Bilingual Rhotics in Two Island Communities of the Archipelago of San Andres, Colombia

LANGUAGE CONTACT AND CONVERGING PATHS OF VARIATION: BILINGUAL RHOTICS IN TWO ISLAND COMMUNITIES OF THE ARCHIPELAGO OF SAN ANDRES, COLOMBIA

FALCON D. RESTREPO-RAMOS

A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA

2019

Jezu, Ufam Tobie!

ACKNOWLEDGMENTS

I would like to start by thanking my dissertation committee for believing in this project and for their invaluable support throughout all stages. My mentor, Dr. Aaron, for her insightful guidance and knowleadgeable comments and suggestions, which assured the successful completion of this dissertation. I hope I can carry on your example. Dr. Gillian Lord for being a beacon of light and hope through all these years. Dr. Valdés-Kroff for always looking forward to the success of this work. Your guidance will have a lasting impact on my career. Dr. Essegbey for being a voice of reason and being readily available for help.

It goes without saying that I am greatly indebted to all the spectacular faculty, staff, and colleagues at the Department of Spanish and Portugues Studies for all their support and consideration. Five years ago, I arrived at the University of Florida with many dreams and barely enough to support myself, but I was welcomed with a smile and the opportunity to become part of the Department. It was there that I had a sense of the great things that were coming in my future. I look back and I can’t thank you enough Prof. Gillian Lord. It was in great part thanks to you that I was able to fulfill my dreams in a great place. Precisely through this academic setting,

I met my lovely girlfriend, Nofiya Denbaum, whose priceless encouragement, confidence and support made these years even more wonderful. Those Thanksgivings with you and your family gave me the warmest memories throughout these years. Thanks for opening the doors of your house and heart.

I also appreciate the help that I received from external faculty. Dr. Stefanie Wulff, for always being available to discuss how to improve the project. Dr. Aaron Broadwell for advising my own fieldwork. Dr. Andrew Lotto, who introduced me to phonetic analyses that were new to me. I would also like to thank the Center for Latin American studies for partially funding this project through a Tinker Foundation grant. Finally, I want to recognize the marvelous people of

the Archipelago of San Andres for whom this work has been done. I hope this project benefits these communities and that more scholars carry on what I and many others have started. Last but not least, my family and friends for their love, friendship and prayers.

TABLE OF CONTENTS

page

ACKNOWLEDGMENTS ...... 4

LIST OF TABLES ...... 8

LIST OF FIGURES ...... 12

ABSTRACT ...... 16

CHAPTER

1 INTRODUCTION ...... 18 1.1 Historical and Sociolinguistic Background ...... 18 1.1.1 The Archipelago Of San Andres Between 1618-1810 ...... 20 1.1.2 The Archipelago Of San Andres Between 1810 And 1886 ...... 21 1.1.3 The Archipelago Of San Andres Between 1886 And The Present ...... 22 1.1.4 The Past And Present Of Islander Creole ...... 28 1.2 The Current Study ...... 36 1.2.1 Identification Of The Linguistic Varieties Under Study ...... 37 1.2.2 Studying Sound Variation And Change In This Contact Scenario: Rhotics ...... 38

2 REVIEW OF THE LITERATURE ...... 41 2.1 Language Contact and the Comparative Method in the Archipelago ...... 42 2.2 Rhotic Variation In Monolingual And Bilingual Settings ...... 53 2.2.1 Variable Production Of Rhotics In Contact Situations: How To Account For The Articulation And Distribution Of Non-Vibrant Rhotics In The Archipelago Of San Andres ...... 59 2.2.2 Enhancing Our Understanding Of Contact Phonologies In Spanish Vibrant Rhotics ...... 64 2.3 Summary ...... 68

3 METHODOLOGY ...... 69 3.1 Data Collection...... 69 3.1.1 Participants ...... 70 3.1.2 Data Elicitation Tasks ...... 71 3.2 Data Analysis ...... 72 3.2.1 Preparation Of The Datasets For The Speech Analysis ...... 72 3.2.2 Phonetic Analysis...... 73 3.2.3 Variable Rule Analysis ...... 74 3.2.4 Envelope Of Variation ...... 75 3.2.5 Coding Of The Sociolinguistic Data ...... 80

4 A SOCIOPHONETIC ANALYSIS OF NON-VIBRANT RHOTICS IN THE ARCHIPELAGO OF SAN ANDRES ...... 83

4.1 Acoustic Properties of the Non-Vibrant Rhotics in the Varieties of the Archipelago ...... 86 4.1.1 The Non-Vibrant Rhotics In The Archipelago ...... 88 4.1.2 Spectral Moments In The Non-Vibrant Rhotics Of The Archipelago ...... 89 4.1.3 Formant Frequencies In The Rhotics Of The Archipelago ...... 94 4.1.4 Segmental Duration In The Rhotics Of The Archipelago ...... 98 4.1.4.1 A comparison of segmental duration in the rhotics of the Archipelago ...98 4.1.4.2 A comparison of segmental duration in non-vibrant taps and trills between Raizal and Continental Spanish ...... 105 4.2 Discriminant Function Analysis ...... 114 4.3 Correlation Analysis Between Generations Of Raizal Spanish And Continental Spanish/Raizal Creole...... 122

5 COMPARATIVE VARIATIONIST ANALYSIS OF RHOTIC VARIATION IN THE ARCHIPELAGO OF SAN ANDRES ...... 140 5.1 Distribution Of Rhotic Variants ...... 142 5.1.1 Distribution Of Rhotic Variants In Raizal Spanish...... 143 5.1.2 Distribution Of Rhotic Variants In Continental Spanish ...... 147 5.2 Variable Rule Analysis ...... 150 5.2.1 Preparation Of The Datasets ...... 150 5.2.2 First Generation ...... 155 5.2.2 Second Generation ...... 166 5.2.3 Third Generation ...... 175 5.2.4 Continental Spanish ...... 185 5.3 A Tale Of Distant Stories: Bilingual Raizal Spanish And Monolingual Continental Spanish Compared ...... 198

6 DISCUSSION ...... 209 6.1 Variation And Change In Non-Vibrant Rhotic Production...... 209 6.2 Variation And Change In Vibrant Tap And Trill Production ...... 217

7 CONCLUSION...... 223 7.1 Theoretical Implications...... 223 7.2 Limitations ...... 226 7.3 Future directions...... 227

APPENDIX

FIELDWORK AND MATERIALS ...... 231 Fieldwork ...... 231 Materials ...... 232

LIST OF REFERENCES ...... 234

BIOGRAPHICAL SKETCH ...... 248

LIST OF TABLES

Table page

3-1 Generations of Raizal Spanish informants ...... 71

3-2 Contextual distribution of rhotics in Spanish...... 75

3-3 Distribution of canonical and non-canonical rhotics according to the position in the word in Raizal Spanish...... 78

3-4 Linguistic variables and categories included in the quantitative analysis of vibrant rhotics...... 81

3-5 Extralinguistic variables and categories included in the quantitative analysis of vibrant rhotics...... 82

4-1 Mean values and standard deviations for spectral moment measurements in all varieties...... 88

4-2 Mean values and standard deviations for formant frequencies measurements in all varieties...... 89

4-3 A comparison of COG values reported for English sibilants, Spanish fricative trills and the rhotics in the Archipelago...... 91

4-4 A comparison of skewness values reported for English sibilants, Spanish fricative trills and the rhotics in the Archipelago...... 92

4-5 Intervocalic rhotics in Raizal Spanish based on rhotic stress...... 107

4-6 Intervocalic taps and trills in Raizal Spanish based on rhotic stress...... 108

4-7 ANOVA output of duration and the interaction between the type of rhotic and stress. ..109

4-8 Tukey Posthoc of the interaction effect between the type of rhotic and word stress across generations...... 109

4-9 Results of the Tukey’s Post Hoc tests for each generation and the non-significant p- values for each combination of taps and trills in their stress context...... 112

4-10 Tests of equality of group means ...... 115

4-11 Test of null hypothesis of equal population covariance matrices with a significant value lesser than 0.001...... 116

4-12 Structure matrix of individual predictors with weighted predicting capabilities...... 117

4-13 Final classification results...... 119

4-14 Linear, mixed effects model of segmental duration between Raizal Creole approximants and Spanish trills in generations of Raizal Spanish informants and Continental Spanish...... 127

4-15 Linear, mixed effects models of F3 between Raizal Creole approximants and Spanish taps (left) and trills (right) in generations of Raizal Spanish informants and Continental Spanish...... 130

4-16 Linear, mixed effects models of the distance between F3-F2 between Raizal Creole approximants and Spanish taps (left) and trills (right) in generations of Raizal Spanish informants and Continental Spanish...... 131

4-17 Linear, mixed effects models of F3 frequencies between Raizal Creole approximants and Spanish trills in generations of Raizal Spanish informants and Continental Spanish...... 134

4-18 Linear, mixed effects models of F3-F2 distance frequencies between Raizal Creole approximants and Spanish taps in generations of Raizal Spanish informants and Continental Spanish in Old Providence and San Andres...... 136

4-19 Linear, mixed effects models of F3-F2 distance frequencies between Raizal Creole approximants and Spanish trills in generations of Raizal Spanish informants and Continental Spanish in Old Providence and San Andres...... 138

5-1 Overall distribution of canonical and non-canonical trills and taps in Raizal Spanish. ..143

5-2 Relative frequencies of canonical and non-canonical rhotics by generation...... 145

5-3 Overall distribution of canonical and non-canonical trills and taps in Continental Spanish...... 147

5-4 Interaction between linguistic variable pairs in the tap and trill datasets...... 151

5-5 Simplified codification of linguistic factor variables...... 152

5-6 Interaction between linguistic variable pairs in the tap and trill datasets...... 153

5-7 Final test of Interaction between linguistic variable pairs in the tap and trill datasets. ...154

5-8 Significant linguistic factors contributing to canonical taps in first generation informants ...... 156

5-9 Non-significant linguistic factors contributing to canonical taps in first generation informants...... 157

5-10 Linguistic factors contributing to vibrant trills in first generation informants...... 161

5-11 Non-significant linguistic factors contributing to vibrant trills in first generation informants...... 163

5-12 Extralinguistic factors contributing to canonical taps in first generation informants...... 164

5-13 Social factors contributing to vibrant trills in first generation informants...... 165

5-14 Significant linguistic factors contributing to canonical taps in second generation informants...... 167

5-15 Non-significant linguistic factors contributing to canonical taps in second generation informants...... 168

5-16 Significant linguistic factors contributing to vibrant trills in second generation informants...... 170

5-17 Non-significant linguistic factors contributing to vibrant trills in second generation informants...... 172

5-18 Social factors contributing to canonical taps in second generation informants...... 173

5-19 Social factors contributing to vibrant trills in second generation informants...... 174

5-20 Significant Linguistic factors contributing to canonical taps in third generation informants...... 176

5-21 Non-significant Linguistic factors contributing to canonical taps in third generation informants...... 178

5-22 Significant linguistic factors contributing to vibrant trills in third generation informants...... 179

5-23 Non-significant linguistic factors contributing to vibrant trills in third generation informants...... 180

5-24 Non-significant linguistic factors contributing to canonical trills in third generation informants...... 181

5-25 Non-significant linguistic factors contributing to canonical trills in third generation informants...... 182

5-26 Significant extralinguistic factors contributing to canonical taps in third generation informants...... 183

5-27 Significant extralinguistic factors contributing to vibrant trills in third generation informants...... 184

5-28 Significant linguistic factors contributing to canonical taps in Continental Spanish...... 186

5-29 Non-significant linguistic factors contributing to canonical taps in Continental Spanish...... 187

5-30 Mixed-effects model of the non-significant linguistic factors contributing to vibrant trills in Continental Spanish...... 190

5-31 Fixed effect model of the significant linguistic factors contributing to vibrant trills in Continental Spanish...... 191

5-32 Significant linguistic factors contributing to canonical trills in Continental Spanish. ....192

5-33 Non-significant linguistic factors contributing to canonical trills in Continental Spanish...... 193

5-34 Significant social factors contributing to canonical taps in Continental Spanish...... 195

5-35 Significant social factors contributing to vibrant trills in Continental Spanish...... 196

5-36 Variable rule analyses of the contribution of factors selected as significant to occurrence of vibrant taps in generations of bilingual Raizales and Continental Spanish...... 199

5-37 Factors' direction of effect by data set (vibrant taps)...... 200

5-38 Variable rule analyses of the contribution of extralinguistic factors selected as significant favoring the occurrence of vibrant taps in generations of bilingual Raizales and Continental Spanish...... 202

5-39 Factors' direction of effect by data set (social variables for taps)...... 203

5-40 Variable rule analyses of the contribution of factors selected as significant to occurrence of vibrant trills in generations of bilingual Raizales and Continental Spanish...... 204

5-41 Factors' direction of effect by data set (vibrant trills)...... 205

5-42 Variable rule analyses of the contribution of factors selected as significant to occurrence of vibrant trills in generations of bilingual Raizales and Continental Spanish...... 206

5-43 Factors' direction of effect by data set (social variables for vibrant trills)...... 207

LIST OF FIGURES

Figure page

1-1 (Left) Archipelago of San Andres in the Western Caribbean. (Right) A close-up of the islands of the Archipelago (Google Earth, 2018)...... 19

1-2 The Archipelago of San Andres and the Corn Islands...... 23

1-3 Main traffic ports of slaves and the main local languages (Holm, 1978; Le page and De Camp, 1960, taken from Dittman, 1992) ...... 31

1-4 Island of San Andres with the location of the ethnic neighborhoods and the downtown areas...... 33

1-5 Island of Old Providence with the location of the ethnic neighborhoods and the downtown areas...... 35

3-1 Geographical distribution of the informants ...... 69

3-3 Tap with a normative lingual closure (left). Tap displaying no tongue vibration (right)...... 76

3-4 Trill with the normative lingual closures (left). Trill displaying no tongue vibration (right)...... 76

3-5 Trill displaying one lingual closure with a subsequent formant structure...... 77

3-6 Visual representation of the distribution of canonical and non-canonical rhotics according to the position in the word in Raizal Spanish...... 79

4-1 Praat capture of the rhotic produced in initial position in the word [ɻif] ‘reef’ in Islander Creole...... 84

4-2 Praat capture of the zero-occlusion (i.e., non-vibrant) rhotic produced in initial position in the word [ɹajsal] ‘raizal’ in Continental Spanish...... 85

4-3 Praat capture of the rhotic produced in initial position in the word [ɹekwedo] ‘recuerdo’ in Raizal Spanish...... 85

4-4 Boxplot with mean values of the first spectral moment (COG) in all varieties...... 90

4-5 Density estimate and mean values of the second spectral moment...... 91

4-6 Boxplot with mean values for Kurtosis in the three varieties of the Archipelago...... 93

4-7 Boxplot with mean values of formant frequencies in all varieties...... 95

4-8 Scatterplot with regression line of formant frequencies in Raizal Spanish, Raizal Creole, and Continental Spanish in San Andres and Old Providence...... 96

4-9 Boxplot with mean values for segmental duration in all varieties in all word contexts. ...99

4-10 Mean duration in milliseconds across population groups and position of the rhotic within the word...... 100

4-11 Spectogram of word-final rhotic in Raizal Creole with different contiguous elements. .102

4-12 Differences in mean rhotic duration based on word stress across the varieties of the Archipelago...... 103

4-13 Tap/trill contrast in Continental Spanish and Raizal Spanish...... 106

4-14 Mean duration of approximant taps and trills in intervocalic position in Raizal Spanish...... 106

4-15 Intervocalic rhotics in Raizal Spanish according to word stress...... 108

4-16 Mean duration of approximant taps and trills in generations of bilingual Raizales and Continental Spanish according to word stress...... 110

4-17 Hierarchy of non-significant factor combinations of stress and tap/trill segments...... 113

4-18 Graphic representation of population differences in terms of the acoustic correlates submitted for the analysis...... 118

4-19 Distribution of non-vibrant rhotics in the varieties of the Archipelago...... 124

4-20 Distribution of non-vibrant taps and trills in Raizal Spanish in both islands of the Archipelago: Old Providence and San Andres...... 125

4-21 A comparison of rhotic duration between varieties...... 125

4-22 A comparison of tap and trill duration between Spanish varieties and Raizal Creole. ....126

4-23 A comparison of F3 and F3-F2 frequencies between Spanish varieties and Raizal Creole...... 128

4-24 A comparison of F3 and F3-F2 frequencies in taps and trills between Spanish varieties and Raizal Creole (left for taps and right for trills)...... 129

4-25 A comparison of F3 frequencies in trills between Spanish varieties and Raizal Creole (left for taps and right for trills) in San Andres and Old Providence...... 133

4-26 A comparison of F3-F2 distance in taps between Spanish varieties and Raizal Creole in San Andres and Old Providence...... 135

4-27 A comparison of F3-F2 distance in trills between Spanish varieties and Raizal Creole in San Andres and Old Providence...... 137

5-1 Overall distribution of rhoticity and non-rhoticity categories in the varieties of the Archipelago...... 142

5-2 Distribution of canonical (left) and non-canonical (right) trills and taps in Raizal Spanish across generations...... 144

5-3 Distribution of canonical and non-canonical trills and taps in Raizal Spanish between islands...... 145

5-4 Distribution of canonical and non-canonical trills and taps in Raizal Spanish based on sex...... 146

5-5 Distribution of canonical (left) and non-canonical (right) trills and taps in Continental Spanish between age groups and place of birth...... 148

5-6 Distribution of canonical and non-canonical trills and taps in Continental Spanish based on sex...... 149

5-7 Conditional inference tree of the significant linguistic factors for taps in first generation Raizales...... 158

5-8 Conditional inference trees of the significant linguistic factors for trills in first generation Raizales...... 162

5-9 Conditional inference tree of the significant social factors for taps in first generation Raizales...... 164

5-10 Conditional inference tree of the significant social factors for trills in first generation Raizales...... 166

5-11 Conditional inference tree of the significant social factors for taps in second generation Raizales...... 169

5-12 Conditional inference tree of the significant linguistic factors for trills in second generation Raizales...... 171

5-13 Conditional inference tree of the significant social factors for taps in second generation Raizales...... 173

5-14 Conditional inference tree of the significant social factors for trills in second generation Raizales...... 175

5-15 Conditional inference tree of the significant linguistic factors for taps in third generation Raizales...... 177

5-16 Conditional inference tree of the significant linguistic factors for trills in third generation Raizales...... 179

5-17 Conditional inference tree of the significant social factors for trills in third generation Raizales...... 184

5-18 Conditional inference tree of the significant linguistic factors for taps in Continental Spanish...... 187

5-19 Distribution of trills according to the number of apical occlusions in Continental Spanish...... 188

5-20 Conditional inference tree of the significant linguistic factors for vibrant trills in Continental Spanish...... 192

5-21 Conditional inference tree of the significant social factors for trills in Continental Spanish...... 197

5-22 A comparison of the proportions or vibrant and non-vibrant rhotics in in the Archipelago...... 198

A-1 Informants completing the Jigsaw task (Thoms J., Liao J. & Szuztak A., 2005) during fieldwork in May 2017...... 231

A-2 Informants completing the Diapix task (Baker & Hazan, 2011) during fieldwork in May 2017...... 231

B-1 Frog, Where Are You? (Adapted from Mayer, 1969) ...... 232

B-2 Jigsaw Task. (Adapted from Thoms J., Liao J. & Szuztak A., 2005) ...... 232

B-3 Diapix. (Adapted from Baker & Hazan, 2011) ...... 233

Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy

LANGUAGE CONTACT AND CONVERGING PATHS OF VARIATION: BILINGUAL RHOTICS IN TWO ISLAND COMMUNITIES OF THE ARCHIPELAGO OF SAN ANDRES, COLOMBIA

Falcon D. Restrepo-Ramos

August 2019

Chair: Jessica E. Aaron Major: Romance Languages

In the Caribbean Archipelago of San Andres, Colombia, Spanish coexists with an

English-based Creole known as Islander. This dissertation examines the outcomes of language contact in terms of the variable production of bilingual rhotics in the local self-denominated

Raizal communities. Speech data was collected by means of sociolinguistic interviews and other- speech elicitation tasks from the three linguistic varieties under study: Islander Creole (referred in this study as Raizal Creole), Raizal Spanish – the bilingual Spanish variety, and Continental

Spanish – the immigrant variety). These varieties were compared in terms of 1) the acoustic predictors of non-vibrant rhotics, and 2) the relative frequencies and linguistic constraints that condition vibrant tap/trill variation.

The sociophonetic analysis of non-vibrant rhotics identified an approximant manner of articulation while an analysis of formant frequencies revealed different places of articulation for

Continental Spanish and Raizal Creole, where the former produces and alveolar non-vibrant rhotic and the latter a post-alveolar zero-occlusion /r/. Furthermore, it was found that F3 frequencies and the F3-F2 distance increasingly resemble the values in Islander Creole approximants, whereas younger generations are more closely associated to Continental Spanish.

Results show a change in progress, where non-vibrant approximants in younger generations are converging in the direction of Continental Spanish, while formant frequencies in seniors are more associated with Raizal Creole.

Although the frequency of use of vibrant taps and trills in younger generations increasingly resembles those presented in the monolingual variety, the behavior of rhotic variation is different in both Spanish varieties. Results of the cross-variety comparison show that

Raizal Spanish displays a generational continuity where a restructuring of the constraint ordering starts in second generation and is completed in younger Raizales. On the contrary, Continental

Spanish behaves differently in terms of the systematic linguistic conditionings. The evidence suggests that rhotic variation has changed internally within both varieties.

Documentation of rhotic variation of these islands sheds light on the effects of contact, and the particular social situation of these communities, providing an account on how language contact in the Caribbean, and more specifically, the Western Caribbean, is occurring in local communities coexisting with a national non-lexifier language.

CHAPTER 1 INTRODUCTION

This dissertation examines the variable realization of rhotics in three language varieties in contact in the Caribbean Archipelago of San Andres, a Colombian territory, composed of a set of three islands: San Andres, Old Providence, and Santa Catalina. In this Archipelago, rhotics in the three main languages in contact are investigated: an English-based Creole known as Islander;

Continental Spanish, the immigrant Spanish variety from mainland Colombia; and the bilingual

Spanish variety spoken by Islander speakers. Little is known regarding the outcomes of contact at the sociophonetic level of a non-lexifier language in contact with an English-based Creole in this part of the Caribbean, and this is particularly the case of bilingual rhotics in these island communities. The current study addresses this void in the field by examining the highly variable use of /r/ in the Spanish variety spoken by generations of bilingual Raizales emerged from extended contact between Continental Spanish and Islander Creole. This study examines /r/ production across generations and its association with the languages in contact, as a window to the effects of language contact in the local communities of this Archipelago.

This chapter begins with an account of the historical and sociolinguistic background of the islands. In doing this, I explore the historical events that shaped the current sociolinguistic situation of the communities of the Archipelago. This is followed by a discussion of the population groups and the sound variables under study. The chapter concludes with an outline of the subsequent chapters of this dissertation.

1.1 Historical and Sociolinguistic Background

The archipelago of San Andres is located in the southwestern area of the Caribbean, 445 miles from the coast of Colombia and 120 miles offshore of Nicaragua. San Andres is the biggest island of the Archipelago and the center of the provincial government, while Old Providence and

Santa Catalina are the smaller conjoined islands of the Archipelago. Old Providence and Santa

Catalina are connected by a small bridge of roughly 215 meters (705 feet), known as El puente de los enamorados or the Lovers’ Bridge. For the most part, Santa Catalina is treated under the jurisdiction of Old Providence, since the strip of land facing towards Old Providence is the only small portion of Santa Catalina that is inhabited, leaving the rest of the island absent of people and suited for avid hikers and tourists (See Figure 1-1).

Figure 1-1. (Left) Archipelago of San Andres in the Western Caribbean. (Right) A close-up of the islands of the Archipelago (Google Earth, 2018).

Throughout this work, I will be treating Santa Catalina as part of Old Providence. These three islands along with many uninhabited banks, keys, and reefs constitute this Southwestern

Caribbean Archipelago. Despite being the smallest province of Colombia in terms of land size

(52.5 square miles, San Andres 26 miles, Old Providence 17.2 miles, and Santa Catalina 1 mile), its total area, including its maritime territory, comprises a vast and strategic territory of 128,000 square miles. The historical events that brought about today’s social, cultural, and linguistic situation of the Archipelago are as complex as they are fascinating. Disputed by the British and the Spanish for over 150 years, claimed by Colombia during the Independence wars in South

America, and positioned in the turmoil of two religions (catholic and protestants) and three languages (English, Creole, and Spanish), the history of the Archipelago of San Andres is intertwined with its present situation.

1.1.1 The Archipelago Of San Andres Between 1618-1810

Historically, the communities of this minority Creole language have coexisted with

Colombian migrants for almost two centuries, after Colombia gained independence from Spain and appended the islands in 1822 (Morren, 2001). However, The Archipelago of San Andres has a long history that goes back some two hundred years before this event. During the Thirty Years’

War in Europe (1618-1648), English sailors were sent to the Caribbean to attack and loot the

Spanish galleons that passed through this section of the Caribbean en route to Spain (Morren,

2001). Although there is not an exact date, it is presumed that these English expeditions lead by

Captain Sussex Cammock found these islands for the first time by Europeans somewhere between the church calendar day of Saint Catherine and Saint Andrew, the latter which happens five days later (Parson, 1956 in Ross, 2007; Morren, 2001). However, it is believed that the first route to the islands was established by the various indigenous groups of Miskito Coast in Central

America (Ross, 2007). Further historical accounts maintain that the first European settlers of the islands consisted of Puritan colonists who arrived in San Andres for the first time in 1627 from

Bermuda. The incoming of settlers continued for four more years and in 1629 merchants, most of whom professed the Puritan reform. The further arrival of the first 90 Puritan colonists who came on the Seaflower ship directly from England coincides with the abandonment of the Island of

San Andres because of the lack of sweet water reservoirs (Bartens, 2013), and the subsequent founding of Old Providence in 1631, the first British colony in the Western Caribbean (Patiño-

Roselli, 1992).

Records suggest that by “1635 there were 500 white men, 40 white women and a few children living in Providencia” (Ross, 2007, p. 8). Due to the physical demands of tropical agriculture and the building of fortifications, the first black slaves were bought from the Dutch and brought to the islands in 1633 (Ross, 2007; Patiño-Roselli, 1992), and quickly outnumbered

the white plantation owners (Morren, 2001). Because of the harsh conditions of slavery, by 1638 a rebellion broke out, which was barely controlled, and served as a prelude for a Spanish attack three years later (Ross, 2007). Aside from a brief phase when Captain Morgan held Old

Providence as his base for his notorious raids and sacking of Panama and the fact that San

Andres served as a staging post for a few colonists and slaves that managed to escape from

Jamaica to the Corn Islands and Miskito Coast (Nicaragua), the islands remain mostly uninhabited until 1677.

The period from 1677 to 1780 is known as ‘the forgotten century’, as there is virtually nothing known about the history of these islands (Vollmer, 1997 in Bartens, 2013). Despite the lack of written recorded history, it is generally assumed that the foundations of the population came from Jamaica. There are reports of Jamaicans visiting the islands in the search for fine woods (i.e., cedar) since the 1680s and resettling the islands around 1730 along with colonists from the British Isles (mainly from Scotland and Ireland) and West Africa. The census of 1793 registered a population of 391 inhabitants in the Archipelago. By 1806, thanks to decreeing land grants to settlers from Jamaica, a sharp increase in the population occurred, rising to 1150 inhabitants, from which 850 were slaves (de Friedemann, 1989, p. 149).

1.1.2 The Archipelago Of San Andres Between 1810 And 1886

The war for independence had already started in 1810 between Nueva Granada and

Spain. Bolivar’s forces gained independence from Spain in 1821 and the Federation of La Gran

Colombia was established: a territory stretching from Central America to Cape Gracias a Dios

(Nicaragua), and comprising Venezuela, Nueva Granada (Colombia), and Ecuador in the south.

The official version is that the islands adhered voluntarily to the newly created Republic and that representatives of the islands were present in the Congress of Cucuta in 1822. However, this account is challenged by Islander oral history (Bartens, 2013; Ross, 2007). Haitian Republican

Captain, Juan Faiquere, was met with the resistance of 53 men and 300 slaves on his arrival to the islands. Bolivar died in 1830 and his dream of a republican union fell apart with the dissolution of La Gran Colombia in 1831. Soon after, La Nueva Granada was created with the secession of Venezuela and Ecuador, while keeping Panama and parts of Nicaragua. For the most part and until the creation of a new Constitution in 1886, the Archipelago remained autonomous due to its distance with the central government of Nueva Granada.

In 1868, Bogota, the capital of the Republic, took over the administration of the islands from Cartagena, but was later returned to the Province of Bolivar in 1888, becoming the Canton of Providencia. Some years before the installation of a new Constitution in 1886, there were plans to establish schools functioning in Spanish with the aim of allowing the Islanders to run the local administration while being able to with maintain political ties with Continental Colombia.

It was later in 1886 that the new Constitution would serve as the main acculturation tool for the imposition of Spanish and the Catholic religion in a process known as ‘Colombianisation’.

1.1.3 The Archipelago Of San Andres Between 1886 And The Present

Prior to the new Constitution of 1886, the Colombian government had left the islands to function mostly autonomously. With the new Constitution the central government posture changed. The Colombian State soon established public schools on the islands promoting Spanish and Catholicism, along with orders from Bogota to appoint officials external to the islands. For the Afro-Caribbean Islanders, the imposition of Colombian values and culture and the domination by a foreign minority who were strange to their community was perceived as a form of suppressing their African roots and Anglo heritage and an attempt to implant the Colombian culture. As soon as the Nicaraguan government took over the Atlantic coast from the Miskito

Indians in 1894, a new geopolitical situation divided the Islander diaspora: The Corn Islands and

the Miskito coast were taken by the Nicaraguan government and San Andres and Old Providence remained a Colombian territory (see Figure 1-2). This was formally ratified later through the

Esguerra-Barcenas treaty of 1928.

Figure 1-2. The Archipelago of San Andres, and the Nicaraguan territories (i.e., the Corn Islands, Bluefields, Pearl Lagoon, and Miskito Cay).

Backed by the US and promised with the construction of a new canal that would connect both the Atlantic and the Pacific oceans, Panama seceded from Colombia in 1903. The same year, Panamanian officials aboard the US warship Nashville arrived in San Andres offering the islands to break away from Colombia and join the US. The offer was rejected, but American influence still remained in the region in the form of cultural exports, such as baseball, boxing and basketball, and through the spread of Protestant churches. The Colombian government established control over its maritime territory by establishing a provincial system. Despite this, policies still emanated from Bogota, including the implementation of migratory and trading controls on the islands. As a result, the free movement of islanders with contiguous Afro-

Caribbean settlements was restricted and administration of the islands relied on mainland

Colombian officials. To this point, the push by the national government for public schools under the Catholic system was fruitful, thanks to being run by fellow Brothers from the Mill Hill

Mission from London, who ran the school system in English from 1913 to 1926, gaining great acceptance and respect from the native community. The harmonious relationship that the Baptist church had with the Catholic church came to an end in 1926, when the Mill Hill Mission was replaced by the Spanish Capuchins, sent from the highlands of Colombia, more specifically from the Province of Antioquia.

The Capuchins implemented a new education system solely in Spanish and were involved in a civilising mission, mainly the forcing of the Spanish language and Catholicism on the minority communities (Ross, 2007). However, the islanders, well educated, versed in English as their vernacular, and sharing a world-wide Christian faith, were more educated than many parts of mainland Colombia. The combined effect of the economic depression of the 1930s, the plague that befell the coconut plantations in 1931 and the rat plague in 1932, devastated the economy of the islands, and thus, forcing migration of the 20% of the population to Colon in Panama and

Cartagena (Bartens, 2013). In the meantime, masses of islanders became Catholic and the new generations who passed through the school system now spoke Spanish as their second language.

Their economy relied on artisan fishing and small farming production for the upcoming years.

Although their population remain largely unmixed and their vernacular English unchanged for the most part (Ross, 2007), the first half of century resulted in the preliminary phases of incorporation into the Colombian State.

The most significant economic and social changes started out in 1953, when President

Gustavo Rojas Pinillas declared San Andres a duty-free port by Decree 2966, bringing huge commercial expansion to the island of San Andres, provoking substantial influx of Continental

Colombians immigrants, and stripping the native communities from vast portions of their territories. By doing this, Colombia was emulating the Spanish Royal Order of May 20, 1792, which gave San Andres “the rank of minor port and freedom from all import and export duties”

(Ross, 2007, p. 11). The declaration of the Freeport came also with two other main events that had huge implications for the current situation of the Archipelago. First, an airport was built named under the president of the time. This caused a major impact on commerce and migration to and from the islands, as the week-long trip by ship was shortened by a few hours on plane.

Similarly, during the same time, the Colombian Government and the Pope Pius XII signed an amendment to the Concordat stipulating the banning of all non-Catholic Christian activities in

Mission areas, including the Archipelago. Granted power over the public-school system, the

Catholic Mission in San Andres had already banned English as the language of instruction in public education in 1946, and in March 1954, the same stipulation barred English from private schools, including 3 Baptist and 2 Adventist schools (Bartens, 2013; Ross, 2007). Six hundred children were left without education and those who entered the public system had to start over from primary levels. With these new events, the struggle intensified between the agricultural and fishing community of English, Creole-speaking, Protestant islanders and the newcomers:

Spanish-speaking, business-savvy, Catholic immigrants from Continental Colombia. Not only mainlanders arrived in masses to the islands, but also other foreign merchants from middle east, the Turcos, who took advantage of the incentives offered by the government in 1959 to commercially develop the Freeport. According to records of the Chamber of Commerce, by 1965 there were 356 business, of which 189 were registered to continental Colombians, 115 to foreign nationals and only 53 to Islanders. Another consequence of the commercial expansion resulted in the dispossessing of land from the Islanders. Land rights had been passed to islanders through

generations, many of which had suffered under slavery, and the pressure to sell was higher than ever, due to the commercial value that acquired each meter of land and the higher tax demands.

During the period of years between 1962 and 1977, immigration was at its peak and a census in 1973 reported that from a population of 31,000 the islanders became for the first time a minority in their own territory. Moreover, a whole generation of islanders had grown up under the instruction of Spanish in schools. This represented a turning point in the islands and the

Islander community. As uncontrolled development and foreign investment increased, more immigrants from Colombia were constantly arriving to the Islands, and thus, most of them remained in the islands mixing with the natives and having inter-ethnic offsprings. By then, a new generation of islanders emerged, islander by birth, but with ties with Continental Colombia and spoke Spanish in many households, mainly facilitated when the mother was a Spanish- speaking immigrant. From then on, the term islander started failing to have the same connotation for some and the question of who an islander was became problematic for many. The new islanders, those with symbolic roots or origins in their Afro-Caribbean ancestry and English,

Creole languages, chose to be called ‘Raizales’. This new ethnic marker made the distinction between islanders by birth with no Creole heritage and ‘pañas’, or Spanish-Speaking, Colombian immigrants, and the minority, native Islander community, now ‘Raizales’. This categorization remains untouched until the present work and will be used as the main term to distinguish between the groups under study.

A new hope for recognition of Raizal issues by the Colombian government was possible thanks to the new Constitution of 1991. The Colombianisation policies enforced in the

Constitution of 1886 were now abolished by the creation of new laws that gave more autonomy to marginalized communities, such as Raizales. Article 42 gave the president the legal power to

decree laws to control the overflowing population and the protection of the natural resources of the islands in 1991. As a result, the OCCRE (Oficina de Control, Circulación y Residencia) was created by Presidential Decree 2762 of 1991 with the aim of controlling immigration to the islands and granting formal residency to Raizales and immigrants who meet the permanency test.

Despite the creation of the OCCRE, the irregular influx of new immigrant increased. According to official estimates, the total population of the Archipelago is 59,573 and some 23,396 ethnically self-describe as Raizales (DANE, 2005). However, unofficial accounts indicate that over 100,000 people live in this territory of less than 17 square miles, from which 20,000 are

Raizal in origin (Enciso-Patiño, 2004). It is believed that the population of newly arrived immigrants is doubled every 10 years with an estimated population in 2017 of 130,000 people.

The island of San Andres is the island with the sixth highest population worldwide and the highest population density per square mile in the Caribbean (Instituto de Investigaciones Marinas y Costeras “José Benito Vives de Andréis”, 2012). A new census was conducted by DANE in

2018 but updated demographic data have yet to be publicly available. Despite the discrepancies in population data, the immigrant population expansion of the Islands has an effect on the maintenance and vitality of the language and culture of these communities. Residents of the islands, both Raizales and established immigrants, consider the overpopulation crisis as the main problem associated with violence, territorial and cultural displacement in the Archipelago.

In 2001 Nicaragua brought forward a suit against Colombia questioning its sovereignty over the Archipelago of San Andres. Years of poor State preparation and planning from the

Colombian government resulted in the loss of maritime waters to Nicaragua, which was dictated by the International Court of Justice in the Hague in 2012. Although the islands remained part of

Colombia, the loss of maritime waters meant the loss of substantial portion of fishing grounds for

Raizales. Despite a history of acculturation, Raizales still claimed loyalty to Colombia, while, at the same time, more and more Creole organizations advocating for the Raizal identity and heritage were established under the umbrella term, Raizal Movement (Gonzalez-Palacios, 2016;

Ranocchiari and Calabresi, 2016). This has allowed the community to exert pressure on the local government of the archipelago, opposing policies with little benefit to the Raizal community. As the community seems to realize that their culture and heritage is in jeopardy, notorious initiatives for the visibility of their Creole language have emerged. Funded by the Colombian Ministry of

Culture, the local TV network airs trilingual content: in addition to English and Spanish, Islander is spoken regularly in live talk shows; the local news has two daily broadcasts, one in Spanish, and right after, one in English. More recently, the only Colombian motion picture spoken solely in a language other than Spanish was released in Creole (Bad Lucky Goat) in 2017, having a great impact on the visibility of the language. Social media has also been a hub for native

Islanders to express themselves in their dominant Creole language and clearly more research needs to be done in these virtual spaces.

The history of the people of the Archipelago of San Andres has remained under the influence of opposing forces, one representing old and new colonialism and the other representing the diaspora of the Creole people of the Caribbean. The nature of these opposing forces resulted in a series of contrasts in the Archipelago between master and slave, white vs black, Catholic vs Protestant, Hispanophone vs Anglophone/Creole speaker, Paña vs Raizal.

1.1.4 The Past And Present Of Islander Creole

Islander Creole has a direct kinship with the diaspora of the English-based Creoles of the

Caribbean. Its most recent origins can be traced back to the second half of 1700s, with the arrival of colonists from parts of the British Caribbean, particularly from Jamaica. By the late 1800s, trade with Jamaica was prohibited by the Colombian government, and thus, halting the direct

influence from Jamaican Creole. It is believed that during the second half of the 19th century,

Islander Creole must have crystallized into a separate mesolect (i.e., the intermediate variety between a Creole from the colonization era to the modern and more standardized variety), which gave birth to the acrolect or standard variety spoken nowadays (Bartens, 2013; Dittman, 1992;

Bickerton 1973). Therefore, Islander Creole is directly derived from Jamaican Creole (Bartens,

2013). Washabaug’s study (1974) on the island of Old Providence proposes that Islander Creole is decreolizing and suggests that the entire system is moving towards Standard English.

However, as Aceto (1996) maintains, the decreolization model or continuum have limited applications in scenarios where a Creole variety is in direct contact with a national non-lexifier language, such as Spanish. Either way, we know that by the beginning of the 20th century, most inhabitants of the islands were direct descendants of Creole-speaking Afro-Caribbean people

(Patiño-Roselli, 1992), and as we also know, from then on multiple social, economic, and political events produced the current ethnolinguistic situation of the islands.

When deconstructing the history of Islander Creole even further, we must go back during the period of the Atlantic slave trade. During the 16th century, European powers were able to expand their empires overseas, thanks to the European discovery of the ‘New World’ in 1492 by

Christopher Columbus and his crew in the three ships that sailed from the Spanish port of Los

Palos. This discovery led to the inhumane kidnaping and subsequent trafficking of human labor from Africa by the most prolific maritime powers of the time, the Portuguese and the British fleet. For approximately 300 years, slaves were imported to the Americas by a system of human trafficking from the coasts of West Africa. It is assumed that an Afro-Portuguese and an Afro-

English proto-pidgins were originated in the trading feitorias or factories of West Africa and were brought to the Americas by slave traders (Lipski, 2009; Holm, 2000; McWhorter, 1997;

Berlin, 1996). This primordial reduced language (i.e., pidgin) resulted from extended contact between the many languages of the African slaves and their European masters and traders. The steady contact produced a stabilized make-shift language with fixed yet simplified norms of communication, which dropped structural and lexical forms of the languages in contact, such inflections and word endings (Holm, 2000). The influence that the more powerful language (i.e., superstrate) exerts over the languages with less power (i.e., substrate) often results in the accommodation and the use of words from the superstrate language but with the adaptation of the linguistic norms of the substrate language. At this point, this pidgin has a very restricted and limited domain confined most likely to trade, and it has not become no one’s first language yet

(Holm, 2000).

The pidginized African ancestor of Islander Creole can be reconstructed according to the slave trading history to Jamaica of the seventieth century up to the ninetieth century. From 1655 to 1725, a quarter of the slaves brought to Jamaica came from the region of Gambia and Sierra

Leona. Based on the map on Figure 1-3, this would correspond to zones 1 and 2 according to the main traffic ports of slaves in West Africa numerated from 1 to 9 (Dittman, 1992). The African languages imported from this region include Mandinga and Mende or Bende, which according to

Holm (1978), represent the structural base for the Portuguese-based and English-based Creoles in the Caribbean and Krio in Sierra Leona. Similarly, another 25% of the slaves imported to the

Americas came from zone 5, which is known as the British colony of the Gold Coast, where

Akan is the major linguistic group.

Figure 1-3. Main traffic ports of slaves and the main local languages (Holm, 1978; Le page and De Camp, 1960, taken from Dittman, 1992)

The same region became a source for a greater percentage of the slave trade to the Americas from 1725 to 1775. At the same time other areas of West Africa saw an increasing number of slaves being embarked from the Slave Coast (Nigeria) on zones 6 and 7 from 1726 to 1808, which pertained to the language groups of Yoruba, Igbo, Efik, and Ibo. The last stage of the slave trade involved the importation of more slaves from the center of the continent (Congo) and further south to the countries of Angola and Zambia on zones 8 and 9 from 1776 to 1808

(Dittman, 1992). The phase and location in which this English-based proto pidgin turn into a full-fledged nativized Creole language is uncertain and “mostly based on speculations” (Holm,

2000, p. 8). A Creole, a term which used to denote customs and speech of Africans and

Europeans in the New World, expanded its meaning to refer to the genesis and development of a language with a pidgin ancestry, whose speakers were “displaced geographically and ties with the original language and sociocultural identity were partly broken”, which were often the result of slavery (Holm, 2000, p. 6). Precisely, Edwards (1974) revealed that over 17% of the lexical

items in Islander Creole, which also appear in Jamaican Creole, were of African origin. The common ancestry of these lexicon comes from the Twi language, a variety of Akan, spoken in what was known as the African Gold Coast. Another notorious substrate lexifier is Mende from the Southwest coast of Gambia, formerly known as Windward Coast during the time of the slave trade, and Sierra Leona (Dittman, 1992). Commonly, Mende or Bende is sometimes used by

Raizal locals to refer to their Creole language, while Islander Creole is a term used to contrast with the language of the mainlanders (i.e., Spanish). Recall that at the onset of the slave trade in

1655 importation of slaves from the feitorias started in the region of current Gambia and Sierra

Leona and the Gold Coast (current Nigeria) and reached its peak until 1775. Precisely, the ancestors of current Raizales, and thus, the English-based proto-pidgin that originated the offshoot of Jamaican Creole, is what we now know as Islander Creole.

The ethnolinguistic vitality of Raizales and their Creole language has changed dramatically during the last 100 years. At the beginning of the 20th century, it was estimated that a total of 8,000 Raizales lived in the Archipelago (Dittman, 1992). In 2005 the official number reached 23,396 (DANE, 2005). The commercial expansion of the islands in 1953 resulting from the Freeport declaration brought an ever-increasing influx of business-oriented immigrants until restriction policies were put into place. Consequently, the old sites where Raizales have traditionally established their communities have now been displaced by hotels, an airport, and an array of multiple businesses. The conflict for land and means of living have displaced this community to live outside of the commercial spheres of economic and political significance of the islands, and settled in ethnic neighborhoods, where Islander Creole maintains oral transmission and community ascription. Both Raizal communities in San Andres and Old

Providence have remained in close-knit communities somewhat distant from the fast-paced life

of the urban area. In San Andres, Raizales tend to live in ethnic neighborhoods, such as The Hill,

San Luis and South End. However, the Hill seems to be the stronghold of the Raizal identity, particularly because this part of the island has little economic interest for commerce and tourism, and it has been historically the location of the First Baptist Church and other English-speaking churches (see Figure 1-4).

Figure 1-4. Island of San Andres with the location of the ethnic neighborhoods and the downtown areas.

As a result, Islander Creole is more preserved in this area of the island. In contrast, there are several popular beaches and keys in San Luis which attract more tourists and commerce and have resulted in more immigrant households. The only tourist attraction in South End is a blowhole naturally formed in the rocky cliffs that protects the south corner of the island from coastal erosion. By 2005 the three ethnic neighborhoods of the island were primarily composed of 5,140 Raizal natives from a population of 6,979 (DANE, 2005). The rest of the rural area of the island is also predominantly Raizal, with some 5,088 out of 7,545 belonging to this ethnic

group (DANE, 2005). Conversely, the center of commerce and tourism, as well as the local government and other administrative buildings, are in the North End. Official numbers estimate a population of 40,902 in this part of the island, from which 9,523 are of Raizal ancestry (DANE,

2005). In other words, the immigrant population outnumbers Raizales by a ratio of 4 to 1 in

North End. It is worth noting that these estimates correspond to the 2005 Colombian census and it is probable that the population disparity has continued to increase. A new census in 2018 has been implemented by the national government, but official demographic numbers have yet to become available to the public. It is not surprising that bank, restaurant, hotel, and any other type of business-related activities are mainly carried out in Spanish. During her fieldwork in 1989,

Dittman (1992) noted that the use of Continental Spanish alternates with Islander Creole during linguistic relationships in public offices when in the presence of Continental immigrants. In practice, administrative issues are primarily conducted in Spanish in the local governmental building even between Raizales, despite the fact that local officials are required to have knowledge of Islander Creole to hold public offices. However, it was also observed that Islander

Creole is preferred in informal conversations between Raizal public officials, something also noted by Dittman (1992).

