 Bagneres-de- Luchon French Indo-European: Romance Attraction (6.2) Grammont 1933, Blevins and G arrett 1998 Banawa Arauan Stress-attraction (6.2) Buller, Buller and Everett 1993, Everett 1996 Basque Spanish Indo-European: Romance Aspiration (5.2.1) Gonzalez 2002b Canadian English Indo-European: Germanic Flapping (4.2) De Hasebe Ludt 1987 Capanahua Panoan Deletion (7.1) Metathesis (8) Loos 1969, 1999, Safir 1973, Gonzalez 2002a Cayuga Northern Iroquoian Deletion (7.1) Chafe 1977, Foster 1982, Michelson 1988, Dougherty 1993, Blevins and G arrett 1998 Chali (Tati) Indo-Iranian Post-aspiration (5.2.1) Deletion (7.1) Yar-Shater 1969 Cocopa Hokan Fortition (4.3) Wares 1968 Colville Interior Salish Attraction (6.1) Mattina 1979 Coast Tshimshian Penutian Glottalization (5.3) Attraction (6.1) Dunn and Hayes 1983, Dunn 1995, Kehrein 2001, Howe and Pulleyblank 2001 Copala Trique Otomanguean Lengthening (4.1) Lenition (4.4) Voicing (5.1) Hollenbach 1977 Danish West Scandinavian Attraction (6.1) Fisher-Jorgersen 1989, Zee 1988 Diegueno Hokan Fortition (4.3) Wares 1968 Djabugai Pama-Nyungan Flapping (4.2) Voicing (5.1) Patz 1991 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 372 Dutch Indo-European: Germanic Epenthesis (7.2) Booij 1995, Smith 2002 English Indo-European: Germanic Lengthening (4.1) Lenition (4.4) Voicing (5.1) Post-aspiration (5.2.1) Deletion (7.1) Borowsky 1986, Clements 1987, Hayes 1995, Dilley and Shattuck-Hufnagel 1996, Hammond 1999, Davis 1999,2002, Lavoie 2001 Faroese West Scandinavian Pre-aspiration (5.2.2) Petersen et alia 1998 Farsi Indo-Iranian Fortition (4.3) Voicing (5.1) Post-aspiration (5.2.1) Samareh 1977 German Indo-European: Germanic Post- aspiration (5.2.1) Epenthesis (7.2) Kohler 1977, Gegerich 1989, Iverson and Salmons 1995, Alber to appear Gitksan Penutian Glottalization (5.3) Rigsby and Ingram 1990, Kehrein 2001 Gothic Eastern Germanic Dissimilation (9) Mosse 1956, Collinge 1985, Chomsky and Halle 1968, W right 1957 Greek Indo-European: Greek Lengthening (4.1) Botinis, Fourakis and Bannert 2001 Guayabero Guahiban Sec. articulation (4.5) Fortition (4.3) Lenition (4.4) Keels 1985 Guelavia Zapotec Otomanguean Lengthening (4.1) Jones and Rnudson 1977 Gundidj Victorian Gemination (4.1) Hercus 1986 Huariapano Panoan Epenthesis (7.2) Parker 1994, 1998 Icelandic West Scandinavian Pre-aspiration (5.2.2) Thrainsson 1974, Silverman 1997 Ingush North-East Caucasian Pre-aspiration (5.2.2) Nichols 1994 Iowa-Oto Siouan Stress attraction (6.2) Downing 1998 Irish Celtic Pre-aspiration (5.2.2) Ni Chasaide and Docherty 1984 Italian Indo-European: Romance Gemination (4. 1) Chierchia 1982, Loporcaro 1999, Saltarelli in press Kiliwa Hokan Fortition (4.3) Wares 1968 Kupia Indo-Iranian Flapping (4. 2) Lenition (4.4) Christmas and Christmas 1975 Le Havre French Indo-European: Romance Metathesis (8) Grammont 1909, Blevins and G arrett 1998 Lilloet Salishan Glottalization (5.3) Howe and Pulleyblank 2001 Maori Austronesian Lengthening (4.1) Fortition (4.3) Post-aspiration (5.2.1) Bauer 1993 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 373 Maricopa Hokan Fortition (4.3) Wares 1968 Middle English Indo-European: Germanic Voicing (5.1) Kabell and Lauridsen 1984 Mohave Hokan Fortition (4.3) Wares 1968 Mohawk Northern Iroquoian Metathesis (8) Michelson 1988 Narinari Victorian Gemination (4.1) Hercus 1986 Nganasan Uralic Lenition (4.4) Helimsky 1998, Vaysman 2002 North Yuman Hokan Fortition (4.3) Wares 1968 Northern Faroese West Scandinavian Pre-aspiration (5.2.2) Petersen et alia 1998, Kehrein 2001, Ladefoged and Maddieson 1996 Norton Sound Yupik Central Yupik Gemination (4.1) Fortition (4.3) Jakobson 1985, Leer 1985, Van de Vivjer 1998 Ojibwa Algonquian Pre-aspiration (5.2.1) Bloomfield 1956 Old-English Indo-European: Germanic Deletion (7.1) Kabell and Lauridsen 1984 Oneida Northern Iroquoian Deletion (7.1) Metathesis (8) Lounsbury 1942, Michelson 1988, Chafe 1977, Blevins and G arrett 1998 Paamese Austronesian Lenition (4.4) Voicing (5.1) Crowley 1982 Paipai Hokan Fortition (4.3) Epenthesis (7.2) Wares 1968 Pattani Sino-Tibetan Post-aspiration (5.2.1) Sarma 1982 Piraha Mura Stress attraction (6.2) Everett and Everett 1984ab, Everett 1988 Popoloca Otomanguean Gemination (4.1) Voicing (5.1) Glottalization (5.3) Stark and Machin 1977, V eerman-Leichsenring 1984 Proto-Germanic Proto-Germanic Voicing (5.1) Wright 1957 Romance Indo-European Dissimilation Grammont 1933, Posner 1961 Saanich Salishan Glottalization (5.3) Howe and Pulleyblank 2001 Scots Gaelic Celtic Pre-aspiration (5.2.1) Ni Chasaide and Dochartaigh 1984 Senoufo Niger-Congo Lengthening (4.1) Flapping (4.2) Lenition (4.4) Sec. articulation (4.5) Voicing (5.1) Mills (1984) Shuswap Interior Salish Attraction (6.1) Kuipers 1974, Steriade 1997 Silacayoapan Mixteco Otomanguean Gemination (4. 1) Lenition (4.4) North and Shields 1977 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 374 Voicing (5.1) Post-aspiration (5.2.1) Somali Cushitic Gemination (4.1) Lenition (4.4) Armstrong (1964) Spanish Indo-European: Romance Lengthening (4.1) Lenition (4.4) Quilis 1981 Squamish Central Salish Fortition (4.3) Post-aspiration (5.2.1) Kuipers 1967 Standard Faroese West Scandinavian Pre-aspiration (5.2.1) Kehrein 2001, Ladefoged and Maddieson 1996 Taipai Hokan Fortition (4.3) Wares 1968 Tarascan Tarascan Pre-aspiration (5.2.1) Foster 1969, Silverman 2002 Thompson River Salish Interior Salish Attraction (6.1) Dunn and Hays 1983 Toreva Hopi Uto-Aztecan Pre-aspiration (5.2.2) Whorf 1946 Turkish Altaic Lengthening (4.1) Jonathan Barnes p.c. Twana Central Salish Attraction (6.1) Thompson 1979, Blevins and G arrett 1998 Urubu-Kaapor Tupi-Guarani Lengthening (4.1) Voicing (5.1) Kakamasu 1986 Walapai Hokan Fortition (4.3) Wares 1968 Wasco-Wisram Penutian Voicing (5.1) Sapir 1925 Wembawemba Victorian Voicing (5.1) Hercus 1986 Wergaia Victorian Gemination (4.1) Voicing (5.1) Hercus 1986 West Tarangan Austronesian Fortition (4.3) Nivens 1992 Yuman Hokan Fortition (4.3) Wares 1968 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