Old Providence presents a different sociolinguistic panorama mainly due to its distance from San Andres that has allowed Old Providence to be more culturally and linguistically preserved. There are 56 miles of maritime waters between them, and contrary to San Andres, Old

Providence was not included in the expansionist commercial project of a duty-free port.

Estimates place the population of Old Providence at 5,000 inhabitants, from which 89% are composed of Raizales (Enciso-Patiño 2004). Official numbers seem to support this information with 3,645 Islanders accounting for 88% of the entire population, mostly living in rural areas

(DANE, 2005). Precisely, these rural areas are connected by the main road of the island, which traverse around the coastline, and has allowed for small settlements forming on both sides of the road and in small neighborhoods.

Figure 1-5. Island of Old Providence with the location of the ethnic neighborhoods and the downtown areas.

Figure 1-5 shows the distribution of the Raizal population in ethnic neighborhoods around the island and the main commercial hub of Town, where the port and the mayor’s office are also located. Official estimates of the Raizal population outside of Town is 2,138 out of

2,408 people and 1,507 out of 1,739 inside Town (DANE, 2005). In a speech community of this size, it is safe to assume a greater maintenance of Islander Creole permeated less by the presence of Spanish in comparison to San Andres. In fact, speakers of Island Creole have far fewer daily interactions with Spanish speakers, due to more rigid protective measures against new immigrants, the insistence of the locals for their English legacy (Bartens, 2013), and the value of

English as the international language of tourism (Flórez, 2006). As a result, Islander is still

prevalent in most aspects of the lives of Raizales (Moya-Chaves, 2014; Bartens, 2013; Flórez,

2006; Bartens, 2002; Morren, 2001), and has helped to maintain the prestige on this Islander variety.

In what follows, I will circumscribe the population groups for this study and their corresponding linguistic varieties in a description of their sociolinguistic background.

1.2 The Current Study

There are three main languages coexisting in this Western Caribbean Archipelago:

Islander Creole, Colombian Spanish, and English. The first two languages have been integrated into the linguistic repertoire of the indigenous bilingual, self-denominated Raizales (Bartens,

2013). Although, it has been noted that English can be occasionally spoken by Raizal speakers who have worked in cruise ships or lived in English-speaking countries (Bartens, 2009), it is commonly absent in everyday interactions unless used in touristic domains or the Baptist church

(Morren, 2011; Grimes, 2000). Marginally, there are other minor immigrant languages from

Muslim countries present in the Archipelago. The Turcos, now in control of the islands economy, and until recently, holding office in major political positions (Governor Housni 2016-

2018, revoked under corruption charges), have safeguarded their culture and Middle Eastern languages from the influence of Islander Creole and Colombian culture, by receiving their own education in private schools or in their own countries (Ross, 2007). Some of them have even born in the islands but have been reluctant to be absorbed by the Christian faith. Since the Turcos represent a small immigrant minority in the islands and contact has been mostly reserved for commercial settings, the interrelation of these languages with the larger perspective of contact falls outside the scope of this research. However, further studies should examine their influence in the Archipelago.

1.2.1 Identification Of The Linguistic Varieties Under Study

For this study, the Raizal ethnonym will be used to contrast between the three varieties under examination. As have been noted previously, this term (Raiz ‘root’ and -al ‘belonging to’

(belonging to the root)) corresponds to the denomination of the ethnic group of the archipelago of San Andres whose genetics, history, culture, identity and language can be traced back to their

African roots. The Raizal identification is widely used in juxtaposition from the relatively

‘newly’ arrived immigrants known as Pañas, who come mainly from coastal areas of continental

Colombia. Precisely, the main focus of this work involves the cross-examination of the Spanish variety spoken by generations of bilingual Raizales emerged from extended contact between

Colombian Spanish and Islander Creole. In doing so, I distinguish two linguistic varieties spoken by the same Raizal population. The first variety involves the Raizal mother-tongue: the English- based Creole, known as Islander Creole. This variety will be categorized as Raizal Creole due to being the vernacular of the Raizal ethnic group and will be used in this study henceforth. The second variety correspond to the bilingual Spanish spoken by Raizal informants, Raizal Spanish.

This variety will be examined across three generations of speakers in both Islands of the

Archipelago, San Andres and Old Providence (which includes Santa Catalina). In addition,

Continental Spanish, the monolingual Spanish spoken by immigrants who arrived in the islands from coastal Colombia, will also be compared. In conclusion, the three linguistic varieties that will be examined in this study are: Raizal Creole, Raizal Spanish, and Continental Spanish.

Precisely, Raizal Spanish will be the main focus of attention for this study, as it will be analyzed between generations through the comparative method (Tagliamonte, 2012) with the other Spanish variety of the islands, Continental Spanish. The comparative sociolinguistics method studies the behavior of linguistic variation across data sets with the aim of determining the underlying grammatical relationships between dialects or languages (Tagliamonte, 2012;

Poplack and Tagliamonte, 1999). This method uses quantitative approaches to compare the variation patterns of linguistic features from one source to another (Tagliamonte, 2012). I will examine this approach further in the subsequent section.

1.2.2 Studying Sound Variation And Change In This Contact Scenario: Rhotics

The main goal of this study is to examine the variable production of two types of rhotics in the bilingual speech communities of the Archipelago of San Andres as a window to understand the effects of language contact: vibrant and non-vibrant rhotics. The term rhotics is derived from the Greek letter rho to characterize a collection of r-like sounds, categorized according to the manner of articulation in different languages. This collection of sounds can be realized as taps, trills, approximants or fricatives (Thomas, 2011). Precisely, vibrant rhotics refer to the use of a mouth articulator (i.e., the apical portion of the tongue or the uvula) to produce a vibration or bouncing against another surface (i.e., the palate or tongue dorsum). In normative terms, vibrant rhotics are classified based on the number of vibrations and the phonotactics of the language (Zahler and Dainone, 2014; Bradley and Willis, 2012; Diaz-Campos, 2008; Hualde,

2005). Vibrant rhotics realized with one apical closure in the alveolar region are known as taps.

Vibrant rhotics that are produced with the realization of multiple vibrations of the tongue in the alveolar region or with the uvula in the tongue dorsum are known as trills. Contrary to vibrant rhotics, non-vibrant rhotics are produced with other articulatory gestures that don’t require any vibration of a mouth articulator. Such sounds are realized with a pharyngeal constriction (i.e., approximant) or with fricative noise (i.e., fricative or assibilated).

Phonological variants are very susceptible to language variation and change, as the properties of sounds provide speakers with a vast array of options to adapt and accommodate to social situations (Thomas, 2011). Particularly, rhotics represent a highly variable segment without a single physical property that unites them all (i.e., they are produced with different

articulatory mechanisms) (Thomas, 2011). The lack of unifying articulatory properties makes this sound particularly different (both perceptually and articulatorily). Rhotics sounds are very diverse in their acoustic properties cross linguistically and can provide a wealth of research in contact situations, especially in scenarios where the languages that coexist are lexically unrelated. Precisely, the sociophonetic study of rhotics in a contact scenario, such as the

Archipelago of San Andres, represents a fertile ground for studying sound variation and change in bilingual communities. Compared to Spanish, Raizal Creole, which is lexically derived from

English, has been reported to make use of a non-vibrant English-like /r/ (Bartens, 2013).

However, the exact phonetic cues for determining its nature are unknown. Spanish, on the other hand, realizes its rhotic inventories in the form of vibrant taps or trills. However, weakened rhotic variants (i.e., non-vibrant) are increasingly being reported in different monolingual varieties of Spanish, and there seems to be a trend toward this innovative production in spontaneous speech (Diaz-Campos, 2008; Henriksen, 2014). The extent to which this is present in bilingual varieties is still fully unexplored. Hence, the Archipelago of San Andres, an insular territory where an English-based Creole and a national non-lexifier language (i.e., Spanish) are in contact, provides a natural context where vibrant and non-vibrant rhotic variation and change can be studied across generations of bilingual communities.

As a result, non-vibrant rhotic segments are analyzed in terms of their acoustic properties in Raizal Creole, Continental Spanish, and generations of Raizal Spanish. This sociophonetic analysis is presented in Chapter 4. Vibrant rhotics are examined in terms of the variable production of lingual closures or vibrations between the Spanish varieties in contact (i.e., Raizal

Spanish and Continental Spanish). This analysis includes a comparison of the linguistic and

social conditionings that constraint the variable production of vibrant rhotics in generations of

Raizales and Continental Spanish. Results are presented in Chapter 5.

The current study incorporates a comparative approach under the sociolinguistic framework to uncover sound variation across generations of Raizal Spanish and its association with rhotic production in Raizal Creole and Continental Spanish. The purpose of this project is to examine this contact situation in light of the potential variable influence that Spanish and

Islander has had on rhotics in the vernacular Spanish spoken by the bilingual community in San

Andres, Old Providence and Santa Catalina, as a window to establish the outcomes of contact in these island communities. This project integrates two perspectives from which I aim to analyze sound variation at the community level: 1) a sociophonetic analysis will delve into the patterns of variation of non-vibrant rhotics, and 2) linguistic and social background measures will provide an analysis of vibrant rhotic production across varieties in the Archipelago.

The dissertation is organized as follows: Chapter 2 presents a review of literature that deals with influential research on language contact, variationist sociolinguistics, and rhotic variation. Chapter 3 examines the methodological approaches used in this study. Chapter 4 presents the results on non-vibrant rhotic variation and its acoustic association with one of the contact languages, Continental Spanish and Islander Creole. Chapter 5 shows the findings of the comparative variationist analysis of vibrant rhotics in the Archipelago. Chapter 6 presents a discussion of the findings. Lastly, Chapter 7 includes the conclusion of the study.

CHAPTER 2 REVIEW OF THE LITERATURE

This part of the dissertation is devoted to examining previous research that supports and justifies the analysis of the data collected for this work. This literature review is composed of two sections. First, I will detail research that has dealt with language contact and the comparative approach used for this study. The next section analyzes literature on rhotic variation in monolingual and bilingual settings.

The findings of previous research on language contact, and rhotic variation and change will serve to contextualize the research questions that guide the current study. The first research question refers to the analysis of non-vibrant rhotics and is enunciated as follows: How do the acoustic properties of non-vibrant rhotics compare across Raizal Spanish, Raizal Creole, and

Continental Spanish? In order to answer the first research question, three other sub-questions follow up: (1) What are the acoustic correlates that characterize non-vibrant rhotics in the varieties of the Archipelago? (2) What are the acoustic correlates that best predict discrimination between varieties? (3) Is there a change in progress between non-vibrant rhotics across generations of Raizal Spanish speakers? The last question in the series encompass the overarching response to the first research question and opens the door for another inquiry: If there is a change in progress, is there any generation that converges in the direction of one of the languages in contact, either Raizal Creole or Continental Spanish?

Subsequently, the second research question guiding this study refers to the examination of vibrant rhotics: How does vibrant rhotic variation differ in Raizal and Continental Spanish?

Two other inquiries are necessary to answer the overarching question on vibrant rhotics: (1)

What are the sociolinguistic factors that account for sound variation in rhotics across generations

of the vernacular Spanish of Raizales, and Continental Spanish? (2) Is speech variation spreading socially in the direction of Continental Spanish motivated by new generation of speakers?

With these research questions in mind, I aim to contribute to current research on contact phenomena and sound variation and change. This original research project delves further into areas of the Hispanic world where linguistic and cultural tensions have arisen, due to asymmetrical linguistic relationships.

2.1 Language Contact and the Comparative Method in the Archipelago

In this study, I am analyzing a case of language contact in the Caribbean Archipelago of

San Andres, Colombia. A language contact phenomenon corresponds to a situation in which languages coexists within the same space and are used concurrently in the same society at the same time (Thomason, 2001, p. 1; Klee and Lynch, 2009, p. 1). Expanding this basic definition, such conditions will imply situations of bilingualism, in which anyone who uses two languages for functional purposes regardless of proficiency is considered a bilingual individual (Thomason

2001, p. 3). In such scenarios, the languages in contact might exert an influence in the bilingual speech of these individuals and changes in the languages could arise. The development of a theory and methodology of studies on language contact has focused on the dichotomy between language-internal change vs. language external change. This contrast arises from two points of view: whether we attribute a structural change (i.e., of phonological/phonetic, syntactic, morphological, etc., nature) to the gradual development of the language (i.e., internal), or to interlinguistic influence due to the (stable) contact between languages (external). However, both internal and external factors (i.e., sociolinguistic factors) interact in many ways to cause linguistic change and it is well accepted that this dichotomy is not mutually exclusive(Klee and

Lynch, 2009; Aikhenvald, 2006; Heine and Kuteva, 2003; Sankoff, 2002; Thomason 2001;

Silva-Corvalán, 1994; Cedergren, 1973, among others). For this reason, the interplay of multiple

complex interactions from sociolinguistic factors might be in place in any language contact situation (Heine and Kuteva, 2003).

In general terms, when we assume a linguistic feature has been an outcome of language contact, two main processes are in play: transfer (Weinreich,1953), whether direct or indirect

(Silva-Corvalán, 1994), and simplification (Silva-Corvalán, 1994; Winford, 2005; McWhorter,

2007). In transfer, a structural feature of a source language is simply transferred to the receiving language, producing the subsequent restructuration of the corresponding subsystems (Wienreich,

1953, Klee and Lynch, 2009). In direct transfer, Silva-Corvalán (1994) distinguishes between substitution or incorporation of a linguistic form in the receiving language due to its use in the source language. Indirect transfer occurs from frequent use or loss of a form in the receiving language due to categorical presence or absence of another form in the source language. For this research, I use the term transfer to explain any “interlinguistic influence of structural nature”, including phonological/phonetical elements (Klee and Lynch, 2009, p. 15).

Another process that characterizes contact-induced change concerns simplification. This process arises when there has been transfer from incomplete acquisition of a community in contact with another language. This term involves simplification, reduction, and omission of forms through gradient levels of simplified forms. As a result, the less marked more basic forms tend to be used. For instance, this is seen in the overgeneralization of less-complex clitic forms in

Andean Spanish (i.e., leismo), and simplification of the L1 as a result of incomplete acquisition as is the case of Spanish in the US (Klee and Lynch, 2009).

As we will see below, the concept of transfer has been studied the most, and this concept is also most relevant to the current study. Thomason and Kaufman (1988) distinguish borrowing and linguistic interference in scenarios derived from transfer. Mainly, these describe the two

outcomes of structural change that may occur. Commonly in a borrowing situation, new lexical elements from the contact language are incorporated into the L1 speech. In more extended bilingual settings, structural transfers of linguistic features are integrated more gradually than lexical borrowings. Again, these structural transfers are capitalized in incomplete acquisition situations of the L2 among members of a large speech community, due to causes involving social distances among subsets of the population or the lack of instruction in the L2. If structural transfer is persistent and ‘anomalies’ spread gradually across generations within the community, then this leads to alterations of the L2 grammar or language change.

There are several steps to determine whether a structural transfer involving a phonological change is caused due to language contact. According to Thomason (2001), a contact-induced change can be attributed whenever we can discard its occurrence in a monolingual situation. Thomason and Kaufman (1988) make a distinction for determining whether a linguistic change should be ruled out as a language-internal phenomenon or establish socially driven linguistic change instead. Specifically, if a linguistic form x is considered to take place due to contact, other linguistic forms w, y and z should have surfaced as well due to influence of the same contact language, usually in other subsystems different than the one of x in the recipient language. Parallel to Thomason and Kaufman (1988), Van Coetsem (1988) refers to the notion of agentivity of the source and recipient language, to account for transfer in highly structural domains (specially phonology) compared to borrowing of items in less structured domains (notably lexicon). Basically, borrowing occurs when material from an external source language or L2 is imported into the recipient or most dominant language. This is referred to as recipient language agentivity (or RL Agentivity), and lead to little or no modification in the structure of the RL. In contrast, direct imposition of structural elements results from transfer of

material from the dominant L1 to the recipient L2, or the language the speaker is less proficient

(Source Language or SL agentivity). This latter case would correspond to the language contact situation of the Archipelago of San Andres, where I analyze the imposition of phonological segments into the L2 of Raizales. As complemented by Winford (2003), SL agentivity can significantly affect the structure of the recipient L2 that is created by learners of the SL-dominant bilinguals.

However, one problem with the framework that Coetsem (1988) and Winford (2003) propose involves determining the dominant language of a bilingual. In most studies, the dominant language in a language contact situation is assumed without empirical foundations, and hence, establishing the exact types of agentivity proposed might become vague (Klee and Lynch,

2009). Establishing a contact-induced change is usually easy with loanwords, as their language origin can be tracked directly. Contrary to this, establishing the same for structural transfers might become precarious, and ‘even close to impossible’ (Thomason, 2001, p. 91). For making a clear case on structural change due to contact, Thomason (2001, 2010) highlights the need to look at the language as a whole, looking for possible cases of shift-induced interference elsewhere in the language. In her own words, “no case for structural change will be fully convincing if we cannot point to other instances of structural interference of the same source language in the same receiving language” (p. 91). In addition to this, she points out four other social factors to potentially account for contact-induced change: (1) identifying a source language: it is necessary to demonstrate the degree of contact, intensive enough to produce a situation of contact-induce change; (2) finding shared structural features in both the source and recipient language: there is no need for the shared features to be identical in everything, and in most cases the interference features and their distributions differ in the donor language; (3)

demonstrating the interference features were not in the receiving language prior to come into intimate contact with the source language: proving that the innovative feature was absent in the receiving language; (4) attesting the presence of the innovative features in the source language before it came into contact with the recipient language. Thomason (2001) points out that in long- established linguistic areas the last two requirements are hard to determine, and in some cases impossible to verify. Differing from her previous work (Thomason and Kaufman, 1988), she now insists in the need to look for internal motivations for the changes, such as universal structural tendencies and markedness considerations, as multiple causation is always a real possibility (Thomason, 2001). In this sense, Thomason and Kaufmann (1988) pointed out two linguistic factors that commonly condition variation and change in language contact situations.

Mainly, the typological distance between the languages in contact, and markedness of the forms that are more feasible to be transferred and structurally integrated. Essentially, depending on how typologically related the languages in contact are, the more probable the linguistic features are transferred. In the same line, transfer of linguistic forms depends on the difficulty to being acquired. It is less likely that more marked forms are transferred in a linguistic interference situation.

Accordingly, with these approaches in mind, one might hypothesize that the Archipelago is a prime site to explore language variation and change. However, a main critical point that

Thomason and Kaufman (1988) underestimate in their hypothesis involves the internal linguistic restrictions, which work together with external conditions to determine the linguistic outcomes resulting from contact. Silva-Corvalán (1994) and Sankoff (2002) contest Thomason and

Kaufmann (1988) assumptions, and they argue that even in intense contact situations transfer of

structural elements are conditioned by the recipient language itself, disallowing radical changes in the structure of the language.

It seems that we face a cumbersome situation when we are dealing with the burden of determining contact-induced change in the speech data. Consequently, this poses an important question: if, according to Thomason and Kaufman (1988), linguistic transfer resulting from contact is so difficult to determine, then why is it so important to establish contact-induced change? The answer to this relates to the purpose of sociolinguistic inquiry itself. We observe language use and document linguistic phenomena to test assumptions and theories about language itself and the individuals in the society, including tracking its variation and change over time (variationist sociolinguistics). Moreover, we go deeper into the powerful language-internal mechanisms and the historical and contemporary social processes involved in the development of the languages.

As we have noted previously in this work (Chapter 1), there are three main languages in contact in the Archipelago: Islander Creole, Colombian Spanish, and English. Most Raizales are fully fluent in Islander and Spanish, while English is the language of sermon in some churches, particularly from the Baptist religion, which makes use of the British variety. Except in generations of 60 years of age and older, most Raizales are passive bilinguals in this sense, understanding the language but not fully fluent in English, mostly thanks to the typological closeness of Raizal Creole, an English-based Creole, and English. Despite being absent in the language repertoire of most Raizales, English has been regarded by some members of the community as the language they identify the most, as a way to politically oppose Spanish.

English seems to be the language of international business and prestige, mainly due to national demands on bilingualism, but also from American popular culture in movies, internet, and

media. Bartens (2009) reports difficulties in making aware some parts of the Raizal community that English and Raizal Creole are two different languages. It seems that this situation happens in bilingual Raizales of low socio-economical class, but primarily in high school students who take

English as a subject matter.

The case of the Archipelago of San Andres is special due to several historical events that transformed the society of the islands, as were summarized in Chapter 1, and because Spanish is a European language, typologically distant from the local Creole, that migrated and took over the political and economic control of the islands. A continua model has been attributed to the sequential changes in Creole languages in contact scenarios where a Creole and a lexifier/non- lexifier language (i.e., lexically (un)related) coexist. DeCamp (1971) and Bickerton (1973) distinguish a Creole continuum in contact situations where the Creole may undergo three sequential stages: a basilect extreme, where the Creole is maintained closer to the speech of the colonial times; a mesolect or intermediate Creole sequence that is diverging from the basilect and approaching a more standard variety; and the acrolect or the educated standard, the extreme in which, the Creole tends to become more similar to its lexifier (Patrick, 1999). The ongoing changes from the basilect toward the acrolect has been known to involve a process of decreolization, which implies that a Creole is changing in a way that is increasingly resembling its lexifier (Snow, 2000).

Changes in Creole and lexifier/non-lexifier languages have been defined under two main categories: spontaneous and non-spontaneous (Snow, 2000). In spontaneous changes, a Creole language produces internally motivated innovations, whereas in non-spontaneous changes, the language undergoes changes driven by contact. Both types of linguistic change have a direct relationship with the literature of language change and contact posited previously and have been

applied to contacts situations in which a Creole language is involved. The overapplication of the continua model and the decreolization process in cases where a Creole is in close contact with a national non-lexifier language has been commonly used to explain this type of contact phenomena. For instance, Washabaugh (1974) indicated that Old Providence Creole was decreolizing and moving toward British English, a non-spontaneous change. However, as Snow

(2000) suggested a Creole in close contact with a non-lexifier national language “cannot be said to be decreolizing”, but it is certainly experiencing some type of linguistic change, be it spontaneous or non-spontaneous (p. 341). Herzfeld (1977) suggested that Spanish has displaced

English as the acrolect language in the continuum by means of process labeled ‘second language acrolect displacement’ in Limon Creole speakers, an English-based Creole spoken in Costa Rica

(p. 145). In a case that resembles more a language shift, Le page and Tabouret-Keller (1985) suggest that the French-based Creole spoken in St. Lucia is undergoing a continuum of varieties that range from the basilect French-based Creole to a relexified creolized English at the other end.

According to DeCamp (1971) a Creole continuum can only occur during direct contact between a Creole and its lexifier, and thus, decreolization would only arise under such circumstances (Rickford, 1987; Bickerton, 1980). Similarly, Aceto (1999) and Snow (2000) argue that Creoles in contact with lexically unrelated languages will undergo some type of change that would arise differently under lexifier/Creole contact situations. As a result, it seems that intimate contact between a national non-lexifier language, such as Spanish, and an English- based Creole may suggest a different situation than other scenarios were the Creole is in direct contact with its lexifier (Snow, 2000). A decreolization scenario in the Archipelago of San

Andres is partially plausible, but very unlikely from a gradient of Islander to Spanish as

suggested by Forbes (1989). On the one hand, English is the prestigious acrolect in the Raizal community and the language of the bible and the Protestant churches in ethnic neighborhoods.

On the other hand, English seems to be used only on these specific contexts and during interactions with non-Spanish speaking tourists. This linguistic separation based on the domain of use resembles a case of diglossia and a stable relationship between English and Islander

(Ferguson, 1959). In contrast, there seems to be an unstable relationship between Spanish and

Islander, where Spanish has increasingly been used in contexts reserved only for Islander Creole in the past, such as the nuclear family and peer groups (Bartens, 2013). Some researchers even report pessimistically intergenerational halt in the transmission of Creole among Raizales

(Bartens, 2013). Although, I argue that Islander Creole is still passed on from generation to generation in certain areas densely populated by Raizales with a strong Raizal identity (i.e., La

Loma, San Luis and ethnic neighborhoods of Old Providence), there is extended bilingualism in the Raizal population, which may signal the first stages of language shift rather than diglossia

(Snow, 2000).

Given this perspective, the bilingual speech community of the Raizales presents a great opportunity to study the outcomes of a Creole and non-lexifier language contact phenomenon, particularly from the variable production of a phonological structure. Precisely, variationist sociolinguistics has been built upon the study of phonological variation (Tagliamonte, 2012).

The idea that systematic, rule-governed variation is an inherent part of languages is the foundational maxim of language variation, and it is through the study of sound change that important contributions to a theory of language change have been produced (Labov, 2010, 2001,

1994, 1963). Weinrich, Labov, and Herzog (1968) address the notion of an innovative form taking the place of another, either through time or along some geographic or social dimension as

the “possibility of describing orderly differentiation” (Tagliamonte, 2012, p. 2; Sankoff and

Thibault, 1981). The orderly differentiation that has been pinpointed by these influential figures corresponds to the anomalies or non-standard forms that appear in a language in seemingly unsystematic ways, but to the eyes of the sociolinguistic researcher, its complex nature can be tested empirically through a data-based approach and using statistical methods. At the end, such linguistic patterns can only be understood by means of a strong interpretative component grounded in the researcher’s ethnographic observations of the real-world use of language within the speech community under investigation (Tagliamonte, 2012). Precisely, it has been noted that phonological variation is very sensitive to the influence of social factors (Tagliamonte, 2012), and the language contact situation in the Archipelago is a prime case for these factors. In such contexts, the comparison of linguistic varieties under the comparative sociolinguistic framework has been useful in language contact scenarios (Tagliamonte 2012; Meyerhoff, 2009;

Tagliamonte, 1997; Bickerton, 1975). The comparative method tests the connections implied in linguistic variation within the possible sources of influence (Tagliamonte, 2012) and the genetical relationship between languages (Ferguson, 1990). The systematic comparison of a linguistic form and its language behavior across datasets allows for the interpretation of the relationship of the linguistic systems involved. Based on the variationist framework, when the conditioning effects of the variable linguistic feature show patterns similar to a language source, then it can be interpreted that such structure is drawn from a specific linguistic system if it is determined that it existed before contact. On the contrary, when the conditioning effects are dissimilar, then we can conclude that the variable phenomenon belongs to a different linguistic system (Tagliamonte, 2012).

For this work, once determining that a potential transfer of phonological features is in place from the source language to the recipient vernacular L2 language, the second step consists in confirming that the social factors that condition transfer of such phonological features are spread across generations of bilingual Raizales, as opposed to marginal cases of substratum interference. Precisely, Van Coetsem (1988) argues that bilingual speakers are the agents of change in a linguistic transfer situation. Mainly, the direction of transfer is always carried out from the L1 or SL to the L2 or RL, and the agents of the transfer might involve either the dominant speakers of the L1 or the L2-dominant speakers. Following Thomason (2001), it is necessary to also look into the presence and variable production of the same phonological feature in the bilingual variety and, subsequently, Raizal Creole and Continental Spanish. More precisely, the analysis of the production of rhotics in the Raizal Creole variety needs to be related to the envelope of variation or the overlapping space where all variants may occur in the interlanguage (i.e., the vernacular in sociolinguistics terms) of the speech data (Aaron, 2010). For the time being, this rhotic realization is not present in the Spanish variety in contact with the

Raizal speech community, and thus, we can assume that transfer will be available through imposition of forms from the L1 (Van Coetsem, 1988).

After we establish a case for potential structural transfer of phonological forms, then we might consider the influence of social factors in contact-induced change. I have collected data from different age groups of Raizal speakers in two islands with a different sociolinguistic situation, in terms of its influence and contact with Spanish. I have also identified several social factors that were observed through attentive fieldwork, which might condition phonological variation in these communities. These factors together allow researchers to track the spread of variation, and more importantly, the direction of change across generations of bilinguals in two

comparable communities of the same raizal affiliation. Ultimately, the incorporation of a strong statistical component of sociophonetics and variationist research and the ethnographic survey of the islands will act jointly to interpret the evidence and make clear the nature of the potential ongoing phonological change. I look into both internal and external factors conditioning language change in the speech communities of the archipelago of San Andres. I consider the bilingual L2 Spanish speakers as legitimate speakers of either language who are agents of the potential structural change in progress. Moreover, I argue for examining the potential phonological outcomes of contact through the lens of comparative linguistics, in which I compare the sound variables produced in the languages in contact in the Archipelago. This allow me to corroborate a possible phonological interference in the L2 from the L1 of the informants.

Language contact and the processes of change are better understood examining the historical social forces playing a significant role in a speech community. To this aim, a closer ethnographic exploration and careful sociolinguistic fieldwork were indispensable to determine the multiple factors working in the potential linguistic transfer and the nature of the ongoing change. In this project, I recognize that adducing any linguistic change to the simple contact between languages is a big enterprise. Rather, we have seen that both changes are due to language-internal processes (i.e., spontaneous) and to external factors resulting from contact

(non-spontaneous). The case of the Archipelago of San Andres will be methodologically tested to determine any potential influence from the languages in contact.

2.2 Rhotic Variation In Monolingual And Bilingual Settings

I now turn to the analysis of the sound category selected for this work, rhotics. These sounds can be produced as trills, taps, approximants or fricatives (Thomas, 2011, p. 129;

Ladefoged and Madiesson, 1996). In what follows, I will examine the common acoustic cues that

describe these sounds in Spanish and English, and the research that have dealt with rhotic variation in monolingual and bilingual settings.

In Spanish, rhotic sounds are contrastive in word-medial intervocalic contexts, and they are traditionally described as either alveolar taps or trills (Navarro-Tomás 1999; Hualde, 2005).

In alveolar taps, the apical portion of the tongue bounces a single time against the alveolar ridge or palate, while alveolar trills require at least two of these articulatory gestures (three times in normative terms). In addition, there are cases in which the uvula vibrates against the tongue dorsum producing a uvular trill in some languages (i.e., French and German) and even some

Spanish varieties (Willis, 2006). This study will focus on alveolar trills as no uvular segments were attested in the data. The distinction between taps and trills is remarkably easy to recognize on a spectrogram, as the tongue articulator produces either one vibration or bouncing for taps, and at least two for trills. Acoustically, these sounds could be potentially measured according to how the oral cavity is configured to modify the airstream produced from the lungs. Based on this conceptualization, the mouth is seen a filter and the oral cavity is viewed as consisting of one or more tubes that favors certain high-frequency resonances or formants. This model of human speech is known as the source-filter theory, and it is important for the study of vowels and certain consonants (Kent and Read, 2002). The measurement of formants is a common practice for researchers looking for the configurations of rhotics cross linguistically. Typically, the three first formants are informative enough to determine the characteristics of vowels and liquids

(Kent and Read, 2002): low frequency formants are in the range of 100 to 1000 Hertz (F1) and describes the height of the tongue; mid-range frequency formants show the frontness or backness of the tongue (F2); and F3 formants dwell normally in the range of 2000-3000 Hertz and refer to the length of the vocal tract.

For our interests, however, these measurements are more suitable for English rhotics than for Spanish trills and taps, as their unifying features are hard to determine based on formant configuration, since apical rhotics show weak or absent F2 formant values (Thomas, 2010, p.129) and there is interruption of the formant lines due to lingual occlusions. Despite this, there are other phonetic cues for differentiating these Spanish segments, in addition to the measurement of visible apical closures in a spectrogram. Mainly, the segmental duration has also been used as a reliable measurement, perceptible enough to distinguish between taps and trills in monolingual and bilingual communities. In a study of a monolingual variety, Bradley and Willis

(2012) investigated the variable production of rhotics and the correlates for the tap-trill contrast in the Spanish of Veracruz, Mexico. They found that these variants present reduction and elision of lingual closures, but those with a measurable contact are produced long enough to be measured in terms of segmental duration. Overall, trills are produced in this variety with one or two lingual contacts with a pre-or post-approximant phase, and taps are produced with no visible contact in most cases. In another acoustical study of a bilingual speech community in the greater area of Chicagoland, Henriksen (2015) analyzed the potential loss of the phonological trill-tap contrast in 16 first- and second-generation Spanish-English bilinguals. He found that most of these bilinguals produce the phonemic trill either with one apical occlusion, or as a non-trilled r- colored variant, potentially resulting in loss of contrast between the two Spanish rhotics.

However, he found that almost all speakers realize the Spanish tap-trill contrast by means of segmental duration. What’s relevant in these studies is that speakers maintain the contrastive distinction just by using different production cues. Later in this section, I will come back to the production of non-prescriptive rhotic variants in spontaneous speech and the implications for this research.

English, on the other hand, has been suggested to have different allophonic variants that are hard to describe and label phonetically (Thomas, 2010, p. 133). However, they do have one thing in common: a lowered F3 value. For the sake of convenience, I will label the English rhotics as a retroflex approximant /ɹ/, which encompass those identified as ‘bunched-tongue’ approximant, either the tip-down bunched or the tip-up bunched types (Espy-Wilson at al, 2000).

English /r/ requires a maneuver involving the retraction of the tongue and the constriction of the pharynx. Formant values are sufficient to distinguish acoustically English retroflex /ɹ/. This approximant has the lowest F3 among English sounds, even to the extent that F3 patterns might be narrowly separated from F2 (Kent and Read, 2002). Hagiwara (1995) reports F3 mean measurements of 1500Hz among adult males, and 1950Hz in adult females. Generally, F3 formant values are distributed to the contiguous vowel segments producing, what is known, as r- coloring, which is associated as well with a low F3 closer to F2. On the contrary, Spanish rhotic articulation requires specialized articulatory gestures and aerodynamic conditions to produce a prototypical trill. These conditions include a stable apico-alveolar closure with more apical retraction and predorsum lowering than the tap (Recasens and Pallares, 1999). Solé (2002) also notices that trills require higher pressure build-up, greater magnitude of linguo-palatal contact and longer duration of the first closure to be produced (p. 686). According to this author, small variations in oral pressure results in no trilling or devoicing.

At first sight, when cross linguistically contrasting the acoustic nature of rhotics in

Spanish and English, we encounter a clear way to distinguish both segments in terms of their articulatory production. Moreover, the Spanish alveolar tap and the American English alveolar tap produced as an allophone of /t/ and /d/ are nearly identical, although they are not mapped equally due to orthographic interference (Face, 2006). As a result, in terms of associating

Spanish and English rhotics, there is a clear distinction between the Spanish alveolar trill and the

English retroflex /ɹ/.

Given the above, there are problems with the traditional categorization of Spanish trills and taps in terms of number of closures: they are not being uniformly produced across Spanish dialects with a trend towards innovative weakened variants. In other words, there is high variability in the production of Spanish trills and taps in monolingual varieties of Spanish. As noted by numerous authors cited here and elsewhere (Henriksen (2014; 2015); Bradley and

Willis (2012); Lipski, 2011; Thomas, 2010; Willis, 2007; Colantoni, 2006), Spanish rhotics are realized through multiple allophonic variants, conditioned by sociolinguistic factors, such as social strata, age, regional dialect, and gender. In particular, the vibrante multiple /r/ might be realized as a trill, fricative, or approximant.

Innovative features of Spanish rhotic production are being increasingly documented in the Hispanic world. Henriksen (2014) describes innovative weakened variants with zero-apical occlusions in the Spanish of Central and Northern Spain. Similar results were found in Southern

Spain (Zahler & Daidone, 2014). Henriksen (2015) later studied a bilingual community in the

Chicago area reporting trills with reduced number of closures and zero-apical occlusions for both trills and taps. When the zero-closure variant appeared, it was classified as either an alveolar approximant or an assibilated /r/. Similarly, Thomas (2010) also presents a voiced retroflex sibilant in intervocalic position in some parts of Latin America. Bradley (1999) also describes an assibilated rhotic in Ecuadorian Spanish produced in informal registers. In some parts of

Dominican Republic, a pre-breathy voice trill has also been studied (Willis, 2007). Approximant and fricative variants have also been reported with an extensive period of r-coloring into the following vowel, in some instances when no occlusion was produced (Bradley and Willis, 2012).

The same authors also claim that all informants realized the non-normative trill with different frequencies in women and men (i.e., females ranged from 14% to 59%, and males from 38% to

90%). Diaz-Campos (2008) also presented an acoustic analysis on Venezuelan Spanish that revealed the common production of approximants when compared to typical trills. Moreover, in a phonetic analysis of intervocalic trills in the Spanish of Argentina, Colantoni (2006) found that the normative trills are not common in semi-informal registers, and there seems to exist a continuum from fricatives to approximants, similar to other Spanish varieties. These studies reporting the variable production of Spanish rhotics in different Spanish varieties raise ample questions regarding the sociolinguistic implications of this ongoing sound change (Diaz-Campos,

2008), and the reliable acoustic measurements for determining the exact nature of non-vibrant rhotics in the bilingual Spanish speech of the communities of San Andres and Old Providence.

Clearly, an additional analysis of these rhotic variants is required to correctly identify and label the fricatives and approximant segments that are potentially distributed in the data collected.

As we can see from the description of the rhotics in both languages, the realization of rhotics in English and Spanish is quite different in terms of the articulatory conditions necessary for their production, even allowing the perceptual distinction among speakers of Spanish and

English. In terms of acoustic measures, F3 formant patterns in English suffice to characterize the retroflex approximant. In contrast, the classification of Spanish rhotics face two distinct approaches. First, vibrant rhotic realizations can be examined by means of a spectrographic analysis of apical closures and segmental duration. Subsequently, the nature of non-vibrant variants containing zero-apical occlusions needs to be determined. Several studies have pointed out that these variants can be classified as either assibilated rhotics or approximants. However, the degree of resemblance of these Spanish approximants to the description of English retroflex

approximants is yet to be determined. In what follows, I will propose different acoustic measures to establish the nature of the potential zero-closure rhotics in the speech data collected.

2.2.1 Variable Production Of Rhotics In Contact Situations: How To Account For The Articulation And Distribution Of Non-Vibrant Rhotics In The Archipelago Of San Andres

Recall that we have mentioned at the beginning of this subsection that rhotic production differs according to the manner of articulation. Precisely, rhotics can be further categorized according to the vibration or lack thereof of a speech organ against another given certain specific aerodynamics conditions. Ladefoged and Maddieson (1996) describe the production of a trill as the vibration of a movable part of the vocal tract placed close enough to another surface that a current of air with the right force narrowly passes through this configuration, producing the repeating closing and opening of the channel of air. When this vocal configuration produces a periodic vibration, then a trilling occurs. The Spanish language makes a distinction in the number of vibrations of the articulator (i.e., the tongue) to distinguish between two types of vibrant /r/s.

R-sounds with a single short vibration are classified as taps, while longer multiple vibrations of the tongue with the upper mouth surface are known as trills. Technically, both sounds can be identified as trills, but Spanish phonemically contrasts taps and trills in intervocalic positions.

The opposite of a vibrant trill is a non-vibrant rhotic. These types of sounds are produced with no vibration of the articulators and are classified as either approximants or fricatives, due to not displaying any type of occlusive phase.

A survey of the literature on rhotic variation in bilingual speech reveal the need for more studies on communities where languages are coexisting in the same environment. In a study on bilingual speakers, Henriksen (2015) described the zero-apical occlusion (i.e., non-vibrant rhotic) as “auditorily […] similar (but not identical) to the American English approximant rhotic” (p.

297). He later omitted to determine the manner of articulation of this variant, arguing that it was

either produced as an assibilated rhotic or as a weakened approximant. Aikhenvald (2006) investigated the confluence of indigenous languages in the Vaupés region (Colombia) limiting with the Amazonian basin, which have resulted in accommodation of different rhotic-like sounds due to contact. In particular, the East Tucanoan family of languages has exerted an influence in the Tariana flapped /ɾ/ in contexts where it should appear, resulting in a reduction of the system and accommodation of variants when two flaps are present in the same word (i.e., warikiri

‘young man’ is produced in the surface as walíkiri) (p.40). Weisglass (2015) has shown that

Spanish-Basque bilinguals produced mostly taps with only a few trills in onset cluster where a rhotic is present. She also demonstrated that voicing of the preceding consonant, rather than place of articulation and the subsequent vowel, had an effect on the realization of rhotics by these bilinguals. Likewise, she also reports that monolinguals produced mostly approximants, when the normative trill would be expected in monolingual production. Research on other phonetic variants has also shown effects in bilingual varieties. An early study on bilingual phonology was published by Flege (1987), who analyzed VOTs of initial /t/ by French-English bilinguals living in the United States for 12 years. When contrasted with the VOTs of French and

English monolinguals, the bilinguals produced intermediate VOT values higher than the French monolinguals. In what seems a case of convergence (Klee and Lynch, 2009, p. 30; Bullock and

Toribio, 2004), this research documented that bilinguals were able to detect the phonetic differences in the VOTs of the two languages but failed to establish a new phonetic category for the L2 English /t/.

The question remains as to the exact nature of non-vibrant rhotics produced by bilingual

Raizal speakers that can be traced to a common language source. What this review of studies conducted on contact communities shows is that phonological change in bilingual speech follow

a trajectory in which phonetic variants gradually coincide with those allophones of the dominant language, which constitutes a case of phonological interference (Thomason, 2001; Winford,

2003). Therefore, the driving assumption on the distribution of non-vibrant rhotics in the variety of Spanish spoken in the Archipelago involves the transfer of phonetic features in the bilingual speech that resemble those of the donor language. In other words, the realization of these rhotics in the vernacular speech will potentially correlate to instabilities in the production of rhotics across generations of Raizal Creole speakers, constrained by different degrees of contact in the

Archipelago. Specifically, young Raizal informants in San Andres might show signs of convergence more related to the monolingual Spanish variety depending on sociolinguistic factors, such as education and age. Old Providence, on the other hand, presents a different sociolinguistic situation, due to historical events that left this island out of the commercial expansion project of the mid-1950s. As a result, this Raizal community might exhibit a greater level of phonological interference from the L1 English-based Islander Creole. Regarding a similar contact situation, a retroflex rhotic variant has been documented in Belizean Spanish

(Balam, 2013) and in Limon Creole/Spanish bilinguals (Zimmer, 2011). It should be noted, however, that the reports on the Limon Creole retroflex approximant were mainly impressionistic. Bartens (2009) also noted a retroflex post-alveolar variant in young Islander natives, but this information was anecdotal. Therefore, the variability of rhotic production in these communities merits further empirical study.

For our purpose, it would be crucial to test this hypothesis with the appropriate phonetic measures to determine the correct acoustic correlates that characterize the production of non- vibrant rhotics in both the monolingual and bilingual varieties of these communities, including the immigrant Colombian Spanish speakers. A great deal of non-trilled realizations in Central

American Spanish and certain Colombian Spanish varieties have been documented, suggesting that multiple-contact trills are rare in general. In a survey of rhotic production in Central and

South American Spanish varieties, Bradley (2006) only found that 16% of syllable-initial tokens could be classified as trills (containing two lingual closures) in /sr/ clusters across word boundaries, and only one token presented the prescribed three lingual-contact closure.

Colombian informants realized non-trilled variants in a 4 to 1 proportion compared to normative trills. From these, 60% percent constituted approximant realizations and another 20% fricative rhotics. We must note that the informants selected for this study exhibited zero /s/ reduction in post nuclear position, which differs from the description of the vast lowland immigrant residents in the Archipelago. Fricative realizations in syllable-initial position have also been documented in the central highlands of Colombia conditioned by sociolinguistic factors (Lipski, 1994).

Crucially for this project, the distribution of both vibrant and non-vibrant rhotics will depend not only on the sociolinguistic background of the informants, but also on language-internal constraints, including the phonological context. Similarly, it seems that rhotic stridency is significantly dependent of the phonological context. In the same study by Bradley (2006), strident rhotics were favored after /s/ where no pause is present, whereas non-strident realizations are favored after /n/ in postvocalic positions. A common trend was found between strident and non-strident realizations in word-medial intervocalic contexts. In sum, these results suggest that

Latin American varieties, including the Colombian Spanish variety with no /s/ reduction, exhibit the production of rhotics that fall outside the classification of trills. Among the non-trilled realizations two rhotic categories are found in different phonological contexts: fricatives and approximants.

In order to determine the degree of phonetic transfer from a language source, the sociophonetic analysis of non-vibrant rhotics requires accurate identification of the acoustic correlates of the rhotics produced in the three varieties. First, it has been noted that the degree of periodicity, or the measurement of repeating patterns of the waveform, is a reliable acoustic measure to distinguish between fricative and approximants, as assibilated rhotics require more constriction to produce the turbulent airstream particular to fricative sounds (Colantoni, 2006). In a study on Argentina Spanish, for example, Colantoni (2006) showed that the measurement of the cepstral peak can reveal the degree of periodicity that correspond to both assibilated and approximant rhotic variants, resulting in the identification of more periodic approximant rhotics and the less periodic assibilated segments. Acoustically, both standard trills and approximant rhotics, are similar with regards to the degree of periodicity. Another subsegmental approach to measure spectral properties involves the center of gravity (COG). COG measures the maximal frequency at which “sonic energy is maximally concentrated” (Erker, 2010, p. 13). Consistent higher COG values correspond to English post-alveolar sibilants and Spanish fricatives, but lower values will show less assibilated backed realizations. Similarly, other spectral analysis could be carried out to measure the backness of rhotic sounds: kurtosis and skewness. Positive kurtosis and higher skewedness values are related to backed sounds (Jongman et al, 2000), suggesting a possible postalveolar approximant rhotic variant.

In addition, and as pointed out previously, formant trajectories are mostly sufficient to account for English-like retroflex variants. In addition to a lowered F3 (Hagiwara, 1995), it has been suggested that manner of articulation of English [ɹ] can be calculated by measuring the distances between F4 and F5: a greater distance between F4 and F5 in the vicinity of 1400Hz indicates a more retroflex articulation, while a smaller distance, around 700Hz, indicates a

bunched-like rhotic (Zhou et al, 2008). Olsen (2012) found F4-F5 separation values ranging from

525Hz and 1603 Hz with a mean of 1052Hz within a continuum of retroflex and bunched [ɹ] in

L1 English-speaking males learning Spanish as their L2. Such measures were also replicated in a contact Spanish variety spoken in Northern Belize (Balam, 2013).

In sum, the distribution of non-vibrant rhotics in the vernacular Spanish speech of the bilingual Raizal members of San Andres and Old Providence needs to be acoustically compared with non-trilled production of Islander and Continental Spanish. The acoustic correlates of the non-vibrant rhotics produced in these contact languages will be used to compare the samples of the rhotic segments across population groups. As it has been suggested in language contact scenarios (Tagliamonte, 2012), the transfer of phonological features can be traced back according to generational changes and across datasets. Indeed, the variation patterns of non- vibrant rhotics in the Archipelago will be compared across generations and between language sources in this sociophonetic analysis.

2.2.2 Enhancing Our Understanding Of Contact Phonologies In Spanish Vibrant Rhotics

We have covered a description of this project in terms of the acoustic analysis that will be carried out in the non-vibrant rhotics of the varieties in contact. Likewise, I have shown the methods for establishing phonological transfer from contact. However, we have not yet addressed the analysis of vibrant rhotic production in the Archipelago. In this study, a vibrant rhotic is defined as an alveolar tap or trill that presents single or periodic lingual closures with the surface of the palate. Such rhotic characterization only pertains to the Spanish varieties present in the islands. As a result, the analysis of vibrant rhotics will only consider the influence of the monolingual Continental Spanish in the bilingual Spanish speech of Raizales, as Raizal

Creole presents non-vibrant rhotic realizations in their phonological inventory, and therefore, would not be expected to show evidence of influence from Spanish in this context.

While the examination of non-vibrant rhotics consists of the analysis of acoustic measurements produced by different population groups, vibrant rhotic variation in the Spanish varieties of the Archipelago will be studied by analyzing the effects of linguistic and extralinguistic factors. As is well understood in language contact scenarios (Aikhenvald, 2006;

Klee and Lynch, 2009; Sankoff, 2002; Silva-Corvalán, 1994; Thomason 2001), language-internal constraints might be in play conditioning the variable production of non-normative trills, along with social factors thoroughly documented during fieldwork. Precisely, several studies have found a confluence of linguistic and extralinguistic factors conditioning vibrant rhotic production in Spanish varieties. For instance, the phonological context might influence the variable production of Spanish vibrant rhotics, including syllable-final in prepausal position, consonant clusters, and intervocalic positions, the contrastive contexts where trills and taps alternate. Balam

(2013) found that the intervocalic trill position seemed the most vulnerable context, where

Belizean Spanish speakers mostly produced the retroflex variant or a voiced tap. In Spanish, trills and taps alternate contrastively in word-medial intervocalic position and word-initial contexts where a vowel is preceding (e.g. pero ‘but’ vs perro ‘dog’; a Roma ‘to Rome’ vs. aroma

‘aroma’). In all other contexts, the tap and the trill occur in complementary distribution (for a complete illustration see Henriksen, 2015). The phonological context, syllable stress and the grammatical category have also been used to examine trill variation in Spanish (Henriksen and

Willis, 2010; Diaz-Campos, 2008; Lastra and Buitragueño, 2006; Lewis, 2004; Diez-Canseco,

1997). For instance, a trill production was found to be more likely before /l/, /n/, vowels or a pause compared to contexts followed by a /s/ (Lewis, 2004; Diez-Canseco, 1997). Diaz-Campos

(2008) also reported a greater effect for voiced alveolar trills in word-initial positions as opposed to word-internal positions. Vibrant rhotics in the Spanish spoken in Jerez, Spain, also seem to

favor trill production in unstressed syllables (Henriksen and Willis, 2010), while longer words over 4 syllables appear significant for the realization of trills. Finally, Diaz-Campos (2008) also reported that adjectives and verbs favored trills.

As rhotic production seems to be constrained by certain linguistic factors, other socially conditioned factors also appear to condition a vibrant realization, such as sex, age, education, social class, origin, and speech style. For example, Diez-Canseco (1997) found greater production of trills during a word-naming task compared to interviews and conversations. It is generally understood that the ‘observer’s paradox’ poses a challenge for collecting vernacular speech (Meyerhoff et al, 2012), particularly in communities where certain variants have been stigmatized (Wardhaugh, 2006, p. 19). As such, the retroflex approximant has been noted to be a stigmatized variant in Costa Rica (Adams, 2002; Vásquez Carranza, 2006), and it might be probable that non-standard variants might be neutralized during fieldwork in the Archipelago.

For this reason, it might be reasonable in this study to compare naturalistic data with other speech production elicited either through story telling or collaboration tasks, to be certain of the description of this rhotic variant. For sex, females seem more prone to a standard trill production than males, a pattern also attested with other phonological variables in other Spanish varieties

(Holmquist, 2011). In contrast, results for age seem to be fluctuating between young and old speakers, as in Jerez younger speakers seem to favor a trilled variant (Henriksen and Willis,

2010), while in Caracas senior speakers favored trills (Diaz-Campos, 2008). On the other hand, a steadier pattern is found in the middle classes, who favored a trill production in Cuzco (Diez-

Canseco, 1997) and Caracas (Diaz-Campos, 2008), but appear to favor an assibilated trill variant in Mexico City (Lastra and Buitragueño, 2006). Finally, speakers in Cuzco urban settings favored a trilled variant compared to speakers of rural areas (Diez-Canseco, 1997).

As we have seen that trill production is constrained by a number of linguistic and extralinguistic factors, I will be testing the research questions on language variation and change posited in this work. Mainly, I will examine structural changes at the phonological level induced partly from external sources. Investigating this phenomenon will offer insight in synchronic variation in unstable areas of the language of bilingual Raizal speakers, vulnerable to external interference. As a result, studying the variable production of vibrant rhotics presents another valuable opportunity to study the sociolinguistic implications of potential language change in speech communities in contact with Spanish. At this point, I suggest that extralinguistic factors work together with language-internal constraints to induce transfer of phonological features in the contact variety. To my knowledge, no variationist study has looked into this phenomenon in the Spanish spoken by the bilingual Raizal community of the Archipelago of San Andres. By looking at the factors conditioning the production of these rhotics and providing statistical tests of the internal and external factors involved in their variable realization, I will contribute to understanding the outcomes of contact at the phonological level. These outcomes will be visible in a linguistic subsystem vulnerable to structural transfer, such as the bilingual phonology

Raizales (Thomason, 2001). This seems particularly true in older generations of islanders, who have acquired Spanish late in their lives with limited formal instruction in standard Spanish, a case that might resemble that of current Belize (Balam, 2013). In addition, the intensity of contact and the level of bilingualism might condition the degree of language change in these communities.

In sum, research on rhotics has shown that trills with zero-apical occlusions are very common in all the Spanish varieties surveyed, and from these, fricatives and approximants account for these innovative realizations. Although not empirically, studies that examine the

sociophonetics of bilingual speech communities in contact situations report transfer of phonological features in the vernacular speech that bear similarity with the source language

(Balam, 2013; Zimmer, 2011, Bartens, 2009). This suggests that a transfer hypothesis for the articulation and distribution of rhotics across generations of Raizales in the two islands under study is possible. I argue for the examination of this phenomena as a valuable opportunity to study the unexplored sociolinguistic outcomes of contact at the phonological level in these two communities. Moreover, vibrant rhotics also present an opportunity to examine the nature of variation across generations of Raizales in contact with Continental Spanish.

2.3 Summary

In this chapter, I have examined literature on two main issues concerning the analysis of the contact situation present in the Archipelago. The extensive research done on other different contact scenarios has paved the way to determine the complex influence of different linguistic and social factors in the rhotics produced in the Raizal communities of San Andres and Old

Providence and the languages coexisting in these territories. By dividing the analysis of sound variation according to the production of vibrant and non-vibrant rhotic realizations, I aim to uncover the acoustic and phonological patterns of variation between varieties and the spread of change across generations of bilingual Raizales. Through such empirical approach, this work is the first to look at the sociophonetics of non-vibrant rhotics and offer a variationist account of vibrant rhotics in the Archipelago of San Andres. The present study aims to expand our knowledge of contact situations involving an English-based Creole coexisting with a national non-lexifier language in the Western Caribbean.

CHAPTER 3 METHODOLOGY

This chapter illustrates the methods and procedures employed for the collection and analysis of the data for this study. I have divided this section in two parts: first I will explain the methods used for the collection of the data during fieldwork; then I will examine the procedures that are used for analyzing the speech data.

3.1 Data Collection

Fieldwork was conducted in the islands of San Andres and Old Providence (including

Santa Catalina) over two consecutive visits in Spring and Summer 2017. Over 70 hours of speech data were collected through sociolinguistic interviews, a picture story narration task

(Mayer, 1969), and two information gap activities (Baker & Hazan, 2011; Thomas & Szuztak,

2005), both of them intended to elicit specific phonetic segments that were deemed as potential focus of variation. Once permission from the IRB and informed consent was obtained, each recording took place in informal non-controlled environments, such as the dwellings of the informants or in public spaces, including beaches or streets. However, effort was made to obtain noise-attenuated recordings in closed spaces.

Purple icons: 1st Generation Yellow icons: 2nd Generation Green icons: 3rd Generation Blue icons: Continentals

San Andres Old Providence and Santa Catalina

Figure 3-1. Geographical distribution of the informants.

The locations within the islands mainly included ethnic neighborhoods and these are detailed in Figure 3-1. Speech data was recorded with a Zoom H4N Pro sampled at 44100Hz/32- bit and a Shure omnidirectional lavalier microphone with a mounted preamplifier placed as close to the mouth as possible. Sociolinguistic interviews were the main data collection technique used for this study. These were conducted in Spanish, lasted between 60 to 90 minutes and covered a range of familiar topics for the informants, including work, education, family, life in the

Archipelago, children, society, religion, and ethnic and territorial issues. There were two older informants who requested the interview to be conducted in English/Creole. Once the speech data were saved in a personal password-protected laptop, each recording was labeled and metadata information was obtained from each informant, including occupation, family, language background, education and age.

3.1.1 Participants

In order to examine the potential process of sound change, speech data from 45 island- born, long-time resident Raizal informants were collected with ages ranging from 18 to 89. In total, the informants included 23 in San Andres, 22 in Providencia, and 24 were males and 21 were females. Another eight speech samples were taken from monolingual Continental Spanish speakers and were either born or have lived in the islands for the majority of their lives. In order to observe generational changes in the bilingual speech data collected, each recording was classified according to three age groups in both research sites: first generation, second generation, and third generation. This classification was determined due to the number and ages of the informants. Thus, five recordings were obtained for each generational group in both San

Andres and Old Providence accounting for a total of 30 recordings of Raizal Spanish.

Table 3-1. Generations of Raizal Spanish informants. Table 3-1 presents a description of the Raizal Spanish informants recruited for this study. Generation Age Median Island Sex 1st Generation 72 San Andres 3 females, 2 males 2nd Generation 46 San Andres 3 females, 2 males 3rd Generation 28 San Andres 2 females, 3 males 1st Generation 61.5 Old Providence 3 females, 2 males 2nd Generation 43 Old Providence 3 females, 2 males 3rd Generation 26 Old Providence 2 females, 3 males

Speech data for islander Creole were obtained from 5 informants, some of whom were also interviewed in Spanish, but were asked to narrate the Frog story in Islander (Mercer, 1969), interacted with other Raizal informants or recounted Anansi stories. Anansi stories or the stories of the trickster spider are traditional folk stories from the Gold Coast in West Africa. Most

Raizal informants are acquainted with these stories and some female informants still tell these stories to their children. Continental Spanish informants were mostly recruited in the Island of

San Andres and only one was recruited in Old Providence (N = 8). For both Raizal Creole and

Continental Spanish informants, no generational classification was followed, as only representative samples of language were required to compare with Raizal Spanish.

3.1.2 Data Elicitation Tasks

One of the additional tasks included a narration of the picture book Frog, Where are you? by Mayer Mercer (1969) and it was implemented in 23 interviews. Bilingual Raizal informants were asked to tell the story in Spanish and Raizal Creole. This was done in tandem whenever possible. The other two tasks involved adapted versions of the jigsaw task by Thoms, Liao and

Szuztak (2005), and the Diapix task by Baker and Hazan (2011). Informants engaged in an information-seeking interaction, where they were requested to collaborate together to find the missing words in their graphics. Words were controlled to elicit trill /r/, word-medial stops /b d g/, and coda /s/ (see Appendix). These tasks were conducted in tandem, whenever two

informants where available on site, mainly informants that were previously interviewed and their relatives and friends being present in the household. All the informants reported that they previously knew each other and good rapport between them was observed. In addition, informants were asked to interact in the language they felt more comfortable. As a result, several language pairs were recorded, including monolingual and bilingual Creole and Spanish interactions. An additional lavalier microphone was used following the same recording process as in the sociolinguistic interviews. A total of 14 interactions were recorded, including 8 interactions in San Andres, and 6 interactions in Old Providence.

3.2 Data Analysis

3.2.1 Preparation Of The Datasets For The Speech Analysis

Before the analysis of the speech data could be conducted, several datasets were prepared for Raizal Creole, Continental Spanish, and each generation of Raizal Spanish. In total 40 recordings were selected (Raizal Spanish = 30, according to the generational groups in both islands, Raizal Creole = 5, Continental Spanish = 5) for both analysis of vibrant and non-vibrant rhotics. For each recording, approximately 120 rhotic tokens were obtained from a transcribed sample of 15 minutes of speech selected 15 minutes after initiated the recording. This number of tokens was established in order to maintain a comparable distribution per participant and per age group. In addition, the selection and timing of recording segments was established based on the assumption that rapport with the participant could be obtained 15 minutes after initiated the interview. Each individual token was visualized in a spectrogram and their phonetic boundaries were labeled and segmented in Praat (Boersma and Weenik, 2018). In total 4,266 tokens were obtained for all varieties (Raizal Spanish = 3,408, Raizal Creole = 328, Continental Spanish =

530). Additionally, each token and its corresponding contexts of use were classified in independent spreadsheets according to the population groups. Two sources of data were

compiled in each spreadsheet for two separate analyses: 1) acoustic information on the phonetic properties of rhotics and 2) a sociolinguistic classification of each token.

3.2.2 Phonetic Analysis

Three speech properties of rhotics were obtained for each individual token: duration, spectral moments, and formant frequencies. Since obtaining this information manually, given the sheer amount of data would represent a major endeavor, three Praat scripts were used to automatically extract the acoustic information of the labeled segments in the TextGrid, a separate space for annotating linguistic information in Praat. Duration was calculated for each segment in milliseconds from the rhotic segments previously labeled in the TextGrid and results were saved in a text file (Lennes, 2002). Spectral properties give information about the manner of articulation of rhotics and were obtained through three spectral moments: the mean of the frequency or center of gravity (COG), which provides information about the average concentration of energy in the spectrum, the tilt of the energy distribution (skewness), and the peakedness of the energy distribution (kurtosis) (Colantoni, 2006; Jongman et al, 2000; Forrest et al. 1988). These measurements were automatically obtained using a Praat script that extracts the spectra information of labeled rhotic segments and results were saved on a text file (DiCanio et al, 2013). Finally, the same procedure was used to obtain formant frequencies of rhotic segments.

This time, the Praat script was modified to extend the scope of the measurements and include F4 and F5 frequencies (Kawahara, 2010) as well as F2 and F3 values. Along with F2 and F3, F4 and

F5 provides information about the degree of resonance of postalveolar approximant rhotics, and thus, identifies a retroflex from a bunched-tip variant in English rhotics (Zhou et al, 2008). Once obtained, these measurements were saved on a spreadsheet and were further analyzed to determine the acoustic properties of non-vibrant rhotics produced in the different varieties of the

Archipelago. Data was submitted to R for statistical analysis (R Core Team, 2013). R packages

used in the analysis include: Tidyverse (Wickham, 2017), Rbrul (Johnson, 2009). Results are presented in Chapter 4.

3.2.3 Variable Rule Analysis

The second step in the preparation of the datasets involves the sociolinguistic classification of each token. The essential goal of this procedure is to prepare the datasets so a quantitative analysis of vibrant rhotic variation could be conducted. This analysis is done with the aim of understanding the behavior of dependent variables according to independent linguistic and social factors that condition variation (Tagliamonte, 2012; Sankoff, 1988). The statistical modeling to mathematically assess the relationship between the dependent and the independent variables is carried out by means of a multivariate regression that presents the output in the form of probabilities or factor weights, known as the variable rule analysis (i.e., Varbrul). The software employed by variationist sociolinguists to produce such analysis is called the variable rule program (Tagliamonte, 2012). For this work, I used a software package known as Rbrul

(Johnson, 2009), which runs on R, a scientific programming language (R CoreTeam, 2013).

Rbrul allows to account for the effect of random variables in the multiple logistic regression model, which are factors unlikely to be replicable and drawn from larger populations in highly unbalanced datasets (Johnson, 2009; Baayen, 2008). Such characteristics are the staple of sociolinguistic fieldwork, where data consists of unbalanced groups of lexical items not sampled across datasets and collected from individuals with highly variable production differences. For our purposes, I included the random effect of lexical frequency (token) and informant in the

Varbrul models of rhotic variation across datasets (File-Muriel, 2009; Bybee, 2001). In order to safely determine the application value for our analysis and the dyads that comprise the dependent variables, it is of utmost importance to establish the envelope of variation and the behavior of different rhotic variants across the independent variables (Aaron, 2010).

3.2.4 Envelope Of Variation

The realization patterns of taps and trills respond to an asymmetrical distribution based on their contextual position within the word. Taps and trills are mutually exclusive in only two positions: taps appear exclusively in complex-syllable onset positions, while word-initial contexts correspond exclusively to trills. Table 3-2 illustrates the general distribution of Spanish taps and trills.

Table 3-2. Contextual distribution of rhotics in Spanish. Context Rhotic Use Examples − Word Initial Trill /roxo/ − Complex Tap /trompo/ Syllable Onset − Preconsonantal Variable /barko/ or Word Medial tap/trill /baɾko/ − Word Final Variable /kantar/ or tap/trill /kantaɾ/ − Intervocalic Contrastive /karo/ vs tap/trill /kaɾo/

Although both segments may also appear complementarily in preconsonantal word- medial and word-final positions, taps generally take this position in spontaneous speech. On the contrary, taps and trills appear in phonemic contrast in intervocalic position. This contrast is supported by around 30 minimal pairs in the Spanish language (Bradley and Willis, 2012). The distinction between trills and taps becomes clear by the word environment in which they either mutually exclude each other, share the same allophonic context, or contrast phonemically.

The first step to determine the envelope of variation involves identifying the rhotic variants present in the bilingual and monolingual varieties of Spanish. As a result, it has been observed in the datasets three categories of production in the Spanish rhotics in the Archipelago.

First, segments can be produced with the normative count of lingual closures for both taps and

trills. On the contrary, there are instance where no apical occlusion is found in these segments.

Finally, there is an intermediate category, particularly, for trills, where a weakened variant shows only one lingual closure and an approximant phase. Acoustically, this is better observed through the spectrogram of each variant.

Figure 3-3. Tap with a normative lingual closure (left). Tap displaying no tongue vibration (right).

Figure 3-4. Trill with the normative lingual closures (left). Trill displaying no tongue vibration (right).

Figures 3-3 to 3-4 show the spectrogram of taps and trills produced with either the normative lingual vibration(s) or segments without any visible vibration. Similarly, it has also been documented in the data trills without the normative number of apical occlusions. In other words, trills with only one tongue vibration and additional formant structure after the lingual occlusion.

Figure 3-5. Trill displaying one lingual closure with a subsequent formant structure.

This can be seen in Figure 3-5, where the intervocalic trill is realized with only one lingual closure and with a continuous formant structure. It is worth noting that this phenomenon is only possible within trills, as taps with one vibration will be regarded as normative taps. All these rhotic variants show that the /r/ in the Archipelago is variable produced with two distinctive acoustical properties: rhotics realized with the normative lingual closures, and rhotics produced either without any tongue vibration or trills without the normative apical occlusion.

Accordingly, four different rhotic variants are classified in the Archipelago:

1. A tap with one normative lingual closure. This variant will be classified as a Canonical Tap.

2. A trill with two or more normative lingual closures. This variant will be classified as a Canonical Trill.

3. A non-normative tap that lacks any visible lingual closure. This variant will be classified as a Non-Canonical Tap.

4. A non-normative trill that lacks either any visible lingual closure or is produced with one tongue vibration. This variant will be classified as a Non-Canonical Trill.

While this classification makes use of spectrographic evidence and word context (for classification 3 in taps and 4 for trills), it needs to hold the principle of accountability in our data.

In other words, we need to determine whether this classification behaves similarly and shows frequency patterns that correspond to the canonical versus non-canonical distinction. To do this,

I cross-tabulated the patterns of use of all the four rhotic variants present in the data according to the position within the word. This is done to determine whether the behavior of the rhotic variants are proportional to both canonical and non-canonical categories in the same word positions.

Table 3-3. Distribution of canonical and non-canonical rhotics according to the position in the word in Raizal Spanish.

Variant Pos-word N Freq % cano-tap complex 239 0.167 14.8 Variant Pos-word N Freq % cano-tap final 208 0.145 12.9 cano-trill final 3 0.016 0.18 cano-tap intervocalic 824 0.577 51.3 cano-trill initial 115 0.646 7.16 cano-tap medial 156 0.109 9.7 cano-trill intervocalic 56 0.314 3.48 non-tap complex 373 0.338 20.6 cano-trill medial 4 0.022 0.24 non-tap final 113 0.102 6.2 non-trill initial 409 0.585 22.6 non-tap intervocalic 505 0.457 28 non-trill intervocalic 290 0.415 16 non-tap medial 113 0.102 6.2

As can be seen in Table 3-3, the overall distribution of taps and trills correspond to similar patterns of occurrence in both their canonical and non-canonical forms. That is, taps have the same proportions of occurrence in both classifications and the same can be said about trill.

For instance, taps in word-medial positions account for the lowest frequency of occurrence in both canonical and non-canonical classifications. In contrast, the highest frequency appears in intervocalic taps, while complex onsets and word-final have intermediate frequencies. This is consistent for both canonical and non-canonical taps. For trills, only seven tokens in word- medial and word-final positions appear in the dataset, while word-initial positions have the highest frequency of occurrence in the data for both canonical and non-canonical variants. These patterns are seen more clearly in Figure 3-6, where canonical and non-canonical variants show the same comparable proportions across word positions.

Figure 3-6. Visual representation of the distribution of canonical and non-canonical rhotics according to the position in the word in Raizal Spanish.

However, when taps and trills are compared in their respective canonical and non- canonical forms, it appears that canonical variants are more frequent in taps (N=1427) rather than trills (N=178). In contrast, this seems more frequently realized in their non-canonical forms

(N=699). Despite presenting different frequencies, these results suggest that the proportions based on the categories of canonical and non-canonical forms behave similarly and uniformly across rhotic variants in their corresponding word positions. While taps present higher counts compared to trills in both categories, trills appear more frequently in the form of a variant containing less than two lingual closures.

Now that we have established a similar behavior between canonical and non-canonical variants, the next step involves determining whether examining trills and taps conjointly, or on the contrary, conduct separate analysis of rhotic types. As it has been observed before (Table 3-

2), only word-final and word-medial contexts allows for taps and trills to appear non- contrastively and non-mutually exclusive. As observed in Table 3-3, trills in this dataset only appear in intervocalic and word-initial position, with the exception of 7 tokens that were

produced in word-final (N=3) and word-medial (N=4) contexts. This suggests that the envelope of variation can only be attributed to taps and trills in their respective canonical and non- canonical forms rather than the position in the word. Another reason to conduct a separate analysis of taps and trills according to their canonical and non-canonical forms involves the fact that these variants vastly coincide only in intervocalic positions where taps and trills are in contrastive distribution. In other words, no tap or trill are competing variants in any word position in Raizal Spanish. In contrast, both canonical and non-canonical forms appear in all word positions where a tap and a trill occur in their corresponding word context.

The previous analysis has achieved two important milestones for this study. First, we have identified the variable context where rhotic variants compete. That is, rather than being the case where either tap and trill variants are displacing each other, canonical and non-canonical taps co-occur in the same contexts, and thus, both variants are competing for the same position.

In the same manner, canonical and non-canonical trills also appear to be overlapping within the same contexts of occurrence. Secondly, these results indicate that the results of the Varbrul analysis, to be discussed in Chapter 5, will consist of two independent tests for each generation of Raizal Spanish and for the Continental Spanish dataset: 1) an analysis of tap variants, and 2) another analysis of trill variation. As a result, the application values for each test will involve: 1) canonical taps, and 2) canonical trills.

3.2.5 Coding Of The Sociolinguistic Data

After I have identified the dependent variable for the Varbrul analysis and the corresponding application values, the datasets were manually coded according to several independent variables. As detailed in Chapter 3, several linguistic predictors that condition rhotic variation have been identified in previous research (Zahler and Daidone, 2014; Henriksen and

Willis, 2010; Diaz-Campos, 2008; Lastra and Buitragueño, 2006; Lewis, 2004; Diez-Canseco,

1997), and these were used to determine the factor groups. The coding schema for the sociolinguistic analysis of vibrant rhotics is summarized in Tables 3-4 and Table 3-5.

Table 3-4. Linguistic variables and categories included in the quantitative analysis of vibrant rhotics. Variables Categories Categorical Position in the word Word-initial Intervocalic Complex onset Word-medial Word-Final Preceding segment High vowel Liquid Mid vowel Sibilant Low Vowel Stop Pause Nasal Following segment High vowel Liquid Mid vowel Sibilant Low Vowel Stop Pause Nasal Stress Posttonic Pretonic Tonic Number of Syllables Two- Three Three+ Grammatical Category Adjective Noun Adverb Preposition Conjunction Verb Continuous: F3 and Duration Random: Token and Informant

The factor groups or variables included for the linguistic analysis correspond to the word context, phonological environment, position of the rhotic according to word stress, number of syllables, and grammatical category. It is hypothesized that segments in word-initial and intervocalic positions, and hence rhotics followed by vocalic segments, would favor a canonical production due to showing longer durations (Diaz-Campos, 2008; Willis, 2007; Diez-Canseco,

1997). Likewise, unstressed syllables, words with more than three syllables, and adjectives and verbs seem to favor the production of canonical variants (Henriksen and Willis, 2010; Diaz-

Campos, 2008). Along with the random variables of token and informant, F3 and duration were

also included in the analysis. Since rhotics are phonetic segments realized, in essence, through acoustic correlates, continuous variables, such as F3 and duration, might show in more detail the variation behavior of vibrant rhotics across datasets. This is one of the benefits of using a Rbrul, as it also allows the submission of continuous variables into the model. The social variables coded for the analysis are detailed in Table 3-5.

Table 3-5. Social variables and categories included in the quantitative analysis of vibrant rhotics. Variables Categories Generation First Second Third Sex Male Female Education Level Secondary Tertiary Island of Dwelling Old Providence San Andres Occupation Home Informal Employed Student Speech style Task Interview

Based on previous research (Balam, 2013; Henriksen and Willis, 2010; Diaz-Campos,

2008; Lastra and Buitragueño, 2006; Diez-Canseco, 1997) and the ethnographic fieldwork conducted in the Archipelago, several social factors were identified and coded for each individual informant recruited for the study. Informants were classified according to age, sex, education level, island of dwelling, occupation, and speech style. It is assumed that younger generations and educated/formally employed women would favor a canonical realization

(Henriksen and Willis, 2010). Once the data were coded, it was submitted for statistical analysis on Rbrul and compared across datasets. Results of the quantitative variationist analysis are presented on Chapter 5.

CHAPTER 4 A SOCIOPHONETIC ANALYSIS OF NON-VIBRANT RHOTICS IN THE ARCHIPELAGO OF SAN ANDRES

In this section, I present the results of the sociophonetic analysis of non-vibrant rhotics produced between the population groups under study in the Archipelago of San Andres. I have divided this section in three parts. This section begins with the acoustic analysis of non-vibrant rhotics in terms of the spectral moments (center of gravity, skewedness, and kurtosis), formant frequencies (F2, F3, F4, and F5) and duration that characterize the rhotics in Continental

Spanish, Raizal Creole, and Raizal Spanish. The aim of this analysis is to determine the type of non-vibrant rhotics produced between the population groups in terms of the place of articulation, either (Spanish) alveolar or (English) post-alveolar, and manner of articulation, either fricative or approximant. Once the acoustic properties have been compared in each variety, I conduct a discriminant function analysis to determine the best acoustic predictors that better discriminate between Raizal Spanish, Raizal Creole, and Continental Spanish. Finally, I test the correlation of non-vibrant rhotics in Raizal Spanish across three generations (i.e., young, adult, and senior) using the best acoustic predictors found in the discriminant function analysis. These acoustic correlates are compared cross linguistically with the aim of examining a potential association of a Raizal generation influenced by contact with Spanish or Creole.

This first analysis aims at determining the acoustic similarities and differences in the rhotics produced in the local Spanish variety (i.e., Raizal Spanish) with respect to Continental

Spanish (i.e., the coastal regional Spanish variety) and the English-based Creole language (i.e.,

Raizal Creole). As noted earlier in the methodology section, the rhotics for this analysis involve the segments produced with zero occlusion in the three varieties. In Raizal Spanish and

Continental Spanish, these segments were chosen based on the absence of occlusion in the spectrogram and inspecting the abrupt changes in the waveform.

Likewise, rhotics were extracted in Creole speech by inspecting changes in the waveform and the lowering of F3 in the spectrogram. Attention was paid to formant transitions from vowels to liquid and r-coloring of contiguous vowels (Morgan and Sessarego, 2016; Bradley and Willis,

2012). Likewise, the relative low onset frequency for F3 also served as a cue to identify this phone (Kent and Read, 2002). Examples of how these rhotics are seen in Praat are presented in

Figures 4-1. to 4-3, which show sample rhotics in the three varieties.

Relative Low Onset Frequency for F3

Figure 4-1. Praat capture of the rhotic produced in initial position in the word [ɻif] ‘reef’ in Islander Creole. Note the lack of occlusion and the F3 lowering curve getting closer to the second formant (F2). There is also coarticulation of the preceding vowel segment (i.e., /i/), which surrounds the rhotic between vowels. Due to these characteristics, this rhotic is preliminarily classified as a postalveolar [ɹ̠ ], and tentatively a retroflex with no lingual palatalization [ɻ].

Abrupt change in the waveform

Figure 4-2. Praat capture of the zero-occlusion (i.e., non-vibrant) rhotic produced in initial position in the word [ɹajsal] ‘raizal’ in Continental Spanish. Note the lack of occlusions, the steady F3 line and the lowering of the second formant (F2). This rhotic is preceded and followed by vowel segments. Due to these characteristics, this rhotic is preliminarily classified as an alveolar [ɹ].

Abrupt constriction of the waveform

Figure 4-3. Praat capture of the rhotic produced in initial position in the word [ɹekwedo] ‘recuerdo’ in Raizal Spanish. Note the lack of occlusions and the steady F3 and F2 lines. There is also elision of the medial rhotic preceding the dental-alveolar stop (i.e., /d/), something quite common in this dialect. There is also coarticulation of the preceding vowel segment (i.e., ), which surrounds the rhotic between vowels. Due to these characteristics, this rhotic is preliminary classified as an alveolar rhotic.

After visually inspecting these spectrograms, it becomes puzzling whether the bilingual rhotics under investigation are converging in the direction of one of the varieties in contact

(Continental rhotics and Creole rhotics) or, on the contrary, have a stable interlanguage impermeable to contact. In order to test this, several acoustic correlates found to be significant to study rhotic liquids were extracted (Blecua et al, 2014; Erker, 2010; Colantoni, 2006; Jongman et al, 2000). These correlates involve: duration (measured in milliseconds), spectral moments

(center of gravity (COG), Skeweness, and Kurtosis), and formant frequencies (F2, F3, F4, and

F5). In total, 1928 tokens were labeled as non-vibrant rhotics in Praat, from which 1450 were extracted from bilingual informants (75%), 328 tokens from Creole speech (17%), and 150 from

Continental speakers (8%). The lack of occlusions served as the criteria to select these rhotics in

Continental and Raizal Spanish. Rhotics with one occlusion and with an approximant r-coloring phase were excluded from the analysis.

I now present the analysis of the acoustic properties of the rhotics in the three varieties of the Archipelago. First, in section 4.1, I examine the overall acoustic measurement in all varieties with the aim of: (1) establishing the essential differences between the two languages in contact

(Spanish and Creole); (2) setting the baseline for establishing a linguistic difference; and (3) identifying the acoustic correlates of Raizal Spanish with regards to the two varieties in contact.

Then, the acoustic correlates that best predict a discrimination between varieties will be selected in a discriminant function analysis in Section 4.2. Finally, I will present in Section 4.3 a correlation analysis to determine whether a generation of Raizal Spanish is associated with either

Creole or Continental Spanish based on the place of dwelling (San Andres or Old Providence).

4.1 Acoustic Properties of the Non-Vibrant Rhotics in the Varieties of the Archipelago

This section compares the overall acoustic properties associated with the rhotics in each variety. First, I provide an analysis of the acoustic differences of rhotics in terms of 1) spectral

moments, 2) formant frequencies, and 3) duration. The analysis of spectral values tests whether these rhotics are produced with a high degree of assibilation properties. Likewise, these measurements allow to determine how backed these rhotics are realized in these varieties. In other words, these measurements provide an acoustic measurement to differentiate between places of articulation (i.e., alveolar and post-alveolar) and between the two manner of articulation: approximant or fricative. Determining the manner of articulation allows to test whether an innovative assibilated variant is present in the Spanish speech of bilingual Raizales.

With respect to the place of articulation, the more backed the realization would refer to a postalveolar rhotic in contrast to an alveolar rhotic, an allophone expected in the monolingual

Spanish variety. The comparison of the formant frequencies also renders visible the differences between postalveolar and alveolar rhotics. As examined in previous chapters, the lowering of F3 frequencies and the separation of F3-F2 are reliable measures to identify the approximant liquid

/r/ (Espy-Wilson, 1992; Lehiste, 1964). English has been reported to present the lowest F3 frequency (Kent and Read, 2003; Hagiwara, 1995). As such, it is expected that the English-based

Creole would display lower F3 frequencies and smaller F3-F2 separations than bilingual rhotics and an even greater difference with Continental rhotics. Likewise, it is also hypothesized that the rhotics in the archipelago are produced with non-sibilant properties, as this was not attested perceptually during fieldwork.

Based on Hammond’s (1999) finding that canonical /r/ production in native Spanish speakers is rare, segmental duration has been used determine whether phonemic contrast between taps and trills is being maintained or lost in different Spanish varieties (Bradley and

Willis, 2012; Willis and Bradley; 2008). Moreover, rhotic duration has been documented to be useful to determine neutralization of the intervocalic trill/tap realization in understudied Spanish

dialects with divergent distribution patterns and coexisting with other languages (Balam, 2013).

Therefore, I examine the duration of the rhotic segments in the three varieties to determine

production differences. While Creole rhotics are compared in all contexts, I focus on taps and

trills in both varieties of Spanish (Raizal and Continental).

4.1.1 The Non-Vibrant Rhotics In The Archipelago

Overall, it was found that the acoustic values of the rhotics in the Archipelago differ in

terms of spectral moments and formants frequencies, which suggest that these rhotics are

produced differently in these varieties. The analysis of spectral moments reveals the manner and

place of articulation, while a comparison of formant frequencies is useful for identifying the

degree of alveolar/postalveolar/retroflex realization between rhotics in the varieties. Table 4-1

shows the token count, mean values and standard deviations for all the spectral windows

measured in the varieties under study.

Table 4-1. Mean values and standard deviations for spectral moment measurements in all varieties. Population Total Mean SD Mean SD Mean SD Count COG COG Skewness Skewness Kurtosis Kurtosis Raizal Creole 328/523 777 213 14.5 8.72 518 610

Raizal Spanish 1450/3925 755 230 15.6 8.94 542 719

Continental Spanish 150/690 861 337 12.5 7.73 351 483

COG captures the mean frequency of the rhotic realization, while skewness and kurtosis

obtain global aspects of the speech segments, such as spectral tilt and peakedness, respectively.

On average, Raizal Creole and Raizal Spanish values seem to be closer to each other (COG=777-

755; Skewness=14.5-15.6; kurtosis=518-542, respectively) while diverging from Continental

Spanish measurements in all spectral measurements (COG=861; Skewness=12.5; kurtosis=351).

Likewise, formant frequencies also display different values for all population groups. This is summarized in Table 4-2.

Table 4-2. Mean values and standard deviations for formant frequencies measurements in all varieties. Population Mean SD F2 Mean F3 SD F3 Mean F3- SD F3-F2 F2 F2 Distance Distance Raizal Creole 1334 193 1993 209 659 228 Raizal Spanish 1436 212 2247 329 810 303 Continental Spanish 1521 261 2633 290 1112 321

Rather than converging toward one of the varieties in contact, formant frequencies in the bilingual Raizal Spanish rhotics seem to be distributed between Raizal Creole and Continental

Spanish (F2=1436; F3=2247; F3-F2 distance=810). Overall, it seems that these acoustic predictors differ greatly in Raizal Creole and Continental Spanish, while Raizal Spanish appears to be between the two contact languages, suggesting that certain generation of Raizales may be converging toward either Islander or the monolingual Spanish variety. In the following subsections, a detailed description of the acoustic measurements divided by population groups will be presented.

4.1.2 Spectral Moments In The Non-Vibrant Rhotics Of The Archipelago

In general, the spectral moment measurements evidence a great deal of variability, showing outlier values across all varieties. From these, bilingual rhotics presents the most variability, displaying data points across a wide range of frequencies. This suggests that there are some individual differences within some bilingual informants, which might be associated with their level of proficiency or based on generation. For the purpose of this sociolinguistics study, we check this assumption of generation in further subsections.

On average, the mean high frequencies of the spectrum (COG) are different in the three varieties. As can be seen in Figure 4-4, the concentration of energy seems to be in the lower frequencies in Raizal Spanish and Islander Creole compared to Continental Spanish (755Hz,

777Hz, versus 861 Hz, respectively).

Figure 4-4. Boxplot with mean values of the first spectral moment (COG) in all varieties.

As can be seen in Figure 4-4, the concentration of energy seems to be in the lower frequencies in Raizal Spanish and Islander Creole compared to Continental Spanish (755Hz,

777Hz, versus 861 Hz, respectively). The mean ranges in COG frequencies below the 1000 hertz suggest that the majority of these rhotics are produced with a low degree of assibilation properties in all varieties. Colantoni (2006) reports significant inter-speaker differences for all eight informants of her study on assibilated rhotics in Argentina, ranging from 1300Hz to

5500Hz. Colantoni argues that overall lower values in the mean frequency may correspond to more retracted articulation as compared to English post-alveolar segments. Table 4-3 shows a comparison of COG values reported in previous studies for English sibilants, Spanish fricative trills, and the values obtained for the Rhotics of the Archipelago of San Andres.

Table 4-3 A comparison of COG values reported for English sibilants, Spanish fricative trills and the rhotics in the Archipelago. English Sibilants COG Spanish Fricative COG Rhotics in the COG (Jongman et al, Trill (Colantoni, Archipelago 2000) 2006) /f, v/ 5108 [ř] 1300Hz - /r/ Continental 861Hz 5500HZ /s, z/ 6133 -- -- /r/ Raizal Creole 777Hz /ʃ, ʒ/ 4239 -- -- /r/ Raizal Spanish 755Hz

When compared to each other, bilingual and Creole rhotics have more similar values, while Continental rhotics have seem to center higher on the spectrum. A one-way ANOVA confirms that there are significant differences between the groups in terms of COG means

(F=13.9, p= 9.8e-07). Tukey post hoc tests indicated a significant difference between two groups:

Raizal Creole and Continental Spanish (p= 0.001), and Continental Spanish and Raizal Spanish

(p = 6e-6). In contrast, the only group that appears to have similar mean frequencies are the Raizal

Creole and Raizal Spanish (p=0.26).

Mean values of the second spectral moment, skewness, represent another example of acoustic differences between varieties.

Figure 4-5. Density estimate and mean values of the second spectral moment (Skewness) in Continental Spanish, Raizal Creole, and Raizal Spanish (i.e., Means = 11.84567, 13.61165, and 14.45552, respectively).

Figure 4-5 shows an asymmetrical distribution of the spectral tilt in the speech signal and the probability distribution of the population groups. All varieties display a skewness mean above 10, which indicates a concentration of energy in the lower frequency. Due to the contrast in population size (N=1450, 328, 150 for Raizal Spanish, Raizal Creole, and Continental

Spanish, respectively), the density estimate for Raizal Spanish appears higher compared to the other two varieties. Positive skewness values for Spanish fricative trills have been reported in the range of 0.8 to 4.8 (Colantoni, 2006; Quilis and Carril, 1971). For English alveolar /s/ and postalveolar / ʃ, ʒ/ sibilants, mean values have been reported at -0.229 and 0.693, respectively

(Jongman et al, 2000). Table 4-4 shows the mean values for skewness reported in previous works compared to the values of the rhotics in the Archipelago.

Table 4-4 A comparison of skewness values reported for English sibilants, Spanish fricative trills and the rhotics in the Archipelago. English Sibilants Skewness Spanish Skewness Rhotics in the Skewness (Jongman et al, Fricative Trill Archipelago 2000) (Colantoni, 2006) /f, v/ 0.077 [ř] 0.8 -5.8 /r/ Continental 11.84567 /s, z/ -0.229 -- -- /r/ Raizal Creole 13.61165 /ʃ, ʒ/ 0.693 -- -- /r/ Raizal 14.45552 Spanish

The values previously reported indicate a high concentration of energy in the lower frequencies. The positive values reported in this study differ considerably in all varieties, which suggest a manner of articulation that diverges from sibilant consonants. For the three varieties in the Archipelago, the values are increasingly distant from 0, the mean of the spectral tilt, with a mean ranging between 11.8 and 14.5. This greatly differs from the higher frequency produced by the hissing noise typical of alveolar and post-alveolar fricatives, which gravitate around the negative tilt (-0.229), and Spanish fricative trills, which reach to 6 in extreme outliers (Colantoni,

2006). Articulatorily, the measurements for the rhotics in the Archipelago represent a vocal

gesture where rhotics are produced with less constriction than sibilant but not quite reaching lingual closure. As a result, these values suggest that the articulatory gestures to produce these rhotics correspond to an approximant manner of articulation not reaching yet the lower frequencies necessary to produce an assibilated rhotic.

Subsequently, a one-way ANOVA was conducted to determine if there were significant differences in skewness mean values between the population groups of the Archipelago. Results from the ANOVA revealed a between groups effect (F=9.816, p= 4.74e-05), and Tukey post hoc tests further compared the significant differences between groups. This output indicates that

Raizal Creole and Continental Spanish present significant differences(p=0.0476178), and a greater effect was found between Raizal Spanish and Continental Spanish (p=0.0000947). No significant difference was found between Raizal Spanish and Raizal Creole (p=0.1039952) in terms of skewness.

The third spectral moment, kurtosis, provides information about the properties of the distribution peaks (Forrest et al, 1988).

Figure 4-6. Boxplot with mean values for Kurtosis in the three varieties of the Archipelago.

The positive values reported here for kurtosis are an indication of a clearly defined spectrum with well resolved peaks, which are related to more backed realizations (Jongman et al,

2000). Figure 4-6 shows that Raizal Creole and Raizal Spanish seem to differ from Continental

Spanish based on the means of this acoustic value, suggesting that the rhotics in these varieties are produced with more backed realizations. Although these mean values relate to the overall production of zero-occlusion rhotics without regards to generation, kurtosis values are significantly different between varieties (F=5.281, p= 0.00516). Tukey post hoc tests further found a greater significant difference between Raizal Spanish and Continental Spanish (p=

0.0034053) and a significant effect between Raizal Creole and Continental Spanish (p=

0.0366817). In contrast, Raizal Spanish and Raizal Creole appear to be similar to each other based on spectral peakedness (p= 0.8309804).

So far, the analysis of the spectral properties in the rhotics of these varieties suggest that these segments are produced differently across population groups. Moreover, these values also indicate a manner of articulation that differs from fricative measurements reported in previous studies. Now, I turn to the analysis of formant frequencies which correspond to a reliable measurement of place of articulation for approximant postalveolar rhotics in English varieties.

4.1.3 Formant Frequencies In The Rhotics Of The Archipelago

The boxplots on Figure 4-7 reveal that there is greater variability in the bilingual groups compared to the Continental and Creole groups. Despite the number of outliers, the means of all data sets seem practically equal to the medians for all formant measurements. Continental

Spanish appear to have higher formant values on average, whereas Creole rhotics present the lowest means. Hagiwara (1995) reported F3 measurements for English /r/ in the range of 1300Hz to 2200Hz, which seem to be in the vicinity for Raizal Creole rhotics (~1993HZ).

Figure 4-7. Boxplot with mean values of formant frequencies in all varieties. The formants seen here correspond to the second and third formant (F2 and F3, correspondingly) from left to right. A final measurement corresponds to the distance of F3 minus F2.

The same author also reports a relatively smaller F3-F2 distance pattern. Again, the values reported here are also consistent with a narrower F3-F2 separation when compared to the other two varieties due to F3 lowering. Lower F3 frequencies indicate a shorter vocal tract length, while F2 increases as tongue moves forward (Kent and Read, 2002). As happens with

Creole rhotics, the lower F3 and F2 values suggest a more backed rhotic realization than

Bilingual and Continental rhotics, and it seems that the lingual gesture takes place in the postalveolar region. Zhou et al (2008) distinguish between a bunch and a retroflex realization of

English /r/ based on F5-F4 frequency configuration. In particular, higher F5-F4 distances of about 1300Hz are typical of retroflex ‘tongue dorsum down’ gestures, while lower separation of these formants below 1150Hz indicate a bunched ‘tongue tip down’ shape (p. 4478). Although, these tongue shape configurations seem to be irrelevant for phonemic contrast, they do seem to be associated to speaker’s identity.

Continental rhotics are produced with a mean around the 2600 Hz, close to the value reported by Borzone de Manrique (1980) for Spanish trills (2500Hz). Similarly, the F3 and F2 values for Continental rhotics appear relatively higher compared to all varieties, which suggest a more fronted place of articulation for Spanish rhotics. As revealed by the trends found in the

analysis of the three spectral moments, there are no indications that Continental rhotics, or the varieties spoken by the Raizales are produced with the presence of a turbulent airstream mainly due to an ‘insufficient narrow constriction’, but rather they lack the precise aerodynamic condition to produce a trill or tap (Colantoni, 2006, p. 22). In other words, the acoustic correlates for the rhotics in these varieties suggest an approximant manner of articulation rather than assibilated rhotic realization.

Formant values for Raizal Spanish rhotics seem to fall between those of Creole and

Spanish, which supports our assumption that bilingual rhotics are gravitating between the two languages. This pattern is better observed in Figure 4-8 where a comparison of F3 and F2 formants is displayed between all varieties in the two islands of the Archipelago. The trend lines seen in this Figure reveal that Raizal Spanish rhotics are realized with formant frequencies that lie in between Continental Spanish and Raizal Creole. Moreover, they are produced with greater variability in the F3 measurement than in the other two varieties, showing a more tilted regression line approaching either Continental rhotics or Creole rhotics at both ends of the slope respectively.

Figure 4-8. Scatterplot with regression line of formant frequencies in Raizal Spanish, Raizal Creole, and Continental Spanish in San Andres and Old Providence.

Results of the Pearson correlation test indicated that there is a significant positive association between F3 and F2 that increases across varieties first in Raizal Spanish (r(1448) =

.44, p = < 2.2e-16), then in Raizal Creole (r(326) = .36, p = < 2.2e-11), and in last place Continental

Spanish (r(148) = .32, p = < 4.467e-05). As hypothesized in this study, it appears that rhotics produced within certain subgroups of Raizal Spanish are increasingly resembling the formant frequencies present in the two contact varieties. However, this effect seems more robust in San

Andres than in Old Providence, where the slope of the trend line is more pronounced.

A simple linear regression was calculated to predict the formant frequency values across population groups. Results indicated that there was a significant effect between F3 frequencies and population (F(2, 1925) = 226.2, p < 0.001, R2 = .19) The individual predictors were examined further and indicated that Continental Spanish (t = 104.42, p = 2e-16), Raizal Creole (t =

-21.02, p = 2e-16), and Raizal Spanish (t = -14.57, p = 2e-16), were significant predictors in the model. Similarly, a simple linear regression was conducted between F2 frequencies and population (F(2, 1925) =47.25, p=< 2e-16, R2 = .046), showing significant differences between

Continental Spanish, Raizal Creole, and Raizal Spanish (t = 87.335, p = <2e-16, t = -8.909, p =

<2e-16, and t = -4.633, p = 3.84e-06 respectively). The same statistical analysis was conducted for the distance between F3 and F2 between population groups (F(2, 1925) =122.7, p=< 2e-16, R2 =

.1122), revealing significant differences in all varieties (Continental Spanish t = 46.43, p=<2e-16,

Raizal Creole t = -15.66, p = <2e-16, and Raizal Spanish t = -11.98, p=<2e-16).

Although these tests reveal significant differences between varieties, the visual display in

Figure 4-8 indicates that Raizal bilinguals might be associating their rhotic production with either Continental Spanish or Raizal Creole. As a result, we should analyze these formant frequencies further, examining the effect of age in the speech production of rhotics between San

Andres and Old Providence. Generational differences might reveal a change in progress in non- vibrant rhotics in the Archipelago. This is explored in detail in section 4.3.

4.1.4 Segmental Duration In The Rhotics Of The Archipelago

The final acoustic analysis involves the measurements of segmental duration. As opposed to the segmental values reported earlier, the duration for non-vibrant approximant rhotic segments have implications for the classification of Spanish non-vibrant trills and taps. It has been reported that the contrast between trill /r/ and tap /ɾ/ can be maintained through the overall duration of both segments when no lingual closure is present (Henriksen, 2015, in Chicagoland;

Balam, 2013, in Belize Spanish; Bradley and Willis, 2011, in Veracruz Spanish; Willis and

Bradley, 2008, in Dominican Spanish). Turning to the phonetic data from the Archipelago of San

Andres, I distinguish two methods for the analysis of the segmental duration of these approximant rhotics. First, I present an overall comparison of segmental duration between varieties. As I am interested in overall differences, no distinction between trills and taps will be made. Then, I contrast the two Spanish varieties, Continental Spanish and Raizal Spanish, in terms of segmental duration between trills and taps. As it has been noted in contact varieties

(Balam, 2013), neutralization of /r/ and /ɾ/ may occur in intervocalic contexts in bilingual speech production. However, Henriksen (2015) also found that heritage speakers in the Chicagoland variety maintained the tap/trill contrast in terms of segmental duration. As a result, I test both potential scenarios based on the position of the rhotic in the word.

4.1.4.1 A comparison of segmental duration in the rhotics of the Archipelago

For this comparison, it worth noting that Raizal Creole makes no phonemic distinction between tap and trills compared to Spanish, and thus, the comparison between varieties might reveal significant differences in word contexts. As a result, this comparison will serve the purpose of exploring general patterns of segmental duration across varieties. Figure 4-9 displays

the means duration of approximant rhotics in the Archipelago. Overall, Raizal Creole presents higher mean durations for rhotic realizations.

Figure 4-9. Boxplot with mean values for segmental duration in all varieties in all word contexts.

In contrast, Raizal Spanish seems to be converging towards the Continental variety with a difference of 0.0104 ms, compared to the mean distance in Raizal Creole (0.0378 ms). The monolingual Spanish variety shows shorter segmental duration for rhotics in all contexts. Now, when look more closely into these differences in terms of word context, we find that these linguistic varieties vary according to the position in the word. As seen in Figure 4-10, duration of the rhotic segment is dependent on word context and language. For instance, Islander Creole has longer segmental duration in all positions within words, while Spanish, either the Raizal or the

Continental variety, behave with similar patterns across different positions in the word. In general, rhotics used in complex onsets present the shortest durations in Continental Spanish,

Raizal Spanish and Raizal Creole, ranging from 0.019 ms and 0.035 ms to 0.055 ms,

respectively. In contrast, approximant rhotics in initial position have the longest duration in the

Spanish population groups, and the second longest segment in Raizal Creole.

Figure 4-10. Mean duration in milliseconds across population groups and position of the rhotic within the word.

A two-way ANOVA comparison indicated that segmental duration is associated with different population groups (p=< 2e-16) and correlated with different word positions (p=< 2e-16).

Likewise, a significant interaction was found between population and word position (p=< 2.75e-

14), which shows that the relationship between population groups and word position depends on the duration of the rhotic segment.

These results reveal that rhotic duration is significantly dependent on language variety and the position of the rhotic within the word, which confirms the patterns found in Figure 4-10.

In terms of the differences in rhotic duration within the word, a Tukey post hoc test indicated that rhotics in medial and final position (p=0.5775459) and in medial and intervocalic position

(p=0.5143831) present no significant differences in segmental duration. In other words, these

100

word positions fail to show significant differences, and thus, rhotic durations significantly differ in word-initial position and complex onset syllables.

Contrary to phonemic trills in intervocalic position, approximants in this context present the second shortest duration in most varieties, except in Continental Spanish, which presents one of the shortest durations right before medial positions. It has been reported that Spanish trills have a duration ranging from 60 to 90 ms (Henriksen, 2015; Bradley and Willis, 2012), considerably higher than the approximants reported here. The differences in duration suggest different articulation processes in the aerodynamic conditions that support the trill realization and the subsequent lingual closures that ensue. Solé (2002) showed that specific aerodynamic conditions need to be met in order to maintain the appropriate air pressure to produce the voiced trills. Any variation in the magnitude of 2.5 to 5 cm H2O ‘impaired voiced and voiceless trills’

(p. 686). Any larger drop of the oropharyngeal and subglottal pressure will result into an approximant realization (p. 682). Thus, the articulatory gestures to attain a successful tongue-tip trilling require precise aerodynamic conditions and lingual positionings that contrast with the conditions present in an approximant allophone.

These aerodynamic conditions partly explain the differences in mean durations in intervocalic position between the Spanish approximants in the Archipelago and trills. The longer durations for Raizal Creole rhotics seem to be language-specific, a pattern that Raizal informants seem to fully differentiate. Despite appearing with longer durations across all word contexts

(except for word-final positions), segmental durations in Raizal Spanish are approaching

Continental Spanish rather than diverging towards Raizal Creole. Final-word contexts seem to be the only position different in both Spanish varieties and Raizal Creole with greater variability displayed through the standard error bars. Figure 4-11 shows the contrast between coarticulation

101

of the vowel segment and the abrupt segmentation of the contiguous stop consonant in Islander word-final rhotics.

Abrupt change in the formant frequencies 0.034 ms Coarticulation of formant line

Figure 4-11. Spectrogram of word-final rhotic in Raizal Creole with different contiguous elements. Right: /r/ followed by a stop consonant and the abrupt end of the formant frequency lines. Left: coarticulation of /r/ with the contiguous vocalic element. Note the continuation of the formant lines until F3 and F2 begin to separate some 0.034 milliseconds later.

Although, final-word contexts are the second longest in both varieties of Spanish,

Islander Creole presents the greatest differences in segmental duration in this position. A closer inspection of the data revealed that over 28% and 21% of word-final rhotics in Creole were followed by a consonant and a pause, respectively, with an average duration of 0.091 ms.

Conversely, 50% of word-final rhotics were followed by vowel with an average duration of

0.081 ms, 0.010 milliseconds shorter in average compared to the previous contexts. This difference in mean duration might be to coarticulation with the preceding vocalic element, where the /r/ segment boundaries become indistinguishable through a smooth sequence. It is worth noting that all consonantal elements following the word-final approximant consisted of stop sounds (i.e., /d m p/ and one fricative /ʒ/), segments that have a distinctive abrupt transition in the formant frequency visible in the spectrograph, due to the occlusive phase.

These differences in the nature of contiguous elements might explain the greater variability in segmental duration in final-word position, due to being the only position permitting either a contiguous pause or a stop consonant and, and thus, visible formant transitions and

102

longer and distinguishable rhotic sounds. Likewise, it is worth noting that this particular word context displays highly frequent cases of non-rhoticity, and thus, one of the contexts most permeable to rhotic erosion. It is possible that Creole speakers have two distinct allophonic realizations in this position, one that requires additional articulatory effort and, thus, more duration, and a contrastive production that can either merge or coarticulate with the preceding vowel transforming itself into a shwa-like vowel, a phenomenon reported in other English dialects (Zhou et al, 2008). In terms of segmental duration differences according to word stress,

Figure 4-12 shows rhotic duration in three stress categories: tonic, pre-tonic and post-tonic.

Figure 4-12. Differences in mean rhotic duration based on word stress across the varieties of the Archipelago.

Language differences appear between the Spanish varieties and Creole, where Islander contrast by its higher duration in post-tonic rhotics and the Spanish varieties appear in the opposite spectrum. In contrast, pre-tonic rhotics have the shortest duration in Raizal Creole, but one of the highest in both Raizal and Continental Spanish. As expected, these two varieties of

Spanish seem to have the same rhotic duration patterns in terms of word stress. Overall, pre-tonic and tonic rhotics have the same duration pattern in Spanish, and a preference for shortest post-

103

tonic durations. Despite the longer rhotic duration across stress categories, Raizal Spanish shares the similar duration measurements as the monolingual variety and appears to approach the

Continental Spanish in all stress contexts. The only exception appears in tonic position, due to a slightly longer duration compared to Continental Spanish, a pattern also found in other monolingual Spanish varieties (Colantoni, 2006 for Corrientes Spanish, Argentina). Results from a two-way ANOVA show that segmental duration significantly differs according to population

(p=< 2e-16) and stress (p=5.07e-07). A significant interaction between population and position of the rhotic based on word stress was also found (3.51e-07), which suggests that the varieties of the

Archipelago differ on rhotic duration according to word stress. A Tukey post hoc test also indicates that these linguistic varieties contrast between posttonic (p=0.0001326) versus tonic and pretonic rhotics (p=0.9452959). In other words, both Raizal Creole and the Spanish varieties contrast their rhotics based on word stress, but with different magnitudes. While Islander have longer durations in post-tonic rhotics and shorter durations in pre-tonic and tonic positions,

Spanish prefers shorter durations in post-tonic rhotics and longer segments in pre-tonic and tonic positions.

In sum, we can see that Raizal Creole and Continental Spanish differ on the overall segmental duration of their approximant rhotics. Longer rhotic segments are found in Raizal

Creole while Spanish has shorter versions. These differences are also evident in terms of word context and word stress. The varieties of the Archipelago present contrastive rhotic durations in word-initial and complex onset contexts, and in post-tonic positions. I now turn to a comparison of trill/tap durations between Raizal Spanish and Continental Spanish. This is done with the aim of determining a potential neutralization of the tap/trill contrast that might be affected by word context and stress in the two Spanish varieties of the Archipelago.

104

4.1.4.2 A comparison of segmental duration in non-vibrant taps and trills between Raizal and Continental Spanish

In this subsection, Continental Spanish and Raizal Spanish are compared in terms of the duration of the approximant rhotics that present no vibration in their respective contrastive context: intervocalic, word-medial position (i.e., carro ‘car’ vs caro ‘expensive’). In this study, no instance of a syllable-initial, word-internal tap that contrasted with a trill in this position was found1. The tap/trill contrast has been reported to be neutralized in some bilingual dialects

(Balam, 2013), but maintained in terms of segmental duration in other bilingual and monolingual varieties (Henriksen, 2015; Bradley and Willis, 2012). Neutralization of the tap/trill contrast in approximant rhotics might be a signal of an ongoing linguistic change in progress resulting from diverging variation patterns of segmental duration in generations of bilingual Raizales. I test this assumption in the rhotics of the Archipelago exploring the effect of segmental variation and linguistic variables, such as word stress across generations of Raizales.

A comparison of the tap/trill contrast between the bilingual and monolingual Spanish varieties is presented in Figure 4-13, where the mean duration of tap and trills are contrasted in both Spanish varieties. Continental Spanish defines the boundaries between taps and trills with a visibly duration contrast. Raizal Spanish uses the same duration cues to contrast between taps and trills, but this effect seems to be weaker compared to the monolingual variety. Although these intervocalic rhotics present duration differences in both Spanish varieties, Raizal Spanish shows closer duration values between taps and trills compared to Continental Spanish.

1 In some instances, a word-initial trill may contrast with a syllable-initial tap in word internal position (a Roma ‘to Rome’ vs aroma ‘aroma’) (Hualde, 2005, p. 183, in Henriksen, 2015).

105

Figure 4-13. Tap/trill contrast in Continental Spanish and Raizal Spanish.

In other words, Continental Spanish makes use of segmental duration to successfully contrast between approximant taps and trills in intervocalic position, while this contrast seems less clear in Raizal Spanish. Based on this interpretation, a comparison of the segmental duration of non-vibrant rhotics is warranted in order to determine whether there is neutralization of the tap/trill contrast in the bilingual Spanish variety.

A subsequent visualization of segmental duration across generations of Raizal bilinguals gives further detail about the differences in rhotic variation in this population.

Figure 4-14. Mean duration of approximant taps and trills in intervocalic position in Raizal Spanish.

106

This is seen in Figure 4-14, where the values for segmental duration of Spanish taps and trills are visualized. It can be observed that mean duration values in bilingual Raizal Spanish seem to fluctuate across age groups. The closer distance between taps and trills appears in first generation (0.041 ms for taps and 0.057 for trills), while third generation presents a greater mean distance between both segments (0.032 ms for taps and 0.060 ms for trills). Second generation displays intermediate mean values between both first and third generations (0.037 ms for taps and 0.058 ms for trills). However, it can also be observed that segmental duration in bilingual taps and trills present a great deal of variability, as there are certain taps that appear with duration values that approximate trills in first and second generation rhotics, which seems to hint to a possible neutralization of taps and trills in specific contexts.

On the contrary, there seem to be clearer differences and less variation in third generation

Raizales. In order to test the hypothesis that the contrast of taps and trills is permeable to neutralization across generation of bilinguals, it is of special interest to check whether stress has an effect on the trill/tap contrast in intervocalic approximant rhotics. Due to the articulatory emphasis employed in stressed syllables, rhotics in these contexts might be produced with a higher production effort, and thus, longer segmental duration. Table 4-5 summarizes the duration values of rhotics in intervocalic position according to syllabic stress. While post-tonic rhotics have the highest count, they have the lowest mean duration (0.03905). Conversely, tonic rhotics present the longest duration (0.05039), and pre-tonic rhotics appear with a mean duration of

0.041 milliseconds.

Table 4-5. Intervocalic rhotics in Raizal Spanish based on rhotic stress. Population Stress Mean SD N SE Raizal Spanish Pre-tonic 0.04174 0.01775 53 0.00244 Raizal Spanish Tonic 0.05039 0.02316 142 0.00194 Raizal Spanish Post-tonic 0.03905 0.01673 424 0.00081

107

These mean duration differences and the fluctuation in the standard deviation throughout all stress contexts suggest that segmental duration in taps and trills might be conditioned by syllabic stress. For the purpose of this linguistic enquiry, I will test whether the variable neutralization of the tap/trill contrast in Raizal Spanish is present according to word stress.

Table 4-6. Intervocalic taps and trills in Raizal Spanish based on rhotic stress. Population Trill/Tap Stress Mean SD N SE Raizal Spanish Tap Pre-tonic 0.03717 0.01652 38 0.00268 Raizal Spanish Tap Tonic 0.04381 0.02007 102 0.00199 Raizal Spanish Tap Post-tonic 0.03667 0.01477 365 0.00077 Raizal Spanish Trill Pre-tonic 0.05333 0.01574 15 0.00406 Raizal Spanish Trill Tonic 0.06717 0.0222 40 0.00351 Raizal Spanish Trill Post-tonic 0.05374 0.02042 59 0.00266

Overall, mean duration times appear shorter in taps than in trills as seen on Table 4-6, which at first sight, shows that Raizal bilinguals contrast between these two phonemic rhotic variants. However, standard deviation measurements show that duration in taps through all word stress contexts vary to a degree that approximates trills.

Similarly, Figure 4-15 graphically displays these duration differences in terms of word stress.

Figure 4-15. Intervocalic rhotics in Raizal Spanish according to word stress.

108

Both taps and trills show the same pattern of stress variation, where pre-tonic and post-

tonic rhotics appear with similar duration. In contrast, tonic contexts display the highest duration

values in both taps and trills. It seems that, if neutralization of the tap/trill contrast exist, tonic

taps and pre-tonic/post-tonic trills might be approaching each other and present overlapping

duration values. A three-way ANOVA was conducted to examine the significant differences

between the type of rhotic and stress in Raizal Spanish.

Table 4-7. ANOVA output of duration and the interaction between the type of rhotic and stress. Df Sum Sq Mean Sq F value Pr(>F) Type of Rhotic 2 0.00768 0.00384 13.694 1.52e-06 *** Stress 1 0.03648 0.03648 130.071 < 2e-16 *** Type_rhotic:Stress 2 0.00089 0.00044 1.579 0.207 Residuals 611 0.17138 0.00028

Significant differences were found between segmental duration in taps and trills and

across different word stress contexts. In contrast, the interaction effect between type of rhotic and

word stress resulted in no significant correlation, which suggest that the joint effect of these

variables is not significant for the model. Since I am interested in determining whether there is a

context in which neutralization of the tap/trill contrast may occur, an additional Tukey HSD

Posthoc test was carried out, revealing that pre-tonic trills approach the durations of the

approximant taps in tonic positions.

Table 4-8. Tukey Posthoc of the interaction effect between the type of rhotic and word stress across generations. diff lwr upr p adj Trill:posttonic-Tap:posttonic 0.0162919529 0.009573823 0.023010083 0.0000000 Tap:pretonic-Tap:posttonic 0.0012458519 -0.006915314 0.009407018 0.9979940 Trill:pretonic-Tap:posttonic 0.0170716896 0.004458130 0.029685249 0.0016797 Tap:tonic-Tap:posttonic 0.0068466666 0.001484382 0.012208951 0.0038482 Trill:tonic-Tap:posttonic 0.0297206985 0.021746464 0.037694933 0.0000000 Tap:pretonic-Trill:posttonic -0.0150461010 -0.025004877 -0.005087325 0.0002633 Trill:pretonic-Trill:posttonic 0.0007797367 -0.013064918 0.014624392 0.9999850 Tap:tonic-Trill:posttonic -0.0094452863 -0.017276422 -0.001614150 0.0079106 Trill:tonic-Trill:posttonic 0.0134287456 0.003622573 0.023234918 0.0014044

109

Table 4-8. Continued. diff lwr upr p adj Trill:pretonic-Tap:pretonic 0.0158258377 0.001226323 0.030425353 0.0247390 Tap:tonic-Tap:pretonic 0.0056008147 -0.003498529 0.014700158 0.4928300 Trill:tonic-Tap:pretonic 0.0284748466 0.017629001 0.039320692 0.0000000 Tap:tonic-Trill:pretonic -0.0102250230 -0.023464936 0.003014890 0.2355584 Trill:tonic-Trill:pretonic 0.0126490089 -0.001846839 0.027144857 0.1272445 Trill:tonic-Tap:tonic 0.0228740319 0.013941964 0.031806100 0.000000

These results indicate that neutralization of the tap/trill might be occurring according to

location of word stress among some generations of bilingual Raizales. The lack of significant p-

values in specific word stress contexts is seen in Table 4-8, where tonic taps fail to contrast with

pre-tonic trills and it seems that only this particular stress context is vulnerable to neutralization.

However, a closer inspection of segmental duration in these contexts shows a great deal of

variability in stressed contexts across generations of bilingual Raizales. At first sight, it seems

that first generation bilinguals tend to realize approximant trills in pre-tonic positions with

similar values than approximant taps in tonic positions.

Figure 4-16. Mean duration of approximant taps and trills in generations of bilingual Raizales and Continental Spanish according to word stress.

110

This pattern is visible in Figure 4-16, where error bars enhanced the variability of the plotted data, and thus, showing the range of variability across type of rhotics. Duration values in taps and trills seem to overlap in pre-tonic and tonic contexts in first generation, while pre-tonic taps and trills appear with closer duration values in second generation. On the contrary, only post-tonic trills seem to have greater variability in third generation with no clear overlap with other stress contexts. Continental Spanish have greater mean durations in trills across all stress contexts while taps remain under the 0.035ms mark. The difference between taps and trills in first generation Raizales appear to be less marked.

Precisely, pre-tonic trills and tonic taps in intervocalic positions (i.e., verano ‘Summer’ vs

Barranquilla ‘Barranquilla’) seem to have overlapping durations, where taps are produced with increasingly higher durations than trills. A detailed inspection of the dataset for first generation show that forty-seven tonic taps in first generation Raizales were realized with a mean duration of 0.0514 milliseconds with values fluctuating 0.019 milliseconds (sd=0.019). In contrast, the sample for pre-tonic trills consisted of 6 tokens with a mean duration of 0.0474 milliseconds and values even closer to mean in the range of 0.00684 ms, which indicates the low variability in segmental duration in this particular context. Despite the low frequency and the fact that only two words with pretonic intervocalic trills appeared in the corpus (Barranquilla and mataratón

‘medium-sized tree whose properties are used for pest control’), three first-generation informants used these lexical items with ranging duration values from 0.039ms to 0.057 ms, which suggest that this specific context is vulnerable to short duration trills in older Raizales.

So far, the graphical visualization of these differences suggest that variable neutralization of the tap/trill contrast occur in older generation of Raizales in specific contexts, while the speech of the younger generations seems to be less permeable to this phenomenon. However, to

111

exactly determine the nature of segmental durations in taps and trills, we need to examine the statistical differences across all Raizal generations. For this purpose, three ANOVA tests were conducted (one per generation) to examine the interaction effect between the type of rhotic (i.e., taps and trills) and stress in segmental duration. Table 4-9 shows the results of the Tukey’s Post

Hoc tests for each age group and the non-significant p-values for each combination of taps and trills in their corresponding stress context.

Table 4-9. Results of the Tukey’s Post Hoc tests for each generation and the non-significant p- values for each combination of taps and trills in their stress context. Df Sum Sq Mean Sq F value Pr(>F) type_rhotic:stress 2 0.00175 0.000875 3.363 0.036113 * Factors (1st Gen) diff lwr upr p adj tap:tonic-trill:pretonic 0.004001241 -0.016077563 0.0240800445 0.992 trill:pretonic-tap:pretonic 0.005744896 -0.018665409 0.0301552003 0.984 trill:pretonic-tap:posttonic 0.008093672 -0.011141435 0.0273287787 0.832 trill:tonic-tap:tonic 0.007210001 -0.006195585 0.0206155869 0.636 tap:tonic-trill:posttonic -0.008745595 -0.020365406 0.0028742165 0.259 trill:tonic-tap:pretonic 0.016956137 -0.002341903 0.0362541777 0.121 Df Sum Sq Mean Sq F value Pr(>F) type_rhotic:stress 2 0.00509 0.002546 7.468 0.000734 *** Factors (2nd Gen) diff lwr upr p adj trill:pretonic-tap:pretonic 0.0146530535 -0.0120395210 0.041345628 0.613 tap:tonic-trill:pretonic -0.0150193242 -0.0403649094 0.010326261 0.530 tap:pretonic-trill:posttonic -0.0103002397 -0.0259746591 0.005374180 0.411 trill:pretonic-tap:posttonic 0.0159938599 -0.0082946471 0.040282367 0.408 tap:tonic-trill:posttonic -0.0106665104 -0.0239176453 0.002584625 0.192 Df Sum Sq Mean Sq F value Pr(>F) type_rhotic:stress 2 0.000212 0.000106 0.601 0.550 Factors (3rd Gen) diff lwr upr p adj tap:tonic-trill:posttonic -0.0173846174 -0.0353815 0.0006120106 0.064

First generation displays six combinations of factors that fail to reach significant differences between taps and trills. There are no significant differences in terms of segmental duration between tonic (p = 0.99), pre-tonic (p = 0.98), and post-tonic taps (p = 0.83) with pretonic trills. Other non-significant duration differences appear between tonic taps and tonic (p =

0.636), and post-tonic trills (p = 0.26), and pre-tonic taps and tonic trills (p = 0.12). Moreover,

112

second generation also seems to neutralize the tap/trill contrast, but to a lesser extent. Again, the main non-significant differences were found between pre-tonic trills and pre-tonic (p = 0.61), tonic (p = 0.53), and post-tonic (p = 0.40) taps. While post-tonic trills and pre-tonic (p = 0.41) and tonic taps (p = 0.19) failed to reach significant differences based on segmental duration, the third generation only presented one context between taps and trills approaching significance (i.e., tonic taps and post-tonic trills at p = 0.064). The hierarchy of non-significant p-values suggests that neutralization of the tap/trill contrast is increasingly occurring in older generations, while younger Raizales can differentiate between the two segments using durational cues to contrast between taps and trills. This pattern is clearly visualized in Figure 4-17, which displays the hierarchy of non-significant factor combinations.

Figure 4-17. Hierarchy of non-significant factor combinations of stress and tap/trill segments.

Non-significant p-values are higher in senior Raizales, while decreasing in second generation, but barely appearing in third generation. In order to confirm that bilingual Raizales are the only population exhibiting this phenomenon, the same Post Hoc test was conducted in

113

Continental Spanish, resulting in significant differences across all factor combinations (p <

0.001).

I interpret this finding as a sign of a change in progress in the Archipelago of San Andres, where older Raizal speakers neutralize the phonemic contrast between tap and trills in specific stressed positions, while younger generations realize the contrast with durations consistent with the monolingual variety, basically across all contexts. In this sense, it seems to be the case that approximant rhotics in intervocalic position are converging toward Spanish in younger generations of Raizales, distancing themselves from older generations. For this reason, a one- way ANOVA was conducted with the aim of testing a correlation between young islander bilinguals with Continental Spanish in terms of segmental duration of the intervocalic rhotic contrast. No significant differences were found between these two groups (p=0.147).

Furthermore, a subsequent test with second and first generation Raizales revealed increasingly significant differences in intervocalic tap/trills with Continental Spanish (p=0.00027, and p=3.37e-07, respectively). These results indicate that rhotic duration in intervocalic contexts differ according to generation of Raizal Spanish speakers. Younger generations of Raizales are converging in the direction of Continental Spanish and older generations are gradually diverging from the Spanish norm, resulting in the extreme case of neutralization of the tap-trill contrast in the oldest generation. These results indicate an ongoing change in progress of the tap/trill contrast in the Archipelago of San Andres.

4.2 Discriminant Function Analysis

Once the examination of the acoustic correlates in the rhotics of the Archipelago has been presented, a discriminant function analysis (DFA) ensued to determine the best predictors that discriminate between linguistic groups. This analysis renders visible the relationship between the acoustic correlates in the rhotics of the Archipelago and the three varieties in contact, with the

114

aim of predicting a group classification based on acoustic correlates (Grinstead et al, 2013).

More importantly, the results of this test allows to determine the most important acoustic predictors for a correlation analysis in section 4.3.

Table 4-10. Tests of equality of group means. Tests of Equality of Group Means Wilks' Lambda F df1 df2 Sig. duration .782 268.232 2 1925 .000 COG .986 13.937 2 1925 .000 Skewness .990 9.816 2 1925 .000 Kurtosis .995 5.281 2 1925 .005 F2 .953 47.254 2 1925 .000 F3 .810 226.216 2 1925 .000 F3F2_distance .887 122.742 2 1925 .000 F4mean .991 8.928 2 1925 .000 F5mean .953 47.458 2 1925 .000 F4F5_distance .937 64.726 2 1925 .000

The first analysis involved the classification of all the linguistic groups of the

Archipelago based on the combination of the quantitative acoustic measurements obtained (i.e., spectral moments, formant frequencies, and segmental duration). First, I am interested in determining the extent to which these acoustic correlates or predictor variables are significantly different in all linguistic groups. Table 4-10 shows that all predictor variables are different at a statistically significant level. These results indicate that these predictor variables seem to be properly discriminating between linguistic groups. Then, a Box’s M test of equal group variance revealed a significant value lower than p=0.001, suggesting an unequal group variance (Table 4-

11). A visual inspection of the data in the form of histograms revealed a non-normal distribution of the acoustic measurements. Extreme outliers were found in the data that either fail to represent the speakers’ measurements or could be a sign of individual differences worth keeping in the statistical analysis.

115

Table 4-11. Test of null hypothesis of equal population covariance matrices with a significant value lesser than 0.001. Box’s M Test Results Box's M 843.791 F Approx. 19.877 df1 42 df2 618976.170 Sig. .000

It is worth mentioning that some of the assumptions of this analysis involves the equal variance of the linguistic groups and the normal distribution of the data. Failing to satisfy these assumptions might become a potential limitation of this analysis. Although the sample size for each group is unequal (Islander = 328, Continental Spanish = 150, Raizal Spanish= 1450), extreme outliers that fall outside the interquartile range were excluded for this specific analysis in order to satisfy the assumption of normality (Poulsen and French, 2008). The occurrence of these datapoints is mainly due to the nature of the fieldwork procedures and the rationale of this sociolinguistics project of collecting naturalistic data on site2. It is believed that this situation produced the collection of extreme outliers that highly diverged from the measurement centers.

As a result, the data cleaning procedure involved the removal of outliers that fall outside the 1.5 interquartile range (1.5*IQR). In other words, this procedure keeps the values that remain in the inner quartiles spread within 25% below and above the mean, excluding outliers in the low and high end. As a result, the violation of the normality assumption is not fatal, and the ensuing results of this test are still reliable as non-normality is caused by data skewness and not outliers

(Tabachnick and Fidell 1996). The relationship between the predictor variables and the

2 Although, every effort was made to collect the acoustic samples in a relatively noise-reduced environment, this was not always possible due to the informal setting of the research site, where data were collected sporadically in the informants’ dwelling located next to busy streets

116

predictive model is reported with a correlation coefficient of R2=0.31 for the first function and a

R2= 0.10 for the second function, but with a significant value of p=0.000 at the Wilk’s Lambda test for both functions, indicating that this group of predictor variables make predictions that are statistically accurate. After establishing that the predictive model satisfactorily meets the validity criteria, the analysis additionally identified the individual acoustic measurements with the highest predicting capabilities for group membership. This structure matrix is presented in Table

4-12.

Table 4-12. Structure matrix of individual predictors with weighted predicting capabilities. Structure Matrix Function 1 2 duration .778 .338 F3 -.687 .498 F3-F2_distance -.461 .373 F4-F5_distance .375 .225 F2 -.361 .209 F5 mean .334 .048 F4 mean -.143 .052 COG -.042 .350* Skewness .053 -.281* Kurtosis .050 -.223*

A minus sign indicates a discriminant function coefficient with a negative correlation.

The highest coefficient loadings determine those predictor variables with the most abilities to predict group membership. As can be seen in Table 4-12, duration and formant frequencies

(mainly F3 and the distance between F3-F2) present the highest predicting values. This resonates with our analysis of formant frequencies and duration in previous sections (Sections 4.1.3 and

4.1.4 respectively). Both Creole and Spanish show highly distinctive segmental durations, as well as formant frequencies of approximant rhotics. Likewise, it has been reported that alveolar and post alveolar approximants present different F3 frequencies (Zhou et al, 2008; Kent and

117

Read, 2002; Hagiwara, 1995). As such, these results are consistent with previous findings and the differences are expected when comparing an English-based Creole and Spanish. In contrast, spectral moments seem to have the lowest predicting effect in this analysis.

The function of these measurements corresponds to analyzing the manner of articulation of the non-vibrant rhotics in each population group. The lowest predicting capabilities of spectral moments is due to fact that all three linguistic varieties share the same approximant manner of rhotic realization. Since no significant differences were found in the manner of articulation of these rhotics either in Raizal Creole, Continental Spanish, and Raizal Spanish, spectral moments are the predictor variables with the lowest predicting capabilities to assign group membership from all the acoustic measurements. Although some acoustic correlates present better classification capabilities than others (i.e., duration and formant frequencies vs. spectral moments) for all linguistic varieties, altogether they have significant predicting capabilities to differentiate between linguistic groups.

Figure 4-18. Graphic representation of population differences in terms of the acoustic correlates submitted for the analysis. Population group 1 refers to Raizal Creole, group 2 identifies Raizal Spanish, and group 3 is assigned to Continental Spanish.

118

Visually, this can be seen in Figure 4-18 where the predictor variables submitted for the analysis accurately differentiate between all linguistic groups. Raizal Creole and Continental

Spanish appear to be in opposite positions, while Raizal Spanish remains in between both

Islander and Spanish, suggesting that the bilingual variety gravitates in an intermediate position.

The last table resulting from the analysis involves the final classification results on Table 4-13.

Table 4-13. Final classification results. a. 62.3% of original grouped cases correctly classified. b. Cross validation is done only for those cases in the analysis. In cross validation, each case is classified by the functions derived from all cases other than that case. c. 62.0% of cross-validated grouped cases correctly classified. Classification Resultsa,c Population Predicted Group Membership Total Creole Raizal Sp. Cont. Sp. Original Count Creole 237 82 9 328 Raizal Sp. 258 850 342 1450 Cont. Sp. 7 29 114 150 % Creole 72.3 25.0 2.7 100.0 Raizal Sp. 17.8 58.6 23.6 100.0 Cont. Sp. 4.7 19.3 76.0 100.0 Cross-validatedb Count Creole 237 82 9 328 Raizal Sp. 260 846 344 1450 Cont. Sp. 7 30 113 150 % Creole 72.3 25.0 2.7 100.0 Raizal Sp. 17.9 58.3 23.7 100.0 Cont. Sp. 4.7 20.0 75.3 100.0

This Table shows the degree of accuracy of our model to predict group membership. In other words, these results show how well the predictor variables are suited to discern between population groups. In the cross-validated section of Table 4-13, 72.3%, 58.3%, and 75.3% of the acoustic measurements in Raizal Creole, Raizal Spanish, and Continental Spanish, respectively, correctly predict their corresponding population group. The high prediction rate in Raizal Creole and Continental Spanish compared to Raizal Spanish indicates that the acoustic correlates included in the analysis present more variability in the bilingual variety, and thus, the model struggles to accurately predict Raizal Spanish membership. Despite this, this analysis revealed

119

the strongest predictors for discriminating group membership in all linguistic populations. As such, the use of all these predictor variables, particularly, segmental duration and formant frequencies, produce an accurate and statistically significant model.

Interim summary of findings: Thus far, the overall analysis of the acoustic properties of the non-vibrant rhotics of the Archipelago of San Andres has revealed significant differences in the three classified linguistic groups. First, I examined the overall differences in spectral moments. Values for COG, Kurtosis, and Skewness appear similar in the rhotics of the

Archipelago. A comparison with English sibilant and Spanish sibilant rhotics (Colantoni, 2006;

Jongman et al, 2000) showed decreasing spectral moment values for the rhotics realized in

Raizal Creole, Raizal Spanish, and Continental Spanish, indicating a manner of articulation consistent with an approximant production that lacks the fricative properties of a sibilant rhotic.

Approximant rhotics showed significant differences in terms of formant frequencies.

Higher F3 and F2 values for Continental Spanish in the range of 2600Hz and 1500Hz, respectively, contrast with lower values in Raizal Creole (i.e., 1993HZ for F3 and 1334Hz for

F2) and a shorter distance between F3-F2 in the vicinity of 650Hz. Such measurements suggest a rhotic realization consistent with a postalveolar production in Raizal Creole and an alveolar approximant in Continental Spanish. Articulatorily, these two variants differ in both languages:

Raizal Creole produces a successful variant of their phonemic rhotic, a postalveolar approximant with a lingual retraction resulting in lower F3 due to shorter vocal length tract and consistent with what is expected of an English-based Creole; while Continental Spanish produces an allophonic variant of the tap/trill alveolar rhotic that lacks the aerodynamic conditions to produce at least one lingual closure. While for Raizal Creole this rhotic is the natural production of a postalveolar approximant, for Continental Spanish an alveolar approximant rhotic is an

120

incomplete realization of a vibrant /r/. On the contrary, the bilingual Raizal Spanish variety lies in between both languages in contact and between the formant values of Raizal Creole and

Continental Spanish. As seen in the scatterplot of Figure 4-8, F3 in Raizal Spanish presents a more pronounced regression line that approximates either Continental Spanish or Raizal Creole at both ends. This suggests that certain bilingual informants are approaching the F3 frequencies of the variants produced in either Spanish or Islander. It is yet to establish whether a cross generational pattern in this formant frequency is present in San Andres and Old Providence that might signal a change in progress in the Archipelago of San Andres.

Certainly, the patterns of segmental duration present robust differences between languages. Duration values for Raizal Creole appear significantly higher in all word contexts and rhotic stress, while Continental Spanish and Raizal Spanish are more similar in their segmental duration. However, a detailed comparison of segmental duration between Raizal Spanish and

Continental Spanish showed that the tap/trill contrast is clearly defined in monolingual informants, but the segmental boundaries appear weaker in the bilingual Spanish variety. A subsequent analysis on Raizal Spanish revealed that the tap/trill contrast is neutralized in older generations, particularly in specific word stress contexts. Furthermore, such finding correlates with the strongest predictor of group membership in an ensuing discriminant function analysis.

Along with formant frequencies, segmental duration appears to have the best predicting capabilities to classify between linguistic groups. In contrast, spectral moments present the lowest predicting weight in this analysis, due to the lack of differences in the manner of articulation across all varieties. As a result, segmental duration and formant frequencies allow for the successful classification of the three varieties under study and serve as the main acoustic correlates that help differentiate between population groups. It is with these acoustic predictors in

121

mind that the ensuing analysis will consist of determining a correlation between these rhotic approximants and the linguistic varieties produced in the Archipelago. Precisely, one of the research questions that have guided this study consists of examining the acoustic correlates in the three bilingual generations of Raizal Spanish and test whether there is an age group that converges towards Raizal Creole or Continental Spanish rhotics. In the next section, I conduct this analysis with the aim of further unveiling the outcomes of contact in the Archipelago.

4.3 Correlation Analysis Between Generations Of Raizal Spanish And Continental Spanish/Raizal Creole.

So far, the analysis of the acoustic properties of these segments has shown that rhotics are realized differently in these varieties. As hypothesized, the greatest difference was found between Continental Spanish and Raizal Creole. This corroborates our assumption that Raizal

Creole and Spanish have different allophonic variants for rhotics. Namely, this Creole produces a postalveolar /r/ similar to the same English retroflex liquid. On the other hand, Continental

Spanish produced their rhotics with different acoustic cues: on average COG values are higher, and Skewness and kurtosis measurements were lower than bilingual and Creole rhotics. In addition, formant and duration measurements in Continental Spanish contrast sharply with Raizal

Creole, suggesting a more fronted realization. As a result, non-vibrant rhotics in Continental

Spanish were classified as alveolar approximants. As observed previously, bilingual Raizal rhotics lie in the intersection of Continental Spanish and Raizal Creole measurements. As expected in a language contact situation, the variability observed in the acoustic properties of bilingual rhotics presuppose that Raizal Spanish might be heterogeneously converging toward one of the varieties in contact. As stated in the hypotheses of this work, generational groups of

Raizal speakers might be producing rhotics with similar acoustic correlates to Raizal Creole in older generations and to Continental Spanish in younger generations. In the following analysis, I

122

test the correlation between generations of Raizal speakers in the two islands of the Archipelago with either Raizal Creole or Continental Spanish. The acoustic measurements that are used in this analysis correspond to the significant correlates found in the discriminant function analysis of section 4.2: duration, F3, and F3-F2 distance. Since the measurement of the F5 and F4 frequencies correspond to the second and third resonances of the back cavity, respectively, the distance between these two formant frequencies are relevant for determining the degree of narrowing in the laryngeal region and tongue shapes for English /r/ (Zhou et al, 2008). As a such, the shifts in the spacing of these two formant frequencies are used for determining a bunch-tip variant from a retroflex realization in approximant English-like rhotics, both of which are produced in the backed postalveolar region. As a result, the correlation models of the F5-F4 distance will be excluded from this analysis, as it has been determined a fronted alveolar realization in the non-vibrant monolingual Spanish rhotics of the Archipelago.

In total, 1,928 approximant tokens were collected for this analysis: 328 from five Raizal

Creole speakers, 1,450 from thirty Raizal Spanish informants, and 150 tokens from five

Continental Spanish participants. A first glance at the distribution of non-vibrant rhotics between

Raizal generations and the two contact languages (i.e., Raizal Creole and Continental Spanish) show a decreasing trend of approximant production in third generation Raizales that approximates the rates in monolingual Spanish. This is summarized in Figure 4-19.

123

Figure 4-19. Distribution of non-vibrant rhotics in the varieties of the Archipelago. R1, R2, and R3 correspond to the first, second, and third generations of bilingual Spanish Raizales, respectively.

On the contrary, senior Raizal informants present higher rates of production between generations, while Raizal Creole also shows the same trend, as approximant rhotics appear over

60% in this dataset compared to non-rhoticity. As a result, there seems to be an association between certain Raizal generations and either Raizal Creole or Continental Spanish in terms of the proportions of occurrence of non-vibrant rhotics. Rates of use in second generation appear in an intermediate position between first and third generation. As we know, Spanish combines rhotics into two phonemic categories (i.e., taps and trills), which, in normative use, are realized with the periodic vibration of the apical region of the tongue. In this case, this articulatory gesture is unsuccessful in the two bilingual and monolingual Spanish varieties, but the tap/trill distinction is still relevant in Spanish and is mostly realized through segmental duration as seen in section 4.1.4. Figure 4-20 summarizes the occurrences of taps and trills in the bilingual Raizal population, the biggest dataset in the analysis. In general, taps appear more frequently in both islands compared to trills, mainly due to appearing in all word contexts, except in initial-word positions.

124

Figure 4-20. Distribution of non-vibrant taps and trills in Raizal Spanish in both islands of the Archipelago: Old Providence and San Andres.

When we compare the segmental duration of rhotics across generation of Raizales with the two contact varieties, a clear distinction between Spanish and Creole is observed, as approximant post-alveolar Creole rhotics are realized with longer durational cues.

Figure 4-21. A comparison of rhotic duration between varieties.

125

These differences seem to be language specific, since both bilingual monolingual varieties of Spanish appear with similar duration values, as seen in Figure 4-21. This resonates with the findings in Figure 4-9, where Raizal Spanish and Continental Spanish produce shorter rhotic segments compared to Raizal Creole.

Figure 4-22. A comparison of tap and trill duration between Spanish varieties and Raizal Creole.

The differences between taps and trills and Creole approximants are further detailed in

Figure 4-22. Taps are clearly realized with shorter duration values, while trills appear longer than taps and closer to Creole rhotics. Precisely, the fact that trills are produced with longer durations than taps signal a longer trill realization. Despite this, approximant rhotics in Raizal Creole still appear with greater variation and with higher mean durations in most generations in both taps and trills. In what appears the approaching of trill duration with a Creole standard, first and third generation seem to have duration means similar to Islander.

As a result, a simple, linear, mixed effect test was conducted to determine whether significant differences exist in mean durations in Raizal Creole approximants and trills across generation of Raizal Spanish speakers and Continental Spanish. Results in Table 4-14 show that

126

differences in trill duration in all Spanish varieties are statistically significant when compared to approximant Creole rhotics, except for third generation Raizales (p = 0.090501).

Table 4-14. Linear, mixed effects model of segmental duration between Raizal Creole approximants and Spanish trills in generations of Raizal Spanish informants and Continental Spanish. Random Variables Dependent variable: Informant and Token Duration Raizal – First Generation Estimate: -0.016 (p = 0.027186) * Raizal – Second Generation Estimate: -0.030 (p = 0.000273) ** Raizal – Third Generation Estimate: -0.013 (p = 0.090501) Continental Spanish Estimate: -0.020 (p = 0.040258) * Intercept Estimate: 0.085 Observations 702 Log Likelihood 1,437.33 Akaike Inf. Crit. -2,858.66 Bayesian Inf. Crit. -2,822.23 Note: *p = 0.05 **p = 0.01 ***p<0.01

For all these tests, the intercept consists of Raizal Creole measurements. Although approaching significance, this generation produces mean trill durations that increasingly resemble those values presented in Raizal Creole. Since no other generation displays this effect and the estimate in both Continental Spanish (i.e., 0.065 ms) and Raizal Creole (i.e., 0.085 ms) appear similar, it seems that the Spanish varieties are using segmental cues in the form of higher trill mean durations to produce the phonemic contrast with taps, which are visible shorter than trills, rather than correlating to the segmental durations in RaizalCreole.

This result resonates with our findings in section 4.1.4.2, where it was found that younger

Raizal and Continental Spanish speakers use durational cues to contrast between non-vibrant taps

127

and trills. Accordingly, the results on segmental duration between the population groups under study are interpreted as occurring independently in the Spanish language. The second and third acoustic predictors involve the lowering of F3 and the spacing of the F3-F2 frequencies. As noted previously in section 4.1.3, the formant frequencies in the Archipelago differ depending on the rhotics of each variety under study. Thus, the formant-lowering effect of F3 and the narrowing space of F3-F2 of rhotics will be tested across generations of bilingual Raizal Spanish to determine a correlation with Raizal Creole or Continental Spanish.

Figure 4-23. A comparison of F3 and F3-F2 frequencies between Spanish varieties and Raizal Creole.

Accordingly, when we visualize the mean formant frequencies in these population groups in Figure 4-23, both formant measurements follow the same increasing trend between generations toward Continental Spanish, suggesting differences between older and younger

Raizales. While third generation approaches the values encountered in Continental Spanish, the older first generation resembles more Raizal Creole formants. In order to visualize further these differences and similarities, rhotics in these population groups were divided in terms of F3 and the F3-F2 spacing realized independently between taps and trills.

128

Figure 4-24. A comparison of F3 and F3-F2 frequencies in taps and trills between Spanish varieties and Raizal Creole (left for taps and right for trills).

Figure 4-24 shows the relationship between F3 and the formant measurements found in the distance between F3 and F2 in all groups. Positive correlation lines seen in this Figure indicate that there is a linear relationship in F3 frequencies and F3-F2 distance in trills (Pearson r(700) = .82, p = < 2.2e-16) and taps (Pearson r(1552) = .75, p = < 2.2e-16) between all population groups. Likewise, it can be observed that Continental Spanish and Raizal Creole present regression lines that lie on two opposite sides of the plot, while generations of Raizal

Creole gravitate between both languages. In addition, it is worth noting that, in general terms, third generation seem to approach Continental Spanish, while first and second generation present lower F3 values, and thus, smaller F3-F2 spacing. Whether these observations corroborate statistical differences between a Raizal generation with Raizal Creole is tested with two linear mixed effects models for taps and trills. Our null hypothesis states that there are no significant differences between older generations of Raizal Spanish and Raizal Creole in terms of F3 and

F3-F2 distance frequencies of non-vibrant trills and taps. Precisely, failing to reject the null hypothesis in any of the following analysis will indicate that no statistically significant differences exist between Creole approximants and non-vibrant trills/taps in senior Raizales, and thus, this will suggest generational transfer of phonological features in Raizal bilinguals.

129

Table 4-15. Linear, mixed effects models of F3 between Raizal Creole approximants and Spanish taps (left) and trills (right) in generations of Raizal Spanish informants and Continental Spanish. Random Variables Dependent variable: Random Variables Dependent variable: Token and Informant F3 in Taps Token and Informant F3 in Trills Raizal – First Estimate: 213.27 Raizal – First Estimate: 201.39 Generation (p = 0.0102) ** Generation (p = 0.06285) Raizal – Second Estimate: 278.31 Raizal – Second Estimate: 382.60 Generation (p = 0.0014) ** Generation (p = 0.00136) ** Raizal – Third Estimate: 368.63 Raizal – Third Estimate: 500.91 Generation (p = 4.91e-05) *** Generation (p = 4.34e-05) *** Estimate: 595.41 Estimate: 688.50 Continental Spanish *** Continental Spanish (p = 4.91e-05) (p = 4.41e-06) *** Intercept Estimate: 2,010.19 Intercept 2,009.12 Observations 1,554 Observations 702 Log Likelihood -10,828.06 Log Likelihood -4,834.83 Akaike Inf. Crit. 21,672.13 Akaike Inf. Crit. 9,685.66 Bayesian Inf. Crit. 21,714.91 Bayesian Inf. Crit. 9,722.09 Note: *p = 0.05 **p = 0.01 ***p<0.01 Note: *p = 0.05 **p = 0.01 ***p<0.01

Results for taps in the left side of Table 4-15 shows that all F3 values reach statistical significance at p = 0.05, and thus, taps in Continental Spanish and Raizal Spanish significantly differ from Raizal Creole approximants. However, the estimates of younger Raizales seem to increasingly converge towards Continental Spanish, while taps in senior informants present lower F3 values that approach those of Raizal Creole. Similarly, results displayed for trills in

Table 4-15 show the same pattern, where formant values in younger generations are diverging toward Continental Spanish. On the contrary, trills in first generation present values that fall barely outside the significance threshold, and thus, failing to reject our null hypothesis for trills as this is the only age group that presents no statistical difference in terms of F3 frequencies with

Raizal Creole. Such finding could suggest a change in progress in non-vibrant trills across generations of Raizal Spanish, where the lowering effect of F3 indicates a point of articulation in first generation speakers that differs from the trills produced in third generation Raizales.

130

Although the differences in F3 approaches a significance level (p = 0.06285) and the estimate with Raizal Creole is still distant (positive estimate at 201.39 Hz), it appears that trills in this generation are realized with a different articulatory configuration that resemble less the tongue gestures of younger generations. While a lowered F3 typically characterizes an English/Creole- like rhotic, it has shown here and elsewhere that the decreasing values of this resonance often approaches F2 (Zhou et al, 2008; Lehiste, 1964; Dalston, 1975; Espy- Wilson, 1987).

Table 4-16. Linear, mixed effects models of the distance between F3-F2 between Raizal Creole approximants and Spanish taps (left) and trills (right) in generations of Raizal Spanish informants and Continental Spanish. Dependent variable: Dependent variable: Random Variables Random Variables F3-F2 Distance in F3-F2 Distance in Token and Informant Token and Informant Taps Trills Raizal – First Estimate: 72.22 Raizal – First Estimate: 72.69 Generation (p = 0.362130) Generation (p = 0.499553) Raizal – Second Estimate: 132.56 Raizal – Second Estimate: 255.85 Generation (p = 0.107456) Generation (p = 0.028260) * Raizal – Third Estimate: 185.32 Raizal – Third Estimate: 379.97 Generation (p = 0.026419) * Generation (p = 0.001614) *** Estimate: 399.01 Estimate: 557.55 Continental Spanish Continental Spanish (p = 0.000135) *** (p = 0.000164) *** Intercept Estimate: 695.14 Intercept Estimate: 696.91 Observations 1,554 Observations 702 Log Likelihood -10,791.52 Log Likelihood -4,852.91 Akaike Inf. Crit. 21,599.04 Akaike Inf. Crit. 9,721.81 Bayesian Inf. Crit. 21,641.83 Bayesian Inf. Crit. 9,758.24 Note: *p = 0.05 **p = 0.01 ***p<0.01 Note: *p = 0.05 **p = 0.01 ***p<0.01

Precisely, the same test for significance was conducted in tap and trills and results are summarized in Table 4-163. Results indicate that the F3-F2 distance in non-vibrant taps and trills

3 The intercept consists of Raizal Creole measurements for all the statistical tests in this analysis.

131

is statistically significant in third generation and Continental Spanish, which suggest that this formant spacing in the youngest Raizales bear less resemblance to Raizal Creole approximants.

The key piece of evidence comes from older generation of bilingual Raizales, in which no significant differences were found with Creole post-alveolar approximants in the F3-F2 spacing, and thus, they seem to be approaching the frequencies of a Creole-like rhotic rather than an alveolar non-vibrant Spanish realization. As such, not only the lack of effect is seen in taps produced by senior (p = 0.362130) and second generation Raizales (p = 0.107456), but less prevalent in trills only in first generation (p = 0.499553). Furthermore, Continental Spanish displays the opposite pattern, as these rhotics are further dissociated with Creole approximants, suggesting that the properties of Raizal Spanish rhotics change across age groups. The fact that first generation Raizales in both taps and trills were not significant predictors in the model due to the high p-value (p = 0.362130 for taps and p = 0.499553 for trills) indicate that the variation patterns are derived from the transfer of phonological features from the Creole language and dependent upon generational differences. These findings confirm the hypotheses posited in this study, as there are generational differences in rhotic production that are converging towards the languages in contact in the Archipelago.

A further point of interest consists of determining whether these patterns depart from a

Creole variety in either San Andres or Old Providence. As the previous results have shown, we have already established that the rates of use of non-vibrant rhotics and the variation in the form of F3 and F3-F2 frequencies across generations of Raizales is not random, but rather these are correlated with one of the two varieties in contact. For the last part of this section, I have divided the analysis according to the measurements for F3 and F3-F2 distance in both Islands of the

Archipelago: San Andres and Old Providence. Figure 4-25 displays the mean F3 values across

132

population groups found in both islands. While Raizal Creole appear with the lowest F3 values and Continental Spanish with the highest frequencies, the different generations of Raizales remain between both contact languages at varying distances.

Figure 4-25. A comparison of F3 frequencies in trills between Spanish varieties and Raizal Creole (left for taps and right for trills) in San Andres and Old Providence.

The heterogeneity observed in F3 frequencies between generations suggest that older generations in both islands are approaching the rhotic frequencies of Raizal Creole on one end, but younger speakers are converging toward Continental Spanish on the other end. The trend is particularly more visible in first and second generation in Old Providence and only in first generation in San Andres. In order to measure the effect of these differences in both islands, two linear mixed effects models were conducted to test whether significant differences exist between population groups and F3 frequencies in Raizal Creole. Results are summarized in Table 4-17.

Indeed, F3 frequencies in first generation Raizales present the lowest estimates and p-values across all groups indicating non-significant differences with Raizal Creole. As the estimates in younger generations increasingly resemble Continental Spanish, senior informants present F3

133

frequencies that fail to reach statistical significance, and thus, indicating converging values in the

direction of Raizal Creole.

Table 4-17. Linear, mixed effects models of F3 frequencies between Raizal Creole approximants and Spanish trills in generations of Raizal Spanish informants and Continental Spanish.

Random Variables Dependent variable: Random Variables Dependent variable: Token and Informant F3 SAI Token and Informant F3 Old Providence Estimate: 169.49 Estimate: 221.40 Raizal – First Generation Raizal – First Generation (p = 0.12180) (p = 0.0782) Estimate: 401.57*** Estimate: 310.49 Raizal – Second Generation Raizal – Second Generation * (p = 0.00171) (p = 0.0249) Estimate: 363.96*** Estimate: 619.27 Raizal – Third Generation Raizal – Third Generation (p = 0.00392) (p =5.94e-05) *** Estimate: 678.94*** Estimate: 702.99 Continental Spanish Continental Spanish (p = 4.65e-06) (p =1.38e-05)*** Intercept Estimate: 2,008.21*** Intercept Estimate: 2,013.94 Observations 539 Observations 853 Log Likelihood -3,672.50 Log Likelihood -5,712.36 Akaike Inf. Crit. 7,360.99 Akaike Inf. Crit. 11,440.72 Bayesian Inf. Crit. 7,395.31 Bayesian Inf. Crit. 11,478.71 Note: *p = 0.05 **p = 0.01 ***p<0.01 Note: *p = 0.05 **p = 0.01 ***p<0.01

Since the p-values in first generation fail to reach statistical significance and maintain a

level that approaches significance, there is weak evidence that F3 frequencies in this population

are different than those of Raizal Creole, particularly in San Andres. In other words, if we were

to submit more independent variables, first generation would have to be excluded, since it is not

a significant predictor for F3 variation, due to its similarity with Raizal Creole. Additional

evidence for a change in progress in non-vibrant rhotics in the Archipelago, comes in the form of

the differences in the narrowing of F3-F2 between all population groups under study. Recall that

we found statistical evidence for a non-significant effect in the F3-F2 spacing in taps and trills in

first generation informants. A graphical comparison between islands also gives a hint about the

convergence of formant frequencies in bilingual taps toward one of the contact languages.

134

Figure 4-26. A comparison of F3-F2 distance in taps between Spanish varieties and Raizal Creole in San Andres and Old Providence.

Figure 4-26 presents the mean F3-F2 spacing in taps between population groups in San

Andres and Old Providence. It seems that the narrowing of space between F3 and F2 is lower in

San Andres and more similar to Raizal Creole than in Old Providence. Precisely, these formant values seem to be lower in first and second generation in Old Providence and converging toward

Raizal Creole, while third generation presents higher frequencies closer to Continental Spanish.

Similarly, two linear regression tests were conducted to determine the relationship between F3-

F2 mean spacing in Raizal Creole approximants and the mean values in taps in generations of bilingual Raizal Spanish and Continental Spanish.

Results are summarized in Table 4-18. Indeed, no significant differences were found in first and second generation Raizales in in both islands. However, not only the same insignificant effect was found in older generations but also in younger Raizales in San Andres.

135

Table 4-18. Linear, mixed effects models of F3-F2 distance frequencies between Raizal Creole approximants and Spanish taps in generations of Raizal Spanish informants and Continental Spanish in Old Providence and San Andres. Random Variables Dependent variable: Random Variables Dependent variable: Token and Informant F3-F2 in SAI Token and Informant F3-F2 in Old Providence Estimate: 63.61 Estimate: 68.50 Raizal – First Generation Raizal – First Generation (p = 0.452677) (p = 0.45854) Estimate: 144.91* Estimate: 104.26 Raizal – Second Generation Raizal – Second Generation (p = 0.109424) (p = 0.27926) Estimate: 81.95 Estimate: 277.93 Raizal – Third Generation Raizal – Third Generation *** (p = 0.353243) (p = 0.00692) Estimate: 402.88*** Estimate: 380.29 Continental Spanish Continental Spanish *** (p = 0.000171) (p = 0.00053) Intercept 693.41*** Intercept 709.81 Observations 1,033 Observations 1,308 Log Likelihood -7,126.87 Log Likelihood -8,970.41 Akaike Inf. Crit. 14,269.73 Akaike Inf. Crit. 17,956.83 Bayesian Inf. Crit. 14,309.25 Bayesian Inf. Crit. 17,998.24 Note: *p = 0.05 **p = 0.01 ***p<0.01 Note: *p = 0.05 **p = 0.01 ***p<0.01

This unexpected result might challenge our assumption that older generations are

producing a tap that is realized with an increasingly different point of articulation that

approximates Creole rhotics. An inspection of the overall F3 formant values in taps in San

Andres indicates that third generation present the highest F3 mean frequency across all

generations (2378 Hz). Likewise, front vowels make up 33% (N=141/421) of the preceding

segments and 31% (N=131/421) of following segments of the tap’s dataset in third generation.

Back vowels only accounts for 15% (N=65/421) and 17% (N=74/421) of preceding and

following segments, respectively. This suggest that F2 in third generation is reflecting the high

proportion of front vowels that surround taps. These results indicate that F2 might be increasing

rather than having the case where F3 is lowering, and thus, the tongue gesture is still realized

reflecting an alveolar point of articulation.

136

Similarly, F3-F2 spacing in trills present a similar pattern than taps, where older generations are closer to Creole approximants. This is seen in Figure 4-27.

Figure 4-27. A comparison of F3-F2 distance in trills between Spanish varieties and Raizal Creole in San Andres and Old Providence.

Precisely, older generations in Old Providence present mean values closer to Raizal

Creole, while younger informants are closer to Continental Spanish. In San Andres the pattern is similar but second and third generation depart from Raizal Creole more visibly than first generation. Recall that a similar pattern was found with taps, as younger generations in San

Andres were also approximating Continental Spanish. In a similar manner, two mixed effects tests were conducted to determine the relation of mean F3-F2 values in Raizal Creole approximants and bilingual and monolingual Spanish approximant trills. Results are summarized in Table 4-19. This final analysis demonstrates that the F3-F2 distance in trills in third generation and Continental Spanish are significantly different to Raizal Creole rhotics.

137

Table 4-19. Linear, mixed effects models of F3-F2 distance frequencies between Raizal Creole approximants and Spanish trills in generations of Raizal Spanish informants and Continental Spanish in Old Providence and San Andres.

Random Variables Dependent variable: Random Variables Dependent variable: Token and Informant F3-F2 in SAI Token and Informants F3-F2 in Old Providence Raizal – First Generation Estimate: 37.75 Raizal – First Generation Estimate: 81.17 (p = 0.64364) (p = 0.553609) Raizal – Second Generation Estimate: 263.42 Raizal – Second Generation Estimate: 197.23 (p = 0.00721) *** (p = 0.188485) Raizal – Third Generation Estimate: 205.09 Raizal – Third Generation Estimate: 525.32 (p = 0.03417) * (p = 0.001219) *** Continental Spanish Estimate: 537.21 Continental Spanish Estimate: 589.80 (p = 7.40e-06) *** (p = 0.000466) *** Intercept Estimate: 695.13*** Intercept Estimate: 701.74 Observations 539 Observations 853 Log Likelihood -3,687.50 Log Likelihood -5,760.56 Akaike Inf. Crit. 7,390.99 Akaike Inf. Crit. 11,537.13 Bayesian Inf. Crit. 7,425.31 Bayesian Inf. Crit. 11,575.12 Note: *p = 0.05 **p = 0.01 ***p<0.01 Note: *p = 0.05 **p = 0.01 ***p<0.01

On the contrary, these formant frequencies in first generation Raizales failed to reach a significant effect with high p-values, suggesting a null effect and non-significant differences between senior trills and Creole approximants. This null effect is displayed in both islands and include second generation to a lesser degree in Old Providence only (p= 0.188). In sum, we have established that the formant frequencies in older generations are increasingly resembling those of Raizal Creole. On the other hand, third generation and

Continental Spanish appear highly uncorrelated with Raizal Creole approximants showing converging formant values.

138

The analysis of non-vibrant rhotics has addressed the research questions regarding the nature of non-vibrant rhotic variation and change in the Archipelago. The next Chapter examines the outcomes of contact in Raizal generations and vibrant rhotics.

139

CHAPTER 5 COMPARATIVE VARIATIONIST ANALYSIS OF RHOTIC VARIATION IN THE ARCHIPELAGO OF SAN ANDRES

Considering the findings of the previous chapter concerning the sociophonetics of non- vibrant rhotics and its implications for language change, this section investigates the patterns of highly structured vibrant rhotic variation across generations of Raizales in Old Providence and

San Andres. Moreover, through a side-by-side comparison of Raizal Spanish with monolingual

Spanish, I examine the cross-linguistic influence of the Spanish varieties coexisting in the islands, and test whether a competing rhotic variant is being constrained by the effects of language contact. To do so, I divide this Chapter in three subsections, as follows: first, I examine the distribution of the vibrant rhotic variants in Raizal Spanish and Continental Spanish based on context of occurrence. In doing so, I present a frequency analysis of rhotic elision in Raizal

Spanish, Continental Spanish, and Raizal Creole. Next, I present a quantitative analysis of the linguistic and social constraints of rhotic variation circumscribed in the Spanish varieties under study. This statistical modelling is conducted on the statistical tool Rbrul, a variable rule program

(i.e., Varbrul) that allows for running statistical tests with fixed and random effects on complex and unbalanced data sets (Johnson, 2009). This analysis allows for the assessment of the relationship between linguistic and social variables on the rhotic variants, while including the random effect of the individual and the tokens where the rhotic is produced. Finally, I compare the factors that contribute the greatest effect to variation in each variety, with the aim of reconstructing the development of the rhotics in the Raizal Spanish variety and Continental

Spanish, and potentially linking a generational change in the Archipelago. The aim of this section is to provide evidence to determine whether sound change is occurring by examining vibrant rhotic production across generations of Raizal Spanish.

140

The datasets that comprise this analysis involves rhotic tokens from the two Spanish varieties in contact: Raizal Spanish and Continental Spanish. We have previously identified that

Raizal Creole produces a post-alveolar approximant. The articulation of a Creole rhotic is such that no apical vibration occurs in the post-alveolar palatal region. As described in Chapter 4, non-vibrant rhotics are also present in the monolingual and bilingual variety of Spanish.

However, as we will see in Table 5-1 and Figure 5-1, the vast majority of the rhotics produced in

Raizal Spanish and Continental Spanish are realized with tongue tip vibrations. As such, Raizal

Creole rhotics present acoustic properties that differ phonemically from Spanish vibrant /r/, including non-vibration, contrasting formant frequencies, segmental duration differences, and to a minor degree, spectral moment values, that bars the production of a Creole-like /r/ with vibration properties in Spanish. At the same time, a systematic vibrant Spanish-like /r/ has not been documented in Raizal Creole in this study, and to my knowledge, it has not been reported in previous studies (Bartens, 2013). For this reason, no rhotic variant can be attributed to a comparable ‘conflict site’ where all the three linguistic varieties differ (Tagliamonte, 2012, p. 13) functionally and structurally. The only exception is r-elision. R-elision involves a situation where the rhotic variant is not uttered or ‘dropped’ from being produced by the speaker. In this study, r- elision or non-rhoticity has been found to coexist with the uttered rhotic variant not only in both monolingual and bilingual Spanish, but also in Raizal Creole, and corresponds to a comparable linguistic phenomenon across the varieties of the Archipelago. However, this analysis focuses on the variable production of vibrant rhotics across Spanish varieties, and thus, non-rhoticity falls outside the scope of this study1. Given these considerations, this section ends with two

1 It is worth mentioning that this study assumes non-rhoticity to be a phenomenon present in this English-based Creole. However, the assumption that Raizal Creole has as an underlying elided form of a fully produced /r/ phoneme has yet to be tested empirically. Despite this, non-rhoticity is a phenomenon that deserves a future thorough study.

141

comparative assessments: An analysis of the Varbrul results on the linguistic and social factors that condition vibrant rhotic variation across generations of Raizal Spanish and Continental

Spanish. A comparative assessment of the Varbrul results of vibrant rhotic variation across generations of Raizal Spanish and Continental Spanish. By following this comparative approach, this chapter will gradually examine the systematic patterns of rhotic variation under the variationist framework across two varieties. Let’s now examine the distribution of the rhotic variants in our datasets.

5.1 Distribution Of Rhotic Variants

Rhotics in the Archipelago of San Andres can be divided into two different categories: rhotics realized as different variants, either as approximants, taps, or trills, and another category in which complete reduction and further non-rhoticity segment takes place (Labov, 1966). While most of the data in this study reflects the production of a specific rhotic variant in each variety under study, it has also been observed cases in which non-rhoticity occurs. As can be seen in

Figure 5-1, these two rhotic categories follow an uneven distributional pattern, in which r-elision or non-rhoticity changes in the data sets from 12.9% in Raizal Spanish and 23% in Continental

Spanish, to 43% in Raizal Creole. This higher proportion of rhotic elision in Raizal Creole suggests a pattern that sharply differs from the Spanish varieties.

Figure 5-1. Overall distribution of rhoticity and non-rhoticity categories in the varieties of the Archipelago.

As such, this phenomenon merits an analysis on its own but falls outside of the scope of this study. The following subsections of this chapter will examine the distribution patterns of

142

fully produced rhotics in Raizal Spanish and Continental Spanish. In doing this, I present frequency tables and figures that shows the distribution of canonical and non-canonical rhotic variants across different factor groups.

5.1.1 Distribution Of Rhotic Variants In Raizal Spanish

Raizal Spanish is the biggest data set in the corpus. Representative samples were methodologically obtained from three generations of Raizal informants in both islands reaching

3,408 tokens of taps and trills in the whole dataset. Table 5-1 illustrates the overall trends of canonical and non-canonical forms in Raizal Spanish in two types of Spanish rhotics, those with one simple vibration (i.e., taps), and those with multiple apical vibrations (i.e., trills). The highest proportion of occurrence in the data correspond to taps, either produced with one apical closure or canonical (42%), or with zero-occlusion or non-canonical (32.5%). Canonical taps comprise the type of rhotic with the highest proportion in the data with 42% of occurrences in the dataset.

Table 5-1. Overall distribution of canonical and non-canonical trills and taps in Raizal Spanish. DV Type N Freq Average Canonical Trills 176 0.109 5.16% Canonical Tap 1429 0.890 41.93% Non-canonical Trills 697 0.386 20.45% Non-canonical Tap 1106 0.613 32.45%

On the contrary, trills appear more frequently in their non-canonical form (0.386) rather than when produced with more than one apical occlusion (0.109). Overall, non-canonical forms appear 53% in the data, and taps appear more than trills (74%). The proportion of trills produced in their non-canonical forms compared to the vast majority of canonical taps correspond to an asymmetrical realization that reflects the variants usage in this bilingual population. In the first place, canonical taps are produced with one apical occlusion, whereas trills require longer segments with the sufficient aerodynamical conditions to be realized with at least a second apical

143

closure. Whether this articulatory restriction is more prevalent across generations and island of dwelling can be also observed in our data.

Figure 5-2. Distribution of canonical (left) and non-canonical (right) trills and taps in Raizal Spanish across generations.

Figure 5-2 displays the generational differences across generations of Raizales. First generation informants appear with the lowest count of canonical rhotic for both trills (N=13) and taps (N=650), while third generation present the highest numbers. Second generation informants lie between both generations. Likewise, an increasing pattern of canonical taps across generations suggests that taps present better conditions for its canonical production in the

Archipelago than its phonemic counterpart. With regards to non-canonical trills and taps, younger generations lead the change with lower counts of non-canonical rhotics. On the contrary, the oldest generation produce higher numbers of non-canonical rhotic realizations, except for trills, where second generation Raizales produce higher counts. This contrastive trend indicates that younger generations produce more canonical realizations in the Archipelago, while older Raizales produce less canonical taps and trills. Generational differences are further visible in Table 5-2.

144

Table 5-2. Relative frequencies of canonical and non-canonical rhotics by generation. Generation DV N Average in DV N Average in Cano:Non Canonical Non-Canonical Ratio First Cano 320 19.94% Non 826 45.8% 1:2.5 Second Cano 525 32.7% Non 586 32.5% 1:1.12 Third Cano 760 47.35% Non 391 21.7% 1:0.5

Senior and younger Raizales display sharp and almost identical contrastive differences between canonical and non-canonical rhotic production. While non-canonical rhotic realizations surpasses canonical rhotics in first generation informants with a ratio of 1:2.5, these values drop considerably in the youngest Raizal age group who use canonical rhotics with a ratio of 1 every

0.5. Less notably are the differences in the intermediate second generation, who remain stable between canonical and non-canonical taps and trills and refrain from choosing one variant over the other. Differences across islands are also are also visible in what seems to be a gradual change across generations. Figure 5-3 shows that Raizales in Old Providence present higher counts of canonical taps compared to San Andres, but lower numbers in canonical trill productions. Differences in non-canonical rhotics appear less pronounced, as both islands tend to produce non-canonical rhotics with similar rates.

Figure 5-3. Distribution of canonical and non-canonical trills and taps in Raizal Spanish between islands.

145

In contrast, canonical forms appear highly preferred by females in both trills and taps.

Figure 5-4 shows the distribution asymmetries between males and females for both canonical and non-canonical taps and trills. From the 1,693 tokens collected from female Raizales, almost

1,000 rhotics are either canonical trills (n=104, 6%) or canonical taps (n=846, 50%). Similarly, a comparable number of tokens from male informants was also collected (n=1,715).

Figure 5-4. Distribution of canonical and non-canonical trills and taps in Raizal Spanish based on sex.

However, contrary to females, male informants appear to realize canonical rhotic forms less frequently than females, while tend to produce more non-canonical variants in both trills

(N=397, 23%) and taps (N=663, 38%). In contrast, females display lower rates of non-canonical production (N=743, 43%) compared to males (N=1060, 62%), which suggest their high preference for more prestigious canonical forms.

These results point to a drastic change in rhotic distribution in both ends of Raizal generations. Based on these results, younger generations are dropping the zero-occlusion rhotics, a characteristic prevalent in Raizal Creole, and adopting the standard trill and tap production of

Spanish. Moreover, the rate of use for canonical forms in the Archipelago seem to be led by women, who seem to prefer more prestigious canonical forms across rhotic types. Finally, while

146

the differences between islands is less robust as with sex and generation, Old Providence appear to be the island from which the change is being originated at least for canonical taps. Although, these results are based on number of tokens and frequency rates, the data was collected in a way that it allows to make valid inferences across social factors, due to the balanced distribution of the dataset. Further in this Chapter, a Varbrul analysis will test the inferences posited in this preliminary subsection.

5.1.2 Distribution Of Rhotic Variants In Continental Spanish

We have seen so far that the bilingual Raizal Spanish variety appears to be undergoing a change in frequency towards a more canonical rhotic variant. However, we need to determine whether these rates are similar in the monolingual variety. It is worth noting that while the phonetic data for Raizal Spanish were collected following a rigorous methodological practice, the present data set for the monolingual Continental Spanish variety lack the same procedures.

Moreover, monolingual Spanish informants come from different backgrounds, and thus, the data were collected from random adult informants of any age and sex. In total 530, taps and trills tokens were collected from Continental Spanish informants who were either born in Coastal

Colombia but have lived in the Archipelago at least for 2/3 of their lifetime, or informants who were born in San Andres but without working proficiency of Raizal Creole.

Table 5-3. Overall distribution of canonical and non-canonical trills and taps in Continental Spanish. DV Type N Freq Average Cano Trill 86 0.238 16.2% Cano Tap 275 0.762 51.8% Non Trill 49 0.289 9.3% Non Tap 120 0.711 22.7%

A summary of the Continental Spanish dataset can be observed in Table 5-2. A first glance at this Table shows that at least 68% of Continental Spanish taps (n=275, 51.8%) and

147

trills (n=86, 16.2%) are produced with a canonical realization. On the one hand, taps compose the highest token proportion in the dataset with over 73% (n=395) and they seem more likely to be produced with one lingual closure (0.762). On the other hand, trills comprise the lowest rate in the data for both canonical (n=86, 16.2%) and non-canonical (n=49, 9.3%) rhotics and they seem less likely to be produced with a canonical variant (0.238). Now, we can further examine the differences across age groups and place of birth. Figure 5-5 shows a visual comparison across generations and place of birth, either in San Andres or in Coastal Colombia.

Figure 5-5. Distribution of canonical (left) and non-canonical (right) trills and taps in Continental Spanish between age groups and place of birth.

From 530 tokens from Continental Spanish, 182 rhotic samples belong to senior informants who were born in Coastal Colombia. The rest (n=348) corresponds to Spanish monolingual adults of second generation who were born in San Andres. Due to the uneven sample size of Continental Spanish informants, the patterns shown here are descriptive in nature.

However, they display general trends in the data that provide insights of rhotic distribution across social factors. For instance, older Costeños produced fewer canonical trills and taps than the younger generation. Similarly, the older generation also seems to produce a lower proportion of non-canonical rhotics, a contradictory pattern in the data.

148

Figure 5-6. Distribution of canonical and non-canonical trills and taps in Continental Spanish based on sex.

In terms of sex differences, 210 tokens were collected from female informants while 320 tokens were obtained from male informants (see Figure 5-6). As such, females appear less prone to produce canonical rhotics as shown in this dataset (only 120 tokens compared to 241 in males). In contrast, females and males display a similar use of non-canonical taps (60 and 60, respectively) and trills (30 and 19, respectively). Overall, the description of the Continental

Spanish population shows that trills are less common in the dataset than taps in both canonical and non-canonical productions. When the sample is divided by generation and place of birth, it was observed that seniors born in Coastal Colombia realize less canonical variants. At the same time, this population also have a lower count with non-canonical rhotics. The data also show differences between males and female, as males have more canonical taps and trills than females.

Both sexes have similar amounts of non-canonical rhotics in this data. As far as a comparison of social factors in Continental Spanish, it is best to consider the data in its entirety without additional categorizations, due to the uneven sample distribution across generations, place of birth, and sex. As a result, the Varbrul analysis described in the next section includes a comparison of Raizal Spanish across generations with the complete Continental Spanish dataset.

149

5.2 Variable Rule Analysis

The examination of the frequency of use of canonical and non-canonical rhotics has shown what appears to be a gradual generational change in Raizal Spanish. However, such limited analysis can only superficially show the nature of this change. To better understand, we need to establish the linguistic context where this change is permeable and whether such contexts have been favored over generations. Moreover, it has been observed that social factors such as sex and island of dwelling seem to favor one rhotic production over the other, but whether this preference is significantly present across generations and other social factors, such as education level and job marketplace, needs to be further examined.

Given the differences in frequency in Raizal Spanish, it is plausible to consider a change in the internal structure of rhotic variation in this variety that contrasts in certain generations with

Continental Spanish. The thesis posited in this study involves a connection of structural linguistic constraints between Raizal Spanish and Continental Spanish driven by younger social groups in the Archipelago. The prediction is that the internal instability of constraint hierarchies will be greater in the intermediate generation, since they are in between the maintenance of rhotic variants produced by senior Raizales and the emerging preference for innovative rhotics in younger generation due to increasing contact with Continental Spanish.

5.2.1 Preparation Of The Datasets

As examined in the previous Methodology chapter, I have established the envelope of variation where canonical and non-canonical taps and trills compete with each other. The structure of the following Varbrul analysis will examine tap and trills independently in each generation according to their canonical and non-canonical variants.

Recall that Tables 3-4 and 3-5 illustrated the independent variables and the factor levels included in this study. This was done with the aim of facilitating a comprehensive analysis of

150

this phonological phenomenon and the factors that condition tap and trill variation. However, a crosstabulation of different factor groups showed small numbers when contrasted with the tap and trill production, suggesting that the independent linguistic predictors needed to be simplified.

Tagliamonte (2012) suggests a minimum of 10 tokens per cell to reduce the likelihood of random fluctuation by 90%. A 100% reduction is possible with 35 tokens per cell. The same author also indicates that regardless of the statistical procedure, submitting multiple factors into a Varbrul program increases the chance of overtaxing the regression model, and thus, compromising the analysis. The statistical program will likely not produce optimal results if it contains too many factor levels within factor groups (p. 135).

As a standard pre-Varbrul procedure, a test for factor dependency that shows significant factor group interaction was conducted in Rbrul. In total, 15 different linguistic pairs for each tap and trill datasets were tested, resulting in 13 pairs with significant interactions within taps and six pairs within trills. Table 5-4 displays the factor groups with significant interactions.

Table 5-4. Interaction between linguistic variable pairs in the tap and trill datasets. Pairs in Taps p-Value Pairs in Trills p-Value prec_segment:stress 0.0653 prec_segment:stress 6.47e-04 foll_segment:prec_segment 7.77e-04 foll_segment:pos_word 8.38e-04 foll_segment:gram_cat 5.31e-03 foll_segment:num_syl 0 foll_segment:num_syl 0.0101 foll_segment:prec_segment 0 foll_segment:pos_word 0.0103 gram_cat:prec_segment 0 foll_segment:stress 0.0445 gram_cat:num_syl 4.83e-03 gram_cat:stress 2.50e-03 gram_cat:num_syl 0.0176 gram_cat:prec_segment 0 gram_cat:pos_word 1.01e-04 num_syl:prec_segment 1.13e-03 num_syl:pos_word 2.35e-03 pos_word:stress 0.0166

Overall, there are two types of variable pairs that present interactions: those with a categorical distribution within factors (i.e., knockouts), and those with small number of tokens

151

per cell. In both taps and trills, the phonological context has a large effect on rhotic distribution.

It seems that phonotactic constraints play a role in the interaction effects found in Table 5-4.

Taps, for instance, never appear immediately after a pause, liquid or nasal or nasal consonant, whereas trills more frequently appear before vowels rather than consonants. On the other hand, certain factor group combinations appear with multiple knockouts, and a subsequent p-value of zero, as there is nothing to measure. For instance, the low count of words with five or more syllables means that the contingent table will show no values for any combinations. This is also seen with monosyllabic words, which can only have tonic rhotics, as there is only one possible stressed context.

As a result, before the modeling of the datasets is conducted in the program, it becomes necessary to reduce the number of factor levels for each linguistic predictor, to avoid complex interactions with these unbalanced data. Accordingly, Table 5-5 shows the new simplified codification of the linguistic factor groups.

Table 5-5. Simplified codification of linguistic factor variables. Variables Categories Categorical Position in the word Word-initial Intervocalic (pos_word) Complex onset Word-medial Word-Final Preceding segment Vowel (prec_segment) Consonant Following segment Vowel (foll_segment) Consonant Stress (stress) Stressed Unstressed Number of Syllables Two- (num_syl) Three+ Grammatical Category Grammatical (gram_cat) Lexical Modifier

152

Subsequently, a test for interactions with the new coding scheme was carried out and results are visible on Table 5-6. Note that for taps the number of factor group combinations that resulted statistically significant was reduced from 13 to four.

Table 5-6. Interaction between linguistic variable pairs in the tap and trill datasets. Pairs in Taps p-Value Pairs in Trills p-Value gram_cat:pos_word 7.35e-03 gram_cat:num_syl 8.40e-04 foll_segment:gram_cat 7.97e-03 foll_segment:pos_word 1.71e-03 gram_cat:prec_segment 0.0123 prec_segment:stress 3.61e-03 foll_segment:num_syl 0.0497 foll_segment:gram_cat 0.0361 pos_word:prec_segment 0

For trills, the reduction of interaction in variable pairs was reduced to five combinations, with only position in the word and preceding segments variables displaying significant number of knockouts. Similarly, it can be observed that following and preceding segment displays significant interaction effects with word position. This is because vocalic and consonantal elements are highly dependent to word context. For instance, every rhotic in intervocalic position is surrounded by vowels (i.e., carrera); in complex onsets, a rhotic is preceded by a consonant

(i.e., Providencia) and followed by a vowel; in word-final positions they are might be preceded by a vowel (i.e., mar ‘sea’) or a consonant (i.e., compre ‘buy’); and a rhotic is always prevocalic in word-initial contexts (i.e., ron ‘rum’). All these factors are correlated and are dependent on each other, and thus, they cannot be included within the same model. As a result, preceding segment and following segment were analyzed independently with tap and trills production to determine the greatest contribution to the Varbrul models.

The first comparison with tap variation showed that word position appears with lower deviance and Akaike information criterion or AIC2 (2631.135, 2643.135, respectively) compared

2 The Akaike information criterion or AIC is a test that indicates the relative quality of the statistical model. The lower the score the better the model.

153

to a model including following and preceding segment (Dev: 2637.62, AIC: 2647.62), which suggest that a Varbrul regression with word position slightly improves the relative quality of the statistical model. Similarly, these factors were inspected in the trill dataset. Results show that a model with preceding and following segment improves the Varbrul model (Dev. 534.731, AIC

546.731) compared to including only word position (Dev. 562.515, AIC 570.515). Corroborating the improvement of the models, a test for interaction in both datasets was conducted with all the linguistic predictors with tap and trill production.

Table 5-7. Final test of interaction between linguistic variable pairs in the tap and trill datasets. Pairs in Taps p-Value Pairs in Trills p-Value foll_segment:num_syl 8.62e-04 num_syl:stress 9.19e-03

Table 5-7 summarizes the results of the final test for interaction between linguistic factor pairs. No factor pairs were found with a significant interaction effect for taps. On the contrary, two factor group combinations were found significantly correlated in trills: following segment and number of syllables, and number of syllables and stress. At this point, we have simplified the overall complexity of our data as far as possible and now whether further reduction of the number of factor groups is necessary, then it is contingent to the detailed examination of each individual generation. In conclusion, the final linguistic factors that will be submitted for independent analysis of taps and trills are the following:

Taps (6 factors) Word position, stress, number of syllables, grammatical category, F3, Duration

Trills (7 factors) Preceding segment, following segment, stress, number of syllables, grammatical category, F3, Duration

Finally, extralinguistic factors were also analyzed to determine interaction effects between highly correlated variables. A significant interaction between sex and education level as

154

a mismatch is present in the proportions hierarchy for high school educated informants and primary and tertiary education levels (p=0). Further inspection of the crosstabulated data shows a categorical distribution, where multiple knockouts are present as no females compose primary- educated informants. Furthermore, there is only one male with tertiary education in the sample.

These interactions correspond to a limitation in the data collection procedures and the fact that the samples were primarily collected based on the generation of the informants in both islands. A comparison between these two variables in terms of quality improvement of the model was conducted, revealing a higher deviance (1065) and a higher AIC (1073) when education level is correlated with tap production, as compared to sex (Dev.: 1061, AIC: 1069). The same pattern was visible with trills. The variable sex improved the relative quality of the statistical model by a slight margin. Compare the deviance (573) and AIC (581) of the latter with that of education level (Dev. 579, AIC 587). Consequently, education level will be excluded from the present

Varbrul analysis of extralinguistic variables.

With this simplified version of the factor levels and the reduction of the interaction effects within the datasets, we now turn to the analysis of the datasets for each generation according to the canonical and non-canonical classification in both taps and trills.

5.2.2 First Generation

Previously, it was noted that older Raizales have a notable preference for non-canonical forms as compared with other younger generations based on the distribution and frequency of the

Raizal Spanish dataset. Table 5-4 and Table 5-5 show the results of the mixed effects logistic regression for this generation in the variable rule program Rbrul (Johnson, 2009). In total, 1,146 tokens were selected from third generation, accounting for a third of the Raizal Spanish dataset

(33.6%). Two separate models were run that distinguishes taps from trills. As noted in chapter 4, taps and trills occur in exclusive word contexts and contrast phonemically in intervocalic

155

position. In addition, just 7 trill tokens were found within the only variable contexts that share with taps (i.e., word-medial and word-final positions), which renders a contrast between taps and trills inadequate for a statistical analysis.

For our first regression model, the dependent variable involved canonical vs non- canonical taps, and the application value consist of canonical taps. The selection of the independent variables was contingent to the degree of interaction between factors. From the 4 linguistic predictors, no factor pair interaction was found significant at p = 0.05, and thus all predictors were submitted for analysis.

Table 5-8. Significant linguistic factors contributing to canonical taps in first generation informants Application Value: Canonical Tap Random Intercept: Token and Informant F3 = p = 3.01e-14 Log Odds 0.00294 Word Position = p = 8.65e-08 Factors Log Odds N Proportion Weight Intervocalic 0.645 428 0.467 0.656 Final 0.541 119 0.370 0.632 Medial -0.439 101 0.208 0.392 Complex -0.747 250 0.168 0.321 Range 33 Model Total = 898 DF = 7 Input Prob. = <.001 Overall R2 = 0.45 Dev. 893.368 AIC: 907

The quantitative analysis of taps revealed that the existence of phonetic patterns and the occurrence of canonical taps is not random for senior Raizales. As is shown in Table 5-8, both step-up and step-down model matched, and the program chose two predictors as significant: F3 and position in the word. F3 frequencies display the high statistical significance in rhotic production (p = 3.01e-14). Although the positive effect is marginal (log odds = 0.00294), canonical taps seem to be realized with higher F3 frequencies. In other words, higher first resonance values in the vocal cavity is correlated with consistent lingual closures, and thus, the production of a canonical tap. In addition, word position appears as the most powerful logistic

156

predictor, indicated by its lowest p-value and its range between factors (33). A detailed categorization of the word position factor group also shows a significant effect in rhotic variation

(p= 8.65e-08). Taps in intervocalic and final positions are more favorable to canonical productions (Fw= 0.656, and Fw=0.632, respectively), while preconsonantal medial and complex syllable onsets positions show a contributing effect for non-canonical taps (Fw= 0.392, and

Fw=0.321, respectively).

Table 5-9. Non-significant linguistic factors contributing to canonical taps in first generation informants. Application Value: Canonical Random Intercept: Token and Informant Grammatical Category = (Not significant) p = 0.352 Factors Log Odds N Proportion Weight Modifier 0.23875 130 0.408 [0.559] Lexical -0.00675 638 0.343 [0.498] Grammatical -0.232 130 0.269 [0.442] Range 11 Duration = (Not significant) p = 0.568 Log Odds -3.653 Stress = (Not significant) p = 0.634 Factors Log Odds N Proportion Weight Stressed 0.0422 366 0.352 [0.511] Unstressed -0.0422 532 0.335 [0.489] Range 3 Number of Syllables = (Not significant) p = 0.938 Factors Log Odds N Proportion Weight Three+ 0.00723 458 0.36 [0.502] Two- -0.00723 440 0.323 [0.498] Range 1 Model Total = 898 DF = 8 Input Prob. = 0.32 Overall R2 = 0.33 Dev.: 1021.634 AIC: 1037

While the regression model in Table 5-8 displays the significant factor groups in first generation for taps, Table 5-9 shows the output the rest of the predictors that fail to reach statistical significance, including grammatical category, duration, stress, and number of syllables.

The main observation from this analysis involves that none of these linguistic predictors reach statistical significance, and their effect is rather small due to the short range between factors.

157

While the model presented in Table 5-8 and Table 5-9 represent the linguistic constraints that condition the production of canonical and non-canonical taps, this model also considers the random effect of individual differences and lexical density. The inclusion of two random intercepts allowed the models to account for the effect of individuals in the dataset and the production of repetitive tokens in the data, resulting in a regression model that analyses the within-group correlation of balanced and unbalanced grouped data, such as this linguistic dataset.

The effect of the significant predictor in tap variation in this generation can also be seen graphically. This is presented by means of a conditional inference tree produced in R (Liaw &

Wiener, 2002).

Figure 5-7. Conditional inference tree of the significant linguistic factors for taps in first generation Raizales.

As shown in our model from Table 5-7, the random forest in Figure 5-7 selected F3 as the most significant predictor at the top of the tree, while rhotic position in the word appear significant for rhotics with a F3 frequency over 2367.309 Hz. Likewise, word position is also significant when it gravitates between F3 frequencies of 2367 HZ and 2086.25 Hz. Word position only appear as the lowest significant linguistic factor for F3 frequencies equal or lower than 2086. Taps are more frequently non-canonical in final and intervocalic position below the

158

2086 Hz range. As F3 frequencies increase, canonical variants are more frequent, particularly in final and intervocalic positions. These findings indicate that the lower the F3 frequencies the more likely to encounter a non-canonical variant in all word positions. In contrast, canonical taps seem more likely to occur with increasing F3 frequencies over the 2086 Hz mark. However, this effect is more robust when the F3 values rise above 2367 Hz, a pattern seen in Table 5-8.

The second analysis within this generation corresponds to trill variation. Trills in this dataset comprise much smaller number of observations. In total, 248 tokens correspond to trills in this generation. More remarkably, there were only 13 instances of canonical trills in this dataset. This suggests that trills with the normative lingual closures are highly infrequent in this population, and almost not present in their phonetic inventory. A closer look into the distribution of this rhotic shows that 8 out 13 instances of trill variant appeared in initial position.

Additionally, they occur exclusively in prevocalic position, and more frequently within nouns with less than 3 syllables long.

This situation poses a problem to our analysis due to the low number of tokens for the canonical trill, which would be insufficient for statistical significance. In resolving this problem, it is important to distinguish the acoustic properties of the trill variants. Recall that canonical variants were selected based on whether they were realized with two or more lingual occlusion, while non-canonical trills consisted of segments with zero visible lingual closure and continuous formant structure. In addition, trills with one apical occlusion and a prolonged opening phase or r-coloring (Bradley and Willis, 2012; Henriksen and Willis, 2010) have also been observed in this data set and were also classified as non-canonical variants. This type of trill, where approximant phases accompany lingual contacts, has been argued to exist in all languages that have trills (Blecua, 2001; Ladefoged and Madison, 1996). Similarly, we have established in

159

chapter 5 that the manner of articulation of non-vibrant rhotics in the Archipelago of San Andres correspond to an approximant realization, for both Spanish and Raizal Creole varieties. The most important difference between these two languages is the lack of any lingual contact in Creole, while any realization with no apical occlusion in Spanish correspond to an approximant realization of taps and trills. As a result, it is safe to assume that vibrant trills with one lingual closure in Raizal Spanish deviates from the Raizal Creole norm even if they are not produced with the normative two lingual closures. Precisely, when we filter trills with only one apical closure and trills with zero occlusions, 0.36% of tokens (n = 91) appear for the former and 0.58% for the latter (n = 144) , which suggest that both variants present different rates of occurrence, and thus behave different in this generation.

For this reason, a second Varbrul model was run following a different categorization of the dependent variable for trills. To start, a distinction was made between trills that show any visible lingual contact and those that appear as approximant trills (i.e., zero lingual contact).

Then, a new classification emerges for the dependent variable of the trill dataset that follows a corresponding distinction with taps: vibrant (N=104) and non-vibrant (N=144). As a result, a more balanced distribution was obtained, and the dataset was ready to be submitted to the regression analysis with Vibrant Trill as the application value. Two significant interactions were found in a preliminary model: following segment and grammatical category (p= 0.0241), and grammatical category and stress (p=0.0443). Subsequently, these factor groups were correlated individually with the dependent variable resulting in all the factors significantly predicting trill variation. As a result, five independent linguistic predictors and two acoustic correlates were submitted for analysis: stress, position in the word, following segment, preceding segment, grammatical category, duration and F3.

160

Table 5-10 shows the results of the regression model with the new classification for trills and the significant factor groups conditioning trill variation.

Table 5-10. Linguistic factors contributing to vibrant trills in first generation informants. Application Value: Vibrant Random Intercept: Token and Informant Preceding Segment = p = 1.63e-05 Factors Log Odds N Proportion Weight Vowel 1.211 204 0.475 0.77 Consonant -1.211 42 0.143 0.23 Range 54 F3 = p = 1.93e-03 Log Odds 0.00252 Grammatical Category = p = 0.0102 Factors Log Odds N Proportion Weight Noun 0.508 157 0.459 0.624 Other -0.508 89 0.348 0.376 Range 25 Model Total = 246 DF = 6 Input Prob. = <.001 Overall R2 = 0.638 Dev.: 231.973 AIC: 243.9

The most telling result in this model is that preceding segment and F3 are significant factor groups for vibrant trill production, while grammatical category also has a significant effect on vibrant trill realization. Lingual contacts in contexts where a vowel is preceding the trill highly predict a vibrant realization. It is not clear, however, if the effect is more frequent in intervocalic or word-initial contexts when coarticulated with a vowel in a preceding word.

Moreover, higher F3 frequencies represent the acoustic cue that signals a vibrant realization.

Nouns also appear more likely to contain a vibrant realization. These results are visually enhanced in Figure 5-8, which show two inference trees with preceding segment and position in the word.

161

Figure 5-8. Conditional inference trees of the significant linguistic factors for trills in first generation Raizales.

F3 values over 2128 Hz appear as the main continuous predictor of vibrant trills. The relationship between rhotic position within the word and preceding segment is correlated to F3 values below 2128 Hz mark. First, no vibrant trill was found with a preceding consonant when

F3 drops below 2128 Hz. However, when this drop in F3 is maintained, intervocalic positions favor more vibrant realizations assisted mainly by the preceding vowel segment than in word- initial position. This suggest that the drop in F3 highly predicts that a vibrant trill is less likely to be realized unless a vowel precedes the trill segment. Nevertheless, the phonological vowel context might subsidize the lack of aerodynamic conditions for the realization of a vibrant trill.

On the contrary, the higher the F3 values, the more likely a vibrant trill is produced, and thus, a canonical variant occurs. This acoustic pattern is also visible in taps, mainly in intervocalic positions. Non-significant factors were also correlated with trill variation and results are presented in Table 5-11. Duration appears with a significant effect and a negative correlation with vibrant trills, suggesting that this generation produces shorter vibrant segments in trill contexts than non-vibrant trills.

162

Table 5-11. Non-significant linguistic factors contributing to vibrant trills in first generation informants. Application Value: Vibrant Trill Random Intercept: Token Duration = p = 7.02e-03 Log Odds -15.127 Stress = p = 0.0762 Factors Log Odds N Proportion Weight Stressed 0.274 85 0.518 [0.568] Unstressed -0.274 161 0.366 [0.432] Range 13 Following Segment = p = 0.28 Factors Log Odds N Proportion Weight Other 0.16 91 0.505 [0.54] Mid -0.16 155 0.368 [0.46] Range 8 Number of Syllables = p = 0.764 Factors Log Odds N Proportion Weight Three+ 0.0428 149 0.423 [0.511] Two- -0.0428 97 0.412 [0.489] Range 3 Model Total = 246 DF = 6 Input Prob. = 0.68 Overall R2 = 0.0781 Dev.: 319.581 AIC: 331.5

This resonates with our findings in Chapter 4 with regards to the neutralization of taps and trills in intervocalic positions in older Raizales. In other observations, while stress approaches significance, other linguistic variables appear far from significantly predicting trill variation.

Similarly, a subsequent analysis of social factors that contribute to rhotic variation in first generation was conducted. First, it was observed that genre and job marketplace presented significant interaction effects between other factor group pairs. As a result, it was further determined to exclude these two variables, since they fail to significantly predict tap production

(p=0.826, and p=0.577, respectively). Results of the Varbrul analysis can be seen in Table 5-12 for taps and Table 5-13 for trills. The model for taps selected sex and island of dwelling as stastically significant.

163

Table 5-12. Extralinguistic factors contributing to canonical taps in first generation informants. Application Value: Canonical Random Intercept: Token and Informant Sex = p = 8.71e-04 Factors Log Odds N Proportion Weight Female 0.726 466 0.483 0.674 Male -0.726 432 0.19 0.326 Range 35 Island of Dwelling = p = 0.0101 Factors Log Odds N Proportion Weight Old Prov. 0.493 462 0.452 0.621 San Andres -0.493 436 0.225 0.379 Range 25 Model Total = 898 DF= 5 Input Prob. = 0.28 Overall R2 = 0.363 Dev. = 1009.3 AIC = 1019.3

First generation females appear to highly favor the realization of canonical taps (Fw =

0.674), while males favor the non-canonical variant (Fw = 0.326). In addition, canonical taps are favored in the island of Old Providence (Fw= 0.621), while San Andres appear to favor non- canonical productions (0.379). These results are displayed graphically in the random tree of

Figure 5-9.

Figure 5-9. Conditional inference tree of the significant social factors for taps in first generation Raizales.

164

There are significant differences between males and females in both San Andres and Old

Providence. Females in San Andres, but mainly in Old Providence, seem to be leading the change for more canonical tap variants compared to males. Until we analyze the Continental

Spanish dataset, this finding seems to be challenging our assumption that contact with Spanish in

San Andres correspond to the leading cause of a potential rhotic change. A further examination of the linguistic background of these female informants suggest that they have a great deal of interaction with Spanish speaking foreigners: two of them work or were formerly employed in one of the hotels of Old Providence, one as a cook and the other as a receptionist; the final female informant is a homemaker, and is proficient in Gipsy and English, which suggest that she might have been in contact with foreigners to the island. This background experience suggest that contact with Spanish is still plausible but it is hard to attest up to this point. Let’s now turn to the extralinguistic factors that condition trill variation in first generation Raizales.

A check-up of the dataset revealed, two signification interaction pairs between genre and job marketplace (p=0.0377), and genre and island of dwelling (p=0). When correlated individually with trill production, it was determined that all these variables failed to reach statistical significance, and thus, they were excluded from the analysis.

Table 5-13. Social factors contributing to vibrant trills in first generation informants. Application Value: Vibrant Trill Random Intercept: Token and Informant Sex = p = 6.53e-03 Factors Log Odds N Proportion Weight Female 1.332 120 0.608 0.791 Male -1.332 126 0.238 0.209 Range 59 Model Total = 246 DF = 4 Input Prob. = 0.362 Overall R2 = 0.266 Dev. = 262.8 AIC = 270.8

Table 5-13 displays the results of the extralinguistic varables that predict trill variation in first generation. Trill variation appears to be conditioned only by sex. Again, females favored a

165

vibrant version of the trill (0.791). A conditional inference tree in Figure 5-10 renders graphically these results and includes island of dwelling for comparison with taps.

Figure 5-10. Conditional inference tree of the significant social factors for trills in first generation Raizales.

As noted for taps, females are more likely to favor a vibrant realization than males.

However, the effect of the island of dwelling is only visible within females (0.73%, N =

225/307), which shows why this variables was excluded by the model in Table 5-13. While rhotic variation is contingent to a combination of acoustic cues (i.e., F3) and linguistic predictors, namely word position in taps and preceding segment in trills, it is yet to determine whether the same contraints hold for younger generations, and if females in Old Providence continue leading the preference for a canonical tap and vibrant trill. Let’s now turn to the Varbrul analysis of second generation Raizales.

5.2.2 Second Generation

Second generation represents the transitional age group between first generation and third generation. On the one hand, they are the offspring of senior Raizales. On the other hand, they are the parents of younger Raizales. Moreover, they had the burden of maintaining their ancestral

166

language, but also in keeping up with the increasingly important position of Spanish after the declaration of the San Andres as a free-duty port. If there is generational breaking point, this generation should display the divergence in the linguistic behavior of taps and trills.

The analysis for this generation comprises 1,111 tokens (32.6% of the whole data set).

This data were submitted for analysis and the Varbrul results are presented from Table 5-14 to

Table 5-20. In order to reach the results for taps, 772 tokens were selected for this variant and a test for interacting factors was conducted revealing one significant interacting factor pair within this dataset (i.e., grammatical category and position in the word at p = 0.032). In order to determine the level of significance of these factors, an individual correlation analysis was carried out, indicating word position as the only significant factor predicting tap variation (p = 1.3e-04).

Consequently, a mixed-effect regression run on Rbrul with token and informant as random intercepts was submitted with all the significant linguistic predictors with the exception of grammatical category. A detailed description of this test is shown in Table 5-14.

Table 5-14. Significant linguistic factors contributing to canonical taps in second generation informants. Application Value: Canonical Tap Random Intercept: Token and Informant F3 = p = 2.46e-14 Log Odds 0.00246 Duration = p = 2.34e-04 Log Odds 43.684 Word Position = p = 0.0104 Factors Log Odds N Proportion Weight Final 0.579 77 0.779 0.641 Medial 0.322 76 0.711 0.58 Intervocalic -0.171 447 0.633 0.457 Complex -0.73 172 0.424 0.325 Range 32 Model Total = 772 DF = 8 Input Prob. = 0.001 Overall R2 = 0.684 Dev.: 663.774 AIC = 679.7

167

The test showed one significant linguistic predictor (i.e., word position), and two significant continuous acoustic correlates: duration and F3. For this generation, positive values in acoustic cues, such as duration and F3 predict a canonical tap production. This resonates with previous findings from first generation, as higher F3 frequencies signal a canonical production.

Longer durations also indicate that Raizales are actually producing a canonical variant, as the vibrant realization takes longer times to be realized (Colantoni, 2006; Mendoza et al, 2003;

Colantoni, 2001). Postvocalic positions in word-medial and word-final contexts also favored a tap with a lingual contact. This time around, intervocalic positions favor less a canonical tap in this generation.

Table 5-15. Non-significant linguistic factors contributing to canonical taps in second generation informants. Application Value: Canonical Tap Random Intercept: Token and Informant Stress = p = 0.0326 Factors Log Odds N Proportion Weight Stressed 0.248 313 0.668 [0.562] Unstressed -0.248 459 0.569 [0.438] Range 13 Number of Syllables = p = 0.866 Factors Log Odds N Proportion Weight Two- 0.0239 634 0.62 [0.506] Three+ -0.0239 138 0.558 [0.494] Range 1 Model Total = 772 DF = 4 Input Prob. = 0.61 Overall R2 = 0.644 Dev.: 752.835 AIC: 762.8

A second regression model presenting the factors that failed to reach statistical significance in the first model is presented in Table 5-15. While stress becomes significant without word position in the model (p=0.0326), the number of syllables appear far less correlated with canonical taps (p= 0.866). The higher number of words with more than three syllables created a new factor level that compares with monosyllables, disyllables, and three-syllable words. However, this variable is yet to produce a significant effect in tap production. The

168

findings for this generation suggest that the linguistic and acoustic constraints for tap variation are changing order, mainly because tap position within the word is being favored by other contexts, rather than intervocalic and final positions. Graphically, the new word context for canonical taps is displayed in the conditional tree in Figure 5-11.

Figure 5-11. Conditional inference tree of the significant social factors for taps in second generation Raizales.

F3 frequencies appear as the most significant in the upper part of the tree. Then, a distinction is made for taps that have F3 frequencies either below or above 2079 Hz values.

When frequencies decrease this value, canonical final-word taps are more frequently than other rhotic positions. However, higher F3 frequencies in final-word, medial, and intervocalic positions correspond to increasingly numbers of canonical taps. Only when F3 rises over 2348

Hz, then consonant clusters in syllable-initial position show more canonical variants. Finally, segmental durations of over 0.028 milliseconds also correlate with higher F3 values when these reach over 2452 Hz. In sum, the higher the F3 and the longer the segmental duration, the more canonical taps are realized in this generation, which signals that these informants are producing vibrant taps.

169

To test whether a change in hierarchical constraint order in taps has also its corresponding effect in trills within this generation, I extracted 339 tokens for bilingual trills. An inspection of the distribution of canonical and non-canonical trills also revealed a disproportionate pattern, where trills with two or more lingual contacts comprise 15.6% of the dataset (n=53). Subsequently, it was observed that, while trills with multiple lingual contacts are scarce in the dataset, variants with one lingual closure appear with an average of 45.2% (n=154).

As a result, the same distinction as in first generation between vibrant and non-vibrant trills was implemented for this analysis (n=207 and n=132, respectively). Although the vibrant/non-vibrant comparison is different in this generation by a ratio of 1.6 to 1 when compared to first generation

(1 to 1.3), the distinction still captures the frequency differences between these two trill categories.

As has been done previously, a test for interaction between linguistic factors pairs was conducted and one factor combination was found to have significant interactions between p=

<0.04. After individual correlation tests were carried out for each individual factor, preceding segment appeared significant for trill variation (p= 4.75e-05).

Table 5-16. Significant linguistic factors contributing to vibrant trills in second generation informants. Application Value: Vibrant Trill Random Intercept: Token and Informant Preceding Segment = p = 1.15e-05 Factors Log Odds N Proportion Weight Vowel 1.098 280 0.657 0.75 Consonant -1.098 55 0.364 0.25 Range 50 Duration = p = 0.0122 Log Odds 20.639 F3 = p = 0.0382 Log Odds 0.0013 Model Total = 335 DF = 6 Input Prob. = 0.012 Overall R2 = 0.789 Dev.: 205.78 AIC: 217.7

170

As a result, number of syllables was excluded from the model and four linguistic variables and two acoustic predictors were submitted into the model. Results of the Varbrul analysis for trills is shown in Table 5-16. The presence of a vibrant trill appears significantly favored when a vowel is preceding the segment, while longer duration (log odds = 20.639) and higher F3 values (log odds = 0.00138) also significantly contribute to its realization. When these results are contrasted with senior Raizales, grammatical category disappears as a linguistic constraint for second generation informants, remaining only F3 and the preceding segment, and segmental duration emerging as an acoustic predictor used for vibrant trills. This resonates with

Colantoni (2006) and Mendoza et al (2003), who indicate that canonical trills present longer segments than non-vibrant rhotics (i.e., approximants).

Figure 5-12. Conditional inference tree of the significant linguistic factors for trills in second generation Raizales.

A visualization of the linguistic nature of vibrant trills can be seen in Figure 5-12. Over

85% of trills realized with F3 values below 2016 Hz are non-vibrant. In contrast, the effect of the preceding segment and duration appear on trills produced with F3 frequencies over 2016 Hz: lingual contacts are more likely to be produced whenever a trill is preceded by a vowel. On the

171

contrary, the conditions required for a word-initial vibrating trill preceded by a consonant to appear with lingual contacts, correspond to durations over 0.065 milliseconds. In other words, vocalic environments are more conducive to a vibrant realization, while preconsonantal trills require acoustic cues to produce lingual closures.

In a separate analysis, Table 5-17 summarizes the linguistic factors that appear as non- significant for trill production. Despite the effect fails to approach significance, high and low

(i.e., Other) vowels seem more likely to occur with a vibrant production compared to mid vowels. Stress and grammatical category also appear as non-significant for vibrant trill production in second generation Raizales.

Table 5-17. Non-significant linguistic factors contributing to vibrant trills in second generation informants. Application Value: Vibrant Trill Random Intercept: Token and Informant Following Segment = p = 0.13 Factors Log Odds N Proportion Weight Other 0.371 135 0.652 [0.592] Mid -0.371 200 0.58 [0.408] Range 19 Stress = p = 0.189 Factors Log Odds N Proportion Weight Stressed 0.317 155 0.613 [0.579] Unstressed -0.317 180 0.606 [0.421] Range 15 Grammatical Category = p = 0.379 Factors Log Odds N Proportion Weight Verb 0.229 57 0.702 [0.557] Other -0.229 278 0.59 [0.443] Range 11 Model Total = 335 DF = 6 Input Prob. = 0.713 Overall R2 = 0.79 Dev.: 225.271 AIC: 237.27

Results on extralinguistic variables for taps and trills were also analyzed for this generation. The Varbrul results for extralinguistic factors contributing to tap variation are presented in Table 5-18. A test for interaction showed significant effects in two factor pairs and

172

so island of dwelling and job marketplace were excluded as the non-correlated factors in tap production.

Table 5-18. Social factors contributing to canonical taps in second generation informants. Application Value: Canonical Tap Random Intercept: Token and Informant Sex = p = 0.0416 Factors Log Odds N Proportion Weight Female 1.357 462 0.732 0.795 Male -1.357 310 0.426 0.205 Range 59 Genre = p = 0.0244 Factors Log Odds N Proportion Weight Narration 0.433 86 0.814 0.607 Interview -0.433 686 0.583 0.393 Range 21 Model Total = 772 DF = 5 Input Prob. = 0.62 Overall R2 = 0.63 Dev.: 748.133 AIC: 758.13

As seen by the magnitude effect in the range of the factor levels in sex, female informants highly favored a canonical tap realization. Likewise, the method for collecting the data appear significant for this generation, which suggests that bilingual Raizales were paying more attention to their pronunciation during the narration of the Frogstory (Mercer, 1969) but were producing more casual speech during the interview.

Figure 5-13. Conditional inference tree of the significant social factors for taps in second generation Raizales.

173

Subsequently, a conditional random tree displaying the significant extralinguistic factors influencing tap production is presented in Figure 5-13. As can be seen in this visual representation, males exhibit competing values between a canonical and a non-canonical realization. In females, however, tap realization is dependent on the mode of production, either through the sociolinguistic interview or during the narration of the graphical story, as slightly more canonical taps are realized during this task. Thanks to this Figure, it can be observed that females seem to be paying more attention to their pronunciation during the narration task and maintain the prestigious vibrant variant more frequently compared to males.

Finally, the extralinguistic factors contributing to trill production were also submitted for analysis. First, it was noted a significant interaction between job marketplace and sex (p= 0), and it was subsequently determined that only sex significantly predicted trill variation (p= 0.0277).

As a result, three extralinguistic variables were submitted for analysis: sex, genre, and island of dwelling.

Table 5-19. Social factors contributing to vibrant trills in second generation informants. Application Value: Vibrant Trill Random Intercept: Token and Informant Sex = p = 0.0269 Factors Log Odds N Proportion Weight Female 2.194 191 0.822 0.9 Male -2.194 144 0.326 0.1 Range 80 Model Total = 335 DF = 4 Input Prob. = 0.54 Overall R2 = 0.786 Dev.: 227.429 AIC: 235.42

Table 5-19 shows the results of the Varbrul model. Females favor a vibrant trill in significant proportions, as opposed to males who prefer a non-vibrant realization. No significant effect was found according to the island of dwelling or genre. When we visualize the nature of these extralinguistic constraints in Figure 5-14, the random forest displays the behavior of trills in males and females.

174

Figure 5-14. Conditional inference tree of the significant social factors for trills in second generation Raizales.

For males, non-vibrant trills are the norm, contrary to females, who are more likely to produce a trill with a lingual contact, a pattern also seen with taps. In sum, the Varbrul analysis of the second generation Raizal bilinguals has established several findings. First, acoustic cues, as segmental duration and F3, appear increasingly more important for predicting both tap and trill variation in this dataset. While position in the word seems significant for taps, the vocalic segment preceding the rhotic has a more significant effect for trill production. In addition, females continue to spearhead the production of canonical taps and vibrant trills in the

Archipelago. However, informants appear more aware of their speech and produced a canonical tap according to the genre of the data elicitation method, either during the interview or during the narration of the Frogstory. This effect was not visible in trills. Let’s now continue with our final

Varbrul analysis involving the third generation Raizales.

5.2.3 Third Generation

Third generation represents the youngest informants recruited for this study and the population who were born during the continental immigration expansion of the decade of 1990.

For this analysis, 1,151 tokens were collected from this generation accounting for 33.8% of the whole dataset for bilingual Raizal Spanish. Taps account for 75% of the sample (N=861), while

175

the rest comprises trills (N=290). For taps, a test for interaction in factor group pairs was conducted revealing no significant interaction between variable combinations. As a result, a first regression of the model included four linguistic variables and two acoustic predictors.

Table 5-20. Significant linguistic factors contributing to canonical taps in third generation informants. Application Value: Canonical Taps Random Intercept: Token and Informant F3 = p = 1.24e-11 Log Odds 0.0025 Word Position = p = 2.28e-05 Factors Log Odds N Proportion Weight Medial 0.675 92 0.88 0.663 Final 0.523 125 0.832 0.628 Intervocalic -0.322 454 0.751 0.42 Complex -0.876 190 0.653 0.294 Range 36 Model Total = 861 DF = 7 Input Prob. = 0.01 Overall R2 = 0.317 Dev.: 848.331 AIC: 862.33

Results of the significant factors selected in the model are presented on Table 5-20. In total, two predictors were selected significant in the model: F3 and word position. As expected, when compared to a non-canonical tap, slightly higher F3 values signal a canonical realization with a positive log odds of 0.0025. Similarly, the phonological environment that surrounds the tap segments is highly predictive of a canonical variant. Taps preceded by a vowel in preconsonantal position are more likely to be produced with a lingual contact. Similarly, word- final contexts contribute to a canonical tap in this generation. Recall that adult informants within second generation started favoring canonical taps in final and word-medial contexts, while senior

Raizales preferred prevocalic taps in intervocalic and final positions, younger population completed this change. Whether the preference for canonical taps in these positions correlates with the monolingual Spanish variety will be tested in Section 5-3.

176

Figure 5-15. Conditional inference tree of the significant linguistic factors for taps in third generation Raizales.

Graphically, a random tree in Figure 5-15 shows the effect between the acoustic predictor and the linguistic variable. Taps with F3 values higher than 2621Hz seem to be unaffected by any linguistic predictor and show significantly more canonical taps than variants without any lingual occlusion. In contrast, segments with lower F3 values than the 2621 Hz mark seem to be particularly influenced by word position. Greater proportions of canonical taps appear post- vocalically in preconsonantal medial position. In comparison, intervocalic and word-final taps are produced with slightly decreasing amounts in their non-canonical variant. Similarly, taps in syllable-initial, consonant clusters appear affected by the F3 frequencies produced: F3 frequencies below 2419 Hz are more likely to be realized as an approximant compared to variants over this mark. These findings suggest that lower F3 frequencies are consistently predicting a non-vibrant variant, a pattern also found in the previous generations.

Resuming with the analysis of non-vibrant tap variants, Table 5-21 show the results of a second regression model that includes the three independent linguistic variables that fail to reach statistical significance in the main model.

177

Table 5-21. Non-significant linguistic factors contributing to canonical taps in third generation informants. Application Value: Canonical Taps Random Intercept: Token and Informant Stress = p = 7.02e-03 Factors Log Odds N Proportion Weight Stressed 0.316 378 0.807 [0.578] Unstressed -0.316 483 0.714 [0.422] Range 15 Duration = (Not Significant) p = 0.213 Log Odds 12.688 Grammatical Category = p = 0.344 Factors Log Odds N Proportion Weight Open 0.211 686 0.773 [0.552] Closed -0.211 175 0.686 [0.448] Range 11 Number of Syllables = (Not Significant) p = 0.49 Factors Log Odds N Proportion Weight Four+ 0.103 131 0.794 [0.526] Three- -0.103 730 0.748 [0.474] Range 5 Model Total = 861 DF = 7 Input Prob. 0.74 Overall R2: 0.295 Dev.: 907.844 AIC: 921.844

In this model, stress becomes statistically significant when word position is excluded.

Taps in stressed syllables is correlated with a canonical realization. On the contrary, duration, the grammatical category, and the number of syllables fail to reach statistical significance for tap production.

Similarly, trills offer an opportunity to check whether these linguistic constraints are consistent across allophonic variants. In this case, it was observed that zero-occlusion tokens make up 24% of the dataset (N=70). Moreover, 110 tokens with one lingual contact were found

(38%), and the sample also contains 110 trills with multiple lingual contacts (38%). Since there is a disproportion in approximant trills compared to segments with any visible apical closure, two models will be compared based on two distinctions. Trill variants in the first model will contrast vibrant and non-vibrant productions. The second model will be dividing trills into canonical and non-canonical realizations. In other words, I will test trills with two or more

178

lingual contacts, and non-canonical trills, consisting of zero occlusions and only one apical closure.

A test for interaction in the first model was carried out and found that one significant effect in factor group pairs. Consequently, these factor groups were individually correlated with the dependent variable and it was found that both stress and grammatical category appear as non- significant predictors for trill production (p = 0.715 and p = 0.453, respectively). Subsequently, the Varbrul test was run with all the other linguistic and acoustic predictors. The model selected one significant linguistic factor and no acoustic predictors.

Table 5-22. Significant linguistic factors contributing to vibrant trills in third generation informants. Application Value: Vibrant Trills Random Intercept: Token and Informant Preceding Segment = p = 3.05e-04 Factors Log Odds N Proportion Weight Vowel 0.772 231 0.814 0.684 Consonant -0.772 55 0.527 0.316 Range 37 Model Total = 286 DF = 4 Input Prob. = 0.72 Overall R2 = 0.335 Dev.: 278.452 AIC: 286.45

As seen in Table 5-22, vibrant trills are more favorable in postvocalic contexts (Fw =

0.689) rather than postconsonantal positions (Fw = 0.311). Correspondingly, Figure 5-16 displays the effect of the preceding vocalic segment in the production of a vibrant trill.

Figure 5-16. Conditional inference tree of the significant linguistic factors for trills in third generation Raizales.

179

Postvocalic trills are more likely to be produced with a lingual contact than preconsonantal trills. Moreover, these postvocalic segments appeared more frequently in the dataset than their postconsonantal counterparts. The number of syllables was excluded from the random tree, because it only reaches significance when random variables are included in the model. In a related analysis, Table 5-23 shows the factors that fail to reach statistical significance in our first model.

Table 5-23. Non-significant linguistic factors contributing to vibrant trills in third generation informants. Application Value: Vibrant Trills Random Intercept: Token and Informant F3 = (Not Significant) p = 0.125 Log Odds 0.00109 Number of Syllables = (Not Significant) p = 0.472 Factors Log Odds N Proportion Weight Three+ 0.153 149 0.785 [0.538] Two- -0.153 137 0.73 [0.462] Range 7 Following Segment = (Not Significant) p = 0.629 Factors Log Odds N Proportion Weight High 0.148 35 0.771 [0.537] Other -0.148 251 0.757 [0.463] Range 7 Duration = (Not Significant) p = 0.661 Log Odds -2.916 Model Total = 286 DF = 7 Input Prob. = 0.26 Overall R2 = 0.375 Dev.: 287.793 AIC: 301.79

No other linguistic variable nor acoustic predictor was found to be significant in the

Varbrul program, suggesting that vibrant trills are favored with specific linguistic constraints relating to the phonological context.

Since we have established that it is possible to classify trills on canonical and non- canonical, thanks to the increasing production of vibrant trills in this generation, I will compare common patterns in linguistic constraints for both canonical and vibrant categorizations of the

180

dependent variables. For this analysis, a test for factor group interactions ensued, resulting in two pairs of variables with a significant interaction effect (i.e., following segment and grammatical category at p= 0.0169, and following segment and preceding segment at p= 0.022). As a result, each individual interacting factor group was correlated with trill variation and only the grammatical category factor group appear as a significant predictor for the model (p = 0.0196).

Subsequently, three linguistic variables (i.e., stress, grammatical category, and number of syllables) and two acoustic predictors (i.e., duration and F3) were submitted for analysis.

Table 5-24. Non-significant linguistic factors contributing to canonical trills in third generation informants. Application Value: Canonical Trill Random Intercept: Token and Informant Duration = p = 4.24e-11 Log Odds 47.489 Model Total = 286 DF = 4 Input Prob. = 0.012 Overall R2 0.653 Dev.: 250.39 AIC: 258.39

Results are presented in Table 5-24. The Varbrul model of this categorization only distinguished an acoustic cue contributing to a canonical trill, suggesting that longer rhotics are conducive to lingual contacts. When compared with the model of the first categorization, it can be seen that a distinction between vibrant and non-vibrant trills produces a more complex model that includes linguistic constraints that significantly predict a vibrant trill.

This is also visible when we test the non-significant factors that fail to be included in the first model. These non-significant factor groups are presented in Table 5-25. No factor group was selected significant in the model and none approached significance, except for number of syllables. These findings suggest that when both categorization of the independent variable are compared, the distinction between canonical and non-canonical trills fail to capture the complexity of the predictors involved in trill variation.

181

Table 5-25. Non-significant linguistic factors contributing to canonical trills in third generation informants. Application Value: Canonical Trill Random Intercept: Token and Informant Number of Syllables = (Not Significant) p = 0.0853 Factors Log Odds N Proportion Weight Three- 0.477 137 0.46 [0.617] Three 0.326 88 0.352 [0.581] Three+ -0.803 61 0.213 [0.309] Range 31 Grammatical Category = (Not Significant) p = 0.158 Factors Log Odds N Proportion Weight Noun 0.325 176 0.443 0.581 Other -0.325 110 0.264 0.419 Range 17 F3 = (Not Significant) p = 0.51 Log Odds 0.000513 Stress = (Not Significant) p = 0.796 Factors Log Odds N Proportion Weight Unstressed 0.0567 190 0.353 [0.514] Stressed -0.0567 96 0.417 [0.486] Range 3 Model Total = 286 DF = 8 Input Prob. = 0.089 Overall R2 = 0.59 Dev.: 282.614 AIC: 298.61

Not only vibrant and non-vibrant rhotics are conditioned by more linguistic variables, but they are also produced with different articulatory gestures, as approximant variants differ to vibrant rhotics in the absence of airstream blockade between the tip of the tongue and the alveolar region. For these reasons, and for the sake of comparison with the other generations, I will use the vibrant and non-vibrant categorization as the baseline for comparison with trills in

Continental Spanish in Section 6.2.5.

Finally, I compared the extralinguistic variables that contribute to tap and trill variation in the youngest generation of Raizales. A first test for factor group interactions in tap variation produced no significant result. Consequently, all extralinguistic predictors were included in the

Varbrul analysis.

182

Table 5-26. Significant extralinguistic factors contributing to canonical taps in third generation informants. Application Value: Canonical Tap Random Intercept: Token and Informant Job Marketplace = p = 0.0355 Factors Log Odds N Proportion Weight Employed 0.38 342 0.813 0.594 Informal -0.38 519 0.717 0.406 Range 19 Model Total = 861 DF = 4 Input Prob. = 0.83 Overall R2 = 0.267 Dev.: 914.489 AIC: 922.48

The model selected job marketplace as the only extralinguistic significant factor for tap variation. Table 5-26 summarizes these results. Informants with an employment favor a canonical realization than Raizales without a formal occupation. In most cases, informants would require instruction in technical or professional education in order to opt for formal employment in the few organizations of the islands, mainly belonging to the local government. An inspection of the dataset revealed that three of the informants have completed tertiary education in higher education institutions in Continental Colombia, while the rest of the informants plan to pursue a similar academic path. Professional education in Colombia requires at least 4 years of formal studies, and thus almost the same time living in major cities of the country, where Raizales are in contact with regional linguistic traits of Colombian Spanish. Vibrant production of a tap seems to be associated with education level and the corresponding time of dwelling in regional areas of

Colombia.

Additionally, trill production was also analyzed in terms of the extralinguistic constraints that condition variation. A pre-statistical test for group interaction was conducted and two factor group pairs were found to be significantly correlated. As a result, all these factor groups were individually correlated with trill production and only genre was found to be statistically significant (p = 0.0399), while sex (p = 0.187) and job marketplace (p = 0.0774) were excluded from the model.

183

Table 5-27. Significant extralinguistic factors contributing to vibrant trills in third generation informants. Application Value: Vibrant Trill Random Intercept: Token and Informant Genre = p = 0.0431 Factors Log Odds N Proportion Weight Task 0.611 77 0.909 0.648 Interview -0.611 209 0.703 0.352 Range 29 Island of Dwelling= (Not Significant) p = 0.461 Factors Log Odds N Proportion Weight San Andres 0.216 141 0.801 [0.554] Old Providence -0.216 145 0.717 [0.446] Range 11 Model Total = 286 DF = 5 Input Prob. 0.855 Overall R2 = 0.314 Dev.: 286.726 AIC: 296.72

Subsequently, the extralinguistic variables submitted for analysis consisted of genre and island of dwelling. Results of the Varbrul test are presented in Table 5-27. Recall that job marketplace was the only extralinguistic variable that reached statistical significance for tap variation. This time, genre appears as the only significant extralinguistic predictor for trill production.

Figure 5-17. Conditional inference tree of the significant social factors for trills in third generation Raizales.

184

This can be seen in Figure 5-17, where more vibrant trills were realized during the semi structured elicitation tasks, which suggests that informants were producing more careful speech during the narration and interaction tasks. During the interview, it seems that their speech was produced without attention to rhotic production.

Rhotics in third generation make use of a set of linguistic constraints similar to the previous two generations, while excluding acoustic predictors that were significant for older

Raizales, except for taps that use F3 frequencies to differentiate between tap variants. First, the effect of the phonological context and the position in the word is prevalent for both taps and trills. Canonical taps are favored in postvocalic contexts in word-medial and word-final positions, whereas the occurrence of vibrant trills is more likely with a preceding vowel.

Additional linguistic constraints also condition vibrant trills: vibrant variants are favored in trills appearing in three-syllable words. With respect to the extralinguistic factors that condition rhotic variation, the type of occupation and the method of data elicitation had an effect on tap and trill production, respectively. Informants with formal employment and tertiary education favor a canonical tap variant, while vibrant trills were favored in careful speech during linguistic tasks. I will now follow with the analysis of rhotics in the monolingual Spanish variety to determine whether rhotic variation in Continental Spanish correlates with a generation of Raizales.

5.2.4 Continental Spanish

The dataset for Continental Spanish is composed of 530 tokens divided in 395 taps

(74.5%) and 135 trills (25.5%) collected from 5 monolingual Spanish speakers, who were born in Coastal Colombia, but have lived in the Archipelago for the majority of their lives. This population represents the inhabitants of the islands that are more in contact with Raizales, due to the opening of San Andres as a commercial free-duty port in 1953 and to a high immigratory influx of the 1990 to the present.

185

In this section, I present the Varbrul analysis of tap and trill variation in this population with the purpose of comparing the effects of the significant linguistic and social predictors across generations of Raizales with Continental Spanish. With the aim of selecting the linguistic and acoustic predictors for the first model for taps, a test for interaction between all the linguistic predictors was conducted and no significant interaction effect was found. However, for this model F3, duration, and word position were submitted for analysis in order to compare this model with tap variation across Raizal generations. Table 5-28 summarizes these results.

Table 5-28. Significant linguistic factors contributing to canonical taps in Continental Spanish. Application Value: Canonical Taps Random Intercept: Token and Informant F3 = p = 0.0171 Log Odds 0.00106 Word Position = (Not Significant) p = 0.0853 Factors Log Odds N Proportion Weight Complex 0.44 108 0.75 [0.608] Intervocalic 0.251 201 0.711 [0.562] Final -0.34 40 0.6 [0.416] Medial -0.351 46 0.587 [0.413] Range 19 Duration = p = 0.663 Log Odds 3.875 Model Total = 395 DF = 7 Input Prob. = 0.097 Overall R2 = 0.06 Dev.: 472.373 AIC: 486.37

The model only selected F3 as a correlating with tap variation. Higher F3 frequencies increase the likelihood of a vibrant tap. While word position approached significance for tap variation, duration appear as a highly uncorrelated predictor. A random forest in Figure 5-18 further displays this pattern, where the F3 makes a distinction between taps produced with a frequency above 2621 Hz and other segments that were realized with a F3 below 2621 Hz. Only taps below this frequency value seem to be correlated with their position in the word. Taps in

186

complex onset are more frequently realized with a lingual occlusion when the F3 goes above

2419 Hz but below 2621 Hz.

Figure 5-18. Conditional inference tree of the significant linguistic factors for taps in Continental Spanish.

Similarly, taps in medial contexts appear more frequently in their canonical form compared to taps in word-final and intervocalic position. Consistently, higher F3 are significantly correlated with a vibrant variant.

Table 5-29. Non-significant linguistic factors contributing to canonical taps in Continental Spanish. Application Value: Canonical Tap Random Intercept: Token and Informant Stress = (Not Significant) p = 0.45 Factors Log Odds N Proportion Weight Unstressed 0.103 242 0.698 [0.526] Stressed -0.103 153 0.693 [0.474] Range 5 Number of Syllables = (Not Significant) p = 0.506 Factors Log Odds N Proportion Weight Two- 0.0824 196 0.699 [0.521] Three+ -0.0824 199 0.693 [0.479] Range 5 Grammatical Category = (Not Significant) p = 0.559 Factors Log Odds N Proportion Weight Lexical 0.0998 255 0.714 [0.525] Other -0.0998 140 0.664 [0.475] Range 5 Model Total = 395 DF = 6 Input Prob. = 0.717 Overall R2 = 0.145 Dev.: 479.09 AIC: 491.09

187

Although several factors failed to approach significance, they also shed light on the patterns of tap variation. Table 5-29 presents the non-significant predictors that were excluded from the first model. No factor group approached significance, and thus, no linguistic variable seems to predict a canonical variant, which indicates that the significant predictors on Table 5-28 are unambiguously set apart from these other variables.

In a separate analysis, trills were also examined in this population. As noted previously, trills comprise 25.5% of the dataset for this generation and includes 135 tokens. When the data set was further inspected, only 29 instances of zero-occlusion trills were found (21.5%). It was later observed that trills with multiple lingual closures outnumber approximant trills. The distribution of trills according to the number of apical occlusions is presented in Figure 5-19.

Figure 5-19. Distribution of trills according to the number of apical occlusions in Continental Spanish.

This distribution resembles the one found for third generation Raizales, where 24% of the trill dataset contained non-vibrant trills. However, there are only 18 trills with one lingual contact

(13%) as opposed to 110 tokens in third generation (38%). Trills with multiple occlusions reach

188

86 observations accounting for 64%. This asymmetrical distribution of trills based on the number of occlusions suggest that Continental speakers produce more frequently trill variants that are realized with more than one lingual contact. This is something that is expected from a monolingual variety.

Due to the low number of observations of approximant trills two separate regressions will be conducted: 1) a first analysis that distinguishes zero-occlusion trills, and 2) another one that compares non-canonical trills. In this first analysis, no interaction was found between factor group pairs, and thus, all the linguistic variables for vibrant trills (see Table 3-4), including the variables that were significant for all Raizal generations were submitted to the regression model, as follows: duration, F3, preceding segment, grammatical category, and number of syllables.

However, once the Varbrul program completed the analysis, no fixed predictor was selected as significant.

Consequently, the mixed effect model and a fixed effect analysis without random variables were compared, with the aim of determining whether these random factors were overtaxing the model due to the few number of tokens available in the dataset. It was found that the inclusion of the two random effects contribute to the relative quality of the model (AIC:

123.866) compared to a model with only fixed effects (AIC: 127.291). However, the inclusion of these random intercepts produced a model without significant factor groups.

As a result, Table 5-30 shows the Varbrul results with all the variables that failed to reach statistical significance for trill variation. The only variable closer to reach significance was grammatical category and the rest of the linguistic predictors appear less correlated with trill variation. Likewise, no acoustic predictor such as F3 and duration contributed to vibrant trill variation in a statically significant value.

189

Table 5-30. Mixed-effects model of the non-significant linguistic factors contributing to vibrant trills in Continental Spanish. Application Value: Vibrant Trill Random Intercept: Token and Informant Grammatical Category = p = 0.166 Factors Log Odds N Proportion Weight Lexical 0.563 111 0.784 [0.637] Other -0.563 22 0.773 [0.363] Range 27 Stress = p = 0.342 Factors Log Odds N Proportion Weight Stressed 0.389 57 0.912 [0.596] Unstressed -0.389 76 0.684 [0.404] Range 19 Duration = p = 0.363 Log Odds 9.163 F3 = p = 0.436 Log Odds -0.000797 Number of Syllables = p = 0.49 Factors Log Odds N Proportion Weight Two- 0.235 62 0.903 [0.558] Three+ -0.235 71 0.676 [0.442] Range 11 Following Segment = p = 0.492 Factors Log Odds N Proportion Weight Other 0.204 47 0.851 [0.551] Mid -0.204 86 0.744 [0.449] Range 11 Preceding Segment = p = 0.802 Factors Log Odds N Proportion Weight Vowel 0.0976 119 0.798 [0.524] Consonant -0.0976 14 0.643 [0.476] Range 5 Model Total = 133 DF = 10 Input Prob. = 0.927 Overall R2 = 0.35 Dev.: 106.074 AIC = 115.43

When the variables that conditioned trill variation in previous Raizal generations are submitted in a fixed effect model with no random variable, several predictors reach statistical significance. This time, F3 and number of syllables appear significant for trill variation. Results are shown in Table 5-31.

190

Table 5-31. Fixed effect model of the significant linguistic factors contributing to vibrant trills in Continental Spanish. Application Value: Vibrant Trill F3 = p = 5.09e-03 Log Odds -0.00224 Number of Syllables = p = 0.0471 Factors Log Odds N Proportion Weight Two- 0.557 62 0.903 0.636 Three+ -0.557 71 0.676 0.364 Range 27 Duration = p = 0.074 Log Odds 14.754 Preceding Segment = (Not significant) p = 0.278 Factors Log Odds N Proportion Weight Vowel 0.365 119 0.798 [0.59] Consonant -0.365 14 0.643 [0.41] Range 18 Grammatical Category = (Not significant) p = 0.44 Factors Log Odds N Proportion Weight Lexical 0.247 111 0.784 [0.561] Other -0.247 22 0.773 [0.439] Range 16 Model Total = 133 DF = 6 Input Prob. = 0.997 Overall R2 = 0.302 Dev.: 115.291 AIC = 127.29

Compared to the mixed-effects model in Table 5-31, the fixed effect model returns factor groups that have been significant for previous generations of Raizales with similar effects: larger

F3 frequencies, longer segments, and two syllable words or shorter favor a vibrant realization.

This time, the preceding segment or grammatical category were outside the significant statistical threshold.

When these variables were correlated with trill production in a conditional inference tree, only duration is displayed conditioning trill variation. Longer segments over 0.054ms appear highly frequent to be realized as a vibrant trill.

191

Figure 5-20. Conditional inference tree of the significant linguistic factors for vibrant trills in Continental Spanish.

Indeed, Figure 5-20 displays the role of duration and vibrant trills, despite only approaching statistical significance as seen in the model in Table 5-30 (p = 0.074). In a separate analysis, approximant trills were combined with segments produced with one lingual contact, in a category known as canonical trills. Based on a pre-statistical test for factor interaction, no factor group were excluded from the analysis. Subsequently, a Varbrul regression was conducted and one acoustic predictor was selected as significant predictor for a canonical trill production.

This model is presented in Table 5-32.

Table 5-32. Significant linguistic factors contributing to canonical trills in Continental Spanish. Application Value: Canonical Trill Random Intercept: Token and Informant Duration = p = 4.77e-04 Log Odds 30.245 Model Total = 133 DF = 4 Input Prob. = 0.119 Overall R2 = 0.624 AIC = 117.269

For this model, only one acoustic predictor was selected as significant (i.e., duration). As such, segmental duration distinguishes between trills produced with two or more lingual contacts and those segments that don’t reach a substantial articulatory gesture. As presented in Table 5-

33, the rest of the linguistic and acoustic predictors appear as non-significant for trill production.

192

Table 5-33. Non-significant linguistic factors contributing to canonical trills in Continental Spanish. Application Value: Canonical Trill Random Intercept: Token and Informant Number of Syllables = (Not Significant) p = 0.12 Factors Log Odds N Proportion Weight Two- 0.49 100 0.72 [0.62] Three+ -0.49 33 0.424 [0.38] Range 24 Stress = (Not Significant) p = 0.144 Factors Log Odds N Proportion Weight Stressed 0.497 57 0.842 [0.622] Unstressed -0.497 76 0.5 [0.378] Range 25 Preceding Segment = (Not Significant) p = 0.407 Factors Log Odds N Proportion Weight Vowel 0.363 119 0.672 [0.59] Consonant -0.363 14 0.429 [0.41] Range 18 F3 = (Not Significant) p = 0.444 Log Odds -0.000885 Grammatical Category = (Not Significant) p = 0.496 Factors Log Odds N Proportion Weight Noun 0.335 75 0.733 [0.583] Others 0.135 22 0.727 [0.534] Verb -0.47 36 0.417 [0.385] Range 20 Following Segment = (Not Significant) p = 0.603 Factors Log Odds N Proportion Weight Other 0.148 47 0.723 [0.537] Mid -0.148 86 0.605 [0.463] Range 7 Model Total = 133 DF = 10 Input Prob. = 0.89 Overall R2 = 0.615 Dev.: 110.294 AIC: 130.29

These findings tell us about to the nature of vibrant trills and the corresponding classification used for vibrant and canonical trills. On one hand, rather than using linguistic factors to differentiate a canonical from a non-canonical variant, trills variation is conditioned according to durational cues. On the other hand, approximant trills make use of F3, duration, and the number of syllables to predict trill variation. The use of different types of correlates for both categorizations (i.e., an acoustic cue for a canonical distinction and a linguistic variable for a

193

vibrant trill) and the fact that a zero-occlusion trill is shorter and far less frequent in this population compared to vibrant trills, which are produced with longer segmental duration and lingual contacts, suggest that non-vibrant trill correspond to a different phonetic category than vibrant trills. Rather than assuming a canonical distinction in Spanish trills, it seems that approximant rhotics are represented by different phonological features in the phonemic trill inventory, an observation also posited in previous research (Picard, 1997). Precisely, the properties of this allophonic trill variant are more related to the acoustic nature of the speech segment and the number of lingual closures. The evidence suggests that a [±] vibrant phonological feature could represent this variant more accurately in the phonological system of

Spanish. This panorama sets a baseline for our study and indicates that a trill categorization based on the number of lingual contacts rather than a canonical arbitrary distinction holds the linguistic differences of allophonic variants.

Now, I present the final analysis of the extralinguistic variables that condition tap and trill variation in Continental Spanish. These results shed light on the social factors that significantly predict rhotic production in the monolingual Spanish variety. First, a crosstabulation of the tap’s dataset revealed that data for females was categorically distributed across other social predictors, and thus no variation was present. When sex was correlated with tap production, it was found no significant effect with this predictor (p = 0.234). As a result, sex was excluded from the analysis.

Likewise, island of dwelling was also excluded, due to collecting data from only one informant living in Old Providence and to its non-significant predicting level (p = 0.0936). As a result, generation, education level, job marketplace, and genre were submitted for analysis. Results of the model are presented on Table 5-34. Generation and education level were selected as significant factors.

194

Table 5-34. Significant social factors contributing to canonical taps in Continental Spanish. Application Value: Canonical Tap Random Intercept: Token and Informant Generation = p = 7.24e-03 Factors Log Odds N Proportion Weight Second 0.38 150 0.753 0.594 Third -0.38 245 0.661 0.406 Range 19 Education Level = p = 0.032 Factors Log Odds N Proportion Weight Secondary 0.283 221 0.729 0.57 Tertiary -0.283 174 0.655 0.43 Range 14 Model Total = 395 DF = 5 Input Prob. = 0.74 Overall R2 = 0.17 Dev.: 471.61 AIC: 481.61

Mainly, second generation informants appear favoring a canonical tap variant, which suggest that older speakers tend to keep the tap vibration compared to younger speakers. In addition, informants with secondary education also appear to retain a variant with a lingual closure. This seems in juxtaposition to the what has been found in previous literature with regards to the role of higher educated speakers and prestigious variants (Moreno-Fernandez,

2008). Nonetheless, the data presented here includes informants who are young and have attended a tertiary education institution. Precisely, all informants in the third-generation group have received advanced education and they all happen to be females.

Finally, trill variation was also submitted for analysis. Again, sex and island of dwelling were excluded from the model and the rest of the social predictors were included in the Varbrul program. In addition, genre and job marketplace were also excluded from the model due to interacting significantly with education level. As a result, generation and education level were submitted for analysis. Table 5-35 shows that only education level was highly significant for predicting trill variation.

195

Table 5-35. Significant social factors contributing to vibrant trills in Continental Spanish. Application Value: Vibrant Trill Random Intercept: Token and Informant Education Level = p = 6.38e-04 Factors Log Odds N Proportion Weight Secondary 1.462 98 0.918 0.812 Tertiary -1.462 35 0.4 0.188 Range 63 Generation = p = 0.0678 Factors Log Odds N Proportion Weight Third 0.528 101 0.822 0.629 Second -0.528 32 0.656 0.371 Range 23 Model Total = 133 DF = 5 Input Prob. = 0.688 Overall R2 = 0.384

Informants with secondary education highly favor a vibrant trill production, while younger speakers also favor this variant. A closer inspection of the dataset shows that only three males appear with secondary education and two of them are young adults with stable jobs and finished high school. The oldest of these informants happens to live in Old Providence, attended school until sixth grade and works as a construction worker. In contrast, females make up the only population with tertiary education. When correlated individually with trill production, sex appear statistically significant for the trill production (p = 0.22e-03) and males highly favored a vibrant variant (Fw = 0.804) compared to females (Fw = 0.196).

These findings indicate that males highly favor a vibrant trill, while tertiary-educated females favor a non-vibrant realization. However, it should be noted again that these results might not portray an accurate trill variation pattern due to the distribution of the informants.

Results can be graphically seen in Figure 5-21, where education level and sex are identically correlated.

196

Figure 5-21. Conditional inference tree of the significant social factors for trills in Continental Spanish.

The effect of generation, education level, and sex in rhotic production seem to overlap in taps and trills. Overall, informants who attended secondary education favor a vibrant trill realization while second and third generation favor a vibrant tap and trill, respectively. The role of genre, island of dwelling or occupation was either disfavored for statistical significance or was highly correlated with other variables. These findings reflect on the nature of this dataset and the limitations of this analysis mainly due to the low number of tokens for trills and the uneven distribution of the data collection. Future studies should address these limitations and apply a more distributed data collection procedure.

However, the analysis of Continental Spanish rhotics has shed light on the variation patterns of taps and trills and it has allowed to produce a comparison between the bilingual and the monolingual Spanish variety for a subsequent section of this Chapter. Precisely, it was found that vibrant variants are realized depending on the phonological context and the number of syllables: for taps, a prevocalic context, and for trills, two-syllable words or shorter. Results showed that higher F3 frequencies and longer segments predict a vibrant segment. In addition, it was noted that a distinction between vibrant and non-vibrant trills represents more precisely the linguistic differences between these two types of trills, rather than producing an arbitrary categorization between canonical and non-canonical variants. Despite the asymmetrical

197

distribution across several extralinguistic factors, it was found that generation and education level significantly predict tap and trill production. In what follows, I will compare the bilingual

Spanish variety across generations of Raizales and Continental Spanish with the aim of determining the effect of contact and rhotic variation.

5.3 A Tale Of Distant Stories: Bilingual Raizal Spanish And Monolingual Continental Spanish Compared

Having examined the linguistic factors that predict tap and trill variation and the specific constraints that favor a vibrant realization by generation in bilingual Raizales and monolingual informants, it is time to compare the behavior of rhotic production in both Spanish varieties. For this comparison, a canonical tap will be equated with a vibrant tap. In its linguistic essence, the label ‘canonical’ relates to a tap with one lingual contact, a vibration of the tongue against the alveo-palatal region, while a non-canonical variant reflects an approximant tap. Since it was found that the number of lingual closures was a more appropriate categorization of the linguistic reality of this rhotics, the canonical distinction is no longer necessary.

Figure 5-22. A comparison of the proportions or vibrant and non-vibrant rhotics in in the Archipelago.

198

Figure 5-22 recapitulates the proportions of vibrant and non-vibrant variants found in each generation of Raizal Spanish and Continental Spanish. When the rate of production is compared across generations of bilingual Raizales, senior informants appear with proportions that contrast with the other population groups. For instance, the lowest proportions appear with vibrant realizations, and the higher rates of non-vibrant rhotics are produced by this generation.

Similarly, second generation speakers appear with intermediate rates between senior and younger

Raizales in both taps and trills. On the contrary, Continental Spanish and younger Raizales appear with similar mean values across all rhotic realizations, which suggests that, at least in terms of rates of production, the monolingual Spanish variety and the youngest Raizal informants differ very little.

Table 5-36. Variable rule analyses of the contribution of factors selected as significant to occurrence of vibrant taps in generations of bilingual Raizales and Continental Spanish. 1st Gen. input probability: <.001 (34%), N = 307/898 2nd Gen.: 0.00101 (61%), N = 470/772 3rd Gen.: 0.0108 (75%), N = 650/861 Continental: 0.0821 (70%), N = 275/395 Factor Group Relative Frequency Vibrant Probability 1st Gen 2nd Gen 3rd Gen Cont. 1st Gen 2nd Gen 3rd Gen Cont. Word Position Final 37 78 83 60 .63 .64 .63 [.41] ** Medial 21 71 88 59 .39 .58 .66 [.41] Intervocalic 46 63 75 71 .65 .45 .42 [.56] Complex 17 42 65 75 .32 .32 .29 [.60] Range 33 32 37 19 F3 Log odds .00294 .00246 .0025 .00106 Duration Log odds [-3.653] 43.684 [12.688] 3.875 1st Gen.: Overall R2 = 0.45; AIC 907.368 2nd Gen.: Overall R2 = 0.684; AIC 679.774 3rd Gen.: Overall R2 = 0.317; AIC 862.331 Continental: Overall R2 = 0.0604; AIC 486.373 ** Square brackets [ ] indicate that this effect does not achieve statistical significance but approaches significance at p = 0.0853.

199

The linguistic properties of this behavior are clearly detailed in Tables 5-36 and 5-37, where the Varbrul analyses for all taps and the corresponding factor’s direction of effect are contrasted. The first four columns display the proportional trend in favoring the vibrant tap. Taps appearing in intervocalic and complex-onset positions display increasing proportions from first generation to the monolingual variety. On the contrary, word-final and word-medial taps start being favored in second and third generation, but lack preference in Continental Spanish. The column for probability shows that vibrant taps are maintained in intervocalic position only in first generation and monolingual Spanish, while word-final positions are retained in all bilingual generations. Vibrant taps emerge in word-medial, preconsonantal contexts for second and third generation Raizales. However, Continental informants retain vibrant realizations in complex- onset syllables and disfavor final-word contexts. Of all the varieties under study, word position appears significant, except in Continental Spanish, where it reaches significance at p= 0.0853.

Statistically significant acoustic cues are also shared in all varieties. Except for duration in first generation, higher F3 frequencies and longer segments are correlated with a vibrant realization, particularly within bilingual Raizales.

Table 5-37. Factors' direction of effect by data set (vibrant taps). Factor Group 1st Gen 2nd Gen 3rd Gen Cont. Word Position Final    ⨂ Medial ⨂   ⨂ Intervocalic  ⨂ ⨂  Complex ⨂ ⨂ ⨂  F3 Log odds + + + + Duration Log odds - + + +  = tends to favor a Vibrant Tap more ⨂ = tends to favor a Vibrant Tap less + = a positive correlation with a Vibrant Tap (acoustic variable) - = a negative correlation with a Vibrant Tap (acoustic variable)

200

While the evidence suggest that same factor group is significant across varieties, the constraint hierarchy differs between models. Following Poplack and Tagliamonte (in Aaron,

2006), these differences between groups and the factor’s direction of effect are graphically represented in Table 5-37. As noted previously, word-final and positions are linked to a vibrant tap across generations of Raizales. Word-medial contexts emerge in the second generation and is maintained in younger Raizales. Intervocalic vibrant taps are retained in senior informants but disfavored in subsequent generations. Taps in syllable-initial consonant clusters (i.e., Complex) are relevant for vibrant variants in Continental Spanish. Intervocalic positions are the only context that is shared between Continental Spanish and bilingual Raizales in the first-generation subgroup. The mostly stable nature of tap variation in Raizal Spanish suggest that the use of a vibrant realization is the result of generational continuity arisen through community-internal transmission. As a result, the observation that intervocalic position was favored in first generation but displaced by word-medial contexts in the second and third shows a restructuring of factor constraints from the same language-internal mechanism that started in second generation and has finalized in the youngest generation.

On the contrary, Continental Spanish is patterning in a different way, preferring intervocalic and complex contexts. This leads to the conclusion that tap variation in both Spanish varieties are changing independently with a sharp contrast on the contexts where vibrant taps are produced. While the bilingual variety emphasizes vibrant realizations in post-vocalic contexts, monolingual Spanish disfavor these contexts for a non-vibrant production or even a non-rhotic production. In fact, it was found that non-rhoticity accounts for 23% of the whole Continental

Spanish dataset, and thus, it is not an uncommon feature of the Costeño Spanish dialect.

Precisely, the extralinguistic variables that condition tap variation were also compared across

201

generations of Raizales and Continental Spanish in Table 5-38. The island of dwelling appears significant in bilingual Spanish but with a different constraint ranking: vibrant taps are preferred in Old Providence for first and second generation but disfavored in third generation.

Table 5-38. Variable rule analyses of the contribution of extralinguistic factors selected as significant favoring the occurrence of vibrant taps in generations of bilingual Raizales and Continental Spanish. 1st Gen. input probability: 0.287 (34%), N = 307/898 2nd Gen.: 0.628 (61%), N = 470/772 3rd Gen.: 0.834 (75%), N = 650/861 Continental: 0.741 (70%), N = 275/395 Factor Group Relative Frequency Vibrant Probability 1st Gen 2nd Gen 3rd Gen Cont. 1st Gen 2nd Gen 3rd Gen C ont. Island of Dwelling Old Providence 45 66 72 80 .62 .55 .45 [.59] San Andres 38 55 79 68 .38 .45 .55 [.41] Range 25 10 10 18 Sex Female 48 73 78 65 .67 .79 [.54]* [45] Male 19 43 73 73 .32 .20 [.46] [55] Range 35 59 37 10 Job Marketplace Employed 41 70 81 73 [56] [68] 59 [55] Informal 32 53 72 67 [44] [31] 41 [45] Range 12 37 18 10 Genre Interview 34 58 74 70 [53] 39 [44] [52] Task 28 81 87 69 [47] 61 [56] [48] Range 6 22 12 4 1st Gen.: Overall R2 = 0.363; AIC 1019.34 2nd Gen.: Overall R2 = 0.634; AIC 758.133 3rd Gen.: Overall R2 = 0.267; AIC 922.489 Continental: Overall R2 = 0.171; AIC 481.61 * Square brackets [ ] indicate that this effect does not achieve statistical significance.

The island of dwelling appears significant in bilingual Spanish but with a different constraint ranking: vibrant taps are preferred in Old Providence for first and second generation but disfavored in third generation. Tap production also seemed to be significantly correlated with females in first and second generation. However, note that when the ranking order changed in the

202

first factor group, the role of females in tap production stopped being significant in third generation. Tap variation is significantly conditioned by the type of occupation in third- generation informants and the genre of the data collection procedure became significant in second generation alone.

Table 5-39. Factors' direction of effect by data set (social variables for taps). Factor Group 1st Gen 2nd Gen 3rd Gen Cont. Island of Dwelling Old Providence   ⨂  San Andres ⨂ ⨂  ⨂ Sex Female    ⨂ Male ⨂ ⨂ ⨂  Job Marketplace Employed     Informal ⨂ ⨂ ⨂ ⨂ Genre Interview  ⨂ ⨂  Task ⨂   ⨂

 = tends to favor a Vibrant Tap more ⨂ = tends to favor a Vibrant Tap less

Table 5-39 visually summarizes the relative strength of social factors and presents the patterns of the effects within factor groups. With some exceptions third generation and

Continental Spanish, we see vibrant taps consistently being favored in Old Providence, by females and by informants with formal employment. While second and third generation favored a tap vibrant during the task procedure, the interview favored lingual closures in first generation and Continental Spanish. As vibrant realizations are correlated to Old Providence, females and formally employed Raizales, most of whom tend to have tertiary education, lead the increasing use of vibrant taps in the Archipelago.

Further evidence supporting the observation of independent patterns of rhotic variation is found in trill production and displayed in Table 5-40. No significant factor group was shared

203

between the two Spanish varieties. The role of the preceding phonological environment appears stable in bilingual Raizales and remains significant across all generations while grammatical category only became significant in first generation.

Table 5-40. Variable rule analyses of the contribution of factors selected as significant to occurrence of vibrant trills in generations of bilingual Raizales and Continental Spanish. 1st Gen. input probability: <.001 (42%), N = 103/246 2nd Gen.: 0.0125 (61%), N = 204/335 3rd Gen.: 0.716 (76%), N = 217/286 Continental: 0.997 (78%), N = 104/133 Factor Group Relative Frequency Vibrant Probability 1st Gen 2nd Gen 3rd Gen Cont. 1st Gen 2nd Gen 3rd Gen Cont. Preceding Segment Vowel 47 66 81 80 .77 .75 .69 [.59]* Consonant 14 36 53 64 .23 .25 .31 [.41] Range 54 50 38 18 Grammatical Category Noun 46 70 77 78 .62 [.55] [.54] [.56] Other 35 59 73 77 .38 [.45] [.46] [.44] Range 24 11 9 12 Number of Syllables Three+ 42 56 78 67 [.51] [.46] [.54] .36 Two- 41 65 73 90 [.49] [.54] [.46] .64 Range 2 12 8 28 F3 Log odds .00252 .0013 [.00111] -.00224 Duration Log odds [-15.127] 20.639 [-4.089] 14.754 1st Gen.: Overall R2 = 0.638; AIC 243.973 2nd Gen.: Overall R2 = 0.789; AIC 217.78 3rd Gen.: Overall R2 = 0.35; AIC 284.334 Continental: Overall R2 = 0.302; AIC 127.291 * Square brackets [ ] indicate that this effect does not achieve statistical significance.

Despite this, the ordering of the factors is the same across all generations and varieties in these factor groups: vibrant trills are favored with preceding vowels and nouns. The number of syllables is statistically significant for Continental Spanish and presents a different ordering of factors compared to bilingual Raizales. Likewise, acoustic cues appear unstable across datasets.

On the one hand, higher F3 frequencies signal a vibrant trill until third generation, where it stops

204

being significant, and in Continental Spanish it is negatively correlated with a vibrant realization.

On the other hand, duration also presents a significant predicting value for trill variation but only for second generation and Continental Spanish. In first and third generation, duration fails to reach statistical significance and appears negatively correlated. Table 5-41 shows a visual representation of the direction of effect in trill variation in all datasets, regardless of statistical significance.

Table 5-41. Factors' direction of effect by data set (Vibrant trills). Factor Group 1st Gen 2nd Gen 3rd Gen Cont. Preceding Segment Vowel     Consonant ⨂ ⨂ ⨂ ⨂ Grammatical Category Noun     Other ⨂ ⨂ ⨂ ⨂ Number of Syllables Three+  ⨂  ⨂ Two- ⨂  ⨂  F3 Log odds + + + - Duration Log odds - + - +  = tends to favor a Vibrant Trill more ⨂ = tends to favor a Vibrant Trill less + = a positive correlation with a Vibrant Trill (acoustic variable) - = a negative correlation with a Vibrant Trill (acoustic variable)

As noted previously, no variation in the ordering of factors is found, except in the number of syllables. This unbalanced parallel of significant linguistic predictors and the consistent ordering of factors in preceding segment and grammatical category (except for number of syllables) suggest that trill variation is constrained independently in Raizal and Continental

Spanish. Due to the more unstable behavior in terms of the relative strength of factors, trill variation appears different compared to tap production.

205

Similarly, tap variation also seem conditioned by extralinguistic factors in certain Raizal generations. Tables 5-42 displays the results of the Varbrul analysis for all datasets. It was observed that sex and the data collection procedure had an effect in trill variation for certain generation of Raizales.

Table 5-42. Variable rule analyses of the contribution of factors selected as significant to occurrence of vibrant trills in generations of bilingual Raizales and Continental Spanish. 1st Gen. input probability: <.001 (42%), N = 103/246 2nd Gen.: 0.0125 (61%), N = 204/335 3rd Gen.: 0.716 (76%), N = 217/286 Continental: 0.997 (78%), N = 104/133 Factor Group Relative Frequency Vibrant Probability 1st Gen 2nd Gen 3rd Gen Cont. 1st Gen 2nd Gen 3rd Gen Cont. Sex Female 61 82 78 40 .79 .90 [.55]* [.20] Male 24 33 73 92 .21 .10 [.45] [.80] Range 57 80 10 60 Genre Task 43 70 90 93 [.60] [.61] .65 [.59] Interview 41 53 70 58 [.40] [.39] .35 [.41] Range 20 22 30 18 1st Gen.: Overall R2 = 0.509; AIC 262.887 2nd Gen.: Overall R2 = 0.786; AIC 235.429 3rd Gen.: Overall R2 = 0.314; AIC 296.726 Continental: Overall R2 = 0.384; AIC 109.193 * Square brackets [ ] indicate that this effect does not achieve statistical significance.

When these results are graphically summarized in Table 5-43 regardless of the significance level, it can be seen that females are leading the change for vibrant trill across generations of Raizales. The pattern is different in Continental Spanish, as it is males who prefer a vibrant variant. Since the effect failed to reach statistical significance and samples from

Continental Spanish was collected without regards to any age group, the effects based on the sex of the informants are tentative at least. For all groups, a vibrant trill was favored during the

206

narration task, which suggest that more effort was placed to produce the lingual closures during semi-directed speech.

Table 5-43. Factors' direction of effect by data set (Social variables for vibrant trills). Factor Group 1st Gen 2nd Gen 3rd Gen Cont. Sex Female    ⨂ Male ⨂ ⨂ ⨂  Genre Task     Interview ⨂ ⨂ ⨂ ⨂

 = tends to favor a Vibrant Tap more ⨂ = tends to favor a Vibrant Tap less

In this section, we have conducted a cross-variety comparison between generations of

Raizales and Continental Spanish to uncover common patterns of rhotic variation between the two Spanish varieties, and thus, determine a potential contact-induced change. Although the frequency of use of vibrant and non-vibrant realizations in younger generations increasingly resemble those presented in the monolingual variety, the behavior of rhotic variation is different in both Spanish varieties. While the same factor group is significant in all tap datasets, the ordering of the factors is different in Raizal Spanish and Continental Spanish. A similar case is found in trills, as it was observed a homogenous pattern of phonological variation across all

Raizal age groups, while Continental Spanish displays the same constraint hierarchy when statistical significance is disregarded. These results point to the independent nature of the varieties of Spanish spoken in the Archipelago. First, Raizal Spanish displays a generational continuity where a restructuring of the constraint ordering starts in second generation and is completed in younger Raizales. On the contrary, Continental Spanish behaves differently in terms of the systematic linguistic conditionings and the evidence suggest that rhotic variation has changed internally within both varieties. As a result, we dismiss the hypothesis of contact-

207

induced change in rhotic variation in Raizal Spanish. We will discuss these findings in detail in the next Chapter.

208

CHAPTER 6 DISCUSSION

As the analysis of the data has furnished results on rhotic variation and change in the

Archipelago of San Andres, this chapter will examine the implications of these findings in light of the research questions that guided this study. First, I review the results of the sociophonetic analysis of non-vibrant rhotics and the phonetic association of generations of Raizal Spanish with the languages in contact. Then, I discuss the findings of the comparative variationist analysis of vibrant rhotics and its consequences for testing contact-derived change.

6.1 Variation And Change In Non-Vibrant Rhotic Production

Chapter 4 presented the quantitative examination of non-vibrant rhotic production in the

Archipelago of San Andres. In doing so, the acoustic correlates that characterize zero-occlusion rhotics in the languages under study were examined and the best predictors were selected for testing a potential correlation across generations of Raizal Spanish and either Raizal Creole or

Continental Spanish. The first part of the analysis consisted of comparing the overall acoustic properties of non-vibrant rhotics between all the varieties in the islands. First, it was observed that these variants appear frequently across all datasets when compared with non-rhoticity and the occurrence of vibrant rhotics. Almost two thirds of the tokens in the Raizal Creole dataset

(62.7%) are fully produced variants, while non-rhoticity is present in the rest of the dataset. Non- vibrant rhotics in Raizal Spanish account for 36.9%, while the same variant appears in 21.7% of the Continental Spanish data. Not only the distribution of non-vibrant rhotics in Raizal Spanish appeared in an intermediate position between Raizal Creole and Continental Spanish, but also the acoustic values examined mostly seemed to lie between the two languages in contact.

Several acoustic correlates were examined to determine the nature of these rhotic variants, such as duration, formant frequencies (F2, F3, F4, and F5), and spectral moments,

209

including COG, skewness, and kurtosis. Precisely, mean duration values in Raizal Spanish always appeared greater than Continental Spanish, but far lower than Raizal Creole. These values were also confirmed when compared in more specific contexts, such as word stress and word position. Formant frequencies also revealed mean F2 and F3 values that remain between the same measurements for Raizal Creole and Continental Spanish. The same pattern was also found when the distance between F3-F2 was examined. As expected in English-like rhotics, formant frequencies in Raizal Creole always appeared with the lowest values across all varieties.

Overall, spectral moments showed great variability within varieties, but the measurements fell within a range that differs from values found for English sibilants and Spanish fricative trills found in previous studies (Colantoni, 2006; Jongman et al, 2000). Since non-vibrant variants were on average produced with acoustic properties that diverged from assibilated sounds, it was determined that the non-vibrant rhotics of the Archipelago are realized as approximant variants.

Moreover, the differences in segmental duration and formant frequencies between languages and the fact that Raizal Spanish gravitates between both Raizal Creole and Continental

Spanish suggest a place of articulation that diverge between varieties. A more backed realization is found in Raizal Creole and a more fronted production is realized in Continental Spanish. As a result, non-vibrant rhotics in Continental Spanish were classified as an alveolar approximant, while Raizal Creole makes use of a post-alveolar approximant. At this point, the place of articulation of approximant rhotics in Raizal Spanish vary between both alveolar and post- alveolar variants. This first finding responds partially to the first research question posited for this study, as it has determined the acoustic correlates that characterize non-vibrant rhotics in the varieties of the Archipelago.

210

Due to the number of acoustic predictors used for analyzing the properties of approximant rhotics in these varieties, it became necessary to discern the best correlates for determining the association of a generation of Raizal Spanish with the languages in contact, either Creole or Spanish. Consequently, a discriminant function analysis was conducted revealing that duration values and formant frequencies best predicted the differences between varieties. This finding directly answers the second part of our first research question on the acoustic correlates that best predict discrimination between varieties. Precisely, the acoustic measurement with the greatest discrimination capabilities involved the segmental duration. Due to the fact that duration best predicts crosslinguistic differences between rhotics, it was further decided to focus on the analysis of a phenomenon known to exist in bilingual and monolingual

Spanish settings, the neutralization or maintenance of the tap/trill contrast in non-vibrant rhotics

(Balam, 2013; Henriksen, 2015; Bradley and Willis, 2012). While it was observed that

Continental Spanish maintain the tap/trill contrast by means of segmental duration, the bilingual

Raizal variety exhibited the erosion of the durational differences between taps and trills across generations. Results showed that neutralization takes place across word stress in both taps and trills and that the effect of this phenomenon is most prevalent in first generation. The neutralization pattern is also present in second generation informants, but almost non-existent in younger, third generation Raizales. This is the first result that indicated a diachronic change in progress in the Archipelago of San Andres, as some bilingual taps and trills conditioned by word stress remain indistinguishable in terms of segmental duration in senior Raizales and second- generation informants. On the contrary, the tap/trill contrast is maintained in younger generations. Such finding presents evidence to partially respond the last part of our first research question on non-vibrant rhotic variation and change in the Archipelago.

211

The final analysis consisted of contrasting the second predictors found to best discriminate between linguistic varieties, formant frequencies. The aim of this analysis was to finally determine the place of articulation of Raizal Spanish rhotics and statistically measure an association of their formant frequencies with the contact languages. For this comparison, it was determined to contrast F3 and the distance between F3-F2, as higher resonances of the vocal tract

(i.e., F4 and F5) have been used to determine approximant variants in English rhotics (Zhou et al,

2008) but not Spanish. Correspondingly, Spanish taps and trills were compared in terms of these formant measurements with Raizal Creole and it was found that F3 frequencies in trills in senior generations barely approximated the statistically significant threshold at p = 0.05. In addition, when compared with Old Providence a slightly higher non-significant effect was found in trills in

San Andres island, as F3 values failed to reject the null hypothesis by reaching non-significant statistical differences with Raizal Creole approximants (p = 0.12). While these results show that significant differences in terms of F3 frequencies with Raizal Creole vary across generations of

Raizales, they also show that younger generations presented similar p-values with Continental

Spanish, which in turn, greatly diverged from senior informants. However, the greatest non- significant effect was found when the distance of F3-F2 was contrasted in taps and trills between generations of Raizal Spanish and Raizal Creole. Taps in both islands failed to reach statistical differences with Raizal Creole between almost all generations, which challenged the assumption that only approximant taps in older generations are approaching the F3-F2 distance frequencies of Raizal Creole. While it is possible that, in fact, taps failed to reach statistical significance regardless of the generations of Raizal Spanish, it seems the case that F2 is increasing rather than

F3 decreasing in younger generations due to the high proportion of front vowels that surround taps in the dataset for third generation. Further evidence to this point comes when comparing the

212

overall F3-F2 distance without contrasting between islands, as third generation Raizales failed to reject the hypothesis of null differences at a statistically significant level of p =0.026 (Table 4-

16). On the contrary, trills show more straightforward differences between older/younger informants, since it seems that this variant is increasingly showing non-statistically significant differences in older generations of Old Providence informants at p-values that exceed p = 0.50.

For instance, the intermediate second generation exhibits non-significant differences much lower than first generation (p=0.18 and p=0.55, respectively) (see Table 4-19). While this generational effect is not present in San Andres, significant differences in F3-F2 distance frequencies in senior Raizales are virtually non-existent with p-values of p=0.64).

All these findings have shown that formant frequencies in taps and trills across generations of Raizales are associated with either Raizal Creole or Continental Spanish, suggesting that both taps and trills in older generations appear to be realized with a place of articulation that increasingly converges in the direction of Creole post-alveolar approximants.

Likewise, the contrary can also be said about younger generations producing bilingual Spanish rhotics, as these formants are diverging from the senior standard and converging in the direction of Continental Spanish alveolar approximants. With these results on the generational differences of formant frequencies, the evidence suggests that there is a change in progress in the form of the place of articulation of Raizal Spanish approximants. Moreover, it is further corroborated that formant frequencies in approximant taps and trills realized in younger generations are closely associated with Continental Spanish. The comparison of non-significant p-values makes a clear case for the association of a generation of Raizal Spanish with one of the languages in contact, which directly answers the last part of the first research question posited for this study on the change in progress and the subsequent convergence of a generation with either Raizal Creole or

213

Continental Spanish. Finally, there are mixed results in terms of which island is leading a generational change, as San Andres seems to be leading the change in terms of F3 in trills and

F3-F2 in taps, but Old Providence presents a gradual generational change of the F3-F2 distance in trills.

The ongoing change of non-vibrant rhotics of the Archipelago resonates with the framework discussed on contact situations where a Creole language is in direct contact with a national non-lexifier language (Snow, 2000; Aceto, 1999; DeCamp, 1971). If we apply the explanation of a Creole continua to the processes of change in this bilingual variety in a contact scenario, we would have to attribute any potential linguistic influence on the languages in contact. In this study, the case of non-vibrant rhotics presents evidence against a Creole-like continuum of Raizal Spanish rhotics, as these segments do not present gradual changes or steady behavior in the direction of Raizal Creole, but rather are becoming more similar to the national non-lexifier language in contact. If the concept of decreolization implies a Creole continuum in which only a Creole basilect and the lexifier acrolect exist in the two extremes of the spectrum, then clearly there is a different situation reflected in the production of Raizal Spanish rhotics, which suggest an influence in some nature by the existence of the non-lexifier language. First, approximant rhotics in younger generations are converging toward Continental Spanish, and thus, diverging from the English-like acrolect. Secondly, these findings suggest that transfer of phonetic features from the source language to the recipient bilingual Spanish is in place, which seems to indicate that the change has been partially produced due to the influence of the languages in contact. A possible influence from the contact languages is further supported by the phonological interference of the neutralization and maintenance of the tap/trill contrast across generations of Raizal Spanish. The maintenance of the tap/trill contrast by younger Raizales

214

seems to map the same phonological categorization present in the monolingual Spanish variety.

In contrast, senior Raizales present a case were erosion of the tap/trill contrast is taking place due to the unstable phonological categorization of Spanish rhotics. Precisely, the discriminant function analysis conducted in the datasets for each variety selected duration and formant frequencies as the specific acoustic predictors that best classified the languages of the

Archipelago. However, the analysis also showed less predicting capabilities in the bilingual

Spanish variety, which suggest a greater variation effect in the unstable phonology of Raizal

Spanish.

The findings of these tests have shown that a change in progress is taking place in the approximant rhotics of the bilingual Spanish variety spoken by generations of Raizales.

Although, there is evidence in the form of the tap/trill contrast and the correlation of formant frequencies that suggest the ascription to one of the contact languages, this partially explains the reason behind the mechanisms of linguistic change in the rhotics of the Archipelago. Rather, it seems that the maintenance and neutralization of the tap/trill contrast is a sign of balanced bilingualism in the younger Raizal Population. Moreover, attributing a change in progress completely to the contact languages would imply that a post-alveolar approximant would have never been arisen in monolingual situations. However, it has been documented dialectal variation of Spanish rhotics in monolingual varieties that differ from the apico-alveolar trill variants. For instance, several studies have identified a post-alveolar approximant rhotic in place of an apico- alveolar trill in Costa Rican Spanish (Lipski, 2011), an assibilated variant in Central Valley

Costa Rican Spanish (Adams, 2002) and in Highland Ecuadorian Spanish (Bradley, 2004, 1999), and a velarized production in Puerto Rican Spanish (Campos-Astorkiza, 2012). The phenomenon of rhotic variation is not exclusive to monolingual scenarios. Retroflex approximants have been

215

reported in bilingual varieties of Spanish in direct contact with English-based Creoles (Balam,

2012; Zimmer, 2011; Hagerty, 1979), coexisting with US English (Ramos-Pellicia, 2007; Lastra de Suarez, 1975; Sanchez, 1973; Cassano, 1977), and with indigenous languages in the Yucatan peninsula (Lope-Blanch, 1975).

Instead of the prevalence of an innovative variant, the bilingual approximant rhotics of the Archipelago of San Andres presents a generational change in frequency toward a normative alveolar monolingual approximant. The process is accelerated among second and third generation Raizales who seem to be more balanced bilinguals educated in the Spanish-only education system of the islands and Continental Colombia. Precisely, nine third-generation informants recruited for this study have finished high school and three have completed undergraduate education in Colombia. Moreover, two young Raizales have expressed their desires to pursue tertiary education in Medellin, Colombia. All of them have received early education in Spanish during childhood in preschool. In contrast, senior Raizales reported to have been exposed to Spanish once they entered primary school, as their Creole was the only language spoken in the household. Some have even learned Spanish during adolescence, since they were enrolled in private Baptist primary schools when the language of instruction was English. No informant reported to have pursued tertiary education outside the islands and three did not finished high school. The contrast in the acquisition process of Spanish across generations is reflected on the realization of an approximant similar to the monolingual variant in a context where there is normative pressure on the bilingual variety due to contact with Continental

Spanish. The apparent rhotic changes in monolingual and bilingual varieties toward an assibilated (Adams, 2002) or a backed innovative variant (Campos-Astorkiza, 2012; Balam,

2012; Zimmer, 2011; Lipski, 2011; Ramos-Pellicia, 2007; Lastra de Suarez, 2006; Bradley,

216

2004, 1999; Hagerty, 1979; Cassano, 1977; Lope-Blanch, 1975; Sanchez, 1973) is reversed in

Raizal Spanish and accelerated among second and third generation speakers due to stable bilingualism and the pressure to produce an apico-alveolar approximant as close as possible to that of the prestigious Spanish norm. Similar to other phonological variables in bilingual varieties of Spanish in contact with English (Lynch, 2009), normative variants seem to be retained due to ascription to a social group and issues of personal identity, while morphosyntactic phenomena in contact situations seem to be less permeable to external social factors, as ongoing language change has been suggested to be accelerated due to the absence of normative pressure of the language and more restricted contexts of use, particularly, in the US

(Silva-Corvalán, 1994). Precisely, external social concerns play a crucial role in the nature of rhotic variation in the Archipelago of San Andres, as it seems that younger Raizal generations mark their speech with phonological variables more in line with the prestigious national language. Further evidence of this ‘reverse’ language change is presented in the variationist comparative analysis of vibrant rhotics.

6.2 Variation And Change In Vibrant Tap And Trill Production

The second question that guided this work concerns the nature in the production of vibrant rhotics in the Spanish varieties of the Archipelago. This analysis complements the findings in the previous Chapter on non-vibrant rhotics and further our understanding of the mechanisms involved in the processes of rhotic change. This comparison analysis involved the datasets of vibrant taps and trills produced in Raizal Spanish and Continental Spanish with the aim of quantitatively assess the structural changes across generations of Raizales and corroborate the assumption of change derived by contact with Continental Spanish. The analysis was further divided into two comparisons. First, I conducted three independent variable rule analysis

(Varbrul) separated by datasets of taps and trills across generations of Raizal Spanish. This was

217

complemented with a final Varbrul model of tap/trill production in Continental Spanish. Lastly, I compared the relationship between varieties by contrasting the factor groups in the statistical models produced in the previous Varbrul analyses. The cross-variety comparison (i.e., comparative method) aimed at assessing the similarities and differences in the phonological patterns of taps and trills in Raizal Spanish and Continental Spanish, and thus, determine whether there is a correspondence between Spanish varieties in vibrant rhotic production in the

Archipelago (Tagliamonte, 2012).

Before the Varbrul analyses could take place, it was observed that a canonical and non- canonical distinction in the bilingual tap/trill production in Raizal Spanish presented several issues for the analysis. Canonical taps comprise variants with one apical occlusion, while trills require two or more lingual closures. As opposed to a similar distinction in previous studies

(Lopez-Alonzo, 2016; Henriksen, 2014; Zahler and Daidone, 2014; Weissglass, 2014), the frequencies of trill variants with two or more apical occlusion in senior Raizales comprised very few instances (n = 13), which posed a problem for the analysis due to the low count not conducive for statistical significance. Further tests using the canonical/non-canonical categorization with the other generations of Raizales also failed to capture the complexity of the predictors involved in trill variation. As a result, it was determined that trills will be compared in terms of the absence or presence of tongue vibration with the alveolar ridge rather than normative counts of lingual closures. Such categorization served as the application value for our

Varbrul models and it seems to represent a natural phonemic feature of rhotics (even across languages of the world) than the arbitrary normative classification of Spanish canonical and non- canonical trills (Picard, 1997; Ladefoged and Maddieson, 1996). When compared with second and third generation, trills with two or more lingual closures accounted for 15.6% and 38%,

218

respectively, while variants with one apical occlusion appeared with a proportion of 45.2% in second generation and 38% in third generation. The unbalanced proportion between these two generations suggests that rhotic trilling is a phonological feature that seems to be increasing across generations of Raizales and associated with younger and more balanced bilingual speakers. The frequency pattern found in young Raizales resembles the trends for multiple trilling in Continental Spanish, which accounts for 64% of the whole dataset, while variants with one apical occlusion reaches 13%. The distributional frequencies seem to be increased by greater levels of bilingualism in younger speakers, who appear to be acquiring the vibrant phonological feature in comparable proportions to the monolingual variety.

Once the distribution of taps and trills was documented and the vibrant categorization was adopted, the tap and trill datasets were analyzed separately across generations of Raizales.

The quantitative models for taps predicted vibrant taps to be favored by the same factor group

(i.e., Word Position) and acoustic predictors (i.e., Duration and F3) across all generations.

Moreover, the ordering of the factor seems to be consistent in all datasets, as word-final and word-medial position favor a vibrant tap, except for intervocalic position in first generation.

When compared with Continental Spanish, word position failed to become significant for vibrant taps and intervocalic and complex-onset contexts favored the production of this variant, the exact opposite of Raizal Spanish (see Table 5-36). In fact, this dataset only displayed the word position factor group approaching significance for tap production at p = 0.0853. The extralinguistic factors also exhibited two factor groups that appeared statistically significant in Raizal Spanish:

Island of Dwelling and Sex. Likewise, females and speakers in Old Providence seem to favor a vibrant tap across all generations, except for third generation informants in San Andres. Again, no correspondence was found between extralinguistic factor groups in Continental Spanish. The

219

statistical significance and the ordering of the same factor groups between all generation of

Raizales indicates a generational continuity that became restructured in second generation when all factors had the same relative strength and that have arisen through community-internal transmission. In contrast, the lack of correspondence in terms of the distributional frequency, the statistical significance, and ranking of the factor groups between Raizal Spanish and the contact monolingual Spanish variety suggest that both phonological structures have developed independently from one another. In sum, these findings suggest that the systematic pattern of the variable use of vibrant taps have arisen independently in these Spanish dialects. Although this finding contrasts with the move toward the monolingual variety in terms of formant frequencies in younger generations, the acoustic properties of approximant rhotics are still in line with the fronted realization of an alveolar Spanish /r/. In other words, both non-vibrant and vibrant rhotics use acoustic correlates to resemble more a prototypical Spanish rhotic, even though the linguistic behavior of vibrant taps and trills is different than that of the contact monolingual variety.

Furthermore, this also implies that that the acoustic correlates and linguistic conditioning of

Spanish rhotics overlap regardless of the linguistic variety.

The case of trill production is by no means different. Only preceding segment was found to be statistically significant between all generations, while grammatical category appeared significant for first generation alone. The hierarchy of the constraints showed that prevocalic sounds favor a vibrant production across all generations. In terms of the statistical significance of constraints between varieties, only the number of syllables was significant in Continental

Spanish. When the direction of effect was compared in Table 5-41, Raizal Spanish and the contact variety exhibit the same ordering of factors, favoring preceding vowels and nouns for trill production. Although the same factors were ranked equally, except for number of syllables,

220

the phonological context and grammatical category seem to be indicative of universal mechanisms of trills production. As the lack of correspondence in terms of statistical significance, the ranking of constraints, and the relative strength of factors between the bilingual and the monolingual variety seem to suggest, there is no indication of a systematic organization inherited from a common source, and thus, no shared phonological structure is present in the datasets. Rather than an internal parallel, it seems that these varieties are influenced by pervasive universal mechanisms within the language. Although only significant for first and second generation, extralinguistic factors point to the generational transmission of constraints by women between all Raizal generations, while men seem to favor the vibrant trill production in

Continental Spanish. Only the influence of careful speech was found in trill production during tasks, but this effect failed to reach statistical significance across all datasets.

In sum, the evidence points to a lack of shared significant constraints and the interplay of internal linguistic mechanisms in tap/trill production and respond to the research question dealing with speech variation and change in the vibrant rhotics of Raizal Spanish. The systematic variation constrained by the linguistic factors of word position in taps and the preceding segment in trills seemed to be retained across generations of Raizales but deviates from the pattern found in the contact variety. Although the distributional frequency of vibrant rhotics is converging toward the monolingual variety, the nature of tap and trill variation diverges from the behavior of taps and trills in monolingual Spanish, suggesting an implausible connection between these

Spanish varieties. Supporting this interpretation comes from the stable constraint hierarchy found in second generation and its transmission to third generation. I interpret the findings in terms of the diachronic retention of linguistic factors and the lack of correspondence with Continental

Spanish as evidence discounting the assumption of a potential contact-derived change.

221

The question remains as to how the systematic patterns of vibrant tap/trill production arose in the Raizal community. To start, second generation and younger Raizales seem to be more educated than senior Raizales, which implies that many speakers have studied in

Continental Colombia and abroad. Moreover, the educational system in the Archipelago has adopted Spanish as the language of instruction in public institutions, while private schools have gradually changed to use Spanish for all purposes. Similarly, even though the Costeño dialect is the most frequent, there is a confluence of other regional Colombian dialects. Moreover, television, radio and newspapers are predominantly in Spanish. Since Spanish has pervasively employed for almost all communications in the islands, the impact in rhotic variation is noticeably in line with the monolingual standard. The language-internal mechanisms and the use of Spanish in almost all social contexts point in the direction of stable bilingualism in younger generations. As it seems, decades of imposition of the Colombian system and identity for over 66 years in these Caribbean islands has brought clear implications in the ethnolinguistic vitality of the local Raizal people that is visible in the ‘reverse’ change of non-vibrant rhotics, the level of bilingualism present in young educated Raizales, and the patterns of use and the restructuring of the linguistic conditionings in vibrant rhotic production.

222

CHAPTER 7 CONCLUSION

This project has presented two analyses that examine the language contact situation of a

Western Caribbean Creole coexisting with a national non-lexifier language in the Archipelago of

San Andres, Colombia. The first investigated the acoustic realization of non-vibrant rhotic realization in the three main varieties spoken in the islands: Raizal Creole, Raizal Spanish, and

Continental Spanish. The second focused on the behavior of tap/trill variation in Raizal Spanish and Continental Spanish. The primary objective of these two analyses was to reveal the mechanisms of rhotic variation and potential change across generations of Raizales and determine the influence of the contact languages, Raizal Creole and Continental Spanish, in the rhotic production of Raizal Spanish.

In all, the results of this dissertation have consequences regarding the nature of bilingual rhotics in Spanish, as a well as, the outcomes of contact in the two island communities under study. This chapter reviews the theoretical implications in section 7.1. Then, I address the limitations of this sociolinguistic research in section 7.2. Finally, section 7.3 points out some considerations for the expansion of this study and for future research on similar contact scenarios.

7.1 Theoretical Implications

The evidence posited here suggests transfer of phonetic and phonological features in non- vibrant rhotics, and thus, shows the effects that contact with Spanish has had on generations of

Raizales. The distribution of the datasets show that non-vibrant realizations are more frequent in senior Raizales, while second and younger generations are gradually increasing the production of vibrant rhotics to monolingual rates. Additionally, a comparison of the acoustic measurements best predicting association to a variety suggest that first generation Raizales are linked to the

223

realization of Raizal Creole approximants, while the contrary occurs with younger generations.

The role of bilingualism is consistent with this interpretation. New generations of Raizales have pursued at younger ages education in the Spanish-only education system of the islands. This is seen in the gradual changes of approximant production from second to third generation.

Moreover, the phonological distinction between taps and trills gives further evidence to support the interpretation of more stable bilingualism in younger generations that affects the realization of non-vibrant approximants and clearly distinguish the new generations from the older ones.

However, claiming a change derived by contact might be partially true. Due to the fact that backed realizations of non-vibrant rhotics in other monolingual varieties of Spanish have been documented (Campos-Astorkiza, 2012; Lipski, 2011), and also in bilingual varieties (Balam,

2012; Zimmer, 2011; Hagerty, 1979; Ramos-Pellicia, 2007; Lastra de Suarez, 2006; Sanchez,

1973; Cassano, 1977; Lope-Blanch, 1975), the interpretation of contact-induced change is partially plausible but not compelling. Moreover, the pattern of rhotic variation seems to be gradually converging across generations towards a fronted alveolar realization similar to that of monolingual Spanish, which suggest a reverse change towards a standard variant already present in the inventory of Spanish. Thus, contact seems to be an external mechanism that is affecting the rate of change, rather than being the sole precursor of change. All this evidence points to non-vibrant rhotic change accelerated by contact and maintained by stable bilingualism in younger generations of Raizales due to internal social pressures to produce a normative apico- alveolar approximant as close as possible to the socially-salient features of Spanish rhotics

(Lynch, 2009; Aaron and Hernandez, 2007).

In contrast, we need to address the fact that the constraints for production of vibrant taps and trills do not align with Continental Spanish. This is explained by an external influence

224

exerted on Raizal Spanish different than the monolingual Spanish variety from Coastal Colombia used for comparison in this study. Both rhotic types are converging toward a monolingual standard but are not necessarily behaving the same way as the main immigrant Spanish variety.

The findings of this work seem to corroborate that multiples sources of influence have shaped

Raizal Spanish to its current rhotic production. Lynch (2009) explains that, contrary to morphosyntactic change in contact situations (Silva-Corvalan, 1994), external social factors play a more vital role in phonological phenomena, where linguistic change responds to group identity, normative pressure on the subordinate language, and speaker’s attitude toward the prestigious language. Indeed, although most young bilingual speakers still ethnically identify themselves as

Raizales, fluency in Spanish is seen as a ticket to being educated in the Colombian tertiary education system, and thus, business-oriented and economically successful, which is associated with the pressure of being balanced bilingual in Spanish, without leaving their linguistic and cultural Afro-Caribbean heritage.

Whether these findings signal the first stages of language shift is open to debate. Several authors have suggested that the contact situation between Raizal Creole and Continental Spanish implies a decreolization of the local language, and thus a post-Creole continuum (Forbes, 1989;

Washabaugh, 1977), while other set the alarms to an imminent death of Raizal Creole (Bartens,

2013; Andrade-Arbelaez, 2006; Bartens, 2002). I argue that decreolization process is very unlikely in languages that are not lexically related, which also resonates with previous theoretical approaches (Snow, 2000; Aceto, 1996; Rickford, 1987; Bickerton, 1980; DeCamp, 1971). Snow

(2000) suggests that unstable relationships between a Creole language and a national non-lexifier language, where diglossia is non-existent and higher proportions of bilingualism is present in the

Creole community, is conducive to language shift. The present results on sound variation points

225

toward the preliminary stages of a potential language shift. Fishman’s (1991) Graded

Intergenerational Disruption Scale or GIDS (1991) and its Expanded version or EGIDS (Lewis and Simons 2010) are used to measure the scale of language shift. Bartens (2013) locates Raizal

Creole between stages 5 and 7 in the GIDS, or between the process of developing its own literature at stage 5 to language shift at stage 7, where the disruption of the generational transmission of the language to children is taking place. The EGIDS is more specific between stages 5 to 7 and adds two extra layers at stage 6 including vigorous (Stage 6a) and threatened

(Stage 6b). At stage 6a there is a diglossic situation, where the language is used for oral communication and by all generations, resulting in a sustainable situation. At stage 6b the same situation applies as in 6a, except that the language is losing users instead of signaling a sustainable scenario. Given the results of this study, I place Raizal Creole at stage 6a in Old

Providence and 6b in San Andres. Precisely, the situation in the Archipelago appears to be worse in San Andres, due to the social and economic implications brought about by the overpopulation of this island. Contrary to San Andres, Raizales compose the vast majority of the population of

Old Providence, and Raizal Creole maintains its allocated niche within the Raizal household.

The situation in San Andres is aggravated by the unstable relationship between Raizal Creole and

Continental Spanish, where Raizal Creole is losing contexts of use, and increased by the new generations of more proficient Creole-Spanish bilinguals. Yet, the sense of impeding danger has contributed to exalt the Raizal identity and has supported the struggle for recognition and protection of this ethnic group, which seems to have partially allowed the language to still being transmitted to younger generations, and thus, provisionally halted the processes leading to shift.

7.2 Limitations

This dissertation only focused on one aspect of phonetic variation to account for the processes of sound change in the bilingual speech of Raizales. It is well known that the effects of

226

contact are visible in other areas of language of grammar and in multiple phonological structures

(Thomason, 2001, 2010; Thomason and Kauffman 1989), and as such, only rhoticity was examined in this study. In addition, the analysis of non-rhoticity, which was documented in all varieties, might have enhanced the findings of this dissertation by comparing its behavior with fully produced variants (providing this phonological variant exist in Raizal Creole). Similarly, individual sociophonetic analysis of Raizal Creole, Raizal Spanish, and Continental Spanish might have shed light on the distribution of the acoustic predictors according to extralinguistic factors. This was missing from the sociophonetic analysis of non-vibrant rhotics. This study also provides evidence for a potential language shift in the Archipelago, but the findings only suggest this scenario as potentially possible given the evidence at both the Raizal individuals, community, and the society of the Archipelago. Further analysis on Raizal Spanish variation might provide further evidence that points to the more specific mechanisms involved in language variation and change.

Nonetheless, this work has contributed greatly with research on a little-known contact situation in the Western Caribbean by using quantitative measures to study sound variation in the

Western Caribbean.

7.3 Future directions

This geographical area is ripe for continuing research on language contact phenomena between the diaspora of English-based Creoles of the Western Caribbean and Spanish. There are several Creole communities in the Atlantic coasts of Central American countries, some of which have been studied previously but need to be revisited, and others that need further documentation. Herzfeld (1980), Aceto (1996) and Snow (2000) reported a contact situation in the area of Bocas del Toro, Panama, more precisely in the island of Bastimentos. Herzfeld (1977,

2003), Zimmer (2011) and others (Winkler and Obeng, 2000; Aguilar-Sánchez, 2005) have

227

reported studies on Limon Creole in Costa Rica. The Creole spoken in the Miskito coast of

Nicaragua has been examined by Holm (1982, 1978), Decker and Keener (1998), Bartens

(2009), and the endangered Creole of Rama Cay, an offshoot of Miskito Creole, has also been documented (Assadi, 1983; Craig 1992). The Spanish spoken in Bluefields, Nicaragua, in contact with an English-based Creole and indigenous languages has been also studied (Lopez-Alonzo,

2016). In Honduras, the Creole spoken in Bay Islands has also been investigated (Graham, 2005;

Warantz, 1983). Belizean Creole, in contact with the lexifier language, has been examined in terms of the effects of contact in the bilingual Spanish variety (Balam, 2013) and in terms of its sociolinguistic situation (Escure, 2013). Finally, the first enclave of many English-based Creoles,

Jamaican Creole, which still remains in contact with its lexifier has been extensively investigated

(Lalla and d’Costa, 2009; Hinrichs, 2006; Patrick, 1999; Bailey, 1971, 1966). All these Creole languages share a common English lexifier and history that extends from Western Africa, the transatlantic slave trade, and Caribbean slave posts since colonial times, to its diaspora across the

Atlantic coast of Central America and the islands along the coast, including the Archipelago of

San Andres. Precisely, this study has pointed out one of many cases of contact between these

Creole languages and the superstrate lexically unrelated language of Spanish in contexts where the unstable relationship between these languages has provoked important changes in the local communities. Future research should look into such phenomena in communities in need of revitalization and maintenance efforts, while contributing with theoretical applications to the field. A social network analysis might have enhanced the results of this study by determining the exact nature of the interactions between Raizales and Continentales. This study assumed that demographic data for Costeño immigrants had a direct relationship with the level of interaction between population groups. Although, this might be true for younger Raizales who attend school

228

in centric institutions, older Raizales seem to maintain different social networks that partially exclude Costeño immigrants. As a result, further studies should address this methodological gap.

Thinking at the macro level and in the spirit of enhancing the impact of such research, it seems to me that these Creole languages are lacking additional and current documentation on publicly available platforms. There is a need of an online corpus that encompass the usage of these languages and that historically traces the diaspora of these communities. Projects such as the Slave Voyages Database (https://www.slavevoyages.org/), created through the collaboration of numerous scholars, could be replicated in this part of the Caribbean by tracing the family names and relationships within the communities of the Western Caribbean. Such visionary project (i.e., potential entitled The Western Caribbean Creoles Corpus – WCCC) might have a profound impact in the countries where Spanish and English coexist with Creole languages with the aim of the protection and preservation of the linguistic and cultural heritage of Creole communities.

Methodologically, this research has introduced acoustic approaches to measure the properties of rhoticity in the three varieties under study. I encourage future phonetic research to incorporate acoustic measurements to study language contact scenarios, since it provides a vast array of data to accurately identify the nature of sound variation across varieties and between extralinguistic factors. Moreover, non-rhoticity in Western Caribbean Creole varieties is a phenomenon that merits further study in order to determine whether a non-rhotic variant is present in the phonetic inventory of these languages. In addition, future studies should examine the linguistic landscape in these language contact scenarios. Such studies might shed light on the dynamics of language practices in the public sphere of linguistically disputed territories, while informing results on language variation and change. Precisely, additional linguistics landscape

229

data collected in this scenario will serve to examine the language policies and visibility of the languages in contact (Restrepo-Ramos, 2019).

More research should address potential courses of action for the recognition and protection of multilingual Caribbean enclaves, and thus, other linguistic landscapes, such as

Catalan in Spain, could offer an alternative look on how the politics of language could be addressed differently to recognize the multilingual setting and communities of this Archipelago, thanks to a more stable situation in terms of public display of written signs in touristic and urban settings, and protective language laws and governmental recognition (Martinez-Ibarra, 2016).

This study has contributed to deepen our understanding of sound variation and change in language contact settings (Van Mensel et al, 2012), and opens the path for more research in the

Archipelago and other areas of the Caribbean. This dissertation has revealed the generational outcomes of contact at the phonetic/phonological level in Caribbean islands where a Creole language and Spanish coexist.

230

APPENDIX FIELDWORK AND MATERIALS

Fieldwork

Figure A-1. Informants completing the Jigsaw task (Thoms J., Liao J. & Szuztak A., 2005) during fieldwork in May 2017.

Figure A-2. Informants completing the Diapix task (Baker & Hazan, 2011) during fieldwork in May 2017.

231

Materials

Figure B-1. Frog, Where Are You? (Adapted from Mayer, 1969)

Figure B-2. Jigsaw task. (Adapted from Thoms J., Liao J. & Szuztak A., 2005)

232

Figure B-3. Diapix. (Adapted from Baker & Hazan, 2011)

233

LIST OF REFERENCES

Aaron, J. E. (2006). Variation and change in Spanish future temporal expression: Rates, constraints, and grammaticization (Doctoral dissertation, University of New Mexico). Aaron, J. E. (2010). Pushing the envelope: Looking beyond the variable context. Language variation and change, 22(1), 1-36. Aaron, J. E., & Hernández, J. E. (2007). Quantitative evidence for contact-induced accommodation. Spanish in contact: Policy, social and linguistic inquiries, 329. Aceto, M. (1996). Syntactic innovation in a Caribbean Creole: The Bastimentos variety of Panamanian Creole English. English World-Wide, 17(1), 43–61. Aceto, M. (1999). Looking beyond decreolization as an explanatory model of language change in Creole-speaking communities. JPCL, 14(1), 93–119. Adams, C. (2002). Strong assibilation and prestige: A sociolinguistic study in the Central Valley of Costa Rica. Ph.D. Dissertation, University of California Aguilar‐Sánchez, J. (2005). English in Costa Rica. World Englishes, 24(2), 161-172. Aikhenvald, A. Y. (2006). Grammars in contact: A cross-linguistic perspective. In Aikhenvald, A. & Dixon, R (Eds.). Grammars in contact: A cross-linguistic typology, Oxford University Press, p. 1-54. Andrade-Arbeláez, J. (2006). Estudio sociolingüístico de San Andrés, isla: un aporte a la cultura sanandresana. Cuadernos del Caribe, 4(8), 42-55. Assadi, B. (1983). Rama Cay Creole English. In Holm (ed.), Central American English (Vol. 2). John Benjamins Publishing. 115–122. Baayen, R. H. (2008). Analyzing linguistic data: A practical introduction to statistics using R. Cambridge University Press. Backhaus, P. (Ed.). (2007). Linguistic landscapes: A comparative study of urban multilingualism in Tokyo (Vol. 136). Multilingual matters. Bad Lucky Goat. [The Guardian: Movie Article] Accessed on January 2018. Retrieved from: https://www.theguardian.com/world/2017/nov/09/colombia-film-english-old-providence Bailey, B. L., & Bailey, B. L. (1966). Jamaican Creole syntax. Cambridge University Press. Bailey, B. L. (1971). Jamaican Creole. Pidginization and creolization of languages, 341. Baker, R., & Hazan, V. (2011). DiapixUK: task materials for the elicitation of multiple spontaneous speech dialogs. Behavior Research Methods, 43(3), 761-770. Balam, O. (2013). Neutralization of the intervocalic rhotic contrast in Northern Belizean Spanish. Borealis–An International Journal of Hispanic Linguistics, 2(2), 285-315.

234

Barni, M. & Bagna, C. (2009). A mapping technique and the linguistic landscape. In Linguistic Lanscape: Expanding the Scenery. Eds. Shohamy & Gorter. Routledge, NY. Bartens, A. (2002). Another short note on Creoles in contact with non‐lexifier prestige languages. Journal of Pidgin and Creole Languages, 17(2), 273-278. Bartens, A. (2009). A comparison of the English-based Creoles of Nicaragua and San Andres and Old Providence. Neuphilologische Mitteilungen, 110(3), 299-318. Bartens, A. (2013). San Andres Creole English. The Survey of Pidgin and Creole Languages, 1, 101-114. Berlin, I. (1996). From Creole to African: Atlantic Creoles and the origins of African-American society in mainland North America. William and Mary Quarterly, 53, 251–288. Bickerton, D. (1973). The nature of a Creole continuum. Language, 640-669. Bickerton, D. (1975). Dynamics of a Creole system (pp. 1-164). Cambridge: Cambridge University Press. Bickerton, D. (1980). Decreolization and the Creole continuum. In A. Valdman & A. Highfield (Eds.), Theoretical orientations in Creole studies (pp. 109–127). New York: Academic Press. Blecua, B. (2001). Las vibrantes del español: manifestaciones acústicas y procesos fonéticos. Barcelona: Universitat Autònoma de Barcelona dissertation. Blecua, B., Cicres, J., & Gil, J. (2014). Variación en las róticas del español y su implicación en la identificación del locutor. Revista de Filología Románica, 31, 13. Blommaert, J. (2013). Ethnography, superdiversity and linguistic landscapes: Chronicles of complexity. Bristol, United Kingdom: Multilingual Matters. Boersma, P., & Weenink, D. (2018). Praat: doing phonetics by computer (Version 6.0.40) [Computer program]. Retrieved May, 2018. Borzone de Manrique, A. M. (1980). Manual de fonética acústica. Buenos Aires: Editorial Hachette. Bradley, T. G. (1999). Assibilation in Ecuadorian Spanish. In Formal Perspectives on Romance Linguistics. Selected Papers from the 28th Linguistics Symposium on Romance Languages. Amsterdam, John Benjamins Publishing Company (pp. 57-71). Bradley, T. G. (2004). Gestural timing and rhotic variation in Spanish. Laboratory approaches to Spanish phonology, 7, 195. Bradley, T. G. (2006). Spanish rhotics and Dominican hypercorrect/s. Probus, 18(1), 1-33. Bradley, T. G., & Willis, E. W. (2012). Rhotic variation and contrast in Veracruz Mexican Spanish. Estudios de fonética experimental, 21, 43-74.

235

Bullock, B. E., & Toribio, A. J. (2004). Introduction: Convergence as an emergent property in bilingual speech. Bilingualism: Language and Cognition, 7(2), 91-93. Bybee, J. L., & Hopper, P. J. (Eds.). (2001). Frequency and the emergence of linguistic structure (Vol. 45). John Benjamins Publishing. Campos-Astorkiza, R. (2012). The phonemes of Spanish. In The Handbook of Hispanic Linguistics, 69, 89. Wiley and Sons. Cassano, P. V. (1977). Problems in language borrowing and lending exemplified by American Spanish phonology. Orbis, 36(1), pp. 149-163. Cedergren, H. 1973. The interplay of social and linguistic factors in Panama. Doctoral dissertation, Cornell University. Chambers, Jack K. & Trudgill, Peter J. (1980). Dialectology. Cambridge: Cambridge University Press. Chatfield C., Collins A.J. (1980) Principal component analysis. In Introduction to Multivariate Analysis. Springer, Boston, MA. Colantoni, L. M. (2001). Mergers, chain shifts and dissimilatory processes: Palatals and rhotics in Argentine Spanish (Doctoral dissertation, University of Minnesota). Colantoni, L. (2006). Increasing periodicity to reduce similarity: an acoustic account of deassibilation in rhotics. In Selected proceedings of the 2nd Conference on Laboratory Approaches to Spanish Phonetics and Phonology (pp. 22-34). Somerville, MA: Cascadilla Proceedings Project. Constitución Política de Colombia – Artículo 10 (1991). [Online] Accessed on January 2018. Retrieved from: http://www.corteconstitucional.gov.co/inicio/Constitucion%20politica%20de%20Colomb ia.pdf Coulmas, F. (2009). Linguistic landscaping and the seed of public sphere. In Linguistic Lanscape: Expanding the Scenery. Eds. Shohamy & Gorter. Routledge, NY. Craig, C. G. (1992). Language shift and language death: the case of Rama in Nicaragua. International journal of the sociology of language, 93(1), 11-26. Dalston, R. M. (1975). Acoustic characteristics of English/w, r, l/spoken correctly by young children and adults. The Journal of the Acoustical Society of America, 57(2), 462-469. DANE -The National Administrative Department of Stastistics (2005). 2005 Census of Colombia. [Online] Accessed on October 2017. Available on the Web: http://systema59.dane.gov.co/cgibin/RpWebEngine.exe/PortalAction?&MODE=MAIN& BASE=CG2005BASICO&MAIN=WebServerMain.inl

236

DeCamp, D. (1971). Toward a generative analysis of the post-Creole continuum. In D. Hymes (Ed.), Pidginization and creolization of languages (pp. 349–370). Cambridge, UK: Cambridge University Press. De Friedemann, Nina (1989). Religión y tradición oral en San Andrés y Providencia. In San Andrés y Providencia: Tradiciones culturales y coyuntura política, edited by Isabel Clemente. Bogotá: Ediciones Uniandes. Decreto Presidencial 2762 (1991). [Online] Accessed on April 2019. Retrieved from: http://www.camarasai.org/images/documentos/normatividad-especial/2-Decreto-2762-de- 1991.pdf Decreto Presidencial 2548 (1999). [Online] Accessed on January 2019. Retrieved from: http://www.suin-juriscol.gov.co/viewDocument.asp?id=1891717 Decker, K., & Keener, A. (1998). A Report on the English-Lexifier Creole of Nicaragua, also known as Miskito Coast Creole, with special reference to Bluefields and the Corn Islands. Summer Institute of Linguistics. Díaz-Campos, M. (2008). Variable production of the trill in spontaneous speech: Sociolinguistic implications. In Selected proceedings of the 3rd Conference on Laboratory Approaches to Spanish Phonology (pp. 47-58). Diez-Canseco, S. (1997). Language variation: The influence of speakers' attitudes and gender on sociolinguistic variables in the Spanish of Cusco, Peru (Doctoral dissertation, University of Pittsburgh). DiCanio, C., Haskins Laboratories & SUNY Buffalo (2013) Praat Script for Automatic Extraction of Spectral Moments [Online] Accessed on January 2018. Retrieved from: https://www.acsu.buffalo.edu/~cdicanio/scripts.html Dittmann, M. (1992). El criollo sanandresano: Lengua y cultura. Cali: Universidad del Valle. Edwards, Jay. 1974. African influences on the English of San Andres Island, Colombia. In Pidgins and Creoles: Current Trends and Prospects. D. DeCamp and I. Hancock (eds.) Washington, D.C.: Georgetown University Press, 1–26. Enciso-Patiño, P. (2004). Estado del arte de la etnoeducación en Colombia con énfasis en política pública. Colombia. Ministerio de Educación Nacional. Erker, D. 2010. A subsegmental approach to coda /s/ weakening in Dominican Spanish. International Journal of the Sociology of Language, 203, 9–26. Escure, G. (2013). Belizean Creole. St. Kitts and the Atlantic Creoles: The Texts of Samuel Augustus Mathews in perspective, 92-100. Espy-Wilson, C. Y., Boyce, S. E., Jackson, M., Narayanan, S., & Alwan, A. (2000). Acoustic modeling of American English/r. The Journal of the Acoustical Society of America, 108(1), 343-356.

237

Espy-Wilson, C. Y. (1987). An acoustic-phonetic approach to speech recognition: application to the semivowels. [Online] Accessed on November 2018. Retrieved from: dspace.mit.edu Espy‐Wilson, C. Y. (1992). Acoustic measures for linguistic features distinguishing the semivowels/wjrl/in American English. The Journal of the Acoustical Society of America, 92(2), 736-757. ESRI (2011). ArcGIS Desktop: Release 10. Redlands, CA: Environmental Systems Research Institute. Face, Timothy L. (2006). “Intervocalic Rhotic Pronunciation by Adult Learners of Spanish as a Second Language.” 7th Conference on the Acquisition of Spanish and Portuguese as First and Second Languages. Ed. Carol. A. Klee and Timothy L. Face. Somerville, MA: Cascadilla. 47–58. Print. Ferguson, C. A. (1990). From esses to aitches: identifying pathways of diachronic change. Studies in Typology and Diachrony for Joseph H. Greenberg. Amsterdam/Philadelphia: John Benjamins, 59-78. Ferguson, C. (1959). Diglossia. Word, 15, 325–340. Fishman, J. (1991). Reversing language shift: Theoretical and empirical foundations of assistance to threatened languages. Clevedon, Avon: Multilingual Matters. File-Muriel, R. J. (2009). The role of lexical frequency in the weakening of syllable-final lexical/s/in the Spanish of Barranquilla, Colombia. Hispania, 348-360. Flege, J. E. (1987). The production of “new” and “similar” phones in a foreign language: Evidence for the effect of equivalence classification. Journal of phonetics, 15(1), 47-65. Florez, S (2006). A Study of language attitudes in two Creole-speaking islands: San Andres and Providence (Colombia). Ikala, Revista de lenguaje y cultura, 11(17) pp.119-147. Forbes, Oakley (1989). Aproximaciones sociolingüísticas en torno a la realidad de las lenguas en contacto en las islas de San Andrés y Providencia: Bilingüismo y diglosia." In San Andrés y Providencia: Tradiciones culturales y coyuntura política, edited by Isabel Clemente. Bogotá: Ediciones Uniandes. Forrest, K., Weismer, G., Milenkovic, P., & Dougall, R. N. (1988). Statistical analysis of word‐ initial voiceless obstruents: preliminary data. The Journal of the Acoustical Society of America, 84(1), 115-123. Foulkes, P., & Docherty, G. J. (2000). Another chapter in the story of/r: ‘Labiodental’variants in British English. Journal of sociolinguistics, 4(1), 30-59. García León, D. L. (2014). Minorías lingúísticas en Colombia. Acercamiento desde las políticas públicas y los derechos humanos. Cuadernos de Lingüística Hispánica, 24, 163-182. González Palacios, M. A. (2016). Identity and social mobilization of the raizal people from the San Andres Archipelago in the Colombian Caribbean. Análisis Político, 29(86), 141-158.

238

Google Earth v 7.3.1. (2018). Archipelago of San Andres, Colombia. Lat.12.549969°, Long. - 81.718758°. Viewed January 2018. Graham, R. (2005). Partial creolization, restructuring and convergence in Bay Islands Englishes. English World-Wide, 26(1), 43-76. Grimes, B.F. (2000). Ethnologue: Languages of the world.14th edition. Dallas: SIL International. Grinstead, J., Baron, A., Vega-Mendoza, M., De la Mora, J., Cantú-Sánchez, M., & Flores, B. (2013). Tense marking and spontaneous speech measures in Spanish specific language impairment: A discriminant function analysis. Journal of Speech, Language, and Hearing Research. Gorter, D. (2013). Linguistic landscapes in a multilingual world. Annual Review of Applied Linguistics, 33, 190-212. Hagerty, T. W. (1979). Phonological Analysis of the Spanish of Belize. Unpublished Dissertation, University of California, Los Angeles. Hagiwara, R. E., 1995, Acoustic realizations of American /r/ as produced by women and men. UCLA Working Papers in Phonetics No. 90. Los Angeles, CA: University of California Hammond, Robert. 1999. On the non-occurrence of the phone [r)] in the Spanish sound system. Advances in Hispanic Linguistics, ed. by Javier Gutiérrez-Rexach and Fernando Martínez-Gil, 135-151. Somerville, MA: Cascadilla Press. Heine, B., & Kuteva, T. (2003). On contact-induced grammaticalization. Studies in Language. International Journal sponsored by the Foundation “Foundations of Language”, 27(3), 529-572. Henriksen, N. C., & Willis, E. W. (2010). Acoustic characterization of phonemic trill production in Jerezano Andalusian Spanish. In Selected proceedings of the 4th Conference on Laboratory Approaches to Spanish Phonology (pp. 115-127). Somerville, MA: Cascadilla Proceedings Project. Henriksen, N. (2014). Sociophonetic analysis of phonemic trill variation in two sub-varieties of Peninsular Spanish. Journal of Linguistic Geography, 2(1), 4-24. Henriksen, N. (2015). Acoustic analysis of the rhotic contrast in Chicagoland Spanish: An intergenerational study. Linguistic Approaches to Bilingualism, 5(3), 285-321. Hernandez-Campoy, J. M. (2003). Exposure to contact and the geographical adoption of standard features: Two complementary approaches. Language in Society, 32(2), 227–255. Hernandez-Campoy, J. M., & Conde-Silvestre, J. C. (2005). Sociolinguistic and Geolinguistic Approaches to the Historical Diffusion of Linguistic Innovations: Incipient Standardisation in Late Middle English. International Journal of English Studies, 5(1), 101–134.

239

Herzfeld, A. (1977). Second language acrolect replacement in Limón Creole. Kansas Working Papers in Linguistics, 2, 193–222. Herzfeld, A. (1980). Limon Creole and Panamanian Creole: comparison and contrast. Mid- America Linguistics Conference. [Online] Accessed on February 2019. Retrieved from: kuscholarworks.ku.edu Hickey, R. (2014). The sound structure of Modern Irish (Vol. 47). Walter de Gruyter GmbH & Co KG. Hinrichs, L. (2006). Codeswitching on the web: English and Jamaican Creole in e-mail communication (Vol. 147). John Benjamins Publishing. Holm, J. A. (1978). The Creole English of Nicaragua's Coast: Its Sociolinguistic History and a Comparative Study of Its Lexicon and Syntax (Doctoral dissertation, University College, London). Holm, J. (1982). Central American English (Vol. 2). John Benjamins Publishing. Holm, J. (2000). An introduction to pidgins and Creoles. Cambridge University Press. Holmquist, J. (2011). Gender and Variation: Word-final/s/in Men’s and Women’s Speech in Puerto Rico’s Western Highlands. Handbook of Hispanic Sociolinguistics, 230-243. Huebner, T. (2009). A framework for the linguistic analysis of linguistic landscapes. In Linguistic landscape: Expanding the scenery. Eds. Shohamy & Gorter. Routledge, NY. Hualde, J. I. (2005). The sounds of Spanish. Cambridge: Cambridge University Press. Hult, F. (2009). Language ecology and linguistic landscape analisys. In Linguistic Lanscape: Expanding the Scenery. Eds. Shohamy & Gorter. Routledge, NY. Instituto de Investigaciones Marinas y Costeras “José Benito Vives de Andrés” (2012). Atlas de la reserva de biósfera Seaflower: Archipiélago de San Andrés, Providencia y Santa Catalina. In Gómez-López et al. (Eds.). San Andres Isla, Colombia. Retrieved from: http://www.observatorio.biosferaseaflower.org/index.php/es/publicaciones/item/361- atlas-de-la-reserva-debiosfera-seaflower Johnson, D. E. (2009). Getting off the GoldVarb standard: Introducing Rbrul for mixed‐effects variable rule analysis. Language and linguistics compass, 3(1), 359-383. Jolliffe, I. T. (1986). Principal component analysis and factor analysis. In Principal component analysis (pp. 115-128). Springer, New York, NY. Jongman, A., Wayland, R., & Wong, S. (2000). Acoustic characteristics of English fricatives. The Journal of the Acoustical Society of America, 108(3), 1252-1263. Kawahara, S (2010) Praat Script for Automatic Extraction of Formant Frequencies [Online] Accessed on January 2018. Retrieved from: http://user.keio.ac.jp/~kawahara/resource.html

240

Kent, R. D., Read, C. (2001). The acoustic analysis of speech. Cengage Learning, 2nd Edition. Klee C. & Lynch A. (2009). Capítulo 1: Introducción. El español en contacto con otras lenguas. Georgetown University Press. P. 1-24. Labov, W. (1963). The social motivation of a sound change. Word, 19(3), 273-309. Labov, W. (1966). The social stratification of English in New York city. [Online] Accessed on January 2019. Retrieved from: https://eric.ed.gov/?id=ED012927 Labov, W. (1994). Principles of linguistic change: Internal factors. Oxford, UK: Basil Blackwell. Labov, W. (2001). Principles of linguistic change, Volume 2: Social factors. Language in Society Oxford, 29. Labov, W. (2010). Principles of linguistic change, volume 3: Cognitive and cultural factors (Vol. 36). John Wiley & Sons. Ladefoged, P., & Maddieson, I. (1996). The sounds of the world's languages (Vol. 1012). Oxford: Blackwell. Lalla, B., & d'Costa, J. (2009). Language in exile: Three hundred years of Jamaican Creole. University of Alabama Press. Landry, R., & Bourhis, R. Y. (1997). Linguistic landscape and ethnolinguistic vitality: An empirical study. Journal of language and social psychology, 16(1), 23-49. Lastra de Suárez, Y. (1975). El habla y la educación de los niños de origen mexicano en Los Angeles, in E. Hernández-Chávez, A. Cohen and A. Beltramo (eds.), El lenguaje de los chicanos: regional and social characteristics of language used by Mexican Americans. Arlington, Center for Applied Linguistics, pp. 61-69. Lastra, Y., & Butragueño, P. M. (2006). Un posible cambio en curso: el caso de las vibrantes en la ciudad de México. Estudios sociolingüísticos del español de España y América, 35-68. Leeman, J., & Modan, G. (2009). Commodified language in Chinatown: A contextualized approach to linguistic landscape. Journal of Sociolinguistics, 13(3), 332-362. Lehiste, I. (1964). Acoustical characteristics of selected English consonants. Research Center in Anthropology. Folklore and Linguistics, Indiana University, Bloomington. Lennes, M (2002) Praat Script for Automatic Extraction of Segmental Duration [Online] Accessed on January 2018. Retrieved from: http://phonetics.linguistics.ucla.edu/facilities/acoustic/praat.html Le Page, R. B., & Tabouret-Keller, A. (1985). Acts of identity. Creole-based approaches to language and ethnicity. Cambridge, UK: Cambridge University Press

241

Le Page, R. B., & De Camp, D. (1960). Jamaican Creole: An Historical Introduction to Jamaican Creole: Four Jamaican Creole Texts with Introduction, Phonemic Transcriptions and Glosses. Macmillan. Lewis, M. P., & Simons, G. F. (2010). Assessing endangerment: expanding Fishman’s GIDS. Revue Roumaine de Linguistique, 55(2), 103-120. Lewis, A. M. (2004). Coarticulatory effects on Spanish trill production. In Proceedings of the 2003 Texas Linguistics Society Conference (Vol. 116, p. 127). Ley de lenguas nativas 1381 de 2010. [Online] Accessed on January 2018. Retrieved from: http://www.mincultura.gov.co/areas/poblaciones/publicaciones/Documents/Ley%20de%2 0Lenguas%202013.pdf Ley 47 de 1993. [Online] Accessed on January 2018. Retrieved from: https://www.ramajudicial.gov.co/documents/573203/887553/Ley+47+de+1993+%28Org aniza+el+Archipielago+de+San+Andre%CC%81s%29.pdf/7fd85fe1-5581-43e1-a89a- 1e82df9fc4e6 Liaw, A., & Wiener, M. (2002). Classification and Regression by random Forest. R News, 2(3), 18--22.

Lipski, J. (1994). Latin American Spanish. Addison-Wesley. Lipski, J (2009). Afro-Choteño speech: Towards the (re)creation of ‘Black Spanish’. Negritud, 2, 99–120.

Lipski, J. (2011). Socio-phonological variation in Latin American Spanish. The Handbook of Hispanic Sociolinguistics, 72-97. Lope Blanch, J. M. (1975). Un caso de posible influencia maya en el español mexicano. Nueva Revista de Filología Hispánica, 24, 89-100. Lopez Alonzo, K. I. (2016). Rhotic production in the Spanish of Bluefields, Nicaragua, a language contact situation (Doctoral dissertation, The Ohio State University). Lou, J. J. (2016). The linguistic landscape of Chinatown: A sociolinguistic ethnography. Bristol, United Kingdom: Multilingual Matters. Lynch, A. (2009). A sociolinguistic analysis of final/s/in Miami Cuban Spanish. Language Sciences, 31(6), 766-790. Marten, H. F., Van Mensel, L., & Gorter, D. (2012). Studying minority languages in the linguistic landscape. In Minority Languages in the Linguistic Landscape (pp. 1-15). Palgrave Macmillan UK. Martínez-Ibarra, F. M. (2016). Spanish and Valencian in contact. In Sessarego, S. & Tejedo- Herrero, F. (Eds.), Spanish Language and Sociolinguistic Analysis, 8, (pp.135-153). John Benjamin Publishing.

242

Mayer, M. (1969). Frog, where are you? New York: Dial Press. McWhorter, J. (1997). It happened at Cormantin: Locating the origin of the Atlantic English- based Creoles. Journal of Pidgin and Creole Languages, 12(1), 59-102. McWhorter, J. (2007). Language interrupted: Signs of non-native acquisition in standard language grammars. Oxford University Press. Mendoza, E., Carballo, A., Cruz, M, Muñoz, J., & Marrero, V. (2003). Temporal variability in speech segments of Spanish: Context and speaker related differences. Speech Communication, 40. 431-47. Meyerhoff, M. (2009). Replication, transfer, and calquing: Using variation as a tool in the study of language contact. Language Variation and Change, 21(3), 297-317. Meyerhoff, M., Adachi, C., Nanbakhsh, G., & Strycharz, A. (2012). Sociolinguistic fieldwork. In Thieberger, N (Ed.), The Oxford Handbook of Linguistic Fieldwork. Oxford. Moreno-Fernández, F. (2008). Introduction: a sociolinguistic panorama of Spanish in Spain. International Journal of the Sociology of Language, 2008(193/194), 13-20. Morgan, T. A., & Sessarego, S. (2016). A phonetic analysis of intervocalic/r/in Highland Bolivian Spanish. Spanish in Context, 13(2), 195-211. Morren R. (2001). Creole-based trilingual education in the Caribbean archipelago of San Andres, Providence and Santa Catalina. Journal of Multilingual and Multicultural Development, 22(3), 227-241. Moya-Chaves, D. S. (2014). La situación sociolingüística de la lengua Creole de San Andrés Isla: el caso de San Luis. Colombian Applied Linguistics Journal, 16(1), 55-66. Navarro-Tomás, T. (1999). Manual de pronunciación española. Madrid: Consejo Superior de Investigaciones Científicas. Nerbonne, J., Colen, R., Gooskens, C., Kleiweg, P., & Leinonen, T. (2011). Gabmap-a web application for dialectology. Dialectologia: Revista Electrònica, 65-89. Olsen, M. K. (2012). The L2 acquisition of Spanish rhotics by L1 English speakers: The effect of L1 articulatory routines and phonetic context for allophonic variation. Hispania, 95(1), 65-82. Patiño Rosselli, C. (1992). La criollística y las lenguas criollas de Colombia. Thesaurus: Boletín del Instituto Caro y Cuervo, 47(2), 233-264. Patrick, P. L. (1999). Urban Jamaican Creole: variation in the mesolect. John Benjamins Publishing. Penny, R. J. (2000). Variation and change in Spanish. Cambridge University Press. Picard, M. (1997). English Flapping and the feature [vibrant]. English Language & Linguistics, 1(2), 285-294.

243

Poplack, S. & Tagliamonte, S. (1999) The grammaticalization of going to in (African American) English. Language Variation and Change, 11(3), 315–342. Poulsen, J., & French, A. (2008). Discriminant function analysis. [Online] Accessed on November 2018. Retrieved from: https://pdfs.semanticscholar.org/8e10/85b33d84d7015174e30e3e911272b6e1c2fe.pdf Quilis, A., & Carril, R. B. (1971). Análisis acústico de [ř] en algunas zonas de Hispanoamérica. Revista de Filología Española, 54(3/4), 271-316. Ramos-Pellicia, M. F. (2007). Lorain Puerto Rican Spanish and ‘r’ in Three Generations, in J. Holmquist, A. Lorenzino & L. Sayahi (eds.), Selected Proceedings of the Third Workshop on Spanish Sociolinguistics. Somerville, MA, Cascadilla Proceedings Project, pp. 53-60. Ranocchiari, D., & Calabresi, G. (2016). Ethnicity and Religion in the Archipelago of San Andrés, Providencia and Santa Catalina. Bulletin of Latin American Research, 35(4), 481-495. R Core Team (2013). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/. Recasens, D., & Pallarès, M. D. (1999). A study of/J/and/r/in the light of the “DAC” coarticulation model. Journal of Phonetics, 27(2), 143-169. Restrepo Ramos, F. (2019). Public signage in a Multilingual Caribbean enclave: The linguistic landscape of Old Providence and Santa Catalina, Colombia. In Pascual y Cabo D., Elola I. (Eds.), Issues in Hispanic and Lusophone Linguistics. Benjamin Publishings. (In press) Rickford, J. (1987). Dimensions of a Creole continuum: History, texts, and analysis of Guyanese Creole. Stanford, CA: Stanford University Press. Ross, J. (2007). Routes for roots: entering the 21st century in San Andres Island, Colombia. Caribbean Studies, 3-36. Sánchez, R. (1973). Nuestra circunstancia lingüística. Voices. Readings from El Grito. A Journal of Mexican American Thought 1967-1973. Berkeley, Quinto Sol Publications, pp. 420- 449. Sankoff, G. (2002). Linguistic outcomes of language contact. In J. K. Chambers, Peter Trudgill and Natalie Schilling-Estes (Eds.), The Handbook of Language Variation and Change. Wiley Online Library. Sankoff, D., & Thibault, P. (1981). Weak complementarity: Tense and aspect in Montreal French. Syntactic change, 25, 205-216. Sankoff, D. (1988). Sociolinguistics and syntactic variation. Linguistics: The Cambridge survey, 4, 140-161.

244

Silva-Corvalán, C. (1994). Language Contact and Change: Spanish in Los Angeles. Oxford University Press. Slave Voyages Database [Online] Accessed on March 2019. Retrieved from the Web: https://www.slavevoyages.org/ Snow, P. (2000). Caribbean Creole/non-lexifier contact situations: A provisional survey. Journal of Pidgin and Creole languages, 15(2), 339-343. Solé, Maria-Josep. 2002. Aerodynamic characteristics of trills and phonological patterning. Journal of Phonetics, 30, 655-88. Spolsky, B. (2009). Prolegomena to a sociolinguistic theory of public signage. In Shohamy, E., & Gorter, D. (Eds.), Linguistic Landscape: Expanding the Scenery (pp. 25-39). Routledge, NY. Spruit, M. R. (2006). Measuring syntactic variation in Dutch dialects. Literary and linguistic computing, 21(4), 493-506. Szabó, T. P., & Troyer, R. (2017). Inclusive ethnographies: Beyond the binaries of observer and observed in linguistic landscape studies. Linguistic Landscape, 3, 306-326. Tabachnick, B. G., & Fidell, L. S. (1996). Using multivariate statistics. 3rd edition. NY: Harper Collins. Tagliamonte, S. A. (1997). Obsolescence in the English perfect? Evidence from Samaná English. American Speech, 33-68. Tagliamonte, S. A. (2012). Variationist Sociolinguistics: Change. Observation, Interpretation. John Wiley & Sons. Thomas, E. (2011). Sociophonetics: An introduction. Palgrave Macmillan. Thomas, J., Jianling L, & Anja S. (2005). The Use of L1 in a L2 Online Chat Activity. Canadian Modern Language Review/Revue canadienne des langues vivantes, 62(1). Thomason, S., & Terrence K. (1988). Language contact, creolization, and genetic linguistics. Berkeley: U. of California Press. Thomason, S. G. (2001). Language contact: An introduction. Georgetown University Press. Thomason, S. (2010). Contact explanations in linguistics. In Hickey, R (Ed.), The handbook of language contact. John Wiley & Sons. Van Coetsem, F. (1988). Loan phonology and the two transfer types in language contact (Vol. 27). Walter de Gruyter GmbH & Co KG. Vásquez Carranza, L. M. (2006). On the phonetic realization and distribution of Costa Rican Rhotics. Filología y Lingüística, 32(2), pp. 291-309.

245

Van Mensel, L., Marten, H. F., & Gorter, D. (2012). Minority languages through the lens of the linguistic landscape. In Minority Languages in the Linguistic Landscape. (pp. 319-323). Palgrave Macmillan UK. Warantz, E. (1983). The Bay Islands English of Honduras. In Holm (ed.), Central American English (Vol. 2). John Benjamins Publishing. 71-94. Wardhaugh, R. (2006). An Introduction to Sociolinguistics. Oxford, Blackwell Publishing. Washabaugh, W. (1974). Variability in decreolization on Providence Island, Colombia. Unpublished doctoral dissertation, Wayne State University, Detroit, MI. Washabaugh, W. (1977). Constraining variation in decreolization. Language, 329-352. Weinreich, U. (1953). Languages in contact. The Hague: Mouton. Weinreich, U., Labov, W., & Herzog, M. (1968). Empirical foundations for a theory of language change. In Winfred P. Lehmann and Yakov Malkiel (eds.), Directions for Historical Linguistics (95-195). University of Texas Press. Weissglass, C. (2011). An acoustic study of rhotics in onset clusters in La Rioja. In Romance Linguistics 2010: Selected Papers from the 40th Linguistic Symposium on Romance Languages (LSRL), Seattle, Washington, March 2010 (Vol. 318, p. 49). John Benjamins Publishing. Wickham, H. (2017). Tidyverse: Easily Install and Load the 'Tidyverse'. R package version 1.2.1. https://CRAN.R-project.org/package=tidyverse Willis, E. W. (2006). Trill variation in Dominican Spanish: An acoustic examination and comparative analysis. In Selected proceedings of the 9th Hispanic Linguistics Symposium (Vol. 1, pp. 121-131). Somerville, MA: Cascadilla Proceedings Project. Willis, E. W. (2007). An acoustic study of the ‘pre-aspirated trill’in narrative Cibaeño Dominican Spanish. Journal of the International Phonetic Association, 37(01), 33-49. Willis, E. W., & Bradley, T. G. (2008). Contrast maintenance of taps and trills in Dominican Spanish: Data and analysis. In Selected proceedings of the 3rd Conference on Laboratory Approaches to Spanish Phonology (pp. 87-100). Somerville, MA: Cascadilla Proceedings Project. Winford, D. (2003). An introduction to contact linguistics. Malden, MA: Blackwell Pub. Winford, D. (2005). Contact-induced changes: Classification and processes. Diachronica, 22(2), 373-427. Winkler, E., & Obeng, S. (2000). West Africanisms in Limonese Creole English. World Englishes, 19(2), 155-171. Zahler, S., & Daidone, D. (2014). A variationist account of trill/r/usage in the Spanish of Málaga. IULC Working Papers, 14(2).

246

Zimmer, T. (2011). El español hablado por los afrocostarricenses: estudio lingüístico y sociolingüístico. Kassel university press GmbH. Zhou, X., Espy-Wilson, C. Y., Boyce, S., Tiede, M., Holland, C., & Choe, A. (2008). A magnetic resonance imaging-based articulatory and acoustic study of “retroflex” and “bunched” American English/r. The Journal of the Acoustical Society of America, 123(6), 4466-4481.

247

BIOGRAPHICAL SKETCH

Falcon Restrepo is a doctoral candidate in Hispanic Linguistics in the Department of

Spanish and Portuguese Studies at the University of Florida. He has received an M.A. in

Linguistics and TESOL from West Virginia University and a B.A. in Translation from

Universidad de Antioquia, Colombia. Besides his work on the Caribbean Archipelago of San

Andres, his has been recently involved in projects on language policies in the linguistic landscape of contact scenarios, the Voseo in Latin America, and research on Spanish L2 acquistion and teaching. He is currently a Visiting Assistant Professor at the University of

Central Florida for the academic year of 2019-2020.

248