On the identification of null arguments

PDF

The fine structure of negative polarity items in Chinese

PDF

The independence of binding and intensification

PDF

Linguistic attitudes in the Basque Country: The social acceptance of a new variety

PDF

Two types of dependency

PDF

The beginnings of language loss in discourse: A study of American Lithuanian

PDF

The effect of non-native speaker status on the use of linguistic accommodation by native speakers of Japanese: Implications for communication accommodation theory

PDF

Syntactic and non -syntactic scope

PDF

Seeing ghosts: Readings on the effectivity of phantasms

PDF

Form and meaning: Negation and question in Chinese

PDF

The face of others, the taste of things: Photogenie and sensation in silent cinema

PDF

Overt focus movement and minimal information marking

PDF

Discourse functional units: A re-examination of discourse markers with particular reference to Spanish

PDF

At wit's end: The rhetoric of humor and the ends of talk

PDF

Second language acquisition of Spanish morpho -syntax by Quechua -speaking children

PDF

The modal preterite phenomenon (MPP) in colloquial American English: A diachronic and synchronic analysis

PDF

The problem of space remains: How Colette and Virginia Woolf use text, narrative, and style to rewrite gendered space

PDF

Natan Altman and the problem of Jewish art in Russia in the 1910s

PDF

Pushkin's taste and taste for Pushkin: Toward a reconstruction of taste mechanisms

PDF

A descriptive grammar of San Bartolome Zoogocho Zapotec

Asset Metadata

Creator Gonzalez, Carolina (author)

Core Title The effect of stress and foot structure on consonantal processes

Degree Doctor of Philosophy

Degree Program Linguistics

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag language, linguistics,OAI-PMH Harvest

Language English

Contributor Digitized by ProQuest (provenance)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-364564

Unique identifier UC11339741

Identifier 3103894.pdf (filename),usctheses-c16-364564 (legacy record id)

Legacy Identifier 3103894.pdf

Dmrecord 364564

Document Type Dissertation

Rights Gonzalez, Carolina

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA